Index: head/sys/contrib/ipfilter/netinet/ip_rpcb_pxy.c =================================================================== --- head/sys/contrib/ipfilter/netinet/ip_rpcb_pxy.c (revision 336675) +++ head/sys/contrib/ipfilter/netinet/ip_rpcb_pxy.c (revision 336676) @@ -1,1475 +1,1475 @@ /* * Copyright (C) 2002-2012 by Ryan Beasley * * See the IPFILTER.LICENCE file for details on licencing. */ /* * Overview: * This is an in-kernel application proxy for Sun's RPCBIND (nee portmap) * protocol as defined in RFC1833. It is far from complete, mostly * lacking in less-likely corner cases, but it's definitely functional. * * Invocation: * rdr /32 port -> port udp proxy rpcbu * * If the host running IP Filter is the same as the RPC server, it's * perfectly legal for both the internal and external addresses and ports * to match. * * When triggered by appropriate IP NAT rules, this proxy works by * examining data contained in received packets. Requests and replies are * modified, NAT and state table entries created, etc., as necessary. */ /* * TODO / NOTES * * o Must implement locking to protect proxy session data. * o Fragmentation isn't supported. * o Only supports UDP. * o Doesn't support multiple RPC records in a single request. * o Errors should be more fine-grained. (e.g., malloc failure vs. * illegal RPCB request / reply) * o Even with the limit on the total amount of recorded transactions, * should there be a timeout on transaction removal? * o There is a potential collision between cloning, wildcard NAT and * state entries. There should be an appr_getport routine for * to avoid this. * o The enclosed hack of STREAMS support is pretty sick and most likely * broken. * * $Id$ */ #define IPF_RPCB_PROXY /* * Function prototypes */ void ipf_p_rpcb_main_load __P((void)); void ipf_p_rpcb_main_unload __P((void)); int ipf_p_rpcb_new __P((void *, fr_info_t *, ap_session_t *, nat_t *)); void ipf_p_rpcb_del __P((ipf_main_softc_t *, ap_session_t *)); int ipf_p_rpcb_in __P((void *, fr_info_t *, ap_session_t *, nat_t *)); int ipf_p_rpcb_out __P((void *, fr_info_t *, ap_session_t *, nat_t *)); static void ipf_p_rpcb_flush __P((rpcb_session_t *)); static int ipf_p_rpcb_decodereq __P((fr_info_t *, nat_t *, rpcb_session_t *, rpc_msg_t *)); static int ipf_p_rpcb_skipauth __P((rpc_msg_t *, xdr_auth_t *, u_32_t **)); static int ipf_p_rpcb_insert __P((rpcb_session_t *, rpcb_xact_t *)); static int ipf_p_rpcb_xdrrpcb __P((rpc_msg_t *, u_32_t *, rpcb_args_t *)); static int ipf_p_rpcb_getuaddr __P((rpc_msg_t *, xdr_uaddr_t *, u_32_t **)); static u_int ipf_p_rpcb_atoi __P((char *)); static int ipf_p_rpcb_modreq __P((fr_info_t *, nat_t *, rpc_msg_t *, mb_t *, u_int)); static int ipf_p_rpcb_decoderep __P((fr_info_t *, nat_t *, rpcb_session_t *, rpc_msg_t *, rpcb_xact_t **)); static rpcb_xact_t * ipf_p_rpcb_lookup __P((rpcb_session_t *, u_32_t)); static void ipf_p_rpcb_deref __P((rpcb_session_t *, rpcb_xact_t *)); static int ipf_p_rpcb_getproto __P((rpc_msg_t *, xdr_proto_t *, u_32_t **)); static int ipf_p_rpcb_getnat __P((fr_info_t *, nat_t *, u_int, u_int)); static int ipf_p_rpcb_modv3 __P((fr_info_t *, nat_t *, rpc_msg_t *, mb_t *, u_int)); static int ipf_p_rpcb_modv4 __P((fr_info_t *, nat_t *, rpc_msg_t *, mb_t *, u_int)); static void ipf_p_rpcb_fixlen __P((fr_info_t *, int)); /* * Global variables */ static frentry_t rpcbfr; /* Skeleton rule for reference by entities this proxy creates. */ -static VNET_DEFINE(int, rpcbcnt); +VNET_DEFINE_STATIC(int, rpcbcnt); #define V_rpcbcnt VNET(rpcbcnt) /* Upper bound of allocated RPCB sessions. */ /* XXX rpcbcnt still requires locking. */ static int rpcb_proxy_init = 0; /* * Since rpc_msg contains only pointers, one should use this macro as a * handy way to get to the goods. (In case you're wondering about the name, * this started as BYTEREF -> BREF -> B.) */ #define B(r) (u_32_t)ntohl(*(r)) /* * Public subroutines */ /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_main_load */ /* Returns: void */ /* Parameters: (void) */ /* */ /* Initialize the filter rule entry and session limiter. */ /* -------------------------------------------------------------------- */ void ipf_p_rpcb_main_load() { V_rpcbcnt = 0; bzero((char *)&rpcbfr, sizeof(rpcbfr)); rpcbfr.fr_ref = 1; rpcbfr.fr_flags = FR_PASS|FR_QUICK|FR_KEEPSTATE; MUTEX_INIT(&rpcbfr.fr_lock, "ipf Sun RPCB proxy rule lock"); rpcb_proxy_init = 1; } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_main_unload */ /* Returns: void */ /* Parameters: (void) */ /* */ /* Destroy rpcbfr's mutex to avoid a lock leak. */ /* -------------------------------------------------------------------- */ void ipf_p_rpcb_main_unload() { if (rpcb_proxy_init == 1) { MUTEX_DESTROY(&rpcbfr.fr_lock); rpcb_proxy_init = 0; } } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_new */ /* Returns: int - -1 == failure, 0 == success */ /* Parameters: fin(I) - pointer to packet information */ /* aps(I) - pointer to proxy session structure */ /* nat(I) - pointer to NAT session structure */ /* */ /* Allocate resources for per-session proxy structures. */ /* -------------------------------------------------------------------- */ int ipf_p_rpcb_new(arg, fin, aps, nat) void *arg; fr_info_t *fin; ap_session_t *aps; nat_t *nat; { rpcb_session_t *rs; nat = nat; /* LINT */ if (fin->fin_v != 4) return -1; KMALLOC(rs, rpcb_session_t *); if (rs == NULL) return(-1); bzero((char *)rs, sizeof(*rs)); MUTEX_INIT(&rs->rs_rxlock, "ipf Sun RPCB proxy session lock"); aps->aps_data = rs; return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_del */ /* Returns: void */ /* Parameters: aps(I) - pointer to proxy session structure */ /* */ /* Free up a session's list of RPCB requests. */ /* -------------------------------------------------------------------- */ void ipf_p_rpcb_del(softc, aps) ipf_main_softc_t *softc; ap_session_t *aps; { rpcb_session_t *rs; rs = (rpcb_session_t *)aps->aps_data; MUTEX_ENTER(&rs->rs_rxlock); ipf_p_rpcb_flush(rs); MUTEX_EXIT(&rs->rs_rxlock); MUTEX_DESTROY(&rs->rs_rxlock); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_in */ /* Returns: int - APR_ERR(1) == drop the packet, */ /* APR_ERR(2) == kill the proxy session, */ /* else change in packet length (in bytes) */ /* Parameters: fin(I) - pointer to packet information */ /* ip(I) - pointer to packet header */ /* aps(I) - pointer to proxy session structure */ /* nat(I) - pointer to NAT session structure */ /* */ /* Given a presumed RPCB request, perform some minor tests and pass off */ /* for decoding. Also pass packet off for a rewrite if necessary. */ /* -------------------------------------------------------------------- */ int ipf_p_rpcb_in(arg, fin, aps, nat) void *arg; fr_info_t *fin; ap_session_t *aps; nat_t *nat; { rpc_msg_t rpcmsg, *rm; rpcb_session_t *rs; u_int off, dlen; mb_t *m; int rv; /* Disallow fragmented or illegally short packets. */ if ((fin->fin_flx & (FI_FRAG|FI_SHORT)) != 0) return(APR_ERR(1)); /* Perform basic variable initialization. */ rs = (rpcb_session_t *)aps->aps_data; m = fin->fin_m; off = (char *)fin->fin_dp - (char *)fin->fin_ip; off += sizeof(udphdr_t) + fin->fin_ipoff; dlen = fin->fin_dlen - sizeof(udphdr_t); /* Disallow packets outside legal range for supported requests. */ if ((dlen < RPCB_REQMIN) || (dlen > RPCB_REQMAX)) return(APR_ERR(1)); /* Copy packet over to convenience buffer. */ rm = &rpcmsg; bzero((char *)rm, sizeof(*rm)); COPYDATA(m, off, dlen, (caddr_t)&rm->rm_msgbuf); rm->rm_buflen = dlen; /* Send off to decode request. */ rv = ipf_p_rpcb_decodereq(fin, nat, rs, rm); switch(rv) { case -1: return(APR_ERR(1)); /*NOTREACHED*/ break; case 0: break; case 1: rv = ipf_p_rpcb_modreq(fin, nat, rm, m, off); break; default: /*CONSTANTCONDITION*/ IPF_PANIC(1, ("illegal rv %d (ipf_p_rpcb_req)", rv)); } return(rv); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_out */ /* Returns: int - APR_ERR(1) == drop the packet, */ /* APR_ERR(2) == kill the proxy session, */ /* else change in packet length (in bytes) */ /* Parameters: fin(I) - pointer to packet information */ /* ip(I) - pointer to packet header */ /* aps(I) - pointer to proxy session structure */ /* nat(I) - pointer to NAT session structure */ /* */ /* Given a presumed RPCB reply, perform some minor tests and pass off */ /* for decoding. If the message indicates a successful request with */ /* valid addressing information, create NAT and state structures to */ /* allow direct communication between RPC client and server. */ /* -------------------------------------------------------------------- */ int ipf_p_rpcb_out(arg, fin, aps, nat) void *arg; fr_info_t *fin; ap_session_t *aps; nat_t *nat; { rpc_msg_t rpcmsg, *rm; rpcb_session_t *rs; rpcb_xact_t *rx; u_int off, dlen; int rv, diff; mb_t *m; /* Disallow fragmented or illegally short packets. */ if ((fin->fin_flx & (FI_FRAG|FI_SHORT)) != 0) return(APR_ERR(1)); /* Perform basic variable initialization. */ rs = (rpcb_session_t *)aps->aps_data; rx = NULL; m = fin->fin_m; off = (char *)fin->fin_dp - (char *)fin->fin_ip; off += sizeof(udphdr_t) + fin->fin_ipoff; dlen = fin->fin_dlen - sizeof(udphdr_t); diff = 0; /* Disallow packets outside legal range for supported requests. */ if ((dlen < RPCB_REPMIN) || (dlen > RPCB_REPMAX)) return(APR_ERR(1)); /* Copy packet over to convenience buffer. */ rm = &rpcmsg; bzero((char *)rm, sizeof(*rm)); COPYDATA(m, off, dlen, (caddr_t)&rm->rm_msgbuf); rm->rm_buflen = dlen; rx = NULL; /* XXX gcc */ /* Send off to decode reply. */ rv = ipf_p_rpcb_decoderep(fin, nat, rs, rm, &rx); switch(rv) { case -1: /* Bad packet */ if (rx != NULL) { MUTEX_ENTER(&rs->rs_rxlock); ipf_p_rpcb_deref(rs, rx); MUTEX_EXIT(&rs->rs_rxlock); } return(APR_ERR(1)); /*NOTREACHED*/ break; case 0: /* Negative reply / request rejected */ break; case 1: /* Positive reply */ /* * With the IP address embedded in a GETADDR(LIST) reply, * we'll need to rewrite the packet in the very possible * event that the internal & external addresses aren't the * same. (i.e., this box is either a router or rpcbind * only listens on loopback.) */ if (nat->nat_odstaddr != nat->nat_ndstaddr) { if (rx->rx_type == RPCB_RES_STRING) diff = ipf_p_rpcb_modv3(fin, nat, rm, m, off); else if (rx->rx_type == RPCB_RES_LIST) diff = ipf_p_rpcb_modv4(fin, nat, rm, m, off); } break; default: /*CONSTANTCONDITION*/ IPF_PANIC(1, ("illegal rv %d (ipf_p_rpcb_decoderep)", rv)); } if (rx != NULL) { MUTEX_ENTER(&rs->rs_rxlock); /* XXX Gross hack - I'm overloading the reference * counter to deal with both threads and retransmitted * requests. One deref signals that this thread is * finished with rx, and the other signals that we've * processed its reply. */ ipf_p_rpcb_deref(rs, rx); ipf_p_rpcb_deref(rs, rx); MUTEX_EXIT(&rs->rs_rxlock); } return(diff); } /* * Private support subroutines */ /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_flush */ /* Returns: void */ /* Parameters: rs(I) - pointer to RPCB session structure */ /* */ /* Simply flushes the list of outstanding transactions, if any. */ /* -------------------------------------------------------------------- */ static void ipf_p_rpcb_flush(rs) rpcb_session_t *rs; { rpcb_xact_t *r1, *r2; r1 = rs->rs_rxlist; if (r1 == NULL) return; while (r1 != NULL) { r2 = r1; r1 = r1->rx_next; KFREE(r2); } } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_decodereq */ /* Returns: int - -1 == bad request or critical failure, */ /* 0 == request successfully decoded, */ /* 1 == request successfully decoded; requires */ /* address rewrite/modification */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT session structure */ /* rs(I) - pointer to RPCB session structure */ /* rm(I) - pointer to RPC message structure */ /* */ /* Take a presumed RPCB request, decode it, and store the results in */ /* the transaction list. If the internal target address needs to be */ /* modified, store its location in ptr. */ /* WARNING: It's the responsibility of the caller to make sure there */ /* is enough room in rs_buf for the basic RPC message "preamble". */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_decodereq(fin, nat, rs, rm) fr_info_t *fin; nat_t *nat; rpcb_session_t *rs; rpc_msg_t *rm; { rpcb_args_t *ra; u_32_t xdr, *p; rpc_call_t *rc; rpcb_xact_t rx; int mod; p = (u_32_t *)rm->rm_msgbuf; mod = 0; bzero((char *)&rx, sizeof(rx)); rc = &rm->rm_call; rm->rm_xid = p; rx.rx_xid = B(p++); /* Record this message's XID. */ /* Parse out and test the RPC header. */ if ((B(p++) != RPCB_CALL) || (B(p++) != RPCB_MSG_VERSION) || (B(p++) != RPCB_PROG)) return(-1); /* Record the RPCB version and procedure. */ rc->rc_vers = p++; rc->rc_proc = p++; /* Bypass RPC authentication stuff. */ if (ipf_p_rpcb_skipauth(rm, &rc->rc_authcred, &p) != 0) return(-1); if (ipf_p_rpcb_skipauth(rm, &rc->rc_authverf, &p) != 0) return(-1); /* Compare RPCB version and procedure numbers. */ switch(B(rc->rc_vers)) { case 2: /* This proxy only supports PMAP_GETPORT. */ if (B(rc->rc_proc) != RPCB_GETPORT) return(-1); /* Portmap requests contain four 4 byte parameters. */ if (RPCB_BUF_EQ(rm, p, 16) == 0) return(-1); p += 2; /* Skip requested program and version numbers. */ /* Sanity check the requested protocol. */ xdr = B(p); if (!(xdr == IPPROTO_UDP || xdr == IPPROTO_TCP)) return(-1); rx.rx_type = RPCB_RES_PMAP; rx.rx_proto = xdr; break; case 3: case 4: /* GETADDRLIST is exclusive to v4; GETADDR for v3 & v4 */ switch(B(rc->rc_proc)) { case RPCB_GETADDR: rx.rx_type = RPCB_RES_STRING; rx.rx_proto = (u_int)fin->fin_p; break; case RPCB_GETADDRLIST: if (B(rc->rc_vers) != 4) return(-1); rx.rx_type = RPCB_RES_LIST; break; default: return(-1); } ra = &rc->rc_rpcbargs; /* Decode the 'struct rpcb' request. */ if (ipf_p_rpcb_xdrrpcb(rm, p, ra) != 0) return(-1); /* Are the target address & port valid? */ if ((ra->ra_maddr.xu_ip != nat->nat_ndstaddr) || (ra->ra_maddr.xu_port != nat->nat_ndport)) return(-1); /* Do we need to rewrite this packet? */ if ((nat->nat_ndstaddr != nat->nat_odstaddr) || (nat->nat_ndport != nat->nat_odport)) mod = 1; break; default: return(-1); } MUTEX_ENTER(&rs->rs_rxlock); if (ipf_p_rpcb_insert(rs, &rx) != 0) { MUTEX_EXIT(&rs->rs_rxlock); return(-1); } MUTEX_EXIT(&rs->rs_rxlock); return(mod); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_skipauth */ /* Returns: int -- -1 == illegal auth parameters (lengths) */ /* 0 == valid parameters, pointer advanced */ /* Parameters: rm(I) - pointer to RPC message structure */ /* auth(I) - pointer to RPC auth structure */ /* buf(IO) - pointer to location within convenience buffer */ /* */ /* Record auth data length & location of auth data, then advance past */ /* it. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_skipauth(rm, auth, buf) rpc_msg_t *rm; xdr_auth_t *auth; u_32_t **buf; { u_32_t *p, xdr; p = *buf; /* Make sure we have enough space for expected fixed auth parms. */ if (RPCB_BUF_GEQ(rm, p, 8) == 0) return(-1); p++; /* We don't care about auth_flavor. */ auth->xa_string.xs_len = p; xdr = B(p++); /* Length of auth_data */ /* Test for absurdity / illegality of auth_data length. */ if ((XDRALIGN(xdr) < xdr) || (RPCB_BUF_GEQ(rm, p, XDRALIGN(xdr)) == 0)) return(-1); auth->xa_string.xs_str = (char *)p; p += XDRALIGN(xdr); /* Advance our location. */ *buf = (u_32_t *)p; return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_insert */ /* Returns: int -- -1 == list insertion failed, */ /* 0 == item successfully added */ /* Parameters: rs(I) - pointer to RPCB session structure */ /* rx(I) - pointer to RPCB transaction structure */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_insert(rs, rx) rpcb_session_t *rs; rpcb_xact_t *rx; { rpcb_xact_t *rxp; rxp = ipf_p_rpcb_lookup(rs, rx->rx_xid); if (rxp != NULL) { ++rxp->rx_ref; return(0); } if (V_rpcbcnt == RPCB_MAXREQS) return(-1); KMALLOC(rxp, rpcb_xact_t *); if (rxp == NULL) return(-1); bcopy((char *)rx, (char *)rxp, sizeof(*rx)); if (rs->rs_rxlist != NULL) rs->rs_rxlist->rx_pnext = &rxp->rx_next; rxp->rx_pnext = &rs->rs_rxlist; rxp->rx_next = rs->rs_rxlist; rs->rs_rxlist = rxp; rxp->rx_ref = 1; ++V_rpcbcnt; return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_xdrrpcb */ /* Returns: int -- -1 == failure to properly decode the request */ /* 0 == rpcb successfully decoded */ /* Parameters: rs(I) - pointer to RPCB session structure */ /* p(I) - pointer to location within session buffer */ /* rpcb(O) - pointer to rpcb (xdr type) structure */ /* */ /* Decode a XDR encoded rpcb structure and record its contents in rpcb */ /* within only the context of TCP/UDP over IP networks. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_xdrrpcb(rm, p, ra) rpc_msg_t *rm; u_32_t *p; rpcb_args_t *ra; { if (!RPCB_BUF_GEQ(rm, p, 20)) return(-1); /* Bypass target program & version. */ p += 2; /* Decode r_netid. Must be "tcp" or "udp". */ if (ipf_p_rpcb_getproto(rm, &ra->ra_netid, &p) != 0) return(-1); /* Decode r_maddr. */ if (ipf_p_rpcb_getuaddr(rm, &ra->ra_maddr, &p) != 0) return(-1); /* Advance to r_owner and make sure it's empty. */ if (!RPCB_BUF_EQ(rm, p, 4) || (B(p) != 0)) return(-1); return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_getuaddr */ /* Returns: int -- -1 == illegal string, */ /* 0 == string parsed; contents recorded */ /* Parameters: rm(I) - pointer to RPC message structure */ /* xu(I) - pointer to universal address structure */ /* p(IO) - pointer to location within message buffer */ /* */ /* Decode the IP address / port at p and record them in xu. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_getuaddr(rm, xu, p) rpc_msg_t *rm; xdr_uaddr_t *xu; u_32_t **p; { char *c, *i, *b, *pp; u_int d, dd, l, t; char uastr[24]; /* Test for string length. */ if (!RPCB_BUF_GEQ(rm, *p, 4)) return(-1); xu->xu_xslen = (*p)++; xu->xu_xsstr = (char *)*p; /* Length check */ l = B(xu->xu_xslen); if (l < 11 || l > 23 || !RPCB_BUF_GEQ(rm, *p, XDRALIGN(l))) return(-1); /* Advance p */ *(char **)p += XDRALIGN(l); /* Copy string to local buffer & terminate C style */ bcopy(xu->xu_xsstr, uastr, l); uastr[l] = '\0'; i = (char *)&xu->xu_ip; pp = (char *)&xu->xu_port; /* * Expected format: a.b.c.d.e.f where [a-d] correspond to bytes of * an IP address and [ef] are the bytes of a L4 port. */ if (!(ISDIGIT(uastr[0]) && ISDIGIT(uastr[l-1]))) return(-1); b = uastr; for (c = &uastr[1], d = 0, dd = 0; c < &uastr[l-1]; c++) { if (ISDIGIT(*c)) { dd = 0; continue; } if (*c == '.') { if (dd != 0) return(-1); /* Check for ASCII byte. */ *c = '\0'; t = ipf_p_rpcb_atoi(b); if (t > 255) return(-1); /* Aim b at beginning of the next byte. */ b = c + 1; /* Switch off IP addr vs port parsing. */ if (d < 4) i[d++] = t & 0xff; else pp[d++ - 4] = t & 0xff; dd = 1; continue; } return(-1); } if (d != 5) /* String must contain exactly 5 periods. */ return(-1); /* Handle the last byte (port low byte) */ t = ipf_p_rpcb_atoi(b); if (t > 255) return(-1); pp[d - 4] = t & 0xff; return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_atoi (XXX should be generic for all proxies) */ /* Returns: int -- integer representation of supplied string */ /* Parameters: ptr(I) - input string */ /* */ /* Simple version of atoi(3) ripped from ip_rcmd_pxy.c. */ /* -------------------------------------------------------------------- */ static u_int ipf_p_rpcb_atoi(ptr) char *ptr; { register char *s = ptr, c; register u_int i = 0; while (((c = *s++) != '\0') && ISDIGIT(c)) { i *= 10; i += c - '0'; } return i; } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_modreq */ /* Returns: int -- change in datagram length */ /* APR_ERR(2) - critical failure */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT session */ /* rm(I) - pointer to RPC message structure */ /* m(I) - pointer to mbuf chain */ /* off(I) - current offset within mbuf chain */ /* */ /* When external and internal addresses differ, we rewrite the former */ /* with the latter. (This is exclusive to protocol versions 3 & 4). */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_modreq(fin, nat, rm, m, off) fr_info_t *fin; nat_t *nat; rpc_msg_t *rm; mb_t *m; u_int off; { u_int len, xlen, pos, bogo; rpcb_args_t *ra; char uaddr[24]; udphdr_t *udp; char *i, *p; int diff; ra = &rm->rm_call.rc_rpcbargs; i = (char *)&nat->nat_odstaddr; p = (char *)&nat->nat_odport; /* Form new string. */ bzero(uaddr, sizeof(uaddr)); /* Just in case we need padding. */ #if defined(SNPRINTF) && defined(_KERNEL) SNPRINTF(uaddr, sizeof(uaddr), #else (void) sprintf(uaddr, #endif "%u.%u.%u.%u.%u.%u", i[0] & 0xff, i[1] & 0xff, i[2] & 0xff, i[3] & 0xff, p[0] & 0xff, p[1] & 0xff); len = strlen(uaddr); xlen = XDRALIGN(len); /* Determine mbuf offset to start writing to. */ pos = (char *)ra->ra_maddr.xu_xslen - rm->rm_msgbuf; off += pos; /* Write new string length. */ bogo = htonl(len); COPYBACK(m, off, 4, (caddr_t)&bogo); off += 4; /* Write new string. */ COPYBACK(m, off, xlen, uaddr); off += xlen; /* Write in zero r_owner. */ bogo = 0; COPYBACK(m, off, 4, (caddr_t)&bogo); /* Determine difference in data lengths. */ diff = xlen - XDRALIGN(B(ra->ra_maddr.xu_xslen)); /* * If our new string has a different length, make necessary * adjustments. */ if (diff != 0) { udp = fin->fin_dp; udp->uh_ulen = htons(ntohs(udp->uh_ulen) + diff); fin->fin_plen += diff; fin->fin_ip->ip_len = htons(fin->fin_plen); fin->fin_dlen += diff; /* XXX Storage lengths. */ } return(diff); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_decoderep */ /* Returns: int - -1 == bad request or critical failure, */ /* 0 == valid, negative reply */ /* 1 == vaddlid, positive reply; needs no changes */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT session structure */ /* rs(I) - pointer to RPCB session structure */ /* rm(I) - pointer to RPC message structure */ /* rxp(O) - pointer to RPCB transaction structure */ /* */ /* Take a presumed RPCB reply, extract the XID, search for the original */ /* request information, and determine whether the request was accepted */ /* or rejected. With a valid accepted reply, go ahead and create NAT */ /* and state entries, and finish up by rewriting the packet as */ /* required. */ /* */ /* WARNING: It's the responsibility of the caller to make sure there */ /* is enough room in rs_buf for the basic RPC message "preamble". */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_decoderep(fin, nat, rs, rm, rxp) fr_info_t *fin; nat_t *nat; rpcb_session_t *rs; rpc_msg_t *rm; rpcb_xact_t **rxp; { rpcb_listp_t *rl; rpcb_entry_t *re; rpcb_xact_t *rx; u_32_t xdr, *p; rpc_resp_t *rr; int rv, cnt; p = (u_32_t *)rm->rm_msgbuf; bzero((char *)&rx, sizeof(rx)); rr = &rm->rm_resp; rm->rm_xid = p; xdr = B(p++); /* Record this message's XID. */ /* Lookup XID */ MUTEX_ENTER(&rs->rs_rxlock); if ((rx = ipf_p_rpcb_lookup(rs, xdr)) == NULL) { MUTEX_EXIT(&rs->rs_rxlock); return(-1); } ++rx->rx_ref; /* per thread reference */ MUTEX_EXIT(&rs->rs_rxlock); *rxp = rx; /* Test call vs reply */ if (B(p++) != RPCB_REPLY) return(-1); /* Test reply_stat */ switch(B(p++)) { case RPCB_MSG_DENIED: return(0); case RPCB_MSG_ACCEPTED: break; default: return(-1); } /* Bypass RPC authentication stuff. */ if (ipf_p_rpcb_skipauth(rm, &rr->rr_authverf, &p) != 0) return(-1); /* Test accept status */ if (!RPCB_BUF_GEQ(rm, p, 4)) return(-1); if (B(p++) != 0) return(0); /* Parse out the expected reply */ switch(rx->rx_type) { case RPCB_RES_PMAP: /* There must be only one 4 byte argument. */ if (!RPCB_BUF_EQ(rm, p, 4)) return(-1); rr->rr_v2 = p; xdr = B(rr->rr_v2); /* Reply w/ a 0 port indicates service isn't registered */ if (xdr == 0) return(0); /* Is the value sane? */ if (xdr > 65535) return(-1); /* Create NAT & state table entries. */ if (ipf_p_rpcb_getnat(fin, nat, rx->rx_proto, (u_int)xdr) != 0) return(-1); break; case RPCB_RES_STRING: /* Expecting a XDR string; need 4 bytes for length */ if (!RPCB_BUF_GEQ(rm, p, 4)) return(-1); rr->rr_v3.xu_str.xs_len = p++; rr->rr_v3.xu_str.xs_str = (char *)p; xdr = B(rr->rr_v3.xu_xslen); /* A null string indicates an unregistered service */ if ((xdr == 0) && RPCB_BUF_EQ(rm, p, 0)) return(0); /* Decode the target IP address / port. */ if (ipf_p_rpcb_getuaddr(rm, &rr->rr_v3, &p) != 0) return(-1); /* Validate the IP address and port contained. */ if (nat->nat_odstaddr != rr->rr_v3.xu_ip) return(-1); /* Create NAT & state table entries. */ if (ipf_p_rpcb_getnat(fin, nat, rx->rx_proto, (u_int)rr->rr_v3.xu_port) != 0) return(-1); break; case RPCB_RES_LIST: if (!RPCB_BUF_GEQ(rm, p, 4)) return(-1); /* rpcb_entry_list_ptr */ switch(B(p)) { case 0: return(0); /*NOTREACHED*/ break; case 1: break; default: return(-1); } rl = &rr->rr_v4; rl->rl_list = p++; cnt = 0; for(;;) { re = &rl->rl_entries[rl->rl_cnt]; if (ipf_p_rpcb_getuaddr(rm, &re->re_maddr, &p) != 0) return(-1); if (ipf_p_rpcb_getproto(rm, &re->re_netid, &p) != 0) return(-1); /* re_semantics & re_pfamily length */ if (!RPCB_BUF_GEQ(rm, p, 12)) return(-1); p++; /* Skipping re_semantics. */ xdr = B(p++); if ((xdr != 4) || strncmp((char *)p, "inet", 4)) return(-1); p++; if (ipf_p_rpcb_getproto(rm, &re->re_proto, &p) != 0) return(-1); if (!RPCB_BUF_GEQ(rm, p, 4)) return(-1); re->re_more = p; if (B(re->re_more) > 1) /* 0,1 only legal values */ return(-1); ++rl->rl_cnt; ++cnt; if (B(re->re_more) == 0) break; /* Replies in max out at 2; TCP and/or UDP */ if (cnt > 2) return(-1); p++; } for(rl->rl_cnt = 0; rl->rl_cnt < cnt; rl->rl_cnt++) { re = &rl->rl_entries[rl->rl_cnt]; rv = ipf_p_rpcb_getnat(fin, nat, re->re_proto.xp_proto, (u_int)re->re_maddr.xu_port); if (rv != 0) return(-1); } break; default: /*CONSTANTCONDITION*/ IPF_PANIC(1, ("illegal rx_type %d", rx->rx_type)); } return(1); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_lookup */ /* Returns: rpcb_xact_t * - NULL == no matching record, */ /* else pointer to relevant entry */ /* Parameters: rs(I) - pointer to RPCB session */ /* xid(I) - XID to look for */ /* -------------------------------------------------------------------- */ static rpcb_xact_t * ipf_p_rpcb_lookup(rs, xid) rpcb_session_t *rs; u_32_t xid; { rpcb_xact_t *rx; if (rs->rs_rxlist == NULL) return(NULL); for (rx = rs->rs_rxlist; rx != NULL; rx = rx->rx_next) if (rx->rx_xid == xid) break; return(rx); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_deref */ /* Returns: (void) */ /* Parameters: rs(I) - pointer to RPCB session */ /* rx(I) - pointer to RPC transaction struct to remove */ /* force(I) - indicates to delete entry regardless of */ /* reference count */ /* Locking: rs->rs_rxlock must be held write only */ /* */ /* Free the RPCB transaction record rx from the chain of entries. */ /* -------------------------------------------------------------------- */ static void ipf_p_rpcb_deref(rs, rx) rpcb_session_t *rs; rpcb_xact_t *rx; { rs = rs; /* LINT */ if (rx == NULL) return; if (--rx->rx_ref != 0) return; if (rx->rx_next != NULL) rx->rx_next->rx_pnext = rx->rx_pnext; *rx->rx_pnext = rx->rx_next; KFREE(rx); --V_rpcbcnt; } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_getproto */ /* Returns: int - -1 == illegal protocol/netid, */ /* 0 == legal protocol/netid */ /* Parameters: rm(I) - pointer to RPC message structure */ /* xp(I) - pointer to netid structure */ /* p(IO) - pointer to location within packet buffer */ /* */ /* Decode netid/proto stored at p and record its numeric value. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_getproto(rm, xp, p) rpc_msg_t *rm; xdr_proto_t *xp; u_32_t **p; { u_int len; /* Must have 4 bytes for length & 4 bytes for "tcp" or "udp". */ if (!RPCB_BUF_GEQ(rm, p, 8)) return(-1); xp->xp_xslen = (*p)++; xp->xp_xsstr = (char *)*p; /* Test the string length. */ len = B(xp->xp_xslen); if (len != 3) return(-1); /* Test the actual string & record the protocol accordingly. */ if (!strncmp((char *)xp->xp_xsstr, "tcp\0", 4)) xp->xp_proto = IPPROTO_TCP; else if (!strncmp((char *)xp->xp_xsstr, "udp\0", 4)) xp->xp_proto = IPPROTO_UDP; else { return(-1); } /* Advance past the string. */ (*p)++; return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_getnat */ /* Returns: int -- -1 == failed to create table entries, */ /* 0 == success */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT table entry */ /* proto(I) - transport protocol for new entries */ /* port(I) - new port to use w/ wildcard table entries */ /* */ /* Create state and NAT entries to handle an anticipated connection */ /* attempt between RPC client and server. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_getnat(fin, nat, proto, port) fr_info_t *fin; nat_t *nat; u_int proto; u_int port; { ipf_main_softc_t *softc = fin->fin_main_soft; ipnat_t *ipn, ipnat; tcphdr_t tcp; ipstate_t *is; fr_info_t fi; nat_t *natl; int nflags; ipn = nat->nat_ptr; /* Generate dummy fr_info */ bcopy((char *)fin, (char *)&fi, sizeof(fi)); fi.fin_out = 0; fi.fin_p = proto; fi.fin_sport = 0; fi.fin_dport = port & 0xffff; fi.fin_flx |= FI_IGNORE; fi.fin_saddr = nat->nat_osrcaddr; fi.fin_daddr = nat->nat_odstaddr; bzero((char *)&tcp, sizeof(tcp)); tcp.th_dport = htons(port); if (proto == IPPROTO_TCP) { tcp.th_win = htons(8192); TCP_OFF_A(&tcp, sizeof(tcphdr_t) >> 2); fi.fin_dlen = sizeof(tcphdr_t); tcp.th_flags = TH_SYN; nflags = NAT_TCP; } else { fi.fin_dlen = sizeof(udphdr_t); nflags = NAT_UDP; } nflags |= SI_W_SPORT|NAT_SEARCH; fi.fin_dp = &tcp; fi.fin_plen = fi.fin_hlen + fi.fin_dlen; /* * Search for existing NAT & state entries. Pay close attention to * mutexes / locks grabbed from lookup routines, as not doing so could * lead to bad things. * * If successful, fr_stlookup returns with ipf_state locked. We have * no use for this lock, so simply unlock it if necessary. */ is = ipf_state_lookup(&fi, &tcp, NULL); if (is != NULL) { RWLOCK_EXIT(&softc->ipf_state); } RWLOCK_EXIT(&softc->ipf_nat); WRITE_ENTER(&softc->ipf_nat); natl = ipf_nat_inlookup(&fi, nflags, proto, fi.fin_src, fi.fin_dst); if ((natl != NULL) && (is != NULL)) { MUTEX_DOWNGRADE(&softc->ipf_nat); return(0); } /* Slightly modify the following structures for actual use in creating * NAT and/or state entries. We're primarily concerned with stripping * flags that may be detrimental to the creation process or simply * shouldn't be associated with a table entry. */ fi.fin_fr = &rpcbfr; fi.fin_flx &= ~FI_IGNORE; nflags &= ~NAT_SEARCH; if (natl == NULL) { #ifdef USE_MUTEXES ipf_nat_softc_t *softn = softc->ipf_nat_soft; #endif /* XXX Since we're just copying the original ipn contents * back, would we be better off just sending a pointer to * the 'temp' copy off to nat_new instead? */ /* Generate template/bogus NAT rule. */ bcopy((char *)ipn, (char *)&ipnat, sizeof(ipnat)); ipn->in_flags = nflags & IPN_TCPUDP; ipn->in_apr = NULL; ipn->in_pr[0] = proto; ipn->in_pr[1] = proto; ipn->in_dpmin = fi.fin_dport; ipn->in_dpmax = fi.fin_dport; ipn->in_dpnext = fi.fin_dport; ipn->in_space = 1; ipn->in_ippip = 1; if (ipn->in_flags & IPN_FILTER) { ipn->in_scmp = 0; ipn->in_dcmp = 0; } ipn->in_plabel = -1; /* Create NAT entry. return NULL if this fails. */ MUTEX_ENTER(&softn->ipf_nat_new); natl = ipf_nat_add(&fi, ipn, NULL, nflags|SI_CLONE|NAT_SLAVE, NAT_INBOUND); MUTEX_EXIT(&softn->ipf_nat_new); bcopy((char *)&ipnat, (char *)ipn, sizeof(ipnat)); if (natl == NULL) { MUTEX_DOWNGRADE(&softc->ipf_nat); return(-1); } natl->nat_ptr = ipn; fi.fin_saddr = natl->nat_nsrcaddr; fi.fin_daddr = natl->nat_ndstaddr; ipn->in_use++; (void) ipf_nat_proto(&fi, natl, nflags); MUTEX_ENTER(&natl->nat_lock); ipf_nat_update(&fi, natl); MUTEX_EXIT(&natl->nat_lock); } MUTEX_DOWNGRADE(&softc->ipf_nat); if (is == NULL) { /* Create state entry. Return NULL if this fails. */ fi.fin_flx |= FI_NATED; fi.fin_flx &= ~FI_STATE; nflags &= NAT_TCPUDP; nflags |= SI_W_SPORT|SI_CLONE; if (ipf_state_add(softc, &fi, NULL, nflags) != 0) { /* * XXX nat_delete is private to ip_nat.c. Should * check w/ Darren about this one. * * nat_delete(natl, NL_EXPIRE); */ return(-1); } } return(0); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_modv3 */ /* Returns: int -- change in packet length */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT session */ /* rm(I) - pointer to RPC message structure */ /* m(I) - pointer to mbuf chain */ /* off(I) - offset within mbuf chain */ /* */ /* Write a new universal address string to this packet, adjusting */ /* lengths as necessary. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_modv3(fin, nat, rm, m, off) fr_info_t *fin; nat_t *nat; rpc_msg_t *rm; mb_t *m; u_int off; { u_int len, xlen, pos, bogo; rpc_resp_t *rr; char uaddr[24]; char *i, *p; int diff; rr = &rm->rm_resp; i = (char *)&nat->nat_ndstaddr; p = (char *)&rr->rr_v3.xu_port; /* Form new string. */ bzero(uaddr, sizeof(uaddr)); /* Just in case we need padding. */ #if defined(SNPRINTF) && defined(_KERNEL) SNPRINTF(uaddr, sizeof(uaddr), #else (void) sprintf(uaddr, #endif "%u.%u.%u.%u.%u.%u", i[0] & 0xff, i[1] & 0xff, i[2] & 0xff, i[3] & 0xff, p[0] & 0xff, p[1] & 0xff); len = strlen(uaddr); xlen = XDRALIGN(len); /* Determine mbuf offset to write to. */ pos = (char *)rr->rr_v3.xu_xslen - rm->rm_msgbuf; off += pos; /* Write new string length. */ bogo = htonl(len); COPYBACK(m, off, 4, (caddr_t)&bogo); off += 4; /* Write new string. */ COPYBACK(m, off, xlen, uaddr); /* Determine difference in data lengths. */ diff = xlen - XDRALIGN(B(rr->rr_v3.xu_xslen)); /* * If our new string has a different length, make necessary * adjustments. */ if (diff != 0) ipf_p_rpcb_fixlen(fin, diff); return(diff); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_modv4 */ /* Returns: int -- change in packet length */ /* Parameters: fin(I) - pointer to packet information */ /* nat(I) - pointer to NAT session */ /* rm(I) - pointer to RPC message structure */ /* m(I) - pointer to mbuf chain */ /* off(I) - offset within mbuf chain */ /* */ /* Write new rpcb_entry list, adjusting lengths as necessary. */ /* -------------------------------------------------------------------- */ static int ipf_p_rpcb_modv4(fin, nat, rm, m, off) fr_info_t *fin; nat_t *nat; rpc_msg_t *rm; mb_t *m; u_int off; { u_int len, xlen, pos, bogo; rpcb_listp_t *rl; rpcb_entry_t *re; rpc_resp_t *rr; char uaddr[24]; int diff, cnt; char *i, *p; diff = 0; rr = &rm->rm_resp; rl = &rr->rr_v4; i = (char *)&nat->nat_ndstaddr; /* Determine mbuf offset to write to. */ re = &rl->rl_entries[0]; pos = (char *)re->re_maddr.xu_xslen - rm->rm_msgbuf; off += pos; for (cnt = 0; cnt < rl->rl_cnt; cnt++) { re = &rl->rl_entries[cnt]; p = (char *)&re->re_maddr.xu_port; /* Form new string. */ bzero(uaddr, sizeof(uaddr)); /* Just in case we need padding. */ #if defined(SNPRINTF) && defined(_KERNEL) SNPRINTF(uaddr, sizeof(uaddr), #else (void) sprintf(uaddr, #endif "%u.%u.%u.%u.%u.%u", i[0] & 0xff, i[1] & 0xff, i[2] & 0xff, i[3] & 0xff, p[0] & 0xff, p[1] & 0xff); len = strlen(uaddr); xlen = XDRALIGN(len); /* Write new string length. */ bogo = htonl(len); COPYBACK(m, off, 4, (caddr_t)&bogo); off += 4; /* Write new string. */ COPYBACK(m, off, xlen, uaddr); off += xlen; /* Record any change in length. */ diff += xlen - XDRALIGN(B(re->re_maddr.xu_xslen)); /* If the length changed, copy back the rest of this entry. */ len = ((char *)re->re_more + 4) - (char *)re->re_netid.xp_xslen; if (diff != 0) { COPYBACK(m, off, len, (caddr_t)re->re_netid.xp_xslen); } off += len; } /* * If our new string has a different length, make necessary * adjustments. */ if (diff != 0) ipf_p_rpcb_fixlen(fin, diff); return(diff); } /* -------------------------------------------------------------------- */ /* Function: ipf_p_rpcb_fixlen */ /* Returns: (void) */ /* Parameters: fin(I) - pointer to packet information */ /* len(I) - change in packet length */ /* */ /* Adjust various packet related lengths held in structure and packet */ /* header fields. */ /* -------------------------------------------------------------------- */ static void ipf_p_rpcb_fixlen(fin, len) fr_info_t *fin; int len; { udphdr_t *udp; udp = fin->fin_dp; udp->uh_ulen = htons(ntohs(udp->uh_ulen) + len); fin->fin_plen += len; fin->fin_ip->ip_len = htons(fin->fin_plen); fin->fin_dlen += len; } #undef B Index: head/sys/net/bpf.c =================================================================== --- head/sys/net/bpf.c (revision 336675) +++ head/sys/net/bpf.c (revision 336676) @@ -1,3080 +1,3080 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from the Stanford/CMU enet packet filter, * (net/enet.c) distributed as part of 4.3BSD, and code contributed * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence * Berkeley Laboratory. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)bpf.c 8.4 (Berkeley) 1/9/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_bpf.h" #include "opt_ddb.h" #include "opt_netgraph.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #ifdef BPF_JITTER #include #endif #include #include #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_BPF, "BPF", "BPF data"); static struct bpf_if_ext dead_bpf_if = { .bif_dlist = LIST_HEAD_INITIALIZER() }; struct bpf_if { #define bif_next bif_ext.bif_next #define bif_dlist bif_ext.bif_dlist struct bpf_if_ext bif_ext; /* public members */ u_int bif_dlt; /* link layer type */ u_int bif_hdrlen; /* length of link header */ struct ifnet *bif_ifp; /* corresponding interface */ struct rwlock bif_lock; /* interface lock */ LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */ int bif_flags; /* Interface flags */ struct bpf_if **bif_bpf; /* Pointer to pointer to us */ }; CTASSERT(offsetof(struct bpf_if, bif_ext) == 0); #define BPFIF_RLOCK(bif) rw_rlock(&(bif)->bif_lock) #define BPFIF_RUNLOCK(bif) rw_runlock(&(bif)->bif_lock) #define BPFIF_WLOCK(bif) rw_wlock(&(bif)->bif_lock) #define BPFIF_WUNLOCK(bif) rw_wunlock(&(bif)->bif_lock) #if defined(DEV_BPF) || defined(NETGRAPH_BPF) #define PRINET 26 /* interruptible */ #define SIZEOF_BPF_HDR(type) \ (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) #ifdef COMPAT_FREEBSD32 #include #include #define BPF_ALIGNMENT32 sizeof(int32_t) #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32) #ifndef BURN_BRIDGES /* * 32-bit version of structure prepended to each packet. We use this header * instead of the standard one for 32-bit streams. We mark the a stream as * 32-bit the first time we see a 32-bit compat ioctl request. */ struct bpf_hdr32 { struct timeval32 bh_tstamp; /* time stamp */ uint32_t bh_caplen; /* length of captured portion */ uint32_t bh_datalen; /* original length of packet */ uint16_t bh_hdrlen; /* length of bpf header (this struct plus alignment padding) */ }; #endif struct bpf_program32 { u_int bf_len; uint32_t bf_insns; }; struct bpf_dltlist32 { u_int bfl_len; u_int bfl_list; }; #define BIOCSETF32 _IOW('B', 103, struct bpf_program32) #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) #endif #define BPF_LOCK() sx_xlock(&bpf_sx) #define BPF_UNLOCK() sx_xunlock(&bpf_sx) #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED) /* * bpf_iflist is a list of BPF interface structures, each corresponding to a * specific DLT. The same network interface might have several BPF interface * structures registered by different layers in the stack (i.e., 802.11 * frames, ethernet frames, etc). */ static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist; static struct sx bpf_sx; /* bpf global lock */ static int bpf_bpfd_cnt; static void bpf_attachd(struct bpf_d *, struct bpf_if *); static void bpf_detachd(struct bpf_d *); static void bpf_detachd_locked(struct bpf_d *); static void bpf_freed(struct bpf_d *); static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, struct sockaddr *, int *, struct bpf_d *); static int bpf_setif(struct bpf_d *, struct ifreq *); static void bpf_timed_out(void *); static __inline void bpf_wakeup(struct bpf_d *); static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), struct bintime *); static void reset_d(struct bpf_d *); static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); static int bpf_setdlt(struct bpf_d *, u_int); static void filt_bpfdetach(struct knote *); static int filt_bpfread(struct knote *, long); static void bpf_drvinit(void *); static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); int bpf_maxinsns = BPF_MAXINSNS; SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, &bpf_maxinsns, 0, "Maximum bpf program instructions"); static int bpf_zerocopy_enable = 0; SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, bpf_stats_sysctl, "bpf statistics portal"); -static VNET_DEFINE(int, bpf_optimize_writers) = 0; +VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0; #define V_bpf_optimize_writers VNET(bpf_optimize_writers) SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0, "Do not send packets until BPF program is set"); static d_open_t bpfopen; static d_read_t bpfread; static d_write_t bpfwrite; static d_ioctl_t bpfioctl; static d_poll_t bpfpoll; static d_kqfilter_t bpfkqfilter; static struct cdevsw bpf_cdevsw = { .d_version = D_VERSION, .d_open = bpfopen, .d_read = bpfread, .d_write = bpfwrite, .d_ioctl = bpfioctl, .d_poll = bpfpoll, .d_name = "bpf", .d_kqfilter = bpfkqfilter, }; static struct filterops bpfread_filtops = { .f_isfd = 1, .f_detach = filt_bpfdetach, .f_event = filt_bpfread, }; eventhandler_tag bpf_ifdetach_cookie = NULL; /* * LOCKING MODEL USED BY BPF: * Locks: * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal, * some global counters and every bpf_if reference. * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters. * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields * used by bpf_mtap code. * * Lock order: * * Global lock, interface lock, descriptor lock * * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2] * working model. In many places (like bpf_detachd) we start with BPF descriptor * (and we need to at least rlock it to get reliable interface pointer). This * gives us potential LOR. As a result, we use global lock to protect from bpf_if * change in every such place. * * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and * 3) descriptor main wlock. * Reading bd_bif can be protected by any of these locks, typically global lock. * * Changing read/write BPF filter is protected by the same three locks, * the same applies for reading. * * Sleeping in global lock is not allowed due to bpfdetach() using it. */ /* * Wrapper functions for various buffering methods. If the set of buffer * modes expands, we will probably want to introduce a switch data structure * similar to protosw, et. */ static void bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, u_int len) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_append_bytes(d, buf, offset, src, len)); case BPF_BUFMODE_ZBUF: counter_u64_add(d->bd_zcopy, 1); return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); default: panic("bpf_buf_append_bytes"); } } static void bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, u_int len) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); case BPF_BUFMODE_ZBUF: counter_u64_add(d->bd_zcopy, 1); return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); default: panic("bpf_buf_append_mbuf"); } } /* * This function gets called when the free buffer is re-assigned. */ static void bpf_buf_reclaimed(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return; case BPF_BUFMODE_ZBUF: bpf_zerocopy_buf_reclaimed(d); return; default: panic("bpf_buf_reclaimed"); } } /* * If the buffer mechanism has a way to decide that a held buffer can be made * free, then it is exposed via the bpf_canfreebuf() interface. (1) is * returned if the buffer can be discarded, (0) is returned if it cannot. */ static int bpf_canfreebuf(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_canfreebuf(d)); } return (0); } /* * Allow the buffer model to indicate that the current store buffer is * immutable, regardless of the appearance of space. Return (1) if the * buffer is writable, and (0) if not. */ static int bpf_canwritebuf(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_canwritebuf(d)); } return (1); } /* * Notify buffer model that an attempt to write to the store buffer has * resulted in a dropped packet, in which case the buffer may be considered * full. */ static void bpf_buffull(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: bpf_zerocopy_buffull(d); break; } } /* * Notify the buffer model that a buffer has moved into the hold position. */ void bpf_bufheld(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: bpf_zerocopy_bufheld(d); break; } } static void bpf_free(struct bpf_d *d) { switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_free(d)); case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_free(d)); default: panic("bpf_buf_free"); } } static int bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) { if (d->bd_bufmode != BPF_BUFMODE_BUFFER) return (EOPNOTSUPP); return (bpf_buffer_uiomove(d, buf, len, uio)); } static int bpf_ioctl_sblen(struct bpf_d *d, u_int *i) { if (d->bd_bufmode != BPF_BUFMODE_BUFFER) return (EOPNOTSUPP); return (bpf_buffer_ioctl_sblen(d, i)); } static int bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_getzmax(td, d, i)); } static int bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); } static int bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); } /* * General BPF functions. */ static int bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, struct sockaddr *sockp, int *hdrlen, struct bpf_d *d) { const struct ieee80211_bpf_params *p; struct ether_header *eh; struct mbuf *m; int error; int len; int hlen; int slen; /* * Build a sockaddr based on the data link layer type. * We do this at this level because the ethernet header * is copied directly into the data field of the sockaddr. * In the case of SLIP, there is no header and the packet * is forwarded as is. * Also, we are careful to leave room at the front of the mbuf * for the link level header. */ switch (linktype) { case DLT_SLIP: sockp->sa_family = AF_INET; hlen = 0; break; case DLT_EN10MB: sockp->sa_family = AF_UNSPEC; /* XXX Would MAXLINKHDR be better? */ hlen = ETHER_HDR_LEN; break; case DLT_FDDI: sockp->sa_family = AF_IMPLINK; hlen = 0; break; case DLT_RAW: sockp->sa_family = AF_UNSPEC; hlen = 0; break; case DLT_NULL: /* * null interface types require a 4 byte pseudo header which * corresponds to the address family of the packet. */ sockp->sa_family = AF_UNSPEC; hlen = 4; break; case DLT_ATM_RFC1483: /* * en atm driver requires 4-byte atm pseudo header. * though it isn't standard, vpi:vci needs to be * specified anyway. */ sockp->sa_family = AF_UNSPEC; hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ break; case DLT_PPP: sockp->sa_family = AF_UNSPEC; hlen = 4; /* This should match PPP_HDRLEN */ break; case DLT_IEEE802_11: /* IEEE 802.11 wireless */ sockp->sa_family = AF_IEEE80211; hlen = 0; break; case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ sockp->sa_family = AF_IEEE80211; sockp->sa_len = 12; /* XXX != 0 */ hlen = sizeof(struct ieee80211_bpf_params); break; default: return (EIO); } len = uio->uio_resid; if (len < hlen || len - hlen > ifp->if_mtu) return (EMSGSIZE); m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); if (m == NULL) return (EIO); m->m_pkthdr.len = m->m_len = len; *mp = m; error = uiomove(mtod(m, u_char *), len, uio); if (error) goto bad; slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len); if (slen == 0) { error = EPERM; goto bad; } /* Check for multicast destination */ switch (linktype) { case DLT_EN10MB: eh = mtod(m, struct ether_header *); if (ETHER_IS_MULTICAST(eh->ether_dhost)) { if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, ETHER_ADDR_LEN) == 0) m->m_flags |= M_BCAST; else m->m_flags |= M_MCAST; } if (d->bd_hdrcmplt == 0) { memcpy(eh->ether_shost, IF_LLADDR(ifp), sizeof(eh->ether_shost)); } break; } /* * Make room for link header, and copy it to sockaddr */ if (hlen != 0) { if (sockp->sa_family == AF_IEEE80211) { /* * Collect true length from the parameter header * NB: sockp is known to be zero'd so if we do a * short copy unspecified parameters will be * zero. * NB: packet may not be aligned after stripping * bpf params * XXX check ibp_vers */ p = mtod(m, const struct ieee80211_bpf_params *); hlen = p->ibp_len; if (hlen > sizeof(sockp->sa_data)) { error = EINVAL; goto bad; } } bcopy(mtod(m, const void *), sockp->sa_data, hlen); } *hdrlen = hlen; return (0); bad: m_freem(m); return (error); } /* * Attach file to the bpf interface, i.e. make d listen on bp. */ static void bpf_attachd(struct bpf_d *d, struct bpf_if *bp) { int op_w; BPF_LOCK_ASSERT(); /* * Save sysctl value to protect from sysctl change * between reads */ op_w = V_bpf_optimize_writers || d->bd_writer; if (d->bd_bif != NULL) bpf_detachd_locked(d); /* * Point d at bp, and add d to the interface's list. * Since there are many applications using BPF for * sending raw packets only (dhcpd, cdpd are good examples) * we can delay adding d to the list of active listeners until * some filter is configured. */ BPFIF_WLOCK(bp); BPFD_LOCK(d); d->bd_bif = bp; if (op_w != 0) { /* Add to writers-only list */ LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); /* * We decrement bd_writer on every filter set operation. * First BIOCSETF is done by pcap_open_live() to set up * snap length. After that appliation usually sets its own filter */ d->bd_writer = 2; } else LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); bpf_bpfd_cnt++; CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); if (op_w == 0) EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); } /* * Check if we need to upgrade our descriptor @d from write-only mode. */ static int bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen) { int is_snap, need_upgrade; /* * Check if we've already upgraded or new filter is empty. */ if (d->bd_writer == 0 || fcode == NULL) return (0); need_upgrade = 0; /* * Check if cmd looks like snaplen setting from * pcap_bpf.c:pcap_open_live(). * Note we're not checking .k value here: * while pcap_open_live() definitely sets to non-zero value, * we'd prefer to treat k=0 (deny ALL) case the same way: e.g. * do not consider upgrading immediately */ if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K)) is_snap = 1; else is_snap = 0; if (is_snap == 0) { /* * We're setting first filter and it doesn't look like * setting snaplen. We're probably using bpf directly. * Upgrade immediately. */ need_upgrade = 1; } else { /* * Do not require upgrade by first BIOCSETF * (used to set snaplen) by pcap_open_live(). */ if (--d->bd_writer == 0) { /* * First snaplen filter has already * been set. This is probably catch-all * filter */ need_upgrade = 1; } } CTR5(KTR_NET, "%s: filter function set by pid %d, " "bd_writer counter %d, snap %d upgrade %d", __func__, d->bd_pid, d->bd_writer, is_snap, need_upgrade); return (need_upgrade); } /* * Add d to the list of active bp filters. * Requires bpf_attachd() to be called before. */ static void bpf_upgraded(struct bpf_d *d) { struct bpf_if *bp; BPF_LOCK_ASSERT(); bp = d->bd_bif; /* * Filter can be set several times without specifying interface. * Mark d as reader and exit. */ if (bp == NULL) { BPFD_LOCK(d); d->bd_writer = 0; BPFD_UNLOCK(d); return; } BPFIF_WLOCK(bp); BPFD_LOCK(d); /* Remove from writers-only list */ LIST_REMOVE(d, bd_next); LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); /* Mark d as reader */ d->bd_writer = 0; BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); } /* * Detach a file from its interface. */ static void bpf_detachd(struct bpf_d *d) { BPF_LOCK(); bpf_detachd_locked(d); BPF_UNLOCK(); } static void bpf_detachd_locked(struct bpf_d *d) { int error; struct bpf_if *bp; struct ifnet *ifp; CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); BPF_LOCK_ASSERT(); /* Check if descriptor is attached */ if ((bp = d->bd_bif) == NULL) return; BPFIF_WLOCK(bp); BPFD_LOCK(d); /* Save bd_writer value */ error = d->bd_writer; /* * Remove d from the interface's descriptor list. */ LIST_REMOVE(d, bd_next); ifp = bp->bif_ifp; d->bd_bif = NULL; BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); bpf_bpfd_cnt--; /* Call event handler iff d is attached */ if (error == 0) EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); /* * Check if this descriptor had requested promiscuous mode. * If so, turn it off. */ if (d->bd_promisc) { d->bd_promisc = 0; CURVNET_SET(ifp->if_vnet); error = ifpromisc(ifp, 0); CURVNET_RESTORE(); if (error != 0 && error != ENXIO) { /* * ENXIO can happen if a pccard is unplugged * Something is really wrong if we were able to put * the driver into promiscuous mode, but can't * take it out. */ if_printf(bp->bif_ifp, "bpf_detach: ifpromisc failed (%d)\n", error); } } } /* * Close the descriptor by detaching it from its interface, * deallocating its buffers, and marking it free. */ static void bpf_dtor(void *data) { struct bpf_d *d = data; BPFD_LOCK(d); if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; BPFD_UNLOCK(d); funsetown(&d->bd_sigio); bpf_detachd(d); #ifdef MAC mac_bpfdesc_destroy(d); #endif /* MAC */ seldrain(&d->bd_sel); knlist_destroy(&d->bd_sel.si_note); callout_drain(&d->bd_callout); bpf_freed(d); free(d, M_BPF); } /* * Open ethernet device. Returns ENXIO for illegal minor device number, * EBUSY if file is open by another process. */ /* ARGSUSED */ static int bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) { struct bpf_d *d; int error; d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); error = devfs_set_cdevpriv(d, bpf_dtor); if (error != 0) { free(d, M_BPF); return (error); } /* Setup counters */ d->bd_rcount = counter_u64_alloc(M_WAITOK); d->bd_dcount = counter_u64_alloc(M_WAITOK); d->bd_fcount = counter_u64_alloc(M_WAITOK); d->bd_wcount = counter_u64_alloc(M_WAITOK); d->bd_wfcount = counter_u64_alloc(M_WAITOK); d->bd_wdcount = counter_u64_alloc(M_WAITOK); d->bd_zcopy = counter_u64_alloc(M_WAITOK); /* * For historical reasons, perform a one-time initialization call to * the buffer routines, even though we're not yet committed to a * particular buffer method. */ bpf_buffer_init(d); if ((flags & FREAD) == 0) d->bd_writer = 2; d->bd_hbuf_in_use = 0; d->bd_bufmode = BPF_BUFMODE_BUFFER; d->bd_sig = SIGIO; d->bd_direction = BPF_D_INOUT; BPF_PID_REFRESH(d, td); #ifdef MAC mac_bpfdesc_init(d); mac_bpfdesc_create(td->td_ucred, d); #endif mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); return (0); } /* * bpfread - read next chunk of packets from buffers */ static int bpfread(struct cdev *dev, struct uio *uio, int ioflag) { struct bpf_d *d; int error; int non_block; int timed_out; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); /* * Restrict application to use a buffer the same size as * as kernel buffers. */ if (uio->uio_resid != d->bd_bufsize) return (EINVAL); non_block = ((ioflag & O_NONBLOCK) != 0); BPFD_LOCK(d); BPF_PID_REFRESH_CUR(d); if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { BPFD_UNLOCK(d); return (EOPNOTSUPP); } if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); timed_out = (d->bd_state == BPF_TIMED_OUT); d->bd_state = BPF_IDLE; while (d->bd_hbuf_in_use) { error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET|PCATCH, "bd_hbuf", 0); if (error != 0) { BPFD_UNLOCK(d); return (error); } } /* * If the hold buffer is empty, then do a timed sleep, which * ends when the timeout expires or when enough packets * have arrived to fill the store buffer. */ while (d->bd_hbuf == NULL) { if (d->bd_slen != 0) { /* * A packet(s) either arrived since the previous * read or arrived while we were asleep. */ if (d->bd_immediate || non_block || timed_out) { /* * Rotate the buffers and return what's here * if we are in immediate mode, non-blocking * flag is set, or this descriptor timed out. */ ROTATE_BUFFERS(d); break; } } /* * No data is available, check to see if the bpf device * is still pointed at a real interface. If not, return * ENXIO so that the userland process knows to rebind * it before using it again. */ if (d->bd_bif == NULL) { BPFD_UNLOCK(d); return (ENXIO); } if (non_block) { BPFD_UNLOCK(d); return (EWOULDBLOCK); } error = msleep(d, &d->bd_lock, PRINET|PCATCH, "bpf", d->bd_rtout); if (error == EINTR || error == ERESTART) { BPFD_UNLOCK(d); return (error); } if (error == EWOULDBLOCK) { /* * On a timeout, return what's in the buffer, * which may be nothing. If there is something * in the store buffer, we can rotate the buffers. */ if (d->bd_hbuf) /* * We filled up the buffer in between * getting the timeout and arriving * here, so we don't need to rotate. */ break; if (d->bd_slen == 0) { BPFD_UNLOCK(d); return (0); } ROTATE_BUFFERS(d); break; } } /* * At this point, we know we have something in the hold slot. */ d->bd_hbuf_in_use = 1; BPFD_UNLOCK(d); /* * Move data from hold buffer into user space. * We know the entire buffer is transferred since * we checked above that the read buffer is bpf_bufsize bytes. * * We do not have to worry about simultaneous reads because * we waited for sole access to the hold buffer above. */ error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); BPFD_LOCK(d); KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); d->bd_hbuf_in_use = 0; wakeup(&d->bd_hbuf_in_use); BPFD_UNLOCK(d); return (error); } /* * If there are processes sleeping on this descriptor, wake them up. */ static __inline void bpf_wakeup(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); if (d->bd_state == BPF_WAITING) { callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; } wakeup(d); if (d->bd_async && d->bd_sig && d->bd_sigio) pgsigio(&d->bd_sigio, d->bd_sig, 0); selwakeuppri(&d->bd_sel, PRINET); KNOTE_LOCKED(&d->bd_sel.si_note, 0); } static void bpf_timed_out(void *arg) { struct bpf_d *d = (struct bpf_d *)arg; BPFD_LOCK_ASSERT(d); if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) return; if (d->bd_state == BPF_WAITING) { d->bd_state = BPF_TIMED_OUT; if (d->bd_slen != 0) bpf_wakeup(d); } } static int bpf_ready(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); if (!bpf_canfreebuf(d) && d->bd_hlen != 0) return (1); if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && d->bd_slen != 0) return (1); return (0); } static int bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) { struct bpf_d *d; struct ifnet *ifp; struct mbuf *m, *mc; struct sockaddr dst; struct route ro; int error, hlen; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); BPF_PID_REFRESH_CUR(d); counter_u64_add(d->bd_wcount, 1); /* XXX: locking required */ if (d->bd_bif == NULL) { counter_u64_add(d->bd_wdcount, 1); return (ENXIO); } ifp = d->bd_bif->bif_ifp; if ((ifp->if_flags & IFF_UP) == 0) { counter_u64_add(d->bd_wdcount, 1); return (ENETDOWN); } if (uio->uio_resid == 0) { counter_u64_add(d->bd_wdcount, 1); return (0); } bzero(&dst, sizeof(dst)); m = NULL; hlen = 0; /* XXX: bpf_movein() can sleep */ error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, &m, &dst, &hlen, d); if (error) { counter_u64_add(d->bd_wdcount, 1); return (error); } counter_u64_add(d->bd_wfcount, 1); if (d->bd_hdrcmplt) dst.sa_family = pseudo_AF_HDRCMPLT; if (d->bd_feedback) { mc = m_dup(m, M_NOWAIT); if (mc != NULL) mc->m_pkthdr.rcvif = ifp; /* Set M_PROMISC for outgoing packets to be discarded. */ if (d->bd_direction == BPF_D_INOUT) m->m_flags |= M_PROMISC; } else mc = NULL; m->m_pkthdr.len -= hlen; m->m_len -= hlen; m->m_data += hlen; /* XXX */ CURVNET_SET(ifp->if_vnet); #ifdef MAC BPFD_LOCK(d); mac_bpfdesc_create_mbuf(d, m); if (mc != NULL) mac_bpfdesc_create_mbuf(d, mc); BPFD_UNLOCK(d); #endif bzero(&ro, sizeof(ro)); if (hlen != 0) { ro.ro_prepend = (u_char *)&dst.sa_data; ro.ro_plen = hlen; ro.ro_flags = RT_HAS_HEADER; } error = (*ifp->if_output)(ifp, m, &dst, &ro); if (error) counter_u64_add(d->bd_wdcount, 1); if (mc != NULL) { if (error == 0) (*ifp->if_input)(ifp, mc); else m_freem(mc); } CURVNET_RESTORE(); return (error); } /* * Reset a descriptor by flushing its packet buffer and clearing the receive * and drop counts. This is doable for kernel-only buffers, but with * zero-copy buffers, we can't write to (or rotate) buffers that are * currently owned by userspace. It would be nice if we could encapsulate * this logic in the buffer code rather than here. */ static void reset_d(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); while (d->bd_hbuf_in_use) mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, "bd_hbuf", 0); if ((d->bd_hbuf != NULL) && (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { /* Free the hold buffer. */ d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); } if (bpf_canwritebuf(d)) d->bd_slen = 0; counter_u64_zero(d->bd_rcount); counter_u64_zero(d->bd_dcount); counter_u64_zero(d->bd_fcount); counter_u64_zero(d->bd_wcount); counter_u64_zero(d->bd_wfcount); counter_u64_zero(d->bd_wdcount); counter_u64_zero(d->bd_zcopy); } /* * FIONREAD Check for read packet available. * BIOCGBLEN Get buffer len [for read()]. * BIOCSETF Set read filter. * BIOCSETFNR Set read filter without resetting descriptor. * BIOCSETWF Set write filter. * BIOCFLUSH Flush read packet buffer. * BIOCPROMISC Put interface into promiscuous mode. * BIOCGDLT Get link layer type. * BIOCGETIF Get interface name. * BIOCSETIF Set interface. * BIOCSRTIMEOUT Set read timeout. * BIOCGRTIMEOUT Get read timeout. * BIOCGSTATS Get packet stats. * BIOCIMMEDIATE Set immediate mode. * BIOCVERSION Get filter language version. * BIOCGHDRCMPLT Get "header already complete" flag * BIOCSHDRCMPLT Set "header already complete" flag * BIOCGDIRECTION Get packet direction flag * BIOCSDIRECTION Set packet direction flag * BIOCGTSTAMP Get time stamp format and resolution. * BIOCSTSTAMP Set time stamp format and resolution. * BIOCLOCK Set "locked" flag * BIOCFEEDBACK Set packet feedback mode. * BIOCSETZBUF Set current zero-copy buffer locations. * BIOCGETZMAX Get maximum zero-copy buffer size. * BIOCROTZBUF Force rotation of zero-copy buffer * BIOCSETBUFMODE Set buffer mode. * BIOCGETBUFMODE Get current buffer mode. */ /* ARGSUSED */ static int bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) { struct bpf_d *d; int error; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); /* * Refresh PID associated with this descriptor. */ BPFD_LOCK(d); BPF_PID_REFRESH(d, td); if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; BPFD_UNLOCK(d); if (d->bd_locked == 1) { switch (cmd) { case BIOCGBLEN: case BIOCFLUSH: case BIOCGDLT: case BIOCGDLTLIST: #ifdef COMPAT_FREEBSD32 case BIOCGDLTLIST32: #endif case BIOCGETIF: case BIOCGRTIMEOUT: #if defined(COMPAT_FREEBSD32) && defined(__amd64__) case BIOCGRTIMEOUT32: #endif case BIOCGSTATS: case BIOCVERSION: case BIOCGRSIG: case BIOCGHDRCMPLT: case BIOCSTSTAMP: case BIOCFEEDBACK: case FIONREAD: case BIOCLOCK: case BIOCSRTIMEOUT: #if defined(COMPAT_FREEBSD32) && defined(__amd64__) case BIOCSRTIMEOUT32: #endif case BIOCIMMEDIATE: case TIOCGPGRP: case BIOCROTZBUF: break; default: return (EPERM); } } #ifdef COMPAT_FREEBSD32 /* * If we see a 32-bit compat ioctl, mark the stream as 32-bit so * that it will get 32-bit packet headers. */ switch (cmd) { case BIOCSETF32: case BIOCSETFNR32: case BIOCSETWF32: case BIOCGDLTLIST32: case BIOCGRTIMEOUT32: case BIOCSRTIMEOUT32: if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { BPFD_LOCK(d); d->bd_compat32 = 1; BPFD_UNLOCK(d); } } #endif CURVNET_SET(TD_TO_VNET(td)); switch (cmd) { default: error = EINVAL; break; /* * Check for read packet available. */ case FIONREAD: { int n; BPFD_LOCK(d); n = d->bd_slen; while (d->bd_hbuf_in_use) mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, "bd_hbuf", 0); if (d->bd_hbuf) n += d->bd_hlen; BPFD_UNLOCK(d); *(int *)addr = n; break; } /* * Get buffer len [for read()]. */ case BIOCGBLEN: BPFD_LOCK(d); *(u_int *)addr = d->bd_bufsize; BPFD_UNLOCK(d); break; /* * Set buffer length. */ case BIOCSBLEN: error = bpf_ioctl_sblen(d, (u_int *)addr); break; /* * Set link layer read filter. */ case BIOCSETF: case BIOCSETFNR: case BIOCSETWF: #ifdef COMPAT_FREEBSD32 case BIOCSETF32: case BIOCSETFNR32: case BIOCSETWF32: #endif error = bpf_setf(d, (struct bpf_program *)addr, cmd); break; /* * Flush read packet buffer. */ case BIOCFLUSH: BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); break; /* * Put interface into promiscuous mode. */ case BIOCPROMISC: if (d->bd_bif == NULL) { /* * No interface attached yet. */ error = EINVAL; break; } if (d->bd_promisc == 0) { error = ifpromisc(d->bd_bif->bif_ifp, 1); if (error == 0) d->bd_promisc = 1; } break; /* * Get current data link type. */ case BIOCGDLT: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else *(u_int *)addr = d->bd_bif->bif_dlt; BPF_UNLOCK(); break; /* * Get a list of supported data link types. */ #ifdef COMPAT_FREEBSD32 case BIOCGDLTLIST32: { struct bpf_dltlist32 *list32; struct bpf_dltlist dltlist; list32 = (struct bpf_dltlist32 *)addr; dltlist.bfl_len = list32->bfl_len; dltlist.bfl_list = PTRIN(list32->bfl_list); BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else { error = bpf_getdltlist(d, &dltlist); if (error == 0) list32->bfl_len = dltlist.bfl_len; } BPF_UNLOCK(); break; } #endif case BIOCGDLTLIST: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); BPF_UNLOCK(); break; /* * Set data link type. */ case BIOCSDLT: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else error = bpf_setdlt(d, *(u_int *)addr); BPF_UNLOCK(); break; /* * Get interface name. */ case BIOCGETIF: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else { struct ifnet *const ifp = d->bd_bif->bif_ifp; struct ifreq *const ifr = (struct ifreq *)addr; strlcpy(ifr->ifr_name, ifp->if_xname, sizeof(ifr->ifr_name)); } BPF_UNLOCK(); break; /* * Set interface. */ case BIOCSETIF: { int alloc_buf, size; /* * Behavior here depends on the buffering model. If * we're using kernel memory buffers, then we can * allocate them here. If we're using zero-copy, * then the user process must have registered buffers * by the time we get here. */ alloc_buf = 0; BPFD_LOCK(d); if (d->bd_bufmode == BPF_BUFMODE_BUFFER && d->bd_sbuf == NULL) alloc_buf = 1; BPFD_UNLOCK(d); if (alloc_buf) { size = d->bd_bufsize; error = bpf_buffer_ioctl_sblen(d, &size); if (error != 0) break; } BPF_LOCK(); error = bpf_setif(d, (struct ifreq *)addr); BPF_UNLOCK(); break; } /* * Set read timeout. */ case BIOCSRTIMEOUT: #if defined(COMPAT_FREEBSD32) && defined(__amd64__) case BIOCSRTIMEOUT32: #endif { struct timeval *tv = (struct timeval *)addr; #if defined(COMPAT_FREEBSD32) && !defined(__mips__) struct timeval32 *tv32; struct timeval tv64; if (cmd == BIOCSRTIMEOUT32) { tv32 = (struct timeval32 *)addr; tv = &tv64; tv->tv_sec = tv32->tv_sec; tv->tv_usec = tv32->tv_usec; } else #endif tv = (struct timeval *)addr; /* * Subtract 1 tick from tvtohz() since this isn't * a one-shot timer. */ if ((error = itimerfix(tv)) == 0) d->bd_rtout = tvtohz(tv) - 1; break; } /* * Get read timeout. */ case BIOCGRTIMEOUT: #if defined(COMPAT_FREEBSD32) && defined(__amd64__) case BIOCGRTIMEOUT32: #endif { struct timeval *tv; #if defined(COMPAT_FREEBSD32) && defined(__amd64__) struct timeval32 *tv32; struct timeval tv64; if (cmd == BIOCGRTIMEOUT32) tv = &tv64; else #endif tv = (struct timeval *)addr; tv->tv_sec = d->bd_rtout / hz; tv->tv_usec = (d->bd_rtout % hz) * tick; #if defined(COMPAT_FREEBSD32) && defined(__amd64__) if (cmd == BIOCGRTIMEOUT32) { tv32 = (struct timeval32 *)addr; tv32->tv_sec = tv->tv_sec; tv32->tv_usec = tv->tv_usec; } #endif break; } /* * Get packet stats. */ case BIOCGSTATS: { struct bpf_stat *bs = (struct bpf_stat *)addr; /* XXXCSJP overflow */ bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount); bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount); break; } /* * Set immediate mode. */ case BIOCIMMEDIATE: BPFD_LOCK(d); d->bd_immediate = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCVERSION: { struct bpf_version *bv = (struct bpf_version *)addr; bv->bv_major = BPF_MAJOR_VERSION; bv->bv_minor = BPF_MINOR_VERSION; break; } /* * Get "header already complete" flag */ case BIOCGHDRCMPLT: BPFD_LOCK(d); *(u_int *)addr = d->bd_hdrcmplt; BPFD_UNLOCK(d); break; /* * Set "header already complete" flag */ case BIOCSHDRCMPLT: BPFD_LOCK(d); d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; BPFD_UNLOCK(d); break; /* * Get packet direction flag */ case BIOCGDIRECTION: BPFD_LOCK(d); *(u_int *)addr = d->bd_direction; BPFD_UNLOCK(d); break; /* * Set packet direction flag */ case BIOCSDIRECTION: { u_int direction; direction = *(u_int *)addr; switch (direction) { case BPF_D_IN: case BPF_D_INOUT: case BPF_D_OUT: BPFD_LOCK(d); d->bd_direction = direction; BPFD_UNLOCK(d); break; default: error = EINVAL; } } break; /* * Get packet timestamp format and resolution. */ case BIOCGTSTAMP: BPFD_LOCK(d); *(u_int *)addr = d->bd_tstamp; BPFD_UNLOCK(d); break; /* * Set packet timestamp format and resolution. */ case BIOCSTSTAMP: { u_int func; func = *(u_int *)addr; if (BPF_T_VALID(func)) d->bd_tstamp = func; else error = EINVAL; } break; case BIOCFEEDBACK: BPFD_LOCK(d); d->bd_feedback = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCLOCK: BPFD_LOCK(d); d->bd_locked = 1; BPFD_UNLOCK(d); break; case FIONBIO: /* Non-blocking I/O */ break; case FIOASYNC: /* Send signal on receive packets */ BPFD_LOCK(d); d->bd_async = *(int *)addr; BPFD_UNLOCK(d); break; case FIOSETOWN: /* * XXX: Add some sort of locking here? * fsetown() can sleep. */ error = fsetown(*(int *)addr, &d->bd_sigio); break; case FIOGETOWN: BPFD_LOCK(d); *(int *)addr = fgetown(&d->bd_sigio); BPFD_UNLOCK(d); break; /* This is deprecated, FIOSETOWN should be used instead. */ case TIOCSPGRP: error = fsetown(-(*(int *)addr), &d->bd_sigio); break; /* This is deprecated, FIOGETOWN should be used instead. */ case TIOCGPGRP: *(int *)addr = -fgetown(&d->bd_sigio); break; case BIOCSRSIG: /* Set receive signal */ { u_int sig; sig = *(u_int *)addr; if (sig >= NSIG) error = EINVAL; else { BPFD_LOCK(d); d->bd_sig = sig; BPFD_UNLOCK(d); } break; } case BIOCGRSIG: BPFD_LOCK(d); *(u_int *)addr = d->bd_sig; BPFD_UNLOCK(d); break; case BIOCGETBUFMODE: BPFD_LOCK(d); *(u_int *)addr = d->bd_bufmode; BPFD_UNLOCK(d); break; case BIOCSETBUFMODE: /* * Allow the buffering mode to be changed as long as we * haven't yet committed to a particular mode. Our * definition of commitment, for now, is whether or not a * buffer has been allocated or an interface attached, since * that's the point where things get tricky. */ switch (*(u_int *)addr) { case BPF_BUFMODE_BUFFER: break; case BPF_BUFMODE_ZBUF: if (bpf_zerocopy_enable) break; /* FALLSTHROUGH */ default: CURVNET_RESTORE(); return (EINVAL); } BPFD_LOCK(d); if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || d->bd_fbuf != NULL || d->bd_bif != NULL) { BPFD_UNLOCK(d); CURVNET_RESTORE(); return (EBUSY); } d->bd_bufmode = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCGETZMAX: error = bpf_ioctl_getzmax(td, d, (size_t *)addr); break; case BIOCSETZBUF: error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); break; case BIOCROTZBUF: error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); break; } CURVNET_RESTORE(); return (error); } /* * Set d's packet filter program to fp. If this file already has a filter, * free it and replace it. Returns EINVAL for bogus requests. * * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls * since reading d->bd_bif can't be protected by d or interface lock due to * lock order. * * Additionally, we have to acquire interface write lock due to bpf_mtap() uses * interface read lock to read all filers. * */ static int bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) { #ifdef COMPAT_FREEBSD32 struct bpf_program fp_swab; struct bpf_program32 *fp32; #endif struct bpf_insn *fcode, *old; #ifdef BPF_JITTER bpf_jit_filter *jfunc, *ofunc; #endif size_t size; u_int flen; int need_upgrade; #ifdef COMPAT_FREEBSD32 switch (cmd) { case BIOCSETF32: case BIOCSETWF32: case BIOCSETFNR32: fp32 = (struct bpf_program32 *)fp; fp_swab.bf_len = fp32->bf_len; fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; fp = &fp_swab; switch (cmd) { case BIOCSETF32: cmd = BIOCSETF; break; case BIOCSETWF32: cmd = BIOCSETWF; break; } break; } #endif fcode = NULL; #ifdef BPF_JITTER jfunc = ofunc = NULL; #endif need_upgrade = 0; /* * Check new filter validness before acquiring any locks. * Allocate memory for new filter, if needed. */ flen = fp->bf_len; if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) return (EINVAL); size = flen * sizeof(*fp->bf_insns); if (size > 0) { /* We're setting up new filter. Copy and check actual data. */ fcode = malloc(size, M_BPF, M_WAITOK); if (copyin(fp->bf_insns, fcode, size) != 0 || !bpf_validate(fcode, flen)) { free(fcode, M_BPF); return (EINVAL); } #ifdef BPF_JITTER if (cmd != BIOCSETWF) { /* * Filter is copied inside fcode and is * perfectly valid. */ jfunc = bpf_jitter(fcode, flen); } #endif } BPF_LOCK(); /* * Set up new filter. * Protect filter change by interface lock. * Additionally, we are protected by global lock here. */ if (d->bd_bif != NULL) BPFIF_WLOCK(d->bd_bif); BPFD_LOCK(d); if (cmd == BIOCSETWF) { old = d->bd_wfilter; d->bd_wfilter = fcode; } else { old = d->bd_rfilter; d->bd_rfilter = fcode; #ifdef BPF_JITTER ofunc = d->bd_bfilter; d->bd_bfilter = jfunc; #endif if (cmd == BIOCSETF) reset_d(d); need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen); } BPFD_UNLOCK(d); if (d->bd_bif != NULL) BPFIF_WUNLOCK(d->bd_bif); if (old != NULL) free(old, M_BPF); #ifdef BPF_JITTER if (ofunc != NULL) bpf_destroy_jit_filter(ofunc); #endif /* Move d to active readers list. */ if (need_upgrade != 0) bpf_upgraded(d); BPF_UNLOCK(); return (0); } /* * Detach a file from its current interface (if attached at all) and attach * to the interface indicated by the name stored in ifr. * Return an errno or 0. */ static int bpf_setif(struct bpf_d *d, struct ifreq *ifr) { struct bpf_if *bp; struct ifnet *theywant; BPF_LOCK_ASSERT(); theywant = ifunit(ifr->ifr_name); if (theywant == NULL || theywant->if_bpf == NULL) return (ENXIO); bp = theywant->if_bpf; /* Check if interface is not being detached from BPF */ BPFIF_RLOCK(bp); if (bp->bif_flags & BPFIF_FLAG_DYING) { BPFIF_RUNLOCK(bp); return (ENXIO); } BPFIF_RUNLOCK(bp); /* * At this point, we expect the buffer is already allocated. If not, * return an error. */ switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: case BPF_BUFMODE_ZBUF: if (d->bd_sbuf == NULL) return (EINVAL); break; default: panic("bpf_setif: bufmode %d", d->bd_bufmode); } if (bp != d->bd_bif) bpf_attachd(d, bp); BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); return (0); } /* * Support for select() and poll() system calls * * Return true iff the specific operation will not block indefinitely. * Otherwise, return false but make a note that a selwakeup() must be done. */ static int bpfpoll(struct cdev *dev, int events, struct thread *td) { struct bpf_d *d; int revents; if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) return (events & (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); /* * Refresh PID associated with this descriptor. */ revents = events & (POLLOUT | POLLWRNORM); BPFD_LOCK(d); BPF_PID_REFRESH(d, td); if (events & (POLLIN | POLLRDNORM)) { if (bpf_ready(d)) revents |= events & (POLLIN | POLLRDNORM); else { selrecord(td, &d->bd_sel); /* Start the read timeout if necessary. */ if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { callout_reset(&d->bd_callout, d->bd_rtout, bpf_timed_out, d); d->bd_state = BPF_WAITING; } } } BPFD_UNLOCK(d); return (revents); } /* * Support for kevent() system call. Register EVFILT_READ filters and * reject all others. */ int bpfkqfilter(struct cdev *dev, struct knote *kn) { struct bpf_d *d; if (devfs_get_cdevpriv((void **)&d) != 0 || kn->kn_filter != EVFILT_READ) return (1); /* * Refresh PID associated with this descriptor. */ BPFD_LOCK(d); BPF_PID_REFRESH_CUR(d); kn->kn_fop = &bpfread_filtops; kn->kn_hook = d; knlist_add(&d->bd_sel.si_note, kn, 1); BPFD_UNLOCK(d); return (0); } static void filt_bpfdetach(struct knote *kn) { struct bpf_d *d = (struct bpf_d *)kn->kn_hook; knlist_remove(&d->bd_sel.si_note, kn, 0); } static int filt_bpfread(struct knote *kn, long hint) { struct bpf_d *d = (struct bpf_d *)kn->kn_hook; int ready; BPFD_LOCK_ASSERT(d); ready = bpf_ready(d); if (ready) { kn->kn_data = d->bd_slen; /* * Ignore the hold buffer if it is being copied to user space. */ if (!d->bd_hbuf_in_use && d->bd_hbuf) kn->kn_data += d->bd_hlen; } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { callout_reset(&d->bd_callout, d->bd_rtout, bpf_timed_out, d); d->bd_state = BPF_WAITING; } return (ready); } #define BPF_TSTAMP_NONE 0 #define BPF_TSTAMP_FAST 1 #define BPF_TSTAMP_NORMAL 2 #define BPF_TSTAMP_EXTERN 3 static int bpf_ts_quality(int tstype) { if (tstype == BPF_T_NONE) return (BPF_TSTAMP_NONE); if ((tstype & BPF_T_FAST) != 0) return (BPF_TSTAMP_FAST); return (BPF_TSTAMP_NORMAL); } static int bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) { struct m_tag *tag; int quality; quality = bpf_ts_quality(tstype); if (quality == BPF_TSTAMP_NONE) return (quality); if (m != NULL) { tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); if (tag != NULL) { *bt = *(struct bintime *)(tag + 1); return (BPF_TSTAMP_EXTERN); } } if (quality == BPF_TSTAMP_NORMAL) binuptime(bt); else getbinuptime(bt); return (quality); } /* * Incoming linkage from device drivers. Process the packet pkt, of length * pktlen, which is stored in a contiguous buffer. The packet is parsed * by each process' filter, and if accepted, stashed into the corresponding * buffer. */ void bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) { struct bintime bt; struct bpf_d *d; #ifdef BPF_JITTER bpf_jit_filter *bf; #endif u_int slen; int gottime; gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { /* * We are not using any locks for d here because: * 1) any filter change is protected by interface * write lock * 2) destroying/detaching d is protected by interface * write lock, too */ counter_u64_add(d->bd_rcount, 1); /* * NB: We dont call BPF_CHECK_DIRECTION() here since there is no * way for the caller to indiciate to us whether this packet * is inbound or outbound. In the bpf_mtap() routines, we use * the interface pointers on the mbuf to figure it out. */ #ifdef BPF_JITTER bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; if (bf != NULL) slen = (*(bf->func))(pkt, pktlen, pktlen); else #endif slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); if (slen != 0) { /* * Filter matches. Let's to acquire write lock. */ BPFD_LOCK(d); counter_u64_add(d->bd_fcount, 1); if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, pkt, pktlen, slen, bpf_append_bytes, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } #define BPF_CHECK_DIRECTION(d, r, i) \ (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ ((d)->bd_direction == BPF_D_OUT && (r) == (i))) /* * Incoming linkage from device drivers, when packet is in an mbuf chain. * Locking model is explained in bpf_tap(). */ void bpf_mtap(struct bpf_if *bp, struct mbuf *m) { struct bintime bt; struct bpf_d *d; #ifdef BPF_JITTER bpf_jit_filter *bf; #endif u_int pktlen, slen; int gottime; /* Skip outgoing duplicate packets. */ if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { m->m_flags &= ~M_PROMISC; return; } pktlen = m_length(m, NULL); gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) continue; counter_u64_add(d->bd_rcount, 1); #ifdef BPF_JITTER bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; /* XXX We cannot handle multiple mbufs. */ if (bf != NULL && m->m_next == NULL) slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); else #endif slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); if (slen != 0) { BPFD_LOCK(d); counter_u64_add(d->bd_fcount, 1); if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, m); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, (u_char *)m, pktlen, slen, bpf_append_mbuf, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } /* * Incoming linkage from device drivers, when packet is in * an mbuf chain and to be prepended by a contiguous header. */ void bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) { struct bintime bt; struct mbuf mb; struct bpf_d *d; u_int pktlen, slen; int gottime; /* Skip outgoing duplicate packets. */ if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { m->m_flags &= ~M_PROMISC; return; } pktlen = m_length(m, NULL); /* * Craft on-stack mbuf suitable for passing to bpf_filter. * Note that we cut corners here; we only setup what's * absolutely needed--this mbuf should never go anywhere else. */ mb.m_next = m; mb.m_data = data; mb.m_len = dlen; pktlen += dlen; gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) continue; counter_u64_add(d->bd_rcount, 1); slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); if (slen != 0) { BPFD_LOCK(d); counter_u64_add(d->bd_fcount, 1); if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, m); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, (u_char *)&mb, pktlen, slen, bpf_append_mbuf, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } #undef BPF_CHECK_DIRECTION #undef BPF_TSTAMP_NONE #undef BPF_TSTAMP_FAST #undef BPF_TSTAMP_NORMAL #undef BPF_TSTAMP_EXTERN static int bpf_hdrlen(struct bpf_d *d) { int hdrlen; hdrlen = d->bd_bif->bif_hdrlen; #ifndef BURN_BRIDGES if (d->bd_tstamp == BPF_T_NONE || BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); else #endif hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); else #endif hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) hdrlen = BPF_WORDALIGN32(hdrlen); else #endif hdrlen = BPF_WORDALIGN(hdrlen); return (hdrlen - d->bd_bif->bif_hdrlen); } static void bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) { struct bintime bt2, boottimebin; struct timeval tsm; struct timespec tsn; if ((tstype & BPF_T_MONOTONIC) == 0) { bt2 = *bt; getboottimebin(&boottimebin); bintime_add(&bt2, &boottimebin); bt = &bt2; } switch (BPF_T_FORMAT(tstype)) { case BPF_T_MICROTIME: bintime2timeval(bt, &tsm); ts->bt_sec = tsm.tv_sec; ts->bt_frac = tsm.tv_usec; break; case BPF_T_NANOTIME: bintime2timespec(bt, &tsn); ts->bt_sec = tsn.tv_sec; ts->bt_frac = tsn.tv_nsec; break; case BPF_T_BINTIME: ts->bt_sec = bt->sec; ts->bt_frac = bt->frac; break; } } /* * Move the packet data from interface memory (pkt) into the * store buffer. "cpfn" is the routine called to do the actual data * transfer. bcopy is passed in to copy contiguous chunks, while * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, * pkt is really an mbuf. */ static void catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), struct bintime *bt) { struct bpf_xhdr hdr; #ifndef BURN_BRIDGES struct bpf_hdr hdr_old; #ifdef COMPAT_FREEBSD32 struct bpf_hdr32 hdr32_old; #endif #endif int caplen, curlen, hdrlen, totlen; int do_wakeup = 0; int do_timestamp; int tstype; BPFD_LOCK_ASSERT(d); /* * Detect whether user space has released a buffer back to us, and if * so, move it from being a hold buffer to a free buffer. This may * not be the best place to do it (for example, we might only want to * run this check if we need the space), but for now it's a reliable * spot to do it. */ if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); } /* * Figure out how many bytes to move. If the packet is * greater or equal to the snapshot length, transfer that * much. Otherwise, transfer the whole packet (unless * we hit the buffer size limit). */ hdrlen = bpf_hdrlen(d); totlen = hdrlen + min(snaplen, pktlen); if (totlen > d->bd_bufsize) totlen = d->bd_bufsize; /* * Round up the end of the previous packet to the next longword. * * Drop the packet if there's no room and no hope of room * If the packet would overflow the storage buffer or the storage * buffer is considered immutable by the buffer model, try to rotate * the buffer and wakeup pending processes. */ #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) curlen = BPF_WORDALIGN32(d->bd_slen); else #endif curlen = BPF_WORDALIGN(d->bd_slen); if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { if (d->bd_fbuf == NULL) { /* * There's no room in the store buffer, and no * prospect of room, so drop the packet. Notify the * buffer model. */ bpf_buffull(d); counter_u64_add(d->bd_dcount, 1); return; } KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); ROTATE_BUFFERS(d); do_wakeup = 1; curlen = 0; } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) /* * Immediate mode is set, or the read timeout has already * expired during a select call. A packet arrived, so the * reader should be woken up. */ do_wakeup = 1; caplen = totlen - hdrlen; tstype = d->bd_tstamp; do_timestamp = tstype != BPF_T_NONE; #ifndef BURN_BRIDGES if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { struct bpf_ts ts; if (do_timestamp) bpf_bintime2ts(bt, &ts, tstype); #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) { bzero(&hdr32_old, sizeof(hdr32_old)); if (do_timestamp) { hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; } hdr32_old.bh_datalen = pktlen; hdr32_old.bh_hdrlen = hdrlen; hdr32_old.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, sizeof(hdr32_old)); goto copy; } #endif bzero(&hdr_old, sizeof(hdr_old)); if (do_timestamp) { hdr_old.bh_tstamp.tv_sec = ts.bt_sec; hdr_old.bh_tstamp.tv_usec = ts.bt_frac; } hdr_old.bh_datalen = pktlen; hdr_old.bh_hdrlen = hdrlen; hdr_old.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, sizeof(hdr_old)); goto copy; } #endif /* * Append the bpf header. Note we append the actual header size, but * move forward the length of the header plus padding. */ bzero(&hdr, sizeof(hdr)); if (do_timestamp) bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); hdr.bh_datalen = pktlen; hdr.bh_hdrlen = hdrlen; hdr.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); /* * Copy the packet data into the store buffer and update its length. */ #ifndef BURN_BRIDGES copy: #endif (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); d->bd_slen = curlen + totlen; if (do_wakeup) bpf_wakeup(d); } /* * Free buffers currently in use by a descriptor. * Called on close. */ static void bpf_freed(struct bpf_d *d) { /* * We don't need to lock out interrupts since this descriptor has * been detached from its interface and it yet hasn't been marked * free. */ bpf_free(d); if (d->bd_rfilter != NULL) { free((caddr_t)d->bd_rfilter, M_BPF); #ifdef BPF_JITTER if (d->bd_bfilter != NULL) bpf_destroy_jit_filter(d->bd_bfilter); #endif } if (d->bd_wfilter != NULL) free((caddr_t)d->bd_wfilter, M_BPF); mtx_destroy(&d->bd_lock); counter_u64_free(d->bd_rcount); counter_u64_free(d->bd_dcount); counter_u64_free(d->bd_fcount); counter_u64_free(d->bd_wcount); counter_u64_free(d->bd_wfcount); counter_u64_free(d->bd_wdcount); counter_u64_free(d->bd_zcopy); } /* * Attach an interface to bpf. dlt is the link layer type; hdrlen is the * fixed size of the link header (variable length headers not yet supported). */ void bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) { bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); } /* * Attach an interface to bpf. ifp is a pointer to the structure * defining the interface to be attached, dlt is the link layer type, * and hdrlen is the fixed size of the link header (variable length * headers are not yet supporrted). */ void bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) { struct bpf_if *bp; bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); if (bp == NULL) panic("bpfattach"); LIST_INIT(&bp->bif_dlist); LIST_INIT(&bp->bif_wlist); bp->bif_ifp = ifp; bp->bif_dlt = dlt; rw_init(&bp->bif_lock, "bpf interface lock"); KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); bp->bif_bpf = driverp; *driverp = bp; BPF_LOCK(); LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); BPF_UNLOCK(); bp->bif_hdrlen = hdrlen; if (bootverbose && IS_DEFAULT_VNET(curvnet)) if_printf(ifp, "bpf attached\n"); } #ifdef VIMAGE /* * When moving interfaces between vnet instances we need a way to * query the dlt and hdrlen before detach so we can re-attch the if_bpf * after the vmove. We unfortunately have no device driver infrastructure * to query the interface for these values after creation/attach, thus * add this as a workaround. */ int bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen) { if (bp == NULL) return (ENXIO); if (bif_dlt == NULL && bif_hdrlen == NULL) return (0); if (bif_dlt != NULL) *bif_dlt = bp->bif_dlt; if (bif_hdrlen != NULL) *bif_hdrlen = bp->bif_hdrlen; return (0); } #endif /* * Detach bpf from an interface. This involves detaching each descriptor * associated with the interface. Notify each descriptor as it's detached * so that any sleepers wake up and get ENXIO. */ void bpfdetach(struct ifnet *ifp) { struct bpf_if *bp, *bp_temp; struct bpf_d *d; int ndetached; ndetached = 0; BPF_LOCK(); /* Find all bpf_if struct's which reference ifp and detach them. */ LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { if (ifp != bp->bif_ifp) continue; LIST_REMOVE(bp, bif_next); /* Add to to-be-freed list */ LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); ndetached++; /* * Delay freeing bp till interface is detached * and all routes through this interface are removed. * Mark bp as detached to restrict new consumers. */ BPFIF_WLOCK(bp); bp->bif_flags |= BPFIF_FLAG_DYING; *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if; BPFIF_WUNLOCK(bp); CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", __func__, bp->bif_dlt, bp, ifp); /* Free common descriptors */ while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { bpf_detachd_locked(d); BPFD_LOCK(d); bpf_wakeup(d); BPFD_UNLOCK(d); } /* Free writer-only descriptors */ while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { bpf_detachd_locked(d); BPFD_LOCK(d); bpf_wakeup(d); BPFD_UNLOCK(d); } } BPF_UNLOCK(); #ifdef INVARIANTS if (ndetached == 0) printf("bpfdetach: %s was not attached\n", ifp->if_xname); #endif } /* * Interface departure handler. * Note departure event does not guarantee interface is going down. * Interface renaming is currently done via departure/arrival event set. * * Departure handled is called after all routes pointing to * given interface are removed and interface is in down state * restricting any packets to be sent/received. We assume it is now safe * to free data allocated by BPF. */ static void bpf_ifdetach(void *arg __unused, struct ifnet *ifp) { struct bpf_if *bp, *bp_temp; int nmatched = 0; /* Ignore ifnet renaming. */ if (ifp->if_flags & IFF_RENAMING) return; BPF_LOCK(); /* * Find matching entries in free list. * Nothing should be found if bpfdetach() was not called. */ LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { if (ifp != bp->bif_ifp) continue; CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", __func__, bp, ifp); LIST_REMOVE(bp, bif_next); rw_destroy(&bp->bif_lock); free(bp, M_BPF); nmatched++; } BPF_UNLOCK(); } /* * Get a list of available data link type of the interface. */ static int bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) { struct ifnet *ifp; struct bpf_if *bp; u_int *lst; int error, n, n1; BPF_LOCK_ASSERT(); ifp = d->bd_bif->bif_ifp; again: n1 = 0; LIST_FOREACH(bp, &bpf_iflist, bif_next) { if (bp->bif_ifp == ifp) n1++; } if (bfl->bfl_list == NULL) { bfl->bfl_len = n1; return (0); } if (n1 > bfl->bfl_len) return (ENOMEM); BPF_UNLOCK(); lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); n = 0; BPF_LOCK(); LIST_FOREACH(bp, &bpf_iflist, bif_next) { if (bp->bif_ifp != ifp) continue; if (n >= n1) { free(lst, M_TEMP); goto again; } lst[n] = bp->bif_dlt; n++; } BPF_UNLOCK(); error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); free(lst, M_TEMP); BPF_LOCK(); bfl->bfl_len = n; return (error); } /* * Set the data link type of a BPF instance. */ static int bpf_setdlt(struct bpf_d *d, u_int dlt) { int error, opromisc; struct ifnet *ifp; struct bpf_if *bp; BPF_LOCK_ASSERT(); if (d->bd_bif->bif_dlt == dlt) return (0); ifp = d->bd_bif->bif_ifp; LIST_FOREACH(bp, &bpf_iflist, bif_next) { if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) break; } if (bp != NULL) { opromisc = d->bd_promisc; bpf_attachd(d, bp); BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); if (opromisc) { error = ifpromisc(bp->bif_ifp, 1); if (error) if_printf(bp->bif_ifp, "bpf_setdlt: ifpromisc failed (%d)\n", error); else d->bd_promisc = 1; } } return (bp == NULL ? EINVAL : 0); } static void bpf_drvinit(void *unused) { struct cdev *dev; sx_init(&bpf_sx, "bpf global lock"); LIST_INIT(&bpf_iflist); LIST_INIT(&bpf_freelist); dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); /* For compatibility */ make_dev_alias(dev, "bpf0"); /* Register interface departure handler */ bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( ifnet_departure_event, bpf_ifdetach, NULL, EVENTHANDLER_PRI_ANY); } /* * Zero out the various packet counters associated with all of the bpf * descriptors. At some point, we will probably want to get a bit more * granular and allow the user to specify descriptors to be zeroed. */ static void bpf_zero_counters(void) { struct bpf_if *bp; struct bpf_d *bd; BPF_LOCK(); LIST_FOREACH(bp, &bpf_iflist, bif_next) { BPFIF_RLOCK(bp); LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { BPFD_LOCK(bd); counter_u64_zero(bd->bd_rcount); counter_u64_zero(bd->bd_dcount); counter_u64_zero(bd->bd_fcount); counter_u64_zero(bd->bd_wcount); counter_u64_zero(bd->bd_wfcount); counter_u64_zero(bd->bd_zcopy); BPFD_UNLOCK(bd); } BPFIF_RUNLOCK(bp); } BPF_UNLOCK(); } /* * Fill filter statistics */ static void bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) { bzero(d, sizeof(*d)); BPFD_LOCK_ASSERT(bd); d->bd_structsize = sizeof(*d); /* XXX: reading should be protected by global lock */ d->bd_immediate = bd->bd_immediate; d->bd_promisc = bd->bd_promisc; d->bd_hdrcmplt = bd->bd_hdrcmplt; d->bd_direction = bd->bd_direction; d->bd_feedback = bd->bd_feedback; d->bd_async = bd->bd_async; d->bd_rcount = counter_u64_fetch(bd->bd_rcount); d->bd_dcount = counter_u64_fetch(bd->bd_dcount); d->bd_fcount = counter_u64_fetch(bd->bd_fcount); d->bd_sig = bd->bd_sig; d->bd_slen = bd->bd_slen; d->bd_hlen = bd->bd_hlen; d->bd_bufsize = bd->bd_bufsize; d->bd_pid = bd->bd_pid; strlcpy(d->bd_ifname, bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); d->bd_locked = bd->bd_locked; d->bd_wcount = counter_u64_fetch(bd->bd_wcount); d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount); d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount); d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy); d->bd_bufmode = bd->bd_bufmode; } /* * Handle `netstat -B' stats request */ static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) { static const struct xbpf_d zerostats; struct xbpf_d *xbdbuf, *xbd, tempstats; int index, error; struct bpf_if *bp; struct bpf_d *bd; /* * XXX This is not technically correct. It is possible for non * privileged users to open bpf devices. It would make sense * if the users who opened the devices were able to retrieve * the statistics for them, too. */ error = priv_check(req->td, PRIV_NET_BPF); if (error) return (error); /* * Check to see if the user is requesting that the counters be * zeroed out. Explicitly check that the supplied data is zeroed, * as we aren't allowing the user to set the counters currently. */ if (req->newptr != NULL) { if (req->newlen != sizeof(tempstats)) return (EINVAL); memset(&tempstats, 0, sizeof(tempstats)); error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); if (error) return (error); if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) return (EINVAL); bpf_zero_counters(); return (0); } if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); if (bpf_bpfd_cnt == 0) return (SYSCTL_OUT(req, 0, 0)); xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); BPF_LOCK(); if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { BPF_UNLOCK(); free(xbdbuf, M_BPF); return (ENOMEM); } index = 0; LIST_FOREACH(bp, &bpf_iflist, bif_next) { BPFIF_RLOCK(bp); /* Send writers-only first */ LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { xbd = &xbdbuf[index++]; BPFD_LOCK(bd); bpfstats_fill_xbpf(xbd, bd); BPFD_UNLOCK(bd); } LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { xbd = &xbdbuf[index++]; BPFD_LOCK(bd); bpfstats_fill_xbpf(xbd, bd); BPFD_UNLOCK(bd); } BPFIF_RUNLOCK(bp); } BPF_UNLOCK(); error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); free(xbdbuf, M_BPF); return (error); } SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); #else /* !DEV_BPF && !NETGRAPH_BPF */ /* * NOP stubs to allow bpf-using drivers to load and function. * * A 'better' implementation would allow the core bpf functionality * to be loaded at runtime. */ void bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) { } void bpf_mtap(struct bpf_if *bp, struct mbuf *m) { } void bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) { } void bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) { bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); } void bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) { *driverp = (struct bpf_if *)&dead_bpf_if; } void bpfdetach(struct ifnet *ifp) { } u_int bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) { return -1; /* "no filter" behaviour */ } int bpf_validate(const struct bpf_insn *f, int len) { return 0; /* false */ } #endif /* !DEV_BPF && !NETGRAPH_BPF */ #ifdef DDB static void bpf_show_bpf_if(struct bpf_if *bpf_if) { if (bpf_if == NULL) return; db_printf("%p:\n", bpf_if); #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e); /* bif_ext.bif_next */ /* bif_ext.bif_dlist */ BPF_DB_PRINTF("%#x", bif_dlt); BPF_DB_PRINTF("%u", bif_hdrlen); BPF_DB_PRINTF("%p", bif_ifp); /* bif_lock */ /* bif_wlist */ BPF_DB_PRINTF("%#x", bif_flags); } DB_SHOW_COMMAND(bpf_if, db_show_bpf_if) { if (!have_addr) { db_printf("usage: show bpf_if \n"); return; } bpf_show_bpf_if((struct bpf_if *)addr); } #endif Index: head/sys/net/ieee8023ad_lacp.c =================================================================== --- head/sys/net/ieee8023ad_lacp.c (revision 336675) +++ head/sys/net/ieee8023ad_lacp.c (revision 336676) @@ -1,2113 +1,2113 @@ /* $NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c)2005 YAMAMOTO Takashi, * Copyright (c)2008 Andrew Thompson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_ratelimit.h" #include #include #include #include #include #include #include /* hz */ #include /* for net/if.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * actor system priority and port priority. * XXX should be configurable. */ #define LACP_SYSTEM_PRIO 0x8000 #define LACP_PORT_PRIO 0x8000 const uint8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] = { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 }; static const struct tlv_template lacp_info_tlv_template[] = { { LACP_TYPE_ACTORINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, { LACP_TYPE_PARTNERINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_peerinfo) }, { LACP_TYPE_COLLECTORINFO, sizeof(struct tlvhdr) + sizeof(struct lacp_collectorinfo) }, { 0, 0 }, }; static const struct tlv_template marker_info_tlv_template[] = { { MARKER_TYPE_INFO, sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, { 0, 0 }, }; static const struct tlv_template marker_response_tlv_template[] = { { MARKER_TYPE_RESPONSE, sizeof(struct tlvhdr) + sizeof(struct lacp_markerinfo) }, { 0, 0 }, }; typedef void (*lacp_timer_func_t)(struct lacp_port *); static void lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *); static void lacp_fill_markerinfo(struct lacp_port *, struct lacp_markerinfo *); static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *); static void lacp_suppress_distributing(struct lacp_softc *, struct lacp_aggregator *); static void lacp_transit_expire(void *); static void lacp_update_portmap(struct lacp_softc *); static void lacp_select_active_aggregator(struct lacp_softc *); static uint16_t lacp_compose_key(struct lacp_port *); static int tlv_check(const void *, size_t, const struct tlvhdr *, const struct tlv_template *, boolean_t); static void lacp_tick(void *); static void lacp_fill_aggregator_id(struct lacp_aggregator *, const struct lacp_port *); static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *, const struct lacp_peerinfo *); static int lacp_aggregator_is_compatible(const struct lacp_aggregator *, const struct lacp_port *); static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *, const struct lacp_peerinfo *); static struct lacp_aggregator *lacp_aggregator_get(struct lacp_softc *, struct lacp_port *); static void lacp_aggregator_addref(struct lacp_softc *, struct lacp_aggregator *); static void lacp_aggregator_delref(struct lacp_softc *, struct lacp_aggregator *); /* receive machine */ static int lacp_pdu_input(struct lacp_port *, struct mbuf *); static int lacp_marker_input(struct lacp_port *, struct mbuf *); static void lacp_sm_rx(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_timer(struct lacp_port *); static void lacp_sm_rx_set_expired(struct lacp_port *); static void lacp_sm_rx_update_ntt(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_record_pdu(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_update_selected(struct lacp_port *, const struct lacpdu *); static void lacp_sm_rx_record_default(struct lacp_port *); static void lacp_sm_rx_update_default_selected(struct lacp_port *); static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *, const struct lacp_peerinfo *); /* mux machine */ static void lacp_sm_mux(struct lacp_port *); static void lacp_set_mux(struct lacp_port *, enum lacp_mux_state); static void lacp_sm_mux_timer(struct lacp_port *); /* periodic transmit machine */ static void lacp_sm_ptx_update_timeout(struct lacp_port *, uint8_t); static void lacp_sm_ptx_tx_schedule(struct lacp_port *); static void lacp_sm_ptx_timer(struct lacp_port *); /* transmit machine */ static void lacp_sm_tx(struct lacp_port *); static void lacp_sm_assert_ntt(struct lacp_port *); static void lacp_run_timers(struct lacp_port *); static int lacp_compare_peerinfo(const struct lacp_peerinfo *, const struct lacp_peerinfo *); static int lacp_compare_systemid(const struct lacp_systemid *, const struct lacp_systemid *); static void lacp_port_enable(struct lacp_port *); static void lacp_port_disable(struct lacp_port *); static void lacp_select(struct lacp_port *); static void lacp_unselect(struct lacp_port *); static void lacp_disable_collecting(struct lacp_port *); static void lacp_enable_collecting(struct lacp_port *); static void lacp_disable_distributing(struct lacp_port *); static void lacp_enable_distributing(struct lacp_port *); static int lacp_xmit_lacpdu(struct lacp_port *); static int lacp_xmit_marker(struct lacp_port *); /* Debugging */ static void lacp_dump_lacpdu(const struct lacpdu *); static const char *lacp_format_partner(const struct lacp_peerinfo *, char *, size_t); static const char *lacp_format_lagid(const struct lacp_peerinfo *, const struct lacp_peerinfo *, char *, size_t); static const char *lacp_format_lagid_aggregator(const struct lacp_aggregator *, char *, size_t); static const char *lacp_format_state(uint8_t, char *, size_t); static const char *lacp_format_mac(const uint8_t *, char *, size_t); static const char *lacp_format_systemid(const struct lacp_systemid *, char *, size_t); static const char *lacp_format_portid(const struct lacp_portid *, char *, size_t); static void lacp_dprintf(const struct lacp_port *, const char *, ...) __attribute__((__format__(__printf__, 2, 3))); -static VNET_DEFINE(int, lacp_debug); +VNET_DEFINE_STATIC(int, lacp_debug); #define V_lacp_debug VNET(lacp_debug) SYSCTL_NODE(_net_link_lagg, OID_AUTO, lacp, CTLFLAG_RD, 0, "ieee802.3ad"); SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, debug, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_debug), 0, "Enable LACP debug logging (1=debug, 2=trace)"); -static VNET_DEFINE(int, lacp_default_strict_mode) = 1; +VNET_DEFINE_STATIC(int, lacp_default_strict_mode) = 1; SYSCTL_INT(_net_link_lagg_lacp, OID_AUTO, default_strict_mode, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(lacp_default_strict_mode), 0, "LACP strict protocol compliance default"); #define LACP_DPRINTF(a) if (V_lacp_debug & 0x01) { lacp_dprintf a ; } #define LACP_TRACE(a) if (V_lacp_debug & 0x02) { lacp_dprintf(a,"%s\n",__func__); } #define LACP_TPRINTF(a) if (V_lacp_debug & 0x04) { lacp_dprintf a ; } /* * partner administration variables. * XXX should be configurable. */ static const struct lacp_peerinfo lacp_partner_admin_optimistic = { .lip_systemid = { .lsi_prio = 0xffff }, .lip_portid = { .lpi_prio = 0xffff }, .lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION | LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING, }; static const struct lacp_peerinfo lacp_partner_admin_strict = { .lip_systemid = { .lsi_prio = 0xffff }, .lip_portid = { .lpi_prio = 0xffff }, .lip_state = 0, }; static const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = { [LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer, [LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer, [LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer, }; struct mbuf * lacp_input(struct lagg_port *lgp, struct mbuf *m) { struct lacp_port *lp = LACP_PORT(lgp); uint8_t subtype; if (m->m_pkthdr.len < sizeof(struct ether_header) + sizeof(subtype)) { m_freem(m); return (NULL); } m_copydata(m, sizeof(struct ether_header), sizeof(subtype), &subtype); switch (subtype) { case SLOWPROTOCOLS_SUBTYPE_LACP: lacp_pdu_input(lp, m); return (NULL); case SLOWPROTOCOLS_SUBTYPE_MARKER: lacp_marker_input(lp, m); return (NULL); } /* Not a subtype we are interested in */ return (m); } /* * lacp_pdu_input: process lacpdu */ static int lacp_pdu_input(struct lacp_port *lp, struct mbuf *m) { struct lacp_softc *lsc = lp->lp_lsc; struct lacpdu *du; int error = 0; if (m->m_pkthdr.len != sizeof(*du)) { goto bad; } if ((m->m_flags & M_MCAST) == 0) { goto bad; } if (m->m_len < sizeof(*du)) { m = m_pullup(m, sizeof(*du)); if (m == NULL) { return (ENOMEM); } } du = mtod(m, struct lacpdu *); if (memcmp(&du->ldu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { goto bad; } /* * ignore the version for compatibility with * the future protocol revisions. */ #if 0 if (du->ldu_sph.sph_version != 1) { goto bad; } #endif /* * ignore tlv types for compatibility with * the future protocol revisions. */ if (tlv_check(du, sizeof(*du), &du->ldu_tlv_actor, lacp_info_tlv_template, FALSE)) { goto bad; } if (V_lacp_debug > 0) { lacp_dprintf(lp, "lacpdu receive\n"); lacp_dump_lacpdu(du); } if ((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_rx_test) { LACP_TPRINTF((lp, "Dropping RX PDU\n")); goto bad; } LACP_LOCK(lsc); lacp_sm_rx(lp, du); LACP_UNLOCK(lsc); m_freem(m); return (error); bad: m_freem(m); return (EINVAL); } static void lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO); memcpy(&info->lip_systemid.lsi_mac, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO); info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index); info->lip_state = lp->lp_state; } static void lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info) { struct ifnet *ifp = lp->lp_ifp; /* Fill in the port index and system id (encoded as the MAC) */ info->mi_rq_port = htons(ifp->if_index); memcpy(&info->mi_rq_system, lp->lp_systemid.lsi_mac, ETHER_ADDR_LEN); info->mi_rq_xid = htonl(0); } static int lacp_xmit_lacpdu(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct mbuf *m; struct lacpdu *du; int error; LACP_LOCK_ASSERT(lp->lp_lsc); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { return (ENOMEM); } m->m_len = m->m_pkthdr.len = sizeof(*du); du = mtod(m, struct lacpdu *); memset(du, 0, sizeof(*du)); memcpy(&du->ldu_eh.ether_dhost, ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&du->ldu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); du->ldu_eh.ether_type = htons(ETHERTYPE_SLOW); du->ldu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_LACP; du->ldu_sph.sph_version = 1; TLV_SET(&du->ldu_tlv_actor, LACP_TYPE_ACTORINFO, sizeof(du->ldu_actor)); du->ldu_actor = lp->lp_actor; TLV_SET(&du->ldu_tlv_partner, LACP_TYPE_PARTNERINFO, sizeof(du->ldu_partner)); du->ldu_partner = lp->lp_partner; TLV_SET(&du->ldu_tlv_collector, LACP_TYPE_COLLECTORINFO, sizeof(du->ldu_collector)); du->ldu_collector.lci_maxdelay = 0; if (V_lacp_debug > 0) { lacp_dprintf(lp, "lacpdu transmit\n"); lacp_dump_lacpdu(du); } m->m_flags |= M_MCAST; /* * XXX should use higher priority queue. * otherwise network congestion can break aggregation. */ error = lagg_enqueue(lp->lp_ifp, m); return (error); } static int lacp_xmit_marker(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct mbuf *m; struct markerdu *mdu; int error; LACP_LOCK_ASSERT(lp->lp_lsc); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { return (ENOMEM); } m->m_len = m->m_pkthdr.len = sizeof(*mdu); mdu = mtod(m, struct markerdu *); memset(mdu, 0, sizeof(*mdu)); memcpy(&mdu->mdu_eh.ether_dhost, ethermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); mdu->mdu_eh.ether_type = htons(ETHERTYPE_SLOW); mdu->mdu_sph.sph_subtype = SLOWPROTOCOLS_SUBTYPE_MARKER; mdu->mdu_sph.sph_version = 1; /* Bump the transaction id and copy over the marker info */ lp->lp_marker.mi_rq_xid = htonl(ntohl(lp->lp_marker.mi_rq_xid) + 1); TLV_SET(&mdu->mdu_tlv, MARKER_TYPE_INFO, sizeof(mdu->mdu_info)); mdu->mdu_info = lp->lp_marker; LACP_DPRINTF((lp, "marker transmit, port=%u, sys=%6D, id=%u\n", ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", ntohl(mdu->mdu_info.mi_rq_xid))); m->m_flags |= M_MCAST; error = lagg_enqueue(lp->lp_ifp, m); return (error); } void lacp_linkstate(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; struct ifnet *ifp = lgp->lp_ifp; struct ifmediareq ifmr; int error = 0; u_int media; uint8_t old_state; uint16_t old_key; bzero((char *)&ifmr, sizeof(ifmr)); error = (*ifp->if_ioctl)(ifp, SIOCGIFXMEDIA, (caddr_t)&ifmr); if (error != 0) { bzero((char *)&ifmr, sizeof(ifmr)); error = (*ifp->if_ioctl)(ifp, SIOCGIFMEDIA, (caddr_t)&ifmr); } if (error != 0) return; LACP_LOCK(lsc); media = ifmr.ifm_active; LACP_DPRINTF((lp, "media changed 0x%x -> 0x%x, ether = %d, fdx = %d, " "link = %d\n", lp->lp_media, media, IFM_TYPE(media) == IFM_ETHER, (media & IFM_FDX) != 0, ifp->if_link_state == LINK_STATE_UP)); old_state = lp->lp_state; old_key = lp->lp_key; lp->lp_media = media; /* * If the port is not an active full duplex Ethernet link then it can * not be aggregated. */ if (IFM_TYPE(media) != IFM_ETHER || (media & IFM_FDX) == 0 || ifp->if_link_state != LINK_STATE_UP) { lacp_port_disable(lp); } else { lacp_port_enable(lp); } lp->lp_key = lacp_compose_key(lp); if (old_state != lp->lp_state || old_key != lp->lp_key) { LACP_DPRINTF((lp, "-> UNSELECTED\n")); lp->lp_selected = LACP_UNSELECTED; } LACP_UNLOCK(lsc); } static void lacp_tick(void *arg) { struct lacp_softc *lsc = arg; struct lacp_port *lp; LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) continue; CURVNET_SET(lp->lp_ifp->if_vnet); lacp_run_timers(lp); lacp_select(lp); lacp_sm_mux(lp); lacp_sm_tx(lp); lacp_sm_ptx_tx_schedule(lp); CURVNET_RESTORE(); } callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); } int lacp_port_create(struct lagg_port *lgp) { struct lagg_softc *sc = lgp->lp_softc; struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_port *lp; struct ifnet *ifp = lgp->lp_ifp; struct sockaddr_dl sdl; struct ifmultiaddr *rifma = NULL; int error; link_init_sdl(ifp, (struct sockaddr *)&sdl, IFT_ETHER); sdl.sdl_alen = ETHER_ADDR_LEN; bcopy(ðermulticastaddr_slowprotocols, LLADDR(&sdl), ETHER_ADDR_LEN); error = if_addmulti(ifp, (struct sockaddr *)&sdl, &rifma); if (error) { printf("%s: ADDMULTI failed on %s\n", __func__, lgp->lp_ifp->if_xname); return (error); } lp = malloc(sizeof(struct lacp_port), M_DEVBUF, M_NOWAIT|M_ZERO); if (lp == NULL) return (ENOMEM); LACP_LOCK(lsc); lgp->lp_psc = lp; lp->lp_ifp = ifp; lp->lp_lagg = lgp; lp->lp_lsc = lsc; lp->lp_ifma = rifma; LIST_INSERT_HEAD(&lsc->lsc_ports, lp, lp_next); lacp_fill_actorinfo(lp, &lp->lp_actor); lacp_fill_markerinfo(lp, &lp->lp_marker); lp->lp_state = LACP_STATE_ACTIVITY; lp->lp_aggregator = NULL; lacp_sm_rx_set_expired(lp); LACP_UNLOCK(lsc); lacp_linkstate(lgp); return (0); } void lacp_port_destroy(struct lagg_port *lgp) { struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; int i; LACP_LOCK(lsc); for (i = 0; i < LACP_NTIMER; i++) { LACP_TIMER_DISARM(lp, i); } lacp_disable_collecting(lp); lacp_disable_distributing(lp); lacp_unselect(lp); LIST_REMOVE(lp, lp_next); LACP_UNLOCK(lsc); /* The address may have already been removed by if_purgemaddrs() */ if (!lgp->lp_detaching) if_delmulti_ifma(lp->lp_ifma); free(lp, M_DEVBUF); } void lacp_req(struct lagg_softc *sc, void *data) { struct lacp_opreq *req = (struct lacp_opreq *)data; struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_aggregator *la; bzero(req, sizeof(struct lacp_opreq)); /* * If the LACP softc is NULL, return with the opreq structure full of * zeros. It is normal for the softc to be NULL while the lagg is * being destroyed. */ if (NULL == lsc) return; la = lsc->lsc_active_aggregator; LACP_LOCK(lsc); if (la != NULL) { req->actor_prio = ntohs(la->la_actor.lip_systemid.lsi_prio); memcpy(&req->actor_mac, &la->la_actor.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->actor_key = ntohs(la->la_actor.lip_key); req->actor_portprio = ntohs(la->la_actor.lip_portid.lpi_prio); req->actor_portno = ntohs(la->la_actor.lip_portid.lpi_portno); req->actor_state = la->la_actor.lip_state; req->partner_prio = ntohs(la->la_partner.lip_systemid.lsi_prio); memcpy(&req->partner_mac, &la->la_partner.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->partner_key = ntohs(la->la_partner.lip_key); req->partner_portprio = ntohs(la->la_partner.lip_portid.lpi_prio); req->partner_portno = ntohs(la->la_partner.lip_portid.lpi_portno); req->partner_state = la->la_partner.lip_state; } LACP_UNLOCK(lsc); } void lacp_portreq(struct lagg_port *lgp, void *data) { struct lacp_opreq *req = (struct lacp_opreq *)data; struct lacp_port *lp = LACP_PORT(lgp); struct lacp_softc *lsc = lp->lp_lsc; LACP_LOCK(lsc); req->actor_prio = ntohs(lp->lp_actor.lip_systemid.lsi_prio); memcpy(&req->actor_mac, &lp->lp_actor.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->actor_key = ntohs(lp->lp_actor.lip_key); req->actor_portprio = ntohs(lp->lp_actor.lip_portid.lpi_prio); req->actor_portno = ntohs(lp->lp_actor.lip_portid.lpi_portno); req->actor_state = lp->lp_actor.lip_state; req->partner_prio = ntohs(lp->lp_partner.lip_systemid.lsi_prio); memcpy(&req->partner_mac, &lp->lp_partner.lip_systemid.lsi_mac, ETHER_ADDR_LEN); req->partner_key = ntohs(lp->lp_partner.lip_key); req->partner_portprio = ntohs(lp->lp_partner.lip_portid.lpi_prio); req->partner_portno = ntohs(lp->lp_partner.lip_portid.lpi_portno); req->partner_state = lp->lp_partner.lip_state; LACP_UNLOCK(lsc); } static void lacp_disable_collecting(struct lacp_port *lp) { LACP_DPRINTF((lp, "collecting disabled\n")); lp->lp_state &= ~LACP_STATE_COLLECTING; } static void lacp_enable_collecting(struct lacp_port *lp) { LACP_DPRINTF((lp, "collecting enabled\n")); lp->lp_state |= LACP_STATE_COLLECTING; } static void lacp_disable_distributing(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; struct lacp_softc *lsc = lp->lp_lsc; struct lagg_softc *sc = lsc->lsc_softc; char buf[LACP_LAGIDSTR_MAX+1]; LACP_LOCK_ASSERT(lsc); if (la == NULL || (lp->lp_state & LACP_STATE_DISTRIBUTING) == 0) { return; } KASSERT(!TAILQ_EMPTY(&la->la_ports), ("no aggregator ports")); KASSERT(la->la_nports > 0, ("nports invalid (%d)", la->la_nports)); KASSERT(la->la_refcnt >= la->la_nports, ("aggregator refcnt invalid")); LACP_DPRINTF((lp, "disable distributing on aggregator %s, " "nports %d -> %d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), la->la_nports, la->la_nports - 1)); TAILQ_REMOVE(&la->la_ports, lp, lp_dist_q); la->la_nports--; sc->sc_active = la->la_nports; if (lsc->lsc_active_aggregator == la) { lacp_suppress_distributing(lsc, la); lacp_select_active_aggregator(lsc); /* regenerate the port map, the active aggregator has changed */ lacp_update_portmap(lsc); } lp->lp_state &= ~LACP_STATE_DISTRIBUTING; } static void lacp_enable_distributing(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; struct lacp_softc *lsc = lp->lp_lsc; struct lagg_softc *sc = lsc->lsc_softc; char buf[LACP_LAGIDSTR_MAX+1]; LACP_LOCK_ASSERT(lsc); if ((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0) { return; } LACP_DPRINTF((lp, "enable distributing on aggregator %s, " "nports %d -> %d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), la->la_nports, la->la_nports + 1)); KASSERT(la->la_refcnt > la->la_nports, ("aggregator refcnt invalid")); TAILQ_INSERT_HEAD(&la->la_ports, lp, lp_dist_q); la->la_nports++; sc->sc_active = la->la_nports; lp->lp_state |= LACP_STATE_DISTRIBUTING; if (lsc->lsc_active_aggregator == la) { lacp_suppress_distributing(lsc, la); lacp_update_portmap(lsc); } else /* try to become the active aggregator */ lacp_select_active_aggregator(lsc); } static void lacp_transit_expire(void *vp) { struct lacp_softc *lsc = vp; LACP_LOCK_ASSERT(lsc); CURVNET_SET(lsc->lsc_softc->sc_ifp->if_vnet); LACP_TRACE(NULL); CURVNET_RESTORE(); lsc->lsc_suppress_distributing = FALSE; } void lacp_attach(struct lagg_softc *sc) { struct lacp_softc *lsc; lsc = malloc(sizeof(struct lacp_softc), M_DEVBUF, M_WAITOK | M_ZERO); sc->sc_psc = lsc; lsc->lsc_softc = sc; lsc->lsc_hashkey = m_ether_tcpip_hash_init(); lsc->lsc_active_aggregator = NULL; lsc->lsc_strict_mode = VNET(lacp_default_strict_mode); LACP_LOCK_INIT(lsc); TAILQ_INIT(&lsc->lsc_aggregators); LIST_INIT(&lsc->lsc_ports); callout_init_mtx(&lsc->lsc_transit_callout, &lsc->lsc_mtx, 0); callout_init_mtx(&lsc->lsc_callout, &lsc->lsc_mtx, 0); /* if the lagg is already up then do the same */ if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) lacp_init(sc); } void lacp_detach(void *psc) { struct lacp_softc *lsc = (struct lacp_softc *)psc; KASSERT(TAILQ_EMPTY(&lsc->lsc_aggregators), ("aggregators still active")); KASSERT(lsc->lsc_active_aggregator == NULL, ("aggregator still attached")); callout_drain(&lsc->lsc_transit_callout); callout_drain(&lsc->lsc_callout); LACP_LOCK_DESTROY(lsc); free(lsc, M_DEVBUF); } void lacp_init(struct lagg_softc *sc) { struct lacp_softc *lsc = LACP_SOFTC(sc); LACP_LOCK(lsc); callout_reset(&lsc->lsc_callout, hz, lacp_tick, lsc); LACP_UNLOCK(lsc); } void lacp_stop(struct lagg_softc *sc) { struct lacp_softc *lsc = LACP_SOFTC(sc); LACP_LOCK(lsc); callout_stop(&lsc->lsc_transit_callout); callout_stop(&lsc->lsc_callout); LACP_UNLOCK(lsc); } struct lagg_port * lacp_select_tx_port(struct lagg_softc *sc, struct mbuf *m) { struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_portmap *pm; struct lacp_port *lp; uint32_t hash; if (__predict_false(lsc->lsc_suppress_distributing)) { LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); return (NULL); } pm = &lsc->lsc_pmap[lsc->lsc_activemap]; if (pm->pm_count == 0) { LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); return (NULL); } if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) hash = m->m_pkthdr.flowid >> sc->flowid_shift; else hash = m_ether_tcpip_hash(sc->sc_flags, m, lsc->lsc_hashkey); hash %= pm->pm_count; lp = pm->pm_map[hash]; KASSERT((lp->lp_state & LACP_STATE_DISTRIBUTING) != 0, ("aggregated port is not distributing")); return (lp->lp_lagg); } #ifdef RATELIMIT struct lagg_port * lacp_select_tx_port_by_hash(struct lagg_softc *sc, uint32_t flowid) { struct lacp_softc *lsc = LACP_SOFTC(sc); struct lacp_portmap *pm; struct lacp_port *lp; uint32_t hash; if (__predict_false(lsc->lsc_suppress_distributing)) { LACP_DPRINTF((NULL, "%s: waiting transit\n", __func__)); return (NULL); } pm = &lsc->lsc_pmap[lsc->lsc_activemap]; if (pm->pm_count == 0) { LACP_DPRINTF((NULL, "%s: no active aggregator\n", __func__)); return (NULL); } hash = flowid >> sc->flowid_shift; hash %= pm->pm_count; lp = pm->pm_map[hash]; return (lp->lp_lagg); } #endif /* * lacp_suppress_distributing: drop transmit packets for a while * to preserve packet ordering. */ static void lacp_suppress_distributing(struct lacp_softc *lsc, struct lacp_aggregator *la) { struct lacp_port *lp; if (lsc->lsc_active_aggregator != la) { return; } LACP_TRACE(NULL); lsc->lsc_suppress_distributing = TRUE; /* send a marker frame down each port to verify the queues are empty */ LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) { lp->lp_flags |= LACP_PORT_MARK; lacp_xmit_marker(lp); } /* set a timeout for the marker frames */ callout_reset(&lsc->lsc_transit_callout, LACP_TRANSIT_DELAY * hz / 1000, lacp_transit_expire, lsc); } static int lacp_compare_peerinfo(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b) { return (memcmp(a, b, offsetof(struct lacp_peerinfo, lip_state))); } static int lacp_compare_systemid(const struct lacp_systemid *a, const struct lacp_systemid *b) { return (memcmp(a, b, sizeof(*a))); } #if 0 /* unused */ static int lacp_compare_portid(const struct lacp_portid *a, const struct lacp_portid *b) { return (memcmp(a, b, sizeof(*a))); } #endif static uint64_t lacp_aggregator_bandwidth(struct lacp_aggregator *la) { struct lacp_port *lp; uint64_t speed; lp = TAILQ_FIRST(&la->la_ports); if (lp == NULL) { return (0); } speed = ifmedia_baudrate(lp->lp_media); speed *= la->la_nports; if (speed == 0) { LACP_DPRINTF((lp, "speed 0? media=0x%x nports=%d\n", lp->lp_media, la->la_nports)); } return (speed); } /* * lacp_select_active_aggregator: select an aggregator to be used to transmit * packets from lagg(4) interface. */ static void lacp_select_active_aggregator(struct lacp_softc *lsc) { struct lacp_aggregator *la; struct lacp_aggregator *best_la = NULL; uint64_t best_speed = 0; char buf[LACP_LAGIDSTR_MAX+1]; LACP_TRACE(NULL); TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { uint64_t speed; if (la->la_nports == 0) { continue; } speed = lacp_aggregator_bandwidth(la); LACP_DPRINTF((NULL, "%s, speed=%jd, nports=%d\n", lacp_format_lagid_aggregator(la, buf, sizeof(buf)), speed, la->la_nports)); /* * This aggregator is chosen if the partner has a better * system priority or, the total aggregated speed is higher * or, it is already the chosen aggregator */ if ((best_la != NULL && LACP_SYS_PRI(la->la_partner) < LACP_SYS_PRI(best_la->la_partner)) || speed > best_speed || (speed == best_speed && la == lsc->lsc_active_aggregator)) { best_la = la; best_speed = speed; } } KASSERT(best_la == NULL || best_la->la_nports > 0, ("invalid aggregator refcnt")); KASSERT(best_la == NULL || !TAILQ_EMPTY(&best_la->la_ports), ("invalid aggregator list")); if (lsc->lsc_active_aggregator != best_la) { LACP_DPRINTF((NULL, "active aggregator changed\n")); LACP_DPRINTF((NULL, "old %s\n", lacp_format_lagid_aggregator(lsc->lsc_active_aggregator, buf, sizeof(buf)))); } else { LACP_DPRINTF((NULL, "active aggregator not changed\n")); } LACP_DPRINTF((NULL, "new %s\n", lacp_format_lagid_aggregator(best_la, buf, sizeof(buf)))); if (lsc->lsc_active_aggregator != best_la) { lsc->lsc_active_aggregator = best_la; lacp_update_portmap(lsc); if (best_la) { lacp_suppress_distributing(lsc, best_la); } } } /* * Updated the inactive portmap array with the new list of ports and * make it live. */ static void lacp_update_portmap(struct lacp_softc *lsc) { struct lagg_softc *sc = lsc->lsc_softc; struct lacp_aggregator *la; struct lacp_portmap *p; struct lacp_port *lp; uint64_t speed; u_int newmap; int i; newmap = lsc->lsc_activemap == 0 ? 1 : 0; p = &lsc->lsc_pmap[newmap]; la = lsc->lsc_active_aggregator; speed = 0; bzero(p, sizeof(struct lacp_portmap)); if (la != NULL && la->la_nports > 0) { p->pm_count = la->la_nports; i = 0; TAILQ_FOREACH(lp, &la->la_ports, lp_dist_q) p->pm_map[i++] = lp; KASSERT(i == p->pm_count, ("Invalid port count")); speed = lacp_aggregator_bandwidth(la); } sc->sc_ifp->if_baudrate = speed; /* switch the active portmap over */ atomic_store_rel_int(&lsc->lsc_activemap, newmap); LACP_DPRINTF((NULL, "Set table %d with %d ports\n", lsc->lsc_activemap, lsc->lsc_pmap[lsc->lsc_activemap].pm_count)); } static uint16_t lacp_compose_key(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; u_int media = lp->lp_media; uint16_t key; if ((lp->lp_state & LACP_STATE_AGGREGATION) == 0) { /* * non-aggregatable links should have unique keys. * * XXX this isn't really unique as if_index is 16 bit. */ /* bit 0..14: (some bits of) if_index of this port */ key = lp->lp_ifp->if_index; /* bit 15: 1 */ key |= 0x8000; } else { u_int subtype = IFM_SUBTYPE(media); KASSERT(IFM_TYPE(media) == IFM_ETHER, ("invalid media type")); KASSERT((media & IFM_FDX) != 0, ("aggregating HDX interface")); /* bit 0..4: IFM_SUBTYPE modulo speed */ switch (subtype) { case IFM_10_T: case IFM_10_2: case IFM_10_5: case IFM_10_STP: case IFM_10_FL: key = IFM_10_T; break; case IFM_100_TX: case IFM_100_FX: case IFM_100_T4: case IFM_100_VG: case IFM_100_T2: case IFM_100_T: key = IFM_100_TX; break; case IFM_1000_SX: case IFM_1000_LX: case IFM_1000_CX: case IFM_1000_T: case IFM_1000_KX: case IFM_1000_SGMII: case IFM_1000_CX_SGMII: key = IFM_1000_SX; break; case IFM_10G_LR: case IFM_10G_SR: case IFM_10G_CX4: case IFM_10G_TWINAX: case IFM_10G_TWINAX_LONG: case IFM_10G_LRM: case IFM_10G_T: case IFM_10G_KX4: case IFM_10G_KR: case IFM_10G_CR1: case IFM_10G_ER: case IFM_10G_SFI: case IFM_10G_AOC: key = IFM_10G_LR; break; case IFM_20G_KR2: key = IFM_20G_KR2; break; case IFM_2500_KX: case IFM_2500_T: key = IFM_2500_KX; break; case IFM_5000_T: key = IFM_5000_T; break; case IFM_50G_PCIE: case IFM_50G_CR2: case IFM_50G_KR2: key = IFM_50G_PCIE; break; case IFM_56G_R4: key = IFM_56G_R4; break; case IFM_25G_PCIE: case IFM_25G_CR: case IFM_25G_KR: case IFM_25G_SR: case IFM_25G_LR: case IFM_25G_ACC: case IFM_25G_AOC: key = IFM_25G_PCIE; break; case IFM_40G_CR4: case IFM_40G_SR4: case IFM_40G_LR4: case IFM_40G_XLPPI: case IFM_40G_KR4: key = IFM_40G_CR4; break; case IFM_100G_CR4: case IFM_100G_SR4: case IFM_100G_KR4: case IFM_100G_LR4: key = IFM_100G_CR4; break; default: key = subtype; break; } /* bit 5..14: (some bits of) if_index of lagg device */ key |= 0x7fe0 & ((sc->sc_ifp->if_index) << 5); /* bit 15: 0 */ } return (htons(key)); } static void lacp_aggregator_addref(struct lacp_softc *lsc, struct lacp_aggregator *la) { char buf[LACP_LAGIDSTR_MAX+1]; LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_refcnt, la->la_refcnt + 1)); KASSERT(la->la_refcnt > 0, ("refcount <= 0")); la->la_refcnt++; KASSERT(la->la_refcnt > la->la_nports, ("invalid refcount")); } static void lacp_aggregator_delref(struct lacp_softc *lsc, struct lacp_aggregator *la) { char buf[LACP_LAGIDSTR_MAX+1]; LACP_DPRINTF((NULL, "%s: lagid=%s, refcnt %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_refcnt, la->la_refcnt - 1)); KASSERT(la->la_refcnt > la->la_nports, ("invalid refcnt")); la->la_refcnt--; if (la->la_refcnt > 0) { return; } KASSERT(la->la_refcnt == 0, ("refcount not zero")); KASSERT(lsc->lsc_active_aggregator != la, ("aggregator active")); TAILQ_REMOVE(&lsc->lsc_aggregators, la, la_q); free(la, M_DEVBUF); } /* * lacp_aggregator_get: allocate an aggregator. */ static struct lacp_aggregator * lacp_aggregator_get(struct lacp_softc *lsc, struct lacp_port *lp) { struct lacp_aggregator *la; la = malloc(sizeof(*la), M_DEVBUF, M_NOWAIT); if (la) { la->la_refcnt = 1; la->la_nports = 0; TAILQ_INIT(&la->la_ports); la->la_pending = 0; TAILQ_INSERT_TAIL(&lsc->lsc_aggregators, la, la_q); } return (la); } /* * lacp_fill_aggregator_id: setup a newly allocated aggregator from a port. */ static void lacp_fill_aggregator_id(struct lacp_aggregator *la, const struct lacp_port *lp) { lacp_fill_aggregator_id_peer(&la->la_partner, &lp->lp_partner); lacp_fill_aggregator_id_peer(&la->la_actor, &lp->lp_actor); la->la_actor.lip_state = lp->lp_state & LACP_STATE_AGGREGATION; } static void lacp_fill_aggregator_id_peer(struct lacp_peerinfo *lpi_aggr, const struct lacp_peerinfo *lpi_port) { memset(lpi_aggr, 0, sizeof(*lpi_aggr)); lpi_aggr->lip_systemid = lpi_port->lip_systemid; lpi_aggr->lip_key = lpi_port->lip_key; } /* * lacp_aggregator_is_compatible: check if a port can join to an aggregator. */ static int lacp_aggregator_is_compatible(const struct lacp_aggregator *la, const struct lacp_port *lp) { if (!(lp->lp_state & LACP_STATE_AGGREGATION) || !(lp->lp_partner.lip_state & LACP_STATE_AGGREGATION)) { return (0); } if (!(la->la_actor.lip_state & LACP_STATE_AGGREGATION)) { return (0); } if (!lacp_peerinfo_is_compatible(&la->la_partner, &lp->lp_partner)) { return (0); } if (!lacp_peerinfo_is_compatible(&la->la_actor, &lp->lp_actor)) { return (0); } return (1); } static int lacp_peerinfo_is_compatible(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b) { if (memcmp(&a->lip_systemid, &b->lip_systemid, sizeof(a->lip_systemid))) { return (0); } if (memcmp(&a->lip_key, &b->lip_key, sizeof(a->lip_key))) { return (0); } return (1); } static void lacp_port_enable(struct lacp_port *lp) { lp->lp_state |= LACP_STATE_AGGREGATION; } static void lacp_port_disable(struct lacp_port *lp) { lacp_set_mux(lp, LACP_MUX_DETACHED); lp->lp_state &= ~LACP_STATE_AGGREGATION; lp->lp_selected = LACP_UNSELECTED; lacp_sm_rx_record_default(lp); lp->lp_partner.lip_state &= ~LACP_STATE_AGGREGATION; lp->lp_state &= ~LACP_STATE_EXPIRED; } /* * lacp_select: select an aggregator. create one if necessary. */ static void lacp_select(struct lacp_port *lp) { struct lacp_softc *lsc = lp->lp_lsc; struct lacp_aggregator *la; char buf[LACP_LAGIDSTR_MAX+1]; if (lp->lp_aggregator) { return; } /* If we haven't heard from our peer, skip this step. */ if (lp->lp_state & LACP_STATE_DEFAULTED) return; KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); LACP_DPRINTF((lp, "port lagid=%s\n", lacp_format_lagid(&lp->lp_actor, &lp->lp_partner, buf, sizeof(buf)))); TAILQ_FOREACH(la, &lsc->lsc_aggregators, la_q) { if (lacp_aggregator_is_compatible(la, lp)) { break; } } if (la == NULL) { la = lacp_aggregator_get(lsc, lp); if (la == NULL) { LACP_DPRINTF((lp, "aggregator creation failed\n")); /* * will retry on the next tick. */ return; } lacp_fill_aggregator_id(la, lp); LACP_DPRINTF((lp, "aggregator created\n")); } else { LACP_DPRINTF((lp, "compatible aggregator found\n")); if (la->la_refcnt == LACP_MAX_PORTS) return; lacp_aggregator_addref(lsc, la); } LACP_DPRINTF((lp, "aggregator lagid=%s\n", lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)))); lp->lp_aggregator = la; lp->lp_selected = LACP_SELECTED; } /* * lacp_unselect: finish unselect/detach process. */ static void lacp_unselect(struct lacp_port *lp) { struct lacp_softc *lsc = lp->lp_lsc; struct lacp_aggregator *la = lp->lp_aggregator; KASSERT(!LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); if (la == NULL) { return; } lp->lp_aggregator = NULL; lacp_aggregator_delref(lsc, la); } /* mux machine */ static void lacp_sm_mux(struct lacp_port *lp) { struct lagg_port *lgp = lp->lp_lagg; struct lagg_softc *sc = lgp->lp_softc; enum lacp_mux_state new_state; boolean_t p_sync = (lp->lp_partner.lip_state & LACP_STATE_SYNC) != 0; boolean_t p_collecting = (lp->lp_partner.lip_state & LACP_STATE_COLLECTING) != 0; enum lacp_selected selected = lp->lp_selected; struct lacp_aggregator *la; if (V_lacp_debug > 1) lacp_dprintf(lp, "%s: state= 0x%x, selected= 0x%x, " "p_sync= 0x%x, p_collecting= 0x%x\n", __func__, lp->lp_mux_state, selected, p_sync, p_collecting); re_eval: la = lp->lp_aggregator; KASSERT(lp->lp_mux_state == LACP_MUX_DETACHED || la != NULL, ("MUX not detached")); new_state = lp->lp_mux_state; switch (lp->lp_mux_state) { case LACP_MUX_DETACHED: if (selected != LACP_UNSELECTED) { new_state = LACP_MUX_WAITING; } break; case LACP_MUX_WAITING: KASSERT(la->la_pending > 0 || !LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE), ("timer_wait_while still active")); if (selected == LACP_SELECTED && la->la_pending == 0) { new_state = LACP_MUX_ATTACHED; } else if (selected == LACP_UNSELECTED) { new_state = LACP_MUX_DETACHED; } break; case LACP_MUX_ATTACHED: if (selected == LACP_SELECTED && p_sync) { new_state = LACP_MUX_COLLECTING; } else if (selected != LACP_SELECTED) { new_state = LACP_MUX_DETACHED; } break; case LACP_MUX_COLLECTING: if (selected == LACP_SELECTED && p_sync && p_collecting) { new_state = LACP_MUX_DISTRIBUTING; } else if (selected != LACP_SELECTED || !p_sync) { new_state = LACP_MUX_ATTACHED; } break; case LACP_MUX_DISTRIBUTING: if (selected != LACP_SELECTED || !p_sync || !p_collecting) { new_state = LACP_MUX_COLLECTING; lacp_dprintf(lp, "Interface stopped DISTRIBUTING, possible flapping\n"); sc->sc_flapping++; } break; default: panic("%s: unknown state", __func__); } if (lp->lp_mux_state == new_state) { return; } lacp_set_mux(lp, new_state); goto re_eval; } static void lacp_set_mux(struct lacp_port *lp, enum lacp_mux_state new_state) { struct lacp_aggregator *la = lp->lp_aggregator; if (lp->lp_mux_state == new_state) { return; } switch (new_state) { case LACP_MUX_DETACHED: lp->lp_state &= ~LACP_STATE_SYNC; lacp_disable_distributing(lp); lacp_disable_collecting(lp); lacp_sm_assert_ntt(lp); /* cancel timer */ if (LACP_TIMER_ISARMED(lp, LACP_TIMER_WAIT_WHILE)) { KASSERT(la->la_pending > 0, ("timer_wait_while not active")); la->la_pending--; } LACP_TIMER_DISARM(lp, LACP_TIMER_WAIT_WHILE); lacp_unselect(lp); break; case LACP_MUX_WAITING: LACP_TIMER_ARM(lp, LACP_TIMER_WAIT_WHILE, LACP_AGGREGATE_WAIT_TIME); la->la_pending++; break; case LACP_MUX_ATTACHED: lp->lp_state |= LACP_STATE_SYNC; lacp_disable_collecting(lp); lacp_sm_assert_ntt(lp); break; case LACP_MUX_COLLECTING: lacp_enable_collecting(lp); lacp_disable_distributing(lp); lacp_sm_assert_ntt(lp); break; case LACP_MUX_DISTRIBUTING: lacp_enable_distributing(lp); break; default: panic("%s: unknown state", __func__); } LACP_DPRINTF((lp, "mux_state %d -> %d\n", lp->lp_mux_state, new_state)); lp->lp_mux_state = new_state; } static void lacp_sm_mux_timer(struct lacp_port *lp) { struct lacp_aggregator *la = lp->lp_aggregator; char buf[LACP_LAGIDSTR_MAX+1]; KASSERT(la->la_pending > 0, ("no pending event")); LACP_DPRINTF((lp, "%s: aggregator %s, pending %d -> %d\n", __func__, lacp_format_lagid(&la->la_actor, &la->la_partner, buf, sizeof(buf)), la->la_pending, la->la_pending - 1)); la->la_pending--; } /* periodic transmit machine */ static void lacp_sm_ptx_update_timeout(struct lacp_port *lp, uint8_t oldpstate) { if (LACP_STATE_EQ(oldpstate, lp->lp_partner.lip_state, LACP_STATE_TIMEOUT)) { return; } LACP_DPRINTF((lp, "partner timeout changed\n")); /* * FAST_PERIODIC -> SLOW_PERIODIC * or * SLOW_PERIODIC (-> PERIODIC_TX) -> FAST_PERIODIC * * let lacp_sm_ptx_tx_schedule to update timeout. */ LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); /* * if timeout has been shortened, assert NTT. */ if ((lp->lp_partner.lip_state & LACP_STATE_TIMEOUT)) { lacp_sm_assert_ntt(lp); } } static void lacp_sm_ptx_tx_schedule(struct lacp_port *lp) { int timeout; if (!(lp->lp_state & LACP_STATE_ACTIVITY) && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) { /* * NO_PERIODIC */ LACP_TIMER_DISARM(lp, LACP_TIMER_PERIODIC); return; } if (LACP_TIMER_ISARMED(lp, LACP_TIMER_PERIODIC)) { return; } timeout = (lp->lp_partner.lip_state & LACP_STATE_TIMEOUT) ? LACP_FAST_PERIODIC_TIME : LACP_SLOW_PERIODIC_TIME; LACP_TIMER_ARM(lp, LACP_TIMER_PERIODIC, timeout); } static void lacp_sm_ptx_timer(struct lacp_port *lp) { lacp_sm_assert_ntt(lp); } static void lacp_sm_rx(struct lacp_port *lp, const struct lacpdu *du) { int timeout; /* * check LACP_DISABLED first */ if (!(lp->lp_state & LACP_STATE_AGGREGATION)) { return; } /* * check loopback condition. */ if (!lacp_compare_systemid(&du->ldu_actor.lip_systemid, &lp->lp_actor.lip_systemid)) { return; } /* * EXPIRED, DEFAULTED, CURRENT -> CURRENT */ lacp_sm_rx_update_selected(lp, du); lacp_sm_rx_update_ntt(lp, du); lacp_sm_rx_record_pdu(lp, du); timeout = (lp->lp_state & LACP_STATE_TIMEOUT) ? LACP_SHORT_TIMEOUT_TIME : LACP_LONG_TIMEOUT_TIME; LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, timeout); lp->lp_state &= ~LACP_STATE_EXPIRED; /* * kick transmit machine without waiting the next tick. */ lacp_sm_tx(lp); } static void lacp_sm_rx_set_expired(struct lacp_port *lp) { lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; lp->lp_partner.lip_state |= LACP_STATE_TIMEOUT; LACP_TIMER_ARM(lp, LACP_TIMER_CURRENT_WHILE, LACP_SHORT_TIMEOUT_TIME); lp->lp_state |= LACP_STATE_EXPIRED; } static void lacp_sm_rx_timer(struct lacp_port *lp) { if ((lp->lp_state & LACP_STATE_EXPIRED) == 0) { /* CURRENT -> EXPIRED */ LACP_DPRINTF((lp, "%s: CURRENT -> EXPIRED\n", __func__)); lacp_sm_rx_set_expired(lp); } else { /* EXPIRED -> DEFAULTED */ LACP_DPRINTF((lp, "%s: EXPIRED -> DEFAULTED\n", __func__)); lacp_sm_rx_update_default_selected(lp); lacp_sm_rx_record_default(lp); lp->lp_state &= ~LACP_STATE_EXPIRED; } } static void lacp_sm_rx_record_pdu(struct lacp_port *lp, const struct lacpdu *du) { boolean_t active; uint8_t oldpstate; char buf[LACP_STATESTR_MAX+1]; LACP_TRACE(lp); oldpstate = lp->lp_partner.lip_state; active = (du->ldu_actor.lip_state & LACP_STATE_ACTIVITY) || ((lp->lp_state & LACP_STATE_ACTIVITY) && (du->ldu_partner.lip_state & LACP_STATE_ACTIVITY)); lp->lp_partner = du->ldu_actor; if (active && ((LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, LACP_STATE_AGGREGATION) && !lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner)) || (du->ldu_partner.lip_state & LACP_STATE_AGGREGATION) == 0)) { /* * XXX Maintain legacy behavior of leaving the * LACP_STATE_SYNC bit unchanged from the partner's * advertisement if lsc_strict_mode is false. * TODO: We should re-examine the concept of the "strict mode" * to ensure it makes sense to maintain a non-strict mode. */ if (lp->lp_lsc->lsc_strict_mode) lp->lp_partner.lip_state |= LACP_STATE_SYNC; } else { lp->lp_partner.lip_state &= ~LACP_STATE_SYNC; } lp->lp_state &= ~LACP_STATE_DEFAULTED; if (oldpstate != lp->lp_partner.lip_state) { LACP_DPRINTF((lp, "old pstate %s\n", lacp_format_state(oldpstate, buf, sizeof(buf)))); LACP_DPRINTF((lp, "new pstate %s\n", lacp_format_state(lp->lp_partner.lip_state, buf, sizeof(buf)))); } lacp_sm_ptx_update_timeout(lp, oldpstate); } static void lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du) { LACP_TRACE(lp); if (lacp_compare_peerinfo(&lp->lp_actor, &du->ldu_partner) || !LACP_STATE_EQ(lp->lp_state, du->ldu_partner.lip_state, LACP_STATE_ACTIVITY | LACP_STATE_SYNC | LACP_STATE_AGGREGATION)) { LACP_DPRINTF((lp, "%s: assert ntt\n", __func__)); lacp_sm_assert_ntt(lp); } } static void lacp_sm_rx_record_default(struct lacp_port *lp) { uint8_t oldpstate; LACP_TRACE(lp); oldpstate = lp->lp_partner.lip_state; if (lp->lp_lsc->lsc_strict_mode) lp->lp_partner = lacp_partner_admin_strict; else lp->lp_partner = lacp_partner_admin_optimistic; lp->lp_state |= LACP_STATE_DEFAULTED; lacp_sm_ptx_update_timeout(lp, oldpstate); } static void lacp_sm_rx_update_selected_from_peerinfo(struct lacp_port *lp, const struct lacp_peerinfo *info) { LACP_TRACE(lp); if (lacp_compare_peerinfo(&lp->lp_partner, info) || !LACP_STATE_EQ(lp->lp_partner.lip_state, info->lip_state, LACP_STATE_AGGREGATION)) { lp->lp_selected = LACP_UNSELECTED; /* mux machine will clean up lp->lp_aggregator */ } } static void lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du) { LACP_TRACE(lp); lacp_sm_rx_update_selected_from_peerinfo(lp, &du->ldu_actor); } static void lacp_sm_rx_update_default_selected(struct lacp_port *lp) { LACP_TRACE(lp); if (lp->lp_lsc->lsc_strict_mode) lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin_strict); else lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin_optimistic); } /* transmit machine */ static void lacp_sm_tx(struct lacp_port *lp) { int error = 0; if (!(lp->lp_state & LACP_STATE_AGGREGATION) #if 1 || (!(lp->lp_state & LACP_STATE_ACTIVITY) && !(lp->lp_partner.lip_state & LACP_STATE_ACTIVITY)) #endif ) { lp->lp_flags &= ~LACP_PORT_NTT; } if (!(lp->lp_flags & LACP_PORT_NTT)) { return; } /* Rate limit to 3 PDUs per LACP_FAST_PERIODIC_TIME */ if (ppsratecheck(&lp->lp_last_lacpdu, &lp->lp_lacpdu_sent, (3 / LACP_FAST_PERIODIC_TIME)) == 0) { LACP_DPRINTF((lp, "rate limited pdu\n")); return; } if (((1 << lp->lp_ifp->if_dunit) & lp->lp_lsc->lsc_debug.lsc_tx_test) == 0) { error = lacp_xmit_lacpdu(lp); } else { LACP_TPRINTF((lp, "Dropping TX PDU\n")); } if (error == 0) { lp->lp_flags &= ~LACP_PORT_NTT; } else { LACP_DPRINTF((lp, "lacpdu transmit failure, error %d\n", error)); } } static void lacp_sm_assert_ntt(struct lacp_port *lp) { lp->lp_flags |= LACP_PORT_NTT; } static void lacp_run_timers(struct lacp_port *lp) { int i; for (i = 0; i < LACP_NTIMER; i++) { KASSERT(lp->lp_timer[i] >= 0, ("invalid timer value %d", lp->lp_timer[i])); if (lp->lp_timer[i] == 0) { continue; } else if (--lp->lp_timer[i] <= 0) { if (lacp_timer_funcs[i]) { (*lacp_timer_funcs[i])(lp); } } } } int lacp_marker_input(struct lacp_port *lp, struct mbuf *m) { struct lacp_softc *lsc = lp->lp_lsc; struct lagg_port *lgp = lp->lp_lagg; struct lacp_port *lp2; struct markerdu *mdu; int error = 0; int pending = 0; if (m->m_pkthdr.len != sizeof(*mdu)) { goto bad; } if ((m->m_flags & M_MCAST) == 0) { goto bad; } if (m->m_len < sizeof(*mdu)) { m = m_pullup(m, sizeof(*mdu)); if (m == NULL) { return (ENOMEM); } } mdu = mtod(m, struct markerdu *); if (memcmp(&mdu->mdu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN)) { goto bad; } if (mdu->mdu_sph.sph_version != 1) { goto bad; } switch (mdu->mdu_tlv.tlv_type) { case MARKER_TYPE_INFO: if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, marker_info_tlv_template, TRUE)) { goto bad; } mdu->mdu_tlv.tlv_type = MARKER_TYPE_RESPONSE; memcpy(&mdu->mdu_eh.ether_dhost, ðermulticastaddr_slowprotocols, ETHER_ADDR_LEN); memcpy(&mdu->mdu_eh.ether_shost, lgp->lp_lladdr, ETHER_ADDR_LEN); error = lagg_enqueue(lp->lp_ifp, m); break; case MARKER_TYPE_RESPONSE: if (tlv_check(mdu, sizeof(*mdu), &mdu->mdu_tlv, marker_response_tlv_template, TRUE)) { goto bad; } LACP_DPRINTF((lp, "marker response, port=%u, sys=%6D, id=%u\n", ntohs(mdu->mdu_info.mi_rq_port), mdu->mdu_info.mi_rq_system, ":", ntohl(mdu->mdu_info.mi_rq_xid))); /* Verify that it is the last marker we sent out */ if (memcmp(&mdu->mdu_info, &lp->lp_marker, sizeof(struct lacp_markerinfo))) goto bad; LACP_LOCK(lsc); lp->lp_flags &= ~LACP_PORT_MARK; if (lsc->lsc_suppress_distributing) { /* Check if any ports are waiting for a response */ LIST_FOREACH(lp2, &lsc->lsc_ports, lp_next) { if (lp2->lp_flags & LACP_PORT_MARK) { pending = 1; break; } } if (pending == 0) { /* All interface queues are clear */ LACP_DPRINTF((NULL, "queue flush complete\n")); lsc->lsc_suppress_distributing = FALSE; } } LACP_UNLOCK(lsc); m_freem(m); break; default: goto bad; } return (error); bad: LACP_DPRINTF((lp, "bad marker frame\n")); m_freem(m); return (EINVAL); } static int tlv_check(const void *p, size_t size, const struct tlvhdr *tlv, const struct tlv_template *tmpl, boolean_t check_type) { while (/* CONSTCOND */ 1) { if ((const char *)tlv - (const char *)p + sizeof(*tlv) > size) { return (EINVAL); } if ((check_type && tlv->tlv_type != tmpl->tmpl_type) || tlv->tlv_length != tmpl->tmpl_length) { return (EINVAL); } if (tmpl->tmpl_type == 0) { break; } tlv = (const struct tlvhdr *) ((const char *)tlv + tlv->tlv_length); tmpl++; } return (0); } /* Debugging */ const char * lacp_format_mac(const uint8_t *mac, char *buf, size_t buflen) { snprintf(buf, buflen, "%02X-%02X-%02X-%02X-%02X-%02X", (int)mac[0], (int)mac[1], (int)mac[2], (int)mac[3], (int)mac[4], (int)mac[5]); return (buf); } const char * lacp_format_systemid(const struct lacp_systemid *sysid, char *buf, size_t buflen) { char macbuf[LACP_MACSTR_MAX+1]; snprintf(buf, buflen, "%04X,%s", ntohs(sysid->lsi_prio), lacp_format_mac(sysid->lsi_mac, macbuf, sizeof(macbuf))); return (buf); } const char * lacp_format_portid(const struct lacp_portid *portid, char *buf, size_t buflen) { snprintf(buf, buflen, "%04X,%04X", ntohs(portid->lpi_prio), ntohs(portid->lpi_portno)); return (buf); } const char * lacp_format_partner(const struct lacp_peerinfo *peer, char *buf, size_t buflen) { char sysid[LACP_SYSTEMIDSTR_MAX+1]; char portid[LACP_PORTIDSTR_MAX+1]; snprintf(buf, buflen, "(%s,%04X,%s)", lacp_format_systemid(&peer->lip_systemid, sysid, sizeof(sysid)), ntohs(peer->lip_key), lacp_format_portid(&peer->lip_portid, portid, sizeof(portid))); return (buf); } const char * lacp_format_lagid(const struct lacp_peerinfo *a, const struct lacp_peerinfo *b, char *buf, size_t buflen) { char astr[LACP_PARTNERSTR_MAX+1]; char bstr[LACP_PARTNERSTR_MAX+1]; #if 0 /* * there's a convention to display small numbered peer * in the left. */ if (lacp_compare_peerinfo(a, b) > 0) { const struct lacp_peerinfo *t; t = a; a = b; b = t; } #endif snprintf(buf, buflen, "[%s,%s]", lacp_format_partner(a, astr, sizeof(astr)), lacp_format_partner(b, bstr, sizeof(bstr))); return (buf); } const char * lacp_format_lagid_aggregator(const struct lacp_aggregator *la, char *buf, size_t buflen) { if (la == NULL) { return ("(none)"); } return (lacp_format_lagid(&la->la_actor, &la->la_partner, buf, buflen)); } const char * lacp_format_state(uint8_t state, char *buf, size_t buflen) { snprintf(buf, buflen, "%b", state, LACP_STATE_BITS); return (buf); } static void lacp_dump_lacpdu(const struct lacpdu *du) { char buf[LACP_PARTNERSTR_MAX+1]; char buf2[LACP_STATESTR_MAX+1]; printf("actor=%s\n", lacp_format_partner(&du->ldu_actor, buf, sizeof(buf))); printf("actor.state=%s\n", lacp_format_state(du->ldu_actor.lip_state, buf2, sizeof(buf2))); printf("partner=%s\n", lacp_format_partner(&du->ldu_partner, buf, sizeof(buf))); printf("partner.state=%s\n", lacp_format_state(du->ldu_partner.lip_state, buf2, sizeof(buf2))); printf("maxdelay=%d\n", ntohs(du->ldu_collector.lci_maxdelay)); } static void lacp_dprintf(const struct lacp_port *lp, const char *fmt, ...) { va_list va; if (lp) { printf("%s: ", lp->lp_ifp->if_xname); } va_start(va, fmt); vprintf(fmt, va); va_end(va); } Index: head/sys/net/if.c =================================================================== --- head/sys/net/if.c (revision 336675) +++ head/sys/net/if.c (revision 336676) @@ -1,4579 +1,4579 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1980, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)if.c 8.5 (Berkeley) 1/9/95 * $FreeBSD$ */ #include "opt_inet6.h" #include "opt_inet.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #include #include #include #ifdef INET #include #include #endif /* INET */ #ifdef INET6 #include #include #endif /* INET6 */ #endif /* INET || INET6 */ #include /* * Consumers of struct ifreq such as tcpdump assume no pad between ifr_name * and ifr_ifru when it is used in SIOCGIFCONF. */ _Static_assert(sizeof(((struct ifreq *)0)->ifr_name) == offsetof(struct ifreq, ifr_ifru), "gap between ifr_name and ifr_ifru"); __read_mostly epoch_t net_epoch_preempt; __read_mostly epoch_t net_epoch; #ifdef COMPAT_FREEBSD32 #include #include struct ifreq_buffer32 { uint32_t length; /* (size_t) */ uint32_t buffer; /* (void *) */ }; /* * Interface request structure used for socket * ioctl's. All interface ioctl's must have parameter * definitions which begin with ifr_name. The * remainder may be interface specific. */ struct ifreq32 { char ifr_name[IFNAMSIZ]; /* if name, e.g. "en0" */ union { struct sockaddr ifru_addr; struct sockaddr ifru_dstaddr; struct sockaddr ifru_broadaddr; struct ifreq_buffer32 ifru_buffer; short ifru_flags[2]; short ifru_index; int ifru_jid; int ifru_metric; int ifru_mtu; int ifru_phys; int ifru_media; uint32_t ifru_data; int ifru_cap[2]; u_int ifru_fib; u_char ifru_vlan_pcp; } ifr_ifru; }; CTASSERT(sizeof(struct ifreq) == sizeof(struct ifreq32)); CTASSERT(__offsetof(struct ifreq, ifr_ifru) == __offsetof(struct ifreq32, ifr_ifru)); struct ifgroupreq32 { char ifgr_name[IFNAMSIZ]; u_int ifgr_len; union { char ifgru_group[IFNAMSIZ]; uint32_t ifgru_groups; } ifgr_ifgru; }; struct ifmediareq32 { char ifm_name[IFNAMSIZ]; int ifm_current; int ifm_mask; int ifm_status; int ifm_active; int ifm_count; uint32_t ifm_ulist; /* (int *) */ }; #define SIOCGIFMEDIA32 _IOC_NEWTYPE(SIOCGIFMEDIA, struct ifmediareq32) #define SIOCGIFXMEDIA32 _IOC_NEWTYPE(SIOCGIFXMEDIA, struct ifmediareq32) #define _CASE_IOC_IFGROUPREQ_32(cmd) \ case _IOC_NEWTYPE((cmd), struct ifgroupreq32): #else /* !COMPAT_FREEBSD32 */ #define _CASE_IOC_IFGROUPREQ_32(cmd) #endif /* !COMPAT_FREEBSD32 */ #define CASE_IOC_IFGROUPREQ(cmd) \ _CASE_IOC_IFGROUPREQ_32(cmd) \ case (cmd) union ifreq_union { struct ifreq ifr; #ifdef COMPAT_FREEBSD32 struct ifreq32 ifr32; #endif }; union ifgroupreq_union { struct ifgroupreq ifgr; #ifdef COMPAT_FREEBSD32 struct ifgroupreq32 ifgr32; #endif }; SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers"); SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management"); SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN, &ifqmaxlen, 0, "max send queue size"); /* Log link state change events */ static int log_link_state_change = 1; SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW, &log_link_state_change, 0, "log interface link state change events"); /* Log promiscuous mode change events */ static int log_promisc_mode_change = 1; SYSCTL_INT(_net_link, OID_AUTO, log_promisc_mode_change, CTLFLAG_RDTUN, &log_promisc_mode_change, 1, "log promiscuous mode change events"); /* Interface description */ static unsigned int ifdescr_maxlen = 1024; SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW, &ifdescr_maxlen, 0, "administrative maximum length for interface description"); static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions"); /* global sx for non-critical path ifdescr */ static struct sx ifdescr_sx; SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr"); void (*ng_ether_link_state_p)(struct ifnet *ifp, int state); void (*lagg_linkstate_p)(struct ifnet *ifp, int state); /* These are external hooks for CARP. */ void (*carp_linkstate_p)(struct ifnet *ifp); void (*carp_demote_adj_p)(int, char *); int (*carp_master_p)(struct ifaddr *); #if defined(INET) || defined(INET6) int (*carp_forus_p)(struct ifnet *ifp, u_char *dhost); int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa); int (*carp_ioctl_p)(struct ifreq *, u_long, struct thread *); int (*carp_attach_p)(struct ifaddr *, int); void (*carp_detach_p)(struct ifaddr *, bool); #endif #ifdef INET int (*carp_iamatch_p)(struct ifaddr *, uint8_t **); #endif #ifdef INET6 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6); caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr); #endif struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL; /* * XXX: Style; these should be sorted alphabetically, and unprototyped * static functions should be prototyped. Currently they are sorted by * declaration order. */ static void if_attachdomain(void *); static void if_attachdomain1(struct ifnet *); static int ifconf(u_long, caddr_t); static void *if_grow(void); static void if_input_default(struct ifnet *, struct mbuf *); static int if_requestencap_default(struct ifnet *, struct if_encap_req *); static void if_route(struct ifnet *, int flag, int fam); static int if_setflag(struct ifnet *, int, int, int *, int); static int if_transmit(struct ifnet *ifp, struct mbuf *m); static void if_unroute(struct ifnet *, int flag, int fam); static void link_rtrequest(int, struct rtentry *, struct rt_addrinfo *); static int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *); static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int); static void do_link_state_change(void *, int); static int if_getgroup(struct ifgroupreq *, struct ifnet *); static int if_getgroupmembers(struct ifgroupreq *); static void if_delgroups(struct ifnet *); static void if_attach_internal(struct ifnet *, int, struct if_clone *); static int if_detach_internal(struct ifnet *, int, struct if_clone **); #ifdef VIMAGE static void if_vmove(struct ifnet *, struct vnet *); #endif #ifdef INET6 /* * XXX: declare here to avoid to include many inet6 related files.. * should be more generalized? */ extern void nd6_setmtu(struct ifnet *); #endif /* ipsec helper hooks */ VNET_DEFINE(struct hhook_head *, ipsec_hhh_in[HHOOK_IPSEC_COUNT]); VNET_DEFINE(struct hhook_head *, ipsec_hhh_out[HHOOK_IPSEC_COUNT]); VNET_DEFINE(int, if_index); int ifqmaxlen = IFQ_MAXLEN; VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */ VNET_DEFINE(struct ifgrouphead, ifg_head); -static VNET_DEFINE(int, if_indexlim) = 8; +VNET_DEFINE_STATIC(int, if_indexlim) = 8; /* Table of ifnet by index. */ VNET_DEFINE(struct ifnet **, ifindex_table); #define V_if_indexlim VNET(if_indexlim) #define V_ifindex_table VNET(ifindex_table) /* * The global network interface list (V_ifnet) and related state (such as * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and * an rwlock. Either may be acquired shared to stablize the list, but both * must be acquired writable to modify the list. This model allows us to * both stablize the interface list during interrupt thread processing, but * also to stablize it over long-running ioctls, without introducing priority * inversions and deadlocks. */ struct rwlock ifnet_rwlock; RW_SYSINIT_FLAGS(ifnet_rw, &ifnet_rwlock, "ifnet_rw", RW_RECURSE); struct sx ifnet_sxlock; SX_SYSINIT_FLAGS(ifnet_sx, &ifnet_sxlock, "ifnet_sx", SX_RECURSE); /* * The allocation of network interfaces is a rather non-atomic affair; we * need to select an index before we are ready to expose the interface for * use, so will use this pointer value to indicate reservation. */ #define IFNET_HOLD (void *)(uintptr_t)(-1) static if_com_alloc_t *if_com_alloc[256]; static if_com_free_t *if_com_free[256]; static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals"); MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address"); MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address"); struct ifnet * ifnet_byindex_locked(u_short idx) { if (idx > V_if_index) return (NULL); if (V_ifindex_table[idx] == IFNET_HOLD) return (NULL); return (V_ifindex_table[idx]); } struct ifnet * ifnet_byindex(u_short idx) { struct ifnet *ifp; ifp = ifnet_byindex_locked(idx); return (ifp); } struct ifnet * ifnet_byindex_ref(u_short idx) { struct ifnet *ifp; IFNET_RLOCK_NOSLEEP(); ifp = ifnet_byindex_locked(idx); if (ifp == NULL || (ifp->if_flags & IFF_DYING)) { IFNET_RUNLOCK_NOSLEEP(); return (NULL); } if_ref(ifp); IFNET_RUNLOCK_NOSLEEP(); return (ifp); } /* * Allocate an ifindex array entry; return 0 on success or an error on * failure. */ static u_short ifindex_alloc(void **old) { u_short idx; IFNET_WLOCK_ASSERT(); /* * Try to find an empty slot below V_if_index. If we fail, take the * next slot. */ for (idx = 1; idx <= V_if_index; idx++) { if (V_ifindex_table[idx] == NULL) break; } /* Catch if_index overflow. */ if (idx >= V_if_indexlim) { *old = if_grow(); return (USHRT_MAX); } if (idx > V_if_index) V_if_index = idx; return (idx); } static void ifindex_free_locked(u_short idx) { IFNET_WLOCK_ASSERT(); V_ifindex_table[idx] = NULL; while (V_if_index > 0 && V_ifindex_table[V_if_index] == NULL) V_if_index--; } static void ifindex_free(u_short idx) { IFNET_WLOCK(); ifindex_free_locked(idx); IFNET_WUNLOCK(); } static void ifnet_setbyindex(u_short idx, struct ifnet *ifp) { V_ifindex_table[idx] = ifp; } struct ifaddr * ifaddr_byindex(u_short idx) { struct ifnet *ifp; struct ifaddr *ifa = NULL; IFNET_RLOCK_NOSLEEP(); ifp = ifnet_byindex_locked(idx); if (ifp != NULL && (ifa = ifp->if_addr) != NULL) ifa_ref(ifa); IFNET_RUNLOCK_NOSLEEP(); return (ifa); } /* * Network interface utility routines. * * Routines with ifa_ifwith* names take sockaddr *'s as * parameters. */ static void vnet_if_init(const void *unused __unused) { void *old; CK_STAILQ_INIT(&V_ifnet); CK_STAILQ_INIT(&V_ifg_head); IFNET_WLOCK(); old = if_grow(); /* create initial table */ IFNET_WUNLOCK(); epoch_wait_preempt(net_epoch_preempt); free(old, M_IFNET); vnet_if_clone_init(); } VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init, NULL); #ifdef VIMAGE static void vnet_if_uninit(const void *unused __unused) { VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p " "not empty", __func__, __LINE__, &V_ifnet)); VNET_ASSERT(CK_STAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p " "not empty", __func__, __LINE__, &V_ifg_head)); free((caddr_t)V_ifindex_table, M_IFNET); } VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST, vnet_if_uninit, NULL); static void vnet_if_return(const void *unused __unused) { struct ifnet *ifp, *nifp; /* Return all inherited interfaces to their parent vnets. */ CK_STAILQ_FOREACH_SAFE(ifp, &V_ifnet, if_link, nifp) { if (ifp->if_home_vnet != ifp->if_vnet) if_vmove(ifp, ifp->if_home_vnet); } } VNET_SYSUNINIT(vnet_if_return, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_if_return, NULL); #endif static void * if_grow(void) { int oldlim; u_int n; struct ifnet **e; void *old; old = NULL; IFNET_WLOCK_ASSERT(); oldlim = V_if_indexlim; IFNET_WUNLOCK(); n = (oldlim << 1) * sizeof(*e); e = malloc(n, M_IFNET, M_WAITOK | M_ZERO); IFNET_WLOCK(); if (V_if_indexlim != oldlim) { free(e, M_IFNET); return (NULL); } if (V_ifindex_table != NULL) { memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2); old = V_ifindex_table; } V_if_indexlim <<= 1; V_ifindex_table = e; return (old); } /* * Allocate a struct ifnet and an index for an interface. A layer 2 * common structure will also be allocated if an allocation routine is * registered for the passed type. */ struct ifnet * if_alloc(u_char type) { struct ifnet *ifp; u_short idx; void *old; ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO); restart: IFNET_WLOCK(); idx = ifindex_alloc(&old); if (__predict_false(idx == USHRT_MAX)) { IFNET_WUNLOCK(); epoch_wait_preempt(net_epoch_preempt); free(old, M_IFNET); goto restart; } ifnet_setbyindex(idx, IFNET_HOLD); IFNET_WUNLOCK(); ifp->if_index = idx; ifp->if_type = type; ifp->if_alloctype = type; #ifdef VIMAGE ifp->if_vnet = curvnet; #endif if (if_com_alloc[type] != NULL) { ifp->if_l2com = if_com_alloc[type](type, ifp); if (ifp->if_l2com == NULL) { free(ifp, M_IFNET); ifindex_free(idx); return (NULL); } } IF_ADDR_LOCK_INIT(ifp); TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp); ifp->if_afdata_initialized = 0; IF_AFDATA_LOCK_INIT(ifp); CK_STAILQ_INIT(&ifp->if_addrhead); CK_STAILQ_INIT(&ifp->if_multiaddrs); CK_STAILQ_INIT(&ifp->if_groups); #ifdef MAC mac_ifnet_init(ifp); #endif ifq_init(&ifp->if_snd, ifp); refcount_init(&ifp->if_refcount, 1); /* Index reference. */ for (int i = 0; i < IFCOUNTERS; i++) ifp->if_counters[i] = counter_u64_alloc(M_WAITOK); ifp->if_get_counter = if_get_counter_default; ifp->if_pcp = IFNET_PCP_NONE; ifnet_setbyindex(ifp->if_index, ifp); return (ifp); } /* * Do the actual work of freeing a struct ifnet, and layer 2 common * structure. This call is made when the last reference to an * interface is released. */ static void if_free_internal(struct ifnet *ifp) { KASSERT((ifp->if_flags & IFF_DYING), ("if_free_internal: interface not dying")); if (if_com_free[ifp->if_alloctype] != NULL) if_com_free[ifp->if_alloctype](ifp->if_l2com, ifp->if_alloctype); #ifdef MAC mac_ifnet_destroy(ifp); #endif /* MAC */ if (ifp->if_description != NULL) free(ifp->if_description, M_IFDESCR); IF_AFDATA_DESTROY(ifp); IF_ADDR_LOCK_DESTROY(ifp); ifq_delete(&ifp->if_snd); for (int i = 0; i < IFCOUNTERS; i++) counter_u64_free(ifp->if_counters[i]); free(ifp, M_IFNET); } static void if_destroy(epoch_context_t ctx) { struct ifnet *ifp; ifp = __containerof(ctx, struct ifnet, if_epoch_ctx); if_free_internal(ifp); } /* * Deregister an interface and free the associated storage. */ void if_free(struct ifnet *ifp) { ifp->if_flags |= IFF_DYING; /* XXX: Locking */ CURVNET_SET_QUIET(ifp->if_vnet); IFNET_WLOCK(); KASSERT(ifp == ifnet_byindex_locked(ifp->if_index), ("%s: freeing unallocated ifnet", ifp->if_xname)); ifindex_free_locked(ifp->if_index); IFNET_WUNLOCK(); if (refcount_release(&ifp->if_refcount)) epoch_call(net_epoch_preempt, &ifp->if_epoch_ctx, if_destroy); CURVNET_RESTORE(); } /* * Interfaces to keep an ifnet type-stable despite the possibility of the * driver calling if_free(). If there are additional references, we defer * freeing the underlying data structure. */ void if_ref(struct ifnet *ifp) { /* We don't assert the ifnet list lock here, but arguably should. */ refcount_acquire(&ifp->if_refcount); } void if_rele(struct ifnet *ifp) { if (!refcount_release(&ifp->if_refcount)) return; epoch_call(net_epoch_preempt, &ifp->if_epoch_ctx, if_destroy); } void ifq_init(struct ifaltq *ifq, struct ifnet *ifp) { mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF); if (ifq->ifq_maxlen == 0) ifq->ifq_maxlen = ifqmaxlen; ifq->altq_type = 0; ifq->altq_disc = NULL; ifq->altq_flags &= ALTQF_CANTCHANGE; ifq->altq_tbr = NULL; ifq->altq_ifp = ifp; } void ifq_delete(struct ifaltq *ifq) { mtx_destroy(&ifq->ifq_mtx); } /* * Perform generic interface initialization tasks and attach the interface * to the list of "active" interfaces. If vmove flag is set on entry * to if_attach_internal(), perform only a limited subset of initialization * tasks, given that we are moving from one vnet to another an ifnet which * has already been fully initialized. * * Note that if_detach_internal() removes group membership unconditionally * even when vmove flag is set, and if_attach_internal() adds only IFG_ALL. * Thus, when if_vmove() is applied to a cloned interface, group membership * is lost while a cloned one always joins a group whose name is * ifc->ifc_name. To recover this after if_detach_internal() and * if_attach_internal(), the cloner should be specified to * if_attach_internal() via ifc. If it is non-NULL, if_attach_internal() * attempts to join a group whose name is ifc->ifc_name. * * XXX: * - The decision to return void and thus require this function to * succeed is questionable. * - We should probably do more sanity checking. For instance we don't * do anything to insure if_xname is unique or non-empty. */ void if_attach(struct ifnet *ifp) { if_attach_internal(ifp, 0, NULL); } /* * Compute the least common TSO limit. */ void if_hw_tsomax_common(if_t ifp, struct ifnet_hw_tsomax *pmax) { /* * 1) If there is no limit currently, take the limit from * the network adapter. * * 2) If the network adapter has a limit below the current * limit, apply it. */ if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 && ifp->if_hw_tsomax < pmax->tsomaxbytes)) { pmax->tsomaxbytes = ifp->if_hw_tsomax; } if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 && ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) { pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; } if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 && ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) { pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } /* * Update TSO limit of a network adapter. * * Returns zero if no change. Else non-zero. */ int if_hw_tsomax_update(if_t ifp, struct ifnet_hw_tsomax *pmax) { int retval = 0; if (ifp->if_hw_tsomax != pmax->tsomaxbytes) { ifp->if_hw_tsomax = pmax->tsomaxbytes; retval++; } if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) { ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize; retval++; } if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) { ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount; retval++; } return (retval); } static void if_attach_internal(struct ifnet *ifp, int vmove, struct if_clone *ifc) { unsigned socksize, ifasize; int namelen, masklen; struct sockaddr_dl *sdl; struct ifaddr *ifa; if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index)) panic ("%s: BUG: if_attach called without if_alloc'd input()\n", ifp->if_xname); #ifdef VIMAGE ifp->if_vnet = curvnet; if (ifp->if_home_vnet == NULL) ifp->if_home_vnet = curvnet; #endif if_addgroup(ifp, IFG_ALL); /* Restore group membership for cloned interfaces. */ if (vmove && ifc != NULL) if_clone_addgroup(ifp, ifc); getmicrotime(&ifp->if_lastchange); ifp->if_epoch = time_uptime; KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) || (ifp->if_transmit != NULL && ifp->if_qflush != NULL), ("transmit and qflush must both either be set or both be NULL")); if (ifp->if_transmit == NULL) { ifp->if_transmit = if_transmit; ifp->if_qflush = if_qflush; } if (ifp->if_input == NULL) ifp->if_input = if_input_default; if (ifp->if_requestencap == NULL) ifp->if_requestencap = if_requestencap_default; if (!vmove) { #ifdef MAC mac_ifnet_create(ifp); #endif /* * Create a Link Level name for this device. */ namelen = strlen(ifp->if_xname); /* * Always save enough space for any possiable name so we * can do a rename in place later. */ masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ; socksize = masklen + ifp->if_addrlen; if (socksize < sizeof(*sdl)) socksize = sizeof(*sdl); socksize = roundup2(socksize, sizeof(long)); ifasize = sizeof(*ifa) + 2 * socksize; ifa = ifa_alloc(ifasize, M_WAITOK); sdl = (struct sockaddr_dl *)(ifa + 1); sdl->sdl_len = socksize; sdl->sdl_family = AF_LINK; bcopy(ifp->if_xname, sdl->sdl_data, namelen); sdl->sdl_nlen = namelen; sdl->sdl_index = ifp->if_index; sdl->sdl_type = ifp->if_type; ifp->if_addr = ifa; ifa->ifa_ifp = ifp; ifa->ifa_rtrequest = link_rtrequest; ifa->ifa_addr = (struct sockaddr *)sdl; sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl); ifa->ifa_netmask = (struct sockaddr *)sdl; sdl->sdl_len = masklen; while (namelen != 0) sdl->sdl_data[--namelen] = 0xff; CK_STAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link); /* Reliably crash if used uninitialized. */ ifp->if_broadcastaddr = NULL; if (ifp->if_type == IFT_ETHER) { ifp->if_hw_addr = malloc(ifp->if_addrlen, M_IFADDR, M_WAITOK | M_ZERO); } #if defined(INET) || defined(INET6) /* Use defaults for TSO, if nothing is set */ if (ifp->if_hw_tsomax == 0 && ifp->if_hw_tsomaxsegcount == 0 && ifp->if_hw_tsomaxsegsize == 0) { /* * The TSO defaults needs to be such that an * NFS mbuf list of 35 mbufs totalling just * below 64K works and that a chain of mbufs * can be defragged into at most 32 segments: */ ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN)); ifp->if_hw_tsomaxsegcount = 35; ifp->if_hw_tsomaxsegsize = 2048; /* 2K */ /* XXX some drivers set IFCAP_TSO after ethernet attach */ if (ifp->if_capabilities & IFCAP_TSO) { if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n", ifp->if_hw_tsomax, ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize); } } #endif } #ifdef VIMAGE else { /* * Update the interface index in the link layer address * of the interface. */ for (ifa = ifp->if_addr; ifa != NULL; ifa = CK_STAILQ_NEXT(ifa, ifa_link)) { if (ifa->ifa_addr->sa_family == AF_LINK) { sdl = (struct sockaddr_dl *)ifa->ifa_addr; sdl->sdl_index = ifp->if_index; } } } #endif IFNET_WLOCK(); CK_STAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link); #ifdef VIMAGE curvnet->vnet_ifcnt++; #endif IFNET_WUNLOCK(); if (domain_init_status >= 2) if_attachdomain1(ifp); EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp); if (IS_DEFAULT_VNET(curvnet)) devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL); /* Announce the interface. */ rt_ifannouncemsg(ifp, IFAN_ARRIVAL); } static void if_epochalloc(void *dummy __unused) { net_epoch_preempt = epoch_alloc(EPOCH_PREEMPT); net_epoch = epoch_alloc(0); } SYSINIT(ifepochalloc, SI_SUB_TASKQ + 1, SI_ORDER_ANY, if_epochalloc, NULL); static void if_attachdomain(void *dummy) { struct ifnet *ifp; CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) if_attachdomain1(ifp); } SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND, if_attachdomain, NULL); static void if_attachdomain1(struct ifnet *ifp) { struct domain *dp; /* * Since dp->dom_ifattach calls malloc() with M_WAITOK, we * cannot lock ifp->if_afdata initialization, entirely. */ IF_AFDATA_LOCK(ifp); if (ifp->if_afdata_initialized >= domain_init_status) { IF_AFDATA_UNLOCK(ifp); log(LOG_WARNING, "%s called more than once on %s\n", __func__, ifp->if_xname); return; } ifp->if_afdata_initialized = domain_init_status; IF_AFDATA_UNLOCK(ifp); /* address family dependent data region */ bzero(ifp->if_afdata, sizeof(ifp->if_afdata)); for (dp = domains; dp; dp = dp->dom_next) { if (dp->dom_ifattach) ifp->if_afdata[dp->dom_family] = (*dp->dom_ifattach)(ifp); } } /* * Remove any unicast or broadcast network addresses from an interface. */ void if_purgeaddrs(struct ifnet *ifp) { struct ifaddr *ifa, *next; NET_EPOCH_ENTER(); CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) { if (ifa->ifa_addr->sa_family == AF_LINK) continue; #ifdef INET /* XXX: Ugly!! ad hoc just for INET */ if (ifa->ifa_addr->sa_family == AF_INET) { struct ifaliasreq ifr; bzero(&ifr, sizeof(ifr)); ifr.ifra_addr = *ifa->ifa_addr; if (ifa->ifa_dstaddr) ifr.ifra_broadaddr = *ifa->ifa_dstaddr; if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp, NULL) == 0) continue; } #endif /* INET */ #ifdef INET6 if (ifa->ifa_addr->sa_family == AF_INET6) { in6_purgeaddr(ifa); /* ifp_addrhead is already updated */ continue; } #endif /* INET6 */ IF_ADDR_WLOCK(ifp); CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link); IF_ADDR_WUNLOCK(ifp); ifa_free(ifa); } NET_EPOCH_EXIT(); } /* * Remove any multicast network addresses from an interface when an ifnet * is going away. */ static void if_purgemaddrs(struct ifnet *ifp) { struct ifmultiaddr *ifma; IF_ADDR_WLOCK(ifp); while (!CK_STAILQ_EMPTY(&ifp->if_multiaddrs)) { ifma = CK_STAILQ_FIRST(&ifp->if_multiaddrs); CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); if_delmulti_locked(ifp, ifma, 1); } IF_ADDR_WUNLOCK(ifp); } /* * Detach an interface, removing it from the list of "active" interfaces. * If vmove flag is set on entry to if_detach_internal(), perform only a * limited subset of cleanup tasks, given that we are moving an ifnet from * one vnet to another, where it must be fully operational. * * XXXRW: There are some significant questions about event ordering, and * how to prevent things from starting to use the interface during detach. */ void if_detach(struct ifnet *ifp) { CURVNET_SET_QUIET(ifp->if_vnet); if_detach_internal(ifp, 0, NULL); CURVNET_RESTORE(); } /* * The vmove flag, if set, indicates that we are called from a callpath * that is moving an interface to a different vnet instance. * * The shutdown flag, if set, indicates that we are called in the * process of shutting down a vnet instance. Currently only the * vnet_if_return SYSUNINIT function sets it. Note: we can be called * on a vnet instance shutdown without this flag being set, e.g., when * the cloned interfaces are destoyed as first thing of teardown. */ static int if_detach_internal(struct ifnet *ifp, int vmove, struct if_clone **ifcp) { struct ifaddr *ifa; int i; struct domain *dp; struct ifnet *iter; int found = 0; #ifdef VIMAGE int shutdown; shutdown = (ifp->if_vnet->vnet_state > SI_SUB_VNET && ifp->if_vnet->vnet_state < SI_SUB_VNET_DONE) ? 1 : 0; #endif IFNET_WLOCK(); CK_STAILQ_FOREACH(iter, &V_ifnet, if_link) if (iter == ifp) { CK_STAILQ_REMOVE(&V_ifnet, ifp, ifnet, if_link); found = 1; break; } IFNET_WUNLOCK(); if (!found) { /* * While we would want to panic here, we cannot * guarantee that the interface is indeed still on * the list given we don't hold locks all the way. */ return (ENOENT); #if 0 if (vmove) panic("%s: ifp=%p not on the ifnet tailq %p", __func__, ifp, &V_ifnet); else return; /* XXX this should panic as well? */ #endif } /* * At this point we know the interface still was on the ifnet list * and we removed it so we are in a stable state. */ #ifdef VIMAGE curvnet->vnet_ifcnt--; #endif epoch_wait_preempt(net_epoch_preempt); /* * In any case (destroy or vmove) detach us from the groups * and remove/wait for pending events on the taskq. * XXX-BZ in theory an interface could still enqueue a taskq change? */ if_delgroups(ifp); taskqueue_drain(taskqueue_swi, &ifp->if_linktask); /* * Check if this is a cloned interface or not. Must do even if * shutting down as a if_vmove_reclaim() would move the ifp and * the if_clone_addgroup() will have a corrupted string overwise * from a gibberish pointer. */ if (vmove && ifcp != NULL) *ifcp = if_clone_findifc(ifp); if_down(ifp); #ifdef VIMAGE /* * On VNET shutdown abort here as the stack teardown will do all * the work top-down for us. */ if (shutdown) { /* * In case of a vmove we are done here without error. * If we would signal an error it would lead to the same * abort as if we did not find the ifnet anymore. * if_detach() calls us in void context and does not care * about an early abort notification, so life is splendid :) */ goto finish_vnet_shutdown; } #endif /* * At this point we are not tearing down a VNET and are either * going to destroy or vmove the interface and have to cleanup * accordingly. */ /* * Remove routes and flush queues. */ #ifdef ALTQ if (ALTQ_IS_ENABLED(&ifp->if_snd)) altq_disable(&ifp->if_snd); if (ALTQ_IS_ATTACHED(&ifp->if_snd)) altq_detach(&ifp->if_snd); #endif if_purgeaddrs(ifp); #ifdef INET in_ifdetach(ifp); #endif #ifdef INET6 /* * Remove all IPv6 kernel structs related to ifp. This should be done * before removing routing entries below, since IPv6 interface direct * routes are expected to be removed by the IPv6-specific kernel API. * Otherwise, the kernel will detect some inconsistency and bark it. */ in6_ifdetach(ifp); #endif if_purgemaddrs(ifp); /* Announce that the interface is gone. */ rt_ifannouncemsg(ifp, IFAN_DEPARTURE); EVENTHANDLER_INVOKE(ifnet_departure_event, ifp); if (IS_DEFAULT_VNET(curvnet)) devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL); if (!vmove) { /* * Prevent further calls into the device driver via ifnet. */ if_dead(ifp); /* * Remove link ifaddr pointer and maybe decrement if_index. * Clean up all addresses. */ free(ifp->if_hw_addr, M_IFADDR); ifp->if_hw_addr = NULL; ifp->if_addr = NULL; /* We can now free link ifaddr. */ IF_ADDR_WLOCK(ifp); if (!CK_STAILQ_EMPTY(&ifp->if_addrhead)) { ifa = CK_STAILQ_FIRST(&ifp->if_addrhead); CK_STAILQ_REMOVE(&ifp->if_addrhead, ifa, ifaddr, ifa_link); IF_ADDR_WUNLOCK(ifp); ifa_free(ifa); } else IF_ADDR_WUNLOCK(ifp); } rt_flushifroutes(ifp); #ifdef VIMAGE finish_vnet_shutdown: #endif /* * We cannot hold the lock over dom_ifdetach calls as they might * sleep, for example trying to drain a callout, thus open up the * theoretical race with re-attaching. */ IF_AFDATA_LOCK(ifp); i = ifp->if_afdata_initialized; ifp->if_afdata_initialized = 0; IF_AFDATA_UNLOCK(ifp); for (dp = domains; i > 0 && dp; dp = dp->dom_next) { if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family]) { (*dp->dom_ifdetach)(ifp, ifp->if_afdata[dp->dom_family]); ifp->if_afdata[dp->dom_family] = NULL; } } return (0); } #ifdef VIMAGE /* * if_vmove() performs a limited version of if_detach() in current * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg. * An attempt is made to shrink if_index in current vnet, find an * unused if_index in target vnet and calls if_grow() if necessary, * and finally find an unused if_xname for the target vnet. */ static void if_vmove(struct ifnet *ifp, struct vnet *new_vnet) { struct if_clone *ifc; u_int bif_dlt, bif_hdrlen; void *old; int rc; /* * if_detach_internal() will call the eventhandler to notify * interface departure. That will detach if_bpf. We need to * safe the dlt and hdrlen so we can re-attach it later. */ bpf_get_bp_params(ifp->if_bpf, &bif_dlt, &bif_hdrlen); /* * Detach from current vnet, but preserve LLADDR info, do not * mark as dead etc. so that the ifnet can be reattached later. * If we cannot find it, we lost the race to someone else. */ rc = if_detach_internal(ifp, 1, &ifc); if (rc != 0) return; /* * Unlink the ifnet from ifindex_table[] in current vnet, and shrink * the if_index for that vnet if possible. * * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized, * or we'd lock on one vnet and unlock on another. */ IFNET_WLOCK(); ifindex_free_locked(ifp->if_index); IFNET_WUNLOCK(); /* * Perform interface-specific reassignment tasks, if provided by * the driver. */ if (ifp->if_reassign != NULL) ifp->if_reassign(ifp, new_vnet, NULL); /* * Switch to the context of the target vnet. */ CURVNET_SET_QUIET(new_vnet); restart: IFNET_WLOCK(); ifp->if_index = ifindex_alloc(&old); if (__predict_false(ifp->if_index == USHRT_MAX)) { IFNET_WUNLOCK(); epoch_wait_preempt(net_epoch_preempt); free(old, M_IFNET); goto restart; } ifnet_setbyindex(ifp->if_index, ifp); IFNET_WUNLOCK(); if_attach_internal(ifp, 1, ifc); if (ifp->if_bpf == NULL) bpfattach(ifp, bif_dlt, bif_hdrlen); CURVNET_RESTORE(); } /* * Move an ifnet to or from another child prison/vnet, specified by the jail id. */ static int if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid) { struct prison *pr; struct ifnet *difp; int shutdown; /* Try to find the prison within our visibility. */ sx_slock(&allprison_lock); pr = prison_find_child(td->td_ucred->cr_prison, jid); sx_sunlock(&allprison_lock); if (pr == NULL) return (ENXIO); prison_hold_locked(pr); mtx_unlock(&pr->pr_mtx); /* Do not try to move the iface from and to the same prison. */ if (pr->pr_vnet == ifp->if_vnet) { prison_free(pr); return (EEXIST); } /* Make sure the named iface does not exists in the dst. prison/vnet. */ /* XXX Lock interfaces to avoid races. */ CURVNET_SET_QUIET(pr->pr_vnet); difp = ifunit(ifname); if (difp != NULL) { CURVNET_RESTORE(); prison_free(pr); return (EEXIST); } /* Make sure the VNET is stable. */ shutdown = (ifp->if_vnet->vnet_state > SI_SUB_VNET && ifp->if_vnet->vnet_state < SI_SUB_VNET_DONE) ? 1 : 0; if (shutdown) { CURVNET_RESTORE(); prison_free(pr); return (EBUSY); } CURVNET_RESTORE(); /* Move the interface into the child jail/vnet. */ if_vmove(ifp, pr->pr_vnet); /* Report the new if_xname back to the userland. */ sprintf(ifname, "%s", ifp->if_xname); prison_free(pr); return (0); } static int if_vmove_reclaim(struct thread *td, char *ifname, int jid) { struct prison *pr; struct vnet *vnet_dst; struct ifnet *ifp; int shutdown; /* Try to find the prison within our visibility. */ sx_slock(&allprison_lock); pr = prison_find_child(td->td_ucred->cr_prison, jid); sx_sunlock(&allprison_lock); if (pr == NULL) return (ENXIO); prison_hold_locked(pr); mtx_unlock(&pr->pr_mtx); /* Make sure the named iface exists in the source prison/vnet. */ CURVNET_SET(pr->pr_vnet); ifp = ifunit(ifname); /* XXX Lock to avoid races. */ if (ifp == NULL) { CURVNET_RESTORE(); prison_free(pr); return (ENXIO); } /* Do not try to move the iface from and to the same prison. */ vnet_dst = TD_TO_VNET(td); if (vnet_dst == ifp->if_vnet) { CURVNET_RESTORE(); prison_free(pr); return (EEXIST); } /* Make sure the VNET is stable. */ shutdown = (ifp->if_vnet->vnet_state > SI_SUB_VNET && ifp->if_vnet->vnet_state < SI_SUB_VNET_DONE) ? 1 : 0; if (shutdown) { CURVNET_RESTORE(); prison_free(pr); return (EBUSY); } /* Get interface back from child jail/vnet. */ if_vmove(ifp, vnet_dst); CURVNET_RESTORE(); /* Report the new if_xname back to the userland. */ sprintf(ifname, "%s", ifp->if_xname); prison_free(pr); return (0); } #endif /* VIMAGE */ /* * Add a group to an interface */ int if_addgroup(struct ifnet *ifp, const char *groupname) { struct ifg_list *ifgl; struct ifg_group *ifg = NULL; struct ifg_member *ifgm; int new = 0; if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' && groupname[strlen(groupname) - 1] <= '9') return (EINVAL); IFNET_WLOCK(); CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) { IFNET_WUNLOCK(); return (EEXIST); } if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP, M_NOWAIT)) == NULL) { IFNET_WUNLOCK(); return (ENOMEM); } if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member), M_TEMP, M_NOWAIT)) == NULL) { free(ifgl, M_TEMP); IFNET_WUNLOCK(); return (ENOMEM); } CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) if (!strcmp(ifg->ifg_group, groupname)) break; if (ifg == NULL) { if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group), M_TEMP, M_NOWAIT)) == NULL) { free(ifgl, M_TEMP); free(ifgm, M_TEMP); IFNET_WUNLOCK(); return (ENOMEM); } strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group)); ifg->ifg_refcnt = 0; CK_STAILQ_INIT(&ifg->ifg_members); CK_STAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next); new = 1; } ifg->ifg_refcnt++; ifgl->ifgl_group = ifg; ifgm->ifgm_ifp = ifp; IF_ADDR_WLOCK(ifp); CK_STAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next); CK_STAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next); IF_ADDR_WUNLOCK(ifp); IFNET_WUNLOCK(); if (new) EVENTHANDLER_INVOKE(group_attach_event, ifg); EVENTHANDLER_INVOKE(group_change_event, groupname); return (0); } /* * Remove a group from an interface */ int if_delgroup(struct ifnet *ifp, const char *groupname) { struct ifg_list *ifgl; struct ifg_member *ifgm; int freeifgl; IFNET_WLOCK(); CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) break; if (ifgl == NULL) { IFNET_WUNLOCK(); return (ENOENT); } freeifgl = 0; IF_ADDR_WLOCK(ifp); CK_STAILQ_REMOVE(&ifp->if_groups, ifgl, ifg_list, ifgl_next); IF_ADDR_WUNLOCK(ifp); CK_STAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) if (ifgm->ifgm_ifp == ifp) break; if (ifgm != NULL) CK_STAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifg_member, ifgm_next); if (--ifgl->ifgl_group->ifg_refcnt == 0) { CK_STAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_group, ifg_next); freeifgl = 1; } IFNET_WUNLOCK(); epoch_wait_preempt(net_epoch_preempt); if (freeifgl) { EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group); free(ifgl->ifgl_group, M_TEMP); } free(ifgm, M_TEMP); free(ifgl, M_TEMP); EVENTHANDLER_INVOKE(group_change_event, groupname); return (0); } /* * Remove an interface from all groups */ static void if_delgroups(struct ifnet *ifp) { struct ifg_list *ifgl; struct ifg_member *ifgm; char groupname[IFNAMSIZ]; int ifglfree; IFNET_WLOCK(); while (!CK_STAILQ_EMPTY(&ifp->if_groups)) { ifgl = CK_STAILQ_FIRST(&ifp->if_groups); strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ); IF_ADDR_WLOCK(ifp); CK_STAILQ_REMOVE(&ifp->if_groups, ifgl, ifg_list, ifgl_next); IF_ADDR_WUNLOCK(ifp); CK_STAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) if (ifgm->ifgm_ifp == ifp) break; if (ifgm != NULL) CK_STAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifg_member, ifgm_next); ifglfree = 0; if (--ifgl->ifgl_group->ifg_refcnt == 0) { CK_STAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_group, ifg_next); ifglfree = 1; } IFNET_WUNLOCK(); epoch_wait_preempt(net_epoch_preempt); free(ifgm, M_TEMP); if (ifglfree) { EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group); free(ifgl->ifgl_group, M_TEMP); } EVENTHANDLER_INVOKE(group_change_event, groupname); IFNET_WLOCK(); } IFNET_WUNLOCK(); } static char * ifgr_group_get(void *ifgrp) { union ifgroupreq_union *ifgrup; ifgrup = ifgrp; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) return (&ifgrup->ifgr32.ifgr_ifgru.ifgru_group[0]); #endif return (&ifgrup->ifgr.ifgr_ifgru.ifgru_group[0]); } static struct ifg_req * ifgr_groups_get(void *ifgrp) { union ifgroupreq_union *ifgrup; ifgrup = ifgrp; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) return ((struct ifg_req *)(uintptr_t) ifgrup->ifgr32.ifgr_ifgru.ifgru_groups); #endif return (ifgrup->ifgr.ifgr_ifgru.ifgru_groups); } /* * Stores all groups from an interface in memory pointed to by ifgr. */ static int if_getgroup(struct ifgroupreq *ifgr, struct ifnet *ifp) { int len, error; struct ifg_list *ifgl; struct ifg_req ifgrq, *ifgp; if (ifgr->ifgr_len == 0) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) ifgr->ifgr_len += sizeof(struct ifg_req); IF_ADDR_RUNLOCK(ifp); return (0); } len = ifgr->ifgr_len; ifgp = ifgr_groups_get(ifgr); /* XXX: wire */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) { if (len < sizeof(ifgrq)) { IF_ADDR_RUNLOCK(ifp); return (EINVAL); } bzero(&ifgrq, sizeof ifgrq); strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group, sizeof(ifgrq.ifgrq_group)); if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) { IF_ADDR_RUNLOCK(ifp); return (error); } len -= sizeof(ifgrq); ifgp++; } IF_ADDR_RUNLOCK(ifp); return (0); } /* * Stores all members of a group in memory pointed to by igfr */ static int if_getgroupmembers(struct ifgroupreq *ifgr) { struct ifg_group *ifg; struct ifg_member *ifgm; struct ifg_req ifgrq, *ifgp; int len, error; IFNET_RLOCK(); CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) if (!strcmp(ifg->ifg_group, ifgr->ifgr_name)) break; if (ifg == NULL) { IFNET_RUNLOCK(); return (ENOENT); } if (ifgr->ifgr_len == 0) { CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) ifgr->ifgr_len += sizeof(ifgrq); IFNET_RUNLOCK(); return (0); } len = ifgr->ifgr_len; ifgp = ifgr_groups_get(ifgr); CK_STAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) { if (len < sizeof(ifgrq)) { IFNET_RUNLOCK(); return (EINVAL); } bzero(&ifgrq, sizeof ifgrq); strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname, sizeof(ifgrq.ifgrq_member)); if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) { IFNET_RUNLOCK(); return (error); } len -= sizeof(ifgrq); ifgp++; } IFNET_RUNLOCK(); return (0); } /* * Return counter values from counter(9)s stored in ifnet. */ uint64_t if_get_counter_default(struct ifnet *ifp, ift_counter cnt) { KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); return (counter_u64_fetch(ifp->if_counters[cnt])); } /* * Increase an ifnet counter. Usually used for counters shared * between the stack and a driver, but function supports them all. */ void if_inc_counter(struct ifnet *ifp, ift_counter cnt, int64_t inc) { KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); counter_u64_add(ifp->if_counters[cnt], inc); } /* * Copy data from ifnet to userland API structure if_data. */ void if_data_copy(struct ifnet *ifp, struct if_data *ifd) { ifd->ifi_type = ifp->if_type; ifd->ifi_physical = 0; ifd->ifi_addrlen = ifp->if_addrlen; ifd->ifi_hdrlen = ifp->if_hdrlen; ifd->ifi_link_state = ifp->if_link_state; ifd->ifi_vhid = 0; ifd->ifi_datalen = sizeof(struct if_data); ifd->ifi_mtu = ifp->if_mtu; ifd->ifi_metric = ifp->if_metric; ifd->ifi_baudrate = ifp->if_baudrate; ifd->ifi_hwassist = ifp->if_hwassist; ifd->ifi_epoch = ifp->if_epoch; ifd->ifi_lastchange = ifp->if_lastchange; ifd->ifi_ipackets = ifp->if_get_counter(ifp, IFCOUNTER_IPACKETS); ifd->ifi_ierrors = ifp->if_get_counter(ifp, IFCOUNTER_IERRORS); ifd->ifi_opackets = ifp->if_get_counter(ifp, IFCOUNTER_OPACKETS); ifd->ifi_oerrors = ifp->if_get_counter(ifp, IFCOUNTER_OERRORS); ifd->ifi_collisions = ifp->if_get_counter(ifp, IFCOUNTER_COLLISIONS); ifd->ifi_ibytes = ifp->if_get_counter(ifp, IFCOUNTER_IBYTES); ifd->ifi_obytes = ifp->if_get_counter(ifp, IFCOUNTER_OBYTES); ifd->ifi_imcasts = ifp->if_get_counter(ifp, IFCOUNTER_IMCASTS); ifd->ifi_omcasts = ifp->if_get_counter(ifp, IFCOUNTER_OMCASTS); ifd->ifi_iqdrops = ifp->if_get_counter(ifp, IFCOUNTER_IQDROPS); ifd->ifi_oqdrops = ifp->if_get_counter(ifp, IFCOUNTER_OQDROPS); ifd->ifi_noproto = ifp->if_get_counter(ifp, IFCOUNTER_NOPROTO); } /* * Wrapper functions for struct ifnet address list locking macros. These are * used by kernel modules to avoid encoding programming interface or binary * interface assumptions that may be violated when kernel-internal locking * approaches change. */ void if_addr_rlock(struct ifnet *ifp) { MPASS(*(uint64_t *)&ifp->if_addr_et == 0); epoch_enter_preempt(net_epoch_preempt, &ifp->if_addr_et); } void if_addr_runlock(struct ifnet *ifp) { epoch_exit_preempt(net_epoch_preempt, &ifp->if_addr_et); #ifdef INVARIANTS bzero(&ifp->if_addr_et, sizeof(struct epoch_tracker)); #endif } void if_maddr_rlock(if_t ifp) { MPASS(*(uint64_t *)&ifp->if_maddr_et == 0); epoch_enter_preempt(net_epoch_preempt, &ifp->if_maddr_et); } void if_maddr_runlock(if_t ifp) { epoch_exit_preempt(net_epoch_preempt, &ifp->if_maddr_et); #ifdef INVARIANTS bzero(&ifp->if_maddr_et, sizeof(struct epoch_tracker)); #endif } /* * Initialization, destruction and refcounting functions for ifaddrs. */ struct ifaddr * ifa_alloc(size_t size, int flags) { struct ifaddr *ifa; KASSERT(size >= sizeof(struct ifaddr), ("%s: invalid size %zu", __func__, size)); ifa = malloc(size, M_IFADDR, M_ZERO | flags); if (ifa == NULL) return (NULL); if ((ifa->ifa_opackets = counter_u64_alloc(flags)) == NULL) goto fail; if ((ifa->ifa_ipackets = counter_u64_alloc(flags)) == NULL) goto fail; if ((ifa->ifa_obytes = counter_u64_alloc(flags)) == NULL) goto fail; if ((ifa->ifa_ibytes = counter_u64_alloc(flags)) == NULL) goto fail; refcount_init(&ifa->ifa_refcnt, 1); return (ifa); fail: /* free(NULL) is okay */ counter_u64_free(ifa->ifa_opackets); counter_u64_free(ifa->ifa_ipackets); counter_u64_free(ifa->ifa_obytes); counter_u64_free(ifa->ifa_ibytes); free(ifa, M_IFADDR); return (NULL); } void ifa_ref(struct ifaddr *ifa) { refcount_acquire(&ifa->ifa_refcnt); } static void ifa_destroy(epoch_context_t ctx) { struct ifaddr *ifa; ifa = __containerof(ctx, struct ifaddr, ifa_epoch_ctx); counter_u64_free(ifa->ifa_opackets); counter_u64_free(ifa->ifa_ipackets); counter_u64_free(ifa->ifa_obytes); counter_u64_free(ifa->ifa_ibytes); free(ifa, M_IFADDR); } void ifa_free(struct ifaddr *ifa) { if (refcount_release(&ifa->ifa_refcnt)) epoch_call(net_epoch_preempt, &ifa->ifa_epoch_ctx, ifa_destroy); } static int ifa_maintain_loopback_route(int cmd, const char *otype, struct ifaddr *ifa, struct sockaddr *ia) { int error; struct rt_addrinfo info; struct sockaddr_dl null_sdl; struct ifnet *ifp; ifp = ifa->ifa_ifp; bzero(&info, sizeof(info)); if (cmd != RTM_DELETE) info.rti_ifp = V_loif; info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC | RTF_PINNED; info.rti_info[RTAX_DST] = ia; info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl; link_init_sdl(ifp, (struct sockaddr *)&null_sdl, ifp->if_type); error = rtrequest1_fib(cmd, &info, NULL, ifp->if_fib); if (error != 0 && !(cmd == RTM_ADD && error == EEXIST) && !(cmd == RTM_DELETE && error == ENOENT)) if_printf(ifp, "%s failed: %d\n", otype, error); return (error); } int ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) { return (ifa_maintain_loopback_route(RTM_ADD, "insertion", ifa, ia)); } int ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) { return (ifa_maintain_loopback_route(RTM_DELETE, "deletion", ifa, ia)); } int ifa_switch_loopback_route(struct ifaddr *ifa, struct sockaddr *ia) { return (ifa_maintain_loopback_route(RTM_CHANGE, "switch", ifa, ia)); } /* * XXX: Because sockaddr_dl has deeper structure than the sockaddr * structs used to represent other address families, it is necessary * to perform a different comparison. */ #define sa_dl_equal(a1, a2) \ ((((const struct sockaddr_dl *)(a1))->sdl_len == \ ((const struct sockaddr_dl *)(a2))->sdl_len) && \ (bcmp(CLLADDR((const struct sockaddr_dl *)(a1)), \ CLLADDR((const struct sockaddr_dl *)(a2)), \ ((const struct sockaddr_dl *)(a1))->sdl_alen) == 0)) /* * Locate an interface based on a complete address. */ /*ARGSUSED*/ struct ifaddr * ifa_ifwithaddr(const struct sockaddr *addr) { struct ifnet *ifp; struct ifaddr *ifa; MPASS(in_epoch(net_epoch_preempt)); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != addr->sa_family) continue; if (sa_equal(addr, ifa->ifa_addr)) { goto done; } /* IP6 doesn't have broadcast */ if ((ifp->if_flags & IFF_BROADCAST) && ifa->ifa_broadaddr && ifa->ifa_broadaddr->sa_len != 0 && sa_equal(ifa->ifa_broadaddr, addr)) { goto done; } } } ifa = NULL; done: return (ifa); } int ifa_ifwithaddr_check(const struct sockaddr *addr) { int rc; NET_EPOCH_ENTER(); rc = (ifa_ifwithaddr(addr) != NULL); NET_EPOCH_EXIT(); return (rc); } /* * Locate an interface based on the broadcast address. */ /* ARGSUSED */ struct ifaddr * ifa_ifwithbroadaddr(const struct sockaddr *addr, int fibnum) { struct ifnet *ifp; struct ifaddr *ifa; MPASS(in_epoch(net_epoch_preempt)); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) continue; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != addr->sa_family) continue; if ((ifp->if_flags & IFF_BROADCAST) && ifa->ifa_broadaddr && ifa->ifa_broadaddr->sa_len != 0 && sa_equal(ifa->ifa_broadaddr, addr)) { goto done; } } } ifa = NULL; done: return (ifa); } /* * Locate the point to point interface with a given destination address. */ /*ARGSUSED*/ struct ifaddr * ifa_ifwithdstaddr(const struct sockaddr *addr, int fibnum) { struct ifnet *ifp; struct ifaddr *ifa; MPASS(in_epoch(net_epoch_preempt)); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if ((ifp->if_flags & IFF_POINTOPOINT) == 0) continue; if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) continue; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != addr->sa_family) continue; if (ifa->ifa_dstaddr != NULL && sa_equal(addr, ifa->ifa_dstaddr)) { goto done; } } } ifa = NULL; done: return (ifa); } /* * Find an interface on a specific network. If many, choice * is most specific found. */ struct ifaddr * ifa_ifwithnet(const struct sockaddr *addr, int ignore_ptp, int fibnum) { struct ifnet *ifp; struct ifaddr *ifa; struct ifaddr *ifa_maybe = NULL; u_int af = addr->sa_family; const char *addr_data = addr->sa_data, *cplim; MPASS(in_epoch(net_epoch_preempt)); /* * AF_LINK addresses can be looked up directly by their index number, * so do that if we can. */ if (af == AF_LINK) { const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)addr; if (sdl->sdl_index && sdl->sdl_index <= V_if_index) return (ifaddr_byindex(sdl->sdl_index)); } /* * Scan though each interface, looking for ones that have addresses * in this address family and the requested fib. Maintain a reference * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that * kept it stable when we move onto the next interface. */ CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum)) continue; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { const char *cp, *cp2, *cp3; if (ifa->ifa_addr->sa_family != af) next: continue; if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) { /* * This is a bit broken as it doesn't * take into account that the remote end may * be a single node in the network we are * looking for. * The trouble is that we don't know the * netmask for the remote end. */ if (ifa->ifa_dstaddr != NULL && sa_equal(addr, ifa->ifa_dstaddr)) { goto done; } } else { /* * Scan all the bits in the ifa's address. * If a bit dissagrees with what we are * looking for, mask it with the netmask * to see if it really matters. * (A byte at a time) */ if (ifa->ifa_netmask == 0) continue; cp = addr_data; cp2 = ifa->ifa_addr->sa_data; cp3 = ifa->ifa_netmask->sa_data; cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; while (cp3 < cplim) if ((*cp++ ^ *cp2++) & *cp3++) goto next; /* next address! */ /* * If the netmask of what we just found * is more specific than what we had before * (if we had one), or if the virtual status * of new prefix is better than of the old one, * then remember the new one before continuing * to search for an even better one. */ if (ifa_maybe == NULL || ifa_preferred(ifa_maybe, ifa) || rn_refines((caddr_t)ifa->ifa_netmask, (caddr_t)ifa_maybe->ifa_netmask)) { ifa_maybe = ifa; } } } } ifa = ifa_maybe; ifa_maybe = NULL; done: return (ifa); } /* * Find an interface address specific to an interface best matching * a given address. */ struct ifaddr * ifaof_ifpforaddr(const struct sockaddr *addr, struct ifnet *ifp) { struct ifaddr *ifa; const char *cp, *cp2, *cp3; char *cplim; struct ifaddr *ifa_maybe = NULL; u_int af = addr->sa_family; if (af >= AF_MAX) return (NULL); MPASS(in_epoch(net_epoch_preempt)); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != af) continue; if (ifa_maybe == NULL) ifa_maybe = ifa; if (ifa->ifa_netmask == 0) { if (sa_equal(addr, ifa->ifa_addr) || (ifa->ifa_dstaddr && sa_equal(addr, ifa->ifa_dstaddr))) goto done; continue; } if (ifp->if_flags & IFF_POINTOPOINT) { if (sa_equal(addr, ifa->ifa_dstaddr)) goto done; } else { cp = addr->sa_data; cp2 = ifa->ifa_addr->sa_data; cp3 = ifa->ifa_netmask->sa_data; cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask; for (; cp3 < cplim; cp3++) if ((*cp++ ^ *cp2++) & *cp3) break; if (cp3 == cplim) goto done; } } ifa = ifa_maybe; done: return (ifa); } /* * See whether new ifa is better than current one: * 1) A non-virtual one is preferred over virtual. * 2) A virtual in master state preferred over any other state. * * Used in several address selecting functions. */ int ifa_preferred(struct ifaddr *cur, struct ifaddr *next) { return (cur->ifa_carp && (!next->ifa_carp || ((*carp_master_p)(next) && !(*carp_master_p)(cur)))); } #include /* * Default action when installing a route with a Link Level gateway. * Lookup an appropriate real ifa to point to. * This should be moved to /sys/net/link.c eventually. */ static void link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info) { struct ifaddr *ifa, *oifa; struct sockaddr *dst; struct ifnet *ifp; if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == NULL) || ((ifp = ifa->ifa_ifp) == NULL) || ((dst = rt_key(rt)) == NULL)) return; NET_EPOCH_ENTER(); ifa = ifaof_ifpforaddr(dst, ifp); if (ifa) { oifa = rt->rt_ifa; if (oifa != ifa) { ifa_free(oifa); ifa_ref(ifa); } rt->rt_ifa = ifa; if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest) ifa->ifa_rtrequest(cmd, rt, info); } NET_EPOCH_EXIT(); } struct sockaddr_dl * link_alloc_sdl(size_t size, int flags) { return (malloc(size, M_TEMP, flags)); } void link_free_sdl(struct sockaddr *sa) { free(sa, M_TEMP); } /* * Fills in given sdl with interface basic info. * Returns pointer to filled sdl. */ struct sockaddr_dl * link_init_sdl(struct ifnet *ifp, struct sockaddr *paddr, u_char iftype) { struct sockaddr_dl *sdl; sdl = (struct sockaddr_dl *)paddr; memset(sdl, 0, sizeof(struct sockaddr_dl)); sdl->sdl_len = sizeof(struct sockaddr_dl); sdl->sdl_family = AF_LINK; sdl->sdl_index = ifp->if_index; sdl->sdl_type = iftype; return (sdl); } /* * Mark an interface down and notify protocols of * the transition. */ static void if_unroute(struct ifnet *ifp, int flag, int fam) { struct ifaddr *ifa; KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP")); ifp->if_flags &= ~flag; getmicrotime(&ifp->if_lastchange); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) pfctlinput(PRC_IFDOWN, ifa->ifa_addr); ifp->if_qflush(ifp); if (ifp->if_carp) (*carp_linkstate_p)(ifp); rt_ifmsg(ifp); } /* * Mark an interface up and notify protocols of * the transition. */ static void if_route(struct ifnet *ifp, int flag, int fam) { struct ifaddr *ifa; KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP")); ifp->if_flags |= flag; getmicrotime(&ifp->if_lastchange); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family)) pfctlinput(PRC_IFUP, ifa->ifa_addr); if (ifp->if_carp) (*carp_linkstate_p)(ifp); rt_ifmsg(ifp); #ifdef INET6 in6_if_up(ifp); #endif } void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */ void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */ struct ifnet *(*vlan_trunkdev_p)(struct ifnet *); struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t); int (*vlan_tag_p)(struct ifnet *, uint16_t *); int (*vlan_setcookie_p)(struct ifnet *, void *); void *(*vlan_cookie_p)(struct ifnet *); /* * Handle a change in the interface link state. To avoid LORs * between driver lock and upper layer locks, as well as possible * recursions, we post event to taskqueue, and all job * is done in static do_link_state_change(). */ void if_link_state_change(struct ifnet *ifp, int link_state) { /* Return if state hasn't changed. */ if (ifp->if_link_state == link_state) return; ifp->if_link_state = link_state; taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask); } static void do_link_state_change(void *arg, int pending) { struct ifnet *ifp = (struct ifnet *)arg; int link_state = ifp->if_link_state; CURVNET_SET(ifp->if_vnet); /* Notify that the link state has changed. */ rt_ifmsg(ifp); if (ifp->if_vlantrunk != NULL) (*vlan_link_state_p)(ifp); if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) && ifp->if_l2com != NULL) (*ng_ether_link_state_p)(ifp, link_state); if (ifp->if_carp) (*carp_linkstate_p)(ifp); if (ifp->if_bridge) ifp->if_bridge_linkstate(ifp); if (ifp->if_lagg) (*lagg_linkstate_p)(ifp, link_state); if (IS_DEFAULT_VNET(curvnet)) devctl_notify("IFNET", ifp->if_xname, (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL); if (pending > 1) if_printf(ifp, "%d link states coalesced\n", pending); if (log_link_state_change) if_printf(ifp, "link state changed to %s\n", (link_state == LINK_STATE_UP) ? "UP" : "DOWN" ); EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state); CURVNET_RESTORE(); } /* * Mark an interface down and notify protocols of * the transition. */ void if_down(struct ifnet *ifp) { EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN); if_unroute(ifp, IFF_UP, AF_UNSPEC); } /* * Mark an interface up and notify protocols of * the transition. */ void if_up(struct ifnet *ifp) { if_route(ifp, IFF_UP, AF_UNSPEC); EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP); } /* * Flush an interface queue. */ void if_qflush(struct ifnet *ifp) { struct mbuf *m, *n; struct ifaltq *ifq; ifq = &ifp->if_snd; IFQ_LOCK(ifq); #ifdef ALTQ if (ALTQ_IS_ENABLED(ifq)) ALTQ_PURGE(ifq); #endif n = ifq->ifq_head; while ((m = n) != NULL) { n = m->m_nextpkt; m_freem(m); } ifq->ifq_head = 0; ifq->ifq_tail = 0; ifq->ifq_len = 0; IFQ_UNLOCK(ifq); } /* * Map interface name to interface structure pointer, with or without * returning a reference. */ struct ifnet * ifunit_ref(const char *name) { struct ifnet *ifp; IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 && !(ifp->if_flags & IFF_DYING)) break; } if (ifp != NULL) if_ref(ifp); IFNET_RUNLOCK_NOSLEEP(); return (ifp); } struct ifnet * ifunit(const char *name) { struct ifnet *ifp; IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0) break; } IFNET_RUNLOCK_NOSLEEP(); return (ifp); } static void * ifr_buffer_get_buffer(void *data) { union ifreq_union *ifrup; ifrup = data; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) return ((void *)(uintptr_t) ifrup->ifr32.ifr_ifru.ifru_buffer.buffer); #endif return (ifrup->ifr.ifr_ifru.ifru_buffer.buffer); } static void ifr_buffer_set_buffer_null(void *data) { union ifreq_union *ifrup; ifrup = data; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) ifrup->ifr32.ifr_ifru.ifru_buffer.buffer = 0; else #endif ifrup->ifr.ifr_ifru.ifru_buffer.buffer = NULL; } static size_t ifr_buffer_get_length(void *data) { union ifreq_union *ifrup; ifrup = data; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) return (ifrup->ifr32.ifr_ifru.ifru_buffer.length); #endif return (ifrup->ifr.ifr_ifru.ifru_buffer.length); } static void ifr_buffer_set_length(void *data, size_t len) { union ifreq_union *ifrup; ifrup = data; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) ifrup->ifr32.ifr_ifru.ifru_buffer.length = len; else #endif ifrup->ifr.ifr_ifru.ifru_buffer.length = len; } void * ifr_data_get_ptr(void *ifrp) { union ifreq_union *ifrup; ifrup = ifrp; #ifdef COMPAT_FREEBSD32 if (SV_CURPROC_FLAG(SV_ILP32)) return ((void *)(uintptr_t) ifrup->ifr32.ifr_ifru.ifru_data); #endif return (ifrup->ifr.ifr_ifru.ifru_data); } /* * Hardware specific interface ioctls. */ static int ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td) { struct ifreq *ifr; int error = 0, do_ifup = 0; int new_flags, temp_flags; size_t namelen, onamelen; size_t descrlen; char *descrbuf, *odescrbuf; char new_name[IFNAMSIZ]; struct ifaddr *ifa; struct sockaddr_dl *sdl; ifr = (struct ifreq *)data; switch (cmd) { case SIOCGIFINDEX: ifr->ifr_index = ifp->if_index; break; case SIOCGIFFLAGS: temp_flags = ifp->if_flags | ifp->if_drv_flags; ifr->ifr_flags = temp_flags & 0xffff; ifr->ifr_flagshigh = temp_flags >> 16; break; case SIOCGIFCAP: ifr->ifr_reqcap = ifp->if_capabilities; ifr->ifr_curcap = ifp->if_capenable; break; #ifdef MAC case SIOCGIFMAC: error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp); break; #endif case SIOCGIFMETRIC: ifr->ifr_metric = ifp->if_metric; break; case SIOCGIFMTU: ifr->ifr_mtu = ifp->if_mtu; break; case SIOCGIFPHYS: /* XXXGL: did this ever worked? */ ifr->ifr_phys = 0; break; case SIOCGIFDESCR: error = 0; sx_slock(&ifdescr_sx); if (ifp->if_description == NULL) error = ENOMSG; else { /* space for terminating nul */ descrlen = strlen(ifp->if_description) + 1; if (ifr_buffer_get_length(ifr) < descrlen) ifr_buffer_set_buffer_null(ifr); else error = copyout(ifp->if_description, ifr_buffer_get_buffer(ifr), descrlen); ifr_buffer_set_length(ifr, descrlen); } sx_sunlock(&ifdescr_sx); break; case SIOCSIFDESCR: error = priv_check(td, PRIV_NET_SETIFDESCR); if (error) return (error); /* * Copy only (length-1) bytes to make sure that * if_description is always nul terminated. The * length parameter is supposed to count the * terminating nul in. */ if (ifr_buffer_get_length(ifr) > ifdescr_maxlen) return (ENAMETOOLONG); else if (ifr_buffer_get_length(ifr) == 0) descrbuf = NULL; else { descrbuf = malloc(ifr_buffer_get_length(ifr), M_IFDESCR, M_WAITOK | M_ZERO); error = copyin(ifr_buffer_get_buffer(ifr), descrbuf, ifr_buffer_get_length(ifr) - 1); if (error) { free(descrbuf, M_IFDESCR); break; } } sx_xlock(&ifdescr_sx); odescrbuf = ifp->if_description; ifp->if_description = descrbuf; sx_xunlock(&ifdescr_sx); getmicrotime(&ifp->if_lastchange); free(odescrbuf, M_IFDESCR); break; case SIOCGIFFIB: ifr->ifr_fib = ifp->if_fib; break; case SIOCSIFFIB: error = priv_check(td, PRIV_NET_SETIFFIB); if (error) return (error); if (ifr->ifr_fib >= rt_numfibs) return (EINVAL); ifp->if_fib = ifr->ifr_fib; break; case SIOCSIFFLAGS: error = priv_check(td, PRIV_NET_SETIFFLAGS); if (error) return (error); /* * Currently, no driver owned flags pass the IFF_CANTCHANGE * check, so we don't need special handling here yet. */ new_flags = (ifr->ifr_flags & 0xffff) | (ifr->ifr_flagshigh << 16); if (ifp->if_flags & IFF_UP && (new_flags & IFF_UP) == 0) { if_down(ifp); } else if (new_flags & IFF_UP && (ifp->if_flags & IFF_UP) == 0) { do_ifup = 1; } /* See if permanently promiscuous mode bit is about to flip */ if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) { if (new_flags & IFF_PPROMISC) ifp->if_flags |= IFF_PROMISC; else if (ifp->if_pcount == 0) ifp->if_flags &= ~IFF_PROMISC; if (log_promisc_mode_change) if_printf(ifp, "permanently promiscuous mode %s\n", ((new_flags & IFF_PPROMISC) ? "enabled" : "disabled")); } ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) | (new_flags &~ IFF_CANTCHANGE); if (ifp->if_ioctl) { (void) (*ifp->if_ioctl)(ifp, cmd, data); } if (do_ifup) if_up(ifp); getmicrotime(&ifp->if_lastchange); break; case SIOCSIFCAP: error = priv_check(td, PRIV_NET_SETIFCAP); if (error) return (error); if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); if (ifr->ifr_reqcap & ~ifp->if_capabilities) return (EINVAL); error = (*ifp->if_ioctl)(ifp, cmd, data); if (error == 0) getmicrotime(&ifp->if_lastchange); break; #ifdef MAC case SIOCSIFMAC: error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp); break; #endif case SIOCSIFNAME: error = priv_check(td, PRIV_NET_SETIFNAME); if (error) return (error); error = copyinstr(ifr_data_get_ptr(ifr), new_name, IFNAMSIZ, NULL); if (error != 0) return (error); if (new_name[0] == '\0') return (EINVAL); if (new_name[IFNAMSIZ-1] != '\0') { new_name[IFNAMSIZ-1] = '\0'; if (strlen(new_name) == IFNAMSIZ-1) return (EINVAL); } if (ifunit(new_name) != NULL) return (EEXIST); /* * XXX: Locking. Nothing else seems to lock if_flags, * and there are numerous other races with the * ifunit() checks not being atomic with namespace * changes (renames, vmoves, if_attach, etc). */ ifp->if_flags |= IFF_RENAMING; /* Announce the departure of the interface. */ rt_ifannouncemsg(ifp, IFAN_DEPARTURE); EVENTHANDLER_INVOKE(ifnet_departure_event, ifp); if_printf(ifp, "changing name to '%s'\n", new_name); IF_ADDR_WLOCK(ifp); strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname)); ifa = ifp->if_addr; sdl = (struct sockaddr_dl *)ifa->ifa_addr; namelen = strlen(new_name); onamelen = sdl->sdl_nlen; /* * Move the address if needed. This is safe because we * allocate space for a name of length IFNAMSIZ when we * create this in if_attach(). */ if (namelen != onamelen) { bcopy(sdl->sdl_data + onamelen, sdl->sdl_data + namelen, sdl->sdl_alen); } bcopy(new_name, sdl->sdl_data, namelen); sdl->sdl_nlen = namelen; sdl = (struct sockaddr_dl *)ifa->ifa_netmask; bzero(sdl->sdl_data, onamelen); while (namelen != 0) sdl->sdl_data[--namelen] = 0xff; IF_ADDR_WUNLOCK(ifp); EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp); /* Announce the return of the interface. */ rt_ifannouncemsg(ifp, IFAN_ARRIVAL); ifp->if_flags &= ~IFF_RENAMING; break; #ifdef VIMAGE case SIOCSIFVNET: error = priv_check(td, PRIV_NET_SETIFVNET); if (error) return (error); error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid); break; #endif case SIOCSIFMETRIC: error = priv_check(td, PRIV_NET_SETIFMETRIC); if (error) return (error); ifp->if_metric = ifr->ifr_metric; getmicrotime(&ifp->if_lastchange); break; case SIOCSIFPHYS: error = priv_check(td, PRIV_NET_SETIFPHYS); if (error) return (error); if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); error = (*ifp->if_ioctl)(ifp, cmd, data); if (error == 0) getmicrotime(&ifp->if_lastchange); break; case SIOCSIFMTU: { u_long oldmtu = ifp->if_mtu; error = priv_check(td, PRIV_NET_SETIFMTU); if (error) return (error); if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) return (EINVAL); if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); error = (*ifp->if_ioctl)(ifp, cmd, data); if (error == 0) { getmicrotime(&ifp->if_lastchange); rt_ifmsg(ifp); #ifdef INET NETDUMP_REINIT(ifp); #endif } /* * If the link MTU changed, do network layer specific procedure. */ if (ifp->if_mtu != oldmtu) { #ifdef INET6 nd6_setmtu(ifp); #endif rt_updatemtu(ifp); } break; } case SIOCADDMULTI: case SIOCDELMULTI: if (cmd == SIOCADDMULTI) error = priv_check(td, PRIV_NET_ADDMULTI); else error = priv_check(td, PRIV_NET_DELMULTI); if (error) return (error); /* Don't allow group membership on non-multicast interfaces. */ if ((ifp->if_flags & IFF_MULTICAST) == 0) return (EOPNOTSUPP); /* Don't let users screw up protocols' entries. */ if (ifr->ifr_addr.sa_family != AF_LINK) return (EINVAL); if (cmd == SIOCADDMULTI) { struct ifmultiaddr *ifma; /* * Userland is only permitted to join groups once * via the if_addmulti() KPI, because it cannot hold * struct ifmultiaddr * between calls. It may also * lose a race while we check if the membership * already exists. */ IF_ADDR_RLOCK(ifp); ifma = if_findmulti(ifp, &ifr->ifr_addr); IF_ADDR_RUNLOCK(ifp); if (ifma != NULL) error = EADDRINUSE; else error = if_addmulti(ifp, &ifr->ifr_addr, &ifma); } else { error = if_delmulti(ifp, &ifr->ifr_addr); } if (error == 0) getmicrotime(&ifp->if_lastchange); break; case SIOCSIFPHYADDR: case SIOCDIFPHYADDR: #ifdef INET6 case SIOCSIFPHYADDR_IN6: #endif case SIOCSIFMEDIA: case SIOCSIFGENERIC: error = priv_check(td, PRIV_NET_HWIOCTL); if (error) return (error); if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); error = (*ifp->if_ioctl)(ifp, cmd, data); if (error == 0) getmicrotime(&ifp->if_lastchange); break; case SIOCGIFSTATUS: case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: case SIOCGIFMEDIA: case SIOCGIFXMEDIA: case SIOCGIFGENERIC: case SIOCGIFRSSKEY: case SIOCGIFRSSHASH: if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); error = (*ifp->if_ioctl)(ifp, cmd, data); break; case SIOCSIFLLADDR: error = priv_check(td, PRIV_NET_SETLLADDR); if (error) return (error); error = if_setlladdr(ifp, ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len); break; case SIOCGHWADDR: error = if_gethwaddr(ifp, ifr); break; CASE_IOC_IFGROUPREQ(SIOCAIFGROUP): error = priv_check(td, PRIV_NET_ADDIFGROUP); if (error) return (error); if ((error = if_addgroup(ifp, ifgr_group_get((struct ifgroupreq *)data)))) return (error); break; CASE_IOC_IFGROUPREQ(SIOCGIFGROUP): if ((error = if_getgroup((struct ifgroupreq *)data, ifp))) return (error); break; CASE_IOC_IFGROUPREQ(SIOCDIFGROUP): error = priv_check(td, PRIV_NET_DELIFGROUP); if (error) return (error); if ((error = if_delgroup(ifp, ifgr_group_get((struct ifgroupreq *)data)))) return (error); break; default: error = ENOIOCTL; break; } return (error); } #ifdef COMPAT_FREEBSD32 struct ifconf32 { int32_t ifc_len; union { uint32_t ifcu_buf; uint32_t ifcu_req; } ifc_ifcu; }; #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32) #endif #ifdef COMPAT_FREEBSD32 static void ifmr_init(struct ifmediareq *ifmr, caddr_t data) { struct ifmediareq32 *ifmr32; ifmr32 = (struct ifmediareq32 *)data; memcpy(ifmr->ifm_name, ifmr32->ifm_name, sizeof(ifmr->ifm_name)); ifmr->ifm_current = ifmr32->ifm_current; ifmr->ifm_mask = ifmr32->ifm_mask; ifmr->ifm_status = ifmr32->ifm_status; ifmr->ifm_active = ifmr32->ifm_active; ifmr->ifm_count = ifmr32->ifm_count; ifmr->ifm_ulist = (int *)(uintptr_t)ifmr32->ifm_ulist; } static void ifmr_update(const struct ifmediareq *ifmr, caddr_t data) { struct ifmediareq32 *ifmr32; ifmr32 = (struct ifmediareq32 *)data; ifmr32->ifm_current = ifmr->ifm_current; ifmr32->ifm_mask = ifmr->ifm_mask; ifmr32->ifm_status = ifmr->ifm_status; ifmr32->ifm_active = ifmr->ifm_active; ifmr32->ifm_count = ifmr->ifm_count; } #endif /* * Interface ioctls. */ int ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td) { #ifdef COMPAT_FREEBSD32 caddr_t saved_data = NULL; struct ifmediareq ifmr; struct ifmediareq *ifmrp; #endif struct ifnet *ifp; struct ifreq *ifr; int error; int oif_flags; #ifdef VIMAGE int shutdown; #endif CURVNET_SET(so->so_vnet); #ifdef VIMAGE /* Make sure the VNET is stable. */ shutdown = (so->so_vnet->vnet_state > SI_SUB_VNET && so->so_vnet->vnet_state < SI_SUB_VNET_DONE) ? 1 : 0; if (shutdown) { CURVNET_RESTORE(); return (EBUSY); } #endif switch (cmd) { case SIOCGIFCONF: error = ifconf(cmd, data); CURVNET_RESTORE(); return (error); #ifdef COMPAT_FREEBSD32 case SIOCGIFCONF32: { struct ifconf32 *ifc32; struct ifconf ifc; ifc32 = (struct ifconf32 *)data; ifc.ifc_len = ifc32->ifc_len; ifc.ifc_buf = PTRIN(ifc32->ifc_buf); error = ifconf(SIOCGIFCONF, (void *)&ifc); CURVNET_RESTORE(); if (error == 0) ifc32->ifc_len = ifc.ifc_len; return (error); } #endif } #ifdef COMPAT_FREEBSD32 ifmrp = NULL; switch (cmd) { case SIOCGIFMEDIA32: case SIOCGIFXMEDIA32: ifmrp = &ifmr; ifmr_init(ifmrp, data); cmd = _IOC_NEWTYPE(cmd, struct ifmediareq); saved_data = data; data = (caddr_t)ifmrp; } #endif ifr = (struct ifreq *)data; switch (cmd) { #ifdef VIMAGE case SIOCSIFRVNET: error = priv_check(td, PRIV_NET_SETIFVNET); if (error == 0) error = if_vmove_reclaim(td, ifr->ifr_name, ifr->ifr_jid); goto out_noref; #endif case SIOCIFCREATE: case SIOCIFCREATE2: error = priv_check(td, PRIV_NET_IFCREATE); if (error == 0) error = if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name), cmd == SIOCIFCREATE2 ? ifr_data_get_ptr(ifr) : NULL); goto out_noref; case SIOCIFDESTROY: error = priv_check(td, PRIV_NET_IFDESTROY); if (error == 0) error = if_clone_destroy(ifr->ifr_name); goto out_noref; case SIOCIFGCLONERS: error = if_clone_list((struct if_clonereq *)data); goto out_noref; CASE_IOC_IFGROUPREQ(SIOCGIFGMEMB): error = if_getgroupmembers((struct ifgroupreq *)data); goto out_noref; #if defined(INET) || defined(INET6) case SIOCSVH: case SIOCGVH: if (carp_ioctl_p == NULL) error = EPROTONOSUPPORT; else error = (*carp_ioctl_p)(ifr, cmd, td); goto out_noref; #endif } ifp = ifunit_ref(ifr->ifr_name); if (ifp == NULL) { error = ENXIO; goto out_noref; } error = ifhwioctl(cmd, ifp, data, td); if (error != ENOIOCTL) goto out_ref; oif_flags = ifp->if_flags; if (so->so_proto == NULL) { error = EOPNOTSUPP; goto out_ref; } /* * Pass the request on to the socket control method, and if the * latter returns EOPNOTSUPP, directly to the interface. * * Make an exception for the legacy SIOCSIF* requests. Drivers * trust SIOCSIFADDR et al to come from an already privileged * layer, and do not perform any credentials checks or input * validation. */ error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd, data, ifp, td)); if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL && cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR && cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK) error = (*ifp->if_ioctl)(ifp, cmd, data); if ((oif_flags ^ ifp->if_flags) & IFF_UP) { #ifdef INET6 if (ifp->if_flags & IFF_UP) in6_if_up(ifp); #endif } out_ref: if_rele(ifp); out_noref: #ifdef COMPAT_FREEBSD32 if (ifmrp != NULL) { KASSERT((cmd == SIOCGIFMEDIA || cmd == SIOCGIFXMEDIA), ("ifmrp non-NULL, but cmd is not an ifmedia req 0x%lx", cmd)); data = saved_data; ifmr_update(ifmrp, data); } #endif CURVNET_RESTORE(); return (error); } /* * The code common to handling reference counted flags, * e.g., in ifpromisc() and if_allmulti(). * The "pflag" argument can specify a permanent mode flag to check, * such as IFF_PPROMISC for promiscuous mode; should be 0 if none. * * Only to be used on stack-owned flags, not driver-owned flags. */ static int if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch) { struct ifreq ifr; int error; int oldflags, oldcount; /* Sanity checks to catch programming errors */ KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0, ("%s: setting driver-owned flag %d", __func__, flag)); if (onswitch) KASSERT(*refcount >= 0, ("%s: increment negative refcount %d for flag %d", __func__, *refcount, flag)); else KASSERT(*refcount > 0, ("%s: decrement non-positive refcount %d for flag %d", __func__, *refcount, flag)); /* In case this mode is permanent, just touch refcount */ if (ifp->if_flags & pflag) { *refcount += onswitch ? 1 : -1; return (0); } /* Save ifnet parameters for if_ioctl() may fail */ oldcount = *refcount; oldflags = ifp->if_flags; /* * See if we aren't the only and touching refcount is enough. * Actually toggle interface flag if we are the first or last. */ if (onswitch) { if ((*refcount)++) return (0); ifp->if_flags |= flag; } else { if (--(*refcount)) return (0); ifp->if_flags &= ~flag; } /* Call down the driver since we've changed interface flags */ if (ifp->if_ioctl == NULL) { error = EOPNOTSUPP; goto recover; } ifr.ifr_flags = ifp->if_flags & 0xffff; ifr.ifr_flagshigh = ifp->if_flags >> 16; error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); if (error) goto recover; /* Notify userland that interface flags have changed */ rt_ifmsg(ifp); return (0); recover: /* Recover after driver error */ *refcount = oldcount; ifp->if_flags = oldflags; return (error); } /* * Set/clear promiscuous mode on interface ifp based on the truth value * of pswitch. The calls are reference counted so that only the first * "on" request actually has an effect, as does the final "off" request. * Results are undefined if the "off" and "on" requests are not matched. */ int ifpromisc(struct ifnet *ifp, int pswitch) { int error; int oldflags = ifp->if_flags; error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC, &ifp->if_pcount, pswitch); /* If promiscuous mode status has changed, log a message */ if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC) && log_promisc_mode_change) if_printf(ifp, "promiscuous mode %s\n", (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled"); return (error); } /* * Return interface configuration * of system. List may be used * in later ioctl's (above) to get * other information. */ /*ARGSUSED*/ static int ifconf(u_long cmd, caddr_t data) { struct ifconf *ifc = (struct ifconf *)data; struct ifnet *ifp; struct ifaddr *ifa; struct ifreq ifr; struct sbuf *sb; int error, full = 0, valid_len, max_len; /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */ max_len = MAXPHYS - 1; /* Prevent hostile input from being able to crash the system */ if (ifc->ifc_len <= 0) return (EINVAL); again: if (ifc->ifc_len <= max_len) { max_len = ifc->ifc_len; full = 1; } sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN); max_len = 0; valid_len = 0; IFNET_RLOCK(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { int addrs; /* * Zero the ifr to make sure we don't disclose the contents * of the stack. */ memset(&ifr, 0, sizeof(ifr)); if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name)) >= sizeof(ifr.ifr_name)) { sbuf_delete(sb); IFNET_RUNLOCK(); return (ENAMETOOLONG); } addrs = 0; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct sockaddr *sa = ifa->ifa_addr; if (prison_if(curthread->td_ucred, sa) != 0) continue; addrs++; if (sa->sa_len <= sizeof(*sa)) { if (sa->sa_len < sizeof(*sa)) { memset(&ifr.ifr_ifru.ifru_addr, 0, sizeof(ifr.ifr_ifru.ifru_addr)); memcpy(&ifr.ifr_ifru.ifru_addr, sa, sa->sa_len); } else ifr.ifr_ifru.ifru_addr = *sa; sbuf_bcat(sb, &ifr, sizeof(ifr)); max_len += sizeof(ifr); } else { sbuf_bcat(sb, &ifr, offsetof(struct ifreq, ifr_addr)); max_len += offsetof(struct ifreq, ifr_addr); sbuf_bcat(sb, sa, sa->sa_len); max_len += sa->sa_len; } if (sbuf_error(sb) == 0) valid_len = sbuf_len(sb); } IF_ADDR_RUNLOCK(ifp); if (addrs == 0) { sbuf_bcat(sb, &ifr, sizeof(ifr)); max_len += sizeof(ifr); if (sbuf_error(sb) == 0) valid_len = sbuf_len(sb); } } IFNET_RUNLOCK(); /* * If we didn't allocate enough space (uncommon), try again. If * we have already allocated as much space as we are allowed, * return what we've got. */ if (valid_len != max_len && !full) { sbuf_delete(sb); goto again; } ifc->ifc_len = valid_len; sbuf_finish(sb); error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len); sbuf_delete(sb); return (error); } /* * Just like ifpromisc(), but for all-multicast-reception mode. */ int if_allmulti(struct ifnet *ifp, int onswitch) { return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch)); } struct ifmultiaddr * if_findmulti(struct ifnet *ifp, const struct sockaddr *sa) { struct ifmultiaddr *ifma; IF_ADDR_LOCK_ASSERT(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (sa->sa_family == AF_LINK) { if (sa_dl_equal(ifma->ifma_addr, sa)) break; } else { if (sa_equal(ifma->ifma_addr, sa)) break; } } return ifma; } /* * Allocate a new ifmultiaddr and initialize based on passed arguments. We * make copies of passed sockaddrs. The ifmultiaddr will not be added to * the ifnet multicast address list here, so the caller must do that and * other setup work (such as notifying the device driver). The reference * count is initialized to 1. */ static struct ifmultiaddr * if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa, int mflags) { struct ifmultiaddr *ifma; struct sockaddr *dupsa; ifma = malloc(sizeof *ifma, M_IFMADDR, mflags | M_ZERO); if (ifma == NULL) return (NULL); dupsa = malloc(sa->sa_len, M_IFMADDR, mflags); if (dupsa == NULL) { free(ifma, M_IFMADDR); return (NULL); } bcopy(sa, dupsa, sa->sa_len); ifma->ifma_addr = dupsa; ifma->ifma_ifp = ifp; ifma->ifma_refcount = 1; ifma->ifma_protospec = NULL; if (llsa == NULL) { ifma->ifma_lladdr = NULL; return (ifma); } dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags); if (dupsa == NULL) { free(ifma->ifma_addr, M_IFMADDR); free(ifma, M_IFMADDR); return (NULL); } bcopy(llsa, dupsa, llsa->sa_len); ifma->ifma_lladdr = dupsa; return (ifma); } /* * if_freemulti: free ifmultiaddr structure and possibly attached related * addresses. The caller is responsible for implementing reference * counting, notifying the driver, handling routing messages, and releasing * any dependent link layer state. */ #ifdef MCAST_VERBOSE extern void kdb_backtrace(void); #endif static void if_freemulti_internal(struct ifmultiaddr *ifma) { KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d", ifma->ifma_refcount)); if (ifma->ifma_lladdr != NULL) free(ifma->ifma_lladdr, M_IFMADDR); #ifdef MCAST_VERBOSE kdb_backtrace(); printf("%s freeing ifma: %p\n", __func__, ifma); #endif free(ifma->ifma_addr, M_IFMADDR); free(ifma, M_IFMADDR); } static void if_destroymulti(epoch_context_t ctx) { struct ifmultiaddr *ifma; ifma = __containerof(ctx, struct ifmultiaddr, ifma_epoch_ctx); if_freemulti_internal(ifma); } void if_freemulti(struct ifmultiaddr *ifma) { KASSERT(ifma->ifma_refcount == 0, ("if_freemulti_epoch: refcount %d", ifma->ifma_refcount)); epoch_call(net_epoch_preempt, &ifma->ifma_epoch_ctx, if_destroymulti); } /* * Register an additional multicast address with a network interface. * * - If the address is already present, bump the reference count on the * address and return. * - If the address is not link-layer, look up a link layer address. * - Allocate address structures for one or both addresses, and attach to the * multicast address list on the interface. If automatically adding a link * layer address, the protocol address will own a reference to the link * layer address, to be freed when it is freed. * - Notify the network device driver of an addition to the multicast address * list. * * 'sa' points to caller-owned memory with the desired multicast address. * * 'retifma' will be used to return a pointer to the resulting multicast * address reference, if desired. */ int if_addmulti(struct ifnet *ifp, struct sockaddr *sa, struct ifmultiaddr **retifma) { struct ifmultiaddr *ifma, *ll_ifma; struct sockaddr *llsa; struct sockaddr_dl sdl; int error; #ifdef INET IN_MULTI_LIST_UNLOCK_ASSERT(); #endif #ifdef INET6 IN6_MULTI_LIST_UNLOCK_ASSERT(); #endif /* * If the address is already present, return a new reference to it; * otherwise, allocate storage and set up a new address. */ IF_ADDR_WLOCK(ifp); ifma = if_findmulti(ifp, sa); if (ifma != NULL) { ifma->ifma_refcount++; if (retifma != NULL) *retifma = ifma; IF_ADDR_WUNLOCK(ifp); return (0); } /* * The address isn't already present; resolve the protocol address * into a link layer address, and then look that up, bump its * refcount or allocate an ifma for that also. * Most link layer resolving functions returns address data which * fits inside default sockaddr_dl structure. However callback * can allocate another sockaddr structure, in that case we need to * free it later. */ llsa = NULL; ll_ifma = NULL; if (ifp->if_resolvemulti != NULL) { /* Provide called function with buffer size information */ sdl.sdl_len = sizeof(sdl); llsa = (struct sockaddr *)&sdl; error = ifp->if_resolvemulti(ifp, &llsa, sa); if (error) goto unlock_out; } /* * Allocate the new address. Don't hook it up yet, as we may also * need to allocate a link layer multicast address. */ ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT); if (ifma == NULL) { error = ENOMEM; goto free_llsa_out; } /* * If a link layer address is found, we'll need to see if it's * already present in the address list, or allocate is as well. * When this block finishes, the link layer address will be on the * list. */ if (llsa != NULL) { ll_ifma = if_findmulti(ifp, llsa); if (ll_ifma == NULL) { ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT); if (ll_ifma == NULL) { --ifma->ifma_refcount; if_freemulti(ifma); error = ENOMEM; goto free_llsa_out; } CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma, ifma_link); } else ll_ifma->ifma_refcount++; ifma->ifma_llifma = ll_ifma; } /* * We now have a new multicast address, ifma, and possibly a new or * referenced link layer address. Add the primary address to the * ifnet address list. */ CK_STAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link); if (retifma != NULL) *retifma = ifma; /* * Must generate the message while holding the lock so that 'ifma' * pointer is still valid. */ rt_newmaddrmsg(RTM_NEWMADDR, ifma); IF_ADDR_WUNLOCK(ifp); /* * We are certain we have added something, so call down to the * interface to let them know about it. */ if (ifp->if_ioctl != NULL) { (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0); } if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl)) link_free_sdl(llsa); return (0); free_llsa_out: if ((llsa != NULL) && (llsa != (struct sockaddr *)&sdl)) link_free_sdl(llsa); unlock_out: IF_ADDR_WUNLOCK(ifp); return (error); } /* * Delete a multicast group membership by network-layer group address. * * Returns ENOENT if the entry could not be found. If ifp no longer * exists, results are undefined. This entry point should only be used * from subsystems which do appropriate locking to hold ifp for the * duration of the call. * Network-layer protocol domains must use if_delmulti_ifma(). */ int if_delmulti(struct ifnet *ifp, struct sockaddr *sa) { struct ifmultiaddr *ifma; int lastref; #ifdef INVARIANTS struct ifnet *oifp; IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(oifp, &V_ifnet, if_link) if (ifp == oifp) break; if (ifp != oifp) ifp = NULL; IFNET_RUNLOCK_NOSLEEP(); KASSERT(ifp != NULL, ("%s: ifnet went away", __func__)); #endif if (ifp == NULL) return (ENOENT); IF_ADDR_WLOCK(ifp); lastref = 0; ifma = if_findmulti(ifp, sa); if (ifma != NULL) lastref = if_delmulti_locked(ifp, ifma, 0); IF_ADDR_WUNLOCK(ifp); if (ifma == NULL) return (ENOENT); if (lastref && ifp->if_ioctl != NULL) { (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0); } return (0); } /* * Delete all multicast group membership for an interface. * Should be used to quickly flush all multicast filters. */ void if_delallmulti(struct ifnet *ifp) { struct ifmultiaddr *ifma; struct ifmultiaddr *next; IF_ADDR_WLOCK(ifp); CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) if_delmulti_locked(ifp, ifma, 0); IF_ADDR_WUNLOCK(ifp); } void if_delmulti_ifma(struct ifmultiaddr *ifma) { if_delmulti_ifma_flags(ifma, 0); } /* * Delete a multicast group membership by group membership pointer. * Network-layer protocol domains must use this routine. * * It is safe to call this routine if the ifp disappeared. */ void if_delmulti_ifma_flags(struct ifmultiaddr *ifma, int flags) { struct ifnet *ifp; int lastref; MCDPRINTF("%s freeing ifma: %p\n", __func__, ifma); #ifdef INET IN_MULTI_LIST_UNLOCK_ASSERT(); #endif ifp = ifma->ifma_ifp; #ifdef DIAGNOSTIC if (ifp == NULL) { printf("%s: ifma_ifp seems to be detached\n", __func__); } else { struct ifnet *oifp; IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(oifp, &V_ifnet, if_link) if (ifp == oifp) break; if (ifp != oifp) ifp = NULL; IFNET_RUNLOCK_NOSLEEP(); } #endif /* * If and only if the ifnet instance exists: Acquire the address lock. */ if (ifp != NULL) IF_ADDR_WLOCK(ifp); lastref = if_delmulti_locked(ifp, ifma, flags); if (ifp != NULL) { /* * If and only if the ifnet instance exists: * Release the address lock. * If the group was left: update the hardware hash filter. */ IF_ADDR_WUNLOCK(ifp); if (lastref && ifp->if_ioctl != NULL) { (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0); } } } /* * Perform deletion of network-layer and/or link-layer multicast address. * * Return 0 if the reference count was decremented. * Return 1 if the final reference was released, indicating that the * hardware hash filter should be reprogrammed. */ static int if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching) { struct ifmultiaddr *ll_ifma; if (ifp != NULL && ifma->ifma_ifp != NULL) { KASSERT(ifma->ifma_ifp == ifp, ("%s: inconsistent ifp %p", __func__, ifp)); IF_ADDR_WLOCK_ASSERT(ifp); } ifp = ifma->ifma_ifp; MCDPRINTF("%s freeing %p from %s \n", __func__, ifma, ifp ? ifp->if_xname : ""); /* * If the ifnet is detaching, null out references to ifnet, * so that upper protocol layers will notice, and not attempt * to obtain locks for an ifnet which no longer exists. The * routing socket announcement must happen before the ifnet * instance is detached from the system. */ if (detaching) { #ifdef DIAGNOSTIC printf("%s: detaching ifnet instance %p\n", __func__, ifp); #endif /* * ifp may already be nulled out if we are being reentered * to delete the ll_ifma. */ if (ifp != NULL) { rt_newmaddrmsg(RTM_DELMADDR, ifma); ifma->ifma_ifp = NULL; } } if (--ifma->ifma_refcount > 0) return 0; if (ifp != NULL && detaching == 0) CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); /* * If this ifma is a network-layer ifma, a link-layer ifma may * have been associated with it. Release it first if so. */ ll_ifma = ifma->ifma_llifma; if (ll_ifma != NULL) { KASSERT(ifma->ifma_lladdr != NULL, ("%s: llifma w/o lladdr", __func__)); if (detaching) ll_ifma->ifma_ifp = NULL; /* XXX */ if (--ll_ifma->ifma_refcount == 0) { if (ifp != NULL) { CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); } if_freemulti(ll_ifma); } } #ifdef INVARIANTS if (ifp) { struct ifmultiaddr *ifmatmp; CK_STAILQ_FOREACH(ifmatmp, &ifp->if_multiaddrs, ifma_link) MPASS(ifma != ifmatmp); } #endif if_freemulti(ifma); /* * The last reference to this instance of struct ifmultiaddr * was released; the hardware should be notified of this change. */ return 1; } /* * Set the link layer address on an interface. * * At this time we only support certain types of interfaces, * and we don't allow the length of the address to change. * * Set noinline to be dtrace-friendly */ __noinline int if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len) { struct sockaddr_dl *sdl; struct ifaddr *ifa; struct ifreq ifr; int rc; rc = 0; NET_EPOCH_ENTER(); ifa = ifp->if_addr; if (ifa == NULL) { rc = EINVAL; goto out; } sdl = (struct sockaddr_dl *)ifa->ifa_addr; if (sdl == NULL) { rc = EINVAL; goto out; } if (len != sdl->sdl_alen) { /* don't allow length to change */ rc = EINVAL; goto out; } switch (ifp->if_type) { case IFT_ETHER: case IFT_XETHER: case IFT_L2VLAN: case IFT_BRIDGE: case IFT_IEEE8023ADLAG: bcopy(lladdr, LLADDR(sdl), len); break; default: rc = ENODEV; goto out; } /* * If the interface is already up, we need * to re-init it in order to reprogram its * address filter. */ NET_EPOCH_EXIT(); if ((ifp->if_flags & IFF_UP) != 0) { if (ifp->if_ioctl) { ifp->if_flags &= ~IFF_UP; ifr.ifr_flags = ifp->if_flags & 0xffff; ifr.ifr_flagshigh = ifp->if_flags >> 16; (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); ifp->if_flags |= IFF_UP; ifr.ifr_flags = ifp->if_flags & 0xffff; ifr.ifr_flagshigh = ifp->if_flags >> 16; (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr); } } EVENTHANDLER_INVOKE(iflladdr_event, ifp); return (0); out: NET_EPOCH_EXIT(); return (rc); } /* * Compat function for handling basic encapsulation requests. * Not converted stacks (FDDI, IB, ..) supports traditional * output model: ARP (and other similar L2 protocols) are handled * inside output routine, arpresolve/nd6_resolve() returns MAC * address instead of full prepend. * * This function creates calculated header==MAC for IPv4/IPv6 and * returns EAFNOSUPPORT (which is then handled in ARP code) for other * address families. */ static int if_requestencap_default(struct ifnet *ifp, struct if_encap_req *req) { if (req->rtype != IFENCAP_LL) return (EOPNOTSUPP); if (req->bufsize < req->lladdr_len) return (ENOMEM); switch (req->family) { case AF_INET: case AF_INET6: break; default: return (EAFNOSUPPORT); } /* Copy lladdr to storage as is */ memmove(req->buf, req->lladdr, req->lladdr_len); req->bufsize = req->lladdr_len; req->lladdr_off = 0; return (0); } /* * Tunnel interfaces can nest, also they may cause infinite recursion * calls when misconfigured. We'll prevent this by detecting loops. * High nesting level may cause stack exhaustion. We'll prevent this * by introducing upper limit. * * Return 0, if tunnel nesting count is equal or less than limit. */ int if_tunnel_check_nesting(struct ifnet *ifp, struct mbuf *m, uint32_t cookie, int limit) { struct m_tag *mtag; int count; count = 1; mtag = NULL; while ((mtag = m_tag_locate(m, cookie, 0, mtag)) != NULL) { if (*(struct ifnet **)(mtag + 1) == ifp) { log(LOG_NOTICE, "%s: loop detected\n", if_name(ifp)); return (EIO); } count++; } if (count > limit) { log(LOG_NOTICE, "%s: if_output recursively called too many times(%d)\n", if_name(ifp), count); return (EIO); } mtag = m_tag_alloc(cookie, 0, sizeof(struct ifnet *), M_NOWAIT); if (mtag == NULL) return (ENOMEM); *(struct ifnet **)(mtag + 1) = ifp; m_tag_prepend(m, mtag); return (0); } /* * Get the link layer address that was read from the hardware at attach. * * This is only set by Ethernet NICs (IFT_ETHER), but laggX interfaces re-type * their component interfaces as IFT_IEEE8023ADLAG. */ int if_gethwaddr(struct ifnet *ifp, struct ifreq *ifr) { if (ifp->if_hw_addr == NULL) return (ENODEV); switch (ifp->if_type) { case IFT_ETHER: case IFT_IEEE8023ADLAG: bcopy(ifp->if_hw_addr, ifr->ifr_addr.sa_data, ifp->if_addrlen); return (0); default: return (ENODEV); } } /* * The name argument must be a pointer to storage which will last as * long as the interface does. For physical devices, the result of * device_get_name(dev) is a good choice and for pseudo-devices a * static string works well. */ void if_initname(struct ifnet *ifp, const char *name, int unit) { ifp->if_dname = name; ifp->if_dunit = unit; if (unit != IF_DUNIT_NONE) snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit); else strlcpy(ifp->if_xname, name, IFNAMSIZ); } int if_printf(struct ifnet *ifp, const char *fmt, ...) { char if_fmt[256]; va_list ap; snprintf(if_fmt, sizeof(if_fmt), "%s: %s", ifp->if_xname, fmt); va_start(ap, fmt); vlog(LOG_INFO, if_fmt, ap); va_end(ap); return (0); } void if_start(struct ifnet *ifp) { (*(ifp)->if_start)(ifp); } /* * Backwards compatibility interface for drivers * that have not implemented it */ static int if_transmit(struct ifnet *ifp, struct mbuf *m) { int error; IFQ_HANDOFF(ifp, m, error); return (error); } static void if_input_default(struct ifnet *ifp __unused, struct mbuf *m) { m_freem(m); } int if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust) { int active = 0; IF_LOCK(ifq); if (_IF_QFULL(ifq)) { IF_UNLOCK(ifq); if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); m_freem(m); return (0); } if (ifp != NULL) { if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len + adjust); if (m->m_flags & (M_BCAST|M_MCAST)) if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); active = ifp->if_drv_flags & IFF_DRV_OACTIVE; } _IF_ENQUEUE(ifq, m); IF_UNLOCK(ifq); if (ifp != NULL && !active) (*(ifp)->if_start)(ifp); return (1); } void if_register_com_alloc(u_char type, if_com_alloc_t *a, if_com_free_t *f) { KASSERT(if_com_alloc[type] == NULL, ("if_register_com_alloc: %d already registered", type)); KASSERT(if_com_free[type] == NULL, ("if_register_com_alloc: %d free already registered", type)); if_com_alloc[type] = a; if_com_free[type] = f; } void if_deregister_com_alloc(u_char type) { KASSERT(if_com_alloc[type] != NULL, ("if_deregister_com_alloc: %d not registered", type)); KASSERT(if_com_free[type] != NULL, ("if_deregister_com_alloc: %d free not registered", type)); if_com_alloc[type] = NULL; if_com_free[type] = NULL; } /* API for driver access to network stack owned ifnet.*/ uint64_t if_setbaudrate(struct ifnet *ifp, uint64_t baudrate) { uint64_t oldbrate; oldbrate = ifp->if_baudrate; ifp->if_baudrate = baudrate; return (oldbrate); } uint64_t if_getbaudrate(if_t ifp) { return (((struct ifnet *)ifp)->if_baudrate); } int if_setcapabilities(if_t ifp, int capabilities) { ((struct ifnet *)ifp)->if_capabilities = capabilities; return (0); } int if_setcapabilitiesbit(if_t ifp, int setbit, int clearbit) { ((struct ifnet *)ifp)->if_capabilities |= setbit; ((struct ifnet *)ifp)->if_capabilities &= ~clearbit; return (0); } int if_getcapabilities(if_t ifp) { return ((struct ifnet *)ifp)->if_capabilities; } int if_setcapenable(if_t ifp, int capabilities) { ((struct ifnet *)ifp)->if_capenable = capabilities; return (0); } int if_setcapenablebit(if_t ifp, int setcap, int clearcap) { if(setcap) ((struct ifnet *)ifp)->if_capenable |= setcap; if(clearcap) ((struct ifnet *)ifp)->if_capenable &= ~clearcap; return (0); } const char * if_getdname(if_t ifp) { return ((struct ifnet *)ifp)->if_dname; } int if_togglecapenable(if_t ifp, int togglecap) { ((struct ifnet *)ifp)->if_capenable ^= togglecap; return (0); } int if_getcapenable(if_t ifp) { return ((struct ifnet *)ifp)->if_capenable; } /* * This is largely undesirable because it ties ifnet to a device, but does * provide flexiblity for an embedded product vendor. Should be used with * the understanding that it violates the interface boundaries, and should be * a last resort only. */ int if_setdev(if_t ifp, void *dev) { return (0); } int if_setdrvflagbits(if_t ifp, int set_flags, int clear_flags) { ((struct ifnet *)ifp)->if_drv_flags |= set_flags; ((struct ifnet *)ifp)->if_drv_flags &= ~clear_flags; return (0); } int if_getdrvflags(if_t ifp) { return ((struct ifnet *)ifp)->if_drv_flags; } int if_setdrvflags(if_t ifp, int flags) { ((struct ifnet *)ifp)->if_drv_flags = flags; return (0); } int if_setflags(if_t ifp, int flags) { ((struct ifnet *)ifp)->if_flags = flags; return (0); } int if_setflagbits(if_t ifp, int set, int clear) { ((struct ifnet *)ifp)->if_flags |= set; ((struct ifnet *)ifp)->if_flags &= ~clear; return (0); } int if_getflags(if_t ifp) { return ((struct ifnet *)ifp)->if_flags; } int if_clearhwassist(if_t ifp) { ((struct ifnet *)ifp)->if_hwassist = 0; return (0); } int if_sethwassistbits(if_t ifp, int toset, int toclear) { ((struct ifnet *)ifp)->if_hwassist |= toset; ((struct ifnet *)ifp)->if_hwassist &= ~toclear; return (0); } int if_sethwassist(if_t ifp, int hwassist_bit) { ((struct ifnet *)ifp)->if_hwassist = hwassist_bit; return (0); } int if_gethwassist(if_t ifp) { return ((struct ifnet *)ifp)->if_hwassist; } int if_setmtu(if_t ifp, int mtu) { ((struct ifnet *)ifp)->if_mtu = mtu; return (0); } int if_getmtu(if_t ifp) { return ((struct ifnet *)ifp)->if_mtu; } int if_getmtu_family(if_t ifp, int family) { struct domain *dp; for (dp = domains; dp; dp = dp->dom_next) { if (dp->dom_family == family && dp->dom_ifmtu != NULL) return (dp->dom_ifmtu((struct ifnet *)ifp)); } return (((struct ifnet *)ifp)->if_mtu); } int if_setsoftc(if_t ifp, void *softc) { ((struct ifnet *)ifp)->if_softc = softc; return (0); } void * if_getsoftc(if_t ifp) { return ((struct ifnet *)ifp)->if_softc; } void if_setrcvif(struct mbuf *m, if_t ifp) { m->m_pkthdr.rcvif = (struct ifnet *)ifp; } void if_setvtag(struct mbuf *m, uint16_t tag) { m->m_pkthdr.ether_vtag = tag; } uint16_t if_getvtag(struct mbuf *m) { return (m->m_pkthdr.ether_vtag); } int if_sendq_empty(if_t ifp) { return IFQ_DRV_IS_EMPTY(&((struct ifnet *)ifp)->if_snd); } struct ifaddr * if_getifaddr(if_t ifp) { return ((struct ifnet *)ifp)->if_addr; } int if_getamcount(if_t ifp) { return ((struct ifnet *)ifp)->if_amcount; } int if_setsendqready(if_t ifp) { IFQ_SET_READY(&((struct ifnet *)ifp)->if_snd); return (0); } int if_setsendqlen(if_t ifp, int tx_desc_count) { IFQ_SET_MAXLEN(&((struct ifnet *)ifp)->if_snd, tx_desc_count); ((struct ifnet *)ifp)->if_snd.ifq_drv_maxlen = tx_desc_count; return (0); } int if_vlantrunkinuse(if_t ifp) { return ((struct ifnet *)ifp)->if_vlantrunk != NULL?1:0; } int if_input(if_t ifp, struct mbuf* sendmp) { (*((struct ifnet *)ifp)->if_input)((struct ifnet *)ifp, sendmp); return (0); } /* XXX */ #ifndef ETH_ADDR_LEN #define ETH_ADDR_LEN 6 #endif int if_setupmultiaddr(if_t ifp, void *mta, int *cnt, int max) { struct ifmultiaddr *ifma; uint8_t *lmta = (uint8_t *)mta; int mcnt = 0; CK_STAILQ_FOREACH(ifma, &((struct ifnet *)ifp)->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; if (mcnt == max) break; bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), &lmta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN); mcnt++; } *cnt = mcnt; return (0); } int if_multiaddr_array(if_t ifp, void *mta, int *cnt, int max) { int error; if_maddr_rlock(ifp); error = if_setupmultiaddr(ifp, mta, cnt, max); if_maddr_runlock(ifp); return (error); } int if_multiaddr_count(if_t ifp, int max) { struct ifmultiaddr *ifma; int count; count = 0; if_maddr_rlock(ifp); CK_STAILQ_FOREACH(ifma, &((struct ifnet *)ifp)->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; count++; if (count == max) break; } if_maddr_runlock(ifp); return (count); } int if_multi_apply(struct ifnet *ifp, int (*filter)(void *, struct ifmultiaddr *, int), void *arg) { struct ifmultiaddr *ifma; int cnt = 0; if_maddr_rlock(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) cnt += filter(arg, ifma, cnt); if_maddr_runlock(ifp); return (cnt); } struct mbuf * if_dequeue(if_t ifp) { struct mbuf *m; IFQ_DRV_DEQUEUE(&((struct ifnet *)ifp)->if_snd, m); return (m); } int if_sendq_prepend(if_t ifp, struct mbuf *m) { IFQ_DRV_PREPEND(&((struct ifnet *)ifp)->if_snd, m); return (0); } int if_setifheaderlen(if_t ifp, int len) { ((struct ifnet *)ifp)->if_hdrlen = len; return (0); } caddr_t if_getlladdr(if_t ifp) { return (IF_LLADDR((struct ifnet *)ifp)); } void * if_gethandle(u_char type) { return (if_alloc(type)); } void if_bpfmtap(if_t ifh, struct mbuf *m) { struct ifnet *ifp = (struct ifnet *)ifh; BPF_MTAP(ifp, m); } void if_etherbpfmtap(if_t ifh, struct mbuf *m) { struct ifnet *ifp = (struct ifnet *)ifh; ETHER_BPF_MTAP(ifp, m); } void if_vlancap(if_t ifh) { struct ifnet *ifp = (struct ifnet *)ifh; VLAN_CAPABILITIES(ifp); } int if_sethwtsomax(if_t ifp, u_int if_hw_tsomax) { ((struct ifnet *)ifp)->if_hw_tsomax = if_hw_tsomax; return (0); } int if_sethwtsomaxsegcount(if_t ifp, u_int if_hw_tsomaxsegcount) { ((struct ifnet *)ifp)->if_hw_tsomaxsegcount = if_hw_tsomaxsegcount; return (0); } int if_sethwtsomaxsegsize(if_t ifp, u_int if_hw_tsomaxsegsize) { ((struct ifnet *)ifp)->if_hw_tsomaxsegsize = if_hw_tsomaxsegsize; return (0); } u_int if_gethwtsomax(if_t ifp) { return (((struct ifnet *)ifp)->if_hw_tsomax); } u_int if_gethwtsomaxsegcount(if_t ifp) { return (((struct ifnet *)ifp)->if_hw_tsomaxsegcount); } u_int if_gethwtsomaxsegsize(if_t ifp) { return (((struct ifnet *)ifp)->if_hw_tsomaxsegsize); } void if_setinitfn(if_t ifp, void (*init_fn)(void *)) { ((struct ifnet *)ifp)->if_init = init_fn; } void if_setioctlfn(if_t ifp, int (*ioctl_fn)(if_t, u_long, caddr_t)) { ((struct ifnet *)ifp)->if_ioctl = (void *)ioctl_fn; } void if_setstartfn(if_t ifp, void (*start_fn)(if_t)) { ((struct ifnet *)ifp)->if_start = (void *)start_fn; } void if_settransmitfn(if_t ifp, if_transmit_fn_t start_fn) { ((struct ifnet *)ifp)->if_transmit = start_fn; } void if_setqflushfn(if_t ifp, if_qflush_fn_t flush_fn) { ((struct ifnet *)ifp)->if_qflush = flush_fn; } void if_setgetcounterfn(if_t ifp, if_get_counter_t fn) { ifp->if_get_counter = fn; } /* Revisit these - These are inline functions originally. */ int drbr_inuse_drv(if_t ifh, struct buf_ring *br) { return drbr_inuse(ifh, br); } struct mbuf* drbr_dequeue_drv(if_t ifh, struct buf_ring *br) { return drbr_dequeue(ifh, br); } int drbr_needs_enqueue_drv(if_t ifh, struct buf_ring *br) { return drbr_needs_enqueue(ifh, br); } int drbr_enqueue_drv(if_t ifh, struct buf_ring *br, struct mbuf *m) { return drbr_enqueue(ifh, br, m); } Index: head/sys/net/if_bridge.c =================================================================== --- head/sys/net/if_bridge.c (revision 336675) +++ head/sys/net/if_bridge.c (revision 336676) @@ -1,3610 +1,3610 @@ /* $NetBSD: if_bridge.c,v 1.31 2005/06/01 19:45:34 jdc Exp $ */ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright 2001 Wasabi Systems, Inc. * All rights reserved. * * Written by Jason R. Thorpe for Wasabi Systems, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed for the NetBSD Project by * Wasabi Systems, Inc. * 4. The name of Wasabi Systems, Inc. may not be used to endorse * or promote products derived from this software without specific prior * written permission. * * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ /* * Copyright (c) 1999, 2000 Jason L. Wright (jason@thought.net) * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * OpenBSD: if_bridge.c,v 1.60 2001/06/15 03:38:33 itojun Exp */ /* * Network interface bridge support. * * TODO: * * - Currently only supports Ethernet-like interfaces (Ethernet, * 802.11, VLANs on Ethernet, etc.) Figure out a nice way * to bridge other types of interfaces (maybe consider * heterogeneous bridges). */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include /* for net/if.h */ #include #include /* string functions */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #endif #if defined(INET) || defined(INET6) #include #endif #include #include #include #include #include #include #include /* * Size of the route hash table. Must be a power of two. */ #ifndef BRIDGE_RTHASH_SIZE #define BRIDGE_RTHASH_SIZE 1024 #endif #define BRIDGE_RTHASH_MASK (BRIDGE_RTHASH_SIZE - 1) /* * Default maximum number of addresses to cache. */ #ifndef BRIDGE_RTABLE_MAX #define BRIDGE_RTABLE_MAX 2000 #endif /* * Timeout (in seconds) for entries learned dynamically. */ #ifndef BRIDGE_RTABLE_TIMEOUT #define BRIDGE_RTABLE_TIMEOUT (20 * 60) /* same as ARP */ #endif /* * Number of seconds between walks of the route list. */ #ifndef BRIDGE_RTABLE_PRUNE_PERIOD #define BRIDGE_RTABLE_PRUNE_PERIOD (5 * 60) #endif /* * List of capabilities to possibly mask on the member interface. */ #define BRIDGE_IFCAPS_MASK (IFCAP_TOE|IFCAP_TSO|IFCAP_TXCSUM|\ IFCAP_TXCSUM_IPV6) /* * List of capabilities to strip */ #define BRIDGE_IFCAPS_STRIP IFCAP_LRO /* * Bridge interface list entry. */ struct bridge_iflist { LIST_ENTRY(bridge_iflist) bif_next; struct ifnet *bif_ifp; /* member if */ struct bstp_port bif_stp; /* STP state */ uint32_t bif_flags; /* member if flags */ int bif_savedcaps; /* saved capabilities */ uint32_t bif_addrmax; /* max # of addresses */ uint32_t bif_addrcnt; /* cur. # of addresses */ uint32_t bif_addrexceeded;/* # of address violations */ }; /* * Bridge route node. */ struct bridge_rtnode { LIST_ENTRY(bridge_rtnode) brt_hash; /* hash table linkage */ LIST_ENTRY(bridge_rtnode) brt_list; /* list linkage */ struct bridge_iflist *brt_dst; /* destination if */ unsigned long brt_expire; /* expiration time */ uint8_t brt_flags; /* address flags */ uint8_t brt_addr[ETHER_ADDR_LEN]; uint16_t brt_vlan; /* vlan id */ }; #define brt_ifp brt_dst->bif_ifp /* * Software state for each bridge. */ struct bridge_softc { struct ifnet *sc_ifp; /* make this an interface */ LIST_ENTRY(bridge_softc) sc_list; struct mtx sc_mtx; struct cv sc_cv; uint32_t sc_brtmax; /* max # of addresses */ uint32_t sc_brtcnt; /* cur. # of addresses */ uint32_t sc_brttimeout; /* rt timeout in seconds */ struct callout sc_brcallout; /* bridge callout */ uint32_t sc_iflist_ref; /* refcount for sc_iflist */ uint32_t sc_iflist_xcnt; /* refcount for sc_iflist */ LIST_HEAD(, bridge_iflist) sc_iflist; /* member interface list */ LIST_HEAD(, bridge_rtnode) *sc_rthash; /* our forwarding table */ LIST_HEAD(, bridge_rtnode) sc_rtlist; /* list version of above */ uint32_t sc_rthash_key; /* key for hash */ LIST_HEAD(, bridge_iflist) sc_spanlist; /* span ports list */ struct bstp_state sc_stp; /* STP state */ uint32_t sc_brtexceeded; /* # of cache drops */ struct ifnet *sc_ifaddr; /* member mac copied from */ u_char sc_defaddr[6]; /* Default MAC address */ }; -static VNET_DEFINE(struct mtx, bridge_list_mtx); +VNET_DEFINE_STATIC(struct mtx, bridge_list_mtx); #define V_bridge_list_mtx VNET(bridge_list_mtx) static eventhandler_tag bridge_detach_cookie; int bridge_rtable_prune_period = BRIDGE_RTABLE_PRUNE_PERIOD; uma_zone_t bridge_rtnode_zone; static int bridge_clone_create(struct if_clone *, int, caddr_t); static void bridge_clone_destroy(struct ifnet *); static int bridge_ioctl(struct ifnet *, u_long, caddr_t); static void bridge_mutecaps(struct bridge_softc *); static void bridge_set_ifcap(struct bridge_softc *, struct bridge_iflist *, int); static void bridge_ifdetach(void *arg __unused, struct ifnet *); static void bridge_init(void *); static void bridge_dummynet(struct mbuf *, struct ifnet *); static void bridge_stop(struct ifnet *, int); static int bridge_transmit(struct ifnet *, struct mbuf *); static void bridge_qflush(struct ifnet *); static struct mbuf *bridge_input(struct ifnet *, struct mbuf *); static int bridge_output(struct ifnet *, struct mbuf *, struct sockaddr *, struct rtentry *); static int bridge_enqueue(struct bridge_softc *, struct ifnet *, struct mbuf *); static void bridge_rtdelete(struct bridge_softc *, struct ifnet *ifp, int); static void bridge_forward(struct bridge_softc *, struct bridge_iflist *, struct mbuf *m); static void bridge_timer(void *); static void bridge_broadcast(struct bridge_softc *, struct ifnet *, struct mbuf *, int); static void bridge_span(struct bridge_softc *, struct mbuf *); static int bridge_rtupdate(struct bridge_softc *, const uint8_t *, uint16_t, struct bridge_iflist *, int, uint8_t); static struct ifnet *bridge_rtlookup(struct bridge_softc *, const uint8_t *, uint16_t); static void bridge_rttrim(struct bridge_softc *); static void bridge_rtage(struct bridge_softc *); static void bridge_rtflush(struct bridge_softc *, int); static int bridge_rtdaddr(struct bridge_softc *, const uint8_t *, uint16_t); static void bridge_rtable_init(struct bridge_softc *); static void bridge_rtable_fini(struct bridge_softc *); static int bridge_rtnode_addr_cmp(const uint8_t *, const uint8_t *); static struct bridge_rtnode *bridge_rtnode_lookup(struct bridge_softc *, const uint8_t *, uint16_t); static int bridge_rtnode_insert(struct bridge_softc *, struct bridge_rtnode *); static void bridge_rtnode_destroy(struct bridge_softc *, struct bridge_rtnode *); static void bridge_rtable_expire(struct ifnet *, int); static void bridge_state_change(struct ifnet *, int); static struct bridge_iflist *bridge_lookup_member(struct bridge_softc *, const char *name); static struct bridge_iflist *bridge_lookup_member_if(struct bridge_softc *, struct ifnet *ifp); static void bridge_delete_member(struct bridge_softc *, struct bridge_iflist *, int); static void bridge_delete_span(struct bridge_softc *, struct bridge_iflist *); static int bridge_ioctl_add(struct bridge_softc *, void *); static int bridge_ioctl_del(struct bridge_softc *, void *); static int bridge_ioctl_gifflags(struct bridge_softc *, void *); static int bridge_ioctl_sifflags(struct bridge_softc *, void *); static int bridge_ioctl_scache(struct bridge_softc *, void *); static int bridge_ioctl_gcache(struct bridge_softc *, void *); static int bridge_ioctl_gifs(struct bridge_softc *, void *); static int bridge_ioctl_rts(struct bridge_softc *, void *); static int bridge_ioctl_saddr(struct bridge_softc *, void *); static int bridge_ioctl_sto(struct bridge_softc *, void *); static int bridge_ioctl_gto(struct bridge_softc *, void *); static int bridge_ioctl_daddr(struct bridge_softc *, void *); static int bridge_ioctl_flush(struct bridge_softc *, void *); static int bridge_ioctl_gpri(struct bridge_softc *, void *); static int bridge_ioctl_spri(struct bridge_softc *, void *); static int bridge_ioctl_ght(struct bridge_softc *, void *); static int bridge_ioctl_sht(struct bridge_softc *, void *); static int bridge_ioctl_gfd(struct bridge_softc *, void *); static int bridge_ioctl_sfd(struct bridge_softc *, void *); static int bridge_ioctl_gma(struct bridge_softc *, void *); static int bridge_ioctl_sma(struct bridge_softc *, void *); static int bridge_ioctl_sifprio(struct bridge_softc *, void *); static int bridge_ioctl_sifcost(struct bridge_softc *, void *); static int bridge_ioctl_sifmaxaddr(struct bridge_softc *, void *); static int bridge_ioctl_addspan(struct bridge_softc *, void *); static int bridge_ioctl_delspan(struct bridge_softc *, void *); static int bridge_ioctl_gbparam(struct bridge_softc *, void *); static int bridge_ioctl_grte(struct bridge_softc *, void *); static int bridge_ioctl_gifsstp(struct bridge_softc *, void *); static int bridge_ioctl_sproto(struct bridge_softc *, void *); static int bridge_ioctl_stxhc(struct bridge_softc *, void *); static int bridge_pfil(struct mbuf **, struct ifnet *, struct ifnet *, int); static int bridge_ip_checkbasic(struct mbuf **mp); #ifdef INET6 static int bridge_ip6_checkbasic(struct mbuf **mp); #endif /* INET6 */ static int bridge_fragment(struct ifnet *, struct mbuf **mp, struct ether_header *, int, struct llc *); static void bridge_linkstate(struct ifnet *ifp); static void bridge_linkcheck(struct bridge_softc *sc); /* The default bridge vlan is 1 (IEEE 802.1Q-2003 Table 9-2) */ #define VLANTAGOF(_m) \ (_m->m_flags & M_VLANTAG) ? EVL_VLANOFTAG(_m->m_pkthdr.ether_vtag) : 1 static struct bstp_cb_ops bridge_ops = { .bcb_state = bridge_state_change, .bcb_rtage = bridge_rtable_expire }; SYSCTL_DECL(_net_link); static SYSCTL_NODE(_net_link, IFT_BRIDGE, bridge, CTLFLAG_RW, 0, "Bridge"); /* only pass IP[46] packets when pfil is enabled */ -static VNET_DEFINE(int, pfil_onlyip) = 1; +VNET_DEFINE_STATIC(int, pfil_onlyip) = 1; #define V_pfil_onlyip VNET(pfil_onlyip) SYSCTL_INT(_net_link_bridge, OID_AUTO, pfil_onlyip, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(pfil_onlyip), 0, "Only pass IP packets when pfil is enabled"); /* run pfil hooks on the bridge interface */ -static VNET_DEFINE(int, pfil_bridge) = 1; +VNET_DEFINE_STATIC(int, pfil_bridge) = 1; #define V_pfil_bridge VNET(pfil_bridge) SYSCTL_INT(_net_link_bridge, OID_AUTO, pfil_bridge, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(pfil_bridge), 0, "Packet filter on the bridge interface"); /* layer2 filter with ipfw */ -static VNET_DEFINE(int, pfil_ipfw); +VNET_DEFINE_STATIC(int, pfil_ipfw); #define V_pfil_ipfw VNET(pfil_ipfw) /* layer2 ARP filter with ipfw */ -static VNET_DEFINE(int, pfil_ipfw_arp); +VNET_DEFINE_STATIC(int, pfil_ipfw_arp); #define V_pfil_ipfw_arp VNET(pfil_ipfw_arp) SYSCTL_INT(_net_link_bridge, OID_AUTO, ipfw_arp, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(pfil_ipfw_arp), 0, "Filter ARP packets through IPFW layer2"); /* run pfil hooks on the member interface */ -static VNET_DEFINE(int, pfil_member) = 1; +VNET_DEFINE_STATIC(int, pfil_member) = 1; #define V_pfil_member VNET(pfil_member) SYSCTL_INT(_net_link_bridge, OID_AUTO, pfil_member, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(pfil_member), 0, "Packet filter on the member interface"); /* run pfil hooks on the physical interface for locally destined packets */ -static VNET_DEFINE(int, pfil_local_phys); +VNET_DEFINE_STATIC(int, pfil_local_phys); #define V_pfil_local_phys VNET(pfil_local_phys) SYSCTL_INT(_net_link_bridge, OID_AUTO, pfil_local_phys, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(pfil_local_phys), 0, "Packet filter on the physical interface for locally destined packets"); /* log STP state changes */ -static VNET_DEFINE(int, log_stp); +VNET_DEFINE_STATIC(int, log_stp); #define V_log_stp VNET(log_stp) SYSCTL_INT(_net_link_bridge, OID_AUTO, log_stp, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(log_stp), 0, "Log STP state changes"); /* share MAC with first bridge member */ -static VNET_DEFINE(int, bridge_inherit_mac); +VNET_DEFINE_STATIC(int, bridge_inherit_mac); #define V_bridge_inherit_mac VNET(bridge_inherit_mac) SYSCTL_INT(_net_link_bridge, OID_AUTO, inherit_mac, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(bridge_inherit_mac), 0, "Inherit MAC address from the first bridge member"); -static VNET_DEFINE(int, allow_llz_overlap) = 0; +VNET_DEFINE_STATIC(int, allow_llz_overlap) = 0; #define V_allow_llz_overlap VNET(allow_llz_overlap) SYSCTL_INT(_net_link_bridge, OID_AUTO, allow_llz_overlap, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(allow_llz_overlap), 0, "Allow overlap of link-local scope " "zones of a bridge interface and the member interfaces"); struct bridge_control { int (*bc_func)(struct bridge_softc *, void *); int bc_argsize; int bc_flags; }; #define BC_F_COPYIN 0x01 /* copy arguments in */ #define BC_F_COPYOUT 0x02 /* copy arguments out */ #define BC_F_SUSER 0x04 /* do super-user check */ const struct bridge_control bridge_control_table[] = { { bridge_ioctl_add, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_del, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gifflags, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_COPYOUT }, { bridge_ioctl_sifflags, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_scache, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gcache, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_gifs, sizeof(struct ifbifconf), BC_F_COPYIN|BC_F_COPYOUT }, { bridge_ioctl_rts, sizeof(struct ifbaconf), BC_F_COPYIN|BC_F_COPYOUT }, { bridge_ioctl_saddr, sizeof(struct ifbareq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_sto, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gto, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_daddr, sizeof(struct ifbareq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_flush, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gpri, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_spri, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_ght, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_sht, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gfd, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_sfd, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gma, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_sma, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_sifprio, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_sifcost, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_addspan, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_delspan, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_gbparam, sizeof(struct ifbropreq), BC_F_COPYOUT }, { bridge_ioctl_grte, sizeof(struct ifbrparam), BC_F_COPYOUT }, { bridge_ioctl_gifsstp, sizeof(struct ifbpstpconf), BC_F_COPYIN|BC_F_COPYOUT }, { bridge_ioctl_sproto, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_stxhc, sizeof(struct ifbrparam), BC_F_COPYIN|BC_F_SUSER }, { bridge_ioctl_sifmaxaddr, sizeof(struct ifbreq), BC_F_COPYIN|BC_F_SUSER }, }; const int bridge_control_table_size = nitems(bridge_control_table); -static VNET_DEFINE(LIST_HEAD(, bridge_softc), bridge_list); +VNET_DEFINE_STATIC(LIST_HEAD(, bridge_softc), bridge_list); #define V_bridge_list VNET(bridge_list) #define BRIDGE_LIST_LOCK_INIT(x) mtx_init(&V_bridge_list_mtx, \ "if_bridge list", NULL, MTX_DEF) #define BRIDGE_LIST_LOCK_DESTROY(x) mtx_destroy(&V_bridge_list_mtx) #define BRIDGE_LIST_LOCK(x) mtx_lock(&V_bridge_list_mtx) #define BRIDGE_LIST_UNLOCK(x) mtx_unlock(&V_bridge_list_mtx) -static VNET_DEFINE(struct if_clone *, bridge_cloner); +VNET_DEFINE_STATIC(struct if_clone *, bridge_cloner); #define V_bridge_cloner VNET(bridge_cloner) static const char bridge_name[] = "bridge"; static void vnet_bridge_init(const void *unused __unused) { BRIDGE_LIST_LOCK_INIT(); LIST_INIT(&V_bridge_list); V_bridge_cloner = if_clone_simple(bridge_name, bridge_clone_create, bridge_clone_destroy, 0); } VNET_SYSINIT(vnet_bridge_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_bridge_init, NULL); static void vnet_bridge_uninit(const void *unused __unused) { if_clone_detach(V_bridge_cloner); V_bridge_cloner = NULL; BRIDGE_LIST_LOCK_DESTROY(); } VNET_SYSUNINIT(vnet_bridge_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_bridge_uninit, NULL); static int bridge_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: bridge_rtnode_zone = uma_zcreate("bridge_rtnode", sizeof(struct bridge_rtnode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); bridge_dn_p = bridge_dummynet; bridge_detach_cookie = EVENTHANDLER_REGISTER( ifnet_departure_event, bridge_ifdetach, NULL, EVENTHANDLER_PRI_ANY); break; case MOD_UNLOAD: EVENTHANDLER_DEREGISTER(ifnet_departure_event, bridge_detach_cookie); uma_zdestroy(bridge_rtnode_zone); bridge_dn_p = NULL; break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t bridge_mod = { "if_bridge", bridge_modevent, 0 }; DECLARE_MODULE(if_bridge, bridge_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_bridge, 1); MODULE_DEPEND(if_bridge, bridgestp, 1, 1, 1); /* * handler for net.link.bridge.ipfw */ static int sysctl_pfil_ipfw(SYSCTL_HANDLER_ARGS) { int enable = V_pfil_ipfw; int error; error = sysctl_handle_int(oidp, &enable, 0, req); enable &= 1; if (enable != V_pfil_ipfw) { V_pfil_ipfw = enable; /* * Disable pfil so that ipfw doesnt run twice, if the user * really wants both then they can re-enable pfil_bridge and/or * pfil_member. Also allow non-ip packets as ipfw can filter by * layer2 type. */ if (V_pfil_ipfw) { V_pfil_onlyip = 0; V_pfil_bridge = 0; V_pfil_member = 0; } } return (error); } SYSCTL_PROC(_net_link_bridge, OID_AUTO, ipfw, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(pfil_ipfw), 0, &sysctl_pfil_ipfw, "I", "Layer2 filter with IPFW"); /* * bridge_clone_create: * * Create a new bridge instance. */ static int bridge_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct bridge_softc *sc, *sc2; struct ifnet *bifp, *ifp; int fb, retry; unsigned long hostid; sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { free(sc, M_DEVBUF); return (ENOSPC); } BRIDGE_LOCK_INIT(sc); sc->sc_brtmax = BRIDGE_RTABLE_MAX; sc->sc_brttimeout = BRIDGE_RTABLE_TIMEOUT; /* Initialize our routing table. */ bridge_rtable_init(sc); callout_init_mtx(&sc->sc_brcallout, &sc->sc_mtx, 0); LIST_INIT(&sc->sc_iflist); LIST_INIT(&sc->sc_spanlist); ifp->if_softc = sc; if_initname(ifp, bridge_name, unit); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = bridge_ioctl; ifp->if_transmit = bridge_transmit; ifp->if_qflush = bridge_qflush; ifp->if_init = bridge_init; ifp->if_type = IFT_BRIDGE; /* * Generate an ethernet address with a locally administered address. * * Since we are using random ethernet addresses for the bridge, it is * possible that we might have address collisions, so make sure that * this hardware address isn't already in use on another bridge. * The first try uses the hostid and falls back to arc4rand(). */ fb = 0; getcredhostid(curthread->td_ucred, &hostid); do { if (fb || hostid == 0) { arc4rand(sc->sc_defaddr, ETHER_ADDR_LEN, 1); sc->sc_defaddr[0] &= ~1;/* clear multicast bit */ sc->sc_defaddr[0] |= 2; /* set the LAA bit */ } else { sc->sc_defaddr[0] = 0x2; sc->sc_defaddr[1] = (hostid >> 24) & 0xff; sc->sc_defaddr[2] = (hostid >> 16) & 0xff; sc->sc_defaddr[3] = (hostid >> 8 ) & 0xff; sc->sc_defaddr[4] = hostid & 0xff; sc->sc_defaddr[5] = ifp->if_dunit & 0xff; } fb = 1; retry = 0; BRIDGE_LIST_LOCK(); LIST_FOREACH(sc2, &V_bridge_list, sc_list) { bifp = sc2->sc_ifp; if (memcmp(sc->sc_defaddr, IF_LLADDR(bifp), ETHER_ADDR_LEN) == 0) { retry = 1; break; } } BRIDGE_LIST_UNLOCK(); } while (retry == 1); bstp_attach(&sc->sc_stp, &bridge_ops); ether_ifattach(ifp, sc->sc_defaddr); /* Now undo some of the damage... */ ifp->if_baudrate = 0; ifp->if_type = IFT_BRIDGE; BRIDGE_LIST_LOCK(); LIST_INSERT_HEAD(&V_bridge_list, sc, sc_list); BRIDGE_LIST_UNLOCK(); return (0); } /* * bridge_clone_destroy: * * Destroy a bridge instance. */ static void bridge_clone_destroy(struct ifnet *ifp) { struct bridge_softc *sc = ifp->if_softc; struct bridge_iflist *bif; BRIDGE_LOCK(sc); bridge_stop(ifp, 1); ifp->if_flags &= ~IFF_UP; while ((bif = LIST_FIRST(&sc->sc_iflist)) != NULL) bridge_delete_member(sc, bif, 0); while ((bif = LIST_FIRST(&sc->sc_spanlist)) != NULL) { bridge_delete_span(sc, bif); } BRIDGE_UNLOCK(sc); callout_drain(&sc->sc_brcallout); BRIDGE_LIST_LOCK(); LIST_REMOVE(sc, sc_list); BRIDGE_LIST_UNLOCK(); bstp_detach(&sc->sc_stp); ether_ifdetach(ifp); if_free(ifp); /* Tear down the routing table. */ bridge_rtable_fini(sc); BRIDGE_LOCK_DESTROY(sc); free(sc, M_DEVBUF); } /* * bridge_ioctl: * * Handle a control request from the operator. */ static int bridge_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct bridge_softc *sc = ifp->if_softc; struct ifreq *ifr = (struct ifreq *)data; struct bridge_iflist *bif; struct thread *td = curthread; union { struct ifbreq ifbreq; struct ifbifconf ifbifconf; struct ifbareq ifbareq; struct ifbaconf ifbaconf; struct ifbrparam ifbrparam; struct ifbropreq ifbropreq; } args; struct ifdrv *ifd = (struct ifdrv *) data; const struct bridge_control *bc; int error = 0; switch (cmd) { case SIOCADDMULTI: case SIOCDELMULTI: break; case SIOCGDRVSPEC: case SIOCSDRVSPEC: if (ifd->ifd_cmd >= bridge_control_table_size) { error = EINVAL; break; } bc = &bridge_control_table[ifd->ifd_cmd]; if (cmd == SIOCGDRVSPEC && (bc->bc_flags & BC_F_COPYOUT) == 0) { error = EINVAL; break; } else if (cmd == SIOCSDRVSPEC && (bc->bc_flags & BC_F_COPYOUT) != 0) { error = EINVAL; break; } if (bc->bc_flags & BC_F_SUSER) { error = priv_check(td, PRIV_NET_BRIDGE); if (error) break; } if (ifd->ifd_len != bc->bc_argsize || ifd->ifd_len > sizeof(args)) { error = EINVAL; break; } bzero(&args, sizeof(args)); if (bc->bc_flags & BC_F_COPYIN) { error = copyin(ifd->ifd_data, &args, ifd->ifd_len); if (error) break; } BRIDGE_LOCK(sc); error = (*bc->bc_func)(sc, &args); BRIDGE_UNLOCK(sc); if (error) break; if (bc->bc_flags & BC_F_COPYOUT) error = copyout(&args, ifd->ifd_data, ifd->ifd_len); break; case SIOCSIFFLAGS: if (!(ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked down and it is running, * then stop and disable it. */ BRIDGE_LOCK(sc); bridge_stop(ifp, 1); BRIDGE_UNLOCK(sc); } else if ((ifp->if_flags & IFF_UP) && !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked up and it is stopped, then * start it. */ (*ifp->if_init)(sc); } break; case SIOCSIFMTU: if (ifr->ifr_mtu < 576) { error = EINVAL; break; } if (LIST_EMPTY(&sc->sc_iflist)) { sc->sc_ifp->if_mtu = ifr->ifr_mtu; break; } BRIDGE_LOCK(sc); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (bif->bif_ifp->if_mtu != ifr->ifr_mtu) { log(LOG_NOTICE, "%s: invalid MTU: %u(%s)" " != %d\n", sc->sc_ifp->if_xname, bif->bif_ifp->if_mtu, bif->bif_ifp->if_xname, ifr->ifr_mtu); error = EINVAL; break; } } if (!error) sc->sc_ifp->if_mtu = ifr->ifr_mtu; BRIDGE_UNLOCK(sc); break; default: /* * drop the lock as ether_ioctl() will call bridge_start() and * cause the lock to be recursed. */ error = ether_ioctl(ifp, cmd, data); break; } return (error); } /* * bridge_mutecaps: * * Clear or restore unwanted capabilities on the member interface */ static void bridge_mutecaps(struct bridge_softc *sc) { struct bridge_iflist *bif; int enabled, mask; /* Initial bitmask of capabilities to test */ mask = BRIDGE_IFCAPS_MASK; LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { /* Every member must support it or its disabled */ mask &= bif->bif_savedcaps; } BRIDGE_XLOCK(sc); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { enabled = bif->bif_ifp->if_capenable; enabled &= ~BRIDGE_IFCAPS_STRIP; /* strip off mask bits and enable them again if allowed */ enabled &= ~BRIDGE_IFCAPS_MASK; enabled |= mask; BRIDGE_UNLOCK(sc); bridge_set_ifcap(sc, bif, enabled); BRIDGE_LOCK(sc); } BRIDGE_XDROP(sc); } static void bridge_set_ifcap(struct bridge_softc *sc, struct bridge_iflist *bif, int set) { struct ifnet *ifp = bif->bif_ifp; struct ifreq ifr; int error; BRIDGE_UNLOCK_ASSERT(sc); bzero(&ifr, sizeof(ifr)); ifr.ifr_reqcap = set; if (ifp->if_capenable != set) { error = (*ifp->if_ioctl)(ifp, SIOCSIFCAP, (caddr_t)&ifr); if (error) if_printf(sc->sc_ifp, "error setting capabilities on %s: %d\n", ifp->if_xname, error); if ((ifp->if_capenable & ~set) != 0) if_printf(sc->sc_ifp, "can't disable some capabilities on %s: 0x%x\n", ifp->if_xname, ifp->if_capenable & ~set); } } /* * bridge_lookup_member: * * Lookup a bridge member interface. */ static struct bridge_iflist * bridge_lookup_member(struct bridge_softc *sc, const char *name) { struct bridge_iflist *bif; struct ifnet *ifp; BRIDGE_LOCK_ASSERT(sc); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { ifp = bif->bif_ifp; if (strcmp(ifp->if_xname, name) == 0) return (bif); } return (NULL); } /* * bridge_lookup_member_if: * * Lookup a bridge member interface by ifnet*. */ static struct bridge_iflist * bridge_lookup_member_if(struct bridge_softc *sc, struct ifnet *member_ifp) { struct bridge_iflist *bif; BRIDGE_LOCK_ASSERT(sc); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (bif->bif_ifp == member_ifp) return (bif); } return (NULL); } /* * bridge_delete_member: * * Delete the specified member interface. */ static void bridge_delete_member(struct bridge_softc *sc, struct bridge_iflist *bif, int gone) { struct ifnet *ifs = bif->bif_ifp; struct ifnet *fif = NULL; BRIDGE_LOCK_ASSERT(sc); if (bif->bif_flags & IFBIF_STP) bstp_disable(&bif->bif_stp); ifs->if_bridge = NULL; BRIDGE_XLOCK(sc); LIST_REMOVE(bif, bif_next); BRIDGE_XDROP(sc); /* * If removing the interface that gave the bridge its mac address, set * the mac address of the bridge to the address of the next member, or * to its default address if no members are left. */ if (V_bridge_inherit_mac && sc->sc_ifaddr == ifs) { if (LIST_EMPTY(&sc->sc_iflist)) { bcopy(sc->sc_defaddr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); sc->sc_ifaddr = NULL; } else { fif = LIST_FIRST(&sc->sc_iflist)->bif_ifp; bcopy(IF_LLADDR(fif), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); sc->sc_ifaddr = fif; } EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } bridge_linkcheck(sc); bridge_mutecaps(sc); /* recalcuate now this interface is removed */ bridge_rtdelete(sc, ifs, IFBF_FLUSHALL); KASSERT(bif->bif_addrcnt == 0, ("%s: %d bridge routes referenced", __func__, bif->bif_addrcnt)); ifs->if_bridge_output = NULL; ifs->if_bridge_input = NULL; ifs->if_bridge_linkstate = NULL; BRIDGE_UNLOCK(sc); if (!gone) { switch (ifs->if_type) { case IFT_ETHER: case IFT_L2VLAN: /* * Take the interface out of promiscuous mode, but only * if it was promiscuous in the first place. It might * not be if we're in the bridge_ioctl_add() error path. */ if (ifs->if_flags & IFF_PROMISC) (void) ifpromisc(ifs, 0); break; case IFT_GIF: break; default: #ifdef DIAGNOSTIC panic("bridge_delete_member: impossible"); #endif break; } /* reneable any interface capabilities */ bridge_set_ifcap(sc, bif, bif->bif_savedcaps); } bstp_destroy(&bif->bif_stp); /* prepare to free */ BRIDGE_LOCK(sc); free(bif, M_DEVBUF); } /* * bridge_delete_span: * * Delete the specified span interface. */ static void bridge_delete_span(struct bridge_softc *sc, struct bridge_iflist *bif) { BRIDGE_LOCK_ASSERT(sc); KASSERT(bif->bif_ifp->if_bridge == NULL, ("%s: not a span interface", __func__)); LIST_REMOVE(bif, bif_next); free(bif, M_DEVBUF); } static int bridge_ioctl_add(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif = NULL; struct ifnet *ifs; int error = 0; ifs = ifunit(req->ifbr_ifsname); if (ifs == NULL) return (ENOENT); if (ifs->if_ioctl == NULL) /* must be supported */ return (EINVAL); /* If it's in the span list, it can't be a member. */ LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) if (ifs == bif->bif_ifp) return (EBUSY); if (ifs->if_bridge == sc) return (EEXIST); if (ifs->if_bridge != NULL) return (EBUSY); switch (ifs->if_type) { case IFT_ETHER: case IFT_L2VLAN: case IFT_GIF: /* permitted interface types */ break; default: return (EINVAL); } #ifdef INET6 /* * Two valid inet6 addresses with link-local scope must not be * on the parent interface and the member interfaces at the * same time. This restriction is needed to prevent violation * of link-local scope zone. Attempts to add a member * interface which has inet6 addresses when the parent has * inet6 triggers removal of all inet6 addresses on the member * interface. */ /* Check if the parent interface has a link-local scope addr. */ if (V_allow_llz_overlap == 0 && in6ifa_llaonifp(sc->sc_ifp) != NULL) { /* * If any, remove all inet6 addresses from the member * interfaces. */ BRIDGE_XLOCK(sc); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (in6ifa_llaonifp(bif->bif_ifp)) { BRIDGE_UNLOCK(sc); in6_ifdetach(bif->bif_ifp); BRIDGE_LOCK(sc); if_printf(sc->sc_ifp, "IPv6 addresses on %s have been removed " "before adding it as a member to prevent " "IPv6 address scope violation.\n", bif->bif_ifp->if_xname); } } BRIDGE_XDROP(sc); if (in6ifa_llaonifp(ifs)) { BRIDGE_UNLOCK(sc); in6_ifdetach(ifs); BRIDGE_LOCK(sc); if_printf(sc->sc_ifp, "IPv6 addresses on %s have been removed " "before adding it as a member to prevent " "IPv6 address scope violation.\n", ifs->if_xname); } } #endif /* Allow the first Ethernet member to define the MTU */ if (LIST_EMPTY(&sc->sc_iflist)) sc->sc_ifp->if_mtu = ifs->if_mtu; else if (sc->sc_ifp->if_mtu != ifs->if_mtu) { if_printf(sc->sc_ifp, "invalid MTU: %u(%s) != %u\n", ifs->if_mtu, ifs->if_xname, sc->sc_ifp->if_mtu); return (EINVAL); } bif = malloc(sizeof(*bif), M_DEVBUF, M_NOWAIT|M_ZERO); if (bif == NULL) return (ENOMEM); bif->bif_ifp = ifs; bif->bif_flags = IFBIF_LEARNING | IFBIF_DISCOVER; bif->bif_savedcaps = ifs->if_capenable; /* * Assign the interface's MAC address to the bridge if it's the first * member and the MAC address of the bridge has not been changed from * the default randomly generated one. */ if (V_bridge_inherit_mac && LIST_EMPTY(&sc->sc_iflist) && !memcmp(IF_LLADDR(sc->sc_ifp), sc->sc_defaddr, ETHER_ADDR_LEN)) { bcopy(IF_LLADDR(ifs), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); sc->sc_ifaddr = ifs; EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } ifs->if_bridge = sc; ifs->if_bridge_output = bridge_output; ifs->if_bridge_input = bridge_input; ifs->if_bridge_linkstate = bridge_linkstate; bstp_create(&sc->sc_stp, &bif->bif_stp, bif->bif_ifp); /* * XXX: XLOCK HERE!?! * * NOTE: insert_***HEAD*** should be safe for the traversals. */ LIST_INSERT_HEAD(&sc->sc_iflist, bif, bif_next); /* Set interface capabilities to the intersection set of all members */ bridge_mutecaps(sc); bridge_linkcheck(sc); /* Place the interface into promiscuous mode */ switch (ifs->if_type) { case IFT_ETHER: case IFT_L2VLAN: BRIDGE_UNLOCK(sc); error = ifpromisc(ifs, 1); BRIDGE_LOCK(sc); break; } if (error) bridge_delete_member(sc, bif, 0); return (error); } static int bridge_ioctl_del(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); bridge_delete_member(sc, bif, 0); return (0); } static int bridge_ioctl_gifflags(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; struct bstp_port *bp; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); bp = &bif->bif_stp; req->ifbr_ifsflags = bif->bif_flags; req->ifbr_state = bp->bp_state; req->ifbr_priority = bp->bp_priority; req->ifbr_path_cost = bp->bp_path_cost; req->ifbr_portno = bif->bif_ifp->if_index & 0xfff; req->ifbr_proto = bp->bp_protover; req->ifbr_role = bp->bp_role; req->ifbr_stpflags = bp->bp_flags; req->ifbr_addrcnt = bif->bif_addrcnt; req->ifbr_addrmax = bif->bif_addrmax; req->ifbr_addrexceeded = bif->bif_addrexceeded; /* Copy STP state options as flags */ if (bp->bp_operedge) req->ifbr_ifsflags |= IFBIF_BSTP_EDGE; if (bp->bp_flags & BSTP_PORT_AUTOEDGE) req->ifbr_ifsflags |= IFBIF_BSTP_AUTOEDGE; if (bp->bp_ptp_link) req->ifbr_ifsflags |= IFBIF_BSTP_PTP; if (bp->bp_flags & BSTP_PORT_AUTOPTP) req->ifbr_ifsflags |= IFBIF_BSTP_AUTOPTP; if (bp->bp_flags & BSTP_PORT_ADMEDGE) req->ifbr_ifsflags |= IFBIF_BSTP_ADMEDGE; if (bp->bp_flags & BSTP_PORT_ADMCOST) req->ifbr_ifsflags |= IFBIF_BSTP_ADMCOST; return (0); } static int bridge_ioctl_sifflags(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; struct bstp_port *bp; int error; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); bp = &bif->bif_stp; if (req->ifbr_ifsflags & IFBIF_SPAN) /* SPAN is readonly */ return (EINVAL); if (req->ifbr_ifsflags & IFBIF_STP) { if ((bif->bif_flags & IFBIF_STP) == 0) { error = bstp_enable(&bif->bif_stp); if (error) return (error); } } else { if ((bif->bif_flags & IFBIF_STP) != 0) bstp_disable(&bif->bif_stp); } /* Pass on STP flags */ bstp_set_edge(bp, req->ifbr_ifsflags & IFBIF_BSTP_EDGE ? 1 : 0); bstp_set_autoedge(bp, req->ifbr_ifsflags & IFBIF_BSTP_AUTOEDGE ? 1 : 0); bstp_set_ptp(bp, req->ifbr_ifsflags & IFBIF_BSTP_PTP ? 1 : 0); bstp_set_autoptp(bp, req->ifbr_ifsflags & IFBIF_BSTP_AUTOPTP ? 1 : 0); /* Save the bits relating to the bridge */ bif->bif_flags = req->ifbr_ifsflags & IFBIFMASK; return (0); } static int bridge_ioctl_scache(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; sc->sc_brtmax = param->ifbrp_csize; bridge_rttrim(sc); return (0); } static int bridge_ioctl_gcache(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; param->ifbrp_csize = sc->sc_brtmax; return (0); } static int bridge_ioctl_gifs(struct bridge_softc *sc, void *arg) { struct ifbifconf *bifc = arg; struct bridge_iflist *bif; struct ifbreq breq; char *buf, *outbuf; int count, buflen, len, error = 0; count = 0; LIST_FOREACH(bif, &sc->sc_iflist, bif_next) count++; LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) count++; buflen = sizeof(breq) * count; if (bifc->ifbic_len == 0) { bifc->ifbic_len = buflen; return (0); } BRIDGE_UNLOCK(sc); outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); BRIDGE_LOCK(sc); count = 0; buf = outbuf; len = min(bifc->ifbic_len, buflen); bzero(&breq, sizeof(breq)); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (len < sizeof(breq)) break; strlcpy(breq.ifbr_ifsname, bif->bif_ifp->if_xname, sizeof(breq.ifbr_ifsname)); /* Fill in the ifbreq structure */ error = bridge_ioctl_gifflags(sc, &breq); if (error) break; memcpy(buf, &breq, sizeof(breq)); count++; buf += sizeof(breq); len -= sizeof(breq); } LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) { if (len < sizeof(breq)) break; strlcpy(breq.ifbr_ifsname, bif->bif_ifp->if_xname, sizeof(breq.ifbr_ifsname)); breq.ifbr_ifsflags = bif->bif_flags; breq.ifbr_portno = bif->bif_ifp->if_index & 0xfff; memcpy(buf, &breq, sizeof(breq)); count++; buf += sizeof(breq); len -= sizeof(breq); } BRIDGE_UNLOCK(sc); bifc->ifbic_len = sizeof(breq) * count; error = copyout(outbuf, bifc->ifbic_req, bifc->ifbic_len); BRIDGE_LOCK(sc); free(outbuf, M_TEMP); return (error); } static int bridge_ioctl_rts(struct bridge_softc *sc, void *arg) { struct ifbaconf *bac = arg; struct bridge_rtnode *brt; struct ifbareq bareq; char *buf, *outbuf; int count, buflen, len, error = 0; if (bac->ifbac_len == 0) return (0); count = 0; LIST_FOREACH(brt, &sc->sc_rtlist, brt_list) count++; buflen = sizeof(bareq) * count; BRIDGE_UNLOCK(sc); outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); BRIDGE_LOCK(sc); count = 0; buf = outbuf; len = min(bac->ifbac_len, buflen); bzero(&bareq, sizeof(bareq)); LIST_FOREACH(brt, &sc->sc_rtlist, brt_list) { if (len < sizeof(bareq)) goto out; strlcpy(bareq.ifba_ifsname, brt->brt_ifp->if_xname, sizeof(bareq.ifba_ifsname)); memcpy(bareq.ifba_dst, brt->brt_addr, sizeof(brt->brt_addr)); bareq.ifba_vlan = brt->brt_vlan; if ((brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC && time_uptime < brt->brt_expire) bareq.ifba_expire = brt->brt_expire - time_uptime; else bareq.ifba_expire = 0; bareq.ifba_flags = brt->brt_flags; memcpy(buf, &bareq, sizeof(bareq)); count++; buf += sizeof(bareq); len -= sizeof(bareq); } out: BRIDGE_UNLOCK(sc); bac->ifbac_len = sizeof(bareq) * count; error = copyout(outbuf, bac->ifbac_req, bac->ifbac_len); BRIDGE_LOCK(sc); free(outbuf, M_TEMP); return (error); } static int bridge_ioctl_saddr(struct bridge_softc *sc, void *arg) { struct ifbareq *req = arg; struct bridge_iflist *bif; int error; bif = bridge_lookup_member(sc, req->ifba_ifsname); if (bif == NULL) return (ENOENT); error = bridge_rtupdate(sc, req->ifba_dst, req->ifba_vlan, bif, 1, req->ifba_flags); return (error); } static int bridge_ioctl_sto(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; sc->sc_brttimeout = param->ifbrp_ctime; return (0); } static int bridge_ioctl_gto(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; param->ifbrp_ctime = sc->sc_brttimeout; return (0); } static int bridge_ioctl_daddr(struct bridge_softc *sc, void *arg) { struct ifbareq *req = arg; return (bridge_rtdaddr(sc, req->ifba_dst, req->ifba_vlan)); } static int bridge_ioctl_flush(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; bridge_rtflush(sc, req->ifbr_ifsflags); return (0); } static int bridge_ioctl_gpri(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; struct bstp_state *bs = &sc->sc_stp; param->ifbrp_prio = bs->bs_bridge_priority; return (0); } static int bridge_ioctl_spri(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_priority(&sc->sc_stp, param->ifbrp_prio)); } static int bridge_ioctl_ght(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; struct bstp_state *bs = &sc->sc_stp; param->ifbrp_hellotime = bs->bs_bridge_htime >> 8; return (0); } static int bridge_ioctl_sht(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_htime(&sc->sc_stp, param->ifbrp_hellotime)); } static int bridge_ioctl_gfd(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; struct bstp_state *bs = &sc->sc_stp; param->ifbrp_fwddelay = bs->bs_bridge_fdelay >> 8; return (0); } static int bridge_ioctl_sfd(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_fdelay(&sc->sc_stp, param->ifbrp_fwddelay)); } static int bridge_ioctl_gma(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; struct bstp_state *bs = &sc->sc_stp; param->ifbrp_maxage = bs->bs_bridge_max_age >> 8; return (0); } static int bridge_ioctl_sma(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_maxage(&sc->sc_stp, param->ifbrp_maxage)); } static int bridge_ioctl_sifprio(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); return (bstp_set_port_priority(&bif->bif_stp, req->ifbr_priority)); } static int bridge_ioctl_sifcost(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); return (bstp_set_path_cost(&bif->bif_stp, req->ifbr_path_cost)); } static int bridge_ioctl_sifmaxaddr(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; bif = bridge_lookup_member(sc, req->ifbr_ifsname); if (bif == NULL) return (ENOENT); bif->bif_addrmax = req->ifbr_addrmax; return (0); } static int bridge_ioctl_addspan(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif = NULL; struct ifnet *ifs; ifs = ifunit(req->ifbr_ifsname); if (ifs == NULL) return (ENOENT); LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) if (ifs == bif->bif_ifp) return (EBUSY); if (ifs->if_bridge != NULL) return (EBUSY); switch (ifs->if_type) { case IFT_ETHER: case IFT_GIF: case IFT_L2VLAN: break; default: return (EINVAL); } bif = malloc(sizeof(*bif), M_DEVBUF, M_NOWAIT|M_ZERO); if (bif == NULL) return (ENOMEM); bif->bif_ifp = ifs; bif->bif_flags = IFBIF_SPAN; LIST_INSERT_HEAD(&sc->sc_spanlist, bif, bif_next); return (0); } static int bridge_ioctl_delspan(struct bridge_softc *sc, void *arg) { struct ifbreq *req = arg; struct bridge_iflist *bif; struct ifnet *ifs; ifs = ifunit(req->ifbr_ifsname); if (ifs == NULL) return (ENOENT); LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) if (ifs == bif->bif_ifp) break; if (bif == NULL) return (ENOENT); bridge_delete_span(sc, bif); return (0); } static int bridge_ioctl_gbparam(struct bridge_softc *sc, void *arg) { struct ifbropreq *req = arg; struct bstp_state *bs = &sc->sc_stp; struct bstp_port *root_port; req->ifbop_maxage = bs->bs_bridge_max_age >> 8; req->ifbop_hellotime = bs->bs_bridge_htime >> 8; req->ifbop_fwddelay = bs->bs_bridge_fdelay >> 8; root_port = bs->bs_root_port; if (root_port == NULL) req->ifbop_root_port = 0; else req->ifbop_root_port = root_port->bp_ifp->if_index; req->ifbop_holdcount = bs->bs_txholdcount; req->ifbop_priority = bs->bs_bridge_priority; req->ifbop_protocol = bs->bs_protover; req->ifbop_root_path_cost = bs->bs_root_pv.pv_cost; req->ifbop_bridgeid = bs->bs_bridge_pv.pv_dbridge_id; req->ifbop_designated_root = bs->bs_root_pv.pv_root_id; req->ifbop_designated_bridge = bs->bs_root_pv.pv_dbridge_id; req->ifbop_last_tc_time.tv_sec = bs->bs_last_tc_time.tv_sec; req->ifbop_last_tc_time.tv_usec = bs->bs_last_tc_time.tv_usec; return (0); } static int bridge_ioctl_grte(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; param->ifbrp_cexceeded = sc->sc_brtexceeded; return (0); } static int bridge_ioctl_gifsstp(struct bridge_softc *sc, void *arg) { struct ifbpstpconf *bifstp = arg; struct bridge_iflist *bif; struct bstp_port *bp; struct ifbpstpreq bpreq; char *buf, *outbuf; int count, buflen, len, error = 0; count = 0; LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if ((bif->bif_flags & IFBIF_STP) != 0) count++; } buflen = sizeof(bpreq) * count; if (bifstp->ifbpstp_len == 0) { bifstp->ifbpstp_len = buflen; return (0); } BRIDGE_UNLOCK(sc); outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); BRIDGE_LOCK(sc); count = 0; buf = outbuf; len = min(bifstp->ifbpstp_len, buflen); bzero(&bpreq, sizeof(bpreq)); LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (len < sizeof(bpreq)) break; if ((bif->bif_flags & IFBIF_STP) == 0) continue; bp = &bif->bif_stp; bpreq.ifbp_portno = bif->bif_ifp->if_index & 0xfff; bpreq.ifbp_fwd_trans = bp->bp_forward_transitions; bpreq.ifbp_design_cost = bp->bp_desg_pv.pv_cost; bpreq.ifbp_design_port = bp->bp_desg_pv.pv_port_id; bpreq.ifbp_design_bridge = bp->bp_desg_pv.pv_dbridge_id; bpreq.ifbp_design_root = bp->bp_desg_pv.pv_root_id; memcpy(buf, &bpreq, sizeof(bpreq)); count++; buf += sizeof(bpreq); len -= sizeof(bpreq); } BRIDGE_UNLOCK(sc); bifstp->ifbpstp_len = sizeof(bpreq) * count; error = copyout(outbuf, bifstp->ifbpstp_req, bifstp->ifbpstp_len); BRIDGE_LOCK(sc); free(outbuf, M_TEMP); return (error); } static int bridge_ioctl_sproto(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_protocol(&sc->sc_stp, param->ifbrp_proto)); } static int bridge_ioctl_stxhc(struct bridge_softc *sc, void *arg) { struct ifbrparam *param = arg; return (bstp_set_holdcount(&sc->sc_stp, param->ifbrp_txhc)); } /* * bridge_ifdetach: * * Detach an interface from a bridge. Called when a member * interface is detaching. */ static void bridge_ifdetach(void *arg __unused, struct ifnet *ifp) { struct bridge_softc *sc = ifp->if_bridge; struct bridge_iflist *bif; if (ifp->if_flags & IFF_RENAMING) return; if (V_bridge_cloner == NULL) { /* * This detach handler can be called after * vnet_bridge_uninit(). Just return in that case. */ return; } /* Check if the interface is a bridge member */ if (sc != NULL) { BRIDGE_LOCK(sc); bif = bridge_lookup_member_if(sc, ifp); if (bif != NULL) bridge_delete_member(sc, bif, 1); BRIDGE_UNLOCK(sc); return; } /* Check if the interface is a span port */ BRIDGE_LIST_LOCK(); LIST_FOREACH(sc, &V_bridge_list, sc_list) { BRIDGE_LOCK(sc); LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) if (ifp == bif->bif_ifp) { bridge_delete_span(sc, bif); break; } BRIDGE_UNLOCK(sc); } BRIDGE_LIST_UNLOCK(); } /* * bridge_init: * * Initialize a bridge interface. */ static void bridge_init(void *xsc) { struct bridge_softc *sc = (struct bridge_softc *)xsc; struct ifnet *ifp = sc->sc_ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; BRIDGE_LOCK(sc); callout_reset(&sc->sc_brcallout, bridge_rtable_prune_period * hz, bridge_timer, sc); ifp->if_drv_flags |= IFF_DRV_RUNNING; bstp_init(&sc->sc_stp); /* Initialize Spanning Tree */ BRIDGE_UNLOCK(sc); } /* * bridge_stop: * * Stop the bridge interface. */ static void bridge_stop(struct ifnet *ifp, int disable) { struct bridge_softc *sc = ifp->if_softc; BRIDGE_LOCK_ASSERT(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; callout_stop(&sc->sc_brcallout); bstp_stop(&sc->sc_stp); bridge_rtflush(sc, IFBF_FLUSHDYN); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; } /* * bridge_enqueue: * * Enqueue a packet on a bridge member interface. * */ static int bridge_enqueue(struct bridge_softc *sc, struct ifnet *dst_ifp, struct mbuf *m) { int len, err = 0; short mflags; struct mbuf *m0; /* We may be sending a fragment so traverse the mbuf */ for (; m; m = m0) { m0 = m->m_nextpkt; m->m_nextpkt = NULL; len = m->m_pkthdr.len; mflags = m->m_flags; /* * If underlying interface can not do VLAN tag insertion itself * then attach a packet tag that holds it. */ if ((m->m_flags & M_VLANTAG) && (dst_ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) { m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); if (m == NULL) { if_printf(dst_ifp, "unable to prepend VLAN header\n"); if_inc_counter(dst_ifp, IFCOUNTER_OERRORS, 1); continue; } m->m_flags &= ~M_VLANTAG; } M_ASSERTPKTHDR(m); /* We shouldn't transmit mbuf without pkthdr */ if ((err = dst_ifp->if_transmit(dst_ifp, m))) { m_freem(m0); if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); break; } if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, len); if (mflags & M_MCAST) if_inc_counter(sc->sc_ifp, IFCOUNTER_OMCASTS, 1); } return (err); } /* * bridge_dummynet: * * Receive a queued packet from dummynet and pass it on to the output * interface. * * The mbuf has the Ethernet header already attached. */ static void bridge_dummynet(struct mbuf *m, struct ifnet *ifp) { struct bridge_softc *sc; sc = ifp->if_bridge; /* * The packet didnt originate from a member interface. This should only * ever happen if a member interface is removed while packets are * queued for it. */ if (sc == NULL) { m_freem(m); return; } if (PFIL_HOOKED(&V_inet_pfil_hook) #ifdef INET6 || PFIL_HOOKED(&V_inet6_pfil_hook) #endif ) { if (bridge_pfil(&m, sc->sc_ifp, ifp, PFIL_OUT) != 0) return; if (m == NULL) return; } bridge_enqueue(sc, ifp, m); } /* * bridge_output: * * Send output from a bridge member interface. This * performs the bridging function for locally originated * packets. * * The mbuf has the Ethernet header already attached. We must * enqueue or free the mbuf before returning. */ static int bridge_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa, struct rtentry *rt) { struct ether_header *eh; struct ifnet *dst_if; struct bridge_softc *sc; uint16_t vlan; if (m->m_len < ETHER_HDR_LEN) { m = m_pullup(m, ETHER_HDR_LEN); if (m == NULL) return (0); } eh = mtod(m, struct ether_header *); sc = ifp->if_bridge; vlan = VLANTAGOF(m); BRIDGE_LOCK(sc); /* * If bridge is down, but the original output interface is up, * go ahead and send out that interface. Otherwise, the packet * is dropped below. */ if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { dst_if = ifp; goto sendunicast; } /* * If the packet is a multicast, or we don't know a better way to * get there, send to all interfaces. */ if (ETHER_IS_MULTICAST(eh->ether_dhost)) dst_if = NULL; else dst_if = bridge_rtlookup(sc, eh->ether_dhost, vlan); if (dst_if == NULL) { struct bridge_iflist *bif; struct mbuf *mc; int error = 0, used = 0; bridge_span(sc, m); BRIDGE_LOCK2REF(sc, error); if (error) { m_freem(m); return (0); } LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { dst_if = bif->bif_ifp; if (dst_if->if_type == IFT_GIF) continue; if ((dst_if->if_drv_flags & IFF_DRV_RUNNING) == 0) continue; /* * If this is not the original output interface, * and the interface is participating in spanning * tree, make sure the port is in a state that * allows forwarding. */ if (dst_if != ifp && (bif->bif_flags & IFBIF_STP) && bif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) continue; if (LIST_NEXT(bif, bif_next) == NULL) { used = 1; mc = m; } else { mc = m_copypacket(m, M_NOWAIT); if (mc == NULL) { if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); continue; } } bridge_enqueue(sc, dst_if, mc); } if (used == 0) m_freem(m); BRIDGE_UNREF(sc); return (0); } sendunicast: /* * XXX Spanning tree consideration here? */ bridge_span(sc, m); if ((dst_if->if_drv_flags & IFF_DRV_RUNNING) == 0) { m_freem(m); BRIDGE_UNLOCK(sc); return (0); } BRIDGE_UNLOCK(sc); bridge_enqueue(sc, dst_if, m); return (0); } /* * bridge_transmit: * * Do output on a bridge. * */ static int bridge_transmit(struct ifnet *ifp, struct mbuf *m) { struct bridge_softc *sc; struct ether_header *eh; struct ifnet *dst_if; int error = 0; sc = ifp->if_softc; ETHER_BPF_MTAP(ifp, m); eh = mtod(m, struct ether_header *); BRIDGE_LOCK(sc); if (((m->m_flags & (M_BCAST|M_MCAST)) == 0) && (dst_if = bridge_rtlookup(sc, eh->ether_dhost, 1)) != NULL) { BRIDGE_UNLOCK(sc); error = bridge_enqueue(sc, dst_if, m); } else bridge_broadcast(sc, ifp, m, 0); return (error); } /* * The ifp->if_qflush entry point for if_bridge(4) is no-op. */ static void bridge_qflush(struct ifnet *ifp __unused) { } /* * bridge_forward: * * The forwarding function of the bridge. * * NOTE: Releases the lock on return. */ static void bridge_forward(struct bridge_softc *sc, struct bridge_iflist *sbif, struct mbuf *m) { struct bridge_iflist *dbif; struct ifnet *src_if, *dst_if, *ifp; struct ether_header *eh; uint16_t vlan; uint8_t *dst; int error; src_if = m->m_pkthdr.rcvif; ifp = sc->sc_ifp; if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); vlan = VLANTAGOF(m); if ((sbif->bif_flags & IFBIF_STP) && sbif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) goto drop; eh = mtod(m, struct ether_header *); dst = eh->ether_dhost; /* If the interface is learning, record the address. */ if (sbif->bif_flags & IFBIF_LEARNING) { error = bridge_rtupdate(sc, eh->ether_shost, vlan, sbif, 0, IFBAF_DYNAMIC); /* * If the interface has addresses limits then deny any source * that is not in the cache. */ if (error && sbif->bif_addrmax) goto drop; } if ((sbif->bif_flags & IFBIF_STP) != 0 && sbif->bif_stp.bp_state == BSTP_IFSTATE_LEARNING) goto drop; /* * At this point, the port either doesn't participate * in spanning tree or it is in the forwarding state. */ /* * If the packet is unicast, destined for someone on * "this" side of the bridge, drop it. */ if ((m->m_flags & (M_BCAST|M_MCAST)) == 0) { dst_if = bridge_rtlookup(sc, dst, vlan); if (src_if == dst_if) goto drop; } else { /* * Check if its a reserved multicast address, any address * listed in 802.1D section 7.12.6 may not be forwarded by the * bridge. * This is currently 01-80-C2-00-00-00 to 01-80-C2-00-00-0F */ if (dst[0] == 0x01 && dst[1] == 0x80 && dst[2] == 0xc2 && dst[3] == 0x00 && dst[4] == 0x00 && dst[5] <= 0x0f) goto drop; /* ...forward it to all interfaces. */ if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1); dst_if = NULL; } /* * If we have a destination interface which is a member of our bridge, * OR this is a unicast packet, push it through the bpf(4) machinery. * For broadcast or multicast packets, don't bother because it will * be reinjected into ether_input. We do this before we pass the packets * through the pfil(9) framework, as it is possible that pfil(9) will * drop the packet, or possibly modify it, making it difficult to debug * firewall issues on the bridge. */ if (dst_if != NULL || (m->m_flags & (M_BCAST | M_MCAST)) == 0) ETHER_BPF_MTAP(ifp, m); /* run the packet filter */ if (PFIL_HOOKED(&V_inet_pfil_hook) #ifdef INET6 || PFIL_HOOKED(&V_inet6_pfil_hook) #endif ) { BRIDGE_UNLOCK(sc); if (bridge_pfil(&m, ifp, src_if, PFIL_IN) != 0) return; if (m == NULL) return; BRIDGE_LOCK(sc); } if (dst_if == NULL) { bridge_broadcast(sc, src_if, m, 1); return; } /* * At this point, we're dealing with a unicast frame * going to a different interface. */ if ((dst_if->if_drv_flags & IFF_DRV_RUNNING) == 0) goto drop; dbif = bridge_lookup_member_if(sc, dst_if); if (dbif == NULL) /* Not a member of the bridge (anymore?) */ goto drop; /* Private segments can not talk to each other */ if (sbif->bif_flags & dbif->bif_flags & IFBIF_PRIVATE) goto drop; if ((dbif->bif_flags & IFBIF_STP) && dbif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) goto drop; BRIDGE_UNLOCK(sc); if (PFIL_HOOKED(&V_inet_pfil_hook) #ifdef INET6 || PFIL_HOOKED(&V_inet6_pfil_hook) #endif ) { if (bridge_pfil(&m, ifp, dst_if, PFIL_OUT) != 0) return; if (m == NULL) return; } bridge_enqueue(sc, dst_if, m); return; drop: BRIDGE_UNLOCK(sc); m_freem(m); } /* * bridge_input: * * Receive input from a member interface. Queue the packet for * bridging if it is not for us. */ static struct mbuf * bridge_input(struct ifnet *ifp, struct mbuf *m) { struct bridge_softc *sc = ifp->if_bridge; struct bridge_iflist *bif, *bif2; struct ifnet *bifp; struct ether_header *eh; struct mbuf *mc, *mc2; uint16_t vlan; int error; if ((sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return (m); bifp = sc->sc_ifp; vlan = VLANTAGOF(m); /* * Implement support for bridge monitoring. If this flag has been * set on this interface, discard the packet once we push it through * the bpf(4) machinery, but before we do, increment the byte and * packet counters associated with this interface. */ if ((bifp->if_flags & IFF_MONITOR) != 0) { m->m_pkthdr.rcvif = bifp; ETHER_BPF_MTAP(bifp, m); if_inc_counter(bifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(bifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); m_freem(m); return (NULL); } BRIDGE_LOCK(sc); bif = bridge_lookup_member_if(sc, ifp); if (bif == NULL) { BRIDGE_UNLOCK(sc); return (m); } eh = mtod(m, struct ether_header *); bridge_span(sc, m); if (m->m_flags & (M_BCAST|M_MCAST)) { /* Tap off 802.1D packets; they do not get forwarded. */ if (memcmp(eh->ether_dhost, bstp_etheraddr, ETHER_ADDR_LEN) == 0) { bstp_input(&bif->bif_stp, ifp, m); /* consumes mbuf */ BRIDGE_UNLOCK(sc); return (NULL); } if ((bif->bif_flags & IFBIF_STP) && bif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) { BRIDGE_UNLOCK(sc); return (m); } /* * Make a deep copy of the packet and enqueue the copy * for bridge processing; return the original packet for * local processing. */ mc = m_dup(m, M_NOWAIT); if (mc == NULL) { BRIDGE_UNLOCK(sc); return (m); } /* Perform the bridge forwarding function with the copy. */ bridge_forward(sc, bif, mc); /* * Reinject the mbuf as arriving on the bridge so we have a * chance at claiming multicast packets. We can not loop back * here from ether_input as a bridge is never a member of a * bridge. */ KASSERT(bifp->if_bridge == NULL, ("loop created in bridge_input")); mc2 = m_dup(m, M_NOWAIT); if (mc2 != NULL) { /* Keep the layer3 header aligned */ int i = min(mc2->m_pkthdr.len, max_protohdr); mc2 = m_copyup(mc2, i, ETHER_ALIGN); } if (mc2 != NULL) { mc2->m_pkthdr.rcvif = bifp; (*bifp->if_input)(bifp, mc2); } /* Return the original packet for local processing. */ return (m); } if ((bif->bif_flags & IFBIF_STP) && bif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) { BRIDGE_UNLOCK(sc); return (m); } #if (defined(INET) || defined(INET6)) # define OR_CARP_CHECK_WE_ARE_DST(iface) \ || ((iface)->if_carp \ && (*carp_forus_p)((iface), eh->ether_dhost)) # define OR_CARP_CHECK_WE_ARE_SRC(iface) \ || ((iface)->if_carp \ && (*carp_forus_p)((iface), eh->ether_shost)) #else # define OR_CARP_CHECK_WE_ARE_DST(iface) # define OR_CARP_CHECK_WE_ARE_SRC(iface) #endif #ifdef INET6 # define OR_PFIL_HOOKED_INET6 \ || PFIL_HOOKED(&V_inet6_pfil_hook) #else # define OR_PFIL_HOOKED_INET6 #endif #define GRAB_OUR_PACKETS(iface) \ if ((iface)->if_type == IFT_GIF) \ continue; \ /* It is destined for us. */ \ if (memcmp(IF_LLADDR((iface)), eh->ether_dhost, ETHER_ADDR_LEN) == 0 \ OR_CARP_CHECK_WE_ARE_DST((iface)) \ ) { \ if ((iface)->if_type == IFT_BRIDGE) { \ ETHER_BPF_MTAP(iface, m); \ if_inc_counter(iface, IFCOUNTER_IPACKETS, 1); \ if_inc_counter(iface, IFCOUNTER_IBYTES, m->m_pkthdr.len); \ /* Filter on the physical interface. */ \ if (V_pfil_local_phys && \ (PFIL_HOOKED(&V_inet_pfil_hook) \ OR_PFIL_HOOKED_INET6)) { \ if (bridge_pfil(&m, NULL, ifp, \ PFIL_IN) != 0 || m == NULL) { \ BRIDGE_UNLOCK(sc); \ return (NULL); \ } \ eh = mtod(m, struct ether_header *); \ } \ } \ if (bif->bif_flags & IFBIF_LEARNING) { \ error = bridge_rtupdate(sc, eh->ether_shost, \ vlan, bif, 0, IFBAF_DYNAMIC); \ if (error && bif->bif_addrmax) { \ BRIDGE_UNLOCK(sc); \ m_freem(m); \ return (NULL); \ } \ } \ m->m_pkthdr.rcvif = iface; \ BRIDGE_UNLOCK(sc); \ return (m); \ } \ \ /* We just received a packet that we sent out. */ \ if (memcmp(IF_LLADDR((iface)), eh->ether_shost, ETHER_ADDR_LEN) == 0 \ OR_CARP_CHECK_WE_ARE_SRC((iface)) \ ) { \ BRIDGE_UNLOCK(sc); \ m_freem(m); \ return (NULL); \ } /* * Unicast. Make sure it's not for the bridge. */ do { GRAB_OUR_PACKETS(bifp) } while (0); /* * Give a chance for ifp at first priority. This will help when the * packet comes through the interface like VLAN's with the same MACs * on several interfaces from the same bridge. This also will save * some CPU cycles in case the destination interface and the input * interface (eq ifp) are the same. */ do { GRAB_OUR_PACKETS(ifp) } while (0); /* Now check the all bridge members. */ LIST_FOREACH(bif2, &sc->sc_iflist, bif_next) { GRAB_OUR_PACKETS(bif2->bif_ifp) } #undef OR_CARP_CHECK_WE_ARE_DST #undef OR_CARP_CHECK_WE_ARE_SRC #undef OR_PFIL_HOOKED_INET6 #undef GRAB_OUR_PACKETS /* Perform the bridge forwarding function. */ bridge_forward(sc, bif, m); return (NULL); } /* * bridge_broadcast: * * Send a frame to all interfaces that are members of * the bridge, except for the one on which the packet * arrived. * * NOTE: Releases the lock on return. */ static void bridge_broadcast(struct bridge_softc *sc, struct ifnet *src_if, struct mbuf *m, int runfilt) { struct bridge_iflist *dbif, *sbif; struct mbuf *mc; struct ifnet *dst_if; int error = 0, used = 0, i; sbif = bridge_lookup_member_if(sc, src_if); BRIDGE_LOCK2REF(sc, error); if (error) { m_freem(m); return; } /* Filter on the bridge interface before broadcasting */ if (runfilt && (PFIL_HOOKED(&V_inet_pfil_hook) #ifdef INET6 || PFIL_HOOKED(&V_inet6_pfil_hook) #endif )) { if (bridge_pfil(&m, sc->sc_ifp, NULL, PFIL_OUT) != 0) goto out; if (m == NULL) goto out; } LIST_FOREACH(dbif, &sc->sc_iflist, bif_next) { dst_if = dbif->bif_ifp; if (dst_if == src_if) continue; /* Private segments can not talk to each other */ if (sbif && (sbif->bif_flags & dbif->bif_flags & IFBIF_PRIVATE)) continue; if ((dbif->bif_flags & IFBIF_STP) && dbif->bif_stp.bp_state == BSTP_IFSTATE_DISCARDING) continue; if ((dbif->bif_flags & IFBIF_DISCOVER) == 0 && (m->m_flags & (M_BCAST|M_MCAST)) == 0) continue; if ((dst_if->if_drv_flags & IFF_DRV_RUNNING) == 0) continue; if (LIST_NEXT(dbif, bif_next) == NULL) { mc = m; used = 1; } else { mc = m_dup(m, M_NOWAIT); if (mc == NULL) { if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); continue; } } /* * Filter on the output interface. Pass a NULL bridge interface * pointer so we do not redundantly filter on the bridge for * each interface we broadcast on. */ if (runfilt && (PFIL_HOOKED(&V_inet_pfil_hook) #ifdef INET6 || PFIL_HOOKED(&V_inet6_pfil_hook) #endif )) { if (used == 0) { /* Keep the layer3 header aligned */ i = min(mc->m_pkthdr.len, max_protohdr); mc = m_copyup(mc, i, ETHER_ALIGN); if (mc == NULL) { if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); continue; } } if (bridge_pfil(&mc, NULL, dst_if, PFIL_OUT) != 0) continue; if (mc == NULL) continue; } bridge_enqueue(sc, dst_if, mc); } if (used == 0) m_freem(m); out: BRIDGE_UNREF(sc); } /* * bridge_span: * * Duplicate a packet out one or more interfaces that are in span mode, * the original mbuf is unmodified. */ static void bridge_span(struct bridge_softc *sc, struct mbuf *m) { struct bridge_iflist *bif; struct ifnet *dst_if; struct mbuf *mc; if (LIST_EMPTY(&sc->sc_spanlist)) return; LIST_FOREACH(bif, &sc->sc_spanlist, bif_next) { dst_if = bif->bif_ifp; if ((dst_if->if_drv_flags & IFF_DRV_RUNNING) == 0) continue; mc = m_copypacket(m, M_NOWAIT); if (mc == NULL) { if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); continue; } bridge_enqueue(sc, dst_if, mc); } } /* * bridge_rtupdate: * * Add a bridge routing entry. */ static int bridge_rtupdate(struct bridge_softc *sc, const uint8_t *dst, uint16_t vlan, struct bridge_iflist *bif, int setflags, uint8_t flags) { struct bridge_rtnode *brt; int error; BRIDGE_LOCK_ASSERT(sc); /* Check the source address is valid and not multicast. */ if (ETHER_IS_MULTICAST(dst) || (dst[0] == 0 && dst[1] == 0 && dst[2] == 0 && dst[3] == 0 && dst[4] == 0 && dst[5] == 0) != 0) return (EINVAL); /* 802.1p frames map to vlan 1 */ if (vlan == 0) vlan = 1; /* * A route for this destination might already exist. If so, * update it, otherwise create a new one. */ if ((brt = bridge_rtnode_lookup(sc, dst, vlan)) == NULL) { if (sc->sc_brtcnt >= sc->sc_brtmax) { sc->sc_brtexceeded++; return (ENOSPC); } /* Check per interface address limits (if enabled) */ if (bif->bif_addrmax && bif->bif_addrcnt >= bif->bif_addrmax) { bif->bif_addrexceeded++; return (ENOSPC); } /* * Allocate a new bridge forwarding node, and * initialize the expiration time and Ethernet * address. */ brt = uma_zalloc(bridge_rtnode_zone, M_NOWAIT | M_ZERO); if (brt == NULL) return (ENOMEM); if (bif->bif_flags & IFBIF_STICKY) brt->brt_flags = IFBAF_STICKY; else brt->brt_flags = IFBAF_DYNAMIC; memcpy(brt->brt_addr, dst, ETHER_ADDR_LEN); brt->brt_vlan = vlan; if ((error = bridge_rtnode_insert(sc, brt)) != 0) { uma_zfree(bridge_rtnode_zone, brt); return (error); } brt->brt_dst = bif; bif->bif_addrcnt++; } if ((brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC && brt->brt_dst != bif) { brt->brt_dst->bif_addrcnt--; brt->brt_dst = bif; brt->brt_dst->bif_addrcnt++; } if ((flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC) brt->brt_expire = time_uptime + sc->sc_brttimeout; if (setflags) brt->brt_flags = flags; return (0); } /* * bridge_rtlookup: * * Lookup the destination interface for an address. */ static struct ifnet * bridge_rtlookup(struct bridge_softc *sc, const uint8_t *addr, uint16_t vlan) { struct bridge_rtnode *brt; BRIDGE_LOCK_ASSERT(sc); if ((brt = bridge_rtnode_lookup(sc, addr, vlan)) == NULL) return (NULL); return (brt->brt_ifp); } /* * bridge_rttrim: * * Trim the routine table so that we have a number * of routing entries less than or equal to the * maximum number. */ static void bridge_rttrim(struct bridge_softc *sc) { struct bridge_rtnode *brt, *nbrt; BRIDGE_LOCK_ASSERT(sc); /* Make sure we actually need to do this. */ if (sc->sc_brtcnt <= sc->sc_brtmax) return; /* Force an aging cycle; this might trim enough addresses. */ bridge_rtage(sc); if (sc->sc_brtcnt <= sc->sc_brtmax) return; LIST_FOREACH_SAFE(brt, &sc->sc_rtlist, brt_list, nbrt) { if ((brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC) { bridge_rtnode_destroy(sc, brt); if (sc->sc_brtcnt <= sc->sc_brtmax) return; } } } /* * bridge_timer: * * Aging timer for the bridge. */ static void bridge_timer(void *arg) { struct bridge_softc *sc = arg; BRIDGE_LOCK_ASSERT(sc); bridge_rtage(sc); if (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) callout_reset(&sc->sc_brcallout, bridge_rtable_prune_period * hz, bridge_timer, sc); } /* * bridge_rtage: * * Perform an aging cycle. */ static void bridge_rtage(struct bridge_softc *sc) { struct bridge_rtnode *brt, *nbrt; BRIDGE_LOCK_ASSERT(sc); LIST_FOREACH_SAFE(brt, &sc->sc_rtlist, brt_list, nbrt) { if ((brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC) { if (time_uptime >= brt->brt_expire) bridge_rtnode_destroy(sc, brt); } } } /* * bridge_rtflush: * * Remove all dynamic addresses from the bridge. */ static void bridge_rtflush(struct bridge_softc *sc, int full) { struct bridge_rtnode *brt, *nbrt; BRIDGE_LOCK_ASSERT(sc); LIST_FOREACH_SAFE(brt, &sc->sc_rtlist, brt_list, nbrt) { if (full || (brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC) bridge_rtnode_destroy(sc, brt); } } /* * bridge_rtdaddr: * * Remove an address from the table. */ static int bridge_rtdaddr(struct bridge_softc *sc, const uint8_t *addr, uint16_t vlan) { struct bridge_rtnode *brt; int found = 0; BRIDGE_LOCK_ASSERT(sc); /* * If vlan is zero then we want to delete for all vlans so the lookup * may return more than one. */ while ((brt = bridge_rtnode_lookup(sc, addr, vlan)) != NULL) { bridge_rtnode_destroy(sc, brt); found = 1; } return (found ? 0 : ENOENT); } /* * bridge_rtdelete: * * Delete routes to a speicifc member interface. */ static void bridge_rtdelete(struct bridge_softc *sc, struct ifnet *ifp, int full) { struct bridge_rtnode *brt, *nbrt; BRIDGE_LOCK_ASSERT(sc); LIST_FOREACH_SAFE(brt, &sc->sc_rtlist, brt_list, nbrt) { if (brt->brt_ifp == ifp && (full || (brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC)) bridge_rtnode_destroy(sc, brt); } } /* * bridge_rtable_init: * * Initialize the route table for this bridge. */ static void bridge_rtable_init(struct bridge_softc *sc) { int i; sc->sc_rthash = malloc(sizeof(*sc->sc_rthash) * BRIDGE_RTHASH_SIZE, M_DEVBUF, M_WAITOK); for (i = 0; i < BRIDGE_RTHASH_SIZE; i++) LIST_INIT(&sc->sc_rthash[i]); sc->sc_rthash_key = arc4random(); LIST_INIT(&sc->sc_rtlist); } /* * bridge_rtable_fini: * * Deconstruct the route table for this bridge. */ static void bridge_rtable_fini(struct bridge_softc *sc) { KASSERT(sc->sc_brtcnt == 0, ("%s: %d bridge routes referenced", __func__, sc->sc_brtcnt)); free(sc->sc_rthash, M_DEVBUF); } /* * The following hash function is adapted from "Hash Functions" by Bob Jenkins * ("Algorithm Alley", Dr. Dobbs Journal, September 1997). */ #define mix(a, b, c) \ do { \ a -= b; a -= c; a ^= (c >> 13); \ b -= c; b -= a; b ^= (a << 8); \ c -= a; c -= b; c ^= (b >> 13); \ a -= b; a -= c; a ^= (c >> 12); \ b -= c; b -= a; b ^= (a << 16); \ c -= a; c -= b; c ^= (b >> 5); \ a -= b; a -= c; a ^= (c >> 3); \ b -= c; b -= a; b ^= (a << 10); \ c -= a; c -= b; c ^= (b >> 15); \ } while (/*CONSTCOND*/0) static __inline uint32_t bridge_rthash(struct bridge_softc *sc, const uint8_t *addr) { uint32_t a = 0x9e3779b9, b = 0x9e3779b9, c = sc->sc_rthash_key; b += addr[5] << 8; b += addr[4]; a += addr[3] << 24; a += addr[2] << 16; a += addr[1] << 8; a += addr[0]; mix(a, b, c); return (c & BRIDGE_RTHASH_MASK); } #undef mix static int bridge_rtnode_addr_cmp(const uint8_t *a, const uint8_t *b) { int i, d; for (i = 0, d = 0; i < ETHER_ADDR_LEN && d == 0; i++) { d = ((int)a[i]) - ((int)b[i]); } return (d); } /* * bridge_rtnode_lookup: * * Look up a bridge route node for the specified destination. Compare the * vlan id or if zero then just return the first match. */ static struct bridge_rtnode * bridge_rtnode_lookup(struct bridge_softc *sc, const uint8_t *addr, uint16_t vlan) { struct bridge_rtnode *brt; uint32_t hash; int dir; BRIDGE_LOCK_ASSERT(sc); hash = bridge_rthash(sc, addr); LIST_FOREACH(brt, &sc->sc_rthash[hash], brt_hash) { dir = bridge_rtnode_addr_cmp(addr, brt->brt_addr); if (dir == 0 && (brt->brt_vlan == vlan || vlan == 0)) return (brt); if (dir > 0) return (NULL); } return (NULL); } /* * bridge_rtnode_insert: * * Insert the specified bridge node into the route table. We * assume the entry is not already in the table. */ static int bridge_rtnode_insert(struct bridge_softc *sc, struct bridge_rtnode *brt) { struct bridge_rtnode *lbrt; uint32_t hash; int dir; BRIDGE_LOCK_ASSERT(sc); hash = bridge_rthash(sc, brt->brt_addr); lbrt = LIST_FIRST(&sc->sc_rthash[hash]); if (lbrt == NULL) { LIST_INSERT_HEAD(&sc->sc_rthash[hash], brt, brt_hash); goto out; } do { dir = bridge_rtnode_addr_cmp(brt->brt_addr, lbrt->brt_addr); if (dir == 0 && brt->brt_vlan == lbrt->brt_vlan) return (EEXIST); if (dir > 0) { LIST_INSERT_BEFORE(lbrt, brt, brt_hash); goto out; } if (LIST_NEXT(lbrt, brt_hash) == NULL) { LIST_INSERT_AFTER(lbrt, brt, brt_hash); goto out; } lbrt = LIST_NEXT(lbrt, brt_hash); } while (lbrt != NULL); #ifdef DIAGNOSTIC panic("bridge_rtnode_insert: impossible"); #endif out: LIST_INSERT_HEAD(&sc->sc_rtlist, brt, brt_list); sc->sc_brtcnt++; return (0); } /* * bridge_rtnode_destroy: * * Destroy a bridge rtnode. */ static void bridge_rtnode_destroy(struct bridge_softc *sc, struct bridge_rtnode *brt) { BRIDGE_LOCK_ASSERT(sc); LIST_REMOVE(brt, brt_hash); LIST_REMOVE(brt, brt_list); sc->sc_brtcnt--; brt->brt_dst->bif_addrcnt--; uma_zfree(bridge_rtnode_zone, brt); } /* * bridge_rtable_expire: * * Set the expiry time for all routes on an interface. */ static void bridge_rtable_expire(struct ifnet *ifp, int age) { struct bridge_softc *sc = ifp->if_bridge; struct bridge_rtnode *brt; BRIDGE_LOCK(sc); /* * If the age is zero then flush, otherwise set all the expiry times to * age for the interface */ if (age == 0) bridge_rtdelete(sc, ifp, IFBF_FLUSHDYN); else { LIST_FOREACH(brt, &sc->sc_rtlist, brt_list) { /* Cap the expiry time to 'age' */ if (brt->brt_ifp == ifp && brt->brt_expire > time_uptime + age && (brt->brt_flags & IFBAF_TYPEMASK) == IFBAF_DYNAMIC) brt->brt_expire = time_uptime + age; } } BRIDGE_UNLOCK(sc); } /* * bridge_state_change: * * Callback from the bridgestp code when a port changes states. */ static void bridge_state_change(struct ifnet *ifp, int state) { struct bridge_softc *sc = ifp->if_bridge; static const char *stpstates[] = { "disabled", "listening", "learning", "forwarding", "blocking", "discarding" }; CURVNET_SET(ifp->if_vnet); if (V_log_stp) log(LOG_NOTICE, "%s: state changed to %s on %s\n", sc->sc_ifp->if_xname, stpstates[state], ifp->if_xname); CURVNET_RESTORE(); } /* * Send bridge packets through pfil if they are one of the types pfil can deal * with, or if they are ARP or REVARP. (pfil will pass ARP and REVARP without * question.) If *bifp or *ifp are NULL then packet filtering is skipped for * that interface. */ static int bridge_pfil(struct mbuf **mp, struct ifnet *bifp, struct ifnet *ifp, int dir) { int snap, error, i, hlen; struct ether_header *eh1, eh2; struct ip *ip; struct llc llc1; u_int16_t ether_type; snap = 0; error = -1; /* Default error if not error == 0 */ #if 0 /* we may return with the IP fields swapped, ensure its not shared */ KASSERT(M_WRITABLE(*mp), ("%s: modifying a shared mbuf", __func__)); #endif if (V_pfil_bridge == 0 && V_pfil_member == 0 && V_pfil_ipfw == 0) return (0); /* filtering is disabled */ i = min((*mp)->m_pkthdr.len, max_protohdr); if ((*mp)->m_len < i) { *mp = m_pullup(*mp, i); if (*mp == NULL) { printf("%s: m_pullup failed\n", __func__); return (-1); } } eh1 = mtod(*mp, struct ether_header *); ether_type = ntohs(eh1->ether_type); /* * Check for SNAP/LLC. */ if (ether_type < ETHERMTU) { struct llc *llc2 = (struct llc *)(eh1 + 1); if ((*mp)->m_len >= ETHER_HDR_LEN + 8 && llc2->llc_dsap == LLC_SNAP_LSAP && llc2->llc_ssap == LLC_SNAP_LSAP && llc2->llc_control == LLC_UI) { ether_type = htons(llc2->llc_un.type_snap.ether_type); snap = 1; } } /* * If we're trying to filter bridge traffic, don't look at anything * other than IP and ARP traffic. If the filter doesn't understand * IPv6, don't allow IPv6 through the bridge either. This is lame * since if we really wanted, say, an AppleTalk filter, we are hosed, * but of course we don't have an AppleTalk filter to begin with. * (Note that since pfil doesn't understand ARP it will pass *ALL* * ARP traffic.) */ switch (ether_type) { case ETHERTYPE_ARP: case ETHERTYPE_REVARP: if (V_pfil_ipfw_arp == 0) return (0); /* Automatically pass */ break; case ETHERTYPE_IP: #ifdef INET6 case ETHERTYPE_IPV6: #endif /* INET6 */ break; default: /* * Check to see if the user wants to pass non-ip * packets, these will not be checked by pfil(9) and * passed unconditionally so the default is to drop. */ if (V_pfil_onlyip) goto bad; } /* Run the packet through pfil before stripping link headers */ if (PFIL_HOOKED(&V_link_pfil_hook) && V_pfil_ipfw != 0 && dir == PFIL_OUT && ifp != NULL) { error = pfil_run_hooks(&V_link_pfil_hook, mp, ifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* packet consumed by filter */ return (error); } /* Strip off the Ethernet header and keep a copy. */ m_copydata(*mp, 0, ETHER_HDR_LEN, (caddr_t) &eh2); m_adj(*mp, ETHER_HDR_LEN); /* Strip off snap header, if present */ if (snap) { m_copydata(*mp, 0, sizeof(struct llc), (caddr_t) &llc1); m_adj(*mp, sizeof(struct llc)); } /* * Check the IP header for alignment and errors */ if (dir == PFIL_IN) { switch (ether_type) { case ETHERTYPE_IP: error = bridge_ip_checkbasic(mp); break; #ifdef INET6 case ETHERTYPE_IPV6: error = bridge_ip6_checkbasic(mp); break; #endif /* INET6 */ default: error = 0; } if (error) goto bad; } error = 0; /* * Run the packet through pfil */ switch (ether_type) { case ETHERTYPE_IP: /* * Run pfil on the member interface and the bridge, both can * be skipped by clearing pfil_member or pfil_bridge. * * Keep the order: * in_if -> bridge_if -> out_if */ if (V_pfil_bridge && dir == PFIL_OUT && bifp != NULL) error = pfil_run_hooks(&V_inet_pfil_hook, mp, bifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* filter may consume */ break; if (V_pfil_member && ifp != NULL) error = pfil_run_hooks(&V_inet_pfil_hook, mp, ifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* filter may consume */ break; if (V_pfil_bridge && dir == PFIL_IN && bifp != NULL) error = pfil_run_hooks(&V_inet_pfil_hook, mp, bifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* filter may consume */ break; /* check if we need to fragment the packet */ /* bridge_fragment generates a mbuf chain of packets */ /* that already include eth headers */ if (V_pfil_member && ifp != NULL && dir == PFIL_OUT) { i = (*mp)->m_pkthdr.len; if (i > ifp->if_mtu) { error = bridge_fragment(ifp, mp, &eh2, snap, &llc1); return (error); } } /* Recalculate the ip checksum. */ ip = mtod(*mp, struct ip *); hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) goto bad; if (hlen > (*mp)->m_len) { if ((*mp = m_pullup(*mp, hlen)) == NULL) goto bad; ip = mtod(*mp, struct ip *); if (ip == NULL) goto bad; } ip->ip_sum = 0; if (hlen == sizeof(struct ip)) ip->ip_sum = in_cksum_hdr(ip); else ip->ip_sum = in_cksum(*mp, hlen); break; #ifdef INET6 case ETHERTYPE_IPV6: if (V_pfil_bridge && dir == PFIL_OUT && bifp != NULL) error = pfil_run_hooks(&V_inet6_pfil_hook, mp, bifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* filter may consume */ break; if (V_pfil_member && ifp != NULL) error = pfil_run_hooks(&V_inet6_pfil_hook, mp, ifp, dir, 0, NULL); if (*mp == NULL || error != 0) /* filter may consume */ break; if (V_pfil_bridge && dir == PFIL_IN && bifp != NULL) error = pfil_run_hooks(&V_inet6_pfil_hook, mp, bifp, dir, 0, NULL); break; #endif default: error = 0; break; } if (*mp == NULL) return (error); if (error != 0) goto bad; error = -1; /* * Finally, put everything back the way it was and return */ if (snap) { M_PREPEND(*mp, sizeof(struct llc), M_NOWAIT); if (*mp == NULL) return (error); bcopy(&llc1, mtod(*mp, caddr_t), sizeof(struct llc)); } M_PREPEND(*mp, ETHER_HDR_LEN, M_NOWAIT); if (*mp == NULL) return (error); bcopy(&eh2, mtod(*mp, caddr_t), ETHER_HDR_LEN); return (0); bad: m_freem(*mp); *mp = NULL; return (error); } /* * Perform basic checks on header size since * pfil assumes ip_input has already processed * it for it. Cut-and-pasted from ip_input.c. * Given how simple the IPv6 version is, * does the IPv4 version really need to be * this complicated? * * XXX Should we update ipstat here, or not? * XXX Right now we update ipstat but not * XXX csum_counter. */ static int bridge_ip_checkbasic(struct mbuf **mp) { struct mbuf *m = *mp; struct ip *ip; int len, hlen; u_short sum; if (*mp == NULL) return (-1); if (IP_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0) { if ((m = m_copyup(m, sizeof(struct ip), (max_linkhdr + 3) & ~3)) == NULL) { /* XXXJRT new stat, please */ KMOD_IPSTAT_INC(ips_toosmall); goto bad; } } else if (__predict_false(m->m_len < sizeof (struct ip))) { if ((m = m_pullup(m, sizeof (struct ip))) == NULL) { KMOD_IPSTAT_INC(ips_toosmall); goto bad; } } ip = mtod(m, struct ip *); if (ip == NULL) goto bad; if (ip->ip_v != IPVERSION) { KMOD_IPSTAT_INC(ips_badvers); goto bad; } hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) { /* minimum header length */ KMOD_IPSTAT_INC(ips_badhlen); goto bad; } if (hlen > m->m_len) { if ((m = m_pullup(m, hlen)) == NULL) { KMOD_IPSTAT_INC(ips_badhlen); goto bad; } ip = mtod(m, struct ip *); if (ip == NULL) goto bad; } if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); } else { if (hlen == sizeof(struct ip)) { sum = in_cksum_hdr(ip); } else { sum = in_cksum(m, hlen); } } if (sum) { KMOD_IPSTAT_INC(ips_badsum); goto bad; } /* Retrieve the packet length. */ len = ntohs(ip->ip_len); /* * Check for additional length bogosity */ if (len < hlen) { KMOD_IPSTAT_INC(ips_badlen); goto bad; } /* * Check that the amount of data in the buffers * is as at least much as the IP header would have us expect. * Drop packet if shorter than we expect. */ if (m->m_pkthdr.len < len) { KMOD_IPSTAT_INC(ips_tooshort); goto bad; } /* Checks out, proceed */ *mp = m; return (0); bad: *mp = m; return (-1); } #ifdef INET6 /* * Same as above, but for IPv6. * Cut-and-pasted from ip6_input.c. * XXX Should we update ip6stat, or not? */ static int bridge_ip6_checkbasic(struct mbuf **mp) { struct mbuf *m = *mp; struct ip6_hdr *ip6; /* * If the IPv6 header is not aligned, slurp it up into a new * mbuf with space for link headers, in the event we forward * it. Otherwise, if it is aligned, make sure the entire base * IPv6 header is in the first mbuf of the chain. */ if (IP6_HDR_ALIGNED_P(mtod(m, caddr_t)) == 0) { struct ifnet *inifp = m->m_pkthdr.rcvif; if ((m = m_copyup(m, sizeof(struct ip6_hdr), (max_linkhdr + 3) & ~3)) == NULL) { /* XXXJRT new stat, please */ IP6STAT_INC(ip6s_toosmall); in6_ifstat_inc(inifp, ifs6_in_hdrerr); goto bad; } } else if (__predict_false(m->m_len < sizeof(struct ip6_hdr))) { struct ifnet *inifp = m->m_pkthdr.rcvif; if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) { IP6STAT_INC(ip6s_toosmall); in6_ifstat_inc(inifp, ifs6_in_hdrerr); goto bad; } } ip6 = mtod(m, struct ip6_hdr *); if ((ip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) { IP6STAT_INC(ip6s_badvers); in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_hdrerr); goto bad; } /* Checks out, proceed */ *mp = m; return (0); bad: *mp = m; return (-1); } #endif /* INET6 */ /* * bridge_fragment: * * Fragment mbuf chain in multiple packets and prepend ethernet header. */ static int bridge_fragment(struct ifnet *ifp, struct mbuf **mp, struct ether_header *eh, int snap, struct llc *llc) { struct mbuf *m = *mp, *nextpkt = NULL, *mprev = NULL, *mcur = NULL; struct ip *ip; int error = -1; if (m->m_len < sizeof(struct ip) && (m = m_pullup(m, sizeof(struct ip))) == NULL) goto dropit; ip = mtod(m, struct ip *); m->m_pkthdr.csum_flags |= CSUM_IP; error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist); if (error) goto dropit; /* * Walk the chain and re-add the Ethernet header for * each mbuf packet. */ for (mcur = m; mcur; mcur = mcur->m_nextpkt) { nextpkt = mcur->m_nextpkt; mcur->m_nextpkt = NULL; if (snap) { M_PREPEND(mcur, sizeof(struct llc), M_NOWAIT); if (mcur == NULL) { error = ENOBUFS; if (mprev != NULL) mprev->m_nextpkt = nextpkt; goto dropit; } bcopy(llc, mtod(mcur, caddr_t),sizeof(struct llc)); } M_PREPEND(mcur, ETHER_HDR_LEN, M_NOWAIT); if (mcur == NULL) { error = ENOBUFS; if (mprev != NULL) mprev->m_nextpkt = nextpkt; goto dropit; } bcopy(eh, mtod(mcur, caddr_t), ETHER_HDR_LEN); /* * The previous two M_PREPEND could have inserted one or two * mbufs in front so we have to update the previous packet's * m_nextpkt. */ mcur->m_nextpkt = nextpkt; if (mprev != NULL) mprev->m_nextpkt = mcur; else { /* The first mbuf in the original chain needs to be * updated. */ *mp = mcur; } mprev = mcur; } KMOD_IPSTAT_INC(ips_fragmented); return (error); dropit: for (mcur = *mp; mcur; mcur = m) { /* droping the full packet chain */ m = mcur->m_nextpkt; m_freem(mcur); } return (error); } static void bridge_linkstate(struct ifnet *ifp) { struct bridge_softc *sc = ifp->if_bridge; struct bridge_iflist *bif; BRIDGE_LOCK(sc); bif = bridge_lookup_member_if(sc, ifp); if (bif == NULL) { BRIDGE_UNLOCK(sc); return; } bridge_linkcheck(sc); BRIDGE_UNLOCK(sc); bstp_linkstate(&bif->bif_stp); } static void bridge_linkcheck(struct bridge_softc *sc) { struct bridge_iflist *bif; int new_link, hasls; BRIDGE_LOCK_ASSERT(sc); new_link = LINK_STATE_DOWN; hasls = 0; /* Our link is considered up if at least one of our ports is active */ LIST_FOREACH(bif, &sc->sc_iflist, bif_next) { if (bif->bif_ifp->if_capabilities & IFCAP_LINKSTATE) hasls++; if (bif->bif_ifp->if_link_state == LINK_STATE_UP) { new_link = LINK_STATE_UP; break; } } if (!LIST_EMPTY(&sc->sc_iflist) && !hasls) { /* If no interfaces support link-state then we default to up */ new_link = LINK_STATE_UP; } if_link_state_change(sc->sc_ifp, new_link); } Index: head/sys/net/if_clone.c =================================================================== --- head/sys/net/if_clone.c (revision 336675) +++ head/sys/net/if_clone.c (revision 336676) @@ -1,756 +1,756 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2012 Gleb Smirnoff * Copyright (c) 1980, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)if.c 8.5 (Berkeley) 1/9/95 * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Current IF_MAXUNIT expands maximum to 5 characters. */ #define IFCLOSIZ (IFNAMSIZ - 5) /* * Structure describing a `cloning' interface. * * List of locks * (c) const until freeing * (d) driver specific data, may need external protection. * (e) locked by if_cloners_mtx * (i) locked by ifc_mtx mtx */ struct if_clone { char ifc_name[IFCLOSIZ]; /* (c) Name of device, e.g. `gif' */ struct unrhdr *ifc_unrhdr; /* (c) alloc_unr(9) header */ int ifc_maxunit; /* (c) maximum unit number */ int ifc_flags; long ifc_refcnt; /* (i) Reference count. */ LIST_HEAD(, ifnet) ifc_iflist; /* (i) List of cloned interfaces */ struct mtx ifc_mtx; /* Mutex to protect members. */ enum { SIMPLE, ADVANCED } ifc_type; /* (c) */ /* (c) Driver specific cloning functions. Called with no locks held. */ union { struct { /* advanced cloner */ ifc_match_t *_ifc_match; ifc_create_t *_ifc_create; ifc_destroy_t *_ifc_destroy; } A; struct { /* simple cloner */ ifcs_create_t *_ifcs_create; ifcs_destroy_t *_ifcs_destroy; int _ifcs_minifs; /* minimum ifs */ } S; } U; #define ifc_match U.A._ifc_match #define ifc_create U.A._ifc_create #define ifc_destroy U.A._ifc_destroy #define ifcs_create U.S._ifcs_create #define ifcs_destroy U.S._ifcs_destroy #define ifcs_minifs U.S._ifcs_minifs LIST_ENTRY(if_clone) ifc_list; /* (e) On list of cloners */ }; static void if_clone_free(struct if_clone *ifc); static int if_clone_createif(struct if_clone *ifc, char *name, size_t len, caddr_t params); static int ifc_simple_match(struct if_clone *, const char *); static int ifc_simple_create(struct if_clone *, char *, size_t, caddr_t); static int ifc_simple_destroy(struct if_clone *, struct ifnet *); static struct mtx if_cloners_mtx; MTX_SYSINIT(if_cloners_lock, &if_cloners_mtx, "if_cloners lock", MTX_DEF); -static VNET_DEFINE(int, if_cloners_count); +VNET_DEFINE_STATIC(int, if_cloners_count); VNET_DEFINE(LIST_HEAD(, if_clone), if_cloners); #define V_if_cloners_count VNET(if_cloners_count) #define V_if_cloners VNET(if_cloners) #define IF_CLONERS_LOCK_ASSERT() mtx_assert(&if_cloners_mtx, MA_OWNED) #define IF_CLONERS_LOCK() mtx_lock(&if_cloners_mtx) #define IF_CLONERS_UNLOCK() mtx_unlock(&if_cloners_mtx) #define IF_CLONE_LOCK_INIT(ifc) \ mtx_init(&(ifc)->ifc_mtx, "if_clone lock", NULL, MTX_DEF) #define IF_CLONE_LOCK_DESTROY(ifc) mtx_destroy(&(ifc)->ifc_mtx) #define IF_CLONE_LOCK_ASSERT(ifc) mtx_assert(&(ifc)->ifc_mtx, MA_OWNED) #define IF_CLONE_LOCK(ifc) mtx_lock(&(ifc)->ifc_mtx) #define IF_CLONE_UNLOCK(ifc) mtx_unlock(&(ifc)->ifc_mtx) #define IF_CLONE_ADDREF(ifc) \ do { \ IF_CLONE_LOCK(ifc); \ IF_CLONE_ADDREF_LOCKED(ifc); \ IF_CLONE_UNLOCK(ifc); \ } while (0) #define IF_CLONE_ADDREF_LOCKED(ifc) \ do { \ IF_CLONE_LOCK_ASSERT(ifc); \ KASSERT((ifc)->ifc_refcnt >= 0, \ ("negative refcnt %ld", (ifc)->ifc_refcnt)); \ (ifc)->ifc_refcnt++; \ } while (0) #define IF_CLONE_REMREF(ifc) \ do { \ IF_CLONE_LOCK(ifc); \ IF_CLONE_REMREF_LOCKED(ifc); \ } while (0) #define IF_CLONE_REMREF_LOCKED(ifc) \ do { \ IF_CLONE_LOCK_ASSERT(ifc); \ KASSERT((ifc)->ifc_refcnt > 0, \ ("bogus refcnt %ld", (ifc)->ifc_refcnt)); \ if (--(ifc)->ifc_refcnt == 0) { \ IF_CLONE_UNLOCK(ifc); \ if_clone_free(ifc); \ } else { \ /* silently free the lock */ \ IF_CLONE_UNLOCK(ifc); \ } \ } while (0) #define IFC_IFLIST_INSERT(_ifc, _ifp) \ LIST_INSERT_HEAD(&_ifc->ifc_iflist, _ifp, if_clones) #define IFC_IFLIST_REMOVE(_ifc, _ifp) \ LIST_REMOVE(_ifp, if_clones) static MALLOC_DEFINE(M_CLONE, "clone", "interface cloning framework"); void vnet_if_clone_init(void) { LIST_INIT(&V_if_cloners); } /* * Lookup and create a clone network interface. */ int if_clone_create(char *name, size_t len, caddr_t params) { struct if_clone *ifc; /* Try to find an applicable cloner for this request */ IF_CLONERS_LOCK(); LIST_FOREACH(ifc, &V_if_cloners, ifc_list) if (ifc->ifc_type == SIMPLE) { if (ifc_simple_match(ifc, name)) break; } else { if (ifc->ifc_match(ifc, name)) break; } #ifdef VIMAGE if (ifc == NULL && !IS_DEFAULT_VNET(curvnet)) { CURVNET_SET_QUIET(vnet0); LIST_FOREACH(ifc, &V_if_cloners, ifc_list) if (ifc->ifc_type == SIMPLE) { if (ifc_simple_match(ifc, name)) break; } else { if (ifc->ifc_match(ifc, name)) break; } CURVNET_RESTORE(); } #endif IF_CLONERS_UNLOCK(); if (ifc == NULL) return (EINVAL); return (if_clone_createif(ifc, name, len, params)); } /* * Create a clone network interface. */ static int if_clone_createif(struct if_clone *ifc, char *name, size_t len, caddr_t params) { int err; struct ifnet *ifp; if (ifunit(name) != NULL) return (EEXIST); if (ifc->ifc_type == SIMPLE) err = ifc_simple_create(ifc, name, len, params); else err = (*ifc->ifc_create)(ifc, name, len, params); if (!err) { ifp = ifunit(name); if (ifp == NULL) panic("%s: lookup failed for %s", __func__, name); if ((ifc->ifc_flags & IFC_NOGROUP) == 0) if_addgroup(ifp, ifc->ifc_name); IF_CLONE_LOCK(ifc); IFC_IFLIST_INSERT(ifc, ifp); IF_CLONE_UNLOCK(ifc); } return (err); } /* * Lookup and destroy a clone network interface. */ int if_clone_destroy(const char *name) { int err; struct if_clone *ifc; struct ifnet *ifp; ifp = ifunit_ref(name); if (ifp == NULL) return (ENXIO); /* Find the cloner for this interface */ IF_CLONERS_LOCK(); LIST_FOREACH(ifc, &V_if_cloners, ifc_list) { if (strcmp(ifc->ifc_name, ifp->if_dname) == 0) { break; } } #ifdef VIMAGE if (ifc == NULL && !IS_DEFAULT_VNET(curvnet)) { CURVNET_SET_QUIET(vnet0); LIST_FOREACH(ifc, &V_if_cloners, ifc_list) if (ifc->ifc_type == SIMPLE) { if (ifc_simple_match(ifc, name)) break; } else { if (ifc->ifc_match(ifc, name)) break; } CURVNET_RESTORE(); } #endif IF_CLONERS_UNLOCK(); if (ifc == NULL) { if_rele(ifp); return (EINVAL); } err = if_clone_destroyif(ifc, ifp); if_rele(ifp); return err; } /* * Destroy a clone network interface. */ int if_clone_destroyif(struct if_clone *ifc, struct ifnet *ifp) { int err; struct ifnet *ifcifp; if (ifc->ifc_type == ADVANCED && ifc->ifc_destroy == NULL) return(EOPNOTSUPP); /* * Given that the cloned ifnet might be attached to a different * vnet from where its cloner was registered, we have to * switch to the vnet context of the target vnet. */ CURVNET_SET_QUIET(ifp->if_vnet); IF_CLONE_LOCK(ifc); LIST_FOREACH(ifcifp, &ifc->ifc_iflist, if_clones) { if (ifcifp == ifp) { IFC_IFLIST_REMOVE(ifc, ifp); break; } } IF_CLONE_UNLOCK(ifc); if (ifcifp == NULL) { CURVNET_RESTORE(); return (ENXIO); /* ifp is not on the list. */ } if ((ifc->ifc_flags & IFC_NOGROUP) == 0) if_delgroup(ifp, ifc->ifc_name); if (ifc->ifc_type == SIMPLE) err = ifc_simple_destroy(ifc, ifp); else err = (*ifc->ifc_destroy)(ifc, ifp); if (err != 0) { if ((ifc->ifc_flags & IFC_NOGROUP) == 0) if_addgroup(ifp, ifc->ifc_name); IF_CLONE_LOCK(ifc); IFC_IFLIST_INSERT(ifc, ifp); IF_CLONE_UNLOCK(ifc); } CURVNET_RESTORE(); return (err); } static struct if_clone * if_clone_alloc(const char *name, int maxunit) { struct if_clone *ifc; KASSERT(name != NULL, ("%s: no name\n", __func__)); ifc = malloc(sizeof(struct if_clone), M_CLONE, M_WAITOK | M_ZERO); strncpy(ifc->ifc_name, name, IFCLOSIZ-1); IF_CLONE_LOCK_INIT(ifc); IF_CLONE_ADDREF(ifc); ifc->ifc_maxunit = maxunit ? maxunit : IF_MAXUNIT; ifc->ifc_unrhdr = new_unrhdr(0, ifc->ifc_maxunit, &ifc->ifc_mtx); LIST_INIT(&ifc->ifc_iflist); return (ifc); } static int if_clone_attach(struct if_clone *ifc) { struct if_clone *ifc1; IF_CLONERS_LOCK(); LIST_FOREACH(ifc1, &V_if_cloners, ifc_list) if (strcmp(ifc->ifc_name, ifc1->ifc_name) == 0) { IF_CLONERS_UNLOCK(); IF_CLONE_REMREF(ifc); return (EEXIST); } LIST_INSERT_HEAD(&V_if_cloners, ifc, ifc_list); V_if_cloners_count++; IF_CLONERS_UNLOCK(); return (0); } struct if_clone * if_clone_advanced(const char *name, u_int maxunit, ifc_match_t match, ifc_create_t create, ifc_destroy_t destroy) { struct if_clone *ifc; ifc = if_clone_alloc(name, maxunit); ifc->ifc_type = ADVANCED; ifc->ifc_match = match; ifc->ifc_create = create; ifc->ifc_destroy = destroy; if (if_clone_attach(ifc) != 0) return (NULL); EVENTHANDLER_INVOKE(if_clone_event, ifc); return (ifc); } struct if_clone * if_clone_simple(const char *name, ifcs_create_t create, ifcs_destroy_t destroy, u_int minifs) { struct if_clone *ifc; u_int unit; ifc = if_clone_alloc(name, 0); ifc->ifc_type = SIMPLE; ifc->ifcs_create = create; ifc->ifcs_destroy = destroy; ifc->ifcs_minifs = minifs; if (if_clone_attach(ifc) != 0) return (NULL); for (unit = 0; unit < minifs; unit++) { char name[IFNAMSIZ]; int error __unused; snprintf(name, IFNAMSIZ, "%s%d", ifc->ifc_name, unit); error = if_clone_createif(ifc, name, IFNAMSIZ, NULL); KASSERT(error == 0, ("%s: failed to create required interface %s", __func__, name)); } EVENTHANDLER_INVOKE(if_clone_event, ifc); return (ifc); } /* * Unregister a network interface cloner. */ void if_clone_detach(struct if_clone *ifc) { IF_CLONERS_LOCK(); LIST_REMOVE(ifc, ifc_list); V_if_cloners_count--; IF_CLONERS_UNLOCK(); /* Allow all simples to be destroyed */ if (ifc->ifc_type == SIMPLE) ifc->ifcs_minifs = 0; /* destroy all interfaces for this cloner */ while (!LIST_EMPTY(&ifc->ifc_iflist)) if_clone_destroyif(ifc, LIST_FIRST(&ifc->ifc_iflist)); IF_CLONE_REMREF(ifc); } static void if_clone_free(struct if_clone *ifc) { KASSERT(LIST_EMPTY(&ifc->ifc_iflist), ("%s: ifc_iflist not empty", __func__)); IF_CLONE_LOCK_DESTROY(ifc); delete_unrhdr(ifc->ifc_unrhdr); free(ifc, M_CLONE); } /* * Provide list of interface cloners to userspace. */ int if_clone_list(struct if_clonereq *ifcr) { char *buf, *dst, *outbuf = NULL; struct if_clone *ifc; int buf_count, count, err = 0; if (ifcr->ifcr_count < 0) return (EINVAL); IF_CLONERS_LOCK(); /* * Set our internal output buffer size. We could end up not * reporting a cloner that is added between the unlock and lock * below, but that's not a major problem. Not caping our * allocation to the number of cloners actually in the system * could be because that would let arbitrary users cause us to * allocate arbitrary amounts of kernel memory. */ buf_count = (V_if_cloners_count < ifcr->ifcr_count) ? V_if_cloners_count : ifcr->ifcr_count; IF_CLONERS_UNLOCK(); outbuf = malloc(IFNAMSIZ*buf_count, M_CLONE, M_WAITOK | M_ZERO); IF_CLONERS_LOCK(); ifcr->ifcr_total = V_if_cloners_count; if ((dst = ifcr->ifcr_buffer) == NULL) { /* Just asking how many there are. */ goto done; } count = (V_if_cloners_count < buf_count) ? V_if_cloners_count : buf_count; for (ifc = LIST_FIRST(&V_if_cloners), buf = outbuf; ifc != NULL && count != 0; ifc = LIST_NEXT(ifc, ifc_list), count--, buf += IFNAMSIZ) { strlcpy(buf, ifc->ifc_name, IFNAMSIZ); } done: IF_CLONERS_UNLOCK(); if (err == 0 && dst != NULL) err = copyout(outbuf, dst, buf_count*IFNAMSIZ); if (outbuf != NULL) free(outbuf, M_CLONE); return (err); } /* * if_clone_findifc() looks up ifnet from the current * cloner list, and returns ifc if found. Note that ifc_refcnt * is incremented. */ struct if_clone * if_clone_findifc(struct ifnet *ifp) { struct if_clone *ifc, *ifc0; struct ifnet *ifcifp; ifc0 = NULL; IF_CLONERS_LOCK(); LIST_FOREACH(ifc, &V_if_cloners, ifc_list) { IF_CLONE_LOCK(ifc); LIST_FOREACH(ifcifp, &ifc->ifc_iflist, if_clones) { if (ifp == ifcifp) { ifc0 = ifc; IF_CLONE_ADDREF_LOCKED(ifc); break; } } IF_CLONE_UNLOCK(ifc); if (ifc0 != NULL) break; } IF_CLONERS_UNLOCK(); return (ifc0); } /* * if_clone_addgroup() decrements ifc_refcnt because it is called after * if_clone_findifc(). */ void if_clone_addgroup(struct ifnet *ifp, struct if_clone *ifc) { if ((ifc->ifc_flags & IFC_NOGROUP) == 0) { if_addgroup(ifp, ifc->ifc_name); IF_CLONE_REMREF(ifc); } } /* * A utility function to extract unit numbers from interface names of * the form name###. * * Returns 0 on success and an error on failure. */ int ifc_name2unit(const char *name, int *unit) { const char *cp; int cutoff = INT_MAX / 10; int cutlim = INT_MAX % 10; for (cp = name; *cp != '\0' && (*cp < '0' || *cp > '9'); cp++); if (*cp == '\0') { *unit = -1; } else if (cp[0] == '0' && cp[1] != '\0') { /* Disallow leading zeroes. */ return (EINVAL); } else { for (*unit = 0; *cp != '\0'; cp++) { if (*cp < '0' || *cp > '9') { /* Bogus unit number. */ return (EINVAL); } if (*unit > cutoff || (*unit == cutoff && *cp - '0' > cutlim)) return (EINVAL); *unit = (*unit * 10) + (*cp - '0'); } } return (0); } static int ifc_alloc_unit_specific(struct if_clone *ifc, int *unit) { char name[IFNAMSIZ]; if (*unit > ifc->ifc_maxunit) return (ENOSPC); if (alloc_unr_specific(ifc->ifc_unrhdr, *unit) == -1) return (EEXIST); snprintf(name, IFNAMSIZ, "%s%d", ifc->ifc_name, *unit); if (ifunit(name) != NULL) { free_unr(ifc->ifc_unrhdr, *unit); return (EEXIST); } IF_CLONE_ADDREF(ifc); return (0); } static int ifc_alloc_unit_next(struct if_clone *ifc, int *unit) { int error; *unit = alloc_unr(ifc->ifc_unrhdr); if (*unit == -1) return (ENOSPC); free_unr(ifc->ifc_unrhdr, *unit); for (;;) { error = ifc_alloc_unit_specific(ifc, unit); if (error != EEXIST) break; (*unit)++; } return (error); } int ifc_alloc_unit(struct if_clone *ifc, int *unit) { if (*unit < 0) return (ifc_alloc_unit_next(ifc, unit)); else return (ifc_alloc_unit_specific(ifc, unit)); } void ifc_free_unit(struct if_clone *ifc, int unit) { free_unr(ifc->ifc_unrhdr, unit); IF_CLONE_REMREF(ifc); } static int ifc_simple_match(struct if_clone *ifc, const char *name) { const char *cp; int i; /* Match the name */ for (cp = name, i = 0; i < strlen(ifc->ifc_name); i++, cp++) { if (ifc->ifc_name[i] != *cp) return (0); } /* Make sure there's a unit number or nothing after the name */ for (; *cp != '\0'; cp++) { if (*cp < '0' || *cp > '9') return (0); } return (1); } static int ifc_simple_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) { char *dp; int wildcard; int unit; int err; err = ifc_name2unit(name, &unit); if (err != 0) return (err); wildcard = (unit < 0); err = ifc_alloc_unit(ifc, &unit); if (err != 0) return (err); err = ifc->ifcs_create(ifc, unit, params); if (err != 0) { ifc_free_unit(ifc, unit); return (err); } /* In the wildcard case, we need to update the name. */ if (wildcard) { for (dp = name; *dp != '\0'; dp++); if (snprintf(dp, len - (dp-name), "%d", unit) > len - (dp-name) - 1) { /* * This can only be a programmer error and * there's no straightforward way to recover if * it happens. */ panic("if_clone_create(): interface name too long"); } } return (0); } static int ifc_simple_destroy(struct if_clone *ifc, struct ifnet *ifp) { int unit; unit = ifp->if_dunit; if (unit < ifc->ifcs_minifs) return (EINVAL); ifc->ifcs_destroy(ifp); ifc_free_unit(ifc, unit); return (0); } const char * ifc_name(struct if_clone *ifc) { return (ifc->ifc_name); } void ifc_flags_set(struct if_clone *ifc, int flags) { ifc->ifc_flags = flags; } int ifc_flags_get(struct if_clone *ifc) { return (ifc->ifc_flags); } Index: head/sys/net/if_disc.c =================================================================== --- head/sys/net/if_disc.c (revision 336675) +++ head/sys/net/if_disc.c (revision 336676) @@ -1,246 +1,246 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)if_loop.c 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ /* * Discard interface driver for protocol testing and timing. * (Based on the loopback.) */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_inet.h" #include "opt_inet6.h" #ifdef TINY_DSMTU #define DSMTU (1024+512) #else #define DSMTU 65532 #endif struct disc_softc { struct ifnet *sc_ifp; }; static int discoutput(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static int discioctl(struct ifnet *, u_long, caddr_t); static int disc_clone_create(struct if_clone *, int, caddr_t); static void disc_clone_destroy(struct ifnet *); static const char discname[] = "disc"; static MALLOC_DEFINE(M_DISC, discname, "Discard interface"); -static VNET_DEFINE(struct if_clone *, disc_cloner); +VNET_DEFINE_STATIC(struct if_clone *, disc_cloner); #define V_disc_cloner VNET(disc_cloner) static int disc_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ifnet *ifp; struct disc_softc *sc; sc = malloc(sizeof(struct disc_softc), M_DISC, M_WAITOK | M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_LOOP); if (ifp == NULL) { free(sc, M_DISC); return (ENOSPC); } ifp->if_softc = sc; if_initname(ifp, discname, unit); ifp->if_mtu = DSMTU; /* * IFF_LOOPBACK should not be removed from disc's flags because * it controls what PF-specific routes are magically added when * a network address is assigned to the interface. Things just * won't work as intended w/o such routes because the output * interface selection for a packet is totally route-driven. * A valid alternative to IFF_LOOPBACK can be IFF_BROADCAST or * IFF_POINTOPOINT, but it would result in different properties * of the interface. */ ifp->if_flags = IFF_LOOPBACK | IFF_MULTICAST; ifp->if_drv_flags = IFF_DRV_RUNNING; ifp->if_ioctl = discioctl; ifp->if_output = discoutput; ifp->if_hdrlen = 0; ifp->if_addrlen = 0; ifp->if_snd.ifq_maxlen = 20; if_attach(ifp); bpfattach(ifp, DLT_NULL, sizeof(u_int32_t)); return (0); } static void disc_clone_destroy(struct ifnet *ifp) { struct disc_softc *sc; sc = ifp->if_softc; bpfdetach(ifp); if_detach(ifp); if_free(ifp); free(sc, M_DISC); } static void vnet_disc_init(const void *unused __unused) { V_disc_cloner = if_clone_simple(discname, disc_clone_create, disc_clone_destroy, 0); } VNET_SYSINIT(vnet_disc_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_disc_init, NULL); static void vnet_disc_uninit(const void *unused __unused) { if_clone_detach(V_disc_cloner); } VNET_SYSUNINIT(vnet_disc_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_disc_uninit, NULL); static int disc_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t disc_mod = { "if_disc", disc_modevent, NULL }; DECLARE_MODULE(if_disc, disc_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); static int discoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { u_int32_t af; M_ASSERTPKTHDR(m); /* BPF writes need to be handled specially. */ if (dst->sa_family == AF_UNSPEC) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; if (bpf_peers_present(ifp->if_bpf)) bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m); m->m_pkthdr.rcvif = ifp; if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); m_freem(m); return (0); } /* * Process an ioctl request. */ static int discioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; int error = 0; switch (cmd) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; /* * Everything else is done at a higher level. */ break; case SIOCADDMULTI: case SIOCDELMULTI: if (ifr == NULL) { error = EAFNOSUPPORT; /* XXX */ break; } switch (ifr->ifr_addr.sa_family) { #ifdef INET case AF_INET: break; #endif #ifdef INET6 case AF_INET6: break; #endif default: error = EAFNOSUPPORT; break; } break; case SIOCSIFMTU: ifp->if_mtu = ifr->ifr_mtu; break; default: error = EINVAL; } return (error); } Index: head/sys/net/if_edsc.c =================================================================== --- head/sys/net/if_edsc.c (revision 336675) +++ head/sys/net/if_edsc.c (revision 336676) @@ -1,371 +1,371 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following edsclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following edsclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE EDSCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)if_loop.c 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ /* * Discard interface driver for protocol testing and timing. * Mimics an Ethernet device so that VLANs can be attached to it etc. */ #include /* types, important constants */ #include /* SYSINIT for load-time initializations */ #include /* malloc(9) */ #include /* module(9) */ #include /* mbuf(9) */ #include /* struct ifreq */ #include /* socket ioctl's */ /* #include if you need printf(9) or other all-purpose globals */ #include /* bpf(9) */ #include /* Ethernet related constants and types */ #include #include /* basic part of ifnet(9) */ #include /* network interface cloning */ #include /* IFT_ETHER and friends */ #include /* kernel-only part of ifnet(9) */ #include static const char edscname[] = "edsc"; /* * Software configuration of an interface specific to this device type. */ struct edsc_softc { struct ifnet *sc_ifp; /* ptr to generic interface configuration */ /* * A non-null driver can keep various things here, for instance, * the hardware revision, cached values of write-only registers, etc. */ }; /* * Attach to the interface cloning framework. */ -static VNET_DEFINE(struct if_clone *, edsc_cloner); +VNET_DEFINE_STATIC(struct if_clone *, edsc_cloner); #define V_edsc_cloner VNET(edsc_cloner) static int edsc_clone_create(struct if_clone *, int, caddr_t); static void edsc_clone_destroy(struct ifnet *); /* * Interface driver methods. */ static void edsc_init(void *dummy); /* static void edsc_input(struct ifnet *ifp, struct mbuf *m); would be here */ static int edsc_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data); static void edsc_start(struct ifnet *ifp); /* * We'll allocate softc instances from this. */ static MALLOC_DEFINE(M_EDSC, edscname, "Ethernet discard interface"); /* * Create an interface instance. */ static int edsc_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct edsc_softc *sc; struct ifnet *ifp; static u_char eaddr[ETHER_ADDR_LEN]; /* 0:0:0:0:0:0 */ /* * Allocate soft and ifnet structures. Link each to the other. */ sc = malloc(sizeof(struct edsc_softc), M_EDSC, M_WAITOK | M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { free(sc, M_EDSC); return (ENOSPC); } ifp->if_softc = sc; /* * Get a name for this particular interface in its ifnet structure. */ if_initname(ifp, edscname, unit); /* * Typical Ethernet interface flags: we can do broadcast and * multicast but can't hear our own broadcasts or multicasts. */ ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX; /* * We can pretent we have the whole set of hardware features * because we just discard all packets we get from the upper layer. * However, the features are disabled initially. They can be * enabled via edsc_ioctl() when needed. */ ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | IFCAP_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU; ifp->if_capenable = 0; /* * Set the interface driver methods. */ ifp->if_init = edsc_init; /* ifp->if_input = edsc_input; */ ifp->if_ioctl = edsc_ioctl; ifp->if_start = edsc_start; /* * Set the maximum output queue length from the global parameter. */ ifp->if_snd.ifq_maxlen = ifqmaxlen; /* * Do ifnet initializations common to all Ethernet drivers * and attach to the network interface framework. * TODO: Pick a non-zero link level address. */ ether_ifattach(ifp, eaddr); /* * Now we can mark the interface as running, i.e., ready * for operation. */ ifp->if_drv_flags |= IFF_DRV_RUNNING; return (0); } /* * Destroy an interface instance. */ static void edsc_clone_destroy(struct ifnet *ifp) { struct edsc_softc *sc = ifp->if_softc; /* * Detach from the network interface framework. */ ether_ifdetach(ifp); /* * Free memory occupied by ifnet and softc. */ if_free(ifp); free(sc, M_EDSC); } /* * This method is invoked from ether_ioctl() when it's time * to bring up the hardware. */ static void edsc_init(void *dummy) { #if 0 /* what a hardware driver would do here... */ struct edsc_soft *sc = (struct edsc_softc *)dummy; struct ifnet *ifp = sc->sc_ifp; /* blah-blah-blah */ #endif } /* * Network interfaces are controlled via the ioctl(2) syscall. */ static int edsc_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; switch (cmd) { case SIOCSIFCAP: #if 1 /* * Just turn on any capabilities requested. * The generic ifioctl() function has already made sure * that they are supported, i.e., set in if_capabilities. */ ifp->if_capenable = ifr->ifr_reqcap; #else /* * A h/w driver would need to analyze the requested * bits and program the hardware, e.g.: */ mask = ifp->if_capenable ^ ifr->ifr_reqcap; if (mask & IFCAP_VLAN_HWTAGGING) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) /* blah-blah-blah */ else /* etc-etc-etc */ } #endif break; default: /* * Offload the rest onto the common Ethernet handler. */ return (ether_ioctl(ifp, cmd, data)); } return (0); } /* * Process the output queue. */ static void edsc_start(struct ifnet *ifp) { struct mbuf *m; /* * A hardware interface driver can set IFF_DRV_OACTIVE * in ifp->if_drv_flags: * * ifp->if_drv_flags |= IFF_DRV_OACTIVE; * * to prevent if_start from being invoked again while the * transmission is under way. The flag is to protect the * device's transmitter, not the method itself. The output * queue is locked and several threads can process it in * parallel safely, so the driver can use other means to * serialize access to the transmitter. * * If using IFF_DRV_OACTIVE, the driver should clear the flag * not earlier than the current transmission is complete, e.g., * upon an interrupt from the device, not just before returning * from if_start. This method merely starts the transmission, * which may proceed asynchronously. */ /* * We loop getting packets from the queue until it's empty. * A h/w driver would loop until the device can accept more * data into its buffer, or while there are free transmit * descriptors, or whatever. */ for (;;) { /* * Try to dequeue one packet. Stop if the queue is empty. * Use IF_DEQUEUE() here if ALTQ(9) support is unneeded. */ IFQ_DEQUEUE(&ifp->if_snd, m); if (m == NULL) break; /* * Let bpf(9) at the packet. */ BPF_MTAP(ifp, m); /* * Update the interface counters. */ if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* * Finally, just drop the packet. * TODO: Reply to ARP requests unless IFF_NOARP is set. */ m_freem(m); } /* * ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; * would be here only if the transmission were synchronous. */ } static void vnet_edsc_init(const void *unused __unused) { /* * Connect to the network interface cloning framework. * The last argument is the number of units to be created * from the outset. It's also the minimum number of units * allowed. We don't want any units created as soon as the * driver is loaded. */ V_edsc_cloner = if_clone_simple(edscname, edsc_clone_create, edsc_clone_destroy, 0); } VNET_SYSINIT(vnet_edsc_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_edsc_init, NULL); static void vnet_edsc_uninit(const void *unused __unused) { /* * Disconnect from the cloning framework. * Existing interfaces will be disposed of properly. */ if_clone_detach(V_edsc_cloner); } VNET_SYSUNINIT(vnet_edsc_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_edsc_uninit, NULL); /* * This function provides handlers for module events, namely load and unload. */ static int edsc_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: /* * There are other event types, but we don't handle them. * See module(9). */ return (EOPNOTSUPP); } return (0); } static moduledata_t edsc_mod = { "if_edsc", /* name */ edsc_modevent, /* event handler */ NULL /* additional data */ }; DECLARE_MODULE(if_edsc, edsc_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); Index: head/sys/net/if_enc.c =================================================================== --- head/sys/net/if_enc.c (revision 336675) +++ head/sys/net/if_enc.c (revision 336676) @@ -1,445 +1,445 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2006 The FreeBSD Project. * Copyright (c) 2015 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #endif #include #include #define ENCMTU (1024+512) /* XXX this define must have the same value as in OpenBSD */ #define M_CONF 0x0400 /* payload was encrypted (ESP-transport) */ #define M_AUTH 0x0800 /* payload was authenticated (AH or ESP auth) */ #define M_AUTH_AH 0x2000 /* header was authenticated (AH) */ struct enchdr { u_int32_t af; u_int32_t spi; u_int32_t flags; }; struct enc_softc { struct ifnet *sc_ifp; }; -static VNET_DEFINE(struct enc_softc *, enc_sc); +VNET_DEFINE_STATIC(struct enc_softc *, enc_sc); #define V_enc_sc VNET(enc_sc) -static VNET_DEFINE(struct if_clone *, enc_cloner); +VNET_DEFINE_STATIC(struct if_clone *, enc_cloner); #define V_enc_cloner VNET(enc_cloner) static int enc_ioctl(struct ifnet *, u_long, caddr_t); static int enc_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static int enc_clone_create(struct if_clone *, int, caddr_t); static void enc_clone_destroy(struct ifnet *); static int enc_add_hhooks(struct enc_softc *); static void enc_remove_hhooks(struct enc_softc *); static const char encname[] = "enc"; #define IPSEC_ENC_AFTER_PFIL 0x04 /* * Before and after are relative to when we are stripping the * outer IP header. * * AFTER_PFIL flag used only for bpf_mask_*. It enables BPF capturing * after PFIL hook execution. It might be useful when PFIL hook does * some changes to the packet, e.g. address translation. If PFIL hook * consumes mbuf, nothing will be captured. */ -static VNET_DEFINE(int, filter_mask_in) = IPSEC_ENC_BEFORE; -static VNET_DEFINE(int, bpf_mask_in) = IPSEC_ENC_BEFORE; -static VNET_DEFINE(int, filter_mask_out) = IPSEC_ENC_BEFORE; -static VNET_DEFINE(int, bpf_mask_out) = IPSEC_ENC_BEFORE | IPSEC_ENC_AFTER; +VNET_DEFINE_STATIC(int, filter_mask_in) = IPSEC_ENC_BEFORE; +VNET_DEFINE_STATIC(int, bpf_mask_in) = IPSEC_ENC_BEFORE; +VNET_DEFINE_STATIC(int, filter_mask_out) = IPSEC_ENC_BEFORE; +VNET_DEFINE_STATIC(int, bpf_mask_out) = IPSEC_ENC_BEFORE | IPSEC_ENC_AFTER; #define V_filter_mask_in VNET(filter_mask_in) #define V_bpf_mask_in VNET(bpf_mask_in) #define V_filter_mask_out VNET(filter_mask_out) #define V_bpf_mask_out VNET(bpf_mask_out) static SYSCTL_NODE(_net, OID_AUTO, enc, CTLFLAG_RW, 0, "enc sysctl"); static SYSCTL_NODE(_net_enc, OID_AUTO, in, CTLFLAG_RW, 0, "enc input sysctl"); static SYSCTL_NODE(_net_enc, OID_AUTO, out, CTLFLAG_RW, 0, "enc output sysctl"); SYSCTL_INT(_net_enc_in, OID_AUTO, ipsec_filter_mask, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(filter_mask_in), 0, "IPsec input firewall filter mask"); SYSCTL_INT(_net_enc_in, OID_AUTO, ipsec_bpf_mask, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(bpf_mask_in), 0, "IPsec input bpf mask"); SYSCTL_INT(_net_enc_out, OID_AUTO, ipsec_filter_mask, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(filter_mask_out), 0, "IPsec output firewall filter mask"); SYSCTL_INT(_net_enc_out, OID_AUTO, ipsec_bpf_mask, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(bpf_mask_out), 0, "IPsec output bpf mask"); static void enc_clone_destroy(struct ifnet *ifp) { struct enc_softc *sc; sc = ifp->if_softc; KASSERT(sc == V_enc_sc, ("sc != ifp->if_softc")); bpfdetach(ifp); if_detach(ifp); if_free(ifp); free(sc, M_DEVBUF); V_enc_sc = NULL; } static int enc_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ifnet *ifp; struct enc_softc *sc; sc = malloc(sizeof(struct enc_softc), M_DEVBUF, M_WAITOK | M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_ENC); if (ifp == NULL) { free(sc, M_DEVBUF); return (ENOSPC); } if (V_enc_sc != NULL) { if_free(ifp); free(sc, M_DEVBUF); return (EEXIST); } V_enc_sc = sc; if_initname(ifp, encname, unit); ifp->if_mtu = ENCMTU; ifp->if_ioctl = enc_ioctl; ifp->if_output = enc_output; ifp->if_softc = sc; if_attach(ifp); bpfattach(ifp, DLT_ENC, sizeof(struct enchdr)); return (0); } static int enc_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { m_freem(m); return (0); } static int enc_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { if (cmd != SIOCSIFFLAGS) return (EINVAL); if (ifp->if_flags & IFF_UP) ifp->if_drv_flags |= IFF_DRV_RUNNING; else ifp->if_drv_flags &= ~IFF_DRV_RUNNING; return (0); } static void enc_bpftap(struct ifnet *ifp, struct mbuf *m, const struct secasvar *sav, int32_t hhook_type, uint8_t enc, uint8_t af) { struct enchdr hdr; if (hhook_type == HHOOK_TYPE_IPSEC_IN && (enc & V_bpf_mask_in) == 0) return; else if (hhook_type == HHOOK_TYPE_IPSEC_OUT && (enc & V_bpf_mask_out) == 0) return; if (bpf_peers_present(ifp->if_bpf) == 0) return; hdr.af = af; hdr.spi = sav->spi; hdr.flags = 0; if (sav->alg_enc != SADB_EALG_NONE) hdr.flags |= M_CONF; if (sav->alg_auth != SADB_AALG_NONE) hdr.flags |= M_AUTH; bpf_mtap2(ifp->if_bpf, &hdr, sizeof(hdr), m); } /* * One helper hook function is used by any hook points. * + from hhook_type we can determine the packet direction: * HHOOK_TYPE_IPSEC_IN or HHOOK_TYPE_IPSEC_OUT; * + from hhook_id we can determine address family: AF_INET or AF_INET6; * + udata contains pointer to enc_softc; * + ctx_data contains pointer to struct ipsec_ctx_data. */ static int enc_hhook(int32_t hhook_type, int32_t hhook_id, void *udata, void *ctx_data, void *hdata, struct osd *hosd) { struct ipsec_ctx_data *ctx; struct enc_softc *sc; struct ifnet *ifp, *rcvif; struct pfil_head *ph; int pdir; sc = (struct enc_softc *)udata; ifp = sc->sc_ifp; if ((ifp->if_flags & IFF_UP) == 0) return (0); ctx = (struct ipsec_ctx_data *)ctx_data; /* XXX: wrong hook point was used by caller? */ if (ctx->af != hhook_id) return (EPFNOSUPPORT); enc_bpftap(ifp, *ctx->mp, ctx->sav, hhook_type, ctx->enc, ctx->af); switch (hhook_type) { case HHOOK_TYPE_IPSEC_IN: if (ctx->enc == IPSEC_ENC_BEFORE) { /* Do accounting only once */ if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, (*ctx->mp)->m_pkthdr.len); } if ((ctx->enc & V_filter_mask_in) == 0) return (0); /* skip pfil processing */ pdir = PFIL_IN; break; case HHOOK_TYPE_IPSEC_OUT: if (ctx->enc == IPSEC_ENC_BEFORE) { /* Do accounting only once */ if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, (*ctx->mp)->m_pkthdr.len); } if ((ctx->enc & V_filter_mask_out) == 0) return (0); /* skip pfil processing */ pdir = PFIL_OUT; break; default: return (EINVAL); } switch (hhook_id) { #ifdef INET case AF_INET: ph = &V_inet_pfil_hook; break; #endif #ifdef INET6 case AF_INET6: ph = &V_inet6_pfil_hook; break; #endif default: ph = NULL; } if (ph == NULL || !PFIL_HOOKED(ph)) return (0); /* Make a packet looks like it was received on enc(4) */ rcvif = (*ctx->mp)->m_pkthdr.rcvif; (*ctx->mp)->m_pkthdr.rcvif = ifp; if (pfil_run_hooks(ph, ctx->mp, ifp, pdir, 0, ctx->inp) != 0 || *ctx->mp == NULL) { *ctx->mp = NULL; /* consumed by filter */ return (EACCES); } (*ctx->mp)->m_pkthdr.rcvif = rcvif; enc_bpftap(ifp, *ctx->mp, ctx->sav, hhook_type, IPSEC_ENC_AFTER_PFIL, ctx->af); return (0); } static int enc_add_hhooks(struct enc_softc *sc) { struct hookinfo hki; int error; error = EPFNOSUPPORT; hki.hook_func = enc_hhook; hki.hook_helper = NULL; hki.hook_udata = sc; #ifdef INET hki.hook_id = AF_INET; hki.hook_type = HHOOK_TYPE_IPSEC_IN; error = hhook_add_hook(V_ipsec_hhh_in[HHOOK_IPSEC_INET], &hki, HHOOK_WAITOK); if (error != 0) return (error); hki.hook_type = HHOOK_TYPE_IPSEC_OUT; error = hhook_add_hook(V_ipsec_hhh_out[HHOOK_IPSEC_INET], &hki, HHOOK_WAITOK); if (error != 0) return (error); #endif #ifdef INET6 hki.hook_id = AF_INET6; hki.hook_type = HHOOK_TYPE_IPSEC_IN; error = hhook_add_hook(V_ipsec_hhh_in[HHOOK_IPSEC_INET6], &hki, HHOOK_WAITOK); if (error != 0) return (error); hki.hook_type = HHOOK_TYPE_IPSEC_OUT; error = hhook_add_hook(V_ipsec_hhh_out[HHOOK_IPSEC_INET6], &hki, HHOOK_WAITOK); if (error != 0) return (error); #endif return (error); } static void enc_remove_hhooks(struct enc_softc *sc) { struct hookinfo hki; hki.hook_func = enc_hhook; hki.hook_helper = NULL; hki.hook_udata = sc; #ifdef INET hki.hook_id = AF_INET; hki.hook_type = HHOOK_TYPE_IPSEC_IN; hhook_remove_hook(V_ipsec_hhh_in[HHOOK_IPSEC_INET], &hki); hki.hook_type = HHOOK_TYPE_IPSEC_OUT; hhook_remove_hook(V_ipsec_hhh_out[HHOOK_IPSEC_INET], &hki); #endif #ifdef INET6 hki.hook_id = AF_INET6; hki.hook_type = HHOOK_TYPE_IPSEC_IN; hhook_remove_hook(V_ipsec_hhh_in[HHOOK_IPSEC_INET6], &hki); hki.hook_type = HHOOK_TYPE_IPSEC_OUT; hhook_remove_hook(V_ipsec_hhh_out[HHOOK_IPSEC_INET6], &hki); #endif } static void vnet_enc_init(const void *unused __unused) { V_enc_sc = NULL; V_enc_cloner = if_clone_simple(encname, enc_clone_create, enc_clone_destroy, 1); } VNET_SYSINIT(vnet_enc_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_enc_init, NULL); static void vnet_enc_init_proto(void *unused __unused) { KASSERT(V_enc_sc != NULL, ("%s: V_enc_sc is %p\n", __func__, V_enc_sc)); if (enc_add_hhooks(V_enc_sc) != 0) enc_clone_destroy(V_enc_sc->sc_ifp); } VNET_SYSINIT(vnet_enc_init_proto, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_enc_init_proto, NULL); static void vnet_enc_uninit(const void *unused __unused) { KASSERT(V_enc_sc != NULL, ("%s: V_enc_sc is %p\n", __func__, V_enc_sc)); if_clone_detach(V_enc_cloner); } VNET_SYSUNINIT(vnet_enc_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_enc_uninit, NULL); /* * The hhook consumer needs to go before ip[6]_destroy are called on * SI_ORDER_THIRD. */ static void vnet_enc_uninit_hhook(const void *unused __unused) { KASSERT(V_enc_sc != NULL, ("%s: V_enc_sc is %p\n", __func__, V_enc_sc)); enc_remove_hhooks(V_enc_sc); } VNET_SYSUNINIT(vnet_enc_uninit_hhook, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, vnet_enc_uninit_hhook, NULL); static int enc_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t enc_mod = { "if_enc", enc_modevent, 0 }; DECLARE_MODULE(if_enc, enc_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_enc, 1); Index: head/sys/net/if_epair.c =================================================================== --- head/sys/net/if_epair.c (revision 336675) +++ head/sys/net/if_epair.c (revision 336676) @@ -1,1055 +1,1055 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2008 The FreeBSD Foundation * Copyright (c) 2009-2010 Bjoern A. Zeeb * All rights reserved. * * This software was developed by CK Software GmbH under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * A pair of virtual back-to-back connected ethernet like interfaces * (``two interfaces with a virtual cross-over cable''). * * This is mostly intended to be used to provide connectivity between * different virtual network stack instances. */ /* * Things to re-think once we have more experience: * - ifp->if_reassign function once we can test with vimage. Depending on * how if_vmove() is going to be improved. * - Real random etheraddrs that are checked to be uniquish; we would need * to re-do them in case we move the interface between network stacks * in a private if_reassign function. * In case we bridge to a real interface/network or between indepedent * epairs on multiple stacks/machines, we may need this. * For now let the user handle that case. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include SYSCTL_DECL(_net_link); static SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW, 0, "epair sysctl"); #ifdef EPAIR_DEBUG static int epair_debug = 0; SYSCTL_INT(_net_link_epair, OID_AUTO, epair_debug, CTLFLAG_RW, &epair_debug, 0, "if_epair(4) debugging."); #define DPRINTF(fmt, arg...) \ if (epair_debug) \ printf("[%s:%d] " fmt, __func__, __LINE__, ##arg) #else #define DPRINTF(fmt, arg...) #endif static void epair_nh_sintr(struct mbuf *); static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *); static void epair_nh_drainedcpu(u_int); static void epair_start_locked(struct ifnet *); static int epair_media_change(struct ifnet *); static void epair_media_status(struct ifnet *, struct ifmediareq *); static int epair_clone_match(struct if_clone *, const char *); static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t); static int epair_clone_destroy(struct if_clone *, struct ifnet *); static const char epairname[] = "epair"; static unsigned int next_index = 0; /* Netisr related definitions and sysctl. */ static struct netisr_handler epair_nh = { .nh_name = epairname, .nh_proto = NETISR_EPAIR, .nh_policy = NETISR_POLICY_CPU, .nh_handler = epair_nh_sintr, .nh_m2cpuid = epair_nh_m2cpuid, .nh_drainedcpu = epair_nh_drainedcpu, }; static int sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&epair_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&epair_nh, qlimit)); } SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_epair_netisr_maxqlen, "I", "Maximum if_epair(4) netisr \"hw\" queue length"); struct epair_softc { struct ifnet *ifp; /* This ifp. */ struct ifnet *oifp; /* other ifp of pair. */ struct ifmedia media; /* Media config (fake). */ u_int refcount; /* # of mbufs in flight. */ u_int cpuid; /* CPU ID assigned upon creation. */ void (*if_qflush)(struct ifnet *); /* Original if_qflush routine. */ }; /* * Per-CPU list of ifps with data in the ifq that needs to be flushed * to the netisr ``hw'' queue before we allow any further direct queuing * to the ``hw'' queue. */ struct epair_ifp_drain { STAILQ_ENTRY(epair_ifp_drain) ifp_next; struct ifnet *ifp; }; STAILQ_HEAD(eid_list, epair_ifp_drain); #define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \ "if_epair", NULL, MTX_DEF) #define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx) #define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \ MA_OWNED) #define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx) #define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx) #ifdef INVARIANTS #define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v)) #define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r)) #define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r)) #define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p) #else #define EPAIR_REFCOUNT_INIT(r, v) #define EPAIR_REFCOUNT_AQUIRE(r) #define EPAIR_REFCOUNT_RELEASE(r) #define EPAIR_REFCOUNT_ASSERT(a, p) #endif static MALLOC_DEFINE(M_EPAIR, epairname, "Pair of virtual cross-over connected Ethernet-like interfaces"); -static VNET_DEFINE(struct if_clone *, epair_cloner); +VNET_DEFINE_STATIC(struct if_clone *, epair_cloner); #define V_epair_cloner VNET(epair_cloner) /* * DPCPU area and functions. */ struct epair_dpcpu { struct mtx if_epair_mtx; /* Per-CPU locking. */ int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */ struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with * data in the ifq. */ }; DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu); static void epair_dpcpu_init(void) { struct epair_dpcpu *epair_dpcpu; struct eid_list *s; u_int cpuid; CPU_FOREACH(cpuid) { epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); /* Initialize per-cpu lock. */ EPAIR_LOCK_INIT(epair_dpcpu); /* Driver flags are per-cpu as are our netisr "hw" queues. */ epair_dpcpu->epair_drv_flags = 0; /* * Initialize per-cpu drain list. * Manually do what STAILQ_HEAD_INITIALIZER would do. */ s = &epair_dpcpu->epair_ifp_drain_list; s->stqh_first = NULL; s->stqh_last = &s->stqh_first; } } static void epair_dpcpu_detach(void) { struct epair_dpcpu *epair_dpcpu; u_int cpuid; CPU_FOREACH(cpuid) { epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); /* Destroy per-cpu lock. */ EPAIR_LOCK_DESTROY(epair_dpcpu); } } /* * Helper functions. */ static u_int cpuid_from_ifp(struct ifnet *ifp) { struct epair_softc *sc; if (ifp == NULL) return (0); sc = ifp->if_softc; return (sc->cpuid); } /* * Netisr handler functions. */ static void epair_nh_sintr(struct mbuf *m) { struct ifnet *ifp; struct epair_softc *sc __unused; ifp = m->m_pkthdr.rcvif; (*ifp->if_input)(ifp, m); sc = ifp->if_softc; EPAIR_REFCOUNT_RELEASE(&sc->refcount); EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, ("%s: ifp=%p sc->refcount not >= 1: %d", __func__, ifp, sc->refcount)); DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount); } static struct mbuf * epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid) { *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif); return (m); } static void epair_nh_drainedcpu(u_int cpuid) { struct epair_dpcpu *epair_dpcpu; struct epair_ifp_drain *elm, *tvar; struct ifnet *ifp; epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); EPAIR_LOCK(epair_dpcpu); /* * Assume our "hw" queue and possibly ifq will be emptied * again. In case we will overflow the "hw" queue while * draining, epair_start_locked will set IFF_DRV_OACTIVE * again and we will stop and return. */ STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next, tvar) { ifp = elm->ifp; epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; epair_start_locked(ifp); IFQ_LOCK(&ifp->if_snd); if (IFQ_IS_EMPTY(&ifp->if_snd)) { struct epair_softc *sc __unused; STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list, elm, epair_ifp_drain, ifp_next); /* The cached ifp goes off the list. */ sc = ifp->if_softc; EPAIR_REFCOUNT_RELEASE(&sc->refcount); EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, ("%s: ifp=%p sc->refcount not >= 1: %d", __func__, ifp, sc->refcount)); free(elm, M_EPAIR); } IFQ_UNLOCK(&ifp->if_snd); if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) { /* Our "hw"q overflew again. */ epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; DPRINTF("hw queue length overflow at %u\n", epair_nh.nh_qlimit); break; } } EPAIR_UNLOCK(epair_dpcpu); } /* * Network interface (`if') related functions. */ static void epair_remove_ifp_from_draining(struct ifnet *ifp) { struct epair_dpcpu *epair_dpcpu; struct epair_ifp_drain *elm, *tvar; u_int cpuid; CPU_FOREACH(cpuid) { epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); EPAIR_LOCK(epair_dpcpu); STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next, tvar) { if (ifp == elm->ifp) { struct epair_softc *sc __unused; STAILQ_REMOVE( &epair_dpcpu->epair_ifp_drain_list, elm, epair_ifp_drain, ifp_next); /* The cached ifp goes off the list. */ sc = ifp->if_softc; EPAIR_REFCOUNT_RELEASE(&sc->refcount); EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, ("%s: ifp=%p sc->refcount not >= 1: %d", __func__, ifp, sc->refcount)); free(elm, M_EPAIR); } } EPAIR_UNLOCK(epair_dpcpu); } } static int epair_add_ifp_for_draining(struct ifnet *ifp) { struct epair_dpcpu *epair_dpcpu; struct epair_softc *sc; struct epair_ifp_drain *elm = NULL; sc = ifp->if_softc; epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); EPAIR_LOCK_ASSERT(epair_dpcpu); STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next) if (elm->ifp == ifp) break; /* If the ifp is there already, return success. */ if (elm != NULL) return (0); elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO); if (elm == NULL) return (ENOMEM); elm->ifp = ifp; /* Add a reference for the ifp pointer on the list. */ EPAIR_REFCOUNT_AQUIRE(&sc->refcount); STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next); return (0); } static void epair_start_locked(struct ifnet *ifp) { struct epair_dpcpu *epair_dpcpu; struct mbuf *m; struct epair_softc *sc; struct ifnet *oifp; int error; DPRINTF("ifp=%p\n", ifp); sc = ifp->if_softc; epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); EPAIR_LOCK_ASSERT(epair_dpcpu); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; if ((ifp->if_flags & IFF_UP) == 0) return; /* * We get packets here from ether_output via if_handoff() * and need to put them into the input queue of the oifp * and call oifp->if_input() via netisr/epair_sintr(). */ oifp = sc->oifp; sc = oifp->if_softc; for (;;) { IFQ_DEQUEUE(&ifp->if_snd, m); if (m == NULL) break; BPF_MTAP(ifp, m); /* * In case the outgoing interface is not usable, * drop the packet. */ if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (oifp->if_flags & IFF_UP) ==0) { if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); m_freem(m); continue; } DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname); /* * Add a reference so the interface cannot go while the * packet is in transit as we rely on rcvif to stay valid. */ EPAIR_REFCOUNT_AQUIRE(&sc->refcount); m->m_pkthdr.rcvif = oifp; CURVNET_SET_QUIET(oifp->if_vnet); error = netisr_queue(NETISR_EPAIR, m); CURVNET_RESTORE(); if (!error) { if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* Someone else received the packet. */ if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); } else { /* The packet was freed already. */ epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; ifp->if_drv_flags |= IFF_DRV_OACTIVE; (void) epair_add_ifp_for_draining(ifp); if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); EPAIR_REFCOUNT_RELEASE(&sc->refcount); EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, ("%s: ifp=%p sc->refcount not >= 1: %d", __func__, oifp, sc->refcount)); } } } static void epair_start(struct ifnet *ifp) { struct epair_dpcpu *epair_dpcpu; epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu); EPAIR_LOCK(epair_dpcpu); epair_start_locked(ifp); EPAIR_UNLOCK(epair_dpcpu); } static int epair_transmit_locked(struct ifnet *ifp, struct mbuf *m) { struct epair_dpcpu *epair_dpcpu; struct epair_softc *sc; struct ifnet *oifp; int error, len; short mflags; DPRINTF("ifp=%p m=%p\n", ifp, m); sc = ifp->if_softc; epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); EPAIR_LOCK_ASSERT(epair_dpcpu); if (m == NULL) return (0); /* * We are not going to use the interface en/dequeue mechanism * on the TX side. We are called from ether_output_frame() * and will put the packet into the incoming queue of the * other interface of our pair via the netsir. */ if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { m_freem(m); return (ENXIO); } if ((ifp->if_flags & IFF_UP) == 0) { m_freem(m); return (ENETDOWN); } BPF_MTAP(ifp, m); /* * In case the outgoing interface is not usable, * drop the packet. */ oifp = sc->oifp; if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (oifp->if_flags & IFF_UP) ==0) { if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); m_freem(m); return (0); } len = m->m_pkthdr.len; mflags = m->m_flags; DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname); #ifdef ALTQ /* Support ALTQ via the classic if_start() path. */ IF_LOCK(&ifp->if_snd); if (ALTQ_IS_ENABLED(&ifp->if_snd)) { ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error); if (error) if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); IF_UNLOCK(&ifp->if_snd); if (!error) { if_inc_counter(ifp, IFCOUNTER_OBYTES, len); if (mflags & (M_BCAST|M_MCAST)) if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0) epair_start_locked(ifp); else (void)epair_add_ifp_for_draining(ifp); } return (error); } IF_UNLOCK(&ifp->if_snd); #endif if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) { /* * Our hardware queue is full, try to fall back * queuing to the ifq but do not call ifp->if_start. * Either we are lucky or the packet is gone. */ IFQ_ENQUEUE(&ifp->if_snd, m, error); if (!error) (void)epair_add_ifp_for_draining(ifp); return (error); } sc = oifp->if_softc; /* * Add a reference so the interface cannot go while the * packet is in transit as we rely on rcvif to stay valid. */ EPAIR_REFCOUNT_AQUIRE(&sc->refcount); m->m_pkthdr.rcvif = oifp; CURVNET_SET_QUIET(oifp->if_vnet); error = netisr_queue(NETISR_EPAIR, m); CURVNET_RESTORE(); if (!error) { if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* * IFQ_HANDOFF_ADJ/ip_handoff() update statistics, * but as we bypass all this we have to duplicate * the logic another time. */ if_inc_counter(ifp, IFCOUNTER_OBYTES, len); if (mflags & (M_BCAST|M_MCAST)) if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); /* Someone else received the packet. */ if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); } else { /* The packet was freed already. */ epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; ifp->if_drv_flags |= IFF_DRV_OACTIVE; if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); EPAIR_REFCOUNT_RELEASE(&sc->refcount); EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, ("%s: ifp=%p sc->refcount not >= 1: %d", __func__, oifp, sc->refcount)); } return (error); } static int epair_transmit(struct ifnet *ifp, struct mbuf *m) { struct epair_dpcpu *epair_dpcpu; int error; epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu); EPAIR_LOCK(epair_dpcpu); error = epair_transmit_locked(ifp, m); EPAIR_UNLOCK(epair_dpcpu); return (error); } static void epair_qflush(struct ifnet *ifp) { struct epair_softc *sc; sc = ifp->if_softc; KASSERT(sc != NULL, ("%s: ifp=%p, epair_softc gone? sc=%p\n", __func__, ifp, sc)); /* * Remove this ifp from all backpointer lists. The interface will not * usable for flushing anyway nor should it have anything to flush * after if_qflush(). */ epair_remove_ifp_from_draining(ifp); if (sc->if_qflush) sc->if_qflush(ifp); } static int epair_media_change(struct ifnet *ifp __unused) { /* Do nothing. */ return (0); } static void epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr) { imr->ifm_status = IFM_AVALID | IFM_ACTIVE; imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX; } static int epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct epair_softc *sc; struct ifreq *ifr; int error; ifr = (struct ifreq *)data; switch (cmd) { case SIOCSIFFLAGS: case SIOCADDMULTI: case SIOCDELMULTI: error = 0; break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: sc = ifp->if_softc; error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd); break; case SIOCSIFMTU: /* We basically allow all kinds of MTUs. */ ifp->if_mtu = ifr->ifr_mtu; error = 0; break; default: /* Let the common ethernet handler process this. */ error = ether_ioctl(ifp, cmd, data); break; } return (error); } static void epair_init(void *dummy __unused) { } /* * Interface cloning functions. * We use our private ones so that we can create/destroy our secondary * device along with the primary one. */ static int epair_clone_match(struct if_clone *ifc, const char *name) { const char *cp; DPRINTF("name='%s'\n", name); /* * Our base name is epair. * Our interfaces will be named epair[ab]. * So accept anything of the following list: * - epair * - epair * but not the epair[ab] versions. */ if (strncmp(epairname, name, sizeof(epairname)-1) != 0) return (0); for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) { if (*cp < '0' || *cp > '9') return (0); } return (1); } static int epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) { struct epair_softc *sca, *scb; struct ifnet *ifp; char *dp; int error, unit, wildcard; uint64_t hostid; uint32_t key[3]; uint32_t hash; uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ /* * We are abusing params to create our second interface. * Actually we already created it and called if_clone_create() * for it to do the official insertion procedure the moment we knew * it cannot fail anymore. So just do attach it here. */ if (params) { scb = (struct epair_softc *)params; ifp = scb->ifp; /* Copy epairNa etheraddr and change the last byte. */ memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN); eaddr[5] = 0x0b; ether_ifattach(ifp, eaddr); /* Correctly set the name for the cloner list. */ strlcpy(name, ifp->if_xname, len); return (0); } /* Try to see if a special unit was requested. */ error = ifc_name2unit(name, &unit); if (error != 0) return (error); wildcard = (unit < 0); error = ifc_alloc_unit(ifc, &unit); if (error != 0) return (error); /* * If no unit had been given, we need to adjust the ifName. * Also make sure there is space for our extra [ab] suffix. */ for (dp = name; *dp != '\0'; dp++); if (wildcard) { error = snprintf(dp, len - (dp - name), "%d", unit); if (error > len - (dp - name) - 1) { /* ifName too long. */ ifc_free_unit(ifc, unit); return (ENOSPC); } dp += error; } if (len - (dp - name) - 1 < 1) { /* No space left for our [ab] suffix. */ ifc_free_unit(ifc, unit); return (ENOSPC); } *dp = 'b'; /* Must not change dp so we can replace 'a' by 'b' later. */ *(dp+1) = '\0'; /* Check if 'a' and 'b' interfaces already exist. */ if (ifunit(name) != NULL) return (EEXIST); *dp = 'a'; if (ifunit(name) != NULL) return (EEXIST); /* Allocate memory for both [ab] interfaces */ sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); EPAIR_REFCOUNT_INIT(&sca->refcount, 1); sca->ifp = if_alloc(IFT_ETHER); if (sca->ifp == NULL) { free(sca, M_EPAIR); ifc_free_unit(ifc, unit); return (ENOSPC); } scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); EPAIR_REFCOUNT_INIT(&scb->refcount, 1); scb->ifp = if_alloc(IFT_ETHER); if (scb->ifp == NULL) { free(scb, M_EPAIR); if_free(sca->ifp); free(sca, M_EPAIR); ifc_free_unit(ifc, unit); return (ENOSPC); } /* * Cross-reference the interfaces so we will be able to free both. */ sca->oifp = scb->ifp; scb->oifp = sca->ifp; /* * Calculate the cpuid for netisr queueing based on the * ifIndex of the interfaces. As long as we cannot configure * this or use cpuset information easily we cannot guarantee * cache locality but we can at least allow parallelism. */ sca->cpuid = netisr_get_cpuid(sca->ifp->if_index); scb->cpuid = netisr_get_cpuid(scb->ifp->if_index); /* Initialise pseudo media types. */ ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status); ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL); ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T); ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status); ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL); ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T); /* Finish initialization of interface a. */ ifp = sca->ifp; ifp->if_softc = sca; strlcpy(ifp->if_xname, name, IFNAMSIZ); ifp->if_dname = epairname; ifp->if_dunit = unit; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_capabilities = IFCAP_VLAN_MTU; ifp->if_capenable = IFCAP_VLAN_MTU; ifp->if_start = epair_start; ifp->if_ioctl = epair_ioctl; ifp->if_init = epair_init; if_setsendqlen(ifp, ifqmaxlen); if_setsendqready(ifp); /* * Calculate the etheraddr hashing the hostid and the * interface index. The result would be hopefully unique. * Note that the "a" component of an epair instance may get moved * to a different VNET after creation. In that case its index * will be freed and the index can get reused by new epair instance. * Make sure we do not create same etheraddr again. */ getcredhostid(curthread->td_ucred, (unsigned long *)&hostid); if (hostid == 0) arc4rand(&hostid, sizeof(hostid), 0); if (ifp->if_index > next_index) next_index = ifp->if_index; else next_index++; key[0] = (uint32_t)next_index; key[1] = (uint32_t)(hostid & 0xffffffff); key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff); hash = jenkins_hash32(key, 3, 0); eaddr[0] = 0x02; memcpy(&eaddr[1], &hash, 4); eaddr[5] = 0x0a; ether_ifattach(ifp, eaddr); sca->if_qflush = ifp->if_qflush; ifp->if_qflush = epair_qflush; ifp->if_transmit = epair_transmit; ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ /* Swap the name and finish initialization of interface b. */ *dp = 'b'; ifp = scb->ifp; ifp->if_softc = scb; strlcpy(ifp->if_xname, name, IFNAMSIZ); ifp->if_dname = epairname; ifp->if_dunit = unit; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_capabilities = IFCAP_VLAN_MTU; ifp->if_capenable = IFCAP_VLAN_MTU; ifp->if_start = epair_start; ifp->if_ioctl = epair_ioctl; ifp->if_init = epair_init; if_setsendqlen(ifp, ifqmaxlen); if_setsendqready(ifp); /* We need to play some tricks here for the second interface. */ strlcpy(name, epairname, len); error = if_clone_create(name, len, (caddr_t)scb); if (error) panic("%s: if_clone_create() for our 2nd iface failed: %d", __func__, error); scb->if_qflush = ifp->if_qflush; ifp->if_qflush = epair_qflush; ifp->if_transmit = epair_transmit; ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ /* * Restore name to a as the ifp for this will go into the * cloner list for the initial call. */ strlcpy(name, sca->ifp->if_xname, len); DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb); /* Tell the world, that we are ready to rock. */ sca->ifp->if_drv_flags |= IFF_DRV_RUNNING; scb->ifp->if_drv_flags |= IFF_DRV_RUNNING; if_link_state_change(sca->ifp, LINK_STATE_UP); if_link_state_change(scb->ifp, LINK_STATE_UP); return (0); } static int epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) { struct ifnet *oifp; struct epair_softc *sca, *scb; int unit, error; DPRINTF("ifp=%p\n", ifp); /* * In case we called into if_clone_destroyif() ourselves * again to remove the second interface, the softc will be * NULL. In that case so not do anything but return success. */ if (ifp->if_softc == NULL) return (0); unit = ifp->if_dunit; sca = ifp->if_softc; oifp = sca->oifp; scb = oifp->if_softc; DPRINTF("ifp=%p oifp=%p\n", ifp, oifp); if_link_state_change(ifp, LINK_STATE_DOWN); if_link_state_change(oifp, LINK_STATE_DOWN); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; oifp->if_drv_flags &= ~IFF_DRV_RUNNING; /* * Get rid of our second half. As the other of the two * interfaces may reside in a different vnet, we need to * switch before freeing them. */ CURVNET_SET_QUIET(oifp->if_vnet); ether_ifdetach(oifp); /* * Wait for all packets to be dispatched to if_input. * The numbers can only go down as the interface is * detached so there is no need to use atomics. */ DPRINTF("scb refcnt=%u\n", scb->refcount); EPAIR_REFCOUNT_ASSERT(scb->refcount == 1, ("%s: ifp=%p scb->refcount!=1: %d", __func__, oifp, scb->refcount)); oifp->if_softc = NULL; error = if_clone_destroyif(ifc, oifp); if (error) panic("%s: if_clone_destroyif() for our 2nd iface failed: %d", __func__, error); if_free(oifp); ifmedia_removeall(&scb->media); free(scb, M_EPAIR); CURVNET_RESTORE(); ether_ifdetach(ifp); /* * Wait for all packets to be dispatched to if_input. */ DPRINTF("sca refcnt=%u\n", sca->refcount); EPAIR_REFCOUNT_ASSERT(sca->refcount == 1, ("%s: ifp=%p sca->refcount!=1: %d", __func__, ifp, sca->refcount)); if_free(ifp); ifmedia_removeall(&sca->media); free(sca, M_EPAIR); ifc_free_unit(ifc, unit); return (0); } static void vnet_epair_init(const void *unused __unused) { V_epair_cloner = if_clone_advanced(epairname, 0, epair_clone_match, epair_clone_create, epair_clone_destroy); #ifdef VIMAGE netisr_register_vnet(&epair_nh); #endif } VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_epair_init, NULL); static void vnet_epair_uninit(const void *unused __unused) { #ifdef VIMAGE netisr_unregister_vnet(&epair_nh); #endif if_clone_detach(V_epair_cloner); } VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_epair_uninit, NULL); static void epair_uninit(const void *unused __unused) { netisr_unregister(&epair_nh); epair_dpcpu_detach(); if (bootverbose) printf("%s unloaded.\n", epairname); } SYSUNINIT(epair_uninit, SI_SUB_INIT_IF, SI_ORDER_MIDDLE, epair_uninit, NULL); static int epair_modevent(module_t mod, int type, void *data) { int qlimit; switch (type) { case MOD_LOAD: /* For now limit us to one global mutex and one inq. */ epair_dpcpu_init(); epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */ if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit)) epair_nh.nh_qlimit = qlimit; netisr_register(&epair_nh); if (bootverbose) printf("%s initialized.\n", epairname); break; case MOD_UNLOAD: /* Handled in epair_uninit() */ break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t epair_mod = { "if_epair", epair_modevent, 0 }; DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE); MODULE_VERSION(if_epair, 1); Index: head/sys/net/if_ethersubr.c =================================================================== --- head/sys/net/if_ethersubr.c (revision 336675) +++ head/sys/net/if_ethersubr.c (revision 336676) @@ -1,1371 +1,1371 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)if_ethersubr.c 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_netgraph.h" #include "opt_mbuf_profiling.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #include #include #include #endif #ifdef INET6 #include #endif #include #ifdef CTASSERT CTASSERT(sizeof (struct ether_header) == ETHER_ADDR_LEN * 2 + 2); CTASSERT(sizeof (struct ether_addr) == ETHER_ADDR_LEN); #endif VNET_DEFINE(struct pfil_head, link_pfil_hook); /* Packet filter hooks */ /* netgraph node hooks for ng_ether(4) */ void (*ng_ether_input_p)(struct ifnet *ifp, struct mbuf **mp); void (*ng_ether_input_orphan_p)(struct ifnet *ifp, struct mbuf *m); int (*ng_ether_output_p)(struct ifnet *ifp, struct mbuf **mp); void (*ng_ether_attach_p)(struct ifnet *ifp); void (*ng_ether_detach_p)(struct ifnet *ifp); void (*vlan_input_p)(struct ifnet *, struct mbuf *); /* if_bridge(4) support */ void (*bridge_dn_p)(struct mbuf *, struct ifnet *); /* if_lagg(4) support */ struct mbuf *(*lagg_input_p)(struct ifnet *, struct mbuf *); static const u_char etherbroadcastaddr[ETHER_ADDR_LEN] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; static int ether_resolvemulti(struct ifnet *, struct sockaddr **, struct sockaddr *); #ifdef VIMAGE static void ether_reassign(struct ifnet *, struct vnet *, char *); #endif static int ether_requestencap(struct ifnet *, struct if_encap_req *); #define senderr(e) do { error = (e); goto bad;} while (0) static void update_mbuf_csumflags(struct mbuf *src, struct mbuf *dst) { int csum_flags = 0; if (src->m_pkthdr.csum_flags & CSUM_IP) csum_flags |= (CSUM_IP_CHECKED|CSUM_IP_VALID); if (src->m_pkthdr.csum_flags & CSUM_DELAY_DATA) csum_flags |= (CSUM_DATA_VALID|CSUM_PSEUDO_HDR); if (src->m_pkthdr.csum_flags & CSUM_SCTP) csum_flags |= CSUM_SCTP_VALID; dst->m_pkthdr.csum_flags |= csum_flags; if (csum_flags & CSUM_DATA_VALID) dst->m_pkthdr.csum_data = 0xffff; } /* * Handle link-layer encapsulation requests. */ static int ether_requestencap(struct ifnet *ifp, struct if_encap_req *req) { struct ether_header *eh; struct arphdr *ah; uint16_t etype; const u_char *lladdr; if (req->rtype != IFENCAP_LL) return (EOPNOTSUPP); if (req->bufsize < ETHER_HDR_LEN) return (ENOMEM); eh = (struct ether_header *)req->buf; lladdr = req->lladdr; req->lladdr_off = 0; switch (req->family) { case AF_INET: etype = htons(ETHERTYPE_IP); break; case AF_INET6: etype = htons(ETHERTYPE_IPV6); break; case AF_ARP: ah = (struct arphdr *)req->hdata; ah->ar_hrd = htons(ARPHRD_ETHER); switch(ntohs(ah->ar_op)) { case ARPOP_REVREQUEST: case ARPOP_REVREPLY: etype = htons(ETHERTYPE_REVARP); break; case ARPOP_REQUEST: case ARPOP_REPLY: default: etype = htons(ETHERTYPE_ARP); break; } if (req->flags & IFENCAP_FLAG_BROADCAST) lladdr = ifp->if_broadcastaddr; break; default: return (EAFNOSUPPORT); } memcpy(&eh->ether_type, &etype, sizeof(eh->ether_type)); memcpy(eh->ether_dhost, lladdr, ETHER_ADDR_LEN); memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN); req->bufsize = sizeof(struct ether_header); return (0); } static int ether_resolve_addr(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro, u_char *phdr, uint32_t *pflags, struct llentry **plle) { struct ether_header *eh; uint32_t lleflags = 0; int error = 0; #if defined(INET) || defined(INET6) uint16_t etype; #endif if (plle) *plle = NULL; eh = (struct ether_header *)phdr; switch (dst->sa_family) { #ifdef INET case AF_INET: if ((m->m_flags & (M_BCAST | M_MCAST)) == 0) error = arpresolve(ifp, 0, m, dst, phdr, &lleflags, plle); else { if (m->m_flags & M_BCAST) memcpy(eh->ether_dhost, ifp->if_broadcastaddr, ETHER_ADDR_LEN); else { const struct in_addr *a; a = &(((const struct sockaddr_in *)dst)->sin_addr); ETHER_MAP_IP_MULTICAST(a, eh->ether_dhost); } etype = htons(ETHERTYPE_IP); memcpy(&eh->ether_type, &etype, sizeof(etype)); memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN); } break; #endif #ifdef INET6 case AF_INET6: if ((m->m_flags & M_MCAST) == 0) error = nd6_resolve(ifp, 0, m, dst, phdr, &lleflags, plle); else { const struct in6_addr *a6; a6 = &(((const struct sockaddr_in6 *)dst)->sin6_addr); ETHER_MAP_IPV6_MULTICAST(a6, eh->ether_dhost); etype = htons(ETHERTYPE_IPV6); memcpy(&eh->ether_type, &etype, sizeof(etype)); memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN); } break; #endif default: if_printf(ifp, "can't handle af%d\n", dst->sa_family); if (m != NULL) m_freem(m); return (EAFNOSUPPORT); } if (error == EHOSTDOWN) { if (ro != NULL && (ro->ro_flags & RT_HAS_GW) != 0) error = EHOSTUNREACH; } if (error != 0) return (error); *pflags = RT_MAY_LOOP; if (lleflags & LLE_IFADDR) *pflags |= RT_L2_ME; return (0); } /* * Ethernet output routine. * Encapsulate a packet of type family for the local net. * Use trailer local net encapsulation if enough data in first * packet leaves a multiple of 512 bytes of data in remainder. */ int ether_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { int error = 0; char linkhdr[ETHER_HDR_LEN], *phdr; struct ether_header *eh; struct pf_mtag *t; int loop_copy = 1; int hlen; /* link layer header length */ uint32_t pflags; struct llentry *lle = NULL; int addref = 0; phdr = NULL; pflags = 0; if (ro != NULL) { /* XXX BPF uses ro_prepend */ if (ro->ro_prepend != NULL) { phdr = ro->ro_prepend; hlen = ro->ro_plen; } else if (!(m->m_flags & (M_BCAST | M_MCAST))) { if ((ro->ro_flags & RT_LLE_CACHE) != 0) { lle = ro->ro_lle; if (lle != NULL && (lle->la_flags & LLE_VALID) == 0) { LLE_FREE(lle); lle = NULL; /* redundant */ ro->ro_lle = NULL; } if (lle == NULL) { /* if we lookup, keep cache */ addref = 1; } else /* * Notify LLE code that * the entry was used * by datapath. */ llentry_mark_used(lle); } if (lle != NULL) { phdr = lle->r_linkdata; hlen = lle->r_hdrlen; pflags = lle->r_flags; } } } #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error) senderr(error); #endif M_PROFILE(m); if (ifp->if_flags & IFF_MONITOR) senderr(ENETDOWN); if (!((ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING))) senderr(ENETDOWN); if (phdr == NULL) { /* No prepend data supplied. Try to calculate ourselves. */ phdr = linkhdr; hlen = ETHER_HDR_LEN; error = ether_resolve_addr(ifp, m, dst, ro, phdr, &pflags, addref ? &lle : NULL); if (addref && lle != NULL) ro->ro_lle = lle; if (error != 0) return (error == EWOULDBLOCK ? 0 : error); } if ((pflags & RT_L2_ME) != 0) { update_mbuf_csumflags(m, m); return (if_simloop(ifp, m, dst->sa_family, 0)); } loop_copy = pflags & RT_MAY_LOOP; /* * Add local net header. If no space in first mbuf, * allocate another. * * Note that we do prepend regardless of RT_HAS_HEADER flag. * This is done because BPF code shifts m_data pointer * to the end of ethernet header prior to calling if_output(). */ M_PREPEND(m, hlen, M_NOWAIT); if (m == NULL) senderr(ENOBUFS); if ((pflags & RT_HAS_HEADER) == 0) { eh = mtod(m, struct ether_header *); memcpy(eh, phdr, hlen); } /* * If a simplex interface, and the packet is being sent to our * Ethernet address or a broadcast address, loopback a copy. * XXX To make a simplex device behave exactly like a duplex * device, we should copy in the case of sending to our own * ethernet address (thus letting the original actually appear * on the wire). However, we don't do that here for security * reasons and compatibility with the original behavior. */ if ((m->m_flags & M_BCAST) && loop_copy && (ifp->if_flags & IFF_SIMPLEX) && ((t = pf_find_mtag(m)) == NULL || !t->routed)) { struct mbuf *n; /* * Because if_simloop() modifies the packet, we need a * writable copy through m_dup() instead of a readonly * one as m_copy[m] would give us. The alternative would * be to modify if_simloop() to handle the readonly mbuf, * but performancewise it is mostly equivalent (trading * extra data copying vs. extra locking). * * XXX This is a local workaround. A number of less * often used kernel parts suffer from the same bug. * See PR kern/105943 for a proposed general solution. */ if ((n = m_dup(m, M_NOWAIT)) != NULL) { update_mbuf_csumflags(m, n); (void)if_simloop(ifp, n, dst->sa_family, hlen); } else if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1); } /* * Bridges require special output handling. */ if (ifp->if_bridge) { BRIDGE_OUTPUT(ifp, m, error); return (error); } #if defined(INET) || defined(INET6) if (ifp->if_carp && (error = (*carp_output_p)(ifp, m, dst))) goto bad; #endif /* Handle ng_ether(4) processing, if any */ if (ifp->if_l2com != NULL) { KASSERT(ng_ether_output_p != NULL, ("ng_ether_output_p is NULL")); if ((error = (*ng_ether_output_p)(ifp, &m)) != 0) { bad: if (m != NULL) m_freem(m); return (error); } if (m == NULL) return (0); } /* Continue with link-layer output */ return ether_output_frame(ifp, m); } static bool ether_set_pcp(struct mbuf **mp, struct ifnet *ifp, uint8_t pcp) { struct ether_header *eh; eh = mtod(*mp, struct ether_header *); if (ntohs(eh->ether_type) == ETHERTYPE_VLAN || ether_8021q_frame(mp, ifp, ifp, 0, pcp)) return (true); if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (false); } /* * Ethernet link layer output routine to send a raw frame to the device. * * This assumes that the 14 byte Ethernet header is present and contiguous * in the first mbuf (if BRIDGE'ing). */ int ether_output_frame(struct ifnet *ifp, struct mbuf *m) { int error; uint8_t pcp; pcp = ifp->if_pcp; if (pcp != IFNET_PCP_NONE && !ether_set_pcp(&m, ifp, pcp)) return (0); if (PFIL_HOOKED(&V_link_pfil_hook)) { error = pfil_run_hooks(&V_link_pfil_hook, &m, ifp, PFIL_OUT, 0, NULL); if (error != 0) return (EACCES); if (m == NULL) return (0); } /* * Queue message on interface, update output statistics if * successful, and start output if interface not yet active. */ return ((ifp->if_transmit)(ifp, m)); } /* * Process a received Ethernet packet; the packet is in the * mbuf chain m with the ethernet header at the front. */ static void ether_input_internal(struct ifnet *ifp, struct mbuf *m) { struct ether_header *eh; u_short etype; if ((ifp->if_flags & IFF_UP) == 0) { m_freem(m); return; } #ifdef DIAGNOSTIC if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { if_printf(ifp, "discard frame at !IFF_DRV_RUNNING\n"); m_freem(m); return; } #endif if (m->m_len < ETHER_HDR_LEN) { /* XXX maybe should pullup? */ if_printf(ifp, "discard frame w/o leading ethernet " "header (len %u pkt len %u)\n", m->m_len, m->m_pkthdr.len); if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); m_freem(m); return; } eh = mtod(m, struct ether_header *); etype = ntohs(eh->ether_type); random_harvest_queue_ether(m, sizeof(*m), 2); CURVNET_SET_QUIET(ifp->if_vnet); if (ETHER_IS_MULTICAST(eh->ether_dhost)) { if (ETHER_IS_BROADCAST(eh->ether_dhost)) m->m_flags |= M_BCAST; else m->m_flags |= M_MCAST; if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1); } #ifdef MAC /* * Tag the mbuf with an appropriate MAC label before any other * consumers can get to it. */ mac_ifnet_create_mbuf(ifp, m); #endif /* * Give bpf a chance at the packet. */ ETHER_BPF_MTAP(ifp, m); /* * If the CRC is still on the packet, trim it off. We do this once * and once only in case we are re-entered. Nothing else on the * Ethernet receive path expects to see the FCS. */ if (m->m_flags & M_HASFCS) { m_adj(m, -ETHER_CRC_LEN); m->m_flags &= ~M_HASFCS; } if (!(ifp->if_capenable & IFCAP_HWSTATS)) if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); /* Allow monitor mode to claim this frame, after stats are updated. */ if (ifp->if_flags & IFF_MONITOR) { m_freem(m); CURVNET_RESTORE(); return; } /* Handle input from a lagg(4) port */ if (ifp->if_type == IFT_IEEE8023ADLAG) { KASSERT(lagg_input_p != NULL, ("%s: if_lagg not loaded!", __func__)); m = (*lagg_input_p)(ifp, m); if (m != NULL) ifp = m->m_pkthdr.rcvif; else { CURVNET_RESTORE(); return; } } /* * If the hardware did not process an 802.1Q tag, do this now, * to allow 802.1P priority frames to be passed to the main input * path correctly. * TODO: Deal with Q-in-Q frames, but not arbitrary nesting levels. */ if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_VLAN) { struct ether_vlan_header *evl; if (m->m_len < sizeof(*evl) && (m = m_pullup(m, sizeof(*evl))) == NULL) { #ifdef DIAGNOSTIC if_printf(ifp, "cannot pullup VLAN header\n"); #endif if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); CURVNET_RESTORE(); return; } evl = mtod(m, struct ether_vlan_header *); m->m_pkthdr.ether_vtag = ntohs(evl->evl_tag); m->m_flags |= M_VLANTAG; bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN, ETHER_HDR_LEN - ETHER_TYPE_LEN); m_adj(m, ETHER_VLAN_ENCAP_LEN); eh = mtod(m, struct ether_header *); } M_SETFIB(m, ifp->if_fib); /* Allow ng_ether(4) to claim this frame. */ if (ifp->if_l2com != NULL) { KASSERT(ng_ether_input_p != NULL, ("%s: ng_ether_input_p is NULL", __func__)); m->m_flags &= ~M_PROMISC; (*ng_ether_input_p)(ifp, &m); if (m == NULL) { CURVNET_RESTORE(); return; } eh = mtod(m, struct ether_header *); } /* * Allow if_bridge(4) to claim this frame. * The BRIDGE_INPUT() macro will update ifp if the bridge changed it * and the frame should be delivered locally. */ if (ifp->if_bridge != NULL) { m->m_flags &= ~M_PROMISC; BRIDGE_INPUT(ifp, m); if (m == NULL) { CURVNET_RESTORE(); return; } eh = mtod(m, struct ether_header *); } #if defined(INET) || defined(INET6) /* * Clear M_PROMISC on frame so that carp(4) will see it when the * mbuf flows up to Layer 3. * FreeBSD's implementation of carp(4) uses the inprotosw * to dispatch IPPROTO_CARP. carp(4) also allocates its own * Ethernet addresses of the form 00:00:5e:00:01:xx, which * is outside the scope of the M_PROMISC test below. * TODO: Maintain a hash table of ethernet addresses other than * ether_dhost which may be active on this ifp. */ if (ifp->if_carp && (*carp_forus_p)(ifp, eh->ether_dhost)) { m->m_flags &= ~M_PROMISC; } else #endif { /* * If the frame received was not for our MAC address, set the * M_PROMISC flag on the mbuf chain. The frame may need to * be seen by the rest of the Ethernet input path in case of * re-entry (e.g. bridge, vlan, netgraph) but should not be * seen by upper protocol layers. */ if (!ETHER_IS_MULTICAST(eh->ether_dhost) && bcmp(IF_LLADDR(ifp), eh->ether_dhost, ETHER_ADDR_LEN) != 0) m->m_flags |= M_PROMISC; } ether_demux(ifp, m); CURVNET_RESTORE(); } /* * Ethernet input dispatch; by default, direct dispatch here regardless of * global configuration. However, if RSS is enabled, hook up RSS affinity * so that when deferred or hybrid dispatch is enabled, we can redistribute * load based on RSS. * * XXXRW: Would be nice if the ifnet passed up a flag indicating whether or * not it had already done work distribution via multi-queue. Then we could * direct dispatch in the event load balancing was already complete and * handle the case of interfaces with different capabilities better. * * XXXRW: Sort of want an M_DISTRIBUTED flag to avoid multiple distributions * at multiple layers? * * XXXRW: For now, enable all this only if RSS is compiled in, although it * works fine without RSS. Need to characterise the performance overhead * of the detour through the netisr code in the event the result is always * direct dispatch. */ static void ether_nh_input(struct mbuf *m) { M_ASSERTPKTHDR(m); KASSERT(m->m_pkthdr.rcvif != NULL, ("%s: NULL interface pointer", __func__)); ether_input_internal(m->m_pkthdr.rcvif, m); } static struct netisr_handler ether_nh = { .nh_name = "ether", .nh_handler = ether_nh_input, .nh_proto = NETISR_ETHER, #ifdef RSS .nh_policy = NETISR_POLICY_CPU, .nh_dispatch = NETISR_DISPATCH_DIRECT, .nh_m2cpuid = rss_m2cpuid, #else .nh_policy = NETISR_POLICY_SOURCE, .nh_dispatch = NETISR_DISPATCH_DIRECT, #endif }; static void ether_init(__unused void *arg) { netisr_register(ðer_nh); } SYSINIT(ether, SI_SUB_INIT_IF, SI_ORDER_ANY, ether_init, NULL); static void vnet_ether_init(__unused void *arg) { int i; /* Initialize packet filter hooks. */ V_link_pfil_hook.ph_type = PFIL_TYPE_AF; V_link_pfil_hook.ph_af = AF_LINK; if ((i = pfil_head_register(&V_link_pfil_hook)) != 0) printf("%s: WARNING: unable to register pfil link hook, " "error %d\n", __func__, i); #ifdef VIMAGE netisr_register_vnet(ðer_nh); #endif } VNET_SYSINIT(vnet_ether_init, SI_SUB_PROTO_IF, SI_ORDER_ANY, vnet_ether_init, NULL); #ifdef VIMAGE static void vnet_ether_pfil_destroy(__unused void *arg) { int i; if ((i = pfil_head_unregister(&V_link_pfil_hook)) != 0) printf("%s: WARNING: unable to unregister pfil link hook, " "error %d\n", __func__, i); } VNET_SYSUNINIT(vnet_ether_pfil_uninit, SI_SUB_PROTO_PFIL, SI_ORDER_ANY, vnet_ether_pfil_destroy, NULL); static void vnet_ether_destroy(__unused void *arg) { netisr_unregister_vnet(ðer_nh); } VNET_SYSUNINIT(vnet_ether_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, vnet_ether_destroy, NULL); #endif static void ether_input(struct ifnet *ifp, struct mbuf *m) { struct mbuf *mn; /* * The drivers are allowed to pass in a chain of packets linked with * m_nextpkt. We split them up into separate packets here and pass * them up. This allows the drivers to amortize the receive lock. */ while (m) { mn = m->m_nextpkt; m->m_nextpkt = NULL; /* * We will rely on rcvif being set properly in the deferred context, * so assert it is correct here. */ KASSERT(m->m_pkthdr.rcvif == ifp, ("%s: ifnet mismatch m %p " "rcvif %p ifp %p", __func__, m, m->m_pkthdr.rcvif, ifp)); CURVNET_SET_QUIET(ifp->if_vnet); netisr_dispatch(NETISR_ETHER, m); CURVNET_RESTORE(); m = mn; } } /* * Upper layer processing for a received Ethernet packet. */ void ether_demux(struct ifnet *ifp, struct mbuf *m) { struct ether_header *eh; int i, isr; u_short ether_type; KASSERT(ifp != NULL, ("%s: NULL interface pointer", __func__)); /* Do not grab PROMISC frames in case we are re-entered. */ if (PFIL_HOOKED(&V_link_pfil_hook) && !(m->m_flags & M_PROMISC)) { i = pfil_run_hooks(&V_link_pfil_hook, &m, ifp, PFIL_IN, 0, NULL); if (i != 0 || m == NULL) return; } eh = mtod(m, struct ether_header *); ether_type = ntohs(eh->ether_type); /* * If this frame has a VLAN tag other than 0, call vlan_input() * if its module is loaded. Otherwise, drop. */ if ((m->m_flags & M_VLANTAG) && EVL_VLANOFTAG(m->m_pkthdr.ether_vtag) != 0) { if (ifp->if_vlantrunk == NULL) { if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1); m_freem(m); return; } KASSERT(vlan_input_p != NULL,("%s: VLAN not loaded!", __func__)); /* Clear before possibly re-entering ether_input(). */ m->m_flags &= ~M_PROMISC; (*vlan_input_p)(ifp, m); return; } /* * Pass promiscuously received frames to the upper layer if the user * requested this by setting IFF_PPROMISC. Otherwise, drop them. */ if ((ifp->if_flags & IFF_PPROMISC) == 0 && (m->m_flags & M_PROMISC)) { m_freem(m); return; } /* * Reset layer specific mbuf flags to avoid confusing upper layers. * Strip off Ethernet header. */ m->m_flags &= ~M_VLANTAG; m_clrprotoflags(m); m_adj(m, ETHER_HDR_LEN); /* * Dispatch frame to upper layer. */ switch (ether_type) { #ifdef INET case ETHERTYPE_IP: isr = NETISR_IP; break; case ETHERTYPE_ARP: if (ifp->if_flags & IFF_NOARP) { /* Discard packet if ARP is disabled on interface */ m_freem(m); return; } isr = NETISR_ARP; break; #endif #ifdef INET6 case ETHERTYPE_IPV6: isr = NETISR_IPV6; break; #endif default: goto discard; } netisr_dispatch(isr, m); return; discard: /* * Packet is to be discarded. If netgraph is present, * hand the packet to it for last chance processing; * otherwise dispose of it. */ if (ifp->if_l2com != NULL) { KASSERT(ng_ether_input_orphan_p != NULL, ("ng_ether_input_orphan_p is NULL")); /* * Put back the ethernet header so netgraph has a * consistent view of inbound packets. */ M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); (*ng_ether_input_orphan_p)(ifp, m); return; } m_freem(m); } /* * Convert Ethernet address to printable (loggable) representation. * This routine is for compatibility; it's better to just use * * printf("%6D", , ":"); * * since there's no static buffer involved. */ char * ether_sprintf(const u_char *ap) { static char etherbuf[18]; snprintf(etherbuf, sizeof (etherbuf), "%6D", ap, ":"); return (etherbuf); } /* * Perform common duties while attaching to interface list */ void ether_ifattach(struct ifnet *ifp, const u_int8_t *lla) { int i; struct ifaddr *ifa; struct sockaddr_dl *sdl; ifp->if_addrlen = ETHER_ADDR_LEN; ifp->if_hdrlen = ETHER_HDR_LEN; if_attach(ifp); ifp->if_mtu = ETHERMTU; ifp->if_output = ether_output; ifp->if_input = ether_input; ifp->if_resolvemulti = ether_resolvemulti; ifp->if_requestencap = ether_requestencap; #ifdef VIMAGE ifp->if_reassign = ether_reassign; #endif if (ifp->if_baudrate == 0) ifp->if_baudrate = IF_Mbps(10); /* just a default */ ifp->if_broadcastaddr = etherbroadcastaddr; ifa = ifp->if_addr; KASSERT(ifa != NULL, ("%s: no lladdr!\n", __func__)); sdl = (struct sockaddr_dl *)ifa->ifa_addr; sdl->sdl_type = IFT_ETHER; sdl->sdl_alen = ifp->if_addrlen; bcopy(lla, LLADDR(sdl), ifp->if_addrlen); if (ifp->if_hw_addr != NULL) bcopy(lla, ifp->if_hw_addr, ifp->if_addrlen); bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); if (ng_ether_attach_p != NULL) (*ng_ether_attach_p)(ifp); /* Announce Ethernet MAC address if non-zero. */ for (i = 0; i < ifp->if_addrlen; i++) if (lla[i] != 0) break; if (i != ifp->if_addrlen) if_printf(ifp, "Ethernet address: %6D\n", lla, ":"); uuid_ether_add(LLADDR(sdl)); /* Add necessary bits are setup; announce it now. */ EVENTHANDLER_INVOKE(ether_ifattach_event, ifp); if (IS_DEFAULT_VNET(curvnet)) devctl_notify("ETHERNET", ifp->if_xname, "IFATTACH", NULL); } /* * Perform common duties while detaching an Ethernet interface */ void ether_ifdetach(struct ifnet *ifp) { struct sockaddr_dl *sdl; sdl = (struct sockaddr_dl *)(ifp->if_addr->ifa_addr); uuid_ether_del(LLADDR(sdl)); if (ifp->if_l2com != NULL) { KASSERT(ng_ether_detach_p != NULL, ("ng_ether_detach_p is NULL")); (*ng_ether_detach_p)(ifp); } bpfdetach(ifp); if_detach(ifp); } #ifdef VIMAGE void ether_reassign(struct ifnet *ifp, struct vnet *new_vnet, char *unused __unused) { if (ifp->if_l2com != NULL) { KASSERT(ng_ether_detach_p != NULL, ("ng_ether_detach_p is NULL")); (*ng_ether_detach_p)(ifp); } if (ng_ether_attach_p != NULL) { CURVNET_SET_QUIET(new_vnet); (*ng_ether_attach_p)(ifp); CURVNET_RESTORE(); } } #endif SYSCTL_DECL(_net_link); SYSCTL_NODE(_net_link, IFT_ETHER, ether, CTLFLAG_RW, 0, "Ethernet"); #if 0 /* * This is for reference. We have a table-driven version * of the little-endian crc32 generator, which is faster * than the double-loop. */ uint32_t ether_crc32_le(const uint8_t *buf, size_t len) { size_t i; uint32_t crc; int bit; uint8_t data; crc = 0xffffffff; /* initial value */ for (i = 0; i < len; i++) { for (data = *buf++, bit = 0; bit < 8; bit++, data >>= 1) { carry = (crc ^ data) & 1; crc >>= 1; if (carry) crc = (crc ^ ETHER_CRC_POLY_LE); } } return (crc); } #else uint32_t ether_crc32_le(const uint8_t *buf, size_t len) { static const uint32_t crctab[] = { 0x00000000, 0x1db71064, 0x3b6e20c8, 0x26d930ac, 0x76dc4190, 0x6b6b51f4, 0x4db26158, 0x5005713c, 0xedb88320, 0xf00f9344, 0xd6d6a3e8, 0xcb61b38c, 0x9b64c2b0, 0x86d3d2d4, 0xa00ae278, 0xbdbdf21c }; size_t i; uint32_t crc; crc = 0xffffffff; /* initial value */ for (i = 0; i < len; i++) { crc ^= buf[i]; crc = (crc >> 4) ^ crctab[crc & 0xf]; crc = (crc >> 4) ^ crctab[crc & 0xf]; } return (crc); } #endif uint32_t ether_crc32_be(const uint8_t *buf, size_t len) { size_t i; uint32_t crc, carry; int bit; uint8_t data; crc = 0xffffffff; /* initial value */ for (i = 0; i < len; i++) { for (data = *buf++, bit = 0; bit < 8; bit++, data >>= 1) { carry = ((crc & 0x80000000) ? 1 : 0) ^ (data & 0x01); crc <<= 1; if (carry) crc = (crc ^ ETHER_CRC_POLY_BE) | carry; } } return (crc); } int ether_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct ifaddr *ifa = (struct ifaddr *) data; struct ifreq *ifr = (struct ifreq *) data; int error = 0; switch (command) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: ifp->if_init(ifp->if_softc); /* before arpwhohas */ arp_ifinit(ifp, ifa); break; #endif default: ifp->if_init(ifp->if_softc); break; } break; case SIOCGIFADDR: bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0], ETHER_ADDR_LEN); break; case SIOCSIFMTU: /* * Set the interface MTU. */ if (ifr->ifr_mtu > ETHERMTU) { error = EINVAL; } else { ifp->if_mtu = ifr->ifr_mtu; } break; case SIOCSLANPCP: error = priv_check(curthread, PRIV_NET_SETLANPCP); if (error != 0) break; if (ifr->ifr_lan_pcp > 7 && ifr->ifr_lan_pcp != IFNET_PCP_NONE) { error = EINVAL; } else { ifp->if_pcp = ifr->ifr_lan_pcp; /* broadcast event about PCP change */ EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP); } break; case SIOCGLANPCP: ifr->ifr_lan_pcp = ifp->if_pcp; break; default: error = EINVAL; /* XXX netbsd has ENOTTY??? */ break; } return (error); } static int ether_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa, struct sockaddr *sa) { struct sockaddr_dl *sdl; #ifdef INET struct sockaddr_in *sin; #endif #ifdef INET6 struct sockaddr_in6 *sin6; #endif u_char *e_addr; switch(sa->sa_family) { case AF_LINK: /* * No mapping needed. Just check that it's a valid MC address. */ sdl = (struct sockaddr_dl *)sa; e_addr = LLADDR(sdl); if (!ETHER_IS_MULTICAST(e_addr)) return EADDRNOTAVAIL; *llsa = NULL; return 0; #ifdef INET case AF_INET: sin = (struct sockaddr_in *)sa; if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) return EADDRNOTAVAIL; sdl = link_init_sdl(ifp, *llsa, IFT_ETHER); sdl->sdl_alen = ETHER_ADDR_LEN; e_addr = LLADDR(sdl); ETHER_MAP_IP_MULTICAST(&sin->sin_addr, e_addr); *llsa = (struct sockaddr *)sdl; return 0; #endif #ifdef INET6 case AF_INET6: sin6 = (struct sockaddr_in6 *)sa; if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { /* * An IP6 address of 0 means listen to all * of the Ethernet multicast address used for IP6. * (This is used for multicast routers.) */ ifp->if_flags |= IFF_ALLMULTI; *llsa = NULL; return 0; } if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) return EADDRNOTAVAIL; sdl = link_init_sdl(ifp, *llsa, IFT_ETHER); sdl->sdl_alen = ETHER_ADDR_LEN; e_addr = LLADDR(sdl); ETHER_MAP_IPV6_MULTICAST(&sin6->sin6_addr, e_addr); *llsa = (struct sockaddr *)sdl; return 0; #endif default: /* * Well, the text isn't quite right, but it's the name * that counts... */ return EAFNOSUPPORT; } } static moduledata_t ether_mod = { .name = "ether", }; void ether_vlan_mtap(struct bpf_if *bp, struct mbuf *m, void *data, u_int dlen) { struct ether_vlan_header vlan; struct mbuf mv, mb; KASSERT((m->m_flags & M_VLANTAG) != 0, ("%s: vlan information not present", __func__)); KASSERT(m->m_len >= sizeof(struct ether_header), ("%s: mbuf not large enough for header", __func__)); bcopy(mtod(m, char *), &vlan, sizeof(struct ether_header)); vlan.evl_proto = vlan.evl_encap_proto; vlan.evl_encap_proto = htons(ETHERTYPE_VLAN); vlan.evl_tag = htons(m->m_pkthdr.ether_vtag); m->m_len -= sizeof(struct ether_header); m->m_data += sizeof(struct ether_header); /* * If a data link has been supplied by the caller, then we will need to * re-create a stack allocated mbuf chain with the following structure: * * (1) mbuf #1 will contain the supplied data link * (2) mbuf #2 will contain the vlan header * (3) mbuf #3 will contain the original mbuf's packet data * * Otherwise, submit the packet and vlan header via bpf_mtap2(). */ if (data != NULL) { mv.m_next = m; mv.m_data = (caddr_t)&vlan; mv.m_len = sizeof(vlan); mb.m_next = &mv; mb.m_data = data; mb.m_len = dlen; bpf_mtap(bp, &mb); } else bpf_mtap2(bp, &vlan, sizeof(vlan), m); m->m_len += sizeof(struct ether_header); m->m_data -= sizeof(struct ether_header); } struct mbuf * ether_vlanencap(struct mbuf *m, uint16_t tag) { struct ether_vlan_header *evl; M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT); if (m == NULL) return (NULL); /* M_PREPEND takes care of m_len, m_pkthdr.len for us */ if (m->m_len < sizeof(*evl)) { m = m_pullup(m, sizeof(*evl)); if (m == NULL) return (NULL); } /* * Transform the Ethernet header into an Ethernet header * with 802.1Q encapsulation. */ evl = mtod(m, struct ether_vlan_header *); bcopy((char *)evl + ETHER_VLAN_ENCAP_LEN, (char *)evl, ETHER_HDR_LEN - ETHER_TYPE_LEN); evl->evl_encap_proto = htons(ETHERTYPE_VLAN); evl->evl_tag = htons(tag); return (m); } static SYSCTL_NODE(_net_link, IFT_L2VLAN, vlan, CTLFLAG_RW, 0, "IEEE 802.1Q VLAN"); static SYSCTL_NODE(_net_link_vlan, PF_LINK, link, CTLFLAG_RW, 0, "for consistency"); -static VNET_DEFINE(int, soft_pad); +VNET_DEFINE_STATIC(int, soft_pad); #define V_soft_pad VNET(soft_pad) SYSCTL_INT(_net_link_vlan, OID_AUTO, soft_pad, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(soft_pad), 0, "pad short frames before tagging"); /* * For now, make preserving PCP via an mbuf tag optional, as it increases * per-packet memory allocations and frees. In the future, it would be * preferable to reuse ether_vtag for this, or similar. */ int vlan_mtag_pcp = 0; SYSCTL_INT(_net_link_vlan, OID_AUTO, mtag_pcp, CTLFLAG_RW, &vlan_mtag_pcp, 0, "Retain VLAN PCP information as packets are passed up the stack"); bool ether_8021q_frame(struct mbuf **mp, struct ifnet *ife, struct ifnet *p, uint16_t vid, uint8_t pcp) { struct m_tag *mtag; int n; uint16_t tag; static const char pad[8]; /* just zeros */ /* * Pad the frame to the minimum size allowed if told to. * This option is in accord with IEEE Std 802.1Q, 2003 Ed., * paragraph C.4.4.3.b. It can help to work around buggy * bridges that violate paragraph C.4.4.3.a from the same * document, i.e., fail to pad short frames after untagging. * E.g., a tagged frame 66 bytes long (incl. FCS) is OK, but * untagging it will produce a 62-byte frame, which is a runt * and requires padding. There are VLAN-enabled network * devices that just discard such runts instead or mishandle * them somehow. */ if (V_soft_pad && p->if_type == IFT_ETHER) { for (n = ETHERMIN + ETHER_HDR_LEN - (*mp)->m_pkthdr.len; n > 0; n -= sizeof(pad)) { if (!m_append(*mp, min(n, sizeof(pad)), pad)) break; } if (n > 0) { m_freem(*mp); *mp = NULL; if_printf(ife, "cannot pad short frame"); return (false); } } /* * If underlying interface can do VLAN tag insertion itself, * just pass the packet along. However, we need some way to * tell the interface where the packet came from so that it * knows how to find the VLAN tag to use, so we attach a * packet tag that holds it. */ if (vlan_mtag_pcp && (mtag = m_tag_locate(*mp, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL)) != NULL) tag = EVL_MAKETAG(vid, *(uint8_t *)(mtag + 1), 0); else tag = EVL_MAKETAG(vid, pcp, 0); if (p->if_capenable & IFCAP_VLAN_HWTAGGING) { (*mp)->m_pkthdr.ether_vtag = tag; (*mp)->m_flags |= M_VLANTAG; } else { *mp = ether_vlanencap(*mp, tag); if (*mp == NULL) { if_printf(ife, "unable to prepend 802.1Q header"); return (false); } } return (true); } DECLARE_MODULE(ether, ether_mod, SI_SUB_INIT_IF, SI_ORDER_ANY); MODULE_VERSION(ether, 1); Index: head/sys/net/if_gif.c =================================================================== --- head/sys/net/if_gif.c (revision 336675) +++ head/sys/net/if_gif.c (revision 336676) @@ -1,701 +1,701 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * Copyright (c) 2018 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: if_gif.c,v 1.87 2001/10/19 08:50:27 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif /* INET */ #ifdef INET6 #ifndef INET #include #endif #include #include #include #include #endif /* INET6 */ #include #include #include #include #include static const char gifname[] = "gif"; MALLOC_DEFINE(M_GIF, "gif", "Generic Tunnel Interface"); static struct sx gif_ioctl_sx; SX_SYSINIT(gif_ioctl_sx, &gif_ioctl_sx, "gif_ioctl"); void (*ng_gif_input_p)(struct ifnet *ifp, struct mbuf **mp, int af); void (*ng_gif_input_orphan_p)(struct ifnet *ifp, struct mbuf *m, int af); void (*ng_gif_attach_p)(struct ifnet *ifp); void (*ng_gif_detach_p)(struct ifnet *ifp); static void gif_delete_tunnel(struct gif_softc *); static int gif_ioctl(struct ifnet *, u_long, caddr_t); static int gif_transmit(struct ifnet *, struct mbuf *); static void gif_qflush(struct ifnet *); static int gif_clone_create(struct if_clone *, int, caddr_t); static void gif_clone_destroy(struct ifnet *); -static VNET_DEFINE(struct if_clone *, gif_cloner); +VNET_DEFINE_STATIC(struct if_clone *, gif_cloner); #define V_gif_cloner VNET(gif_cloner) SYSCTL_DECL(_net_link); static SYSCTL_NODE(_net_link, IFT_GIF, gif, CTLFLAG_RW, 0, "Generic Tunnel Interface"); #ifndef MAX_GIF_NEST /* * This macro controls the default upper limitation on nesting of gif tunnels. * Since, setting a large value to this macro with a careless configuration * may introduce system crash, we don't allow any nestings by default. * If you need to configure nested gif tunnels, you can define this macro * in your kernel configuration file. However, if you do so, please be * careful to configure the tunnels so that it won't make a loop. */ #define MAX_GIF_NEST 1 #endif -static VNET_DEFINE(int, max_gif_nesting) = MAX_GIF_NEST; +VNET_DEFINE_STATIC(int, max_gif_nesting) = MAX_GIF_NEST; #define V_max_gif_nesting VNET(max_gif_nesting) SYSCTL_INT(_net_link_gif, OID_AUTO, max_nesting, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(max_gif_nesting), 0, "Max nested tunnels"); static int gif_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct gif_softc *sc; sc = malloc(sizeof(struct gif_softc), M_GIF, M_WAITOK | M_ZERO); sc->gif_fibnum = curthread->td_proc->p_fibnum; GIF2IFP(sc) = if_alloc(IFT_GIF); GIF2IFP(sc)->if_softc = sc; if_initname(GIF2IFP(sc), gifname, unit); GIF2IFP(sc)->if_addrlen = 0; GIF2IFP(sc)->if_mtu = GIF_MTU; GIF2IFP(sc)->if_flags = IFF_POINTOPOINT | IFF_MULTICAST; GIF2IFP(sc)->if_ioctl = gif_ioctl; GIF2IFP(sc)->if_transmit = gif_transmit; GIF2IFP(sc)->if_qflush = gif_qflush; GIF2IFP(sc)->if_output = gif_output; GIF2IFP(sc)->if_capabilities |= IFCAP_LINKSTATE; GIF2IFP(sc)->if_capenable |= IFCAP_LINKSTATE; if_attach(GIF2IFP(sc)); bpfattach(GIF2IFP(sc), DLT_NULL, sizeof(u_int32_t)); if (ng_gif_attach_p != NULL) (*ng_gif_attach_p)(GIF2IFP(sc)); return (0); } static void gif_clone_destroy(struct ifnet *ifp) { struct gif_softc *sc; sx_xlock(&gif_ioctl_sx); sc = ifp->if_softc; gif_delete_tunnel(sc); if (ng_gif_detach_p != NULL) (*ng_gif_detach_p)(ifp); bpfdetach(ifp); if_detach(ifp); ifp->if_softc = NULL; sx_xunlock(&gif_ioctl_sx); GIF_WAIT(); if_free(ifp); free(sc, M_GIF); } static void vnet_gif_init(const void *unused __unused) { V_gif_cloner = if_clone_simple(gifname, gif_clone_create, gif_clone_destroy, 0); #ifdef INET in_gif_init(); #endif #ifdef INET6 in6_gif_init(); #endif } VNET_SYSINIT(vnet_gif_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_gif_init, NULL); static void vnet_gif_uninit(const void *unused __unused) { if_clone_detach(V_gif_cloner); #ifdef INET in_gif_uninit(); #endif #ifdef INET6 in6_gif_uninit(); #endif } VNET_SYSUNINIT(vnet_gif_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_gif_uninit, NULL); static int gifmodevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t gif_mod = { "if_gif", gifmodevent, 0 }; DECLARE_MODULE(if_gif, gif_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_gif, 1); struct gif_list * gif_hashinit(void) { struct gif_list *hash; int i; hash = malloc(sizeof(struct gif_list) * GIF_HASH_SIZE, M_GIF, M_WAITOK); for (i = 0; i < GIF_HASH_SIZE; i++) CK_LIST_INIT(&hash[i]); return (hash); } void gif_hashdestroy(struct gif_list *hash) { free(hash, M_GIF); } #define MTAG_GIF 1080679712 static int gif_transmit(struct ifnet *ifp, struct mbuf *m) { struct gif_softc *sc; struct etherip_header *eth; #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; uint32_t t; #endif uint32_t af; uint8_t proto, ecn; int error; #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error) { m_freem(m); goto err; } #endif error = ENETDOWN; GIF_RLOCK(); sc = ifp->if_softc; if ((ifp->if_flags & IFF_MONITOR) != 0 || (ifp->if_flags & IFF_UP) == 0 || sc->gif_family == 0 || (error = if_tunnel_check_nesting(ifp, m, MTAG_GIF, V_max_gif_nesting)) != 0) { m_freem(m); goto err; } /* Now pull back the af that we stashed in the csum_data. */ if (ifp->if_bridge) af = AF_LINK; else af = m->m_pkthdr.csum_data; m->m_flags &= ~(M_BCAST|M_MCAST); M_SETFIB(m, sc->gif_fibnum); BPF_MTAP2(ifp, &af, sizeof(af), m); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); /* inner AF-specific encapsulation */ ecn = 0; switch (af) { #ifdef INET case AF_INET: proto = IPPROTO_IPV4; if (m->m_len < sizeof(struct ip)) m = m_pullup(m, sizeof(struct ip)); if (m == NULL) { error = ENOBUFS; goto err; } ip = mtod(m, struct ip *); ip_ecn_ingress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED: ECN_NOCARE, &ecn, &ip->ip_tos); break; #endif #ifdef INET6 case AF_INET6: proto = IPPROTO_IPV6; if (m->m_len < sizeof(struct ip6_hdr)) m = m_pullup(m, sizeof(struct ip6_hdr)); if (m == NULL) { error = ENOBUFS; goto err; } t = 0; ip6 = mtod(m, struct ip6_hdr *); ip6_ecn_ingress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED: ECN_NOCARE, &t, &ip6->ip6_flow); ecn = (ntohl(t) >> 20) & 0xff; break; #endif case AF_LINK: proto = IPPROTO_ETHERIP; M_PREPEND(m, sizeof(struct etherip_header), M_NOWAIT); if (m == NULL) { error = ENOBUFS; goto err; } eth = mtod(m, struct etherip_header *); eth->eip_resvh = 0; eth->eip_ver = ETHERIP_VERSION; eth->eip_resvl = 0; break; default: error = EAFNOSUPPORT; m_freem(m); goto err; } /* XXX should we check if our outer source is legal? */ /* dispatch to output logic based on outer AF */ switch (sc->gif_family) { #ifdef INET case AF_INET: error = in_gif_output(ifp, m, proto, ecn); break; #endif #ifdef INET6 case AF_INET6: error = in6_gif_output(ifp, m, proto, ecn); break; #endif default: m_freem(m); } err: if (error) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); GIF_RUNLOCK(); return (error); } static void gif_qflush(struct ifnet *ifp __unused) { } int gif_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { uint32_t af; if (dst->sa_family == AF_UNSPEC) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; /* * Now save the af in the inbound pkt csum data, this is a cheat since * we are using the inbound csum_data field to carry the af over to * the gif_transmit() routine, avoiding using yet another mtag. */ m->m_pkthdr.csum_data = af; return (ifp->if_transmit(ifp, m)); } void gif_input(struct mbuf *m, struct ifnet *ifp, int proto, uint8_t ecn) { struct etherip_header *eip; #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; uint32_t t; #endif struct ether_header *eh; struct ifnet *oldifp; int isr, n, af; if (ifp == NULL) { /* just in case */ m_freem(m); return; } m->m_pkthdr.rcvif = ifp; m_clrprotoflags(m); switch (proto) { #ifdef INET case IPPROTO_IPV4: af = AF_INET; if (m->m_len < sizeof(struct ip)) m = m_pullup(m, sizeof(struct ip)); if (m == NULL) goto drop; ip = mtod(m, struct ip *); if (ip_ecn_egress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED: ECN_NOCARE, &ecn, &ip->ip_tos) == 0) { m_freem(m); goto drop; } break; #endif #ifdef INET6 case IPPROTO_IPV6: af = AF_INET6; if (m->m_len < sizeof(struct ip6_hdr)) m = m_pullup(m, sizeof(struct ip6_hdr)); if (m == NULL) goto drop; t = htonl((uint32_t)ecn << 20); ip6 = mtod(m, struct ip6_hdr *); if (ip6_ecn_egress((ifp->if_flags & IFF_LINK1) ? ECN_ALLOWED: ECN_NOCARE, &t, &ip6->ip6_flow) == 0) { m_freem(m); goto drop; } break; #endif case IPPROTO_ETHERIP: af = AF_LINK; break; default: m_freem(m); goto drop; } #ifdef MAC mac_ifnet_create_mbuf(ifp, m); #endif if (bpf_peers_present(ifp->if_bpf)) { uint32_t af1 = af; bpf_mtap2(ifp->if_bpf, &af1, sizeof(af1), m); } if ((ifp->if_flags & IFF_MONITOR) != 0) { if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); m_freem(m); return; } if (ng_gif_input_p != NULL) { (*ng_gif_input_p)(ifp, &m, af); if (m == NULL) goto drop; } /* * Put the packet to the network layer input queue according to the * specified address family. * Note: older versions of gif_input directly called network layer * input functions, e.g. ip6_input, here. We changed the policy to * prevent too many recursive calls of such input functions, which * might cause kernel panic. But the change may introduce another * problem; if the input queue is full, packets are discarded. * The kernel stack overflow really happened, and we believed * queue-full rarely occurs, so we changed the policy. */ switch (af) { #ifdef INET case AF_INET: isr = NETISR_IP; break; #endif #ifdef INET6 case AF_INET6: isr = NETISR_IPV6; break; #endif case AF_LINK: n = sizeof(struct etherip_header) + sizeof(struct ether_header); if (n > m->m_len) m = m_pullup(m, n); if (m == NULL) goto drop; eip = mtod(m, struct etherip_header *); if (eip->eip_ver != ETHERIP_VERSION) { /* discard unknown versions */ m_freem(m); goto drop; } m_adj(m, sizeof(struct etherip_header)); m->m_flags &= ~(M_BCAST|M_MCAST); m->m_pkthdr.rcvif = ifp; if (ifp->if_bridge) { oldifp = ifp; eh = mtod(m, struct ether_header *); if (ETHER_IS_MULTICAST(eh->ether_dhost)) { if (ETHER_IS_BROADCAST(eh->ether_dhost)) m->m_flags |= M_BCAST; else m->m_flags |= M_MCAST; if_inc_counter(ifp, IFCOUNTER_IMCASTS, 1); } BRIDGE_INPUT(ifp, m); if (m != NULL && ifp != oldifp) { /* * The bridge gave us back itself or one of the * members for which the frame is addressed. */ ether_demux(ifp, m); return; } } if (m != NULL) m_freem(m); return; default: if (ng_gif_input_orphan_p != NULL) (*ng_gif_input_orphan_p)(ifp, m, af); else m_freem(m); return; } if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); M_SETFIB(m, ifp->if_fib); netisr_dispatch(isr, m); return; drop: if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); } static int gif_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq*)data; struct gif_softc *sc; u_int options; int error; switch (cmd) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; case SIOCADDMULTI: case SIOCDELMULTI: case SIOCGIFMTU: case SIOCSIFFLAGS: return (0); case SIOCSIFMTU: if (ifr->ifr_mtu < GIF_MTU_MIN || ifr->ifr_mtu > GIF_MTU_MAX) return (EINVAL); else ifp->if_mtu = ifr->ifr_mtu; return (0); } sx_xlock(&gif_ioctl_sx); sc = ifp->if_softc; if (sc == NULL) { error = ENXIO; goto bad; } error = 0; switch (cmd) { case SIOCDIFPHYADDR: if (sc->gif_family == 0) break; gif_delete_tunnel(sc); break; #ifdef INET case SIOCSIFPHYADDR: case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: error = in_gif_ioctl(sc, cmd, data); break; #endif #ifdef INET6 case SIOCSIFPHYADDR_IN6: case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: error = in6_gif_ioctl(sc, cmd, data); break; #endif case SIOCGTUNFIB: ifr->ifr_fib = sc->gif_fibnum; break; case SIOCSTUNFIB: if ((error = priv_check(curthread, PRIV_NET_GIF)) != 0) break; if (ifr->ifr_fib >= rt_numfibs) error = EINVAL; else sc->gif_fibnum = ifr->ifr_fib; break; case GIFGOPTS: options = sc->gif_options; error = copyout(&options, ifr_data_get_ptr(ifr), sizeof(options)); break; case GIFSOPTS: if ((error = priv_check(curthread, PRIV_NET_GIF)) != 0) break; error = copyin(ifr_data_get_ptr(ifr), &options, sizeof(options)); if (error) break; if (options & ~GIF_OPTMASK) { error = EINVAL; break; } if (sc->gif_options != options) { switch (sc->gif_family) { #ifdef INET case AF_INET: error = in_gif_setopts(sc, options); break; #endif #ifdef INET6 case AF_INET6: error = in6_gif_setopts(sc, options); break; #endif default: /* No need to invoke AF-handler */ sc->gif_options = options; } } break; default: error = EINVAL; break; } if (error == 0 && sc->gif_family != 0) { if ( #ifdef INET cmd == SIOCSIFPHYADDR || #endif #ifdef INET6 cmd == SIOCSIFPHYADDR_IN6 || #endif 0) { ifp->if_drv_flags |= IFF_DRV_RUNNING; if_link_state_change(ifp, LINK_STATE_UP); } } bad: sx_xunlock(&gif_ioctl_sx); return (error); } static void gif_delete_tunnel(struct gif_softc *sc) { sx_assert(&gif_ioctl_sx, SA_XLOCKED); if (sc->gif_family != 0) { CK_LIST_REMOVE(sc, chain); /* Wait until it become safe to free gif_hdr */ GIF_WAIT(); free(sc->gif_hdr, M_GIF); } sc->gif_family = 0; GIF2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; if_link_state_change(GIF2IFP(sc), LINK_STATE_DOWN); } Index: head/sys/net/if_gre.c =================================================================== --- head/sys/net/if_gre.c (revision 336675) +++ head/sys/net/if_gre.c (revision 336676) @@ -1,675 +1,675 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1998 The NetBSD Foundation, Inc. * Copyright (c) 2014, 2018 Andrey V. Elsukov * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Heiko W.Rupp * * IPv6-over-GRE contributed by Gert Doering * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * $NetBSD: if_gre.c,v 1.49 2003/12/11 00:22:29 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #include #endif #ifdef INET6 #include #include #include #endif #include #include #include #include #include #define GREMTU 1476 static const char grename[] = "gre"; MALLOC_DEFINE(M_GRE, grename, "Generic Routing Encapsulation"); static struct sx gre_ioctl_sx; SX_SYSINIT(gre_ioctl_sx, &gre_ioctl_sx, "gre_ioctl"); static int gre_clone_create(struct if_clone *, int, caddr_t); static void gre_clone_destroy(struct ifnet *); -static VNET_DEFINE(struct if_clone *, gre_cloner); +VNET_DEFINE_STATIC(struct if_clone *, gre_cloner); #define V_gre_cloner VNET(gre_cloner) static void gre_qflush(struct ifnet *); static int gre_transmit(struct ifnet *, struct mbuf *); static int gre_ioctl(struct ifnet *, u_long, caddr_t); static int gre_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static void gre_delete_tunnel(struct gre_softc *); SYSCTL_DECL(_net_link); static SYSCTL_NODE(_net_link, IFT_TUNNEL, gre, CTLFLAG_RW, 0, "Generic Routing Encapsulation"); #ifndef MAX_GRE_NEST /* * This macro controls the default upper limitation on nesting of gre tunnels. * Since, setting a large value to this macro with a careless configuration * may introduce system crash, we don't allow any nestings by default. * If you need to configure nested gre tunnels, you can define this macro * in your kernel configuration file. However, if you do so, please be * careful to configure the tunnels so that it won't make a loop. */ #define MAX_GRE_NEST 1 #endif -static VNET_DEFINE(int, max_gre_nesting) = MAX_GRE_NEST; +VNET_DEFINE_STATIC(int, max_gre_nesting) = MAX_GRE_NEST; #define V_max_gre_nesting VNET(max_gre_nesting) SYSCTL_INT(_net_link_gre, OID_AUTO, max_nesting, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(max_gre_nesting), 0, "Max nested tunnels"); static void vnet_gre_init(const void *unused __unused) { V_gre_cloner = if_clone_simple(grename, gre_clone_create, gre_clone_destroy, 0); #ifdef INET in_gre_init(); #endif #ifdef INET6 in6_gre_init(); #endif } VNET_SYSINIT(vnet_gre_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_gre_init, NULL); static void vnet_gre_uninit(const void *unused __unused) { if_clone_detach(V_gre_cloner); #ifdef INET in_gre_uninit(); #endif #ifdef INET6 in6_gre_uninit(); #endif } VNET_SYSUNINIT(vnet_gre_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_gre_uninit, NULL); static int gre_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct gre_softc *sc; sc = malloc(sizeof(struct gre_softc), M_GRE, M_WAITOK | M_ZERO); sc->gre_fibnum = curthread->td_proc->p_fibnum; GRE2IFP(sc) = if_alloc(IFT_TUNNEL); GRE2IFP(sc)->if_softc = sc; if_initname(GRE2IFP(sc), grename, unit); GRE2IFP(sc)->if_mtu = GREMTU; GRE2IFP(sc)->if_flags = IFF_POINTOPOINT|IFF_MULTICAST; GRE2IFP(sc)->if_output = gre_output; GRE2IFP(sc)->if_ioctl = gre_ioctl; GRE2IFP(sc)->if_transmit = gre_transmit; GRE2IFP(sc)->if_qflush = gre_qflush; GRE2IFP(sc)->if_capabilities |= IFCAP_LINKSTATE; GRE2IFP(sc)->if_capenable |= IFCAP_LINKSTATE; if_attach(GRE2IFP(sc)); bpfattach(GRE2IFP(sc), DLT_NULL, sizeof(u_int32_t)); return (0); } static void gre_clone_destroy(struct ifnet *ifp) { struct gre_softc *sc; sx_xlock(&gre_ioctl_sx); sc = ifp->if_softc; gre_delete_tunnel(sc); bpfdetach(ifp); if_detach(ifp); ifp->if_softc = NULL; sx_xunlock(&gre_ioctl_sx); GRE_WAIT(); if_free(ifp); free(sc, M_GRE); } static int gre_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; struct gre_softc *sc; uint32_t opt; int error; switch (cmd) { case SIOCSIFMTU: /* XXX: */ if (ifr->ifr_mtu < 576) return (EINVAL); ifp->if_mtu = ifr->ifr_mtu; return (0); case SIOCSIFADDR: ifp->if_flags |= IFF_UP; case SIOCSIFFLAGS: case SIOCADDMULTI: case SIOCDELMULTI: return (0); case GRESADDRS: case GRESADDRD: case GREGADDRS: case GREGADDRD: case GRESPROTO: case GREGPROTO: return (EOPNOTSUPP); } sx_xlock(&gre_ioctl_sx); sc = ifp->if_softc; if (sc == NULL) { error = ENXIO; goto end; } error = 0; switch (cmd) { case SIOCDIFPHYADDR: if (sc->gre_family == 0) break; gre_delete_tunnel(sc); break; #ifdef INET case SIOCSIFPHYADDR: case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: error = in_gre_ioctl(sc, cmd, data); break; #endif #ifdef INET6 case SIOCSIFPHYADDR_IN6: case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: error = in6_gre_ioctl(sc, cmd, data); break; #endif case SIOCGTUNFIB: ifr->ifr_fib = sc->gre_fibnum; break; case SIOCSTUNFIB: if ((error = priv_check(curthread, PRIV_NET_GRE)) != 0) break; if (ifr->ifr_fib >= rt_numfibs) error = EINVAL; else sc->gre_fibnum = ifr->ifr_fib; break; case GRESKEY: case GRESOPTS: if ((error = priv_check(curthread, PRIV_NET_GRE)) != 0) break; if ((error = copyin(ifr_data_get_ptr(ifr), &opt, sizeof(opt))) != 0) break; if (cmd == GRESKEY) { if (sc->gre_key == opt) break; } else if (cmd == GRESOPTS) { if (opt & ~GRE_OPTMASK) { error = EINVAL; break; } if (sc->gre_options == opt) break; } switch (sc->gre_family) { #ifdef INET case AF_INET: in_gre_setopts(sc, cmd, opt); break; #endif #ifdef INET6 case AF_INET6: in6_gre_setopts(sc, cmd, opt); break; #endif default: if (cmd == GRESKEY) sc->gre_key = opt; else sc->gre_options = opt; break; } /* * XXX: Do we need to initiate change of interface * state here? */ break; case GREGKEY: error = copyout(&sc->gre_key, ifr_data_get_ptr(ifr), sizeof(sc->gre_key)); break; case GREGOPTS: error = copyout(&sc->gre_options, ifr_data_get_ptr(ifr), sizeof(sc->gre_options)); break; default: error = EINVAL; break; } if (error == 0 && sc->gre_family != 0) { if ( #ifdef INET cmd == SIOCSIFPHYADDR || #endif #ifdef INET6 cmd == SIOCSIFPHYADDR_IN6 || #endif 0) { ifp->if_drv_flags |= IFF_DRV_RUNNING; if_link_state_change(ifp, LINK_STATE_UP); } } end: sx_xunlock(&gre_ioctl_sx); return (error); } static void gre_delete_tunnel(struct gre_softc *sc) { sx_assert(&gre_ioctl_sx, SA_XLOCKED); if (sc->gre_family != 0) { CK_LIST_REMOVE(sc, chain); GRE_WAIT(); free(sc->gre_hdr, M_GRE); sc->gre_family = 0; } GRE2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; if_link_state_change(GRE2IFP(sc), LINK_STATE_DOWN); } struct gre_list * gre_hashinit(void) { struct gre_list *hash; int i; hash = malloc(sizeof(struct gre_list) * GRE_HASH_SIZE, M_GRE, M_WAITOK); for (i = 0; i < GRE_HASH_SIZE; i++) CK_LIST_INIT(&hash[i]); return (hash); } void gre_hashdestroy(struct gre_list *hash) { free(hash, M_GRE); } void gre_updatehdr(struct gre_softc *sc, struct grehdr *gh) { uint32_t *opts; uint16_t flags; sx_assert(&gre_ioctl_sx, SA_XLOCKED); flags = 0; opts = gh->gre_opts; if (sc->gre_options & GRE_ENABLE_CSUM) { flags |= GRE_FLAGS_CP; sc->gre_hlen += 2 * sizeof(uint16_t); *opts++ = 0; } if (sc->gre_key != 0) { flags |= GRE_FLAGS_KP; sc->gre_hlen += sizeof(uint32_t); *opts++ = htonl(sc->gre_key); } if (sc->gre_options & GRE_ENABLE_SEQ) { flags |= GRE_FLAGS_SP; sc->gre_hlen += sizeof(uint32_t); *opts++ = 0; } else sc->gre_oseq = 0; gh->gre_flags = htons(flags); } int gre_input(struct mbuf *m, int off, int proto, void *arg) { struct gre_softc *sc = arg; struct grehdr *gh; struct ifnet *ifp; uint32_t *opts; #ifdef notyet uint32_t key; #endif uint16_t flags; int hlen, isr, af; ifp = GRE2IFP(sc); hlen = off + sizeof(struct grehdr) + 4 * sizeof(uint32_t); if (m->m_pkthdr.len < hlen) goto drop; if (m->m_len < hlen) { m = m_pullup(m, hlen); if (m == NULL) goto drop; } gh = (struct grehdr *)mtodo(m, off); flags = ntohs(gh->gre_flags); if (flags & ~GRE_FLAGS_MASK) goto drop; opts = gh->gre_opts; hlen = 2 * sizeof(uint16_t); if (flags & GRE_FLAGS_CP) { /* reserved1 field must be zero */ if (((uint16_t *)opts)[1] != 0) goto drop; if (in_cksum_skip(m, m->m_pkthdr.len, off) != 0) goto drop; hlen += 2 * sizeof(uint16_t); opts++; } if (flags & GRE_FLAGS_KP) { #ifdef notyet /* * XXX: The current implementation uses the key only for outgoing * packets. But we can check the key value here, or even in the * encapcheck function. */ key = ntohl(*opts); #endif hlen += sizeof(uint32_t); opts++; } #ifdef notyet } else key = 0; if (sc->gre_key != 0 && (key != sc->gre_key || key != 0)) goto drop; #endif if (flags & GRE_FLAGS_SP) { #ifdef notyet seq = ntohl(*opts); #endif hlen += sizeof(uint32_t); } switch (ntohs(gh->gre_proto)) { case ETHERTYPE_WCCP: /* * For WCCP skip an additional 4 bytes if after GRE header * doesn't follow an IP header. */ if (flags == 0 && (*(uint8_t *)gh->gre_opts & 0xF0) != 0x40) hlen += sizeof(uint32_t); /* FALLTHROUGH */ case ETHERTYPE_IP: isr = NETISR_IP; af = AF_INET; break; case ETHERTYPE_IPV6: isr = NETISR_IPV6; af = AF_INET6; break; default: goto drop; } m_adj(m, off + hlen); m_clrprotoflags(m); m->m_pkthdr.rcvif = ifp; M_SETFIB(m, ifp->if_fib); #ifdef MAC mac_ifnet_create_mbuf(ifp, m); #endif BPF_MTAP2(ifp, &af, sizeof(af), m); if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); if ((ifp->if_flags & IFF_MONITOR) != 0) m_freem(m); else netisr_dispatch(isr, m); return (IPPROTO_DONE); drop: if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); m_freem(m); return (IPPROTO_DONE); } static int gre_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { uint32_t af; if (dst->sa_family == AF_UNSPEC) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; /* * Now save the af in the inbound pkt csum data, this is a cheat since * we are using the inbound csum_data field to carry the af over to * the gre_transmit() routine, avoiding using yet another mtag. */ m->m_pkthdr.csum_data = af; return (ifp->if_transmit(ifp, m)); } static void gre_setseqn(struct grehdr *gh, uint32_t seq) { uint32_t *opts; uint16_t flags; opts = gh->gre_opts; flags = ntohs(gh->gre_flags); KASSERT((flags & GRE_FLAGS_SP) != 0, ("gre_setseqn called, but GRE_FLAGS_SP isn't set ")); if (flags & GRE_FLAGS_CP) opts++; if (flags & GRE_FLAGS_KP) opts++; *opts = htonl(seq); } #define MTAG_GRE 1307983903 static int gre_transmit(struct ifnet *ifp, struct mbuf *m) { struct gre_softc *sc; struct grehdr *gh; uint32_t af; int error, len; uint16_t proto; len = 0; #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error) { m_freem(m); goto drop; } #endif error = ENETDOWN; GRE_RLOCK(); sc = ifp->if_softc; if ((ifp->if_flags & IFF_MONITOR) != 0 || (ifp->if_flags & IFF_UP) == 0 || sc->gre_family == 0 || (error = if_tunnel_check_nesting(ifp, m, MTAG_GRE, V_max_gre_nesting)) != 0) { m_freem(m); goto drop; } af = m->m_pkthdr.csum_data; M_SETFIB(m, sc->gre_fibnum); M_PREPEND(m, sc->gre_hlen, M_NOWAIT); if (m == NULL) { error = ENOBUFS; goto drop; } bcopy(sc->gre_hdr, mtod(m, void *), sc->gre_hlen); /* Determine GRE proto */ switch (af) { #ifdef INET case AF_INET: proto = htons(ETHERTYPE_IP); break; #endif #ifdef INET6 case AF_INET6: proto = htons(ETHERTYPE_IPV6); break; #endif default: m_freem(m); error = ENETDOWN; goto drop; } /* Determine offset of GRE header */ switch (sc->gre_family) { #ifdef INET case AF_INET: len = sizeof(struct ip); break; #endif #ifdef INET6 case AF_INET6: len = sizeof(struct ip6_hdr); break; #endif default: m_freem(m); error = ENETDOWN; goto drop; } gh = (struct grehdr *)mtodo(m, len); gh->gre_proto = proto; if (sc->gre_options & GRE_ENABLE_SEQ) gre_setseqn(gh, sc->gre_oseq++); if (sc->gre_options & GRE_ENABLE_CSUM) { *(uint16_t *)gh->gre_opts = in_cksum_skip(m, m->m_pkthdr.len, len); } len = m->m_pkthdr.len - len; switch (sc->gre_family) { #ifdef INET case AF_INET: error = in_gre_output(m, af, sc->gre_hlen); break; #endif #ifdef INET6 case AF_INET6: error = in6_gre_output(m, af, sc->gre_hlen); break; #endif default: m_freem(m); error = ENETDOWN; } drop: if (error) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); else { if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, len); } GRE_RUNLOCK(); return (error); } static void gre_qflush(struct ifnet *ifp __unused) { } static int gremodevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t gre_mod = { "if_gre", gremodevent, 0 }; DECLARE_MODULE(if_gre, gre_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_gre, 1); Index: head/sys/net/if_ipsec.c =================================================================== --- head/sys/net/if_ipsec.c (revision 336675) +++ head/sys/net/if_ipsec.c (revision 336676) @@ -1,1002 +1,1002 @@ /*- * Copyright (c) 2016 Yandex LLC * Copyright (c) 2016 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include static MALLOC_DEFINE(M_IPSEC, "ipsec", "IPsec Virtual Tunnel Interface"); static const char ipsecname[] = "ipsec"; #if defined(INET) && defined(INET6) #define IPSEC_SPCOUNT 4 #else #define IPSEC_SPCOUNT 2 #endif struct ipsec_softc { struct ifnet *ifp; struct rmlock lock; struct secpolicy *sp[IPSEC_SPCOUNT]; uint32_t reqid; u_int family; u_int fibnum; LIST_ENTRY(ipsec_softc) chain; LIST_ENTRY(ipsec_softc) hash; }; #define IPSEC_LOCK_INIT(sc) rm_init(&(sc)->lock, "if_ipsec softc") #define IPSEC_LOCK_DESTROY(sc) rm_destroy(&(sc)->lock) #define IPSEC_RLOCK_TRACKER struct rm_priotracker ipsec_tracker #define IPSEC_RLOCK(sc) rm_rlock(&(sc)->lock, &ipsec_tracker) #define IPSEC_RUNLOCK(sc) rm_runlock(&(sc)->lock, &ipsec_tracker) #define IPSEC_RLOCK_ASSERT(sc) rm_assert(&(sc)->lock, RA_RLOCKED) #define IPSEC_WLOCK(sc) rm_wlock(&(sc)->lock) #define IPSEC_WUNLOCK(sc) rm_wunlock(&(sc)->lock) #define IPSEC_WLOCK_ASSERT(sc) rm_assert(&(sc)->lock, RA_WLOCKED) static struct rmlock ipsec_sc_lock; RM_SYSINIT(ipsec_sc_lock, &ipsec_sc_lock, "if_ipsec softc list"); #define IPSEC_SC_RLOCK_TRACKER struct rm_priotracker ipsec_sc_tracker #define IPSEC_SC_RLOCK() rm_rlock(&ipsec_sc_lock, &ipsec_sc_tracker) #define IPSEC_SC_RUNLOCK() rm_runlock(&ipsec_sc_lock, &ipsec_sc_tracker) #define IPSEC_SC_RLOCK_ASSERT() rm_assert(&ipsec_sc_lock, RA_RLOCKED) #define IPSEC_SC_WLOCK() rm_wlock(&ipsec_sc_lock) #define IPSEC_SC_WUNLOCK() rm_wunlock(&ipsec_sc_lock) #define IPSEC_SC_WLOCK_ASSERT() rm_assert(&ipsec_sc_lock, RA_WLOCKED) LIST_HEAD(ipsec_iflist, ipsec_softc); -static VNET_DEFINE(struct ipsec_iflist, ipsec_sc_list); -static VNET_DEFINE(struct ipsec_iflist *, ipsec_sc_htbl); -static VNET_DEFINE(u_long, ipsec_sc_hmask); +VNET_DEFINE_STATIC(struct ipsec_iflist, ipsec_sc_list); +VNET_DEFINE_STATIC(struct ipsec_iflist *, ipsec_sc_htbl); +VNET_DEFINE_STATIC(u_long, ipsec_sc_hmask); #define V_ipsec_sc_list VNET(ipsec_sc_list) #define V_ipsec_sc_htbl VNET(ipsec_sc_htbl) #define V_ipsec_sc_hmask VNET(ipsec_sc_hmask) static uint32_t ipsec_hash(uint32_t id) { return (fnv_32_buf(&id, sizeof(id), FNV1_32_INIT)); } #define SCHASH_NHASH_LOG2 5 #define SCHASH_NHASH (1 << SCHASH_NHASH_LOG2) #define SCHASH_HASHVAL(id) (ipsec_hash((id)) & V_ipsec_sc_hmask) #define SCHASH_HASH(id) &V_ipsec_sc_htbl[SCHASH_HASHVAL(id)] /* * ipsec_ioctl_sx protects from concurrent ioctls. */ static struct sx ipsec_ioctl_sx; SX_SYSINIT(ipsec_ioctl_sx, &ipsec_ioctl_sx, "ipsec_ioctl"); static int ipsec_init_reqid(struct ipsec_softc *); static int ipsec_set_tunnel(struct ipsec_softc *, struct sockaddr *, struct sockaddr *, uint32_t); static void ipsec_delete_tunnel(struct ifnet *, int); static int ipsec_set_addresses(struct ifnet *, struct sockaddr *, struct sockaddr *); static int ipsec_set_reqid(struct ifnet *, uint32_t); static int ipsec_ioctl(struct ifnet *, u_long, caddr_t); static int ipsec_transmit(struct ifnet *, struct mbuf *); static int ipsec_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static void ipsec_qflush(struct ifnet *); static int ipsec_clone_create(struct if_clone *, int, caddr_t); static void ipsec_clone_destroy(struct ifnet *); -static VNET_DEFINE(struct if_clone *, ipsec_cloner); +VNET_DEFINE_STATIC(struct if_clone *, ipsec_cloner); #define V_ipsec_cloner VNET(ipsec_cloner) static int ipsec_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ipsec_softc *sc; struct ifnet *ifp; sc = malloc(sizeof(*sc), M_IPSEC, M_WAITOK | M_ZERO); sc->fibnum = curthread->td_proc->p_fibnum; sc->ifp = ifp = if_alloc(IFT_TUNNEL); IPSEC_LOCK_INIT(sc); ifp->if_softc = sc; if_initname(ifp, ipsecname, unit); ifp->if_addrlen = 0; ifp->if_mtu = IPSEC_MTU; ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST; ifp->if_ioctl = ipsec_ioctl; ifp->if_transmit = ipsec_transmit; ifp->if_qflush = ipsec_qflush; ifp->if_output = ipsec_output; if_attach(ifp); bpfattach(ifp, DLT_NULL, sizeof(uint32_t)); IPSEC_SC_WLOCK(); LIST_INSERT_HEAD(&V_ipsec_sc_list, sc, chain); IPSEC_SC_WUNLOCK(); return (0); } static void ipsec_clone_destroy(struct ifnet *ifp) { struct ipsec_softc *sc; sx_xlock(&ipsec_ioctl_sx); sc = ifp->if_softc; IPSEC_SC_WLOCK(); ipsec_delete_tunnel(ifp, 1); LIST_REMOVE(sc, chain); IPSEC_SC_WUNLOCK(); bpfdetach(ifp); if_detach(ifp); ifp->if_softc = NULL; sx_xunlock(&ipsec_ioctl_sx); if_free(ifp); IPSEC_LOCK_DESTROY(sc); free(sc, M_IPSEC); } static void vnet_ipsec_init(const void *unused __unused) { LIST_INIT(&V_ipsec_sc_list); V_ipsec_sc_htbl = hashinit(SCHASH_NHASH, M_IPSEC, &V_ipsec_sc_hmask); V_ipsec_cloner = if_clone_simple(ipsecname, ipsec_clone_create, ipsec_clone_destroy, 0); } VNET_SYSINIT(vnet_ipsec_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_ipsec_init, NULL); static void vnet_ipsec_uninit(const void *unused __unused) { if_clone_detach(V_ipsec_cloner); hashdestroy(V_ipsec_sc_htbl, M_IPSEC, V_ipsec_sc_hmask); } VNET_SYSUNINIT(vnet_ipsec_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_ipsec_uninit, NULL); static struct secpolicy * ipsec_getpolicy(struct ipsec_softc *sc, int dir, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: return (sc->sp[(dir == IPSEC_DIR_INBOUND ? 0: 1)]); #endif #ifdef INET6 case AF_INET6: return (sc->sp[(dir == IPSEC_DIR_INBOUND ? 0: 1) #ifdef INET + 2 #endif ]); #endif } return (NULL); } static struct secasindex * ipsec_getsaidx(struct ipsec_softc *sc, int dir, sa_family_t af) { struct secpolicy *sp; sp = ipsec_getpolicy(sc, dir, af); if (sp == NULL) return (NULL); return (&sp->req[0]->saidx); } static int ipsec_transmit(struct ifnet *ifp, struct mbuf *m) { IPSEC_RLOCK_TRACKER; struct ipsec_softc *sc; struct secpolicy *sp; struct ip *ip; uint32_t af; int error; #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error) { m_freem(m); goto err; } #endif error = ENETDOWN; sc = ifp->if_softc; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (ifp->if_flags & IFF_MONITOR) != 0 || (ifp->if_flags & IFF_UP) == 0) { m_freem(m); goto err; } /* Determine address family to correctly handle packet in BPF */ ip = mtod(m, struct ip *); switch (ip->ip_v) { #ifdef INET case IPVERSION: af = AF_INET; break; #endif #ifdef INET6 case (IPV6_VERSION >> 4): af = AF_INET6; break; #endif default: error = EAFNOSUPPORT; m_freem(m); goto err; } /* * Loop prevention. * XXX: for now just check presence of IPSEC_OUT_DONE mbuf tag. * We can read full chain and compare destination address, * proto and mode from xform_history with values from softc. */ if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL) { m_freem(m); goto err; } IPSEC_RLOCK(sc); if (sc->family == 0) { IPSEC_RUNLOCK(sc); m_freem(m); goto err; } sp = ipsec_getpolicy(sc, IPSEC_DIR_OUTBOUND, af); key_addref(sp); M_SETFIB(m, sc->fibnum); IPSEC_RUNLOCK(sc); BPF_MTAP2(ifp, &af, sizeof(af), m); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); switch (af) { #ifdef INET case AF_INET: error = ipsec4_process_packet(m, sp, NULL); break; #endif #ifdef INET6 case AF_INET6: error = ipsec6_process_packet(m, sp, NULL); break; #endif default: panic("%s: unknown address family\n", __func__); } err: if (error != 0) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (error); } static void ipsec_qflush(struct ifnet *ifp __unused) { } static int ipsec_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { return (ifp->if_transmit(ifp, m)); } int ipsec_if_input(struct mbuf *m, struct secasvar *sav, uint32_t af) { IPSEC_SC_RLOCK_TRACKER; struct secasindex *saidx; struct ipsec_softc *sc; struct ifnet *ifp; if (sav->state != SADB_SASTATE_MATURE && sav->state != SADB_SASTATE_DYING) { m_freem(m); return (ENETDOWN); } if (sav->sah->saidx.mode != IPSEC_MODE_TUNNEL || sav->sah->saidx.proto != IPPROTO_ESP) return (0); IPSEC_SC_RLOCK(); /* * We only acquire SC_RLOCK() while we are doing search in * ipsec_sc_htbl. It is safe, because removing softc or changing * of reqid/addresses requires removing from hash table. */ LIST_FOREACH(sc, SCHASH_HASH(sav->sah->saidx.reqid), hash) { saidx = ipsec_getsaidx(sc, IPSEC_DIR_INBOUND, sav->sah->saidx.src.sa.sa_family); /* SA's reqid should match reqid in SP */ if (saidx == NULL || sav->sah->saidx.reqid != saidx->reqid) continue; /* SAH's addresses should match tunnel endpoints. */ if (key_sockaddrcmp(&sav->sah->saidx.dst.sa, &saidx->dst.sa, 0) != 0) continue; if (key_sockaddrcmp(&sav->sah->saidx.src.sa, &saidx->src.sa, 0) == 0) break; } if (sc == NULL) { IPSEC_SC_RUNLOCK(); /* Tunnel was not found. Nothing to do. */ return (0); } ifp = sc->ifp; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (ifp->if_flags & IFF_UP) == 0) { IPSEC_SC_RUNLOCK(); m_freem(m); return (ENETDOWN); } /* * We found matching and working tunnel. * Set its ifnet as receiving interface. */ m->m_pkthdr.rcvif = ifp; IPSEC_SC_RUNLOCK(); m_clrprotoflags(m); M_SETFIB(m, ifp->if_fib); BPF_MTAP2(ifp, &af, sizeof(af), m); if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); if ((ifp->if_flags & IFF_MONITOR) != 0) { m_freem(m); return (ENETDOWN); } return (0); } /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */ int ipsec_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { IPSEC_RLOCK_TRACKER; struct ifreq *ifr = (struct ifreq*)data; struct sockaddr *dst, *src; struct ipsec_softc *sc; struct secasindex *saidx; #ifdef INET struct sockaddr_in *sin = NULL; #endif #ifdef INET6 struct sockaddr_in6 *sin6 = NULL; #endif uint32_t reqid; int error; switch (cmd) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; case SIOCADDMULTI: case SIOCDELMULTI: case SIOCGIFMTU: case SIOCSIFFLAGS: return (0); case SIOCSIFMTU: if (ifr->ifr_mtu < IPSEC_MTU_MIN || ifr->ifr_mtu > IPSEC_MTU_MAX) return (EINVAL); else ifp->if_mtu = ifr->ifr_mtu; return (0); } sx_xlock(&ipsec_ioctl_sx); sc = ifp->if_softc; /* Check that softc is still here */ if (sc == NULL) { error = ENXIO; goto bad; } error = 0; switch (cmd) { case SIOCSIFPHYADDR: #ifdef INET6 case SIOCSIFPHYADDR_IN6: #endif error = EINVAL; switch (cmd) { #ifdef INET case SIOCSIFPHYADDR: src = (struct sockaddr *) &(((struct in_aliasreq *)data)->ifra_addr); dst = (struct sockaddr *) &(((struct in_aliasreq *)data)->ifra_dstaddr); break; #endif #ifdef INET6 case SIOCSIFPHYADDR_IN6: src = (struct sockaddr *) &(((struct in6_aliasreq *)data)->ifra_addr); dst = (struct sockaddr *) &(((struct in6_aliasreq *)data)->ifra_dstaddr); break; #endif default: goto bad; } /* sa_family must be equal */ if (src->sa_family != dst->sa_family || src->sa_len != dst->sa_len) goto bad; /* validate sa_len */ switch (src->sa_family) { #ifdef INET case AF_INET: if (src->sa_len != sizeof(struct sockaddr_in)) goto bad; break; #endif #ifdef INET6 case AF_INET6: if (src->sa_len != sizeof(struct sockaddr_in6)) goto bad; break; #endif default: error = EAFNOSUPPORT; goto bad; } /* check sa_family looks sane for the cmd */ error = EAFNOSUPPORT; switch (cmd) { #ifdef INET case SIOCSIFPHYADDR: if (src->sa_family == AF_INET) break; goto bad; #endif #ifdef INET6 case SIOCSIFPHYADDR_IN6: if (src->sa_family == AF_INET6) break; goto bad; #endif } error = EADDRNOTAVAIL; switch (src->sa_family) { #ifdef INET case AF_INET: if (satosin(src)->sin_addr.s_addr == INADDR_ANY || satosin(dst)->sin_addr.s_addr == INADDR_ANY) goto bad; break; #endif #ifdef INET6 case AF_INET6: if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(src)->sin6_addr) || IN6_IS_ADDR_UNSPECIFIED(&satosin6(dst)->sin6_addr)) goto bad; /* * Check validity of the scope zone ID of the * addresses, and convert it into the kernel * internal form if necessary. */ error = sa6_embedscope(satosin6(src), 0); if (error != 0) goto bad; error = sa6_embedscope(satosin6(dst), 0); if (error != 0) goto bad; #endif }; error = ipsec_set_addresses(ifp, src, dst); break; case SIOCDIFPHYADDR: ipsec_delete_tunnel(ifp, 0); break; case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: #ifdef INET6 case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: #endif IPSEC_RLOCK(sc); if (sc->family == 0) { IPSEC_RUNLOCK(sc); error = EADDRNOTAVAIL; break; } saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sc->family); switch (cmd) { #ifdef INET case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: if (saidx->src.sa.sa_family != AF_INET) { error = EADDRNOTAVAIL; break; } sin = (struct sockaddr_in *)&ifr->ifr_addr; memset(sin, 0, sizeof(*sin)); sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); break; #endif #ifdef INET6 case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: if (saidx->src.sa.sa_family != AF_INET6) { error = EADDRNOTAVAIL; break; } sin6 = (struct sockaddr_in6 *) &(((struct in6_ifreq *)data)->ifr_addr); memset(sin6, 0, sizeof(*sin6)); sin6->sin6_family = AF_INET6; sin6->sin6_len = sizeof(*sin6); break; #endif default: error = EAFNOSUPPORT; } if (error == 0) { switch (cmd) { #ifdef INET case SIOCGIFPSRCADDR: sin->sin_addr = saidx->src.sin.sin_addr; break; case SIOCGIFPDSTADDR: sin->sin_addr = saidx->dst.sin.sin_addr; break; #endif #ifdef INET6 case SIOCGIFPSRCADDR_IN6: sin6->sin6_addr = saidx->src.sin6.sin6_addr; break; case SIOCGIFPDSTADDR_IN6: sin6->sin6_addr = saidx->dst.sin6.sin6_addr; break; #endif } } IPSEC_RUNLOCK(sc); if (error != 0) break; switch (cmd) { #ifdef INET case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: error = prison_if(curthread->td_ucred, (struct sockaddr *)sin); if (error != 0) memset(sin, 0, sizeof(*sin)); break; #endif #ifdef INET6 case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: error = prison_if(curthread->td_ucred, (struct sockaddr *)sin6); if (error == 0) error = sa6_recoverscope(sin6); if (error != 0) memset(sin6, 0, sizeof(*sin6)); #endif } break; case SIOCGTUNFIB: ifr->ifr_fib = sc->fibnum; break; case SIOCSTUNFIB: if ((error = priv_check(curthread, PRIV_NET_SETIFFIB)) != 0) break; if (ifr->ifr_fib >= rt_numfibs) error = EINVAL; else sc->fibnum = ifr->ifr_fib; break; case IPSECGREQID: reqid = sc->reqid; error = copyout(&reqid, ifr_data_get_ptr(ifr), sizeof(reqid)); break; case IPSECSREQID: if ((error = priv_check(curthread, PRIV_NET_SETIFCAP)) != 0) break; error = copyin(ifr_data_get_ptr(ifr), &reqid, sizeof(reqid)); if (error != 0) break; error = ipsec_set_reqid(ifp, reqid); break; default: error = EINVAL; break; } bad: sx_xunlock(&ipsec_ioctl_sx); return (error); } /* * Allocate new private security policies for tunneling interface. * Each tunneling interface has following security policies for * both AF: * 0.0.0.0/0[any] 0.0.0.0/0[any] -P in \ * ipsec esp/tunnel/RemoteIP-LocalIP/unique:reqid * 0.0.0.0/0[any] 0.0.0.0/0[any] -P out \ * ipsec esp/tunnel/LocalIP-RemoteIP/unique:reqid */ static int ipsec_newpolicies(struct ipsec_softc *sc, struct secpolicy *sp[IPSEC_SPCOUNT], const struct sockaddr *src, const struct sockaddr *dst, uint32_t reqid) { struct ipsecrequest *isr; int i; memset(sp, 0, sizeof(struct secpolicy *) * IPSEC_SPCOUNT); for (i = 0; i < IPSEC_SPCOUNT; i++) { if ((sp[i] = key_newsp()) == NULL) goto fail; if ((isr = ipsec_newisr()) == NULL) goto fail; sp[i]->policy = IPSEC_POLICY_IPSEC; sp[i]->state = IPSEC_SPSTATE_DEAD; sp[i]->req[sp[i]->tcount++] = isr; sp[i]->created = time_second; /* Use priority field to store if_index */ sp[i]->priority = sc->ifp->if_index; isr->level = IPSEC_LEVEL_UNIQUE; isr->saidx.proto = IPPROTO_ESP; isr->saidx.mode = IPSEC_MODE_TUNNEL; isr->saidx.reqid = reqid; if (i % 2 == 0) { sp[i]->spidx.dir = IPSEC_DIR_INBOUND; bcopy(src, &isr->saidx.dst, src->sa_len); bcopy(dst, &isr->saidx.src, dst->sa_len); } else { sp[i]->spidx.dir = IPSEC_DIR_OUTBOUND; bcopy(src, &isr->saidx.src, src->sa_len); bcopy(dst, &isr->saidx.dst, dst->sa_len); } sp[i]->spidx.ul_proto = IPSEC_ULPROTO_ANY; #ifdef INET if (i < 2) { sp[i]->spidx.src.sa.sa_family = sp[i]->spidx.dst.sa.sa_family = AF_INET; sp[i]->spidx.src.sa.sa_len = sp[i]->spidx.dst.sa.sa_len = sizeof(struct sockaddr_in); continue; } #endif #ifdef INET6 sp[i]->spidx.src.sa.sa_family = sp[i]->spidx.dst.sa.sa_family = AF_INET6; sp[i]->spidx.src.sa.sa_len = sp[i]->spidx.dst.sa.sa_len = sizeof(struct sockaddr_in6); #endif } return (0); fail: for (i = 0; i < IPSEC_SPCOUNT; i++) key_freesp(&sp[i]); return (ENOMEM); } static int ipsec_check_reqid(uint32_t reqid) { struct ipsec_softc *sc; IPSEC_SC_RLOCK_ASSERT(); LIST_FOREACH(sc, &V_ipsec_sc_list, chain) { if (sc->reqid == reqid) return (EEXIST); } return (0); } /* * We use key_newreqid() to automatically obtain unique reqid. * Then we check that given id is unique, i.e. it is not used by * another if_ipsec(4) interface. This macro limits the number of * tries to get unique id. */ #define IPSEC_REQID_TRYCNT 64 static int ipsec_init_reqid(struct ipsec_softc *sc) { uint32_t reqid; int trycount; IPSEC_SC_RLOCK_ASSERT(); if (sc->reqid != 0) /* already initialized */ return (0); trycount = IPSEC_REQID_TRYCNT; while (--trycount > 0) { reqid = key_newreqid(); if (ipsec_check_reqid(reqid) == 0) break; } if (trycount == 0) return (EEXIST); sc->reqid = reqid; return (0); } /* * Set or update reqid for given tunneling interface. * When specified reqid is zero, generate new one. * We are protected by ioctl_sx lock from concurrent id generation. * Also softc would not disappear while we hold ioctl_sx lock. */ static int ipsec_set_reqid(struct ifnet *ifp, uint32_t reqid) { IPSEC_SC_RLOCK_TRACKER; struct ipsec_softc *sc; struct secasindex *saidx; sx_assert(&ipsec_ioctl_sx, SA_XLOCKED); sc = ifp->if_softc; if (sc->reqid == reqid && reqid != 0) return (0); IPSEC_SC_RLOCK(); if (reqid != 0) { /* Check that specified reqid doesn't exist */ if (ipsec_check_reqid(reqid) != 0) { IPSEC_SC_RUNLOCK(); return (EEXIST); } sc->reqid = reqid; } else { /* Generate new reqid */ if (ipsec_init_reqid(sc) != 0) { IPSEC_SC_RUNLOCK(); return (EEXIST); } } IPSEC_SC_RUNLOCK(); /* Tunnel isn't fully configured, just return. */ if (sc->family == 0) return (0); saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sc->family); KASSERT(saidx != NULL, ("saidx is NULL, but family is %d", sc->family)); return (ipsec_set_tunnel(sc, &saidx->src.sa, &saidx->dst.sa, sc->reqid)); } /* * Set tunnel endpoints addresses. */ static int ipsec_set_addresses(struct ifnet *ifp, struct sockaddr *src, struct sockaddr *dst) { IPSEC_SC_RLOCK_TRACKER; struct ipsec_softc *sc, *tsc; struct secasindex *saidx; sx_assert(&ipsec_ioctl_sx, SA_XLOCKED); sc = ifp->if_softc; if (sc->family != 0) { saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, src->sa_family); if (saidx != NULL && saidx->reqid == sc->reqid && key_sockaddrcmp(&saidx->src.sa, src, 0) == 0 && key_sockaddrcmp(&saidx->dst.sa, dst, 0) == 0) return (0); /* Nothing has been changed. */ } /* * We cannot service IPsec tunnel when source address is * not our own. */ #ifdef INET if (src->sa_family == AF_INET && in_localip(satosin(src)->sin_addr) == 0) return (EADDRNOTAVAIL); #endif #ifdef INET6 /* * NOTE: IPv6 addresses are in kernel internal form with * embedded scope zone id. */ if (src->sa_family == AF_INET6 && in6_localip(&satosin6(src)->sin6_addr) == 0) return (EADDRNOTAVAIL); #endif /* Check that given addresses aren't already configured */ IPSEC_SC_RLOCK(); LIST_FOREACH(tsc, &V_ipsec_sc_list, chain) { if (tsc == sc || tsc->family != src->sa_family) continue; saidx = ipsec_getsaidx(tsc, IPSEC_DIR_OUTBOUND, tsc->family); if (key_sockaddrcmp(&saidx->src.sa, src, 0) == 0 && key_sockaddrcmp(&saidx->dst.sa, dst, 0) == 0) { /* We already have tunnel with such addresses */ IPSEC_SC_RUNLOCK(); return (EADDRNOTAVAIL); } } /* If reqid is not set, generate new one. */ if (ipsec_init_reqid(sc) != 0) { IPSEC_SC_RUNLOCK(); return (EEXIST); } IPSEC_SC_RUNLOCK(); return (ipsec_set_tunnel(sc, src, dst, sc->reqid)); } static int ipsec_set_tunnel(struct ipsec_softc *sc, struct sockaddr *src, struct sockaddr *dst, uint32_t reqid) { struct secpolicy *sp[IPSEC_SPCOUNT]; struct secpolicy *oldsp[IPSEC_SPCOUNT]; int i, f; sx_assert(&ipsec_ioctl_sx, SA_XLOCKED); /* Allocate SP with new addresses. */ if (ipsec_newpolicies(sc, sp, src, dst, reqid) == 0) { /* Add new policies to SPDB */ if (key_register_ifnet(sp, IPSEC_SPCOUNT) != 0) { for (i = 0; i < IPSEC_SPCOUNT; i++) key_freesp(&sp[i]); return (EAGAIN); } IPSEC_SC_WLOCK(); if ((f = sc->family) != 0) LIST_REMOVE(sc, hash); IPSEC_WLOCK(sc); for (i = 0; i < IPSEC_SPCOUNT; i++) { oldsp[i] = sc->sp[i]; sc->sp[i] = sp[i]; } sc->family = src->sa_family; IPSEC_WUNLOCK(sc); LIST_INSERT_HEAD(SCHASH_HASH(sc->reqid), sc, hash); IPSEC_SC_WUNLOCK(); } else { sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING; return (ENOMEM); } sc->ifp->if_drv_flags |= IFF_DRV_RUNNING; if (f != 0) { key_unregister_ifnet(oldsp, IPSEC_SPCOUNT); for (i = 0; i < IPSEC_SPCOUNT; i++) key_freesp(&oldsp[i]); } return (0); } static void ipsec_delete_tunnel(struct ifnet *ifp, int locked) { struct ipsec_softc *sc = ifp->if_softc; struct secpolicy *oldsp[IPSEC_SPCOUNT]; int i; sx_assert(&ipsec_ioctl_sx, SA_XLOCKED); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; if (sc->family != 0) { if (!locked) IPSEC_SC_WLOCK(); /* Remove from hash table */ LIST_REMOVE(sc, hash); IPSEC_WLOCK(sc); for (i = 0; i < IPSEC_SPCOUNT; i++) { oldsp[i] = sc->sp[i]; sc->sp[i] = NULL; } sc->family = 0; IPSEC_WUNLOCK(sc); if (!locked) IPSEC_SC_WUNLOCK(); key_unregister_ifnet(oldsp, IPSEC_SPCOUNT); for (i = 0; i < IPSEC_SPCOUNT; i++) key_freesp(&oldsp[i]); } } Index: head/sys/net/if_lagg.c =================================================================== --- head/sys/net/if_lagg.c (revision 336675) +++ head/sys/net/if_lagg.c (revision 336676) @@ -1,2193 +1,2193 @@ /* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ /* * Copyright (c) 2005, 2006 Reyk Floeter * Copyright (c) 2007 Andrew Thompson * Copyright (c) 2014, 2016 Marcelo Araujo * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #endif #ifdef INET #include #include #endif #ifdef INET6 #include #include #include #endif #include #include #include #define LAGG_RLOCK() struct epoch_tracker lagg_et; epoch_enter_preempt(net_epoch_preempt, &lagg_et) #define LAGG_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &lagg_et) #define LAGG_RLOCK_ASSERT() MPASS(in_epoch(net_epoch_preempt)) #define LAGG_UNLOCK_ASSERT() MPASS(!in_epoch(net_epoch_preempt)) #define LAGG_SX_INIT(_sc) sx_init(&(_sc)->sc_sx, "if_lagg sx") #define LAGG_SX_DESTROY(_sc) sx_destroy(&(_sc)->sc_sx) #define LAGG_XLOCK(_sc) sx_xlock(&(_sc)->sc_sx) #define LAGG_XUNLOCK(_sc) sx_xunlock(&(_sc)->sc_sx) #define LAGG_SXLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_LOCKED) #define LAGG_XLOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_sx, SA_XLOCKED) /* Special flags we should propagate to the lagg ports. */ static struct { int flag; int (*func)(struct ifnet *, int); } lagg_pflags[] = { {IFF_PROMISC, ifpromisc}, {IFF_ALLMULTI, if_allmulti}, {0, NULL} }; VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ #define V_lagg_list VNET(lagg_list) -static VNET_DEFINE(struct mtx, lagg_list_mtx); +VNET_DEFINE_STATIC(struct mtx, lagg_list_mtx); #define V_lagg_list_mtx VNET(lagg_list_mtx) #define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ "if_lagg list", NULL, MTX_DEF) #define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) #define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) #define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) eventhandler_tag lagg_detach_cookie = NULL; static int lagg_clone_create(struct if_clone *, int, caddr_t); static void lagg_clone_destroy(struct ifnet *); -static VNET_DEFINE(struct if_clone *, lagg_cloner); +VNET_DEFINE_STATIC(struct if_clone *, lagg_cloner); #define V_lagg_cloner VNET(lagg_cloner) static const char laggname[] = "lagg"; static void lagg_capabilities(struct lagg_softc *); static int lagg_port_create(struct lagg_softc *, struct ifnet *); static int lagg_port_destroy(struct lagg_port *, int); static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); static void lagg_linkstate(struct lagg_softc *); static void lagg_port_state(struct ifnet *, int); static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); static int lagg_port_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); #ifdef LAGG_PORT_STACKING static int lagg_port_checkstacking(struct lagg_softc *); #endif static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); static void lagg_init(void *); static void lagg_stop(struct lagg_softc *); static int lagg_ioctl(struct ifnet *, u_long, caddr_t); #ifdef RATELIMIT static int lagg_snd_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *, struct m_snd_tag **); #endif static int lagg_setmulti(struct lagg_port *); static int lagg_clrmulti(struct lagg_port *); static int lagg_setcaps(struct lagg_port *, int cap); static int lagg_setflag(struct lagg_port *, int, int, int (*func)(struct ifnet *, int)); static int lagg_setflags(struct lagg_port *, int status); static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt); static int lagg_transmit(struct ifnet *, struct mbuf *); static void lagg_qflush(struct ifnet *); static int lagg_media_change(struct ifnet *); static void lagg_media_status(struct ifnet *, struct ifmediareq *); static struct lagg_port *lagg_link_active(struct lagg_softc *, struct lagg_port *); /* Simple round robin */ static void lagg_rr_attach(struct lagg_softc *); static int lagg_rr_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* Active failover */ static int lagg_fail_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* Loadbalancing */ static void lagg_lb_attach(struct lagg_softc *); static void lagg_lb_detach(struct lagg_softc *); static int lagg_lb_port_create(struct lagg_port *); static void lagg_lb_port_destroy(struct lagg_port *); static int lagg_lb_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); /* Broadcast */ static int lagg_bcast_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_bcast_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); /* 802.3ad LACP */ static void lagg_lacp_attach(struct lagg_softc *); static void lagg_lacp_detach(struct lagg_softc *); static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, struct mbuf *); static void lagg_lacp_lladdr(struct lagg_softc *); /* lagg protocol table */ static const struct lagg_proto { lagg_proto pr_num; void (*pr_attach)(struct lagg_softc *); void (*pr_detach)(struct lagg_softc *); int (*pr_start)(struct lagg_softc *, struct mbuf *); struct mbuf * (*pr_input)(struct lagg_softc *, struct lagg_port *, struct mbuf *); int (*pr_addport)(struct lagg_port *); void (*pr_delport)(struct lagg_port *); void (*pr_linkstate)(struct lagg_port *); void (*pr_init)(struct lagg_softc *); void (*pr_stop)(struct lagg_softc *); void (*pr_lladdr)(struct lagg_softc *); void (*pr_request)(struct lagg_softc *, void *); void (*pr_portreq)(struct lagg_port *, void *); } lagg_protos[] = { { .pr_num = LAGG_PROTO_NONE }, { .pr_num = LAGG_PROTO_ROUNDROBIN, .pr_attach = lagg_rr_attach, .pr_start = lagg_rr_start, .pr_input = lagg_rr_input, }, { .pr_num = LAGG_PROTO_FAILOVER, .pr_start = lagg_fail_start, .pr_input = lagg_fail_input, }, { .pr_num = LAGG_PROTO_LOADBALANCE, .pr_attach = lagg_lb_attach, .pr_detach = lagg_lb_detach, .pr_start = lagg_lb_start, .pr_input = lagg_lb_input, .pr_addport = lagg_lb_port_create, .pr_delport = lagg_lb_port_destroy, }, { .pr_num = LAGG_PROTO_LACP, .pr_attach = lagg_lacp_attach, .pr_detach = lagg_lacp_detach, .pr_start = lagg_lacp_start, .pr_input = lagg_lacp_input, .pr_addport = lacp_port_create, .pr_delport = lacp_port_destroy, .pr_linkstate = lacp_linkstate, .pr_init = lacp_init, .pr_stop = lacp_stop, .pr_lladdr = lagg_lacp_lladdr, .pr_request = lacp_req, .pr_portreq = lacp_portreq, }, { .pr_num = LAGG_PROTO_BROADCAST, .pr_start = lagg_bcast_start, .pr_input = lagg_bcast_input, }, }; SYSCTL_DECL(_net_link); SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, "Link Aggregation"); /* Allow input on any failover links */ -static VNET_DEFINE(int, lagg_failover_rx_all); +VNET_DEFINE_STATIC(int, lagg_failover_rx_all); #define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(lagg_failover_rx_all), 0, "Accept input from any interface in a failover lagg"); /* Default value for using flowid */ -static VNET_DEFINE(int, def_use_flowid) = 0; +VNET_DEFINE_STATIC(int, def_use_flowid) = 0; #define V_def_use_flowid VNET(def_use_flowid) SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, &VNET_NAME(def_use_flowid), 0, "Default setting for using flow id for load sharing"); /* Default value for flowid shift */ -static VNET_DEFINE(int, def_flowid_shift) = 16; +VNET_DEFINE_STATIC(int, def_flowid_shift) = 16; #define V_def_flowid_shift VNET(def_flowid_shift) SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, &VNET_NAME(def_flowid_shift), 0, "Default setting for flowid shift for load sharing"); static void vnet_lagg_init(const void *unused __unused) { LAGG_LIST_LOCK_INIT(); SLIST_INIT(&V_lagg_list); V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, lagg_clone_destroy, 0); } VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_lagg_init, NULL); static void vnet_lagg_uninit(const void *unused __unused) { if_clone_detach(V_lagg_cloner); LAGG_LIST_LOCK_DESTROY(); } VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_lagg_uninit, NULL); static int lagg_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: lagg_input_p = lagg_input; lagg_linkstate_p = lagg_port_state; lagg_detach_cookie = EVENTHANDLER_REGISTER( ifnet_departure_event, lagg_port_ifdetach, NULL, EVENTHANDLER_PRI_ANY); break; case MOD_UNLOAD: EVENTHANDLER_DEREGISTER(ifnet_departure_event, lagg_detach_cookie); lagg_input_p = NULL; lagg_linkstate_p = NULL; break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t lagg_mod = { "if_lagg", lagg_modevent, 0 }; DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_lagg, 1); static void lagg_proto_attach(struct lagg_softc *sc, lagg_proto pr) { LAGG_XLOCK_ASSERT(sc); KASSERT(sc->sc_proto == LAGG_PROTO_NONE, ("%s: sc %p has proto", __func__, sc)); if (sc->sc_ifflags & IFF_DEBUG) if_printf(sc->sc_ifp, "using proto %u\n", pr); if (lagg_protos[pr].pr_attach != NULL) lagg_protos[pr].pr_attach(sc); sc->sc_proto = pr; } static void lagg_proto_detach(struct lagg_softc *sc) { lagg_proto pr; LAGG_XLOCK_ASSERT(sc); pr = sc->sc_proto; sc->sc_proto = LAGG_PROTO_NONE; if (lagg_protos[pr].pr_detach != NULL) lagg_protos[pr].pr_detach(sc); } static int lagg_proto_start(struct lagg_softc *sc, struct mbuf *m) { return (lagg_protos[sc->sc_proto].pr_start(sc, m)); } static struct mbuf * lagg_proto_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { return (lagg_protos[sc->sc_proto].pr_input(sc, lp, m)); } static int lagg_proto_addport(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_addport == NULL) return (0); else return (lagg_protos[sc->sc_proto].pr_addport(lp)); } static void lagg_proto_delport(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_delport != NULL) lagg_protos[sc->sc_proto].pr_delport(lp); } static void lagg_proto_linkstate(struct lagg_softc *sc, struct lagg_port *lp) { if (lagg_protos[sc->sc_proto].pr_linkstate != NULL) lagg_protos[sc->sc_proto].pr_linkstate(lp); } static void lagg_proto_init(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_init != NULL) lagg_protos[sc->sc_proto].pr_init(sc); } static void lagg_proto_stop(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_stop != NULL) lagg_protos[sc->sc_proto].pr_stop(sc); } static void lagg_proto_lladdr(struct lagg_softc *sc) { if (lagg_protos[sc->sc_proto].pr_lladdr != NULL) lagg_protos[sc->sc_proto].pr_lladdr(sc); } static void lagg_proto_request(struct lagg_softc *sc, void *v) { if (lagg_protos[sc->sc_proto].pr_request != NULL) lagg_protos[sc->sc_proto].pr_request(sc, v); } static void lagg_proto_portreq(struct lagg_softc *sc, struct lagg_port *lp, void *v) { if (lagg_protos[sc->sc_proto].pr_portreq != NULL) lagg_protos[sc->sc_proto].pr_portreq(lp, v); } /* * This routine is run via an vlan * config EVENT */ static void lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) { struct lagg_softc *sc = ifp->if_softc; struct lagg_port *lp; if (ifp->if_softc != arg) /* Not our event */ return; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); LAGG_RUNLOCK(); } /* * This routine is run via an vlan * unconfig EVENT */ static void lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) { struct lagg_softc *sc = ifp->if_softc; struct lagg_port *lp; if (ifp->if_softc != arg) /* Not our event */ return; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); LAGG_RUNLOCK(); } static int lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct lagg_softc *sc; struct ifnet *ifp; static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); ifp = sc->sc_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { free(sc, M_DEVBUF); return (ENOSPC); } LAGG_SX_INIT(sc); LAGG_XLOCK(sc); if (V_def_use_flowid) sc->sc_opts |= LAGG_OPT_USE_FLOWID; sc->flowid_shift = V_def_flowid_shift; /* Hash all layers by default */ sc->sc_flags = MBUF_HASHFLAG_L2|MBUF_HASHFLAG_L3|MBUF_HASHFLAG_L4; lagg_proto_attach(sc, LAGG_PROTO_DEFAULT); CK_SLIST_INIT(&sc->sc_ports); /* Initialise pseudo media types */ ifmedia_init(&sc->sc_media, 0, lagg_media_change, lagg_media_status); ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); if_initname(ifp, laggname, unit); ifp->if_softc = sc; ifp->if_transmit = lagg_transmit; ifp->if_qflush = lagg_qflush; ifp->if_init = lagg_init; ifp->if_ioctl = lagg_ioctl; ifp->if_get_counter = lagg_get_counter; ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; #ifdef RATELIMIT ifp->if_snd_tag_alloc = lagg_snd_tag_alloc; ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS | IFCAP_TXRTLMT; #else ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; #endif /* * Attach as an ordinary ethernet device, children will be attached * as special device IFT_IEEE8023ADLAG. */ ether_ifattach(ifp, eaddr); sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); /* Insert into the global list of laggs */ LAGG_LIST_LOCK(); SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); LAGG_LIST_UNLOCK(); LAGG_XUNLOCK(sc); return (0); } static void lagg_clone_destroy(struct ifnet *ifp) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_port *lp; LAGG_XLOCK(sc); sc->sc_destroying = 1; lagg_stop(sc); ifp->if_flags &= ~IFF_UP; EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); /* Shutdown and remove lagg ports */ while ((lp = CK_SLIST_FIRST(&sc->sc_ports)) != NULL) lagg_port_destroy(lp, 1); /* Unhook the aggregation protocol */ lagg_proto_detach(sc); LAGG_XUNLOCK(sc); ifmedia_removeall(&sc->sc_media); ether_ifdetach(ifp); if_free(ifp); LAGG_LIST_LOCK(); SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); LAGG_LIST_UNLOCK(); LAGG_SX_DESTROY(sc); free(sc, M_DEVBUF); } static void lagg_capabilities(struct lagg_softc *sc) { struct lagg_port *lp; int cap, ena, pena; uint64_t hwa; struct ifnet_hw_tsomax hw_tsomax; LAGG_XLOCK_ASSERT(sc); /* Get common enabled capabilities for the lagg ports */ ena = ~0; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) ena &= lp->lp_ifp->if_capenable; ena = (ena == ~0 ? 0 : ena); /* * Apply common enabled capabilities back to the lagg ports. * May require several iterations if they are dependent. */ do { pena = ena; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_setcaps(lp, ena); ena &= lp->lp_ifp->if_capenable; } } while (pena != ena); /* Get other capabilities from the lagg ports */ cap = ~0; hwa = ~(uint64_t)0; memset(&hw_tsomax, 0, sizeof(hw_tsomax)); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { cap &= lp->lp_ifp->if_capabilities; hwa &= lp->lp_ifp->if_hwassist; if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); } cap = (cap == ~0 ? 0 : cap); hwa = (hwa == ~(uint64_t)0 ? 0 : hwa); if (sc->sc_ifp->if_capabilities != cap || sc->sc_ifp->if_capenable != ena || sc->sc_ifp->if_hwassist != hwa || if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { sc->sc_ifp->if_capabilities = cap; sc->sc_ifp->if_capenable = ena; sc->sc_ifp->if_hwassist = hwa; getmicrotime(&sc->sc_ifp->if_lastchange); if (sc->sc_ifflags & IFF_DEBUG) if_printf(sc->sc_ifp, "capabilities 0x%08x enabled 0x%08x\n", cap, ena); } } static int lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) { struct lagg_softc *sc_ptr; struct lagg_port *lp, *tlp; int error, i; uint64_t *pval; LAGG_XLOCK_ASSERT(sc); /* Limit the maximal number of lagg ports */ if (sc->sc_count >= LAGG_MAX_PORTS) return (ENOSPC); /* Check if port has already been associated to a lagg */ if (ifp->if_lagg != NULL) { /* Port is already in the current lagg? */ lp = (struct lagg_port *)ifp->if_lagg; if (lp->lp_softc == sc) return (EEXIST); return (EBUSY); } /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ if (ifp->if_type != IFT_ETHER && ifp->if_type != IFT_L2VLAN) return (EPROTONOSUPPORT); /* Allow the first Ethernet member to define the MTU */ if (CK_SLIST_EMPTY(&sc->sc_ports)) sc->sc_ifp->if_mtu = ifp->if_mtu; else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { if_printf(sc->sc_ifp, "invalid MTU for %s\n", ifp->if_xname); return (EINVAL); } lp = malloc(sizeof(struct lagg_port), M_DEVBUF, M_WAITOK|M_ZERO); lp->lp_softc = sc; /* Check if port is a stacked lagg */ LAGG_LIST_LOCK(); SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { if (ifp == sc_ptr->sc_ifp) { LAGG_LIST_UNLOCK(); free(lp, M_DEVBUF); return (EINVAL); /* XXX disable stacking for the moment, its untested */ #ifdef LAGG_PORT_STACKING lp->lp_flags |= LAGG_PORT_STACK; if (lagg_port_checkstacking(sc_ptr) >= LAGG_MAX_STACKING) { LAGG_LIST_UNLOCK(); free(lp, M_DEVBUF); return (E2BIG); } #endif } } LAGG_LIST_UNLOCK(); if_ref(ifp); lp->lp_ifp = ifp; bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); lp->lp_ifcapenable = ifp->if_capenable; if (CK_SLIST_EMPTY(&sc->sc_ports)) { bcopy(IF_LLADDR(ifp), IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); lagg_proto_lladdr(sc); EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } else { if_setlladdr(ifp, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); } lagg_setflags(lp, 1); if (CK_SLIST_EMPTY(&sc->sc_ports)) sc->sc_primary = lp; /* Change the interface type */ lp->lp_iftype = ifp->if_type; ifp->if_type = IFT_IEEE8023ADLAG; ifp->if_lagg = lp; lp->lp_ioctl = ifp->if_ioctl; ifp->if_ioctl = lagg_port_ioctl; lp->lp_output = ifp->if_output; ifp->if_output = lagg_port_output; /* Read port counters */ pval = lp->port_counters.val; for (i = 0; i < IFCOUNTERS; i++, pval++) *pval = ifp->if_get_counter(ifp, i); /* * Insert into the list of ports. * Keep ports sorted by if_index. It is handy, when configuration * is predictable and `ifconfig laggN create ...` command * will lead to the same result each time. */ LAGG_RLOCK(); CK_SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { if (tlp->lp_ifp->if_index < ifp->if_index && ( CK_SLIST_NEXT(tlp, lp_entries) == NULL || ((struct lagg_port*)CK_SLIST_NEXT(tlp, lp_entries))->lp_ifp->if_index > ifp->if_index)) break; } LAGG_RUNLOCK(); if (tlp != NULL) CK_SLIST_INSERT_AFTER(tlp, lp, lp_entries); else CK_SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); sc->sc_count++; lagg_setmulti(lp); if ((error = lagg_proto_addport(sc, lp)) != 0) { /* Remove the port, without calling pr_delport. */ lagg_port_destroy(lp, 0); return (error); } /* Update lagg capabilities */ lagg_capabilities(sc); lagg_linkstate(sc); return (0); } #ifdef LAGG_PORT_STACKING static int lagg_port_checkstacking(struct lagg_softc *sc) { struct lagg_softc *sc_ptr; struct lagg_port *lp; int m = 0; LAGG_SXLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_flags & LAGG_PORT_STACK) { sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; m = MAX(m, lagg_port_checkstacking(sc_ptr)); } } return (m + 1); } #endif static void lagg_port_destroy_cb(epoch_context_t ec) { struct lagg_port *lp; struct ifnet *ifp; lp = __containerof(ec, struct lagg_port, lp_epoch_ctx); ifp = lp->lp_ifp; if_rele(ifp); free(lp, M_DEVBUF); } static int lagg_port_destroy(struct lagg_port *lp, int rundelport) { struct lagg_softc *sc = lp->lp_softc; struct lagg_port *lp_ptr, *lp0; struct ifnet *ifp = lp->lp_ifp; uint64_t *pval, vdiff; int i; LAGG_XLOCK_ASSERT(sc); if (rundelport) lagg_proto_delport(sc, lp); if (lp->lp_detaching == 0) lagg_clrmulti(lp); /* Restore interface */ ifp->if_type = lp->lp_iftype; ifp->if_ioctl = lp->lp_ioctl; ifp->if_output = lp->lp_output; ifp->if_lagg = NULL; /* Update detached port counters */ pval = lp->port_counters.val; for (i = 0; i < IFCOUNTERS; i++, pval++) { vdiff = ifp->if_get_counter(ifp, i) - *pval; sc->detached_counters.val[i] += vdiff; } /* Finally, remove the port from the lagg */ CK_SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); sc->sc_count--; /* Update the primary interface */ if (lp == sc->sc_primary) { uint8_t lladdr[ETHER_ADDR_LEN]; if ((lp0 = CK_SLIST_FIRST(&sc->sc_ports)) == NULL) bzero(&lladdr, ETHER_ADDR_LEN); else bcopy(lp0->lp_lladdr, lladdr, ETHER_ADDR_LEN); sc->sc_primary = lp0; if (sc->sc_destroying == 0) { bcopy(lladdr, IF_LLADDR(sc->sc_ifp), ETHER_ADDR_LEN); lagg_proto_lladdr(sc); EVENTHANDLER_INVOKE(iflladdr_event, sc->sc_ifp); } /* * Update lladdr for each port (new primary needs update * as well, to switch from old lladdr to its 'real' one) */ CK_SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) if_setlladdr(lp_ptr->lp_ifp, lladdr, ETHER_ADDR_LEN); } if (lp->lp_ifflags) if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); if (lp->lp_detaching == 0) { lagg_setflags(lp, 0); lagg_setcaps(lp, lp->lp_ifcapenable); if_setlladdr(ifp, lp->lp_lladdr, ETHER_ADDR_LEN); } /* * free port and release it's ifnet reference after a grace period has * elapsed. */ epoch_call(net_epoch_preempt, &lp->lp_epoch_ctx, lagg_port_destroy_cb); /* Update lagg capabilities */ lagg_capabilities(sc); lagg_linkstate(sc); return (0); } static int lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct lagg_reqport *rp = (struct lagg_reqport *)data; struct lagg_softc *sc; struct lagg_port *lp = NULL; int error = 0; /* Should be checked by the caller */ if (ifp->if_type != IFT_IEEE8023ADLAG || (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) goto fallback; switch (cmd) { case SIOCGLAGGPORT: if (rp->rp_portname[0] == '\0' || ifunit(rp->rp_portname) != ifp) { error = EINVAL; break; } LAGG_RLOCK(); if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_RUNLOCK(); break; } lagg_port2req(lp, rp); LAGG_RUNLOCK(); break; case SIOCSIFCAP: if (lp->lp_ioctl == NULL) { error = EINVAL; break; } error = (*lp->lp_ioctl)(ifp, cmd, data); if (error) break; /* Update lagg interface capabilities */ LAGG_XLOCK(sc); lagg_capabilities(sc); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(sc->sc_ifp); break; case SIOCSIFMTU: /* Do not allow the MTU to be changed once joined */ error = EINVAL; break; default: goto fallback; } return (error); fallback: if (lp != NULL && lp->lp_ioctl != NULL) return ((*lp->lp_ioctl)(ifp, cmd, data)); return (EINVAL); } /* * Requests counter @cnt data. * * Counter value is calculated the following way: * 1) for each port, sum difference between current and "initial" measurements. * 2) add lagg logical interface counters. * 3) add data from detached_counters array. * * We also do the following things on ports attach/detach: * 1) On port attach we store all counters it has into port_counter array. * 2) On port detach we add the different between "initial" and * current counters data to detached_counters array. */ static uint64_t lagg_get_counter(struct ifnet *ifp, ift_counter cnt) { struct lagg_softc *sc; struct lagg_port *lp; struct ifnet *lpifp; uint64_t newval, oldval, vsum; /* Revise this when we've got non-generic counters. */ KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt)); sc = (struct lagg_softc *)ifp->if_softc; vsum = 0; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { /* Saved attached value */ oldval = lp->port_counters.val[cnt]; /* current value */ lpifp = lp->lp_ifp; newval = lpifp->if_get_counter(lpifp, cnt); /* Calculate diff and save new */ vsum += newval - oldval; } LAGG_RUNLOCK(); /* * Add counter data which might be added by upper * layer protocols operating on logical interface. */ vsum += if_get_counter_default(ifp, cnt); /* * Add counter data from detached ports counters */ vsum += sc->detached_counters.val[cnt]; return (vsum); } /* * For direct output to child ports. */ static int lagg_port_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { struct lagg_port *lp = ifp->if_lagg; switch (dst->sa_family) { case pseudo_AF_HDRCMPLT: case AF_UNSPEC: return ((*lp->lp_output)(ifp, m, dst, ro)); } /* drop any other frames */ m_freem(m); return (ENETDOWN); } static void lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) { struct lagg_port *lp; struct lagg_softc *sc; if ((lp = ifp->if_lagg) == NULL) return; /* If the ifnet is just being renamed, don't do anything. */ if (ifp->if_flags & IFF_RENAMING) return; sc = lp->lp_softc; LAGG_XLOCK(sc); lp->lp_detaching = 1; lagg_port_destroy(lp, 1); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(sc->sc_ifp); } static void lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) { struct lagg_softc *sc = lp->lp_softc; strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); rp->rp_prio = lp->lp_prio; rp->rp_flags = lp->lp_flags; lagg_proto_portreq(sc, lp, &rp->rp_psc); /* Add protocol specific flags */ switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: if (lp == sc->sc_primary) rp->rp_flags |= LAGG_PORT_MASTER; if (lp == lagg_link_active(sc, sc->sc_primary)) rp->rp_flags |= LAGG_PORT_ACTIVE; break; case LAGG_PROTO_ROUNDROBIN: case LAGG_PROTO_LOADBALANCE: case LAGG_PROTO_BROADCAST: if (LAGG_PORTACTIVE(lp)) rp->rp_flags |= LAGG_PORT_ACTIVE; break; case LAGG_PROTO_LACP: /* LACP has a different definition of active */ if (lacp_isactive(lp)) rp->rp_flags |= LAGG_PORT_ACTIVE; if (lacp_iscollecting(lp)) rp->rp_flags |= LAGG_PORT_COLLECTING; if (lacp_isdistributing(lp)) rp->rp_flags |= LAGG_PORT_DISTRIBUTING; break; } } static void lagg_init(void *xsc) { struct lagg_softc *sc = (struct lagg_softc *)xsc; struct ifnet *ifp = sc->sc_ifp; struct lagg_port *lp; LAGG_XLOCK(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { LAGG_XUNLOCK(sc); return; } ifp->if_drv_flags |= IFF_DRV_RUNNING; /* * Update the port lladdrs if needed. * This might be if_setlladdr() notification * that lladdr has been changed. */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (memcmp(IF_LLADDR(ifp), IF_LLADDR(lp->lp_ifp), ETHER_ADDR_LEN) != 0) if_setlladdr(lp->lp_ifp, IF_LLADDR(ifp), ETHER_ADDR_LEN); } lagg_proto_init(sc); LAGG_XUNLOCK(sc); } static void lagg_stop(struct lagg_softc *sc) { struct ifnet *ifp = sc->sc_ifp; LAGG_XLOCK_ASSERT(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; lagg_proto_stop(sc); } static int lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_reqall *ra = (struct lagg_reqall *)data; struct lagg_reqopts *ro = (struct lagg_reqopts *)data; struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; struct lagg_reqflags *rf = (struct lagg_reqflags *)data; struct ifreq *ifr = (struct ifreq *)data; struct lagg_port *lp; struct ifnet *tpif; struct thread *td = curthread; char *buf, *outbuf; int count, buflen, len, error = 0; bzero(&rpbuf, sizeof(rpbuf)); switch (cmd) { case SIOCGLAGG: LAGG_XLOCK(sc); buflen = sc->sc_count * sizeof(struct lagg_reqport); outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); ra->ra_proto = sc->sc_proto; lagg_proto_request(sc, &ra->ra_psc); count = 0; buf = outbuf; len = min(ra->ra_size, buflen); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (len < sizeof(rpbuf)) break; lagg_port2req(lp, &rpbuf); memcpy(buf, &rpbuf, sizeof(rpbuf)); count++; buf += sizeof(rpbuf); len -= sizeof(rpbuf); } LAGG_XUNLOCK(sc); ra->ra_ports = count; ra->ra_size = count * sizeof(rpbuf); error = copyout(outbuf, ra->ra_port, ra->ra_size); free(outbuf, M_TEMP); break; case SIOCSLAGG: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (ra->ra_proto >= LAGG_PROTO_MAX) { error = EPROTONOSUPPORT; break; } LAGG_XLOCK(sc); lagg_proto_detach(sc); LAGG_UNLOCK_ASSERT(); lagg_proto_attach(sc, ra->ra_proto); LAGG_XUNLOCK(sc); break; case SIOCGLAGGOPTS: LAGG_XLOCK(sc); ro->ro_opts = sc->sc_opts; if (sc->sc_proto == LAGG_PROTO_LACP) { struct lacp_softc *lsc; lsc = (struct lacp_softc *)sc->sc_psc; if (lsc->lsc_debug.lsc_tx_test != 0) ro->ro_opts |= LAGG_OPT_LACP_TXTEST; if (lsc->lsc_debug.lsc_rx_test != 0) ro->ro_opts |= LAGG_OPT_LACP_RXTEST; if (lsc->lsc_strict_mode != 0) ro->ro_opts |= LAGG_OPT_LACP_STRICT; if (lsc->lsc_fast_timeout != 0) ro->ro_opts |= LAGG_OPT_LACP_TIMEOUT; ro->ro_active = sc->sc_active; } else { ro->ro_active = 0; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) ro->ro_active += LAGG_PORTACTIVE(lp); } ro->ro_bkt = sc->sc_bkt; ro->ro_flapping = sc->sc_flapping; ro->ro_flowid_shift = sc->flowid_shift; LAGG_XUNLOCK(sc); break; case SIOCSLAGGOPTS: if (sc->sc_proto == LAGG_PROTO_ROUNDROBIN) { if (ro->ro_bkt == 0) sc->sc_bkt = 1; // Minimum 1 packet per iface. else sc->sc_bkt = ro->ro_bkt; } error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (ro->ro_opts == 0) break; /* * Set options. LACP options are stored in sc->sc_psc, * not in sc_opts. */ int valid, lacp; switch (ro->ro_opts) { case LAGG_OPT_USE_FLOWID: case -LAGG_OPT_USE_FLOWID: case LAGG_OPT_FLOWIDSHIFT: valid = 1; lacp = 0; break; case LAGG_OPT_LACP_TXTEST: case -LAGG_OPT_LACP_TXTEST: case LAGG_OPT_LACP_RXTEST: case -LAGG_OPT_LACP_RXTEST: case LAGG_OPT_LACP_STRICT: case -LAGG_OPT_LACP_STRICT: case LAGG_OPT_LACP_TIMEOUT: case -LAGG_OPT_LACP_TIMEOUT: valid = lacp = 1; break; default: valid = lacp = 0; break; } LAGG_XLOCK(sc); if (valid == 0 || (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { /* Invalid combination of options specified. */ error = EINVAL; LAGG_XUNLOCK(sc); break; /* Return from SIOCSLAGGOPTS. */ } /* * Store new options into sc->sc_opts except for * FLOWIDSHIFT and LACP options. */ if (lacp == 0) { if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) sc->flowid_shift = ro->ro_flowid_shift; else if (ro->ro_opts > 0) sc->sc_opts |= ro->ro_opts; else sc->sc_opts &= ~ro->ro_opts; } else { struct lacp_softc *lsc; struct lacp_port *lp; lsc = (struct lacp_softc *)sc->sc_psc; switch (ro->ro_opts) { case LAGG_OPT_LACP_TXTEST: lsc->lsc_debug.lsc_tx_test = 1; break; case -LAGG_OPT_LACP_TXTEST: lsc->lsc_debug.lsc_tx_test = 0; break; case LAGG_OPT_LACP_RXTEST: lsc->lsc_debug.lsc_rx_test = 1; break; case -LAGG_OPT_LACP_RXTEST: lsc->lsc_debug.lsc_rx_test = 0; break; case LAGG_OPT_LACP_STRICT: lsc->lsc_strict_mode = 1; break; case -LAGG_OPT_LACP_STRICT: lsc->lsc_strict_mode = 0; break; case LAGG_OPT_LACP_TIMEOUT: LACP_LOCK(lsc); LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) lp->lp_state |= LACP_STATE_TIMEOUT; LACP_UNLOCK(lsc); lsc->lsc_fast_timeout = 1; break; case -LAGG_OPT_LACP_TIMEOUT: LACP_LOCK(lsc); LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) lp->lp_state &= ~LACP_STATE_TIMEOUT; LACP_UNLOCK(lsc); lsc->lsc_fast_timeout = 0; break; } } LAGG_XUNLOCK(sc); break; case SIOCGLAGGFLAGS: rf->rf_flags = 0; LAGG_XLOCK(sc); if (sc->sc_flags & MBUF_HASHFLAG_L2) rf->rf_flags |= LAGG_F_HASHL2; if (sc->sc_flags & MBUF_HASHFLAG_L3) rf->rf_flags |= LAGG_F_HASHL3; if (sc->sc_flags & MBUF_HASHFLAG_L4) rf->rf_flags |= LAGG_F_HASHL4; LAGG_XUNLOCK(sc); break; case SIOCSLAGGHASH: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { error = EINVAL; break; } LAGG_XLOCK(sc); sc->sc_flags = 0; if (rf->rf_flags & LAGG_F_HASHL2) sc->sc_flags |= MBUF_HASHFLAG_L2; if (rf->rf_flags & LAGG_F_HASHL3) sc->sc_flags |= MBUF_HASHFLAG_L3; if (rf->rf_flags & LAGG_F_HASHL4) sc->sc_flags |= MBUF_HASHFLAG_L4; LAGG_XUNLOCK(sc); break; case SIOCGLAGGPORT: if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } LAGG_RLOCK(); if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_RUNLOCK(); if_rele(tpif); break; } lagg_port2req(lp, rp); LAGG_RUNLOCK(); if_rele(tpif); break; case SIOCSLAGGPORT: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } #ifdef INET6 /* * A laggport interface should not have inet6 address * because two interfaces with a valid link-local * scope zone must not be merged in any form. This * restriction is needed to prevent violation of * link-local scope zone. Attempts to add a laggport * interface which has inet6 addresses triggers * removal of all inet6 addresses on the member * interface. */ if (in6ifa_llaonifp(tpif)) { in6_ifdetach(tpif); if_printf(sc->sc_ifp, "IPv6 addresses on %s have been removed " "before adding it as a member to prevent " "IPv6 address scope violation.\n", tpif->if_xname); } #endif LAGG_XLOCK(sc); error = lagg_port_create(sc, tpif); LAGG_XUNLOCK(sc); if_rele(tpif); VLAN_CAPABILITIES(ifp); break; case SIOCSLAGGDELPORT: error = priv_check(td, PRIV_NET_LAGG); if (error) break; if (rp->rp_portname[0] == '\0' || (tpif = ifunit_ref(rp->rp_portname)) == NULL) { error = EINVAL; break; } LAGG_XLOCK(sc); if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || lp->lp_softc != sc) { error = ENOENT; LAGG_XUNLOCK(sc); if_rele(tpif); break; } error = lagg_port_destroy(lp, 1); LAGG_XUNLOCK(sc); if_rele(tpif); VLAN_CAPABILITIES(ifp); break; case SIOCSIFFLAGS: /* Set flags on ports too */ LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_setflags(lp, 1); } if (!(ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked down and it is running, * then stop and disable it. */ lagg_stop(sc); LAGG_XUNLOCK(sc); } else if ((ifp->if_flags & IFF_UP) && !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { /* * If interface is marked up and it is stopped, then * start it. */ LAGG_XUNLOCK(sc); (*ifp->if_init)(sc); } else LAGG_XUNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { lagg_clrmulti(lp); lagg_setmulti(lp); } LAGG_XUNLOCK(sc); error = 0; break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); break; case SIOCSIFCAP: LAGG_XLOCK(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ioctl != NULL) (*lp->lp_ioctl)(lp->lp_ifp, cmd, data); } lagg_capabilities(sc); LAGG_XUNLOCK(sc); VLAN_CAPABILITIES(ifp); error = 0; break; case SIOCSIFMTU: /* Do not allow the MTU to be directly changed */ error = EINVAL; break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } #ifdef RATELIMIT static int lagg_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_port *lp; struct lagg_lb *lb; uint32_t p; switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: lp = lagg_link_active(sc, sc->sc_primary); break; case LAGG_PROTO_LOADBALANCE: if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || params->hdr.flowtype == M_HASHTYPE_NONE) return (EOPNOTSUPP); p = params->hdr.flowid >> sc->flowid_shift; p %= sc->sc_count; lb = (struct lagg_lb *)sc->sc_psc; lp = lb->lb_ports[p]; lp = lagg_link_active(sc, lp); break; case LAGG_PROTO_LACP: if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) == 0 || params->hdr.flowtype == M_HASHTYPE_NONE) return (EOPNOTSUPP); lp = lacp_select_tx_port_by_hash(sc, params->hdr.flowid); break; default: return (EOPNOTSUPP); } if (lp == NULL) return (EOPNOTSUPP); ifp = lp->lp_ifp; if (ifp == NULL || ifp->if_snd_tag_alloc == NULL || (ifp->if_capenable & IFCAP_TXRTLMT) == 0) return (EOPNOTSUPP); /* forward allocation request */ return (ifp->if_snd_tag_alloc(ifp, params, ppmt)); } #endif static int lagg_setmulti(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; struct ifnet *ifp = lp->lp_ifp; struct ifnet *scifp = sc->sc_ifp; struct lagg_mc *mc; struct ifmultiaddr *ifma; int error; IF_ADDR_WLOCK(scifp); CK_STAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); if (mc == NULL) { IF_ADDR_WUNLOCK(scifp); return (ENOMEM); } bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len); mc->mc_addr.sdl_index = ifp->if_index; mc->mc_ifma = NULL; SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); } IF_ADDR_WUNLOCK(scifp); SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { error = if_addmulti(ifp, (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); if (error) return (error); } return (0); } static int lagg_clrmulti(struct lagg_port *lp) { struct lagg_mc *mc; LAGG_XLOCK_ASSERT(lp->lp_softc); while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); if (mc->mc_ifma && lp->lp_detaching == 0) if_delmulti_ifma(mc->mc_ifma); free(mc, M_DEVBUF); } return (0); } static int lagg_setcaps(struct lagg_port *lp, int cap) { struct ifreq ifr; if (lp->lp_ifp->if_capenable == cap) return (0); if (lp->lp_ioctl == NULL) return (ENXIO); ifr.ifr_reqcap = cap; return ((*lp->lp_ioctl)(lp->lp_ifp, SIOCSIFCAP, (caddr_t)&ifr)); } /* Handle a ref counted flag that should be set on the lagg port as well */ static int lagg_setflag(struct lagg_port *lp, int flag, int status, int (*func)(struct ifnet *, int)) { struct lagg_softc *sc = lp->lp_softc; struct ifnet *scifp = sc->sc_ifp; struct ifnet *ifp = lp->lp_ifp; int error; LAGG_XLOCK_ASSERT(sc); status = status ? (scifp->if_flags & flag) : 0; /* Now "status" contains the flag value or 0 */ /* * See if recorded ports status is different from what * we want it to be. If it is, flip it. We record ports * status in lp_ifflags so that we won't clear ports flag * we haven't set. In fact, we don't clear or set ports * flags directly, but get or release references to them. * That's why we can be sure that recorded flags still are * in accord with actual ports flags. */ if (status != (lp->lp_ifflags & flag)) { error = (*func)(ifp, status); if (error) return (error); lp->lp_ifflags &= ~flag; lp->lp_ifflags |= status; } return (0); } /* * Handle IFF_* flags that require certain changes on the lagg port * if "status" is true, update ports flags respective to the lagg * if "status" is false, forcedly clear the flags set on port. */ static int lagg_setflags(struct lagg_port *lp, int status) { int error, i; for (i = 0; lagg_pflags[i].flag; i++) { error = lagg_setflag(lp, lagg_pflags[i].flag, status, lagg_pflags[i].func); if (error) return (error); } return (0); } static int lagg_transmit(struct ifnet *ifp, struct mbuf *m) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; int error; LAGG_RLOCK(); /* We need a Tx algorithm and at least one port */ if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { LAGG_RUNLOCK(); m_freem(m); if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (ENXIO); } ETHER_BPF_MTAP(ifp, m); error = lagg_proto_start(sc, m); LAGG_RUNLOCK(); if (error != 0) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); return (error); } /* * The ifp->if_qflush entry point for lagg(4) is no-op. */ static void lagg_qflush(struct ifnet *ifp __unused) { } static struct mbuf * lagg_input(struct ifnet *ifp, struct mbuf *m) { struct lagg_port *lp = ifp->if_lagg; struct lagg_softc *sc = lp->lp_softc; struct ifnet *scifp = sc->sc_ifp; LAGG_RLOCK(); if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || lp->lp_detaching != 0 || sc->sc_proto == LAGG_PROTO_NONE) { LAGG_RUNLOCK(); m_freem(m); return (NULL); } ETHER_BPF_MTAP(scifp, m); m = lagg_proto_input(sc, lp, m); if (m != NULL && (scifp->if_flags & IFF_MONITOR) != 0) { m_freem(m); m = NULL; } LAGG_RUNLOCK(); return (m); } static int lagg_media_change(struct ifnet *ifp) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; if (sc->sc_ifflags & IFF_DEBUG) printf("%s\n", __func__); /* Ignore */ return (0); } static void lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) { struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; struct lagg_port *lp; imr->ifm_status = IFM_AVALID; imr->ifm_active = IFM_ETHER | IFM_AUTO; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (LAGG_PORTACTIVE(lp)) imr->ifm_status |= IFM_ACTIVE; } LAGG_RUNLOCK(); } static void lagg_linkstate(struct lagg_softc *sc) { struct lagg_port *lp; int new_link = LINK_STATE_DOWN; uint64_t speed; LAGG_XLOCK_ASSERT(sc); /* Our link is considered up if at least one of our ports is active */ LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (lp->lp_ifp->if_link_state == LINK_STATE_UP) { new_link = LINK_STATE_UP; break; } } LAGG_RUNLOCK(); if_link_state_change(sc->sc_ifp, new_link); /* Update if_baudrate to reflect the max possible speed */ switch (sc->sc_proto) { case LAGG_PROTO_FAILOVER: sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? sc->sc_primary->lp_ifp->if_baudrate : 0; break; case LAGG_PROTO_ROUNDROBIN: case LAGG_PROTO_LOADBALANCE: case LAGG_PROTO_BROADCAST: speed = 0; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) speed += lp->lp_ifp->if_baudrate; LAGG_RUNLOCK(); sc->sc_ifp->if_baudrate = speed; break; case LAGG_PROTO_LACP: /* LACP updates if_baudrate itself */ break; } } static void lagg_port_state(struct ifnet *ifp, int state) { struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; struct lagg_softc *sc = NULL; if (lp != NULL) sc = lp->lp_softc; if (sc == NULL) return; LAGG_XLOCK(sc); lagg_linkstate(sc); lagg_proto_linkstate(sc, lp); LAGG_XUNLOCK(sc); } struct lagg_port * lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) { struct lagg_port *lp_next, *rval = NULL; struct epoch_tracker net_et; /* * Search a port which reports an active link state. */ if (lp == NULL) goto search; if (LAGG_PORTACTIVE(lp)) { rval = lp; goto found; } if ((lp_next = CK_SLIST_NEXT(lp, lp_entries)) != NULL && LAGG_PORTACTIVE(lp_next)) { rval = lp_next; goto found; } search: epoch_enter_preempt(net_epoch_preempt, &net_et); CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { if (LAGG_PORTACTIVE(lp_next)) { epoch_exit_preempt(net_epoch_preempt, &net_et); return (lp_next); } } epoch_exit_preempt(net_epoch_preempt, &net_et); found: return (rval); } int lagg_enqueue(struct ifnet *ifp, struct mbuf *m) { return (ifp->if_transmit)(ifp, m); } /* * Simple round robin aggregation */ static void lagg_rr_attach(struct lagg_softc *sc) { sc->sc_seq = 0; sc->sc_bkt_count = sc->sc_bkt; } static int lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; uint32_t p; if (sc->sc_bkt_count == 0 && sc->sc_bkt > 0) sc->sc_bkt_count = sc->sc_bkt; if (sc->sc_bkt > 0) { atomic_subtract_int(&sc->sc_bkt_count, 1); if (atomic_cmpset_int(&sc->sc_bkt_count, 0, sc->sc_bkt)) p = atomic_fetchadd_32(&sc->sc_seq, 1); else p = sc->sc_seq; } else p = atomic_fetchadd_32(&sc->sc_seq, 1); p %= sc->sc_count; lp = CK_SLIST_FIRST(&sc->sc_ports); while (p--) lp = CK_SLIST_NEXT(lp, lp_entries); /* * Check the port's link state. This will return the next active * port if the link is down or the port is NULL. */ if ((lp = lagg_link_active(sc, lp)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * Broadcast mode */ static int lagg_bcast_start(struct lagg_softc *sc, struct mbuf *m) { int active_ports = 0; int errors = 0; int ret; struct lagg_port *lp, *last = NULL; struct mbuf *m0; LAGG_RLOCK(); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { if (!LAGG_PORTACTIVE(lp)) continue; active_ports++; if (last != NULL) { m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (m0 == NULL) { ret = ENOBUFS; errors++; break; } ret = lagg_enqueue(last->lp_ifp, m0); if (ret != 0) errors++; } last = lp; } LAGG_RUNLOCK(); if (last == NULL) { m_freem(m); return (ENOENT); } if ((last = lagg_link_active(sc, last)) == NULL) { m_freem(m); return (ENETDOWN); } ret = lagg_enqueue(last->lp_ifp, m); if (ret != 0) errors++; if (errors == 0) return (ret); return (0); } static struct mbuf* lagg_bcast_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * Active failover */ static int lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; /* Use the master port if active or the next available port */ if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; struct lagg_port *tmp_tp; if (lp == sc->sc_primary || V_lagg_failover_rx_all) { m->m_pkthdr.rcvif = ifp; return (m); } if (!LAGG_PORTACTIVE(sc->sc_primary)) { tmp_tp = lagg_link_active(sc, sc->sc_primary); /* * If tmp_tp is null, we've received a packet when all * our links are down. Weird, but process it anyways. */ if ((tmp_tp == NULL || tmp_tp == lp)) { m->m_pkthdr.rcvif = ifp; return (m); } } m_freem(m); return (NULL); } /* * Loadbalancing */ static void lagg_lb_attach(struct lagg_softc *sc) { struct lagg_port *lp; struct lagg_lb *lb; LAGG_XLOCK_ASSERT(sc); lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO); lb->lb_key = m_ether_tcpip_hash_init(); sc->sc_psc = lb; CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lagg_lb_port_create(lp); } static void lagg_lb_detach(struct lagg_softc *sc) { struct lagg_lb *lb; lb = (struct lagg_lb *)sc->sc_psc; if (lb != NULL) free(lb, M_DEVBUF); } static int lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) { struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; struct lagg_port *lp_next; int i = 0; bzero(&lb->lb_ports, sizeof(lb->lb_ports)); LAGG_RLOCK(); CK_SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { if (lp_next == lp) continue; if (i >= LAGG_MAX_PORTS) return (EINVAL); if (sc->sc_ifflags & IFF_DEBUG) printf("%s: port %s at index %d\n", sc->sc_ifname, lp_next->lp_ifp->if_xname, i); lb->lb_ports[i++] = lp_next; } LAGG_RUNLOCK(); return (0); } static int lagg_lb_port_create(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; return (lagg_lb_porttable(sc, NULL)); } static void lagg_lb_port_destroy(struct lagg_port *lp) { struct lagg_softc *sc = lp->lp_softc; lagg_lb_porttable(sc, lp); } static int lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; struct lagg_port *lp = NULL; uint32_t p = 0; if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) p = m->m_pkthdr.flowid >> sc->flowid_shift; else p = m_ether_tcpip_hash(sc->sc_flags, m, lb->lb_key); p %= sc->sc_count; lp = lb->lb_ports[p]; /* * Check the port's link state. This will return the next active * port if the link is down or the port is NULL. */ if ((lp = lagg_link_active(sc, lp)) == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; /* Just pass in the packet to our lagg device */ m->m_pkthdr.rcvif = ifp; return (m); } /* * 802.3ad LACP */ static void lagg_lacp_attach(struct lagg_softc *sc) { struct lagg_port *lp; lacp_attach(sc); LAGG_XLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_create(lp); } static void lagg_lacp_detach(struct lagg_softc *sc) { struct lagg_port *lp; void *psc; LAGG_XLOCK_ASSERT(sc); CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_destroy(lp); psc = sc->sc_psc; sc->sc_psc = NULL; lacp_detach(psc); } static void lagg_lacp_lladdr(struct lagg_softc *sc) { struct lagg_port *lp; LAGG_SXLOCK_ASSERT(sc); /* purge all the lacp ports */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_destroy(lp); /* add them back in */ CK_SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) lacp_port_create(lp); } static int lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) { struct lagg_port *lp; lp = lacp_select_tx_port(sc, m); if (lp == NULL) { m_freem(m); return (ENETDOWN); } /* Send mbuf */ return (lagg_enqueue(lp->lp_ifp, m)); } static struct mbuf * lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) { struct ifnet *ifp = sc->sc_ifp; struct ether_header *eh; u_short etype; eh = mtod(m, struct ether_header *); etype = ntohs(eh->ether_type); /* Tap off LACP control messages */ if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { m = lacp_input(lp, m); if (m == NULL) return (NULL); } /* * If the port is not collecting or not in the active aggregator then * free and return. */ if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { m_freem(m); return (NULL); } m->m_pkthdr.rcvif = ifp; return (m); } Index: head/sys/net/if_llatbl.c =================================================================== --- head/sys/net/if_llatbl.c (revision 336675) +++ head/sys/net/if_llatbl.c (revision 336676) @@ -1,966 +1,966 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2004 Luigi Rizzo, Alessandro Cerri. All rights reserved. * Copyright (c) 2004-2008 Qing Li. All rights reserved. * Copyright (c) 2008 Kip Macy. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_LLTABLE, "lltable", "link level address tables"); -static VNET_DEFINE(SLIST_HEAD(, lltable), lltables) = +VNET_DEFINE_STATIC(SLIST_HEAD(, lltable), lltables) = SLIST_HEAD_INITIALIZER(lltables); #define V_lltables VNET(lltables) static struct rwlock lltable_list_lock; RW_SYSINIT(lltable_list_lock, &lltable_list_lock, "lltable_list_lock"); #define LLTABLE_LIST_RLOCK() rw_rlock(&lltable_list_lock) #define LLTABLE_LIST_RUNLOCK() rw_runlock(&lltable_list_lock) #define LLTABLE_LIST_WLOCK() rw_wlock(&lltable_list_lock) #define LLTABLE_LIST_WUNLOCK() rw_wunlock(&lltable_list_lock) #define LLTABLE_LIST_LOCK_ASSERT() rw_assert(&lltable_list_lock, RA_LOCKED) static void lltable_unlink(struct lltable *llt); static void llentries_unlink(struct lltable *llt, struct llentries *head); static void htable_unlink_entry(struct llentry *lle); static void htable_link_entry(struct lltable *llt, struct llentry *lle); static int htable_foreach_lle(struct lltable *llt, llt_foreach_cb_t *f, void *farg); /* * Dump lle state for a specific address family. */ static int lltable_dump_af(struct lltable *llt, struct sysctl_req *wr) { int error; LLTABLE_LIST_LOCK_ASSERT(); if (llt->llt_ifp->if_flags & IFF_LOOPBACK) return (0); error = 0; IF_AFDATA_RLOCK(llt->llt_ifp); error = lltable_foreach_lle(llt, (llt_foreach_cb_t *)llt->llt_dump_entry, wr); IF_AFDATA_RUNLOCK(llt->llt_ifp); return (error); } /* * Dump arp state for a specific address family. */ int lltable_sysctl_dumparp(int af, struct sysctl_req *wr) { struct lltable *llt; int error = 0; LLTABLE_LIST_RLOCK(); SLIST_FOREACH(llt, &V_lltables, llt_link) { if (llt->llt_af == af) { error = lltable_dump_af(llt, wr); if (error != 0) goto done; } } done: LLTABLE_LIST_RUNLOCK(); return (error); } /* * Common function helpers for chained hash table. */ /* * Runs specified callback for each entry in @llt. * Caller does the locking. * */ static int htable_foreach_lle(struct lltable *llt, llt_foreach_cb_t *f, void *farg) { struct llentry *lle, *next; int i, error; error = 0; for (i = 0; i < llt->llt_hsize; i++) { CK_LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) { error = f(llt, lle, farg); if (error != 0) break; } } return (error); } static void htable_link_entry(struct lltable *llt, struct llentry *lle) { struct llentries *lleh; uint32_t hashidx; if ((lle->la_flags & LLE_LINKED) != 0) return; IF_AFDATA_WLOCK_ASSERT(llt->llt_ifp); hashidx = llt->llt_hash(lle, llt->llt_hsize); lleh = &llt->lle_head[hashidx]; lle->lle_tbl = llt; lle->lle_head = lleh; lle->la_flags |= LLE_LINKED; CK_LIST_INSERT_HEAD(lleh, lle, lle_next); } static void htable_unlink_entry(struct llentry *lle) { if ((lle->la_flags & LLE_LINKED) != 0) { IF_AFDATA_WLOCK_ASSERT(lle->lle_tbl->llt_ifp); CK_LIST_REMOVE(lle, lle_next); lle->la_flags &= ~(LLE_VALID | LLE_LINKED); #if 0 lle->lle_tbl = NULL; lle->lle_head = NULL; #endif } } struct prefix_match_data { const struct sockaddr *addr; const struct sockaddr *mask; struct llentries dchain; u_int flags; }; static int htable_prefix_free_cb(struct lltable *llt, struct llentry *lle, void *farg) { struct prefix_match_data *pmd; pmd = (struct prefix_match_data *)farg; if (llt->llt_match_prefix(pmd->addr, pmd->mask, pmd->flags, lle)) { LLE_WLOCK(lle); CK_LIST_INSERT_HEAD(&pmd->dchain, lle, lle_chain); } return (0); } static void htable_prefix_free(struct lltable *llt, const struct sockaddr *addr, const struct sockaddr *mask, u_int flags) { struct llentry *lle, *next; struct prefix_match_data pmd; bzero(&pmd, sizeof(pmd)); pmd.addr = addr; pmd.mask = mask; pmd.flags = flags; CK_LIST_INIT(&pmd.dchain); IF_AFDATA_WLOCK(llt->llt_ifp); /* Push matching lles to chain */ lltable_foreach_lle(llt, htable_prefix_free_cb, &pmd); llentries_unlink(llt, &pmd.dchain); IF_AFDATA_WUNLOCK(llt->llt_ifp); CK_LIST_FOREACH_SAFE(lle, &pmd.dchain, lle_chain, next) lltable_free_entry(llt, lle); } static void htable_free_tbl(struct lltable *llt) { free(llt->lle_head, M_LLTABLE); free(llt, M_LLTABLE); } static void llentries_unlink(struct lltable *llt, struct llentries *head) { struct llentry *lle, *next; CK_LIST_FOREACH_SAFE(lle, head, lle_chain, next) llt->llt_unlink_entry(lle); } /* * Helper function used to drop all mbufs in hold queue. * * Returns the number of held packets, if any, that were dropped. */ size_t lltable_drop_entry_queue(struct llentry *lle) { size_t pkts_dropped; struct mbuf *next; LLE_WLOCK_ASSERT(lle); pkts_dropped = 0; while ((lle->la_numheld > 0) && (lle->la_hold != NULL)) { next = lle->la_hold->m_nextpkt; m_freem(lle->la_hold); lle->la_hold = next; lle->la_numheld--; pkts_dropped++; } KASSERT(lle->la_numheld == 0, ("%s: la_numheld %d > 0, pkts_droped %zd", __func__, lle->la_numheld, pkts_dropped)); return (pkts_dropped); } void lltable_set_entry_addr(struct ifnet *ifp, struct llentry *lle, const char *linkhdr, size_t linkhdrsize, int lladdr_off) { memcpy(lle->r_linkdata, linkhdr, linkhdrsize); lle->r_hdrlen = linkhdrsize; lle->ll_addr = &lle->r_linkdata[lladdr_off]; lle->la_flags |= LLE_VALID; lle->r_flags |= RLLE_VALID; } /* * Tries to update @lle link-level address. * Since update requires AFDATA WLOCK, function * drops @lle lock, acquires AFDATA lock and then acquires * @lle lock to maintain lock order. * * Returns 1 on success. */ int lltable_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, const char *linkhdr, size_t linkhdrsize, int lladdr_off) { /* Perform real LLE update */ /* use afdata WLOCK to update fields */ LLE_WLOCK_ASSERT(lle); LLE_ADDREF(lle); LLE_WUNLOCK(lle); IF_AFDATA_WLOCK(ifp); LLE_WLOCK(lle); /* * Since we droppped LLE lock, other thread might have deleted * this lle. Check and return */ if ((lle->la_flags & LLE_DELETED) != 0) { IF_AFDATA_WUNLOCK(ifp); LLE_FREE_LOCKED(lle); return (0); } /* Update data */ lltable_set_entry_addr(ifp, lle, linkhdr, linkhdrsize, lladdr_off); IF_AFDATA_WUNLOCK(ifp); LLE_REMREF(lle); return (1); } /* * Helper function used to pre-compute full/partial link-layer * header data suitable for feeding into if_output(). */ int lltable_calc_llheader(struct ifnet *ifp, int family, char *lladdr, char *buf, size_t *bufsize, int *lladdr_off) { struct if_encap_req ereq; int error; bzero(buf, *bufsize); bzero(&ereq, sizeof(ereq)); ereq.buf = buf; ereq.bufsize = *bufsize; ereq.rtype = IFENCAP_LL; ereq.family = family; ereq.lladdr = lladdr; ereq.lladdr_len = ifp->if_addrlen; error = ifp->if_requestencap(ifp, &ereq); if (error == 0) { *bufsize = ereq.bufsize; *lladdr_off = ereq.lladdr_off; } return (error); } /* * Update link-layer header for given @lle after * interface lladdr was changed. */ static int llentry_update_ifaddr(struct lltable *llt, struct llentry *lle, void *farg) { struct ifnet *ifp; u_char linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; u_char *lladdr; int lladdr_off; ifp = (struct ifnet *)farg; lladdr = lle->ll_addr; LLE_WLOCK(lle); if ((lle->la_flags & LLE_VALID) == 0) { LLE_WUNLOCK(lle); return (0); } if ((lle->la_flags & LLE_IFADDR) != 0) lladdr = IF_LLADDR(ifp); linkhdrsize = sizeof(linkhdr); lltable_calc_llheader(ifp, llt->llt_af, lladdr, linkhdr, &linkhdrsize, &lladdr_off); memcpy(lle->r_linkdata, linkhdr, linkhdrsize); LLE_WUNLOCK(lle); return (0); } /* * Update all calculated headers for given @llt */ void lltable_update_ifaddr(struct lltable *llt) { if (llt->llt_ifp->if_flags & IFF_LOOPBACK) return; IF_AFDATA_WLOCK(llt->llt_ifp); lltable_foreach_lle(llt, llentry_update_ifaddr, llt->llt_ifp); IF_AFDATA_WUNLOCK(llt->llt_ifp); } /* * * Performs generic cleanup routines and frees lle. * * Called for non-linked entries, with callouts and * other AF-specific cleanups performed. * * @lle must be passed WLOCK'ed * * Returns the number of held packets, if any, that were dropped. */ size_t llentry_free(struct llentry *lle) { size_t pkts_dropped; LLE_WLOCK_ASSERT(lle); KASSERT((lle->la_flags & LLE_LINKED) == 0, ("freeing linked lle")); pkts_dropped = lltable_drop_entry_queue(lle); /* cancel timer */ if (callout_stop(&lle->lle_timer) > 0) LLE_REMREF(lle); LLE_FREE_LOCKED(lle); return (pkts_dropped); } /* * (al)locate an llentry for address dst (equivalent to rtalloc for new-arp). * * If found the llentry * is returned referenced and unlocked. */ struct llentry * llentry_alloc(struct ifnet *ifp, struct lltable *lt, struct sockaddr_storage *dst) { struct llentry *la, *la_tmp; IF_AFDATA_RLOCK(ifp); la = lla_lookup(lt, LLE_EXCLUSIVE, (struct sockaddr *)dst); IF_AFDATA_RUNLOCK(ifp); if (la != NULL) { LLE_ADDREF(la); LLE_WUNLOCK(la); return (la); } if ((ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) { la = lltable_alloc_entry(lt, 0, (struct sockaddr *)dst); if (la == NULL) return (NULL); IF_AFDATA_WLOCK(ifp); LLE_WLOCK(la); /* Prefer any existing LLE over newly-created one */ la_tmp = lla_lookup(lt, LLE_EXCLUSIVE, (struct sockaddr *)dst); if (la_tmp == NULL) lltable_link_entry(lt, la); IF_AFDATA_WUNLOCK(ifp); if (la_tmp != NULL) { lltable_free_entry(lt, la); la = la_tmp; } LLE_ADDREF(la); LLE_WUNLOCK(la); } return (la); } /* * Free all entries from given table and free itself. */ static int lltable_free_cb(struct lltable *llt, struct llentry *lle, void *farg) { struct llentries *dchain; dchain = (struct llentries *)farg; LLE_WLOCK(lle); CK_LIST_INSERT_HEAD(dchain, lle, lle_chain); return (0); } /* * Free all entries from given table and free itself. */ void lltable_free(struct lltable *llt) { struct llentry *lle, *next; struct llentries dchain; KASSERT(llt != NULL, ("%s: llt is NULL", __func__)); lltable_unlink(llt); CK_LIST_INIT(&dchain); IF_AFDATA_WLOCK(llt->llt_ifp); /* Push all lles to @dchain */ lltable_foreach_lle(llt, lltable_free_cb, &dchain); llentries_unlink(llt, &dchain); IF_AFDATA_WUNLOCK(llt->llt_ifp); CK_LIST_FOREACH_SAFE(lle, &dchain, lle_chain, next) { llentry_free(lle); } llt->llt_free_tbl(llt); } #if 0 void lltable_drain(int af) { struct lltable *llt; struct llentry *lle; int i; LLTABLE_LIST_RLOCK(); SLIST_FOREACH(llt, &V_lltables, llt_link) { if (llt->llt_af != af) continue; for (i=0; i < llt->llt_hsize; i++) { CK_LIST_FOREACH(lle, &llt->lle_head[i], lle_next) { LLE_WLOCK(lle); if (lle->la_hold) { m_freem(lle->la_hold); lle->la_hold = NULL; } LLE_WUNLOCK(lle); } } } LLTABLE_LIST_RUNLOCK(); } #endif /* * Deletes an address from given lltable. * Used for userland interaction to remove * individual entries. Skips entries added by OS. */ int lltable_delete_addr(struct lltable *llt, u_int flags, const struct sockaddr *l3addr) { struct llentry *lle; struct ifnet *ifp; ifp = llt->llt_ifp; IF_AFDATA_WLOCK(ifp); lle = lla_lookup(llt, LLE_EXCLUSIVE, l3addr); if (lle == NULL) { IF_AFDATA_WUNLOCK(ifp); return (ENOENT); } if ((lle->la_flags & LLE_IFADDR) != 0 && (flags & LLE_IFADDR) == 0) { IF_AFDATA_WUNLOCK(ifp); LLE_WUNLOCK(lle); return (EPERM); } lltable_unlink_entry(llt, lle); IF_AFDATA_WUNLOCK(ifp); llt->llt_delete_entry(llt, lle); return (0); } void lltable_prefix_free(int af, struct sockaddr *addr, struct sockaddr *mask, u_int flags) { struct lltable *llt; LLTABLE_LIST_RLOCK(); SLIST_FOREACH(llt, &V_lltables, llt_link) { if (llt->llt_af != af) continue; llt->llt_prefix_free(llt, addr, mask, flags); } LLTABLE_LIST_RUNLOCK(); } struct lltable * lltable_allocate_htbl(uint32_t hsize) { struct lltable *llt; int i; llt = malloc(sizeof(struct lltable), M_LLTABLE, M_WAITOK | M_ZERO); llt->llt_hsize = hsize; llt->lle_head = malloc(sizeof(struct llentries) * hsize, M_LLTABLE, M_WAITOK | M_ZERO); for (i = 0; i < llt->llt_hsize; i++) CK_LIST_INIT(&llt->lle_head[i]); /* Set some default callbacks */ llt->llt_link_entry = htable_link_entry; llt->llt_unlink_entry = htable_unlink_entry; llt->llt_prefix_free = htable_prefix_free; llt->llt_foreach_entry = htable_foreach_lle; llt->llt_free_tbl = htable_free_tbl; return (llt); } /* * Links lltable to global llt list. */ void lltable_link(struct lltable *llt) { LLTABLE_LIST_WLOCK(); SLIST_INSERT_HEAD(&V_lltables, llt, llt_link); LLTABLE_LIST_WUNLOCK(); } static void lltable_unlink(struct lltable *llt) { LLTABLE_LIST_WLOCK(); SLIST_REMOVE(&V_lltables, llt, lltable, llt_link); LLTABLE_LIST_WUNLOCK(); } /* * External methods used by lltable consumers */ int lltable_foreach_lle(struct lltable *llt, llt_foreach_cb_t *f, void *farg) { return (llt->llt_foreach_entry(llt, f, farg)); } struct llentry * lltable_alloc_entry(struct lltable *llt, u_int flags, const struct sockaddr *l3addr) { return (llt->llt_alloc_entry(llt, flags, l3addr)); } void lltable_free_entry(struct lltable *llt, struct llentry *lle) { llt->llt_free_entry(llt, lle); } void lltable_link_entry(struct lltable *llt, struct llentry *lle) { llt->llt_link_entry(llt, lle); } void lltable_unlink_entry(struct lltable *llt, struct llentry *lle) { llt->llt_unlink_entry(lle); } void lltable_fill_sa_entry(const struct llentry *lle, struct sockaddr *sa) { struct lltable *llt; llt = lle->lle_tbl; llt->llt_fill_sa_entry(lle, sa); } struct ifnet * lltable_get_ifp(const struct lltable *llt) { return (llt->llt_ifp); } int lltable_get_af(const struct lltable *llt) { return (llt->llt_af); } /* * Called in route_output when rtm_flags contains RTF_LLDATA. */ int lla_rt_output(struct rt_msghdr *rtm, struct rt_addrinfo *info) { struct sockaddr_dl *dl = (struct sockaddr_dl *)info->rti_info[RTAX_GATEWAY]; struct sockaddr *dst = (struct sockaddr *)info->rti_info[RTAX_DST]; struct ifnet *ifp; struct lltable *llt; struct llentry *lle, *lle_tmp; uint8_t linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; int lladdr_off; u_int laflags = 0; int error; KASSERT(dl != NULL && dl->sdl_family == AF_LINK, ("%s: invalid dl\n", __func__)); ifp = ifnet_byindex(dl->sdl_index); if (ifp == NULL) { log(LOG_INFO, "%s: invalid ifp (sdl_index %d)\n", __func__, dl->sdl_index); return EINVAL; } /* XXX linked list may be too expensive */ LLTABLE_LIST_RLOCK(); SLIST_FOREACH(llt, &V_lltables, llt_link) { if (llt->llt_af == dst->sa_family && llt->llt_ifp == ifp) break; } LLTABLE_LIST_RUNLOCK(); KASSERT(llt != NULL, ("Yep, ugly hacks are bad\n")); error = 0; switch (rtm->rtm_type) { case RTM_ADD: /* Add static LLE */ laflags = 0; if (rtm->rtm_rmx.rmx_expire == 0) laflags = LLE_STATIC; lle = lltable_alloc_entry(llt, laflags, dst); if (lle == NULL) return (ENOMEM); linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, dst->sa_family, LLADDR(dl), linkhdr, &linkhdrsize, &lladdr_off) != 0) return (EINVAL); lltable_set_entry_addr(ifp, lle, linkhdr, linkhdrsize, lladdr_off); if ((rtm->rtm_flags & RTF_ANNOUNCE)) lle->la_flags |= LLE_PUB; lle->la_expire = rtm->rtm_rmx.rmx_expire; laflags = lle->la_flags; /* Try to link new entry */ lle_tmp = NULL; IF_AFDATA_WLOCK(ifp); LLE_WLOCK(lle); lle_tmp = lla_lookup(llt, LLE_EXCLUSIVE, dst); if (lle_tmp != NULL) { /* Check if we are trying to replace immutable entry */ if ((lle_tmp->la_flags & LLE_IFADDR) != 0) { IF_AFDATA_WUNLOCK(ifp); LLE_WUNLOCK(lle_tmp); lltable_free_entry(llt, lle); return (EPERM); } /* Unlink existing entry from table */ lltable_unlink_entry(llt, lle_tmp); } lltable_link_entry(llt, lle); IF_AFDATA_WUNLOCK(ifp); if (lle_tmp != NULL) { EVENTHANDLER_INVOKE(lle_event, lle_tmp,LLENTRY_EXPIRED); lltable_free_entry(llt, lle_tmp); } /* * By invoking LLE handler here we might get * two events on static LLE entry insertion * in routing socket. However, since we might have * other subscribers we need to generate this event. */ EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED); LLE_WUNLOCK(lle); #ifdef INET /* gratuitous ARP */ if ((laflags & LLE_PUB) && dst->sa_family == AF_INET) arprequest(ifp, &((struct sockaddr_in *)dst)->sin_addr, &((struct sockaddr_in *)dst)->sin_addr, (u_char *)LLADDR(dl)); #endif break; case RTM_DELETE: return (lltable_delete_addr(llt, 0, dst)); default: error = EINVAL; } return (error); } #ifdef DDB struct llentry_sa { struct llentry base; struct sockaddr l3_addr; }; static void llatbl_lle_show(struct llentry_sa *la) { struct llentry *lle; uint8_t octet[6]; lle = &la->base; db_printf("lle=%p\n", lle); db_printf(" lle_next=%p\n", lle->lle_next.cle_next); db_printf(" lle_lock=%p\n", &lle->lle_lock); db_printf(" lle_tbl=%p\n", lle->lle_tbl); db_printf(" lle_head=%p\n", lle->lle_head); db_printf(" la_hold=%p\n", lle->la_hold); db_printf(" la_numheld=%d\n", lle->la_numheld); db_printf(" la_expire=%ju\n", (uintmax_t)lle->la_expire); db_printf(" la_flags=0x%04x\n", lle->la_flags); db_printf(" la_asked=%u\n", lle->la_asked); db_printf(" la_preempt=%u\n", lle->la_preempt); db_printf(" ln_state=%d\n", lle->ln_state); db_printf(" ln_router=%u\n", lle->ln_router); db_printf(" ln_ntick=%ju\n", (uintmax_t)lle->ln_ntick); db_printf(" lle_refcnt=%d\n", lle->lle_refcnt); bcopy(lle->ll_addr, octet, sizeof(octet)); db_printf(" ll_addr=%02x:%02x:%02x:%02x:%02x:%02x\n", octet[0], octet[1], octet[2], octet[3], octet[4], octet[5]); db_printf(" lle_timer=%p\n", &lle->lle_timer); switch (la->l3_addr.sa_family) { #ifdef INET case AF_INET: { struct sockaddr_in *sin; char l3s[INET_ADDRSTRLEN]; sin = (struct sockaddr_in *)&la->l3_addr; inet_ntoa_r(sin->sin_addr, l3s); db_printf(" l3_addr=%s\n", l3s); break; } #endif #ifdef INET6 case AF_INET6: { struct sockaddr_in6 *sin6; char l3s[INET6_ADDRSTRLEN]; sin6 = (struct sockaddr_in6 *)&la->l3_addr; ip6_sprintf(l3s, &sin6->sin6_addr); db_printf(" l3_addr=%s\n", l3s); break; } #endif default: db_printf(" l3_addr=N/A (af=%d)\n", la->l3_addr.sa_family); break; } } DB_SHOW_COMMAND(llentry, db_show_llentry) { if (!have_addr) { db_printf("usage: show llentry \n"); return; } llatbl_lle_show((struct llentry_sa *)addr); } static void llatbl_llt_show(struct lltable *llt) { int i; struct llentry *lle; db_printf("llt=%p llt_af=%d llt_ifp=%p\n", llt, llt->llt_af, llt->llt_ifp); for (i = 0; i < llt->llt_hsize; i++) { CK_LIST_FOREACH(lle, &llt->lle_head[i], lle_next) { llatbl_lle_show((struct llentry_sa *)lle); if (db_pager_quit) return; } } } DB_SHOW_COMMAND(lltable, db_show_lltable) { if (!have_addr) { db_printf("usage: show lltable \n"); return; } llatbl_llt_show((struct lltable *)addr); } DB_SHOW_ALL_COMMAND(lltables, db_show_all_lltables) { VNET_ITERATOR_DECL(vnet_iter); struct lltable *llt; VNET_FOREACH(vnet_iter) { CURVNET_SET_QUIET(vnet_iter); #ifdef VIMAGE db_printf("vnet=%p\n", curvnet); #endif SLIST_FOREACH(llt, &V_lltables, llt_link) { db_printf("llt=%p llt_af=%d llt_ifp=%p(%s)\n", llt, llt->llt_af, llt->llt_ifp, (llt->llt_ifp != NULL) ? llt->llt_ifp->if_xname : "?"); if (have_addr && addr != 0) /* verbose */ llatbl_llt_show(llt); if (db_pager_quit) { CURVNET_RESTORE(); return; } } CURVNET_RESTORE(); } } #endif Index: head/sys/net/if_loop.c =================================================================== --- head/sys/net/if_loop.c (revision 336675) +++ head/sys/net/if_loop.c (revision 336676) @@ -1,456 +1,456 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)if_loop.c 8.2 (Berkeley) 1/9/95 * $FreeBSD$ */ /* * Loopback interface driver for protocol testing and timing. */ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif #ifdef INET6 #ifndef INET #include #endif #include #include #endif #include #ifdef TINY_LOMTU #define LOMTU (1024+512) #elif defined(LARGE_LOMTU) #define LOMTU 131072 #else #define LOMTU 16384 #endif #define LO_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP) #define LO_CSUM_FEATURES6 (CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6) #define LO_CSUM_SET (CSUM_DATA_VALID | CSUM_DATA_VALID_IPV6 | \ CSUM_PSEUDO_HDR | \ CSUM_IP_CHECKED | CSUM_IP_VALID | \ CSUM_SCTP_VALID) int loioctl(struct ifnet *, u_long, caddr_t); int looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro); static int lo_clone_create(struct if_clone *, int, caddr_t); static void lo_clone_destroy(struct ifnet *); VNET_DEFINE(struct ifnet *, loif); /* Used externally */ #ifdef VIMAGE -static VNET_DEFINE(struct if_clone *, lo_cloner); +VNET_DEFINE_STATIC(struct if_clone *, lo_cloner); #define V_lo_cloner VNET(lo_cloner) #endif static struct if_clone *lo_cloner; static const char loname[] = "lo"; static void lo_clone_destroy(struct ifnet *ifp) { #ifndef VIMAGE /* XXX: destroying lo0 will lead to panics. */ KASSERT(V_loif != ifp, ("%s: destroying lo0", __func__)); #endif bpfdetach(ifp); if_detach(ifp); if_free(ifp); } static int lo_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ifnet *ifp; ifp = if_alloc(IFT_LOOP); if (ifp == NULL) return (ENOSPC); if_initname(ifp, loname, unit); ifp->if_mtu = LOMTU; ifp->if_flags = IFF_LOOPBACK | IFF_MULTICAST; ifp->if_ioctl = loioctl; ifp->if_output = looutput; ifp->if_snd.ifq_maxlen = ifqmaxlen; ifp->if_capabilities = ifp->if_capenable = IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 | IFCAP_LINKSTATE; ifp->if_hwassist = LO_CSUM_FEATURES | LO_CSUM_FEATURES6; if_attach(ifp); bpfattach(ifp, DLT_NULL, sizeof(u_int32_t)); if (V_loif == NULL) V_loif = ifp; return (0); } static void vnet_loif_init(const void *unused __unused) { #ifdef VIMAGE lo_cloner = if_clone_simple(loname, lo_clone_create, lo_clone_destroy, 1); V_lo_cloner = lo_cloner; #else lo_cloner = if_clone_simple(loname, lo_clone_create, lo_clone_destroy, 1); #endif } VNET_SYSINIT(vnet_loif_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_loif_init, NULL); #ifdef VIMAGE static void vnet_loif_uninit(const void *unused __unused) { if_clone_detach(V_lo_cloner); V_loif = NULL; } VNET_SYSUNINIT(vnet_loif_uninit, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_loif_uninit, NULL); #endif static int loop_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: break; case MOD_UNLOAD: printf("loop module unload - not possible for this module type\n"); return (EINVAL); default: return (EOPNOTSUPP); } return (0); } static moduledata_t loop_mod = { "if_lo", loop_modevent, 0 }; DECLARE_MODULE(if_lo, loop_mod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); int looutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { u_int32_t af; #ifdef MAC int error; #endif M_ASSERTPKTHDR(m); /* check if we have the packet header */ #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error) { m_freem(m); return (error); } #endif if (ro != NULL && ro->ro_flags & (RT_REJECT|RT_BLACKHOLE)) { m_freem(m); return (ro->ro_flags & RT_BLACKHOLE ? 0 : EHOSTUNREACH); } if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); #ifdef RSS M_HASHTYPE_CLEAR(m); #endif /* BPF writes need to be handled specially. */ if (dst->sa_family == AF_UNSPEC || dst->sa_family == pseudo_AF_HDRCMPLT) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; #if 1 /* XXX */ switch (af) { case AF_INET: if (ifp->if_capenable & IFCAP_RXCSUM) { m->m_pkthdr.csum_data = 0xffff; m->m_pkthdr.csum_flags = LO_CSUM_SET; } m->m_pkthdr.csum_flags &= ~LO_CSUM_FEATURES; break; case AF_INET6: #if 0 /* * XXX-BZ for now always claim the checksum is good despite * any interface flags. This is a workaround for 9.1-R and * a proper solution ought to be sought later. */ if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) { m->m_pkthdr.csum_data = 0xffff; m->m_pkthdr.csum_flags = LO_CSUM_SET; } #else m->m_pkthdr.csum_data = 0xffff; m->m_pkthdr.csum_flags = LO_CSUM_SET; #endif m->m_pkthdr.csum_flags &= ~LO_CSUM_FEATURES6; break; default: printf("looutput: af=%d unexpected\n", af); m_freem(m); return (EAFNOSUPPORT); } #endif return (if_simloop(ifp, m, af, 0)); } /* * if_simloop() * * This function is to support software emulation of hardware loopback, * i.e., for interfaces with the IFF_SIMPLEX attribute. Since they can't * hear their own broadcasts, we create a copy of the packet that we * would normally receive via a hardware loopback. * * This function expects the packet to include the media header of length hlen. */ int if_simloop(struct ifnet *ifp, struct mbuf *m, int af, int hlen) { int isr; M_ASSERTPKTHDR(m); m_tag_delete_nonpersistent(m); m->m_pkthdr.rcvif = ifp; #ifdef MAC mac_ifnet_create_mbuf(ifp, m); #endif /* * Let BPF see incoming packet in the following manner: * - Emulated packet loopback for a simplex interface * (net/if_ethersubr.c) * -> passes it to ifp's BPF * - IPv4/v6 multicast packet loopback (netinet(6)/ip(6)_output.c) * -> not passes it to any BPF * - Normal packet loopback from myself to myself (net/if_loop.c) * -> passes to lo0's BPF (even in case of IPv6, where ifp!=lo0) */ if (hlen > 0) { if (bpf_peers_present(ifp->if_bpf)) { bpf_mtap(ifp->if_bpf, m); } } else { if (bpf_peers_present(V_loif->if_bpf)) { if ((m->m_flags & M_MCAST) == 0 || V_loif == ifp) { /* XXX beware sizeof(af) != 4 */ u_int32_t af1 = af; /* * We need to prepend the address family. */ bpf_mtap2(V_loif->if_bpf, &af1, sizeof(af1), m); } } } /* Strip away media header */ if (hlen > 0) { m_adj(m, hlen); #ifndef __NO_STRICT_ALIGNMENT /* * Some archs do not like unaligned data, so * we move data down in the first mbuf. */ if (mtod(m, vm_offset_t) & 3) { KASSERT(hlen >= 3, ("if_simloop: hlen too small")); bcopy(m->m_data, (char *)(mtod(m, vm_offset_t) - (mtod(m, vm_offset_t) & 3)), m->m_len); m->m_data -= (mtod(m,vm_offset_t) & 3); } #endif } /* Deliver to upper layer protocol */ switch (af) { #ifdef INET case AF_INET: isr = NETISR_IP; break; #endif #ifdef INET6 case AF_INET6: m->m_flags |= M_LOOP; isr = NETISR_IPV6; break; #endif default: printf("if_simloop: can't handle af=%d\n", af); m_freem(m); return (EAFNOSUPPORT); } if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); netisr_queue(isr, m); /* mbuf is free'd on failure. */ return (0); } /* * Process an ioctl request. */ /* ARGSUSED */ int loioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; int error = 0, mask; switch (cmd) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; ifp->if_drv_flags |= IFF_DRV_RUNNING; if_link_state_change(ifp, LINK_STATE_UP); /* * Everything else is done at a higher level. */ break; case SIOCADDMULTI: case SIOCDELMULTI: if (ifr == NULL) { error = EAFNOSUPPORT; /* XXX */ break; } switch (ifr->ifr_addr.sa_family) { #ifdef INET case AF_INET: break; #endif #ifdef INET6 case AF_INET6: break; #endif default: error = EAFNOSUPPORT; break; } break; case SIOCSIFMTU: ifp->if_mtu = ifr->ifr_mtu; break; case SIOCSIFFLAGS: if_link_state_change(ifp, (ifp->if_flags & IFF_UP) ? LINK_STATE_UP: LINK_STATE_DOWN); break; case SIOCSIFCAP: mask = ifp->if_capenable ^ ifr->ifr_reqcap; if ((mask & IFCAP_RXCSUM) != 0) ifp->if_capenable ^= IFCAP_RXCSUM; if ((mask & IFCAP_TXCSUM) != 0) ifp->if_capenable ^= IFCAP_TXCSUM; if ((mask & IFCAP_RXCSUM_IPV6) != 0) { #if 0 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; #else error = EOPNOTSUPP; break; #endif } if ((mask & IFCAP_TXCSUM_IPV6) != 0) { #if 0 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; #else error = EOPNOTSUPP; break; #endif } ifp->if_hwassist = 0; if (ifp->if_capenable & IFCAP_TXCSUM) ifp->if_hwassist = LO_CSUM_FEATURES; #if 0 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) ifp->if_hwassist |= LO_CSUM_FEATURES6; #endif break; default: error = EINVAL; } return (error); } Index: head/sys/net/if_me.c =================================================================== --- head/sys/net/if_me.c (revision 336675) +++ head/sys/net/if_me.c (revision 336676) @@ -1,604 +1,604 @@ /*- * Copyright (c) 2014, 2018 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define MEMTU (1500 - sizeof(struct mobhdr)) static const char mename[] = "me"; static MALLOC_DEFINE(M_IFME, mename, "Minimal Encapsulation for IP"); /* Minimal forwarding header RFC 2004 */ struct mobhdr { uint8_t mob_proto; /* protocol */ uint8_t mob_flags; /* flags */ #define MOB_FLAGS_SP 0x80 /* source present */ uint16_t mob_csum; /* header checksum */ struct in_addr mob_dst; /* original destination address */ struct in_addr mob_src; /* original source addr (optional) */ } __packed; struct me_softc { struct ifnet *me_ifp; u_int me_fibnum; struct in_addr me_src; struct in_addr me_dst; CK_LIST_ENTRY(me_softc) chain; }; CK_LIST_HEAD(me_list, me_softc); #define ME2IFP(sc) ((sc)->me_ifp) #define ME_READY(sc) ((sc)->me_src.s_addr != 0) #define ME_RLOCK() struct epoch_tracker me_et; epoch_enter_preempt(net_epoch_preempt, &me_et) #define ME_RUNLOCK() epoch_exit_preempt(net_epoch_preempt, &me_et) #define ME_WAIT() epoch_wait_preempt(net_epoch_preempt) #ifndef ME_HASH_SIZE #define ME_HASH_SIZE (1 << 4) #endif -static VNET_DEFINE(struct me_list *, me_hashtbl) = NULL; +VNET_DEFINE_STATIC(struct me_list *, me_hashtbl) = NULL; #define V_me_hashtbl VNET(me_hashtbl) #define ME_HASH(src, dst) (V_me_hashtbl[\ me_hashval((src), (dst)) & (ME_HASH_SIZE - 1)]) static struct sx me_ioctl_sx; SX_SYSINIT(me_ioctl_sx, &me_ioctl_sx, "me_ioctl"); static int me_clone_create(struct if_clone *, int, caddr_t); static void me_clone_destroy(struct ifnet *); -static VNET_DEFINE(struct if_clone *, me_cloner); +VNET_DEFINE_STATIC(struct if_clone *, me_cloner); #define V_me_cloner VNET(me_cloner) static void me_qflush(struct ifnet *); static int me_transmit(struct ifnet *, struct mbuf *); static int me_ioctl(struct ifnet *, u_long, caddr_t); static int me_output(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static int me_input(struct mbuf *, int, int, void *); static int me_set_tunnel(struct me_softc *, in_addr_t, in_addr_t); static void me_delete_tunnel(struct me_softc *); SYSCTL_DECL(_net_link); static SYSCTL_NODE(_net_link, IFT_TUNNEL, me, CTLFLAG_RW, 0, "Minimal Encapsulation for IP (RFC 2004)"); #ifndef MAX_ME_NEST #define MAX_ME_NEST 1 #endif -static VNET_DEFINE(int, max_me_nesting) = MAX_ME_NEST; +VNET_DEFINE_STATIC(int, max_me_nesting) = MAX_ME_NEST; #define V_max_me_nesting VNET(max_me_nesting) SYSCTL_INT(_net_link_me, OID_AUTO, max_nesting, CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(max_me_nesting), 0, "Max nested tunnels"); static uint32_t me_hashval(in_addr_t src, in_addr_t dst) { uint32_t ret; ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT); return (fnv_32_buf(&dst, sizeof(dst), ret)); } static struct me_list * me_hashinit(void) { struct me_list *hash; int i; hash = malloc(sizeof(struct me_list) * ME_HASH_SIZE, M_IFME, M_WAITOK); for (i = 0; i < ME_HASH_SIZE; i++) CK_LIST_INIT(&hash[i]); return (hash); } static void vnet_me_init(const void *unused __unused) { V_me_cloner = if_clone_simple(mename, me_clone_create, me_clone_destroy, 0); } VNET_SYSINIT(vnet_me_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_me_init, NULL); static void vnet_me_uninit(const void *unused __unused) { if (V_me_hashtbl != NULL) free(V_me_hashtbl, M_IFME); if_clone_detach(V_me_cloner); } VNET_SYSUNINIT(vnet_me_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_me_uninit, NULL); static int me_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct me_softc *sc; sc = malloc(sizeof(struct me_softc), M_IFME, M_WAITOK | M_ZERO); sc->me_fibnum = curthread->td_proc->p_fibnum; ME2IFP(sc) = if_alloc(IFT_TUNNEL); ME2IFP(sc)->if_softc = sc; if_initname(ME2IFP(sc), mename, unit); ME2IFP(sc)->if_mtu = MEMTU;; ME2IFP(sc)->if_flags = IFF_POINTOPOINT|IFF_MULTICAST; ME2IFP(sc)->if_output = me_output; ME2IFP(sc)->if_ioctl = me_ioctl; ME2IFP(sc)->if_transmit = me_transmit; ME2IFP(sc)->if_qflush = me_qflush; ME2IFP(sc)->if_capabilities |= IFCAP_LINKSTATE; ME2IFP(sc)->if_capenable |= IFCAP_LINKSTATE; if_attach(ME2IFP(sc)); bpfattach(ME2IFP(sc), DLT_NULL, sizeof(u_int32_t)); return (0); } static void me_clone_destroy(struct ifnet *ifp) { struct me_softc *sc; sx_xlock(&me_ioctl_sx); sc = ifp->if_softc; me_delete_tunnel(sc); bpfdetach(ifp); if_detach(ifp); ifp->if_softc = NULL; sx_xunlock(&me_ioctl_sx); ME_WAIT(); if_free(ifp); free(sc, M_IFME); } static int me_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; struct sockaddr_in *src, *dst; struct me_softc *sc; int error; switch (cmd) { case SIOCSIFMTU: if (ifr->ifr_mtu < 576) return (EINVAL); ifp->if_mtu = ifr->ifr_mtu; return (0); case SIOCSIFADDR: ifp->if_flags |= IFF_UP; case SIOCSIFFLAGS: case SIOCADDMULTI: case SIOCDELMULTI: return (0); } sx_xlock(&me_ioctl_sx); sc = ifp->if_softc; if (sc == NULL) { error = ENXIO; goto end; } error = 0; switch (cmd) { case SIOCSIFPHYADDR: src = &((struct in_aliasreq *)data)->ifra_addr; dst = &((struct in_aliasreq *)data)->ifra_dstaddr; if (src->sin_family != dst->sin_family || src->sin_family != AF_INET || src->sin_len != dst->sin_len || src->sin_len != sizeof(struct sockaddr_in)) { error = EINVAL; break; } if (src->sin_addr.s_addr == INADDR_ANY || dst->sin_addr.s_addr == INADDR_ANY) { error = EADDRNOTAVAIL; break; } error = me_set_tunnel(sc, src->sin_addr.s_addr, dst->sin_addr.s_addr); break; case SIOCDIFPHYADDR: me_delete_tunnel(sc); break; case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: if (!ME_READY(sc)) { error = EADDRNOTAVAIL; break; } src = (struct sockaddr_in *)&ifr->ifr_addr; memset(src, 0, sizeof(*src)); src->sin_family = AF_INET; src->sin_len = sizeof(*src); switch (cmd) { case SIOCGIFPSRCADDR: src->sin_addr = sc->me_src; break; case SIOCGIFPDSTADDR: src->sin_addr = sc->me_dst; break; } error = prison_if(curthread->td_ucred, sintosa(src)); if (error != 0) memset(src, 0, sizeof(*src)); break; case SIOCGTUNFIB: ifr->ifr_fib = sc->me_fibnum; break; case SIOCSTUNFIB: if ((error = priv_check(curthread, PRIV_NET_GRE)) != 0) break; if (ifr->ifr_fib >= rt_numfibs) error = EINVAL; else sc->me_fibnum = ifr->ifr_fib; break; default: error = EINVAL; break; } end: sx_xunlock(&me_ioctl_sx); return (error); } static int me_lookup(const struct mbuf *m, int off, int proto, void **arg) { const struct ip *ip; struct me_softc *sc; if (V_me_hashtbl == NULL) return (0); MPASS(in_epoch(net_epoch_preempt)); ip = mtod(m, const struct ip *); CK_LIST_FOREACH(sc, &ME_HASH(ip->ip_dst.s_addr, ip->ip_src.s_addr), chain) { if (sc->me_src.s_addr == ip->ip_dst.s_addr && sc->me_dst.s_addr == ip->ip_src.s_addr) { if ((ME2IFP(sc)->if_flags & IFF_UP) == 0) return (0); *arg = sc; return (ENCAP_DRV_LOOKUP); } } return (0); } static int me_set_tunnel(struct me_softc *sc, in_addr_t src, in_addr_t dst) { struct me_softc *tmp; sx_assert(&me_ioctl_sx, SA_XLOCKED); if (V_me_hashtbl == NULL) V_me_hashtbl = me_hashinit(); if (sc->me_src.s_addr == src && sc->me_dst.s_addr == dst) return (0); CK_LIST_FOREACH(tmp, &ME_HASH(src, dst), chain) { if (tmp == sc) continue; if (tmp->me_src.s_addr == src && tmp->me_dst.s_addr == dst) return (EADDRNOTAVAIL); } me_delete_tunnel(sc); sc->me_dst.s_addr = dst; sc->me_src.s_addr = src; CK_LIST_INSERT_HEAD(&ME_HASH(src, dst), sc, chain); ME2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING; if_link_state_change(ME2IFP(sc), LINK_STATE_UP); return (0); } static void me_delete_tunnel(struct me_softc *sc) { sx_assert(&me_ioctl_sx, SA_XLOCKED); if (ME_READY(sc)) { CK_LIST_REMOVE(sc, chain); ME_WAIT(); sc->me_src.s_addr = 0; sc->me_dst.s_addr = 0; ME2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; if_link_state_change(ME2IFP(sc), LINK_STATE_DOWN); } } static uint16_t me_in_cksum(uint16_t *p, int nwords) { uint32_t sum = 0; while (nwords-- > 0) sum += *p++; sum = (sum >> 16) + (sum & 0xffff); sum += (sum >> 16); return (~sum); } static int me_input(struct mbuf *m, int off, int proto, void *arg) { struct me_softc *sc = arg; struct mobhdr *mh; struct ifnet *ifp; struct ip *ip; int hlen; ifp = ME2IFP(sc); /* checks for short packets */ hlen = sizeof(struct mobhdr); if (m->m_pkthdr.len < sizeof(struct ip) + hlen) hlen -= sizeof(struct in_addr); if (m->m_len < sizeof(struct ip) + hlen) m = m_pullup(m, sizeof(struct ip) + hlen); if (m == NULL) goto drop; mh = (struct mobhdr *)mtodo(m, sizeof(struct ip)); /* check for wrong flags */ if (mh->mob_flags & (~MOB_FLAGS_SP)) { m_freem(m); goto drop; } if (mh->mob_flags) { if (hlen != sizeof(struct mobhdr)) { m_freem(m); goto drop; } } else hlen = sizeof(struct mobhdr) - sizeof(struct in_addr); /* check mobile header checksum */ if (me_in_cksum((uint16_t *)mh, hlen / sizeof(uint16_t)) != 0) { m_freem(m); goto drop; } #ifdef MAC mac_ifnet_create_mbuf(ifp, m); #endif ip = mtod(m, struct ip *); ip->ip_dst = mh->mob_dst; ip->ip_p = mh->mob_proto; ip->ip_sum = 0; ip->ip_len = htons(m->m_pkthdr.len - hlen); if (mh->mob_flags) ip->ip_src = mh->mob_src; memmove(mtodo(m, hlen), ip, sizeof(struct ip)); m_adj(m, hlen); m_clrprotoflags(m); m->m_pkthdr.rcvif = ifp; m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); M_SETFIB(m, ifp->if_fib); hlen = AF_INET; BPF_MTAP2(ifp, &hlen, sizeof(hlen), m); if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); if ((ifp->if_flags & IFF_MONITOR) != 0) m_freem(m); else netisr_dispatch(NETISR_IP, m); return (IPPROTO_DONE); drop: if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); return (IPPROTO_DONE); } static int me_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro __unused) { uint32_t af; if (dst->sa_family == AF_UNSPEC) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; m->m_pkthdr.csum_data = af; return (ifp->if_transmit(ifp, m)); } #define MTAG_ME 1414491977 static int me_transmit(struct ifnet *ifp, struct mbuf *m) { struct mobhdr mh; struct me_softc *sc; struct ip *ip; uint32_t af; int error, hlen, plen; #ifdef MAC error = mac_ifnet_check_transmit(ifp, m); if (error != 0) goto drop; #endif error = ENETDOWN; ME_RLOCK(); sc = ifp->if_softc; if (sc == NULL || !ME_READY(sc) || (ifp->if_flags & IFF_MONITOR) != 0 || (ifp->if_flags & IFF_UP) == 0 || (error = if_tunnel_check_nesting(ifp, m, MTAG_ME, V_max_me_nesting)) != 0) { m_freem(m); goto drop; } af = m->m_pkthdr.csum_data; if (af != AF_INET) { error = EAFNOSUPPORT; m_freem(m); goto drop; } if (m->m_len < sizeof(struct ip)) m = m_pullup(m, sizeof(struct ip)); if (m == NULL) { error = ENOBUFS; goto drop; } ip = mtod(m, struct ip *); /* Fragmented datagramms shouldn't be encapsulated */ if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { error = EINVAL; m_freem(m); goto drop; } mh.mob_proto = ip->ip_p; mh.mob_src = ip->ip_src; mh.mob_dst = ip->ip_dst; if (in_hosteq(sc->me_src, ip->ip_src)) { hlen = sizeof(struct mobhdr) - sizeof(struct in_addr); mh.mob_flags = 0; } else { hlen = sizeof(struct mobhdr); mh.mob_flags = MOB_FLAGS_SP; } BPF_MTAP2(ifp, &af, sizeof(af), m); plen = m->m_pkthdr.len; ip->ip_src = sc->me_src; ip->ip_dst = sc->me_dst; m->m_flags &= ~(M_BCAST|M_MCAST); M_SETFIB(m, sc->me_fibnum); M_PREPEND(m, hlen, M_NOWAIT); if (m == NULL) { error = ENOBUFS; goto drop; } if (m->m_len < sizeof(struct ip) + hlen) m = m_pullup(m, sizeof(struct ip) + hlen); if (m == NULL) { error = ENOBUFS; goto drop; } memmove(mtod(m, void *), mtodo(m, hlen), sizeof(struct ip)); ip = mtod(m, struct ip *); ip->ip_len = htons(m->m_pkthdr.len); ip->ip_p = IPPROTO_MOBILE; ip->ip_sum = 0; mh.mob_csum = 0; mh.mob_csum = me_in_cksum((uint16_t *)&mh, hlen / sizeof(uint16_t)); bcopy(&mh, mtodo(m, sizeof(struct ip)), hlen); error = ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL); drop: if (error) if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); else { if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, plen); } ME_RUNLOCK(); return (error); } static void me_qflush(struct ifnet *ifp __unused) { } static const struct encaptab *ecookie = NULL; static const struct encap_config me_encap_cfg = { .proto = IPPROTO_MOBILE, .min_length = sizeof(struct ip) + sizeof(struct mobhdr) - sizeof(in_addr_t), .exact_match = ENCAP_DRV_LOOKUP, .lookup = me_lookup, .input = me_input }; static int memodevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: ecookie = ip_encap_attach(&me_encap_cfg, NULL, M_WAITOK); break; case MOD_UNLOAD: ip_encap_detach(ecookie); break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t me_mod = { "if_me", memodevent, 0 }; DECLARE_MODULE(if_me, me_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_me, 1); Index: head/sys/net/if_vlan.c =================================================================== --- head/sys/net/if_vlan.c (revision 336675) +++ head/sys/net/if_vlan.c (revision 336676) @@ -1,1988 +1,1988 @@ /*- * Copyright 1998 Massachusetts Institute of Technology * Copyright 2012 ADARA Networks, Inc. * Copyright 2017 Dell EMC Isilon * * Portions of this software were developed by Robert N. M. Watson under * contract to ADARA Networks, Inc. * * Permission to use, copy, modify, and distribute this software and * its documentation for any purpose and without fee is hereby * granted, provided that both the above copyright notice and this * permission notice appear in all copies, that both the above * copyright notice and this permission notice appear in all * supporting documentation, and that the name of M.I.T. not be used * in advertising or publicity pertaining to distribution of the * software without specific, written prior permission. M.I.T. makes * no representations about the suitability of this software for any * purpose. It is provided "as is" without express or implied * warranty. * * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * if_vlan.c - pseudo-device driver for IEEE 802.1Q virtual LANs. * This is sort of sneaky in the implementation, since * we need to pretend to be enough of an Ethernet implementation * to make arp work. The way we do this is by telling everyone * that we are an Ethernet, and then catch the packets that * ether_output() sends to us via if_transmit(), rewrite them for * use by the real outgoing interface, and ask it to send them. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_vlan.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif #define VLAN_DEF_HWIDTH 4 #define VLAN_IFFLAGS (IFF_BROADCAST | IFF_MULTICAST) #define UP_AND_RUNNING(ifp) \ ((ifp)->if_flags & IFF_UP && (ifp)->if_drv_flags & IFF_DRV_RUNNING) LIST_HEAD(ifvlanhead, ifvlan); struct ifvlantrunk { struct ifnet *parent; /* parent interface of this trunk */ struct rmlock lock; #ifdef VLAN_ARRAY #define VLAN_ARRAY_SIZE (EVL_VLID_MASK + 1) struct ifvlan *vlans[VLAN_ARRAY_SIZE]; /* static table */ #else struct ifvlanhead *hash; /* dynamic hash-list table */ uint16_t hmask; uint16_t hwidth; #endif int refcnt; }; /* * This macro provides a facility to iterate over every vlan on a trunk with * the assumption that none will be added/removed during iteration. */ #ifdef VLAN_ARRAY #define VLAN_FOREACH(_ifv, _trunk) \ size_t _i; \ for (_i = 0; _i < VLAN_ARRAY_SIZE; _i++) \ if (((_ifv) = (_trunk)->vlans[_i]) != NULL) #else /* VLAN_ARRAY */ #define VLAN_FOREACH(_ifv, _trunk) \ struct ifvlan *_next; \ size_t _i; \ for (_i = 0; _i < (1 << (_trunk)->hwidth); _i++) \ LIST_FOREACH_SAFE((_ifv), &(_trunk)->hash[_i], ifv_list, _next) #endif /* VLAN_ARRAY */ /* * This macro provides a facility to iterate over every vlan on a trunk while * also modifying the number of vlans on the trunk. The iteration continues * until some condition is met or there are no more vlans on the trunk. */ #ifdef VLAN_ARRAY /* The VLAN_ARRAY case is simple -- just a for loop using the condition. */ #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \ size_t _i; \ for (_i = 0; !(_cond) && _i < VLAN_ARRAY_SIZE; _i++) \ if (((_ifv) = (_trunk)->vlans[_i])) #else /* VLAN_ARRAY */ /* * The hash table case is more complicated. We allow for the hash table to be * modified (i.e. vlans removed) while we are iterating over it. To allow for * this we must restart the iteration every time we "touch" something during * the iteration, since removal will resize the hash table and invalidate our * current position. If acting on the touched element causes the trunk to be * emptied, then iteration also stops. */ #define VLAN_FOREACH_UNTIL_SAFE(_ifv, _trunk, _cond) \ size_t _i; \ bool _touch = false; \ for (_i = 0; \ !(_cond) && _i < (1 << (_trunk)->hwidth); \ _i = (_touch && ((_trunk) != NULL) ? 0 : _i + 1), _touch = false) \ if (((_ifv) = LIST_FIRST(&(_trunk)->hash[_i])) != NULL && \ (_touch = true)) #endif /* VLAN_ARRAY */ struct vlan_mc_entry { struct sockaddr_dl mc_addr; SLIST_ENTRY(vlan_mc_entry) mc_entries; }; struct ifvlan { struct ifvlantrunk *ifv_trunk; struct ifnet *ifv_ifp; #define TRUNK(ifv) ((ifv)->ifv_trunk) #define PARENT(ifv) ((ifv)->ifv_trunk->parent) void *ifv_cookie; int ifv_pflags; /* special flags we have set on parent */ int ifv_capenable; struct ifv_linkmib { int ifvm_encaplen; /* encapsulation length */ int ifvm_mtufudge; /* MTU fudged by this much */ int ifvm_mintu; /* min transmission unit */ uint16_t ifvm_proto; /* encapsulation ethertype */ uint16_t ifvm_tag; /* tag to apply on packets leaving if */ uint16_t ifvm_vid; /* VLAN ID */ uint8_t ifvm_pcp; /* Priority Code Point (PCP). */ } ifv_mib; struct task lladdr_task; SLIST_HEAD(, vlan_mc_entry) vlan_mc_listhead; #ifndef VLAN_ARRAY LIST_ENTRY(ifvlan) ifv_list; #endif }; #define ifv_proto ifv_mib.ifvm_proto #define ifv_tag ifv_mib.ifvm_tag #define ifv_vid ifv_mib.ifvm_vid #define ifv_pcp ifv_mib.ifvm_pcp #define ifv_encaplen ifv_mib.ifvm_encaplen #define ifv_mtufudge ifv_mib.ifvm_mtufudge #define ifv_mintu ifv_mib.ifvm_mintu /* Special flags we should propagate to parent. */ static struct { int flag; int (*func)(struct ifnet *, int); } vlan_pflags[] = { {IFF_PROMISC, ifpromisc}, {IFF_ALLMULTI, if_allmulti}, {0, NULL} }; extern int vlan_mtag_pcp; static const char vlanname[] = "vlan"; static MALLOC_DEFINE(M_VLAN, vlanname, "802.1Q Virtual LAN Interface"); static eventhandler_tag ifdetach_tag; static eventhandler_tag iflladdr_tag; /* * if_vlan uses two module-level locks to allow concurrent modification of vlan * interfaces and (mostly) allow for vlans to be destroyed while they are being * used for tx/rx. To accomplish this in a way that has acceptable performance * and cooperation with other parts of the network stack there is a * non-sleepable rmlock(9) and an sx(9). Both locks are exclusively acquired * when destroying a vlan interface, i.e. when the if_vlantrunk field of struct * ifnet is de-allocated and NULL'd. Thus a reader holding either lock has a * guarantee that the struct ifvlantrunk references a valid vlan trunk. * * The performance-sensitive paths that warrant using the rmlock(9) are * vlan_transmit and vlan_input. Both have to check for the vlan interface's * existence using if_vlantrunk, and being in the network tx/rx paths the use * of an rmlock(9) gives a measureable improvement in performance. * * The reason for having an sx(9) is mostly because there are still areas that * must be sleepable and also have safe concurrent access to a vlan interface. * Since the sx(9) exists, it is used by default in most paths unless sleeping * is not permitted, or if it is not clear whether sleeping is permitted. * * Note that despite these protections, there is still an inherent race in the * destruction of vlans since there's no guarantee that the ifnet hasn't been * freed/reused when the tx/rx functions are called by the stack. This can only * be fixed by addressing ifnet's lifetime issues. */ #define _VLAN_RM_ID ifv_rm_lock #define _VLAN_SX_ID ifv_sx static struct rmlock _VLAN_RM_ID; static struct sx _VLAN_SX_ID; #define VLAN_LOCKING_INIT() \ rm_init(&_VLAN_RM_ID, "vlan_rm"); \ sx_init(&_VLAN_SX_ID, "vlan_sx") #define VLAN_LOCKING_DESTROY() \ rm_destroy(&_VLAN_RM_ID); \ sx_destroy(&_VLAN_SX_ID) #define _VLAN_RM_TRACKER _vlan_rm_tracker #define VLAN_RLOCK() rm_rlock(&_VLAN_RM_ID, \ &_VLAN_RM_TRACKER) #define VLAN_RUNLOCK() rm_runlock(&_VLAN_RM_ID, \ &_VLAN_RM_TRACKER) #define VLAN_WLOCK() rm_wlock(&_VLAN_RM_ID) #define VLAN_WUNLOCK() rm_wunlock(&_VLAN_RM_ID) #define VLAN_RLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_RLOCKED) #define VLAN_WLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_WLOCKED) #define VLAN_RWLOCK_ASSERT() rm_assert(&_VLAN_RM_ID, RA_LOCKED) #define VLAN_LOCK_READER struct rm_priotracker _VLAN_RM_TRACKER #define VLAN_SLOCK() sx_slock(&_VLAN_SX_ID) #define VLAN_SUNLOCK() sx_sunlock(&_VLAN_SX_ID) #define VLAN_XLOCK() sx_xlock(&_VLAN_SX_ID) #define VLAN_XUNLOCK() sx_xunlock(&_VLAN_SX_ID) #define VLAN_SLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_SLOCKED) #define VLAN_XLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_XLOCKED) #define VLAN_SXLOCK_ASSERT() sx_assert(&_VLAN_SX_ID, SA_LOCKED) /* * We also have a per-trunk rmlock(9), that is locked shared on packet * processing and exclusive when configuration is changed. Note: This should * only be acquired while there is a shared lock on either of the global locks * via VLAN_SLOCK or VLAN_RLOCK. Thus, an exclusive lock on the global locks * makes a call to TRUNK_RLOCK/TRUNK_WLOCK technically superfluous. */ #define _TRUNK_RM_TRACKER _trunk_rm_tracker #define TRUNK_LOCK_INIT(trunk) rm_init(&(trunk)->lock, vlanname) #define TRUNK_LOCK_DESTROY(trunk) rm_destroy(&(trunk)->lock) #define TRUNK_RLOCK(trunk) rm_rlock(&(trunk)->lock, \ &_TRUNK_RM_TRACKER) #define TRUNK_WLOCK(trunk) rm_wlock(&(trunk)->lock) #define TRUNK_RUNLOCK(trunk) rm_runlock(&(trunk)->lock, \ &_TRUNK_RM_TRACKER) #define TRUNK_WUNLOCK(trunk) rm_wunlock(&(trunk)->lock) #define TRUNK_RLOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_RLOCKED) #define TRUNK_LOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_LOCKED) #define TRUNK_WLOCK_ASSERT(trunk) rm_assert(&(trunk)->lock, RA_WLOCKED) #define TRUNK_LOCK_READER struct rm_priotracker _TRUNK_RM_TRACKER /* * The VLAN_ARRAY substitutes the dynamic hash with a static array * with 4096 entries. In theory this can give a boost in processing, * however in practice it does not. Probably this is because the array * is too big to fit into CPU cache. */ #ifndef VLAN_ARRAY static void vlan_inithash(struct ifvlantrunk *trunk); static void vlan_freehash(struct ifvlantrunk *trunk); static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv); static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv); static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch); static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid); #endif static void trunk_destroy(struct ifvlantrunk *trunk); static void vlan_init(void *foo); static void vlan_input(struct ifnet *ifp, struct mbuf *m); static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr); #ifdef RATELIMIT static int vlan_snd_tag_alloc(struct ifnet *, union if_snd_tag_alloc_params *, struct m_snd_tag **); #endif static void vlan_qflush(struct ifnet *ifp); static int vlan_setflag(struct ifnet *ifp, int flag, int status, int (*func)(struct ifnet *, int)); static int vlan_setflags(struct ifnet *ifp, int status); static int vlan_setmulti(struct ifnet *ifp); static int vlan_transmit(struct ifnet *ifp, struct mbuf *m); static void vlan_unconfig(struct ifnet *ifp); static void vlan_unconfig_locked(struct ifnet *ifp, int departing); static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t tag); static void vlan_link_state(struct ifnet *ifp); static void vlan_capabilities(struct ifvlan *ifv); static void vlan_trunk_capabilities(struct ifnet *ifp); static struct ifnet *vlan_clone_match_ethervid(const char *, int *); static int vlan_clone_match(struct if_clone *, const char *); static int vlan_clone_create(struct if_clone *, char *, size_t, caddr_t); static int vlan_clone_destroy(struct if_clone *, struct ifnet *); static void vlan_ifdetach(void *arg, struct ifnet *ifp); static void vlan_iflladdr(void *arg, struct ifnet *ifp); static void vlan_lladdr_fn(void *arg, int pending); static struct if_clone *vlan_cloner; #ifdef VIMAGE -static VNET_DEFINE(struct if_clone *, vlan_cloner); +VNET_DEFINE_STATIC(struct if_clone *, vlan_cloner); #define V_vlan_cloner VNET(vlan_cloner) #endif #ifndef VLAN_ARRAY #define HASH(n, m) ((((n) >> 8) ^ ((n) >> 4) ^ (n)) & (m)) static void vlan_inithash(struct ifvlantrunk *trunk) { int i, n; /* * The trunk must not be locked here since we call malloc(M_WAITOK). * It is OK in case this function is called before the trunk struct * gets hooked up and becomes visible from other threads. */ KASSERT(trunk->hwidth == 0 && trunk->hash == NULL, ("%s: hash already initialized", __func__)); trunk->hwidth = VLAN_DEF_HWIDTH; n = 1 << trunk->hwidth; trunk->hmask = n - 1; trunk->hash = malloc(sizeof(struct ifvlanhead) * n, M_VLAN, M_WAITOK); for (i = 0; i < n; i++) LIST_INIT(&trunk->hash[i]); } static void vlan_freehash(struct ifvlantrunk *trunk) { #ifdef INVARIANTS int i; KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); for (i = 0; i < (1 << trunk->hwidth); i++) KASSERT(LIST_EMPTY(&trunk->hash[i]), ("%s: hash table not empty", __func__)); #endif free(trunk->hash, M_VLAN); trunk->hash = NULL; trunk->hwidth = trunk->hmask = 0; } static int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) { int i, b; struct ifvlan *ifv2; TRUNK_WLOCK_ASSERT(trunk); KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); b = 1 << trunk->hwidth; i = HASH(ifv->ifv_vid, trunk->hmask); LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) if (ifv->ifv_vid == ifv2->ifv_vid) return (EEXIST); /* * Grow the hash when the number of vlans exceeds half of the number of * hash buckets squared. This will make the average linked-list length * buckets/2. */ if (trunk->refcnt > (b * b) / 2) { vlan_growhash(trunk, 1); i = HASH(ifv->ifv_vid, trunk->hmask); } LIST_INSERT_HEAD(&trunk->hash[i], ifv, ifv_list); trunk->refcnt++; return (0); } static int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) { int i, b; struct ifvlan *ifv2; TRUNK_WLOCK_ASSERT(trunk); KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); b = 1 << trunk->hwidth; i = HASH(ifv->ifv_vid, trunk->hmask); LIST_FOREACH(ifv2, &trunk->hash[i], ifv_list) if (ifv2 == ifv) { trunk->refcnt--; LIST_REMOVE(ifv2, ifv_list); if (trunk->refcnt < (b * b) / 2) vlan_growhash(trunk, -1); return (0); } panic("%s: vlan not found\n", __func__); return (ENOENT); /*NOTREACHED*/ } /* * Grow the hash larger or smaller if memory permits. */ static void vlan_growhash(struct ifvlantrunk *trunk, int howmuch) { struct ifvlan *ifv; struct ifvlanhead *hash2; int hwidth2, i, j, n, n2; TRUNK_WLOCK_ASSERT(trunk); KASSERT(trunk->hwidth > 0, ("%s: hwidth not positive", __func__)); if (howmuch == 0) { /* Harmless yet obvious coding error */ printf("%s: howmuch is 0\n", __func__); return; } hwidth2 = trunk->hwidth + howmuch; n = 1 << trunk->hwidth; n2 = 1 << hwidth2; /* Do not shrink the table below the default */ if (hwidth2 < VLAN_DEF_HWIDTH) return; /* M_NOWAIT because we're called with trunk mutex held */ hash2 = malloc(sizeof(struct ifvlanhead) * n2, M_VLAN, M_NOWAIT); if (hash2 == NULL) { printf("%s: out of memory -- hash size not changed\n", __func__); return; /* We can live with the old hash table */ } for (j = 0; j < n2; j++) LIST_INIT(&hash2[j]); for (i = 0; i < n; i++) while ((ifv = LIST_FIRST(&trunk->hash[i])) != NULL) { LIST_REMOVE(ifv, ifv_list); j = HASH(ifv->ifv_vid, n2 - 1); LIST_INSERT_HEAD(&hash2[j], ifv, ifv_list); } free(trunk->hash, M_VLAN); trunk->hash = hash2; trunk->hwidth = hwidth2; trunk->hmask = n2 - 1; if (bootverbose) if_printf(trunk->parent, "VLAN hash table resized from %d to %d buckets\n", n, n2); } static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) { struct ifvlan *ifv; TRUNK_RLOCK_ASSERT(trunk); LIST_FOREACH(ifv, &trunk->hash[HASH(vid, trunk->hmask)], ifv_list) if (ifv->ifv_vid == vid) return (ifv); return (NULL); } #if 0 /* Debugging code to view the hashtables. */ static void vlan_dumphash(struct ifvlantrunk *trunk) { int i; struct ifvlan *ifv; for (i = 0; i < (1 << trunk->hwidth); i++) { printf("%d: ", i); LIST_FOREACH(ifv, &trunk->hash[i], ifv_list) printf("%s ", ifv->ifv_ifp->if_xname); printf("\n"); } } #endif /* 0 */ #else static __inline struct ifvlan * vlan_gethash(struct ifvlantrunk *trunk, uint16_t vid) { return trunk->vlans[vid]; } static __inline int vlan_inshash(struct ifvlantrunk *trunk, struct ifvlan *ifv) { if (trunk->vlans[ifv->ifv_vid] != NULL) return EEXIST; trunk->vlans[ifv->ifv_vid] = ifv; trunk->refcnt++; return (0); } static __inline int vlan_remhash(struct ifvlantrunk *trunk, struct ifvlan *ifv) { trunk->vlans[ifv->ifv_vid] = NULL; trunk->refcnt--; return (0); } static __inline void vlan_freehash(struct ifvlantrunk *trunk) { } static __inline void vlan_inithash(struct ifvlantrunk *trunk) { } #endif /* !VLAN_ARRAY */ static void trunk_destroy(struct ifvlantrunk *trunk) { VLAN_XLOCK_ASSERT(); VLAN_WLOCK_ASSERT(); vlan_freehash(trunk); trunk->parent->if_vlantrunk = NULL; TRUNK_LOCK_DESTROY(trunk); if_rele(trunk->parent); free(trunk, M_VLAN); } /* * Program our multicast filter. What we're actually doing is * programming the multicast filter of the parent. This has the * side effect of causing the parent interface to receive multicast * traffic that it doesn't really want, which ends up being discarded * later by the upper protocol layers. Unfortunately, there's no way * to avoid this: there really is only one physical interface. */ static int vlan_setmulti(struct ifnet *ifp) { struct ifnet *ifp_p; struct ifmultiaddr *ifma; struct ifvlan *sc; struct vlan_mc_entry *mc; int error; /* * XXX This stupidly needs the rmlock to avoid sleeping while holding * the in6_multi_mtx (see in6_mc_join_locked). */ VLAN_RWLOCK_ASSERT(); /* Find the parent. */ sc = ifp->if_softc; TRUNK_WLOCK_ASSERT(TRUNK(sc)); ifp_p = PARENT(sc); CURVNET_SET_QUIET(ifp_p->if_vnet); /* First, remove any existing filter entries. */ while ((mc = SLIST_FIRST(&sc->vlan_mc_listhead)) != NULL) { SLIST_REMOVE_HEAD(&sc->vlan_mc_listhead, mc_entries); (void)if_delmulti(ifp_p, (struct sockaddr *)&mc->mc_addr); free(mc, M_VLAN); } /* Now program new ones. */ IF_ADDR_WLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; mc = malloc(sizeof(struct vlan_mc_entry), M_VLAN, M_NOWAIT); if (mc == NULL) { IF_ADDR_WUNLOCK(ifp); return (ENOMEM); } bcopy(ifma->ifma_addr, &mc->mc_addr, ifma->ifma_addr->sa_len); mc->mc_addr.sdl_index = ifp_p->if_index; SLIST_INSERT_HEAD(&sc->vlan_mc_listhead, mc, mc_entries); } IF_ADDR_WUNLOCK(ifp); SLIST_FOREACH (mc, &sc->vlan_mc_listhead, mc_entries) { error = if_addmulti(ifp_p, (struct sockaddr *)&mc->mc_addr, NULL); if (error) return (error); } CURVNET_RESTORE(); return (0); } /* * A handler for parent interface link layer address changes. * If the parent interface link layer address is changed we * should also change it on all children vlans. */ static void vlan_iflladdr(void *arg __unused, struct ifnet *ifp) { struct ifvlan *ifv; struct ifnet *ifv_ifp; struct ifvlantrunk *trunk; struct sockaddr_dl *sdl; VLAN_LOCK_READER; /* Need the rmlock since this is run on taskqueue_swi. */ VLAN_RLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_RUNLOCK(); return; } /* * OK, it's a trunk. Loop over and change all vlan's lladdrs on it. * We need an exclusive lock here to prevent concurrent SIOCSIFLLADDR * ioctl calls on the parent garbling the lladdr of the child vlan. */ TRUNK_WLOCK(trunk); VLAN_FOREACH(ifv, trunk) { /* * Copy new new lladdr into the ifv_ifp, enqueue a task * to actually call if_setlladdr. if_setlladdr needs to * be deferred to a taskqueue because it will call into * the if_vlan ioctl path and try to acquire the global * lock. */ ifv_ifp = ifv->ifv_ifp; bcopy(IF_LLADDR(ifp), IF_LLADDR(ifv_ifp), ifp->if_addrlen); sdl = (struct sockaddr_dl *)ifv_ifp->if_addr->ifa_addr; sdl->sdl_alen = ifp->if_addrlen; taskqueue_enqueue(taskqueue_thread, &ifv->lladdr_task); } TRUNK_WUNLOCK(trunk); VLAN_RUNLOCK(); } /* * A handler for network interface departure events. * Track departure of trunks here so that we don't access invalid * pointers or whatever if a trunk is ripped from under us, e.g., * by ejecting its hot-plug card. However, if an ifnet is simply * being renamed, then there's no need to tear down the state. */ static void vlan_ifdetach(void *arg __unused, struct ifnet *ifp) { struct ifvlan *ifv; struct ifvlantrunk *trunk; /* If the ifnet is just being renamed, don't do anything. */ if (ifp->if_flags & IFF_RENAMING) return; VLAN_XLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_XUNLOCK(); return; } /* * OK, it's a trunk. Loop over and detach all vlan's on it. * Check trunk pointer after each vlan_unconfig() as it will * free it and set to NULL after the last vlan was detached. */ VLAN_FOREACH_UNTIL_SAFE(ifv, ifp->if_vlantrunk, ifp->if_vlantrunk == NULL) vlan_unconfig_locked(ifv->ifv_ifp, 1); /* Trunk should have been destroyed in vlan_unconfig(). */ KASSERT(ifp->if_vlantrunk == NULL, ("%s: purge failed", __func__)); VLAN_XUNLOCK(); } /* * Return the trunk device for a virtual interface. */ static struct ifnet * vlan_trunkdev(struct ifnet *ifp) { struct ifvlan *ifv; VLAN_LOCK_READER; if (ifp->if_type != IFT_L2VLAN) return (NULL); /* Not clear if callers are sleepable, so acquire the rmlock. */ VLAN_RLOCK(); ifv = ifp->if_softc; ifp = NULL; if (ifv->ifv_trunk) ifp = PARENT(ifv); VLAN_RUNLOCK(); return (ifp); } /* * Return the 12-bit VLAN VID for this interface, for use by external * components such as Infiniband. * * XXXRW: Note that the function name here is historical; it should be named * vlan_vid(). */ static int vlan_tag(struct ifnet *ifp, uint16_t *vidp) { struct ifvlan *ifv; if (ifp->if_type != IFT_L2VLAN) return (EINVAL); ifv = ifp->if_softc; *vidp = ifv->ifv_vid; return (0); } /* * Return a driver specific cookie for this interface. Synchronization * with setcookie must be provided by the driver. */ static void * vlan_cookie(struct ifnet *ifp) { struct ifvlan *ifv; if (ifp->if_type != IFT_L2VLAN) return (NULL); ifv = ifp->if_softc; return (ifv->ifv_cookie); } /* * Store a cookie in our softc that drivers can use to store driver * private per-instance data in. */ static int vlan_setcookie(struct ifnet *ifp, void *cookie) { struct ifvlan *ifv; if (ifp->if_type != IFT_L2VLAN) return (EINVAL); ifv = ifp->if_softc; ifv->ifv_cookie = cookie; return (0); } /* * Return the vlan device present at the specific VID. */ static struct ifnet * vlan_devat(struct ifnet *ifp, uint16_t vid) { struct ifvlantrunk *trunk; struct ifvlan *ifv; VLAN_LOCK_READER; TRUNK_LOCK_READER; /* Not clear if callers are sleepable, so acquire the rmlock. */ VLAN_RLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_RUNLOCK(); return (NULL); } ifp = NULL; TRUNK_RLOCK(trunk); ifv = vlan_gethash(trunk, vid); if (ifv) ifp = ifv->ifv_ifp; TRUNK_RUNLOCK(trunk); VLAN_RUNLOCK(); return (ifp); } /* * Recalculate the cached VLAN tag exposed via the MIB. */ static void vlan_tag_recalculate(struct ifvlan *ifv) { ifv->ifv_tag = EVL_MAKETAG(ifv->ifv_vid, ifv->ifv_pcp, 0); } /* * VLAN support can be loaded as a module. The only place in the * system that's intimately aware of this is ether_input. We hook * into this code through vlan_input_p which is defined there and * set here. No one else in the system should be aware of this so * we use an explicit reference here. */ extern void (*vlan_input_p)(struct ifnet *, struct mbuf *); /* For if_link_state_change() eyes only... */ extern void (*vlan_link_state_p)(struct ifnet *); static int vlan_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: ifdetach_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, vlan_ifdetach, NULL, EVENTHANDLER_PRI_ANY); if (ifdetach_tag == NULL) return (ENOMEM); iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event, vlan_iflladdr, NULL, EVENTHANDLER_PRI_ANY); if (iflladdr_tag == NULL) return (ENOMEM); VLAN_LOCKING_INIT(); vlan_input_p = vlan_input; vlan_link_state_p = vlan_link_state; vlan_trunk_cap_p = vlan_trunk_capabilities; vlan_trunkdev_p = vlan_trunkdev; vlan_cookie_p = vlan_cookie; vlan_setcookie_p = vlan_setcookie; vlan_tag_p = vlan_tag; vlan_devat_p = vlan_devat; #ifndef VIMAGE vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match, vlan_clone_create, vlan_clone_destroy); #endif if (bootverbose) printf("vlan: initialized, using " #ifdef VLAN_ARRAY "full-size arrays" #else "hash tables with chaining" #endif "\n"); break; case MOD_UNLOAD: #ifndef VIMAGE if_clone_detach(vlan_cloner); #endif EVENTHANDLER_DEREGISTER(ifnet_departure_event, ifdetach_tag); EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_tag); vlan_input_p = NULL; vlan_link_state_p = NULL; vlan_trunk_cap_p = NULL; vlan_trunkdev_p = NULL; vlan_tag_p = NULL; vlan_cookie_p = NULL; vlan_setcookie_p = NULL; vlan_devat_p = NULL; VLAN_LOCKING_DESTROY(); if (bootverbose) printf("vlan: unloaded\n"); break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t vlan_mod = { "if_vlan", vlan_modevent, 0 }; DECLARE_MODULE(if_vlan, vlan_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); MODULE_VERSION(if_vlan, 3); #ifdef VIMAGE static void vnet_vlan_init(const void *unused __unused) { vlan_cloner = if_clone_advanced(vlanname, 0, vlan_clone_match, vlan_clone_create, vlan_clone_destroy); V_vlan_cloner = vlan_cloner; } VNET_SYSINIT(vnet_vlan_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, vnet_vlan_init, NULL); static void vnet_vlan_uninit(const void *unused __unused) { if_clone_detach(V_vlan_cloner); } VNET_SYSUNINIT(vnet_vlan_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST, vnet_vlan_uninit, NULL); #endif /* * Check for . style interface names. */ static struct ifnet * vlan_clone_match_ethervid(const char *name, int *vidp) { char ifname[IFNAMSIZ]; char *cp; struct ifnet *ifp; int vid; strlcpy(ifname, name, IFNAMSIZ); if ((cp = strchr(ifname, '.')) == NULL) return (NULL); *cp = '\0'; if ((ifp = ifunit_ref(ifname)) == NULL) return (NULL); /* Parse VID. */ if (*++cp == '\0') { if_rele(ifp); return (NULL); } vid = 0; for(; *cp >= '0' && *cp <= '9'; cp++) vid = (vid * 10) + (*cp - '0'); if (*cp != '\0') { if_rele(ifp); return (NULL); } if (vidp != NULL) *vidp = vid; return (ifp); } static int vlan_clone_match(struct if_clone *ifc, const char *name) { const char *cp; if (vlan_clone_match_ethervid(name, NULL) != NULL) return (1); if (strncmp(vlanname, name, strlen(vlanname)) != 0) return (0); for (cp = name + 4; *cp != '\0'; cp++) { if (*cp < '0' || *cp > '9') return (0); } return (1); } static int vlan_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) { char *dp; int wildcard; int unit; int error; int vid; struct ifvlan *ifv; struct ifnet *ifp; struct ifnet *p; struct ifaddr *ifa; struct sockaddr_dl *sdl; struct vlanreq vlr; static const u_char eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ /* * There are 3 (ugh) ways to specify the cloned device: * o pass a parameter block with the clone request. * o specify parameters in the text of the clone device name * o specify no parameters and get an unattached device that * must be configured separately. * The first technique is preferred; the latter two are * supported for backwards compatibility. * * XXXRW: Note historic use of the word "tag" here. New ioctls may be * called for. */ if (params) { error = copyin(params, &vlr, sizeof(vlr)); if (error) return error; p = ifunit_ref(vlr.vlr_parent); if (p == NULL) return (ENXIO); error = ifc_name2unit(name, &unit); if (error != 0) { if_rele(p); return (error); } vid = vlr.vlr_tag; wildcard = (unit < 0); } else if ((p = vlan_clone_match_ethervid(name, &vid)) != NULL) { unit = -1; wildcard = 0; } else { p = NULL; error = ifc_name2unit(name, &unit); if (error != 0) return (error); wildcard = (unit < 0); } error = ifc_alloc_unit(ifc, &unit); if (error != 0) { if (p != NULL) if_rele(p); return (error); } /* In the wildcard case, we need to update the name. */ if (wildcard) { for (dp = name; *dp != '\0'; dp++); if (snprintf(dp, len - (dp-name), "%d", unit) > len - (dp-name) - 1) { panic("%s: interface name too long", __func__); } } ifv = malloc(sizeof(struct ifvlan), M_VLAN, M_WAITOK | M_ZERO); ifp = ifv->ifv_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { ifc_free_unit(ifc, unit); free(ifv, M_VLAN); if (p != NULL) if_rele(p); return (ENOSPC); } SLIST_INIT(&ifv->vlan_mc_listhead); ifp->if_softc = ifv; /* * Set the name manually rather than using if_initname because * we don't conform to the default naming convention for interfaces. */ strlcpy(ifp->if_xname, name, IFNAMSIZ); ifp->if_dname = vlanname; ifp->if_dunit = unit; /* NB: flags are not set here */ ifp->if_linkmib = &ifv->ifv_mib; ifp->if_linkmiblen = sizeof(ifv->ifv_mib); /* NB: mtu is not set here */ ifp->if_init = vlan_init; ifp->if_transmit = vlan_transmit; ifp->if_qflush = vlan_qflush; ifp->if_ioctl = vlan_ioctl; #ifdef RATELIMIT ifp->if_snd_tag_alloc = vlan_snd_tag_alloc; #endif ifp->if_flags = VLAN_IFFLAGS; ether_ifattach(ifp, eaddr); /* Now undo some of the damage... */ ifp->if_baudrate = 0; ifp->if_type = IFT_L2VLAN; ifp->if_hdrlen = ETHER_VLAN_ENCAP_LEN; ifa = ifp->if_addr; sdl = (struct sockaddr_dl *)ifa->ifa_addr; sdl->sdl_type = IFT_L2VLAN; if (p != NULL) { error = vlan_config(ifv, p, vid); if_rele(p); if (error != 0) { /* * Since we've partially failed, we need to back * out all the way, otherwise userland could get * confused. Thus, we destroy the interface. */ ether_ifdetach(ifp); vlan_unconfig(ifp); if_free(ifp); ifc_free_unit(ifc, unit); free(ifv, M_VLAN); return (error); } } return (0); } static int vlan_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) { struct ifvlan *ifv = ifp->if_softc; int unit = ifp->if_dunit; ether_ifdetach(ifp); /* first, remove it from system-wide lists */ vlan_unconfig(ifp); /* now it can be unconfigured and freed */ /* * We should have the only reference to the ifv now, so we can now * drain any remaining lladdr task before freeing the ifnet and the * ifvlan. */ taskqueue_drain(taskqueue_thread, &ifv->lladdr_task); if_free(ifp); free(ifv, M_VLAN); ifc_free_unit(ifc, unit); return (0); } /* * The ifp->if_init entry point for vlan(4) is a no-op. */ static void vlan_init(void *foo __unused) { } /* * The if_transmit method for vlan(4) interface. */ static int vlan_transmit(struct ifnet *ifp, struct mbuf *m) { struct ifvlan *ifv; struct ifnet *p; int error, len, mcast; VLAN_LOCK_READER; VLAN_RLOCK(); ifv = ifp->if_softc; if (TRUNK(ifv) == NULL) { if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); VLAN_RUNLOCK(); m_freem(m); return (ENETDOWN); } p = PARENT(ifv); len = m->m_pkthdr.len; mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; BPF_MTAP(ifp, m); /* * Do not run parent's if_transmit() if the parent is not up, * or parent's driver will cause a system crash. */ if (!UP_AND_RUNNING(p)) { if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); VLAN_RUNLOCK(); m_freem(m); return (ENETDOWN); } if (!ether_8021q_frame(&m, ifp, p, ifv->ifv_vid, ifv->ifv_pcp)) { if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); VLAN_RUNLOCK(); return (0); } /* * Send it, precisely as ether_output() would have. */ error = (p->if_transmit)(p, m); if (error == 0) { if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_OBYTES, len); if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast); } else if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); VLAN_RUNLOCK(); return (error); } /* * The ifp->if_qflush entry point for vlan(4) is a no-op. */ static void vlan_qflush(struct ifnet *ifp __unused) { } static void vlan_input(struct ifnet *ifp, struct mbuf *m) { struct ifvlantrunk *trunk; struct ifvlan *ifv; VLAN_LOCK_READER; TRUNK_LOCK_READER; struct m_tag *mtag; uint16_t vid, tag; VLAN_RLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_RUNLOCK(); m_freem(m); return; } if (m->m_flags & M_VLANTAG) { /* * Packet is tagged, but m contains a normal * Ethernet frame; the tag is stored out-of-band. */ tag = m->m_pkthdr.ether_vtag; m->m_flags &= ~M_VLANTAG; } else { struct ether_vlan_header *evl; /* * Packet is tagged in-band as specified by 802.1q. */ switch (ifp->if_type) { case IFT_ETHER: if (m->m_len < sizeof(*evl) && (m = m_pullup(m, sizeof(*evl))) == NULL) { if_printf(ifp, "cannot pullup VLAN header\n"); VLAN_RUNLOCK(); return; } evl = mtod(m, struct ether_vlan_header *); tag = ntohs(evl->evl_tag); /* * Remove the 802.1q header by copying the Ethernet * addresses over it and adjusting the beginning of * the data in the mbuf. The encapsulated Ethernet * type field is already in place. */ bcopy((char *)evl, (char *)evl + ETHER_VLAN_ENCAP_LEN, ETHER_HDR_LEN - ETHER_TYPE_LEN); m_adj(m, ETHER_VLAN_ENCAP_LEN); break; default: #ifdef INVARIANTS panic("%s: %s has unsupported if_type %u", __func__, ifp->if_xname, ifp->if_type); #endif if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1); VLAN_RUNLOCK(); m_freem(m); return; } } vid = EVL_VLANOFTAG(tag); TRUNK_RLOCK(trunk); ifv = vlan_gethash(trunk, vid); if (ifv == NULL || !UP_AND_RUNNING(ifv->ifv_ifp)) { TRUNK_RUNLOCK(trunk); if_inc_counter(ifp, IFCOUNTER_NOPROTO, 1); VLAN_RUNLOCK(); m_freem(m); return; } TRUNK_RUNLOCK(trunk); if (vlan_mtag_pcp) { /* * While uncommon, it is possible that we will find a 802.1q * packet encapsulated inside another packet that also had an * 802.1q header. For example, ethernet tunneled over IPSEC * arriving over ethernet. In that case, we replace the * existing 802.1q PCP m_tag value. */ mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); if (mtag == NULL) { mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_IN, sizeof(uint8_t), M_NOWAIT); if (mtag == NULL) { if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); VLAN_RUNLOCK(); m_freem(m); return; } m_tag_prepend(m, mtag); } *(uint8_t *)(mtag + 1) = EVL_PRIOFTAG(tag); } m->m_pkthdr.rcvif = ifv->ifv_ifp; if_inc_counter(ifv->ifv_ifp, IFCOUNTER_IPACKETS, 1); VLAN_RUNLOCK(); /* Pass it back through the parent's input routine. */ (*ifv->ifv_ifp->if_input)(ifv->ifv_ifp, m); } static void vlan_lladdr_fn(void *arg, int pending __unused) { struct ifvlan *ifv; struct ifnet *ifp; ifv = (struct ifvlan *)arg; ifp = ifv->ifv_ifp; /* The ifv_ifp already has the lladdr copied in. */ if_setlladdr(ifp, IF_LLADDR(ifp), ifp->if_addrlen); } static int vlan_config(struct ifvlan *ifv, struct ifnet *p, uint16_t vid) { struct ifvlantrunk *trunk; struct ifnet *ifp; int error = 0; /* * We can handle non-ethernet hardware types as long as * they handle the tagging and headers themselves. */ if (p->if_type != IFT_ETHER && (p->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) return (EPROTONOSUPPORT); if ((p->if_flags & VLAN_IFFLAGS) != VLAN_IFFLAGS) return (EPROTONOSUPPORT); /* * Don't let the caller set up a VLAN VID with * anything except VLID bits. * VID numbers 0x0 and 0xFFF are reserved. */ if (vid == 0 || vid == 0xFFF || (vid & ~EVL_VLID_MASK)) return (EINVAL); if (ifv->ifv_trunk) return (EBUSY); /* Acquire rmlock after the branch so we can M_WAITOK. */ VLAN_XLOCK(); if (p->if_vlantrunk == NULL) { trunk = malloc(sizeof(struct ifvlantrunk), M_VLAN, M_WAITOK | M_ZERO); vlan_inithash(trunk); TRUNK_LOCK_INIT(trunk); VLAN_WLOCK(); TRUNK_WLOCK(trunk); p->if_vlantrunk = trunk; trunk->parent = p; if_ref(trunk->parent); } else { VLAN_WLOCK(); trunk = p->if_vlantrunk; TRUNK_WLOCK(trunk); } ifv->ifv_vid = vid; /* must set this before vlan_inshash() */ ifv->ifv_pcp = 0; /* Default: best effort delivery. */ vlan_tag_recalculate(ifv); error = vlan_inshash(trunk, ifv); if (error) goto done; ifv->ifv_proto = ETHERTYPE_VLAN; ifv->ifv_encaplen = ETHER_VLAN_ENCAP_LEN; ifv->ifv_mintu = ETHERMIN; ifv->ifv_pflags = 0; ifv->ifv_capenable = -1; /* * If the parent supports the VLAN_MTU capability, * i.e. can Tx/Rx larger than ETHER_MAX_LEN frames, * use it. */ if (p->if_capenable & IFCAP_VLAN_MTU) { /* * No need to fudge the MTU since the parent can * handle extended frames. */ ifv->ifv_mtufudge = 0; } else { /* * Fudge the MTU by the encapsulation size. This * makes us incompatible with strictly compliant * 802.1Q implementations, but allows us to use * the feature with other NetBSD implementations, * which might still be useful. */ ifv->ifv_mtufudge = ifv->ifv_encaplen; } ifv->ifv_trunk = trunk; ifp = ifv->ifv_ifp; /* * Initialize fields from our parent. This duplicates some * work with ether_ifattach() but allows for non-ethernet * interfaces to also work. */ ifp->if_mtu = p->if_mtu - ifv->ifv_mtufudge; ifp->if_baudrate = p->if_baudrate; ifp->if_output = p->if_output; ifp->if_input = p->if_input; ifp->if_resolvemulti = p->if_resolvemulti; ifp->if_addrlen = p->if_addrlen; ifp->if_broadcastaddr = p->if_broadcastaddr; /* * Copy only a selected subset of flags from the parent. * Other flags are none of our business. */ #define VLAN_COPY_FLAGS (IFF_SIMPLEX) ifp->if_flags &= ~VLAN_COPY_FLAGS; ifp->if_flags |= p->if_flags & VLAN_COPY_FLAGS; #undef VLAN_COPY_FLAGS ifp->if_link_state = p->if_link_state; vlan_capabilities(ifv); /* * Set up our interface address to reflect the underlying * physical interface's. */ bcopy(IF_LLADDR(p), IF_LLADDR(ifp), p->if_addrlen); ((struct sockaddr_dl *)ifp->if_addr->ifa_addr)->sdl_alen = p->if_addrlen; /* * Configure multicast addresses that may already be * joined on the vlan device. */ (void)vlan_setmulti(ifp); TASK_INIT(&ifv->lladdr_task, 0, vlan_lladdr_fn, ifv); /* We are ready for operation now. */ ifp->if_drv_flags |= IFF_DRV_RUNNING; /* Update flags on the parent, if necessary. */ vlan_setflags(ifp, 1); done: /* * We need to drop the non-sleepable rmlock so that the underlying * devices can sleep in their vlan_config hooks. */ TRUNK_WUNLOCK(trunk); VLAN_WUNLOCK(); if (error == 0) EVENTHANDLER_INVOKE(vlan_config, p, ifv->ifv_vid); VLAN_XUNLOCK(); return (error); } static void vlan_unconfig(struct ifnet *ifp) { VLAN_XLOCK(); vlan_unconfig_locked(ifp, 0); VLAN_XUNLOCK(); } static void vlan_unconfig_locked(struct ifnet *ifp, int departing) { struct ifvlantrunk *trunk; struct vlan_mc_entry *mc; struct ifvlan *ifv; struct ifnet *parent; int error; VLAN_XLOCK_ASSERT(); ifv = ifp->if_softc; trunk = ifv->ifv_trunk; parent = NULL; if (trunk != NULL) { /* * Both vlan_transmit and vlan_input rely on the trunk fields * being NULL to determine whether to bail, so we need to get * an exclusive lock here to prevent them from using bad * ifvlans. */ VLAN_WLOCK(); parent = trunk->parent; /* * Since the interface is being unconfigured, we need to * empty the list of multicast groups that we may have joined * while we were alive from the parent's list. */ while ((mc = SLIST_FIRST(&ifv->vlan_mc_listhead)) != NULL) { /* * If the parent interface is being detached, * all its multicast addresses have already * been removed. Warn about errors if * if_delmulti() does fail, but don't abort as * all callers expect vlan destruction to * succeed. */ if (!departing) { error = if_delmulti(parent, (struct sockaddr *)&mc->mc_addr); if (error) if_printf(ifp, "Failed to delete multicast address from parent: %d\n", error); } SLIST_REMOVE_HEAD(&ifv->vlan_mc_listhead, mc_entries); free(mc, M_VLAN); } vlan_setflags(ifp, 0); /* clear special flags on parent */ /* * The trunk lock isn't actually required here, but * vlan_remhash expects it. */ TRUNK_WLOCK(trunk); vlan_remhash(trunk, ifv); TRUNK_WUNLOCK(trunk); ifv->ifv_trunk = NULL; /* * Check if we were the last. */ if (trunk->refcnt == 0) { parent->if_vlantrunk = NULL; trunk_destroy(trunk); } VLAN_WUNLOCK(); } /* Disconnect from parent. */ if (ifv->ifv_pflags) if_printf(ifp, "%s: ifv_pflags unclean\n", __func__); ifp->if_mtu = ETHERMTU; ifp->if_link_state = LINK_STATE_UNKNOWN; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; /* * Only dispatch an event if vlan was * attached, otherwise there is nothing * to cleanup anyway. */ if (parent != NULL) EVENTHANDLER_INVOKE(vlan_unconfig, parent, ifv->ifv_vid); } /* Handle a reference counted flag that should be set on the parent as well */ static int vlan_setflag(struct ifnet *ifp, int flag, int status, int (*func)(struct ifnet *, int)) { struct ifvlan *ifv; int error; VLAN_SXLOCK_ASSERT(); ifv = ifp->if_softc; status = status ? (ifp->if_flags & flag) : 0; /* Now "status" contains the flag value or 0 */ /* * See if recorded parent's status is different from what * we want it to be. If it is, flip it. We record parent's * status in ifv_pflags so that we won't clear parent's flag * we haven't set. In fact, we don't clear or set parent's * flags directly, but get or release references to them. * That's why we can be sure that recorded flags still are * in accord with actual parent's flags. */ if (status != (ifv->ifv_pflags & flag)) { error = (*func)(PARENT(ifv), status); if (error) return (error); ifv->ifv_pflags &= ~flag; ifv->ifv_pflags |= status; } return (0); } /* * Handle IFF_* flags that require certain changes on the parent: * if "status" is true, update parent's flags respective to our if_flags; * if "status" is false, forcedly clear the flags set on parent. */ static int vlan_setflags(struct ifnet *ifp, int status) { int error, i; for (i = 0; vlan_pflags[i].flag; i++) { error = vlan_setflag(ifp, vlan_pflags[i].flag, status, vlan_pflags[i].func); if (error) return (error); } return (0); } /* Inform all vlans that their parent has changed link state */ static void vlan_link_state(struct ifnet *ifp) { struct ifvlantrunk *trunk; struct ifvlan *ifv; VLAN_LOCK_READER; /* Called from a taskqueue_swi task, so we cannot sleep. */ VLAN_RLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_RUNLOCK(); return; } TRUNK_WLOCK(trunk); VLAN_FOREACH(ifv, trunk) { ifv->ifv_ifp->if_baudrate = trunk->parent->if_baudrate; if_link_state_change(ifv->ifv_ifp, trunk->parent->if_link_state); } TRUNK_WUNLOCK(trunk); VLAN_RUNLOCK(); } static void vlan_capabilities(struct ifvlan *ifv) { struct ifnet *p; struct ifnet *ifp; struct ifnet_hw_tsomax hw_tsomax; int cap = 0, ena = 0, mena; u_long hwa = 0; VLAN_SXLOCK_ASSERT(); TRUNK_WLOCK_ASSERT(TRUNK(ifv)); p = PARENT(ifv); ifp = ifv->ifv_ifp; /* Mask parent interface enabled capabilities disabled by user. */ mena = p->if_capenable & ifv->ifv_capenable; /* * If the parent interface can do checksum offloading * on VLANs, then propagate its hardware-assisted * checksumming flags. Also assert that checksum * offloading requires hardware VLAN tagging. */ if (p->if_capabilities & IFCAP_VLAN_HWCSUM) cap |= p->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6); if (p->if_capenable & IFCAP_VLAN_HWCSUM && p->if_capenable & IFCAP_VLAN_HWTAGGING) { ena |= mena & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6); if (ena & IFCAP_TXCSUM) hwa |= p->if_hwassist & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP); if (ena & IFCAP_TXCSUM_IPV6) hwa |= p->if_hwassist & (CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_SCTP_IPV6); } /* * If the parent interface can do TSO on VLANs then * propagate the hardware-assisted flag. TSO on VLANs * does not necessarily require hardware VLAN tagging. */ memset(&hw_tsomax, 0, sizeof(hw_tsomax)); if_hw_tsomax_common(p, &hw_tsomax); if_hw_tsomax_update(ifp, &hw_tsomax); if (p->if_capabilities & IFCAP_VLAN_HWTSO) cap |= p->if_capabilities & IFCAP_TSO; if (p->if_capenable & IFCAP_VLAN_HWTSO) { ena |= mena & IFCAP_TSO; if (ena & IFCAP_TSO) hwa |= p->if_hwassist & CSUM_TSO; } /* * If the parent interface can do LRO and checksum offloading on * VLANs, then guess it may do LRO on VLANs. False positive here * cost nothing, while false negative may lead to some confusions. */ if (p->if_capabilities & IFCAP_VLAN_HWCSUM) cap |= p->if_capabilities & IFCAP_LRO; if (p->if_capenable & IFCAP_VLAN_HWCSUM) ena |= p->if_capenable & IFCAP_LRO; /* * If the parent interface can offload TCP connections over VLANs then * propagate its TOE capability to the VLAN interface. * * All TOE drivers in the tree today can deal with VLANs. If this * changes then IFCAP_VLAN_TOE should be promoted to a full capability * with its own bit. */ #define IFCAP_VLAN_TOE IFCAP_TOE if (p->if_capabilities & IFCAP_VLAN_TOE) cap |= p->if_capabilities & IFCAP_TOE; if (p->if_capenable & IFCAP_VLAN_TOE) { TOEDEV(ifp) = TOEDEV(p); ena |= mena & IFCAP_TOE; } /* * If the parent interface supports dynamic link state, so does the * VLAN interface. */ cap |= (p->if_capabilities & IFCAP_LINKSTATE); ena |= (mena & IFCAP_LINKSTATE); #ifdef RATELIMIT /* * If the parent interface supports ratelimiting, so does the * VLAN interface. */ cap |= (p->if_capabilities & IFCAP_TXRTLMT); ena |= (mena & IFCAP_TXRTLMT); #endif ifp->if_capabilities = cap; ifp->if_capenable = ena; ifp->if_hwassist = hwa; } static void vlan_trunk_capabilities(struct ifnet *ifp) { struct ifvlantrunk *trunk; struct ifvlan *ifv; VLAN_SLOCK(); trunk = ifp->if_vlantrunk; if (trunk == NULL) { VLAN_SUNLOCK(); return; } TRUNK_WLOCK(trunk); VLAN_FOREACH(ifv, trunk) { vlan_capabilities(ifv); } TRUNK_WUNLOCK(trunk); VLAN_SUNLOCK(); } static int vlan_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct ifnet *p; struct ifreq *ifr; struct ifaddr *ifa; struct ifvlan *ifv; struct ifvlantrunk *trunk; struct vlanreq vlr; int error = 0; VLAN_LOCK_READER; ifr = (struct ifreq *)data; ifa = (struct ifaddr *) data; ifv = ifp->if_softc; switch (cmd) { case SIOCSIFADDR: ifp->if_flags |= IFF_UP; #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) arp_ifinit(ifp, ifa); #endif break; case SIOCGIFADDR: bcopy(IF_LLADDR(ifp), &ifr->ifr_addr.sa_data[0], ifp->if_addrlen); break; case SIOCGIFMEDIA: VLAN_SLOCK(); if (TRUNK(ifv) != NULL) { p = PARENT(ifv); if_ref(p); error = (*p->if_ioctl)(p, SIOCGIFMEDIA, data); if_rele(p); /* Limit the result to the parent's current config. */ if (error == 0) { struct ifmediareq *ifmr; ifmr = (struct ifmediareq *)data; if (ifmr->ifm_count >= 1 && ifmr->ifm_ulist) { ifmr->ifm_count = 1; error = copyout(&ifmr->ifm_current, ifmr->ifm_ulist, sizeof(int)); } } } else { error = EINVAL; } VLAN_SUNLOCK(); break; case SIOCSIFMEDIA: error = EINVAL; break; case SIOCSIFMTU: /* * Set the interface MTU. */ VLAN_SLOCK(); trunk = TRUNK(ifv); if (trunk != NULL) { TRUNK_WLOCK(trunk); if (ifr->ifr_mtu > (PARENT(ifv)->if_mtu - ifv->ifv_mtufudge) || ifr->ifr_mtu < (ifv->ifv_mintu - ifv->ifv_mtufudge)) error = EINVAL; else ifp->if_mtu = ifr->ifr_mtu; TRUNK_WUNLOCK(trunk); } else error = EINVAL; VLAN_SUNLOCK(); break; case SIOCSETVLAN: #ifdef VIMAGE /* * XXXRW/XXXBZ: The goal in these checks is to allow a VLAN * interface to be delegated to a jail without allowing the * jail to change what underlying interface/VID it is * associated with. We are not entirely convinced that this * is the right way to accomplish that policy goal. */ if (ifp->if_vnet != ifp->if_home_vnet) { error = EPERM; break; } #endif error = copyin(ifr_data_get_ptr(ifr), &vlr, sizeof(vlr)); if (error) break; if (vlr.vlr_parent[0] == '\0') { vlan_unconfig(ifp); break; } p = ifunit_ref(vlr.vlr_parent); if (p == NULL) { error = ENOENT; break; } error = vlan_config(ifv, p, vlr.vlr_tag); if_rele(p); break; case SIOCGETVLAN: #ifdef VIMAGE if (ifp->if_vnet != ifp->if_home_vnet) { error = EPERM; break; } #endif bzero(&vlr, sizeof(vlr)); VLAN_SLOCK(); if (TRUNK(ifv) != NULL) { strlcpy(vlr.vlr_parent, PARENT(ifv)->if_xname, sizeof(vlr.vlr_parent)); vlr.vlr_tag = ifv->ifv_vid; } VLAN_SUNLOCK(); error = copyout(&vlr, ifr_data_get_ptr(ifr), sizeof(vlr)); break; case SIOCSIFFLAGS: /* * We should propagate selected flags to the parent, * e.g., promiscuous mode. */ VLAN_XLOCK(); if (TRUNK(ifv) != NULL) error = vlan_setflags(ifp, 1); VLAN_XUNLOCK(); break; case SIOCADDMULTI: case SIOCDELMULTI: /* * If we don't have a parent, just remember the membership for * when we do. * * XXX We need the rmlock here to avoid sleeping while * holding in6_multi_mtx. */ VLAN_RLOCK(); trunk = TRUNK(ifv); if (trunk != NULL) { TRUNK_WLOCK(trunk); error = vlan_setmulti(ifp); TRUNK_WUNLOCK(trunk); } VLAN_RUNLOCK(); break; case SIOCGVLANPCP: #ifdef VIMAGE if (ifp->if_vnet != ifp->if_home_vnet) { error = EPERM; break; } #endif ifr->ifr_vlan_pcp = ifv->ifv_pcp; break; case SIOCSVLANPCP: #ifdef VIMAGE if (ifp->if_vnet != ifp->if_home_vnet) { error = EPERM; break; } #endif error = priv_check(curthread, PRIV_NET_SETVLANPCP); if (error) break; if (ifr->ifr_vlan_pcp > 7) { error = EINVAL; break; } ifv->ifv_pcp = ifr->ifr_vlan_pcp; vlan_tag_recalculate(ifv); /* broadcast event about PCP change */ EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_PCP); break; case SIOCSIFCAP: VLAN_SLOCK(); ifv->ifv_capenable = ifr->ifr_reqcap; trunk = TRUNK(ifv); if (trunk != NULL) { TRUNK_WLOCK(trunk); vlan_capabilities(ifv); TRUNK_WUNLOCK(trunk); } VLAN_SUNLOCK(); break; default: error = EINVAL; break; } return (error); } #ifdef RATELIMIT static int vlan_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { /* get trunk device */ ifp = vlan_trunkdev(ifp); if (ifp == NULL || (ifp->if_capenable & IFCAP_TXRTLMT) == 0) return (EOPNOTSUPP); /* forward allocation request */ return (ifp->if_snd_tag_alloc(ifp, params, ppmt)); } #endif Index: head/sys/net/netisr.c =================================================================== --- head/sys/net/netisr.c (revision 336675) +++ head/sys/net/netisr.c (revision 336676) @@ -1,1533 +1,1533 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2007-2009 Robert N. M. Watson * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * This software was developed by Robert N. M. Watson under contract * to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * netisr is a packet dispatch service, allowing synchronous (directly * dispatched) and asynchronous (deferred dispatch) processing of packets by * registered protocol handlers. Callers pass a protocol identifier and * packet to netisr, along with a direct dispatch hint, and work will either * be immediately processed by the registered handler, or passed to a * software interrupt (SWI) thread for deferred dispatch. Callers will * generally select one or the other based on: * * - Whether directly dispatching a netisr handler lead to code reentrance or * lock recursion, such as entering the socket code from the socket code. * - Whether directly dispatching a netisr handler lead to recursive * processing, such as when decapsulating several wrapped layers of tunnel * information (IPSEC within IPSEC within ...). * * Maintaining ordering for protocol streams is a critical design concern. * Enforcing ordering limits the opportunity for concurrency, but maintains * the strong ordering requirements found in some protocols, such as TCP. Of * related concern is CPU affinity--it is desirable to process all data * associated with a particular stream on the same CPU over time in order to * avoid acquiring locks associated with the connection on different CPUs, * keep connection data in one cache, and to generally encourage associated * user threads to live on the same CPU as the stream. It's also desirable * to avoid lock migration and contention where locks are associated with * more than one flow. * * netisr supports several policy variations, represented by the * NETISR_POLICY_* constants, allowing protocols to play various roles in * identifying flows, assigning work to CPUs, etc. These are described in * netisr.h. */ #include "opt_ddb.h" #include "opt_device_polling.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #define _WANT_NETISR_INTERNAL /* Enable definitions from netisr_internal.h */ #include #include #include #include #include /*- * Synchronize use and modification of the registered netisr data structures; * acquire a read lock while modifying the set of registered protocols to * prevent partially registered or unregistered protocols from being run. * * The following data structures and fields are protected by this lock: * * - The netisr_proto array, including all fields of struct netisr_proto. * - The nws array, including all fields of struct netisr_worker. * - The nws_array array. * * Note: the NETISR_LOCKING define controls whether read locks are acquired * in packet processing paths requiring netisr registration stability. This * is disabled by default as it can lead to measurable performance * degradation even with rmlocks (3%-6% for loopback ping-pong traffic), and * because netisr registration and unregistration is extremely rare at * runtime. If it becomes more common, this decision should be revisited. * * XXXRW: rmlocks don't support assertions. */ static struct rmlock netisr_rmlock; #define NETISR_LOCK_INIT() rm_init_flags(&netisr_rmlock, "netisr", \ RM_NOWITNESS) #define NETISR_LOCK_ASSERT() #define NETISR_RLOCK(tracker) rm_rlock(&netisr_rmlock, (tracker)) #define NETISR_RUNLOCK(tracker) rm_runlock(&netisr_rmlock, (tracker)) #define NETISR_WLOCK() rm_wlock(&netisr_rmlock) #define NETISR_WUNLOCK() rm_wunlock(&netisr_rmlock) /* #define NETISR_LOCKING */ static SYSCTL_NODE(_net, OID_AUTO, isr, CTLFLAG_RW, 0, "netisr"); /*- * Three global direct dispatch policies are supported: * * NETISR_DISPATCH_DEFERRED: All work is deferred for a netisr, regardless of * context (may be overriden by protocols). * * NETISR_DISPATCH_HYBRID: If the executing context allows direct dispatch, * and we're running on the CPU the work would be performed on, then direct * dispatch it if it wouldn't violate ordering constraints on the workstream. * * NETISR_DISPATCH_DIRECT: If the executing context allows direct dispatch, * always direct dispatch. (The default.) * * Notice that changing the global policy could lead to short periods of * misordered processing, but this is considered acceptable as compared to * the complexity of enforcing ordering during policy changes. Protocols can * override the global policy (when they're not doing that, they select * NETISR_DISPATCH_DEFAULT). */ #define NETISR_DISPATCH_POLICY_DEFAULT NETISR_DISPATCH_DIRECT #define NETISR_DISPATCH_POLICY_MAXSTR 20 /* Used for temporary buffers. */ static u_int netisr_dispatch_policy = NETISR_DISPATCH_POLICY_DEFAULT; static int sysctl_netisr_dispatch_policy(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_isr, OID_AUTO, dispatch, CTLTYPE_STRING | CTLFLAG_RWTUN, 0, 0, sysctl_netisr_dispatch_policy, "A", "netisr dispatch policy"); /* * Allow the administrator to limit the number of threads (CPUs) to use for * netisr. We don't check netisr_maxthreads before creating the thread for * CPU 0. This must be set at boot. We will create at most one thread per CPU. * By default we initialize this to 1 which would assign just 1 cpu (cpu0) and * therefore only 1 workstream. If set to -1, netisr would use all cpus * (mp_ncpus) and therefore would have those many workstreams. One workstream * per thread (CPU). */ static int netisr_maxthreads = 1; /* Max number of threads. */ SYSCTL_INT(_net_isr, OID_AUTO, maxthreads, CTLFLAG_RDTUN, &netisr_maxthreads, 0, "Use at most this many CPUs for netisr processing"); static int netisr_bindthreads = 0; /* Bind threads to CPUs. */ SYSCTL_INT(_net_isr, OID_AUTO, bindthreads, CTLFLAG_RDTUN, &netisr_bindthreads, 0, "Bind netisr threads to CPUs."); /* * Limit per-workstream mbuf queue limits s to at most net.isr.maxqlimit, * both for initial configuration and later modification using * netisr_setqlimit(). */ #define NETISR_DEFAULT_MAXQLIMIT 10240 static u_int netisr_maxqlimit = NETISR_DEFAULT_MAXQLIMIT; SYSCTL_UINT(_net_isr, OID_AUTO, maxqlimit, CTLFLAG_RDTUN, &netisr_maxqlimit, 0, "Maximum netisr per-protocol, per-CPU queue depth."); /* * The default per-workstream mbuf queue limit for protocols that don't * initialize the nh_qlimit field of their struct netisr_handler. If this is * set above netisr_maxqlimit, we truncate it to the maximum during boot. */ #define NETISR_DEFAULT_DEFAULTQLIMIT 256 static u_int netisr_defaultqlimit = NETISR_DEFAULT_DEFAULTQLIMIT; SYSCTL_UINT(_net_isr, OID_AUTO, defaultqlimit, CTLFLAG_RDTUN, &netisr_defaultqlimit, 0, "Default netisr per-protocol, per-CPU queue limit if not set by protocol"); /* * Store and export the compile-time constant NETISR_MAXPROT limit on the * number of protocols that can register with netisr at a time. This is * required for crashdump analysis, as it sizes netisr_proto[]. */ static u_int netisr_maxprot = NETISR_MAXPROT; SYSCTL_UINT(_net_isr, OID_AUTO, maxprot, CTLFLAG_RD, &netisr_maxprot, 0, "Compile-time limit on the number of protocols supported by netisr."); /* * The netisr_proto array describes all registered protocols, indexed by * protocol number. See netisr_internal.h for more details. */ static struct netisr_proto netisr_proto[NETISR_MAXPROT]; #ifdef VIMAGE /* * The netisr_enable array describes a per-VNET flag for registered * protocols on whether this netisr is active in this VNET or not. * netisr_register() will automatically enable the netisr for the * default VNET and all currently active instances. * netisr_unregister() will disable all active VNETs, including vnet0. * Individual network stack instances can be enabled/disabled by the * netisr_(un)register _vnet() functions. * With this we keep the one netisr_proto per protocol but add a * mechanism to stop netisr processing for vnet teardown. * Apart from that we expect a VNET to always be enabled. */ -static VNET_DEFINE(u_int, netisr_enable[NETISR_MAXPROT]); +VNET_DEFINE_STATIC(u_int, netisr_enable[NETISR_MAXPROT]); #define V_netisr_enable VNET(netisr_enable) #endif /* * Per-CPU workstream data. See netisr_internal.h for more details. */ DPCPU_DEFINE(struct netisr_workstream, nws); /* * Map contiguous values between 0 and nws_count into CPU IDs appropriate for * accessing workstreams. This allows constructions of the form * DPCPU_ID_GET(nws_array[arbitraryvalue % nws_count], nws). */ static u_int nws_array[MAXCPU]; /* * Number of registered workstreams. Will be at most the number of running * CPUs once fully started. */ static u_int nws_count; SYSCTL_UINT(_net_isr, OID_AUTO, numthreads, CTLFLAG_RD, &nws_count, 0, "Number of extant netisr threads."); /* * Synchronization for each workstream: a mutex protects all mutable fields * in each stream, including per-protocol state (mbuf queues). The SWI is * woken up if asynchronous dispatch is required. */ #define NWS_LOCK(s) mtx_lock(&(s)->nws_mtx) #define NWS_LOCK_ASSERT(s) mtx_assert(&(s)->nws_mtx, MA_OWNED) #define NWS_UNLOCK(s) mtx_unlock(&(s)->nws_mtx) #define NWS_SIGNAL(s) swi_sched((s)->nws_swi_cookie, 0) /* * Utility routines for protocols that implement their own mapping of flows * to CPUs. */ u_int netisr_get_cpucount(void) { return (nws_count); } u_int netisr_get_cpuid(u_int cpunumber) { return (nws_array[cpunumber % nws_count]); } /* * The default implementation of flow -> CPU ID mapping. * * Non-static so that protocols can use it to map their own work to specific * CPUs in a manner consistent to netisr for affinity purposes. */ u_int netisr_default_flow2cpu(u_int flowid) { return (nws_array[flowid % nws_count]); } /* * Dispatch tunable and sysctl configuration. */ struct netisr_dispatch_table_entry { u_int ndte_policy; const char *ndte_policy_str; }; static const struct netisr_dispatch_table_entry netisr_dispatch_table[] = { { NETISR_DISPATCH_DEFAULT, "default" }, { NETISR_DISPATCH_DEFERRED, "deferred" }, { NETISR_DISPATCH_HYBRID, "hybrid" }, { NETISR_DISPATCH_DIRECT, "direct" }, }; static void netisr_dispatch_policy_to_str(u_int dispatch_policy, char *buffer, u_int buflen) { const struct netisr_dispatch_table_entry *ndtep; const char *str; u_int i; str = "unknown"; for (i = 0; i < nitems(netisr_dispatch_table); i++) { ndtep = &netisr_dispatch_table[i]; if (ndtep->ndte_policy == dispatch_policy) { str = ndtep->ndte_policy_str; break; } } snprintf(buffer, buflen, "%s", str); } static int netisr_dispatch_policy_from_str(const char *str, u_int *dispatch_policyp) { const struct netisr_dispatch_table_entry *ndtep; u_int i; for (i = 0; i < nitems(netisr_dispatch_table); i++) { ndtep = &netisr_dispatch_table[i]; if (strcmp(ndtep->ndte_policy_str, str) == 0) { *dispatch_policyp = ndtep->ndte_policy; return (0); } } return (EINVAL); } static int sysctl_netisr_dispatch_policy(SYSCTL_HANDLER_ARGS) { char tmp[NETISR_DISPATCH_POLICY_MAXSTR]; u_int dispatch_policy; int error; netisr_dispatch_policy_to_str(netisr_dispatch_policy, tmp, sizeof(tmp)); error = sysctl_handle_string(oidp, tmp, sizeof(tmp), req); if (error == 0 && req->newptr != NULL) { error = netisr_dispatch_policy_from_str(tmp, &dispatch_policy); if (error == 0 && dispatch_policy == NETISR_DISPATCH_DEFAULT) error = EINVAL; if (error == 0) netisr_dispatch_policy = dispatch_policy; } return (error); } /* * Register a new netisr handler, which requires initializing per-protocol * fields for each workstream. All netisr work is briefly suspended while * the protocol is installed. */ void netisr_register(const struct netisr_handler *nhp) { VNET_ITERATOR_DECL(vnet_iter); struct netisr_work *npwp; const char *name; u_int i, proto; proto = nhp->nh_proto; name = nhp->nh_name; /* * Test that the requested registration is valid. */ KASSERT(nhp->nh_name != NULL, ("%s: nh_name NULL for %u", __func__, proto)); KASSERT(nhp->nh_handler != NULL, ("%s: nh_handler NULL for %s", __func__, name)); KASSERT(nhp->nh_policy == NETISR_POLICY_SOURCE || nhp->nh_policy == NETISR_POLICY_FLOW || nhp->nh_policy == NETISR_POLICY_CPU, ("%s: unsupported nh_policy %u for %s", __func__, nhp->nh_policy, name)); KASSERT(nhp->nh_policy == NETISR_POLICY_FLOW || nhp->nh_m2flow == NULL, ("%s: nh_policy != FLOW but m2flow defined for %s", __func__, name)); KASSERT(nhp->nh_policy == NETISR_POLICY_CPU || nhp->nh_m2cpuid == NULL, ("%s: nh_policy != CPU but m2cpuid defined for %s", __func__, name)); KASSERT(nhp->nh_policy != NETISR_POLICY_CPU || nhp->nh_m2cpuid != NULL, ("%s: nh_policy == CPU but m2cpuid not defined for %s", __func__, name)); KASSERT(nhp->nh_dispatch == NETISR_DISPATCH_DEFAULT || nhp->nh_dispatch == NETISR_DISPATCH_DEFERRED || nhp->nh_dispatch == NETISR_DISPATCH_HYBRID || nhp->nh_dispatch == NETISR_DISPATCH_DIRECT, ("%s: invalid nh_dispatch (%u)", __func__, nhp->nh_dispatch)); KASSERT(proto < NETISR_MAXPROT, ("%s(%u, %s): protocol too big", __func__, proto, name)); /* * Test that no existing registration exists for this protocol. */ NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_name == NULL, ("%s(%u, %s): name present", __func__, proto, name)); KASSERT(netisr_proto[proto].np_handler == NULL, ("%s(%u, %s): handler present", __func__, proto, name)); netisr_proto[proto].np_name = name; netisr_proto[proto].np_handler = nhp->nh_handler; netisr_proto[proto].np_m2flow = nhp->nh_m2flow; netisr_proto[proto].np_m2cpuid = nhp->nh_m2cpuid; netisr_proto[proto].np_drainedcpu = nhp->nh_drainedcpu; if (nhp->nh_qlimit == 0) netisr_proto[proto].np_qlimit = netisr_defaultqlimit; else if (nhp->nh_qlimit > netisr_maxqlimit) { printf("%s: %s requested queue limit %u capped to " "net.isr.maxqlimit %u\n", __func__, name, nhp->nh_qlimit, netisr_maxqlimit); netisr_proto[proto].np_qlimit = netisr_maxqlimit; } else netisr_proto[proto].np_qlimit = nhp->nh_qlimit; netisr_proto[proto].np_policy = nhp->nh_policy; netisr_proto[proto].np_dispatch = nhp->nh_dispatch; CPU_FOREACH(i) { npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto]; bzero(npwp, sizeof(*npwp)); npwp->nw_qlimit = netisr_proto[proto].np_qlimit; } #ifdef VIMAGE /* * Test that we are in vnet0 and have a curvnet set. */ KASSERT(curvnet != NULL, ("%s: curvnet is NULL", __func__)); KASSERT(IS_DEFAULT_VNET(curvnet), ("%s: curvnet %p is not vnet0 %p", __func__, curvnet, vnet0)); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); V_netisr_enable[proto] = 1; CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); #endif NETISR_WUNLOCK(); } /* * Clear drop counters across all workstreams for a protocol. */ void netisr_clearqdrops(const struct netisr_handler *nhp) { struct netisr_work *npwp; #ifdef INVARIANTS const char *name; #endif u_int i, proto; proto = nhp->nh_proto; #ifdef INVARIANTS name = nhp->nh_name; #endif KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, name)); NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, name)); CPU_FOREACH(i) { npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto]; npwp->nw_qdrops = 0; } NETISR_WUNLOCK(); } /* * Query current drop counters across all workstreams for a protocol. */ void netisr_getqdrops(const struct netisr_handler *nhp, u_int64_t *qdropp) { struct netisr_work *npwp; struct rm_priotracker tracker; #ifdef INVARIANTS const char *name; #endif u_int i, proto; *qdropp = 0; proto = nhp->nh_proto; #ifdef INVARIANTS name = nhp->nh_name; #endif KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, name)); NETISR_RLOCK(&tracker); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, name)); CPU_FOREACH(i) { npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto]; *qdropp += npwp->nw_qdrops; } NETISR_RUNLOCK(&tracker); } /* * Query current per-workstream queue limit for a protocol. */ void netisr_getqlimit(const struct netisr_handler *nhp, u_int *qlimitp) { struct rm_priotracker tracker; #ifdef INVARIANTS const char *name; #endif u_int proto; proto = nhp->nh_proto; #ifdef INVARIANTS name = nhp->nh_name; #endif KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, name)); NETISR_RLOCK(&tracker); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, name)); *qlimitp = netisr_proto[proto].np_qlimit; NETISR_RUNLOCK(&tracker); } /* * Update the queue limit across per-workstream queues for a protocol. We * simply change the limits, and don't drain overflowed packets as they will * (hopefully) take care of themselves shortly. */ int netisr_setqlimit(const struct netisr_handler *nhp, u_int qlimit) { struct netisr_work *npwp; #ifdef INVARIANTS const char *name; #endif u_int i, proto; if (qlimit > netisr_maxqlimit) return (EINVAL); proto = nhp->nh_proto; #ifdef INVARIANTS name = nhp->nh_name; #endif KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, name)); NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, name)); netisr_proto[proto].np_qlimit = qlimit; CPU_FOREACH(i) { npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto]; npwp->nw_qlimit = qlimit; } NETISR_WUNLOCK(); return (0); } /* * Drain all packets currently held in a particular protocol work queue. */ static void netisr_drain_proto(struct netisr_work *npwp) { struct mbuf *m; /* * We would assert the lock on the workstream but it's not passed in. */ while ((m = npwp->nw_head) != NULL) { npwp->nw_head = m->m_nextpkt; m->m_nextpkt = NULL; if (npwp->nw_head == NULL) npwp->nw_tail = NULL; npwp->nw_len--; m_freem(m); } KASSERT(npwp->nw_tail == NULL, ("%s: tail", __func__)); KASSERT(npwp->nw_len == 0, ("%s: len", __func__)); } /* * Remove the registration of a network protocol, which requires clearing * per-protocol fields across all workstreams, including freeing all mbufs in * the queues at time of unregister. All work in netisr is briefly suspended * while this takes place. */ void netisr_unregister(const struct netisr_handler *nhp) { VNET_ITERATOR_DECL(vnet_iter); struct netisr_work *npwp; #ifdef INVARIANTS const char *name; #endif u_int i, proto; proto = nhp->nh_proto; #ifdef INVARIANTS name = nhp->nh_name; #endif KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, name)); NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, name)); #ifdef VIMAGE VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); V_netisr_enable[proto] = 0; CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); #endif netisr_proto[proto].np_name = NULL; netisr_proto[proto].np_handler = NULL; netisr_proto[proto].np_m2flow = NULL; netisr_proto[proto].np_m2cpuid = NULL; netisr_proto[proto].np_qlimit = 0; netisr_proto[proto].np_policy = 0; CPU_FOREACH(i) { npwp = &(DPCPU_ID_PTR(i, nws))->nws_work[proto]; netisr_drain_proto(npwp); bzero(npwp, sizeof(*npwp)); } NETISR_WUNLOCK(); } #ifdef VIMAGE void netisr_register_vnet(const struct netisr_handler *nhp) { u_int proto; proto = nhp->nh_proto; KASSERT(curvnet != NULL, ("%s: curvnet is NULL", __func__)); KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, nhp->nh_name)); NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, nhp->nh_name)); V_netisr_enable[proto] = 1; NETISR_WUNLOCK(); } static void netisr_drain_proto_vnet(struct vnet *vnet, u_int proto) { struct netisr_workstream *nwsp; struct netisr_work *npwp; struct mbuf *m, *mp, *n, *ne; u_int i; KASSERT(vnet != NULL, ("%s: vnet is NULL", __func__)); NETISR_LOCK_ASSERT(); CPU_FOREACH(i) { nwsp = DPCPU_ID_PTR(i, nws); if (nwsp->nws_intr_event == NULL) continue; npwp = &nwsp->nws_work[proto]; NWS_LOCK(nwsp); /* * Rather than dissecting and removing mbufs from the middle * of the chain, we build a new chain if the packet stays and * update the head and tail pointers at the end. All packets * matching the given vnet are freed. */ m = npwp->nw_head; n = ne = NULL; while (m != NULL) { mp = m; m = m->m_nextpkt; mp->m_nextpkt = NULL; if (mp->m_pkthdr.rcvif->if_vnet != vnet) { if (n == NULL) { n = ne = mp; } else { ne->m_nextpkt = mp; ne = mp; } continue; } /* This is a packet in the selected vnet. Free it. */ npwp->nw_len--; m_freem(mp); } npwp->nw_head = n; npwp->nw_tail = ne; NWS_UNLOCK(nwsp); } } void netisr_unregister_vnet(const struct netisr_handler *nhp) { u_int proto; proto = nhp->nh_proto; KASSERT(curvnet != NULL, ("%s: curvnet is NULL", __func__)); KASSERT(proto < NETISR_MAXPROT, ("%s(%u): protocol too big for %s", __func__, proto, nhp->nh_name)); NETISR_WLOCK(); KASSERT(netisr_proto[proto].np_handler != NULL, ("%s(%u): protocol not registered for %s", __func__, proto, nhp->nh_name)); V_netisr_enable[proto] = 0; netisr_drain_proto_vnet(curvnet, proto); NETISR_WUNLOCK(); } #endif /* * Compose the global and per-protocol policies on dispatch, and return the * dispatch policy to use. */ static u_int netisr_get_dispatch(struct netisr_proto *npp) { /* * Protocol-specific configuration overrides the global default. */ if (npp->np_dispatch != NETISR_DISPATCH_DEFAULT) return (npp->np_dispatch); return (netisr_dispatch_policy); } /* * Look up the workstream given a packet and source identifier. Do this by * checking the protocol's policy, and optionally call out to the protocol * for assistance if required. */ static struct mbuf * netisr_select_cpuid(struct netisr_proto *npp, u_int dispatch_policy, uintptr_t source, struct mbuf *m, u_int *cpuidp) { struct ifnet *ifp; u_int policy; NETISR_LOCK_ASSERT(); /* * In the event we have only one worker, shortcut and deliver to it * without further ado. */ if (nws_count == 1) { *cpuidp = nws_array[0]; return (m); } /* * What happens next depends on the policy selected by the protocol. * If we want to support per-interface policies, we should do that * here first. */ policy = npp->np_policy; if (policy == NETISR_POLICY_CPU) { m = npp->np_m2cpuid(m, source, cpuidp); if (m == NULL) return (NULL); /* * It's possible for a protocol not to have a good idea about * where to process a packet, in which case we fall back on * the netisr code to decide. In the hybrid case, return the * current CPU ID, which will force an immediate direct * dispatch. In the queued case, fall back on the SOURCE * policy. */ if (*cpuidp != NETISR_CPUID_NONE) { *cpuidp = netisr_get_cpuid(*cpuidp); return (m); } if (dispatch_policy == NETISR_DISPATCH_HYBRID) { *cpuidp = netisr_get_cpuid(curcpu); return (m); } policy = NETISR_POLICY_SOURCE; } if (policy == NETISR_POLICY_FLOW) { if (M_HASHTYPE_GET(m) == M_HASHTYPE_NONE && npp->np_m2flow != NULL) { m = npp->np_m2flow(m, source); if (m == NULL) return (NULL); } if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { *cpuidp = netisr_default_flow2cpu(m->m_pkthdr.flowid); return (m); } policy = NETISR_POLICY_SOURCE; } KASSERT(policy == NETISR_POLICY_SOURCE, ("%s: invalid policy %u for %s", __func__, npp->np_policy, npp->np_name)); ifp = m->m_pkthdr.rcvif; if (ifp != NULL) *cpuidp = nws_array[(ifp->if_index + source) % nws_count]; else *cpuidp = nws_array[source % nws_count]; return (m); } /* * Process packets associated with a workstream and protocol. For reasons of * fairness, we process up to one complete netisr queue at a time, moving the * queue to a stack-local queue for processing, but do not loop refreshing * from the global queue. The caller is responsible for deciding whether to * loop, and for setting the NWS_RUNNING flag. The passed workstream will be * locked on entry and relocked before return, but will be released while * processing. The number of packets processed is returned. */ static u_int netisr_process_workstream_proto(struct netisr_workstream *nwsp, u_int proto) { struct netisr_work local_npw, *npwp; u_int handled; struct mbuf *m; NETISR_LOCK_ASSERT(); NWS_LOCK_ASSERT(nwsp); KASSERT(nwsp->nws_flags & NWS_RUNNING, ("%s(%u): not running", __func__, proto)); KASSERT(proto >= 0 && proto < NETISR_MAXPROT, ("%s(%u): invalid proto\n", __func__, proto)); npwp = &nwsp->nws_work[proto]; if (npwp->nw_len == 0) return (0); /* * Move the global work queue to a thread-local work queue. * * Notice that this means the effective maximum length of the queue * is actually twice that of the maximum queue length specified in * the protocol registration call. */ handled = npwp->nw_len; local_npw = *npwp; npwp->nw_head = NULL; npwp->nw_tail = NULL; npwp->nw_len = 0; nwsp->nws_pendingbits &= ~(1 << proto); NWS_UNLOCK(nwsp); while ((m = local_npw.nw_head) != NULL) { local_npw.nw_head = m->m_nextpkt; m->m_nextpkt = NULL; if (local_npw.nw_head == NULL) local_npw.nw_tail = NULL; local_npw.nw_len--; VNET_ASSERT(m->m_pkthdr.rcvif != NULL, ("%s:%d rcvif == NULL: m=%p", __func__, __LINE__, m)); CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); netisr_proto[proto].np_handler(m); CURVNET_RESTORE(); } KASSERT(local_npw.nw_len == 0, ("%s(%u): len %u", __func__, proto, local_npw.nw_len)); if (netisr_proto[proto].np_drainedcpu) netisr_proto[proto].np_drainedcpu(nwsp->nws_cpu); NWS_LOCK(nwsp); npwp->nw_handled += handled; return (handled); } /* * SWI handler for netisr -- processes packets in a set of workstreams that * it owns, woken up by calls to NWS_SIGNAL(). If this workstream is already * being direct dispatched, go back to sleep and wait for the dispatching * thread to wake us up again. */ static void swi_net(void *arg) { #ifdef NETISR_LOCKING struct rm_priotracker tracker; #endif struct netisr_workstream *nwsp; u_int bits, prot; nwsp = arg; #ifdef DEVICE_POLLING KASSERT(nws_count == 1, ("%s: device_polling but nws_count != 1", __func__)); netisr_poll(); #endif #ifdef NETISR_LOCKING NETISR_RLOCK(&tracker); #endif NWS_LOCK(nwsp); KASSERT(!(nwsp->nws_flags & NWS_RUNNING), ("swi_net: running")); if (nwsp->nws_flags & NWS_DISPATCHING) goto out; nwsp->nws_flags |= NWS_RUNNING; nwsp->nws_flags &= ~NWS_SCHEDULED; while ((bits = nwsp->nws_pendingbits) != 0) { while ((prot = ffs(bits)) != 0) { prot--; bits &= ~(1 << prot); (void)netisr_process_workstream_proto(nwsp, prot); } } nwsp->nws_flags &= ~NWS_RUNNING; out: NWS_UNLOCK(nwsp); #ifdef NETISR_LOCKING NETISR_RUNLOCK(&tracker); #endif #ifdef DEVICE_POLLING netisr_pollmore(); #endif } static int netisr_queue_workstream(struct netisr_workstream *nwsp, u_int proto, struct netisr_work *npwp, struct mbuf *m, int *dosignalp) { NWS_LOCK_ASSERT(nwsp); *dosignalp = 0; if (npwp->nw_len < npwp->nw_qlimit) { m->m_nextpkt = NULL; if (npwp->nw_head == NULL) { npwp->nw_head = m; npwp->nw_tail = m; } else { npwp->nw_tail->m_nextpkt = m; npwp->nw_tail = m; } npwp->nw_len++; if (npwp->nw_len > npwp->nw_watermark) npwp->nw_watermark = npwp->nw_len; /* * We must set the bit regardless of NWS_RUNNING, so that * swi_net() keeps calling netisr_process_workstream_proto(). */ nwsp->nws_pendingbits |= (1 << proto); if (!(nwsp->nws_flags & (NWS_RUNNING | NWS_DISPATCHING | NWS_SCHEDULED))) { nwsp->nws_flags |= NWS_SCHEDULED; *dosignalp = 1; /* Defer until unlocked. */ } npwp->nw_queued++; return (0); } else { m_freem(m); npwp->nw_qdrops++; return (ENOBUFS); } } static int netisr_queue_internal(u_int proto, struct mbuf *m, u_int cpuid) { struct netisr_workstream *nwsp; struct netisr_work *npwp; int dosignal, error; #ifdef NETISR_LOCKING NETISR_LOCK_ASSERT(); #endif KASSERT(cpuid <= mp_maxid, ("%s: cpuid too big (%u, %u)", __func__, cpuid, mp_maxid)); KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid)); dosignal = 0; error = 0; nwsp = DPCPU_ID_PTR(cpuid, nws); npwp = &nwsp->nws_work[proto]; NWS_LOCK(nwsp); error = netisr_queue_workstream(nwsp, proto, npwp, m, &dosignal); NWS_UNLOCK(nwsp); if (dosignal) NWS_SIGNAL(nwsp); return (error); } int netisr_queue_src(u_int proto, uintptr_t source, struct mbuf *m) { #ifdef NETISR_LOCKING struct rm_priotracker tracker; #endif u_int cpuid; int error; KASSERT(proto < NETISR_MAXPROT, ("%s: invalid proto %u", __func__, proto)); #ifdef NETISR_LOCKING NETISR_RLOCK(&tracker); #endif KASSERT(netisr_proto[proto].np_handler != NULL, ("%s: invalid proto %u", __func__, proto)); #ifdef VIMAGE if (V_netisr_enable[proto] == 0) { m_freem(m); return (ENOPROTOOPT); } #endif m = netisr_select_cpuid(&netisr_proto[proto], NETISR_DISPATCH_DEFERRED, source, m, &cpuid); if (m != NULL) { KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid)); error = netisr_queue_internal(proto, m, cpuid); } else error = ENOBUFS; #ifdef NETISR_LOCKING NETISR_RUNLOCK(&tracker); #endif return (error); } int netisr_queue(u_int proto, struct mbuf *m) { return (netisr_queue_src(proto, 0, m)); } /* * Dispatch a packet for netisr processing; direct dispatch is permitted by * calling context. */ int netisr_dispatch_src(u_int proto, uintptr_t source, struct mbuf *m) { #ifdef NETISR_LOCKING struct rm_priotracker tracker; #endif struct netisr_workstream *nwsp; struct netisr_proto *npp; struct netisr_work *npwp; int dosignal, error; u_int cpuid, dispatch_policy; KASSERT(proto < NETISR_MAXPROT, ("%s: invalid proto %u", __func__, proto)); #ifdef NETISR_LOCKING NETISR_RLOCK(&tracker); #endif npp = &netisr_proto[proto]; KASSERT(npp->np_handler != NULL, ("%s: invalid proto %u", __func__, proto)); #ifdef VIMAGE if (V_netisr_enable[proto] == 0) { m_freem(m); return (ENOPROTOOPT); } #endif dispatch_policy = netisr_get_dispatch(npp); if (dispatch_policy == NETISR_DISPATCH_DEFERRED) return (netisr_queue_src(proto, source, m)); /* * If direct dispatch is forced, then unconditionally dispatch * without a formal CPU selection. Borrow the current CPU's stats, * even if there's no worker on it. In this case we don't update * nws_flags because all netisr processing will be source ordered due * to always being forced to directly dispatch. */ if (dispatch_policy == NETISR_DISPATCH_DIRECT) { nwsp = DPCPU_PTR(nws); npwp = &nwsp->nws_work[proto]; npwp->nw_dispatched++; npwp->nw_handled++; netisr_proto[proto].np_handler(m); error = 0; goto out_unlock; } KASSERT(dispatch_policy == NETISR_DISPATCH_HYBRID, ("%s: unknown dispatch policy (%u)", __func__, dispatch_policy)); /* * Otherwise, we execute in a hybrid mode where we will try to direct * dispatch if we're on the right CPU and the netisr worker isn't * already running. */ sched_pin(); m = netisr_select_cpuid(&netisr_proto[proto], NETISR_DISPATCH_HYBRID, source, m, &cpuid); if (m == NULL) { error = ENOBUFS; goto out_unpin; } KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid)); if (cpuid != curcpu) goto queue_fallback; nwsp = DPCPU_PTR(nws); npwp = &nwsp->nws_work[proto]; /*- * We are willing to direct dispatch only if three conditions hold: * * (1) The netisr worker isn't already running, * (2) Another thread isn't already directly dispatching, and * (3) The netisr hasn't already been woken up. */ NWS_LOCK(nwsp); if (nwsp->nws_flags & (NWS_RUNNING | NWS_DISPATCHING | NWS_SCHEDULED)) { error = netisr_queue_workstream(nwsp, proto, npwp, m, &dosignal); NWS_UNLOCK(nwsp); if (dosignal) NWS_SIGNAL(nwsp); goto out_unpin; } /* * The current thread is now effectively the netisr worker, so set * the dispatching flag to prevent concurrent processing of the * stream from another thread (even the netisr worker), which could * otherwise lead to effective misordering of the stream. */ nwsp->nws_flags |= NWS_DISPATCHING; NWS_UNLOCK(nwsp); netisr_proto[proto].np_handler(m); NWS_LOCK(nwsp); nwsp->nws_flags &= ~NWS_DISPATCHING; npwp->nw_handled++; npwp->nw_hybrid_dispatched++; /* * If other work was enqueued by another thread while we were direct * dispatching, we need to signal the netisr worker to do that work. * In the future, we might want to do some of that work in the * current thread, rather than trigger further context switches. If * so, we'll want to establish a reasonable bound on the work done in * the "borrowed" context. */ if (nwsp->nws_pendingbits != 0) { nwsp->nws_flags |= NWS_SCHEDULED; dosignal = 1; } else dosignal = 0; NWS_UNLOCK(nwsp); if (dosignal) NWS_SIGNAL(nwsp); error = 0; goto out_unpin; queue_fallback: error = netisr_queue_internal(proto, m, cpuid); out_unpin: sched_unpin(); out_unlock: #ifdef NETISR_LOCKING NETISR_RUNLOCK(&tracker); #endif return (error); } int netisr_dispatch(u_int proto, struct mbuf *m) { return (netisr_dispatch_src(proto, 0, m)); } #ifdef DEVICE_POLLING /* * Kernel polling borrows a netisr thread to run interface polling in; this * function allows kernel polling to request that the netisr thread be * scheduled even if no packets are pending for protocols. */ void netisr_sched_poll(void) { struct netisr_workstream *nwsp; nwsp = DPCPU_ID_PTR(nws_array[0], nws); NWS_SIGNAL(nwsp); } #endif static void netisr_start_swi(u_int cpuid, struct pcpu *pc) { char swiname[12]; struct netisr_workstream *nwsp; int error; KASSERT(!CPU_ABSENT(cpuid), ("%s: CPU %u absent", __func__, cpuid)); nwsp = DPCPU_ID_PTR(cpuid, nws); mtx_init(&nwsp->nws_mtx, "netisr_mtx", NULL, MTX_DEF); nwsp->nws_cpu = cpuid; snprintf(swiname, sizeof(swiname), "netisr %u", cpuid); error = swi_add(&nwsp->nws_intr_event, swiname, swi_net, nwsp, SWI_NET, INTR_MPSAFE, &nwsp->nws_swi_cookie); if (error) panic("%s: swi_add %d", __func__, error); pc->pc_netisr = nwsp->nws_intr_event; if (netisr_bindthreads) { error = intr_event_bind(nwsp->nws_intr_event, cpuid); if (error != 0) printf("%s: cpu %u: intr_event_bind: %d", __func__, cpuid, error); } NETISR_WLOCK(); nws_array[nws_count] = nwsp->nws_cpu; nws_count++; NETISR_WUNLOCK(); } /* * Initialize the netisr subsystem. We rely on BSS and static initialization * of most fields in global data structures. * * Start a worker thread for the boot CPU so that we can support network * traffic immediately in case the network stack is used before additional * CPUs are started (for example, diskless boot). */ static void netisr_init(void *arg) { struct pcpu *pc; NETISR_LOCK_INIT(); if (netisr_maxthreads == 0 || netisr_maxthreads < -1 ) netisr_maxthreads = 1; /* default behavior */ else if (netisr_maxthreads == -1) netisr_maxthreads = mp_ncpus; /* use max cpus */ if (netisr_maxthreads > mp_ncpus) { printf("netisr_init: forcing maxthreads from %d to %d\n", netisr_maxthreads, mp_ncpus); netisr_maxthreads = mp_ncpus; } if (netisr_defaultqlimit > netisr_maxqlimit) { printf("netisr_init: forcing defaultqlimit from %d to %d\n", netisr_defaultqlimit, netisr_maxqlimit); netisr_defaultqlimit = netisr_maxqlimit; } #ifdef DEVICE_POLLING /* * The device polling code is not yet aware of how to deal with * multiple netisr threads, so for the time being compiling in device * polling disables parallel netisr workers. */ if (netisr_maxthreads != 1 || netisr_bindthreads != 0) { printf("netisr_init: forcing maxthreads to 1 and " "bindthreads to 0 for device polling\n"); netisr_maxthreads = 1; netisr_bindthreads = 0; } #endif #ifdef EARLY_AP_STARTUP STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { if (nws_count >= netisr_maxthreads) break; netisr_start_swi(pc->pc_cpuid, pc); } #else pc = get_pcpu(); netisr_start_swi(pc->pc_cpuid, pc); #endif } SYSINIT(netisr_init, SI_SUB_SOFTINTR, SI_ORDER_FIRST, netisr_init, NULL); #ifndef EARLY_AP_STARTUP /* * Start worker threads for additional CPUs. No attempt to gracefully handle * work reassignment, we don't yet support dynamic reconfiguration. */ static void netisr_start(void *arg) { struct pcpu *pc; STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { if (nws_count >= netisr_maxthreads) break; /* Worker will already be present for boot CPU. */ if (pc->pc_netisr != NULL) continue; netisr_start_swi(pc->pc_cpuid, pc); } } SYSINIT(netisr_start, SI_SUB_SMP, SI_ORDER_MIDDLE, netisr_start, NULL); #endif /* * Sysctl monitoring for netisr: query a list of registered protocols. */ static int sysctl_netisr_proto(SYSCTL_HANDLER_ARGS) { struct rm_priotracker tracker; struct sysctl_netisr_proto *snpp, *snp_array; struct netisr_proto *npp; u_int counter, proto; int error; if (req->newptr != NULL) return (EINVAL); snp_array = malloc(sizeof(*snp_array) * NETISR_MAXPROT, M_TEMP, M_ZERO | M_WAITOK); counter = 0; NETISR_RLOCK(&tracker); for (proto = 0; proto < NETISR_MAXPROT; proto++) { npp = &netisr_proto[proto]; if (npp->np_name == NULL) continue; snpp = &snp_array[counter]; snpp->snp_version = sizeof(*snpp); strlcpy(snpp->snp_name, npp->np_name, NETISR_NAMEMAXLEN); snpp->snp_proto = proto; snpp->snp_qlimit = npp->np_qlimit; snpp->snp_policy = npp->np_policy; snpp->snp_dispatch = npp->np_dispatch; if (npp->np_m2flow != NULL) snpp->snp_flags |= NETISR_SNP_FLAGS_M2FLOW; if (npp->np_m2cpuid != NULL) snpp->snp_flags |= NETISR_SNP_FLAGS_M2CPUID; if (npp->np_drainedcpu != NULL) snpp->snp_flags |= NETISR_SNP_FLAGS_DRAINEDCPU; counter++; } NETISR_RUNLOCK(&tracker); KASSERT(counter <= NETISR_MAXPROT, ("sysctl_netisr_proto: counter too big (%d)", counter)); error = SYSCTL_OUT(req, snp_array, sizeof(*snp_array) * counter); free(snp_array, M_TEMP); return (error); } SYSCTL_PROC(_net_isr, OID_AUTO, proto, CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_proto, "S,sysctl_netisr_proto", "Return list of protocols registered with netisr"); /* * Sysctl monitoring for netisr: query a list of workstreams. */ static int sysctl_netisr_workstream(SYSCTL_HANDLER_ARGS) { struct rm_priotracker tracker; struct sysctl_netisr_workstream *snwsp, *snws_array; struct netisr_workstream *nwsp; u_int counter, cpuid; int error; if (req->newptr != NULL) return (EINVAL); snws_array = malloc(sizeof(*snws_array) * MAXCPU, M_TEMP, M_ZERO | M_WAITOK); counter = 0; NETISR_RLOCK(&tracker); CPU_FOREACH(cpuid) { nwsp = DPCPU_ID_PTR(cpuid, nws); if (nwsp->nws_intr_event == NULL) continue; NWS_LOCK(nwsp); snwsp = &snws_array[counter]; snwsp->snws_version = sizeof(*snwsp); /* * For now, we equate workstream IDs and CPU IDs in the * kernel, but expose them independently to userspace in case * that assumption changes in the future. */ snwsp->snws_wsid = cpuid; snwsp->snws_cpu = cpuid; if (nwsp->nws_intr_event != NULL) snwsp->snws_flags |= NETISR_SNWS_FLAGS_INTR; NWS_UNLOCK(nwsp); counter++; } NETISR_RUNLOCK(&tracker); KASSERT(counter <= MAXCPU, ("sysctl_netisr_workstream: counter too big (%d)", counter)); error = SYSCTL_OUT(req, snws_array, sizeof(*snws_array) * counter); free(snws_array, M_TEMP); return (error); } SYSCTL_PROC(_net_isr, OID_AUTO, workstream, CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_workstream, "S,sysctl_netisr_workstream", "Return list of workstreams implemented by netisr"); /* * Sysctl monitoring for netisr: query per-protocol data across all * workstreams. */ static int sysctl_netisr_work(SYSCTL_HANDLER_ARGS) { struct rm_priotracker tracker; struct sysctl_netisr_work *snwp, *snw_array; struct netisr_workstream *nwsp; struct netisr_proto *npp; struct netisr_work *nwp; u_int counter, cpuid, proto; int error; if (req->newptr != NULL) return (EINVAL); snw_array = malloc(sizeof(*snw_array) * MAXCPU * NETISR_MAXPROT, M_TEMP, M_ZERO | M_WAITOK); counter = 0; NETISR_RLOCK(&tracker); CPU_FOREACH(cpuid) { nwsp = DPCPU_ID_PTR(cpuid, nws); if (nwsp->nws_intr_event == NULL) continue; NWS_LOCK(nwsp); for (proto = 0; proto < NETISR_MAXPROT; proto++) { npp = &netisr_proto[proto]; if (npp->np_name == NULL) continue; nwp = &nwsp->nws_work[proto]; snwp = &snw_array[counter]; snwp->snw_version = sizeof(*snwp); snwp->snw_wsid = cpuid; /* See comment above. */ snwp->snw_proto = proto; snwp->snw_len = nwp->nw_len; snwp->snw_watermark = nwp->nw_watermark; snwp->snw_dispatched = nwp->nw_dispatched; snwp->snw_hybrid_dispatched = nwp->nw_hybrid_dispatched; snwp->snw_qdrops = nwp->nw_qdrops; snwp->snw_queued = nwp->nw_queued; snwp->snw_handled = nwp->nw_handled; counter++; } NWS_UNLOCK(nwsp); } KASSERT(counter <= MAXCPU * NETISR_MAXPROT, ("sysctl_netisr_work: counter too big (%d)", counter)); NETISR_RUNLOCK(&tracker); error = SYSCTL_OUT(req, snw_array, sizeof(*snw_array) * counter); free(snw_array, M_TEMP); return (error); } SYSCTL_PROC(_net_isr, OID_AUTO, work, CTLFLAG_RD|CTLTYPE_STRUCT|CTLFLAG_MPSAFE, 0, 0, sysctl_netisr_work, "S,sysctl_netisr_work", "Return list of per-workstream, per-protocol work in netisr"); #ifdef DDB DB_SHOW_COMMAND(netisr, db_show_netisr) { struct netisr_workstream *nwsp; struct netisr_work *nwp; int first, proto; u_int cpuid; db_printf("%3s %6s %5s %5s %5s %8s %8s %8s %8s\n", "CPU", "Proto", "Len", "WMark", "Max", "Disp", "HDisp", "Drop", "Queue"); CPU_FOREACH(cpuid) { nwsp = DPCPU_ID_PTR(cpuid, nws); if (nwsp->nws_intr_event == NULL) continue; first = 1; for (proto = 0; proto < NETISR_MAXPROT; proto++) { if (netisr_proto[proto].np_handler == NULL) continue; nwp = &nwsp->nws_work[proto]; if (first) { db_printf("%3d ", cpuid); first = 0; } else db_printf("%3s ", ""); db_printf( "%6s %5d %5d %5d %8ju %8ju %8ju %8ju\n", netisr_proto[proto].np_name, nwp->nw_len, nwp->nw_watermark, nwp->nw_qlimit, nwp->nw_dispatched, nwp->nw_hybrid_dispatched, nwp->nw_qdrops, nwp->nw_queued); } } } #endif Index: head/sys/net/route.c =================================================================== --- head/sys/net/route.c (revision 336675) +++ head/sys/net/route.c (revision 336676) @@ -1,2266 +1,2266 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1980, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)route.c 8.3.1.1 (Berkeley) 2/23/95 * $FreeBSD$ */ /************************************************************************ * Note: In this file a 'fib' is a "forwarding information base" * * Which is the new name for an in kernel routing (next hop) table. * ***********************************************************************/ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_route.h" #include "opt_sctp.h" #include "opt_mrouting.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include #include #include #define RT_MAXFIBS UINT16_MAX /* Kernel config default option. */ #ifdef ROUTETABLES #if ROUTETABLES <= 0 #error "ROUTETABLES defined too low" #endif #if ROUTETABLES > RT_MAXFIBS #error "ROUTETABLES defined too big" #endif #define RT_NUMFIBS ROUTETABLES #endif /* ROUTETABLES */ /* Initialize to default if not otherwise set. */ #ifndef RT_NUMFIBS #define RT_NUMFIBS 1 #endif #if defined(INET) || defined(INET6) #ifdef SCTP extern void sctp_addr_change(struct ifaddr *ifa, int cmd); #endif /* SCTP */ #endif /* This is read-only.. */ u_int rt_numfibs = RT_NUMFIBS; SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, ""); /* * By default add routes to all fibs for new interfaces. * Once this is set to 0 then only allocate routes on interface * changes for the FIB of the caller when adding a new set of addresses * to an interface. XXX this is a shotgun aproach to a problem that needs * a more fine grained solution.. that will come. * XXX also has the problems getting the FIB from curthread which will not * always work given the fib can be overridden and prefixes can be added * from the network stack context. */ VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1; SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET, &VNET_NAME(rt_add_addr_allfibs), 0, ""); VNET_DEFINE(struct rtstat, rtstat); #define V_rtstat VNET(rtstat) VNET_DEFINE(struct rib_head *, rt_tables); #define V_rt_tables VNET(rt_tables) VNET_DEFINE(int, rttrash); /* routes not in table but not freed */ #define V_rttrash VNET(rttrash) /* * Convert a 'struct radix_node *' to a 'struct rtentry *'. * The operation can be done safely (in this code) because a * 'struct rtentry' starts with two 'struct radix_node''s, the first * one representing leaf nodes in the routing tree, which is * what the code in radix.c passes us as a 'struct radix_node'. * * But because there are a lot of assumptions in this conversion, * do not cast explicitly, but always use the macro below. */ #define RNTORT(p) ((struct rtentry *)(p)) -static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */ +VNET_DEFINE_STATIC(uma_zone_t, rtzone); /* Routing table UMA zone. */ #define V_rtzone VNET(rtzone) static int rtrequest1_fib_change(struct rib_head *, struct rt_addrinfo *, struct rtentry **, u_int); static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *); static int rt_ifdelroute(const struct rtentry *rt, void *arg); static struct rtentry *rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror); static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info); #ifdef RADIX_MPATH static struct radix_node *rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry *rto, int *perror); #endif static int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags); struct if_mtuinfo { struct ifnet *ifp; int mtu; }; static int if_updatemtu_cb(struct radix_node *, void *); /* * handler for net.my_fibnum */ static int sysctl_my_fibnum(SYSCTL_HANDLER_ARGS) { int fibnum; int error; fibnum = curthread->td_proc->p_fibnum; error = sysctl_handle_int(oidp, &fibnum, 0, req); return (error); } SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD, NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller"); static __inline struct rib_head ** rt_tables_get_rnh_ptr(int table, int fam) { struct rib_head **rnh; KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.", __func__)); KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.", __func__)); /* rnh is [fib=0][af=0]. */ rnh = (struct rib_head **)V_rt_tables; /* Get the offset to the requested table and fam. */ rnh += table * (AF_MAX+1) + fam; return (rnh); } struct rib_head * rt_tables_get_rnh(int table, int fam) { return (*rt_tables_get_rnh_ptr(table, fam)); } u_int rt_tables_get_gen(int table, int fam) { struct rib_head *rnh; rnh = *rt_tables_get_rnh_ptr(table, fam); KASSERT(rnh != NULL, ("%s: NULL rib_head pointer table %d fam %d", __func__, table, fam)); return (rnh->rnh_gen); } /* * route initialization must occur before ip6_init2(), which happenas at * SI_ORDER_MIDDLE. */ static void route_init(void) { /* whack the tunable ints into line. */ if (rt_numfibs > RT_MAXFIBS) rt_numfibs = RT_MAXFIBS; if (rt_numfibs == 0) rt_numfibs = 1; } SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, NULL); static int rtentry_zinit(void *mem, int size, int how) { struct rtentry *rt = mem; rt->rt_pksent = counter_u64_alloc(how); if (rt->rt_pksent == NULL) return (ENOMEM); RT_LOCK_INIT(rt); return (0); } static void rtentry_zfini(void *mem, int size) { struct rtentry *rt = mem; RT_LOCK_DESTROY(rt); counter_u64_free(rt->rt_pksent); } static int rtentry_ctor(void *mem, int size, void *arg, int how) { struct rtentry *rt = mem; bzero(rt, offsetof(struct rtentry, rt_endzero)); counter_u64_zero(rt->rt_pksent); rt->rt_chain = NULL; return (0); } static void rtentry_dtor(void *mem, int size, void *arg) { struct rtentry *rt = mem; RT_UNLOCK_COND(rt); } static void vnet_route_init(const void *unused __unused) { struct domain *dom; struct rib_head **rnh; int table; int fam; V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) * sizeof(struct rib_head *), M_RTABLE, M_WAITOK|M_ZERO); V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), rtentry_ctor, rtentry_dtor, rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0); for (dom = domains; dom; dom = dom->dom_next) { if (dom->dom_rtattach == NULL) continue; for (table = 0; table < rt_numfibs; table++) { fam = dom->dom_family; if (table != 0 && fam != AF_INET6 && fam != AF_INET) break; rnh = rt_tables_get_rnh_ptr(table, fam); if (rnh == NULL) panic("%s: rnh NULL", __func__); dom->dom_rtattach((void **)rnh, 0); } } } VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, vnet_route_init, 0); #ifdef VIMAGE static void vnet_route_uninit(const void *unused __unused) { int table; int fam; struct domain *dom; struct rib_head **rnh; for (dom = domains; dom; dom = dom->dom_next) { if (dom->dom_rtdetach == NULL) continue; for (table = 0; table < rt_numfibs; table++) { fam = dom->dom_family; if (table != 0 && fam != AF_INET6 && fam != AF_INET) break; rnh = rt_tables_get_rnh_ptr(table, fam); if (rnh == NULL) panic("%s: rnh NULL", __func__); dom->dom_rtdetach((void **)rnh, 0); } } free(V_rt_tables, M_RTABLE); uma_zdestroy(V_rtzone); } VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, vnet_route_uninit, 0); #endif struct rib_head * rt_table_init(int offset) { struct rib_head *rh; rh = malloc(sizeof(struct rib_head), M_RTABLE, M_WAITOK | M_ZERO); /* TODO: These details should be hidded inside radix.c */ /* Init masks tree */ rn_inithead_internal(&rh->head, rh->rnh_nodes, offset); rn_inithead_internal(&rh->rmhead.head, rh->rmhead.mask_nodes, 0); rh->head.rnh_masks = &rh->rmhead; /* Init locks */ RIB_LOCK_INIT(rh); /* Finally, set base callbacks */ rh->rnh_addaddr = rn_addroute; rh->rnh_deladdr = rn_delete; rh->rnh_matchaddr = rn_match; rh->rnh_lookup = rn_lookup; rh->rnh_walktree = rn_walktree; rh->rnh_walktree_from = rn_walktree_from; return (rh); } static int rt_freeentry(struct radix_node *rn, void *arg) { struct radix_head * const rnh = arg; struct radix_node *x; x = (struct radix_node *)rn_delete(rn + 2, NULL, rnh); if (x != NULL) R_Free(x); return (0); } void rt_table_destroy(struct rib_head *rh) { rn_walktree(&rh->rmhead.head, rt_freeentry, &rh->rmhead.head); /* Assume table is already empty */ RIB_LOCK_DESTROY(rh); free(rh, M_RTABLE); } #ifndef _SYS_SYSPROTO_H_ struct setfib_args { int fibnum; }; #endif int sys_setfib(struct thread *td, struct setfib_args *uap) { if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs) return EINVAL; td->td_proc->p_fibnum = uap->fibnum; return (0); } /* * Packet routing routines. */ void rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum) { struct rtentry *rt; if ((rt = ro->ro_rt) != NULL) { if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP) return; RTFREE(rt); ro->ro_rt = NULL; } ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum); if (ro->ro_rt) RT_UNLOCK(ro->ro_rt); } /* * Look up the route that matches the address given * Or, at least try.. Create a cloned route if needed. * * The returned route, if any, is locked. */ struct rtentry * rtalloc1(struct sockaddr *dst, int report, u_long ignflags) { return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB)); } struct rtentry * rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *newrt; struct rt_addrinfo info; int err = 0, msgtype = RTM_MISS; KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum")); rh = rt_tables_get_rnh(fibnum, dst->sa_family); newrt = NULL; if (rh == NULL) goto miss; /* * Look up the address in the table for that Address Family */ if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RLOCK(rh); #ifdef INVARIANTS else RIB_LOCK_ASSERT(rh); #endif rn = rh->rnh_matchaddr(dst, &rh->head); if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) { newrt = RNTORT(rn); RT_LOCK(newrt); RT_ADDREF(newrt); if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RUNLOCK(rh); return (newrt); } else if ((ignflags & RTF_RNH_LOCKED) == 0) RIB_RUNLOCK(rh); /* * Either we hit the root or could not find any match, * which basically means: "cannot get there from here". */ miss: V_rtstat.rts_unreach++; if (report) { /* * If required, report the failure to the supervising * Authorities. * For a delete, this is not an error. (report == 0) */ bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; rt_missmsg_fib(msgtype, &info, 0, err, fibnum); } return (newrt); } /* * Remove a reference count from an rtentry. * If the count gets low enough, take it out of the routing table */ void rtfree(struct rtentry *rt) { struct rib_head *rnh; KASSERT(rt != NULL,("%s: NULL rt", __func__)); rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family); KASSERT(rnh != NULL,("%s: NULL rnh", __func__)); RT_LOCK_ASSERT(rt); /* * The callers should use RTFREE_LOCKED() or RTFREE(), so * we should come here exactly with the last reference. */ RT_REMREF(rt); if (rt->rt_refcnt > 0) { log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt); goto done; } /* * On last reference give the "close method" a chance * to cleanup private state. This also permits (for * IPv4 and IPv6) a chance to decide if the routing table * entry should be purged immediately or at a later time. * When an immediate purge is to happen the close routine * typically calls rtexpunge which clears the RTF_UP flag * on the entry so that the code below reclaims the storage. */ if (rt->rt_refcnt == 0 && rnh->rnh_close) rnh->rnh_close((struct radix_node *)rt, &rnh->head); /* * If we are no longer "up" (and ref == 0) * then we can free the resources associated * with the route. */ if ((rt->rt_flags & RTF_UP) == 0) { if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT)) panic("rtfree 2"); /* * the rtentry must have been removed from the routing table * so it is represented in rttrash.. remove that now. */ V_rttrash--; #ifdef DIAGNOSTIC if (rt->rt_refcnt < 0) { printf("rtfree: %p not freed (neg refs)\n", rt); goto done; } #endif /* * release references on items we hold them on.. * e.g other routes and ifaddrs. */ if (rt->rt_ifa) ifa_free(rt->rt_ifa); /* * The key is separatly alloc'd so free it (see rt_setgate()). * This also frees the gateway, as they are always malloc'd * together. */ R_Free(rt_key(rt)); /* * and the rtentry itself of course */ uma_zfree(V_rtzone, rt); return; } done: RT_UNLOCK(rt); } /* * Force a routing table entry to the specified * destination to go through the given gateway. * Normally called as a result of a routing redirect * message from the network layer. */ void rtredirect_fib(struct sockaddr *dst, struct sockaddr *gateway, struct sockaddr *netmask, int flags, struct sockaddr *src, u_int fibnum) { struct rtentry *rt; int error = 0; short *stat = NULL; struct rt_addrinfo info; struct ifaddr *ifa; struct rib_head *rnh; ifa = NULL; NET_EPOCH_ENTER(); rnh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rnh == NULL) { error = EAFNOSUPPORT; goto out; } /* verify the gateway is directly reachable */ if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) { error = ENETUNREACH; goto out; } rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */ /* * If the redirect isn't from our current router for this dst, * it's either old or wrong. If it redirects us to ourselves, * we have a routing loop, perhaps as a result of an interface * going down recently. */ if (!(flags & RTF_DONE) && rt) { if (!sa_equal(src, rt->rt_gateway)) { error = EINVAL; goto done; } if (rt->rt_ifa != ifa && ifa->ifa_addr->sa_family != AF_LINK) { error = EINVAL; goto done; } } if ((flags & RTF_GATEWAY) && ifa_ifwithaddr_check(gateway)) { error = EHOSTUNREACH; goto done; } /* * Create a new entry if we just got back a wildcard entry * or the lookup failed. This is necessary for hosts * which use routing redirects generated by smart gateways * to dynamically build the routing tables. */ if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2)) goto create; /* * Don't listen to the redirect if it's * for a route to an interface. */ if (rt->rt_flags & RTF_GATEWAY) { if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) { /* * Changing from route to net => route to host. * Create new route, rather than smashing route to net. */ create: if (rt != NULL) RTFREE_LOCKED(rt); flags |= RTF_DYNAMIC; bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_info[RTAX_NETMASK] = netmask; ifa_ref(ifa); info.rti_ifa = ifa; info.rti_flags = flags; error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum); if (rt != NULL) { RT_LOCK(rt); flags = rt->rt_flags; } stat = &V_rtstat.rts_dynamic; } else { /* * Smash the current notion of the gateway to * this destination. Should check about netmask!!! */ if ((flags & RTF_GATEWAY) == 0) rt->rt_flags &= ~RTF_GATEWAY; rt->rt_flags |= RTF_MODIFIED; flags |= RTF_MODIFIED; stat = &V_rtstat.rts_newgateway; /* * add the key and gateway (in one malloc'd chunk). */ RT_UNLOCK(rt); RIB_WLOCK(rnh); RT_LOCK(rt); rt_setgate(rt, rt_key(rt), gateway); RIB_WUNLOCK(rnh); } } else error = EHOSTUNREACH; done: if (rt) RTFREE_LOCKED(rt); out: NET_EPOCH_EXIT(); if (error) V_rtstat.rts_badredirect++; else if (stat != NULL) (*stat)++; bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_info[RTAX_NETMASK] = netmask; info.rti_info[RTAX_AUTHOR] = src; rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum); } /* * Routing table ioctl interface. */ int rtioctl_fib(u_long req, caddr_t data, u_int fibnum) { /* * If more ioctl commands are added here, make sure the proper * super-user checks are being performed because it is possible for * prison-root to make it this far if raw sockets have been enabled * in jails. */ #ifdef INET /* Multicast goop, grrr... */ return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP; #else /* INET */ return ENXIO; #endif /* INET */ } struct ifaddr * ifa_ifwithroute(int flags, const struct sockaddr *dst, struct sockaddr *gateway, u_int fibnum) { struct ifaddr *ifa; int not_found = 0; MPASS(in_epoch(net_epoch_preempt)); if ((flags & RTF_GATEWAY) == 0) { /* * If we are adding a route to an interface, * and the interface is a pt to pt link * we should search for the destination * as our clue to the interface. Otherwise * we can use the local address. */ ifa = NULL; if (flags & RTF_HOST) ifa = ifa_ifwithdstaddr(dst, fibnum); if (ifa == NULL) ifa = ifa_ifwithaddr(gateway); } else { /* * If we are adding a route to a remote net * or host, the gateway may still be on the * other end of a pt to pt link. */ ifa = ifa_ifwithdstaddr(gateway, fibnum); } if (ifa == NULL) ifa = ifa_ifwithnet(gateway, 0, fibnum); if (ifa == NULL) { struct rtentry *rt; rt = rtalloc1_fib(gateway, 0, flags, fibnum); if (rt == NULL) goto out; /* * dismiss a gateway that is reachable only * through the default router */ switch (gateway->sa_family) { case AF_INET: if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) not_found = 1; break; case AF_INET6: if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr)) not_found = 1; break; default: break; } if (!not_found && rt->rt_ifa != NULL) { ifa = rt->rt_ifa; } RT_REMREF(rt); RT_UNLOCK(rt); if (not_found || ifa == NULL) goto out; } if (ifa->ifa_addr->sa_family != dst->sa_family) { struct ifaddr *oifa = ifa; ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); if (ifa == NULL) ifa = oifa; } out: return (ifa); } /* * Do appropriate manipulations of a routing tree given * all the bits of info needed */ int rtrequest_fib(int req, struct sockaddr *dst, struct sockaddr *gateway, struct sockaddr *netmask, int flags, struct rtentry **ret_nrt, u_int fibnum) { struct rt_addrinfo info; if (dst->sa_len == 0) return(EINVAL); bzero((caddr_t)&info, sizeof(info)); info.rti_flags = flags; info.rti_info[RTAX_DST] = dst; info.rti_info[RTAX_GATEWAY] = gateway; info.rti_info[RTAX_NETMASK] = netmask; return rtrequest1_fib(req, &info, ret_nrt, fibnum); } /* * Copy most of @rt data into @info. * * If @flags contains NHR_COPY, copies dst,netmask and gw to the * pointers specified by @info structure. Assume such pointers * are zeroed sockaddr-like structures with sa_len field initialized * to reflect size of the provided buffer. if no NHR_COPY is specified, * point dst,netmask and gw @info fields to appropriate @rt values. * * if @flags contains NHR_REF, do refcouting on rt_ifp. * * Returns 0 on success. */ int rt_exportinfo(struct rtentry *rt, struct rt_addrinfo *info, int flags) { struct rt_metrics *rmx; struct sockaddr *src, *dst; int sa_len; if (flags & NHR_COPY) { /* Copy destination if dst is non-zero */ src = rt_key(rt); dst = info->rti_info[RTAX_DST]; sa_len = src->sa_len; if (dst != NULL) { if (src->sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_DST; } /* Copy mask if set && dst is non-zero */ src = rt_mask(rt); dst = info->rti_info[RTAX_NETMASK]; if (src != NULL && dst != NULL) { /* * Radix stores different value in sa_len, * assume rt_mask() to have the same length * as rt_key() */ if (sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_NETMASK; } /* Copy gateway is set && dst is non-zero */ src = rt->rt_gateway; dst = info->rti_info[RTAX_GATEWAY]; if ((rt->rt_flags & RTF_GATEWAY) && src != NULL && dst != NULL){ if (src->sa_len > dst->sa_len) return (ENOMEM); memcpy(dst, src, src->sa_len); info->rti_addrs |= RTA_GATEWAY; } } else { info->rti_info[RTAX_DST] = rt_key(rt); info->rti_addrs |= RTA_DST; if (rt_mask(rt) != NULL) { info->rti_info[RTAX_NETMASK] = rt_mask(rt); info->rti_addrs |= RTA_NETMASK; } if (rt->rt_flags & RTF_GATEWAY) { info->rti_info[RTAX_GATEWAY] = rt->rt_gateway; info->rti_addrs |= RTA_GATEWAY; } } rmx = info->rti_rmx; if (rmx != NULL) { info->rti_mflags |= RTV_MTU; rmx->rmx_mtu = rt->rt_mtu; } info->rti_flags = rt->rt_flags; info->rti_ifp = rt->rt_ifp; info->rti_ifa = rt->rt_ifa; ifa_ref(info->rti_ifa); if (flags & NHR_REF) { /* Do 'traditional' refcouting */ if_ref(info->rti_ifp); } return (0); } /* * Lookups up route entry for @dst in RIB database for fib @fibnum. * Exports entry data to @info using rt_exportinfo(). * * if @flags contains NHR_REF, refcouting is performed on rt_ifp. * All references can be released later by calling rib_free_info() * * Returns 0 on success. * Returns ENOENT for lookup failure, ENOMEM for export failure. */ int rib_lookup_info(uint32_t fibnum, const struct sockaddr *dst, uint32_t flags, uint32_t flowid, struct rt_addrinfo *info) { RIB_RLOCK_TRACKER; struct rib_head *rh; struct radix_node *rn; struct rtentry *rt; int error; KASSERT((fibnum < rt_numfibs), ("rib_lookup_rte: bad fibnum")); rh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rh == NULL) return (ENOENT); RIB_RLOCK(rh); rn = rh->rnh_matchaddr(__DECONST(void *, dst), &rh->head); if (rn != NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { rt = RNTORT(rn); /* Ensure route & ifp is UP */ if (RT_LINK_IS_UP(rt->rt_ifp)) { flags = (flags & NHR_REF) | NHR_COPY; error = rt_exportinfo(rt, info, flags); RIB_RUNLOCK(rh); return (error); } } RIB_RUNLOCK(rh); return (ENOENT); } /* * Releases all references acquired by rib_lookup_info() when * called with NHR_REF flags. */ void rib_free_info(struct rt_addrinfo *info) { if_rele(info->rti_ifp); } /* * Iterates over all existing fibs in system calling * @setwa_f function prior to traversing each fib. * Calls @wa_f function for each element in current fib. * If af is not AF_UNSPEC, iterates over fibs in particular * address family. */ void rt_foreach_fib_walk(int af, rt_setwarg_t *setwa_f, rt_walktree_f_t *wa_f, void *arg) { struct rib_head *rnh; uint32_t fibnum; int i; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { /* Do we want some specific family? */ if (af != AF_UNSPEC) { rnh = rt_tables_get_rnh(fibnum, af); if (rnh == NULL) continue; if (setwa_f != NULL) setwa_f(rnh, fibnum, af, arg); RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg); RIB_WUNLOCK(rnh); continue; } for (i = 1; i <= AF_MAX; i++) { rnh = rt_tables_get_rnh(fibnum, i); if (rnh == NULL) continue; if (setwa_f != NULL) setwa_f(rnh, fibnum, i, arg); RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, (walktree_f_t *)wa_f,arg); RIB_WUNLOCK(rnh); } } } struct rt_delinfo { struct rt_addrinfo info; struct rib_head *rnh; struct rtentry *head; }; /* * Conditionally unlinks @rn from radix tree based * on info data passed in @arg. */ static int rt_checkdelroute(struct radix_node *rn, void *arg) { struct rt_delinfo *di; struct rt_addrinfo *info; struct rtentry *rt; int error; di = (struct rt_delinfo *)arg; rt = (struct rtentry *)rn; info = &di->info; error = 0; info->rti_info[RTAX_DST] = rt_key(rt); info->rti_info[RTAX_NETMASK] = rt_mask(rt); info->rti_info[RTAX_GATEWAY] = rt->rt_gateway; rt = rt_unlinkrte(di->rnh, info, &error); if (rt == NULL) { /* Either not allowed or not matched. Skip entry */ return (0); } /* Entry was unlinked. Add to the list and return */ rt->rt_chain = di->head; di->head = rt; return (0); } /* * Iterates over all existing fibs in system. * Deletes each element for which @filter_f function returned * non-zero value. * If @af is not AF_UNSPEC, iterates over fibs in particular * address family. */ void rt_foreach_fib_walk_del(int af, rt_filter_f_t *filter_f, void *arg) { struct rib_head *rnh; struct rt_delinfo di; struct rtentry *rt; uint32_t fibnum; int i, start, end; bzero(&di, sizeof(di)); di.info.rti_filter = filter_f; di.info.rti_filterdata = arg; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { /* Do we want some specific family? */ if (af != AF_UNSPEC) { start = af; end = af; } else { start = 1; end = AF_MAX; } for (i = start; i <= end; i++) { rnh = rt_tables_get_rnh(fibnum, i); if (rnh == NULL) continue; di.rnh = rnh; RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, rt_checkdelroute, &di); RIB_WUNLOCK(rnh); if (di.head == NULL) continue; /* We might have something to reclaim */ while (di.head != NULL) { rt = di.head; di.head = rt->rt_chain; rt->rt_chain = NULL; /* TODO std rt -> rt_addrinfo export */ di.info.rti_info[RTAX_DST] = rt_key(rt); di.info.rti_info[RTAX_NETMASK] = rt_mask(rt); rt_notifydelete(rt, &di.info); RTFREE_LOCKED(rt); } } } } /* * Delete Routes for a Network Interface * * Called for each routing entry via the rnh->rnh_walktree() call above * to delete all route entries referencing a detaching network interface. * * Arguments: * rt pointer to rtentry * arg argument passed to rnh->rnh_walktree() - detaching interface * * Returns: * 0 successful * errno failed - reason indicated */ static int rt_ifdelroute(const struct rtentry *rt, void *arg) { struct ifnet *ifp = arg; if (rt->rt_ifp != ifp) return (0); /* * Protect (sorta) against walktree recursion problems * with cloned routes */ if ((rt->rt_flags & RTF_UP) == 0) return (0); return (1); } /* * Delete all remaining routes using this interface * Unfortuneatly the only way to do this is to slog through * the entire routing table looking for routes which point * to this interface...oh well... */ void rt_flushifroutes_af(struct ifnet *ifp, int af) { KASSERT((af >= 1 && af <= AF_MAX), ("%s: af %d not >= 1 and <= %d", __func__, af, AF_MAX)); rt_foreach_fib_walk_del(af, rt_ifdelroute, ifp); } void rt_flushifroutes(struct ifnet *ifp) { rt_foreach_fib_walk_del(AF_UNSPEC, rt_ifdelroute, ifp); } /* * Conditionally unlinks rtentry matching data inside @info from @rnh. * Returns unlinked, locked and referenced @rtentry on success, * Returns NULL and sets @perror to: * ESRCH - if prefix was not found, * EADDRINUSE - if trying to delete PINNED route without appropriate flag. * ENOENT - if supplied filter function returned 0 (not matched). */ static struct rtentry * rt_unlinkrte(struct rib_head *rnh, struct rt_addrinfo *info, int *perror) { struct sockaddr *dst, *netmask; struct rtentry *rt; struct radix_node *rn; dst = info->rti_info[RTAX_DST]; netmask = info->rti_info[RTAX_NETMASK]; rt = (struct rtentry *)rnh->rnh_lookup(dst, netmask, &rnh->head); if (rt == NULL) { *perror = ESRCH; return (NULL); } if ((info->rti_flags & RTF_PINNED) == 0) { /* Check if target route can be deleted */ if (rt->rt_flags & RTF_PINNED) { *perror = EADDRINUSE; return (NULL); } } if (info->rti_filter != NULL) { if (info->rti_filter(rt, info->rti_filterdata) == 0) { /* Not matched */ *perror = ENOENT; return (NULL); } /* * Filter function requested rte deletion. * Ease the caller work by filling in remaining info * from that particular entry. */ info->rti_info[RTAX_GATEWAY] = rt->rt_gateway; } /* * Remove the item from the tree and return it. * Complain if it is not there and do no more processing. */ *perror = ESRCH; #ifdef RADIX_MPATH if (rt_mpath_capable(rnh)) rn = rt_mpath_unlink(rnh, info, rt, perror); else #endif rn = rnh->rnh_deladdr(dst, netmask, &rnh->head); if (rn == NULL) return (NULL); if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) panic ("rtrequest delete"); rt = RNTORT(rn); RT_LOCK(rt); RT_ADDREF(rt); rt->rt_flags &= ~RTF_UP; *perror = 0; return (rt); } static void rt_notifydelete(struct rtentry *rt, struct rt_addrinfo *info) { struct ifaddr *ifa; /* * give the protocol a chance to keep things in sync. */ ifa = rt->rt_ifa; if (ifa != NULL && ifa->ifa_rtrequest != NULL) ifa->ifa_rtrequest(RTM_DELETE, rt, info); /* * One more rtentry floating around that is not * linked to the routing table. rttrash will be decremented * when RTFREE(rt) is eventually called. */ V_rttrash++; } /* * These (questionable) definitions of apparent local variables apply * to the next two functions. XXXXXX!!! */ #define dst info->rti_info[RTAX_DST] #define gateway info->rti_info[RTAX_GATEWAY] #define netmask info->rti_info[RTAX_NETMASK] #define ifaaddr info->rti_info[RTAX_IFA] #define ifpaddr info->rti_info[RTAX_IFP] #define flags info->rti_flags /* * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined, * it will be referenced so the caller must free it. */ int rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum) { struct ifaddr *ifa; int needref, error; /* * ifp may be specified by sockaddr_dl * when protocol address is ambiguous. */ error = 0; needref = (info->rti_ifa == NULL); NET_EPOCH_ENTER(); if (info->rti_ifp == NULL && ifpaddr != NULL && ifpaddr->sa_family == AF_LINK && (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) { info->rti_ifp = ifa->ifa_ifp; } if (info->rti_ifa == NULL && ifaaddr != NULL) info->rti_ifa = ifa_ifwithaddr(ifaaddr); if (info->rti_ifa == NULL) { struct sockaddr *sa; sa = ifaaddr != NULL ? ifaaddr : (gateway != NULL ? gateway : dst); if (sa != NULL && info->rti_ifp != NULL) info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp); else if (dst != NULL && gateway != NULL) info->rti_ifa = ifa_ifwithroute(flags, dst, gateway, fibnum); else if (sa != NULL) info->rti_ifa = ifa_ifwithroute(flags, sa, sa, fibnum); } if (needref && info->rti_ifa != NULL) { if (info->rti_ifp == NULL) info->rti_ifp = info->rti_ifa->ifa_ifp; ifa_ref(info->rti_ifa); } else error = ENETUNREACH; NET_EPOCH_EXIT(); return (error); } static int if_updatemtu_cb(struct radix_node *rn, void *arg) { struct rtentry *rt; struct if_mtuinfo *ifmtu; rt = (struct rtentry *)rn; ifmtu = (struct if_mtuinfo *)arg; if (rt->rt_ifp != ifmtu->ifp) return (0); if (rt->rt_mtu >= ifmtu->mtu) { /* We have to decrease mtu regardless of flags */ rt->rt_mtu = ifmtu->mtu; return (0); } /* * New MTU is bigger. Check if are allowed to alter it */ if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) { /* * Skip routes with user-supplied MTU and * non-interface routes */ return (0); } /* We are safe to update route MTU */ rt->rt_mtu = ifmtu->mtu; return (0); } void rt_updatemtu(struct ifnet *ifp) { struct if_mtuinfo ifmtu; struct rib_head *rnh; int i, j; ifmtu.ifp = ifp; /* * Try to update rt_mtu for all routes using this interface * Unfortunately the only way to do this is to traverse all * routing tables in all fibs/domains. */ for (i = 1; i <= AF_MAX; i++) { ifmtu.mtu = if_getmtu_family(ifp, i); for (j = 0; j < rt_numfibs; j++) { rnh = rt_tables_get_rnh(j, i); if (rnh == NULL) continue; RIB_WLOCK(rnh); rnh->rnh_walktree(&rnh->head, if_updatemtu_cb, &ifmtu); RIB_WUNLOCK(rnh); } } } #if 0 int p_sockaddr(char *buf, int buflen, struct sockaddr *s); int rt_print(char *buf, int buflen, struct rtentry *rt); int p_sockaddr(char *buf, int buflen, struct sockaddr *s) { void *paddr = NULL; switch (s->sa_family) { case AF_INET: paddr = &((struct sockaddr_in *)s)->sin_addr; break; case AF_INET6: paddr = &((struct sockaddr_in6 *)s)->sin6_addr; break; } if (paddr == NULL) return (0); if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL) return (0); return (strlen(buf)); } int rt_print(char *buf, int buflen, struct rtentry *rt) { struct sockaddr *addr, *mask; int i = 0; addr = rt_key(rt); mask = rt_mask(rt); i = p_sockaddr(buf, buflen, addr); if (!(rt->rt_flags & RTF_HOST)) { buf[i++] = '/'; i += p_sockaddr(buf + i, buflen - i, mask); } if (rt->rt_flags & RTF_GATEWAY) { buf[i++] = '>'; i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway); } return (i); } #endif #ifdef RADIX_MPATH /* * Deletes key for single-path routes, unlinks rtentry with * gateway specified in @info from multi-path routes. * * Returnes unlinked entry. In case of failure, returns NULL * and sets @perror to ESRCH. */ static struct radix_node * rt_mpath_unlink(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry *rto, int *perror) { /* * if we got multipath routes, we require users to specify * a matching RTAX_GATEWAY. */ struct rtentry *rt; // *rto = NULL; struct radix_node *rn; struct sockaddr *gw; gw = info->rti_info[RTAX_GATEWAY]; rt = rt_mpath_matchgate(rto, gw); if (rt == NULL) { *perror = ESRCH; return (NULL); } /* * this is the first entry in the chain */ if (rto == rt) { rn = rn_mpath_next((struct radix_node *)rt); /* * there is another entry, now it's active */ if (rn) { rto = RNTORT(rn); RT_LOCK(rto); rto->rt_flags |= RTF_UP; RT_UNLOCK(rto); } else if (rt->rt_flags & RTF_GATEWAY) { /* * For gateway routes, we need to * make sure that we we are deleting * the correct gateway. * rt_mpath_matchgate() does not * check the case when there is only * one route in the chain. */ if (gw && (rt->rt_gateway->sa_len != gw->sa_len || memcmp(rt->rt_gateway, gw, gw->sa_len))) { *perror = ESRCH; return (NULL); } } /* * use the normal delete code to remove * the first entry */ rn = rnh->rnh_deladdr(dst, netmask, &rnh->head); *perror = 0; return (rn); } /* * if the entry is 2nd and on up */ if (rt_mpath_deldup(rto, rt) == 0) panic ("rtrequest1: rt_mpath_deldup"); *perror = 0; rn = (struct radix_node *)rt; return (rn); } #endif int rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt, u_int fibnum) { int error = 0; struct rtentry *rt, *rt_old; struct radix_node *rn; struct rib_head *rnh; struct ifaddr *ifa; struct sockaddr *ndst; struct sockaddr_storage mdst; KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum")); KASSERT((flags & RTF_RNH_LOCKED) == 0, ("rtrequest1_fib: locked")); switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We support multiple FIBs. */ break; default: fibnum = RT_DEFAULT_FIB; break; } /* * Find the correct routing tree to use for this Address Family */ rnh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rnh == NULL) return (EAFNOSUPPORT); /* * If we are adding a host route then we don't want to put * a netmask in the tree, nor do we want to clone it. */ if (flags & RTF_HOST) netmask = NULL; switch (req) { case RTM_DELETE: if (netmask) { rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask); dst = (struct sockaddr *)&mdst; } RIB_WLOCK(rnh); rt = rt_unlinkrte(rnh, info, &error); RIB_WUNLOCK(rnh); if (error != 0) return (error); rt_notifydelete(rt, info); /* * If the caller wants it, then it can have it, * but it's up to it to free the rtentry as we won't be * doing it. */ if (ret_nrt) { *ret_nrt = rt; RT_UNLOCK(rt); } else RTFREE_LOCKED(rt); break; case RTM_RESOLVE: /* * resolve was only used for route cloning * here for compat */ break; case RTM_ADD: if ((flags & RTF_GATEWAY) && !gateway) return (EINVAL); if (dst && gateway && (dst->sa_family != gateway->sa_family) && (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) return (EINVAL); if (info->rti_ifa == NULL) { error = rt_getifa_fib(info, fibnum); if (error) return (error); } rt = uma_zalloc(V_rtzone, M_NOWAIT); if (rt == NULL) { return (ENOBUFS); } rt->rt_flags = RTF_UP | flags; rt->rt_fibnum = fibnum; /* * Add the gateway. Possibly re-malloc-ing the storage for it. */ if ((error = rt_setgate(rt, dst, gateway)) != 0) { uma_zfree(V_rtzone, rt); return (error); } /* * point to the (possibly newly malloc'd) dest address. */ ndst = (struct sockaddr *)rt_key(rt); /* * make sure it contains the value we want (masked if needed). */ if (netmask) { rt_maskedcopy(dst, ndst, netmask); } else bcopy(dst, ndst, dst->sa_len); /* * We use the ifa reference returned by rt_getifa_fib(). * This moved from below so that rnh->rnh_addaddr() can * examine the ifa and ifa->ifa_ifp if it so desires. */ ifa = info->rti_ifa; ifa_ref(ifa); rt->rt_ifa = ifa; rt->rt_ifp = ifa->ifa_ifp; rt->rt_weight = 1; rt_setmetrics(info, rt); RIB_WLOCK(rnh); RT_LOCK(rt); #ifdef RADIX_MPATH /* do not permit exactly the same dst/mask/gw pair */ if (rt_mpath_capable(rnh) && rt_mpath_conflict(rnh, rt, netmask)) { RIB_WUNLOCK(rnh); ifa_free(rt->rt_ifa); R_Free(rt_key(rt)); uma_zfree(V_rtzone, rt); return (EEXIST); } #endif /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes); rt_old = NULL; if (rn == NULL && (info->rti_flags & RTF_PINNED) != 0) { /* * Force removal and re-try addition * TODO: better multipath&pinned support */ struct sockaddr *info_dst = info->rti_info[RTAX_DST]; info->rti_info[RTAX_DST] = ndst; /* Do not delete existing PINNED(interface) routes */ info->rti_flags &= ~RTF_PINNED; rt_old = rt_unlinkrte(rnh, info, &error); info->rti_flags |= RTF_PINNED; info->rti_info[RTAX_DST] = info_dst; if (rt_old != NULL) rn = rnh->rnh_addaddr(ndst, netmask, &rnh->head, rt->rt_nodes); } RIB_WUNLOCK(rnh); if (rt_old != NULL) RT_UNLOCK(rt_old); /* * If it still failed to go into the tree, * then un-make it (this should be a function) */ if (rn == NULL) { ifa_free(rt->rt_ifa); R_Free(rt_key(rt)); uma_zfree(V_rtzone, rt); return (EEXIST); } if (rt_old != NULL) { rt_notifydelete(rt_old, info); RTFREE(rt_old); } /* * If this protocol has something to add to this then * allow it to do that as well. */ if (ifa->ifa_rtrequest) ifa->ifa_rtrequest(req, rt, info); /* * actually return a resultant rtentry and * give the caller a single reference. */ if (ret_nrt) { *ret_nrt = rt; RT_ADDREF(rt); } rnh->rnh_gen++; /* Routing table updated */ RT_UNLOCK(rt); break; case RTM_CHANGE: RIB_WLOCK(rnh); error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum); RIB_WUNLOCK(rnh); break; default: error = EOPNOTSUPP; } return (error); } #undef dst #undef gateway #undef netmask #undef ifaaddr #undef ifpaddr #undef flags static int rtrequest1_fib_change(struct rib_head *rnh, struct rt_addrinfo *info, struct rtentry **ret_nrt, u_int fibnum) { struct rtentry *rt = NULL; int error = 0; int free_ifa = 0; int family, mtu; struct if_mtuinfo ifmtu; RIB_WLOCK_ASSERT(rnh); rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST], info->rti_info[RTAX_NETMASK], &rnh->head); if (rt == NULL) return (ESRCH); #ifdef RADIX_MPATH /* * If we got multipath routes, * we require users to specify a matching RTAX_GATEWAY. */ if (rt_mpath_capable(rnh)) { rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]); if (rt == NULL) return (ESRCH); } #endif RT_LOCK(rt); rt_setmetrics(info, rt); /* * New gateway could require new ifaddr, ifp; * flags may also be different; ifp may be specified * by ll sockaddr when protocol address is ambiguous */ if (((rt->rt_flags & RTF_GATEWAY) && info->rti_info[RTAX_GATEWAY] != NULL) || info->rti_info[RTAX_IFP] != NULL || (info->rti_info[RTAX_IFA] != NULL && !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) { /* * XXX: Temporarily set RTF_RNH_LOCKED flag in the rti_flags * to avoid rlock in the ifa_ifwithroute(). */ info->rti_flags |= RTF_RNH_LOCKED; error = rt_getifa_fib(info, fibnum); info->rti_flags &= ~RTF_RNH_LOCKED; if (info->rti_ifa != NULL) free_ifa = 1; if (error != 0) goto bad; } /* Check if outgoing interface has changed */ if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa && rt->rt_ifa != NULL) { if (rt->rt_ifa->ifa_rtrequest != NULL) rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info); ifa_free(rt->rt_ifa); rt->rt_ifa = NULL; } /* Update gateway address */ if (info->rti_info[RTAX_GATEWAY] != NULL) { error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]); if (error != 0) goto bad; rt->rt_flags &= ~RTF_GATEWAY; rt->rt_flags |= (RTF_GATEWAY & info->rti_flags); } if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) { ifa_ref(info->rti_ifa); rt->rt_ifa = info->rti_ifa; rt->rt_ifp = info->rti_ifp; } /* Allow some flags to be toggled on change. */ rt->rt_flags &= ~RTF_FMASK; rt->rt_flags |= info->rti_flags & RTF_FMASK; if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL) rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info); /* Alter route MTU if necessary */ if (rt->rt_ifp != NULL) { family = info->rti_info[RTAX_DST]->sa_family; mtu = if_getmtu_family(rt->rt_ifp, family); /* Set default MTU */ if (rt->rt_mtu == 0) rt->rt_mtu = mtu; if (rt->rt_mtu != mtu) { /* Check if we really need to update */ ifmtu.ifp = rt->rt_ifp; ifmtu.mtu = mtu; if_updatemtu_cb(rt->rt_nodes, &ifmtu); } } /* * This route change may have modified the route's gateway. In that * case, any inpcbs that have cached this route need to invalidate their * llentry cache. */ rnh->rnh_gen++; if (ret_nrt) { *ret_nrt = rt; RT_ADDREF(rt); } bad: RT_UNLOCK(rt); if (free_ifa != 0) { ifa_free(info->rti_ifa); info->rti_ifa = NULL; } return (error); } static void rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt) { if (info->rti_mflags & RTV_MTU) { if (info->rti_rmx->rmx_mtu != 0) { /* * MTU was explicitly provided by user. * Keep it. */ rt->rt_flags |= RTF_FIXEDMTU; } else { /* * User explicitly sets MTU to 0. * Assume rollback to default. */ rt->rt_flags &= ~RTF_FIXEDMTU; } rt->rt_mtu = info->rti_rmx->rmx_mtu; } if (info->rti_mflags & RTV_WEIGHT) rt->rt_weight = info->rti_rmx->rmx_weight; /* Kernel -> userland timebase conversion. */ if (info->rti_mflags & RTV_EXPIRE) rt->rt_expire = info->rti_rmx->rmx_expire ? info->rti_rmx->rmx_expire - time_second + time_uptime : 0; } int rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate) { /* XXX dst may be overwritten, can we move this to below */ int dlen = SA_SIZE(dst), glen = SA_SIZE(gate); /* * Prepare to store the gateway in rt->rt_gateway. * Both dst and gateway are stored one after the other in the same * malloc'd chunk. If we have room, we can reuse the old buffer, * rt_gateway already points to the right place. * Otherwise, malloc a new block and update the 'dst' address. */ if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) { caddr_t new; R_Malloc(new, caddr_t, dlen + glen); if (new == NULL) return ENOBUFS; /* * XXX note, we copy from *dst and not *rt_key(rt) because * rt_setgate() can be called to initialize a newly * allocated route entry, in which case rt_key(rt) == NULL * (and also rt->rt_gateway == NULL). * Free()/free() handle a NULL argument just fine. */ bcopy(dst, new, dlen); R_Free(rt_key(rt)); /* free old block, if any */ rt_key(rt) = (struct sockaddr *)new; rt->rt_gateway = (struct sockaddr *)(new + dlen); } /* * Copy the new gateway value into the memory chunk. */ bcopy(gate, rt->rt_gateway, glen); return (0); } void rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask) { u_char *cp1 = (u_char *)src; u_char *cp2 = (u_char *)dst; u_char *cp3 = (u_char *)netmask; u_char *cplim = cp2 + *cp3; u_char *cplim2 = cp2 + *cp1; *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ cp3 += 2; if (cplim > cplim2) cplim = cplim2; while (cp2 < cplim) *cp2++ = *cp1++ & *cp3++; if (cp2 < cplim2) bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2)); } /* * Set up a routing table entry, normally * for an interface. */ #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */ static inline int rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum) { RIB_RLOCK_TRACKER; struct sockaddr *dst; struct sockaddr *netmask; struct rtentry *rt = NULL; struct rt_addrinfo info; int error = 0; int startfib, endfib; char tempbuf[_SOCKADDR_TMPSIZE]; int didwork = 0; int a_failure = 0; static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; struct rib_head *rnh; if (flags & RTF_HOST) { dst = ifa->ifa_dstaddr; netmask = NULL; } else { dst = ifa->ifa_addr; netmask = ifa->ifa_netmask; } if (dst->sa_len == 0) return(EINVAL); switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We support multiple FIBs. */ break; default: fibnum = RT_DEFAULT_FIB; break; } if (fibnum == RT_ALL_FIBS) { if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) startfib = endfib = ifa->ifa_ifp->if_fib; else { startfib = 0; endfib = rt_numfibs - 1; } } else { KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum")); startfib = fibnum; endfib = fibnum; } /* * If it's a delete, check that if it exists, * it's on the correct interface or we might scrub * a route to another ifa which would * be confusing at best and possibly worse. */ if (cmd == RTM_DELETE) { /* * It's a delete, so it should already exist.. * If it's a net, mask off the host bits * (Assuming we have a mask) * XXX this is kinda inet specific.. */ if (netmask != NULL) { rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask); dst = (struct sockaddr *)tempbuf; } } /* * Now go through all the requested tables (fibs) and do the * requested action. Realistically, this will either be fib 0 * for protocols that don't do multiple tables or all the * tables for those that do. */ for ( fibnum = startfib; fibnum <= endfib; fibnum++) { if (cmd == RTM_DELETE) { struct radix_node *rn; /* * Look up an rtentry that is in the routing tree and * contains the correct info. */ rnh = rt_tables_get_rnh(fibnum, dst->sa_family); if (rnh == NULL) /* this table doesn't exist but others might */ continue; RIB_RLOCK(rnh); rn = rnh->rnh_lookup(dst, netmask, &rnh->head); #ifdef RADIX_MPATH if (rt_mpath_capable(rnh)) { if (rn == NULL) error = ESRCH; else { rt = RNTORT(rn); /* * for interface route the * rt->rt_gateway is sockaddr_intf * for cloning ARP entries, so * rt_mpath_matchgate must use the * interface address */ rt = rt_mpath_matchgate(rt, ifa->ifa_addr); if (rt == NULL) error = ESRCH; } } #endif error = (rn == NULL || (rn->rn_flags & RNF_ROOT) || RNTORT(rn)->rt_ifa != ifa); RIB_RUNLOCK(rnh); if (error) { /* this is only an error if bad on ALL tables */ continue; } } /* * Do the actual request */ bzero((caddr_t)&info, sizeof(info)); ifa_ref(ifa); info.rti_ifa = ifa; info.rti_flags = flags | (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED; info.rti_info[RTAX_DST] = dst; /* * doing this for compatibility reasons */ if (cmd == RTM_ADD) info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl; else info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; info.rti_info[RTAX_NETMASK] = netmask; error = rtrequest1_fib(cmd, &info, &rt, fibnum); if (error == 0 && rt != NULL) { /* * notify any listening routing agents of the change */ RT_LOCK(rt); #ifdef RADIX_MPATH /* * in case address alias finds the first address * e.g. ifconfig bge0 192.0.2.246/24 * e.g. ifconfig bge0 192.0.2.247/24 * the address set in the route is 192.0.2.246 * so we need to replace it with 192.0.2.247 */ if (memcmp(rt->rt_ifa->ifa_addr, ifa->ifa_addr, ifa->ifa_addr->sa_len)) { ifa_free(rt->rt_ifa); ifa_ref(ifa); rt->rt_ifp = ifa->ifa_ifp; rt->rt_ifa = ifa; } #endif /* * doing this for compatibility reasons */ if (cmd == RTM_ADD) { ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type = rt->rt_ifp->if_type; ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index = rt->rt_ifp->if_index; } RT_ADDREF(rt); RT_UNLOCK(rt); rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum); RT_LOCK(rt); RT_REMREF(rt); if (cmd == RTM_DELETE) { /* * If we are deleting, and we found an entry, * then it's been removed from the tree.. * now throw it away. */ RTFREE_LOCKED(rt); } else { if (cmd == RTM_ADD) { /* * We just wanted to add it.. * we don't actually need a reference. */ RT_REMREF(rt); } RT_UNLOCK(rt); } didwork = 1; } if (error) a_failure = error; } if (cmd == RTM_DELETE) { if (didwork) { error = 0; } else { /* we only give an error if it wasn't in any table */ error = ((flags & RTF_HOST) ? EHOSTUNREACH : ENETUNREACH); } } else { if (a_failure) { /* return an error if any of them failed */ error = a_failure; } } return (error); } /* * Set up a routing table entry, normally * for an interface. */ int rtinit(struct ifaddr *ifa, int cmd, int flags) { struct sockaddr *dst; int fib = RT_DEFAULT_FIB; if (flags & RTF_HOST) { dst = ifa->ifa_dstaddr; } else { dst = ifa->ifa_addr; } switch (dst->sa_family) { case AF_INET6: case AF_INET: /* We do support multiple FIBs. */ fib = RT_ALL_FIBS; break; } return (rtinit1(ifa, cmd, flags, fib)); } /* * Announce interface address arrival/withdraw * Returns 0 on success. */ int rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %d", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); #if defined(INET) || defined(INET6) #ifdef SCTP /* * notify the SCTP stack * this will only get called when an address is added/deleted * XXX pass the ifaddr struct instead if ifa->ifa_addr... */ sctp_addr_change(ifa, cmd); #endif /* SCTP */ #endif return (rtsock_addrmsg(cmd, ifa, fibnum)); } /* * Announce route addition/removal. * Users of this function MUST validate input data BEFORE calling. * However we have to be able to handle invalid data: * if some userland app sends us "invalid" route message (invalid mask, * no dst, wrong address families, etc...) we need to pass it back * to app (and any other rtsock consumers) with rtm_errno field set to * non-zero value. * Returns 0 on success. */ int rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %d", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__)); return (rtsock_routemsg(cmd, ifp, error, rt, fibnum)); } void rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt) { rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS); } /* * This is called to generate messages from the routing socket * indicating a network interface has had addresses associated with it. */ void rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt, int fibnum) { KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, ("unexpected cmd %u", cmd)); KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); if (cmd == RTM_ADD) { rt_addrmsg(cmd, ifa, fibnum); if (rt != NULL) rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum); } else { if (rt != NULL) rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum); rt_addrmsg(cmd, ifa, fibnum); } } Index: head/sys/net/rtsock.c =================================================================== --- head/sys/net/rtsock.c (revision 336675) +++ head/sys/net/rtsock.c (revision 336676) @@ -1,1974 +1,1974 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1988, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)rtsock.c 8.7 (Berkeley) 10/12/95 * $FreeBSD$ */ #include "opt_mpath.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #endif #ifdef COMPAT_FREEBSD32 #include #include struct if_msghdr32 { uint16_t ifm_msglen; uint8_t ifm_version; uint8_t ifm_type; int32_t ifm_addrs; int32_t ifm_flags; uint16_t ifm_index; struct if_data ifm_data; }; struct if_msghdrl32 { uint16_t ifm_msglen; uint8_t ifm_version; uint8_t ifm_type; int32_t ifm_addrs; int32_t ifm_flags; uint16_t ifm_index; uint16_t _ifm_spare1; uint16_t ifm_len; uint16_t ifm_data_off; struct if_data ifm_data; }; struct ifa_msghdrl32 { uint16_t ifam_msglen; uint8_t ifam_version; uint8_t ifam_type; int32_t ifam_addrs; int32_t ifam_flags; uint16_t ifam_index; uint16_t _ifam_spare1; uint16_t ifam_len; uint16_t ifam_data_off; int32_t ifam_metric; struct if_data ifam_data; }; #define SA_SIZE32(sa) \ ( (((struct sockaddr *)(sa))->sa_len == 0) ? \ sizeof(int) : \ 1 + ( (((struct sockaddr *)(sa))->sa_len - 1) | (sizeof(int) - 1) ) ) #endif /* COMPAT_FREEBSD32 */ MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables"); /* NB: these are not modified */ static struct sockaddr route_src = { 2, PF_ROUTE, }; static struct sockaddr sa_zero = { sizeof(sa_zero), AF_INET, }; /* These are external hooks for CARP. */ int (*carp_get_vhid_p)(struct ifaddr *); /* * Used by rtsock/raw_input callback code to decide whether to filter the update * notification to a socket bound to a particular FIB. */ #define RTS_FILTER_FIB M_PROTO8 typedef struct { int ip_count; /* attached w/ AF_INET */ int ip6_count; /* attached w/ AF_INET6 */ int any_count; /* total attached */ } route_cb_t; -static VNET_DEFINE(route_cb_t, route_cb); +VNET_DEFINE_STATIC(route_cb_t, route_cb); #define V_route_cb VNET(route_cb) struct mtx rtsock_mtx; MTX_SYSINIT(rtsock, &rtsock_mtx, "rtsock route_cb lock", MTX_DEF); #define RTSOCK_LOCK() mtx_lock(&rtsock_mtx) #define RTSOCK_UNLOCK() mtx_unlock(&rtsock_mtx) #define RTSOCK_LOCK_ASSERT() mtx_assert(&rtsock_mtx, MA_OWNED) static SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RD, 0, ""); struct walkarg { int w_tmemsize; int w_op, w_arg; caddr_t w_tmem; struct sysctl_req *w_req; }; static void rts_input(struct mbuf *m); static struct mbuf *rtsock_msg_mbuf(int type, struct rt_addrinfo *rtinfo); static int rtsock_msg_buffer(int type, struct rt_addrinfo *rtinfo, struct walkarg *w, int *plen); static int rt_xaddrs(caddr_t cp, caddr_t cplim, struct rt_addrinfo *rtinfo); static int sysctl_dumpentry(struct radix_node *rn, void *vw); static int sysctl_iflist(int af, struct walkarg *w); static int sysctl_ifmalist(int af, struct walkarg *w); static int route_output(struct mbuf *m, struct socket *so, ...); static void rt_getmetrics(const struct rtentry *rt, struct rt_metrics *out); static void rt_dispatch(struct mbuf *, sa_family_t); static struct sockaddr *rtsock_fix_netmask(struct sockaddr *dst, struct sockaddr *smask, struct sockaddr_storage *dmask); static struct netisr_handler rtsock_nh = { .nh_name = "rtsock", .nh_handler = rts_input, .nh_proto = NETISR_ROUTE, .nh_policy = NETISR_POLICY_SOURCE, }; static int sysctl_route_netisr_maxqlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&rtsock_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&rtsock_nh, qlimit)); } SYSCTL_PROC(_net_route, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_route_netisr_maxqlen, "I", "maximum routing socket dispatch queue length"); static void vnet_rts_init(void) { int tmp; if (IS_DEFAULT_VNET(curvnet)) { if (TUNABLE_INT_FETCH("net.route.netisr_maxqlen", &tmp)) rtsock_nh.nh_qlimit = tmp; netisr_register(&rtsock_nh); } #ifdef VIMAGE else netisr_register_vnet(&rtsock_nh); #endif } VNET_SYSINIT(vnet_rtsock, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, vnet_rts_init, 0); #ifdef VIMAGE static void vnet_rts_uninit(void) { netisr_unregister_vnet(&rtsock_nh); } VNET_SYSUNINIT(vnet_rts_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, vnet_rts_uninit, 0); #endif static int raw_input_rts_cb(struct mbuf *m, struct sockproto *proto, struct sockaddr *src, struct rawcb *rp) { int fibnum; KASSERT(m != NULL, ("%s: m is NULL", __func__)); KASSERT(proto != NULL, ("%s: proto is NULL", __func__)); KASSERT(rp != NULL, ("%s: rp is NULL", __func__)); /* No filtering requested. */ if ((m->m_flags & RTS_FILTER_FIB) == 0) return (0); /* Check if it is a rts and the fib matches the one of the socket. */ fibnum = M_GETFIB(m); if (proto->sp_family != PF_ROUTE || rp->rcb_socket == NULL || rp->rcb_socket->so_fibnum == fibnum) return (0); /* Filtering requested and no match, the socket shall be skipped. */ return (1); } static void rts_input(struct mbuf *m) { struct sockproto route_proto; unsigned short *family; struct m_tag *tag; route_proto.sp_family = PF_ROUTE; tag = m_tag_find(m, PACKET_TAG_RTSOCKFAM, NULL); if (tag != NULL) { family = (unsigned short *)(tag + 1); route_proto.sp_protocol = *family; m_tag_delete(m, tag); } else route_proto.sp_protocol = 0; raw_input_ext(m, &route_proto, &route_src, raw_input_rts_cb); } /* * It really doesn't make any sense at all for this code to share much * with raw_usrreq.c, since its functionality is so restricted. XXX */ static void rts_abort(struct socket *so) { raw_usrreqs.pru_abort(so); } static void rts_close(struct socket *so) { raw_usrreqs.pru_close(so); } /* pru_accept is EOPNOTSUPP */ static int rts_attach(struct socket *so, int proto, struct thread *td) { struct rawcb *rp; int error; KASSERT(so->so_pcb == NULL, ("rts_attach: so_pcb != NULL")); /* XXX */ rp = malloc(sizeof *rp, M_PCB, M_WAITOK | M_ZERO); so->so_pcb = (caddr_t)rp; so->so_fibnum = td->td_proc->p_fibnum; error = raw_attach(so, proto); rp = sotorawcb(so); if (error) { so->so_pcb = NULL; free(rp, M_PCB); return error; } RTSOCK_LOCK(); switch(rp->rcb_proto.sp_protocol) { case AF_INET: V_route_cb.ip_count++; break; case AF_INET6: V_route_cb.ip6_count++; break; } V_route_cb.any_count++; RTSOCK_UNLOCK(); soisconnected(so); so->so_options |= SO_USELOOPBACK; return 0; } static int rts_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return (raw_usrreqs.pru_bind(so, nam, td)); /* xxx just EINVAL */ } static int rts_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (raw_usrreqs.pru_connect(so, nam, td)); /* XXX just EINVAL */ } /* pru_connect2 is EOPNOTSUPP */ /* pru_control is EOPNOTSUPP */ static void rts_detach(struct socket *so) { struct rawcb *rp = sotorawcb(so); KASSERT(rp != NULL, ("rts_detach: rp == NULL")); RTSOCK_LOCK(); switch(rp->rcb_proto.sp_protocol) { case AF_INET: V_route_cb.ip_count--; break; case AF_INET6: V_route_cb.ip6_count--; break; } V_route_cb.any_count--; RTSOCK_UNLOCK(); raw_usrreqs.pru_detach(so); } static int rts_disconnect(struct socket *so) { return (raw_usrreqs.pru_disconnect(so)); } /* pru_listen is EOPNOTSUPP */ static int rts_peeraddr(struct socket *so, struct sockaddr **nam) { return (raw_usrreqs.pru_peeraddr(so, nam)); } /* pru_rcvd is EOPNOTSUPP */ /* pru_rcvoob is EOPNOTSUPP */ static int rts_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { return (raw_usrreqs.pru_send(so, flags, m, nam, control, td)); } /* pru_sense is null */ static int rts_shutdown(struct socket *so) { return (raw_usrreqs.pru_shutdown(so)); } static int rts_sockaddr(struct socket *so, struct sockaddr **nam) { return (raw_usrreqs.pru_sockaddr(so, nam)); } static struct pr_usrreqs route_usrreqs = { .pru_abort = rts_abort, .pru_attach = rts_attach, .pru_bind = rts_bind, .pru_connect = rts_connect, .pru_detach = rts_detach, .pru_disconnect = rts_disconnect, .pru_peeraddr = rts_peeraddr, .pru_send = rts_send, .pru_shutdown = rts_shutdown, .pru_sockaddr = rts_sockaddr, .pru_close = rts_close, }; #ifndef _SOCKADDR_UNION_DEFINED #define _SOCKADDR_UNION_DEFINED /* * The union of all possible address formats we handle. */ union sockaddr_union { struct sockaddr sa; struct sockaddr_in sin; struct sockaddr_in6 sin6; }; #endif /* _SOCKADDR_UNION_DEFINED */ static int rtm_get_jailed(struct rt_addrinfo *info, struct ifnet *ifp, struct rtentry *rt, union sockaddr_union *saun, struct ucred *cred) { /* First, see if the returned address is part of the jail. */ if (prison_if(cred, rt->rt_ifa->ifa_addr) == 0) { info->rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr; return (0); } switch (info->rti_info[RTAX_DST]->sa_family) { #ifdef INET case AF_INET: { struct in_addr ia; struct ifaddr *ifa; int found; found = 0; /* * Try to find an address on the given outgoing interface * that belongs to the jail. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct sockaddr *sa; sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; ia = ((struct sockaddr_in *)sa)->sin_addr; if (prison_check_ip4(cred, &ia) == 0) { found = 1; break; } } IF_ADDR_RUNLOCK(ifp); if (!found) { /* * As a last resort return the 'default' jail address. */ ia = ((struct sockaddr_in *)rt->rt_ifa->ifa_addr)-> sin_addr; if (prison_get_ip4(cred, &ia) != 0) return (ESRCH); } bzero(&saun->sin, sizeof(struct sockaddr_in)); saun->sin.sin_len = sizeof(struct sockaddr_in); saun->sin.sin_family = AF_INET; saun->sin.sin_addr.s_addr = ia.s_addr; info->rti_info[RTAX_IFA] = (struct sockaddr *)&saun->sin; break; } #endif #ifdef INET6 case AF_INET6: { struct in6_addr ia6; struct ifaddr *ifa; int found; found = 0; /* * Try to find an address on the given outgoing interface * that belongs to the jail. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct sockaddr *sa; sa = ifa->ifa_addr; if (sa->sa_family != AF_INET6) continue; bcopy(&((struct sockaddr_in6 *)sa)->sin6_addr, &ia6, sizeof(struct in6_addr)); if (prison_check_ip6(cred, &ia6) == 0) { found = 1; break; } } IF_ADDR_RUNLOCK(ifp); if (!found) { /* * As a last resort return the 'default' jail address. */ ia6 = ((struct sockaddr_in6 *)rt->rt_ifa->ifa_addr)-> sin6_addr; if (prison_get_ip6(cred, &ia6) != 0) return (ESRCH); } bzero(&saun->sin6, sizeof(struct sockaddr_in6)); saun->sin6.sin6_len = sizeof(struct sockaddr_in6); saun->sin6.sin6_family = AF_INET6; bcopy(&ia6, &saun->sin6.sin6_addr, sizeof(struct in6_addr)); if (sa6_recoverscope(&saun->sin6) != 0) return (ESRCH); info->rti_info[RTAX_IFA] = (struct sockaddr *)&saun->sin6; break; } #endif default: return (ESRCH); } return (0); } /*ARGSUSED*/ static int route_output(struct mbuf *m, struct socket *so, ...) { RIB_RLOCK_TRACKER; struct rt_msghdr *rtm = NULL; struct rtentry *rt = NULL; struct rib_head *rnh; struct rt_addrinfo info; struct sockaddr_storage ss; #ifdef INET6 struct sockaddr_in6 *sin6; int i, rti_need_deembed = 0; #endif int alloc_len = 0, len, error = 0, fibnum; struct ifnet *ifp = NULL; union sockaddr_union saun; sa_family_t saf = AF_UNSPEC; struct rawcb *rp = NULL; struct walkarg w; fibnum = so->so_fibnum; #define senderr(e) { error = e; goto flush;} if (m == NULL || ((m->m_len < sizeof(long)) && (m = m_pullup(m, sizeof(long))) == NULL)) return (ENOBUFS); if ((m->m_flags & M_PKTHDR) == 0) panic("route_output"); len = m->m_pkthdr.len; if (len < sizeof(*rtm) || len != mtod(m, struct rt_msghdr *)->rtm_msglen) senderr(EINVAL); /* * Most of current messages are in range 200-240 bytes, * minimize possible re-allocation on reply using larger size * buffer aligned on 1k boundaty. */ alloc_len = roundup2(len, 1024); if ((rtm = malloc(alloc_len, M_TEMP, M_NOWAIT)) == NULL) senderr(ENOBUFS); m_copydata(m, 0, len, (caddr_t)rtm); bzero(&info, sizeof(info)); bzero(&w, sizeof(w)); if (rtm->rtm_version != RTM_VERSION) { /* Do not touch message since format is unknown */ free(rtm, M_TEMP); rtm = NULL; senderr(EPROTONOSUPPORT); } /* * Starting from here, it is possible * to alter original message and insert * caller PID and error value. */ rtm->rtm_pid = curproc->p_pid; info.rti_addrs = rtm->rtm_addrs; info.rti_mflags = rtm->rtm_inits; info.rti_rmx = &rtm->rtm_rmx; /* * rt_xaddrs() performs s6_addr[2] := sin6_scope_id for AF_INET6 * link-local address because rtrequest requires addresses with * embedded scope id. */ if (rt_xaddrs((caddr_t)(rtm + 1), len + (caddr_t)rtm, &info)) senderr(EINVAL); info.rti_flags = rtm->rtm_flags; if (info.rti_info[RTAX_DST] == NULL || info.rti_info[RTAX_DST]->sa_family >= AF_MAX || (info.rti_info[RTAX_GATEWAY] != NULL && info.rti_info[RTAX_GATEWAY]->sa_family >= AF_MAX)) senderr(EINVAL); saf = info.rti_info[RTAX_DST]->sa_family; /* * Verify that the caller has the appropriate privilege; RTM_GET * is the only operation the non-superuser is allowed. */ if (rtm->rtm_type != RTM_GET) { error = priv_check(curthread, PRIV_NET_ROUTE); if (error) senderr(error); } /* * The given gateway address may be an interface address. * For example, issuing a "route change" command on a route * entry that was created from a tunnel, and the gateway * address given is the local end point. In this case the * RTF_GATEWAY flag must be cleared or the destination will * not be reachable even though there is no error message. */ if (info.rti_info[RTAX_GATEWAY] != NULL && info.rti_info[RTAX_GATEWAY]->sa_family != AF_LINK) { struct rt_addrinfo ginfo; struct sockaddr *gdst; bzero(&ginfo, sizeof(ginfo)); bzero(&ss, sizeof(ss)); ss.ss_len = sizeof(ss); ginfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&ss; gdst = info.rti_info[RTAX_GATEWAY]; /* * A host route through the loopback interface is * installed for each interface adddress. In pre 8.0 * releases the interface address of a PPP link type * is not reachable locally. This behavior is fixed as * part of the new L2/L3 redesign and rewrite work. The * signature of this interface address route is the * AF_LINK sa_family type of the rt_gateway, and the * rt_ifp has the IFF_LOOPBACK flag set. */ if (rib_lookup_info(fibnum, gdst, NHR_REF, 0, &ginfo) == 0) { if (ss.ss_family == AF_LINK && ginfo.rti_ifp->if_flags & IFF_LOOPBACK) { info.rti_flags &= ~RTF_GATEWAY; info.rti_flags |= RTF_GWFLAG_COMPAT; } rib_free_info(&ginfo); } } switch (rtm->rtm_type) { struct rtentry *saved_nrt; case RTM_ADD: case RTM_CHANGE: if (rtm->rtm_type == RTM_ADD) { if (info.rti_info[RTAX_GATEWAY] == NULL) senderr(EINVAL); } saved_nrt = NULL; /* support for new ARP code */ if (info.rti_info[RTAX_GATEWAY] != NULL && info.rti_info[RTAX_GATEWAY]->sa_family == AF_LINK && (rtm->rtm_flags & RTF_LLDATA) != 0) { error = lla_rt_output(rtm, &info); #ifdef INET6 if (error == 0) rti_need_deembed = (V_deembed_scopeid) ? 1 : 0; #endif break; } error = rtrequest1_fib(rtm->rtm_type, &info, &saved_nrt, fibnum); if (error == 0 && saved_nrt != NULL) { #ifdef INET6 rti_need_deembed = (V_deembed_scopeid) ? 1 : 0; #endif RT_LOCK(saved_nrt); rtm->rtm_index = saved_nrt->rt_ifp->if_index; RT_REMREF(saved_nrt); RT_UNLOCK(saved_nrt); } break; case RTM_DELETE: saved_nrt = NULL; /* support for new ARP code */ if (info.rti_info[RTAX_GATEWAY] && (info.rti_info[RTAX_GATEWAY]->sa_family == AF_LINK) && (rtm->rtm_flags & RTF_LLDATA) != 0) { error = lla_rt_output(rtm, &info); #ifdef INET6 if (error == 0) rti_need_deembed = (V_deembed_scopeid) ? 1 : 0; #endif break; } error = rtrequest1_fib(RTM_DELETE, &info, &saved_nrt, fibnum); if (error == 0) { RT_LOCK(saved_nrt); rt = saved_nrt; goto report; } #ifdef INET6 /* rt_msg2() will not be used when RTM_DELETE fails. */ rti_need_deembed = (V_deembed_scopeid) ? 1 : 0; #endif break; case RTM_GET: rnh = rt_tables_get_rnh(fibnum, saf); if (rnh == NULL) senderr(EAFNOSUPPORT); RIB_RLOCK(rnh); if (info.rti_info[RTAX_NETMASK] == NULL && rtm->rtm_type == RTM_GET) { /* * Provide longest prefix match for * address lookup (no mask). * 'route -n get addr' */ rt = (struct rtentry *) rnh->rnh_matchaddr( info.rti_info[RTAX_DST], &rnh->head); } else rt = (struct rtentry *) rnh->rnh_lookup( info.rti_info[RTAX_DST], info.rti_info[RTAX_NETMASK], &rnh->head); if (rt == NULL) { RIB_RUNLOCK(rnh); senderr(ESRCH); } #ifdef RADIX_MPATH /* * for RTM_CHANGE/LOCK, if we got multipath routes, * we require users to specify a matching RTAX_GATEWAY. * * for RTM_GET, gate is optional even with multipath. * if gate == NULL the first match is returned. * (no need to call rt_mpath_matchgate if gate == NULL) */ if (rt_mpath_capable(rnh) && (rtm->rtm_type != RTM_GET || info.rti_info[RTAX_GATEWAY])) { rt = rt_mpath_matchgate(rt, info.rti_info[RTAX_GATEWAY]); if (!rt) { RIB_RUNLOCK(rnh); senderr(ESRCH); } } #endif /* * If performing proxied L2 entry insertion, and * the actual PPP host entry is found, perform * another search to retrieve the prefix route of * the local end point of the PPP link. */ if (rtm->rtm_flags & RTF_ANNOUNCE) { struct sockaddr laddr; if (rt->rt_ifp != NULL && rt->rt_ifp->if_type == IFT_PROPVIRTUAL) { struct ifaddr *ifa; NET_EPOCH_ENTER(); ifa = ifa_ifwithnet(info.rti_info[RTAX_DST], 1, RT_ALL_FIBS); if (ifa != NULL) rt_maskedcopy(ifa->ifa_addr, &laddr, ifa->ifa_netmask); NET_EPOCH_EXIT(); } else rt_maskedcopy(rt->rt_ifa->ifa_addr, &laddr, rt->rt_ifa->ifa_netmask); /* * refactor rt and no lock operation necessary */ rt = (struct rtentry *)rnh->rnh_matchaddr(&laddr, &rnh->head); if (rt == NULL) { RIB_RUNLOCK(rnh); senderr(ESRCH); } } RT_LOCK(rt); RT_ADDREF(rt); RIB_RUNLOCK(rnh); report: RT_LOCK_ASSERT(rt); if ((rt->rt_flags & RTF_HOST) == 0 ? jailed_without_vnet(curthread->td_ucred) : prison_if(curthread->td_ucred, rt_key(rt)) != 0) { RT_UNLOCK(rt); senderr(ESRCH); } info.rti_info[RTAX_DST] = rt_key(rt); info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; info.rti_info[RTAX_NETMASK] = rtsock_fix_netmask(rt_key(rt), rt_mask(rt), &ss); info.rti_info[RTAX_GENMASK] = 0; if (rtm->rtm_addrs & (RTA_IFP | RTA_IFA)) { ifp = rt->rt_ifp; if (ifp) { info.rti_info[RTAX_IFP] = ifp->if_addr->ifa_addr; error = rtm_get_jailed(&info, ifp, rt, &saun, curthread->td_ucred); if (error != 0) { RT_UNLOCK(rt); senderr(error); } if (ifp->if_flags & IFF_POINTOPOINT) info.rti_info[RTAX_BRD] = rt->rt_ifa->ifa_dstaddr; rtm->rtm_index = ifp->if_index; } else { info.rti_info[RTAX_IFP] = NULL; info.rti_info[RTAX_IFA] = NULL; } } else if ((ifp = rt->rt_ifp) != NULL) { rtm->rtm_index = ifp->if_index; } /* Check if we need to realloc storage */ rtsock_msg_buffer(rtm->rtm_type, &info, NULL, &len); if (len > alloc_len) { struct rt_msghdr *new_rtm; new_rtm = malloc(len, M_TEMP, M_NOWAIT); if (new_rtm == NULL) { RT_UNLOCK(rt); senderr(ENOBUFS); } bcopy(rtm, new_rtm, rtm->rtm_msglen); free(rtm, M_TEMP); rtm = new_rtm; alloc_len = len; } w.w_tmem = (caddr_t)rtm; w.w_tmemsize = alloc_len; rtsock_msg_buffer(rtm->rtm_type, &info, &w, &len); if (rt->rt_flags & RTF_GWFLAG_COMPAT) rtm->rtm_flags = RTF_GATEWAY | (rt->rt_flags & ~RTF_GWFLAG_COMPAT); else rtm->rtm_flags = rt->rt_flags; rt_getmetrics(rt, &rtm->rtm_rmx); rtm->rtm_addrs = info.rti_addrs; RT_UNLOCK(rt); break; default: senderr(EOPNOTSUPP); } flush: if (rt != NULL) RTFREE(rt); /* * Check to see if we don't want our own messages. */ if ((so->so_options & SO_USELOOPBACK) == 0) { if (V_route_cb.any_count <= 1) { if (rtm != NULL) free(rtm, M_TEMP); m_freem(m); return (error); } /* There is another listener, so construct message */ rp = sotorawcb(so); } if (rtm != NULL) { #ifdef INET6 if (rti_need_deembed) { /* sin6_scope_id is recovered before sending rtm. */ sin6 = (struct sockaddr_in6 *)&ss; for (i = 0; i < RTAX_MAX; i++) { if (info.rti_info[i] == NULL) continue; if (info.rti_info[i]->sa_family != AF_INET6) continue; bcopy(info.rti_info[i], sin6, sizeof(*sin6)); if (sa6_recoverscope(sin6) == 0) bcopy(sin6, info.rti_info[i], sizeof(*sin6)); } } #endif if (error != 0) rtm->rtm_errno = error; else rtm->rtm_flags |= RTF_DONE; m_copyback(m, 0, rtm->rtm_msglen, (caddr_t)rtm); if (m->m_pkthdr.len < rtm->rtm_msglen) { m_freem(m); m = NULL; } else if (m->m_pkthdr.len > rtm->rtm_msglen) m_adj(m, rtm->rtm_msglen - m->m_pkthdr.len); free(rtm, M_TEMP); } if (m != NULL) { M_SETFIB(m, fibnum); m->m_flags |= RTS_FILTER_FIB; if (rp) { /* * XXX insure we don't get a copy by * invalidating our protocol */ unsigned short family = rp->rcb_proto.sp_family; rp->rcb_proto.sp_family = 0; rt_dispatch(m, saf); rp->rcb_proto.sp_family = family; } else rt_dispatch(m, saf); } return (error); } static void rt_getmetrics(const struct rtentry *rt, struct rt_metrics *out) { bzero(out, sizeof(*out)); out->rmx_mtu = rt->rt_mtu; out->rmx_weight = rt->rt_weight; out->rmx_pksent = counter_u64_fetch(rt->rt_pksent); /* Kernel -> userland timebase conversion. */ out->rmx_expire = rt->rt_expire ? rt->rt_expire - time_uptime + time_second : 0; } /* * Extract the addresses of the passed sockaddrs. * Do a little sanity checking so as to avoid bad memory references. * This data is derived straight from userland. */ static int rt_xaddrs(caddr_t cp, caddr_t cplim, struct rt_addrinfo *rtinfo) { struct sockaddr *sa; int i; for (i = 0; i < RTAX_MAX && cp < cplim; i++) { if ((rtinfo->rti_addrs & (1 << i)) == 0) continue; sa = (struct sockaddr *)cp; /* * It won't fit. */ if (cp + sa->sa_len > cplim) return (EINVAL); /* * there are no more.. quit now * If there are more bits, they are in error. * I've seen this. route(1) can evidently generate these. * This causes kernel to core dump. * for compatibility, If we see this, point to a safe address. */ if (sa->sa_len == 0) { rtinfo->rti_info[i] = &sa_zero; return (0); /* should be EINVAL but for compat */ } /* accept it */ #ifdef INET6 if (sa->sa_family == AF_INET6) sa6_embedscope((struct sockaddr_in6 *)sa, V_ip6_use_defzone); #endif rtinfo->rti_info[i] = sa; cp += SA_SIZE(sa); } return (0); } /* * Fill in @dmask with valid netmask leaving original @smask * intact. Mostly used with radix netmasks. */ static struct sockaddr * rtsock_fix_netmask(struct sockaddr *dst, struct sockaddr *smask, struct sockaddr_storage *dmask) { if (dst == NULL || smask == NULL) return (NULL); memset(dmask, 0, dst->sa_len); memcpy(dmask, smask, smask->sa_len); dmask->ss_len = dst->sa_len; dmask->ss_family = dst->sa_family; return ((struct sockaddr *)dmask); } /* * Writes information related to @rtinfo object to newly-allocated mbuf. * Assumes MCLBYTES is enough to construct any message. * Used for OS notifications of vaious events (if/ifa announces,etc) * * Returns allocated mbuf or NULL on failure. */ static struct mbuf * rtsock_msg_mbuf(int type, struct rt_addrinfo *rtinfo) { struct rt_msghdr *rtm; struct mbuf *m; int i; struct sockaddr *sa; #ifdef INET6 struct sockaddr_storage ss; struct sockaddr_in6 *sin6; #endif int len, dlen; switch (type) { case RTM_DELADDR: case RTM_NEWADDR: len = sizeof(struct ifa_msghdr); break; case RTM_DELMADDR: case RTM_NEWMADDR: len = sizeof(struct ifma_msghdr); break; case RTM_IFINFO: len = sizeof(struct if_msghdr); break; case RTM_IFANNOUNCE: case RTM_IEEE80211: len = sizeof(struct if_announcemsghdr); break; default: len = sizeof(struct rt_msghdr); } /* XXXGL: can we use MJUMPAGESIZE cluster here? */ KASSERT(len <= MCLBYTES, ("%s: message too big", __func__)); if (len > MHLEN) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (m); m->m_pkthdr.len = m->m_len = len; rtm = mtod(m, struct rt_msghdr *); bzero((caddr_t)rtm, len); for (i = 0; i < RTAX_MAX; i++) { if ((sa = rtinfo->rti_info[i]) == NULL) continue; rtinfo->rti_addrs |= (1 << i); dlen = SA_SIZE(sa); #ifdef INET6 if (V_deembed_scopeid && sa->sa_family == AF_INET6) { sin6 = (struct sockaddr_in6 *)&ss; bcopy(sa, sin6, sizeof(*sin6)); if (sa6_recoverscope(sin6) == 0) sa = (struct sockaddr *)sin6; } #endif m_copyback(m, len, dlen, (caddr_t)sa); len += dlen; } if (m->m_pkthdr.len != len) { m_freem(m); return (NULL); } rtm->rtm_msglen = len; rtm->rtm_version = RTM_VERSION; rtm->rtm_type = type; return (m); } /* * Writes information related to @rtinfo object to preallocated buffer. * Stores needed size in @plen. If @w is NULL, calculates size without * writing. * Used for sysctl dumps and rtsock answers (RTM_DEL/RTM_GET) generation. * * Returns 0 on success. * */ static int rtsock_msg_buffer(int type, struct rt_addrinfo *rtinfo, struct walkarg *w, int *plen) { int i; int len, buflen = 0, dlen; caddr_t cp = NULL; struct rt_msghdr *rtm = NULL; #ifdef INET6 struct sockaddr_storage ss; struct sockaddr_in6 *sin6; #endif #ifdef COMPAT_FREEBSD32 bool compat32 = false; #endif switch (type) { case RTM_DELADDR: case RTM_NEWADDR: if (w != NULL && w->w_op == NET_RT_IFLISTL) { #ifdef COMPAT_FREEBSD32 if (w->w_req->flags & SCTL_MASK32) { len = sizeof(struct ifa_msghdrl32); compat32 = true; } else #endif len = sizeof(struct ifa_msghdrl); } else len = sizeof(struct ifa_msghdr); break; case RTM_IFINFO: #ifdef COMPAT_FREEBSD32 if (w != NULL && w->w_req->flags & SCTL_MASK32) { if (w->w_op == NET_RT_IFLISTL) len = sizeof(struct if_msghdrl32); else len = sizeof(struct if_msghdr32); compat32 = true; break; } #endif if (w != NULL && w->w_op == NET_RT_IFLISTL) len = sizeof(struct if_msghdrl); else len = sizeof(struct if_msghdr); break; case RTM_NEWMADDR: len = sizeof(struct ifma_msghdr); break; default: len = sizeof(struct rt_msghdr); } if (w != NULL) { rtm = (struct rt_msghdr *)w->w_tmem; buflen = w->w_tmemsize - len; cp = (caddr_t)w->w_tmem + len; } rtinfo->rti_addrs = 0; for (i = 0; i < RTAX_MAX; i++) { struct sockaddr *sa; if ((sa = rtinfo->rti_info[i]) == NULL) continue; rtinfo->rti_addrs |= (1 << i); #ifdef COMPAT_FREEBSD32 if (compat32) dlen = SA_SIZE32(sa); else #endif dlen = SA_SIZE(sa); if (cp != NULL && buflen >= dlen) { #ifdef INET6 if (V_deembed_scopeid && sa->sa_family == AF_INET6) { sin6 = (struct sockaddr_in6 *)&ss; bcopy(sa, sin6, sizeof(*sin6)); if (sa6_recoverscope(sin6) == 0) sa = (struct sockaddr *)sin6; } #endif bcopy((caddr_t)sa, cp, (unsigned)dlen); cp += dlen; buflen -= dlen; } else if (cp != NULL) { /* * Buffer too small. Count needed size * and return with error. */ cp = NULL; } len += dlen; } if (cp != NULL) { dlen = ALIGN(len) - len; if (buflen < dlen) cp = NULL; else buflen -= dlen; } len = ALIGN(len); if (cp != NULL) { /* fill header iff buffer is large enough */ rtm->rtm_version = RTM_VERSION; rtm->rtm_type = type; rtm->rtm_msglen = len; } *plen = len; if (w != NULL && cp == NULL) return (ENOBUFS); return (0); } /* * This routine is called to generate a message from the routing * socket indicating that a redirect has occurred, a routing lookup * has failed, or that a protocol has detected timeouts to a particular * destination. */ void rt_missmsg_fib(int type, struct rt_addrinfo *rtinfo, int flags, int error, int fibnum) { struct rt_msghdr *rtm; struct mbuf *m; struct sockaddr *sa = rtinfo->rti_info[RTAX_DST]; if (V_route_cb.any_count == 0) return; m = rtsock_msg_mbuf(type, rtinfo); if (m == NULL) return; if (fibnum != RT_ALL_FIBS) { KASSERT(fibnum >= 0 && fibnum < rt_numfibs, ("%s: fibnum out " "of range 0 <= %d < %d", __func__, fibnum, rt_numfibs)); M_SETFIB(m, fibnum); m->m_flags |= RTS_FILTER_FIB; } rtm = mtod(m, struct rt_msghdr *); rtm->rtm_flags = RTF_DONE | flags; rtm->rtm_errno = error; rtm->rtm_addrs = rtinfo->rti_addrs; rt_dispatch(m, sa ? sa->sa_family : AF_UNSPEC); } void rt_missmsg(int type, struct rt_addrinfo *rtinfo, int flags, int error) { rt_missmsg_fib(type, rtinfo, flags, error, RT_ALL_FIBS); } /* * This routine is called to generate a message from the routing * socket indicating that the status of a network interface has changed. */ void rt_ifmsg(struct ifnet *ifp) { struct if_msghdr *ifm; struct mbuf *m; struct rt_addrinfo info; if (V_route_cb.any_count == 0) return; bzero((caddr_t)&info, sizeof(info)); m = rtsock_msg_mbuf(RTM_IFINFO, &info); if (m == NULL) return; ifm = mtod(m, struct if_msghdr *); ifm->ifm_index = ifp->if_index; ifm->ifm_flags = ifp->if_flags | ifp->if_drv_flags; if_data_copy(ifp, &ifm->ifm_data); ifm->ifm_addrs = 0; rt_dispatch(m, AF_UNSPEC); } /* * Announce interface address arrival/withdraw. * Please do not call directly, use rt_addrmsg(). * Assume input data to be valid. * Returns 0 on success. */ int rtsock_addrmsg(int cmd, struct ifaddr *ifa, int fibnum) { struct rt_addrinfo info; struct sockaddr *sa; int ncmd; struct mbuf *m; struct ifa_msghdr *ifam; struct ifnet *ifp = ifa->ifa_ifp; struct sockaddr_storage ss; if (V_route_cb.any_count == 0) return (0); ncmd = cmd == RTM_ADD ? RTM_NEWADDR : RTM_DELADDR; bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_IFA] = sa = ifa->ifa_addr; info.rti_info[RTAX_IFP] = ifp->if_addr->ifa_addr; info.rti_info[RTAX_NETMASK] = rtsock_fix_netmask( info.rti_info[RTAX_IFP], ifa->ifa_netmask, &ss); info.rti_info[RTAX_BRD] = ifa->ifa_dstaddr; if ((m = rtsock_msg_mbuf(ncmd, &info)) == NULL) return (ENOBUFS); ifam = mtod(m, struct ifa_msghdr *); ifam->ifam_index = ifp->if_index; ifam->ifam_metric = ifa->ifa_ifp->if_metric; ifam->ifam_flags = ifa->ifa_flags; ifam->ifam_addrs = info.rti_addrs; if (fibnum != RT_ALL_FIBS) { M_SETFIB(m, fibnum); m->m_flags |= RTS_FILTER_FIB; } rt_dispatch(m, sa ? sa->sa_family : AF_UNSPEC); return (0); } /* * Announce route addition/removal. * Please do not call directly, use rt_routemsg(). * Note that @rt data MAY be inconsistent/invalid: * if some userland app sends us "invalid" route message (invalid mask, * no dst, wrong address families, etc...) we need to pass it back * to app (and any other rtsock consumers) with rtm_errno field set to * non-zero value. * * Returns 0 on success. */ int rtsock_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt, int fibnum) { struct rt_addrinfo info; struct sockaddr *sa; struct mbuf *m; struct rt_msghdr *rtm; struct sockaddr_storage ss; if (V_route_cb.any_count == 0) return (0); bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_DST] = sa = rt_key(rt); info.rti_info[RTAX_NETMASK] = rtsock_fix_netmask(sa, rt_mask(rt), &ss); info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; if ((m = rtsock_msg_mbuf(cmd, &info)) == NULL) return (ENOBUFS); rtm = mtod(m, struct rt_msghdr *); rtm->rtm_index = ifp->if_index; rtm->rtm_flags |= rt->rt_flags; rtm->rtm_errno = error; rtm->rtm_addrs = info.rti_addrs; if (fibnum != RT_ALL_FIBS) { M_SETFIB(m, fibnum); m->m_flags |= RTS_FILTER_FIB; } rt_dispatch(m, sa ? sa->sa_family : AF_UNSPEC); return (0); } /* * This is the analogue to the rt_newaddrmsg which performs the same * function but for multicast group memberhips. This is easier since * there is no route state to worry about. */ void rt_newmaddrmsg(int cmd, struct ifmultiaddr *ifma) { struct rt_addrinfo info; struct mbuf *m = NULL; struct ifnet *ifp = ifma->ifma_ifp; struct ifma_msghdr *ifmam; if (V_route_cb.any_count == 0) return; bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_IFA] = ifma->ifma_addr; if (ifp && ifp->if_addr) info.rti_info[RTAX_IFP] = ifp->if_addr->ifa_addr; else info.rti_info[RTAX_IFP] = NULL; /* * If a link-layer address is present, present it as a ``gateway'' * (similarly to how ARP entries, e.g., are presented). */ info.rti_info[RTAX_GATEWAY] = ifma->ifma_lladdr; m = rtsock_msg_mbuf(cmd, &info); if (m == NULL) return; ifmam = mtod(m, struct ifma_msghdr *); KASSERT(ifp != NULL, ("%s: link-layer multicast address w/o ifp\n", __func__)); ifmam->ifmam_index = ifp->if_index; ifmam->ifmam_addrs = info.rti_addrs; rt_dispatch(m, ifma->ifma_addr ? ifma->ifma_addr->sa_family : AF_UNSPEC); } static struct mbuf * rt_makeifannouncemsg(struct ifnet *ifp, int type, int what, struct rt_addrinfo *info) { struct if_announcemsghdr *ifan; struct mbuf *m; if (V_route_cb.any_count == 0) return NULL; bzero((caddr_t)info, sizeof(*info)); m = rtsock_msg_mbuf(type, info); if (m != NULL) { ifan = mtod(m, struct if_announcemsghdr *); ifan->ifan_index = ifp->if_index; strlcpy(ifan->ifan_name, ifp->if_xname, sizeof(ifan->ifan_name)); ifan->ifan_what = what; } return m; } /* * This is called to generate routing socket messages indicating * IEEE80211 wireless events. * XXX we piggyback on the RTM_IFANNOUNCE msg format in a clumsy way. */ void rt_ieee80211msg(struct ifnet *ifp, int what, void *data, size_t data_len) { struct mbuf *m; struct rt_addrinfo info; m = rt_makeifannouncemsg(ifp, RTM_IEEE80211, what, &info); if (m != NULL) { /* * Append the ieee80211 data. Try to stick it in the * mbuf containing the ifannounce msg; otherwise allocate * a new mbuf and append. * * NB: we assume m is a single mbuf. */ if (data_len > M_TRAILINGSPACE(m)) { struct mbuf *n = m_get(M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return; } bcopy(data, mtod(n, void *), data_len); n->m_len = data_len; m->m_next = n; } else if (data_len > 0) { bcopy(data, mtod(m, u_int8_t *) + m->m_len, data_len); m->m_len += data_len; } if (m->m_flags & M_PKTHDR) m->m_pkthdr.len += data_len; mtod(m, struct if_announcemsghdr *)->ifan_msglen += data_len; rt_dispatch(m, AF_UNSPEC); } } /* * This is called to generate routing socket messages indicating * network interface arrival and departure. */ void rt_ifannouncemsg(struct ifnet *ifp, int what) { struct mbuf *m; struct rt_addrinfo info; m = rt_makeifannouncemsg(ifp, RTM_IFANNOUNCE, what, &info); if (m != NULL) rt_dispatch(m, AF_UNSPEC); } static void rt_dispatch(struct mbuf *m, sa_family_t saf) { struct m_tag *tag; /* * Preserve the family from the sockaddr, if any, in an m_tag for * use when injecting the mbuf into the routing socket buffer from * the netisr. */ if (saf != AF_UNSPEC) { tag = m_tag_get(PACKET_TAG_RTSOCKFAM, sizeof(unsigned short), M_NOWAIT); if (tag == NULL) { m_freem(m); return; } *(unsigned short *)(tag + 1) = saf; m_tag_prepend(m, tag); } #ifdef VIMAGE if (V_loif) m->m_pkthdr.rcvif = V_loif; else { m_freem(m); return; } #endif netisr_queue(NETISR_ROUTE, m); /* mbuf is free'd on failure. */ } /* * This is used in dumping the kernel table via sysctl(). */ static int sysctl_dumpentry(struct radix_node *rn, void *vw) { struct walkarg *w = vw; struct rtentry *rt = (struct rtentry *)rn; int error = 0, size; struct rt_addrinfo info; struct sockaddr_storage ss; if (w->w_op == NET_RT_FLAGS && !(rt->rt_flags & w->w_arg)) return 0; if ((rt->rt_flags & RTF_HOST) == 0 ? jailed_without_vnet(w->w_req->td->td_ucred) : prison_if(w->w_req->td->td_ucred, rt_key(rt)) != 0) return (0); bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_DST] = rt_key(rt); info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; info.rti_info[RTAX_NETMASK] = rtsock_fix_netmask(rt_key(rt), rt_mask(rt), &ss); info.rti_info[RTAX_GENMASK] = 0; if (rt->rt_ifp) { info.rti_info[RTAX_IFP] = rt->rt_ifp->if_addr->ifa_addr; info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr; if (rt->rt_ifp->if_flags & IFF_POINTOPOINT) info.rti_info[RTAX_BRD] = rt->rt_ifa->ifa_dstaddr; } if ((error = rtsock_msg_buffer(RTM_GET, &info, w, &size)) != 0) return (error); if (w->w_req && w->w_tmem) { struct rt_msghdr *rtm = (struct rt_msghdr *)w->w_tmem; if (rt->rt_flags & RTF_GWFLAG_COMPAT) rtm->rtm_flags = RTF_GATEWAY | (rt->rt_flags & ~RTF_GWFLAG_COMPAT); else rtm->rtm_flags = rt->rt_flags; rt_getmetrics(rt, &rtm->rtm_rmx); rtm->rtm_index = rt->rt_ifp->if_index; rtm->rtm_errno = rtm->rtm_pid = rtm->rtm_seq = 0; rtm->rtm_addrs = info.rti_addrs; error = SYSCTL_OUT(w->w_req, (caddr_t)rtm, size); return (error); } return (error); } static int sysctl_iflist_ifml(struct ifnet *ifp, const struct if_data *src_ifd, struct rt_addrinfo *info, struct walkarg *w, int len) { struct if_msghdrl *ifm; struct if_data *ifd; ifm = (struct if_msghdrl *)w->w_tmem; #ifdef COMPAT_FREEBSD32 if (w->w_req->flags & SCTL_MASK32) { struct if_msghdrl32 *ifm32; ifm32 = (struct if_msghdrl32 *)ifm; ifm32->ifm_addrs = info->rti_addrs; ifm32->ifm_flags = ifp->if_flags | ifp->if_drv_flags; ifm32->ifm_index = ifp->if_index; ifm32->_ifm_spare1 = 0; ifm32->ifm_len = sizeof(*ifm32); ifm32->ifm_data_off = offsetof(struct if_msghdrl32, ifm_data); ifd = &ifm32->ifm_data; } else #endif { ifm->ifm_addrs = info->rti_addrs; ifm->ifm_flags = ifp->if_flags | ifp->if_drv_flags; ifm->ifm_index = ifp->if_index; ifm->_ifm_spare1 = 0; ifm->ifm_len = sizeof(*ifm); ifm->ifm_data_off = offsetof(struct if_msghdrl, ifm_data); ifd = &ifm->ifm_data; } memcpy(ifd, src_ifd, sizeof(*ifd)); return (SYSCTL_OUT(w->w_req, (caddr_t)ifm, len)); } static int sysctl_iflist_ifm(struct ifnet *ifp, const struct if_data *src_ifd, struct rt_addrinfo *info, struct walkarg *w, int len) { struct if_msghdr *ifm; struct if_data *ifd; ifm = (struct if_msghdr *)w->w_tmem; #ifdef COMPAT_FREEBSD32 if (w->w_req->flags & SCTL_MASK32) { struct if_msghdr32 *ifm32; ifm32 = (struct if_msghdr32 *)ifm; ifm32->ifm_addrs = info->rti_addrs; ifm32->ifm_flags = ifp->if_flags | ifp->if_drv_flags; ifm32->ifm_index = ifp->if_index; ifd = &ifm32->ifm_data; } else #endif { ifm->ifm_addrs = info->rti_addrs; ifm->ifm_flags = ifp->if_flags | ifp->if_drv_flags; ifm->ifm_index = ifp->if_index; ifd = &ifm->ifm_data; } memcpy(ifd, src_ifd, sizeof(*ifd)); return (SYSCTL_OUT(w->w_req, (caddr_t)ifm, len)); } static int sysctl_iflist_ifaml(struct ifaddr *ifa, struct rt_addrinfo *info, struct walkarg *w, int len) { struct ifa_msghdrl *ifam; struct if_data *ifd; ifam = (struct ifa_msghdrl *)w->w_tmem; #ifdef COMPAT_FREEBSD32 if (w->w_req->flags & SCTL_MASK32) { struct ifa_msghdrl32 *ifam32; ifam32 = (struct ifa_msghdrl32 *)ifam; ifam32->ifam_addrs = info->rti_addrs; ifam32->ifam_flags = ifa->ifa_flags; ifam32->ifam_index = ifa->ifa_ifp->if_index; ifam32->_ifam_spare1 = 0; ifam32->ifam_len = sizeof(*ifam32); ifam32->ifam_data_off = offsetof(struct ifa_msghdrl32, ifam_data); ifam32->ifam_metric = ifa->ifa_ifp->if_metric; ifd = &ifam32->ifam_data; } else #endif { ifam->ifam_addrs = info->rti_addrs; ifam->ifam_flags = ifa->ifa_flags; ifam->ifam_index = ifa->ifa_ifp->if_index; ifam->_ifam_spare1 = 0; ifam->ifam_len = sizeof(*ifam); ifam->ifam_data_off = offsetof(struct ifa_msghdrl, ifam_data); ifam->ifam_metric = ifa->ifa_ifp->if_metric; ifd = &ifam->ifam_data; } bzero(ifd, sizeof(*ifd)); ifd->ifi_datalen = sizeof(struct if_data); ifd->ifi_ipackets = counter_u64_fetch(ifa->ifa_ipackets); ifd->ifi_opackets = counter_u64_fetch(ifa->ifa_opackets); ifd->ifi_ibytes = counter_u64_fetch(ifa->ifa_ibytes); ifd->ifi_obytes = counter_u64_fetch(ifa->ifa_obytes); /* Fixup if_data carp(4) vhid. */ if (carp_get_vhid_p != NULL) ifd->ifi_vhid = (*carp_get_vhid_p)(ifa); return (SYSCTL_OUT(w->w_req, w->w_tmem, len)); } static int sysctl_iflist_ifam(struct ifaddr *ifa, struct rt_addrinfo *info, struct walkarg *w, int len) { struct ifa_msghdr *ifam; ifam = (struct ifa_msghdr *)w->w_tmem; ifam->ifam_addrs = info->rti_addrs; ifam->ifam_flags = ifa->ifa_flags; ifam->ifam_index = ifa->ifa_ifp->if_index; ifam->ifam_metric = ifa->ifa_ifp->if_metric; return (SYSCTL_OUT(w->w_req, w->w_tmem, len)); } static int sysctl_iflist(int af, struct walkarg *w) { struct ifnet *ifp; struct ifaddr *ifa; struct if_data ifd; struct rt_addrinfo info; int len, error = 0; struct sockaddr_storage ss; struct epoch_tracker et; bzero((caddr_t)&info, sizeof(info)); bzero(&ifd, sizeof(ifd)); NET_EPOCH_ENTER_ET(et); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if (w->w_arg && w->w_arg != ifp->if_index) continue; if_data_copy(ifp, &ifd); ifa = ifp->if_addr; info.rti_info[RTAX_IFP] = ifa->ifa_addr; error = rtsock_msg_buffer(RTM_IFINFO, &info, w, &len); if (error != 0) goto done; info.rti_info[RTAX_IFP] = NULL; if (w->w_req && w->w_tmem) { if (w->w_op == NET_RT_IFLISTL) error = sysctl_iflist_ifml(ifp, &ifd, &info, w, len); else error = sysctl_iflist_ifm(ifp, &ifd, &info, w, len); if (error) goto done; } while ((ifa = CK_STAILQ_NEXT(ifa, ifa_link)) != NULL) { if (af && af != ifa->ifa_addr->sa_family) continue; if (prison_if(w->w_req->td->td_ucred, ifa->ifa_addr) != 0) continue; info.rti_info[RTAX_IFA] = ifa->ifa_addr; info.rti_info[RTAX_NETMASK] = rtsock_fix_netmask( ifa->ifa_addr, ifa->ifa_netmask, &ss); info.rti_info[RTAX_BRD] = ifa->ifa_dstaddr; error = rtsock_msg_buffer(RTM_NEWADDR, &info, w, &len); if (error != 0) goto done; if (w->w_req && w->w_tmem) { if (w->w_op == NET_RT_IFLISTL) error = sysctl_iflist_ifaml(ifa, &info, w, len); else error = sysctl_iflist_ifam(ifa, &info, w, len); if (error) goto done; } } info.rti_info[RTAX_IFA] = NULL; info.rti_info[RTAX_NETMASK] = NULL; info.rti_info[RTAX_BRD] = NULL; } done: NET_EPOCH_EXIT_ET(et); return (error); } static int sysctl_ifmalist(int af, struct walkarg *w) { struct rt_addrinfo info; struct ifaddr *ifa; struct ifmultiaddr *ifma; struct ifnet *ifp; int error, len; error = 0; bzero((caddr_t)&info, sizeof(info)); IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if (w->w_arg && w->w_arg != ifp->if_index) continue; ifa = ifp->if_addr; info.rti_info[RTAX_IFP] = ifa ? ifa->ifa_addr : NULL; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (af && af != ifma->ifma_addr->sa_family) continue; if (prison_if(w->w_req->td->td_ucred, ifma->ifma_addr) != 0) continue; info.rti_info[RTAX_IFA] = ifma->ifma_addr; info.rti_info[RTAX_GATEWAY] = (ifma->ifma_addr->sa_family != AF_LINK) ? ifma->ifma_lladdr : NULL; error = rtsock_msg_buffer(RTM_NEWMADDR, &info, w, &len); if (error != 0) break; if (w->w_req && w->w_tmem) { struct ifma_msghdr *ifmam; ifmam = (struct ifma_msghdr *)w->w_tmem; ifmam->ifmam_index = ifma->ifma_ifp->if_index; ifmam->ifmam_flags = 0; ifmam->ifmam_addrs = info.rti_addrs; error = SYSCTL_OUT(w->w_req, w->w_tmem, len); if (error != 0) break; } } IF_ADDR_RUNLOCK(ifp); if (error != 0) break; } IFNET_RUNLOCK_NOSLEEP(); return (error); } static int sysctl_rtsock(SYSCTL_HANDLER_ARGS) { RIB_RLOCK_TRACKER; int *name = (int *)arg1; u_int namelen = arg2; struct rib_head *rnh = NULL; /* silence compiler. */ int i, lim, error = EINVAL; int fib = 0; u_char af; struct walkarg w; name ++; namelen--; if (req->newptr) return (EPERM); if (name[1] == NET_RT_DUMP) { if (namelen == 3) fib = req->td->td_proc->p_fibnum; else if (namelen == 4) fib = (name[3] == RT_ALL_FIBS) ? req->td->td_proc->p_fibnum : name[3]; else return ((namelen < 3) ? EISDIR : ENOTDIR); if (fib < 0 || fib >= rt_numfibs) return (EINVAL); } else if (namelen != 3) return ((namelen < 3) ? EISDIR : ENOTDIR); af = name[0]; if (af > AF_MAX) return (EINVAL); bzero(&w, sizeof(w)); w.w_op = name[1]; w.w_arg = name[2]; w.w_req = req; error = sysctl_wire_old_buffer(req, 0); if (error) return (error); /* * Allocate reply buffer in advance. * All rtsock messages has maximum length of u_short. */ w.w_tmemsize = 65536; w.w_tmem = malloc(w.w_tmemsize, M_TEMP, M_WAITOK); switch (w.w_op) { case NET_RT_DUMP: case NET_RT_FLAGS: if (af == 0) { /* dump all tables */ i = 1; lim = AF_MAX; } else /* dump only one table */ i = lim = af; /* * take care of llinfo entries, the caller must * specify an AF */ if (w.w_op == NET_RT_FLAGS && (w.w_arg == 0 || w.w_arg & RTF_LLINFO)) { if (af != 0) error = lltable_sysctl_dumparp(af, w.w_req); else error = EINVAL; break; } /* * take care of routing entries */ for (error = 0; error == 0 && i <= lim; i++) { rnh = rt_tables_get_rnh(fib, i); if (rnh != NULL) { RIB_RLOCK(rnh); error = rnh->rnh_walktree(&rnh->head, sysctl_dumpentry, &w); RIB_RUNLOCK(rnh); } else if (af != 0) error = EAFNOSUPPORT; } break; case NET_RT_IFLIST: case NET_RT_IFLISTL: error = sysctl_iflist(af, &w); break; case NET_RT_IFMALIST: error = sysctl_ifmalist(af, &w); break; } free(w.w_tmem, M_TEMP); return (error); } static SYSCTL_NODE(_net, PF_ROUTE, routetable, CTLFLAG_RD, sysctl_rtsock, ""); /* * Definitions of protocols supported in the ROUTE domain. */ static struct domain routedomain; /* or at least forward */ static struct protosw routesw[] = { { .pr_type = SOCK_RAW, .pr_domain = &routedomain, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_output = route_output, .pr_ctlinput = raw_ctlinput, .pr_init = raw_init, .pr_usrreqs = &route_usrreqs } }; static struct domain routedomain = { .dom_family = PF_ROUTE, .dom_name = "route", .dom_protosw = routesw, .dom_protoswNPROTOSW = &routesw[nitems(routesw)] }; VNET_DOMAIN_SET(route); Index: head/sys/net/vnet.c =================================================================== --- head/sys/net/vnet.c (revision 336675) +++ head/sys/net/vnet.c (revision 336676) @@ -1,805 +1,805 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2004-2009 University of Zagreb * Copyright (c) 2006-2009 FreeBSD Foundation * All rights reserved. * * This software was developed by the University of Zagreb and the * FreeBSD Foundation under sponsorship by the Stichting NLnet and the * FreeBSD Foundation. * * Copyright (c) 2009 Jeffrey Roberson * Copyright (c) 2009 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_kdb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #include #endif #include #include #include /*- * This file implements core functions for virtual network stacks: * * - Virtual network stack management functions. * * - Virtual network stack memory allocator, which virtualizes global * variables in the network stack * * - Virtualized SYSINIT's/SYSUNINIT's, which allow network stack subsystems * to register startup/shutdown events to be run for each virtual network * stack instance. */ FEATURE(vimage, "VIMAGE kernel virtualization"); static MALLOC_DEFINE(M_VNET, "vnet", "network stack control block"); /* * The virtual network stack list has two read-write locks, one sleepable and * the other not, so that the list can be stablized and walked in a variety * of network stack contexts. Both must be acquired exclusively to modify * the list, but a read lock of either lock is sufficient to walk the list. */ struct rwlock vnet_rwlock; struct sx vnet_sxlock; #define VNET_LIST_WLOCK() do { \ sx_xlock(&vnet_sxlock); \ rw_wlock(&vnet_rwlock); \ } while (0) #define VNET_LIST_WUNLOCK() do { \ rw_wunlock(&vnet_rwlock); \ sx_xunlock(&vnet_sxlock); \ } while (0) struct vnet_list_head vnet_head; struct vnet *vnet0; /* * The virtual network stack allocator provides storage for virtualized * global variables. These variables are defined/declared using the * VNET_DEFINE()/VNET_DECLARE() macros, which place them in the 'set_vnet' * linker set. The details of the implementation are somewhat subtle, but * allow the majority of most network subsystems to maintain * virtualization-agnostic. * * The virtual network stack allocator handles variables in the base kernel * vs. modules in similar but different ways. In both cases, virtualized * global variables are marked as such by being declared to be part of the * vnet linker set. These "master" copies of global variables serve two * functions: * * (1) They contain static initialization or "default" values for global * variables which will be propagated to each virtual network stack * instance when created. As with normal global variables, they default * to zero-filled. * * (2) They act as unique global names by which the variable can be referred * to, regardless of network stack instance. The single global symbol * will be used to calculate the location of a per-virtual instance * variable at run-time. * * Each virtual network stack instance has a complete copy of each * virtualized global variable, stored in a malloc'd block of memory * referred to by vnet->vnet_data_mem. Critical to the design is that each * per-instance memory block is laid out identically to the master block so * that the offset of each global variable is the same across all blocks. To * optimize run-time access, a precalculated 'base' address, * vnet->vnet_data_base, is stored in each vnet, and is the amount that can * be added to the address of a 'master' instance of a variable to get to the * per-vnet instance. * * Virtualized global variables are handled in a similar manner, but as each * module has its own 'set_vnet' linker set, and we want to keep all * virtualized globals togther, we reserve space in the kernel's linker set * for potential module variables using a per-vnet character array, * 'modspace'. The virtual network stack allocator maintains a free list to * track what space in the array is free (all, initially) and as modules are * linked, allocates portions of the space to specific globals. The kernel * module linker queries the virtual network stack allocator and will * bind references of the global to the location during linking. It also * calls into the virtual network stack allocator, once the memory is * initialized, in order to propagate the new static initializations to all * existing virtual network stack instances so that the soon-to-be executing * module will find every network stack instance with proper default values. */ /* * Number of bytes of data in the 'set_vnet' linker set, and hence the total * size of all kernel virtualized global variables, and the malloc(9) type * that will be used to allocate it. */ #define VNET_BYTES (VNET_STOP - VNET_START) static MALLOC_DEFINE(M_VNET_DATA, "vnet_data", "VNET data"); /* * VNET_MODMIN is the minimum number of bytes we will reserve for the sum of * global variables across all loaded modules. As this actually sizes an * array declared as a virtualized global variable in the kernel itself, and * we want the virtualized global variable space to be page-sized, we may * have more space than that in practice. */ #define VNET_MODMIN 8192 #define VNET_SIZE roundup2(VNET_BYTES, PAGE_SIZE) /* * Space to store virtualized global variables from loadable kernel modules, * and the free list to manage it. */ -static VNET_DEFINE(char, modspace[VNET_MODMIN]); +VNET_DEFINE_STATIC(char, modspace[VNET_MODMIN]); /* * Global lists of subsystem constructor and destructors for vnets. They are * registered via VNET_SYSINIT() and VNET_SYSUNINIT(). Both lists are * protected by the vnet_sysinit_sxlock global lock. */ static TAILQ_HEAD(vnet_sysinit_head, vnet_sysinit) vnet_constructors = TAILQ_HEAD_INITIALIZER(vnet_constructors); static TAILQ_HEAD(vnet_sysuninit_head, vnet_sysinit) vnet_destructors = TAILQ_HEAD_INITIALIZER(vnet_destructors); struct sx vnet_sysinit_sxlock; #define VNET_SYSINIT_WLOCK() sx_xlock(&vnet_sysinit_sxlock); #define VNET_SYSINIT_WUNLOCK() sx_xunlock(&vnet_sysinit_sxlock); #define VNET_SYSINIT_RLOCK() sx_slock(&vnet_sysinit_sxlock); #define VNET_SYSINIT_RUNLOCK() sx_sunlock(&vnet_sysinit_sxlock); struct vnet_data_free { uintptr_t vnd_start; int vnd_len; TAILQ_ENTRY(vnet_data_free) vnd_link; }; static MALLOC_DEFINE(M_VNET_DATA_FREE, "vnet_data_free", "VNET resource accounting"); static TAILQ_HEAD(, vnet_data_free) vnet_data_free_head = TAILQ_HEAD_INITIALIZER(vnet_data_free_head); static struct sx vnet_data_free_lock; SDT_PROVIDER_DEFINE(vnet); SDT_PROBE_DEFINE1(vnet, functions, vnet_alloc, entry, "int"); SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, alloc, "int", "struct vnet *"); SDT_PROBE_DEFINE2(vnet, functions, vnet_alloc, return, "int", "struct vnet *"); SDT_PROBE_DEFINE2(vnet, functions, vnet_destroy, entry, "int", "struct vnet *"); SDT_PROBE_DEFINE1(vnet, functions, vnet_destroy, return, "int"); #ifdef DDB static void db_show_vnet_print_vs(struct vnet_sysinit *, int); #endif /* * Allocate a virtual network stack. */ struct vnet * vnet_alloc(void) { struct vnet *vnet; SDT_PROBE1(vnet, functions, vnet_alloc, entry, __LINE__); vnet = malloc(sizeof(struct vnet), M_VNET, M_WAITOK | M_ZERO); vnet->vnet_magic_n = VNET_MAGIC_N; vnet->vnet_state = 0; SDT_PROBE2(vnet, functions, vnet_alloc, alloc, __LINE__, vnet); /* * Allocate storage for virtualized global variables and copy in * initial values form our 'master' copy. */ vnet->vnet_data_mem = malloc(VNET_SIZE, M_VNET_DATA, M_WAITOK); memcpy(vnet->vnet_data_mem, (void *)VNET_START, VNET_BYTES); /* * All use of vnet-specific data will immediately subtract VNET_START * from the base memory pointer, so pre-calculate that now to avoid * it on each use. */ vnet->vnet_data_base = (uintptr_t)vnet->vnet_data_mem - VNET_START; /* Initialize / attach vnet module instances. */ CURVNET_SET_QUIET(vnet); vnet_sysinit(); CURVNET_RESTORE(); VNET_LIST_WLOCK(); LIST_INSERT_HEAD(&vnet_head, vnet, vnet_le); VNET_LIST_WUNLOCK(); SDT_PROBE2(vnet, functions, vnet_alloc, return, __LINE__, vnet); return (vnet); } /* * Destroy a virtual network stack. */ void vnet_destroy(struct vnet *vnet) { SDT_PROBE2(vnet, functions, vnet_destroy, entry, __LINE__, vnet); KASSERT(vnet->vnet_sockcnt == 0, ("%s: vnet still has sockets", __func__)); VNET_LIST_WLOCK(); LIST_REMOVE(vnet, vnet_le); VNET_LIST_WUNLOCK(); CURVNET_SET_QUIET(vnet); vnet_sysuninit(); CURVNET_RESTORE(); /* * Release storage for the virtual network stack instance. */ free(vnet->vnet_data_mem, M_VNET_DATA); vnet->vnet_data_mem = NULL; vnet->vnet_data_base = 0; vnet->vnet_magic_n = 0xdeadbeef; free(vnet, M_VNET); SDT_PROBE1(vnet, functions, vnet_destroy, return, __LINE__); } /* * Boot time initialization and allocation of virtual network stacks. */ static void vnet_init_prelink(void *arg __unused) { rw_init(&vnet_rwlock, "vnet_rwlock"); sx_init(&vnet_sxlock, "vnet_sxlock"); sx_init(&vnet_sysinit_sxlock, "vnet_sysinit_sxlock"); LIST_INIT(&vnet_head); } SYSINIT(vnet_init_prelink, SI_SUB_VNET_PRELINK, SI_ORDER_FIRST, vnet_init_prelink, NULL); static void vnet0_init(void *arg __unused) { if (bootverbose) printf("VIMAGE (virtualized network stack) enabled\n"); /* * We MUST clear curvnet in vi_init_done() before going SMP, * otherwise CURVNET_SET() macros would scream about unnecessary * curvnet recursions. */ curvnet = prison0.pr_vnet = vnet0 = vnet_alloc(); } SYSINIT(vnet0_init, SI_SUB_VNET, SI_ORDER_FIRST, vnet0_init, NULL); static void vnet_init_done(void *unused __unused) { curvnet = NULL; } SYSINIT(vnet_init_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_init_done, NULL); /* * Once on boot, initialize the modspace freelist to entirely cover modspace. */ static void vnet_data_startup(void *dummy __unused) { struct vnet_data_free *df; df = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO); df->vnd_start = (uintptr_t)&VNET_NAME(modspace); df->vnd_len = VNET_MODMIN; TAILQ_INSERT_HEAD(&vnet_data_free_head, df, vnd_link); sx_init(&vnet_data_free_lock, "vnet_data alloc lock"); } SYSINIT(vnet_data, SI_SUB_KLD, SI_ORDER_FIRST, vnet_data_startup, NULL); /* Dummy VNET_SYSINIT to make sure we always reach the final end state. */ static void vnet_sysinit_done(void *unused __unused) { return; } VNET_SYSINIT(vnet_sysinit_done, SI_SUB_VNET_DONE, SI_ORDER_ANY, vnet_sysinit_done, NULL); /* * When a module is loaded and requires storage for a virtualized global * variable, allocate space from the modspace free list. This interface * should be used only by the kernel linker. */ void * vnet_data_alloc(int size) { struct vnet_data_free *df; void *s; s = NULL; size = roundup2(size, sizeof(void *)); sx_xlock(&vnet_data_free_lock); TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) { if (df->vnd_len < size) continue; if (df->vnd_len == size) { s = (void *)df->vnd_start; TAILQ_REMOVE(&vnet_data_free_head, df, vnd_link); free(df, M_VNET_DATA_FREE); break; } s = (void *)df->vnd_start; df->vnd_len -= size; df->vnd_start = df->vnd_start + size; break; } sx_xunlock(&vnet_data_free_lock); return (s); } /* * Free space for a virtualized global variable on module unload. */ void vnet_data_free(void *start_arg, int size) { struct vnet_data_free *df; struct vnet_data_free *dn; uintptr_t start; uintptr_t end; size = roundup2(size, sizeof(void *)); start = (uintptr_t)start_arg; end = start + size; /* * Free a region of space and merge it with as many neighbors as * possible. Keeping the list sorted simplifies this operation. */ sx_xlock(&vnet_data_free_lock); TAILQ_FOREACH(df, &vnet_data_free_head, vnd_link) { if (df->vnd_start > end) break; /* * If we expand at the end of an entry we may have to merge * it with the one following it as well. */ if (df->vnd_start + df->vnd_len == start) { df->vnd_len += size; dn = TAILQ_NEXT(df, vnd_link); if (df->vnd_start + df->vnd_len == dn->vnd_start) { df->vnd_len += dn->vnd_len; TAILQ_REMOVE(&vnet_data_free_head, dn, vnd_link); free(dn, M_VNET_DATA_FREE); } sx_xunlock(&vnet_data_free_lock); return; } if (df->vnd_start == end) { df->vnd_start = start; df->vnd_len += size; sx_xunlock(&vnet_data_free_lock); return; } } dn = malloc(sizeof(*df), M_VNET_DATA_FREE, M_WAITOK | M_ZERO); dn->vnd_start = start; dn->vnd_len = size; if (df) TAILQ_INSERT_BEFORE(df, dn, vnd_link); else TAILQ_INSERT_TAIL(&vnet_data_free_head, dn, vnd_link); sx_xunlock(&vnet_data_free_lock); } /* * When a new virtualized global variable has been allocated, propagate its * initial value to each already-allocated virtual network stack instance. */ void vnet_data_copy(void *start, int size) { struct vnet *vnet; VNET_LIST_RLOCK(); LIST_FOREACH(vnet, &vnet_head, vnet_le) memcpy((void *)((uintptr_t)vnet->vnet_data_base + (uintptr_t)start), start, size); VNET_LIST_RUNLOCK(); } /* * Support for special SYSINIT handlers registered via VNET_SYSINIT() * and VNET_SYSUNINIT(). */ void vnet_register_sysinit(void *arg) { struct vnet_sysinit *vs, *vs2; struct vnet *vnet; vs = arg; KASSERT(vs->subsystem > SI_SUB_VNET, ("vnet sysinit too early")); /* Add the constructor to the global list of vnet constructors. */ VNET_SYSINIT_WLOCK(); TAILQ_FOREACH(vs2, &vnet_constructors, link) { if (vs2->subsystem > vs->subsystem) break; if (vs2->subsystem == vs->subsystem && vs2->order > vs->order) break; } if (vs2 != NULL) TAILQ_INSERT_BEFORE(vs2, vs, link); else TAILQ_INSERT_TAIL(&vnet_constructors, vs, link); /* * Invoke the constructor on all the existing vnets when it is * registered. */ VNET_FOREACH(vnet) { CURVNET_SET_QUIET(vnet); vs->func(vs->arg); CURVNET_RESTORE(); } VNET_SYSINIT_WUNLOCK(); } void vnet_deregister_sysinit(void *arg) { struct vnet_sysinit *vs; vs = arg; /* Remove the constructor from the global list of vnet constructors. */ VNET_SYSINIT_WLOCK(); TAILQ_REMOVE(&vnet_constructors, vs, link); VNET_SYSINIT_WUNLOCK(); } void vnet_register_sysuninit(void *arg) { struct vnet_sysinit *vs, *vs2; vs = arg; /* Add the destructor to the global list of vnet destructors. */ VNET_SYSINIT_WLOCK(); TAILQ_FOREACH(vs2, &vnet_destructors, link) { if (vs2->subsystem > vs->subsystem) break; if (vs2->subsystem == vs->subsystem && vs2->order > vs->order) break; } if (vs2 != NULL) TAILQ_INSERT_BEFORE(vs2, vs, link); else TAILQ_INSERT_TAIL(&vnet_destructors, vs, link); VNET_SYSINIT_WUNLOCK(); } void vnet_deregister_sysuninit(void *arg) { struct vnet_sysinit *vs; struct vnet *vnet; vs = arg; /* * Invoke the destructor on all the existing vnets when it is * deregistered. */ VNET_SYSINIT_WLOCK(); VNET_FOREACH(vnet) { CURVNET_SET_QUIET(vnet); vs->func(vs->arg); CURVNET_RESTORE(); } /* Remove the destructor from the global list of vnet destructors. */ TAILQ_REMOVE(&vnet_destructors, vs, link); VNET_SYSINIT_WUNLOCK(); } /* * Invoke all registered vnet constructors on the current vnet. Used during * vnet construction. The caller is responsible for ensuring the new vnet is * the current vnet and that the vnet_sysinit_sxlock lock is locked. */ void vnet_sysinit(void) { struct vnet_sysinit *vs; VNET_SYSINIT_RLOCK(); TAILQ_FOREACH(vs, &vnet_constructors, link) { curvnet->vnet_state = vs->subsystem; vs->func(vs->arg); } VNET_SYSINIT_RUNLOCK(); } /* * Invoke all registered vnet destructors on the current vnet. Used during * vnet destruction. The caller is responsible for ensuring the dying vnet * the current vnet and that the vnet_sysinit_sxlock lock is locked. */ void vnet_sysuninit(void) { struct vnet_sysinit *vs; VNET_SYSINIT_RLOCK(); TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head, link) { curvnet->vnet_state = vs->subsystem; vs->func(vs->arg); } VNET_SYSINIT_RUNLOCK(); } /* * EVENTHANDLER(9) extensions. */ /* * Invoke the eventhandler function originally registered with the possibly * registered argument for all virtual network stack instances. * * This iterator can only be used for eventhandlers that do not take any * additional arguments, as we do ignore the variadic arguments from the * EVENTHANDLER_INVOKE() call. */ void vnet_global_eventhandler_iterator_func(void *arg, ...) { VNET_ITERATOR_DECL(vnet_iter); struct eventhandler_entry_vimage *v_ee; /* * There is a bug here in that we should actually cast things to * (struct eventhandler_entry_ ## name *) but that's not easily * possible in here so just re-using the variadic version we * defined for the generic vimage case. */ v_ee = arg; VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); ((vimage_iterator_func_t)v_ee->func)(v_ee->ee_arg); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); } #ifdef VNET_DEBUG struct vnet_recursion { SLIST_ENTRY(vnet_recursion) vnr_le; const char *prev_fn; const char *where_fn; int where_line; struct vnet *old_vnet; struct vnet *new_vnet; }; static SLIST_HEAD(, vnet_recursion) vnet_recursions = SLIST_HEAD_INITIALIZER(vnet_recursions); static void vnet_print_recursion(struct vnet_recursion *vnr, int brief) { if (!brief) printf("CURVNET_SET() recursion in "); printf("%s() line %d, prev in %s()", vnr->where_fn, vnr->where_line, vnr->prev_fn); if (brief) printf(", "); else printf("\n "); printf("%p -> %p\n", vnr->old_vnet, vnr->new_vnet); } void vnet_log_recursion(struct vnet *old_vnet, const char *old_fn, int line) { struct vnet_recursion *vnr; /* Skip already logged recursion events. */ SLIST_FOREACH(vnr, &vnet_recursions, vnr_le) if (vnr->prev_fn == old_fn && vnr->where_fn == curthread->td_vnet_lpush && vnr->where_line == line && (vnr->old_vnet == vnr->new_vnet) == (curvnet == old_vnet)) return; vnr = malloc(sizeof(*vnr), M_VNET, M_NOWAIT | M_ZERO); if (vnr == NULL) panic("%s: malloc failed", __func__); vnr->prev_fn = old_fn; vnr->where_fn = curthread->td_vnet_lpush; vnr->where_line = line; vnr->old_vnet = old_vnet; vnr->new_vnet = curvnet; SLIST_INSERT_HEAD(&vnet_recursions, vnr, vnr_le); vnet_print_recursion(vnr, 0); #ifdef KDB kdb_backtrace(); #endif } #endif /* VNET_DEBUG */ /* * DDB(4). */ #ifdef DDB static void db_vnet_print(struct vnet *vnet) { db_printf("vnet = %p\n", vnet); db_printf(" vnet_magic_n = %#08x (%s, orig %#08x)\n", vnet->vnet_magic_n, (vnet->vnet_magic_n == VNET_MAGIC_N) ? "ok" : "mismatch", VNET_MAGIC_N); db_printf(" vnet_ifcnt = %u\n", vnet->vnet_ifcnt); db_printf(" vnet_sockcnt = %u\n", vnet->vnet_sockcnt); db_printf(" vnet_data_mem = %p\n", vnet->vnet_data_mem); db_printf(" vnet_data_base = %#jx\n", (uintmax_t)vnet->vnet_data_base); db_printf(" vnet_state = %#08x\n", vnet->vnet_state); db_printf("\n"); } DB_SHOW_ALL_COMMAND(vnets, db_show_all_vnets) { VNET_ITERATOR_DECL(vnet_iter); VNET_FOREACH(vnet_iter) { db_vnet_print(vnet_iter); if (db_pager_quit) break; } } DB_SHOW_COMMAND(vnet, db_show_vnet) { if (!have_addr) { db_printf("usage: show vnet \n"); return; } db_vnet_print((struct vnet *)addr); } static void db_show_vnet_print_vs(struct vnet_sysinit *vs, int ddb) { const char *vsname, *funcname; c_db_sym_t sym; db_expr_t offset; #define xprint(...) \ if (ddb) \ db_printf(__VA_ARGS__); \ else \ printf(__VA_ARGS__) if (vs == NULL) { xprint("%s: no vnet_sysinit * given\n", __func__); return; } sym = db_search_symbol((vm_offset_t)vs, DB_STGY_ANY, &offset); db_symbol_values(sym, &vsname, NULL); sym = db_search_symbol((vm_offset_t)vs->func, DB_STGY_PROC, &offset); db_symbol_values(sym, &funcname, NULL); xprint("%s(%p)\n", (vsname != NULL) ? vsname : "", vs); xprint(" %#08x %#08x\n", vs->subsystem, vs->order); xprint(" %p(%s)(%p)\n", vs->func, (funcname != NULL) ? funcname : "", vs->arg); #undef xprint } DB_SHOW_COMMAND(vnet_sysinit, db_show_vnet_sysinit) { struct vnet_sysinit *vs; db_printf("VNET_SYSINIT vs Name(Ptr)\n"); db_printf(" Subsystem Order\n"); db_printf(" Function(Name)(Arg)\n"); TAILQ_FOREACH(vs, &vnet_constructors, link) { db_show_vnet_print_vs(vs, 1); if (db_pager_quit) break; } } DB_SHOW_COMMAND(vnet_sysuninit, db_show_vnet_sysuninit) { struct vnet_sysinit *vs; db_printf("VNET_SYSUNINIT vs Name(Ptr)\n"); db_printf(" Subsystem Order\n"); db_printf(" Function(Name)(Arg)\n"); TAILQ_FOREACH_REVERSE(vs, &vnet_destructors, vnet_sysuninit_head, link) { db_show_vnet_print_vs(vs, 1); if (db_pager_quit) break; } } #ifdef VNET_DEBUG DB_SHOW_COMMAND(vnetrcrs, db_show_vnetrcrs) { struct vnet_recursion *vnr; SLIST_FOREACH(vnr, &vnet_recursions, vnr_le) vnet_print_recursion(vnr, 1); } #endif #endif /* DDB */ Index: head/sys/netgraph/ng_base.c =================================================================== --- head/sys/netgraph/ng_base.c (revision 336675) +++ head/sys/netgraph/ng_base.c (revision 336676) @@ -1,3850 +1,3850 @@ /*- * Copyright (c) 1996-1999 Whistle Communications, Inc. * All rights reserved. * * Subject to the following obligations and disclaimer of warranty, use and * redistribution of this software, in source or object code forms, with or * without modifications are expressly permitted by Whistle Communications; * provided, however, that: * 1. Any and all reproductions of the source or object code must include the * copyright notice above and the following disclaimer of warranties; and * 2. No rights are granted, in any manner or form, to use Whistle * Communications, Inc. trademarks, including the mark "WHISTLE * COMMUNICATIONS" on advertising, endorsements, or otherwise except as * such appears in the above copyright notice or in the software. * * THIS SOFTWARE IS BEING PROVIDED BY WHISTLE COMMUNICATIONS "AS IS", AND * TO THE MAXIMUM EXTENT PERMITTED BY LAW, WHISTLE COMMUNICATIONS MAKES NO * REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED, REGARDING THIS SOFTWARE, * INCLUDING WITHOUT LIMITATION, ANY AND ALL IMPLIED WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT. * WHISTLE COMMUNICATIONS DOES NOT WARRANT, GUARANTEE, OR MAKE ANY * REPRESENTATIONS REGARDING THE USE OF, OR THE RESULTS OF THE USE OF THIS * SOFTWARE IN TERMS OF ITS CORRECTNESS, ACCURACY, RELIABILITY OR OTHERWISE. * IN NO EVENT SHALL WHISTLE COMMUNICATIONS BE LIABLE FOR ANY DAMAGES * RESULTING FROM OR ARISING OUT OF ANY USE OF THIS SOFTWARE, INCLUDING * WITHOUT LIMITATION, ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * PUNITIVE, OR CONSEQUENTIAL DAMAGES, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES, LOSS OF USE, DATA OR PROFITS, HOWEVER CAUSED AND UNDER ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF WHISTLE COMMUNICATIONS IS ADVISED OF THE POSSIBILITY * OF SUCH DAMAGE. * * Authors: Julian Elischer * Archie Cobbs * * $FreeBSD$ * $Whistle: ng_base.c,v 1.39 1999/01/28 23:54:53 julian Exp $ */ /* * This file implements the base netgraph code. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MODULE_VERSION(netgraph, NG_ABI_VERSION); /* Mutex to protect topology events. */ static struct rwlock ng_topo_lock; #define TOPOLOGY_RLOCK() rw_rlock(&ng_topo_lock) #define TOPOLOGY_RUNLOCK() rw_runlock(&ng_topo_lock) #define TOPOLOGY_WLOCK() rw_wlock(&ng_topo_lock) #define TOPOLOGY_WUNLOCK() rw_wunlock(&ng_topo_lock) #define TOPOLOGY_NOTOWNED() rw_assert(&ng_topo_lock, RA_UNLOCKED) #ifdef NETGRAPH_DEBUG static struct mtx ng_nodelist_mtx; /* protects global node/hook lists */ static struct mtx ngq_mtx; /* protects the queue item list */ static SLIST_HEAD(, ng_node) ng_allnodes; static LIST_HEAD(, ng_node) ng_freenodes; /* in debug, we never free() them */ static SLIST_HEAD(, ng_hook) ng_allhooks; static LIST_HEAD(, ng_hook) ng_freehooks; /* in debug, we never free() them */ static void ng_dumpitems(void); static void ng_dumpnodes(void); static void ng_dumphooks(void); #endif /* NETGRAPH_DEBUG */ /* * DEAD versions of the structures. * In order to avoid races, it is sometimes necessary to point * at SOMETHING even though theoretically, the current entity is * INVALID. Use these to avoid these races. */ struct ng_type ng_deadtype = { NG_ABI_VERSION, "dead", NULL, /* modevent */ NULL, /* constructor */ NULL, /* rcvmsg */ NULL, /* shutdown */ NULL, /* newhook */ NULL, /* findhook */ NULL, /* connect */ NULL, /* rcvdata */ NULL, /* disconnect */ NULL, /* cmdlist */ }; struct ng_node ng_deadnode = { "dead", &ng_deadtype, NGF_INVALID, 0, /* numhooks */ NULL, /* private */ 0, /* ID */ LIST_HEAD_INITIALIZER(ng_deadnode.nd_hooks), {}, /* all_nodes list entry */ {}, /* id hashtable list entry */ { 0, 0, {}, /* should never use! (should hang) */ {}, /* workqueue entry */ STAILQ_HEAD_INITIALIZER(ng_deadnode.nd_input_queue.queue), }, 1, /* refs */ NULL, /* vnet */ #ifdef NETGRAPH_DEBUG ND_MAGIC, __FILE__, __LINE__, {NULL} #endif /* NETGRAPH_DEBUG */ }; struct ng_hook ng_deadhook = { "dead", NULL, /* private */ HK_INVALID | HK_DEAD, 0, /* undefined data link type */ &ng_deadhook, /* Peer is self */ &ng_deadnode, /* attached to deadnode */ {}, /* hooks list */ NULL, /* override rcvmsg() */ NULL, /* override rcvdata() */ 1, /* refs always >= 1 */ #ifdef NETGRAPH_DEBUG HK_MAGIC, __FILE__, __LINE__, {NULL} #endif /* NETGRAPH_DEBUG */ }; /* * END DEAD STRUCTURES */ /* List nodes with unallocated work */ static STAILQ_HEAD(, ng_node) ng_worklist = STAILQ_HEAD_INITIALIZER(ng_worklist); static struct mtx ng_worklist_mtx; /* MUST LOCK NODE FIRST */ /* List of installed types */ static LIST_HEAD(, ng_type) ng_typelist; static struct rwlock ng_typelist_lock; #define TYPELIST_RLOCK() rw_rlock(&ng_typelist_lock) #define TYPELIST_RUNLOCK() rw_runlock(&ng_typelist_lock) #define TYPELIST_WLOCK() rw_wlock(&ng_typelist_lock) #define TYPELIST_WUNLOCK() rw_wunlock(&ng_typelist_lock) /* Hash related definitions. */ LIST_HEAD(nodehash, ng_node); -static VNET_DEFINE(struct nodehash *, ng_ID_hash); -static VNET_DEFINE(u_long, ng_ID_hmask); -static VNET_DEFINE(u_long, ng_nodes); -static VNET_DEFINE(struct nodehash *, ng_name_hash); -static VNET_DEFINE(u_long, ng_name_hmask); -static VNET_DEFINE(u_long, ng_named_nodes); +VNET_DEFINE_STATIC(struct nodehash *, ng_ID_hash); +VNET_DEFINE_STATIC(u_long, ng_ID_hmask); +VNET_DEFINE_STATIC(u_long, ng_nodes); +VNET_DEFINE_STATIC(struct nodehash *, ng_name_hash); +VNET_DEFINE_STATIC(u_long, ng_name_hmask); +VNET_DEFINE_STATIC(u_long, ng_named_nodes); #define V_ng_ID_hash VNET(ng_ID_hash) #define V_ng_ID_hmask VNET(ng_ID_hmask) #define V_ng_nodes VNET(ng_nodes) #define V_ng_name_hash VNET(ng_name_hash) #define V_ng_name_hmask VNET(ng_name_hmask) #define V_ng_named_nodes VNET(ng_named_nodes) static struct rwlock ng_idhash_lock; #define IDHASH_RLOCK() rw_rlock(&ng_idhash_lock) #define IDHASH_RUNLOCK() rw_runlock(&ng_idhash_lock) #define IDHASH_WLOCK() rw_wlock(&ng_idhash_lock) #define IDHASH_WUNLOCK() rw_wunlock(&ng_idhash_lock) /* Method to find a node.. used twice so do it here */ #define NG_IDHASH_FN(ID) ((ID) % (V_ng_ID_hmask + 1)) #define NG_IDHASH_FIND(ID, node) \ do { \ rw_assert(&ng_idhash_lock, RA_LOCKED); \ LIST_FOREACH(node, &V_ng_ID_hash[NG_IDHASH_FN(ID)], \ nd_idnodes) { \ if (NG_NODE_IS_VALID(node) \ && (NG_NODE_ID(node) == ID)) { \ break; \ } \ } \ } while (0) static struct rwlock ng_namehash_lock; #define NAMEHASH_RLOCK() rw_rlock(&ng_namehash_lock) #define NAMEHASH_RUNLOCK() rw_runlock(&ng_namehash_lock) #define NAMEHASH_WLOCK() rw_wlock(&ng_namehash_lock) #define NAMEHASH_WUNLOCK() rw_wunlock(&ng_namehash_lock) /* Internal functions */ static int ng_add_hook(node_p node, const char *name, hook_p * hookp); static int ng_generic_msg(node_p here, item_p item, hook_p lasthook); static ng_ID_t ng_decodeidname(const char *name); static int ngb_mod_event(module_t mod, int event, void *data); static void ng_worklist_add(node_p node); static void ngthread(void *); static int ng_apply_item(node_p node, item_p item, int rw); static void ng_flush_input_queue(node_p node); static node_p ng_ID2noderef(ng_ID_t ID); static int ng_con_nodes(item_p item, node_p node, const char *name, node_p node2, const char *name2); static int ng_con_part2(node_p node, item_p item, hook_p hook); static int ng_con_part3(node_p node, item_p item, hook_p hook); static int ng_mkpeer(node_p node, const char *name, const char *name2, char *type); static void ng_name_rehash(void); static void ng_ID_rehash(void); /* Imported, these used to be externally visible, some may go back. */ void ng_destroy_hook(hook_p hook); int ng_path2noderef(node_p here, const char *path, node_p *dest, hook_p *lasthook); int ng_make_node(const char *type, node_p *nodepp); int ng_path_parse(char *addr, char **node, char **path, char **hook); void ng_rmnode(node_p node, hook_p dummy1, void *dummy2, int dummy3); void ng_unname(node_p node); /* Our own netgraph malloc type */ MALLOC_DEFINE(M_NETGRAPH, "netgraph", "netgraph structures and ctrl messages"); MALLOC_DEFINE(M_NETGRAPH_MSG, "netgraph_msg", "netgraph name storage"); static MALLOC_DEFINE(M_NETGRAPH_HOOK, "netgraph_hook", "netgraph hook structures"); static MALLOC_DEFINE(M_NETGRAPH_NODE, "netgraph_node", "netgraph node structures"); static MALLOC_DEFINE(M_NETGRAPH_ITEM, "netgraph_item", "netgraph item structures"); /* Should not be visible outside this file */ #define _NG_ALLOC_HOOK(hook) \ hook = malloc(sizeof(*hook), M_NETGRAPH_HOOK, M_NOWAIT | M_ZERO) #define _NG_ALLOC_NODE(node) \ node = malloc(sizeof(*node), M_NETGRAPH_NODE, M_NOWAIT | M_ZERO) #define NG_QUEUE_LOCK_INIT(n) \ mtx_init(&(n)->q_mtx, "ng_node", NULL, MTX_DEF) #define NG_QUEUE_LOCK(n) \ mtx_lock(&(n)->q_mtx) #define NG_QUEUE_UNLOCK(n) \ mtx_unlock(&(n)->q_mtx) #define NG_WORKLIST_LOCK_INIT() \ mtx_init(&ng_worklist_mtx, "ng_worklist", NULL, MTX_DEF) #define NG_WORKLIST_LOCK() \ mtx_lock(&ng_worklist_mtx) #define NG_WORKLIST_UNLOCK() \ mtx_unlock(&ng_worklist_mtx) #define NG_WORKLIST_SLEEP() \ mtx_sleep(&ng_worklist, &ng_worklist_mtx, PI_NET, "sleep", 0) #define NG_WORKLIST_WAKEUP() \ wakeup_one(&ng_worklist) #ifdef NETGRAPH_DEBUG /*----------------------------------------------*/ /* * In debug mode: * In an attempt to help track reference count screwups * we do not free objects back to the malloc system, but keep them * in a local cache where we can examine them and keep information safely * after they have been freed. * We use this scheme for nodes and hooks, and to some extent for items. */ static __inline hook_p ng_alloc_hook(void) { hook_p hook; SLIST_ENTRY(ng_hook) temp; mtx_lock(&ng_nodelist_mtx); hook = LIST_FIRST(&ng_freehooks); if (hook) { LIST_REMOVE(hook, hk_hooks); bcopy(&hook->hk_all, &temp, sizeof(temp)); bzero(hook, sizeof(struct ng_hook)); bcopy(&temp, &hook->hk_all, sizeof(temp)); mtx_unlock(&ng_nodelist_mtx); hook->hk_magic = HK_MAGIC; } else { mtx_unlock(&ng_nodelist_mtx); _NG_ALLOC_HOOK(hook); if (hook) { hook->hk_magic = HK_MAGIC; mtx_lock(&ng_nodelist_mtx); SLIST_INSERT_HEAD(&ng_allhooks, hook, hk_all); mtx_unlock(&ng_nodelist_mtx); } } return (hook); } static __inline node_p ng_alloc_node(void) { node_p node; SLIST_ENTRY(ng_node) temp; mtx_lock(&ng_nodelist_mtx); node = LIST_FIRST(&ng_freenodes); if (node) { LIST_REMOVE(node, nd_nodes); bcopy(&node->nd_all, &temp, sizeof(temp)); bzero(node, sizeof(struct ng_node)); bcopy(&temp, &node->nd_all, sizeof(temp)); mtx_unlock(&ng_nodelist_mtx); node->nd_magic = ND_MAGIC; } else { mtx_unlock(&ng_nodelist_mtx); _NG_ALLOC_NODE(node); if (node) { node->nd_magic = ND_MAGIC; mtx_lock(&ng_nodelist_mtx); SLIST_INSERT_HEAD(&ng_allnodes, node, nd_all); mtx_unlock(&ng_nodelist_mtx); } } return (node); } #define NG_ALLOC_HOOK(hook) do { (hook) = ng_alloc_hook(); } while (0) #define NG_ALLOC_NODE(node) do { (node) = ng_alloc_node(); } while (0) #define NG_FREE_HOOK(hook) \ do { \ mtx_lock(&ng_nodelist_mtx); \ LIST_INSERT_HEAD(&ng_freehooks, hook, hk_hooks); \ hook->hk_magic = 0; \ mtx_unlock(&ng_nodelist_mtx); \ } while (0) #define NG_FREE_NODE(node) \ do { \ mtx_lock(&ng_nodelist_mtx); \ LIST_INSERT_HEAD(&ng_freenodes, node, nd_nodes); \ node->nd_magic = 0; \ mtx_unlock(&ng_nodelist_mtx); \ } while (0) #else /* NETGRAPH_DEBUG */ /*----------------------------------------------*/ #define NG_ALLOC_HOOK(hook) _NG_ALLOC_HOOK(hook) #define NG_ALLOC_NODE(node) _NG_ALLOC_NODE(node) #define NG_FREE_HOOK(hook) do { free((hook), M_NETGRAPH_HOOK); } while (0) #define NG_FREE_NODE(node) do { free((node), M_NETGRAPH_NODE); } while (0) #endif /* NETGRAPH_DEBUG */ /*----------------------------------------------*/ /* Set this to kdb_enter("X") to catch all errors as they occur */ #ifndef TRAP_ERROR #define TRAP_ERROR() #endif -static VNET_DEFINE(ng_ID_t, nextID) = 1; +VNET_DEFINE_STATIC(ng_ID_t, nextID) = 1; #define V_nextID VNET(nextID) #ifdef INVARIANTS #define CHECK_DATA_MBUF(m) do { \ struct mbuf *n; \ int total; \ \ M_ASSERTPKTHDR(m); \ for (total = 0, n = (m); n != NULL; n = n->m_next) { \ total += n->m_len; \ if (n->m_nextpkt != NULL) \ panic("%s: m_nextpkt", __func__); \ } \ \ if ((m)->m_pkthdr.len != total) { \ panic("%s: %d != %d", \ __func__, (m)->m_pkthdr.len, total); \ } \ } while (0) #else #define CHECK_DATA_MBUF(m) #endif #define ERROUT(x) do { error = (x); goto done; } while (0) /************************************************************************ Parse type definitions for generic messages ************************************************************************/ /* Handy structure parse type defining macro */ #define DEFINE_PARSE_STRUCT_TYPE(lo, up, args) \ static const struct ng_parse_struct_field \ ng_ ## lo ## _type_fields[] = NG_GENERIC_ ## up ## _INFO args; \ static const struct ng_parse_type ng_generic_ ## lo ## _type = { \ &ng_parse_struct_type, \ &ng_ ## lo ## _type_fields \ } DEFINE_PARSE_STRUCT_TYPE(mkpeer, MKPEER, ()); DEFINE_PARSE_STRUCT_TYPE(connect, CONNECT, ()); DEFINE_PARSE_STRUCT_TYPE(name, NAME, ()); DEFINE_PARSE_STRUCT_TYPE(rmhook, RMHOOK, ()); DEFINE_PARSE_STRUCT_TYPE(nodeinfo, NODEINFO, ()); DEFINE_PARSE_STRUCT_TYPE(typeinfo, TYPEINFO, ()); DEFINE_PARSE_STRUCT_TYPE(linkinfo, LINKINFO, (&ng_generic_nodeinfo_type)); /* Get length of an array when the length is stored as a 32 bit value immediately preceding the array -- as with struct namelist and struct typelist. */ static int ng_generic_list_getLength(const struct ng_parse_type *type, const u_char *start, const u_char *buf) { return *((const u_int32_t *)(buf - 4)); } /* Get length of the array of struct linkinfo inside a struct hooklist */ static int ng_generic_linkinfo_getLength(const struct ng_parse_type *type, const u_char *start, const u_char *buf) { const struct hooklist *hl = (const struct hooklist *)start; return hl->nodeinfo.hooks; } /* Array type for a variable length array of struct namelist */ static const struct ng_parse_array_info ng_nodeinfoarray_type_info = { &ng_generic_nodeinfo_type, &ng_generic_list_getLength }; static const struct ng_parse_type ng_generic_nodeinfoarray_type = { &ng_parse_array_type, &ng_nodeinfoarray_type_info }; /* Array type for a variable length array of struct typelist */ static const struct ng_parse_array_info ng_typeinfoarray_type_info = { &ng_generic_typeinfo_type, &ng_generic_list_getLength }; static const struct ng_parse_type ng_generic_typeinfoarray_type = { &ng_parse_array_type, &ng_typeinfoarray_type_info }; /* Array type for array of struct linkinfo in struct hooklist */ static const struct ng_parse_array_info ng_generic_linkinfo_array_type_info = { &ng_generic_linkinfo_type, &ng_generic_linkinfo_getLength }; static const struct ng_parse_type ng_generic_linkinfo_array_type = { &ng_parse_array_type, &ng_generic_linkinfo_array_type_info }; DEFINE_PARSE_STRUCT_TYPE(typelist, TYPELIST, (&ng_generic_typeinfoarray_type)); DEFINE_PARSE_STRUCT_TYPE(hooklist, HOOKLIST, (&ng_generic_nodeinfo_type, &ng_generic_linkinfo_array_type)); DEFINE_PARSE_STRUCT_TYPE(listnodes, LISTNODES, (&ng_generic_nodeinfoarray_type)); /* List of commands and how to convert arguments to/from ASCII */ static const struct ng_cmdlist ng_generic_cmds[] = { { NGM_GENERIC_COOKIE, NGM_SHUTDOWN, "shutdown", NULL, NULL }, { NGM_GENERIC_COOKIE, NGM_MKPEER, "mkpeer", &ng_generic_mkpeer_type, NULL }, { NGM_GENERIC_COOKIE, NGM_CONNECT, "connect", &ng_generic_connect_type, NULL }, { NGM_GENERIC_COOKIE, NGM_NAME, "name", &ng_generic_name_type, NULL }, { NGM_GENERIC_COOKIE, NGM_RMHOOK, "rmhook", &ng_generic_rmhook_type, NULL }, { NGM_GENERIC_COOKIE, NGM_NODEINFO, "nodeinfo", NULL, &ng_generic_nodeinfo_type }, { NGM_GENERIC_COOKIE, NGM_LISTHOOKS, "listhooks", NULL, &ng_generic_hooklist_type }, { NGM_GENERIC_COOKIE, NGM_LISTNAMES, "listnames", NULL, &ng_generic_listnodes_type /* same as NGM_LISTNODES */ }, { NGM_GENERIC_COOKIE, NGM_LISTNODES, "listnodes", NULL, &ng_generic_listnodes_type }, { NGM_GENERIC_COOKIE, NGM_LISTTYPES, "listtypes", NULL, &ng_generic_typelist_type }, { NGM_GENERIC_COOKIE, NGM_TEXT_CONFIG, "textconfig", NULL, &ng_parse_string_type }, { NGM_GENERIC_COOKIE, NGM_TEXT_STATUS, "textstatus", NULL, &ng_parse_string_type }, { NGM_GENERIC_COOKIE, NGM_ASCII2BINARY, "ascii2binary", &ng_parse_ng_mesg_type, &ng_parse_ng_mesg_type }, { NGM_GENERIC_COOKIE, NGM_BINARY2ASCII, "binary2ascii", &ng_parse_ng_mesg_type, &ng_parse_ng_mesg_type }, { 0 } }; /************************************************************************ Node routines ************************************************************************/ /* * Instantiate a node of the requested type */ int ng_make_node(const char *typename, node_p *nodepp) { struct ng_type *type; int error; /* Check that the type makes sense */ if (typename == NULL) { TRAP_ERROR(); return (EINVAL); } /* Locate the node type. If we fail we return. Do not try to load * module. */ if ((type = ng_findtype(typename)) == NULL) return (ENXIO); /* * If we have a constructor, then make the node and * call the constructor to do type specific initialisation. */ if (type->constructor != NULL) { if ((error = ng_make_node_common(type, nodepp)) == 0) { if ((error = ((*type->constructor)(*nodepp))) != 0) { NG_NODE_UNREF(*nodepp); } } } else { /* * Node has no constructor. We cannot ask for one * to be made. It must be brought into existence by * some external agency. The external agency should * call ng_make_node_common() directly to get the * netgraph part initialised. */ TRAP_ERROR(); error = EINVAL; } return (error); } /* * Generic node creation. Called by node initialisation for externally * instantiated nodes (e.g. hardware, sockets, etc ). * The returned node has a reference count of 1. */ int ng_make_node_common(struct ng_type *type, node_p *nodepp) { node_p node; /* Require the node type to have been already installed */ if (ng_findtype(type->name) == NULL) { TRAP_ERROR(); return (EINVAL); } /* Make a node and try attach it to the type */ NG_ALLOC_NODE(node); if (node == NULL) { TRAP_ERROR(); return (ENOMEM); } node->nd_type = type; #ifdef VIMAGE node->nd_vnet = curvnet; #endif NG_NODE_REF(node); /* note reference */ type->refs++; NG_QUEUE_LOCK_INIT(&node->nd_input_queue); STAILQ_INIT(&node->nd_input_queue.queue); node->nd_input_queue.q_flags = 0; /* Initialize hook list for new node */ LIST_INIT(&node->nd_hooks); /* Get an ID and put us in the hash chain. */ IDHASH_WLOCK(); for (;;) { /* wrap protection, even if silly */ node_p node2 = NULL; node->nd_ID = V_nextID++; /* 137/sec for 1 year before wrap */ /* Is there a problem with the new number? */ NG_IDHASH_FIND(node->nd_ID, node2); /* already taken? */ if ((node->nd_ID != 0) && (node2 == NULL)) { break; } } V_ng_nodes++; if (V_ng_nodes * 2 > V_ng_ID_hmask) ng_ID_rehash(); LIST_INSERT_HEAD(&V_ng_ID_hash[NG_IDHASH_FN(node->nd_ID)], node, nd_idnodes); IDHASH_WUNLOCK(); /* Done */ *nodepp = node; return (0); } /* * Forceably start the shutdown process on a node. Either call * its shutdown method, or do the default shutdown if there is * no type-specific method. * * We can only be called from a shutdown message, so we know we have * a writer lock, and therefore exclusive access. It also means * that we should not be on the work queue, but we check anyhow. * * Persistent node types must have a type-specific method which * allocates a new node in which case, this one is irretrievably going away, * or cleans up anything it needs, and just makes the node valid again, * in which case we allow the node to survive. * * XXX We need to think of how to tell a persistent node that we * REALLY need to go away because the hardware has gone or we * are rebooting.... etc. */ void ng_rmnode(node_p node, hook_p dummy1, void *dummy2, int dummy3) { hook_p hook; /* Check if it's already shutting down */ if ((node->nd_flags & NGF_CLOSING) != 0) return; if (node == &ng_deadnode) { printf ("shutdown called on deadnode\n"); return; } /* Add an extra reference so it doesn't go away during this */ NG_NODE_REF(node); /* * Mark it invalid so any newcomers know not to try use it * Also add our own mark so we can't recurse * note that NGF_INVALID does not do this as it's also set during * creation */ node->nd_flags |= NGF_INVALID|NGF_CLOSING; /* If node has its pre-shutdown method, then call it first*/ if (node->nd_type && node->nd_type->close) (*node->nd_type->close)(node); /* Notify all remaining connected nodes to disconnect */ while ((hook = LIST_FIRST(&node->nd_hooks)) != NULL) ng_destroy_hook(hook); /* * Drain the input queue forceably. * it has no hooks so what's it going to do, bleed on someone? * Theoretically we came here from a queue entry that was added * Just before the queue was closed, so it should be empty anyway. * Also removes us from worklist if needed. */ ng_flush_input_queue(node); /* Ask the type if it has anything to do in this case */ if (node->nd_type && node->nd_type->shutdown) { (*node->nd_type->shutdown)(node); if (NG_NODE_IS_VALID(node)) { /* * Well, blow me down if the node code hasn't declared * that it doesn't want to die. * Presumably it is a persistent node. * If we REALLY want it to go away, * e.g. hardware going away, * Our caller should set NGF_REALLY_DIE in nd_flags. */ node->nd_flags &= ~(NGF_INVALID|NGF_CLOSING); NG_NODE_UNREF(node); /* Assume they still have theirs */ return; } } else { /* do the default thing */ NG_NODE_UNREF(node); } ng_unname(node); /* basically a NOP these days */ /* * Remove extra reference, possibly the last * Possible other holders of references may include * timeout callouts, but theoretically the node's supposed to * have cancelled them. Possibly hardware dependencies may * force a driver to 'linger' with a reference. */ NG_NODE_UNREF(node); } /* * Remove a reference to the node, possibly the last. * deadnode always acts as it it were the last. */ void ng_unref_node(node_p node) { if (node == &ng_deadnode) return; CURVNET_SET(node->nd_vnet); if (refcount_release(&node->nd_refs)) { /* we were the last */ node->nd_type->refs--; /* XXX maybe should get types lock? */ NAMEHASH_WLOCK(); if (NG_NODE_HAS_NAME(node)) { V_ng_named_nodes--; LIST_REMOVE(node, nd_nodes); } NAMEHASH_WUNLOCK(); IDHASH_WLOCK(); V_ng_nodes--; LIST_REMOVE(node, nd_idnodes); IDHASH_WUNLOCK(); mtx_destroy(&node->nd_input_queue.q_mtx); NG_FREE_NODE(node); } CURVNET_RESTORE(); } /************************************************************************ Node ID handling ************************************************************************/ static node_p ng_ID2noderef(ng_ID_t ID) { node_p node; IDHASH_RLOCK(); NG_IDHASH_FIND(ID, node); if (node) NG_NODE_REF(node); IDHASH_RUNLOCK(); return(node); } ng_ID_t ng_node2ID(node_p node) { return (node ? NG_NODE_ID(node) : 0); } /************************************************************************ Node name handling ************************************************************************/ /* * Assign a node a name. */ int ng_name_node(node_p node, const char *name) { uint32_t hash; node_p node2; int i; /* Check the name is valid */ for (i = 0; i < NG_NODESIZ; i++) { if (name[i] == '\0' || name[i] == '.' || name[i] == ':') break; } if (i == 0 || name[i] != '\0') { TRAP_ERROR(); return (EINVAL); } if (ng_decodeidname(name) != 0) { /* valid IDs not allowed here */ TRAP_ERROR(); return (EINVAL); } NAMEHASH_WLOCK(); if (V_ng_named_nodes * 2 > V_ng_name_hmask) ng_name_rehash(); hash = hash32_str(name, HASHINIT) & V_ng_name_hmask; /* Check the name isn't already being used. */ LIST_FOREACH(node2, &V_ng_name_hash[hash], nd_nodes) if (NG_NODE_IS_VALID(node2) && (strcmp(NG_NODE_NAME(node2), name) == 0)) { NAMEHASH_WUNLOCK(); return (EADDRINUSE); } if (NG_NODE_HAS_NAME(node)) LIST_REMOVE(node, nd_nodes); else V_ng_named_nodes++; /* Copy it. */ strlcpy(NG_NODE_NAME(node), name, NG_NODESIZ); /* Update name hash. */ LIST_INSERT_HEAD(&V_ng_name_hash[hash], node, nd_nodes); NAMEHASH_WUNLOCK(); return (0); } /* * Find a node by absolute name. The name should NOT end with ':' * The name "." means "this node" and "[xxx]" means "the node * with ID (ie, at address) xxx". * * Returns the node if found, else NULL. * Eventually should add something faster than a sequential search. * Note it acquires a reference on the node so you can be sure it's still * there. */ node_p ng_name2noderef(node_p here, const char *name) { node_p node; ng_ID_t temp; int hash; /* "." means "this node" */ if (strcmp(name, ".") == 0) { NG_NODE_REF(here); return(here); } /* Check for name-by-ID */ if ((temp = ng_decodeidname(name)) != 0) { return (ng_ID2noderef(temp)); } /* Find node by name. */ hash = hash32_str(name, HASHINIT) & V_ng_name_hmask; NAMEHASH_RLOCK(); LIST_FOREACH(node, &V_ng_name_hash[hash], nd_nodes) if (NG_NODE_IS_VALID(node) && (strcmp(NG_NODE_NAME(node), name) == 0)) { NG_NODE_REF(node); break; } NAMEHASH_RUNLOCK(); return (node); } /* * Decode an ID name, eg. "[f03034de]". Returns 0 if the * string is not valid, otherwise returns the value. */ static ng_ID_t ng_decodeidname(const char *name) { const int len = strlen(name); char *eptr; u_long val; /* Check for proper length, brackets, no leading junk */ if ((len < 3) || (name[0] != '[') || (name[len - 1] != ']') || (!isxdigit(name[1]))) return ((ng_ID_t)0); /* Decode number */ val = strtoul(name + 1, &eptr, 16); if ((eptr - name != len - 1) || (val == ULONG_MAX) || (val == 0)) return ((ng_ID_t)0); return ((ng_ID_t)val); } /* * Remove a name from a node. This should only be called * when shutting down and removing the node. */ void ng_unname(node_p node) { } /* * Allocate a bigger name hash. */ static void ng_name_rehash() { struct nodehash *new; uint32_t hash; u_long hmask; node_p node, node2; int i; new = hashinit_flags((V_ng_name_hmask + 1) * 2, M_NETGRAPH_NODE, &hmask, HASH_NOWAIT); if (new == NULL) return; for (i = 0; i <= V_ng_name_hmask; i++) LIST_FOREACH_SAFE(node, &V_ng_name_hash[i], nd_nodes, node2) { #ifdef INVARIANTS LIST_REMOVE(node, nd_nodes); #endif hash = hash32_str(NG_NODE_NAME(node), HASHINIT) & hmask; LIST_INSERT_HEAD(&new[hash], node, nd_nodes); } hashdestroy(V_ng_name_hash, M_NETGRAPH_NODE, V_ng_name_hmask); V_ng_name_hash = new; V_ng_name_hmask = hmask; } /* * Allocate a bigger ID hash. */ static void ng_ID_rehash() { struct nodehash *new; uint32_t hash; u_long hmask; node_p node, node2; int i; new = hashinit_flags((V_ng_ID_hmask + 1) * 2, M_NETGRAPH_NODE, &hmask, HASH_NOWAIT); if (new == NULL) return; for (i = 0; i <= V_ng_ID_hmask; i++) LIST_FOREACH_SAFE(node, &V_ng_ID_hash[i], nd_idnodes, node2) { #ifdef INVARIANTS LIST_REMOVE(node, nd_idnodes); #endif hash = (node->nd_ID % (hmask + 1)); LIST_INSERT_HEAD(&new[hash], node, nd_idnodes); } hashdestroy(V_ng_ID_hash, M_NETGRAPH_NODE, V_ng_name_hmask); V_ng_ID_hash = new; V_ng_ID_hmask = hmask; } /************************************************************************ Hook routines Names are not optional. Hooks are always connected, except for a brief moment within these routines. On invalidation or during creation they are connected to the 'dead' hook. ************************************************************************/ /* * Remove a hook reference */ void ng_unref_hook(hook_p hook) { if (hook == &ng_deadhook) return; if (refcount_release(&hook->hk_refs)) { /* we were the last */ if (_NG_HOOK_NODE(hook)) /* it'll probably be ng_deadnode */ _NG_NODE_UNREF((_NG_HOOK_NODE(hook))); NG_FREE_HOOK(hook); } } /* * Add an unconnected hook to a node. Only used internally. * Assumes node is locked. (XXX not yet true ) */ static int ng_add_hook(node_p node, const char *name, hook_p *hookp) { hook_p hook; int error = 0; /* Check that the given name is good */ if (name == NULL) { TRAP_ERROR(); return (EINVAL); } if (ng_findhook(node, name) != NULL) { TRAP_ERROR(); return (EEXIST); } /* Allocate the hook and link it up */ NG_ALLOC_HOOK(hook); if (hook == NULL) { TRAP_ERROR(); return (ENOMEM); } hook->hk_refs = 1; /* add a reference for us to return */ hook->hk_flags = HK_INVALID; hook->hk_peer = &ng_deadhook; /* start off this way */ hook->hk_node = node; NG_NODE_REF(node); /* each hook counts as a reference */ /* Set hook name */ strlcpy(NG_HOOK_NAME(hook), name, NG_HOOKSIZ); /* * Check if the node type code has something to say about it * If it fails, the unref of the hook will also unref the node. */ if (node->nd_type->newhook != NULL) { if ((error = (*node->nd_type->newhook)(node, hook, name))) { NG_HOOK_UNREF(hook); /* this frees the hook */ return (error); } } /* * The 'type' agrees so far, so go ahead and link it in. * We'll ask again later when we actually connect the hooks. */ LIST_INSERT_HEAD(&node->nd_hooks, hook, hk_hooks); node->nd_numhooks++; NG_HOOK_REF(hook); /* one for the node */ if (hookp) *hookp = hook; return (0); } /* * Find a hook * * Node types may supply their own optimized routines for finding * hooks. If none is supplied, we just do a linear search. * XXX Possibly we should add a reference to the hook? */ hook_p ng_findhook(node_p node, const char *name) { hook_p hook; if (node->nd_type->findhook != NULL) return (*node->nd_type->findhook)(node, name); LIST_FOREACH(hook, &node->nd_hooks, hk_hooks) { if (NG_HOOK_IS_VALID(hook) && (strcmp(NG_HOOK_NAME(hook), name) == 0)) return (hook); } return (NULL); } /* * Destroy a hook * * As hooks are always attached, this really destroys two hooks. * The one given, and the one attached to it. Disconnect the hooks * from each other first. We reconnect the peer hook to the 'dead' * hook so that it can still exist after we depart. We then * send the peer its own destroy message. This ensures that we only * interact with the peer's structures when it is locked processing that * message. We hold a reference to the peer hook so we are guaranteed that * the peer hook and node are still going to exist until * we are finished there as the hook holds a ref on the node. * We run this same code again on the peer hook, but that time it is already * attached to the 'dead' hook. * * This routine is called at all stages of hook creation * on error detection and must be able to handle any such stage. */ void ng_destroy_hook(hook_p hook) { hook_p peer; node_p node; if (hook == &ng_deadhook) { /* better safe than sorry */ printf("ng_destroy_hook called on deadhook\n"); return; } /* * Protect divorce process with mutex, to avoid races on * simultaneous disconnect. */ TOPOLOGY_WLOCK(); hook->hk_flags |= HK_INVALID; peer = NG_HOOK_PEER(hook); node = NG_HOOK_NODE(hook); if (peer && (peer != &ng_deadhook)) { /* * Set the peer to point to ng_deadhook * from this moment on we are effectively independent it. * send it an rmhook message of its own. */ peer->hk_peer = &ng_deadhook; /* They no longer know us */ hook->hk_peer = &ng_deadhook; /* Nor us, them */ if (NG_HOOK_NODE(peer) == &ng_deadnode) { /* * If it's already divorced from a node, * just free it. */ TOPOLOGY_WUNLOCK(); } else { TOPOLOGY_WUNLOCK(); ng_rmhook_self(peer); /* Send it a surprise */ } NG_HOOK_UNREF(peer); /* account for peer link */ NG_HOOK_UNREF(hook); /* account for peer link */ } else TOPOLOGY_WUNLOCK(); TOPOLOGY_NOTOWNED(); /* * Remove the hook from the node's list to avoid possible recursion * in case the disconnection results in node shutdown. */ if (node == &ng_deadnode) { /* happens if called from ng_con_nodes() */ return; } LIST_REMOVE(hook, hk_hooks); node->nd_numhooks--; if (node->nd_type->disconnect) { /* * The type handler may elect to destroy the node so don't * trust its existence after this point. (except * that we still hold a reference on it. (which we * inherrited from the hook we are destroying) */ (*node->nd_type->disconnect) (hook); } /* * Note that because we will point to ng_deadnode, the original node * is not decremented automatically so we do that manually. */ _NG_HOOK_NODE(hook) = &ng_deadnode; NG_NODE_UNREF(node); /* We no longer point to it so adjust count */ NG_HOOK_UNREF(hook); /* Account for linkage (in list) to node */ } /* * Take two hooks on a node and merge the connection so that the given node * is effectively bypassed. */ int ng_bypass(hook_p hook1, hook_p hook2) { if (hook1->hk_node != hook2->hk_node) { TRAP_ERROR(); return (EINVAL); } TOPOLOGY_WLOCK(); if (NG_HOOK_NOT_VALID(hook1) || NG_HOOK_NOT_VALID(hook2)) { TOPOLOGY_WUNLOCK(); return (EINVAL); } hook1->hk_peer->hk_peer = hook2->hk_peer; hook2->hk_peer->hk_peer = hook1->hk_peer; hook1->hk_peer = &ng_deadhook; hook2->hk_peer = &ng_deadhook; TOPOLOGY_WUNLOCK(); NG_HOOK_UNREF(hook1); NG_HOOK_UNREF(hook2); /* XXX If we ever cache methods on hooks update them as well */ ng_destroy_hook(hook1); ng_destroy_hook(hook2); return (0); } /* * Install a new netgraph type */ int ng_newtype(struct ng_type *tp) { const size_t namelen = strlen(tp->name); /* Check version and type name fields */ if ((tp->version != NG_ABI_VERSION) || (namelen == 0) || (namelen >= NG_TYPESIZ)) { TRAP_ERROR(); if (tp->version != NG_ABI_VERSION) { printf("Netgraph: Node type rejected. ABI mismatch. " "Suggest recompile\n"); } return (EINVAL); } /* Check for name collision */ if (ng_findtype(tp->name) != NULL) { TRAP_ERROR(); return (EEXIST); } /* Link in new type */ TYPELIST_WLOCK(); LIST_INSERT_HEAD(&ng_typelist, tp, types); tp->refs = 1; /* first ref is linked list */ TYPELIST_WUNLOCK(); return (0); } /* * unlink a netgraph type * If no examples exist */ int ng_rmtype(struct ng_type *tp) { /* Check for name collision */ if (tp->refs != 1) { TRAP_ERROR(); return (EBUSY); } /* Unlink type */ TYPELIST_WLOCK(); LIST_REMOVE(tp, types); TYPELIST_WUNLOCK(); return (0); } /* * Look for a type of the name given */ struct ng_type * ng_findtype(const char *typename) { struct ng_type *type; TYPELIST_RLOCK(); LIST_FOREACH(type, &ng_typelist, types) { if (strcmp(type->name, typename) == 0) break; } TYPELIST_RUNLOCK(); return (type); } /************************************************************************ Composite routines ************************************************************************/ /* * Connect two nodes using the specified hooks, using queued functions. */ static int ng_con_part3(node_p node, item_p item, hook_p hook) { int error = 0; /* * When we run, we know that the node 'node' is locked for us. * Our caller has a reference on the hook. * Our caller has a reference on the node. * (In this case our caller is ng_apply_item() ). * The peer hook has a reference on the hook. * We are all set up except for the final call to the node, and * the clearing of the INVALID flag. */ if (NG_HOOK_NODE(hook) == &ng_deadnode) { /* * The node must have been freed again since we last visited * here. ng_destry_hook() has this effect but nothing else does. * We should just release our references and * free anything we can think of. * Since we know it's been destroyed, and it's our caller * that holds the references, just return. */ ERROUT(ENOENT); } if (hook->hk_node->nd_type->connect) { if ((error = (*hook->hk_node->nd_type->connect) (hook))) { ng_destroy_hook(hook); /* also zaps peer */ printf("failed in ng_con_part3()\n"); ERROUT(error); } } /* * XXX this is wrong for SMP. Possibly we need * to separate out 'create' and 'invalid' flags. * should only set flags on hooks we have locked under our node. */ hook->hk_flags &= ~HK_INVALID; done: NG_FREE_ITEM(item); return (error); } static int ng_con_part2(node_p node, item_p item, hook_p hook) { hook_p peer; int error = 0; /* * When we run, we know that the node 'node' is locked for us. * Our caller has a reference on the hook. * Our caller has a reference on the node. * (In this case our caller is ng_apply_item() ). * The peer hook has a reference on the hook. * our node pointer points to the 'dead' node. * First check the hook name is unique. * Should not happen because we checked before queueing this. */ if (ng_findhook(node, NG_HOOK_NAME(hook)) != NULL) { TRAP_ERROR(); ng_destroy_hook(hook); /* should destroy peer too */ printf("failed in ng_con_part2()\n"); ERROUT(EEXIST); } /* * Check if the node type code has something to say about it * If it fails, the unref of the hook will also unref the attached node, * however since that node is 'ng_deadnode' this will do nothing. * The peer hook will also be destroyed. */ if (node->nd_type->newhook != NULL) { if ((error = (*node->nd_type->newhook)(node, hook, hook->hk_name))) { ng_destroy_hook(hook); /* should destroy peer too */ printf("failed in ng_con_part2()\n"); ERROUT(error); } } /* * The 'type' agrees so far, so go ahead and link it in. * We'll ask again later when we actually connect the hooks. */ hook->hk_node = node; /* just overwrite ng_deadnode */ NG_NODE_REF(node); /* each hook counts as a reference */ LIST_INSERT_HEAD(&node->nd_hooks, hook, hk_hooks); node->nd_numhooks++; NG_HOOK_REF(hook); /* one for the node */ /* * We now have a symmetrical situation, where both hooks have been * linked to their nodes, the newhook methods have been called * And the references are all correct. The hooks are still marked * as invalid, as we have not called the 'connect' methods * yet. * We can call the local one immediately as we have the * node locked, but we need to queue the remote one. */ if (hook->hk_node->nd_type->connect) { if ((error = (*hook->hk_node->nd_type->connect) (hook))) { ng_destroy_hook(hook); /* also zaps peer */ printf("failed in ng_con_part2(A)\n"); ERROUT(error); } } /* * Acquire topo mutex to avoid race with ng_destroy_hook(). */ TOPOLOGY_RLOCK(); peer = hook->hk_peer; if (peer == &ng_deadhook) { TOPOLOGY_RUNLOCK(); printf("failed in ng_con_part2(B)\n"); ng_destroy_hook(hook); ERROUT(ENOENT); } TOPOLOGY_RUNLOCK(); if ((error = ng_send_fn2(peer->hk_node, peer, item, &ng_con_part3, NULL, 0, NG_REUSE_ITEM))) { printf("failed in ng_con_part2(C)\n"); ng_destroy_hook(hook); /* also zaps peer */ return (error); /* item was consumed. */ } hook->hk_flags &= ~HK_INVALID; /* need both to be able to work */ return (0); /* item was consumed. */ done: NG_FREE_ITEM(item); return (error); } /* * Connect this node with another node. We assume that this node is * currently locked, as we are only called from an NGM_CONNECT message. */ static int ng_con_nodes(item_p item, node_p node, const char *name, node_p node2, const char *name2) { int error; hook_p hook; hook_p hook2; if (ng_findhook(node2, name2) != NULL) { return(EEXIST); } if ((error = ng_add_hook(node, name, &hook))) /* gives us a ref */ return (error); /* Allocate the other hook and link it up */ NG_ALLOC_HOOK(hook2); if (hook2 == NULL) { TRAP_ERROR(); ng_destroy_hook(hook); /* XXX check ref counts so far */ NG_HOOK_UNREF(hook); /* including our ref */ return (ENOMEM); } hook2->hk_refs = 1; /* start with a reference for us. */ hook2->hk_flags = HK_INVALID; hook2->hk_peer = hook; /* Link the two together */ hook->hk_peer = hook2; NG_HOOK_REF(hook); /* Add a ref for the peer to each*/ NG_HOOK_REF(hook2); hook2->hk_node = &ng_deadnode; strlcpy(NG_HOOK_NAME(hook2), name2, NG_HOOKSIZ); /* * Queue the function above. * Procesing continues in that function in the lock context of * the other node. */ if ((error = ng_send_fn2(node2, hook2, item, &ng_con_part2, NULL, 0, NG_NOFLAGS))) { printf("failed in ng_con_nodes(): %d\n", error); ng_destroy_hook(hook); /* also zaps peer */ } NG_HOOK_UNREF(hook); /* Let each hook go if it wants to */ NG_HOOK_UNREF(hook2); return (error); } /* * Make a peer and connect. * We assume that the local node is locked. * The new node probably doesn't need a lock until * it has a hook, because it cannot really have any work until then, * but we should think about it a bit more. * * The problem may come if the other node also fires up * some hardware or a timer or some other source of activation, * also it may already get a command msg via it's ID. * * We could use the same method as ng_con_nodes() but we'd have * to add ability to remove the node when failing. (Not hard, just * make arg1 point to the node to remove). * Unless of course we just ignore failure to connect and leave * an unconnected node? */ static int ng_mkpeer(node_p node, const char *name, const char *name2, char *type) { node_p node2; hook_p hook1, hook2; int error; if ((error = ng_make_node(type, &node2))) { return (error); } if ((error = ng_add_hook(node, name, &hook1))) { /* gives us a ref */ ng_rmnode(node2, NULL, NULL, 0); return (error); } if ((error = ng_add_hook(node2, name2, &hook2))) { ng_rmnode(node2, NULL, NULL, 0); ng_destroy_hook(hook1); NG_HOOK_UNREF(hook1); return (error); } /* * Actually link the two hooks together. */ hook1->hk_peer = hook2; hook2->hk_peer = hook1; /* Each hook is referenced by the other */ NG_HOOK_REF(hook1); NG_HOOK_REF(hook2); /* Give each node the opportunity to veto the pending connection */ if (hook1->hk_node->nd_type->connect) { error = (*hook1->hk_node->nd_type->connect) (hook1); } if ((error == 0) && hook2->hk_node->nd_type->connect) { error = (*hook2->hk_node->nd_type->connect) (hook2); } /* * drop the references we were holding on the two hooks. */ if (error) { ng_destroy_hook(hook2); /* also zaps hook1 */ ng_rmnode(node2, NULL, NULL, 0); } else { /* As a last act, allow the hooks to be used */ hook1->hk_flags &= ~HK_INVALID; hook2->hk_flags &= ~HK_INVALID; } NG_HOOK_UNREF(hook1); NG_HOOK_UNREF(hook2); return (error); } /************************************************************************ Utility routines to send self messages ************************************************************************/ /* Shut this node down as soon as everyone is clear of it */ /* Should add arg "immediately" to jump the queue */ int ng_rmnode_self(node_p node) { int error; if (node == &ng_deadnode) return (0); node->nd_flags |= NGF_INVALID; if (node->nd_flags & NGF_CLOSING) return (0); error = ng_send_fn(node, NULL, &ng_rmnode, NULL, 0); return (error); } static void ng_rmhook_part2(node_p node, hook_p hook, void *arg1, int arg2) { ng_destroy_hook(hook); return ; } int ng_rmhook_self(hook_p hook) { int error; node_p node = NG_HOOK_NODE(hook); if (node == &ng_deadnode) return (0); error = ng_send_fn(node, hook, &ng_rmhook_part2, NULL, 0); return (error); } /*********************************************************************** * Parse and verify a string of the form: * * Such a string can refer to a specific node or a specific hook * on a specific node, depending on how you look at it. In the * latter case, the PATH component must not end in a dot. * * Both and are optional. The is a string * of hook names separated by dots. This breaks out the original * string, setting *nodep to "NODE" (or NULL if none) and *pathp * to "PATH" (or NULL if degenerate). Also, *hookp will point to * the final hook component of , if any, otherwise NULL. * * This returns -1 if the path is malformed. The char ** are optional. ***********************************************************************/ int ng_path_parse(char *addr, char **nodep, char **pathp, char **hookp) { char *node, *path, *hook; int k; /* * Extract absolute NODE, if any */ for (path = addr; *path && *path != ':'; path++); if (*path) { node = addr; /* Here's the NODE */ *path++ = '\0'; /* Here's the PATH */ /* Node name must not be empty */ if (!*node) return -1; /* A name of "." is OK; otherwise '.' not allowed */ if (strcmp(node, ".") != 0) { for (k = 0; node[k]; k++) if (node[k] == '.') return -1; } } else { node = NULL; /* No absolute NODE */ path = addr; /* Here's the PATH */ } /* Snoop for illegal characters in PATH */ for (k = 0; path[k]; k++) if (path[k] == ':') return -1; /* Check for no repeated dots in PATH */ for (k = 0; path[k]; k++) if (path[k] == '.' && path[k + 1] == '.') return -1; /* Remove extra (degenerate) dots from beginning or end of PATH */ if (path[0] == '.') path++; if (*path && path[strlen(path) - 1] == '.') path[strlen(path) - 1] = 0; /* If PATH has a dot, then we're not talking about a hook */ if (*path) { for (hook = path, k = 0; path[k]; k++) if (path[k] == '.') { hook = NULL; break; } } else path = hook = NULL; /* Done */ if (nodep) *nodep = node; if (pathp) *pathp = path; if (hookp) *hookp = hook; return (0); } /* * Given a path, which may be absolute or relative, and a starting node, * return the destination node. */ int ng_path2noderef(node_p here, const char *address, node_p *destp, hook_p *lasthook) { char fullpath[NG_PATHSIZ]; char *nodename, *path; node_p node, oldnode; /* Initialize */ if (destp == NULL) { TRAP_ERROR(); return EINVAL; } *destp = NULL; /* Make a writable copy of address for ng_path_parse() */ strncpy(fullpath, address, sizeof(fullpath) - 1); fullpath[sizeof(fullpath) - 1] = '\0'; /* Parse out node and sequence of hooks */ if (ng_path_parse(fullpath, &nodename, &path, NULL) < 0) { TRAP_ERROR(); return EINVAL; } /* * For an absolute address, jump to the starting node. * Note that this holds a reference on the node for us. * Don't forget to drop the reference if we don't need it. */ if (nodename) { node = ng_name2noderef(here, nodename); if (node == NULL) { TRAP_ERROR(); return (ENOENT); } } else { if (here == NULL) { TRAP_ERROR(); return (EINVAL); } node = here; NG_NODE_REF(node); } if (path == NULL) { if (lasthook != NULL) *lasthook = NULL; *destp = node; return (0); } /* * Now follow the sequence of hooks * * XXXGL: The path may demolish as we go the sequence, but if * we hold the topology mutex at critical places, then, I hope, * we would always have valid pointers in hand, although the * path behind us may no longer exist. */ for (;;) { hook_p hook; char *segment; /* * Break out the next path segment. Replace the dot we just * found with a NUL; "path" points to the next segment (or the * NUL at the end). */ for (segment = path; *path != '\0'; path++) { if (*path == '.') { *path++ = '\0'; break; } } /* We have a segment, so look for a hook by that name */ hook = ng_findhook(node, segment); TOPOLOGY_WLOCK(); /* Can't get there from here... */ if (hook == NULL || NG_HOOK_PEER(hook) == NULL || NG_HOOK_NOT_VALID(hook) || NG_HOOK_NOT_VALID(NG_HOOK_PEER(hook))) { TRAP_ERROR(); NG_NODE_UNREF(node); TOPOLOGY_WUNLOCK(); return (ENOENT); } /* * Hop on over to the next node * XXX * Big race conditions here as hooks and nodes go away * *** Idea.. store an ng_ID_t in each hook and use that * instead of the direct hook in this crawl? */ oldnode = node; if ((node = NG_PEER_NODE(hook))) NG_NODE_REF(node); /* XXX RACE */ NG_NODE_UNREF(oldnode); /* XXX another race */ if (NG_NODE_NOT_VALID(node)) { NG_NODE_UNREF(node); /* XXX more races */ TOPOLOGY_WUNLOCK(); TRAP_ERROR(); return (ENXIO); } if (*path == '\0') { if (lasthook != NULL) { if (hook != NULL) { *lasthook = NG_HOOK_PEER(hook); NG_HOOK_REF(*lasthook); } else *lasthook = NULL; } TOPOLOGY_WUNLOCK(); *destp = node; return (0); } TOPOLOGY_WUNLOCK(); } } /***************************************************************\ * Input queue handling. * All activities are submitted to the node via the input queue * which implements a multiple-reader/single-writer gate. * Items which cannot be handled immediately are queued. * * read-write queue locking inline functions * \***************************************************************/ static __inline void ng_queue_rw(node_p node, item_p item, int rw); static __inline item_p ng_dequeue(node_p node, int *rw); static __inline item_p ng_acquire_read(node_p node, item_p item); static __inline item_p ng_acquire_write(node_p node, item_p item); static __inline void ng_leave_read(node_p node); static __inline void ng_leave_write(node_p node); /* * Definition of the bits fields in the ng_queue flag word. * Defined here rather than in netgraph.h because no-one should fiddle * with them. * * The ordering here may be important! don't shuffle these. */ /*- Safety Barrier--------+ (adjustable to suit taste) (not used yet) | V +-------+-------+-------+-------+-------+-------+-------+-------+ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |A|c|t|i|v|e| |R|e|a|d|e|r| |C|o|u|n|t| | | | | | | | | |P|A| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |O|W| +-------+-------+-------+-------+-------+-------+-------+-------+ \___________________________ ____________________________/ | | V | | [active reader count] | | | | Operation Pending -------------------------------+ | | Active Writer ---------------------------------------+ Node queue has such semantics: - All flags modifications are atomic. - Reader count can be incremented only if there is no writer or pending flags. As soon as this can't be done with single operation, it is implemented with spin loop and atomic_cmpset(). - Writer flag can be set only if there is no any bits set. It is implemented with atomic_cmpset(). - Pending flag can be set any time, but to avoid collision on queue processing all queue fields are protected by the mutex. - Queue processing thread reads queue holding the mutex, but releases it while processing. When queue is empty pending flag is removed. */ #define WRITER_ACTIVE 0x00000001 #define OP_PENDING 0x00000002 #define READER_INCREMENT 0x00000004 #define READER_MASK 0xfffffffc /* Not valid if WRITER_ACTIVE is set */ #define SAFETY_BARRIER 0x00100000 /* 128K items queued should be enough */ /* Defines of more elaborate states on the queue */ /* Mask of bits a new read cares about */ #define NGQ_RMASK (WRITER_ACTIVE|OP_PENDING) /* Mask of bits a new write cares about */ #define NGQ_WMASK (NGQ_RMASK|READER_MASK) /* Test to decide if there is something on the queue. */ #define QUEUE_ACTIVE(QP) ((QP)->q_flags & OP_PENDING) /* How to decide what the next queued item is. */ #define HEAD_IS_READER(QP) NGI_QUEUED_READER(STAILQ_FIRST(&(QP)->queue)) #define HEAD_IS_WRITER(QP) NGI_QUEUED_WRITER(STAILQ_FIRST(&(QP)->queue)) /* notused */ /* Read the status to decide if the next item on the queue can now run. */ #define QUEUED_READER_CAN_PROCEED(QP) \ (((QP)->q_flags & (NGQ_RMASK & ~OP_PENDING)) == 0) #define QUEUED_WRITER_CAN_PROCEED(QP) \ (((QP)->q_flags & (NGQ_WMASK & ~OP_PENDING)) == 0) /* Is there a chance of getting ANY work off the queue? */ #define NEXT_QUEUED_ITEM_CAN_PROCEED(QP) \ ((HEAD_IS_READER(QP)) ? QUEUED_READER_CAN_PROCEED(QP) : \ QUEUED_WRITER_CAN_PROCEED(QP)) #define NGQRW_R 0 #define NGQRW_W 1 #define NGQ2_WORKQ 0x00000001 /* * Taking into account the current state of the queue and node, possibly take * the next entry off the queue and return it. Return NULL if there was * nothing we could return, either because there really was nothing there, or * because the node was in a state where it cannot yet process the next item * on the queue. */ static __inline item_p ng_dequeue(node_p node, int *rw) { item_p item; struct ng_queue *ngq = &node->nd_input_queue; /* This MUST be called with the mutex held. */ mtx_assert(&ngq->q_mtx, MA_OWNED); /* If there is nothing queued, then just return. */ if (!QUEUE_ACTIVE(ngq)) { CTR4(KTR_NET, "%20s: node [%x] (%p) queue empty; " "queue flags 0x%lx", __func__, node->nd_ID, node, ngq->q_flags); return (NULL); } /* * From here, we can assume there is a head item. * We need to find out what it is and if it can be dequeued, given * the current state of the node. */ if (HEAD_IS_READER(ngq)) { while (1) { long t = ngq->q_flags; if (t & WRITER_ACTIVE) { /* There is writer, reader can't proceed. */ CTR4(KTR_NET, "%20s: node [%x] (%p) queued " "reader can't proceed; queue flags 0x%lx", __func__, node->nd_ID, node, t); return (NULL); } if (atomic_cmpset_acq_int(&ngq->q_flags, t, t + READER_INCREMENT)) break; cpu_spinwait(); } /* We have got reader lock for the node. */ *rw = NGQRW_R; } else if (atomic_cmpset_acq_int(&ngq->q_flags, OP_PENDING, OP_PENDING + WRITER_ACTIVE)) { /* We have got writer lock for the node. */ *rw = NGQRW_W; } else { /* There is somebody other, writer can't proceed. */ CTR4(KTR_NET, "%20s: node [%x] (%p) queued writer can't " "proceed; queue flags 0x%lx", __func__, node->nd_ID, node, ngq->q_flags); return (NULL); } /* * Now we dequeue the request (whatever it may be) and correct the * pending flags and the next and last pointers. */ item = STAILQ_FIRST(&ngq->queue); STAILQ_REMOVE_HEAD(&ngq->queue, el_next); if (STAILQ_EMPTY(&ngq->queue)) atomic_clear_int(&ngq->q_flags, OP_PENDING); CTR6(KTR_NET, "%20s: node [%x] (%p) returning item %p as %s; queue " "flags 0x%lx", __func__, node->nd_ID, node, item, *rw ? "WRITER" : "READER", ngq->q_flags); return (item); } /* * Queue a packet to be picked up later by someone else. * If the queue could be run now, add node to the queue handler's worklist. */ static __inline void ng_queue_rw(node_p node, item_p item, int rw) { struct ng_queue *ngq = &node->nd_input_queue; if (rw == NGQRW_W) NGI_SET_WRITER(item); else NGI_SET_READER(item); item->depth = 1; NG_QUEUE_LOCK(ngq); /* Set OP_PENDING flag and enqueue the item. */ atomic_set_int(&ngq->q_flags, OP_PENDING); STAILQ_INSERT_TAIL(&ngq->queue, item, el_next); CTR5(KTR_NET, "%20s: node [%x] (%p) queued item %p as %s", __func__, node->nd_ID, node, item, rw ? "WRITER" : "READER" ); /* * We can take the worklist lock with the node locked * BUT NOT THE REVERSE! */ if (NEXT_QUEUED_ITEM_CAN_PROCEED(ngq)) ng_worklist_add(node); NG_QUEUE_UNLOCK(ngq); } /* Acquire reader lock on node. If node is busy, queue the packet. */ static __inline item_p ng_acquire_read(node_p node, item_p item) { KASSERT(node != &ng_deadnode, ("%s: working on deadnode", __func__)); /* Reader needs node without writer and pending items. */ for (;;) { long t = node->nd_input_queue.q_flags; if (t & NGQ_RMASK) break; /* Node is not ready for reader. */ if (atomic_cmpset_acq_int(&node->nd_input_queue.q_flags, t, t + READER_INCREMENT)) { /* Successfully grabbed node */ CTR4(KTR_NET, "%20s: node [%x] (%p) acquired item %p", __func__, node->nd_ID, node, item); return (item); } cpu_spinwait(); } /* Queue the request for later. */ ng_queue_rw(node, item, NGQRW_R); return (NULL); } /* Acquire writer lock on node. If node is busy, queue the packet. */ static __inline item_p ng_acquire_write(node_p node, item_p item) { KASSERT(node != &ng_deadnode, ("%s: working on deadnode", __func__)); /* Writer needs completely idle node. */ if (atomic_cmpset_acq_int(&node->nd_input_queue.q_flags, 0, WRITER_ACTIVE)) { /* Successfully grabbed node */ CTR4(KTR_NET, "%20s: node [%x] (%p) acquired item %p", __func__, node->nd_ID, node, item); return (item); } /* Queue the request for later. */ ng_queue_rw(node, item, NGQRW_W); return (NULL); } #if 0 static __inline item_p ng_upgrade_write(node_p node, item_p item) { struct ng_queue *ngq = &node->nd_input_queue; KASSERT(node != &ng_deadnode, ("%s: working on deadnode", __func__)); NGI_SET_WRITER(item); NG_QUEUE_LOCK(ngq); /* * There will never be no readers as we are there ourselves. * Set the WRITER_ACTIVE flags ASAP to block out fast track readers. * The caller we are running from will call ng_leave_read() * soon, so we must account for that. We must leave again with the * READER lock. If we find other readers, then * queue the request for later. However "later" may be rignt now * if there are no readers. We don't really care if there are queued * items as we will bypass them anyhow. */ atomic_add_int(&ngq->q_flags, WRITER_ACTIVE - READER_INCREMENT); if ((ngq->q_flags & (NGQ_WMASK & ~OP_PENDING)) == WRITER_ACTIVE) { NG_QUEUE_UNLOCK(ngq); /* It's just us, act on the item. */ /* will NOT drop writer lock when done */ ng_apply_item(node, item, 0); /* * Having acted on the item, atomically * downgrade back to READER and finish up. */ atomic_add_int(&ngq->q_flags, READER_INCREMENT - WRITER_ACTIVE); /* Our caller will call ng_leave_read() */ return; } /* * It's not just us active, so queue us AT THE HEAD. * "Why?" I hear you ask. * Put us at the head of the queue as we've already been * through it once. If there is nothing else waiting, * set the correct flags. */ if (STAILQ_EMPTY(&ngq->queue)) { /* We've gone from, 0 to 1 item in the queue */ atomic_set_int(&ngq->q_flags, OP_PENDING); CTR3(KTR_NET, "%20s: node [%x] (%p) set OP_PENDING", __func__, node->nd_ID, node); }; STAILQ_INSERT_HEAD(&ngq->queue, item, el_next); CTR4(KTR_NET, "%20s: node [%x] (%p) requeued item %p as WRITER", __func__, node->nd_ID, node, item ); /* Reverse what we did above. That downgrades us back to reader */ atomic_add_int(&ngq->q_flags, READER_INCREMENT - WRITER_ACTIVE); if (QUEUE_ACTIVE(ngq) && NEXT_QUEUED_ITEM_CAN_PROCEED(ngq)) ng_worklist_add(node); NG_QUEUE_UNLOCK(ngq); return; } #endif /* Release reader lock. */ static __inline void ng_leave_read(node_p node) { atomic_subtract_rel_int(&node->nd_input_queue.q_flags, READER_INCREMENT); } /* Release writer lock. */ static __inline void ng_leave_write(node_p node) { atomic_clear_rel_int(&node->nd_input_queue.q_flags, WRITER_ACTIVE); } /* Purge node queue. Called on node shutdown. */ static void ng_flush_input_queue(node_p node) { struct ng_queue *ngq = &node->nd_input_queue; item_p item; NG_QUEUE_LOCK(ngq); while ((item = STAILQ_FIRST(&ngq->queue)) != NULL) { STAILQ_REMOVE_HEAD(&ngq->queue, el_next); if (STAILQ_EMPTY(&ngq->queue)) atomic_clear_int(&ngq->q_flags, OP_PENDING); NG_QUEUE_UNLOCK(ngq); /* If the item is supplying a callback, call it with an error */ if (item->apply != NULL) { if (item->depth == 1) item->apply->error = ENOENT; if (refcount_release(&item->apply->refs)) { (*item->apply->apply)(item->apply->context, item->apply->error); } } NG_FREE_ITEM(item); NG_QUEUE_LOCK(ngq); } NG_QUEUE_UNLOCK(ngq); } /*********************************************************************** * Externally visible method for sending or queueing messages or data. ***********************************************************************/ /* * The module code should have filled out the item correctly by this stage: * Common: * reference to destination node. * Reference to destination rcv hook if relevant. * apply pointer must be or NULL or reference valid struct ng_apply_info. * Data: * pointer to mbuf * Control_Message: * pointer to msg. * ID of original sender node. (return address) * Function: * Function pointer * void * argument * integer argument * * The nodes have several routines and macros to help with this task: */ int ng_snd_item(item_p item, int flags) { hook_p hook; node_p node; int queue, rw; struct ng_queue *ngq; int error = 0; /* We are sending item, so it must be present! */ KASSERT(item != NULL, ("ng_snd_item: item is NULL")); #ifdef NETGRAPH_DEBUG _ngi_check(item, __FILE__, __LINE__); #endif /* Item was sent once more, postpone apply() call. */ if (item->apply) refcount_acquire(&item->apply->refs); node = NGI_NODE(item); /* Node is never optional. */ KASSERT(node != NULL, ("ng_snd_item: node is NULL")); hook = NGI_HOOK(item); /* Valid hook and mbuf are mandatory for data. */ if ((item->el_flags & NGQF_TYPE) == NGQF_DATA) { KASSERT(hook != NULL, ("ng_snd_item: hook for data is NULL")); if (NGI_M(item) == NULL) ERROUT(EINVAL); CHECK_DATA_MBUF(NGI_M(item)); } /* * If the item or the node specifies single threading, force * writer semantics. Similarly, the node may say one hook always * produces writers. These are overrides. */ if (((item->el_flags & NGQF_RW) == NGQF_WRITER) || (node->nd_flags & NGF_FORCE_WRITER) || (hook && (hook->hk_flags & HK_FORCE_WRITER))) { rw = NGQRW_W; } else { rw = NGQRW_R; } /* * If sender or receiver requests queued delivery, or call graph * loops back from outbound to inbound path, or stack usage * level is dangerous - enqueue message. */ if ((flags & NG_QUEUE) || (hook && (hook->hk_flags & HK_QUEUE))) { queue = 1; } else if (hook && (hook->hk_flags & HK_TO_INBOUND) && curthread->td_ng_outbound) { queue = 1; } else { queue = 0; #ifdef GET_STACK_USAGE /* * Most of netgraph nodes have small stack consumption and * for them 25% of free stack space is more than enough. * Nodes/hooks with higher stack usage should be marked as * HI_STACK. For them 50% of stack will be guaranteed then. * XXX: Values 25% and 50% are completely empirical. */ size_t st, su, sl; GET_STACK_USAGE(st, su); sl = st - su; if ((sl * 4 < st) || ((sl * 2 < st) && ((node->nd_flags & NGF_HI_STACK) || (hook && (hook->hk_flags & HK_HI_STACK))))) queue = 1; #endif } if (queue) { /* Put it on the queue for that node*/ ng_queue_rw(node, item, rw); return ((flags & NG_PROGRESS) ? EINPROGRESS : 0); } /* * We already decided how we will be queueud or treated. * Try get the appropriate operating permission. */ if (rw == NGQRW_R) item = ng_acquire_read(node, item); else item = ng_acquire_write(node, item); /* Item was queued while trying to get permission. */ if (item == NULL) return ((flags & NG_PROGRESS) ? EINPROGRESS : 0); NGI_GET_NODE(item, node); /* zaps stored node */ item->depth++; error = ng_apply_item(node, item, rw); /* drops r/w lock when done */ /* If something is waiting on queue and ready, schedule it. */ ngq = &node->nd_input_queue; if (QUEUE_ACTIVE(ngq)) { NG_QUEUE_LOCK(ngq); if (QUEUE_ACTIVE(ngq) && NEXT_QUEUED_ITEM_CAN_PROCEED(ngq)) ng_worklist_add(node); NG_QUEUE_UNLOCK(ngq); } /* * Node may go away as soon as we remove the reference. * Whatever we do, DO NOT access the node again! */ NG_NODE_UNREF(node); return (error); done: /* If was not sent, apply callback here. */ if (item->apply != NULL) { if (item->depth == 0 && error != 0) item->apply->error = error; if (refcount_release(&item->apply->refs)) { (*item->apply->apply)(item->apply->context, item->apply->error); } } NG_FREE_ITEM(item); return (error); } /* * We have an item that was possibly queued somewhere. * It should contain all the information needed * to run it on the appropriate node/hook. * If there is apply pointer and we own the last reference, call apply(). */ static int ng_apply_item(node_p node, item_p item, int rw) { hook_p hook; ng_rcvdata_t *rcvdata; ng_rcvmsg_t *rcvmsg; struct ng_apply_info *apply; int error = 0, depth; /* Node and item are never optional. */ KASSERT(node != NULL, ("ng_apply_item: node is NULL")); KASSERT(item != NULL, ("ng_apply_item: item is NULL")); NGI_GET_HOOK(item, hook); /* clears stored hook */ #ifdef NETGRAPH_DEBUG _ngi_check(item, __FILE__, __LINE__); #endif apply = item->apply; depth = item->depth; switch (item->el_flags & NGQF_TYPE) { case NGQF_DATA: /* * Check things are still ok as when we were queued. */ KASSERT(hook != NULL, ("ng_apply_item: hook for data is NULL")); if (NG_HOOK_NOT_VALID(hook) || NG_NODE_NOT_VALID(node)) { error = EIO; NG_FREE_ITEM(item); break; } /* * If no receive method, just silently drop it. * Give preference to the hook over-ride method. */ if ((!(rcvdata = hook->hk_rcvdata)) && (!(rcvdata = NG_HOOK_NODE(hook)->nd_type->rcvdata))) { error = 0; NG_FREE_ITEM(item); break; } error = (*rcvdata)(hook, item); break; case NGQF_MESG: if (hook && NG_HOOK_NOT_VALID(hook)) { /* * The hook has been zapped then we can't use it. * Immediately drop its reference. * The message may not need it. */ NG_HOOK_UNREF(hook); hook = NULL; } /* * Similarly, if the node is a zombie there is * nothing we can do with it, drop everything. */ if (NG_NODE_NOT_VALID(node)) { TRAP_ERROR(); error = EINVAL; NG_FREE_ITEM(item); break; } /* * Call the appropriate message handler for the object. * It is up to the message handler to free the message. * If it's a generic message, handle it generically, * otherwise call the type's message handler (if it exists). * XXX (race). Remember that a queued message may * reference a node or hook that has just been * invalidated. It will exist as the queue code * is holding a reference, but.. */ if ((NGI_MSG(item)->header.typecookie == NGM_GENERIC_COOKIE) && ((NGI_MSG(item)->header.flags & NGF_RESP) == 0)) { error = ng_generic_msg(node, item, hook); break; } if (((!hook) || (!(rcvmsg = hook->hk_rcvmsg))) && (!(rcvmsg = node->nd_type->rcvmsg))) { TRAP_ERROR(); error = 0; NG_FREE_ITEM(item); break; } error = (*rcvmsg)(node, item, hook); break; case NGQF_FN: case NGQF_FN2: /* * In the case of the shutdown message we allow it to hit * even if the node is invalid. */ if (NG_NODE_NOT_VALID(node) && NGI_FN(item) != &ng_rmnode) { TRAP_ERROR(); error = EINVAL; NG_FREE_ITEM(item); break; } /* Same is about some internal functions and invalid hook. */ if (hook && NG_HOOK_NOT_VALID(hook) && NGI_FN2(item) != &ng_con_part2 && NGI_FN2(item) != &ng_con_part3 && NGI_FN(item) != &ng_rmhook_part2) { TRAP_ERROR(); error = EINVAL; NG_FREE_ITEM(item); break; } if ((item->el_flags & NGQF_TYPE) == NGQF_FN) { (*NGI_FN(item))(node, hook, NGI_ARG1(item), NGI_ARG2(item)); NG_FREE_ITEM(item); } else /* it is NGQF_FN2 */ error = (*NGI_FN2(item))(node, item, hook); break; } /* * We held references on some of the resources * that we took from the item. Now that we have * finished doing everything, drop those references. */ if (hook) NG_HOOK_UNREF(hook); if (rw == NGQRW_R) ng_leave_read(node); else ng_leave_write(node); /* Apply callback. */ if (apply != NULL) { if (depth == 1 && error != 0) apply->error = error; if (refcount_release(&apply->refs)) (*apply->apply)(apply->context, apply->error); } return (error); } /*********************************************************************** * Implement the 'generic' control messages ***********************************************************************/ static int ng_generic_msg(node_p here, item_p item, hook_p lasthook) { int error = 0; struct ng_mesg *msg; struct ng_mesg *resp = NULL; NGI_GET_MSG(item, msg); if (msg->header.typecookie != NGM_GENERIC_COOKIE) { TRAP_ERROR(); error = EINVAL; goto out; } switch (msg->header.cmd) { case NGM_SHUTDOWN: ng_rmnode(here, NULL, NULL, 0); break; case NGM_MKPEER: { struct ngm_mkpeer *const mkp = (struct ngm_mkpeer *) msg->data; if (msg->header.arglen != sizeof(*mkp)) { TRAP_ERROR(); error = EINVAL; break; } mkp->type[sizeof(mkp->type) - 1] = '\0'; mkp->ourhook[sizeof(mkp->ourhook) - 1] = '\0'; mkp->peerhook[sizeof(mkp->peerhook) - 1] = '\0'; error = ng_mkpeer(here, mkp->ourhook, mkp->peerhook, mkp->type); break; } case NGM_CONNECT: { struct ngm_connect *const con = (struct ngm_connect *) msg->data; node_p node2; if (msg->header.arglen != sizeof(*con)) { TRAP_ERROR(); error = EINVAL; break; } con->path[sizeof(con->path) - 1] = '\0'; con->ourhook[sizeof(con->ourhook) - 1] = '\0'; con->peerhook[sizeof(con->peerhook) - 1] = '\0'; /* Don't forget we get a reference.. */ error = ng_path2noderef(here, con->path, &node2, NULL); if (error) break; error = ng_con_nodes(item, here, con->ourhook, node2, con->peerhook); NG_NODE_UNREF(node2); break; } case NGM_NAME: { struct ngm_name *const nam = (struct ngm_name *) msg->data; if (msg->header.arglen != sizeof(*nam)) { TRAP_ERROR(); error = EINVAL; break; } nam->name[sizeof(nam->name) - 1] = '\0'; error = ng_name_node(here, nam->name); break; } case NGM_RMHOOK: { struct ngm_rmhook *const rmh = (struct ngm_rmhook *) msg->data; hook_p hook; if (msg->header.arglen != sizeof(*rmh)) { TRAP_ERROR(); error = EINVAL; break; } rmh->ourhook[sizeof(rmh->ourhook) - 1] = '\0'; if ((hook = ng_findhook(here, rmh->ourhook)) != NULL) ng_destroy_hook(hook); break; } case NGM_NODEINFO: { struct nodeinfo *ni; NG_MKRESPONSE(resp, msg, sizeof(*ni), M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } /* Fill in node info */ ni = (struct nodeinfo *) resp->data; if (NG_NODE_HAS_NAME(here)) strcpy(ni->name, NG_NODE_NAME(here)); strcpy(ni->type, here->nd_type->name); ni->id = ng_node2ID(here); ni->hooks = here->nd_numhooks; break; } case NGM_LISTHOOKS: { const int nhooks = here->nd_numhooks; struct hooklist *hl; struct nodeinfo *ni; hook_p hook; /* Get response struct */ NG_MKRESPONSE(resp, msg, sizeof(*hl) + (nhooks * sizeof(struct linkinfo)), M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } hl = (struct hooklist *) resp->data; ni = &hl->nodeinfo; /* Fill in node info */ if (NG_NODE_HAS_NAME(here)) strcpy(ni->name, NG_NODE_NAME(here)); strcpy(ni->type, here->nd_type->name); ni->id = ng_node2ID(here); /* Cycle through the linked list of hooks */ ni->hooks = 0; LIST_FOREACH(hook, &here->nd_hooks, hk_hooks) { struct linkinfo *const link = &hl->link[ni->hooks]; if (ni->hooks >= nhooks) { log(LOG_ERR, "%s: number of %s changed\n", __func__, "hooks"); break; } if (NG_HOOK_NOT_VALID(hook)) continue; strcpy(link->ourhook, NG_HOOK_NAME(hook)); strcpy(link->peerhook, NG_PEER_HOOK_NAME(hook)); if (NG_PEER_NODE_NAME(hook)[0] != '\0') strcpy(link->nodeinfo.name, NG_PEER_NODE_NAME(hook)); strcpy(link->nodeinfo.type, NG_PEER_NODE(hook)->nd_type->name); link->nodeinfo.id = ng_node2ID(NG_PEER_NODE(hook)); link->nodeinfo.hooks = NG_PEER_NODE(hook)->nd_numhooks; ni->hooks++; } break; } case NGM_LISTNODES: { struct namelist *nl; node_p node; int i; IDHASH_RLOCK(); /* Get response struct. */ NG_MKRESPONSE(resp, msg, sizeof(*nl) + (V_ng_nodes * sizeof(struct nodeinfo)), M_NOWAIT | M_ZERO); if (resp == NULL) { IDHASH_RUNLOCK(); error = ENOMEM; break; } nl = (struct namelist *) resp->data; /* Cycle through the lists of nodes. */ nl->numnames = 0; for (i = 0; i <= V_ng_ID_hmask; i++) { LIST_FOREACH(node, &V_ng_ID_hash[i], nd_idnodes) { struct nodeinfo *const np = &nl->nodeinfo[nl->numnames]; if (NG_NODE_NOT_VALID(node)) continue; if (NG_NODE_HAS_NAME(node)) strcpy(np->name, NG_NODE_NAME(node)); strcpy(np->type, node->nd_type->name); np->id = ng_node2ID(node); np->hooks = node->nd_numhooks; KASSERT(nl->numnames < V_ng_nodes, ("%s: no space", __func__)); nl->numnames++; } } IDHASH_RUNLOCK(); break; } case NGM_LISTNAMES: { struct namelist *nl; node_p node; int i; NAMEHASH_RLOCK(); /* Get response struct. */ NG_MKRESPONSE(resp, msg, sizeof(*nl) + (V_ng_named_nodes * sizeof(struct nodeinfo)), M_NOWAIT); if (resp == NULL) { NAMEHASH_RUNLOCK(); error = ENOMEM; break; } nl = (struct namelist *) resp->data; /* Cycle through the lists of nodes. */ nl->numnames = 0; for (i = 0; i <= V_ng_name_hmask; i++) { LIST_FOREACH(node, &V_ng_name_hash[i], nd_nodes) { struct nodeinfo *const np = &nl->nodeinfo[nl->numnames]; if (NG_NODE_NOT_VALID(node)) continue; strcpy(np->name, NG_NODE_NAME(node)); strcpy(np->type, node->nd_type->name); np->id = ng_node2ID(node); np->hooks = node->nd_numhooks; KASSERT(nl->numnames < V_ng_named_nodes, ("%s: no space", __func__)); nl->numnames++; } } NAMEHASH_RUNLOCK(); break; } case NGM_LISTTYPES: { struct typelist *tl; struct ng_type *type; int num = 0; TYPELIST_RLOCK(); /* Count number of types */ LIST_FOREACH(type, &ng_typelist, types) num++; /* Get response struct */ NG_MKRESPONSE(resp, msg, sizeof(*tl) + (num * sizeof(struct typeinfo)), M_NOWAIT); if (resp == NULL) { TYPELIST_RUNLOCK(); error = ENOMEM; break; } tl = (struct typelist *) resp->data; /* Cycle through the linked list of types */ tl->numtypes = 0; LIST_FOREACH(type, &ng_typelist, types) { struct typeinfo *const tp = &tl->typeinfo[tl->numtypes]; strcpy(tp->type_name, type->name); tp->numnodes = type->refs - 1; /* don't count list */ KASSERT(tl->numtypes < num, ("%s: no space", __func__)); tl->numtypes++; } TYPELIST_RUNLOCK(); break; } case NGM_BINARY2ASCII: { int bufSize = 20 * 1024; /* XXX hard coded constant */ const struct ng_parse_type *argstype; const struct ng_cmdlist *c; struct ng_mesg *binary, *ascii; /* Data area must contain a valid netgraph message */ binary = (struct ng_mesg *)msg->data; if (msg->header.arglen < sizeof(struct ng_mesg) || (msg->header.arglen - sizeof(struct ng_mesg) < binary->header.arglen)) { TRAP_ERROR(); error = EINVAL; break; } /* Get a response message with lots of room */ NG_MKRESPONSE(resp, msg, sizeof(*ascii) + bufSize, M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } ascii = (struct ng_mesg *)resp->data; /* Copy binary message header to response message payload */ bcopy(binary, ascii, sizeof(*binary)); /* Find command by matching typecookie and command number */ for (c = here->nd_type->cmdlist; c != NULL && c->name != NULL; c++) { if (binary->header.typecookie == c->cookie && binary->header.cmd == c->cmd) break; } if (c == NULL || c->name == NULL) { for (c = ng_generic_cmds; c->name != NULL; c++) { if (binary->header.typecookie == c->cookie && binary->header.cmd == c->cmd) break; } if (c->name == NULL) { NG_FREE_MSG(resp); error = ENOSYS; break; } } /* Convert command name to ASCII */ snprintf(ascii->header.cmdstr, sizeof(ascii->header.cmdstr), "%s", c->name); /* Convert command arguments to ASCII */ argstype = (binary->header.flags & NGF_RESP) ? c->respType : c->mesgType; if (argstype == NULL) { *ascii->data = '\0'; } else { if ((error = ng_unparse(argstype, (u_char *)binary->data, ascii->data, bufSize)) != 0) { NG_FREE_MSG(resp); break; } } /* Return the result as struct ng_mesg plus ASCII string */ bufSize = strlen(ascii->data) + 1; ascii->header.arglen = bufSize; resp->header.arglen = sizeof(*ascii) + bufSize; break; } case NGM_ASCII2BINARY: { int bufSize = 20 * 1024; /* XXX hard coded constant */ const struct ng_cmdlist *c; const struct ng_parse_type *argstype; struct ng_mesg *ascii, *binary; int off = 0; /* Data area must contain at least a struct ng_mesg + '\0' */ ascii = (struct ng_mesg *)msg->data; if ((msg->header.arglen < sizeof(*ascii) + 1) || (ascii->header.arglen < 1) || (msg->header.arglen < sizeof(*ascii) + ascii->header.arglen)) { TRAP_ERROR(); error = EINVAL; break; } ascii->data[ascii->header.arglen - 1] = '\0'; /* Get a response message with lots of room */ NG_MKRESPONSE(resp, msg, sizeof(*binary) + bufSize, M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } binary = (struct ng_mesg *)resp->data; /* Copy ASCII message header to response message payload */ bcopy(ascii, binary, sizeof(*ascii)); /* Find command by matching ASCII command string */ for (c = here->nd_type->cmdlist; c != NULL && c->name != NULL; c++) { if (strcmp(ascii->header.cmdstr, c->name) == 0) break; } if (c == NULL || c->name == NULL) { for (c = ng_generic_cmds; c->name != NULL; c++) { if (strcmp(ascii->header.cmdstr, c->name) == 0) break; } if (c->name == NULL) { NG_FREE_MSG(resp); error = ENOSYS; break; } } /* Convert command name to binary */ binary->header.cmd = c->cmd; binary->header.typecookie = c->cookie; /* Convert command arguments to binary */ argstype = (binary->header.flags & NGF_RESP) ? c->respType : c->mesgType; if (argstype == NULL) { bufSize = 0; } else { if ((error = ng_parse(argstype, ascii->data, &off, (u_char *)binary->data, &bufSize)) != 0) { NG_FREE_MSG(resp); break; } } /* Return the result */ binary->header.arglen = bufSize; resp->header.arglen = sizeof(*binary) + bufSize; break; } case NGM_TEXT_CONFIG: case NGM_TEXT_STATUS: /* * This one is tricky as it passes the command down to the * actual node, even though it is a generic type command. * This means we must assume that the item/msg is already freed * when control passes back to us. */ if (here->nd_type->rcvmsg != NULL) { NGI_MSG(item) = msg; /* put it back as we found it */ return((*here->nd_type->rcvmsg)(here, item, lasthook)); } /* Fall through if rcvmsg not supported */ default: TRAP_ERROR(); error = EINVAL; } /* * Sometimes a generic message may be statically allocated * to avoid problems with allocating when in tight memory situations. * Don't free it if it is so. * I break them apart here, because erros may cause a free if the item * in which case we'd be doing it twice. * they are kept together above, to simplify freeing. */ out: NG_RESPOND_MSG(error, here, item, resp); NG_FREE_MSG(msg); return (error); } /************************************************************************ Queue element get/free routines ************************************************************************/ uma_zone_t ng_qzone; uma_zone_t ng_qdzone; static int numthreads = 0; /* number of queue threads */ static int maxalloc = 4096;/* limit the damage of a leak */ static int maxdata = 4096; /* limit the damage of a DoS */ SYSCTL_INT(_net_graph, OID_AUTO, threads, CTLFLAG_RDTUN, &numthreads, 0, "Number of queue processing threads"); SYSCTL_INT(_net_graph, OID_AUTO, maxalloc, CTLFLAG_RDTUN, &maxalloc, 0, "Maximum number of non-data queue items to allocate"); SYSCTL_INT(_net_graph, OID_AUTO, maxdata, CTLFLAG_RDTUN, &maxdata, 0, "Maximum number of data queue items to allocate"); #ifdef NETGRAPH_DEBUG static TAILQ_HEAD(, ng_item) ng_itemlist = TAILQ_HEAD_INITIALIZER(ng_itemlist); static int allocated; /* number of items malloc'd */ #endif /* * Get a queue entry. * This is usually called when a packet first enters netgraph. * By definition, this is usually from an interrupt, or from a user. * Users are not so important, but try be quick for the times that it's * an interrupt. */ static __inline item_p ng_alloc_item(int type, int flags) { item_p item; KASSERT(((type & ~NGQF_TYPE) == 0), ("%s: incorrect item type: %d", __func__, type)); item = uma_zalloc((type == NGQF_DATA) ? ng_qdzone : ng_qzone, ((flags & NG_WAITOK) ? M_WAITOK : M_NOWAIT) | M_ZERO); if (item) { item->el_flags = type; #ifdef NETGRAPH_DEBUG mtx_lock(&ngq_mtx); TAILQ_INSERT_TAIL(&ng_itemlist, item, all); allocated++; mtx_unlock(&ngq_mtx); #endif } return (item); } /* * Release a queue entry */ void ng_free_item(item_p item) { /* * The item may hold resources on its own. We need to free * these before we can free the item. What they are depends upon * what kind of item it is. it is important that nodes zero * out pointers to resources that they remove from the item * or we release them again here. */ switch (item->el_flags & NGQF_TYPE) { case NGQF_DATA: /* If we have an mbuf still attached.. */ NG_FREE_M(_NGI_M(item)); break; case NGQF_MESG: _NGI_RETADDR(item) = 0; NG_FREE_MSG(_NGI_MSG(item)); break; case NGQF_FN: case NGQF_FN2: /* nothing to free really, */ _NGI_FN(item) = NULL; _NGI_ARG1(item) = NULL; _NGI_ARG2(item) = 0; break; } /* If we still have a node or hook referenced... */ _NGI_CLR_NODE(item); _NGI_CLR_HOOK(item); #ifdef NETGRAPH_DEBUG mtx_lock(&ngq_mtx); TAILQ_REMOVE(&ng_itemlist, item, all); allocated--; mtx_unlock(&ngq_mtx); #endif uma_zfree(((item->el_flags & NGQF_TYPE) == NGQF_DATA) ? ng_qdzone : ng_qzone, item); } /* * Change type of the queue entry. * Possibly reallocates it from another UMA zone. */ static __inline item_p ng_realloc_item(item_p pitem, int type, int flags) { item_p item; int from, to; KASSERT((pitem != NULL), ("%s: can't reallocate NULL", __func__)); KASSERT(((type & ~NGQF_TYPE) == 0), ("%s: incorrect item type: %d", __func__, type)); from = ((pitem->el_flags & NGQF_TYPE) == NGQF_DATA); to = (type == NGQF_DATA); if (from != to) { /* If reallocation is required do it and copy item. */ if ((item = ng_alloc_item(type, flags)) == NULL) { ng_free_item(pitem); return (NULL); } *item = *pitem; ng_free_item(pitem); } else item = pitem; item->el_flags = (item->el_flags & ~NGQF_TYPE) | type; return (item); } /************************************************************************ Module routines ************************************************************************/ /* * Handle the loading/unloading of a netgraph node type module */ int ng_mod_event(module_t mod, int event, void *data) { struct ng_type *const type = data; int error = 0; switch (event) { case MOD_LOAD: /* Register new netgraph node type */ if ((error = ng_newtype(type)) != 0) break; /* Call type specific code */ if (type->mod_event != NULL) if ((error = (*type->mod_event)(mod, event, data))) { TYPELIST_WLOCK(); type->refs--; /* undo it */ LIST_REMOVE(type, types); TYPELIST_WUNLOCK(); } break; case MOD_UNLOAD: if (type->refs > 1) { /* make sure no nodes exist! */ error = EBUSY; } else { if (type->refs == 0) /* failed load, nothing to undo */ break; if (type->mod_event != NULL) { /* check with type */ error = (*type->mod_event)(mod, event, data); if (error != 0) /* type refuses.. */ break; } TYPELIST_WLOCK(); LIST_REMOVE(type, types); TYPELIST_WUNLOCK(); } break; default: if (type->mod_event != NULL) error = (*type->mod_event)(mod, event, data); else error = EOPNOTSUPP; /* XXX ? */ break; } return (error); } static void vnet_netgraph_init(const void *unused __unused) { /* We start with small hashes, but they can grow. */ V_ng_ID_hash = hashinit(16, M_NETGRAPH_NODE, &V_ng_ID_hmask); V_ng_name_hash = hashinit(16, M_NETGRAPH_NODE, &V_ng_name_hmask); } VNET_SYSINIT(vnet_netgraph_init, SI_SUB_NETGRAPH, SI_ORDER_FIRST, vnet_netgraph_init, NULL); #ifdef VIMAGE static void vnet_netgraph_uninit(const void *unused __unused) { node_p node = NULL, last_killed = NULL; int i; do { /* Find a node to kill */ IDHASH_RLOCK(); for (i = 0; i <= V_ng_ID_hmask; i++) { LIST_FOREACH(node, &V_ng_ID_hash[i], nd_idnodes) { if (node != &ng_deadnode) { NG_NODE_REF(node); break; } } if (node != NULL) break; } IDHASH_RUNLOCK(); /* Attempt to kill it only if it is a regular node */ if (node != NULL) { if (node == last_killed) { /* This should never happen */ printf("ng node %s needs NGF_REALLY_DIE\n", node->nd_name); if (node->nd_flags & NGF_REALLY_DIE) panic("ng node %s won't die", node->nd_name); node->nd_flags |= NGF_REALLY_DIE; } ng_rmnode(node, NULL, NULL, 0); NG_NODE_UNREF(node); last_killed = node; } } while (node != NULL); hashdestroy(V_ng_name_hash, M_NETGRAPH_NODE, V_ng_name_hmask); hashdestroy(V_ng_ID_hash, M_NETGRAPH_NODE, V_ng_ID_hmask); } VNET_SYSUNINIT(vnet_netgraph_uninit, SI_SUB_NETGRAPH, SI_ORDER_FIRST, vnet_netgraph_uninit, NULL); #endif /* VIMAGE */ /* * Handle loading and unloading for this code. * The only thing we need to link into is the NETISR strucure. */ static int ngb_mod_event(module_t mod, int event, void *data) { struct proc *p; struct thread *td; int i, error = 0; switch (event) { case MOD_LOAD: /* Initialize everything. */ NG_WORKLIST_LOCK_INIT(); rw_init(&ng_typelist_lock, "netgraph types"); rw_init(&ng_idhash_lock, "netgraph idhash"); rw_init(&ng_namehash_lock, "netgraph namehash"); rw_init(&ng_topo_lock, "netgraph topology mutex"); #ifdef NETGRAPH_DEBUG mtx_init(&ng_nodelist_mtx, "netgraph nodelist mutex", NULL, MTX_DEF); mtx_init(&ngq_mtx, "netgraph item list mutex", NULL, MTX_DEF); #endif ng_qzone = uma_zcreate("NetGraph items", sizeof(struct ng_item), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); uma_zone_set_max(ng_qzone, maxalloc); ng_qdzone = uma_zcreate("NetGraph data items", sizeof(struct ng_item), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); uma_zone_set_max(ng_qdzone, maxdata); /* Autoconfigure number of threads. */ if (numthreads <= 0) numthreads = mp_ncpus; /* Create threads. */ p = NULL; /* start with no process */ for (i = 0; i < numthreads; i++) { if (kproc_kthread_add(ngthread, NULL, &p, &td, RFHIGHPID, 0, "ng_queue", "ng_queue%d", i)) { numthreads = i; break; } } break; case MOD_UNLOAD: /* You can't unload it because an interface may be using it. */ error = EBUSY; break; default: error = EOPNOTSUPP; break; } return (error); } static moduledata_t netgraph_mod = { "netgraph", ngb_mod_event, (NULL) }; DECLARE_MODULE(netgraph, netgraph_mod, SI_SUB_NETGRAPH, SI_ORDER_FIRST); SYSCTL_NODE(_net, OID_AUTO, graph, CTLFLAG_RW, 0, "netgraph Family"); SYSCTL_INT(_net_graph, OID_AUTO, abi_version, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_ABI_VERSION,""); SYSCTL_INT(_net_graph, OID_AUTO, msg_version, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_VERSION, ""); #ifdef NETGRAPH_DEBUG void dumphook (hook_p hook, char *file, int line) { printf("hook: name %s, %d refs, Last touched:\n", _NG_HOOK_NAME(hook), hook->hk_refs); printf(" Last active @ %s, line %d\n", hook->lastfile, hook->lastline); if (line) { printf(" problem discovered at file %s, line %d\n", file, line); #ifdef KDB kdb_backtrace(); #endif } } void dumpnode(node_p node, char *file, int line) { printf("node: ID [%x]: type '%s', %d hooks, flags 0x%x, %d refs, %s:\n", _NG_NODE_ID(node), node->nd_type->name, node->nd_numhooks, node->nd_flags, node->nd_refs, node->nd_name); printf(" Last active @ %s, line %d\n", node->lastfile, node->lastline); if (line) { printf(" problem discovered at file %s, line %d\n", file, line); #ifdef KDB kdb_backtrace(); #endif } } void dumpitem(item_p item, char *file, int line) { printf(" ACTIVE item, last used at %s, line %d", item->lastfile, item->lastline); switch(item->el_flags & NGQF_TYPE) { case NGQF_DATA: printf(" - [data]\n"); break; case NGQF_MESG: printf(" - retaddr[%d]:\n", _NGI_RETADDR(item)); break; case NGQF_FN: printf(" - fn@%p (%p, %p, %p, %d (%x))\n", _NGI_FN(item), _NGI_NODE(item), _NGI_HOOK(item), item->body.fn.fn_arg1, item->body.fn.fn_arg2, item->body.fn.fn_arg2); break; case NGQF_FN2: printf(" - fn2@%p (%p, %p, %p, %d (%x))\n", _NGI_FN2(item), _NGI_NODE(item), _NGI_HOOK(item), item->body.fn.fn_arg1, item->body.fn.fn_arg2, item->body.fn.fn_arg2); break; } if (line) { printf(" problem discovered at file %s, line %d\n", file, line); if (_NGI_NODE(item)) { printf("node %p ([%x])\n", _NGI_NODE(item), ng_node2ID(_NGI_NODE(item))); } } } static void ng_dumpitems(void) { item_p item; int i = 1; TAILQ_FOREACH(item, &ng_itemlist, all) { printf("[%d] ", i++); dumpitem(item, NULL, 0); } } static void ng_dumpnodes(void) { node_p node; int i = 1; mtx_lock(&ng_nodelist_mtx); SLIST_FOREACH(node, &ng_allnodes, nd_all) { printf("[%d] ", i++); dumpnode(node, NULL, 0); } mtx_unlock(&ng_nodelist_mtx); } static void ng_dumphooks(void) { hook_p hook; int i = 1; mtx_lock(&ng_nodelist_mtx); SLIST_FOREACH(hook, &ng_allhooks, hk_all) { printf("[%d] ", i++); dumphook(hook, NULL, 0); } mtx_unlock(&ng_nodelist_mtx); } static int sysctl_debug_ng_dump_items(SYSCTL_HANDLER_ARGS) { int error; int val; int i; val = allocated; i = 1; error = sysctl_handle_int(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == 42) { ng_dumpitems(); ng_dumpnodes(); ng_dumphooks(); } return (0); } SYSCTL_PROC(_debug, OID_AUTO, ng_dump_items, CTLTYPE_INT | CTLFLAG_RW, 0, sizeof(int), sysctl_debug_ng_dump_items, "I", "Number of allocated items"); #endif /* NETGRAPH_DEBUG */ /*********************************************************************** * Worklist routines **********************************************************************/ /* * Pick a node off the list of nodes with work, * try get an item to process off it. Remove the node from the list. */ static void ngthread(void *arg) { for (;;) { node_p node; /* Get node from the worklist. */ NG_WORKLIST_LOCK(); while ((node = STAILQ_FIRST(&ng_worklist)) == NULL) NG_WORKLIST_SLEEP(); STAILQ_REMOVE_HEAD(&ng_worklist, nd_input_queue.q_work); NG_WORKLIST_UNLOCK(); CURVNET_SET(node->nd_vnet); CTR3(KTR_NET, "%20s: node [%x] (%p) taken off worklist", __func__, node->nd_ID, node); /* * We have the node. We also take over the reference * that the list had on it. * Now process as much as you can, until it won't * let you have another item off the queue. * All this time, keep the reference * that lets us be sure that the node still exists. * Let the reference go at the last minute. */ for (;;) { item_p item; int rw; NG_QUEUE_LOCK(&node->nd_input_queue); item = ng_dequeue(node, &rw); if (item == NULL) { node->nd_input_queue.q_flags2 &= ~NGQ2_WORKQ; NG_QUEUE_UNLOCK(&node->nd_input_queue); break; /* go look for another node */ } else { NG_QUEUE_UNLOCK(&node->nd_input_queue); NGI_GET_NODE(item, node); /* zaps stored node */ ng_apply_item(node, item, rw); NG_NODE_UNREF(node); } } NG_NODE_UNREF(node); CURVNET_RESTORE(); } } /* * XXX * It's posible that a debugging NG_NODE_REF may need * to be outside the mutex zone */ static void ng_worklist_add(node_p node) { mtx_assert(&node->nd_input_queue.q_mtx, MA_OWNED); if ((node->nd_input_queue.q_flags2 & NGQ2_WORKQ) == 0) { /* * If we are not already on the work queue, * then put us on. */ node->nd_input_queue.q_flags2 |= NGQ2_WORKQ; NG_NODE_REF(node); /* XXX safe in mutex? */ NG_WORKLIST_LOCK(); STAILQ_INSERT_TAIL(&ng_worklist, node, nd_input_queue.q_work); NG_WORKLIST_UNLOCK(); CTR3(KTR_NET, "%20s: node [%x] (%p) put on worklist", __func__, node->nd_ID, node); NG_WORKLIST_WAKEUP(); } else { CTR3(KTR_NET, "%20s: node [%x] (%p) already on worklist", __func__, node->nd_ID, node); } } /*********************************************************************** * Externally useable functions to set up a queue item ready for sending ***********************************************************************/ #ifdef NETGRAPH_DEBUG #define ITEM_DEBUG_CHECKS \ do { \ if (NGI_NODE(item) ) { \ printf("item already has node"); \ kdb_enter(KDB_WHY_NETGRAPH, "has node"); \ NGI_CLR_NODE(item); \ } \ if (NGI_HOOK(item) ) { \ printf("item already has hook"); \ kdb_enter(KDB_WHY_NETGRAPH, "has hook"); \ NGI_CLR_HOOK(item); \ } \ } while (0) #else #define ITEM_DEBUG_CHECKS #endif /* * Put mbuf into the item. * Hook and node references will be removed when the item is dequeued. * (or equivalent) * (XXX) Unsafe because no reference held by peer on remote node. * remote node might go away in this timescale. * We know the hooks can't go away because that would require getting * a writer item on both nodes and we must have at least a reader * here to be able to do this. * Note that the hook loaded is the REMOTE hook. * * This is possibly in the critical path for new data. */ item_p ng_package_data(struct mbuf *m, int flags) { item_p item; if ((item = ng_alloc_item(NGQF_DATA, flags)) == NULL) { NG_FREE_M(m); return (NULL); } ITEM_DEBUG_CHECKS; item->el_flags |= NGQF_READER; NGI_M(item) = m; return (item); } /* * Allocate a queue item and put items into it.. * Evaluate the address as this will be needed to queue it and * to work out what some of the fields should be. * Hook and node references will be removed when the item is dequeued. * (or equivalent) */ item_p ng_package_msg(struct ng_mesg *msg, int flags) { item_p item; if ((item = ng_alloc_item(NGQF_MESG, flags)) == NULL) { NG_FREE_MSG(msg); return (NULL); } ITEM_DEBUG_CHECKS; /* Messages items count as writers unless explicitly exempted. */ if (msg->header.cmd & NGM_READONLY) item->el_flags |= NGQF_READER; else item->el_flags |= NGQF_WRITER; /* * Set the current lasthook into the queue item */ NGI_MSG(item) = msg; NGI_RETADDR(item) = 0; return (item); } #define SET_RETADDR(item, here, retaddr) \ do { /* Data or fn items don't have retaddrs */ \ if ((item->el_flags & NGQF_TYPE) == NGQF_MESG) { \ if (retaddr) { \ NGI_RETADDR(item) = retaddr; \ } else { \ /* \ * The old return address should be ok. \ * If there isn't one, use the address \ * here. \ */ \ if (NGI_RETADDR(item) == 0) { \ NGI_RETADDR(item) \ = ng_node2ID(here); \ } \ } \ } \ } while (0) int ng_address_hook(node_p here, item_p item, hook_p hook, ng_ID_t retaddr) { hook_p peer; node_p peernode; ITEM_DEBUG_CHECKS; /* * Quick sanity check.. * Since a hook holds a reference on its node, once we know * that the peer is still connected (even if invalid,) we know * that the peer node is present, though maybe invalid. */ TOPOLOGY_RLOCK(); if ((hook == NULL) || NG_HOOK_NOT_VALID(hook) || NG_HOOK_NOT_VALID(peer = NG_HOOK_PEER(hook)) || NG_NODE_NOT_VALID(peernode = NG_PEER_NODE(hook))) { NG_FREE_ITEM(item); TRAP_ERROR(); TOPOLOGY_RUNLOCK(); return (ENETDOWN); } /* * Transfer our interest to the other (peer) end. */ NG_HOOK_REF(peer); NG_NODE_REF(peernode); NGI_SET_HOOK(item, peer); NGI_SET_NODE(item, peernode); SET_RETADDR(item, here, retaddr); TOPOLOGY_RUNLOCK(); return (0); } int ng_address_path(node_p here, item_p item, const char *address, ng_ID_t retaddr) { node_p dest = NULL; hook_p hook = NULL; int error; ITEM_DEBUG_CHECKS; /* * Note that ng_path2noderef increments the reference count * on the node for us if it finds one. So we don't have to. */ error = ng_path2noderef(here, address, &dest, &hook); if (error) { NG_FREE_ITEM(item); return (error); } NGI_SET_NODE(item, dest); if (hook) NGI_SET_HOOK(item, hook); SET_RETADDR(item, here, retaddr); return (0); } int ng_address_ID(node_p here, item_p item, ng_ID_t ID, ng_ID_t retaddr) { node_p dest; ITEM_DEBUG_CHECKS; /* * Find the target node. */ dest = ng_ID2noderef(ID); /* GETS REFERENCE! */ if (dest == NULL) { NG_FREE_ITEM(item); TRAP_ERROR(); return(EINVAL); } /* Fill out the contents */ NGI_SET_NODE(item, dest); NGI_CLR_HOOK(item); SET_RETADDR(item, here, retaddr); return (0); } /* * special case to send a message to self (e.g. destroy node) * Possibly indicate an arrival hook too. * Useful for removing that hook :-) */ item_p ng_package_msg_self(node_p here, hook_p hook, struct ng_mesg *msg) { item_p item; /* * Find the target node. * If there is a HOOK argument, then use that in preference * to the address. */ if ((item = ng_alloc_item(NGQF_MESG, NG_NOFLAGS)) == NULL) { NG_FREE_MSG(msg); return (NULL); } /* Fill out the contents */ item->el_flags |= NGQF_WRITER; NG_NODE_REF(here); NGI_SET_NODE(item, here); if (hook) { NG_HOOK_REF(hook); NGI_SET_HOOK(item, hook); } NGI_MSG(item) = msg; NGI_RETADDR(item) = ng_node2ID(here); return (item); } /* * Send ng_item_fn function call to the specified node. */ int ng_send_fn(node_p node, hook_p hook, ng_item_fn *fn, void * arg1, int arg2) { return ng_send_fn1(node, hook, fn, arg1, arg2, NG_NOFLAGS); } int ng_send_fn1(node_p node, hook_p hook, ng_item_fn *fn, void * arg1, int arg2, int flags) { item_p item; if ((item = ng_alloc_item(NGQF_FN, flags)) == NULL) { return (ENOMEM); } item->el_flags |= NGQF_WRITER; NG_NODE_REF(node); /* and one for the item */ NGI_SET_NODE(item, node); if (hook) { NG_HOOK_REF(hook); NGI_SET_HOOK(item, hook); } NGI_FN(item) = fn; NGI_ARG1(item) = arg1; NGI_ARG2(item) = arg2; return(ng_snd_item(item, flags)); } /* * Send ng_item_fn2 function call to the specified node. * * If an optional pitem parameter is supplied, its apply * callback will be copied to the new item. If also NG_REUSE_ITEM * flag is set, no new item will be allocated, but pitem will * be used. */ int ng_send_fn2(node_p node, hook_p hook, item_p pitem, ng_item_fn2 *fn, void *arg1, int arg2, int flags) { item_p item; KASSERT((pitem != NULL || (flags & NG_REUSE_ITEM) == 0), ("%s: NG_REUSE_ITEM but no pitem", __func__)); /* * Allocate a new item if no supplied or * if we can't use supplied one. */ if (pitem == NULL || (flags & NG_REUSE_ITEM) == 0) { if ((item = ng_alloc_item(NGQF_FN2, flags)) == NULL) return (ENOMEM); if (pitem != NULL) item->apply = pitem->apply; } else { if ((item = ng_realloc_item(pitem, NGQF_FN2, flags)) == NULL) return (ENOMEM); } item->el_flags = (item->el_flags & ~NGQF_RW) | NGQF_WRITER; NG_NODE_REF(node); /* and one for the item */ NGI_SET_NODE(item, node); if (hook) { NG_HOOK_REF(hook); NGI_SET_HOOK(item, hook); } NGI_FN2(item) = fn; NGI_ARG1(item) = arg1; NGI_ARG2(item) = arg2; return(ng_snd_item(item, flags)); } /* * Official timeout routines for Netgraph nodes. */ static void ng_callout_trampoline(void *arg) { item_p item = arg; CURVNET_SET(NGI_NODE(item)->nd_vnet); ng_snd_item(item, 0); CURVNET_RESTORE(); } int ng_callout(struct callout *c, node_p node, hook_p hook, int ticks, ng_item_fn *fn, void * arg1, int arg2) { item_p item, oitem; if ((item = ng_alloc_item(NGQF_FN, NG_NOFLAGS)) == NULL) return (ENOMEM); item->el_flags |= NGQF_WRITER; NG_NODE_REF(node); /* and one for the item */ NGI_SET_NODE(item, node); if (hook) { NG_HOOK_REF(hook); NGI_SET_HOOK(item, hook); } NGI_FN(item) = fn; NGI_ARG1(item) = arg1; NGI_ARG2(item) = arg2; oitem = c->c_arg; if (callout_reset(c, ticks, &ng_callout_trampoline, item) == 1 && oitem != NULL) NG_FREE_ITEM(oitem); return (0); } /* A special modified version of untimeout() */ int ng_uncallout(struct callout *c, node_p node) { item_p item; int rval; KASSERT(c != NULL, ("ng_uncallout: NULL callout")); KASSERT(node != NULL, ("ng_uncallout: NULL node")); rval = callout_stop(c); item = c->c_arg; /* Do an extra check */ if ((rval > 0) && (c->c_func == &ng_callout_trampoline) && (item != NULL) && (NGI_NODE(item) == node)) { /* * We successfully removed it from the queue before it ran * So now we need to unreference everything that was * given extra references. (NG_FREE_ITEM does this). */ NG_FREE_ITEM(item); } c->c_arg = NULL; /* * Callers only want to know if the callout was cancelled and * not draining or stopped. */ return (rval > 0); } /* * Set the address, if none given, give the node here. */ void ng_replace_retaddr(node_p here, item_p item, ng_ID_t retaddr) { if (retaddr) { NGI_RETADDR(item) = retaddr; } else { /* * The old return address should be ok. * If there isn't one, use the address here. */ NGI_RETADDR(item) = ng_node2ID(here); } } Index: head/sys/netgraph/ng_eiface.c =================================================================== --- head/sys/netgraph/ng_eiface.c (revision 336675) +++ head/sys/netgraph/ng_eiface.c (revision 336676) @@ -1,686 +1,686 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * * Copyright (c) 1999-2001, Vitaly V Belekhov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static const struct ng_cmdlist ng_eiface_cmdlist[] = { { NGM_EIFACE_COOKIE, NGM_EIFACE_GET_IFNAME, "getifname", NULL, &ng_parse_string_type }, { NGM_EIFACE_COOKIE, NGM_EIFACE_SET, "set", &ng_parse_enaddr_type, NULL }, { 0 } }; /* Node private data */ struct ng_eiface_private { struct ifnet *ifp; /* per-interface network data */ struct ifmedia media; /* (fake) media information */ int link_status; /* fake */ int unit; /* Interface unit number */ node_p node; /* Our netgraph node */ hook_p ether; /* Hook for ethernet stream */ }; typedef struct ng_eiface_private *priv_p; /* Interface methods */ static void ng_eiface_init(void *xsc); static void ng_eiface_start(struct ifnet *ifp); static int ng_eiface_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data); #ifdef DEBUG static void ng_eiface_print_ioctl(struct ifnet *ifp, int cmd, caddr_t data); #endif /* Netgraph methods */ static int ng_eiface_mod_event(module_t, int, void *); static ng_constructor_t ng_eiface_constructor; static ng_rcvmsg_t ng_eiface_rcvmsg; static ng_shutdown_t ng_eiface_rmnode; static ng_newhook_t ng_eiface_newhook; static ng_rcvdata_t ng_eiface_rcvdata; static ng_disconnect_t ng_eiface_disconnect; /* Node type descriptor */ static struct ng_type typestruct = { .version = NG_ABI_VERSION, .name = NG_EIFACE_NODE_TYPE, .mod_event = ng_eiface_mod_event, .constructor = ng_eiface_constructor, .rcvmsg = ng_eiface_rcvmsg, .shutdown = ng_eiface_rmnode, .newhook = ng_eiface_newhook, .rcvdata = ng_eiface_rcvdata, .disconnect = ng_eiface_disconnect, .cmdlist = ng_eiface_cmdlist }; NETGRAPH_INIT(eiface, &typestruct); -static VNET_DEFINE(struct unrhdr *, ng_eiface_unit); +VNET_DEFINE_STATIC(struct unrhdr *, ng_eiface_unit); #define V_ng_eiface_unit VNET(ng_eiface_unit) /************************************************************************ INTERFACE STUFF ************************************************************************/ /* * Process an ioctl for the virtual interface */ static int ng_eiface_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { const priv_p priv = (priv_p)ifp->if_softc; struct ifreq *const ifr = (struct ifreq *)data; int error = 0; #ifdef DEBUG ng_eiface_print_ioctl(ifp, command, data); #endif switch (command) { /* These two are mostly handled at a higher layer */ case SIOCSIFADDR: error = ether_ioctl(ifp, command, data); break; case SIOCGIFADDR: break; /* Set flags */ case SIOCSIFFLAGS: /* * If the interface is marked up and stopped, then start it. * If it is marked down and running, then stop it. */ if (ifp->if_flags & IFF_UP) { if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE); ifp->if_drv_flags |= IFF_DRV_RUNNING; } } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); } break; /* Set the interface MTU */ case SIOCSIFMTU: if (ifr->ifr_mtu > NG_EIFACE_MTU_MAX || ifr->ifr_mtu < NG_EIFACE_MTU_MIN) error = EINVAL; else ifp->if_mtu = ifr->ifr_mtu; break; /* (Fake) media type manipulation */ case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &priv->media, command); break; /* Stuff that's not supported */ case SIOCADDMULTI: case SIOCDELMULTI: error = 0; break; case SIOCSIFPHYS: error = EOPNOTSUPP; break; default: error = EINVAL; break; } return (error); } static void ng_eiface_init(void *xsc) { priv_p sc = xsc; struct ifnet *ifp = sc->ifp; ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } /* * We simply relay the packet to the "ether" hook, if it is connected. * We have been through the netgraph locking and are guaranteed to * be the only code running in this node at this time. */ static void ng_eiface_start2(node_p node, hook_p hook, void *arg1, int arg2) { struct ifnet *ifp = arg1; const priv_p priv = (priv_p)ifp->if_softc; int error = 0; struct mbuf *m; /* Check interface flags */ if (!((ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING))) return; for (;;) { /* * Grab a packet to transmit. */ IF_DEQUEUE(&ifp->if_snd, m); /* If there's nothing to send, break. */ if (m == NULL) break; /* Peel the mbuf off any stale tags */ m_tag_delete_chain(m, NULL); /* * Berkeley packet filter. * Pass packet to bpf if there is a listener. * XXX is this safe? locking? */ BPF_MTAP(ifp, m); if (ifp->if_flags & IFF_MONITOR) { if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); m_freem(m); continue; } /* * Send packet; if hook is not connected, mbuf will get * freed. */ NG_OUTBOUND_THREAD_REF(); NG_SEND_DATA_ONLY(error, priv->ether, m); NG_OUTBOUND_THREAD_UNREF(); /* Update stats */ if (error == 0) if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); else if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); } ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; return; } /* * This routine is called to deliver a packet out the interface. * We simply queue the netgraph version to be called when netgraph locking * allows it to happen. * Until we know what the rest of the networking code is doing for * locking, we don't know how we will interact with it. * Take comfort from the fact that the ifnet struct is part of our * private info and can't go away while we are queued. * [Though we don't know it is still there now....] * it is possible we don't gain anything from this because * we would like to get the mbuf and queue it as data * somehow, but we can't and if we did would we solve anything? */ static void ng_eiface_start(struct ifnet *ifp) { const priv_p priv = (priv_p)ifp->if_softc; /* Don't do anything if output is active */ if (ifp->if_drv_flags & IFF_DRV_OACTIVE) return; ifp->if_drv_flags |= IFF_DRV_OACTIVE; if (ng_send_fn(priv->node, NULL, &ng_eiface_start2, ifp, 0) != 0) ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } #ifdef DEBUG /* * Display an ioctl to the virtual interface */ static void ng_eiface_print_ioctl(struct ifnet *ifp, int command, caddr_t data) { char *str; switch (command & IOC_DIRMASK) { case IOC_VOID: str = "IO"; break; case IOC_OUT: str = "IOR"; break; case IOC_IN: str = "IOW"; break; case IOC_INOUT: str = "IORW"; break; default: str = "IO??"; } log(LOG_DEBUG, "%s: %s('%c', %d, char[%d])\n", ifp->if_xname, str, IOCGROUP(command), command & 0xff, IOCPARM_LEN(command)); } #endif /* DEBUG */ /* * ifmedia stuff */ static int ng_eiface_mediachange(struct ifnet *ifp) { const priv_p priv = (priv_p)ifp->if_softc; struct ifmedia *ifm = &priv->media; if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) ifp->if_baudrate = ifmedia_baudrate(IFM_ETHER | IFM_1000_T); else ifp->if_baudrate = ifmedia_baudrate(ifm->ifm_media); return (0); } static void ng_eiface_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr) { const priv_p priv = (priv_p)ifp->if_softc; struct ifmedia *ifm = &priv->media; if (ifm->ifm_cur->ifm_media == (IFM_ETHER | IFM_AUTO) && (priv->link_status & IFM_ACTIVE)) ifmr->ifm_active = IFM_ETHER | IFM_1000_T | IFM_FDX; else ifmr->ifm_active = ifm->ifm_cur->ifm_media; ifmr->ifm_status = priv->link_status; return; } /************************************************************************ NETGRAPH NODE STUFF ************************************************************************/ /* * Constructor for a node */ static int ng_eiface_constructor(node_p node) { struct ifnet *ifp; priv_p priv; u_char eaddr[6] = {0,0,0,0,0,0}; /* Allocate node and interface private structures */ priv = malloc(sizeof(*priv), M_NETGRAPH, M_WAITOK | M_ZERO); ifp = priv->ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { free(priv, M_NETGRAPH); return (ENOSPC); } /* Link them together */ ifp->if_softc = priv; /* Get an interface unit number */ priv->unit = alloc_unr(V_ng_eiface_unit); /* Link together node and private info */ NG_NODE_SET_PRIVATE(node, priv); priv->node = node; /* Initialize interface structure */ if_initname(ifp, NG_EIFACE_EIFACE_NAME, priv->unit); ifp->if_init = ng_eiface_init; ifp->if_output = ether_output; ifp->if_start = ng_eiface_start; ifp->if_ioctl = ng_eiface_ioctl; ifp->if_snd.ifq_maxlen = ifqmaxlen; ifp->if_flags = (IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST); ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_JUMBO_MTU; ifp->if_capenable = IFCAP_VLAN_MTU | IFCAP_JUMBO_MTU; ifmedia_init(&priv->media, 0, ng_eiface_mediachange, ng_eiface_mediastatus); ifmedia_add(&priv->media, IFM_ETHER | IFM_10_T, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_10_T | IFM_FDX, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_100_TX, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_100_TX | IFM_FDX, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_1000_T, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_10G_T | IFM_FDX, 0, NULL); ifmedia_add(&priv->media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO); priv->link_status = IFM_AVALID; /* Give this node the same name as the interface (if possible) */ if (ng_name_node(node, ifp->if_xname) != 0) log(LOG_WARNING, "%s: can't acquire netgraph name\n", ifp->if_xname); /* Attach the interface */ ether_ifattach(ifp, eaddr); ifp->if_baudrate = ifmedia_baudrate(IFM_ETHER | IFM_1000_T); /* Done */ return (0); } /* * Give our ok for a hook to be added */ static int ng_eiface_newhook(node_p node, hook_p hook, const char *name) { priv_p priv = NG_NODE_PRIVATE(node); struct ifnet *ifp = priv->ifp; if (strcmp(name, NG_EIFACE_HOOK_ETHER)) return (EPFNOSUPPORT); if (priv->ether != NULL) return (EISCONN); priv->ether = hook; NG_HOOK_SET_PRIVATE(hook, &priv->ether); NG_HOOK_SET_TO_INBOUND(hook); priv->link_status |= IFM_ACTIVE; CURVNET_SET_QUIET(ifp->if_vnet); if_link_state_change(ifp, LINK_STATE_UP); CURVNET_RESTORE(); return (0); } /* * Receive a control message */ static int ng_eiface_rcvmsg(node_p node, item_p item, hook_p lasthook) { const priv_p priv = NG_NODE_PRIVATE(node); struct ifnet *const ifp = priv->ifp; struct ng_mesg *resp = NULL; int error = 0; struct ng_mesg *msg; NGI_GET_MSG(item, msg); switch (msg->header.typecookie) { case NGM_EIFACE_COOKIE: switch (msg->header.cmd) { case NGM_EIFACE_SET: { if (msg->header.arglen != ETHER_ADDR_LEN) { error = EINVAL; break; } error = if_setlladdr(priv->ifp, (u_char *)msg->data, ETHER_ADDR_LEN); break; } case NGM_EIFACE_GET_IFNAME: NG_MKRESPONSE(resp, msg, IFNAMSIZ, M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } strlcpy(resp->data, ifp->if_xname, IFNAMSIZ); break; case NGM_EIFACE_GET_IFADDRS: { struct ifaddr *ifa; caddr_t ptr; int buflen; /* Determine size of response and allocate it */ buflen = 0; if_addr_rlock(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) buflen += SA_SIZE(ifa->ifa_addr); NG_MKRESPONSE(resp, msg, buflen, M_NOWAIT); if (resp == NULL) { if_addr_runlock(ifp); error = ENOMEM; break; } /* Add addresses */ ptr = resp->data; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { const int len = SA_SIZE(ifa->ifa_addr); if (buflen < len) { log(LOG_ERR, "%s: len changed?\n", ifp->if_xname); break; } bcopy(ifa->ifa_addr, ptr, len); ptr += len; buflen -= len; } if_addr_runlock(ifp); break; } default: error = EINVAL; break; } /* end of inner switch() */ break; case NGM_FLOW_COOKIE: CURVNET_SET_QUIET(ifp->if_vnet); switch (msg->header.cmd) { case NGM_LINK_IS_UP: priv->link_status |= IFM_ACTIVE; if_link_state_change(ifp, LINK_STATE_UP); break; case NGM_LINK_IS_DOWN: priv->link_status &= ~IFM_ACTIVE; if_link_state_change(ifp, LINK_STATE_DOWN); break; default: break; } CURVNET_RESTORE(); break; default: error = EINVAL; break; } NG_RESPOND_MSG(error, node, item, resp); NG_FREE_MSG(msg); return (error); } /* * Receive data from a hook. Pass the packet to the ether_input routine. */ static int ng_eiface_rcvdata(hook_p hook, item_p item) { const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook)); struct ifnet *const ifp = priv->ifp; struct mbuf *m; NGI_GET_M(item, m); NG_FREE_ITEM(item); if (!((ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING))) { NG_FREE_M(m); return (ENETDOWN); } if (m->m_len < ETHER_HDR_LEN) { m = m_pullup(m, ETHER_HDR_LEN); if (m == NULL) return (EINVAL); } /* Note receiving interface */ m->m_pkthdr.rcvif = ifp; /* Update interface stats */ if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); (*ifp->if_input)(ifp, m); /* Done */ return (0); } /* * Shutdown processing. */ static int ng_eiface_rmnode(node_p node) { const priv_p priv = NG_NODE_PRIVATE(node); struct ifnet *const ifp = priv->ifp; /* * the ifnet may be in a different vnet than the netgraph node, * hence we have to change the current vnet context here. */ CURVNET_SET_QUIET(ifp->if_vnet); ifmedia_removeall(&priv->media); ether_ifdetach(ifp); if_free(ifp); CURVNET_RESTORE(); free_unr(V_ng_eiface_unit, priv->unit); free(priv, M_NETGRAPH); NG_NODE_SET_PRIVATE(node, NULL); NG_NODE_UNREF(node); return (0); } /* * Hook disconnection */ static int ng_eiface_disconnect(hook_p hook) { const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook)); priv->ether = NULL; priv->link_status &= ~IFM_ACTIVE; CURVNET_SET_QUIET(priv->ifp->if_vnet); if_link_state_change(priv->ifp, LINK_STATE_DOWN); CURVNET_RESTORE(); return (0); } /* * Handle loading and unloading for this node type. */ static int ng_eiface_mod_event(module_t mod, int event, void *data) { int error = 0; switch (event) { case MOD_LOAD: case MOD_UNLOAD: break; default: error = EOPNOTSUPP; break; } return (error); } static void vnet_ng_eiface_init(const void *unused) { V_ng_eiface_unit = new_unrhdr(0, 0xffff, NULL); } VNET_SYSINIT(vnet_ng_eiface_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_ng_eiface_init, NULL); static void vnet_ng_eiface_uninit(const void *unused) { delete_unrhdr(V_ng_eiface_unit); } VNET_SYSUNINIT(vnet_ng_eiface_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_ng_eiface_uninit, NULL); Index: head/sys/netgraph/ng_iface.c =================================================================== --- head/sys/netgraph/ng_iface.c (revision 336675) +++ head/sys/netgraph/ng_iface.c (revision 336676) @@ -1,820 +1,820 @@ /* * ng_iface.c */ /*- * Copyright (c) 1996-1999 Whistle Communications, Inc. * All rights reserved. * * Subject to the following obligations and disclaimer of warranty, use and * redistribution of this software, in source or object code forms, with or * without modifications are expressly permitted by Whistle Communications; * provided, however, that: * 1. Any and all reproductions of the source or object code must include the * copyright notice above and the following disclaimer of warranties; and * 2. No rights are granted, in any manner or form, to use Whistle * Communications, Inc. trademarks, including the mark "WHISTLE * COMMUNICATIONS" on advertising, endorsements, or otherwise except as * such appears in the above copyright notice or in the software. * * THIS SOFTWARE IS BEING PROVIDED BY WHISTLE COMMUNICATIONS "AS IS", AND * TO THE MAXIMUM EXTENT PERMITTED BY LAW, WHISTLE COMMUNICATIONS MAKES NO * REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED, REGARDING THIS SOFTWARE, * INCLUDING WITHOUT LIMITATION, ANY AND ALL IMPLIED WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT. * WHISTLE COMMUNICATIONS DOES NOT WARRANT, GUARANTEE, OR MAKE ANY * REPRESENTATIONS REGARDING THE USE OF, OR THE RESULTS OF THE USE OF THIS * SOFTWARE IN TERMS OF ITS CORRECTNESS, ACCURACY, RELIABILITY OR OTHERWISE. * IN NO EVENT SHALL WHISTLE COMMUNICATIONS BE LIABLE FOR ANY DAMAGES * RESULTING FROM OR ARISING OUT OF ANY USE OF THIS SOFTWARE, INCLUDING * WITHOUT LIMITATION, ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, * PUNITIVE, OR CONSEQUENTIAL DAMAGES, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES, LOSS OF USE, DATA OR PROFITS, HOWEVER CAUSED AND UNDER ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF WHISTLE COMMUNICATIONS IS ADVISED OF THE POSSIBILITY * OF SUCH DAMAGE. * * Author: Archie Cobbs * * $FreeBSD$ * $Whistle: ng_iface.c,v 1.33 1999/11/01 09:24:51 julian Exp $ */ /* * This node is also a system networking interface. It has * a hook for each protocol (IP, AppleTalk, etc). Packets * are simply relayed between the interface and the hooks. * * Interfaces are named ng0, ng1, etc. New nodes take the * first available interface name. * * This node also includes Berkeley packet filter support. */ #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef NG_SEPARATE_MALLOC static MALLOC_DEFINE(M_NETGRAPH_IFACE, "netgraph_iface", "netgraph iface node"); #else #define M_NETGRAPH_IFACE M_NETGRAPH #endif /* This struct describes one address family */ struct iffam { sa_family_t family; /* Address family */ const char *hookname; /* Name for hook */ }; typedef const struct iffam *iffam_p; /* List of address families supported by our interface */ const static struct iffam gFamilies[] = { { AF_INET, NG_IFACE_HOOK_INET }, { AF_INET6, NG_IFACE_HOOK_INET6 }, { AF_ATM, NG_IFACE_HOOK_ATM }, { AF_NATM, NG_IFACE_HOOK_NATM }, }; #define NUM_FAMILIES nitems(gFamilies) /* Node private data */ struct ng_iface_private { struct ifnet *ifp; /* Our interface */ int unit; /* Interface unit number */ node_p node; /* Our netgraph node */ hook_p hooks[NUM_FAMILIES]; /* Hook for each address family */ struct rmlock lock; /* Protect private data changes */ }; typedef struct ng_iface_private *priv_p; #define PRIV_RLOCK(priv, t) rm_rlock(&priv->lock, t) #define PRIV_RUNLOCK(priv, t) rm_runlock(&priv->lock, t) #define PRIV_WLOCK(priv) rm_wlock(&priv->lock) #define PRIV_WUNLOCK(priv) rm_wunlock(&priv->lock) /* Interface methods */ static void ng_iface_start(struct ifnet *ifp); static int ng_iface_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data); static int ng_iface_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst, struct route *ro); static void ng_iface_bpftap(struct ifnet *ifp, struct mbuf *m, sa_family_t family); static int ng_iface_send(struct ifnet *ifp, struct mbuf *m, sa_family_t sa); #ifdef DEBUG static void ng_iface_print_ioctl(struct ifnet *ifp, int cmd, caddr_t data); #endif /* Netgraph methods */ static int ng_iface_mod_event(module_t, int, void *); static ng_constructor_t ng_iface_constructor; static ng_rcvmsg_t ng_iface_rcvmsg; static ng_shutdown_t ng_iface_shutdown; static ng_newhook_t ng_iface_newhook; static ng_rcvdata_t ng_iface_rcvdata; static ng_disconnect_t ng_iface_disconnect; /* Helper stuff */ static iffam_p get_iffam_from_af(sa_family_t family); static iffam_p get_iffam_from_hook(priv_p priv, hook_p hook); static iffam_p get_iffam_from_name(const char *name); static hook_p *get_hook_from_iffam(priv_p priv, iffam_p iffam); /* List of commands and how to convert arguments to/from ASCII */ static const struct ng_cmdlist ng_iface_cmds[] = { { NGM_IFACE_COOKIE, NGM_IFACE_GET_IFNAME, "getifname", NULL, &ng_parse_string_type }, { NGM_IFACE_COOKIE, NGM_IFACE_POINT2POINT, "point2point", NULL, NULL }, { NGM_IFACE_COOKIE, NGM_IFACE_BROADCAST, "broadcast", NULL, NULL }, { NGM_IFACE_COOKIE, NGM_IFACE_GET_IFINDEX, "getifindex", NULL, &ng_parse_uint32_type }, { 0 } }; /* Node type descriptor */ static struct ng_type typestruct = { .version = NG_ABI_VERSION, .name = NG_IFACE_NODE_TYPE, .mod_event = ng_iface_mod_event, .constructor = ng_iface_constructor, .rcvmsg = ng_iface_rcvmsg, .shutdown = ng_iface_shutdown, .newhook = ng_iface_newhook, .rcvdata = ng_iface_rcvdata, .disconnect = ng_iface_disconnect, .cmdlist = ng_iface_cmds, }; NETGRAPH_INIT(iface, &typestruct); -static VNET_DEFINE(struct unrhdr *, ng_iface_unit); +VNET_DEFINE_STATIC(struct unrhdr *, ng_iface_unit); #define V_ng_iface_unit VNET(ng_iface_unit) /************************************************************************ HELPER STUFF ************************************************************************/ /* * Get the family descriptor from the family ID */ static __inline iffam_p get_iffam_from_af(sa_family_t family) { iffam_p iffam; int k; for (k = 0; k < NUM_FAMILIES; k++) { iffam = &gFamilies[k]; if (iffam->family == family) return (iffam); } return (NULL); } /* * Get the family descriptor from the hook */ static __inline iffam_p get_iffam_from_hook(priv_p priv, hook_p hook) { int k; for (k = 0; k < NUM_FAMILIES; k++) if (priv->hooks[k] == hook) return (&gFamilies[k]); return (NULL); } /* * Get the hook from the iffam descriptor */ static __inline hook_p * get_hook_from_iffam(priv_p priv, iffam_p iffam) { return (&priv->hooks[iffam - gFamilies]); } /* * Get the iffam descriptor from the name */ static __inline iffam_p get_iffam_from_name(const char *name) { iffam_p iffam; int k; for (k = 0; k < NUM_FAMILIES; k++) { iffam = &gFamilies[k]; if (!strcmp(iffam->hookname, name)) return (iffam); } return (NULL); } /************************************************************************ INTERFACE STUFF ************************************************************************/ /* * Process an ioctl for the virtual interface */ static int ng_iface_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct ifreq *const ifr = (struct ifreq *) data; int error = 0; #ifdef DEBUG ng_iface_print_ioctl(ifp, command, data); #endif switch (command) { /* These two are mostly handled at a higher layer */ case SIOCSIFADDR: ifp->if_flags |= IFF_UP; ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE); break; case SIOCGIFADDR: break; /* Set flags */ case SIOCSIFFLAGS: /* * If the interface is marked up and stopped, then start it. * If it is marked down and running, then stop it. */ if (ifr->ifr_flags & IFF_UP) { if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE); ifp->if_drv_flags |= IFF_DRV_RUNNING; } } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); } break; /* Set the interface MTU */ case SIOCSIFMTU: if (ifr->ifr_mtu > NG_IFACE_MTU_MAX || ifr->ifr_mtu < NG_IFACE_MTU_MIN) error = EINVAL; else ifp->if_mtu = ifr->ifr_mtu; break; /* Stuff that's not supported */ case SIOCADDMULTI: case SIOCDELMULTI: error = 0; break; case SIOCSIFPHYS: error = EOPNOTSUPP; break; default: error = EINVAL; break; } return (error); } /* * This routine is called to deliver a packet out the interface. * We simply look at the address family and relay the packet to * the corresponding hook, if it exists and is connected. */ static int ng_iface_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { struct m_tag *mtag; uint32_t af; int error; /* Check interface flags */ if (!((ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING))) { m_freem(m); return (ENETDOWN); } /* Protect from deadly infinite recursion. */ mtag = NULL; while ((mtag = m_tag_locate(m, MTAG_NGIF, MTAG_NGIF_CALLED, mtag))) { if (*(struct ifnet **)(mtag + 1) == ifp) { log(LOG_NOTICE, "Loop detected on %s\n", ifp->if_xname); m_freem(m); return (EDEADLK); } } mtag = m_tag_alloc(MTAG_NGIF, MTAG_NGIF_CALLED, sizeof(struct ifnet *), M_NOWAIT); if (mtag == NULL) { m_freem(m); return (ENOMEM); } *(struct ifnet **)(mtag + 1) = ifp; m_tag_prepend(m, mtag); /* BPF writes need to be handled specially. */ if (dst->sa_family == AF_UNSPEC) bcopy(dst->sa_data, &af, sizeof(af)); else af = dst->sa_family; /* Berkeley packet filter */ ng_iface_bpftap(ifp, m, af); if (ALTQ_IS_ENABLED(&ifp->if_snd)) { M_PREPEND(m, sizeof(sa_family_t), M_NOWAIT); if (m == NULL) { if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); return (ENOBUFS); } *(sa_family_t *)m->m_data = af; error = (ifp->if_transmit)(ifp, m); } else error = ng_iface_send(ifp, m, af); return (error); } /* * Start method is used only when ALTQ is enabled. */ static void ng_iface_start(struct ifnet *ifp) { struct mbuf *m; sa_family_t sa; KASSERT(ALTQ_IS_ENABLED(&ifp->if_snd), ("%s without ALTQ", __func__)); for(;;) { IFQ_DRV_DEQUEUE(&ifp->if_snd, m); if (m == NULL) break; sa = *mtod(m, sa_family_t *); m_adj(m, sizeof(sa_family_t)); ng_iface_send(ifp, m, sa); } } /* * Flash a packet by the BPF (requires prepending 4 byte AF header) * Note the phoney mbuf; this is OK because BPF treats it read-only. */ static void ng_iface_bpftap(struct ifnet *ifp, struct mbuf *m, sa_family_t family) { KASSERT(family != AF_UNSPEC, ("%s: family=AF_UNSPEC", __func__)); if (bpf_peers_present(ifp->if_bpf)) { int32_t family4 = (int32_t)family; bpf_mtap2(ifp->if_bpf, &family4, sizeof(family4), m); } } /* * This routine does actual delivery of the packet into the * netgraph(4). It is called from ng_iface_start() and * ng_iface_output(). */ static int ng_iface_send(struct ifnet *ifp, struct mbuf *m, sa_family_t sa) { struct rm_priotracker priv_tracker; const priv_p priv = (priv_p) ifp->if_softc; const iffam_p iffam = get_iffam_from_af(sa); hook_p hook; int error; int len; /* Check address family to determine hook (if known) */ if (iffam == NULL) { m_freem(m); log(LOG_WARNING, "%s: can't handle af%d\n", ifp->if_xname, sa); return (EAFNOSUPPORT); } /* Copy length before the mbuf gets invalidated. */ len = m->m_pkthdr.len; PRIV_RLOCK(priv, &priv_tracker); hook = *get_hook_from_iffam(priv, iffam); if (hook == NULL) { NG_FREE_M(m); PRIV_RUNLOCK(priv, &priv_tracker); return ENETDOWN; } NG_HOOK_REF(hook); PRIV_RUNLOCK(priv, &priv_tracker); NG_OUTBOUND_THREAD_REF(); NG_SEND_DATA_ONLY(error, hook, m); NG_OUTBOUND_THREAD_UNREF(); NG_HOOK_UNREF(hook); /* Update stats. */ if (error == 0) { if_inc_counter(ifp, IFCOUNTER_OBYTES, len); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); } return (error); } #ifdef DEBUG /* * Display an ioctl to the virtual interface */ static void ng_iface_print_ioctl(struct ifnet *ifp, int command, caddr_t data) { char *str; switch (command & IOC_DIRMASK) { case IOC_VOID: str = "IO"; break; case IOC_OUT: str = "IOR"; break; case IOC_IN: str = "IOW"; break; case IOC_INOUT: str = "IORW"; break; default: str = "IO??"; } log(LOG_DEBUG, "%s: %s('%c', %d, char[%d])\n", ifp->if_xname, str, IOCGROUP(command), command & 0xff, IOCPARM_LEN(command)); } #endif /* DEBUG */ /************************************************************************ NETGRAPH NODE STUFF ************************************************************************/ /* * Constructor for a node */ static int ng_iface_constructor(node_p node) { struct ifnet *ifp; priv_p priv; /* Allocate node and interface private structures */ priv = malloc(sizeof(*priv), M_NETGRAPH_IFACE, M_WAITOK | M_ZERO); ifp = if_alloc(IFT_PROPVIRTUAL); if (ifp == NULL) { free(priv, M_NETGRAPH_IFACE); return (ENOMEM); } rm_init(&priv->lock, "ng_iface private rmlock"); /* Link them together */ ifp->if_softc = priv; priv->ifp = ifp; /* Get an interface unit number */ priv->unit = alloc_unr(V_ng_iface_unit); /* Link together node and private info */ NG_NODE_SET_PRIVATE(node, priv); priv->node = node; /* Initialize interface structure */ if_initname(ifp, NG_IFACE_IFACE_NAME, priv->unit); ifp->if_output = ng_iface_output; ifp->if_start = ng_iface_start; ifp->if_ioctl = ng_iface_ioctl; ifp->if_mtu = NG_IFACE_MTU_DEFAULT; ifp->if_flags = (IFF_SIMPLEX|IFF_POINTOPOINT|IFF_NOARP|IFF_MULTICAST); ifp->if_type = IFT_PROPVIRTUAL; /* XXX */ ifp->if_addrlen = 0; /* XXX */ ifp->if_hdrlen = 0; /* XXX */ ifp->if_baudrate = 64000; /* XXX */ IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); ifp->if_snd.ifq_drv_maxlen = ifqmaxlen; IFQ_SET_READY(&ifp->if_snd); /* Give this node the same name as the interface (if possible) */ if (ng_name_node(node, ifp->if_xname) != 0) log(LOG_WARNING, "%s: can't acquire netgraph name\n", ifp->if_xname); /* Attach the interface */ if_attach(ifp); bpfattach(ifp, DLT_NULL, sizeof(u_int32_t)); /* Done */ return (0); } /* * Give our ok for a hook to be added */ static int ng_iface_newhook(node_p node, hook_p hook, const char *name) { const iffam_p iffam = get_iffam_from_name(name); const priv_p priv = NG_NODE_PRIVATE(node); hook_p *hookptr; if (iffam == NULL) return (EPFNOSUPPORT); PRIV_WLOCK(priv); hookptr = get_hook_from_iffam(priv, iffam); if (*hookptr != NULL) { PRIV_WUNLOCK(priv); return (EISCONN); } *hookptr = hook; NG_HOOK_HI_STACK(hook); NG_HOOK_SET_TO_INBOUND(hook); PRIV_WUNLOCK(priv); return (0); } /* * Receive a control message */ static int ng_iface_rcvmsg(node_p node, item_p item, hook_p lasthook) { const priv_p priv = NG_NODE_PRIVATE(node); struct ifnet *const ifp = priv->ifp; struct ng_mesg *resp = NULL; int error = 0; struct ng_mesg *msg; NGI_GET_MSG(item, msg); switch (msg->header.typecookie) { case NGM_IFACE_COOKIE: switch (msg->header.cmd) { case NGM_IFACE_GET_IFNAME: NG_MKRESPONSE(resp, msg, IFNAMSIZ, M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } strlcpy(resp->data, ifp->if_xname, IFNAMSIZ); break; case NGM_IFACE_POINT2POINT: case NGM_IFACE_BROADCAST: { /* Deny request if interface is UP */ if ((ifp->if_flags & IFF_UP) != 0) return (EBUSY); /* Change flags */ switch (msg->header.cmd) { case NGM_IFACE_POINT2POINT: ifp->if_flags |= IFF_POINTOPOINT; ifp->if_flags &= ~IFF_BROADCAST; break; case NGM_IFACE_BROADCAST: ifp->if_flags &= ~IFF_POINTOPOINT; ifp->if_flags |= IFF_BROADCAST; break; } break; } case NGM_IFACE_GET_IFINDEX: NG_MKRESPONSE(resp, msg, sizeof(uint32_t), M_NOWAIT); if (resp == NULL) { error = ENOMEM; break; } *((uint32_t *)resp->data) = priv->ifp->if_index; break; default: error = EINVAL; break; } break; case NGM_FLOW_COOKIE: switch (msg->header.cmd) { case NGM_LINK_IS_UP: if_link_state_change(ifp, LINK_STATE_UP); break; case NGM_LINK_IS_DOWN: if_link_state_change(ifp, LINK_STATE_DOWN); break; default: break; } break; default: error = EINVAL; break; } NG_RESPOND_MSG(error, node, item, resp); NG_FREE_MSG(msg); return (error); } /* * Recive data from a hook. Pass the packet to the correct input routine. */ static int ng_iface_rcvdata(hook_p hook, item_p item) { const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook)); const iffam_p iffam = get_iffam_from_hook(priv, hook); struct ifnet *const ifp = priv->ifp; struct mbuf *m; int isr; NGI_GET_M(item, m); NG_FREE_ITEM(item); /* Sanity checks */ KASSERT(iffam != NULL, ("%s: iffam", __func__)); M_ASSERTPKTHDR(m); if ((ifp->if_flags & IFF_UP) == 0) { NG_FREE_M(m); return (ENETDOWN); } /* Update interface stats */ if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); /* Note receiving interface */ m->m_pkthdr.rcvif = ifp; /* Berkeley packet filter */ ng_iface_bpftap(ifp, m, iffam->family); /* Send packet */ switch (iffam->family) { #ifdef INET case AF_INET: isr = NETISR_IP; break; #endif #ifdef INET6 case AF_INET6: isr = NETISR_IPV6; break; #endif default: m_freem(m); return (EAFNOSUPPORT); } random_harvest_queue(m, sizeof(*m), 2, RANDOM_NET_NG); M_SETFIB(m, ifp->if_fib); netisr_dispatch(isr, m); return (0); } /* * Shutdown and remove the node and its associated interface. */ static int ng_iface_shutdown(node_p node) { const priv_p priv = NG_NODE_PRIVATE(node); /* * The ifnet may be in a different vnet than the netgraph node, * hence we have to change the current vnet context here. */ CURVNET_SET_QUIET(priv->ifp->if_vnet); bpfdetach(priv->ifp); if_detach(priv->ifp); if_free(priv->ifp); CURVNET_RESTORE(); priv->ifp = NULL; free_unr(V_ng_iface_unit, priv->unit); rm_destroy(&priv->lock); free(priv, M_NETGRAPH_IFACE); NG_NODE_SET_PRIVATE(node, NULL); NG_NODE_UNREF(node); return (0); } /* * Hook disconnection. Note that we do *not* shutdown when all * hooks have been disconnected. */ static int ng_iface_disconnect(hook_p hook) { const priv_p priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook)); const iffam_p iffam = get_iffam_from_hook(priv, hook); if (iffam == NULL) panic("%s", __func__); PRIV_WLOCK(priv); *get_hook_from_iffam(priv, iffam) = NULL; PRIV_WUNLOCK(priv); return (0); } /* * Handle loading and unloading for this node type. */ static int ng_iface_mod_event(module_t mod, int event, void *data) { int error = 0; switch (event) { case MOD_LOAD: case MOD_UNLOAD: break; default: error = EOPNOTSUPP; break; } return (error); } static void vnet_ng_iface_init(const void *unused) { V_ng_iface_unit = new_unrhdr(0, 0xffff, NULL); } VNET_SYSINIT(vnet_ng_iface_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_ng_iface_init, NULL); static void vnet_ng_iface_uninit(const void *unused) { delete_unrhdr(V_ng_iface_unit); } VNET_SYSUNINIT(vnet_ng_iface_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, vnet_ng_iface_uninit, NULL); Index: head/sys/netinet/cc/cc_cdg.c =================================================================== --- head/sys/netinet/cc/cc_cdg.c (revision 336675) +++ head/sys/netinet/cc/cc_cdg.c (revision 336676) @@ -1,701 +1,701 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2013 * Swinburne University of Technology, Melbourne, Australia * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes, made * possible in part by a gift from The Cisco University Research Program Fund, * a corporate advised fund of Silicon Valley Community Foundation. Development * and testing were further assisted by a grant from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * CAIA Delay-Gradient (CDG) congestion control algorithm * * An implemention of the delay-gradient congestion control algorithm proposed * in the following paper: * * D. A. Hayes and G. Armitage, "Revisiting TCP Congestion Control using Delay * Gradients", in IFIP Networking, Valencia, Spain, 9-13 May 2011. * * Developed as part of the NewTCP research project at Swinburne University of * Technology's Centre for Advanced Internet Architectures, Melbourne, * Australia. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CDG_VERSION "0.1" #define CAST_PTR_INT(X) (*((int*)(X))) /* Private delay-gradient induced congestion control signal. */ #define CC_CDG_DELAY 0x01000000 /* NewReno window deflation factor on loss (as a percentage). */ #define RENO_BETA 50 /* Queue states. */ #define CDG_Q_EMPTY 1 #define CDG_Q_RISING 2 #define CDG_Q_FALLING 3 #define CDG_Q_FULL 4 #define CDG_Q_UNKNOWN 9999 /* Number of bit shifts used in probexp lookup table. */ #define EXP_PREC 15 /* Largest gradient represented in probexp lookup table. */ #define MAXGRAD 5 /* * Delay Precision Enhance - number of bit shifts used for qtrend related * integer arithmetic precision. */ #define D_P_E 7 struct qdiff_sample { long qdiff; STAILQ_ENTRY(qdiff_sample) qdiff_lnk; }; struct cdg { long max_qtrend; long min_qtrend; STAILQ_HEAD(minrtts_head, qdiff_sample) qdiffmin_q; STAILQ_HEAD(maxrtts_head, qdiff_sample) qdiffmax_q; long window_incr; /* rttcount for window increase when in congestion avoidance */ long rtt_count; /* maximum measured rtt within an rtt period */ int maxrtt_in_rtt; /* maximum measured rtt within prev rtt period */ int maxrtt_in_prevrtt; /* minimum measured rtt within an rtt period */ int minrtt_in_rtt; /* minimum measured rtt within prev rtt period */ int minrtt_in_prevrtt; /* consecutive congestion episode counter */ uint32_t consec_cong_cnt; /* when tracking a new reno type loss window */ uint32_t shadow_w; /* maximum number of samples in the moving average queue */ int sample_q_size; /* number of samples in the moving average queue */ int num_samples; /* estimate of the queue state of the path */ int queue_state; }; /* * Lookup table for: * (1 - exp(-x)) << EXP_PREC, where x = [0,MAXGRAD] in 2^-7 increments * * Note: probexp[0] is set to 10 (not 0) as a safety for very low increase * gradients. */ static const int probexp[641] = { 10,255,508,759,1008,1255,1501,1744,1985,2225,2463,2698,2932,3165,3395,3624, 3850,4075,4299,4520,4740,4958,5175,5389,5602,5814,6024,6232,6438,6643,6846, 7048,7248,7447,7644,7839,8033,8226,8417,8606,8794,8981,9166,9350,9532,9713, 9892,10070,10247,10422,10596,10769,10940,11110,11278,11445,11611,11776,11939, 12101,12262,12422,12580,12737,12893,13048,13201,13354,13505,13655,13803,13951, 14097,14243,14387,14530,14672,14813,14952,15091,15229,15365,15500,15635,15768, 15900,16032,16162,16291,16419,16547,16673,16798,16922,17046,17168,17289,17410, 17529,17648,17766,17882,17998,18113,18227,18340,18453,18564,18675,18784,18893, 19001,19108,19215,19320,19425,19529,19632,19734,19835,19936,20036,20135,20233, 20331,20427,20523,20619,20713,20807,20900,20993,21084,21175,21265,21355,21444, 21532,21619,21706,21792,21878,21962,22046,22130,22213,22295,22376,22457,22537, 22617,22696,22774,22852,22929,23006,23082,23157,23232,23306,23380,23453,23525, 23597,23669,23739,23810,23879,23949,24017,24085,24153,24220,24286,24352,24418, 24483,24547,24611,24675,24738,24800,24862,24924,24985,25045,25106,25165,25224, 25283,25341,25399,25456,25513,25570,25626,25681,25737,25791,25846,25899,25953, 26006,26059,26111,26163,26214,26265,26316,26366,26416,26465,26514,26563,26611, 26659,26707,26754,26801,26847,26893,26939,26984,27029,27074,27118,27162,27206, 27249,27292,27335,27377,27419,27460,27502,27543,27583,27624,27664,27703,27743, 27782,27821,27859,27897,27935,27973,28010,28047,28084,28121,28157,28193,28228, 28263,28299,28333,28368,28402,28436,28470,28503,28536,28569,28602,28634,28667, 28699,28730,28762,28793,28824,28854,28885,28915,28945,28975,29004,29034,29063, 29092,29120,29149,29177,29205,29232,29260,29287,29314,29341,29368,29394,29421, 29447,29472,29498,29524,29549,29574,29599,29623,29648,29672,29696,29720,29744, 29767,29791,29814,29837,29860,29882,29905,29927,29949,29971,29993,30014,30036, 30057,30078,30099,30120,30141,30161,30181,30201,30221,30241,30261,30280,30300, 30319,30338,30357,30376,30394,30413,30431,30449,30467,30485,30503,30521,30538, 30555,30573,30590,30607,30624,30640,30657,30673,30690,30706,30722,30738,30753, 30769,30785,30800,30815,30831,30846,30861,30876,30890,30905,30919,30934,30948, 30962,30976,30990,31004,31018,31031,31045,31058,31072,31085,31098,31111,31124, 31137,31149,31162,31174,31187,31199,31211,31223,31235,31247,31259,31271,31283, 31294,31306,31317,31328,31339,31351,31362,31373,31383,31394,31405,31416,31426, 31436,31447,31457,31467,31477,31487,31497,31507,31517,31527,31537,31546,31556, 31565,31574,31584,31593,31602,31611,31620,31629,31638,31647,31655,31664,31673, 31681,31690,31698,31706,31715,31723,31731,31739,31747,31755,31763,31771,31778, 31786,31794,31801,31809,31816,31824,31831,31838,31846,31853,31860,31867,31874, 31881,31888,31895,31902,31908,31915,31922,31928,31935,31941,31948,31954,31960, 31967,31973,31979,31985,31991,31997,32003,32009,32015,32021,32027,32033,32038, 32044,32050,32055,32061,32066,32072,32077,32083,32088,32093,32098,32104,32109, 32114,32119,32124,32129,32134,32139,32144,32149,32154,32158,32163,32168,32173, 32177,32182,32186,32191,32195,32200,32204,32209,32213,32217,32222,32226,32230, 32234,32238,32242,32247,32251,32255,32259,32263,32267,32270,32274,32278,32282, 32286,32290,32293,32297,32301,32304,32308,32311,32315,32318,32322,32325,32329, 32332,32336,32339,32342,32346,32349,32352,32356,32359,32362,32365,32368,32371, 32374,32377,32381,32384,32387,32389,32392,32395,32398,32401,32404,32407,32410, 32412,32415,32418,32421,32423,32426,32429,32431,32434,32437,32439,32442,32444, 32447,32449,32452,32454,32457,32459,32461,32464,32466,32469,32471,32473,32476, 32478,32480,32482,32485,32487,32489,32491,32493,32495,32497,32500,32502,32504, 32506,32508,32510,32512,32514,32516,32518,32520,32522,32524,32526,32527,32529, 32531,32533,32535,32537,32538,32540,32542,32544,32545,32547}; static uma_zone_t qdiffsample_zone; static MALLOC_DEFINE(M_CDG, "cdg data", "Per connection data required for the CDG congestion control algorithm"); static int ertt_id; -static VNET_DEFINE(uint32_t, cdg_alpha_inc); -static VNET_DEFINE(uint32_t, cdg_beta_delay); -static VNET_DEFINE(uint32_t, cdg_beta_loss); -static VNET_DEFINE(uint32_t, cdg_smoothing_factor); -static VNET_DEFINE(uint32_t, cdg_exp_backoff_scale); -static VNET_DEFINE(uint32_t, cdg_consec_cong); -static VNET_DEFINE(uint32_t, cdg_hold_backoff); +VNET_DEFINE_STATIC(uint32_t, cdg_alpha_inc); +VNET_DEFINE_STATIC(uint32_t, cdg_beta_delay); +VNET_DEFINE_STATIC(uint32_t, cdg_beta_loss); +VNET_DEFINE_STATIC(uint32_t, cdg_smoothing_factor); +VNET_DEFINE_STATIC(uint32_t, cdg_exp_backoff_scale); +VNET_DEFINE_STATIC(uint32_t, cdg_consec_cong); +VNET_DEFINE_STATIC(uint32_t, cdg_hold_backoff); #define V_cdg_alpha_inc VNET(cdg_alpha_inc) #define V_cdg_beta_delay VNET(cdg_beta_delay) #define V_cdg_beta_loss VNET(cdg_beta_loss) #define V_cdg_smoothing_factor VNET(cdg_smoothing_factor) #define V_cdg_exp_backoff_scale VNET(cdg_exp_backoff_scale) #define V_cdg_consec_cong VNET(cdg_consec_cong) #define V_cdg_hold_backoff VNET(cdg_hold_backoff) /* Function prototypes. */ static int cdg_mod_init(void); static int cdg_mod_destroy(void); static void cdg_conn_init(struct cc_var *ccv); static int cdg_cb_init(struct cc_var *ccv); static void cdg_cb_destroy(struct cc_var *ccv); static void cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type); static void cdg_ack_received(struct cc_var *ccv, uint16_t ack_type); struct cc_algo cdg_cc_algo = { .name = "cdg", .mod_init = cdg_mod_init, .ack_received = cdg_ack_received, .cb_destroy = cdg_cb_destroy, .cb_init = cdg_cb_init, .conn_init = cdg_conn_init, .cong_signal = cdg_cong_signal, .mod_destroy = cdg_mod_destroy }; /* Vnet created and being initialised. */ static void cdg_init_vnet(const void *unused __unused) { V_cdg_alpha_inc = 0; V_cdg_beta_delay = 70; V_cdg_beta_loss = 50; V_cdg_smoothing_factor = 8; V_cdg_exp_backoff_scale = 3; V_cdg_consec_cong = 5; V_cdg_hold_backoff = 5; } static int cdg_mod_init(void) { VNET_ITERATOR_DECL(v); ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) return (EINVAL); qdiffsample_zone = uma_zcreate("cdg_qdiffsample", sizeof(struct qdiff_sample), NULL, NULL, NULL, NULL, 0, 0); VNET_LIST_RLOCK(); VNET_FOREACH(v) { CURVNET_SET(v); cdg_init_vnet(NULL); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); cdg_cc_algo.post_recovery = newreno_cc_algo.post_recovery; cdg_cc_algo.after_idle = newreno_cc_algo.after_idle; return (0); } static int cdg_mod_destroy(void) { uma_zdestroy(qdiffsample_zone); return (0); } static int cdg_cb_init(struct cc_var *ccv) { struct cdg *cdg_data; cdg_data = malloc(sizeof(struct cdg), M_CDG, M_NOWAIT); if (cdg_data == NULL) return (ENOMEM); cdg_data->shadow_w = 0; cdg_data->max_qtrend = 0; cdg_data->min_qtrend = 0; cdg_data->queue_state = CDG_Q_UNKNOWN; cdg_data->maxrtt_in_rtt = 0; cdg_data->maxrtt_in_prevrtt = 0; cdg_data->minrtt_in_rtt = INT_MAX; cdg_data->minrtt_in_prevrtt = 0; cdg_data->window_incr = 0; cdg_data->rtt_count = 0; cdg_data->consec_cong_cnt = 0; cdg_data->sample_q_size = V_cdg_smoothing_factor; cdg_data->num_samples = 0; STAILQ_INIT(&cdg_data->qdiffmin_q); STAILQ_INIT(&cdg_data->qdiffmax_q); ccv->cc_data = cdg_data; return (0); } static void cdg_conn_init(struct cc_var *ccv) { struct cdg *cdg_data = ccv->cc_data; /* * Initialise the shadow_cwnd in case we are competing with loss based * flows from the start */ cdg_data->shadow_w = CCV(ccv, snd_cwnd); } static void cdg_cb_destroy(struct cc_var *ccv) { struct cdg *cdg_data; struct qdiff_sample *qds, *qds_n; cdg_data = ccv->cc_data; qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); while (qds != NULL) { qds_n = STAILQ_NEXT(qds, qdiff_lnk); uma_zfree(qdiffsample_zone,qds); qds = qds_n; } qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); while (qds != NULL) { qds_n = STAILQ_NEXT(qds, qdiff_lnk); uma_zfree(qdiffsample_zone,qds); qds = qds_n; } free(ccv->cc_data, M_CDG); } static int cdg_beta_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr != NULL && (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100)) return (EINVAL); return (sysctl_handle_int(oidp, arg1, arg2, req)); } static int cdg_exp_backoff_scale_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr != NULL && CAST_PTR_INT(req->newptr) < 1) return (EINVAL); return (sysctl_handle_int(oidp, arg1, arg2, req)); } static inline uint32_t cdg_window_decrease(struct cc_var *ccv, unsigned long owin, unsigned int beta) { return ((ulmin(CCV(ccv, snd_wnd), owin) * beta) / 100); } /* * Window increase function * This window increase function is independent of the initial window size * to ensure small window flows are not discriminated against (i.e. fairness). * It increases at 1pkt/rtt like Reno for alpha_inc rtts, and then 2pkts/rtt for * the next alpha_inc rtts, etc. */ static void cdg_window_increase(struct cc_var *ccv, int new_measurement) { struct cdg *cdg_data; int incr, s_w_incr; cdg_data = ccv->cc_data; incr = s_w_incr = 0; if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { /* Slow start. */ incr = CCV(ccv, t_maxseg); s_w_incr = incr; cdg_data->window_incr = cdg_data->rtt_count = 0; } else { /* Congestion avoidance. */ if (new_measurement) { s_w_incr = CCV(ccv, t_maxseg); if (V_cdg_alpha_inc == 0) { incr = CCV(ccv, t_maxseg); } else { if (++cdg_data->rtt_count >= V_cdg_alpha_inc) { cdg_data->window_incr++; cdg_data->rtt_count = 0; } incr = CCV(ccv, t_maxseg) * cdg_data->window_incr; } } } if (cdg_data->shadow_w > 0) cdg_data->shadow_w = ulmin(cdg_data->shadow_w + s_w_incr, TCP_MAXWIN << CCV(ccv, snd_scale)); CCV(ccv, snd_cwnd) = ulmin(CCV(ccv, snd_cwnd) + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } static void cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type) { struct cdg *cdg_data = ccv->cc_data; switch(signal_type) { case CC_CDG_DELAY: CCV(ccv, snd_ssthresh) = cdg_window_decrease(ccv, CCV(ccv, snd_cwnd), V_cdg_beta_delay); CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); CCV(ccv, snd_recover) = CCV(ccv, snd_max); cdg_data->window_incr = cdg_data->rtt_count = 0; ENTER_CONGRECOVERY(CCV(ccv, t_flags)); break; case CC_NDUPACK: /* * If already responding to congestion OR we have guessed no * queue in the path is full. */ if (IN_CONGRECOVERY(CCV(ccv, t_flags)) || cdg_data->queue_state < CDG_Q_FULL) { CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); CCV(ccv, snd_recover) = CCV(ccv, snd_max); } else { /* * Loss is likely to be congestion related. We have * inferred a queue full state, so have shadow window * react to loss as NewReno would. */ if (cdg_data->shadow_w > 0) cdg_data->shadow_w = cdg_window_decrease(ccv, cdg_data->shadow_w, RENO_BETA); CCV(ccv, snd_ssthresh) = max(cdg_data->shadow_w, cdg_window_decrease(ccv, CCV(ccv, snd_cwnd), V_cdg_beta_loss)); cdg_data->window_incr = cdg_data->rtt_count = 0; } ENTER_RECOVERY(CCV(ccv, t_flags)); break; default: newreno_cc_algo.cong_signal(ccv, signal_type); break; } } /* * Using a negative exponential probabilistic backoff so that sources with * varying RTTs which share the same link will, on average, have the same * probability of backoff over time. * * Prob_backoff = 1 - exp(-qtrend / V_cdg_exp_backoff_scale), where * V_cdg_exp_backoff_scale is the average qtrend for the exponential backoff. */ static inline int prob_backoff(long qtrend) { int backoff, idx, p; backoff = (qtrend > ((MAXGRAD * V_cdg_exp_backoff_scale) << D_P_E)); if (!backoff) { if (V_cdg_exp_backoff_scale > 1) idx = (qtrend + V_cdg_exp_backoff_scale / 2) / V_cdg_exp_backoff_scale; else idx = qtrend; /* Backoff probability proportional to rate of queue growth. */ p = (INT_MAX / (1 << EXP_PREC)) * probexp[idx]; backoff = (random() < p); } return (backoff); } static inline void calc_moving_average(struct cdg *cdg_data, long qdiff_max, long qdiff_min) { struct qdiff_sample *qds; ++cdg_data->num_samples; if (cdg_data->num_samples > cdg_data->sample_q_size) { /* Minimum RTT. */ qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); cdg_data->min_qtrend = cdg_data->min_qtrend + (qdiff_min - qds->qdiff) / cdg_data->sample_q_size; STAILQ_REMOVE_HEAD(&cdg_data->qdiffmin_q, qdiff_lnk); qds->qdiff = qdiff_min; STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, qdiff_lnk); /* Maximum RTT. */ qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); cdg_data->max_qtrend = cdg_data->max_qtrend + (qdiff_max - qds->qdiff) / cdg_data->sample_q_size; STAILQ_REMOVE_HEAD(&cdg_data->qdiffmax_q, qdiff_lnk); qds->qdiff = qdiff_max; STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, qdiff_lnk); --cdg_data->num_samples; } else { qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); if (qds != NULL) { cdg_data->min_qtrend = cdg_data->min_qtrend + qdiff_min / cdg_data->sample_q_size; qds->qdiff = qdiff_min; STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, qdiff_lnk); } qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); if (qds) { cdg_data->max_qtrend = cdg_data->max_qtrend + qdiff_max / cdg_data->sample_q_size; qds->qdiff = qdiff_max; STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, qdiff_lnk); } } } static void cdg_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct cdg *cdg_data; struct ertt *e_t; long qdiff_max, qdiff_min; int congestion, new_measurement, slowstart; cdg_data = ccv->cc_data; e_t = (struct ertt *)khelp_get_osd(CCV(ccv, osd), ertt_id); new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; congestion = 0; cdg_data->maxrtt_in_rtt = imax(e_t->rtt, cdg_data->maxrtt_in_rtt); cdg_data->minrtt_in_rtt = imin(e_t->rtt, cdg_data->minrtt_in_rtt); if (new_measurement) { slowstart = (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)); /* * Update smoothed gradient measurements. Since we are only * using one measurement per RTT, use max or min rtt_in_rtt. * This is also less noisy than a sample RTT measurement. Max * RTT measurements can have trouble due to OS issues. */ if (cdg_data->maxrtt_in_prevrtt) { qdiff_max = ((long)(cdg_data->maxrtt_in_rtt - cdg_data->maxrtt_in_prevrtt) << D_P_E ); qdiff_min = ((long)(cdg_data->minrtt_in_rtt - cdg_data->minrtt_in_prevrtt) << D_P_E ); calc_moving_average(cdg_data, qdiff_max, qdiff_min); /* Probabilistic backoff with respect to gradient. */ if (slowstart && qdiff_min > 0) congestion = prob_backoff(qdiff_min); else if (cdg_data->min_qtrend > 0) congestion = prob_backoff(cdg_data->min_qtrend); else if (slowstart && qdiff_max > 0) congestion = prob_backoff(qdiff_max); else if (cdg_data->max_qtrend > 0) congestion = prob_backoff(cdg_data->max_qtrend); /* Update estimate of queue state. */ if (cdg_data->min_qtrend > 0 && cdg_data->max_qtrend <= 0) { cdg_data->queue_state = CDG_Q_FULL; } else if (cdg_data->min_qtrend >= 0 && cdg_data->max_qtrend < 0) { cdg_data->queue_state = CDG_Q_EMPTY; cdg_data->shadow_w = 0; } else if (cdg_data->min_qtrend > 0 && cdg_data->max_qtrend > 0) { cdg_data->queue_state = CDG_Q_RISING; } else if (cdg_data->min_qtrend < 0 && cdg_data->max_qtrend < 0) { cdg_data->queue_state = CDG_Q_FALLING; } if (cdg_data->min_qtrend < 0 || cdg_data->max_qtrend < 0) cdg_data->consec_cong_cnt = 0; } cdg_data->minrtt_in_prevrtt = cdg_data->minrtt_in_rtt; cdg_data->minrtt_in_rtt = INT_MAX; cdg_data->maxrtt_in_prevrtt = cdg_data->maxrtt_in_rtt; cdg_data->maxrtt_in_rtt = 0; e_t->flags &= ~ERTT_NEW_MEASUREMENT; } if (congestion) { cdg_data->consec_cong_cnt++; if (!IN_RECOVERY(CCV(ccv, t_flags))) { if (cdg_data->consec_cong_cnt <= V_cdg_consec_cong) cdg_cong_signal(ccv, CC_CDG_DELAY); else /* * We have been backing off but the queue is not * falling. Assume we are competing with * loss-based flows and don't back off for the * next V_cdg_hold_backoff RTT periods. */ if (cdg_data->consec_cong_cnt >= V_cdg_consec_cong + V_cdg_hold_backoff) cdg_data->consec_cong_cnt = 0; /* Won't see effect until 2nd RTT. */ cdg_data->maxrtt_in_prevrtt = 0; /* * Resync shadow window in case we are competing with a * loss based flow */ cdg_data->shadow_w = ulmax(CCV(ccv, snd_cwnd), cdg_data->shadow_w); } } else if (ack_type == CC_ACK) cdg_window_increase(ccv, new_measurement); } /* When a vnet is created and being initialised, init the per-stack CDG vars. */ VNET_SYSINIT(cdg_init_vnet, SI_SUB_PROTO_BEGIN, SI_ORDER_FIRST, cdg_init_vnet, NULL); SYSCTL_DECL(_net_inet_tcp_cc_cdg); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, cdg, CTLFLAG_RW, NULL, "CAIA delay-gradient congestion control related settings"); SYSCTL_STRING(_net_inet_tcp_cc_cdg, OID_AUTO, version, CTLFLAG_RD, CDG_VERSION, sizeof(CDG_VERSION) - 1, "Current algorithm/implementation version number"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, alpha_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_alpha_inc), 0, "Increment the window increase factor alpha by 1 MSS segment every " "alpha_inc RTTs during congestion avoidance mode."); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_delay, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_delay), 70, &cdg_beta_handler, "IU", "Delay-based window decrease factor as a percentage " "(on delay-based backoff, w = w * beta_delay / 100)"); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_loss, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_loss), 50, &cdg_beta_handler, "IU", "Loss-based window decrease factor as a percentage " "(on loss-based backoff, w = w * beta_loss / 100)"); SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, exp_backoff_scale, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_exp_backoff_scale), 2, &cdg_exp_backoff_scale_handler, "IU", "Scaling parameter for the probabilistic exponential backoff"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, smoothing_factor, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_smoothing_factor), 8, "Number of samples used for moving average smoothing (0 = no smoothing)"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_consec_cong, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_consec_cong), 5, "Number of consecutive delay-gradient based congestion episodes which will " "trigger loss based CC compatibility"); SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_hold_backoff, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_hold_backoff), 5, "Number of consecutive delay-gradient based congestion episodes to hold " "the window backoff for loss based CC compatibility"); DECLARE_CC_MODULE(cdg, &cdg_cc_algo); MODULE_DEPEND(cdg, ertt, 1, 1, 1); Index: head/sys/netinet/cc/cc_chd.c =================================================================== --- head/sys/netinet/cc/cc_chd.c (revision 336675) +++ head/sys/netinet/cc/cc_chd.c (revision 336676) @@ -1,499 +1,499 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2010 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2010-2011 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes and * Lawrence Stewart, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, Melbourne, Australia by * David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the CAIA-Hamilton delay based congestion control * algorithm, based on "Improved coexistence and loss tolerance for delay based * TCP congestion control" by D. A. Hayes and G. Armitage., in 35th Annual IEEE * Conference on Local Computer Networks (LCN 2010), Denver, Colorado, USA, * 11-14 October 2010. * * Originally released as part of the NewTCP research project at Swinburne * University of Technology's Centre for Advanced Internet Architectures, * Melbourne, Australia, which was made possible in part by a grant from the * Cisco University Research Program Fund at Community Foundation Silicon * Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) /* * Private signal type for rate based congestion signal. * See for appropriate bit-range to use for private signals. */ #define CC_CHD_DELAY 0x02000000 /* Largest possible number returned by random(). */ #define RANDOM_MAX INT_MAX static void chd_ack_received(struct cc_var *ccv, uint16_t ack_type); static void chd_cb_destroy(struct cc_var *ccv); static int chd_cb_init(struct cc_var *ccv); static void chd_cong_signal(struct cc_var *ccv, uint32_t signal_type); static void chd_conn_init(struct cc_var *ccv); static int chd_mod_init(void); struct chd { /* * Shadow window - keeps track of what the NewReno congestion window * would have been if delay-based cwnd backoffs had not been made. This * functionality aids coexistence with loss-based TCP flows which may be * sharing links along the path. */ unsigned long shadow_w; /* * Loss-based TCP compatibility flag - When set, it turns on the shadow * window functionality. */ int loss_compete; /* The maximum round trip time seen within a measured rtt period. */ int maxrtt_in_rtt; /* The previous qdly that caused cwnd to backoff. */ int prev_backoff_qdly; }; static int ertt_id; -static VNET_DEFINE(uint32_t, chd_qmin) = 5; -static VNET_DEFINE(uint32_t, chd_pmax) = 50; -static VNET_DEFINE(uint32_t, chd_loss_fair) = 1; -static VNET_DEFINE(uint32_t, chd_use_max) = 1; -static VNET_DEFINE(uint32_t, chd_qthresh) = 20; +VNET_DEFINE_STATIC(uint32_t, chd_qmin) = 5; +VNET_DEFINE_STATIC(uint32_t, chd_pmax) = 50; +VNET_DEFINE_STATIC(uint32_t, chd_loss_fair) = 1; +VNET_DEFINE_STATIC(uint32_t, chd_use_max) = 1; +VNET_DEFINE_STATIC(uint32_t, chd_qthresh) = 20; #define V_chd_qthresh VNET(chd_qthresh) #define V_chd_qmin VNET(chd_qmin) #define V_chd_pmax VNET(chd_pmax) #define V_chd_loss_fair VNET(chd_loss_fair) #define V_chd_use_max VNET(chd_use_max) static MALLOC_DEFINE(M_CHD, "chd data", "Per connection data required for the CHD congestion control algorithm"); struct cc_algo chd_cc_algo = { .name = "chd", .ack_received = chd_ack_received, .cb_destroy = chd_cb_destroy, .cb_init = chd_cb_init, .cong_signal = chd_cong_signal, .conn_init = chd_conn_init, .mod_init = chd_mod_init }; static __inline void chd_window_decrease(struct cc_var *ccv) { unsigned long win; win = min(CCV(ccv, snd_wnd), CCV(ccv, snd_cwnd)) / CCV(ccv, t_maxseg); win -= max((win / 2), 1); CCV(ccv, snd_ssthresh) = max(win, 2) * CCV(ccv, t_maxseg); } /* * Probabilistic backoff function. Returns 1 if we should backoff or 0 * otherwise. The calculation of p is similar to the calculation of p in cc_hd. */ static __inline int should_backoff(int qdly, int maxqdly, struct chd *chd_data) { unsigned long p, rand; rand = random(); if (qdly < V_chd_qthresh) { chd_data->loss_compete = 0; p = (((RANDOM_MAX / 100) * V_chd_pmax) / (V_chd_qthresh - V_chd_qmin)) * (qdly - V_chd_qmin); } else { if (qdly > V_chd_qthresh) { p = (((RANDOM_MAX / 100) * V_chd_pmax) / (maxqdly - V_chd_qthresh)) * (maxqdly - qdly); if (V_chd_loss_fair && rand < p) chd_data->loss_compete = 1; } else { p = (RANDOM_MAX / 100) * V_chd_pmax; chd_data->loss_compete = 0; } } return (rand < p); } static __inline void chd_window_increase(struct cc_var *ccv, int new_measurement) { struct chd *chd_data; int incr; chd_data = ccv->cc_data; incr = 0; if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { /* Adapted from NewReno slow start. */ if (V_tcp_do_rfc3465) { /* In slow-start with ABC enabled. */ if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) { /* Not due to RTO. */ incr = min(ccv->bytes_this_ack, V_tcp_abc_l_var * CCV(ccv, t_maxseg)); } else { /* Due to RTO. */ incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); } } else incr = CCV(ccv, t_maxseg); } else { /* Congestion avoidance. */ if (V_tcp_do_rfc3465) { if (ccv->flags & CCF_ABC_SENTAWND) { ccv->flags &= ~CCF_ABC_SENTAWND; incr = CCV(ccv, t_maxseg); } } else if (new_measurement) incr = CCV(ccv, t_maxseg); } if (chd_data->shadow_w > 0) { /* Track NewReno window. */ chd_data->shadow_w = min(chd_data->shadow_w + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } CCV(ccv,snd_cwnd) = min(CCV(ccv, snd_cwnd) + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } /* * All ACK signals are used for timing measurements to determine delay-based * congestion. However, window increases are only performed when * ack_type == CC_ACK. */ static void chd_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct chd *chd_data; struct ertt *e_t; int backoff, new_measurement, qdly, rtt; e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); chd_data = ccv->cc_data; new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; backoff = qdly = 0; chd_data->maxrtt_in_rtt = imax(e_t->rtt, chd_data->maxrtt_in_rtt); if (new_measurement) { /* * There is a new per RTT measurement, so check to see if there * is delay based congestion. */ rtt = V_chd_use_max ? chd_data->maxrtt_in_rtt : e_t->rtt; chd_data->maxrtt_in_rtt = 0; if (rtt && e_t->minrtt && !IN_RECOVERY(CCV(ccv, t_flags))) { qdly = rtt - e_t->minrtt; if (qdly > V_chd_qmin) { /* * Probabilistic delay based congestion * indication. */ backoff = should_backoff(qdly, e_t->maxrtt - e_t->minrtt, chd_data); } else chd_data->loss_compete = 0; } /* Reset per RTT measurement flag to start a new measurement. */ e_t->flags &= ~ERTT_NEW_MEASUREMENT; } if (backoff) { /* * Update shadow_w before delay based backoff. */ if (chd_data->loss_compete || qdly > chd_data->prev_backoff_qdly) { /* * Delay is higher than when we backed off previously, * so it is possible that this flow is competing with * loss based flows. */ chd_data->shadow_w = max(CCV(ccv, snd_cwnd), chd_data->shadow_w); } else { /* * Reset shadow_w, as it is probable that this flow is * not competing with loss based flows at the moment. */ chd_data->shadow_w = 0; } chd_data->prev_backoff_qdly = qdly; /* * Send delay-based congestion signal to the congestion signal * handler. */ chd_cong_signal(ccv, CC_CHD_DELAY); } else if (ack_type == CC_ACK) chd_window_increase(ccv, new_measurement); } static void chd_cb_destroy(struct cc_var *ccv) { free(ccv->cc_data, M_CHD); } static int chd_cb_init(struct cc_var *ccv) { struct chd *chd_data; chd_data = malloc(sizeof(struct chd), M_CHD, M_NOWAIT); if (chd_data == NULL) return (ENOMEM); chd_data->shadow_w = 0; ccv->cc_data = chd_data; return (0); } static void chd_cong_signal(struct cc_var *ccv, uint32_t signal_type) { struct ertt *e_t; struct chd *chd_data; int qdly; e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); chd_data = ccv->cc_data; qdly = imax(e_t->rtt, chd_data->maxrtt_in_rtt) - e_t->minrtt; switch(signal_type) { case CC_CHD_DELAY: chd_window_decrease(ccv); /* Set new ssthresh. */ CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); CCV(ccv, snd_recover) = CCV(ccv, snd_max); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); break; case CC_NDUPACK: /* Packet loss. */ /* * Only react to loss as a congestion signal if qdly > * V_chd_qthresh. If qdly is less than qthresh, presume that * this is a non congestion related loss. If qdly is greater * than qthresh, assume that we are competing with loss based * tcp flows and restore window from any unnecessary backoffs, * before the decrease. */ if (!IN_RECOVERY(CCV(ccv, t_flags)) && qdly > V_chd_qthresh) { if (chd_data->loss_compete) { CCV(ccv, snd_cwnd) = max(CCV(ccv, snd_cwnd), chd_data->shadow_w); } chd_window_decrease(ccv); } else { /* * This loss isn't congestion related, or already * recovering from congestion. */ CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); CCV(ccv, snd_recover) = CCV(ccv, snd_max); } if (chd_data->shadow_w > 0) { chd_data->shadow_w = max(chd_data->shadow_w / CCV(ccv, t_maxseg) / 2, 2) * CCV(ccv, t_maxseg); } ENTER_FASTRECOVERY(CCV(ccv, t_flags)); break; default: newreno_cc_algo.cong_signal(ccv, signal_type); } } static void chd_conn_init(struct cc_var *ccv) { struct chd *chd_data; chd_data = ccv->cc_data; chd_data->prev_backoff_qdly = 0; chd_data->maxrtt_in_rtt = 0; chd_data->loss_compete = 0; /* * Initialise the shadow_cwnd to be equal to snd_cwnd in case we are * competing with loss based flows from the start. */ chd_data->shadow_w = CCV(ccv, snd_cwnd); } static int chd_mod_init(void) { ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) { printf("%s: h_ertt module not found\n", __func__); return (ENOENT); } chd_cc_algo.after_idle = newreno_cc_algo.after_idle; chd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; return (0); } static int chd_loss_fair_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_loss_fair; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_chd_loss_fair = new; } return (error); } static int chd_pmax_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_pmax; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100) error = EINVAL; else V_chd_pmax = new; } return (error); } static int chd_qthresh_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_chd_qthresh; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) <= V_chd_qmin) error = EINVAL; else V_chd_qthresh = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_chd); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, chd, CTLFLAG_RW, NULL, "CAIA Hamilton delay-based congestion control related settings"); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, loss_fair, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_loss_fair), 1, &chd_loss_fair_handler, "IU", "Flag to enable shadow window functionality."); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, pmax, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_pmax), 5, &chd_pmax_handler, "IU", "Per RTT maximum backoff probability as a percentage"); SYSCTL_PROC(_net_inet_tcp_cc_chd, OID_AUTO, queue_threshold, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(chd_qthresh), 20, &chd_qthresh_handler, "IU", "Queueing congestion threshold in ticks"); SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, queue_min, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_qmin), 5, "Minimum queueing delay threshold in ticks"); SYSCTL_UINT(_net_inet_tcp_cc_chd, OID_AUTO, use_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(chd_use_max), 1, "Use the maximum RTT seen within the measurement period (RTT) " "as the basic delay measurement for the algorithm."); DECLARE_CC_MODULE(chd, &chd_cc_algo); MODULE_DEPEND(chd, ertt, 1, 1, 1); Index: head/sys/netinet/cc/cc_dctcp.c =================================================================== --- head/sys/netinet/cc/cc_dctcp.c (revision 336675) +++ head/sys/netinet/cc/cc_dctcp.c (revision 336676) @@ -1,471 +1,471 @@ /*- * Copyright (c) 2007-2008 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2014 Midori Kato * Copyright (c) 2014 The FreeBSD Foundation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the DCTCP algorithm for FreeBSD, based on * "Data Center TCP (DCTCP)" by M. Alizadeh, A. Greenberg, D. A. Maltz, * J. Padhye, P. Patel, B. Prabhakar, S. Sengupta, and M. Sridharan., * in ACM Conference on SIGCOMM 2010, New York, USA, * Originally released as the contribution of Microsoft Research project. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) #define MAX_ALPHA_VALUE 1024 -static VNET_DEFINE(uint32_t, dctcp_alpha) = 0; +VNET_DEFINE_STATIC(uint32_t, dctcp_alpha) = 0; #define V_dctcp_alpha VNET(dctcp_alpha) -static VNET_DEFINE(uint32_t, dctcp_shift_g) = 4; +VNET_DEFINE_STATIC(uint32_t, dctcp_shift_g) = 4; #define V_dctcp_shift_g VNET(dctcp_shift_g) -static VNET_DEFINE(uint32_t, dctcp_slowstart) = 0; +VNET_DEFINE_STATIC(uint32_t, dctcp_slowstart) = 0; #define V_dctcp_slowstart VNET(dctcp_slowstart) struct dctcp { int bytes_ecn; /* # of marked bytes during a RTT */ int bytes_total; /* # of acked bytes during a RTT */ int alpha; /* the fraction of marked bytes */ int ce_prev; /* CE state of the last segment */ int save_sndnxt; /* end sequence number of the current window */ int ece_curr; /* ECE flag in this segment */ int ece_prev; /* ECE flag in the last segment */ uint32_t num_cong_events; /* # of congestion events */ }; static MALLOC_DEFINE(M_dctcp, "dctcp data", "Per connection data required for the dctcp algorithm"); static void dctcp_ack_received(struct cc_var *ccv, uint16_t type); static void dctcp_after_idle(struct cc_var *ccv); static void dctcp_cb_destroy(struct cc_var *ccv); static int dctcp_cb_init(struct cc_var *ccv); static void dctcp_cong_signal(struct cc_var *ccv, uint32_t type); static void dctcp_conn_init(struct cc_var *ccv); static void dctcp_post_recovery(struct cc_var *ccv); static void dctcp_ecnpkt_handler(struct cc_var *ccv); static void dctcp_update_alpha(struct cc_var *ccv); struct cc_algo dctcp_cc_algo = { .name = "dctcp", .ack_received = dctcp_ack_received, .cb_destroy = dctcp_cb_destroy, .cb_init = dctcp_cb_init, .cong_signal = dctcp_cong_signal, .conn_init = dctcp_conn_init, .post_recovery = dctcp_post_recovery, .ecnpkt_handler = dctcp_ecnpkt_handler, .after_idle = dctcp_after_idle, }; static void dctcp_ack_received(struct cc_var *ccv, uint16_t type) { struct dctcp *dctcp_data; int bytes_acked = 0; dctcp_data = ccv->cc_data; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) { /* * DCTCP doesn't treat receipt of ECN marked packet as a * congestion event. Thus, DCTCP always executes the ACK * processing out of congestion recovery. */ if (IN_CONGRECOVERY(CCV(ccv, t_flags))) { EXIT_CONGRECOVERY(CCV(ccv, t_flags)); newreno_cc_algo.ack_received(ccv, type); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } else newreno_cc_algo.ack_received(ccv, type); if (type == CC_DUPACK) bytes_acked = CCV(ccv, t_maxseg); if (type == CC_ACK) bytes_acked = ccv->bytes_this_ack; /* Update total bytes. */ dctcp_data->bytes_total += bytes_acked; /* Update total marked bytes. */ if (dctcp_data->ece_curr) { if (!dctcp_data->ece_prev && bytes_acked > CCV(ccv, t_maxseg)) { dctcp_data->bytes_ecn += (bytes_acked - CCV(ccv, t_maxseg)); } else dctcp_data->bytes_ecn += bytes_acked; dctcp_data->ece_prev = 1; } else { if (dctcp_data->ece_prev && bytes_acked > CCV(ccv, t_maxseg)) dctcp_data->bytes_ecn += CCV(ccv, t_maxseg); dctcp_data->ece_prev = 0; } dctcp_data->ece_curr = 0; /* * Update the fraction of marked bytes at the end of * current window size. */ if ((IN_FASTRECOVERY(CCV(ccv, t_flags)) && SEQ_GEQ(ccv->curack, CCV(ccv, snd_recover))) || (!IN_FASTRECOVERY(CCV(ccv, t_flags)) && SEQ_GT(ccv->curack, dctcp_data->save_sndnxt))) dctcp_update_alpha(ccv); } else newreno_cc_algo.ack_received(ccv, type); } static void dctcp_after_idle(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = ccv->cc_data; /* Initialize internal parameters after idle time */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); dctcp_data->alpha = V_dctcp_alpha; dctcp_data->ece_curr = 0; dctcp_data->ece_prev = 0; dctcp_data->num_cong_events = 0; dctcp_cc_algo.after_idle = newreno_cc_algo.after_idle; } static void dctcp_cb_destroy(struct cc_var *ccv) { free(ccv->cc_data, M_dctcp); } static int dctcp_cb_init(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = malloc(sizeof(struct dctcp), M_dctcp, M_NOWAIT|M_ZERO); if (dctcp_data == NULL) return (ENOMEM); /* Initialize some key variables with sensible defaults. */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; /* * When alpha is set to 0 in the beginning, DCTCP sender transfers as * much data as possible until the value converges which may expand the * queueing delay at the switch. When alpha is set to 1, queueing delay * is kept small. * Throughput-sensitive applications should have alpha = 0 * Latency-sensitive applications should have alpha = 1 * * Note: DCTCP draft suggests initial alpha to be 1 but we've decided to * keep it 0 as default. */ dctcp_data->alpha = V_dctcp_alpha; dctcp_data->save_sndnxt = 0; dctcp_data->ce_prev = 0; dctcp_data->ece_curr = 0; dctcp_data->ece_prev = 0; dctcp_data->num_cong_events = 0; ccv->cc_data = dctcp_data; return (0); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void dctcp_cong_signal(struct cc_var *ccv, uint32_t type) { struct dctcp *dctcp_data; u_int win, mss; dctcp_data = ccv->cc_data; win = CCV(ccv, snd_cwnd); mss = CCV(ccv, t_maxseg); switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { CCV(ccv, snd_ssthresh) = mss * max(win / 2 / mss, 2); dctcp_data->num_cong_events++; } else { /* cwnd has already updated as congestion * recovery. Reverse cwnd value using * snd_cwnd_prev and recalculate snd_ssthresh */ win = CCV(ccv, snd_cwnd_prev); CCV(ccv, snd_ssthresh) = max(win / 2 / mss, 2) * mss; } ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: /* * Save current snd_cwnd when the host encounters both * congestion recovery and fast recovery. */ CCV(ccv, snd_cwnd_prev) = win; if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { if (V_dctcp_slowstart && dctcp_data->num_cong_events++ == 0) { CCV(ccv, snd_ssthresh) = mss * max(win / 2 / mss, 2); dctcp_data->alpha = MAX_ALPHA_VALUE; dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } else CCV(ccv, snd_ssthresh) = max((win - ((win * dctcp_data->alpha) >> 11)) / mss, 2) * mss; CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } dctcp_data->ece_curr = 1; break; case CC_RTO: if (CCV(ccv, t_flags) & TF_ECN_PERMIT) { CCV(ccv, t_flags) |= TF_ECN_SND_CWR; dctcp_update_alpha(ccv); dctcp_data->save_sndnxt += CCV(ccv, t_maxseg); dctcp_data->num_cong_events++; } break; } } static void dctcp_conn_init(struct cc_var *ccv) { struct dctcp *dctcp_data; dctcp_data = ccv->cc_data; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } /* * Perform any necessary tasks before we exit congestion recovery. */ static void dctcp_post_recovery(struct cc_var *ccv) { dctcp_cc_algo.post_recovery = newreno_cc_algo.post_recovery; if (CCV(ccv, t_flags) & TF_ECN_PERMIT) dctcp_update_alpha(ccv); } /* * Execute an additional ECN processing using ECN field in IP header and the CWR * bit in TCP header. * * delay_ack == 0 - Delayed ACK disabled * delay_ack == 1 - Delayed ACK enabled */ static void dctcp_ecnpkt_handler(struct cc_var *ccv) { struct dctcp *dctcp_data; uint32_t ccflag; int delay_ack; dctcp_data = ccv->cc_data; ccflag = ccv->flags; delay_ack = 1; /* * DCTCP responses an ACK immediately when the CE state * in between this segment and the last segment is not same. */ if (ccflag & CCF_IPHDR_CE) { if (!dctcp_data->ce_prev && (ccflag & CCF_DELACK)) delay_ack = 0; dctcp_data->ce_prev = 1; CCV(ccv, t_flags) |= TF_ECN_SND_ECE; } else { if (dctcp_data->ce_prev && (ccflag & CCF_DELACK)) delay_ack = 0; dctcp_data->ce_prev = 0; CCV(ccv, t_flags) &= ~TF_ECN_SND_ECE; } /* DCTCP sets delayed ack when this segment sets the CWR flag. */ if ((ccflag & CCF_DELACK) && (ccflag & CCF_TCPHDR_CWR)) delay_ack = 1; if (delay_ack == 0) ccv->flags |= CCF_ACKNOW; else ccv->flags &= ~CCF_ACKNOW; } /* * Update the fraction of marked bytes represented as 'alpha'. * Also initialize several internal parameters at the end of this function. */ static void dctcp_update_alpha(struct cc_var *ccv) { struct dctcp *dctcp_data; int alpha_prev; dctcp_data = ccv->cc_data; alpha_prev = dctcp_data->alpha; dctcp_data->bytes_total = max(dctcp_data->bytes_total, 1); /* * Update alpha: alpha = (1 - g) * alpha + g * F. * Here: * g is weight factor * recommaded to be set to 1/16 * small g = slow convergence between competitive DCTCP flows * large g = impacts low utilization of bandwidth at switches * F is fraction of marked segments in last RTT * updated every RTT * Alpha must be round to 0 - MAX_ALPHA_VALUE. */ dctcp_data->alpha = min(alpha_prev - (alpha_prev >> V_dctcp_shift_g) + (dctcp_data->bytes_ecn << (10 - V_dctcp_shift_g)) / dctcp_data->bytes_total, MAX_ALPHA_VALUE); /* Initialize internal parameters for next alpha calculation */ dctcp_data->bytes_ecn = 0; dctcp_data->bytes_total = 0; dctcp_data->save_sndnxt = CCV(ccv, snd_nxt); } static int dctcp_alpha_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_alpha; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else { if (new > MAX_ALPHA_VALUE) V_dctcp_alpha = MAX_ALPHA_VALUE; else V_dctcp_alpha = new; } } return (error); } static int dctcp_shift_g_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_shift_g; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_dctcp_shift_g = new; } return (error); } static int dctcp_slowstart_handler(SYSCTL_HANDLER_ARGS) { uint32_t new; int error; new = V_dctcp_slowstart; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > 1) error = EINVAL; else V_dctcp_slowstart = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_dctcp); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, dctcp, CTLFLAG_RW, NULL, "dctcp congestion control related settings"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, alpha, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_alpha), 0, &dctcp_alpha_handler, "IU", "dctcp alpha parameter"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, shift_g, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_shift_g), 4, &dctcp_shift_g_handler, "IU", "dctcp shift parameter"); SYSCTL_PROC(_net_inet_tcp_cc_dctcp, OID_AUTO, slowstart, CTLFLAG_VNET|CTLTYPE_UINT|CTLFLAG_RW, &VNET_NAME(dctcp_slowstart), 0, &dctcp_slowstart_handler, "IU", "half CWND reduction after the first slow start"); DECLARE_CC_MODULE(dctcp, &dctcp_cc_algo); Index: head/sys/netinet/cc/cc_hd.c =================================================================== --- head/sys/netinet/cc/cc_hd.c (revision 336675) +++ head/sys/netinet/cc/cc_hd.c (revision 336676) @@ -1,256 +1,256 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2010 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2010 Lawrence Stewart * Copyright (c) 2010-2011 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes and * Lawrence Stewart, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, Melbourne, Australia by * David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the Hamilton Institute's delay-based congestion control * algorithm for FreeBSD, based on "A strategy for fair coexistence of loss and * delay-based congestion control algorithms," by L. Budzisz, R. Stanojevic, R. * Shorten, and F. Baker, IEEE Commun. Lett., vol. 13, no. 7, pp. 555--557, Jul. * 2009. * * Originally released as part of the NewTCP research project at Swinburne * University of Technology's Centre for Advanced Internet Architectures, * Melbourne, Australia, which was made possible in part by a grant from the * Cisco University Research Program Fund at Community Foundation Silicon * Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) /* Largest possible number returned by random(). */ #define RANDOM_MAX INT_MAX static void hd_ack_received(struct cc_var *ccv, uint16_t ack_type); static int hd_mod_init(void); static int ertt_id; -static VNET_DEFINE(uint32_t, hd_qthresh) = 20; -static VNET_DEFINE(uint32_t, hd_qmin) = 5; -static VNET_DEFINE(uint32_t, hd_pmax) = 5; +VNET_DEFINE_STATIC(uint32_t, hd_qthresh) = 20; +VNET_DEFINE_STATIC(uint32_t, hd_qmin) = 5; +VNET_DEFINE_STATIC(uint32_t, hd_pmax) = 5; #define V_hd_qthresh VNET(hd_qthresh) #define V_hd_qmin VNET(hd_qmin) #define V_hd_pmax VNET(hd_pmax) struct cc_algo hd_cc_algo = { .name = "hd", .ack_received = hd_ack_received, .mod_init = hd_mod_init }; /* * Hamilton backoff function. Returns 1 if we should backoff or 0 otherwise. */ static __inline int should_backoff(int qdly, int maxqdly) { unsigned long p; if (qdly < V_hd_qthresh) { p = (((RANDOM_MAX / 100) * V_hd_pmax) / (V_hd_qthresh - V_hd_qmin)) * (qdly - V_hd_qmin); } else { if (qdly > V_hd_qthresh) p = (((RANDOM_MAX / 100) * V_hd_pmax) / (maxqdly - V_hd_qthresh)) * (maxqdly - qdly); else p = (RANDOM_MAX / 100) * V_hd_pmax; } return (random() < p); } /* * If the ack type is CC_ACK, and the inferred queueing delay is greater than * the Qmin threshold, cwnd is reduced probabilistically. When backing off due * to delay, HD behaves like NewReno when an ECN signal is received. HD behaves * as NewReno in all other circumstances. */ static void hd_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct ertt *e_t; int qdly; if (ack_type == CC_ACK) { e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); if (e_t->rtt && e_t->minrtt && V_hd_qthresh > 0) { qdly = e_t->rtt - e_t->minrtt; if (qdly > V_hd_qmin && !IN_RECOVERY(CCV(ccv, t_flags))) { /* Probabilistic backoff of cwnd. */ if (should_backoff(qdly, e_t->maxrtt - e_t->minrtt)) { /* * Update cwnd and ssthresh update to * half cwnd and behave like an ECN (ie * not a packet loss). */ newreno_cc_algo.cong_signal(ccv, CC_ECN); return; } } } } newreno_cc_algo.ack_received(ccv, ack_type); /* As for NewReno. */ } static int hd_mod_init(void) { ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) { printf("%s: h_ertt module not found\n", __func__); return (ENOENT); } hd_cc_algo.after_idle = newreno_cc_algo.after_idle; hd_cc_algo.cong_signal = newreno_cc_algo.cong_signal; hd_cc_algo.post_recovery = newreno_cc_algo.post_recovery; return (0); } static int hd_pmax_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_hd_pmax; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100) error = EINVAL; else V_hd_pmax = new; } return (error); } static int hd_qmin_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_hd_qmin; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) > V_hd_qthresh) error = EINVAL; else V_hd_qmin = new; } return (error); } static int hd_qthresh_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_hd_qthresh; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) < 1 || CAST_PTR_INT(req->newptr) < V_hd_qmin) error = EINVAL; else V_hd_qthresh = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_hd); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, hd, CTLFLAG_RW, NULL, "Hamilton delay-based congestion control related settings"); SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_threshold, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qthresh), 20, &hd_qthresh_handler, "IU", "queueing congestion threshold (qth) in ticks"); SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, pmax, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_pmax), 5, &hd_pmax_handler, "IU", "per packet maximum backoff probability as a percentage"); SYSCTL_PROC(_net_inet_tcp_cc_hd, OID_AUTO, queue_min, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(hd_qmin), 5, &hd_qmin_handler, "IU", "minimum queueing delay threshold (qmin) in ticks"); DECLARE_CC_MODULE(hd, &hd_cc_algo); MODULE_DEPEND(hd, ertt, 1, 1, 1); Index: head/sys/netinet/cc/cc_htcp.c =================================================================== --- head/sys/netinet/cc/cc_htcp.c (revision 336675) +++ head/sys/netinet/cc/cc_htcp.c (revision 336676) @@ -1,527 +1,527 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2007-2008 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart and * James Healy, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the H-TCP congestion control algorithm for FreeBSD, * based on the Internet Draft "draft-leith-tcp-htcp-06.txt" by Leith and * Shorten. Originally released as part of the NewTCP research project at * Swinburne University of Technology's Centre for Advanced Internet * Architectures, Melbourne, Australia, which was made possible in part by a * grant from the Cisco University Research Program Fund at Community Foundation * Silicon Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Fixed point math shifts. */ #define HTCP_SHIFT 8 #define HTCP_ALPHA_INC_SHIFT 4 #define HTCP_INIT_ALPHA 1 #define HTCP_DELTA_L hz /* 1 sec in ticks. */ #define HTCP_MINBETA 128 /* 0.5 << HTCP_SHIFT. */ #define HTCP_MAXBETA 204 /* ~0.8 << HTCP_SHIFT. */ #define HTCP_MINROWE 26 /* ~0.1 << HTCP_SHIFT. */ #define HTCP_MAXROWE 512 /* 2 << HTCP_SHIFT. */ /* RTT_ref (ms) used in the calculation of alpha if RTT scaling is enabled. */ #define HTCP_RTT_REF 100 /* Don't trust SRTT until this many samples have been taken. */ #define HTCP_MIN_RTT_SAMPLES 8 /* * HTCP_CALC_ALPHA performs a fixed point math calculation to determine the * value of alpha, based on the function defined in the HTCP spec. * * i.e. 1 + 10(delta - delta_l) + ((delta - delta_l) / 2) ^ 2 * * "diff" is passed in to the macro as "delta - delta_l" and is expected to be * in units of ticks. * * The joyousnous of fixed point maths means our function implementation looks a * little funky... * * In order to maintain some precision in the calculations, a fixed point shift * HTCP_ALPHA_INC_SHIFT is used to ensure the integer divisions don't * truncate the results too badly. * * The "16" value is the "1" term in the alpha function shifted up by * HTCP_ALPHA_INC_SHIFT * * The "160" value is the "10" multiplier in the alpha function multiplied by * 2^HTCP_ALPHA_INC_SHIFT * * Specifying these as constants reduces the computations required. After * up-shifting all the terms in the function and performing the required * calculations, we down-shift the final result by HTCP_ALPHA_INC_SHIFT to * ensure it is back in the correct range. * * The "hz" terms are required as kernels can be configured to run with * different tick timers, which we have to adjust for in the alpha calculation * (which originally was defined in terms of seconds). * * We also have to be careful to constrain the value of diff such that it won't * overflow whilst performing the calculation. The middle term i.e. (160 * diff) * / hz is the limiting factor in the calculation. We must constrain diff to be * less than the max size of an int divided by the constant 160 figure * i.e. diff < INT_MAX / 160 * * NB: Changing HTCP_ALPHA_INC_SHIFT will require you to MANUALLY update the * constants used in this function! */ #define HTCP_CALC_ALPHA(diff) \ ((\ (16) + \ ((160 * (diff)) / hz) + \ (((diff) / hz) * (((diff) << HTCP_ALPHA_INC_SHIFT) / (4 * hz))) \ ) >> HTCP_ALPHA_INC_SHIFT) static void htcp_ack_received(struct cc_var *ccv, uint16_t type); static void htcp_cb_destroy(struct cc_var *ccv); static int htcp_cb_init(struct cc_var *ccv); static void htcp_cong_signal(struct cc_var *ccv, uint32_t type); static int htcp_mod_init(void); static void htcp_post_recovery(struct cc_var *ccv); static void htcp_recalc_alpha(struct cc_var *ccv); static void htcp_recalc_beta(struct cc_var *ccv); static void htcp_record_rtt(struct cc_var *ccv); static void htcp_ssthresh_update(struct cc_var *ccv); struct htcp { /* cwnd before entering cong recovery. */ unsigned long prev_cwnd; /* cwnd additive increase parameter. */ int alpha; /* cwnd multiplicative decrease parameter. */ int beta; /* Largest rtt seen for the flow. */ int maxrtt; /* Shortest rtt seen for the flow. */ int minrtt; /* Time of last congestion event in ticks. */ int t_last_cong; }; static int htcp_rtt_ref; /* * The maximum number of ticks the value of diff can reach in * htcp_recalc_alpha() before alpha will stop increasing due to overflow. * See comment above HTCP_CALC_ALPHA for more info. */ static int htcp_max_diff = INT_MAX / ((1 << HTCP_ALPHA_INC_SHIFT) * 10); /* Per-netstack vars. */ -static VNET_DEFINE(u_int, htcp_adaptive_backoff) = 0; -static VNET_DEFINE(u_int, htcp_rtt_scaling) = 0; +VNET_DEFINE_STATIC(u_int, htcp_adaptive_backoff) = 0; +VNET_DEFINE_STATIC(u_int, htcp_rtt_scaling) = 0; #define V_htcp_adaptive_backoff VNET(htcp_adaptive_backoff) #define V_htcp_rtt_scaling VNET(htcp_rtt_scaling) static MALLOC_DEFINE(M_HTCP, "htcp data", "Per connection data required for the HTCP congestion control algorithm"); struct cc_algo htcp_cc_algo = { .name = "htcp", .ack_received = htcp_ack_received, .cb_destroy = htcp_cb_destroy, .cb_init = htcp_cb_init, .cong_signal = htcp_cong_signal, .mod_init = htcp_mod_init, .post_recovery = htcp_post_recovery, }; static void htcp_ack_received(struct cc_var *ccv, uint16_t type) { struct htcp *htcp_data; htcp_data = ccv->cc_data; htcp_record_rtt(ccv); /* * Regular ACK and we're not in cong/fast recovery and we're cwnd * limited and we're either not doing ABC or are slow starting or are * doing ABC and we've sent a cwnd's worth of bytes. */ if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && (ccv->flags & CCF_CWND_LIMITED) && (!V_tcp_do_rfc3465 || CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh) || (V_tcp_do_rfc3465 && ccv->flags & CCF_ABC_SENTAWND))) { htcp_recalc_beta(ccv); htcp_recalc_alpha(ccv); /* * Use the logic in NewReno ack_received() for slow start and * for the first HTCP_DELTA_L ticks after either the flow starts * or a congestion event (when alpha equals 1). */ if (htcp_data->alpha == 1 || CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) newreno_cc_algo.ack_received(ccv, type); else { if (V_tcp_do_rfc3465) { /* Increment cwnd by alpha segments. */ CCV(ccv, snd_cwnd) += htcp_data->alpha * CCV(ccv, t_maxseg); ccv->flags &= ~CCF_ABC_SENTAWND; } else /* * Increment cwnd by alpha/cwnd segments to * approximate an increase of alpha segments * per RTT. */ CCV(ccv, snd_cwnd) += (((htcp_data->alpha << HTCP_SHIFT) / (CCV(ccv, snd_cwnd) / CCV(ccv, t_maxseg))) * CCV(ccv, t_maxseg)) >> HTCP_SHIFT; } } } static void htcp_cb_destroy(struct cc_var *ccv) { free(ccv->cc_data, M_HTCP); } static int htcp_cb_init(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = malloc(sizeof(struct htcp), M_HTCP, M_NOWAIT); if (htcp_data == NULL) return (ENOMEM); /* Init some key variables with sensible defaults. */ htcp_data->alpha = HTCP_INIT_ALPHA; htcp_data->beta = HTCP_MINBETA; htcp_data->maxrtt = TCPTV_SRTTBASE; htcp_data->minrtt = TCPTV_SRTTBASE; htcp_data->prev_cwnd = 0; htcp_data->t_last_cong = ticks; ccv->cc_data = htcp_data; return (0); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void htcp_cong_signal(struct cc_var *ccv, uint32_t type) { struct htcp *htcp_data; htcp_data = ccv->cc_data; switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { /* * Apply hysteresis to maxrtt to ensure * reductions in the RTT are reflected in our * measurements. */ htcp_data->maxrtt = (htcp_data->minrtt + (htcp_data->maxrtt - htcp_data->minrtt) * 95) / 100; htcp_ssthresh_update(ccv); htcp_data->t_last_cong = ticks; htcp_data->prev_cwnd = CCV(ccv, snd_cwnd); } ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { /* * Apply hysteresis to maxrtt to ensure reductions in * the RTT are reflected in our measurements. */ htcp_data->maxrtt = (htcp_data->minrtt + (htcp_data->maxrtt - htcp_data->minrtt) * 95) / 100; htcp_ssthresh_update(ccv); CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); htcp_data->t_last_cong = ticks; htcp_data->prev_cwnd = CCV(ccv, snd_cwnd); ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } break; case CC_RTO: /* * Grab the current time and record it so we know when the * most recent congestion event was. Only record it when the * timeout has fired more than once, as there is a reasonable * chance the first one is a false alarm and may not indicate * congestion. */ if (CCV(ccv, t_rxtshift) >= 2) htcp_data->t_last_cong = ticks; break; } } static int htcp_mod_init(void) { htcp_cc_algo.after_idle = newreno_cc_algo.after_idle; /* * HTCP_RTT_REF is defined in ms, and t_srtt in the tcpcb is stored in * units of TCP_RTT_SCALE*hz. Scale HTCP_RTT_REF to be in the same units * as t_srtt. */ htcp_rtt_ref = (HTCP_RTT_REF * TCP_RTT_SCALE * hz) / 1000; return (0); } /* * Perform any necessary tasks before we exit congestion recovery. */ static void htcp_post_recovery(struct cc_var *ccv) { int pipe; struct htcp *htcp_data; pipe = 0; htcp_data = ccv->cc_data; if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { /* * If inflight data is less than ssthresh, set cwnd * conservatively to avoid a burst of data, as suggested in the * NewReno RFC. Otherwise, use the HTCP method. * * XXXLAS: Find a way to do this without needing curack */ if (V_tcp_do_rfc6675_pipe) pipe = tcp_compute_pipe(ccv->ccvc.tcp); else pipe = CCV(ccv, snd_max) - ccv->curack; if (pipe < CCV(ccv, snd_ssthresh)) CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg); else CCV(ccv, snd_cwnd) = max(1, ((htcp_data->beta * htcp_data->prev_cwnd / CCV(ccv, t_maxseg)) >> HTCP_SHIFT)) * CCV(ccv, t_maxseg); } } static void htcp_recalc_alpha(struct cc_var *ccv) { struct htcp *htcp_data; int alpha, diff, now; htcp_data = ccv->cc_data; now = ticks; /* * If ticks has wrapped around (will happen approximately once every 49 * days on a machine with the default kern.hz=1000) and a flow straddles * the wrap point, our alpha calcs will be completely wrong. We cut our * losses and restart alpha from scratch by setting t_last_cong = now - * HTCP_DELTA_L. * * This does not deflate our cwnd at all. It simply slows the rate cwnd * is growing by until alpha regains the value it held prior to taking * this drastic measure. */ if (now < htcp_data->t_last_cong) htcp_data->t_last_cong = now - HTCP_DELTA_L; diff = now - htcp_data->t_last_cong - HTCP_DELTA_L; /* Cap alpha if the value of diff would overflow HTCP_CALC_ALPHA(). */ if (diff < htcp_max_diff) { /* * If it has been more than HTCP_DELTA_L ticks since congestion, * increase alpha according to the function defined in the spec. */ if (diff > 0) { alpha = HTCP_CALC_ALPHA(diff); /* * Adaptive backoff fairness adjustment: * 2 * (1 - beta) * alpha_raw */ if (V_htcp_adaptive_backoff) alpha = max(1, (2 * ((1 << HTCP_SHIFT) - htcp_data->beta) * alpha) >> HTCP_SHIFT); /* * RTT scaling: (RTT / RTT_ref) * alpha * alpha will be the raw value from HTCP_CALC_ALPHA() if * adaptive backoff is off, or the adjusted value if * adaptive backoff is on. */ if (V_htcp_rtt_scaling) alpha = max(1, (min(max(HTCP_MINROWE, (CCV(ccv, t_srtt) << HTCP_SHIFT) / htcp_rtt_ref), HTCP_MAXROWE) * alpha) >> HTCP_SHIFT); } else alpha = 1; htcp_data->alpha = alpha; } } static void htcp_recalc_beta(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* * TCPTV_SRTTBASE is the initialised value of each connection's SRTT, so * we only calc beta if the connection's SRTT has been changed from its * initial value. beta is bounded to ensure it is always between * HTCP_MINBETA and HTCP_MAXBETA. */ if (V_htcp_adaptive_backoff && htcp_data->minrtt != TCPTV_SRTTBASE && htcp_data->maxrtt != TCPTV_SRTTBASE) htcp_data->beta = min(max(HTCP_MINBETA, (htcp_data->minrtt << HTCP_SHIFT) / htcp_data->maxrtt), HTCP_MAXBETA); else htcp_data->beta = HTCP_MINBETA; } /* * Record the minimum and maximum RTT seen for the connection. These are used in * the calculation of beta if adaptive backoff is enabled. */ static void htcp_record_rtt(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* XXXLAS: Should there be some hysteresis for minrtt? */ /* * Record the current SRTT as our minrtt if it's the smallest we've seen * or minrtt is currently equal to its initialised value. Ignore SRTT * until a min number of samples have been taken. */ if ((CCV(ccv, t_srtt) < htcp_data->minrtt || htcp_data->minrtt == TCPTV_SRTTBASE) && (CCV(ccv, t_rttupdated) >= HTCP_MIN_RTT_SAMPLES)) htcp_data->minrtt = CCV(ccv, t_srtt); /* * Record the current SRTT as our maxrtt if it's the largest we've * seen. Ignore SRTT until a min number of samples have been taken. */ if (CCV(ccv, t_srtt) > htcp_data->maxrtt && CCV(ccv, t_rttupdated) >= HTCP_MIN_RTT_SAMPLES) htcp_data->maxrtt = CCV(ccv, t_srtt); } /* * Update the ssthresh in the event of congestion. */ static void htcp_ssthresh_update(struct cc_var *ccv) { struct htcp *htcp_data; htcp_data = ccv->cc_data; /* * On the first congestion event, set ssthresh to cwnd * 0.5, on * subsequent congestion events, set it to cwnd * beta. */ if (CCV(ccv, snd_ssthresh) == TCP_MAXWIN << TCP_MAX_WINSHIFT) CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) * HTCP_MINBETA) >> HTCP_SHIFT; else { htcp_recalc_beta(ccv); CCV(ccv, snd_ssthresh) = ((u_long)CCV(ccv, snd_cwnd) * htcp_data->beta) >> HTCP_SHIFT; } } SYSCTL_DECL(_net_inet_tcp_cc_htcp); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, htcp, CTLFLAG_RW, NULL, "H-TCP related settings"); SYSCTL_UINT(_net_inet_tcp_cc_htcp, OID_AUTO, adaptive_backoff, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(htcp_adaptive_backoff), 0, "enable H-TCP adaptive backoff"); SYSCTL_UINT(_net_inet_tcp_cc_htcp, OID_AUTO, rtt_scaling, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(htcp_rtt_scaling), 0, "enable H-TCP RTT scaling"); DECLARE_CC_MODULE(htcp, &htcp_cc_algo); Index: head/sys/netinet/cc/cc_newreno.c =================================================================== --- head/sys/netinet/cc/cc_newreno.c (revision 336675) +++ head/sys/netinet/cc/cc_newreno.c (revision 336676) @@ -1,392 +1,392 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. * Copyright (c) 2007-2008,2010,2014 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, James * Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This software was first released in 2007 by James Healy and Lawrence Stewart * whilst working on the NewTCP research project at Swinburne University of * Technology's Centre for Advanced Internet Architectures, Melbourne, * Australia, which was made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * More details are available at: * http://caia.swin.edu.au/urp/newtcp/ * * Dec 2014 garmitage@swin.edu.au * Borrowed code fragments from cc_cdg.c to add modifiable beta * via sysctls. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_NEWRENO, "newreno data", "newreno beta values"); #define CAST_PTR_INT(X) (*((int*)(X))) static void newreno_cb_destroy(struct cc_var *ccv); static void newreno_ack_received(struct cc_var *ccv, uint16_t type); static void newreno_after_idle(struct cc_var *ccv); static void newreno_cong_signal(struct cc_var *ccv, uint32_t type); static void newreno_post_recovery(struct cc_var *ccv); static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf); -static VNET_DEFINE(uint32_t, newreno_beta) = 50; -static VNET_DEFINE(uint32_t, newreno_beta_ecn) = 80; +VNET_DEFINE_STATIC(uint32_t, newreno_beta) = 50; +VNET_DEFINE_STATIC(uint32_t, newreno_beta_ecn) = 80; #define V_newreno_beta VNET(newreno_beta) #define V_newreno_beta_ecn VNET(newreno_beta_ecn) struct cc_algo newreno_cc_algo = { .name = "newreno", .cb_destroy = newreno_cb_destroy, .ack_received = newreno_ack_received, .after_idle = newreno_after_idle, .cong_signal = newreno_cong_signal, .post_recovery = newreno_post_recovery, .ctl_output = newreno_ctl_output, }; struct newreno { uint32_t beta; uint32_t beta_ecn; }; static inline struct newreno * newreno_malloc(struct cc_var *ccv) { struct newreno *nreno; nreno = malloc(sizeof(struct newreno), M_NEWRENO, M_NOWAIT); if (nreno != NULL) { /* NB: nreno is not zeroed, so initialise all fields. */ nreno->beta = V_newreno_beta; nreno->beta_ecn = V_newreno_beta_ecn; ccv->cc_data = nreno; } return (nreno); } static void newreno_cb_destroy(struct cc_var *ccv) { free(ccv->cc_data, M_NEWRENO); } static void newreno_ack_received(struct cc_var *ccv, uint16_t type) { if (type == CC_ACK && !IN_RECOVERY(CCV(ccv, t_flags)) && (ccv->flags & CCF_CWND_LIMITED)) { u_int cw = CCV(ccv, snd_cwnd); u_int incr = CCV(ccv, t_maxseg); /* * Regular in-order ACK, open the congestion window. * Method depends on which congestion control state we're * in (slow start or cong avoid) and if ABC (RFC 3465) is * enabled. * * slow start: cwnd <= ssthresh * cong avoid: cwnd > ssthresh * * slow start and ABC (RFC 3465): * Grow cwnd exponentially by the amount of data * ACKed capping the max increment per ACK to * (abc_l_var * maxseg) bytes. * * slow start without ABC (RFC 5681): * Grow cwnd exponentially by maxseg per ACK. * * cong avoid and ABC (RFC 3465): * Grow cwnd linearly by maxseg per RTT for each * cwnd worth of ACKed data. * * cong avoid without ABC (RFC 5681): * Grow cwnd linearly by approximately maxseg per RTT using * maxseg^2 / cwnd per ACK as the increment. * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to * avoid capping cwnd. */ if (cw > CCV(ccv, snd_ssthresh)) { if (V_tcp_do_rfc3465) { if (ccv->flags & CCF_ABC_SENTAWND) ccv->flags &= ~CCF_ABC_SENTAWND; else incr = 0; } else incr = max((incr * incr / cw), 1); } else if (V_tcp_do_rfc3465) { /* * In slow-start with ABC enabled and no RTO in sight? * (Must not use abc_l_var > 1 if slow starting after * an RTO. On RTO, snd_nxt = snd_una, so the * snd_nxt == snd_max check is sufficient to * handle this). * * XXXLAS: Find a way to signal SS after RTO that * doesn't rely on tcpcb vars. */ if (CCV(ccv, snd_nxt) == CCV(ccv, snd_max)) incr = min(ccv->bytes_this_ack, ccv->nsegs * V_tcp_abc_l_var * CCV(ccv, t_maxseg)); else incr = min(ccv->bytes_this_ack, CCV(ccv, t_maxseg)); } /* ABC is on by default, so incr equals 0 frequently. */ if (incr > 0) CCV(ccv, snd_cwnd) = min(cw + incr, TCP_MAXWIN << CCV(ccv, snd_scale)); } } static void newreno_after_idle(struct cc_var *ccv) { int rw; /* * If we've been idle for more than one retransmit timeout the old * congestion window is no longer current and we have to reduce it to * the restart window before we can transmit again. * * The restart window is the initial window or the last CWND, whichever * is smaller. * * This is done to prevent us from flooding the path with a full CWND at * wirespeed, overloading router and switch buffers along the way. * * See RFC5681 Section 4.1. "Restarting Idle Connections". */ if (V_tcp_do_rfc3390) rw = min(4 * CCV(ccv, t_maxseg), max(2 * CCV(ccv, t_maxseg), 4380)); else rw = CCV(ccv, t_maxseg) * 2; CCV(ccv, snd_cwnd) = min(rw, CCV(ccv, snd_cwnd)); } /* * Perform any necessary tasks before we enter congestion recovery. */ static void newreno_cong_signal(struct cc_var *ccv, uint32_t type) { struct newreno *nreno; uint32_t beta, beta_ecn, cwin, factor; u_int mss; cwin = CCV(ccv, snd_cwnd); mss = CCV(ccv, t_maxseg); nreno = ccv->cc_data; beta = (nreno == NULL) ? V_newreno_beta : nreno->beta; beta_ecn = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn; if (V_cc_do_abe && type == CC_ECN) factor = beta_ecn; else factor = beta; /* Catch algos which mistakenly leak private signal types. */ KASSERT((type & CC_SIGPRIVMASK) == 0, ("%s: congestion signal type 0x%08x is private\n", __func__, type)); cwin = max(((uint64_t)cwin * (uint64_t)factor) / (100ULL * (uint64_t)mss), 2) * mss; switch (type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(CCV(ccv, t_flags))) { if (IN_CONGRECOVERY(CCV(ccv, t_flags) && V_cc_do_abe && V_cc_abe_frlossreduce)) { CCV(ccv, snd_ssthresh) = ((uint64_t)CCV(ccv, snd_ssthresh) * (uint64_t)beta) / (100ULL * (uint64_t)beta_ecn); } if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) CCV(ccv, snd_ssthresh) = cwin; ENTER_RECOVERY(CCV(ccv, t_flags)); } break; case CC_ECN: if (!IN_CONGRECOVERY(CCV(ccv, t_flags))) { CCV(ccv, snd_ssthresh) = cwin; CCV(ccv, snd_cwnd) = cwin; ENTER_CONGRECOVERY(CCV(ccv, t_flags)); } break; } } /* * Perform any necessary tasks before we exit congestion recovery. */ static void newreno_post_recovery(struct cc_var *ccv) { int pipe; if (IN_FASTRECOVERY(CCV(ccv, t_flags))) { /* * Fast recovery will conclude after returning from this * function. Window inflation should have left us with * approximately snd_ssthresh outstanding data. But in case we * would be inclined to send a burst, better to do it via the * slow start mechanism. * * XXXLAS: Find a way to do this without needing curack */ if (V_tcp_do_rfc6675_pipe) pipe = tcp_compute_pipe(ccv->ccvc.tcp); else pipe = CCV(ccv, snd_max) - ccv->curack; if (pipe < CCV(ccv, snd_ssthresh)) CCV(ccv, snd_cwnd) = pipe + CCV(ccv, t_maxseg); else CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); } } static int newreno_ctl_output(struct cc_var *ccv, struct sockopt *sopt, void *buf) { struct newreno *nreno; struct cc_newreno_opts *opt; if (sopt->sopt_valsize != sizeof(struct cc_newreno_opts)) return (EMSGSIZE); nreno = ccv->cc_data; opt = buf; switch (sopt->sopt_dir) { case SOPT_SET: /* We cannot set without cc_data memory. */ if (nreno == NULL) { nreno = newreno_malloc(ccv); if (nreno == NULL) return (ENOMEM); } switch (opt->name) { case CC_NEWRENO_BETA: nreno->beta = opt->val; break; case CC_NEWRENO_BETA_ECN: if (!V_cc_do_abe) return (EACCES); nreno->beta_ecn = opt->val; break; default: return (ENOPROTOOPT); } break; case SOPT_GET: switch (opt->name) { case CC_NEWRENO_BETA: opt->val = (nreno == NULL) ? V_newreno_beta : nreno->beta; break; case CC_NEWRENO_BETA_ECN: opt->val = (nreno == NULL) ? V_newreno_beta_ecn : nreno->beta_ecn; break; default: return (ENOPROTOOPT); } break; default: return (EINVAL); } return (0); } static int newreno_beta_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr != NULL ) { if (arg1 == &VNET_NAME(newreno_beta_ecn) && !V_cc_do_abe) return (EACCES); if (CAST_PTR_INT(req->newptr) <= 0 || CAST_PTR_INT(req->newptr) > 100) return (EINVAL); } return (sysctl_handle_int(oidp, arg1, arg2, req)); } SYSCTL_DECL(_net_inet_tcp_cc_newreno); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, newreno, CTLFLAG_RW, NULL, "New Reno related settings"); SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(newreno_beta), 3, &newreno_beta_handler, "IU", "New Reno beta, specified as number between 1 and 100"); SYSCTL_PROC(_net_inet_tcp_cc_newreno, OID_AUTO, beta_ecn, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(newreno_beta_ecn), 3, &newreno_beta_handler, "IU", "New Reno beta ecn, specified as number between 1 and 100"); DECLARE_CC_MODULE(newreno, &newreno_cc_algo); Index: head/sys/netinet/cc/cc_vegas.c =================================================================== --- head/sys/netinet/cc/cc_vegas.c (revision 336675) +++ head/sys/netinet/cc/cc_vegas.c (revision 336676) @@ -1,307 +1,307 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2010 * Swinburne University of Technology, Melbourne, Australia * Copyright (c) 2010 Lawrence Stewart * Copyright (c) 2010-2011 The FreeBSD Foundation * All rights reserved. * * This software was developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by David Hayes and * Lawrence Stewart, made possible in part by a grant from the Cisco University * Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, Melbourne, Australia by * David Hayes under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * An implementation of the Vegas congestion control algorithm for FreeBSD, * based on L. S. Brakmo and L. L. Peterson, "TCP Vegas: end to end congestion * avoidance on a global internet", IEEE J. Sel. Areas Commun., vol. 13, no. 8, * pp. 1465-1480, Oct. 1995. The original Vegas duplicate ack policy has not * been implemented, since clock ticks are not as coarse as they were (i.e. * 500ms) when Vegas was designed. Also, packets are timed once per RTT as in * the original paper. * * Originally released as part of the NewTCP research project at Swinburne * University of Technology's Centre for Advanced Internet Architectures, * Melbourne, Australia, which was made possible in part by a grant from the * Cisco University Research Program Fund at Community Foundation Silicon * Valley. More details are available at: * http://caia.swin.edu.au/urp/newtcp/ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define CAST_PTR_INT(X) (*((int*)(X))) /* * Private signal type for rate based congestion signal. * See for appropriate bit-range to use for private signals. */ #define CC_VEGAS_RATE 0x01000000 static void vegas_ack_received(struct cc_var *ccv, uint16_t ack_type); static void vegas_cb_destroy(struct cc_var *ccv); static int vegas_cb_init(struct cc_var *ccv); static void vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type); static void vegas_conn_init(struct cc_var *ccv); static int vegas_mod_init(void); struct vegas { int slow_start_toggle; }; static int32_t ertt_id; -static VNET_DEFINE(uint32_t, vegas_alpha) = 1; -static VNET_DEFINE(uint32_t, vegas_beta) = 3; +VNET_DEFINE_STATIC(uint32_t, vegas_alpha) = 1; +VNET_DEFINE_STATIC(uint32_t, vegas_beta) = 3; #define V_vegas_alpha VNET(vegas_alpha) #define V_vegas_beta VNET(vegas_beta) static MALLOC_DEFINE(M_VEGAS, "vegas data", "Per connection data required for the Vegas congestion control algorithm"); struct cc_algo vegas_cc_algo = { .name = "vegas", .ack_received = vegas_ack_received, .cb_destroy = vegas_cb_destroy, .cb_init = vegas_cb_init, .cong_signal = vegas_cong_signal, .conn_init = vegas_conn_init, .mod_init = vegas_mod_init }; /* * The vegas window adjustment is done once every RTT, as indicated by the * ERTT_NEW_MEASUREMENT flag. This flag is reset once the new measurment data * has been used. */ static void vegas_ack_received(struct cc_var *ccv, uint16_t ack_type) { struct ertt *e_t; struct vegas *vegas_data; long actual_tx_rate, expected_tx_rate, ndiff; e_t = khelp_get_osd(CCV(ccv, osd), ertt_id); vegas_data = ccv->cc_data; if (e_t->flags & ERTT_NEW_MEASUREMENT) { /* Once per RTT. */ if (e_t->minrtt && e_t->markedpkt_rtt) { expected_tx_rate = e_t->marked_snd_cwnd / e_t->minrtt; actual_tx_rate = e_t->bytes_tx_in_marked_rtt / e_t->markedpkt_rtt; ndiff = (expected_tx_rate - actual_tx_rate) * e_t->minrtt / CCV(ccv, t_maxseg); if (ndiff < V_vegas_alpha) { if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { vegas_data->slow_start_toggle = vegas_data->slow_start_toggle ? 0 : 1; } else { vegas_data->slow_start_toggle = 0; CCV(ccv, snd_cwnd) = min(CCV(ccv, snd_cwnd) + CCV(ccv, t_maxseg), TCP_MAXWIN << CCV(ccv, snd_scale)); } } else if (ndiff > V_vegas_beta) { /* Rate-based congestion. */ vegas_cong_signal(ccv, CC_VEGAS_RATE); vegas_data->slow_start_toggle = 0; } } e_t->flags &= ~ERTT_NEW_MEASUREMENT; } if (vegas_data->slow_start_toggle) newreno_cc_algo.ack_received(ccv, ack_type); } static void vegas_cb_destroy(struct cc_var *ccv) { free(ccv->cc_data, M_VEGAS); } static int vegas_cb_init(struct cc_var *ccv) { struct vegas *vegas_data; vegas_data = malloc(sizeof(struct vegas), M_VEGAS, M_NOWAIT); if (vegas_data == NULL) return (ENOMEM); vegas_data->slow_start_toggle = 1; ccv->cc_data = vegas_data; return (0); } /* * If congestion has been triggered triggered by the Vegas measured rates, it is * handled here, otherwise it falls back to newreno's congestion handling. */ static void vegas_cong_signal(struct cc_var *ccv, uint32_t signal_type) { struct vegas *vegas_data; int presignalrecov; vegas_data = ccv->cc_data; if (IN_RECOVERY(CCV(ccv, t_flags))) presignalrecov = 1; else presignalrecov = 0; switch(signal_type) { case CC_VEGAS_RATE: if (!IN_RECOVERY(CCV(ccv, t_flags))) { CCV(ccv, snd_cwnd) = max(2 * CCV(ccv, t_maxseg), CCV(ccv, snd_cwnd) - CCV(ccv, t_maxseg)); if (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) /* Exit slow start. */ CCV(ccv, snd_ssthresh) = CCV(ccv, snd_cwnd); } break; default: newreno_cc_algo.cong_signal(ccv, signal_type); } if (IN_RECOVERY(CCV(ccv, t_flags)) && !presignalrecov) vegas_data->slow_start_toggle = (CCV(ccv, snd_cwnd) < CCV(ccv, snd_ssthresh)) ? 1 : 0; } static void vegas_conn_init(struct cc_var *ccv) { struct vegas *vegas_data; vegas_data = ccv->cc_data; vegas_data->slow_start_toggle = 1; } static int vegas_mod_init(void) { ertt_id = khelp_get_id("ertt"); if (ertt_id <= 0) { printf("%s: h_ertt module not found\n", __func__); return (ENOENT); } vegas_cc_algo.after_idle = newreno_cc_algo.after_idle; vegas_cc_algo.post_recovery = newreno_cc_algo.post_recovery; return (0); } static int vegas_alpha_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_vegas_alpha; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) < 1 || CAST_PTR_INT(req->newptr) > V_vegas_beta) error = EINVAL; else V_vegas_alpha = new; } return (error); } static int vegas_beta_handler(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_vegas_beta; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr != NULL) { if (CAST_PTR_INT(req->newptr) < 1 || CAST_PTR_INT(req->newptr) < V_vegas_alpha) error = EINVAL; else V_vegas_beta = new; } return (error); } SYSCTL_DECL(_net_inet_tcp_cc_vegas); SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, vegas, CTLFLAG_RW, NULL, "Vegas related settings"); SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, alpha, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(vegas_alpha), 1, &vegas_alpha_handler, "IU", "vegas alpha, specified as number of \"buffers\" (0 < alpha < beta)"); SYSCTL_PROC(_net_inet_tcp_cc_vegas, OID_AUTO, beta, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(vegas_beta), 3, &vegas_beta_handler, "IU", "vegas beta, specified as number of \"buffers\" (0 < alpha < beta)"); DECLARE_CC_MODULE(vegas, &vegas_cc_algo); MODULE_DEPEND(vegas, ertt, 1, 1, 1); Index: head/sys/netinet/if_ether.c =================================================================== --- head/sys/netinet/if_ether.c (revision 336675) +++ head/sys/netinet/if_ether.c (revision 336676) @@ -1,1497 +1,1497 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)if_ether.c 8.1 (Berkeley) 6/10/93 */ /* * Ethernet address resolution protocol. * TODO: * add "inuse/lock" bit (or ref. count) along with valid bit */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #endif #include #define SIN(s) ((const struct sockaddr_in *)(s)) static struct timeval arp_lastlog; static int arp_curpps; static int arp_maxpps = 1; /* Simple ARP state machine */ enum arp_llinfo_state { ARP_LLINFO_INCOMPLETE = 0, /* No LLE data */ ARP_LLINFO_REACHABLE, /* LLE is valid */ ARP_LLINFO_VERIFY, /* LLE is valid, need refresh */ ARP_LLINFO_DELETED, /* LLE is deleted */ }; SYSCTL_DECL(_net_link_ether); static SYSCTL_NODE(_net_link_ether, PF_INET, inet, CTLFLAG_RW, 0, ""); static SYSCTL_NODE(_net_link_ether, PF_ARP, arp, CTLFLAG_RW, 0, ""); /* timer values */ -static VNET_DEFINE(int, arpt_keep) = (20*60); /* once resolved, good for 20 +VNET_DEFINE_STATIC(int, arpt_keep) = (20*60); /* once resolved, good for 20 * minutes */ -static VNET_DEFINE(int, arp_maxtries) = 5; -static VNET_DEFINE(int, arp_proxyall) = 0; -static VNET_DEFINE(int, arpt_down) = 20; /* keep incomplete entries for +VNET_DEFINE_STATIC(int, arp_maxtries) = 5; +VNET_DEFINE_STATIC(int, arp_proxyall) = 0; +VNET_DEFINE_STATIC(int, arpt_down) = 20; /* keep incomplete entries for * 20 seconds */ -static VNET_DEFINE(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/ +VNET_DEFINE_STATIC(int, arpt_rexmit) = 1; /* retransmit arp entries, sec*/ VNET_PCPUSTAT_DEFINE(struct arpstat, arpstat); /* ARP statistics, see if_arp.h */ VNET_PCPUSTAT_SYSINIT(arpstat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(arpstat); #endif /* VIMAGE */ -static VNET_DEFINE(int, arp_maxhold) = 1; +VNET_DEFINE_STATIC(int, arp_maxhold) = 1; #define V_arpt_keep VNET(arpt_keep) #define V_arpt_down VNET(arpt_down) #define V_arpt_rexmit VNET(arpt_rexmit) #define V_arp_maxtries VNET(arp_maxtries) #define V_arp_proxyall VNET(arp_proxyall) #define V_arp_maxhold VNET(arp_maxhold) SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_age, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(arpt_keep), 0, "ARP entry lifetime in seconds"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxtries, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(arp_maxtries), 0, "ARP resolution attempts before returning error"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, proxyall, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(arp_proxyall), 0, "Enable proxy ARP for all suitable requests"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, wait, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(arpt_down), 0, "Incomplete ARP entry lifetime in seconds"); SYSCTL_VNET_PCPUSTAT(_net_link_ether_arp, OID_AUTO, stats, struct arpstat, arpstat, "ARP statistics (struct arpstat, net/if_arp.h)"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, maxhold, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(arp_maxhold), 0, "Number of packets to hold per ARP entry"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, max_log_per_second, CTLFLAG_RW, &arp_maxpps, 0, "Maximum number of remotely triggered ARP messages that can be " "logged per second"); /* * Due to the exponential backoff algorithm used for the interval between GARP * retransmissions, the maximum number of retransmissions is limited for * sanity. This limit corresponds to a maximum interval between retransmissions * of 2^16 seconds ~= 18 hours. * * Making this limit more dynamic is more complicated than worthwhile, * especially since sending out GARPs spaced days apart would be of little * use. A maximum dynamic limit would look something like: * * const int max = fls(INT_MAX / hz) - 1; */ #define MAX_GARP_RETRANSMITS 16 static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS); static int garp_rexmit_count = 0; /* GARP retransmission setting. */ SYSCTL_PROC(_net_link_ether_inet, OID_AUTO, garp_rexmit_count, CTLTYPE_INT|CTLFLAG_RW|CTLFLAG_MPSAFE, &garp_rexmit_count, 0, sysctl_garp_rexmit, "I", "Number of times to retransmit GARP packets;" " 0 to disable, maximum of 16"); #define ARP_LOG(pri, ...) do { \ if (ppsratecheck(&arp_lastlog, &arp_curpps, arp_maxpps)) \ log((pri), "arp: " __VA_ARGS__); \ } while (0) static void arpintr(struct mbuf *); static void arptimer(void *); #ifdef INET static void in_arpinput(struct mbuf *); #endif static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp, int bridged, struct llentry *la); static void arp_mark_lle_reachable(struct llentry *la); static void arp_iflladdr(void *arg __unused, struct ifnet *ifp); static eventhandler_tag iflladdr_tag; static const struct netisr_handler arp_nh = { .nh_name = "arp", .nh_handler = arpintr, .nh_proto = NETISR_ARP, .nh_policy = NETISR_POLICY_SOURCE, }; /* * Timeout routine. Age arp_tab entries periodically. */ static void arptimer(void *arg) { struct llentry *lle = (struct llentry *)arg; struct ifnet *ifp; int r_skip_req; if (lle->la_flags & LLE_STATIC) { return; } LLE_WLOCK(lle); if (callout_pending(&lle->lle_timer)) { /* * Here we are a bit odd here in the treatment of * active/pending. If the pending bit is set, it got * rescheduled before I ran. The active * bit we ignore, since if it was stopped * in ll_tablefree() and was currently running * it would have return 0 so the code would * not have deleted it since the callout could * not be stopped so we want to go through * with the delete here now. If the callout * was restarted, the pending bit will be back on and * we just want to bail since the callout_reset would * return 1 and our reference would have been removed * by arpresolve() below. */ LLE_WUNLOCK(lle); return; } ifp = lle->lle_tbl->llt_ifp; CURVNET_SET(ifp->if_vnet); switch (lle->ln_state) { case ARP_LLINFO_REACHABLE: /* * Expiration time is approaching. * Let's try to refresh entry if it is still * in use. * * Set r_skip_req to get feedback from * fast path. Change state and re-schedule * ourselves. */ LLE_REQ_LOCK(lle); lle->r_skip_req = 1; LLE_REQ_UNLOCK(lle); lle->ln_state = ARP_LLINFO_VERIFY; callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); LLE_WUNLOCK(lle); CURVNET_RESTORE(); return; case ARP_LLINFO_VERIFY: LLE_REQ_LOCK(lle); r_skip_req = lle->r_skip_req; LLE_REQ_UNLOCK(lle); if (r_skip_req == 0 && lle->la_preempt > 0) { /* Entry was used, issue refresh request */ struct in_addr dst; dst = lle->r_l3addr.addr4; lle->la_preempt--; callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); LLE_WUNLOCK(lle); arprequest(ifp, NULL, &dst, NULL); CURVNET_RESTORE(); return; } /* Nothing happened. Reschedule if not too late */ if (lle->la_expire > time_uptime) { callout_schedule(&lle->lle_timer, hz * V_arpt_rexmit); LLE_WUNLOCK(lle); CURVNET_RESTORE(); return; } break; case ARP_LLINFO_INCOMPLETE: case ARP_LLINFO_DELETED: break; } if ((lle->la_flags & LLE_DELETED) == 0) { int evt; if (lle->la_flags & LLE_VALID) evt = LLENTRY_EXPIRED; else evt = LLENTRY_TIMEDOUT; EVENTHANDLER_INVOKE(lle_event, lle, evt); } callout_stop(&lle->lle_timer); /* XXX: LOR avoidance. We still have ref on lle. */ LLE_WUNLOCK(lle); IF_AFDATA_LOCK(ifp); LLE_WLOCK(lle); /* Guard against race with other llentry_free(). */ if (lle->la_flags & LLE_LINKED) { LLE_REMREF(lle); lltable_unlink_entry(lle->lle_tbl, lle); } IF_AFDATA_UNLOCK(ifp); size_t pkts_dropped = llentry_free(lle); ARPSTAT_ADD(dropped, pkts_dropped); ARPSTAT_INC(timeouts); CURVNET_RESTORE(); } /* * Stores link-layer header for @ifp in format suitable for if_output() * into buffer @buf. Resulting header length is stored in @bufsize. * * Returns 0 on success. */ static int arp_fillheader(struct ifnet *ifp, struct arphdr *ah, int bcast, u_char *buf, size_t *bufsize) { struct if_encap_req ereq; int error; bzero(buf, *bufsize); bzero(&ereq, sizeof(ereq)); ereq.buf = buf; ereq.bufsize = *bufsize; ereq.rtype = IFENCAP_LL; ereq.family = AF_ARP; ereq.lladdr = ar_tha(ah); ereq.hdata = (u_char *)ah; if (bcast) ereq.flags = IFENCAP_FLAG_BROADCAST; error = ifp->if_requestencap(ifp, &ereq); if (error == 0) *bufsize = ereq.bufsize; return (error); } /* * Broadcast an ARP request. Caller specifies: * - arp header source ip address * - arp header target ip address * - arp header source ethernet address */ void arprequest(struct ifnet *ifp, const struct in_addr *sip, const struct in_addr *tip, u_char *enaddr) { struct mbuf *m; struct arphdr *ah; struct sockaddr sa; u_char *carpaddr = NULL; uint8_t linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; struct route ro; int error; if (sip == NULL) { /* * The caller did not supply a source address, try to find * a compatible one among those assigned to this interface. */ struct ifaddr *ifa; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; if (ifa->ifa_carp) { if ((*carp_iamatch_p)(ifa, &carpaddr) == 0) continue; sip = &IA_SIN(ifa)->sin_addr; } else { carpaddr = NULL; sip = &IA_SIN(ifa)->sin_addr; } if (0 == ((sip->s_addr ^ tip->s_addr) & IA_MASKSIN(ifa)->sin_addr.s_addr)) break; /* found it. */ } IF_ADDR_RUNLOCK(ifp); if (sip == NULL) { printf("%s: cannot find matching address\n", __func__); return; } } if (enaddr == NULL) enaddr = carpaddr ? carpaddr : (u_char *)IF_LLADDR(ifp); if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL) return; m->m_len = sizeof(*ah) + 2 * sizeof(struct in_addr) + 2 * ifp->if_addrlen; m->m_pkthdr.len = m->m_len; M_ALIGN(m, m->m_len); ah = mtod(m, struct arphdr *); bzero((caddr_t)ah, m->m_len); #ifdef MAC mac_netinet_arp_send(ifp, m); #endif ah->ar_pro = htons(ETHERTYPE_IP); ah->ar_hln = ifp->if_addrlen; /* hardware address length */ ah->ar_pln = sizeof(struct in_addr); /* protocol address length */ ah->ar_op = htons(ARPOP_REQUEST); bcopy(enaddr, ar_sha(ah), ah->ar_hln); bcopy(sip, ar_spa(ah), ah->ar_pln); bcopy(tip, ar_tpa(ah), ah->ar_pln); sa.sa_family = AF_ARP; sa.sa_len = 2; /* Calculate link header for sending frame */ bzero(&ro, sizeof(ro)); linkhdrsize = sizeof(linkhdr); error = arp_fillheader(ifp, ah, 1, linkhdr, &linkhdrsize); if (error != 0 && error != EAFNOSUPPORT) { ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n", if_name(ifp), error); return; } ro.ro_prepend = linkhdr; ro.ro_plen = linkhdrsize; ro.ro_flags = 0; m->m_flags |= M_BCAST; m_clrprotoflags(m); /* Avoid confusing lower layers. */ (*ifp->if_output)(ifp, m, &sa, &ro); ARPSTAT_INC(txrequests); } /* * Resolve an IP address into an ethernet address - heavy version. * Used internally by arpresolve(). * We have already checked than we can't use existing lle without * modification so we have to acquire LLE_EXCLUSIVE lle lock. * * On success, desten and flags are filled in and the function returns 0; * If the packet must be held pending resolution, we return EWOULDBLOCK * On other errors, we return the corresponding error code. * Note that m_freem() handles NULL. */ static int arpresolve_full(struct ifnet *ifp, int is_gw, int flags, struct mbuf *m, const struct sockaddr *dst, u_char *desten, uint32_t *pflags, struct llentry **plle) { struct llentry *la = NULL, *la_tmp; struct mbuf *curr = NULL; struct mbuf *next = NULL; int error, renew; char *lladdr; int ll_len; if (pflags != NULL) *pflags = 0; if (plle != NULL) *plle = NULL; if ((flags & LLE_CREATE) == 0) { IF_AFDATA_RLOCK(ifp); la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); IF_AFDATA_RUNLOCK(ifp); } if (la == NULL && (ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) == 0) { la = lltable_alloc_entry(LLTABLE(ifp), 0, dst); if (la == NULL) { char addrbuf[INET_ADDRSTRLEN]; log(LOG_DEBUG, "arpresolve: can't allocate llinfo for %s on %s\n", inet_ntoa_r(SIN(dst)->sin_addr, addrbuf), if_name(ifp)); m_freem(m); return (EINVAL); } IF_AFDATA_WLOCK(ifp); LLE_WLOCK(la); la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); /* Prefer ANY existing lle over newly-created one */ if (la_tmp == NULL) lltable_link_entry(LLTABLE(ifp), la); IF_AFDATA_WUNLOCK(ifp); if (la_tmp != NULL) { lltable_free_entry(LLTABLE(ifp), la); la = la_tmp; } } if (la == NULL) { m_freem(m); return (EINVAL); } if ((la->la_flags & LLE_VALID) && ((la->la_flags & LLE_STATIC) || la->la_expire > time_uptime)) { if (flags & LLE_ADDRONLY) { lladdr = la->ll_addr; ll_len = ifp->if_addrlen; } else { lladdr = la->r_linkdata; ll_len = la->r_hdrlen; } bcopy(lladdr, desten, ll_len); /* Notify LLE code that the entry was used by datapath */ llentry_mark_used(la); if (pflags != NULL) *pflags = la->la_flags & (LLE_VALID|LLE_IFADDR); if (plle) { LLE_ADDREF(la); *plle = la; } LLE_WUNLOCK(la); return (0); } renew = (la->la_asked == 0 || la->la_expire != time_uptime); /* * There is an arptab entry, but no ethernet address * response yet. Add the mbuf to the list, dropping * the oldest packet if we have exceeded the system * setting. */ if (m != NULL) { if (la->la_numheld >= V_arp_maxhold) { if (la->la_hold != NULL) { next = la->la_hold->m_nextpkt; m_freem(la->la_hold); la->la_hold = next; la->la_numheld--; ARPSTAT_INC(dropped); } } if (la->la_hold != NULL) { curr = la->la_hold; while (curr->m_nextpkt != NULL) curr = curr->m_nextpkt; curr->m_nextpkt = m; } else la->la_hold = m; la->la_numheld++; } /* * Return EWOULDBLOCK if we have tried less than arp_maxtries. It * will be masked by ether_output(). Return EHOSTDOWN/EHOSTUNREACH * if we have already sent arp_maxtries ARP requests. Retransmit the * ARP request, but not faster than one request per second. */ if (la->la_asked < V_arp_maxtries) error = EWOULDBLOCK; /* First request. */ else error = is_gw != 0 ? EHOSTUNREACH : EHOSTDOWN; if (renew) { int canceled; LLE_ADDREF(la); la->la_expire = time_uptime; canceled = callout_reset(&la->lle_timer, hz * V_arpt_down, arptimer, la); if (canceled) LLE_REMREF(la); la->la_asked++; LLE_WUNLOCK(la); arprequest(ifp, NULL, &SIN(dst)->sin_addr, NULL); return (error); } LLE_WUNLOCK(la); return (error); } /* * Resolve an IP address into an ethernet address. */ int arpresolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst, char *desten, uint32_t *pflags, struct llentry **plle) { int error; flags |= LLE_ADDRONLY; error = arpresolve_full(ifp, 0, flags, NULL, dst, desten, pflags, plle); return (error); } /* * Lookups link header based on an IP address. * On input: * ifp is the interface we use * is_gw != 0 if @dst represents gateway to some destination * m is the mbuf. May be NULL if we don't have a packet. * dst is the next hop, * desten is the storage to put LL header. * flags returns subset of lle flags: LLE_VALID | LLE_IFADDR * * On success, full/partial link header and flags are filled in and * the function returns 0. * If the packet must be held pending resolution, we return EWOULDBLOCK * On other errors, we return the corresponding error code. * Note that m_freem() handles NULL. */ int arpresolve(struct ifnet *ifp, int is_gw, struct mbuf *m, const struct sockaddr *dst, u_char *desten, uint32_t *pflags, struct llentry **plle) { struct llentry *la = NULL; if (pflags != NULL) *pflags = 0; if (plle != NULL) *plle = NULL; if (m != NULL) { if (m->m_flags & M_BCAST) { /* broadcast */ (void)memcpy(desten, ifp->if_broadcastaddr, ifp->if_addrlen); return (0); } if (m->m_flags & M_MCAST) { /* multicast */ ETHER_MAP_IP_MULTICAST(&SIN(dst)->sin_addr, desten); return (0); } } IF_AFDATA_RLOCK(ifp); la = lla_lookup(LLTABLE(ifp), plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, dst); if (la != NULL && (la->r_flags & RLLE_VALID) != 0) { /* Entry found, let's copy lle info */ bcopy(la->r_linkdata, desten, la->r_hdrlen); if (pflags != NULL) *pflags = LLE_VALID | (la->r_flags & RLLE_IFADDR); /* Notify the LLE handling code that the entry was used. */ llentry_mark_used(la); if (plle) { LLE_ADDREF(la); *plle = la; LLE_WUNLOCK(la); } IF_AFDATA_RUNLOCK(ifp); return (0); } if (plle && la) LLE_WUNLOCK(la); IF_AFDATA_RUNLOCK(ifp); return (arpresolve_full(ifp, is_gw, la == NULL ? LLE_CREATE : 0, m, dst, desten, pflags, plle)); } /* * Common length and type checks are done here, * then the protocol-specific routine is called. */ static void arpintr(struct mbuf *m) { struct arphdr *ar; struct ifnet *ifp; char *layer; int hlen; ifp = m->m_pkthdr.rcvif; if (m->m_len < sizeof(struct arphdr) && ((m = m_pullup(m, sizeof(struct arphdr))) == NULL)) { ARP_LOG(LOG_NOTICE, "packet with short header received on %s\n", if_name(ifp)); return; } ar = mtod(m, struct arphdr *); /* Check if length is sufficient */ if (m->m_len < arphdr_len(ar)) { m = m_pullup(m, arphdr_len(ar)); if (m == NULL) { ARP_LOG(LOG_NOTICE, "short packet received on %s\n", if_name(ifp)); return; } ar = mtod(m, struct arphdr *); } hlen = 0; layer = ""; switch (ntohs(ar->ar_hrd)) { case ARPHRD_ETHER: hlen = ETHER_ADDR_LEN; /* RFC 826 */ layer = "ethernet"; break; case ARPHRD_INFINIBAND: hlen = 20; /* RFC 4391, INFINIBAND_ALEN */ layer = "infiniband"; break; case ARPHRD_IEEE1394: hlen = 0; /* SHALL be 16 */ /* RFC 2734 */ layer = "firewire"; /* * Restrict too long hardware addresses. * Currently we are capable of handling 20-byte * addresses ( sizeof(lle->ll_addr) ) */ if (ar->ar_hln >= 20) hlen = 16; break; default: ARP_LOG(LOG_NOTICE, "packet with unknown hardware format 0x%02d received on " "%s\n", ntohs(ar->ar_hrd), if_name(ifp)); m_freem(m); return; } if (hlen != 0 && hlen != ar->ar_hln) { ARP_LOG(LOG_NOTICE, "packet with invalid %s address length %d received on %s\n", layer, ar->ar_hln, if_name(ifp)); m_freem(m); return; } ARPSTAT_INC(received); switch (ntohs(ar->ar_pro)) { #ifdef INET case ETHERTYPE_IP: in_arpinput(m); return; #endif } m_freem(m); } #ifdef INET /* * ARP for Internet protocols on 10 Mb/s Ethernet. * Algorithm is that given in RFC 826. * In addition, a sanity check is performed on the sender * protocol address, to catch impersonators. * We no longer handle negotiations for use of trailer protocol: * Formerly, ARP replied for protocol type ETHERTYPE_TRAIL sent * along with IP replies if we wanted trailers sent to us, * and also sent them in response to IP replies. * This allowed either end to announce the desire to receive * trailer packets. * We no longer reply to requests for ETHERTYPE_TRAIL protocol either, * but formerly didn't normally send requests. */ static int log_arp_wrong_iface = 1; static int log_arp_movements = 1; static int log_arp_permanent_modify = 1; static int allow_multicast = 0; SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_wrong_iface, CTLFLAG_RW, &log_arp_wrong_iface, 0, "log arp packets arriving on the wrong interface"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_movements, CTLFLAG_RW, &log_arp_movements, 0, "log arp replies from MACs different than the one in the cache"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, log_arp_permanent_modify, CTLFLAG_RW, &log_arp_permanent_modify, 0, "log arp replies from MACs different than the one in the permanent arp entry"); SYSCTL_INT(_net_link_ether_inet, OID_AUTO, allow_multicast, CTLFLAG_RW, &allow_multicast, 0, "accept multicast addresses"); static void in_arpinput(struct mbuf *m) { struct rm_priotracker in_ifa_tracker; struct arphdr *ah; struct ifnet *ifp = m->m_pkthdr.rcvif; struct llentry *la = NULL, *la_tmp; struct ifaddr *ifa; struct in_ifaddr *ia; struct sockaddr sa; struct in_addr isaddr, itaddr, myaddr; u_int8_t *enaddr = NULL; int op; int bridged = 0, is_bridge = 0; int carped; struct sockaddr_in sin; struct sockaddr *dst; struct nhop4_basic nh4; uint8_t linkhdr[LLE_MAX_LINKHDR]; struct route ro; size_t linkhdrsize; int lladdr_off; int error; char addrbuf[INET_ADDRSTRLEN]; sin.sin_len = sizeof(struct sockaddr_in); sin.sin_family = AF_INET; sin.sin_addr.s_addr = 0; if (ifp->if_bridge) bridged = 1; if (ifp->if_type == IFT_BRIDGE) is_bridge = 1; /* * We already have checked that mbuf contains enough contiguous data * to hold entire arp message according to the arp header. */ ah = mtod(m, struct arphdr *); /* * ARP is only for IPv4 so we can reject packets with * a protocol length not equal to an IPv4 address. */ if (ah->ar_pln != sizeof(struct in_addr)) { ARP_LOG(LOG_NOTICE, "requested protocol length != %zu\n", sizeof(struct in_addr)); goto drop; } if (allow_multicast == 0 && ETHER_IS_MULTICAST(ar_sha(ah))) { ARP_LOG(LOG_NOTICE, "%*D is multicast\n", ifp->if_addrlen, (u_char *)ar_sha(ah), ":"); goto drop; } op = ntohs(ah->ar_op); (void)memcpy(&isaddr, ar_spa(ah), sizeof (isaddr)); (void)memcpy(&itaddr, ar_tpa(ah), sizeof (itaddr)); if (op == ARPOP_REPLY) ARPSTAT_INC(rxreplies); /* * For a bridge, we want to check the address irrespective * of the receive interface. (This will change slightly * when we have clusters of interfaces). */ IN_IFADDR_RLOCK(&in_ifa_tracker); LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) { if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) || ia->ia_ifp == ifp) && itaddr.s_addr == ia->ia_addr.sin_addr.s_addr && (ia->ia_ifa.ifa_carp == NULL || (*carp_iamatch_p)(&ia->ia_ifa, &enaddr))) { ifa_ref(&ia->ia_ifa); IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto match; } } LIST_FOREACH(ia, INADDR_HASH(isaddr.s_addr), ia_hash) if (((bridged && ia->ia_ifp->if_bridge == ifp->if_bridge) || ia->ia_ifp == ifp) && isaddr.s_addr == ia->ia_addr.sin_addr.s_addr) { ifa_ref(&ia->ia_ifa); IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto match; } #define BDG_MEMBER_MATCHES_ARP(addr, ifp, ia) \ (ia->ia_ifp->if_bridge == ifp->if_softc && \ !bcmp(IF_LLADDR(ia->ia_ifp), IF_LLADDR(ifp), ifp->if_addrlen) && \ addr == ia->ia_addr.sin_addr.s_addr) /* * Check the case when bridge shares its MAC address with * some of its children, so packets are claimed by bridge * itself (bridge_input() does it first), but they are really * meant to be destined to the bridge member. */ if (is_bridge) { LIST_FOREACH(ia, INADDR_HASH(itaddr.s_addr), ia_hash) { if (BDG_MEMBER_MATCHES_ARP(itaddr.s_addr, ifp, ia)) { ifa_ref(&ia->ia_ifa); ifp = ia->ia_ifp; IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto match; } } } #undef BDG_MEMBER_MATCHES_ARP IN_IFADDR_RUNLOCK(&in_ifa_tracker); /* * No match, use the first inet address on the receive interface * as a dummy address for the rest of the function. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (ifa->ifa_addr->sa_family == AF_INET && (ifa->ifa_carp == NULL || (*carp_iamatch_p)(ifa, &enaddr))) { ia = ifatoia(ifa); ifa_ref(ifa); IF_ADDR_RUNLOCK(ifp); goto match; } IF_ADDR_RUNLOCK(ifp); /* * If bridging, fall back to using any inet address. */ IN_IFADDR_RLOCK(&in_ifa_tracker); if (!bridged || (ia = CK_STAILQ_FIRST(&V_in_ifaddrhead)) == NULL) { IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto drop; } ifa_ref(&ia->ia_ifa); IN_IFADDR_RUNLOCK(&in_ifa_tracker); match: if (!enaddr) enaddr = (u_int8_t *)IF_LLADDR(ifp); carped = (ia->ia_ifa.ifa_carp != NULL); myaddr = ia->ia_addr.sin_addr; ifa_free(&ia->ia_ifa); if (!bcmp(ar_sha(ah), enaddr, ifp->if_addrlen)) goto drop; /* it's from me, ignore it. */ if (!bcmp(ar_sha(ah), ifp->if_broadcastaddr, ifp->if_addrlen)) { ARP_LOG(LOG_NOTICE, "link address is broadcast for IP address " "%s!\n", inet_ntoa_r(isaddr, addrbuf)); goto drop; } if (ifp->if_addrlen != ah->ar_hln) { ARP_LOG(LOG_WARNING, "from %*D: addr len: new %d, " "i/f %d (ignored)\n", ifp->if_addrlen, (u_char *) ar_sha(ah), ":", ah->ar_hln, ifp->if_addrlen); goto drop; } /* * Warn if another host is using the same IP address, but only if the * IP address isn't 0.0.0.0, which is used for DHCP only, in which * case we suppress the warning to avoid false positive complaints of * potential misconfiguration. */ if (!bridged && !carped && isaddr.s_addr == myaddr.s_addr && myaddr.s_addr != 0) { ARP_LOG(LOG_ERR, "%*D is using my IP address %s on %s!\n", ifp->if_addrlen, (u_char *)ar_sha(ah), ":", inet_ntoa_r(isaddr, addrbuf), ifp->if_xname); itaddr = myaddr; ARPSTAT_INC(dupips); goto reply; } if (ifp->if_flags & IFF_STATICARP) goto reply; bzero(&sin, sizeof(sin)); sin.sin_len = sizeof(struct sockaddr_in); sin.sin_family = AF_INET; sin.sin_addr = isaddr; dst = (struct sockaddr *)&sin; IF_AFDATA_RLOCK(ifp); la = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); IF_AFDATA_RUNLOCK(ifp); if (la != NULL) arp_check_update_lle(ah, isaddr, ifp, bridged, la); else if (itaddr.s_addr == myaddr.s_addr) { /* * Request/reply to our address, but no lle exists yet. * Calculate full link prepend to use in lle. */ linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr, &linkhdrsize, &lladdr_off) != 0) goto reply; /* Allocate new entry */ la = lltable_alloc_entry(LLTABLE(ifp), 0, dst); if (la == NULL) { /* * lle creation may fail if source address belongs * to non-directly connected subnet. However, we * will try to answer the request instead of dropping * frame. */ goto reply; } lltable_set_entry_addr(ifp, la, linkhdr, linkhdrsize, lladdr_off); IF_AFDATA_WLOCK(ifp); LLE_WLOCK(la); la_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); /* * Check if lle still does not exists. * If it does, that means that we either * 1) have configured it explicitly, via * 1a) 'arp -s' static entry or * 1b) interface address static record * or * 2) it was the result of sending first packet to-host * or * 3) it was another arp reply packet we handled in * different thread. * * In all cases except 3) we definitely need to prefer * existing lle. For the sake of simplicity, prefer any * existing lle over newly-create one. */ if (la_tmp == NULL) lltable_link_entry(LLTABLE(ifp), la); IF_AFDATA_WUNLOCK(ifp); if (la_tmp == NULL) { arp_mark_lle_reachable(la); LLE_WUNLOCK(la); } else { /* Free newly-create entry and handle packet */ lltable_free_entry(LLTABLE(ifp), la); la = la_tmp; la_tmp = NULL; arp_check_update_lle(ah, isaddr, ifp, bridged, la); /* arp_check_update_lle() returns @la unlocked */ } la = NULL; } reply: if (op != ARPOP_REQUEST) goto drop; ARPSTAT_INC(rxrequests); if (itaddr.s_addr == myaddr.s_addr) { /* Shortcut.. the receiving interface is the target. */ (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln); } else { struct llentry *lle = NULL; sin.sin_addr = itaddr; IF_AFDATA_RLOCK(ifp); lle = lla_lookup(LLTABLE(ifp), 0, (struct sockaddr *)&sin); IF_AFDATA_RUNLOCK(ifp); if ((lle != NULL) && (lle->la_flags & LLE_PUB)) { (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); (void)memcpy(ar_sha(ah), lle->ll_addr, ah->ar_hln); LLE_RUNLOCK(lle); } else { if (lle != NULL) LLE_RUNLOCK(lle); if (!V_arp_proxyall) goto drop; /* XXX MRT use table 0 for arp reply */ if (fib4_lookup_nh_basic(0, itaddr, 0, 0, &nh4) != 0) goto drop; /* * Don't send proxies for nodes on the same interface * as this one came out of, or we'll get into a fight * over who claims what Ether address. */ if (nh4.nh_ifp == ifp) goto drop; (void)memcpy(ar_tha(ah), ar_sha(ah), ah->ar_hln); (void)memcpy(ar_sha(ah), enaddr, ah->ar_hln); /* * Also check that the node which sent the ARP packet * is on the interface we expect it to be on. This * avoids ARP chaos if an interface is connected to the * wrong network. */ /* XXX MRT use table 0 for arp checks */ if (fib4_lookup_nh_basic(0, isaddr, 0, 0, &nh4) != 0) goto drop; if (nh4.nh_ifp != ifp) { ARP_LOG(LOG_INFO, "proxy: ignoring request" " from %s via %s\n", inet_ntoa_r(isaddr, addrbuf), ifp->if_xname); goto drop; } #ifdef DEBUG_PROXY printf("arp: proxying for %s\n", inet_ntoa_r(itaddr, addrbuf)); #endif } } if (itaddr.s_addr == myaddr.s_addr && IN_LINKLOCAL(ntohl(itaddr.s_addr))) { /* RFC 3927 link-local IPv4; always reply by broadcast. */ #ifdef DEBUG_LINKLOCAL printf("arp: sending reply for link-local addr %s\n", inet_ntoa_r(itaddr, addrbuf)); #endif m->m_flags |= M_BCAST; m->m_flags &= ~M_MCAST; } else { /* default behaviour; never reply by broadcast. */ m->m_flags &= ~(M_BCAST|M_MCAST); } (void)memcpy(ar_tpa(ah), ar_spa(ah), ah->ar_pln); (void)memcpy(ar_spa(ah), &itaddr, ah->ar_pln); ah->ar_op = htons(ARPOP_REPLY); ah->ar_pro = htons(ETHERTYPE_IP); /* let's be sure! */ m->m_len = sizeof(*ah) + (2 * ah->ar_pln) + (2 * ah->ar_hln); m->m_pkthdr.len = m->m_len; m->m_pkthdr.rcvif = NULL; sa.sa_family = AF_ARP; sa.sa_len = 2; /* Calculate link header for sending frame */ bzero(&ro, sizeof(ro)); linkhdrsize = sizeof(linkhdr); error = arp_fillheader(ifp, ah, 0, linkhdr, &linkhdrsize); /* * arp_fillheader() may fail due to lack of support inside encap request * routing. This is not necessary an error, AF_ARP can/should be handled * by if_output(). */ if (error != 0 && error != EAFNOSUPPORT) { ARP_LOG(LOG_ERR, "Failed to calculate ARP header on %s: %d\n", if_name(ifp), error); return; } ro.ro_prepend = linkhdr; ro.ro_plen = linkhdrsize; ro.ro_flags = 0; m_clrprotoflags(m); /* Avoid confusing lower layers. */ (*ifp->if_output)(ifp, m, &sa, &ro); ARPSTAT_INC(txreplies); return; drop: m_freem(m); } #endif /* * Checks received arp data against existing @la. * Updates lle state/performs notification if necessary. */ static void arp_check_update_lle(struct arphdr *ah, struct in_addr isaddr, struct ifnet *ifp, int bridged, struct llentry *la) { struct sockaddr sa; struct mbuf *m_hold, *m_hold_next; uint8_t linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; int lladdr_off; char addrbuf[INET_ADDRSTRLEN]; LLE_WLOCK_ASSERT(la); /* the following is not an error when doing bridging */ if (!bridged && la->lle_tbl->llt_ifp != ifp) { if (log_arp_wrong_iface) ARP_LOG(LOG_WARNING, "%s is on %s " "but got reply from %*D on %s\n", inet_ntoa_r(isaddr, addrbuf), la->lle_tbl->llt_ifp->if_xname, ifp->if_addrlen, (u_char *)ar_sha(ah), ":", ifp->if_xname); LLE_WUNLOCK(la); return; } if ((la->la_flags & LLE_VALID) && bcmp(ar_sha(ah), la->ll_addr, ifp->if_addrlen)) { if (la->la_flags & LLE_STATIC) { LLE_WUNLOCK(la); if (log_arp_permanent_modify) ARP_LOG(LOG_ERR, "%*D attempts to modify " "permanent entry for %s on %s\n", ifp->if_addrlen, (u_char *)ar_sha(ah), ":", inet_ntoa_r(isaddr, addrbuf), ifp->if_xname); return; } if (log_arp_movements) { ARP_LOG(LOG_INFO, "%s moved from %*D " "to %*D on %s\n", inet_ntoa_r(isaddr, addrbuf), ifp->if_addrlen, (u_char *)la->ll_addr, ":", ifp->if_addrlen, (u_char *)ar_sha(ah), ":", ifp->if_xname); } } /* Calculate full link prepend to use in lle */ linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET, ar_sha(ah), linkhdr, &linkhdrsize, &lladdr_off) != 0) return; /* Check if something has changed */ if (memcmp(la->r_linkdata, linkhdr, linkhdrsize) != 0 || (la->la_flags & LLE_VALID) == 0) { /* Try to perform LLE update */ if (lltable_try_set_entry_addr(ifp, la, linkhdr, linkhdrsize, lladdr_off) == 0) return; /* Clear fast path feedback request if set */ la->r_skip_req = 0; } arp_mark_lle_reachable(la); /* * The packets are all freed within the call to the output * routine. * * NB: The lock MUST be released before the call to the * output routine. */ if (la->la_hold != NULL) { m_hold = la->la_hold; la->la_hold = NULL; la->la_numheld = 0; lltable_fill_sa_entry(la, &sa); LLE_WUNLOCK(la); for (; m_hold != NULL; m_hold = m_hold_next) { m_hold_next = m_hold->m_nextpkt; m_hold->m_nextpkt = NULL; /* Avoid confusing lower layers. */ m_clrprotoflags(m_hold); (*ifp->if_output)(ifp, m_hold, &sa, NULL); } } else LLE_WUNLOCK(la); } static void arp_mark_lle_reachable(struct llentry *la) { int canceled, wtime; LLE_WLOCK_ASSERT(la); la->ln_state = ARP_LLINFO_REACHABLE; EVENTHANDLER_INVOKE(lle_event, la, LLENTRY_RESOLVED); if (!(la->la_flags & LLE_STATIC)) { LLE_ADDREF(la); la->la_expire = time_uptime + V_arpt_keep; wtime = V_arpt_keep - V_arp_maxtries * V_arpt_rexmit; if (wtime < 0) wtime = V_arpt_keep; canceled = callout_reset(&la->lle_timer, hz * wtime, arptimer, la); if (canceled) LLE_REMREF(la); } la->la_asked = 0; la->la_preempt = V_arp_maxtries; } /* * Add permanent link-layer record for given interface address. */ static __noinline void arp_add_ifa_lle(struct ifnet *ifp, const struct sockaddr *dst) { struct llentry *lle, *lle_tmp; /* * Interface address LLE record is considered static * because kernel code relies on LLE_STATIC flag to check * if these entries can be rewriten by arp updates. */ lle = lltable_alloc_entry(LLTABLE(ifp), LLE_IFADDR | LLE_STATIC, dst); if (lle == NULL) { log(LOG_INFO, "arp_ifinit: cannot create arp " "entry for interface address\n"); return; } IF_AFDATA_WLOCK(ifp); LLE_WLOCK(lle); /* Unlink any entry if exists */ lle_tmp = lla_lookup(LLTABLE(ifp), LLE_EXCLUSIVE, dst); if (lle_tmp != NULL) lltable_unlink_entry(LLTABLE(ifp), lle_tmp); lltable_link_entry(LLTABLE(ifp), lle); IF_AFDATA_WUNLOCK(ifp); if (lle_tmp != NULL) EVENTHANDLER_INVOKE(lle_event, lle_tmp, LLENTRY_EXPIRED); EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_RESOLVED); LLE_WUNLOCK(lle); if (lle_tmp != NULL) lltable_free_entry(LLTABLE(ifp), lle_tmp); } /* * Handle the garp_rexmit_count. Like sysctl_handle_int(), but limits the range * of valid values. */ static int sysctl_garp_rexmit(SYSCTL_HANDLER_ARGS) { int error; int rexmit_count = *(int *)arg1; error = sysctl_handle_int(oidp, &rexmit_count, 0, req); /* Enforce limits on any new value that may have been set. */ if (!error && req->newptr) { /* A new value was set. */ if (rexmit_count < 0) { rexmit_count = 0; } else if (rexmit_count > MAX_GARP_RETRANSMITS) { rexmit_count = MAX_GARP_RETRANSMITS; } *(int *)arg1 = rexmit_count; } return (error); } /* * Retransmit a Gratuitous ARP (GARP) and, if necessary, schedule a callout to * retransmit it again. A pending callout owns a reference to the ifa. */ static void garp_rexmit(void *arg) { struct in_ifaddr *ia = arg; if (callout_pending(&ia->ia_garp_timer) || !callout_active(&ia->ia_garp_timer)) { IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); ifa_free(&ia->ia_ifa); return; } /* * Drop lock while the ARP request is generated. */ IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); arprequest(ia->ia_ifa.ifa_ifp, &IA_SIN(ia)->sin_addr, &IA_SIN(ia)->sin_addr, IF_LLADDR(ia->ia_ifa.ifa_ifp)); /* * Increment the count of retransmissions. If the count has reached the * maximum value, stop sending the GARP packets. Otherwise, schedule * the callout to retransmit another GARP packet. */ ++ia->ia_garp_count; if (ia->ia_garp_count >= garp_rexmit_count) { ifa_free(&ia->ia_ifa); } else { int rescheduled; IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp); rescheduled = callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz, garp_rexmit, ia); IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); if (rescheduled) { ifa_free(&ia->ia_ifa); } } } /* * Start the GARP retransmit timer. * * A single GARP is always transmitted when an IPv4 address is added * to an interface and that is usually sufficient. However, in some * circumstances, such as when a shared address is passed between * cluster nodes, this single GARP may occasionally be dropped or * lost. This can lead to neighbors on the network link working with a * stale ARP cache and sending packets destined for that address to * the node that previously owned the address, which may not respond. * * To avoid this situation, GARP retransmits can be enabled by setting * the net.link.ether.inet.garp_rexmit_count sysctl to a value greater * than zero. The setting represents the maximum number of * retransmissions. The interval between retransmissions is calculated * using an exponential backoff algorithm, doubling each time, so the * retransmission intervals are: {1, 2, 4, 8, 16, ...} (seconds). */ static void garp_timer_start(struct ifaddr *ifa) { struct in_ifaddr *ia = (struct in_ifaddr *) ifa; IF_ADDR_WLOCK(ia->ia_ifa.ifa_ifp); ia->ia_garp_count = 0; if (callout_reset(&ia->ia_garp_timer, (1 << ia->ia_garp_count) * hz, garp_rexmit, ia) == 0) { ifa_ref(ifa); } IF_ADDR_WUNLOCK(ia->ia_ifa.ifa_ifp); } void arp_ifinit(struct ifnet *ifp, struct ifaddr *ifa) { const struct sockaddr_in *dst_in; const struct sockaddr *dst; if (ifa->ifa_carp != NULL) return; dst = ifa->ifa_addr; dst_in = (const struct sockaddr_in *)dst; if (ntohl(dst_in->sin_addr.s_addr) == INADDR_ANY) return; arp_announce_ifaddr(ifp, dst_in->sin_addr, IF_LLADDR(ifp)); if (garp_rexmit_count > 0) { garp_timer_start(ifa); } arp_add_ifa_lle(ifp, dst); } void arp_announce_ifaddr(struct ifnet *ifp, struct in_addr addr, u_char *enaddr) { if (ntohl(addr.s_addr) != INADDR_ANY) arprequest(ifp, &addr, &addr, enaddr); } /* * Sends gratuitous ARPs for each ifaddr to notify other * nodes about the address change. */ static __noinline void arp_handle_ifllchange(struct ifnet *ifp) { struct ifaddr *ifa; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family == AF_INET) arp_ifinit(ifp, ifa); } } /* * A handler for interface link layer address change event. */ static void arp_iflladdr(void *arg __unused, struct ifnet *ifp) { lltable_update_ifaddr(LLTABLE(ifp)); if ((ifp->if_flags & IFF_UP) != 0) arp_handle_ifllchange(ifp); } static void vnet_arp_init(void) { if (IS_DEFAULT_VNET(curvnet)) { netisr_register(&arp_nh); iflladdr_tag = EVENTHANDLER_REGISTER(iflladdr_event, arp_iflladdr, NULL, EVENTHANDLER_PRI_ANY); } #ifdef VIMAGE else netisr_register_vnet(&arp_nh); #endif } VNET_SYSINIT(vnet_arp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, vnet_arp_init, 0); #ifdef VIMAGE /* * We have to unregister ARP along with IP otherwise we risk doing INADDR_HASH * lookups after destroying the hash. Ideally this would go on SI_ORDER_3.5. */ static void vnet_arp_destroy(__unused void *arg) { netisr_unregister_vnet(&arp_nh); } VNET_SYSUNINIT(vnet_arp_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, vnet_arp_destroy, NULL); #endif Index: head/sys/netinet/igmp.c =================================================================== --- head/sys/netinet/igmp.c (revision 336675) +++ head/sys/netinet/igmp.c (revision 336676) @@ -1,3652 +1,3652 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Bruce Simpson. * Copyright (c) 1988 Stephen Deering. * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)igmp.c 8.1 (Berkeley) 7/19/93 */ /* * Internet Group Management Protocol (IGMP) routines. * [RFC1112, RFC2236, RFC3376] * * Written by Steve Deering, Stanford, May 1988. * Modified by Rosen Sharma, Stanford, Aug 1994. * Modified by Bill Fenner, Xerox PARC, Feb 1995. * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995. * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson. * * MULTICAST Revision: 3.5.1.4 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef KTR_IGMPV3 #define KTR_IGMPV3 KTR_INET #endif static struct igmp_ifsoftc * igi_alloc_locked(struct ifnet *); static void igi_delete_locked(const struct ifnet *); static void igmp_dispatch_queue(struct mbufq *, int, const int); static void igmp_fasttimo_vnet(void); static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); static int igmp_handle_state_change(struct in_multi *, struct igmp_ifsoftc *); static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); static int igmp_input_v1_query(struct ifnet *, const struct ip *, const struct igmp *); static int igmp_input_v2_query(struct ifnet *, const struct ip *, const struct igmp *); static int igmp_input_v3_query(struct ifnet *, const struct ip *, /*const*/ struct igmpv3 *); static int igmp_input_v3_group_query(struct in_multi *, struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *); static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, /*const*/ struct igmp *); static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, /*const*/ struct igmp *); static void igmp_intr(struct mbuf *); static int igmp_isgroupreported(const struct in_addr); static struct mbuf * igmp_ra_alloc(void); #ifdef KTR static char * igmp_rec_type_to_str(const int); #endif static void igmp_set_version(struct igmp_ifsoftc *, const int); static void igmp_slowtimo_vnet(void); static int igmp_v1v2_queue_report(struct in_multi *, const int); static void igmp_v1v2_process_group_timer(struct in_multi *, const int); static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); static void igmp_v2_update_group(struct in_multi *, const int); static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); static struct mbuf * igmp_v3_encap_report(struct ifnet *, struct mbuf *); static int igmp_v3_enqueue_group_record(struct mbufq *, struct in_multi *, const int, const int, const int); static int igmp_v3_enqueue_filter_change(struct mbufq *, struct in_multi *); static void igmp_v3_process_group_timers(struct in_multi_head *, struct mbufq *, struct mbufq *, struct in_multi *, const int); static int igmp_v3_merge_state_changes(struct in_multi *, struct mbufq *); static void igmp_v3_suppress_group_record(struct in_multi *); static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); static const struct netisr_handler igmp_nh = { .nh_name = "igmp", .nh_handler = igmp_intr, .nh_proto = NETISR_IGMP, .nh_policy = NETISR_POLICY_SOURCE, }; /* * System-wide globals. * * Unlocked access to these is OK, except for the global IGMP output * queue. The IGMP subsystem lock ends up being system-wide for the moment, * because all VIMAGEs have to share a global output queue, as netisrs * themselves are not virtualized. * * Locking: * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. * Any may be taken independently; if any are held at the same * time, the above lock order must be followed. * * All output is delegated to the netisr. * Now that Giant has been eliminated, the netisr may be inlined. * * IN_MULTI_LIST_LOCK covers in_multi. * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, * including the output queue. * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of * per-link state iterators. * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, * therefore it is not refcounted. * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. * * Reference counting * * IGMP acquires its own reference every time an in_multi is passed to * it and the group is being joined for the first time. * * IGMP releases its reference(s) on in_multi in a deferred way, * because the operations which process the release run as part of * a loop whose control variables are directly affected by the release * (that, and not recursing on the IF_ADDR_LOCK). * * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds * to a vnet in ifp->if_vnet. * * SMPng: XXX We may potentially race operations on ifma_protospec. * The problem is that we currently lack a clean way of taking the * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, * as anything which modifies ifma needs to be covered by that lock. * So check for ifma_protospec being NULL before proceeding. */ struct mtx igmp_mtx; struct mbuf *m_raopt; /* Router Alert option */ static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); /* * VIMAGE-wide globals. * * The IGMPv3 timers themselves need to run per-image, however, * protosw timers run globally (see tcp). * An ifnet can only be in one vimage at a time, and the loopback * ifnet, loif, is itself virtualized. * It would otherwise be possible to seriously hose IGMP state, * and create inconsistencies in upstream multicast routing, if you have * multiple VIMAGEs running on the same link joining different multicast * groups, UNLESS the "primary IP address" is different. This is because * IGMP for IPv4 does not force link-local addresses to be used for each * node, unlike MLD for IPv6. * Obviously the IGMPv3 per-interface state has per-vimage granularity * also as a result. * * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection * policy to control the address used by IGMP on the link. */ -static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general +VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general * query response */ -static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change +VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change * retransmit */ -static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host +VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host * report; IGMPv3 g/sg * query response */ #define V_interface_timers_running VNET(interface_timers_running) #define V_state_change_timers_running VNET(state_change_timers_running) #define V_current_state_timers_running VNET(current_state_timers_running) -static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head) = +VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) = LIST_HEAD_INITIALIZER(igi_head); -static VNET_DEFINE(struct igmpstat, igmpstat) = { +VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = { .igps_version = IGPS_VERSION_3, .igps_len = sizeof(struct igmpstat), }; -static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0}; +VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0}; #define V_igi_head VNET(igi_head) #define V_igmpstat VNET(igmpstat) #define V_igmp_gsrdelay VNET(igmp_gsrdelay) -static VNET_DEFINE(int, igmp_recvifkludge) = 1; -static VNET_DEFINE(int, igmp_sendra) = 1; -static VNET_DEFINE(int, igmp_sendlocal) = 1; -static VNET_DEFINE(int, igmp_v1enable) = 1; -static VNET_DEFINE(int, igmp_v2enable) = 1; -static VNET_DEFINE(int, igmp_legacysupp); -static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3; +VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1; +VNET_DEFINE_STATIC(int, igmp_sendra) = 1; +VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1; +VNET_DEFINE_STATIC(int, igmp_v1enable) = 1; +VNET_DEFINE_STATIC(int, igmp_v2enable) = 1; +VNET_DEFINE_STATIC(int, igmp_legacysupp); +VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3; #define V_igmp_recvifkludge VNET(igmp_recvifkludge) #define V_igmp_sendra VNET(igmp_sendra) #define V_igmp_sendlocal VNET(igmp_sendlocal) #define V_igmp_v1enable VNET(igmp_v1enable) #define V_igmp_v2enable VNET(igmp_v2enable) #define V_igmp_legacysupp VNET(igmp_legacysupp) #define V_igmp_default_version VNET(igmp_default_version) /* * Virtualized sysctls. */ SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmpstat), igmpstat, ""); SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_recvifkludge), 0, "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_sendra), 0, "Send IP Router Alert option in IGMPv2/v3 messages"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_sendlocal), 0, "Send IGMP membership reports for 224.0.0.0/24 groups"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_v1enable), 0, "Enable backwards compatibility with IGMPv1"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_v2enable), 0, "Enable backwards compatibility with IGMPv2"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_legacysupp), 0, "Allow v1/v2 reports to suppress v3 group responses"); SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", "Default version of IGMP to run on each interface"); SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", "Rate limit for IGMPv3 Group-and-Source queries in seconds"); /* * Non-virtualized sysctls. */ static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, "Per-interface IGMPv3 state"); static __inline void igmp_save_context(struct mbuf *m, struct ifnet *ifp) { #ifdef VIMAGE m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; #endif /* VIMAGE */ m->m_pkthdr.flowid = ifp->if_index; } static __inline void igmp_scrub_context(struct mbuf *m) { m->m_pkthdr.PH_loc.ptr = NULL; m->m_pkthdr.flowid = 0; } /* * Restore context from a queued IGMP output chain. * Return saved ifindex. * * VIMAGE: The assertion is there to make sure that we * actually called CURVNET_SET() with what's in the mbuf chain. */ static __inline uint32_t igmp_restore_context(struct mbuf *m) { #ifdef notyet #if defined(VIMAGE) && defined(INVARIANTS) KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), ("%s: called when curvnet was not restored", __func__)); #endif #endif return (m->m_pkthdr.flowid); } /* * Retrieve or set default IGMP version. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by IGMP lock. */ static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) { int error; int new; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); IGMP_LOCK(); new = V_igmp_default_version; error = sysctl_handle_int(oidp, &new, 0, req); if (error || !req->newptr) goto out_locked; if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { error = EINVAL; goto out_locked; } CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", V_igmp_default_version, new); V_igmp_default_version = new; out_locked: IGMP_UNLOCK(); return (error); } /* * Retrieve or set threshold between group-source queries in seconds. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by IGMP lock. */ static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) { int error; int i; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); IGMP_LOCK(); i = V_igmp_gsrdelay.tv_sec; error = sysctl_handle_int(oidp, &i, 0, req); if (error || !req->newptr) goto out_locked; if (i < -1 || i >= 60) { error = EINVAL; goto out_locked; } CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", V_igmp_gsrdelay.tv_sec, i); V_igmp_gsrdelay.tv_sec = i; out_locked: IGMP_UNLOCK(); return (error); } /* * Expose struct igmp_ifsoftc to userland, keyed by ifindex. * For use by ifmcstat(8). * * SMPng: NOTE: Does an unlocked ifindex space read. * VIMAGE: Assume curvnet set by caller. The node handler itself * is not directly virtualized. */ static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) { int *name; int error; u_int namelen; struct ifnet *ifp; struct igmp_ifsoftc *igi; name = (int *)arg1; namelen = arg2; if (req->newptr != NULL) return (EPERM); if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); if (error) return (error); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); if (name[0] <= 0 || name[0] > V_if_index) { error = ENOENT; goto out_locked; } error = ENOENT; ifp = ifnet_byindex(name[0]); if (ifp == NULL) goto out_locked; LIST_FOREACH(igi, &V_igi_head, igi_link) { if (ifp == igi->igi_ifp) { struct igmp_ifinfo info; info.igi_version = igi->igi_version; info.igi_v1_timer = igi->igi_v1_timer; info.igi_v2_timer = igi->igi_v2_timer; info.igi_v3_timer = igi->igi_v3_timer; info.igi_flags = igi->igi_flags; info.igi_rv = igi->igi_rv; info.igi_qi = igi->igi_qi; info.igi_qri = igi->igi_qri; info.igi_uri = igi->igi_uri; error = SYSCTL_OUT(req, &info, sizeof(info)); break; } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (error); } /* * Dispatch an entire queue of pending packet chains * using the netisr. * VIMAGE: Assumes the vnet pointer has been set. */ static void igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) { struct mbuf *m; while ((m = mbufq_dequeue(mq)) != NULL) { CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); if (loop) m->m_flags |= M_IGMP_LOOP; netisr_dispatch(NETISR_IGMP, m); if (--limit == 0) break; } } /* * Filter outgoing IGMP report state by group. * * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are * disabled for all groups in the 224.0.0.0/24 link-local scope. However, * this may break certain IGMP snooping switches which rely on the old * report behaviour. * * Return zero if the given group is one for which IGMP reports * should be suppressed, or non-zero if reports should be issued. */ static __inline int igmp_isgroupreported(const struct in_addr addr) { if (in_allhosts(addr) || ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) return (0); return (1); } /* * Construct a Router Alert option to use in outgoing packets. */ static struct mbuf * igmp_ra_alloc(void) { struct mbuf *m; struct ipoption *p; m = m_get(M_WAITOK, MT_DATA); p = mtod(m, struct ipoption *); p->ipopt_dst.s_addr = INADDR_ANY; p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */ p->ipopt_list[1] = 0x04; /* 4 bytes long */ p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ p->ipopt_list[3] = 0x00; /* pad byte */ m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; return (m); } /* * Attach IGMP when PF_INET is attached to an interface. */ struct igmp_ifsoftc * igmp_domifattach(struct ifnet *ifp) { struct igmp_ifsoftc *igi; CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK(); igi = igi_alloc_locked(ifp); if (!(ifp->if_flags & IFF_MULTICAST)) igi->igi_flags |= IGIF_SILENT; IGMP_UNLOCK(); return (igi); } /* * VIMAGE: assume curvnet set by caller. */ static struct igmp_ifsoftc * igi_alloc_locked(/*const*/ struct ifnet *ifp) { struct igmp_ifsoftc *igi; IGMP_LOCK_ASSERT(); igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); if (igi == NULL) goto out; igi->igi_ifp = ifp; igi->igi_version = V_igmp_default_version; igi->igi_flags = 0; igi->igi_rv = IGMP_RV_INIT; igi->igi_qi = IGMP_QI_INIT; igi->igi_qri = IGMP_QRI_INIT; igi->igi_uri = IGMP_URI_INIT; mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", ifp, ifp->if_xname); out: return (igi); } /* * Hook for ifdetach. * * NOTE: Some finalization tasks need to run before the protocol domain * is detached, but also before the link layer does its cleanup. * * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). * XXX This is also bitten by unlocked ifma_protospec access. */ void igmp_ifdetach(struct ifnet *ifp) { struct igmp_ifsoftc *igi; struct ifmultiaddr *ifma, *next; struct in_multi *inm; struct in_multi_head inm_free_tmp; CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); SLIST_INIT(&inm_free_tmp); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; if (igi->igi_version == IGMP_VERSION_3) { IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; if (inm->inm_state == IGMP_LEAVING_MEMBER) inm_rele_locked(&inm_free_tmp, inm); inm_clear_recorded(inm); if (__predict_false(ifma_restart)) { ifma_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); inm_release_list_deferred(&inm_free_tmp); } IGMP_UNLOCK(); } /* * Hook for domifdetach. */ void igmp_domifdetach(struct ifnet *ifp) { CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK(); igi_delete_locked(ifp); IGMP_UNLOCK(); } static void igi_delete_locked(const struct ifnet *ifp) { struct igmp_ifsoftc *igi, *tigi; CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK_ASSERT(); LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { if (igi->igi_ifp == ifp) { /* * Free deferred General Query responses. */ mbufq_drain(&igi->igi_gq); LIST_REMOVE(igi, igi_link); free(igi, M_IGMP); return; } } } /* * Process a received IGMPv1 query. * Return non-zero if the message should be dropped. * * VIMAGE: The curvnet pointer is derived from the input ifp. */ static int igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, const struct igmp *igmp) { struct ifmultiaddr *ifma; struct igmp_ifsoftc *igi; struct in_multi *inm; /* * IGMPv1 Host Mmembership Queries SHOULD always be addressed to * 224.0.0.1. They are always treated as General Queries. * igmp_group is always ignored. Do not drop it as a userland * daemon may wish to see it. * XXX SMPng: unlocked increments in igmpstat assumed atomic. */ if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { IGMPSTAT_INC(igps_rcv_badqueries); return (0); } IGMPSTAT_INC(igps_rcv_gen_queries); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Switch to IGMPv1 host compatibility mode. */ igmp_set_version(igi, IGMP_VERSION_1); CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); /* * Start the timers in all of our group records * for the interface on which the query arrived, * except those which are already running. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; if (inm->inm_timer != 0) continue; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY( IGMP_V1V2_MAX_RI * PR_FASTHZ); V_current_state_timers_running = 1; break; case IGMP_LEAVING_MEMBER: break; } } IF_ADDR_RUNLOCK(ifp); out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv2 general or group-specific query. */ static int igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, const struct igmp *igmp) { struct ifmultiaddr *ifma; struct igmp_ifsoftc *igi; struct in_multi *inm; int is_general_query; uint16_t timer; is_general_query = 0; /* * Validate address fields upfront. * XXX SMPng: unlocked increments in igmpstat assumed atomic. */ if (in_nullhost(igmp->igmp_group)) { /* * IGMPv2 General Query. * If this was not sent to the all-hosts group, ignore it. */ if (!in_allhosts(ip->ip_dst)) return (0); IGMPSTAT_INC(igps_rcv_gen_queries); is_general_query = 1; } else { /* IGMPv2 Group-Specific Query. */ IGMPSTAT_INC(igps_rcv_group_queries); } IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Ignore v2 query if in v1 Compatibility Mode. */ if (igi->igi_version == IGMP_VERSION_1) goto out_locked; igmp_set_version(igi, IGMP_VERSION_2); timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; if (timer == 0) timer = 1; if (is_general_query) { /* * For each reporting group joined on this * interface, kick the report timer. */ CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", ifp, ifp->if_xname); IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; igmp_v2_update_group(inm, timer); } IF_ADDR_RUNLOCK(ifp); } else { /* * Group-specific IGMPv2 query, we need only * look up the single group to process it. */ inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { CTR3(KTR_IGMPV3, "process v2 query 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); igmp_v2_update_group(inm, timer); } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Update the report timer on a group in response to an IGMPv2 query. * * If we are becoming the reporting member for this group, start the timer. * If we already are the reporting member for this group, and timer is * below the threshold, reset it. * * We may be updating the group for the first time since we switched * to IGMPv3. If we are, then we must clear any recorded source lists, * and transition to REPORTING state; the group timer is overloaded * for group and group-source query responses. * * Unlike IGMPv3, the delay per group should be jittered * to avoid bursts of IGMPv2 reports. */ static void igmp_v2_update_group(struct in_multi *inm, const int timer) { CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer); IN_MULTI_LIST_LOCK_ASSERT(); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_REPORTING_MEMBER: if (inm->inm_timer != 0 && inm->inm_timer <= timer) { CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " "skipping.", __func__); break; } /* FALLTHROUGH */ case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); inm->inm_state = IGMP_REPORTING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; break; case IGMP_SLEEPING_MEMBER: CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); inm->inm_state = IGMP_AWAKENING_MEMBER; break; case IGMP_LEAVING_MEMBER: break; } } /* * Process a received IGMPv3 general, group-specific or * group-and-source-specific query. * Assumes m has already been pulled up to the full IGMP message length. * Return 0 if successful, otherwise an appropriate error code is returned. */ static int igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, /*const*/ struct igmpv3 *igmpv3) { struct igmp_ifsoftc *igi; struct in_multi *inm; int is_general_query; uint32_t maxresp, nsrc, qqi; uint16_t timer; uint8_t qrv; is_general_query = 0; CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ if (maxresp >= 128) { maxresp = IGMP_MANT(igmpv3->igmp_code) << (IGMP_EXP(igmpv3->igmp_code) + 3); } /* * Robustness must never be less than 2 for on-wire IGMPv3. * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make * an exception for interfaces whose IGMPv3 state changes * are redirected to loopback (e.g. MANET). */ qrv = IGMP_QRV(igmpv3->igmp_misc); if (qrv < 2) { CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, qrv, IGMP_RV_INIT); qrv = IGMP_RV_INIT; } qqi = igmpv3->igmp_qqi; if (qqi >= 128) { qqi = IGMP_MANT(igmpv3->igmp_qqi) << (IGMP_EXP(igmpv3->igmp_qqi) + 3); } timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE; if (timer == 0) timer = 1; nsrc = ntohs(igmpv3->igmp_numsrc); /* * Validate address fields and versions upfront before * accepting v3 query. * XXX SMPng: Unlocked access to igmpstat counters here. */ if (in_nullhost(igmpv3->igmp_group)) { /* * IGMPv3 General Query. * * General Queries SHOULD be directed to 224.0.0.1. * A general query with a source list has undefined * behaviour; discard it. */ IGMPSTAT_INC(igps_rcv_gen_queries); if (!in_allhosts(ip->ip_dst) || nsrc > 0) { IGMPSTAT_INC(igps_rcv_badqueries); return (0); } is_general_query = 1; } else { /* Group or group-source specific query. */ if (nsrc == 0) IGMPSTAT_INC(igps_rcv_group_queries); else IGMPSTAT_INC(igps_rcv_gsr_queries); } IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Discard the v3 query if we're in Compatibility Mode. * The RFC is not obviously worded that hosts need to stay in * compatibility mode until the Old Version Querier Present * timer expires. */ if (igi->igi_version != IGMP_VERSION_3) { CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", igi->igi_version, ifp, ifp->if_xname); goto out_locked; } igmp_set_version(igi, IGMP_VERSION_3); igi->igi_rv = qrv; igi->igi_qi = qqi; igi->igi_qri = maxresp; CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, maxresp); if (is_general_query) { /* * Schedule a current-state report on this ifp for * all groups, possibly containing source lists. * If there is a pending General Query response * scheduled earlier than the selected delay, do * not schedule any other reports. * Otherwise, reset the interface timer. */ CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", ifp, ifp->if_xname); if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); V_interface_timers_running = 1; } } else { /* * Group-source-specific queries are throttled on * a per-group basis to defeat denial-of-service attempts. * Queries for groups we are not a member of on this * link are simply ignored. */ inm = inm_lookup(ifp, igmpv3->igmp_group); if (inm == NULL) goto out_locked; if (nsrc > 0) { if (!ratecheck(&inm->inm_lastgsrtv, &V_igmp_gsrdelay)) { CTR1(KTR_IGMPV3, "%s: GS query throttled.", __func__); IGMPSTAT_INC(igps_drop_gsr_queries); goto out_locked; } } CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)", ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname); /* * If there is a pending General Query response * scheduled sooner than the selected delay, no * further report need be scheduled. * Otherwise, prepare to respond to the * group-specific or group-and-source query. */ if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) igmp_input_v3_group_query(inm, igi, timer, igmpv3); } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv3 group-specific or group-and-source-specific * query. * Return <0 if any error occurred. Currently this is ignored. */ static int igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, int timer, /*const*/ struct igmpv3 *igmpv3) { int retval; uint16_t nsrc; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); retval = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LEAVING_MEMBER: return (retval); break; case IGMP_REPORTING_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: break; } nsrc = ntohs(igmpv3->igmp_numsrc); /* * Deal with group-specific queries upfront. * If any group query is already pending, purge any recorded * source-list state if it exists, and schedule a query response * for this group-specific query. */ if (nsrc == 0) { if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { inm_clear_recorded(inm); timer = min(inm->inm_timer, timer); } inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; return (retval); } /* * Deal with the case where a group-and-source-specific query has * been received but a group-specific query is already pending. */ if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { timer = min(inm->inm_timer, timer); inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; return (retval); } /* * Finally, deal with the case where a group-and-source-specific * query has been received, where a response to a previous g-s-r * query exists, or none exists. * In this case, we need to parse the source-list which the Querier * has provided us with and check if we have any source list filter * entries at T1 for these sources. If we do not, there is no need * schedule a report and the query may be dropped. * If we do, we must record them and schedule a current-state * report for those sources. * FIXME: Handling source lists larger than 1 mbuf requires that * we pass the mbuf chain pointer down to this function, and use * m_getptr() to walk the chain. */ if (inm->inm_nsrc > 0) { const struct in_addr *ap; int i, nrecorded; ap = (const struct in_addr *)(igmpv3 + 1); nrecorded = 0; for (i = 0; i < nsrc; i++, ap++) { retval = inm_record_source(inm, ap->s_addr); if (retval < 0) break; nrecorded += retval; } if (nrecorded > 0) { CTR1(KTR_IGMPV3, "%s: schedule response to SG query", __func__); inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; } } return (retval); } /* * Process a received IGMPv1 host membership report. * * NOTE: 0.0.0.0 workaround breaks const correctness. */ static int igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, /*const*/ struct igmp *igmp) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_multi *inm; IGMPSTAT_INC(igps_rcv_reports); if (ifp->if_flags & IFF_LOOPBACK) return (0); if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || !in_hosteq(igmp->igmp_group, ip->ip_dst)) { IGMPSTAT_INC(igps_rcv_badreports); return (EINVAL); } /* * RFC 3376, Section 4.2.13, 9.2, 9.3: * Booting clients may use the source address 0.0.0.0. Some * IGMP daemons may not know how to use IP_RECVIF to determine * the interface upon which this message was received. * Replace 0.0.0.0 with the subnet address if told to do so. */ if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { NET_EPOCH_ENTER(); IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL) ip->ip_src.s_addr = htonl(ia->ia_subnet); NET_EPOCH_EXIT(); } CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); /* * IGMPv1 report suppression. * If we are a member of this group, and our membership should be * reported, stop our group timer and transition to the 'lazy' state. */ IN_MULTI_LIST_LOCK(); inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { struct igmp_ifsoftc *igi; igi = inm->inm_igi; if (igi == NULL) { KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); goto out_locked; } IGMPSTAT_INC(igps_rcv_ourreports); /* * If we are in IGMPv3 host mode, do not allow the * other host's IGMPv1 report to suppress our reports * unless explicitly configured to do so. */ if (igi->igi_version == IGMP_VERSION_3) { if (V_igmp_legacysupp) igmp_v3_suppress_group_record(inm); goto out_locked; } inm->inm_timer = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); case IGMP_SLEEPING_MEMBER: inm->inm_state = IGMP_SLEEPING_MEMBER; break; case IGMP_REPORTING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); if (igi->igi_version == IGMP_VERSION_1) inm->inm_state = IGMP_LAZY_MEMBER; else if (igi->igi_version == IGMP_VERSION_2) inm->inm_state = IGMP_SLEEPING_MEMBER; break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } out_locked: IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv2 host membership report. * * NOTE: 0.0.0.0 workaround breaks const correctness. */ static int igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, /*const*/ struct igmp *igmp) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_multi *inm; /* * Make sure we don't hear our own membership report. Fast * leave requires knowing that we are the only member of a * group. */ NET_EPOCH_ENTER(); IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { NET_EPOCH_EXIT(); return (0); } IGMPSTAT_INC(igps_rcv_reports); if (ifp->if_flags & IFF_LOOPBACK) { NET_EPOCH_EXIT(); return (0); } if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || !in_hosteq(igmp->igmp_group, ip->ip_dst)) { NET_EPOCH_EXIT(); IGMPSTAT_INC(igps_rcv_badreports); return (EINVAL); } /* * RFC 3376, Section 4.2.13, 9.2, 9.3: * Booting clients may use the source address 0.0.0.0. Some * IGMP daemons may not know how to use IP_RECVIF to determine * the interface upon which this message was received. * Replace 0.0.0.0 with the subnet address if told to do so. */ if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { if (ia != NULL) ip->ip_src.s_addr = htonl(ia->ia_subnet); } NET_EPOCH_EXIT(); CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); /* * IGMPv2 report suppression. * If we are a member of this group, and our membership should be * reported, and our group timer is pending or about to be reset, * stop our group timer by transitioning to the 'lazy' state. */ IN_MULTI_LIST_LOCK(); inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { struct igmp_ifsoftc *igi; igi = inm->inm_igi; KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); IGMPSTAT_INC(igps_rcv_ourreports); /* * If we are in IGMPv3 host mode, do not allow the * other host's IGMPv1 report to suppress our reports * unless explicitly configured to do so. */ if (igi->igi_version == IGMP_VERSION_3) { if (V_igmp_legacysupp) igmp_v3_suppress_group_record(inm); goto out_locked; } inm->inm_timer = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_AWAKENING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); case IGMP_LAZY_MEMBER: inm->inm_state = IGMP_LAZY_MEMBER; break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } out_locked: IN_MULTI_LIST_UNLOCK(); return (0); } int igmp_input(struct mbuf **mp, int *offp, int proto) { int iphlen; struct ifnet *ifp; struct igmp *igmp; struct ip *ip; struct mbuf *m; int igmplen; int minlen; int queryver; CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); m = *mp; ifp = m->m_pkthdr.rcvif; *mp = NULL; IGMPSTAT_INC(igps_rcv_total); ip = mtod(m, struct ip *); iphlen = *offp; igmplen = ntohs(ip->ip_len) - iphlen; /* * Validate lengths. */ if (igmplen < IGMP_MINLEN) { IGMPSTAT_INC(igps_rcv_tooshort); m_freem(m); return (IPPROTO_DONE); } /* * Always pullup to the minimum size for v1/v2 or v3 * to amortize calls to m_pullup(). */ minlen = iphlen; if (igmplen >= IGMP_V3_QUERY_MINLEN) minlen += IGMP_V3_QUERY_MINLEN; else minlen += IGMP_MINLEN; if ((!M_WRITABLE(m) || m->m_len < minlen) && (m = m_pullup(m, minlen)) == NULL) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); /* * Validate checksum. */ m->m_data += iphlen; m->m_len -= iphlen; igmp = mtod(m, struct igmp *); if (in_cksum(m, igmplen)) { IGMPSTAT_INC(igps_rcv_badsum); m_freem(m); return (IPPROTO_DONE); } m->m_data -= iphlen; m->m_len += iphlen; /* * IGMP control traffic is link-scope, and must have a TTL of 1. * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; * probe packets may come from beyond the LAN. */ if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { IGMPSTAT_INC(igps_rcv_badttl); m_freem(m); return (IPPROTO_DONE); } switch (igmp->igmp_type) { case IGMP_HOST_MEMBERSHIP_QUERY: if (igmplen == IGMP_MINLEN) { if (igmp->igmp_code == 0) queryver = IGMP_VERSION_1; else queryver = IGMP_VERSION_2; } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { queryver = IGMP_VERSION_3; } else { IGMPSTAT_INC(igps_rcv_tooshort); m_freem(m); return (IPPROTO_DONE); } switch (queryver) { case IGMP_VERSION_1: IGMPSTAT_INC(igps_rcv_v1v2_queries); if (!V_igmp_v1enable) break; if (igmp_input_v1_query(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_VERSION_2: IGMPSTAT_INC(igps_rcv_v1v2_queries); if (!V_igmp_v2enable) break; if (igmp_input_v2_query(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_VERSION_3: { struct igmpv3 *igmpv3; uint16_t igmpv3len; uint16_t nsrc; IGMPSTAT_INC(igps_rcv_v3_queries); igmpv3 = (struct igmpv3 *)igmp; /* * Validate length based on source count. */ nsrc = ntohs(igmpv3->igmp_numsrc); if (nsrc * sizeof(in_addr_t) > UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } /* * m_pullup() may modify m, so pullup in * this scope. */ igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + sizeof(struct in_addr) * nsrc; if ((!M_WRITABLE(m) || m->m_len < igmpv3len) && (m = m_pullup(m, igmpv3len)) == NULL) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) + iphlen); if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { m_freem(m); return (IPPROTO_DONE); } } break; } break; case IGMP_v1_HOST_MEMBERSHIP_REPORT: if (!V_igmp_v1enable) break; if (igmp_input_v1_report(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_v2_HOST_MEMBERSHIP_REPORT: if (!V_igmp_v2enable) break; if (!ip_checkrouteralert(m)) IGMPSTAT_INC(igps_rcv_nora); if (igmp_input_v2_report(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_v3_HOST_MEMBERSHIP_REPORT: /* * Hosts do not need to process IGMPv3 membership reports, * as report suppression is no longer required. */ if (!ip_checkrouteralert(m)) IGMPSTAT_INC(igps_rcv_nora); break; default: break; } /* * Pass all valid IGMP packets up to any process(es) listening on a * raw IGMP socket. */ *mp = m; return (rip_input(mp, offp, proto)); } /* * Fast timeout handler (global). * VIMAGE: Timeout handlers are expected to service all vimages. */ void igmp_fasttimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); igmp_fasttimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Fast timeout handler (per-vnet). * Sends are shuffled off to a netisr to deal with Giant. * * VIMAGE: Assume caller has set up our curvnet. */ static void igmp_fasttimo_vnet(void) { struct mbufq scq; /* State-change packets */ struct mbufq qrq; /* Query response packets */ struct ifnet *ifp; struct igmp_ifsoftc *igi; struct ifmultiaddr *ifma, *next; struct in_multi *inm; struct in_multi_head inm_free_tmp; int loop, uri_fasthz; loop = 0; uri_fasthz = 0; /* * Quick check to see if any work needs to be done, in order to * minimize the overhead of fasttimo processing. * SMPng: XXX Unlocked reads. */ if (!V_current_state_timers_running && !V_interface_timers_running && !V_state_change_timers_running) return; SLIST_INIT(&inm_free_tmp); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); /* * IGMPv3 General Query response timer processing. */ if (V_interface_timers_running) { CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); V_interface_timers_running = 0; LIST_FOREACH(igi, &V_igi_head, igi_link) { if (igi->igi_v3_timer == 0) { /* Do nothing. */ } else if (--igi->igi_v3_timer == 0) { igmp_v3_dispatch_general_query(igi); } else { V_interface_timers_running = 1; } } } if (!V_current_state_timers_running && !V_state_change_timers_running) goto out_locked; V_current_state_timers_running = 0; V_state_change_timers_running = 0; CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); /* * IGMPv1/v2/v3 host report and state-change timer processing. * Note: Processing a v3 group timer may remove a node. */ LIST_FOREACH(igi, &V_igi_head, igi_link) { ifp = igi->igi_ifp; if (igi->igi_version == IGMP_VERSION_3) { loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * PR_FASTHZ); mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); } IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; switch (igi->igi_version) { case IGMP_VERSION_1: case IGMP_VERSION_2: igmp_v1v2_process_group_timer(inm, igi->igi_version); break; case IGMP_VERSION_3: igmp_v3_process_group_timers(&inm_free_tmp, &qrq, &scq, inm, uri_fasthz); break; } if (__predict_false(ifma_restart)) { ifma_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); if (igi->igi_version == IGMP_VERSION_3) { igmp_dispatch_queue(&qrq, 0, loop); igmp_dispatch_queue(&scq, 0, loop); /* * Free the in_multi reference(s) for this * IGMP lifecycle. */ inm_release_list_deferred(&inm_free_tmp); } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); } /* * Update host report group timer for IGMPv1/v2. * Will update the global pending timer flags. */ static void igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) { int report_timer_expired; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); if (inm->inm_timer == 0) { report_timer_expired = 0; } else if (--inm->inm_timer == 0) { report_timer_expired = 1; } else { V_current_state_timers_running = 1; return; } switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: break; case IGMP_REPORTING_MEMBER: if (report_timer_expired) { inm->inm_state = IGMP_IDLE_MEMBER; (void)igmp_v1v2_queue_report(inm, (version == IGMP_VERSION_2) ? IGMP_v2_HOST_MEMBERSHIP_REPORT : IGMP_v1_HOST_MEMBERSHIP_REPORT); } break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } /* * Update a group's timers for IGMPv3. * Will update the global pending timer flags. * Note: Unlocked read from igi. */ static void igmp_v3_process_group_timers(struct in_multi_head *inmh, struct mbufq *qrq, struct mbufq *scq, struct in_multi *inm, const int uri_fasthz) { int query_response_timer_expired; int state_change_retransmit_timer_expired; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); query_response_timer_expired = 0; state_change_retransmit_timer_expired = 0; /* * During a transition from v1/v2 compatibility mode back to v3, * a group record in REPORTING state may still have its group * timer active. This is a no-op in this function; it is easier * to deal with it here than to complicate the slow-timeout path. */ if (inm->inm_timer == 0) { query_response_timer_expired = 0; } else if (--inm->inm_timer == 0) { query_response_timer_expired = 1; } else { V_current_state_timers_running = 1; } if (inm->inm_sctimer == 0) { state_change_retransmit_timer_expired = 0; } else if (--inm->inm_sctimer == 0) { state_change_retransmit_timer_expired = 1; } else { V_state_change_timers_running = 1; } /* We are in fasttimo, so be quick about it. */ if (!state_change_retransmit_timer_expired && !query_response_timer_expired) return; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: case IGMP_IDLE_MEMBER: break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: /* * Respond to a previously pending Group-Specific * or Group-and-Source-Specific query by enqueueing * the appropriate Current-State report for * immediate transmission. */ if (query_response_timer_expired) { int retval __unused; retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); inm->inm_state = IGMP_REPORTING_MEMBER; /* XXX Clear recorded sources for next time. */ inm_clear_recorded(inm); } /* FALLTHROUGH */ case IGMP_REPORTING_MEMBER: case IGMP_LEAVING_MEMBER: if (state_change_retransmit_timer_expired) { /* * State-change retransmission timer fired. * If there are any further pending retransmissions, * set the global pending state-change flag, and * reset the timer. */ if (--inm->inm_scrv > 0) { inm->inm_sctimer = uri_fasthz; V_state_change_timers_running = 1; } /* * Retransmit the previously computed state-change * report. If there are no further pending * retransmissions, the mbuf queue will be consumed. * Update T0 state to T1 as we have now sent * a state-change. */ (void)igmp_v3_merge_state_changes(inm, scq); inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); /* * If we are leaving the group for good, make sure * we release IGMP's reference to it. * This release must be deferred using a SLIST, * as we are called from a loop which traverses * the in_ifmultiaddr TAILQ. */ if (inm->inm_state == IGMP_LEAVING_MEMBER && inm->inm_scrv == 0) { inm->inm_state = IGMP_NOT_MEMBER; inm_rele_locked(inmh, inm); } } break; } } /* * Suppress a group's pending response to a group or source/group query. * * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. * Do NOT update ST1/ST0 as this operation merely suppresses * the currently pending group record. * Do NOT suppress the response to a general query. It is possible but * it would require adding another state or flag. */ static void igmp_v3_suppress_group_record(struct in_multi *inm) { IN_MULTI_LIST_LOCK_ASSERT(); KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, ("%s: not IGMPv3 mode on link", __func__)); if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) return; if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) inm_clear_recorded(inm); inm->inm_timer = 0; inm->inm_state = IGMP_REPORTING_MEMBER; } /* * Switch to a different IGMP version on the given interface, * as per Section 7.2.1. */ static void igmp_set_version(struct igmp_ifsoftc *igi, const int version) { int old_version_timer; IGMP_LOCK_ASSERT(); CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, version, igi->igi_ifp, igi->igi_ifp->if_xname); if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { /* * Compute the "Older Version Querier Present" timer as per * Section 8.12. */ old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; old_version_timer *= PR_SLOWHZ; if (version == IGMP_VERSION_1) { igi->igi_v1_timer = old_version_timer; igi->igi_v2_timer = 0; } else if (version == IGMP_VERSION_2) { igi->igi_v1_timer = 0; igi->igi_v2_timer = old_version_timer; } } if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { if (igi->igi_version != IGMP_VERSION_2) { igi->igi_version = IGMP_VERSION_2; igmp_v3_cancel_link_timers(igi); } } else if (igi->igi_v1_timer > 0) { if (igi->igi_version != IGMP_VERSION_1) { igi->igi_version = IGMP_VERSION_1; igmp_v3_cancel_link_timers(igi); } } } /* * Cancel pending IGMPv3 timers for the given link and all groups * joined on it; state-change, general-query, and group-query timers. * * Only ever called on a transition from v3 to Compatibility mode. Kill * the timers stone dead (this may be expensive for large N groups), they * will be restarted if Compatibility Mode deems that they must be due to * query processing. */ static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) { struct ifmultiaddr *ifma; struct ifnet *ifp; struct in_multi *inm; struct in_multi_head inm_free_tmp; CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, igi->igi_ifp, igi->igi_ifp->if_xname); IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); SLIST_INIT(&inm_free_tmp); /* * Stop the v3 General Query Response on this link stone dead. * If fasttimo is woken up due to V_interface_timers_running, * the flag will be cleared if there are no pending link timers. */ igi->igi_v3_timer = 0; /* * Now clear the current-state and state-change report timers * for all memberships scoped to this link. */ ifp = igi->igi_ifp; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: /* * These states are either not relevant in v3 mode, * or are unreported. Do nothing. */ break; case IGMP_LEAVING_MEMBER: /* * If we are leaving the group and switching to * compatibility mode, we need to release the final * reference held for issuing the INCLUDE {}, and * transition to REPORTING to ensure the host leave * message is sent upstream to the old querier -- * transition to NOT would lose the leave and race. */ inm_rele_locked(&inm_free_tmp, inm); /* FALLTHROUGH */ case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: inm_clear_recorded(inm); /* FALLTHROUGH */ case IGMP_REPORTING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; break; } /* * Always clear state-change and group report timers. * Free any pending IGMPv3 state-change records. */ inm->inm_sctimer = 0; inm->inm_timer = 0; mbufq_drain(&inm->inm_scq); } IF_ADDR_RUNLOCK(ifp); inm_release_list_deferred(&inm_free_tmp); } /* * Update the Older Version Querier Present timers for a link. * See Section 7.2.1 of RFC 3376. */ static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) { IGMP_LOCK_ASSERT(); if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { /* * IGMPv1 and IGMPv2 Querier Present timers expired. * * Revert to IGMPv3. */ if (igi->igi_version != IGMP_VERSION_3) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_version = IGMP_VERSION_3; } } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { /* * IGMPv1 Querier Present timer expired, * IGMPv2 Querier Present timer running. * If IGMPv2 was disabled since last timeout, * revert to IGMPv3. * If IGMPv2 is enabled, revert to IGMPv2. */ if (!V_igmp_v2enable) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v2_timer = 0; igi->igi_version = IGMP_VERSION_3; } else { --igi->igi_v2_timer; if (igi->igi_version != IGMP_VERSION_2) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_2, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_version = IGMP_VERSION_2; igmp_v3_cancel_link_timers(igi); } } } else if (igi->igi_v1_timer > 0) { /* * IGMPv1 Querier Present timer running. * Stop IGMPv2 timer if running. * * If IGMPv1 was disabled since last timeout, * revert to IGMPv3. * If IGMPv1 is enabled, reset IGMPv2 timer if running. */ if (!V_igmp_v1enable) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v1_timer = 0; igi->igi_version = IGMP_VERSION_3; } else { --igi->igi_v1_timer; } if (igi->igi_v2_timer > 0) { CTR3(KTR_IGMPV3, "%s: cancel v2 timer on %p(%s)", __func__, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v2_timer = 0; } } } /* * Global slowtimo handler. * VIMAGE: Timeout handlers are expected to service all vimages. */ void igmp_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); igmp_slowtimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Per-vnet slowtimo handler. */ static void igmp_slowtimo_vnet(void) { struct igmp_ifsoftc *igi; IGMP_LOCK(); LIST_FOREACH(igi, &V_igi_head, igi_link) { igmp_v1v2_process_querier_timers(igi); } IGMP_UNLOCK(); } /* * Dispatch an IGMPv1/v2 host report or leave message. * These are always small enough to fit inside a single mbuf. */ static int igmp_v1v2_queue_report(struct in_multi *inm, const int type) { struct ifnet *ifp; struct igmp *igmp; struct ip *ip; struct mbuf *m; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); ifp = inm->inm_ifp; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOMEM); M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); m->m_data += sizeof(struct ip); m->m_len = sizeof(struct igmp); igmp = mtod(m, struct igmp *); igmp->igmp_type = type; igmp->igmp_code = 0; igmp->igmp_group = inm->inm_addr; igmp->igmp_cksum = 0; igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); m->m_data -= sizeof(struct ip); m->m_len += sizeof(struct ip); ip = mtod(m, struct ip *); ip->ip_tos = 0; ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); ip->ip_off = 0; ip->ip_p = IPPROTO_IGMP; ip->ip_src.s_addr = INADDR_ANY; if (type == IGMP_HOST_LEAVE_MESSAGE) ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); else ip->ip_dst = inm->inm_addr; igmp_save_context(m, ifp); m->m_flags |= M_IGMPV2; if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) m->m_flags |= M_IGMP_LOOP; CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); netisr_dispatch(NETISR_IGMP, m); return (0); } /* * Process a state change from the upper layer for the given IPv4 group. * * Each socket holds a reference on the in_multi in its own ip_moptions. * The socket layer will have made the necessary updates to.the group * state, it is now up to IGMP to issue a state change report if there * has been any change between T0 (when the last state-change was issued) * and T1 (now). * * We use the IGMPv3 state machine at group level. The IGMP module * however makes the decision as to which IGMP protocol version to speak. * A state change *from* INCLUDE {} always means an initial join. * A state change *to* INCLUDE {} always means a final leave. * * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can * save ourselves a bunch of work; any exclusive mode groups need not * compute source filter lists. * * VIMAGE: curvnet should have been set by caller, as this routine * is called from the socket option handlers. */ int igmp_change_state(struct in_multi *inm) { struct igmp_ifsoftc *igi; struct ifnet *ifp; int error; error = 0; IN_MULTI_LOCK_ASSERT(); /* * Try to detect if the upper layer just asked us to change state * for an interface which has now gone away. */ KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); ifp = inm->inm_ifma->ifma_ifp; /* * Sanity check that netinet's notion of ifp is the * same as net's. */ KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); /* * If we detect a state transition to or from MCAST_UNDEFINED * for this group, then we are starting or finishing an IGMP * life cycle for this group. */ if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_IGMPV3, "%s: initial join", __func__); error = igmp_initial_join(inm, igi); goto out_locked; } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_IGMPV3, "%s: final leave", __func__); igmp_final_leave(inm, igi); goto out_locked; } } else { CTR1(KTR_IGMPV3, "%s: filter set change", __func__); } error = igmp_handle_state_change(inm, igi); out_locked: IGMP_UNLOCK(); return (error); } /* * Perform the initial join for an IGMP group. * * When joining a group: * If the group should have its IGMP traffic suppressed, do nothing. * IGMPv1 starts sending IGMPv1 host membership reports. * IGMPv2 starts sending IGMPv2 host membership reports. * IGMPv3 will schedule an IGMPv3 state-change report containing the * initial state of the membership. */ static int igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) { struct ifnet *ifp; struct mbufq *mq; int error, retval, syncstates; CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); error = 0; syncstates = 1; ifp = inm->inm_ifp; IN_MULTI_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); /* * Groups joined on loopback or marked as 'not reported', * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and * are never reported in any IGMP protocol exchanges. * All other groups enter the appropriate IGMP state machine * for the version in use on this link. * A link marked as IGIF_SILENT causes IGMP to be completely * disabled for the link. */ if ((ifp->if_flags & IFF_LOOPBACK) || (igi->igi_flags & IGIF_SILENT) || !igmp_isgroupreported(inm->inm_addr)) { CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); inm->inm_state = IGMP_SILENT_MEMBER; inm->inm_timer = 0; } else { /* * Deal with overlapping in_multi lifecycle. * If this group was LEAVING, then make sure * we drop the reference we picked up to keep the * group around for the final INCLUDE {} enqueue. */ if (igi->igi_version == IGMP_VERSION_3 && inm->inm_state == IGMP_LEAVING_MEMBER) { MPASS(inm->inm_refcount > 1); inm_rele_locked(NULL, inm); } inm->inm_state = IGMP_REPORTING_MEMBER; switch (igi->igi_version) { case IGMP_VERSION_1: case IGMP_VERSION_2: inm->inm_state = IGMP_IDLE_MEMBER; error = igmp_v1v2_queue_report(inm, (igi->igi_version == IGMP_VERSION_2) ? IGMP_v2_HOST_MEMBERSHIP_REPORT : IGMP_v1_HOST_MEMBERSHIP_REPORT); if (error == 0) { inm->inm_timer = IGMP_RANDOM_DELAY( IGMP_V1V2_MAX_RI * PR_FASTHZ); V_current_state_timers_running = 1; } break; case IGMP_VERSION_3: /* * Defer update of T0 to T1, until the first copy * of the state change has been transmitted. */ syncstates = 0; /* * Immediately enqueue a State-Change Report for * this interface, freeing any previous reports. * Don't kick the timers if there is nothing to do, * or if an error occurred. */ mq = &inm->inm_scq; mbufq_drain(mq); retval = igmp_v3_enqueue_group_record(mq, inm, 1, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) { error = retval * -1; break; } /* * Schedule transmission of pending state-change * report up to RV times for this link. The timer * will fire at the next igmp_fasttimo (~200ms), * giving us an opportunity to merge the reports. */ if (igi->igi_flags & IGIF_LOOPBACK) { inm->inm_scrv = 1; } else { KASSERT(igi->igi_rv > 1, ("%s: invalid robustness %d", __func__, igi->igi_rv)); inm->inm_scrv = igi->igi_rv; } inm->inm_sctimer = 1; V_state_change_timers_running = 1; error = 0; break; } } /* * Only update the T0 state if state change is atomic, * i.e. we don't need to wait for a timer to fire before we * can consider the state change to have been communicated. */ if (syncstates) { inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); } return (error); } /* * Issue an intermediate state change during the IGMP life-cycle. */ static int igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) { struct ifnet *ifp; int retval; CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); ifp = inm->inm_ifp; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); if ((ifp->if_flags & IFF_LOOPBACK) || (igi->igi_flags & IGIF_SILENT) || !igmp_isgroupreported(inm->inm_addr) || (igi->igi_version != IGMP_VERSION_3)) { if (!igmp_isgroupreported(inm->inm_addr)) { CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); } CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); return (0); } mbufq_drain(&inm->inm_scq); retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) return (-retval); /* * If record(s) were enqueued, start the state-change * report timer for this group. */ inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); inm->inm_sctimer = 1; V_state_change_timers_running = 1; return (0); } /* * Perform the final leave for an IGMP group. * * When leaving a group: * IGMPv1 does nothing. * IGMPv2 sends a host leave message, if and only if we are the reporter. * IGMPv3 enqueues a state-change report containing a transition * to INCLUDE {} for immediate transmission. */ static void igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) { int syncstates; syncstates = 1; CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_LEAVING_MEMBER: /* Already leaving or left; do nothing. */ CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: if (igi->igi_version == IGMP_VERSION_2) { #ifdef INVARIANTS if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) panic("%s: IGMPv3 state reached, not IGMPv3 mode", __func__); #endif igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); inm->inm_state = IGMP_NOT_MEMBER; } else if (igi->igi_version == IGMP_VERSION_3) { /* * Stop group timer and all pending reports. * Immediately enqueue a state-change report * TO_IN {} to be sent on the next fast timeout, * giving us an opportunity to merge reports. */ mbufq_drain(&inm->inm_scq); inm->inm_timer = 0; if (igi->igi_flags & IGIF_LOOPBACK) { inm->inm_scrv = 1; } else { inm->inm_scrv = igi->igi_rv; } CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d " "pending retransmissions.", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, inm->inm_scrv); if (inm->inm_scrv == 0) { inm->inm_state = IGMP_NOT_MEMBER; inm->inm_sctimer = 0; } else { int retval __unused; inm_acquire_locked(inm); retval = igmp_v3_enqueue_group_record( &inm->inm_scq, inm, 1, 0, 0); KASSERT(retval != 0, ("%s: enqueue record = %d", __func__, retval)); inm->inm_state = IGMP_LEAVING_MEMBER; inm->inm_sctimer = 1; V_state_change_timers_running = 1; syncstates = 0; } break; } break; case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: /* Our reports are suppressed; do nothing. */ break; } if (syncstates) { inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); } } /* * Enqueue an IGMPv3 group record to the given output queue. * * XXX This function could do with having the allocation code * split out, and the multiple-tree-walks coalesced into a single * routine as has been done in igmp_v3_enqueue_filter_change(). * * If is_state_change is zero, a current-state record is appended. * If is_state_change is non-zero, a state-change report is appended. * * If is_group_query is non-zero, an mbuf packet chain is allocated. * If is_group_query is zero, and if there is a packet with free space * at the tail of the queue, it will be appended to providing there * is enough free space. * Otherwise a new mbuf packet chain is allocated. * * If is_source_query is non-zero, each source is checked to see if * it was recorded for a Group-Source query, and will be omitted if * it is not both in-mode and recorded. * * The function will attempt to allocate leading space in the packet * for the IP/IGMP header to be prepended without fragmenting the chain. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, const int is_state_change, const int is_group_query, const int is_source_query) { struct igmp_grouprec ig; struct igmp_grouprec *pig; struct ifnet *ifp; struct ip_msource *ims, *nims; struct mbuf *m0, *m, *md; int is_filter_list_change; int minrec0len, m0srcs, msrcs, nbytes, off; int record_has_sources; int now; int type; in_addr_t naddr; uint8_t mode; IN_MULTI_LIST_LOCK_ASSERT(); ifp = inm->inm_ifp; is_filter_list_change = 0; m = NULL; m0 = NULL; m0srcs = 0; msrcs = 0; nbytes = 0; nims = NULL; record_has_sources = 1; pig = NULL; type = IGMP_DO_NOTHING; mode = inm->inm_st[1].iss_fmode; /* * If we did not transition out of ASM mode during t0->t1, * and there are no source nodes to process, we can skip * the generation of source records. */ if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && inm->inm_nsrc == 0) record_has_sources = 0; if (is_state_change) { /* * Queue a state change record. * If the mode did not change, and there are non-ASM * listeners or source filters present, * we potentially need to issue two records for the group. * If we are transitioning to MCAST_UNDEFINED, we need * not send any sources. * If there are ASM listeners, and there was no filter * mode transition of any kind, do nothing. */ if (mode != inm->inm_st[0].iss_fmode) { if (mode == MCAST_EXCLUDE) { CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", __func__); type = IGMP_CHANGE_TO_EXCLUDE_MODE; } else { CTR1(KTR_IGMPV3, "%s: change to INCLUDE", __func__); type = IGMP_CHANGE_TO_INCLUDE_MODE; if (mode == MCAST_UNDEFINED) record_has_sources = 0; } } else { if (record_has_sources) { is_filter_list_change = 1; } else { type = IGMP_DO_NOTHING; } } } else { /* * Queue a current state record. */ if (mode == MCAST_EXCLUDE) { type = IGMP_MODE_IS_EXCLUDE; } else if (mode == MCAST_INCLUDE) { type = IGMP_MODE_IS_INCLUDE; KASSERT(inm->inm_st[1].iss_asm == 0, ("%s: inm %p is INCLUDE but ASM count is %d", __func__, inm, inm->inm_st[1].iss_asm)); } } /* * Generate the filter list changes using a separate function. */ if (is_filter_list_change) return (igmp_v3_enqueue_filter_change(mq, inm)); if (type == IGMP_DO_NOTHING) { CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); return (0); } /* * If any sources are present, we must be able to fit at least * one in the trailing space of the tail packet's mbuf, * ideally more. */ minrec0len = sizeof(struct igmp_grouprec); if (record_has_sources) minrec0len += sizeof(in_addr_t); CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__, igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); /* * Check if we have a packet in the tail of the queue for this * group into which the first group record for this group will fit. * Otherwise allocate a new packet. * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. * Note: Group records for G/GSR query responses MUST be sent * in their own packet. */ m0 = mbufq_last(mq); if (!is_group_query && m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && (m0->m_pkthdr.len + minrec0len) < (ifp->if_mtu - IGMP_LEADINGSPACE)) { m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); m = m0; CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); } else { if (mbufq_full(mq)) { CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); return (-ENOMEM); } m = NULL; m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); if (!is_state_change && !is_group_query) { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); igmp_save_context(m, ifp); CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); } /* * Append group record. * If we have sources, we don't know how many yet. */ ig.ig_type = type; ig.ig_datalen = 0; ig.ig_numsrc = 0; ig.ig_group = inm->inm_addr; if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct igmp_grouprec); /* * Append as many sources as will fit in the first packet. * If we are appending to a new packet, the chain allocation * may potentially use clusters; use m_getptr() in this case. * If we are appending to an existing packet, we need to obtain * a pointer to the group record after m_append(), in case a new * mbuf was allocated. * Only append sources which are in-mode at t1. If we are * transitioning to MCAST_UNDEFINED state on the group, do not * include source entries. * Only report recorded sources in our filter set when responding * to a group-source query. */ if (record_has_sources) { if (m == m0) { md = m_last(m); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + md->m_len - nbytes); } else { md = m_getptr(m, 0, &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); } msrcs = 0; RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); if ((now != mode) || (now == mode && mode == MCAST_UNDEFINED)) { CTR1(KTR_IGMPV3, "%s: skip node", __func__); continue; } if (is_source_query && ims->ims_stp == 0) { CTR1(KTR_IGMPV3, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_IGMPV3, "%s: append node", __func__); naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(in_addr_t); ++msrcs; if (msrcs == m0srcs) break; } CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, msrcs); pig->ig_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(in_addr_t)); } if (is_source_query && msrcs == 0) { CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); if (m != m0) m_freem(m); return (0); } /* * We are good to go with first packet. */ if (m != m0) { CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); m->m_pkthdr.PH_vt.vt_nrecs = 1; mbufq_enqueue(mq, m); } else m->m_pkthdr.PH_vt.vt_nrecs++; /* * No further work needed if no source list in packet(s). */ if (!record_has_sources) return (nbytes); /* * Whilst sources remain to be announced, we need to allocate * a new packet and fill out as many sources as will fit. * Always try for a cluster first. */ while (nims != NULL) { if (mbufq_full(mq)) { CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); return (-ENOMEM); } m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); igmp_save_context(m, ifp); md = m_getptr(m, 0, &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 1; nbytes += sizeof(struct igmp_grouprec); m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); msrcs = 0; RB_FOREACH_FROM(ims, ip_msource_tree, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); if ((now != mode) || (now == mode && mode == MCAST_UNDEFINED)) { CTR1(KTR_IGMPV3, "%s: skip node", __func__); continue; } if (is_source_query && ims->ims_stp == 0) { CTR1(KTR_IGMPV3, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_IGMPV3, "%s: append node", __func__); naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } ++msrcs; if (msrcs == m0srcs) break; } pig->ig_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(in_addr_t)); CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); mbufq_enqueue(mq, m); } return (nbytes); } /* * Type used to mark record pass completion. * We exploit the fact we can cast to this easily from the * current filter modes on each ip_msource node. */ typedef enum { REC_NONE = 0x00, /* MCAST_UNDEFINED */ REC_ALLOW = 0x01, /* MCAST_INCLUDE */ REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ REC_FULL = REC_ALLOW | REC_BLOCK } rectype_t; /* * Enqueue an IGMPv3 filter list change to the given output queue. * * Source list filter state is held in an RB-tree. When the filter list * for a group is changed without changing its mode, we need to compute * the deltas between T0 and T1 for each source in the filter set, * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. * * As we may potentially queue two record types, and the entire R-B tree * needs to be walked at once, we break this out into its own function * so we can generate a tightly packed queue of packets. * * XXX This could be written to only use one tree walk, although that makes * serializing into the mbuf chains a bit harder. For now we do two walks * which makes things easier on us, and it may or may not be harder on * the L2 cache. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) { static const int MINRECLEN = sizeof(struct igmp_grouprec) + sizeof(in_addr_t); struct ifnet *ifp; struct igmp_grouprec ig; struct igmp_grouprec *pig; struct ip_msource *ims, *nims; struct mbuf *m, *m0, *md; in_addr_t naddr; int m0srcs, nbytes, npbytes, off, rsrcs, schanged; int nallow, nblock; uint8_t mode, now, then; rectype_t crt, drt, nrt; IN_MULTI_LIST_LOCK_ASSERT(); if (inm->inm_nsrc == 0 || (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) return (0); ifp = inm->inm_ifp; /* interface */ mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ crt = REC_NONE; /* current group record type */ drt = REC_NONE; /* mask of completed group record types */ nrt = REC_NONE; /* record type for current node */ m0srcs = 0; /* # source which will fit in current mbuf chain */ nbytes = 0; /* # of bytes appended to group's state-change queue */ npbytes = 0; /* # of bytes appended this packet */ rsrcs = 0; /* # sources encoded in current record */ schanged = 0; /* # nodes encoded in overall filter change */ nallow = 0; /* # of source entries in ALLOW_NEW */ nblock = 0; /* # of source entries in BLOCK_OLD */ nims = NULL; /* next tree node pointer */ /* * For each possible filter record mode. * The first kind of source we encounter tells us which * is the first kind of record we start appending. * If a node transitioned to UNDEFINED at t1, its mode is treated * as the inverse of the group's filter mode. */ while (drt != REC_FULL) { do { m0 = mbufq_last(mq); if (m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && (m0->m_pkthdr.len + MINRECLEN) < (ifp->if_mtu - IGMP_LEADINGSPACE)) { m = m0; m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); CTR1(KTR_IGMPV3, "%s: use previous packet", __func__); } else { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) { CTR1(KTR_IGMPV3, "%s: m_get*() failed", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 0; igmp_save_context(m, ifp); m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); npbytes = 0; CTR1(KTR_IGMPV3, "%s: allocated new packet", __func__); } /* * Append the IGMP group record header to the * current packet's data area. * Recalculate pointer to free space for next * group record, in case m_append() allocated * a new mbuf or cluster. */ memset(&ig, 0, sizeof(ig)); ig.ig_group = inm->inm_addr; if (!m_append(m, sizeof(ig), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed", __func__); return (-ENOMEM); } npbytes += sizeof(struct igmp_grouprec); if (m != m0) { /* new packet; offset in c hain */ md = m_getptr(m, npbytes - sizeof(struct igmp_grouprec), &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); } else { /* current packet; offset from last append */ md = m_last(m); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + md->m_len - sizeof(struct igmp_grouprec)); } /* * Begin walking the tree for this record type * pass, or continue from where we left off * previously if we had to allocate a new packet. * Only report deltas in-mode at t1. * We need not report included sources as allowed * if we are in inclusive mode on the group, * however the converse is not true. */ rsrcs = 0; if (nims == NULL) nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); RB_FOREACH_FROM(ims, ip_msource_tree, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); then = ims_get_mode(inm, ims, 0); CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", __func__, then, now); if (now == then) { CTR1(KTR_IGMPV3, "%s: skip unchanged", __func__); continue; } if (mode == MCAST_EXCLUDE && now == MCAST_INCLUDE) { CTR1(KTR_IGMPV3, "%s: skip IN src on EX group", __func__); continue; } nrt = (rectype_t)now; if (nrt == REC_NONE) nrt = (rectype_t)(~mode & REC_FULL); if (schanged++ == 0) { crt = nrt; } else if (crt != nrt) continue; naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed", __func__); return (-ENOMEM); } nallow += !!(crt == REC_ALLOW); nblock += !!(crt == REC_BLOCK); if (++rsrcs == m0srcs) break; } /* * If we did not append any tree nodes on this * pass, back out of allocations. */ if (rsrcs == 0) { npbytes -= sizeof(struct igmp_grouprec); if (m != m0) { CTR1(KTR_IGMPV3, "%s: m_free(m)", __func__); m_freem(m); } else { CTR1(KTR_IGMPV3, "%s: m_adj(m, -ig)", __func__); m_adj(m, -((int)sizeof( struct igmp_grouprec))); } continue; } npbytes += (rsrcs * sizeof(in_addr_t)); if (crt == REC_ALLOW) pig->ig_type = IGMP_ALLOW_NEW_SOURCES; else if (crt == REC_BLOCK) pig->ig_type = IGMP_BLOCK_OLD_SOURCES; pig->ig_numsrc = htons(rsrcs); /* * Count the new group record, and enqueue this * packet if it wasn't already queued. */ m->m_pkthdr.PH_vt.vt_nrecs++; if (m != m0) mbufq_enqueue(mq, m); nbytes += npbytes; } while (nims != NULL); drt |= crt; crt = (~crt & REC_FULL); } CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, nallow, nblock); return (nbytes); } static int igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) { struct mbufq *gq; struct mbuf *m; /* pending state-change */ struct mbuf *m0; /* copy of pending state-change */ struct mbuf *mt; /* last state-change in packet */ int docopy, domerge; u_int recslen; docopy = 0; domerge = 0; recslen = 0; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); /* * If there are further pending retransmissions, make a writable * copy of each queued state-change message before merging. */ if (inm->inm_scrv > 0) docopy = 1; gq = &inm->inm_scq; #ifdef KTR if (mbufq_first(gq) == NULL) { CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", __func__, inm); } #endif m = mbufq_first(gq); while (m != NULL) { /* * Only merge the report into the current packet if * there is sufficient space to do so; an IGMPv3 report * packet may only contain 65,535 group records. * Always use a simple mbuf chain concatentation to do this, * as large state changes for single groups may have * allocated clusters. */ domerge = 0; mt = mbufq_last(scq); if (mt != NULL) { recslen = m_length(m, NULL); if ((mt->m_pkthdr.PH_vt.vt_nrecs + m->m_pkthdr.PH_vt.vt_nrecs <= IGMP_V3_REPORT_MAXRECS) && (mt->m_pkthdr.len + recslen <= (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) domerge = 1; } if (!domerge && mbufq_full(gq)) { CTR2(KTR_IGMPV3, "%s: outbound queue full, skipping whole packet %p", __func__, m); mt = m->m_nextpkt; if (!docopy) m_freem(m); m = mt; continue; } if (!docopy) { CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); m0 = mbufq_dequeue(gq); m = m0->m_nextpkt; } else { CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); m0 = m_dup(m, M_NOWAIT); if (m0 == NULL) return (ENOMEM); m0->m_nextpkt = NULL; m = m->m_nextpkt; } if (!domerge) { CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", __func__, m0, scq); mbufq_enqueue(scq, m0); } else { struct mbuf *mtl; /* last mbuf of packet mt */ CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", __func__, m0, mt); mtl = m_last(mt); m0->m_flags &= ~M_PKTHDR; mt->m_pkthdr.len += recslen; mt->m_pkthdr.PH_vt.vt_nrecs += m0->m_pkthdr.PH_vt.vt_nrecs; mtl->m_next = m0; } } return (0); } /* * Respond to a pending IGMPv3 General Query. */ static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) { struct ifmultiaddr *ifma; struct ifnet *ifp; struct in_multi *inm; int retval __unused, loop; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); KASSERT(igi->igi_version == IGMP_VERSION_3, ("%s: called when version %d", __func__, igi->igi_version)); /* * Check that there are some packets queued. If so, send them first. * For large number of groups the reply to general query can take * many packets, we should finish sending them before starting of * queuing the new reply. */ if (mbufq_len(&igi->igi_gq) != 0) goto send; ifp = igi->igi_ifp; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; KASSERT(ifp == inm->inm_ifp, ("%s: inconsistent ifp", __func__)); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; retval = igmp_v3_enqueue_group_record(&igi->igi_gq, inm, 0, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } IF_ADDR_RUNLOCK(ifp); send: loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); /* * Slew transmission of bursts over 500ms intervals. */ if (mbufq_first(&igi->igi_gq) != NULL) { igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( IGMP_RESPONSE_BURST_INTERVAL); V_interface_timers_running = 1; } } /* * Transmit the next pending IGMP message in the output queue. * * We get called from netisr_processqueue(). A mutex private to igmpoq * will be acquired and released around this routine. * * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. * MRT: Nothing needs to be done, as IGMP traffic is always local to * a link and uses a link-scope multicast address. */ static void igmp_intr(struct mbuf *m) { struct ip_moptions imo; struct ifnet *ifp; struct mbuf *ipopts, *m0; int error; uint32_t ifindex; CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); /* * Set VNET image pointer from enqueued mbuf chain * before doing anything else. Whilst we use interface * indexes to guard against interface detach, they are * unique to each VIMAGE and must be retrieved. */ CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); ifindex = igmp_restore_context(m); /* * Check if the ifnet still exists. This limits the scope of * any race in the absence of a global ifp lock for low cost * (an array lookup). */ ifp = ifnet_byindex(ifindex); if (ifp == NULL) { CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", __func__, m, ifindex); m_freem(m); IPSTAT_INC(ips_noroute); goto out; } ipopts = V_igmp_sendra ? m_raopt : NULL; imo.imo_multicast_ttl = 1; imo.imo_multicast_vif = -1; imo.imo_multicast_loop = (V_ip_mrouter != NULL); /* * If the user requested that IGMP traffic be explicitly * redirected to the loopback interface (e.g. they are running a * MANET interface and the routing protocol needs to see the * updates), handle this now. */ if (m->m_flags & M_IGMP_LOOP) imo.imo_multicast_ifp = V_loif; else imo.imo_multicast_ifp = ifp; if (m->m_flags & M_IGMPV2) { m0 = m; } else { m0 = igmp_v3_encap_report(ifp, m); if (m0 == NULL) { CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); m_freem(m); IPSTAT_INC(ips_odropped); goto out; } } igmp_scrub_context(m0); m_clrprotoflags(m); m0->m_pkthdr.rcvif = V_loif; #ifdef MAC mac_netinet_igmp_send(ifp, m0); #endif error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); if (error) { CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); goto out; } IGMPSTAT_INC(igps_snd_reports); out: /* * We must restore the existing vnet pointer before * continuing as we are run from netisr context. */ CURVNET_RESTORE(); } /* * Encapsulate an IGMPv3 report. * * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf * chain has already had its IP/IGMPv3 header prepended. In this case * the function will not attempt to prepend; the lengths and checksums * will however be re-computed. * * Returns a pointer to the new mbuf chain head, or NULL if the * allocation failed. */ static struct mbuf * igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) { struct rm_priotracker in_ifa_tracker; struct igmp_report *igmp; struct ip *ip; int hdrlen, igmpreclen; KASSERT((m->m_flags & M_PKTHDR), ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); igmpreclen = m_length(m, NULL); hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); if (m->m_flags & M_IGMPV3_HDR) { igmpreclen -= hdrlen; } else { M_PREPEND(m, hdrlen, M_NOWAIT); if (m == NULL) return (NULL); m->m_flags |= M_IGMPV3_HDR; } CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); m->m_data += sizeof(struct ip); m->m_len -= sizeof(struct ip); igmp = mtod(m, struct igmp_report *); igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; igmp->ir_rsv1 = 0; igmp->ir_rsv2 = 0; igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs); igmp->ir_cksum = 0; igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); m->m_pkthdr.PH_vt.vt_nrecs = 0; m->m_data -= sizeof(struct ip); m->m_len += sizeof(struct ip); ip = mtod(m, struct ip *); ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; ip->ip_len = htons(hdrlen + igmpreclen); ip->ip_off = htons(IP_DF); ip->ip_p = IPPROTO_IGMP; ip->ip_sum = 0; ip->ip_src.s_addr = INADDR_ANY; if (m->m_flags & M_IGMP_LOOP) { struct in_ifaddr *ia; NET_EPOCH_ENTER(); IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL) ip->ip_src = ia->ia_addr.sin_addr; NET_EPOCH_EXIT(); } ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); return (m); } #ifdef KTR static char * igmp_rec_type_to_str(const int type) { switch (type) { case IGMP_CHANGE_TO_EXCLUDE_MODE: return "TO_EX"; break; case IGMP_CHANGE_TO_INCLUDE_MODE: return "TO_IN"; break; case IGMP_MODE_IS_EXCLUDE: return "MODE_EX"; break; case IGMP_MODE_IS_INCLUDE: return "MODE_IN"; break; case IGMP_ALLOW_NEW_SOURCES: return "ALLOW_NEW"; break; case IGMP_BLOCK_OLD_SOURCES: return "BLOCK_OLD"; break; default: break; } return "unknown"; } #endif #ifdef VIMAGE static void vnet_igmp_init(const void *unused __unused) { netisr_register_vnet(&igmp_nh); } VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_igmp_init, NULL); static void vnet_igmp_uninit(const void *unused __unused) { /* This can happen when we shutdown the entire network stack. */ CTR1(KTR_IGMPV3, "%s: tearing down", __func__); netisr_unregister_vnet(&igmp_nh); } VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_igmp_uninit, NULL); #endif #ifdef DDB DB_SHOW_COMMAND(igi_list, db_show_igi_list) { struct igmp_ifsoftc *igi, *tigi; LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head; if (!have_addr) { db_printf("usage: show igi_list \n"); return; } igi_head = (struct _igi_list *)addr; LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) { db_printf("igmp_ifsoftc %p:\n", igi); db_printf(" ifp %p\n", igi->igi_ifp); db_printf(" version %u\n", igi->igi_version); db_printf(" v1_timer %u\n", igi->igi_v1_timer); db_printf(" v2_timer %u\n", igi->igi_v2_timer); db_printf(" v3_timer %u\n", igi->igi_v3_timer); db_printf(" flags %#x\n", igi->igi_flags); db_printf(" rv %u\n", igi->igi_rv); db_printf(" qi %u\n", igi->igi_qi); db_printf(" qri %u\n", igi->igi_qri); db_printf(" uri %u\n", igi->igi_uri); /* struct mbufq igi_gq; */ db_printf("\n"); } } #endif static int igmp_modevent(module_t mod, int type, void *unused __unused) { switch (type) { case MOD_LOAD: CTR1(KTR_IGMPV3, "%s: initializing", __func__); IGMP_LOCK_INIT(); m_raopt = igmp_ra_alloc(); netisr_register(&igmp_nh); break; case MOD_UNLOAD: CTR1(KTR_IGMPV3, "%s: tearing down", __func__); netisr_unregister(&igmp_nh); m_free(m_raopt); m_raopt = NULL; IGMP_LOCK_DESTROY(); break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t igmp_mod = { "igmp", igmp_modevent, 0 }; DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); Index: head/sys/netinet/in.c =================================================================== --- head/sys/netinet/in.c (revision 336675) +++ head/sys/netinet/in.c (revision 336676) @@ -1,1504 +1,1504 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * Copyright (C) 2001 WIDE Project. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)in.c 8.4 (Berkeley) 1/9/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int in_aifaddr_ioctl(u_long, caddr_t, struct ifnet *, struct thread *); static int in_difaddr_ioctl(u_long, caddr_t, struct ifnet *, struct thread *); static void in_socktrim(struct sockaddr_in *); static void in_purgemaddrs(struct ifnet *); -static VNET_DEFINE(int, nosameprefix); +VNET_DEFINE_STATIC(int, nosameprefix); #define V_nosameprefix VNET(nosameprefix) SYSCTL_INT(_net_inet_ip, OID_AUTO, no_same_prefix, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nosameprefix), 0, "Refuse to create same prefixes on different interfaces"); VNET_DECLARE(struct inpcbinfo, ripcbinfo); #define V_ripcbinfo VNET(ripcbinfo) static struct sx in_control_sx; SX_SYSINIT(in_control_sx, &in_control_sx, "in_control"); /* * Return 1 if an internet address is for a ``local'' host * (one to which we have a connection). */ int in_localaddr(struct in_addr in) { struct rm_priotracker in_ifa_tracker; u_long i = ntohl(in.s_addr); struct in_ifaddr *ia; IN_IFADDR_RLOCK(&in_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { if ((i & ia->ia_subnetmask) == ia->ia_subnet) { IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (1); } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (0); } /* * Return 1 if an internet address is for the local host and configured * on one of its interfaces. */ int in_localip(struct in_addr in) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; IN_IFADDR_RLOCK(&in_ifa_tracker); LIST_FOREACH(ia, INADDR_HASH(in.s_addr), ia_hash) { if (IA_SIN(ia)->sin_addr.s_addr == in.s_addr) { IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (1); } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (0); } /* * Return 1 if an internet address is configured on an interface. */ int in_ifhasaddr(struct ifnet *ifp, struct in_addr in) { struct ifaddr *ifa; struct in_ifaddr *ia; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = (struct in_ifaddr *)ifa; if (ia->ia_addr.sin_addr.s_addr == in.s_addr) { IF_ADDR_RUNLOCK(ifp); return (1); } } IF_ADDR_RUNLOCK(ifp); return (0); } /* * Return a reference to the interface address which is different to * the supplied one but with same IP address value. */ static struct in_ifaddr * in_localip_more(struct in_ifaddr *ia) { struct rm_priotracker in_ifa_tracker; in_addr_t in = IA_SIN(ia)->sin_addr.s_addr; struct in_ifaddr *it; IN_IFADDR_RLOCK(&in_ifa_tracker); LIST_FOREACH(it, INADDR_HASH(in), ia_hash) { if (it != ia && IA_SIN(it)->sin_addr.s_addr == in) { ifa_ref(&it->ia_ifa); IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (it); } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (NULL); } /* * Determine whether an IP address is in a reserved set of addresses * that may not be forwarded, or whether datagrams to that destination * may be forwarded. */ int in_canforward(struct in_addr in) { u_long i = ntohl(in.s_addr); u_long net; if (IN_EXPERIMENTAL(i) || IN_MULTICAST(i) || IN_LINKLOCAL(i)) return (0); if (IN_CLASSA(i)) { net = i & IN_CLASSA_NET; if (net == 0 || net == (IN_LOOPBACKNET << IN_CLASSA_NSHIFT)) return (0); } return (1); } /* * Trim a mask in a sockaddr */ static void in_socktrim(struct sockaddr_in *ap) { char *cplim = (char *) &ap->sin_addr; char *cp = (char *) (&ap->sin_addr + 1); ap->sin_len = 0; while (--cp >= cplim) if (*cp) { (ap)->sin_len = cp - (char *) (ap) + 1; break; } } /* * Generic internet control operations (ioctl's). */ int in_control(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp, struct thread *td) { struct ifreq *ifr = (struct ifreq *)data; struct sockaddr_in *addr = (struct sockaddr_in *)&ifr->ifr_addr; struct ifaddr *ifa; struct in_ifaddr *ia; int error; if (ifp == NULL) return (EADDRNOTAVAIL); /* * Filter out 4 ioctls we implement directly. Forward the rest * to specific functions and ifp->if_ioctl(). */ switch (cmd) { case SIOCGIFADDR: case SIOCGIFBRDADDR: case SIOCGIFDSTADDR: case SIOCGIFNETMASK: break; case SIOCDIFADDR: sx_xlock(&in_control_sx); error = in_difaddr_ioctl(cmd, data, ifp, td); sx_xunlock(&in_control_sx); return (error); case OSIOCAIFADDR: /* 9.x compat */ case SIOCAIFADDR: sx_xlock(&in_control_sx); error = in_aifaddr_ioctl(cmd, data, ifp, td); sx_xunlock(&in_control_sx); return (error); case SIOCSIFADDR: case SIOCSIFBRDADDR: case SIOCSIFDSTADDR: case SIOCSIFNETMASK: /* We no longer support that old commands. */ return (EINVAL); default: if (ifp->if_ioctl == NULL) return (EOPNOTSUPP); return ((*ifp->if_ioctl)(ifp, cmd, data)); } if (addr->sin_addr.s_addr != INADDR_ANY && prison_check_ip4(td->td_ucred, &addr->sin_addr) != 0) return (EADDRNOTAVAIL); /* * Find address for this interface, if it exists. If an * address was specified, find that one instead of the * first one on the interface, if possible. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = (struct in_ifaddr *)ifa; if (ia->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr) break; } if (ifa == NULL) CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (ifa->ifa_addr->sa_family == AF_INET) { ia = (struct in_ifaddr *)ifa; if (prison_check_ip4(td->td_ucred, &ia->ia_addr.sin_addr) == 0) break; } if (ifa == NULL) { IF_ADDR_RUNLOCK(ifp); return (EADDRNOTAVAIL); } error = 0; switch (cmd) { case SIOCGIFADDR: *addr = ia->ia_addr; break; case SIOCGIFBRDADDR: if ((ifp->if_flags & IFF_BROADCAST) == 0) { error = EINVAL; break; } *addr = ia->ia_broadaddr; break; case SIOCGIFDSTADDR: if ((ifp->if_flags & IFF_POINTOPOINT) == 0) { error = EINVAL; break; } *addr = ia->ia_dstaddr; break; case SIOCGIFNETMASK: *addr = ia->ia_sockmask; break; } IF_ADDR_RUNLOCK(ifp); return (error); } static int in_aifaddr_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp, struct thread *td) { const struct in_aliasreq *ifra = (struct in_aliasreq *)data; const struct sockaddr_in *addr = &ifra->ifra_addr; const struct sockaddr_in *broadaddr = &ifra->ifra_broadaddr; const struct sockaddr_in *mask = &ifra->ifra_mask; const struct sockaddr_in *dstaddr = &ifra->ifra_dstaddr; const int vhid = (cmd == SIOCAIFADDR) ? ifra->ifra_vhid : 0; struct ifaddr *ifa; struct in_ifaddr *ia; bool iaIsFirst; int error = 0; error = priv_check(td, PRIV_NET_ADDIFADDR); if (error) return (error); /* * ifra_addr must be present and be of INET family. * ifra_broadaddr/ifra_dstaddr and ifra_mask are optional. */ if (addr->sin_len != sizeof(struct sockaddr_in) || addr->sin_family != AF_INET) return (EINVAL); if (broadaddr->sin_len != 0 && (broadaddr->sin_len != sizeof(struct sockaddr_in) || broadaddr->sin_family != AF_INET)) return (EINVAL); if (mask->sin_len != 0 && (mask->sin_len != sizeof(struct sockaddr_in) || mask->sin_family != AF_INET)) return (EINVAL); if ((ifp->if_flags & IFF_POINTOPOINT) && (dstaddr->sin_len != sizeof(struct sockaddr_in) || dstaddr->sin_addr.s_addr == INADDR_ANY)) return (EDESTADDRREQ); if (vhid > 0 && carp_attach_p == NULL) return (EPROTONOSUPPORT); /* * See whether address already exist. */ iaIsFirst = true; ia = NULL; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct in_ifaddr *it; if (ifa->ifa_addr->sa_family != AF_INET) continue; it = (struct in_ifaddr *)ifa; iaIsFirst = false; if (it->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr && prison_check_ip4(td->td_ucred, &addr->sin_addr) == 0) ia = it; } IF_ADDR_RUNLOCK(ifp); if (ia != NULL) (void )in_difaddr_ioctl(cmd, data, ifp, td); ifa = ifa_alloc(sizeof(struct in_ifaddr), M_WAITOK); ia = (struct in_ifaddr *)ifa; ifa->ifa_addr = (struct sockaddr *)&ia->ia_addr; ifa->ifa_dstaddr = (struct sockaddr *)&ia->ia_dstaddr; ifa->ifa_netmask = (struct sockaddr *)&ia->ia_sockmask; callout_init_rw(&ia->ia_garp_timer, &ifp->if_addr_lock, CALLOUT_RETURNUNLOCKED); ia->ia_ifp = ifp; ia->ia_addr = *addr; if (mask->sin_len != 0) { ia->ia_sockmask = *mask; ia->ia_subnetmask = ntohl(ia->ia_sockmask.sin_addr.s_addr); } else { in_addr_t i = ntohl(addr->sin_addr.s_addr); /* * Be compatible with network classes, if netmask isn't * supplied, guess it based on classes. */ if (IN_CLASSA(i)) ia->ia_subnetmask = IN_CLASSA_NET; else if (IN_CLASSB(i)) ia->ia_subnetmask = IN_CLASSB_NET; else ia->ia_subnetmask = IN_CLASSC_NET; ia->ia_sockmask.sin_addr.s_addr = htonl(ia->ia_subnetmask); } ia->ia_subnet = ntohl(addr->sin_addr.s_addr) & ia->ia_subnetmask; in_socktrim(&ia->ia_sockmask); if (ifp->if_flags & IFF_BROADCAST) { if (broadaddr->sin_len != 0) { ia->ia_broadaddr = *broadaddr; } else if (ia->ia_subnetmask == IN_RFC3021_MASK) { ia->ia_broadaddr.sin_addr.s_addr = INADDR_BROADCAST; ia->ia_broadaddr.sin_len = sizeof(struct sockaddr_in); ia->ia_broadaddr.sin_family = AF_INET; } else { ia->ia_broadaddr.sin_addr.s_addr = htonl(ia->ia_subnet | ~ia->ia_subnetmask); ia->ia_broadaddr.sin_len = sizeof(struct sockaddr_in); ia->ia_broadaddr.sin_family = AF_INET; } } if (ifp->if_flags & IFF_POINTOPOINT) ia->ia_dstaddr = *dstaddr; /* XXXGL: rtinit() needs this strange assignment. */ if (ifp->if_flags & IFF_LOOPBACK) ia->ia_dstaddr = ia->ia_addr; if (vhid != 0) { error = (*carp_attach_p)(&ia->ia_ifa, vhid); if (error) return (error); } /* if_addrhead is already referenced by ifa_alloc() */ IF_ADDR_WLOCK(ifp); CK_STAILQ_INSERT_TAIL(&ifp->if_addrhead, ifa, ifa_link); IF_ADDR_WUNLOCK(ifp); ifa_ref(ifa); /* in_ifaddrhead */ IN_IFADDR_WLOCK(); CK_STAILQ_INSERT_TAIL(&V_in_ifaddrhead, ia, ia_link); LIST_INSERT_HEAD(INADDR_HASH(ia->ia_addr.sin_addr.s_addr), ia, ia_hash); IN_IFADDR_WUNLOCK(); /* * Give the interface a chance to initialize * if this is its first address, * and to validate the address if necessary. */ if (ifp->if_ioctl != NULL) { error = (*ifp->if_ioctl)(ifp, SIOCSIFADDR, (caddr_t)ia); if (error) goto fail1; } /* * Add route for the network. */ if (vhid == 0) { int flags = RTF_UP; if (ifp->if_flags & (IFF_LOOPBACK|IFF_POINTOPOINT)) flags |= RTF_HOST; error = in_addprefix(ia, flags); if (error) goto fail1; } /* * Add a loopback route to self. */ if (vhid == 0 && (ifp->if_flags & IFF_LOOPBACK) == 0 && ia->ia_addr.sin_addr.s_addr != INADDR_ANY && !((ifp->if_flags & IFF_POINTOPOINT) && ia->ia_dstaddr.sin_addr.s_addr == ia->ia_addr.sin_addr.s_addr)) { struct in_ifaddr *eia; eia = in_localip_more(ia); if (eia == NULL) { error = ifa_add_loopback_route((struct ifaddr *)ia, (struct sockaddr *)&ia->ia_addr); if (error) goto fail2; } else ifa_free(&eia->ia_ifa); } if (iaIsFirst && (ifp->if_flags & IFF_MULTICAST)) { struct in_addr allhosts_addr; struct in_ifinfo *ii; ii = ((struct in_ifinfo *)ifp->if_afdata[AF_INET]); allhosts_addr.s_addr = htonl(INADDR_ALLHOSTS_GROUP); error = in_joingroup(ifp, &allhosts_addr, NULL, &ii->ii_allhosts); } EVENTHANDLER_INVOKE(ifaddr_event, ifp); return (error); fail2: if (vhid == 0) (void )in_scrubprefix(ia, LLE_STATIC); fail1: if (ia->ia_ifa.ifa_carp) (*carp_detach_p)(&ia->ia_ifa, false); IF_ADDR_WLOCK(ifp); CK_STAILQ_REMOVE(&ifp->if_addrhead, &ia->ia_ifa, ifaddr, ifa_link); IF_ADDR_WUNLOCK(ifp); ifa_free(&ia->ia_ifa); /* if_addrhead */ IN_IFADDR_WLOCK(); CK_STAILQ_REMOVE(&V_in_ifaddrhead, ia, in_ifaddr, ia_link); LIST_REMOVE(ia, ia_hash); IN_IFADDR_WUNLOCK(); ifa_free(&ia->ia_ifa); /* in_ifaddrhead */ return (error); } static int in_difaddr_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp, struct thread *td) { const struct ifreq *ifr = (struct ifreq *)data; const struct sockaddr_in *addr = (const struct sockaddr_in *) &ifr->ifr_addr; struct ifaddr *ifa; struct in_ifaddr *ia; bool deleteAny, iaIsLast; int error; if (td != NULL) { error = priv_check(td, PRIV_NET_DELIFADDR); if (error) return (error); } if (addr->sin_len != sizeof(struct sockaddr_in) || addr->sin_family != AF_INET) deleteAny = true; else deleteAny = false; iaIsLast = true; ia = NULL; IF_ADDR_WLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct in_ifaddr *it; if (ifa->ifa_addr->sa_family != AF_INET) continue; it = (struct in_ifaddr *)ifa; if (deleteAny && ia == NULL && (td == NULL || prison_check_ip4(td->td_ucred, &it->ia_addr.sin_addr) == 0)) ia = it; if (it->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr && (td == NULL || prison_check_ip4(td->td_ucred, &addr->sin_addr) == 0)) ia = it; if (it != ia) iaIsLast = false; } if (ia == NULL) { IF_ADDR_WUNLOCK(ifp); return (EADDRNOTAVAIL); } CK_STAILQ_REMOVE(&ifp->if_addrhead, &ia->ia_ifa, ifaddr, ifa_link); IF_ADDR_WUNLOCK(ifp); ifa_free(&ia->ia_ifa); /* if_addrhead */ IN_IFADDR_WLOCK(); CK_STAILQ_REMOVE(&V_in_ifaddrhead, ia, in_ifaddr, ia_link); LIST_REMOVE(ia, ia_hash); IN_IFADDR_WUNLOCK(); /* * in_scrubprefix() kills the interface route. */ in_scrubprefix(ia, LLE_STATIC); /* * in_ifadown gets rid of all the rest of * the routes. This is not quite the right * thing to do, but at least if we are running * a routing process they will come back. */ in_ifadown(&ia->ia_ifa, 1); if (ia->ia_ifa.ifa_carp) (*carp_detach_p)(&ia->ia_ifa, (cmd == SIOCDIFADDR) ? false : true); /* * If this is the last IPv4 address configured on this * interface, leave the all-hosts group. * No state-change report need be transmitted. */ if (iaIsLast && (ifp->if_flags & IFF_MULTICAST)) { struct in_ifinfo *ii; ii = ((struct in_ifinfo *)ifp->if_afdata[AF_INET]); if (ii->ii_allhosts) { (void)in_leavegroup(ii->ii_allhosts, NULL); ii->ii_allhosts = NULL; } } IF_ADDR_WLOCK(ifp); if (callout_stop(&ia->ia_garp_timer) == 1) { ifa_free(&ia->ia_ifa); } IF_ADDR_WUNLOCK(ifp); EVENTHANDLER_INVOKE(ifaddr_event, ifp); ifa_free(&ia->ia_ifa); /* in_ifaddrhead */ return (0); } #define rtinitflags(x) \ ((((x)->ia_ifp->if_flags & (IFF_LOOPBACK | IFF_POINTOPOINT)) != 0) \ ? RTF_HOST : 0) /* * Check if we have a route for the given prefix already or add one accordingly. */ int in_addprefix(struct in_ifaddr *target, int flags) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_addr prefix, mask, p, m; int error; if ((flags & RTF_HOST) != 0) { prefix = target->ia_dstaddr.sin_addr; mask.s_addr = 0; } else { prefix = target->ia_addr.sin_addr; mask = target->ia_sockmask.sin_addr; prefix.s_addr &= mask.s_addr; } IN_IFADDR_RLOCK(&in_ifa_tracker); /* Look for an existing address with the same prefix, mask, and fib */ CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { if (rtinitflags(ia)) { p = ia->ia_dstaddr.sin_addr; if (prefix.s_addr != p.s_addr) continue; } else { p = ia->ia_addr.sin_addr; m = ia->ia_sockmask.sin_addr; p.s_addr &= m.s_addr; if (prefix.s_addr != p.s_addr || mask.s_addr != m.s_addr) continue; } if (target->ia_ifp->if_fib != ia->ia_ifp->if_fib) continue; /* * If we got a matching prefix route inserted by other * interface address, we are done here. */ if (ia->ia_flags & IFA_ROUTE) { #ifdef RADIX_MPATH if (ia->ia_addr.sin_addr.s_addr == target->ia_addr.sin_addr.s_addr) { IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (EEXIST); } else break; #endif if (V_nosameprefix) { IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (EEXIST); } else { int fibnum; fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : target->ia_ifp->if_fib; rt_addrmsg(RTM_ADD, &target->ia_ifa, fibnum); IN_IFADDR_RUNLOCK(&in_ifa_tracker); return (0); } } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); /* * No-one seem to have this prefix route, so we try to insert it. */ error = rtinit(&target->ia_ifa, (int)RTM_ADD, flags); if (!error) target->ia_flags |= IFA_ROUTE; return (error); } /* * Removes either all lle entries for given @ia, or lle * corresponding to @ia address. */ static void in_scrubprefixlle(struct in_ifaddr *ia, int all, u_int flags) { struct sockaddr_in addr, mask; struct sockaddr *saddr, *smask; struct ifnet *ifp; saddr = (struct sockaddr *)&addr; bzero(&addr, sizeof(addr)); addr.sin_len = sizeof(addr); addr.sin_family = AF_INET; smask = (struct sockaddr *)&mask; bzero(&mask, sizeof(mask)); mask.sin_len = sizeof(mask); mask.sin_family = AF_INET; mask.sin_addr.s_addr = ia->ia_subnetmask; ifp = ia->ia_ifp; if (all) { /* * Remove all L2 entries matching given prefix. * Convert address to host representation to avoid * doing this on every callback. ia_subnetmask is already * stored in host representation. */ addr.sin_addr.s_addr = ntohl(ia->ia_addr.sin_addr.s_addr); lltable_prefix_free(AF_INET, saddr, smask, flags); } else { /* Remove interface address only */ addr.sin_addr.s_addr = ia->ia_addr.sin_addr.s_addr; lltable_delete_addr(LLTABLE(ifp), LLE_IFADDR, saddr); } } /* * If there is no other address in the system that can serve a route to the * same prefix, remove the route. Hand over the route to the new address * otherwise. */ int in_scrubprefix(struct in_ifaddr *target, u_int flags) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_addr prefix, mask, p, m; int error = 0; /* * Remove the loopback route to the interface address. */ if ((target->ia_addr.sin_addr.s_addr != INADDR_ANY) && !(target->ia_ifp->if_flags & IFF_LOOPBACK) && (flags & LLE_STATIC)) { struct in_ifaddr *eia; /* * XXXME: add fib-aware in_localip. * We definitely don't want to switch between * prefixes in different fibs. */ eia = in_localip_more(target); if (eia != NULL) { error = ifa_switch_loopback_route((struct ifaddr *)eia, (struct sockaddr *)&target->ia_addr); ifa_free(&eia->ia_ifa); } else { error = ifa_del_loopback_route((struct ifaddr *)target, (struct sockaddr *)&target->ia_addr); } } if (rtinitflags(target)) { prefix = target->ia_dstaddr.sin_addr; mask.s_addr = 0; } else { prefix = target->ia_addr.sin_addr; mask = target->ia_sockmask.sin_addr; prefix.s_addr &= mask.s_addr; } if ((target->ia_flags & IFA_ROUTE) == 0) { int fibnum; fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : target->ia_ifp->if_fib; rt_addrmsg(RTM_DELETE, &target->ia_ifa, fibnum); /* * Removing address from !IFF_UP interface or * prefix which exists on other interface (along with route). * No entries should exist here except target addr. * Given that, delete this entry only. */ in_scrubprefixlle(target, 0, flags); return (0); } IN_IFADDR_RLOCK(&in_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { if (rtinitflags(ia)) { p = ia->ia_dstaddr.sin_addr; if (prefix.s_addr != p.s_addr) continue; } else { p = ia->ia_addr.sin_addr; m = ia->ia_sockmask.sin_addr; p.s_addr &= m.s_addr; if (prefix.s_addr != p.s_addr || mask.s_addr != m.s_addr) continue; } if ((ia->ia_ifp->if_flags & IFF_UP) == 0) continue; /* * If we got a matching prefix address, move IFA_ROUTE and * the route itself to it. Make sure that routing daemons * get a heads-up. */ if ((ia->ia_flags & IFA_ROUTE) == 0) { ifa_ref(&ia->ia_ifa); IN_IFADDR_RUNLOCK(&in_ifa_tracker); error = rtinit(&(target->ia_ifa), (int)RTM_DELETE, rtinitflags(target)); if (error == 0) target->ia_flags &= ~IFA_ROUTE; else log(LOG_INFO, "in_scrubprefix: err=%d, old prefix delete failed\n", error); /* Scrub all entries IFF interface is different */ in_scrubprefixlle(target, target->ia_ifp != ia->ia_ifp, flags); error = rtinit(&ia->ia_ifa, (int)RTM_ADD, rtinitflags(ia) | RTF_UP); if (error == 0) ia->ia_flags |= IFA_ROUTE; else log(LOG_INFO, "in_scrubprefix: err=%d, new prefix add failed\n", error); ifa_free(&ia->ia_ifa); return (error); } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); /* * remove all L2 entries on the given prefix */ in_scrubprefixlle(target, 1, flags); /* * As no-one seem to have this prefix, we can remove the route. */ error = rtinit(&(target->ia_ifa), (int)RTM_DELETE, rtinitflags(target)); if (error == 0) target->ia_flags &= ~IFA_ROUTE; else log(LOG_INFO, "in_scrubprefix: err=%d, prefix delete failed\n", error); return (error); } #undef rtinitflags void in_ifscrub_all(void) { struct ifnet *ifp; struct ifaddr *ifa, *nifa; struct ifaliasreq ifr; IFNET_RLOCK(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { /* Cannot lock here - lock recursion. */ /* IF_ADDR_RLOCK(ifp); */ CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, nifa) { if (ifa->ifa_addr->sa_family != AF_INET) continue; /* * This is ugly but the only way for legacy IP to * cleanly remove addresses and everything attached. */ bzero(&ifr, sizeof(ifr)); ifr.ifra_addr = *ifa->ifa_addr; if (ifa->ifa_dstaddr) ifr.ifra_broadaddr = *ifa->ifa_dstaddr; (void)in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp, NULL); } /* IF_ADDR_RUNLOCK(ifp); */ in_purgemaddrs(ifp); igmp_domifdetach(ifp); } IFNET_RUNLOCK(); } int in_ifaddr_broadcast(struct in_addr in, struct in_ifaddr *ia) { return ((in.s_addr == ia->ia_broadaddr.sin_addr.s_addr || /* * Check for old-style (host 0) broadcast, but * taking into account that RFC 3021 obsoletes it. */ (ia->ia_subnetmask != IN_RFC3021_MASK && ntohl(in.s_addr) == ia->ia_subnet)) && /* * Check for an all one subnetmask. These * only exist when an interface gets a secondary * address. */ ia->ia_subnetmask != (u_long)0xffffffff); } /* * Return 1 if the address might be a local broadcast address. */ int in_broadcast(struct in_addr in, struct ifnet *ifp) { struct ifaddr *ifa; int found; if (in.s_addr == INADDR_BROADCAST || in.s_addr == INADDR_ANY) return (1); if ((ifp->if_flags & IFF_BROADCAST) == 0) return (0); found = 0; /* * Look through the list of addresses for a match * with a broadcast address. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (ifa->ifa_addr->sa_family == AF_INET && in_ifaddr_broadcast(in, (struct in_ifaddr *)ifa)) { found = 1; break; } IF_ADDR_RUNLOCK(ifp); return (found); } /* * On interface removal, clean up IPv4 data structures hung off of the ifnet. */ void in_ifdetach(struct ifnet *ifp) { IN_MULTI_LOCK(); in_pcbpurgeif0(&V_ripcbinfo, ifp); in_pcbpurgeif0(&V_udbinfo, ifp); in_pcbpurgeif0(&V_ulitecbinfo, ifp); in_purgemaddrs(ifp); IN_MULTI_UNLOCK(); } /* * Delete all IPv4 multicast address records, and associated link-layer * multicast address records, associated with ifp. * XXX It looks like domifdetach runs AFTER the link layer cleanup. * XXX This should not race with ifma_protospec being set during * a new allocation, if it does, we have bigger problems. */ static void in_purgemaddrs(struct ifnet *ifp) { struct in_multi_head purgeinms; struct in_multi *inm; struct ifmultiaddr *ifma, *next; SLIST_INIT(&purgeinms); IN_MULTI_LIST_LOCK(); /* * Extract list of in_multi associated with the detaching ifp * which the PF_INET layer is about to release. * We need to do this as IF_ADDR_LOCK() may be re-acquired * by code further down. */ IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; inm_rele_locked(&purgeinms, inm); if (__predict_false(ifma_restart)) { ifma_restart = true; goto restart; } } IF_ADDR_WUNLOCK(ifp); inm_release_list_deferred(&purgeinms); igmp_ifdetach(ifp); IN_MULTI_LIST_UNLOCK(); } struct in_llentry { struct llentry base; }; #define IN_LLTBL_DEFAULT_HSIZE 32 #define IN_LLTBL_HASH(k, h) \ (((((((k >> 8) ^ k) >> 8) ^ k) >> 8) ^ k) & ((h) - 1)) /* * Do actual deallocation of @lle. */ static void in_lltable_destroy_lle_unlocked(epoch_context_t ctx) { struct llentry *lle; lle = __containerof(ctx, struct llentry, lle_epoch_ctx); LLE_LOCK_DESTROY(lle); LLE_REQ_DESTROY(lle); free(lle, M_LLTABLE); } /* * Called by the datapath to indicate that * the entry was used. */ static void in_lltable_mark_used(struct llentry *lle) { LLE_REQ_LOCK(lle); lle->r_skip_req = 0; LLE_REQ_UNLOCK(lle); } /* * Called by LLE_FREE_LOCKED when number of references * drops to zero. */ static void in_lltable_destroy_lle(struct llentry *lle) { LLE_WUNLOCK(lle); epoch_call(net_epoch_preempt, &lle->lle_epoch_ctx, in_lltable_destroy_lle_unlocked); } static struct llentry * in_lltable_new(struct in_addr addr4, u_int flags) { struct in_llentry *lle; lle = malloc(sizeof(struct in_llentry), M_LLTABLE, M_NOWAIT | M_ZERO); if (lle == NULL) /* NB: caller generates msg */ return NULL; /* * For IPv4 this will trigger "arpresolve" to generate * an ARP request. */ lle->base.la_expire = time_uptime; /* mark expired */ lle->base.r_l3addr.addr4 = addr4; lle->base.lle_refcnt = 1; lle->base.lle_free = in_lltable_destroy_lle; LLE_LOCK_INIT(&lle->base); LLE_REQ_INIT(&lle->base); callout_init(&lle->base.lle_timer, 1); return (&lle->base); } #define IN_ARE_MASKED_ADDR_EQUAL(d, a, m) ( \ ((((d).s_addr ^ (a).s_addr) & (m).s_addr)) == 0 ) static int in_lltable_match_prefix(const struct sockaddr *saddr, const struct sockaddr *smask, u_int flags, struct llentry *lle) { struct in_addr addr, mask, lle_addr; addr = ((const struct sockaddr_in *)saddr)->sin_addr; mask = ((const struct sockaddr_in *)smask)->sin_addr; lle_addr.s_addr = ntohl(lle->r_l3addr.addr4.s_addr); if (IN_ARE_MASKED_ADDR_EQUAL(lle_addr, addr, mask) == 0) return (0); if (lle->la_flags & LLE_IFADDR) { /* * Delete LLE_IFADDR records IFF address & flag matches. * Note that addr is the interface address within prefix * being matched. * Note also we should handle 'ifdown' cases without removing * ifaddr macs. */ if (addr.s_addr == lle_addr.s_addr && (flags & LLE_STATIC) != 0) return (1); return (0); } /* flags & LLE_STATIC means deleting both dynamic and static entries */ if ((flags & LLE_STATIC) || !(lle->la_flags & LLE_STATIC)) return (1); return (0); } static void in_lltable_free_entry(struct lltable *llt, struct llentry *lle) { size_t pkts_dropped; LLE_WLOCK_ASSERT(lle); KASSERT(llt != NULL, ("lltable is NULL")); /* Unlink entry from table if not already */ if ((lle->la_flags & LLE_LINKED) != 0) { IF_AFDATA_WLOCK_ASSERT(llt->llt_ifp); lltable_unlink_entry(llt, lle); } /* Drop hold queue */ pkts_dropped = llentry_free(lle); ARPSTAT_ADD(dropped, pkts_dropped); } static int in_lltable_rtcheck(struct ifnet *ifp, u_int flags, const struct sockaddr *l3addr) { struct rt_addrinfo info; struct sockaddr_in rt_key, rt_mask; struct sockaddr rt_gateway; int rt_flags; KASSERT(l3addr->sa_family == AF_INET, ("sin_family %d", l3addr->sa_family)); bzero(&rt_key, sizeof(rt_key)); rt_key.sin_len = sizeof(rt_key); bzero(&rt_mask, sizeof(rt_mask)); rt_mask.sin_len = sizeof(rt_mask); bzero(&rt_gateway, sizeof(rt_gateway)); rt_gateway.sa_len = sizeof(rt_gateway); bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key; info.rti_info[RTAX_NETMASK] = (struct sockaddr *)&rt_mask; info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; if (rib_lookup_info(ifp->if_fib, l3addr, NHR_REF, 0, &info) != 0) return (EINVAL); rt_flags = info.rti_flags; /* * If the gateway for an existing host route matches the target L3 * address, which is a special route inserted by some implementation * such as MANET, and the interface is of the correct type, then * allow for ARP to proceed. */ if (rt_flags & RTF_GATEWAY) { if (!(rt_flags & RTF_HOST) || !info.rti_ifp || info.rti_ifp->if_type != IFT_ETHER || (info.rti_ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) != 0 || memcmp(rt_gateway.sa_data, l3addr->sa_data, sizeof(in_addr_t)) != 0) { rib_free_info(&info); return (EINVAL); } } rib_free_info(&info); /* * Make sure that at least the destination address is covered * by the route. This is for handling the case where 2 or more * interfaces have the same prefix. An incoming packet arrives * on one interface and the corresponding outgoing packet leaves * another interface. */ if (!(rt_flags & RTF_HOST) && info.rti_ifp != ifp) { const char *sa, *mask, *addr, *lim; const struct sockaddr_in *l3sin; mask = (const char *)&rt_mask; /* * Just being extra cautious to avoid some custom * code getting into trouble. */ if ((info.rti_addrs & RTA_NETMASK) == 0) return (EINVAL); sa = (const char *)&rt_key; addr = (const char *)l3addr; l3sin = (const struct sockaddr_in *)l3addr; lim = addr + l3sin->sin_len; for ( ; addr < lim; sa++, mask++, addr++) { if ((*sa ^ *addr) & *mask) { #ifdef DIAGNOSTIC char addrbuf[INET_ADDRSTRLEN]; log(LOG_INFO, "IPv4 address: \"%s\" " "is not on the network\n", inet_ntoa_r(l3sin->sin_addr, addrbuf)); #endif return (EINVAL); } } } return (0); } static inline uint32_t in_lltable_hash_dst(const struct in_addr dst, uint32_t hsize) { return (IN_LLTBL_HASH(dst.s_addr, hsize)); } static uint32_t in_lltable_hash(const struct llentry *lle, uint32_t hsize) { return (in_lltable_hash_dst(lle->r_l3addr.addr4, hsize)); } static void in_lltable_fill_sa_entry(const struct llentry *lle, struct sockaddr *sa) { struct sockaddr_in *sin; sin = (struct sockaddr_in *)sa; bzero(sin, sizeof(*sin)); sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = lle->r_l3addr.addr4; } static inline struct llentry * in_lltable_find_dst(struct lltable *llt, struct in_addr dst) { struct llentry *lle; struct llentries *lleh; u_int hashidx; hashidx = in_lltable_hash_dst(dst, llt->llt_hsize); lleh = &llt->lle_head[hashidx]; CK_LIST_FOREACH(lle, lleh, lle_next) { if (lle->la_flags & LLE_DELETED) continue; if (lle->r_l3addr.addr4.s_addr == dst.s_addr) break; } return (lle); } static void in_lltable_delete_entry(struct lltable *llt, struct llentry *lle) { lle->la_flags |= LLE_DELETED; EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_DELETED); #ifdef DIAGNOSTIC log(LOG_INFO, "ifaddr cache = %p is deleted\n", lle); #endif llentry_free(lle); } static struct llentry * in_lltable_alloc(struct lltable *llt, u_int flags, const struct sockaddr *l3addr) { const struct sockaddr_in *sin = (const struct sockaddr_in *)l3addr; struct ifnet *ifp = llt->llt_ifp; struct llentry *lle; char linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; int lladdr_off; KASSERT(l3addr->sa_family == AF_INET, ("sin_family %d", l3addr->sa_family)); /* * A route that covers the given address must have * been installed 1st because we are doing a resolution, * verify this. */ if (!(flags & LLE_IFADDR) && in_lltable_rtcheck(ifp, flags, l3addr) != 0) return (NULL); lle = in_lltable_new(sin->sin_addr, flags); if (lle == NULL) { log(LOG_INFO, "lla_lookup: new lle malloc failed\n"); return (NULL); } lle->la_flags = flags; if (flags & LLE_STATIC) lle->r_flags |= RLLE_VALID; if ((flags & LLE_IFADDR) == LLE_IFADDR) { linkhdrsize = LLE_MAX_LINKHDR; if (lltable_calc_llheader(ifp, AF_INET, IF_LLADDR(ifp), linkhdr, &linkhdrsize, &lladdr_off) != 0) { epoch_call(net_epoch_preempt, &lle->lle_epoch_ctx, in_lltable_destroy_lle_unlocked); return (NULL); } lltable_set_entry_addr(ifp, lle, linkhdr, linkhdrsize, lladdr_off); lle->la_flags |= LLE_STATIC; lle->r_flags |= (RLLE_VALID | RLLE_IFADDR); } return (lle); } /* * Return NULL if not found or marked for deletion. * If found return lle read locked. */ static struct llentry * in_lltable_lookup(struct lltable *llt, u_int flags, const struct sockaddr *l3addr) { const struct sockaddr_in *sin = (const struct sockaddr_in *)l3addr; struct llentry *lle; IF_AFDATA_LOCK_ASSERT(llt->llt_ifp); KASSERT(l3addr->sa_family == AF_INET, ("sin_family %d", l3addr->sa_family)); lle = in_lltable_find_dst(llt, sin->sin_addr); if (lle == NULL) return (NULL); KASSERT((flags & (LLE_UNLOCKED|LLE_EXCLUSIVE)) != (LLE_UNLOCKED|LLE_EXCLUSIVE),("wrong lle request flags: 0x%X", flags)); if (flags & LLE_UNLOCKED) return (lle); if (flags & LLE_EXCLUSIVE) LLE_WLOCK(lle); else LLE_RLOCK(lle); return (lle); } static int in_lltable_dump_entry(struct lltable *llt, struct llentry *lle, struct sysctl_req *wr) { struct ifnet *ifp = llt->llt_ifp; /* XXX stack use */ struct { struct rt_msghdr rtm; struct sockaddr_in sin; struct sockaddr_dl sdl; } arpc; struct sockaddr_dl *sdl; int error; bzero(&arpc, sizeof(arpc)); /* skip deleted entries */ if ((lle->la_flags & LLE_DELETED) == LLE_DELETED) return (0); /* Skip if jailed and not a valid IP of the prison. */ lltable_fill_sa_entry(lle,(struct sockaddr *)&arpc.sin); if (prison_if(wr->td->td_ucred, (struct sockaddr *)&arpc.sin) != 0) return (0); /* * produce a msg made of: * struct rt_msghdr; * struct sockaddr_in; (IPv4) * struct sockaddr_dl; */ arpc.rtm.rtm_msglen = sizeof(arpc); arpc.rtm.rtm_version = RTM_VERSION; arpc.rtm.rtm_type = RTM_GET; arpc.rtm.rtm_flags = RTF_UP; arpc.rtm.rtm_addrs = RTA_DST | RTA_GATEWAY; /* publish */ if (lle->la_flags & LLE_PUB) arpc.rtm.rtm_flags |= RTF_ANNOUNCE; sdl = &arpc.sdl; sdl->sdl_family = AF_LINK; sdl->sdl_len = sizeof(*sdl); sdl->sdl_index = ifp->if_index; sdl->sdl_type = ifp->if_type; if ((lle->la_flags & LLE_VALID) == LLE_VALID) { sdl->sdl_alen = ifp->if_addrlen; bcopy(lle->ll_addr, LLADDR(sdl), ifp->if_addrlen); } else { sdl->sdl_alen = 0; bzero(LLADDR(sdl), ifp->if_addrlen); } arpc.rtm.rtm_rmx.rmx_expire = lle->la_flags & LLE_STATIC ? 0 : lle->la_expire; arpc.rtm.rtm_flags |= (RTF_HOST | RTF_LLDATA); if (lle->la_flags & LLE_STATIC) arpc.rtm.rtm_flags |= RTF_STATIC; if (lle->la_flags & LLE_IFADDR) arpc.rtm.rtm_flags |= RTF_PINNED; arpc.rtm.rtm_index = ifp->if_index; error = SYSCTL_OUT(wr, &arpc, sizeof(arpc)); return (error); } static struct lltable * in_lltattach(struct ifnet *ifp) { struct lltable *llt; llt = lltable_allocate_htbl(IN_LLTBL_DEFAULT_HSIZE); llt->llt_af = AF_INET; llt->llt_ifp = ifp; llt->llt_lookup = in_lltable_lookup; llt->llt_alloc_entry = in_lltable_alloc; llt->llt_delete_entry = in_lltable_delete_entry; llt->llt_dump_entry = in_lltable_dump_entry; llt->llt_hash = in_lltable_hash; llt->llt_fill_sa_entry = in_lltable_fill_sa_entry; llt->llt_free_entry = in_lltable_free_entry; llt->llt_match_prefix = in_lltable_match_prefix; llt->llt_mark_used = in_lltable_mark_used; lltable_link(llt); return (llt); } void * in_domifattach(struct ifnet *ifp) { struct in_ifinfo *ii; ii = malloc(sizeof(struct in_ifinfo), M_IFADDR, M_WAITOK|M_ZERO); ii->ii_llt = in_lltattach(ifp); ii->ii_igmp = igmp_domifattach(ifp); return (ii); } void in_domifdetach(struct ifnet *ifp, void *aux) { struct in_ifinfo *ii = (struct in_ifinfo *)aux; igmp_domifdetach(ifp); lltable_free(ii->ii_llt); free(ii, M_IFADDR); } Index: head/sys/netinet/in_gif.c =================================================================== --- head/sys/netinet/in_gif.c (revision 336675) +++ head/sys/netinet/in_gif.c (revision 336676) @@ -1,407 +1,407 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * Copyright (c) 2018 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: in_gif.c,v 1.54 2001/05/14 14:02:16 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #define GIF_TTL 30 -static VNET_DEFINE(int, ip_gif_ttl) = GIF_TTL; +VNET_DEFINE_STATIC(int, ip_gif_ttl) = GIF_TTL; #define V_ip_gif_ttl VNET(ip_gif_ttl) SYSCTL_INT(_net_inet_ip, IPCTL_GIF_TTL, gifttl, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_gif_ttl), 0, "Default TTL value for encapsulated packets"); /* * We keep interfaces in a hash table using src+dst as key. * Interfaces with GIF_IGNORE_SOURCE flag are linked into plain list. */ -static VNET_DEFINE(struct gif_list *, ipv4_hashtbl) = NULL; -static VNET_DEFINE(struct gif_list, ipv4_list) = CK_LIST_HEAD_INITIALIZER(); +VNET_DEFINE_STATIC(struct gif_list *, ipv4_hashtbl) = NULL; +VNET_DEFINE_STATIC(struct gif_list, ipv4_list) = CK_LIST_HEAD_INITIALIZER(); #define V_ipv4_hashtbl VNET(ipv4_hashtbl) #define V_ipv4_list VNET(ipv4_list) #define GIF_HASH(src, dst) (V_ipv4_hashtbl[\ in_gif_hashval((src), (dst)) & (GIF_HASH_SIZE - 1)]) #define GIF_HASH_SC(sc) GIF_HASH((sc)->gif_iphdr->ip_src.s_addr,\ (sc)->gif_iphdr->ip_dst.s_addr) static uint32_t in_gif_hashval(in_addr_t src, in_addr_t dst) { uint32_t ret; ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT); return (fnv_32_buf(&dst, sizeof(dst), ret)); } static int in_gif_checkdup(const struct gif_softc *sc, in_addr_t src, in_addr_t dst) { struct gif_softc *tmp; if (sc->gif_family == AF_INET && sc->gif_iphdr->ip_src.s_addr == src && sc->gif_iphdr->ip_dst.s_addr == dst) return (EEXIST); CK_LIST_FOREACH(tmp, &GIF_HASH(src, dst), chain) { if (tmp == sc) continue; if (tmp->gif_iphdr->ip_src.s_addr == src && tmp->gif_iphdr->ip_dst.s_addr == dst) return (EADDRNOTAVAIL); } return (0); } static void in_gif_attach(struct gif_softc *sc) { if (sc->gif_options & GIF_IGNORE_SOURCE) CK_LIST_INSERT_HEAD(&V_ipv4_list, sc, chain); else CK_LIST_INSERT_HEAD(&GIF_HASH_SC(sc), sc, chain); } int in_gif_setopts(struct gif_softc *sc, u_int options) { /* NOTE: we are protected with gif_ioctl_sx lock */ MPASS(sc->gif_family == AF_INET); MPASS(sc->gif_options != options); if ((options & GIF_IGNORE_SOURCE) != (sc->gif_options & GIF_IGNORE_SOURCE)) { CK_LIST_REMOVE(sc, chain); sc->gif_options = options; in_gif_attach(sc); } return (0); } int in_gif_ioctl(struct gif_softc *sc, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; struct sockaddr_in *dst, *src; struct ip *ip; int error; /* NOTE: we are protected with gif_ioctl_sx lock */ error = EINVAL; switch (cmd) { case SIOCSIFPHYADDR: src = &((struct in_aliasreq *)data)->ifra_addr; dst = &((struct in_aliasreq *)data)->ifra_dstaddr; /* sanity checks */ if (src->sin_family != dst->sin_family || src->sin_family != AF_INET || src->sin_len != dst->sin_len || src->sin_len != sizeof(*src)) break; if (src->sin_addr.s_addr == INADDR_ANY || dst->sin_addr.s_addr == INADDR_ANY) { error = EADDRNOTAVAIL; break; } if (V_ipv4_hashtbl == NULL) V_ipv4_hashtbl = gif_hashinit(); error = in_gif_checkdup(sc, src->sin_addr.s_addr, dst->sin_addr.s_addr); if (error == EADDRNOTAVAIL) break; if (error == EEXIST) { /* Addresses are the same. Just return. */ error = 0; break; } ip = malloc(sizeof(*ip), M_GIF, M_WAITOK | M_ZERO); ip->ip_src.s_addr = src->sin_addr.s_addr; ip->ip_dst.s_addr = dst->sin_addr.s_addr; if (sc->gif_family != 0) { /* Detach existing tunnel first */ CK_LIST_REMOVE(sc, chain); GIF_WAIT(); free(sc->gif_hdr, M_GIF); /* XXX: should we notify about link state change? */ } sc->gif_family = AF_INET; sc->gif_iphdr = ip; in_gif_attach(sc); break; case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: if (sc->gif_family != AF_INET) { error = EADDRNOTAVAIL; break; } src = (struct sockaddr_in *)&ifr->ifr_addr; memset(src, 0, sizeof(*src)); src->sin_family = AF_INET; src->sin_len = sizeof(*src); src->sin_addr = (cmd == SIOCGIFPSRCADDR) ? sc->gif_iphdr->ip_src: sc->gif_iphdr->ip_dst; error = prison_if(curthread->td_ucred, (struct sockaddr *)src); if (error != 0) memset(src, 0, sizeof(*src)); break; } return (error); } int in_gif_output(struct ifnet *ifp, struct mbuf *m, int proto, uint8_t ecn) { struct gif_softc *sc = ifp->if_softc; struct ip *ip; int len; /* prepend new IP header */ MPASS(in_epoch(net_epoch_preempt)); len = sizeof(struct ip); #ifndef __NO_STRICT_ALIGNMENT if (proto == IPPROTO_ETHERIP) len += ETHERIP_ALIGN; #endif M_PREPEND(m, len, M_NOWAIT); if (m == NULL) return (ENOBUFS); #ifndef __NO_STRICT_ALIGNMENT if (proto == IPPROTO_ETHERIP) { len = mtod(m, vm_offset_t) & 3; KASSERT(len == 0 || len == ETHERIP_ALIGN, ("in_gif_output: unexpected misalignment")); m->m_data += len; m->m_len -= ETHERIP_ALIGN; } #endif ip = mtod(m, struct ip *); MPASS(sc->gif_family == AF_INET); bcopy(sc->gif_iphdr, ip, sizeof(struct ip)); ip->ip_p = proto; /* version will be set in ip_output() */ ip->ip_ttl = V_ip_gif_ttl; ip->ip_len = htons(m->m_pkthdr.len); ip->ip_tos = ecn; return (ip_output(m, NULL, NULL, 0, NULL, NULL)); } static int in_gif_input(struct mbuf *m, int off, int proto, void *arg) { struct gif_softc *sc = arg; struct ifnet *gifp; struct ip *ip; uint8_t ecn; MPASS(in_epoch(net_epoch_preempt)); if (sc == NULL) { m_freem(m); KMOD_IPSTAT_INC(ips_nogif); return (IPPROTO_DONE); } gifp = GIF2IFP(sc); if ((gifp->if_flags & IFF_UP) != 0) { ip = mtod(m, struct ip *); ecn = ip->ip_tos; m_adj(m, off); gif_input(m, gifp, proto, ecn); } else { m_freem(m); KMOD_IPSTAT_INC(ips_nogif); } return (IPPROTO_DONE); } static int in_gif_lookup(const struct mbuf *m, int off, int proto, void **arg) { const struct ip *ip; struct gif_softc *sc; int ret; if (V_ipv4_hashtbl == NULL) return (0); MPASS(in_epoch(net_epoch_preempt)); ip = mtod(m, const struct ip *); /* * NOTE: it is safe to iterate without any locking here, because softc * can be reclaimed only when we are not within net_epoch_preempt * section, but ip_encap lookup+input are executed in epoch section. */ ret = 0; CK_LIST_FOREACH(sc, &GIF_HASH(ip->ip_dst.s_addr, ip->ip_src.s_addr), chain) { /* * This is an inbound packet, its ip_dst is source address * in softc. */ if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr && sc->gif_iphdr->ip_dst.s_addr == ip->ip_src.s_addr) { ret = ENCAP_DRV_LOOKUP; goto done; } } /* * No exact match. * Check the list of interfaces with GIF_IGNORE_SOURCE flag. */ CK_LIST_FOREACH(sc, &V_ipv4_list, chain) { if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr) { ret = 32 + 8; /* src + proto */ goto done; } } return (0); done: if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0) return (0); /* ingress filters on outer source */ if ((GIF2IFP(sc)->if_flags & IFF_LINK2) == 0) { struct nhop4_basic nh4; struct in_addr dst; dst = ip->ip_src; if (fib4_lookup_nh_basic(sc->gif_fibnum, dst, 0, 0, &nh4) != 0) return (0); if (nh4.nh_ifp != m->m_pkthdr.rcvif) return (0); } *arg = sc; return (ret); } static struct { const struct encap_config encap; const struct encaptab *cookie; } ipv4_encap_cfg[] = { { .encap = { .proto = IPPROTO_IPV4, .min_length = 2 * sizeof(struct ip), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in_gif_lookup, .input = in_gif_input }, }, #ifdef INET6 { .encap = { .proto = IPPROTO_IPV6, .min_length = sizeof(struct ip) + sizeof(struct ip6_hdr), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in_gif_lookup, .input = in_gif_input }, }, #endif { .encap = { .proto = IPPROTO_ETHERIP, .min_length = sizeof(struct ip) + sizeof(struct etherip_header) + sizeof(struct ether_header), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in_gif_lookup, .input = in_gif_input }, } }; void in_gif_init(void) { int i; if (!IS_DEFAULT_VNET(curvnet)) return; for (i = 0; i < nitems(ipv4_encap_cfg); i++) ipv4_encap_cfg[i].cookie = ip_encap_attach( &ipv4_encap_cfg[i].encap, NULL, M_WAITOK); } void in_gif_uninit(void) { int i; if (IS_DEFAULT_VNET(curvnet)) { for (i = 0; i < nitems(ipv4_encap_cfg); i++) ip_encap_detach(ipv4_encap_cfg[i].cookie); } if (V_ipv4_hashtbl != NULL) gif_hashdestroy(V_ipv4_hashtbl); } Index: head/sys/netinet/in_pcb.c =================================================================== --- head/sys/netinet/in_pcb.c (revision 336675) +++ head/sys/netinet/in_pcb.c (revision 336676) @@ -1,3453 +1,3453 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1991, 1993, 1995 * The Regents of the University of California. * Copyright (c) 2007-2009 Robert N. M. Watson * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)in_pcb.c 8.4 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_ipsec.h" #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ratelimit.h" #include "opt_pcbgroup.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #include #include #ifdef TCPHPTS #include #endif #include #include #endif #ifdef INET #include #endif #ifdef INET6 #include #include #include #include #endif /* INET6 */ #include #include #define INPCBLBGROUP_SIZMIN 8 #define INPCBLBGROUP_SIZMAX 256 static struct callout ipport_tick_callout; /* * These configure the range of local port addresses assigned to * "unspecified" outgoing connections/packets/whatever. */ VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1; /* 1023 */ VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART; /* 600 */ VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST; /* 10000 */ VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST; /* 65535 */ VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO; /* 49152 */ VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO; /* 65535 */ /* * Reserved ports accessible only to root. There are significant * security considerations that must be accounted for when changing these, * but the security benefits can be great. Please be careful. */ VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1; /* 1023 */ VNET_DEFINE(int, ipport_reservedlow); /* Variables dealing with random ephemeral port allocation. */ VNET_DEFINE(int, ipport_randomized) = 1; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_randomcps) = 10; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_randomtime) = 45; /* user controlled via sysctl */ VNET_DEFINE(int, ipport_stoprandom); /* toggled by ipport_tick */ VNET_DEFINE(int, ipport_tcpallocs); -static VNET_DEFINE(int, ipport_tcplastcount); +VNET_DEFINE_STATIC(int, ipport_tcplastcount); #define V_ipport_tcplastcount VNET(ipport_tcplastcount) static void in_pcbremlists(struct inpcb *inp); #ifdef INET static struct inpcb *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp); #define RANGECHK(var, min, max) \ if ((var) < (min)) { (var) = (min); } \ else if ((var) > (max)) { (var) = (max); } static int sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS) { int error; error = sysctl_handle_int(oidp, arg1, arg2, req); if (error == 0) { RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1); RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1); RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX); RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX); } return (error); } #undef RANGECHK static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0, "IP Ports"); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(ipport_reservedhigh), 0, ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow, CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, ""); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomized), 0, "Enable random port allocation"); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port " "allocations before switching to a sequental one"); SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipport_randomtime), 0, "Minimum time to keep sequental port " "allocation before switching to a random one"); #endif /* INET */ /* * in_pcb.c: manage the Protocol Control Blocks. * * NOTE: It is assumed that most of these functions will be called with * the pcbinfo lock held, and often, the inpcb lock held, as these utility * functions often modify hash chains or addresses in pcbs. */ static struct inpcblbgroup * in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag, uint16_t port, const union in_dependaddr *addr, int size) { struct inpcblbgroup *grp; size_t bytes; bytes = __offsetof(struct inpcblbgroup, il_inp[size]); grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT); if (!grp) return (NULL); grp->il_vflag = vflag; grp->il_lport = port; grp->il_dependladdr = *addr; grp->il_inpsiz = size; LIST_INSERT_HEAD(hdr, grp, il_list); return (grp); } static void in_pcblbgroup_free(struct inpcblbgroup *grp) { LIST_REMOVE(grp, il_list); free(grp, M_TEMP); } static struct inpcblbgroup * in_pcblbgroup_resize(struct inpcblbgrouphead *hdr, struct inpcblbgroup *old_grp, int size) { struct inpcblbgroup *grp; int i; grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag, old_grp->il_lport, &old_grp->il_dependladdr, size); if (!grp) return (NULL); KASSERT(old_grp->il_inpcnt < grp->il_inpsiz, ("invalid new local group size %d and old local group count %d", grp->il_inpsiz, old_grp->il_inpcnt)); for (i = 0; i < old_grp->il_inpcnt; ++i) grp->il_inp[i] = old_grp->il_inp[i]; grp->il_inpcnt = old_grp->il_inpcnt; in_pcblbgroup_free(old_grp); return (grp); } /* * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i] * and shrink group if possible. */ static void in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp, int i) { struct inpcblbgroup *grp = *grpp; for (; i + 1 < grp->il_inpcnt; ++i) grp->il_inp[i] = grp->il_inp[i + 1]; grp->il_inpcnt--; if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN && grp->il_inpcnt <= (grp->il_inpsiz / 4)) { /* Shrink this group. */ struct inpcblbgroup *new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); if (new_grp) *grpp = new_grp; } return; } /* * Add PCB to load balance group for SO_REUSEPORT_LB option. */ static int in_pcbinslbgrouphash(struct inpcb *inp) { struct inpcbinfo *pcbinfo; struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; uint16_t hashmask, lport; uint32_t group_index; struct ucred *cred; static int limit_logged = 0; pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); if (pcbinfo->ipi_lbgrouphashbase == NULL) return (0); hashmask = pcbinfo->ipi_lbgrouphashmask; lport = inp->inp_lport; group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask); hdr = &pcbinfo->ipi_lbgrouphashbase[group_index]; /* * Don't allow jailed socket to join local group. */ if (inp->inp_socket != NULL) cred = inp->inp_socket->so_cred; else cred = NULL; if (cred != NULL && jailed(cred)) return (0); #ifdef INET6 /* * Don't allow IPv4 mapped INET6 wild socket. */ if ((inp->inp_vflag & INP_IPV4) && inp->inp_laddr.s_addr == INADDR_ANY && INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) { return (0); } #endif hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBLBGROUP_PORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; LIST_FOREACH(grp, hdr, il_list) { if (grp->il_vflag == inp->inp_vflag && grp->il_lport == inp->inp_lport && memcmp(&grp->il_dependladdr, &inp->inp_inc.inc_ie.ie_dependladdr, sizeof(grp->il_dependladdr)) == 0) { break; } } if (grp == NULL) { /* Create new load balance group. */ grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag, inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr, INPCBLBGROUP_SIZMIN); if (!grp) return (ENOBUFS); } else if (grp->il_inpcnt == grp->il_inpsiz) { if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { if (!limit_logged) { limit_logged = 1; printf("lb group port %d, limit reached\n", ntohs(grp->il_lport)); } return (0); } /* Expand this local group. */ grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2); if (!grp) return (ENOBUFS); } KASSERT(grp->il_inpcnt < grp->il_inpsiz, ("invalid local group size %d and count %d", grp->il_inpsiz, grp->il_inpcnt)); grp->il_inp[grp->il_inpcnt] = inp; grp->il_inpcnt++; return (0); } /* * Remove PCB from load balance group. */ static void in_pcbremlbgrouphash(struct inpcb *inp) { struct inpcbinfo *pcbinfo; struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; int i; pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); if (pcbinfo->ipi_lbgrouphashbase == NULL) return; hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBLBGROUP_PORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; LIST_FOREACH(grp, hdr, il_list) { for (i = 0; i < grp->il_inpcnt; ++i) { if (grp->il_inp[i] != inp) continue; if (grp->il_inpcnt == 1) { /* We are the last, free this local group. */ in_pcblbgroup_free(grp); } else { /* Pull up inpcbs, shrink group if possible. */ in_pcblbgroup_reorder(hdr, &grp, i); } return; } } } /* * Different protocols initialize their inpcbs differently - giving * different name to the lock. But they all are disposed the same. */ static void inpcb_fini(void *mem, int size) { struct inpcb *inp = mem; INP_LOCK_DESTROY(inp); } /* * Initialize an inpcbinfo -- we should be able to reduce the number of * arguments in time. */ void in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name, struct inpcbhead *listhead, int hash_nelements, int porthash_nelements, char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields) { INP_INFO_LOCK_INIT(pcbinfo, name); INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash"); /* XXXRW: argument? */ INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist"); #ifdef VIMAGE pcbinfo->ipi_vnet = curvnet; #endif pcbinfo->ipi_listhead = listhead; CK_LIST_INIT(pcbinfo->ipi_listhead); pcbinfo->ipi_count = 0; pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB, &pcbinfo->ipi_hashmask); pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB, &pcbinfo->ipi_porthashmask); pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB, &pcbinfo->ipi_lbgrouphashmask); #ifdef PCBGROUP in_pcbgroup_init(pcbinfo, hashfields, hash_nelements); #endif pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb), NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0); uma_zone_set_max(pcbinfo->ipi_zone, maxsockets); uma_zone_set_warning(pcbinfo->ipi_zone, "kern.ipc.maxsockets limit reached"); } /* * Destroy an inpcbinfo. */ void in_pcbinfo_destroy(struct inpcbinfo *pcbinfo) { KASSERT(pcbinfo->ipi_count == 0, ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count)); hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask); hashdestroy(pcbinfo->ipi_porthashbase, M_PCB, pcbinfo->ipi_porthashmask); hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB, pcbinfo->ipi_lbgrouphashmask); #ifdef PCBGROUP in_pcbgroup_destroy(pcbinfo); #endif uma_zdestroy(pcbinfo->ipi_zone); INP_LIST_LOCK_DESTROY(pcbinfo); INP_HASH_LOCK_DESTROY(pcbinfo); INP_INFO_LOCK_DESTROY(pcbinfo); } /* * Allocate a PCB and associate it with the socket. * On success return with the PCB locked. */ int in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo) { struct inpcb *inp; int error; #ifdef INVARIANTS if (pcbinfo == &V_tcbinfo) { INP_INFO_RLOCK_ASSERT(pcbinfo); } else { INP_INFO_WLOCK_ASSERT(pcbinfo); } #endif error = 0; inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); if (inp == NULL) return (ENOBUFS); bzero(&inp->inp_start_zero, inp_zero_size); inp->inp_pcbinfo = pcbinfo; inp->inp_socket = so; inp->inp_cred = crhold(so->so_cred); inp->inp_inc.inc_fibnum = so->so_fibnum; #ifdef MAC error = mac_inpcb_init(inp, M_NOWAIT); if (error != 0) goto out; mac_inpcb_create(so, inp); #endif #if defined(IPSEC) || defined(IPSEC_SUPPORT) error = ipsec_init_pcbpolicy(inp); if (error != 0) { #ifdef MAC mac_inpcb_destroy(inp); #endif goto out; } #endif /*IPSEC*/ #ifdef INET6 if (INP_SOCKAF(so) == AF_INET6) { inp->inp_vflag |= INP_IPV6PROTO; if (V_ip6_v6only) inp->inp_flags |= IN6P_IPV6_V6ONLY; } #endif INP_WLOCK(inp); INP_LIST_WLOCK(pcbinfo); CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list); pcbinfo->ipi_count++; so->so_pcb = (caddr_t)inp; #ifdef INET6 if (V_ip6_auto_flowlabel) inp->inp_flags |= IN6P_AUTOFLOWLABEL; #endif inp->inp_gencnt = ++pcbinfo->ipi_gencnt; refcount_init(&inp->inp_refcount, 1); /* Reference from inpcbinfo */ /* * Routes in inpcb's can cache L2 as well; they are guaranteed * to be cleaned up. */ inp->inp_route.ro_flags = RT_LLE_CACHE; INP_LIST_WUNLOCK(pcbinfo); #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC) out: if (error != 0) { crfree(inp->inp_cred); uma_zfree(pcbinfo->ipi_zone, inp); } #endif return (error); } #ifdef INET int in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { int anonport, error; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) return (EINVAL); anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0; error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr, &inp->inp_lport, cred); if (error) return (error); if (in_pcbinshash(inp) != 0) { inp->inp_laddr.s_addr = INADDR_ANY; inp->inp_lport = 0; return (EAGAIN); } if (anonport) inp->inp_flags |= INP_ANONPORT; return (0); } #endif /* * Select a local port (number) to use. */ #if defined(INET) || defined(INET6) int in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, struct ucred *cred, int lookupflags) { struct inpcbinfo *pcbinfo; struct inpcb *tmpinp; unsigned short *lastport; int count, dorandom, error; u_short aux, first, last, lport; #ifdef INET struct in_addr laddr; #endif pcbinfo = inp->inp_pcbinfo; /* * Because no actual state changes occur here, a global write lock on * the pcbinfo isn't required. */ INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(pcbinfo); if (inp->inp_flags & INP_HIGHPORT) { first = V_ipport_hifirstauto; /* sysctl */ last = V_ipport_hilastauto; lastport = &pcbinfo->ipi_lasthi; } else if (inp->inp_flags & INP_LOWPORT) { error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0); if (error) return (error); first = V_ipport_lowfirstauto; /* 1023 */ last = V_ipport_lowlastauto; /* 600 */ lastport = &pcbinfo->ipi_lastlow; } else { first = V_ipport_firstauto; /* sysctl */ last = V_ipport_lastauto; lastport = &pcbinfo->ipi_lastport; } /* * For UDP(-Lite), use random port allocation as long as the user * allows it. For TCP (and as of yet unknown) connections, * use random port allocation only if the user allows it AND * ipport_tick() allows it. */ if (V_ipport_randomized && (!V_ipport_stoprandom || pcbinfo == &V_udbinfo || pcbinfo == &V_ulitecbinfo)) dorandom = 1; else dorandom = 0; /* * It makes no sense to do random port allocation if * we have the only port available. */ if (first == last) dorandom = 0; /* Make sure to not include UDP(-Lite) packets in the count. */ if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo) V_ipport_tcpallocs++; /* * Instead of having two loops further down counting up or down * make sure that first is always <= last and go with only one * code path implementing all logic. */ if (first > last) { aux = first; first = last; last = aux; } #ifdef INET /* Make the compiler happy. */ laddr.s_addr = 0; if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p", __func__, inp)); laddr = *laddrp; } #endif tmpinp = NULL; /* Make compiler happy. */ lport = *lportp; if (dorandom) *lastport = first + (arc4random() % (last - first)); count = last - first; do { if (count-- < 0) /* completely used? */ return (EADDRNOTAVAIL); ++*lastport; if (*lastport < first || *lastport > last) *lastport = first; lport = htons(*lastport); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) tmpinp = in6_pcblookup_local(pcbinfo, &inp->in6p_laddr, lport, lookupflags, cred); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tmpinp = in_pcblookup_local(pcbinfo, laddr, lport, lookupflags, cred); #endif } while (tmpinp != NULL); #ifdef INET if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) laddrp->s_addr = laddr.s_addr; #endif *lportp = lport; return (0); } /* * Return cached socket options. */ int inp_so_options(const struct inpcb *inp) { int so_options; so_options = 0; if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) so_options |= SO_REUSEPORT_LB; if ((inp->inp_flags2 & INP_REUSEPORT) != 0) so_options |= SO_REUSEPORT; if ((inp->inp_flags2 & INP_REUSEADDR) != 0) so_options |= SO_REUSEADDR; return (so_options); } #endif /* INET || INET6 */ /* * Check if a new BINDMULTI socket is allowed to be created. * * ni points to the new inp. * oi points to the exisitng inp. * * This checks whether the existing inp also has BINDMULTI and * whether the credentials match. */ int in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi) { /* Check permissions match */ if ((ni->inp_flags2 & INP_BINDMULTI) && (ni->inp_cred->cr_uid != oi->inp_cred->cr_uid)) return (0); /* Check the existing inp has BINDMULTI set */ if ((ni->inp_flags2 & INP_BINDMULTI) && ((oi->inp_flags2 & INP_BINDMULTI) == 0)) return (0); /* * We're okay - either INP_BINDMULTI isn't set on ni, or * it is and it matches the checks. */ return (1); } #ifdef INET /* * Set up a bind operation on a PCB, performing port allocation * as required, but do not actually modify the PCB. Callers can * either complete the bind by setting inp_laddr/inp_lport and * calling in_pcbinshash(), or they can just use the resulting * port and address to authorise the sending of a once-off packet. * * On error, the values of *laddrp and *lportp are not changed. */ int in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, u_short *lportp, struct ucred *cred) { struct socket *so = inp->inp_socket; struct sockaddr_in *sin; struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct in_addr laddr; u_short lport = 0; int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT); int error; /* * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here * so that we don't have to add to the (already messy) code below. */ int reuseport_lb = (so->so_options & SO_REUSEPORT_LB); /* * No state changes, so read locks are sufficient here. */ INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(pcbinfo); if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */ return (EADDRNOTAVAIL); laddr.s_addr = *laddrp; if (nam != NULL && laddr.s_addr != INADDR_ANY) return (EINVAL); if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) lookupflags = INPLOOKUP_WILDCARD; if (nam == NULL) { if ((error = prison_local_ip4(cred, &laddr)) != 0) return (error); } else { sin = (struct sockaddr_in *)nam; if (nam->sa_len != sizeof (*sin)) return (EINVAL); #ifdef notdef /* * We should check the family, but old programs * incorrectly fail to initialize it. */ if (sin->sin_family != AF_INET) return (EAFNOSUPPORT); #endif error = prison_local_ip4(cred, &sin->sin_addr); if (error) return (error); if (sin->sin_port != *lportp) { /* Don't allow the port to change. */ if (*lportp != 0) return (EINVAL); lport = sin->sin_port; } /* NB: lport is left as 0 if the port isn't being changed. */ if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) { /* * Treat SO_REUSEADDR as SO_REUSEPORT for multicast; * allow complete duplication of binding if * SO_REUSEPORT is set, or if SO_REUSEADDR is set * and a multicast address is bound on both * new and duplicated sockets. */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0) reuseport = SO_REUSEADDR|SO_REUSEPORT; /* * XXX: How to deal with SO_REUSEPORT_LB here? * Treat same as SO_REUSEPORT for now. */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0) reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB; } else if (sin->sin_addr.s_addr != INADDR_ANY) { sin->sin_port = 0; /* yech... */ bzero(&sin->sin_zero, sizeof(sin->sin_zero)); /* * Is the address a local IP address? * If INP_BINDANY is set, then the socket may be bound * to any endpoint address, local or not. */ if ((inp->inp_flags & INP_BINDANY) == 0 && ifa_ifwithaddr_check((struct sockaddr *)sin) == 0) return (EADDRNOTAVAIL); } laddr = sin->sin_addr; if (lport) { struct inpcb *t; struct tcptw *tw; /* GROSS */ if (ntohs(lport) <= V_ipport_reservedhigh && ntohs(lport) >= V_ipport_reservedlow && priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0)) return (EACCES); if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT, 0) != 0) { t = in_pcblookup_local(pcbinfo, sin->sin_addr, lport, INPLOOKUP_WILDCARD, cred); /* * XXX * This entire block sorely needs a rewrite. */ if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && ((t->inp_flags & INP_TIMEWAIT) == 0) && (so->so_type != SOCK_STREAM || ntohl(t->inp_faddr.s_addr) == INADDR_ANY) && (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || ntohl(t->inp_laddr.s_addr) != INADDR_ANY || (t->inp_flags2 & INP_REUSEPORT) || (t->inp_flags2 & INP_REUSEPORT_LB) == 0) && (inp->inp_cred->cr_uid != t->inp_cred->cr_uid)) return (EADDRINUSE); /* * If the socket is a BINDMULTI socket, then * the credentials need to match and the * original socket also has to have been bound * with BINDMULTI. */ if (t && (! in_pcbbind_check_bindmulti(inp, t))) return (EADDRINUSE); } t = in_pcblookup_local(pcbinfo, sin->sin_addr, lport, lookupflags, cred); if (t && (t->inp_flags & INP_TIMEWAIT)) { /* * XXXRW: If an incpb has had its timewait * state recycled, we treat the address as * being in use (for now). This is better * than a panic, but not desirable. */ tw = intotw(t); if (tw == NULL || ((reuseport & tw->tw_so_options) == 0 && (reuseport_lb & tw->tw_so_options) == 0)) { return (EADDRINUSE); } } else if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) && (reuseport & inp_so_options(t)) == 0 && (reuseport_lb & inp_so_options(t)) == 0) { #ifdef INET6 if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY || ntohl(t->inp_laddr.s_addr) != INADDR_ANY || (inp->inp_vflag & INP_IPV6PROTO) == 0 || (t->inp_vflag & INP_IPV6PROTO) == 0) #endif return (EADDRINUSE); if (t && (! in_pcbbind_check_bindmulti(inp, t))) return (EADDRINUSE); } } } if (*lportp != 0) lport = *lportp; if (lport == 0) { error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags); if (error != 0) return (error); } *laddrp = laddr.s_addr; *lportp = lport; return (0); } /* * Connect from a socket to a specified address. * Both address and port must be specified in argument sin. * If don't have a local address for this socket yet, * then pick one. */ int in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred, struct mbuf *m) { u_short lport, fport; in_addr_t laddr, faddr; int anonport, error; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); lport = inp->inp_lport; laddr = inp->inp_laddr.s_addr; anonport = (lport == 0); error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport, NULL, cred); if (error) return (error); /* Do the initial binding of the local address if required. */ if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { inp->inp_lport = lport; inp->inp_laddr.s_addr = laddr; if (in_pcbinshash(inp) != 0) { inp->inp_laddr.s_addr = INADDR_ANY; inp->inp_lport = 0; return (EAGAIN); } } /* Commit the remaining changes. */ inp->inp_lport = lport; inp->inp_laddr.s_addr = laddr; inp->inp_faddr.s_addr = faddr; inp->inp_fport = fport; in_pcbrehash_mbuf(inp, m); if (anonport) inp->inp_flags |= INP_ANONPORT; return (0); } int in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred) { return (in_pcbconnect_mbuf(inp, nam, cred, NULL)); } /* * Do proper source address selection on an unbound socket in case * of connect. Take jails into account as well. */ int in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr, struct ucred *cred) { struct ifaddr *ifa; struct sockaddr *sa; struct sockaddr_in *sin; struct route sro; int error; KASSERT(laddr != NULL, ("%s: laddr NULL", __func__)); /* * Bypass source address selection and use the primary jail IP * if requested. */ if (cred != NULL && !prison_saddrsel_ip4(cred, laddr)) return (0); error = 0; bzero(&sro, sizeof(sro)); sin = (struct sockaddr_in *)&sro.ro_dst; sin->sin_family = AF_INET; sin->sin_len = sizeof(struct sockaddr_in); sin->sin_addr.s_addr = faddr->s_addr; /* * If route is known our src addr is taken from the i/f, * else punt. * * Find out route to destination. */ if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0) in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum); /* * If we found a route, use the address corresponding to * the outgoing interface. * * Otherwise assume faddr is reachable on a directly connected * network and try to find a corresponding interface to take * the source address from. */ NET_EPOCH_ENTER(); if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) { struct in_ifaddr *ia; struct ifnet *ifp; ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin, inp->inp_socket->so_fibnum)); if (ia == NULL) { ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0, inp->inp_socket->so_fibnum)); } if (ia == NULL) { error = ENETUNREACH; goto done; } if (cred == NULL || !prison_flag(cred, PR_IP4)) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } ifp = ia->ia_ifp; ia = NULL; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } /* * If the outgoing interface on the route found is not * a loopback interface, use the address from that interface. * In case of jails do those three steps: * 1. check if the interface address belongs to the jail. If so use it. * 2. check if we have any address on the outgoing interface * belonging to this jail. If so use it. * 3. as a last resort return the 'default' jail address. */ if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) { struct in_ifaddr *ia; struct ifnet *ifp; /* If not jailed, use the default returned. */ if (cred == NULL || !prison_flag(cred, PR_IP4)) { ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* Jailed. */ /* 1. Check if the iface address belongs to the jail. */ sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa; laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* * 2. Check if we have any address on the outgoing interface * belonging to this jail. */ ia = NULL; ifp = sro.ro_rt->rt_ifp; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } /* * The outgoing interface is marked with 'loopback net', so a route * to ourselves is here. * Try to find the interface of the destination address and then * take the address from there. That interface is not necessarily * a loopback interface. * In case of jails, check that it is an address of the jail * and if we cannot find, fall back to the 'default' jail address. */ if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) { struct sockaddr_in sain; struct in_ifaddr *ia; bzero(&sain, sizeof(struct sockaddr_in)); sain.sin_family = AF_INET; sain.sin_len = sizeof(struct sockaddr_in); sain.sin_addr.s_addr = faddr->s_addr; ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain), inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0, inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithaddr(sintosa(&sain))); if (cred == NULL || !prison_flag(cred, PR_IP4)) { if (ia == NULL) { error = ENETUNREACH; goto done; } laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } /* Jailed. */ if (ia != NULL) { struct ifnet *ifp; ifp = ia->ia_ifp; ia = NULL; CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { sa = ifa->ifa_addr; if (sa->sa_family != AF_INET) continue; sin = (struct sockaddr_in *)sa; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)ifa; break; } } if (ia != NULL) { laddr->s_addr = ia->ia_addr.sin_addr.s_addr; goto done; } } /* 3. As a last resort return the 'default' jail address. */ error = prison_get_ip4(cred, laddr); goto done; } done: NET_EPOCH_EXIT(); if (sro.ro_rt != NULL) RTFREE(sro.ro_rt); return (error); } /* * Set up for a connect from a socket to the specified address. * On entry, *laddrp and *lportp should contain the current local * address and port for the PCB; these are updated to the values * that should be placed in inp_laddr and inp_lport to complete * the connect. * * On success, *faddrp and *fportp will be set to the remote address * and port. These are not updated in the error case. * * If the operation fails because the connection already exists, * *oinpp will be set to the PCB of that connection so that the * caller can decide to override it. In all other cases, *oinpp * is set to NULL. */ int in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp, struct inpcb **oinpp, struct ucred *cred) { struct rm_priotracker in_ifa_tracker; struct sockaddr_in *sin = (struct sockaddr_in *)nam; struct in_ifaddr *ia; struct inpcb *oinp; struct in_addr laddr, faddr; u_short lport, fport; int error; /* * Because a global state change doesn't actually occur here, a read * lock is sufficient. */ INP_LOCK_ASSERT(inp); INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo); if (oinpp != NULL) *oinpp = NULL; if (nam->sa_len != sizeof (*sin)) return (EINVAL); if (sin->sin_family != AF_INET) return (EAFNOSUPPORT); if (sin->sin_port == 0) return (EADDRNOTAVAIL); laddr.s_addr = *laddrp; lport = *lportp; faddr = sin->sin_addr; fport = sin->sin_port; if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) { /* * If the destination address is INADDR_ANY, * use the primary local address. * If the supplied address is INADDR_BROADCAST, * and the primary interface supports broadcast, * choose the broadcast address for that interface. */ if (faddr.s_addr == INADDR_ANY) { IN_IFADDR_RLOCK(&in_ifa_tracker); faddr = IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr; IN_IFADDR_RUNLOCK(&in_ifa_tracker); if (cred != NULL && (error = prison_get_ip4(cred, &faddr)) != 0) return (error); } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) { IN_IFADDR_RLOCK(&in_ifa_tracker); if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags & IFF_BROADCAST) faddr = satosin(&CK_STAILQ_FIRST( &V_in_ifaddrhead)->ia_broadaddr)->sin_addr; IN_IFADDR_RUNLOCK(&in_ifa_tracker); } } if (laddr.s_addr == INADDR_ANY) { error = in_pcbladdr(inp, &faddr, &laddr, cred); /* * If the destination address is multicast and an outgoing * interface has been set as a multicast option, prefer the * address of that interface as our source address. */ if (IN_MULTICAST(ntohl(faddr.s_addr)) && inp->inp_moptions != NULL) { struct ip_moptions *imo; struct ifnet *ifp; imo = inp->inp_moptions; if (imo->imo_multicast_ifp != NULL) { ifp = imo->imo_multicast_ifp; IN_IFADDR_RLOCK(&in_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { if ((ia->ia_ifp == ifp) && (cred == NULL || prison_check_ip4(cred, &ia->ia_addr.sin_addr) == 0)) break; } if (ia == NULL) error = EADDRNOTAVAIL; else { laddr = ia->ia_addr.sin_addr; error = 0; } IN_IFADDR_RUNLOCK(&in_ifa_tracker); } } if (error) return (error); } oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, laddr, lport, 0, NULL); if (oinp != NULL) { if (oinpp != NULL) *oinpp = oinp; return (EADDRINUSE); } if (lport == 0) { error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport, cred); if (error) return (error); } *laddrp = laddr.s_addr; *lportp = lport; *faddrp = faddr.s_addr; *fportp = fport; return (0); } void in_pcbdisconnect(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo); inp->inp_faddr.s_addr = INADDR_ANY; inp->inp_fport = 0; in_pcbrehash(inp); } #endif /* INET */ /* * in_pcbdetach() is responsibe for disassociating a socket from an inpcb. * For most protocols, this will be invoked immediately prior to calling * in_pcbfree(). However, with TCP the inpcb may significantly outlive the * socket, in which case in_pcbfree() is deferred. */ void in_pcbdetach(struct inpcb *inp) { KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__)); #ifdef RATELIMIT if (inp->inp_snd_tag != NULL) in_pcbdetach_txrtlmt(inp); #endif inp->inp_socket->so_pcb = NULL; inp->inp_socket = NULL; } /* * in_pcbref() bumps the reference count on an inpcb in order to maintain * stability of an inpcb pointer despite the inpcb lock being released. This * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded, * but where the inpcb lock may already held, or when acquiring a reference * via a pcbgroup. * * in_pcbref() should be used only to provide brief memory stability, and * must always be followed by a call to INP_WLOCK() and in_pcbrele() to * garbage collect the inpcb if it has been in_pcbfree()'d from another * context. Until in_pcbrele() has returned that the inpcb is still valid, * lock and rele are the *only* safe operations that may be performed on the * inpcb. * * While the inpcb will not be freed, releasing the inpcb lock means that the * connection's state may change, so the caller should be careful to * revalidate any cached state on reacquiring the lock. Drop the reference * using in_pcbrele(). */ void in_pcbref(struct inpcb *inp) { KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); refcount_acquire(&inp->inp_refcount); } /* * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we * return a flag indicating whether or not the inpcb remains valid. If it is * valid, we return with the inpcb lock held. * * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a * reference on an inpcb. Historically more work was done here (actually, in * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the * need for the pcbinfo lock in in_pcbrele(). Deferring the free is entirely * about memory stability (and continued use of the write lock). */ int in_pcbrele_rlocked(struct inpcb *inp) { struct inpcbinfo *pcbinfo; KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); INP_RLOCK_ASSERT(inp); if (refcount_release(&inp->inp_refcount) == 0) { /* * If the inpcb has been freed, let the caller know, even if * this isn't the last reference. */ if (inp->inp_flags2 & INP_FREED) { INP_RUNLOCK(inp); return (1); } return (0); } KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); #ifdef TCPHPTS if (inp->inp_in_hpts || inp->inp_in_input) { struct tcp_hpts_entry *hpts; /* * We should not be on the hpts at * this point in any form. we must * get the lock to be sure. */ hpts = tcp_hpts_lock(inp); if (inp->inp_in_hpts) panic("Hpts:%p inp:%p at free still on hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if (inp->inp_in_input) panic("Hpts:%p inp:%p at free still on input hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); } #endif INP_RUNLOCK(inp); pcbinfo = inp->inp_pcbinfo; uma_zfree(pcbinfo->ipi_zone, inp); return (1); } int in_pcbrele_wlocked(struct inpcb *inp) { struct inpcbinfo *pcbinfo; KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__)); INP_WLOCK_ASSERT(inp); if (refcount_release(&inp->inp_refcount) == 0) { /* * If the inpcb has been freed, let the caller know, even if * this isn't the last reference. */ if (inp->inp_flags2 & INP_FREED) { INP_WUNLOCK(inp); return (1); } return (0); } KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); #ifdef TCPHPTS if (inp->inp_in_hpts || inp->inp_in_input) { struct tcp_hpts_entry *hpts; /* * We should not be on the hpts at * this point in any form. we must * get the lock to be sure. */ hpts = tcp_hpts_lock(inp); if (inp->inp_in_hpts) panic("Hpts:%p inp:%p at free still on hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if (inp->inp_in_input) panic("Hpts:%p inp:%p at free still on input hpts", hpts, inp); mtx_unlock(&hpts->p_mtx); } #endif INP_WUNLOCK(inp); pcbinfo = inp->inp_pcbinfo; uma_zfree(pcbinfo->ipi_zone, inp); return (1); } /* * Temporary wrapper. */ int in_pcbrele(struct inpcb *inp) { return (in_pcbrele_wlocked(inp)); } void in_pcblist_rele_rlocked(epoch_context_t ctx) { struct in_pcblist *il; struct inpcb *inp; struct inpcbinfo *pcbinfo; int i, n; il = __containerof(ctx, struct in_pcblist, il_epoch_ctx); pcbinfo = il->il_pcbinfo; n = il->il_count; INP_INFO_WLOCK(pcbinfo); for (i = 0; i < n; i++) { inp = il->il_inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); free(il, M_TEMP); } static void inpcbport_free(epoch_context_t ctx) { struct inpcbport *phd; phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx); free(phd, M_PCB); } static void in_pcbfree_deferred(epoch_context_t ctx) { struct inpcb *inp; int released __unused; inp = __containerof(ctx, struct inpcb, inp_epoch_ctx); INP_WLOCK(inp); #ifdef INET inp_freemoptions(inp->inp_moptions); inp->inp_moptions = NULL; #endif /* XXXRW: Do as much as possible here. */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (inp->inp_sp != NULL) ipsec_delete_pcbpolicy(inp); #endif #ifdef INET6 if (inp->inp_vflag & INP_IPV6PROTO) { ip6_freepcbopts(inp->in6p_outputopts); ip6_freemoptions(inp->in6p_moptions); inp->in6p_moptions = NULL; } #endif if (inp->inp_options) (void)m_free(inp->inp_options); inp->inp_vflag = 0; crfree(inp->inp_cred); #ifdef MAC mac_inpcb_destroy(inp); #endif released = in_pcbrele_wlocked(inp); MPASS(released); } /* * Unconditionally schedule an inpcb to be freed by decrementing its * reference count, which should occur only after the inpcb has been detached * from its socket. If another thread holds a temporary reference (acquired * using in_pcbref()) then the free is deferred until that reference is * released using in_pcbrele(), but the inpcb is still unlocked. Almost all * work, including removal from global lists, is done in this context, where * the pcbinfo lock is held. */ void in_pcbfree(struct inpcb *inp) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__)); KASSERT((inp->inp_flags2 & INP_FREED) == 0, ("%s: called twice for pcb %p", __func__, inp)); if (inp->inp_flags2 & INP_FREED) { INP_WUNLOCK(inp); return; } #ifdef INVARIANTS if (pcbinfo == &V_tcbinfo) { INP_INFO_LOCK_ASSERT(pcbinfo); } else { INP_INFO_WLOCK_ASSERT(pcbinfo); } #endif INP_WLOCK_ASSERT(inp); INP_LIST_WLOCK(pcbinfo); in_pcbremlists(inp); INP_LIST_WUNLOCK(pcbinfo); RO_INVALIDATE_CACHE(&inp->inp_route); /* mark as destruction in progress */ inp->inp_flags2 |= INP_FREED; INP_WUNLOCK(inp); epoch_call(net_epoch_preempt, &inp->inp_epoch_ctx, in_pcbfree_deferred); } /* * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and * port reservation, and preventing it from being returned by inpcb lookups. * * It is used by TCP to mark an inpcb as unused and avoid future packet * delivery or event notification when a socket remains open but TCP has * closed. This might occur as a result of a shutdown()-initiated TCP close * or a RST on the wire, and allows the port binding to be reused while still * maintaining the invariant that so_pcb always points to a valid inpcb until * in_pcbdetach(). * * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by * in_pcbnotifyall() and in_pcbpurgeif0()? */ void in_pcbdrop(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); #ifdef INVARIANTS if (inp->inp_socket != NULL && inp->inp_ppcb != NULL) MPASS(inp->inp_refcount > 1); #endif /* * XXXRW: Possibly we should protect the setting of INP_DROPPED with * the hash lock...? */ inp->inp_flags |= INP_DROPPED; if (inp->inp_flags & INP_INHASHLIST) { struct inpcbport *phd = inp->inp_phd; INP_HASH_WLOCK(inp->inp_pcbinfo); in_pcbremlbgrouphash(inp); CK_LIST_REMOVE(inp, inp_hash); CK_LIST_REMOVE(inp, inp_portlist); if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { CK_LIST_REMOVE(phd, phd_hash); epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); } INP_HASH_WUNLOCK(inp->inp_pcbinfo); inp->inp_flags &= ~INP_INHASHLIST; #ifdef PCBGROUP in_pcbgroup_remove(inp); #endif } } #ifdef INET /* * Common routines to return the socket addresses associated with inpcbs. */ struct sockaddr * in_sockaddr(in_port_t port, struct in_addr *addr_p) { struct sockaddr_in *sin; sin = malloc(sizeof *sin, M_SONAME, M_WAITOK | M_ZERO); sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = *addr_p; sin->sin_port = port; return (struct sockaddr *)sin; } int in_getsockaddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_lport; addr = inp->inp_laddr; INP_RUNLOCK(inp); *nam = in_sockaddr(port, &addr); return 0; } int in_getpeeraddr(struct socket *so, struct sockaddr **nam) { struct inpcb *inp; struct in_addr addr; in_port_t port; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL")); INP_RLOCK(inp); port = inp->inp_fport; addr = inp->inp_faddr; INP_RUNLOCK(inp); *nam = in_sockaddr(port, &addr); return 0; } void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno, struct inpcb *(*notify)(struct inpcb *, int)) { struct inpcb *inp, *inp_temp; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) { INP_WLOCK(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV4) == 0) { INP_WUNLOCK(inp); continue; } #endif if (inp->inp_faddr.s_addr != faddr.s_addr || inp->inp_socket == NULL) { INP_WUNLOCK(inp); continue; } if ((*notify)(inp, errno)) INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); } void in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp) { struct inpcb *inp; struct ip_moptions *imo; int i, gap; INP_INFO_WLOCK(pcbinfo); CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) { INP_WLOCK(inp); imo = inp->inp_moptions; if ((inp->inp_vflag & INP_IPV4) && imo != NULL) { /* * Unselect the outgoing interface if it is being * detached. */ if (imo->imo_multicast_ifp == ifp) imo->imo_multicast_ifp = NULL; /* * Drop multicast group membership if we joined * through the interface being detached. * * XXX This can all be deferred to an epoch_call */ for (i = 0, gap = 0; i < imo->imo_num_memberships; i++) { if (imo->imo_membership[i]->inm_ifp == ifp) { IN_MULTI_LOCK_ASSERT(); in_leavegroup_locked(imo->imo_membership[i], NULL); gap++; } else if (gap != 0) imo->imo_membership[i - gap] = imo->imo_membership[i]; } imo->imo_num_memberships -= gap; } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(pcbinfo); } /* * Lookup a PCB based on the local address and port. Caller must hold the * hash lock. No inpcb locks or references are acquired. */ #define INP_LOOKUP_MAPPED_PCB_COST 3 struct inpcb * in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr, u_short lport, int lookupflags, struct ucred *cred) { struct inpcb *inp; #ifdef INET6 int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST; #else int matchwild = 3; #endif int wildcard; KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); INP_HASH_LOCK_ASSERT(pcbinfo); if ((lookupflags & INPLOOKUP_WILDCARD) == 0) { struct inpcbhead *head; /* * Look for an unconnected (wildcard foreign addr) PCB that * matches the local address and port we're looking for. */ head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == INADDR_ANY && inp->inp_laddr.s_addr == laddr.s_addr && inp->inp_lport == lport) { /* * Found? */ if (cred == NULL || prison_equal_ip4(cred->cr_prison, inp->inp_cred->cr_prison)) return (inp); } } /* * Not found. */ return (NULL); } else { struct inpcbporthead *porthash; struct inpcbport *phd; struct inpcb *match = NULL; /* * Best fit PCB lookup. * * First see if this local port is in use by looking on the * port hash list. */ porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport, pcbinfo->ipi_porthashmask)]; CK_LIST_FOREACH(phd, porthash, phd_hash) { if (phd->phd_port == lport) break; } if (phd != NULL) { /* * Port is in use by one or more PCBs. Look for best * fit. */ CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) { wildcard = 0; if (cred != NULL && !prison_equal_ip4(inp->inp_cred->cr_prison, cred->cr_prison)) continue; #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; /* * We never select the PCB that has * INP_IPV6 flag and is bound to :: if * we have another PCB which is bound * to 0.0.0.0. If a PCB has the * INP_IPV6 flag, then we set its cost * higher than IPv4 only PCBs. * * Note that the case only happens * when a socket is bound to ::, under * the condition that the use of the * mapped address is allowed. */ if ((inp->inp_vflag & INP_IPV6) != 0) wildcard += INP_LOOKUP_MAPPED_PCB_COST; #endif if (inp->inp_faddr.s_addr != INADDR_ANY) wildcard++; if (inp->inp_laddr.s_addr != INADDR_ANY) { if (laddr.s_addr == INADDR_ANY) wildcard++; else if (inp->inp_laddr.s_addr != laddr.s_addr) continue; } else { if (laddr.s_addr != INADDR_ANY) wildcard++; } if (wildcard < matchwild) { match = inp; matchwild = wildcard; if (matchwild == 0) break; } } } return (match); } } #undef INP_LOOKUP_MAPPED_PCB_COST static struct inpcb * in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo, const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr, uint16_t fport, int lookupflags) { struct inpcb *local_wild = NULL; const struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; struct inpcblbgroup *grp_local_wild; INP_HASH_LOCK_ASSERT(pcbinfo); hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBLBGROUP_PORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)]; /* * Order of socket selection: * 1. non-wild. * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD). * * NOTE: * - Load balanced group does not contain jailed sockets * - Load balanced group does not contain IPv4 mapped INET6 wild sockets */ LIST_FOREACH(grp, hdr, il_list) { #ifdef INET6 if (!(grp->il_vflag & INP_IPV4)) continue; #endif if (grp->il_lport == lport) { uint32_t idx = 0; int pkt_hash = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport); idx = pkt_hash % grp->il_inpcnt; if (grp->il_laddr.s_addr == laddr->s_addr) { return (grp->il_inp[idx]); } else { if (grp->il_laddr.s_addr == INADDR_ANY && (lookupflags & INPLOOKUP_WILDCARD)) { local_wild = grp->il_inp[idx]; grp_local_wild = grp; } } } } if (local_wild != NULL) { return (local_wild); } return (NULL); } #ifdef PCBGROUP /* * Lookup PCB in hash list, using pcbgroup tables. */ static struct inpcb * in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup, struct in_addr faddr, u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp) { struct inpcbhead *head; struct inpcb *inp, *tmpinp; u_short fport = fport_arg, lport = lport_arg; bool locked; /* * First look for an exact match. */ tmpinp = NULL; INP_GROUP_LOCK(pcbgroup); head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, pcbgroup->ipg_hashmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == faddr.s_addr && inp->inp_laddr.s_addr == laddr.s_addr && inp->inp_fport == fport && inp->inp_lport == lport) { /* * XXX We should be able to directly return * the inp here, without any checks. * Well unless both bound with SO_REUSEPORT? */ if (prison_flag(inp->inp_cred, PR_IP4)) goto found; if (tmpinp == NULL) tmpinp = inp; } } if (tmpinp != NULL) { inp = tmpinp; goto found; } #ifdef RSS /* * For incoming connections, we may wish to do a wildcard * match for an RSS-local socket. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif struct inpcb *jail_wild = NULL; struct inpcbhead *head; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbgroup->ipg_hashmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) goto found; else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; #ifdef INET6 if (inp == NULL) inp = local_wild_mapped; #endif if (inp != NULL) goto found; } #endif /* * Then look for a wildcard match, if requested. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif struct inpcb *jail_wild = NULL; struct inpcbhead *head; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->ipi_wildmask)]; CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) goto found; else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ inp = jail_wild; if (inp == NULL) inp = local_exact; if (inp == NULL) inp = local_wild; #ifdef INET6 if (inp == NULL) inp = local_wild_mapped; #endif if (inp != NULL) goto found; } /* if (lookupflags & INPLOOKUP_WILDCARD) */ INP_GROUP_UNLOCK(pcbgroup); return (NULL); found: if (lookupflags & INPLOOKUP_WLOCKPCB) locked = INP_TRY_WLOCK(inp); else if (lookupflags & INPLOOKUP_RLOCKPCB) locked = INP_TRY_RLOCK(inp); else panic("%s: locking bug", __func__); if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) { if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); return (NULL); } else if (!locked) in_pcbref(inp); INP_GROUP_UNLOCK(pcbgroup); if (!locked) { if (lookupflags & INPLOOKUP_WLOCKPCB) { INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) return (NULL); } else { INP_RLOCK(inp); if (in_pcbrele_rlocked(inp)) return (NULL); } } #ifdef INVARIANTS if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WLOCK_ASSERT(inp); else INP_RLOCK_ASSERT(inp); #endif return (inp); } #endif /* PCBGROUP */ /* * Lookup PCB in hash list, using pcbinfo tables. This variation assumes * that the caller has locked the hash list, and will not perform any further * locking or reference operations on either the hash list or the connection. */ static struct inpcb * in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags, struct ifnet *ifp) { struct inpcbhead *head; struct inpcb *inp, *tmpinp; u_short fport = fport_arg, lport = lport_arg; #ifdef INVARIANTS KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); if (!mtx_owned(&pcbinfo->ipi_hash_lock)) MPASS(in_epoch_verbose(net_epoch_preempt, 1)); #endif /* * First look for an exact match. */ tmpinp = NULL; head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr == faddr.s_addr && inp->inp_laddr.s_addr == laddr.s_addr && inp->inp_fport == fport && inp->inp_lport == lport) { /* * XXX We should be able to directly return * the inp here, without any checks. * Well unless both bound with SO_REUSEPORT? */ if (prison_flag(inp->inp_cred, PR_IP4)) return (inp); if (tmpinp == NULL) tmpinp = inp; } } if (tmpinp != NULL) return (tmpinp); /* * Then look in lb group (for wildcard match). */ if (pcbinfo->ipi_lbgrouphashbase != NULL && (lookupflags & INPLOOKUP_WILDCARD)) { inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr, fport, lookupflags); if (inp != NULL) { return (inp); } } /* * Then look for a wildcard match, if requested. */ if ((lookupflags & INPLOOKUP_WILDCARD) != 0) { struct inpcb *local_wild = NULL, *local_exact = NULL; #ifdef INET6 struct inpcb *local_wild_mapped = NULL; #endif struct inpcb *jail_wild = NULL; int injail; /* * Order of socket selection - we always prefer jails. * 1. jailed, non-wild. * 2. jailed, wild. * 3. non-jailed, non-wild. * 4. non-jailed, wild. */ head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport, 0, pcbinfo->ipi_hashmask)]; CK_LIST_FOREACH(inp, head, inp_hash) { #ifdef INET6 /* XXX inp locking */ if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport) continue; injail = prison_flag(inp->inp_cred, PR_IP4); if (injail) { if (prison_check_ip4(inp->inp_cred, &laddr) != 0) continue; } else { if (local_exact != NULL) continue; } if (inp->inp_laddr.s_addr == laddr.s_addr) { if (injail) return (inp); else local_exact = inp; } else if (inp->inp_laddr.s_addr == INADDR_ANY) { #ifdef INET6 /* XXX inp locking, NULL check */ if (inp->inp_vflag & INP_IPV6PROTO) local_wild_mapped = inp; else #endif if (injail) jail_wild = inp; else local_wild = inp; } } /* LIST_FOREACH */ if (jail_wild != NULL) return (jail_wild); if (local_exact != NULL) return (local_exact); if (local_wild != NULL) return (local_wild); #ifdef INET6 if (local_wild_mapped != NULL) return (local_wild_mapped); #endif } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */ return (NULL); } /* * Lookup PCB in hash list, using pcbinfo tables. This variation locks the * hash list lock, and will return the inpcb locked (i.e., requires * INPLOOKUP_LOCKPCB). */ static struct inpcb * in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) { struct inpcb *inp; INP_HASH_RLOCK(pcbinfo); inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp); if (inp != NULL) { if (lookupflags & INPLOOKUP_WLOCKPCB) { INP_WLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_WUNLOCK(inp); inp = NULL; } } else if (lookupflags & INPLOOKUP_RLOCKPCB) { INP_RLOCK(inp); if (__predict_false(inp->inp_flags2 & INP_FREED)) { INP_RUNLOCK(inp); inp = NULL; } } else panic("%s: locking bug", __func__); #ifdef INVARIANTS if (inp != NULL) { if (lookupflags & INPLOOKUP_WLOCKPCB) INP_WLOCK_ASSERT(inp); else INP_RLOCK_ASSERT(inp); } #endif } INP_HASH_RUNLOCK(pcbinfo); return (inp); } /* * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf * from which a pre-calculated hash value may be extracted. * * Possibly more of this logic should be in in_pcbgroup.c. */ struct inpcb * in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp) { #if defined(PCBGROUP) && !defined(RSS) struct inpcbgroup *pcbgroup; #endif KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, ("%s: LOCKPCB not set", __func__)); /* * When not using RSS, use connection groups in preference to the * reservation table when looking up 4-tuples. When using RSS, just * use the reservation table, due to the cost of the Toeplitz hash * in software. * * XXXRW: This policy belongs in the pcbgroup code, as in principle * we could be doing RSS with a non-Toeplitz hash that is affordable * in software. */ #if defined(PCBGROUP) && !defined(RSS) if (in_pcbgroup_enabled(pcbinfo)) { pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); } #endif return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } struct inpcb * in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport, struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp, struct mbuf *m) { #ifdef PCBGROUP struct inpcbgroup *pcbgroup; #endif KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0, ("%s: LOCKPCB not set", __func__)); #ifdef PCBGROUP /* * If we can use a hardware-generated hash to look up the connection * group, use that connection group to find the inpcb. Otherwise * fall back on a software hash -- or the reservation table if we're * using RSS. * * XXXRW: As above, that policy belongs in the pcbgroup code. */ if (in_pcbgroup_enabled(pcbinfo) && !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) { pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m), m->m_pkthdr.flowid); if (pcbgroup != NULL) return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #ifndef RSS pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr, fport); return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport, laddr, lport, lookupflags, ifp)); #endif } #endif return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport, lookupflags, ifp)); } #endif /* INET */ /* * Insert PCB onto various hash lists. */ static int in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update) { struct inpcbhead *pcbhash; struct inpcbporthead *pcbporthash; struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct inpcbport *phd; u_int32_t hashkey_faddr; int so_options; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); KASSERT((inp->inp_flags & INP_INHASHLIST) == 0, ("in_pcbinshash: INP_INHASHLIST")); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); else #endif hashkey_faddr = inp->inp_faddr.s_addr; pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; pcbporthash = &pcbinfo->ipi_porthashbase[ INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)]; /* * Add entry to load balance group. * Only do this if SO_REUSEPORT_LB is set. */ so_options = inp_so_options(inp); if (so_options & SO_REUSEPORT_LB) { int ret = in_pcbinslbgrouphash(inp); if (ret) { /* pcb lb group malloc fail (ret=ENOBUFS). */ return (ret); } } /* * Go through port list and look for a head for this lport. */ CK_LIST_FOREACH(phd, pcbporthash, phd_hash) { if (phd->phd_port == inp->inp_lport) break; } /* * If none exists, malloc one and tack it on. */ if (phd == NULL) { phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT); if (phd == NULL) { return (ENOBUFS); /* XXX */ } bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context)); phd->phd_port = inp->inp_lport; CK_LIST_INIT(&phd->phd_pcblist); CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash); } inp->inp_phd = phd; CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist); CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash); inp->inp_flags |= INP_INHASHLIST; #ifdef PCBGROUP if (do_pcbgroup_update) in_pcbgroup_update(inp); #endif return (0); } /* * For now, there are two public interfaces to insert an inpcb into the hash * lists -- one that does update pcbgroups, and one that doesn't. The latter * is used only in the TCP syncache, where in_pcbinshash is called before the * full 4-tuple is set for the inpcb, and we don't want to install in the * pcbgroup until later. * * XXXRW: This seems like a misfeature. in_pcbinshash should always update * connection groups, and partially initialised inpcbs should not be exposed * to either reservation hash tables or pcbgroups. */ int in_pcbinshash(struct inpcb *inp) { return (in_pcbinshash_internal(inp, 1)); } int in_pcbinshash_nopcbgroup(struct inpcb *inp) { return (in_pcbinshash_internal(inp, 0)); } /* * Move PCB to the proper hash bucket when { faddr, fport } have been * changed. NOTE: This does not handle the case of the lport changing (the * hashed port list would have to be updated as well), so the lport must * not change after in_pcbinshash() has been called. */ void in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; struct inpcbhead *head; u_int32_t hashkey_faddr; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); KASSERT(inp->inp_flags & INP_INHASHLIST, ("in_pcbrehash: !INP_INHASHLIST")); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr); else #endif hashkey_faddr = inp->inp_faddr.s_addr; head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr, inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)]; CK_LIST_REMOVE(inp, inp_hash); CK_LIST_INSERT_HEAD(head, inp, inp_hash); #ifdef PCBGROUP if (m != NULL) in_pcbgroup_update_mbuf(inp, m); else in_pcbgroup_update(inp); #endif } void in_pcbrehash(struct inpcb *inp) { in_pcbrehash_mbuf(inp, NULL); } /* * Remove PCB from various lists. */ static void in_pcbremlists(struct inpcb *inp) { struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; #ifdef INVARIANTS if (pcbinfo == &V_tcbinfo) { INP_INFO_RLOCK_ASSERT(pcbinfo); } else { INP_INFO_WLOCK_ASSERT(pcbinfo); } #endif INP_WLOCK_ASSERT(inp); INP_LIST_WLOCK_ASSERT(pcbinfo); inp->inp_gencnt = ++pcbinfo->ipi_gencnt; if (inp->inp_flags & INP_INHASHLIST) { struct inpcbport *phd = inp->inp_phd; INP_HASH_WLOCK(pcbinfo); /* XXX: Only do if SO_REUSEPORT_LB set? */ in_pcbremlbgrouphash(inp); CK_LIST_REMOVE(inp, inp_hash); CK_LIST_REMOVE(inp, inp_portlist); if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) { CK_LIST_REMOVE(phd, phd_hash); epoch_call(net_epoch_preempt, &phd->phd_epoch_ctx, inpcbport_free); } INP_HASH_WUNLOCK(pcbinfo); inp->inp_flags &= ~INP_INHASHLIST; } CK_LIST_REMOVE(inp, inp_list); pcbinfo->ipi_count--; #ifdef PCBGROUP in_pcbgroup_remove(inp); #endif } /* * Check for alternatives when higher level complains * about service problems. For now, invalidate cached * routing information. If the route was created dynamically * (by a redirect), time to try a default gateway again. */ void in_losing(struct inpcb *inp) { RO_INVALIDATE_CACHE(&inp->inp_route); return; } /* * A set label operation has occurred at the socket layer, propagate the * label change into the in_pcb for the socket. */ void in_pcbsosetlabel(struct socket *so) { #ifdef MAC struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL")); INP_WLOCK(inp); SOCK_LOCK(so); mac_inpcb_sosetlabel(so, inp); SOCK_UNLOCK(so); INP_WUNLOCK(inp); #endif } /* * ipport_tick runs once per second, determining if random port allocation * should be continued. If more than ipport_randomcps ports have been * allocated in the last second, then we return to sequential port * allocation. We return to random allocation only once we drop below * ipport_randomcps for at least ipport_randomtime seconds. */ static void ipport_tick(void *xtp) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */ if (V_ipport_tcpallocs <= V_ipport_tcplastcount + V_ipport_randomcps) { if (V_ipport_stoprandom > 0) V_ipport_stoprandom--; } else V_ipport_stoprandom = V_ipport_randomtime; V_ipport_tcplastcount = V_ipport_tcpallocs; CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL); } static void ip_fini(void *xtp) { callout_stop(&ipport_tick_callout); } /* * The ipport_callout should start running at about the time we attach the * inet or inet6 domains. */ static void ipport_tick_init(const void *unused __unused) { /* Start ipport_tick. */ callout_init(&ipport_tick_callout, 1); callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL); EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL, SHUTDOWN_PRI_DEFAULT); } SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE, ipport_tick_init, NULL); void inp_wlock(struct inpcb *inp) { INP_WLOCK(inp); } void inp_wunlock(struct inpcb *inp) { INP_WUNLOCK(inp); } void inp_rlock(struct inpcb *inp) { INP_RLOCK(inp); } void inp_runlock(struct inpcb *inp) { INP_RUNLOCK(inp); } #ifdef INVARIANT_SUPPORT void inp_lock_assert(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); } void inp_unlock_assert(struct inpcb *inp) { INP_UNLOCK_ASSERT(inp); } #endif void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg) { struct inpcb *inp; INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inp); func(inp, arg); INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); } struct socket * inp_inpcbtosocket(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); return (inp->inp_socket); } struct tcpcb * inp_inpcbtotcpcb(struct inpcb *inp) { INP_WLOCK_ASSERT(inp); return ((struct tcpcb *)inp->inp_ppcb); } int inp_ip_tos_get(const struct inpcb *inp) { return (inp->inp_ip_tos); } void inp_ip_tos_set(struct inpcb *inp, int val) { inp->inp_ip_tos = val; } void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, uint32_t *faddr, uint16_t *fp) { INP_LOCK_ASSERT(inp); *laddr = inp->inp_laddr.s_addr; *faddr = inp->inp_faddr.s_addr; *lp = inp->inp_lport; *fp = inp->inp_fport; } struct inpcb * so_sotoinpcb(struct socket *so) { return (sotoinpcb(so)); } struct tcpcb * so_sototcpcb(struct socket *so) { return (sototcpcb(so)); } /* * Create an external-format (``xinpcb'') structure using the information in * the kernel-format in_pcb structure pointed to by inp. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi) { xi->xi_len = sizeof(struct xinpcb); if (inp->inp_socket) sotoxsocket(inp->inp_socket, &xi->xi_socket); else bzero(&xi->xi_socket, sizeof(struct xsocket)); bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo)); xi->inp_gencnt = inp->inp_gencnt; xi->inp_ppcb = (uintptr_t)inp->inp_ppcb; xi->inp_flow = inp->inp_flow; xi->inp_flowid = inp->inp_flowid; xi->inp_flowtype = inp->inp_flowtype; xi->inp_flags = inp->inp_flags; xi->inp_flags2 = inp->inp_flags2; xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket; xi->in6p_cksum = inp->in6p_cksum; xi->in6p_hops = inp->in6p_hops; xi->inp_ip_tos = inp->inp_ip_tos; xi->inp_vflag = inp->inp_vflag; xi->inp_ip_ttl = inp->inp_ip_ttl; xi->inp_ip_p = inp->inp_ip_p; xi->inp_ip_minttl = inp->inp_ip_minttl; } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent) { char faddr_str[48], laddr_str[48]; db_print_indent(indent); db_printf("%s at %p\n", name, inc); indent += 2; #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { /* IPv6. */ ip6_sprintf(laddr_str, &inc->inc6_laddr); ip6_sprintf(faddr_str, &inc->inc6_faddr); } else #endif { /* IPv4. */ inet_ntoa_r(inc->inc_laddr, laddr_str); inet_ntoa_r(inc->inc_faddr, faddr_str); } db_print_indent(indent); db_printf("inc_laddr %s inc_lport %u\n", laddr_str, ntohs(inc->inc_lport)); db_print_indent(indent); db_printf("inc_faddr %s inc_fport %u\n", faddr_str, ntohs(inc->inc_fport)); } static void db_print_inpflags(int inp_flags) { int comma; comma = 0; if (inp_flags & INP_RECVOPTS) { db_printf("%sINP_RECVOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVRETOPTS) { db_printf("%sINP_RECVRETOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVDSTADDR) { db_printf("%sINP_RECVDSTADDR", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ORIGDSTADDR) { db_printf("%sINP_ORIGDSTADDR", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_HDRINCL) { db_printf("%sINP_HDRINCL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_HIGHPORT) { db_printf("%sINP_HIGHPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_LOWPORT) { db_printf("%sINP_LOWPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ANONPORT) { db_printf("%sINP_ANONPORT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVIF) { db_printf("%sINP_RECVIF", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_MTUDISC) { db_printf("%sINP_MTUDISC", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVTTL) { db_printf("%sINP_RECVTTL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_DONTFRAG) { db_printf("%sINP_DONTFRAG", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_RECVTOS) { db_printf("%sINP_RECVTOS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_IPV6_V6ONLY) { db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_PKTINFO) { db_printf("%sIN6P_PKTINFO", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_HOPLIMIT) { db_printf("%sIN6P_HOPLIMIT", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_HOPOPTS) { db_printf("%sIN6P_HOPOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_DSTOPTS) { db_printf("%sIN6P_DSTOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RTHDR) { db_printf("%sIN6P_RTHDR", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RTHDRDSTOPTS) { db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_TCLASS) { db_printf("%sIN6P_TCLASS", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_AUTOFLOWLABEL) { db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_TIMEWAIT) { db_printf("%sINP_TIMEWAIT", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_ONESBCAST) { db_printf("%sINP_ONESBCAST", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_DROPPED) { db_printf("%sINP_DROPPED", comma ? ", " : ""); comma = 1; } if (inp_flags & INP_SOCKREF) { db_printf("%sINP_SOCKREF", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_RFC2292) { db_printf("%sIN6P_RFC2292", comma ? ", " : ""); comma = 1; } if (inp_flags & IN6P_MTU) { db_printf("IN6P_MTU%s", comma ? ", " : ""); comma = 1; } } static void db_print_inpvflag(u_char inp_vflag) { int comma; comma = 0; if (inp_vflag & INP_IPV4) { db_printf("%sINP_IPV4", comma ? ", " : ""); comma = 1; } if (inp_vflag & INP_IPV6) { db_printf("%sINP_IPV6", comma ? ", " : ""); comma = 1; } if (inp_vflag & INP_IPV6PROTO) { db_printf("%sINP_IPV6PROTO", comma ? ", " : ""); comma = 1; } } static void db_print_inpcb(struct inpcb *inp, const char *name, int indent) { db_print_indent(indent); db_printf("%s at %p\n", name, inp); indent += 2; db_print_indent(indent); db_printf("inp_flow: 0x%x\n", inp->inp_flow); db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent); db_print_indent(indent); db_printf("inp_ppcb: %p inp_pcbinfo: %p inp_socket: %p\n", inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket); db_print_indent(indent); db_printf("inp_label: %p inp_flags: 0x%x (", inp->inp_label, inp->inp_flags); db_print_inpflags(inp->inp_flags); db_printf(")\n"); db_print_indent(indent); db_printf("inp_sp: %p inp_vflag: 0x%x (", inp->inp_sp, inp->inp_vflag); db_print_inpvflag(inp->inp_vflag); db_printf(")\n"); db_print_indent(indent); db_printf("inp_ip_ttl: %d inp_ip_p: %d inp_ip_minttl: %d\n", inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl); db_print_indent(indent); #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { db_printf("in6p_options: %p in6p_outputopts: %p " "in6p_moptions: %p\n", inp->in6p_options, inp->in6p_outputopts, inp->in6p_moptions); db_printf("in6p_icmp6filt: %p in6p_cksum %d " "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum, inp->in6p_hops); } else #endif { db_printf("inp_ip_tos: %d inp_ip_options: %p " "inp_ip_moptions: %p\n", inp->inp_ip_tos, inp->inp_options, inp->inp_moptions); } db_print_indent(indent); db_printf("inp_phd: %p inp_gencnt: %ju\n", inp->inp_phd, (uintmax_t)inp->inp_gencnt); } DB_SHOW_COMMAND(inpcb, db_show_inpcb) { struct inpcb *inp; if (!have_addr) { db_printf("usage: show inpcb \n"); return; } inp = (struct inpcb *)addr; db_print_inpcb(inp, "inpcb", 0); } #endif /* DDB */ #ifdef RATELIMIT /* * Modify TX rate limit based on the existing "inp->inp_snd_tag", * if any. */ int in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate) { union if_snd_tag_modify_params params = { .rate_limit.max_rate = max_pacing_rate, }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_modify == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_modify(mst, ¶ms); } return (error); } /* * Query existing TX rate limit based on the existing * "inp->inp_snd_tag", if any. */ int in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate) { union if_snd_tag_query_params params = { }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_query == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_query(mst, ¶ms); if (error == 0 && p_max_pacing_rate != NULL) *p_max_pacing_rate = params.rate_limit.max_rate; } return (error); } /* * Query existing TX queue level based on the existing * "inp->inp_snd_tag", if any. */ int in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level) { union if_snd_tag_query_params params = { }; struct m_snd_tag *mst; struct ifnet *ifp; int error; mst = inp->inp_snd_tag; if (mst == NULL) return (EINVAL); ifp = mst->ifp; if (ifp == NULL) return (EINVAL); if (ifp->if_snd_tag_query == NULL) return (EOPNOTSUPP); error = ifp->if_snd_tag_query(mst, ¶ms); if (error == 0 && p_txqueue_level != NULL) *p_txqueue_level = params.rate_limit.queue_level; return (error); } /* * Allocate a new TX rate limit send tag from the network interface * given by the "ifp" argument and save it in "inp->inp_snd_tag": */ int in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp, uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate) { union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = (max_pacing_rate == -1U) ? IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = flowid, .rate_limit.hdr.flowtype = flowtype, .rate_limit.max_rate = max_pacing_rate, }; int error; INP_WLOCK_ASSERT(inp); if (inp->inp_snd_tag != NULL) return (EINVAL); if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_alloc(ifp, ¶ms, &inp->inp_snd_tag); /* * At success increment the refcount on * the send tag's network interface: */ if (error == 0) if_ref(inp->inp_snd_tag->ifp); } return (error); } /* * Free an existing TX rate limit tag based on the "inp->inp_snd_tag", * if any: */ void in_pcbdetach_txrtlmt(struct inpcb *inp) { struct m_snd_tag *mst; struct ifnet *ifp; INP_WLOCK_ASSERT(inp); mst = inp->inp_snd_tag; inp->inp_snd_tag = NULL; if (mst == NULL) return; ifp = mst->ifp; if (ifp == NULL) return; /* * If the device was detached while we still had reference(s) * on the ifp, we assume if_snd_tag_free() was replaced with * stubs. */ ifp->if_snd_tag_free(mst); /* release reference count on network interface */ if_rele(ifp); } /* * This function should be called when the INP_RATE_LIMIT_CHANGED flag * is set in the fast path and will attach/detach/modify the TX rate * limit send tag based on the socket's so_max_pacing_rate value. */ void in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb) { struct socket *socket; uint32_t max_pacing_rate; bool did_upgrade; int error; if (inp == NULL) return; socket = inp->inp_socket; if (socket == NULL) return; if (!INP_WLOCKED(inp)) { /* * NOTE: If the write locking fails, we need to bail * out and use the non-ratelimited ring for the * transmit until there is a new chance to get the * write lock. */ if (!INP_TRY_UPGRADE(inp)) return; did_upgrade = 1; } else { did_upgrade = 0; } /* * NOTE: The so_max_pacing_rate value is read unlocked, * because atomic updates are not required since the variable * is checked at every mbuf we send. It is assumed that the * variable read itself will be atomic. */ max_pacing_rate = socket->so_max_pacing_rate; /* * NOTE: When attaching to a network interface a reference is * made to ensure the network interface doesn't go away until * all ratelimit connections are gone. The network interface * pointers compared below represent valid network interfaces, * except when comparing towards NULL. */ if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) { error = 0; } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) { if (inp->inp_snd_tag != NULL) in_pcbdetach_txrtlmt(inp); error = 0; } else if (inp->inp_snd_tag == NULL) { /* * In order to utilize packet pacing with RSS, we need * to wait until there is a valid RSS hash before we * can proceed: */ if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) { error = EAGAIN; } else { error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb), mb->m_pkthdr.flowid, max_pacing_rate); } } else { error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); } if (error == 0 || error == EOPNOTSUPP) inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; if (did_upgrade) INP_DOWNGRADE(inp); } /* * Track route changes for TX rate limiting. */ void in_pcboutput_eagain(struct inpcb *inp) { struct socket *socket; bool did_upgrade; if (inp == NULL) return; socket = inp->inp_socket; if (socket == NULL) return; if (inp->inp_snd_tag == NULL) return; if (!INP_WLOCKED(inp)) { /* * NOTE: If the write locking fails, we need to bail * out and use the non-ratelimited ring for the * transmit until there is a new chance to get the * write lock. */ if (!INP_TRY_UPGRADE(inp)) return; did_upgrade = 1; } else { did_upgrade = 0; } /* detach rate limiting */ in_pcbdetach_txrtlmt(inp); /* make sure new mbuf send tag allocation is made */ inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; if (did_upgrade) INP_DOWNGRADE(inp); } #endif /* RATELIMIT */ Index: head/sys/netinet/ip_carp.c =================================================================== --- head/sys/netinet/ip_carp.c (revision 336675) +++ head/sys/netinet/ip_carp.c (revision 336676) @@ -1,2297 +1,2297 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Michael Shalayeff. * Copyright (c) 2003 Ryan McBride. * Copyright (c) 2011 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_bpf.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #include #include #include #endif #ifdef INET #include #include #endif #ifdef INET6 #include #include #include #include #include #include #endif #include static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses"); struct carp_softc { struct ifnet *sc_carpdev; /* Pointer to parent ifnet. */ struct ifaddr **sc_ifas; /* Our ifaddrs. */ struct sockaddr_dl sc_addr; /* Our link level address. */ struct callout sc_ad_tmo; /* Advertising timeout. */ #ifdef INET struct callout sc_md_tmo; /* Master down timeout. */ #endif #ifdef INET6 struct callout sc_md6_tmo; /* XXX: Master down timeout. */ #endif struct mtx sc_mtx; int sc_vhid; int sc_advskew; int sc_advbase; int sc_naddrs; int sc_naddrs6; int sc_ifasiz; enum { INIT = 0, BACKUP, MASTER } sc_state; int sc_suppress; int sc_sendad_errors; #define CARP_SENDAD_MAX_ERRORS 3 int sc_sendad_success; #define CARP_SENDAD_MIN_SUCCESS 3 int sc_init_counter; uint64_t sc_counter; /* authentication */ #define CARP_HMAC_PAD 64 unsigned char sc_key[CARP_KEY_LEN]; unsigned char sc_pad[CARP_HMAC_PAD]; SHA1_CTX sc_sha1; TAILQ_ENTRY(carp_softc) sc_list; /* On the carp_if list. */ LIST_ENTRY(carp_softc) sc_next; /* On the global list. */ }; struct carp_if { #ifdef INET int cif_naddrs; #endif #ifdef INET6 int cif_naddrs6; #endif TAILQ_HEAD(, carp_softc) cif_vrs; #ifdef INET struct ip_moptions cif_imo; #endif #ifdef INET6 struct ip6_moptions cif_im6o; #endif struct ifnet *cif_ifp; struct mtx cif_mtx; uint32_t cif_flags; #define CIF_PROMISC 0x00000001 }; #define CARP_INET 0 #define CARP_INET6 1 static int proto_reg[] = {-1, -1}; /* * Brief design of carp(4). * * Any carp-capable ifnet may have a list of carp softcs hanging off * its ifp->if_carp pointer. Each softc represents one unique virtual * host id, or vhid. The softc has a back pointer to the ifnet. All * softcs are joined in a global list, which has quite limited use. * * Any interface address that takes part in CARP negotiation has a * pointer to the softc of its vhid, ifa->ifa_carp. That could be either * AF_INET or AF_INET6 address. * * Although, one can get the softc's backpointer to ifnet and traverse * through its ifp->if_addrhead queue to find all interface addresses * involved in CARP, we keep a growable array of ifaddr pointers. This * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that * do calls into the network stack, thus avoiding LORs. * * Locking: * * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(), * callout-driven events and ioctl()s. * * To traverse the list of softcs on an ifnet we use CIF_LOCK() or carp_sx. * To traverse the global list we use the mutex carp_mtx. * * Known issues with locking: * * - Sending ad, we put the pointer to the softc in an mtag, and no reference * counting is done on the softc. * - On module unload we may race (?) with packet processing thread * dereferencing our function pointers. */ /* Accept incoming CARP packets. */ -static VNET_DEFINE(int, carp_allow) = 1; +VNET_DEFINE_STATIC(int, carp_allow) = 1; #define V_carp_allow VNET(carp_allow) /* Set DSCP in outgoing CARP packets. */ -static VNET_DEFINE(int, carp_dscp) = 56; +VNET_DEFINE_STATIC(int, carp_dscp) = 56; #define V_carp_dscp VNET(carp_dscp) /* Preempt slower nodes. */ -static VNET_DEFINE(int, carp_preempt) = 0; +VNET_DEFINE_STATIC(int, carp_preempt) = 0; #define V_carp_preempt VNET(carp_preempt) /* Log level. */ -static VNET_DEFINE(int, carp_log) = 1; +VNET_DEFINE_STATIC(int, carp_log) = 1; #define V_carp_log VNET(carp_log) /* Global advskew demotion. */ -static VNET_DEFINE(int, carp_demotion) = 0; +VNET_DEFINE_STATIC(int, carp_demotion) = 0; #define V_carp_demotion VNET(carp_demotion) /* Send error demotion factor. */ -static VNET_DEFINE(int, carp_senderr_adj) = CARP_MAXSKEW; +VNET_DEFINE_STATIC(int, carp_senderr_adj) = CARP_MAXSKEW; #define V_carp_senderr_adj VNET(carp_senderr_adj) /* Iface down demotion factor. */ -static VNET_DEFINE(int, carp_ifdown_adj) = CARP_MAXSKEW; +VNET_DEFINE_STATIC(int, carp_ifdown_adj) = CARP_MAXSKEW; #define V_carp_ifdown_adj VNET(carp_ifdown_adj) static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS); static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS); static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS); SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW, 0, "CARP"); SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_allow_sysctl, "I", "Accept incoming CARP packets"); SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_dscp_sysctl, "I", "DSCP value for carp packets"); SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode"); SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(carp_log), 0, "CARP log level"); SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_demote_adj_sysctl, "I", "Adjust demotion factor (skew of advskew)"); SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment"); SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(carp_ifdown_adj), 0, "Interface down demotion factor adjustment"); VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats); VNET_PCPUSTAT_SYSINIT(carpstats); VNET_PCPUSTAT_SYSUNINIT(carpstats); #define CARPSTATS_ADD(name, val) \ counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \ sizeof(uint64_t)], (val)) #define CARPSTATS_INC(name) CARPSTATS_ADD(name, 1) SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats, carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)"); #define CARP_LOCK_INIT(sc) mtx_init(&(sc)->sc_mtx, "carp_softc", \ NULL, MTX_DEF) #define CARP_LOCK_DESTROY(sc) mtx_destroy(&(sc)->sc_mtx) #define CARP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) #define CARP_LOCK(sc) mtx_lock(&(sc)->sc_mtx) #define CARP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) #define CIF_LOCK_INIT(cif) mtx_init(&(cif)->cif_mtx, "carp_if", \ NULL, MTX_DEF) #define CIF_LOCK_DESTROY(cif) mtx_destroy(&(cif)->cif_mtx) #define CIF_LOCK_ASSERT(cif) mtx_assert(&(cif)->cif_mtx, MA_OWNED) #define CIF_LOCK(cif) mtx_lock(&(cif)->cif_mtx) #define CIF_UNLOCK(cif) mtx_unlock(&(cif)->cif_mtx) #define CIF_FREE(cif) do { \ CIF_LOCK(cif); \ if (TAILQ_EMPTY(&(cif)->cif_vrs)) \ carp_free_if(cif); \ else \ CIF_UNLOCK(cif); \ } while (0) #define CARP_LOG(...) do { \ if (V_carp_log > 0) \ log(LOG_INFO, "carp: " __VA_ARGS__); \ } while (0) #define CARP_DEBUG(...) do { \ if (V_carp_log > 1) \ log(LOG_DEBUG, __VA_ARGS__); \ } while (0) #define IFNET_FOREACH_IFA(ifp, ifa) \ CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link) \ if ((ifa)->ifa_carp != NULL) #define CARP_FOREACH_IFA(sc, ifa) \ CARP_LOCK_ASSERT(sc); \ for (int _i = 0; \ _i < (sc)->sc_naddrs + (sc)->sc_naddrs6 && \ ((ifa) = sc->sc_ifas[_i]) != NULL; \ ++_i) #define IFNET_FOREACH_CARP(ifp, sc) \ KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) || \ sx_xlocked(&carp_sx), ("cif_vrs not locked")); \ TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list) #define DEMOTE_ADVSKEW(sc) \ (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ? \ CARP_MAXSKEW : ((sc)->sc_advskew + V_carp_demotion)) static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t); static struct carp_softc *carp_alloc(struct ifnet *); static void carp_destroy(struct carp_softc *); static struct carp_if *carp_alloc_if(struct ifnet *); static void carp_free_if(struct carp_if *); static void carp_set_state(struct carp_softc *, int, const char* reason); static void carp_sc_state(struct carp_softc *); static void carp_setrun(struct carp_softc *, sa_family_t); static void carp_master_down(void *); static void carp_master_down_locked(struct carp_softc *, const char* reason); static void carp_send_ad(void *); static void carp_send_ad_locked(struct carp_softc *); static void carp_addroute(struct carp_softc *); static void carp_ifa_addroute(struct ifaddr *); static void carp_delroute(struct carp_softc *); static void carp_ifa_delroute(struct ifaddr *); static void carp_send_ad_all(void *, int); static void carp_demote_adj(int, char *); static LIST_HEAD(, carp_softc) carp_list; static struct mtx carp_mtx; static struct sx carp_sx; static struct task carp_sendall_task = TASK_INITIALIZER(0, carp_send_ad_all, NULL); static void carp_hmac_prepare(struct carp_softc *sc) { uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT; uint8_t vhid = sc->sc_vhid & 0xff; struct ifaddr *ifa; int i, found; #ifdef INET struct in_addr last, cur, in; #endif #ifdef INET6 struct in6_addr last6, cur6, in6; #endif CARP_LOCK_ASSERT(sc); /* Compute ipad from key. */ bzero(sc->sc_pad, sizeof(sc->sc_pad)); bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); for (i = 0; i < sizeof(sc->sc_pad); i++) sc->sc_pad[i] ^= 0x36; /* Precompute first part of inner hash. */ SHA1Init(&sc->sc_sha1); SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version)); SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); #ifdef INET cur.s_addr = 0; do { found = 0; last = cur; cur.s_addr = 0xffffffff; CARP_FOREACH_IFA(sc, ifa) { in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; if (ifa->ifa_addr->sa_family == AF_INET && ntohl(in.s_addr) > ntohl(last.s_addr) && ntohl(in.s_addr) < ntohl(cur.s_addr)) { cur.s_addr = in.s_addr; found++; } } if (found) SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); } while (found); #endif /* INET */ #ifdef INET6 memset(&cur6, 0, sizeof(cur6)); do { found = 0; last6 = cur6; memset(&cur6, 0xff, sizeof(cur6)); CARP_FOREACH_IFA(sc, ifa) { in6 = ifatoia6(ifa)->ia_addr.sin6_addr; if (IN6_IS_SCOPE_EMBED(&in6)) in6.s6_addr16[1] = 0; if (ifa->ifa_addr->sa_family == AF_INET6 && memcmp(&in6, &last6, sizeof(in6)) > 0 && memcmp(&in6, &cur6, sizeof(in6)) < 0) { cur6 = in6; found++; } } if (found) SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); } while (found); #endif /* INET6 */ /* convert ipad to opad */ for (i = 0; i < sizeof(sc->sc_pad); i++) sc->sc_pad[i] ^= 0x36 ^ 0x5c; } static void carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2], unsigned char md[20]) { SHA1_CTX sha1ctx; CARP_LOCK_ASSERT(sc); /* fetch first half of inner hash */ bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); SHA1Final(md, &sha1ctx); /* outer hash */ SHA1Init(&sha1ctx); SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); SHA1Update(&sha1ctx, md, 20); SHA1Final(md, &sha1ctx); } static int carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2], unsigned char md[20]) { unsigned char md2[20]; CARP_LOCK_ASSERT(sc); carp_hmac_generate(sc, counter, md2); return (bcmp(md, md2, sizeof(md2))); } /* * process input packet. * we have rearranged checks order compared to the rfc, * but it seems more efficient this way or not possible otherwise. */ #ifdef INET int carp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct ip *ip = mtod(m, struct ip *); struct carp_header *ch; int iplen, len; iplen = *offp; *mp = NULL; CARPSTATS_INC(carps_ipackets); if (!V_carp_allow) { m_freem(m); return (IPPROTO_DONE); } /* verify that the IP TTL is 255. */ if (ip->ip_ttl != CARP_DFLTTL) { CARPSTATS_INC(carps_badttl); CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__, ip->ip_ttl, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } iplen = ip->ip_hl << 2; if (m->m_pkthdr.len < iplen + sizeof(*ch)) { CARPSTATS_INC(carps_badlen); CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) " "on %s\n", __func__, m->m_len - sizeof(struct ip), m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } if (iplen + sizeof(*ch) < m->m_len) { if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) { CARPSTATS_INC(carps_hdrops); CARP_DEBUG("%s: pullup failed\n", __func__); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); } ch = (struct carp_header *)((char *)ip + iplen); /* * verify that the received packet length is * equal to the CARP header */ len = iplen + sizeof(*ch); if (len > m->m_pkthdr.len) { CARPSTATS_INC(carps_badlen); CARP_DEBUG("%s: packet too short %d on %s\n", __func__, m->m_pkthdr.len, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } if ((m = m_pullup(m, len)) == NULL) { CARPSTATS_INC(carps_hdrops); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); ch = (struct carp_header *)((char *)ip + iplen); /* verify the CARP checksum */ m->m_data += iplen; if (in_cksum(m, len - iplen)) { CARPSTATS_INC(carps_badsum); CARP_DEBUG("%s: checksum failed on %s\n", __func__, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } m->m_data -= iplen; carp_input_c(m, ch, AF_INET); return (IPPROTO_DONE); } #endif #ifdef INET6 int carp6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct carp_header *ch; u_int len; CARPSTATS_INC(carps_ipackets6); if (!V_carp_allow) { m_freem(m); return (IPPROTO_DONE); } /* check if received on a valid carp interface */ if (m->m_pkthdr.rcvif->if_carp == NULL) { CARPSTATS_INC(carps_badif); CARP_DEBUG("%s: packet received on non-carp interface: %s\n", __func__, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } /* verify that the IP TTL is 255 */ if (ip6->ip6_hlim != CARP_DFLTTL) { CARPSTATS_INC(carps_badttl); CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__, ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } /* verify that we have a complete carp packet */ len = m->m_len; IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); if (ch == NULL) { CARPSTATS_INC(carps_badlen); CARP_DEBUG("%s: packet size %u too small\n", __func__, len); return (IPPROTO_DONE); } /* verify the CARP checksum */ m->m_data += *offp; if (in_cksum(m, sizeof(*ch))) { CARPSTATS_INC(carps_badsum); CARP_DEBUG("%s: checksum failed, on %s\n", __func__, m->m_pkthdr.rcvif->if_xname); m_freem(m); return (IPPROTO_DONE); } m->m_data -= *offp; carp_input_c(m, ch, AF_INET6); return (IPPROTO_DONE); } #endif /* INET6 */ /* * This routine should not be necessary at all, but some switches * (VMWare ESX vswitches) can echo our own packets back at us, * and we must ignore them or they will cause us to drop out of * MASTER mode. * * We cannot catch all cases of network loops. Instead, what we * do here is catch any packet that arrives with a carp header * with a VHID of 0, that comes from an address that is our own. * These packets are by definition "from us" (even if they are from * a misconfigured host that is pretending to be us). * * The VHID test is outside this mini-function. */ static int carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af) { #ifdef INET struct ip *ip4; struct in_addr in4; #endif #ifdef INET6 struct ip6_hdr *ip6; struct in6_addr in6; #endif switch (af) { #ifdef INET case AF_INET: ip4 = mtod(m, struct ip *); in4 = ifatoia(ifa)->ia_addr.sin_addr; return (in4.s_addr == ip4->ip_src.s_addr); #endif #ifdef INET6 case AF_INET6: ip6 = mtod(m, struct ip6_hdr *); in6 = ifatoia6(ifa)->ia_addr.sin6_addr; return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0); #endif default: break; } return (0); } static void carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) { struct ifnet *ifp = m->m_pkthdr.rcvif; struct ifaddr *ifa, *match; struct carp_softc *sc; uint64_t tmp_counter; struct timeval sc_tv, ch_tv; int error; /* * Verify that the VHID is valid on the receiving interface. * * There should be just one match. If there are none * the VHID is not valid and we drop the packet. If * there are multiple VHID matches, take just the first * one, for compatibility with previous code. While we're * scanning, check for obvious loops in the network topology * (these should never happen, and as noted above, we may * miss real loops; this is just a double-check). */ IF_ADDR_RLOCK(ifp); error = 0; match = NULL; IFNET_FOREACH_IFA(ifp, ifa) { if (match == NULL && ifa->ifa_carp != NULL && ifa->ifa_addr->sa_family == af && ifa->ifa_carp->sc_vhid == ch->carp_vhid) match = ifa; if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af)) error = ELOOP; } ifa = error ? NULL : match; if (ifa != NULL) ifa_ref(ifa); IF_ADDR_RUNLOCK(ifp); if (ifa == NULL) { if (error == ELOOP) { CARP_DEBUG("dropping looped packet on interface %s\n", ifp->if_xname); CARPSTATS_INC(carps_badif); /* ??? */ } else { CARPSTATS_INC(carps_badvhid); } m_freem(m); return; } /* verify the CARP version. */ if (ch->carp_version != CARP_VERSION) { CARPSTATS_INC(carps_badver); CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname, ch->carp_version); ifa_free(ifa); m_freem(m); return; } sc = ifa->ifa_carp; CARP_LOCK(sc); ifa_free(ifa); if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { CARPSTATS_INC(carps_badauth); CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__, sc->sc_vhid, ifp->if_xname); goto out; } tmp_counter = ntohl(ch->carp_counter[0]); tmp_counter = tmp_counter<<32; tmp_counter += ntohl(ch->carp_counter[1]); /* XXX Replay protection goes here */ sc->sc_init_counter = 0; sc->sc_counter = tmp_counter; sc_tv.tv_sec = sc->sc_advbase; sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256; ch_tv.tv_sec = ch->carp_advbase; ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; switch (sc->sc_state) { case INIT: break; case MASTER: /* * If we receive an advertisement from a master who's going to * be more frequent than us, go into BACKUP state. */ if (timevalcmp(&sc_tv, &ch_tv, >) || timevalcmp(&sc_tv, &ch_tv, ==)) { callout_stop(&sc->sc_ad_tmo); carp_set_state(sc, BACKUP, "more frequent advertisement received"); carp_setrun(sc, 0); carp_delroute(sc); } break; case BACKUP: /* * If we're pre-empting masters who advertise slower than us, * and this one claims to be slower, treat him as down. */ if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) { carp_master_down_locked(sc, "preempting a slower master"); break; } /* * If the master is going to advertise at such a low frequency * that he's guaranteed to time out, we'd might as well just * treat him as timed out now. */ sc_tv.tv_sec = sc->sc_advbase * 3; if (timevalcmp(&sc_tv, &ch_tv, <)) { carp_master_down_locked(sc, "master will time out"); break; } /* * Otherwise, we reset the counter and wait for the next * advertisement. */ carp_setrun(sc, af); break; } out: CARP_UNLOCK(sc); m_freem(m); } static int carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch) { struct m_tag *mtag; if (sc->sc_init_counter) { /* this could also be seconds since unix epoch */ sc->sc_counter = arc4random(); sc->sc_counter = sc->sc_counter << 32; sc->sc_counter += arc4random(); } else sc->sc_counter++; ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); /* Tag packet for carp_output */ if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *), M_NOWAIT)) == NULL) { m_freem(m); CARPSTATS_INC(carps_onomem); return (ENOMEM); } bcopy(&sc, mtag + 1, sizeof(sc)); m_tag_prepend(m, mtag); return (0); } /* * To avoid LORs and possible recursions this function shouldn't * be called directly, but scheduled via taskqueue. */ static void carp_send_ad_all(void *ctx __unused, int pending __unused) { struct carp_softc *sc; mtx_lock(&carp_mtx); LIST_FOREACH(sc, &carp_list, sc_next) if (sc->sc_state == MASTER) { CARP_LOCK(sc); CURVNET_SET(sc->sc_carpdev->if_vnet); carp_send_ad_locked(sc); CURVNET_RESTORE(); CARP_UNLOCK(sc); } mtx_unlock(&carp_mtx); } /* Send a periodic advertisement, executed in callout context. */ static void carp_send_ad(void *v) { struct carp_softc *sc = v; CARP_LOCK_ASSERT(sc); CURVNET_SET(sc->sc_carpdev->if_vnet); carp_send_ad_locked(sc); CURVNET_RESTORE(); CARP_UNLOCK(sc); } static void carp_send_ad_error(struct carp_softc *sc, int error) { if (error) { if (sc->sc_sendad_errors < INT_MAX) sc->sc_sendad_errors++; if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { static const char fmt[] = "send error %d on %s"; char msg[sizeof(fmt) + IFNAMSIZ]; sprintf(msg, fmt, error, sc->sc_carpdev->if_xname); carp_demote_adj(V_carp_senderr_adj, msg); } sc->sc_sendad_success = 0; } else { if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS && ++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) { static const char fmt[] = "send ok on %s"; char msg[sizeof(fmt) + IFNAMSIZ]; sprintf(msg, fmt, sc->sc_carpdev->if_xname); carp_demote_adj(-V_carp_senderr_adj, msg); sc->sc_sendad_errors = 0; } else sc->sc_sendad_errors = 0; } } /* * Pick the best ifaddr on the given ifp for sending CARP * advertisements. * * "Best" here is defined by ifa_preferred(). This function is much * much like ifaof_ifpforaddr() except that we just use ifa_preferred(). * * (This could be simplified to return the actual address, except that * it has a different format in AF_INET and AF_INET6.) */ static struct ifaddr * carp_best_ifa(int af, struct ifnet *ifp) { struct ifaddr *ifa, *best; if (af >= AF_MAX) return (NULL); best = NULL; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family == af && (best == NULL || ifa_preferred(best, ifa))) best = ifa; } IF_ADDR_RUNLOCK(ifp); if (best != NULL) ifa_ref(best); return (best); } static void carp_send_ad_locked(struct carp_softc *sc) { struct carp_header ch; struct timeval tv; struct ifaddr *ifa; struct carp_header *ch_ptr; struct mbuf *m; int len, advskew; CARP_LOCK_ASSERT(sc); advskew = DEMOTE_ADVSKEW(sc); tv.tv_sec = sc->sc_advbase; tv.tv_usec = advskew * 1000000 / 256; ch.carp_version = CARP_VERSION; ch.carp_type = CARP_ADVERTISEMENT; ch.carp_vhid = sc->sc_vhid; ch.carp_advbase = sc->sc_advbase; ch.carp_advskew = advskew; ch.carp_authlen = 7; /* XXX DEFINE */ ch.carp_pad1 = 0; /* must be zero */ ch.carp_cksum = 0; /* XXXGL: OpenBSD picks first ifaddr with needed family. */ #ifdef INET if (sc->sc_naddrs) { struct ip *ip; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { CARPSTATS_INC(carps_onomem); goto resched; } len = sizeof(*ip) + sizeof(ch); m->m_pkthdr.len = len; m->m_pkthdr.rcvif = NULL; m->m_len = len; M_ALIGN(m, m->m_len); m->m_flags |= M_MCAST; ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = sizeof(*ip) >> 2; ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET; ip->ip_len = htons(len); ip->ip_off = htons(IP_DF); ip->ip_ttl = CARP_DFLTTL; ip->ip_p = IPPROTO_CARP; ip->ip_sum = 0; ip_fillid(ip); ifa = carp_best_ifa(AF_INET, sc->sc_carpdev); if (ifa != NULL) { ip->ip_src.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; ifa_free(ifa); } else ip->ip_src.s_addr = 0; ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP); ch_ptr = (struct carp_header *)(&ip[1]); bcopy(&ch, ch_ptr, sizeof(ch)); if (carp_prepare_ad(m, sc, ch_ptr)) goto resched; m->m_data += sizeof(*ip); ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip)); m->m_data -= sizeof(*ip); CARPSTATS_INC(carps_opackets); carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_carpdev->if_carp->cif_imo, NULL)); } #endif /* INET */ #ifdef INET6 if (sc->sc_naddrs6) { struct ip6_hdr *ip6; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { CARPSTATS_INC(carps_onomem); goto resched; } len = sizeof(*ip6) + sizeof(ch); m->m_pkthdr.len = len; m->m_pkthdr.rcvif = NULL; m->m_len = len; M_ALIGN(m, m->m_len); m->m_flags |= M_MCAST; ip6 = mtod(m, struct ip6_hdr *); bzero(ip6, sizeof(*ip6)); ip6->ip6_vfc |= IPV6_VERSION; /* Traffic class isn't defined in ip6 struct instead * it gets offset into flowid field */ ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN + IPTOS_DSCP_OFFSET)); ip6->ip6_hlim = CARP_DFLTTL; ip6->ip6_nxt = IPPROTO_CARP; /* set the source address */ ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev); if (ifa != NULL) { bcopy(IFA_IN6(ifa), &ip6->ip6_src, sizeof(struct in6_addr)); ifa_free(ifa); } else /* This should never happen with IPv6. */ bzero(&ip6->ip6_src, sizeof(struct in6_addr)); /* Set the multicast destination. */ ip6->ip6_dst.s6_addr16[0] = htons(0xff02); ip6->ip6_dst.s6_addr8[15] = 0x12; if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { m_freem(m); CARP_DEBUG("%s: in6_setscope failed\n", __func__); goto resched; } ch_ptr = (struct carp_header *)(&ip6[1]); bcopy(&ch, ch_ptr, sizeof(ch)); if (carp_prepare_ad(m, sc, ch_ptr)) goto resched; m->m_data += sizeof(*ip6); ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6)); m->m_data -= sizeof(*ip6); CARPSTATS_INC(carps_opackets6); carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0, &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL)); } #endif /* INET6 */ resched: callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc); } static void carp_addroute(struct carp_softc *sc) { struct ifaddr *ifa; CARP_FOREACH_IFA(sc, ifa) carp_ifa_addroute(ifa); } static void carp_ifa_addroute(struct ifaddr *ifa) { switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: in_addprefix(ifatoia(ifa), RTF_UP); ifa_add_loopback_route(ifa, (struct sockaddr *)&ifatoia(ifa)->ia_addr); break; #endif #ifdef INET6 case AF_INET6: ifa_add_loopback_route(ifa, (struct sockaddr *)&ifatoia6(ifa)->ia_addr); nd6_add_ifa_lle(ifatoia6(ifa)); break; #endif } } static void carp_delroute(struct carp_softc *sc) { struct ifaddr *ifa; CARP_FOREACH_IFA(sc, ifa) carp_ifa_delroute(ifa); } static void carp_ifa_delroute(struct ifaddr *ifa) { switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: ifa_del_loopback_route(ifa, (struct sockaddr *)&ifatoia(ifa)->ia_addr); in_scrubprefix(ifatoia(ifa), LLE_STATIC); break; #endif #ifdef INET6 case AF_INET6: ifa_del_loopback_route(ifa, (struct sockaddr *)&ifatoia6(ifa)->ia_addr); nd6_rem_ifa_lle(ifatoia6(ifa), 1); break; #endif } } int carp_master(struct ifaddr *ifa) { struct carp_softc *sc = ifa->ifa_carp; return (sc->sc_state == MASTER); } #ifdef INET /* * Broadcast a gratuitous ARP request containing * the virtual router MAC address for each IP address * associated with the virtual router. */ static void carp_send_arp(struct carp_softc *sc) { struct ifaddr *ifa; struct in_addr addr; CARP_FOREACH_IFA(sc, ifa) { if (ifa->ifa_addr->sa_family != AF_INET) continue; addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr; arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr)); } } int carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr) { struct carp_softc *sc = ifa->ifa_carp; if (sc->sc_state == MASTER) { *enaddr = LLADDR(&sc->sc_addr); return (1); } return (0); } #endif #ifdef INET6 static void carp_send_na(struct carp_softc *sc) { static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; struct ifaddr *ifa; struct in6_addr *in6; CARP_FOREACH_IFA(sc, ifa) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; in6 = IFA_IN6(ifa); nd6_na_output(sc->sc_carpdev, &mcast, in6, ND_NA_FLAG_OVERRIDE, 1, NULL); DELAY(1000); /* XXX */ } } /* * Returns ifa in case it's a carp address and it is MASTER, or if the address * matches and is not a carp address. Returns NULL otherwise. */ struct ifaddr * carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr) { struct ifaddr *ifa; ifa = NULL; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) continue; if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER) ifa = NULL; else ifa_ref(ifa); break; } IF_ADDR_RUNLOCK(ifp); return (ifa); } caddr_t carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr) { struct ifaddr *ifa; IF_ADDR_RLOCK(ifp); IFNET_FOREACH_IFA(ifp, ifa) if (ifa->ifa_addr->sa_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) { struct carp_softc *sc = ifa->ifa_carp; struct m_tag *mtag; IF_ADDR_RUNLOCK(ifp); mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *), M_NOWAIT); if (mtag == NULL) /* Better a bit than nothing. */ return (LLADDR(&sc->sc_addr)); bcopy(&sc, mtag + 1, sizeof(sc)); m_tag_prepend(m, mtag); return (LLADDR(&sc->sc_addr)); } IF_ADDR_RUNLOCK(ifp); return (NULL); } #endif /* INET6 */ int carp_forus(struct ifnet *ifp, u_char *dhost) { struct carp_softc *sc; uint8_t *ena = dhost; if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) return (0); CIF_LOCK(ifp->if_carp); IFNET_FOREACH_CARP(ifp, sc) { CARP_LOCK(sc); if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr), ETHER_ADDR_LEN)) { CARP_UNLOCK(sc); CIF_UNLOCK(ifp->if_carp); return (1); } CARP_UNLOCK(sc); } CIF_UNLOCK(ifp->if_carp); return (0); } /* Master down timeout event, executed in callout context. */ static void carp_master_down(void *v) { struct carp_softc *sc = v; CARP_LOCK_ASSERT(sc); CURVNET_SET(sc->sc_carpdev->if_vnet); if (sc->sc_state == BACKUP) { carp_master_down_locked(sc, "master timed out"); } CURVNET_RESTORE(); CARP_UNLOCK(sc); } static void carp_master_down_locked(struct carp_softc *sc, const char *reason) { CARP_LOCK_ASSERT(sc); switch (sc->sc_state) { case BACKUP: carp_set_state(sc, MASTER, reason); carp_send_ad_locked(sc); #ifdef INET carp_send_arp(sc); #endif #ifdef INET6 carp_send_na(sc); #endif carp_setrun(sc, 0); carp_addroute(sc); break; case INIT: case MASTER: #ifdef INVARIANTS panic("carp: VHID %u@%s: master_down event in %s state\n", sc->sc_vhid, sc->sc_carpdev->if_xname, sc->sc_state ? "MASTER" : "INIT"); #endif break; } } /* * When in backup state, af indicates whether to reset the master down timer * for v4 or v6. If it's set to zero, reset the ones which are already pending. */ static void carp_setrun(struct carp_softc *sc, sa_family_t af) { struct timeval tv; CARP_LOCK_ASSERT(sc); if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 || sc->sc_carpdev->if_link_state != LINK_STATE_UP || (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) || !V_carp_allow) return; switch (sc->sc_state) { case INIT: carp_set_state(sc, BACKUP, "initialization complete"); carp_setrun(sc, 0); break; case BACKUP: callout_stop(&sc->sc_ad_tmo); tv.tv_sec = 3 * sc->sc_advbase; tv.tv_usec = sc->sc_advskew * 1000000 / 256; switch (af) { #ifdef INET case AF_INET: callout_reset(&sc->sc_md_tmo, tvtohz(&tv), carp_master_down, sc); break; #endif #ifdef INET6 case AF_INET6: callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), carp_master_down, sc); break; #endif default: #ifdef INET if (sc->sc_naddrs) callout_reset(&sc->sc_md_tmo, tvtohz(&tv), carp_master_down, sc); #endif #ifdef INET6 if (sc->sc_naddrs6) callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), carp_master_down, sc); #endif break; } break; case MASTER: tv.tv_sec = sc->sc_advbase; tv.tv_usec = sc->sc_advskew * 1000000 / 256; callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc); break; } } /* * Setup multicast structures. */ static int carp_multicast_setup(struct carp_if *cif, sa_family_t sa) { struct ifnet *ifp = cif->cif_ifp; int error = 0; switch (sa) { #ifdef INET case AF_INET: { struct ip_moptions *imo = &cif->cif_imo; struct in_addr addr; if (imo->imo_membership) return (0); imo->imo_membership = (struct in_multi **)malloc( (sizeof(struct in_multi *) * IP_MIN_MEMBERSHIPS), M_CARP, M_WAITOK); imo->imo_mfilters = NULL; imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; imo->imo_multicast_vif = -1; addr.s_addr = htonl(INADDR_CARP_GROUP); if ((error = in_joingroup(ifp, &addr, NULL, &imo->imo_membership[0])) != 0) { free(imo->imo_membership, M_CARP); break; } imo->imo_num_memberships++; imo->imo_multicast_ifp = ifp; imo->imo_multicast_ttl = CARP_DFLTTL; imo->imo_multicast_loop = 0; break; } #endif #ifdef INET6 case AF_INET6: { struct ip6_moptions *im6o = &cif->cif_im6o; struct in6_addr in6; struct in6_multi *in6m; if (im6o->im6o_membership) return (0); im6o->im6o_membership = (struct in6_multi **)malloc( (sizeof(struct in6_multi *) * IPV6_MIN_MEMBERSHIPS), M_CARP, M_ZERO | M_WAITOK); im6o->im6o_mfilters = NULL; im6o->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS; im6o->im6o_multicast_hlim = CARP_DFLTTL; im6o->im6o_multicast_ifp = ifp; /* Join IPv6 CARP multicast group. */ bzero(&in6, sizeof(in6)); in6.s6_addr16[0] = htons(0xff02); in6.s6_addr8[15] = 0x12; if ((error = in6_setscope(&in6, ifp, NULL)) != 0) { free(im6o->im6o_membership, M_CARP); break; } in6m = NULL; if ((error = in6_joingroup(ifp, &in6, NULL, &in6m, 0)) != 0) { free(im6o->im6o_membership, M_CARP); break; } im6o->im6o_membership[0] = in6m; im6o->im6o_num_memberships++; /* Join solicited multicast address. */ bzero(&in6, sizeof(in6)); in6.s6_addr16[0] = htons(0xff02); in6.s6_addr32[1] = 0; in6.s6_addr32[2] = htonl(1); in6.s6_addr32[3] = 0; in6.s6_addr8[12] = 0xff; if ((error = in6_setscope(&in6, ifp, NULL)) != 0) { in6_leavegroup(im6o->im6o_membership[0], NULL); free(im6o->im6o_membership, M_CARP); break; } in6m = NULL; if ((error = in6_joingroup(ifp, &in6, NULL, &in6m, 0)) != 0) { in6_leavegroup(im6o->im6o_membership[0], NULL); free(im6o->im6o_membership, M_CARP); break; } im6o->im6o_membership[1] = in6m; im6o->im6o_num_memberships++; break; } #endif } return (error); } /* * Free multicast structures. */ static void carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa) { sx_assert(&carp_sx, SA_XLOCKED); switch (sa) { #ifdef INET case AF_INET: if (cif->cif_naddrs == 0) { struct ip_moptions *imo = &cif->cif_imo; in_leavegroup(imo->imo_membership[0], NULL); KASSERT(imo->imo_mfilters == NULL, ("%s: imo_mfilters != NULL", __func__)); free(imo->imo_membership, M_CARP); imo->imo_membership = NULL; } break; #endif #ifdef INET6 case AF_INET6: if (cif->cif_naddrs6 == 0) { struct ip6_moptions *im6o = &cif->cif_im6o; in6_leavegroup(im6o->im6o_membership[0], NULL); in6_leavegroup(im6o->im6o_membership[1], NULL); KASSERT(im6o->im6o_mfilters == NULL, ("%s: im6o_mfilters != NULL", __func__)); free(im6o->im6o_membership, M_CARP); im6o->im6o_membership = NULL; } break; #endif } } int carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa) { struct m_tag *mtag; struct carp_softc *sc; if (!sa) return (0); switch (sa->sa_family) { #ifdef INET case AF_INET: break; #endif #ifdef INET6 case AF_INET6: break; #endif default: return (0); } mtag = m_tag_find(m, PACKET_TAG_CARP, NULL); if (mtag == NULL) return (0); bcopy(mtag + 1, &sc, sizeof(sc)); /* Set the source MAC address to the Virtual Router MAC Address. */ switch (ifp->if_type) { case IFT_ETHER: case IFT_BRIDGE: case IFT_L2VLAN: { struct ether_header *eh; eh = mtod(m, struct ether_header *); eh->ether_shost[0] = 0; eh->ether_shost[1] = 0; eh->ether_shost[2] = 0x5e; eh->ether_shost[3] = 0; eh->ether_shost[4] = 1; eh->ether_shost[5] = sc->sc_vhid; } break; default: printf("%s: carp is not supported for the %d interface type\n", ifp->if_xname, ifp->if_type); return (EOPNOTSUPP); } return (0); } static struct carp_softc* carp_alloc(struct ifnet *ifp) { struct carp_softc *sc; struct carp_if *cif; sx_assert(&carp_sx, SA_XLOCKED); if ((cif = ifp->if_carp) == NULL) cif = carp_alloc_if(ifp); sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO); sc->sc_advbase = CARP_DFLTINTV; sc->sc_vhid = -1; /* required setting */ sc->sc_init_counter = 1; sc->sc_state = INIT; sc->sc_ifasiz = sizeof(struct ifaddr *); sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO); sc->sc_carpdev = ifp; CARP_LOCK_INIT(sc); #ifdef INET callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); #endif #ifdef INET6 callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); #endif callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); CIF_LOCK(cif); TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list); CIF_UNLOCK(cif); mtx_lock(&carp_mtx); LIST_INSERT_HEAD(&carp_list, sc, sc_next); mtx_unlock(&carp_mtx); return (sc); } static void carp_grow_ifas(struct carp_softc *sc) { struct ifaddr **new; new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO); CARP_LOCK(sc); bcopy(sc->sc_ifas, new, sc->sc_ifasiz); free(sc->sc_ifas, M_CARP); sc->sc_ifas = new; sc->sc_ifasiz *= 2; CARP_UNLOCK(sc); } static void carp_destroy(struct carp_softc *sc) { struct ifnet *ifp = sc->sc_carpdev; struct carp_if *cif = ifp->if_carp; sx_assert(&carp_sx, SA_XLOCKED); if (sc->sc_suppress) carp_demote_adj(-V_carp_ifdown_adj, "vhid removed"); CARP_UNLOCK(sc); CIF_LOCK(cif); TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list); CIF_UNLOCK(cif); mtx_lock(&carp_mtx); LIST_REMOVE(sc, sc_next); mtx_unlock(&carp_mtx); callout_drain(&sc->sc_ad_tmo); #ifdef INET callout_drain(&sc->sc_md_tmo); #endif #ifdef INET6 callout_drain(&sc->sc_md6_tmo); #endif CARP_LOCK_DESTROY(sc); free(sc->sc_ifas, M_CARP); free(sc, M_CARP); } static struct carp_if* carp_alloc_if(struct ifnet *ifp) { struct carp_if *cif; int error; cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO); if ((error = ifpromisc(ifp, 1)) != 0) printf("%s: ifpromisc(%s) failed: %d\n", __func__, ifp->if_xname, error); else cif->cif_flags |= CIF_PROMISC; CIF_LOCK_INIT(cif); cif->cif_ifp = ifp; TAILQ_INIT(&cif->cif_vrs); IF_ADDR_WLOCK(ifp); ifp->if_carp = cif; if_ref(ifp); IF_ADDR_WUNLOCK(ifp); return (cif); } static void carp_free_if(struct carp_if *cif) { struct ifnet *ifp = cif->cif_ifp; CIF_LOCK_ASSERT(cif); KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty", __func__)); IF_ADDR_WLOCK(ifp); ifp->if_carp = NULL; IF_ADDR_WUNLOCK(ifp); CIF_LOCK_DESTROY(cif); if (cif->cif_flags & CIF_PROMISC) ifpromisc(ifp, 0); if_rele(ifp); free(cif, M_CARP); } static void carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv) { CARP_LOCK(sc); carpr->carpr_state = sc->sc_state; carpr->carpr_vhid = sc->sc_vhid; carpr->carpr_advbase = sc->sc_advbase; carpr->carpr_advskew = sc->sc_advskew; if (priv) bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key)); else bzero(carpr->carpr_key, sizeof(carpr->carpr_key)); CARP_UNLOCK(sc); } int carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td) { struct carpreq carpr; struct ifnet *ifp; struct carp_softc *sc = NULL; int error = 0, locked = 0; if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr))) return (error); ifp = ifunit_ref(ifr->ifr_name); if (ifp == NULL) return (ENXIO); switch (ifp->if_type) { case IFT_ETHER: case IFT_L2VLAN: case IFT_BRIDGE: break; default: error = EOPNOTSUPP; goto out; } if ((ifp->if_flags & IFF_MULTICAST) == 0) { error = EADDRNOTAVAIL; goto out; } sx_xlock(&carp_sx); switch (cmd) { case SIOCSVH: if ((error = priv_check(td, PRIV_NETINET_CARP))) break; if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID || carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) { error = EINVAL; break; } if (ifp->if_carp) { IFNET_FOREACH_CARP(ifp, sc) if (sc->sc_vhid == carpr.carpr_vhid) break; } if (sc == NULL) { sc = carp_alloc(ifp); CARP_LOCK(sc); sc->sc_vhid = carpr.carpr_vhid; LLADDR(&sc->sc_addr)[0] = 0; LLADDR(&sc->sc_addr)[1] = 0; LLADDR(&sc->sc_addr)[2] = 0x5e; LLADDR(&sc->sc_addr)[3] = 0; LLADDR(&sc->sc_addr)[4] = 1; LLADDR(&sc->sc_addr)[5] = sc->sc_vhid; } else CARP_LOCK(sc); locked = 1; if (carpr.carpr_advbase > 0) { if (carpr.carpr_advbase > 255 || carpr.carpr_advbase < CARP_DFLTINTV) { error = EINVAL; break; } sc->sc_advbase = carpr.carpr_advbase; } if (carpr.carpr_advskew >= 255) { error = EINVAL; break; } sc->sc_advskew = carpr.carpr_advskew; if (carpr.carpr_key[0] != '\0') { bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); carp_hmac_prepare(sc); } if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { switch (carpr.carpr_state) { case BACKUP: callout_stop(&sc->sc_ad_tmo); carp_set_state(sc, BACKUP, "user requested via ifconfig"); carp_setrun(sc, 0); carp_delroute(sc); break; case MASTER: carp_master_down_locked(sc, "user requested via ifconfig"); break; default: break; } } break; case SIOCGVH: { int priveleged; if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) { error = EINVAL; break; } if (carpr.carpr_count < 1) { error = EMSGSIZE; break; } if (ifp->if_carp == NULL) { error = ENOENT; break; } priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0); if (carpr.carpr_vhid != 0) { IFNET_FOREACH_CARP(ifp, sc) if (sc->sc_vhid == carpr.carpr_vhid) break; if (sc == NULL) { error = ENOENT; break; } carp_carprcp(&carpr, sc, priveleged); error = copyout(&carpr, ifr_data_get_ptr(ifr), sizeof(carpr)); } else { int i, count; count = 0; IFNET_FOREACH_CARP(ifp, sc) count++; if (count > carpr.carpr_count) { CIF_UNLOCK(ifp->if_carp); error = EMSGSIZE; break; } i = 0; IFNET_FOREACH_CARP(ifp, sc) { carp_carprcp(&carpr, sc, priveleged); carpr.carpr_count = count; error = copyout(&carpr, (caddr_t)ifr_data_get_ptr(ifr) + (i * sizeof(carpr)), sizeof(carpr)); if (error) { CIF_UNLOCK(ifp->if_carp); break; } i++; } } break; } default: error = EINVAL; } sx_xunlock(&carp_sx); out: if (locked) CARP_UNLOCK(sc); if_rele(ifp); return (error); } static int carp_get_vhid(struct ifaddr *ifa) { if (ifa == NULL || ifa->ifa_carp == NULL) return (0); return (ifa->ifa_carp->sc_vhid); } int carp_attach(struct ifaddr *ifa, int vhid) { struct ifnet *ifp = ifa->ifa_ifp; struct carp_if *cif = ifp->if_carp; struct carp_softc *sc; int index, error; KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa)); switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: #endif #ifdef INET6 case AF_INET6: #endif break; default: return (EPROTOTYPE); } sx_xlock(&carp_sx); if (ifp->if_carp == NULL) { sx_xunlock(&carp_sx); return (ENOPROTOOPT); } IFNET_FOREACH_CARP(ifp, sc) if (sc->sc_vhid == vhid) break; if (sc == NULL) { sx_xunlock(&carp_sx); return (ENOENT); } error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family); if (error) { CIF_FREE(cif); sx_xunlock(&carp_sx); return (error); } index = sc->sc_naddrs + sc->sc_naddrs6 + 1; if (index > sc->sc_ifasiz / sizeof(struct ifaddr *)) carp_grow_ifas(sc); switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: cif->cif_naddrs++; sc->sc_naddrs++; break; #endif #ifdef INET6 case AF_INET6: cif->cif_naddrs6++; sc->sc_naddrs6++; break; #endif } ifa_ref(ifa); CARP_LOCK(sc); sc->sc_ifas[index - 1] = ifa; ifa->ifa_carp = sc; carp_hmac_prepare(sc); carp_sc_state(sc); CARP_UNLOCK(sc); sx_xunlock(&carp_sx); return (0); } void carp_detach(struct ifaddr *ifa, bool keep_cif) { struct ifnet *ifp = ifa->ifa_ifp; struct carp_if *cif = ifp->if_carp; struct carp_softc *sc = ifa->ifa_carp; int i, index; KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa)); sx_xlock(&carp_sx); CARP_LOCK(sc); /* Shift array. */ index = sc->sc_naddrs + sc->sc_naddrs6; for (i = 0; i < index; i++) if (sc->sc_ifas[i] == ifa) break; KASSERT(i < index, ("%s: %p no backref", __func__, ifa)); for (; i < index - 1; i++) sc->sc_ifas[i] = sc->sc_ifas[i+1]; sc->sc_ifas[index - 1] = NULL; switch (ifa->ifa_addr->sa_family) { #ifdef INET case AF_INET: cif->cif_naddrs--; sc->sc_naddrs--; break; #endif #ifdef INET6 case AF_INET6: cif->cif_naddrs6--; sc->sc_naddrs6--; break; #endif } carp_ifa_delroute(ifa); carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family); ifa->ifa_carp = NULL; ifa_free(ifa); carp_hmac_prepare(sc); carp_sc_state(sc); if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) carp_destroy(sc); else CARP_UNLOCK(sc); if (!keep_cif) CIF_FREE(cif); sx_xunlock(&carp_sx); } static void carp_set_state(struct carp_softc *sc, int state, const char *reason) { CARP_LOCK_ASSERT(sc); if (sc->sc_state != state) { const char *carp_states[] = { CARP_STATES }; char subsys[IFNAMSIZ+5]; snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid, sc->sc_carpdev->if_xname); CARP_LOG("%s: %s -> %s (%s)\n", subsys, carp_states[sc->sc_state], carp_states[state], reason); sc->sc_state = state; devctl_notify("CARP", subsys, carp_states[state], NULL); } } static void carp_linkstate(struct ifnet *ifp) { struct carp_softc *sc; CIF_LOCK(ifp->if_carp); IFNET_FOREACH_CARP(ifp, sc) { CARP_LOCK(sc); carp_sc_state(sc); CARP_UNLOCK(sc); } CIF_UNLOCK(ifp->if_carp); } static void carp_sc_state(struct carp_softc *sc) { CARP_LOCK_ASSERT(sc); if (sc->sc_carpdev->if_link_state != LINK_STATE_UP || !(sc->sc_carpdev->if_flags & IFF_UP) || !V_carp_allow) { callout_stop(&sc->sc_ad_tmo); #ifdef INET callout_stop(&sc->sc_md_tmo); #endif #ifdef INET6 callout_stop(&sc->sc_md6_tmo); #endif carp_set_state(sc, INIT, "hardware interface down"); carp_setrun(sc, 0); if (!sc->sc_suppress) carp_demote_adj(V_carp_ifdown_adj, "interface down"); sc->sc_suppress = 1; } else { carp_set_state(sc, INIT, "hardware interface up"); carp_setrun(sc, 0); if (sc->sc_suppress) carp_demote_adj(-V_carp_ifdown_adj, "interface up"); sc->sc_suppress = 0; } } static void carp_demote_adj(int adj, char *reason) { atomic_add_int(&V_carp_demotion, adj); CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason); taskqueue_enqueue(taskqueue_swi, &carp_sendall_task); } static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS) { int new, error; struct carp_softc *sc; new = V_carp_allow; error = sysctl_handle_int(oidp, &new, 0, req); if (error || !req->newptr) return (error); if (V_carp_allow != new) { V_carp_allow = new; mtx_lock(&carp_mtx); LIST_FOREACH(sc, &carp_list, sc_next) { CARP_LOCK(sc); if (curvnet == sc->sc_carpdev->if_vnet) carp_sc_state(sc); CARP_UNLOCK(sc); } mtx_unlock(&carp_mtx); } return (0); } static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS) { int new, error; new = V_carp_dscp; error = sysctl_handle_int(oidp, &new, 0, req); if (error || !req->newptr) return (error); if (new < 0 || new > 63) return (EINVAL); V_carp_dscp = new; return (0); } static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS) { int new, error; new = V_carp_demotion; error = sysctl_handle_int(oidp, &new, 0, req); if (error || !req->newptr) return (error); carp_demote_adj(new, "sysctl"); return (0); } #ifdef INET extern struct domain inetdomain; static struct protosw in_carp_protosw = { .pr_type = SOCK_RAW, .pr_domain = &inetdomain, .pr_protocol = IPPROTO_CARP, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_input = carp_input, .pr_output = rip_output, .pr_ctloutput = rip_ctloutput, .pr_usrreqs = &rip_usrreqs }; #endif #ifdef INET6 extern struct domain inet6domain; static struct protosw in6_carp_protosw = { .pr_type = SOCK_RAW, .pr_domain = &inet6domain, .pr_protocol = IPPROTO_CARP, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_input = carp6_input, .pr_output = rip6_output, .pr_ctloutput = rip6_ctloutput, .pr_usrreqs = &rip6_usrreqs }; #endif static void carp_mod_cleanup(void) { #ifdef INET if (proto_reg[CARP_INET] == 0) { (void)ipproto_unregister(IPPROTO_CARP); pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW); proto_reg[CARP_INET] = -1; } carp_iamatch_p = NULL; #endif #ifdef INET6 if (proto_reg[CARP_INET6] == 0) { (void)ip6proto_unregister(IPPROTO_CARP); pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW); proto_reg[CARP_INET6] = -1; } carp_iamatch6_p = NULL; carp_macmatch6_p = NULL; #endif carp_ioctl_p = NULL; carp_attach_p = NULL; carp_detach_p = NULL; carp_get_vhid_p = NULL; carp_linkstate_p = NULL; carp_forus_p = NULL; carp_output_p = NULL; carp_demote_adj_p = NULL; carp_master_p = NULL; mtx_unlock(&carp_mtx); taskqueue_drain(taskqueue_swi, &carp_sendall_task); mtx_destroy(&carp_mtx); sx_destroy(&carp_sx); } static int carp_mod_load(void) { int err; mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF); sx_init(&carp_sx, "carp_sx"); LIST_INIT(&carp_list); carp_get_vhid_p = carp_get_vhid; carp_forus_p = carp_forus; carp_output_p = carp_output; carp_linkstate_p = carp_linkstate; carp_ioctl_p = carp_ioctl; carp_attach_p = carp_attach; carp_detach_p = carp_detach; carp_demote_adj_p = carp_demote_adj; carp_master_p = carp_master; #ifdef INET6 carp_iamatch6_p = carp_iamatch6; carp_macmatch6_p = carp_macmatch6; proto_reg[CARP_INET6] = pf_proto_register(PF_INET6, (struct protosw *)&in6_carp_protosw); if (proto_reg[CARP_INET6]) { printf("carp: error %d attaching to PF_INET6\n", proto_reg[CARP_INET6]); carp_mod_cleanup(); return (proto_reg[CARP_INET6]); } err = ip6proto_register(IPPROTO_CARP); if (err) { printf("carp: error %d registering with INET6\n", err); carp_mod_cleanup(); return (err); } #endif #ifdef INET carp_iamatch_p = carp_iamatch; proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw); if (proto_reg[CARP_INET]) { printf("carp: error %d attaching to PF_INET\n", proto_reg[CARP_INET]); carp_mod_cleanup(); return (proto_reg[CARP_INET]); } err = ipproto_register(IPPROTO_CARP); if (err) { printf("carp: error %d registering with INET\n", err); carp_mod_cleanup(); return (err); } #endif return (0); } static int carp_modevent(module_t mod, int type, void *data) { switch (type) { case MOD_LOAD: return carp_mod_load(); /* NOTREACHED */ case MOD_UNLOAD: mtx_lock(&carp_mtx); if (LIST_EMPTY(&carp_list)) carp_mod_cleanup(); else { mtx_unlock(&carp_mtx); return (EBUSY); } break; default: return (EINVAL); } return (0); } static moduledata_t carp_mod = { "carp", carp_modevent, 0 }; DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); Index: head/sys/netinet/ip_divert.c =================================================================== --- head/sys/netinet/ip_divert.c (revision 336675) +++ head/sys/netinet/ip_divert.c (revision 336676) @@ -1,833 +1,833 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_sctp.h" #ifndef INET #error "IPDIVERT requires INET" #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #endif #ifdef SCTP #include #endif #include /* * Divert sockets */ /* * Allocate enough space to hold a full IP packet */ #define DIVSNDQ (65536 + 100) #define DIVRCVQ (65536 + 100) /* * Divert sockets work in conjunction with ipfw or other packet filters, * see the divert(4) manpage for features. * Packets are selected by the packet filter and tagged with an * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by * the packet filter) and information on the matching filter rule for * subsequent reinjection. The divert_port is used to put the packet * on the corresponding divert socket, while the rule number is passed * up (at least partially) as the sin_port in the struct sockaddr. * * Packets written to the divert socket carry in sin_addr a * destination address, and in sin_port the number of the filter rule * after which to continue processing. * If the destination address is INADDR_ANY, the packet is treated as * as outgoing and sent to ip_output(); otherwise it is treated as * incoming and sent to ip_input(). * Further, sin_zero carries some information on the interface, * which can be used in the reinject -- see comments in the code. * * On reinjection, processing in ip_input() and ip_output() * will be exactly the same as for the original packet, except that * packet filter processing will start at the rule number after the one * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0 * will apply the entire ruleset to the packet). */ /* Internal variables. */ -static VNET_DEFINE(struct inpcbhead, divcb); -static VNET_DEFINE(struct inpcbinfo, divcbinfo); +VNET_DEFINE_STATIC(struct inpcbhead, divcb); +VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo); #define V_divcb VNET(divcb) #define V_divcbinfo VNET(divcbinfo) static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ static eventhandler_tag ip_divert_event_tag; /* * Initialize divert connection block queue. */ static void div_zone_change(void *tag) { uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); } static int div_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp = mem; INP_LOCK_INIT(inp, "inp", "divinp"); return (0); } static void div_init(void) { /* * XXX We don't use the hash list for divert IP, but it's easier to * allocate one-entry hash lists than it is to check all over the * place for hashbase == NULL. */ in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb", div_inpcb_init, IPI_HASHFIELDS_NONE); } static void div_destroy(void *unused __unused) { in_pcbinfo_destroy(&V_divcbinfo); } VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, div_destroy, NULL); /* * IPPROTO_DIVERT is not in the real IP protocol number space; this * function should never be called. Just in case, drop any packets. */ static int div_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; KMOD_IPSTAT_INC(ips_noproto); m_freem(m); return (IPPROTO_DONE); } /* * Divert a packet by passing it up to the divert socket at port 'port'. * * Setup generic address and protocol structures for div_input routine, * then pass them along with mbuf chain. */ static void divert_packet(struct mbuf *m, int incoming) { struct ip *ip; struct inpcb *inp; struct socket *sa; u_int16_t nport; struct sockaddr_in divsrc; struct m_tag *mtag; struct epoch_tracker et; mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); if (mtag == NULL) { m_freem(m); return; } /* Assure header */ if (m->m_len < sizeof(struct ip) && (m = m_pullup(m, sizeof(struct ip))) == NULL) return; ip = mtod(m, struct ip *); /* Delayed checksums are currently not compatible with divert. */ if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { in_delayed_cksum(m); m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } #ifdef SCTP if (m->m_pkthdr.csum_flags & CSUM_SCTP) { sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); m->m_pkthdr.csum_flags &= ~CSUM_SCTP; } #endif bzero(&divsrc, sizeof(divsrc)); divsrc.sin_len = sizeof(divsrc); divsrc.sin_family = AF_INET; /* record matching rule, in host format */ divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; /* * Record receive interface address, if any. * But only for incoming packets. */ if (incoming) { struct ifaddr *ifa; struct ifnet *ifp; /* Sanity check */ M_ASSERTPKTHDR(m); /* Find IP address for receive interface */ ifp = m->m_pkthdr.rcvif; if_addr_rlock(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; divsrc.sin_addr = ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; break; } if_addr_runlock(ifp); } /* * Record the incoming interface name whenever we have one. */ if (m->m_pkthdr.rcvif) { /* * Hide the actual interface name in there in the * sin_zero array. XXX This needs to be moved to a * different sockaddr type for divert, e.g. * sockaddr_div with multiple fields like * sockaddr_dl. Presently we have only 7 bytes * but that will do for now as most interfaces * are 4 or less + 2 or less bytes for unit. * There is probably a faster way of doing this, * possibly taking it from the sockaddr_dl on the iface. * This solves the problem of a P2P link and a LAN interface * having the same address, which can result in the wrong * interface being assigned to the packet when fed back * into the divert socket. Theoretically if the daemon saves * and re-uses the sockaddr_in as suggested in the man pages, * this iface name will come along for the ride. * (see div_output for the other half of this.) */ strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, sizeof(divsrc.sin_zero)); } /* Put packet on socket queue, if any */ sa = NULL; nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); INP_INFO_RLOCK_ET(&V_divcbinfo, et); CK_LIST_FOREACH(inp, &V_divcb, inp_list) { /* XXX why does only one socket match? */ if (inp->inp_lport == nport) { INP_RLOCK(inp); sa = inp->inp_socket; SOCKBUF_LOCK(&sa->so_rcv); if (sbappendaddr_locked(&sa->so_rcv, (struct sockaddr *)&divsrc, m, (struct mbuf *)0) == 0) { SOCKBUF_UNLOCK(&sa->so_rcv); sa = NULL; /* force mbuf reclaim below */ } else sorwakeup_locked(sa); INP_RUNLOCK(inp); break; } } INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); if (sa == NULL) { m_freem(m); KMOD_IPSTAT_INC(ips_noproto); KMOD_IPSTAT_DEC(ips_delivered); } } /* * Deliver packet back into the IP processing machinery. * * If no address specified, or address is 0.0.0.0, send to ip_output(); * otherwise, send to ip_input() and mark as having been received on * the interface with that address. */ static int div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, struct mbuf *control) { struct ip *const ip = mtod(m, struct ip *); struct m_tag *mtag; struct ipfw_rule_ref *dt; int error = 0; /* * An mbuf may hasn't come from userland, but we pretend * that it has. */ m->m_pkthdr.rcvif = NULL; m->m_nextpkt = NULL; M_SETFIB(m, so->so_fibnum); if (control) m_freem(control); /* XXX */ mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); if (mtag == NULL) { /* this should be normal */ mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); if (mtag == NULL) { error = ENOBUFS; goto cantsend; } m_tag_prepend(m, mtag); } dt = (struct ipfw_rule_ref *)(mtag+1); /* Loopback avoidance and state recovery */ if (sin) { int i; /* set the starting point. We provide a non-zero slot, * but a non_matching chain_id to skip that info and use * the rulenum/rule_id. */ dt->slot = 1; /* dummy, chain_id is invalid */ dt->chain_id = 0; dt->rulenum = sin->sin_port+1; /* host format ? */ dt->rule_id = 0; /* * Find receive interface with the given name, stuffed * (if it exists) in the sin_zero[] field. * The name is user supplied data so don't trust its size * or that it is zero terminated. */ for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) ; if ( i > 0 && i < sizeof(sin->sin_zero)) m->m_pkthdr.rcvif = ifunit(sin->sin_zero); } /* Reinject packet into the system as incoming or outgoing */ if (!sin || sin->sin_addr.s_addr == 0) { struct mbuf *options = NULL; struct inpcb *inp; dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; inp = sotoinpcb(so); INP_RLOCK(inp); switch (ip->ip_v) { case IPVERSION: /* * Don't allow both user specified and setsockopt * options, and don't allow packet length sizes that * will crash. */ if ((((ip->ip_hl << 2) != sizeof(struct ip)) && inp->inp_options != NULL) || ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { error = EINVAL; INP_RUNLOCK(inp); goto cantsend; } break; #ifdef INET6 case IPV6_VERSION >> 4: { struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); /* Don't allow packet length sizes that will crash */ if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { error = EINVAL; INP_RUNLOCK(inp); goto cantsend; } break; } #endif default: error = EINVAL; INP_RUNLOCK(inp); goto cantsend; } /* Send packet to output processing */ KMOD_IPSTAT_INC(ips_rawout); /* XXX */ #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif /* * Get ready to inject the packet into ip_output(). * Just in case socket options were specified on the * divert socket, we duplicate them. This is done * to avoid having to hold the PCB locks over the call * to ip_output(), as doing this results in a number of * lock ordering complexities. * * Note that we set the multicast options argument for * ip_output() to NULL since it should be invariant that * they are not present. */ KASSERT(inp->inp_moptions == NULL, ("multicast options set on a divert socket")); /* * XXXCSJP: It is unclear to me whether or not it makes * sense for divert sockets to have options. However, * for now we will duplicate them with the INP locks * held so we can use them in ip_output() without * requring a reference to the pcb. */ if (inp->inp_options != NULL) { options = m_dup(inp->inp_options, M_NOWAIT); if (options == NULL) { INP_RUNLOCK(inp); error = ENOBUFS; goto cantsend; } } INP_RUNLOCK(inp); switch (ip->ip_v) { case IPVERSION: error = ip_output(m, options, NULL, ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); break; #ifdef INET6 case IPV6_VERSION >> 4: error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); break; #endif } if (options != NULL) m_freem(options); } else { dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; if (m->m_pkthdr.rcvif == NULL) { /* * No luck with the name, check by IP address. * Clear the port and the ifname to make sure * there are no distractions for ifa_ifwithaddr. */ struct ifaddr *ifa; bzero(sin->sin_zero, sizeof(sin->sin_zero)); sin->sin_port = 0; NET_EPOCH_ENTER(); ifa = ifa_ifwithaddr((struct sockaddr *) sin); if (ifa == NULL) { error = EADDRNOTAVAIL; NET_EPOCH_EXIT(); goto cantsend; } m->m_pkthdr.rcvif = ifa->ifa_ifp; NET_EPOCH_EXIT(); } #ifdef MAC mac_socket_create_mbuf(so, m); #endif /* Send packet to input processing via netisr */ switch (ip->ip_v) { case IPVERSION: /* * Restore M_BCAST flag when destination address is * broadcast. It is expected by ip_tryforward(). */ if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) m->m_flags |= M_MCAST; else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) m->m_flags |= M_BCAST; netisr_queue_src(NETISR_IP, (uintptr_t)so, m); break; #ifdef INET6 case IPV6_VERSION >> 4: netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); break; #endif default: error = EINVAL; goto cantsend; } } return (error); cantsend: m_freem(m); return (error); } static int div_attach(struct socket *so, int proto, struct thread *td) { struct inpcb *inp; int error; inp = sotoinpcb(so); KASSERT(inp == NULL, ("div_attach: inp != NULL")); if (td != NULL) { error = priv_check(td, PRIV_NETINET_DIVERT); if (error) return (error); } error = soreserve(so, div_sendspace, div_recvspace); if (error) return error; INP_INFO_WLOCK(&V_divcbinfo); error = in_pcballoc(so, &V_divcbinfo); if (error) { INP_INFO_WUNLOCK(&V_divcbinfo); return error; } inp = (struct inpcb *)so->so_pcb; INP_INFO_WUNLOCK(&V_divcbinfo); inp->inp_ip_p = proto; inp->inp_vflag |= INP_IPV4; inp->inp_flags |= INP_HDRINCL; INP_WUNLOCK(inp); return 0; } static void div_detach(struct socket *so) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("div_detach: inp == NULL")); INP_INFO_WLOCK(&V_divcbinfo); INP_WLOCK(inp); in_pcbdetach(inp); in_pcbfree(inp); INP_INFO_WUNLOCK(&V_divcbinfo); } static int div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { struct inpcb *inp; int error; inp = sotoinpcb(so); KASSERT(inp != NULL, ("div_bind: inp == NULL")); /* in_pcbbind assumes that nam is a sockaddr_in * and in_pcbbind requires a valid address. Since divert * sockets don't we need to make sure the address is * filled in properly. * XXX -- divert should not be abusing in_pcbind * and should probably have its own family. */ if (nam->sa_family != AF_INET) return EAFNOSUPPORT; ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; INP_INFO_WLOCK(&V_divcbinfo); INP_WLOCK(inp); INP_HASH_WLOCK(&V_divcbinfo); error = in_pcbbind(inp, nam, td->td_ucred); INP_HASH_WUNLOCK(&V_divcbinfo); INP_WUNLOCK(inp); INP_INFO_WUNLOCK(&V_divcbinfo); return error; } static int div_shutdown(struct socket *so) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); INP_WLOCK(inp); socantsendmore(so); INP_WUNLOCK(inp); return 0; } static int div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { /* Packet must have a header (but that's about it) */ if (m->m_len < sizeof (struct ip) && (m = m_pullup(m, sizeof (struct ip))) == NULL) { KMOD_IPSTAT_INC(ips_toosmall); m_freem(m); return EINVAL; } /* Send packet */ return div_output(so, m, (struct sockaddr_in *)nam, control); } static void div_ctlinput(int cmd, struct sockaddr *sa, void *vip) { struct in_addr faddr; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (PRC_IS_REDIRECT(cmd)) return; } static int div_pcblist(SYSCTL_HANDLER_ARGS) { int error, i, n; struct inpcb *inp, **inp_list; inp_gen_t gencnt; struct xinpgen xig; struct epoch_tracker et; /* * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == 0) { n = V_divcbinfo.ipi_count; n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); return 0; } if (req->newptr != 0) return EPERM; /* * OK, now we're committed to doing something. */ INP_INFO_WLOCK(&V_divcbinfo); gencnt = V_divcbinfo.ipi_gencnt; n = V_divcbinfo.ipi_count; INP_INFO_WUNLOCK(&V_divcbinfo); error = sysctl_wire_old_buffer(req, 2 * sizeof(xig) + n*sizeof(struct xinpcb)); if (error != 0) return (error); xig.xig_len = sizeof xig; xig.xig_count = n; xig.xig_gen = gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return error; inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == NULL) return ENOMEM; INP_INFO_RLOCK_ET(&V_divcbinfo, et); for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; inp = CK_LIST_NEXT(inp, inp_list)) { INP_WLOCK(inp); if (inp->inp_gencnt <= gencnt && cr_canseeinpcb(req->td->td_ucred, inp) == 0) { in_pcbref(inp); inp_list[i++] = inp; } INP_WUNLOCK(inp); } INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); n = i; error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (inp->inp_gencnt <= gencnt) { struct xinpcb xi; in_pcbtoxinpcb(inp, &xi); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xi, sizeof xi); } else INP_RUNLOCK(inp); } INP_INFO_WLOCK(&V_divcbinfo); for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_WUNLOCK(&V_divcbinfo); if (!error) { struct epoch_tracker et; /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ INP_INFO_RLOCK_ET(&V_divcbinfo, et); xig.xig_gen = V_divcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_divcbinfo.ipi_count; INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); error = SYSCTL_OUT(req, &xig, sizeof xig); } free(inp_list, M_TEMP); return error; } #ifdef SYSCTL_NODE static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets"); #endif struct pr_usrreqs div_usrreqs = { .pru_attach = div_attach, .pru_bind = div_bind, .pru_control = in_control, .pru_detach = div_detach, .pru_peeraddr = in_getpeeraddr, .pru_send = div_send, .pru_shutdown = div_shutdown, .pru_sockaddr = in_getsockaddr, .pru_sosetlabel = in_pcbsosetlabel }; struct protosw div_protosw = { .pr_type = SOCK_RAW, .pr_protocol = IPPROTO_DIVERT, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_input = div_input, .pr_ctlinput = div_ctlinput, .pr_ctloutput = ip_ctloutput, .pr_init = div_init, .pr_usrreqs = &div_usrreqs }; static int div_modevent(module_t mod, int type, void *unused) { int err = 0; switch (type) { case MOD_LOAD: /* * Protocol will be initialized by pf_proto_register(). * We don't have to register ip_protox because we are not * a true IP protocol that goes over the wire. */ err = pf_proto_register(PF_INET, &div_protosw); if (err != 0) return (err); ip_divert_ptr = divert_packet; ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, NULL, EVENTHANDLER_PRI_ANY); break; case MOD_QUIESCE: /* * IPDIVERT may normally not be unloaded because of the * potential race conditions. Tell kldunload we can't be * unloaded unless the unload is forced. */ err = EPERM; break; case MOD_UNLOAD: /* * Forced unload. * * Module ipdivert can only be unloaded if no sockets are * connected. Maybe this can be changed later to forcefully * disconnect any open sockets. * * XXXRW: Note that there is a slight race here, as a new * socket open request could be spinning on the lock and then * we destroy the lock. */ INP_INFO_WLOCK(&V_divcbinfo); if (V_divcbinfo.ipi_count != 0) { err = EBUSY; INP_INFO_WUNLOCK(&V_divcbinfo); break; } ip_divert_ptr = NULL; err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); INP_INFO_WUNLOCK(&V_divcbinfo); #ifndef VIMAGE div_destroy(NULL); #endif EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); break; default: err = EOPNOTSUPP; break; } return err; } static moduledata_t ipdivertmod = { "ipdivert", div_modevent, 0 }; DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3); MODULE_VERSION(ipdivert, 1); Index: head/sys/netinet/ip_gre.c =================================================================== --- head/sys/netinet/ip_gre.c (revision 336675) +++ head/sys/netinet/ip_gre.c (revision 336676) @@ -1,300 +1,300 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c) 1998 The NetBSD Foundation, Inc. * Copyright (c) 2014, 2018 Andrey V. Elsukov * All rights reserved. * * This code is derived from software contributed to The NetBSD Foundation * by Heiko W.Rupp * * IPv6-over-GRE contributed by Gert Doering * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * $NetBSD: ip_gre.c,v 1.29 2003/09/05 23:02:43 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #define GRE_TTL 30 VNET_DEFINE(int, ip_gre_ttl) = GRE_TTL; #define V_ip_gre_ttl VNET(ip_gre_ttl) SYSCTL_INT(_net_inet_ip, OID_AUTO, grettl, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_gre_ttl), 0, "Default TTL value for encapsulated packets"); -static VNET_DEFINE(struct gre_list *, ipv4_hashtbl) = NULL; +VNET_DEFINE_STATIC(struct gre_list *, ipv4_hashtbl) = NULL; #define V_ipv4_hashtbl VNET(ipv4_hashtbl) #define GRE_HASH(src, dst) (V_ipv4_hashtbl[\ in_gre_hashval((src), (dst)) & (GRE_HASH_SIZE - 1)]) #define GRE_HASH_SC(sc) GRE_HASH((sc)->gre_oip.ip_src.s_addr,\ (sc)->gre_oip.ip_dst.s_addr) static uint32_t in_gre_hashval(in_addr_t src, in_addr_t dst) { uint32_t ret; ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT); return (fnv_32_buf(&dst, sizeof(dst), ret)); } static int in_gre_checkdup(const struct gre_softc *sc, in_addr_t src, in_addr_t dst) { struct gre_softc *tmp; if (sc->gre_family == AF_INET && sc->gre_oip.ip_src.s_addr == src && sc->gre_oip.ip_dst.s_addr == dst) return (EEXIST); CK_LIST_FOREACH(tmp, &GRE_HASH(src, dst), chain) { if (tmp == sc) continue; if (tmp->gre_oip.ip_src.s_addr == src && tmp->gre_oip.ip_dst.s_addr == dst) return (EADDRNOTAVAIL); } return (0); } static int in_gre_lookup(const struct mbuf *m, int off, int proto, void **arg) { const struct ip *ip; struct gre_softc *sc; if (V_ipv4_hashtbl == NULL) return (0); MPASS(in_epoch(net_epoch_preempt)); ip = mtod(m, const struct ip *); CK_LIST_FOREACH(sc, &GRE_HASH(ip->ip_dst.s_addr, ip->ip_src.s_addr), chain) { /* * This is an inbound packet, its ip_dst is source address * in softc. */ if (sc->gre_oip.ip_src.s_addr == ip->ip_dst.s_addr && sc->gre_oip.ip_dst.s_addr == ip->ip_src.s_addr) { if ((GRE2IFP(sc)->if_flags & IFF_UP) == 0) return (0); *arg = sc; return (ENCAP_DRV_LOOKUP); } } return (0); } static void in_gre_attach(struct gre_softc *sc) { sc->gre_hlen = sizeof(struct greip); sc->gre_oip.ip_v = IPVERSION; sc->gre_oip.ip_hl = sizeof(struct ip) >> 2; sc->gre_oip.ip_p = IPPROTO_GRE; gre_updatehdr(sc, &sc->gre_gihdr->gi_gre); CK_LIST_INSERT_HEAD(&GRE_HASH_SC(sc), sc, chain); } void in_gre_setopts(struct gre_softc *sc, u_long cmd, uint32_t value) { MPASS(cmd == GRESKEY || cmd == GRESOPTS); /* NOTE: we are protected with gre_ioctl_sx lock */ MPASS(sc->gre_family == AF_INET); CK_LIST_REMOVE(sc, chain); GRE_WAIT(); if (cmd == GRESKEY) sc->gre_key = value; else sc->gre_options = value; in_gre_attach(sc); } int in_gre_ioctl(struct gre_softc *sc, u_long cmd, caddr_t data) { struct ifreq *ifr = (struct ifreq *)data; struct sockaddr_in *dst, *src; struct ip *ip; int error; /* NOTE: we are protected with gre_ioctl_sx lock */ error = EINVAL; switch (cmd) { case SIOCSIFPHYADDR: src = &((struct in_aliasreq *)data)->ifra_addr; dst = &((struct in_aliasreq *)data)->ifra_dstaddr; /* sanity checks */ if (src->sin_family != dst->sin_family || src->sin_family != AF_INET || src->sin_len != dst->sin_len || src->sin_len != sizeof(*src)) break; if (src->sin_addr.s_addr == INADDR_ANY || dst->sin_addr.s_addr == INADDR_ANY) { error = EADDRNOTAVAIL; break; } if (V_ipv4_hashtbl == NULL) V_ipv4_hashtbl = gre_hashinit(); error = in_gre_checkdup(sc, src->sin_addr.s_addr, dst->sin_addr.s_addr); if (error == EADDRNOTAVAIL) break; if (error == EEXIST) { /* Addresses are the same. Just return. */ error = 0; break; } ip = malloc(sizeof(struct greip) + 3 * sizeof(uint32_t), M_GRE, M_WAITOK | M_ZERO); ip->ip_src.s_addr = src->sin_addr.s_addr; ip->ip_dst.s_addr = dst->sin_addr.s_addr; if (sc->gre_family != 0) { /* Detach existing tunnel first */ CK_LIST_REMOVE(sc, chain); GRE_WAIT(); free(sc->gre_hdr, M_GRE); /* XXX: should we notify about link state change? */ } sc->gre_family = AF_INET; sc->gre_hdr = ip; sc->gre_oseq = 0; sc->gre_iseq = UINT32_MAX; in_gre_attach(sc); break; case SIOCGIFPSRCADDR: case SIOCGIFPDSTADDR: if (sc->gre_family != AF_INET) { error = EADDRNOTAVAIL; break; } src = (struct sockaddr_in *)&ifr->ifr_addr; memset(src, 0, sizeof(*src)); src->sin_family = AF_INET; src->sin_len = sizeof(*src); src->sin_addr = (cmd == SIOCGIFPSRCADDR) ? sc->gre_oip.ip_src: sc->gre_oip.ip_dst; error = prison_if(curthread->td_ucred, (struct sockaddr *)src); if (error != 0) memset(src, 0, sizeof(*src)); break; } return (error); } int in_gre_output(struct mbuf *m, int af, int hlen) { struct greip *gi; gi = mtod(m, struct greip *); switch (af) { case AF_INET: /* * gre_transmit() has used M_PREPEND() that doesn't guarantee * m_data is contiguous more than hlen bytes. Use m_copydata() * here to avoid m_pullup(). */ m_copydata(m, hlen + offsetof(struct ip, ip_tos), sizeof(u_char), &gi->gi_ip.ip_tos); m_copydata(m, hlen + offsetof(struct ip, ip_id), sizeof(u_short), (caddr_t)&gi->gi_ip.ip_id); break; #ifdef INET6 case AF_INET6: gi->gi_ip.ip_tos = 0; /* XXX */ ip_fillid(&gi->gi_ip); break; #endif } gi->gi_ip.ip_ttl = V_ip_gre_ttl; gi->gi_ip.ip_len = htons(m->m_pkthdr.len); return (ip_output(m, NULL, NULL, IP_FORWARDING, NULL, NULL)); } static const struct encaptab *ecookie = NULL; static const struct encap_config ipv4_encap_cfg = { .proto = IPPROTO_GRE, .min_length = sizeof(struct greip) + sizeof(struct ip), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in_gre_lookup, .input = gre_input }; void in_gre_init(void) { if (!IS_DEFAULT_VNET(curvnet)) return; ecookie = ip_encap_attach(&ipv4_encap_cfg, NULL, M_WAITOK); } void in_gre_uninit(void) { if (IS_DEFAULT_VNET(curvnet)) ip_encap_detach(ecookie); if (V_ipv4_hashtbl != NULL) gre_hashdestroy(V_ipv4_hashtbl); } Index: head/sys/netinet/ip_icmp.c =================================================================== --- head/sys/netinet/ip_icmp.c (revision 336675) +++ head/sys/netinet/ip_icmp.c (revision 336676) @@ -1,1055 +1,1055 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_icmp.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif /* INET */ /* * ICMP routines: error generation, receive packet processing, and * routines to turnaround packets back to the originator, and * host table maintenance routines. */ -static VNET_DEFINE(int, icmplim) = 200; +VNET_DEFINE_STATIC(int, icmplim) = 200; #define V_icmplim VNET(icmplim) SYSCTL_INT(_net_inet_icmp, ICMPCTL_ICMPLIM, icmplim, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim), 0, "Maximum number of ICMP responses per second"); -static VNET_DEFINE(int, icmplim_output) = 1; +VNET_DEFINE_STATIC(int, icmplim_output) = 1; #define V_icmplim_output VNET(icmplim_output) SYSCTL_INT(_net_inet_icmp, OID_AUTO, icmplim_output, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmplim_output), 0, "Enable logging of ICMP response rate limiting"); #ifdef INET VNET_PCPUSTAT_DEFINE(struct icmpstat, icmpstat); VNET_PCPUSTAT_SYSINIT(icmpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_icmp, ICMPCTL_STATS, stats, struct icmpstat, icmpstat, "ICMP statistics (struct icmpstat, netinet/icmp_var.h)"); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(icmpstat); #endif /* VIMAGE */ -static VNET_DEFINE(int, icmpmaskrepl) = 0; +VNET_DEFINE_STATIC(int, icmpmaskrepl) = 0; #define V_icmpmaskrepl VNET(icmpmaskrepl) SYSCTL_INT(_net_inet_icmp, ICMPCTL_MASKREPL, maskrepl, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmpmaskrepl), 0, "Reply to ICMP Address Mask Request packets"); -static VNET_DEFINE(u_int, icmpmaskfake) = 0; +VNET_DEFINE_STATIC(u_int, icmpmaskfake) = 0; #define V_icmpmaskfake VNET(icmpmaskfake) SYSCTL_UINT(_net_inet_icmp, OID_AUTO, maskfake, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmpmaskfake), 0, "Fake reply to ICMP Address Mask Request packets"); VNET_DEFINE(int, drop_redirect) = 0; #define V_drop_redirect VNET(drop_redirect) SYSCTL_INT(_net_inet_icmp, OID_AUTO, drop_redirect, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(drop_redirect), 0, "Ignore ICMP redirects"); -static VNET_DEFINE(int, log_redirect) = 0; +VNET_DEFINE_STATIC(int, log_redirect) = 0; #define V_log_redirect VNET(log_redirect) SYSCTL_INT(_net_inet_icmp, OID_AUTO, log_redirect, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(log_redirect), 0, "Log ICMP redirects to the console"); -static VNET_DEFINE(char, reply_src[IFNAMSIZ]); +VNET_DEFINE_STATIC(char, reply_src[IFNAMSIZ]); #define V_reply_src VNET(reply_src) SYSCTL_STRING(_net_inet_icmp, OID_AUTO, reply_src, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(reply_src), IFNAMSIZ, "ICMP reply source for non-local packets"); -static VNET_DEFINE(int, icmp_rfi) = 0; +VNET_DEFINE_STATIC(int, icmp_rfi) = 0; #define V_icmp_rfi VNET(icmp_rfi) SYSCTL_INT(_net_inet_icmp, OID_AUTO, reply_from_interface, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_rfi), 0, "ICMP reply from incoming interface for non-local packets"); /* Router requirements RFC 1812 section 4.3.2.3 requires 576 - 28. */ -static VNET_DEFINE(int, icmp_quotelen) = 548; +VNET_DEFINE_STATIC(int, icmp_quotelen) = 8; #define V_icmp_quotelen VNET(icmp_quotelen) SYSCTL_INT(_net_inet_icmp, OID_AUTO, quotelen, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_quotelen), 0, "Number of bytes from original packet to quote in ICMP reply"); -static VNET_DEFINE(int, icmpbmcastecho) = 0; +VNET_DEFINE_STATIC(int, icmpbmcastecho) = 0; #define V_icmpbmcastecho VNET(icmpbmcastecho) SYSCTL_INT(_net_inet_icmp, OID_AUTO, bmcastecho, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmpbmcastecho), 0, "Reply to multicast ICMP Echo Request and Timestamp packets"); -static VNET_DEFINE(int, icmptstamprepl) = 1; +VNET_DEFINE_STATIC(int, icmptstamprepl) = 1; #define V_icmptstamprepl VNET(icmptstamprepl) SYSCTL_INT(_net_inet_icmp, OID_AUTO, tstamprepl, CTLFLAG_RW, &VNET_NAME(icmptstamprepl), 0, "Respond to ICMP Timestamp packets"); #ifdef ICMPPRINTFS int icmpprintfs = 0; #endif static void icmp_reflect(struct mbuf *); static void icmp_send(struct mbuf *, struct mbuf *); extern struct protosw inetsw[]; /* * Kernel module interface for updating icmpstat. The argument is an index * into icmpstat treated as an array of u_long. While this encodes the * general layout of icmpstat into the caller, it doesn't encode its * location, so that future changes to add, for example, per-CPU stats * support won't cause binary compatibility problems for kernel modules. */ void kmod_icmpstat_inc(int statnum) { counter_u64_add(VNET(icmpstat)[statnum], 1); } /* * Generate an error packet of type error * in response to bad packet ip. */ void icmp_error(struct mbuf *n, int type, int code, uint32_t dest, int mtu) { struct ip *oip, *nip; struct icmp *icp; struct mbuf *m; unsigned icmplen, icmpelen, nlen, oiphlen; KASSERT((u_int)type <= ICMP_MAXTYPE, ("%s: illegal ICMP type", __func__)); if (type != ICMP_REDIRECT) ICMPSTAT_INC(icps_error); /* * Don't send error: * if the original packet was encrypted. * if not the first fragment of message. * in response to a multicast or broadcast packet. * if the old packet protocol was an ICMP error message. */ if (n->m_flags & M_DECRYPTED) goto freeit; if (n->m_flags & (M_BCAST|M_MCAST)) goto freeit; /* Drop if IP header plus 8 bytes is not contiguous in first mbuf. */ if (n->m_len < sizeof(struct ip) + ICMP_MINLEN) goto freeit; oip = mtod(n, struct ip *); oiphlen = oip->ip_hl << 2; if (n->m_len < oiphlen + ICMP_MINLEN) goto freeit; #ifdef ICMPPRINTFS if (icmpprintfs) printf("icmp_error(%p, %x, %d)\n", oip, type, code); #endif if (oip->ip_off & htons(~(IP_MF|IP_DF))) goto freeit; if (oip->ip_p == IPPROTO_ICMP && type != ICMP_REDIRECT && !ICMP_INFOTYPE(((struct icmp *)((caddr_t)oip + oiphlen))->icmp_type)) { ICMPSTAT_INC(icps_oldicmp); goto freeit; } /* * Calculate length to quote from original packet and * prevent the ICMP mbuf from overflowing. * Unfortunately this is non-trivial since ip_forward() * sends us truncated packets. */ nlen = m_length(n, NULL); if (oip->ip_p == IPPROTO_TCP) { struct tcphdr *th; int tcphlen; if (oiphlen + sizeof(struct tcphdr) > n->m_len && n->m_next == NULL) goto stdreply; if (n->m_len < oiphlen + sizeof(struct tcphdr) && (n = m_pullup(n, oiphlen + sizeof(struct tcphdr))) == NULL) goto freeit; oip = mtod(n, struct ip *); th = mtodo(n, oiphlen); tcphlen = th->th_off << 2; if (tcphlen < sizeof(struct tcphdr)) goto freeit; if (ntohs(oip->ip_len) < oiphlen + tcphlen) goto freeit; if (oiphlen + tcphlen > n->m_len && n->m_next == NULL) goto stdreply; if (n->m_len < oiphlen + tcphlen && (n = m_pullup(n, oiphlen + tcphlen)) == NULL) goto freeit; icmpelen = max(tcphlen, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen)); } else if (oip->ip_p == IPPROTO_SCTP) { struct sctphdr *sh; struct sctp_chunkhdr *ch; if (ntohs(oip->ip_len) < oiphlen + sizeof(struct sctphdr)) goto stdreply; if (oiphlen + sizeof(struct sctphdr) > n->m_len && n->m_next == NULL) goto stdreply; if (n->m_len < oiphlen + sizeof(struct sctphdr) && (n = m_pullup(n, oiphlen + sizeof(struct sctphdr))) == NULL) goto freeit; oip = mtod(n, struct ip *); icmpelen = max(sizeof(struct sctphdr), min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen)); sh = mtodo(n, oiphlen); if (ntohl(sh->v_tag) == 0 && ntohs(oip->ip_len) >= oiphlen + sizeof(struct sctphdr) + 8 && (n->m_len >= oiphlen + sizeof(struct sctphdr) + 8 || n->m_next != NULL)) { if (n->m_len < oiphlen + sizeof(struct sctphdr) + 8 && (n = m_pullup(n, oiphlen + sizeof(struct sctphdr) + 8)) == NULL) goto freeit; oip = mtod(n, struct ip *); sh = mtodo(n, oiphlen); ch = (struct sctp_chunkhdr *)(sh + 1); if (ch->chunk_type == SCTP_INITIATION) { icmpelen = max(sizeof(struct sctphdr) + 8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen)); } } } else stdreply: icmpelen = max(8, min(V_icmp_quotelen, ntohs(oip->ip_len) - oiphlen)); icmplen = min(oiphlen + icmpelen, nlen); if (icmplen < sizeof(struct ip)) goto freeit; if (MHLEN > sizeof(struct ip) + ICMP_MINLEN + icmplen) m = m_gethdr(M_NOWAIT, MT_DATA); else m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) goto freeit; #ifdef MAC mac_netinet_icmp_reply(n, m); #endif icmplen = min(icmplen, M_TRAILINGSPACE(m) - sizeof(struct ip) - ICMP_MINLEN); m_align(m, ICMP_MINLEN + icmplen); m->m_len = ICMP_MINLEN + icmplen; /* XXX MRT make the outgoing packet use the same FIB * that was associated with the incoming packet */ M_SETFIB(m, M_GETFIB(n)); icp = mtod(m, struct icmp *); ICMPSTAT_INC(icps_outhist[type]); icp->icmp_type = type; if (type == ICMP_REDIRECT) icp->icmp_gwaddr.s_addr = dest; else { icp->icmp_void = 0; /* * The following assignments assume an overlay with the * just zeroed icmp_void field. */ if (type == ICMP_PARAMPROB) { icp->icmp_pptr = code; code = 0; } else if (type == ICMP_UNREACH && code == ICMP_UNREACH_NEEDFRAG && mtu) { icp->icmp_nextmtu = htons(mtu); } } icp->icmp_code = code; /* * Copy the quotation into ICMP message and * convert quoted IP header back to network representation. */ m_copydata(n, 0, icmplen, (caddr_t)&icp->icmp_ip); nip = &icp->icmp_ip; /* * Set up ICMP message mbuf and copy old IP header (without options * in front of ICMP message. * If the original mbuf was meant to bypass the firewall, the error * reply should bypass as well. */ m->m_flags |= n->m_flags & M_SKIP_FIREWALL; m->m_data -= sizeof(struct ip); m->m_len += sizeof(struct ip); m->m_pkthdr.len = m->m_len; m->m_pkthdr.rcvif = n->m_pkthdr.rcvif; nip = mtod(m, struct ip *); bcopy((caddr_t)oip, (caddr_t)nip, sizeof(struct ip)); nip->ip_len = htons(m->m_len); nip->ip_v = IPVERSION; nip->ip_hl = 5; nip->ip_p = IPPROTO_ICMP; nip->ip_tos = 0; nip->ip_off = 0; icmp_reflect(m); freeit: m_freem(n); } /* * Process a received ICMP message. */ int icmp_input(struct mbuf **mp, int *offp, int proto) { struct icmp *icp; struct in_ifaddr *ia; struct mbuf *m = *mp; struct ip *ip = mtod(m, struct ip *); struct sockaddr_in icmpsrc, icmpdst, icmpgw; int hlen = *offp; int icmplen = ntohs(ip->ip_len) - *offp; int i, code; void (*ctlfunc)(int, struct sockaddr *, void *); int fibnum; *mp = NULL; /* * Locate icmp structure in mbuf, and check * that not corrupted and of at least minimum length. */ #ifdef ICMPPRINTFS if (icmpprintfs) { char srcbuf[INET_ADDRSTRLEN]; char dstbuf[INET_ADDRSTRLEN]; printf("icmp_input from %s to %s, len %d\n", inet_ntoa_r(ip->ip_src, srcbuf), inet_ntoa_r(ip->ip_dst, dstbuf), icmplen); } #endif NET_EPOCH_ENTER(); if (icmplen < ICMP_MINLEN) { ICMPSTAT_INC(icps_tooshort); goto freeit; } i = hlen + min(icmplen, ICMP_ADVLENMIN); if (m->m_len < i && (m = m_pullup(m, i)) == NULL) { ICMPSTAT_INC(icps_tooshort); NET_EPOCH_EXIT(); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); m->m_len -= hlen; m->m_data += hlen; icp = mtod(m, struct icmp *); if (in_cksum(m, icmplen)) { ICMPSTAT_INC(icps_checksum); goto freeit; } m->m_len += hlen; m->m_data -= hlen; #ifdef ICMPPRINTFS if (icmpprintfs) printf("icmp_input, type %d code %d\n", icp->icmp_type, icp->icmp_code); #endif /* * Message type specific processing. */ if (icp->icmp_type > ICMP_MAXTYPE) goto raw; /* Initialize */ bzero(&icmpsrc, sizeof(icmpsrc)); icmpsrc.sin_len = sizeof(struct sockaddr_in); icmpsrc.sin_family = AF_INET; bzero(&icmpdst, sizeof(icmpdst)); icmpdst.sin_len = sizeof(struct sockaddr_in); icmpdst.sin_family = AF_INET; bzero(&icmpgw, sizeof(icmpgw)); icmpgw.sin_len = sizeof(struct sockaddr_in); icmpgw.sin_family = AF_INET; ICMPSTAT_INC(icps_inhist[icp->icmp_type]); code = icp->icmp_code; switch (icp->icmp_type) { case ICMP_UNREACH: switch (code) { case ICMP_UNREACH_NET: case ICMP_UNREACH_HOST: case ICMP_UNREACH_SRCFAIL: case ICMP_UNREACH_NET_UNKNOWN: case ICMP_UNREACH_HOST_UNKNOWN: case ICMP_UNREACH_ISOLATED: case ICMP_UNREACH_TOSNET: case ICMP_UNREACH_TOSHOST: case ICMP_UNREACH_HOST_PRECEDENCE: case ICMP_UNREACH_PRECEDENCE_CUTOFF: code = PRC_UNREACH_NET; break; case ICMP_UNREACH_NEEDFRAG: code = PRC_MSGSIZE; break; /* * RFC 1122, Sections 3.2.2.1 and 4.2.3.9. * Treat subcodes 2,3 as immediate RST */ case ICMP_UNREACH_PROTOCOL: code = PRC_UNREACH_PROTOCOL; break; case ICMP_UNREACH_PORT: code = PRC_UNREACH_PORT; break; case ICMP_UNREACH_NET_PROHIB: case ICMP_UNREACH_HOST_PROHIB: case ICMP_UNREACH_FILTER_PROHIB: code = PRC_UNREACH_ADMIN_PROHIB; break; default: goto badcode; } goto deliver; case ICMP_TIMXCEED: if (code > 1) goto badcode; code += PRC_TIMXCEED_INTRANS; goto deliver; case ICMP_PARAMPROB: if (code > 1) goto badcode; code = PRC_PARAMPROB; deliver: /* * Problem with datagram; advise higher level routines. */ if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) || icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) { ICMPSTAT_INC(icps_badlen); goto freeit; } /* Discard ICMP's in response to multicast packets */ if (IN_MULTICAST(ntohl(icp->icmp_ip.ip_dst.s_addr))) goto badcode; #ifdef ICMPPRINTFS if (icmpprintfs) printf("deliver to protocol %d\n", icp->icmp_ip.ip_p); #endif icmpsrc.sin_addr = icp->icmp_ip.ip_dst; /* * XXX if the packet contains [IPv4 AH TCP], we can't make a * notification to TCP layer. */ i = sizeof(struct ip) + min(icmplen, ICMP_ADVLENPREF(icp)); ip_stripoptions(m); if (m->m_len < i && (m = m_pullup(m, i)) == NULL) { /* This should actually not happen */ ICMPSTAT_INC(icps_tooshort); NET_EPOCH_EXIT(); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); icp = (struct icmp *)(ip + 1); /* * The upper layer handler can rely on: * - The outer IP header has no options. * - The outer IP header, the ICMP header, the inner IP header, * and the first n bytes of the inner payload are contiguous. * n is at least 8, but might be larger based on * ICMP_ADVLENPREF. See its definition in ip_icmp.h. */ ctlfunc = inetsw[ip_protox[icp->icmp_ip.ip_p]].pr_ctlinput; if (ctlfunc) (*ctlfunc)(code, (struct sockaddr *)&icmpsrc, (void *)&icp->icmp_ip); break; badcode: ICMPSTAT_INC(icps_badcode); break; case ICMP_ECHO: if (!V_icmpbmcastecho && (m->m_flags & (M_MCAST | M_BCAST)) != 0) { ICMPSTAT_INC(icps_bmcastecho); break; } if (badport_bandlim(BANDLIM_ICMP_ECHO) < 0) goto freeit; icp->icmp_type = ICMP_ECHOREPLY; goto reflect; case ICMP_TSTAMP: if (V_icmptstamprepl == 0) break; if (!V_icmpbmcastecho && (m->m_flags & (M_MCAST | M_BCAST)) != 0) { ICMPSTAT_INC(icps_bmcasttstamp); break; } if (icmplen < ICMP_TSLEN) { ICMPSTAT_INC(icps_badlen); break; } if (badport_bandlim(BANDLIM_ICMP_TSTAMP) < 0) goto freeit; icp->icmp_type = ICMP_TSTAMPREPLY; icp->icmp_rtime = iptime(); icp->icmp_ttime = icp->icmp_rtime; /* bogus, do later! */ goto reflect; case ICMP_MASKREQ: if (V_icmpmaskrepl == 0) break; /* * We are not able to respond with all ones broadcast * unless we receive it over a point-to-point interface. */ if (icmplen < ICMP_MASKLEN) break; switch (ip->ip_dst.s_addr) { case INADDR_BROADCAST: case INADDR_ANY: icmpdst.sin_addr = ip->ip_src; break; default: icmpdst.sin_addr = ip->ip_dst; } ia = (struct in_ifaddr *)ifaof_ifpforaddr( (struct sockaddr *)&icmpdst, m->m_pkthdr.rcvif); if (ia == NULL) break; if (ia->ia_ifp == NULL) break; icp->icmp_type = ICMP_MASKREPLY; if (V_icmpmaskfake == 0) icp->icmp_mask = ia->ia_sockmask.sin_addr.s_addr; else icp->icmp_mask = V_icmpmaskfake; if (ip->ip_src.s_addr == 0) { if (ia->ia_ifp->if_flags & IFF_BROADCAST) ip->ip_src = satosin(&ia->ia_broadaddr)->sin_addr; else if (ia->ia_ifp->if_flags & IFF_POINTOPOINT) ip->ip_src = satosin(&ia->ia_dstaddr)->sin_addr; } reflect: ICMPSTAT_INC(icps_reflect); ICMPSTAT_INC(icps_outhist[icp->icmp_type]); icmp_reflect(m); NET_EPOCH_EXIT(); return (IPPROTO_DONE); case ICMP_REDIRECT: if (V_log_redirect) { u_long src, dst, gw; src = ntohl(ip->ip_src.s_addr); dst = ntohl(icp->icmp_ip.ip_dst.s_addr); gw = ntohl(icp->icmp_gwaddr.s_addr); printf("icmp redirect from %d.%d.%d.%d: " "%d.%d.%d.%d => %d.%d.%d.%d\n", (int)(src >> 24), (int)((src >> 16) & 0xff), (int)((src >> 8) & 0xff), (int)(src & 0xff), (int)(dst >> 24), (int)((dst >> 16) & 0xff), (int)((dst >> 8) & 0xff), (int)(dst & 0xff), (int)(gw >> 24), (int)((gw >> 16) & 0xff), (int)((gw >> 8) & 0xff), (int)(gw & 0xff)); } /* * RFC1812 says we must ignore ICMP redirects if we * are acting as router. */ if (V_drop_redirect || V_ipforwarding) break; if (code > 3) goto badcode; if (icmplen < ICMP_ADVLENMIN || icmplen < ICMP_ADVLEN(icp) || icp->icmp_ip.ip_hl < (sizeof(struct ip) >> 2)) { ICMPSTAT_INC(icps_badlen); break; } /* * Short circuit routing redirects to force * immediate change in the kernel's routing * tables. The message is also handed to anyone * listening on a raw socket (e.g. the routing * daemon for use in updating its tables). */ icmpgw.sin_addr = ip->ip_src; icmpdst.sin_addr = icp->icmp_gwaddr; #ifdef ICMPPRINTFS if (icmpprintfs) { char dstbuf[INET_ADDRSTRLEN]; char gwbuf[INET_ADDRSTRLEN]; printf("redirect dst %s to %s\n", inet_ntoa_r(icp->icmp_ip.ip_dst, dstbuf), inet_ntoa_r(icp->icmp_gwaddr, gwbuf)); } #endif icmpsrc.sin_addr = icp->icmp_ip.ip_dst; for ( fibnum = 0; fibnum < rt_numfibs; fibnum++) { in_rtredirect((struct sockaddr *)&icmpsrc, (struct sockaddr *)&icmpdst, (struct sockaddr *)0, RTF_GATEWAY | RTF_HOST, (struct sockaddr *)&icmpgw, fibnum); } pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&icmpsrc); break; /* * No kernel processing for the following; * just fall through to send to raw listener. */ case ICMP_ECHOREPLY: case ICMP_ROUTERADVERT: case ICMP_ROUTERSOLICIT: case ICMP_TSTAMPREPLY: case ICMP_IREQREPLY: case ICMP_MASKREPLY: case ICMP_SOURCEQUENCH: default: break; } raw: NET_EPOCH_EXIT(); *mp = m; rip_input(mp, offp, proto); return (IPPROTO_DONE); freeit: NET_EPOCH_EXIT(); m_freem(m); return (IPPROTO_DONE); } /* * Reflect the ip packet back to the source */ static void icmp_reflect(struct mbuf *m) { struct rm_priotracker in_ifa_tracker; struct ip *ip = mtod(m, struct ip *); struct ifaddr *ifa; struct ifnet *ifp; struct in_ifaddr *ia; struct in_addr t; struct nhop4_extended nh_ext; struct mbuf *opts = NULL; int optlen = (ip->ip_hl << 2) - sizeof(struct ip); if (IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || IN_EXPERIMENTAL(ntohl(ip->ip_src.s_addr)) || IN_ZERONET(ntohl(ip->ip_src.s_addr)) ) { m_freem(m); /* Bad return address */ ICMPSTAT_INC(icps_badaddr); goto done; /* Ip_output() will check for broadcast */ } t = ip->ip_dst; ip->ip_dst = ip->ip_src; /* * Source selection for ICMP replies: * * If the incoming packet was addressed directly to one of our * own addresses, use dst as the src for the reply. */ IN_IFADDR_RLOCK(&in_ifa_tracker); LIST_FOREACH(ia, INADDR_HASH(t.s_addr), ia_hash) { if (t.s_addr == IA_SIN(ia)->sin_addr.s_addr) { t = IA_SIN(ia)->sin_addr; IN_IFADDR_RUNLOCK(&in_ifa_tracker); goto match; } } IN_IFADDR_RUNLOCK(&in_ifa_tracker); /* * If the incoming packet was addressed to one of our broadcast * addresses, use the first non-broadcast address which corresponds * to the incoming interface. */ ifp = m->m_pkthdr.rcvif; if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = ifatoia(ifa); if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == t.s_addr) { t = IA_SIN(ia)->sin_addr; IF_ADDR_RUNLOCK(ifp); goto match; } } IF_ADDR_RUNLOCK(ifp); } /* * If the packet was transiting through us, use the address of * the interface the packet came through in. If that interface * doesn't have a suitable IP address, the normal selection * criteria apply. */ if (V_icmp_rfi && ifp != NULL) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = ifatoia(ifa); t = IA_SIN(ia)->sin_addr; IF_ADDR_RUNLOCK(ifp); goto match; } IF_ADDR_RUNLOCK(ifp); } /* * If the incoming packet was not addressed directly to us, use * designated interface for icmp replies specified by sysctl * net.inet.icmp.reply_src (default not set). Otherwise continue * with normal source selection. */ if (V_reply_src[0] != '\0' && (ifp = ifunit(V_reply_src))) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = ifatoia(ifa); t = IA_SIN(ia)->sin_addr; IF_ADDR_RUNLOCK(ifp); goto match; } IF_ADDR_RUNLOCK(ifp); } /* * If the packet was transiting through us, use the address of * the interface that is the closest to the packet source. * When we don't have a route back to the packet source, stop here * and drop the packet. */ if (fib4_lookup_nh_ext(M_GETFIB(m), ip->ip_dst, 0, 0, &nh_ext) != 0) { m_freem(m); ICMPSTAT_INC(icps_noroute); goto done; } t = nh_ext.nh_src; match: #ifdef MAC mac_netinet_icmp_replyinplace(m); #endif ip->ip_src = t; ip->ip_ttl = V_ip_defttl; if (optlen > 0) { u_char *cp; int opt, cnt; u_int len; /* * Retrieve any source routing from the incoming packet; * add on any record-route or timestamp options. */ cp = (u_char *) (ip + 1); if ((opts = ip_srcroute(m)) == NULL && (opts = m_gethdr(M_NOWAIT, MT_DATA))) { opts->m_len = sizeof(struct in_addr); mtod(opts, struct in_addr *)->s_addr = 0; } if (opts) { #ifdef ICMPPRINTFS if (icmpprintfs) printf("icmp_reflect optlen %d rt %d => ", optlen, opts->m_len); #endif for (cnt = optlen; cnt > 0; cnt -= len, cp += len) { opt = cp[IPOPT_OPTVAL]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) len = 1; else { if (cnt < IPOPT_OLEN + sizeof(*cp)) break; len = cp[IPOPT_OLEN]; if (len < IPOPT_OLEN + sizeof(*cp) || len > cnt) break; } /* * Should check for overflow, but it "can't happen" */ if (opt == IPOPT_RR || opt == IPOPT_TS || opt == IPOPT_SECURITY) { bcopy((caddr_t)cp, mtod(opts, caddr_t) + opts->m_len, len); opts->m_len += len; } } /* Terminate & pad, if necessary */ cnt = opts->m_len % 4; if (cnt) { for (; cnt < 4; cnt++) { *(mtod(opts, caddr_t) + opts->m_len) = IPOPT_EOL; opts->m_len++; } } #ifdef ICMPPRINTFS if (icmpprintfs) printf("%d\n", opts->m_len); #endif } ip_stripoptions(m); } m_tag_delete_nonpersistent(m); m->m_flags &= ~(M_BCAST|M_MCAST); icmp_send(m, opts); done: if (opts) (void)m_free(opts); } /* * Send an icmp packet back to the ip level, * after supplying a checksum. */ static void icmp_send(struct mbuf *m, struct mbuf *opts) { struct ip *ip = mtod(m, struct ip *); int hlen; struct icmp *icp; hlen = ip->ip_hl << 2; m->m_data += hlen; m->m_len -= hlen; icp = mtod(m, struct icmp *); icp->icmp_cksum = 0; icp->icmp_cksum = in_cksum(m, ntohs(ip->ip_len) - hlen); m->m_data -= hlen; m->m_len += hlen; m->m_pkthdr.rcvif = (struct ifnet *)0; #ifdef ICMPPRINTFS if (icmpprintfs) { char dstbuf[INET_ADDRSTRLEN]; char srcbuf[INET_ADDRSTRLEN]; printf("icmp_send dst %s src %s\n", inet_ntoa_r(ip->ip_dst, dstbuf), inet_ntoa_r(ip->ip_src, srcbuf)); } #endif (void) ip_output(m, opts, NULL, 0, NULL, NULL); } /* * Return milliseconds since 00:00 UTC in network format. */ uint32_t iptime(void) { struct timeval atv; u_long t; getmicrotime(&atv); t = (atv.tv_sec % (24*60*60)) * 1000 + atv.tv_usec / 1000; return (htonl(t)); } /* * Return the next larger or smaller MTU plateau (table from RFC 1191) * given current value MTU. If DIR is less than zero, a larger plateau * is returned; otherwise, a smaller value is returned. */ int ip_next_mtu(int mtu, int dir) { static int mtutab[] = { 65535, 32000, 17914, 8166, 4352, 2002, 1492, 1280, 1006, 508, 296, 68, 0 }; int i, size; size = (sizeof mtutab) / (sizeof mtutab[0]); if (dir >= 0) { for (i = 0; i < size; i++) if (mtu > mtutab[i]) return mtutab[i]; } else { for (i = size - 1; i >= 0; i--) if (mtu < mtutab[i]) return mtutab[i]; if (mtu == mtutab[0]) return mtutab[0]; } return 0; } #endif /* INET */ /* * badport_bandlim() - check for ICMP bandwidth limit * * Return 0 if it is ok to send an ICMP error response, -1 if we have * hit our bandwidth limit and it is not ok. * * If icmplim is <= 0, the feature is disabled and 0 is returned. * * For now we separate the TCP and UDP subsystems w/ different 'which' * values. We may eventually remove this separation (and simplify the * code further). * * Note that the printing of the error message is delayed so we can * properly print the icmp error rate that the system was trying to do * (i.e. 22000/100 pps, etc...). This can cause long delays in printing * the 'final' error, but it doesn't make sense to solve the printing * delay with more complex code. */ struct icmp_rate { const char *descr; struct counter_rate cr; }; -static VNET_DEFINE(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = { +VNET_DEFINE_STATIC(struct icmp_rate, icmp_rates[BANDLIM_MAX]) = { { "icmp unreach response" }, { "icmp ping response" }, { "icmp tstamp response" }, { "closed port RST response" }, { "open port RST response" }, { "icmp6 unreach response" }, { "sctp ootb response" } }; #define V_icmp_rates VNET(icmp_rates) static void icmp_bandlimit_init(void) { for (int i = 0; i < BANDLIM_MAX; i++) { V_icmp_rates[i].cr.cr_rate = counter_u64_alloc(M_WAITOK); V_icmp_rates[i].cr.cr_ticks = ticks; } } VNET_SYSINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, icmp_bandlimit_init, NULL); static void icmp_bandlimit_uninit(void) { for (int i = 0; i < BANDLIM_MAX; i++) counter_u64_free(V_icmp_rates[i].cr.cr_rate); } VNET_SYSUNINIT(icmp_bandlimit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, icmp_bandlimit_uninit, NULL); int badport_bandlim(int which) { int64_t pps; if (V_icmplim == 0 || which == BANDLIM_UNLIMITED) return (0); KASSERT(which >= 0 && which < BANDLIM_MAX, ("%s: which %d", __func__, which)); pps = counter_ratecheck(&V_icmp_rates[which].cr, V_icmplim); if (pps == -1) return (-1); if (pps > 0 && V_icmplim_output) log(LOG_NOTICE, "Limiting %s from %jd to %d packets/sec\n", V_icmp_rates[which].descr, (intmax_t )pps, V_icmplim); return (0); } Index: head/sys/netinet/ip_id.c =================================================================== --- head/sys/netinet/ip_id.c (revision 336675) +++ head/sys/netinet/ip_id.c (revision 336676) @@ -1,300 +1,300 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2008 Michael J. Silbersack. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * IP ID generation is a fascinating topic. * * In order to avoid ID collisions during packet reassembly, common sense * dictates that the period between reuse of IDs be as large as possible. * This leads to the classic implementation of a system-wide counter, thereby * ensuring that IDs repeat only once every 2^16 packets. * * Subsequent security researchers have pointed out that using a global * counter makes ID values predictable. This predictability allows traffic * analysis, idle scanning, and even packet injection in specific cases. * These results suggest that IP IDs should be as random as possible. * * The "searchable queues" algorithm used in this IP ID implementation was * proposed by Amit Klein. It is a compromise between the above two * viewpoints that has provable behavior that can be tuned to the user's * requirements. * * The basic concept is that we supplement a standard random number generator * with a queue of the last L IDs that we have handed out to ensure that all * IDs have a period of at least L. * * To efficiently implement this idea, we keep two data structures: a * circular array of IDs of size L and a bitstring of 65536 bits. * * To start, we ask the RNG for a new ID. A quick index into the bitstring * is used to determine if this is a recently used value. The process is * repeated until a value is returned that is not in the bitstring. * * Having found a usable ID, we remove the ID stored at the current position * in the queue from the bitstring and replace it with our new ID. Our new * ID is then added to the bitstring and the queue pointer is incremented. * * The lower limit of 512 was chosen because there doesn't seem to be much * point to having a smaller value. The upper limit of 32768 was chosen for * two reasons. First, every step above 32768 decreases the entropy. Taken * to an extreme, 65533 would offer 1 bit of entropy. Second, the number of * attempts it takes the algorithm to find an unused ID drastically * increases, killing performance. The default value of 8192 was chosen * because it provides a good tradeoff between randomness and non-repetition. * * With L=8192, the queue will use 16K of memory. The bitstring always * uses 8K of memory. No memory is allocated until the use of random ids is * enabled. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * By default we generate IP ID only for non-atomic datagrams, as * suggested by RFC6864. We use per-CPU counter for that, or if * user wants to, we can turn on random ID generation. */ -static VNET_DEFINE(int, ip_rfc6864) = 1; -static VNET_DEFINE(int, ip_do_randomid) = 0; +VNET_DEFINE_STATIC(int, ip_rfc6864) = 1; +VNET_DEFINE_STATIC(int, ip_do_randomid) = 0; #define V_ip_rfc6864 VNET(ip_rfc6864) #define V_ip_do_randomid VNET(ip_do_randomid) /* * Random ID state engine. */ static MALLOC_DEFINE(M_IPID, "ipid", "randomized ip id state"); -static VNET_DEFINE(uint16_t *, id_array); -static VNET_DEFINE(bitstr_t *, id_bits); -static VNET_DEFINE(int, array_ptr); -static VNET_DEFINE(int, array_size); -static VNET_DEFINE(int, random_id_collisions); -static VNET_DEFINE(int, random_id_total); -static VNET_DEFINE(struct mtx, ip_id_mtx); +VNET_DEFINE_STATIC(uint16_t *, id_array); +VNET_DEFINE_STATIC(bitstr_t *, id_bits); +VNET_DEFINE_STATIC(int, array_ptr); +VNET_DEFINE_STATIC(int, array_size); +VNET_DEFINE_STATIC(int, random_id_collisions); +VNET_DEFINE_STATIC(int, random_id_total); +VNET_DEFINE_STATIC(struct mtx, ip_id_mtx); #define V_id_array VNET(id_array) #define V_id_bits VNET(id_bits) #define V_array_ptr VNET(array_ptr) #define V_array_size VNET(array_size) #define V_random_id_collisions VNET(random_id_collisions) #define V_random_id_total VNET(random_id_total) #define V_ip_id_mtx VNET(ip_id_mtx) /* * Non-random ID state engine is simply a per-cpu counter. */ -static VNET_DEFINE(counter_u64_t, ip_id); +VNET_DEFINE_STATIC(counter_u64_t, ip_id); #define V_ip_id VNET(ip_id) static int sysctl_ip_randomid(SYSCTL_HANDLER_ARGS); static int sysctl_ip_id_change(SYSCTL_HANDLER_ARGS); static void ip_initid(int); static uint16_t ip_randomid(void); static void ipid_sysinit(void); static void ipid_sysuninit(void); SYSCTL_DECL(_net_inet_ip); SYSCTL_PROC(_net_inet_ip, OID_AUTO, random_id, CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_do_randomid), 0, sysctl_ip_randomid, "IU", "Assign random ip_id values"); SYSCTL_INT(_net_inet_ip, OID_AUTO, rfc6864, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_rfc6864), 0, "Use constant IP ID for atomic datagrams"); SYSCTL_PROC(_net_inet_ip, OID_AUTO, random_id_period, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_VNET, &VNET_NAME(array_size), 0, sysctl_ip_id_change, "IU", "IP ID Array size"); SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id_collisions, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(random_id_collisions), 0, "Count of IP ID collisions"); SYSCTL_INT(_net_inet_ip, OID_AUTO, random_id_total, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(random_id_total), 0, "Count of IP IDs created"); static int sysctl_ip_randomid(SYSCTL_HANDLER_ARGS) { int error, new; new = V_ip_do_randomid; error = sysctl_handle_int(oidp, &new, 0, req); if (error || req->newptr == NULL) return (error); if (new != 0 && new != 1) return (EINVAL); if (new == V_ip_do_randomid) return (0); if (new == 1 && V_ip_do_randomid == 0) ip_initid(8192); /* We don't free memory when turning random ID off, due to race. */ V_ip_do_randomid = new; return (0); } static int sysctl_ip_id_change(SYSCTL_HANDLER_ARGS) { int error, new; new = V_array_size; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new >= 512 && new <= 32768) ip_initid(new); else error = EINVAL; } return (error); } static void ip_initid(int new_size) { uint16_t *new_array; bitstr_t *new_bits; new_array = malloc(new_size * sizeof(uint16_t), M_IPID, M_WAITOK | M_ZERO); new_bits = malloc(bitstr_size(65536), M_IPID, M_WAITOK | M_ZERO); mtx_lock(&V_ip_id_mtx); if (V_id_array != NULL) { free(V_id_array, M_IPID); free(V_id_bits, M_IPID); } V_id_array = new_array; V_id_bits = new_bits; V_array_size = new_size; V_array_ptr = 0; V_random_id_collisions = 0; V_random_id_total = 0; mtx_unlock(&V_ip_id_mtx); } static uint16_t ip_randomid(void) { uint16_t new_id; mtx_lock(&V_ip_id_mtx); /* * To avoid a conflict with the zeros that the array is initially * filled with, we never hand out an id of zero. */ new_id = 0; do { if (new_id != 0) V_random_id_collisions++; arc4rand(&new_id, sizeof(new_id), 0); } while (bit_test(V_id_bits, new_id) || new_id == 0); bit_clear(V_id_bits, V_id_array[V_array_ptr]); bit_set(V_id_bits, new_id); V_id_array[V_array_ptr] = new_id; V_array_ptr++; if (V_array_ptr == V_array_size) V_array_ptr = 0; V_random_id_total++; mtx_unlock(&V_ip_id_mtx); return (new_id); } void ip_fillid(struct ip *ip) { /* * Per RFC6864 Section 4 * * o Atomic datagrams: (DF==1) && (MF==0) && (frag_offset==0) * o Non-atomic datagrams: (DF==0) || (MF==1) || (frag_offset>0) */ if (V_ip_rfc6864 && (ip->ip_off & htons(IP_DF)) == htons(IP_DF)) ip->ip_id = 0; else if (V_ip_do_randomid) ip->ip_id = ip_randomid(); else { counter_u64_add(V_ip_id, 1); /* * There are two issues about this trick, to be kept in mind. * 1) We can migrate between counter_u64_add() and next * line, and grab counter from other CPU, resulting in too * quick ID reuse. This is tolerable in our particular case, * since probability of such event is much lower then reuse * of ID due to legitimate overflow, that at modern Internet * speeds happens all the time. * 2) We are relying on the fact that counter(9) is based on * UMA_ZONE_PCPU uma(9) zone. We also take only last * sixteen bits of a counter, so we don't care about the * fact that machines with 32-bit word update their counters * not atomically. */ ip->ip_id = htons((*(uint64_t *)zpcpu_get(V_ip_id)) & 0xffff); } } static void ipid_sysinit(void) { int i; mtx_init(&V_ip_id_mtx, "ip_id_mtx", NULL, MTX_DEF); V_ip_id = counter_u64_alloc(M_WAITOK); CPU_FOREACH(i) arc4rand(zpcpu_get_cpu(V_ip_id, i), sizeof(uint64_t), 0); } VNET_SYSINIT(ip_id, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipid_sysinit, NULL); static void ipid_sysuninit(void) { if (V_id_array != NULL) { free(V_id_array, M_IPID); free(V_id_bits, M_IPID); } counter_u64_free(V_ip_id); mtx_destroy(&V_ip_id_mtx); } VNET_SYSUNINIT(ip_id, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, ipid_sysuninit, NULL); Index: head/sys/netinet/ip_input.c =================================================================== --- head/sys/netinet/ip_input.c (revision 336675) +++ head/sys/netinet/ip_input.c (revision 336676) @@ -1,1427 +1,1427 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_bootp.h" #include "opt_ipstealth.h" #include "opt_ipsec.h" #include "opt_route.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef CTASSERT CTASSERT(sizeof(struct ip) == 20); #endif /* IP reassembly functions are defined in ip_reass.c. */ extern void ipreass_init(void); extern void ipreass_drain(void); extern void ipreass_slowtimo(void); #ifdef VIMAGE extern void ipreass_destroy(void); #endif struct rmlock in_ifaddr_lock; RM_SYSINIT(in_ifaddr_lock, &in_ifaddr_lock, "in_ifaddr_lock"); VNET_DEFINE(int, rsvp_on); VNET_DEFINE(int, ipforwarding); SYSCTL_INT(_net_inet_ip, IPCTL_FORWARDING, forwarding, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipforwarding), 0, "Enable IP forwarding between interfaces"); -static VNET_DEFINE(int, ipsendredirects) = 1; /* XXX */ +VNET_DEFINE_STATIC(int, ipsendredirects) = 1; /* XXX */ #define V_ipsendredirects VNET(ipsendredirects) SYSCTL_INT(_net_inet_ip, IPCTL_SENDREDIRECTS, redirect, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsendredirects), 0, "Enable sending IP redirects"); /* * XXX - Setting ip_checkinterface mostly implements the receive side of * the Strong ES model described in RFC 1122, but since the routing table * and transmit implementation do not implement the Strong ES model, * setting this to 1 results in an odd hybrid. * * XXX - ip_checkinterface currently must be disabled if you use ipnat * to translate the destination address to another local interface. * * XXX - ip_checkinterface must be disabled if you add IP aliases * to the loopback interface instead of the interface where the * packets for those addresses are received. */ -static VNET_DEFINE(int, ip_checkinterface); +VNET_DEFINE_STATIC(int, ip_checkinterface); #define V_ip_checkinterface VNET(ip_checkinterface) SYSCTL_INT(_net_inet_ip, OID_AUTO, check_interface, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_checkinterface), 0, "Verify packet arrives on correct interface"); VNET_DEFINE(struct pfil_head, inet_pfil_hook); /* Packet filter hooks */ static struct netisr_handler ip_nh = { .nh_name = "ip", .nh_handler = ip_input, .nh_proto = NETISR_IP, #ifdef RSS .nh_m2cpuid = rss_soft_m2cpuid_v4, .nh_policy = NETISR_POLICY_CPU, .nh_dispatch = NETISR_DISPATCH_HYBRID, #else .nh_policy = NETISR_POLICY_FLOW, #endif }; #ifdef RSS /* * Directly dispatched frames are currently assumed * to have a flowid already calculated. * * It should likely have something that assert it * actually has valid flow details. */ static struct netisr_handler ip_direct_nh = { .nh_name = "ip_direct", .nh_handler = ip_direct_input, .nh_proto = NETISR_IP_DIRECT, .nh_m2cpuid = rss_soft_m2cpuid_v4, .nh_policy = NETISR_POLICY_CPU, .nh_dispatch = NETISR_DISPATCH_HYBRID, }; #endif extern struct domain inetdomain; extern struct protosw inetsw[]; u_char ip_protox[IPPROTO_MAX]; VNET_DEFINE(struct in_ifaddrhead, in_ifaddrhead); /* first inet address */ VNET_DEFINE(struct in_ifaddrhashhead *, in_ifaddrhashtbl); /* inet addr hash table */ VNET_DEFINE(u_long, in_ifaddrhmask); /* mask for hash table */ #ifdef IPCTL_DEFMTU SYSCTL_INT(_net_inet_ip, IPCTL_DEFMTU, mtu, CTLFLAG_RW, &ip_mtu, 0, "Default MTU"); #endif #ifdef IPSTEALTH VNET_DEFINE(int, ipstealth); SYSCTL_INT(_net_inet_ip, OID_AUTO, stealth, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipstealth), 0, "IP stealth mode, no TTL decrementation on forwarding"); #endif /* * IP statistics are stored in the "array" of counter(9)s. */ VNET_PCPUSTAT_DEFINE(struct ipstat, ipstat); VNET_PCPUSTAT_SYSINIT(ipstat); SYSCTL_VNET_PCPUSTAT(_net_inet_ip, IPCTL_STATS, stats, struct ipstat, ipstat, "IP statistics (struct ipstat, netinet/ip_var.h)"); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(ipstat); #endif /* VIMAGE */ /* * Kernel module interface for updating ipstat. The argument is an index * into ipstat treated as an array. */ void kmod_ipstat_inc(int statnum) { counter_u64_add(VNET(ipstat)[statnum], 1); } void kmod_ipstat_dec(int statnum) { counter_u64_add(VNET(ipstat)[statnum], -1); } static int sysctl_netinet_intr_queue_maxlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&ip_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&ip_nh, qlimit)); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQMAXLEN, intr_queue_maxlen, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_queue_maxlen, "I", "Maximum size of the IP input queue"); static int sysctl_netinet_intr_queue_drops(SYSCTL_HANDLER_ARGS) { u_int64_t qdrops_long; int error, qdrops; netisr_getqdrops(&ip_nh, &qdrops_long); qdrops = qdrops_long; error = sysctl_handle_int(oidp, &qdrops, 0, req); if (error || !req->newptr) return (error); if (qdrops != 0) return (EINVAL); netisr_clearqdrops(&ip_nh); return (0); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRQDROPS, intr_queue_drops, CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_queue_drops, "I", "Number of packets dropped from the IP input queue"); #ifdef RSS static int sysctl_netinet_intr_direct_queue_maxlen(SYSCTL_HANDLER_ARGS) { int error, qlimit; netisr_getqlimit(&ip_direct_nh, &qlimit); error = sysctl_handle_int(oidp, &qlimit, 0, req); if (error || !req->newptr) return (error); if (qlimit < 1) return (EINVAL); return (netisr_setqlimit(&ip_direct_nh, qlimit)); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRDQMAXLEN, intr_direct_queue_maxlen, CTLTYPE_INT|CTLFLAG_RW, 0, 0, sysctl_netinet_intr_direct_queue_maxlen, "I", "Maximum size of the IP direct input queue"); static int sysctl_netinet_intr_direct_queue_drops(SYSCTL_HANDLER_ARGS) { u_int64_t qdrops_long; int error, qdrops; netisr_getqdrops(&ip_direct_nh, &qdrops_long); qdrops = qdrops_long; error = sysctl_handle_int(oidp, &qdrops, 0, req); if (error || !req->newptr) return (error); if (qdrops != 0) return (EINVAL); netisr_clearqdrops(&ip_direct_nh); return (0); } SYSCTL_PROC(_net_inet_ip, IPCTL_INTRDQDROPS, intr_direct_queue_drops, CTLTYPE_INT|CTLFLAG_RD, 0, 0, sysctl_netinet_intr_direct_queue_drops, "I", "Number of packets dropped from the IP direct input queue"); #endif /* RSS */ /* * IP initialization: fill in IP protocol switch table. * All protocols not implemented in kernel go to raw IP protocol handler. */ void ip_init(void) { struct protosw *pr; int i; CK_STAILQ_INIT(&V_in_ifaddrhead); V_in_ifaddrhashtbl = hashinit(INADDR_NHASH, M_IFADDR, &V_in_ifaddrhmask); /* Initialize IP reassembly queue. */ ipreass_init(); /* Initialize packet filter hooks. */ V_inet_pfil_hook.ph_type = PFIL_TYPE_AF; V_inet_pfil_hook.ph_af = AF_INET; if ((i = pfil_head_register(&V_inet_pfil_hook)) != 0) printf("%s: WARNING: unable to register pfil hook, " "error %d\n", __func__, i); if (hhook_head_register(HHOOK_TYPE_IPSEC_IN, AF_INET, &V_ipsec_hhh_in[HHOOK_IPSEC_INET], HHOOK_WAITOK | HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register input helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_IPSEC_OUT, AF_INET, &V_ipsec_hhh_out[HHOOK_IPSEC_INET], HHOOK_WAITOK | HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register output helper hook\n", __func__); /* Skip initialization of globals for non-default instances. */ #ifdef VIMAGE if (!IS_DEFAULT_VNET(curvnet)) { netisr_register_vnet(&ip_nh); #ifdef RSS netisr_register_vnet(&ip_direct_nh); #endif return; } #endif pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) panic("ip_init: PF_INET not found"); /* Initialize the entire ip_protox[] array to IPPROTO_RAW. */ for (i = 0; i < IPPROTO_MAX; i++) ip_protox[i] = pr - inetsw; /* * Cycle through IP protocols and put them into the appropriate place * in ip_protox[]. */ for (pr = inetdomain.dom_protosw; pr < inetdomain.dom_protoswNPROTOSW; pr++) if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol && pr->pr_protocol != IPPROTO_RAW) { /* Be careful to only index valid IP protocols. */ if (pr->pr_protocol < IPPROTO_MAX) ip_protox[pr->pr_protocol] = pr - inetsw; } netisr_register(&ip_nh); #ifdef RSS netisr_register(&ip_direct_nh); #endif } #ifdef VIMAGE static void ip_destroy(void *unused __unused) { struct ifnet *ifp; int error; #ifdef RSS netisr_unregister_vnet(&ip_direct_nh); #endif netisr_unregister_vnet(&ip_nh); if ((error = pfil_head_unregister(&V_inet_pfil_hook)) != 0) printf("%s: WARNING: unable to unregister pfil hook, " "error %d\n", __func__, error); error = hhook_head_deregister(V_ipsec_hhh_in[HHOOK_IPSEC_INET]); if (error != 0) { printf("%s: WARNING: unable to deregister input helper hook " "type HHOOK_TYPE_IPSEC_IN, id HHOOK_IPSEC_INET: " "error %d returned\n", __func__, error); } error = hhook_head_deregister(V_ipsec_hhh_out[HHOOK_IPSEC_INET]); if (error != 0) { printf("%s: WARNING: unable to deregister output helper hook " "type HHOOK_TYPE_IPSEC_OUT, id HHOOK_IPSEC_INET: " "error %d returned\n", __func__, error); } /* Remove the IPv4 addresses from all interfaces. */ in_ifscrub_all(); /* Make sure the IPv4 routes are gone as well. */ IFNET_RLOCK(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) rt_flushifroutes_af(ifp, AF_INET); IFNET_RUNLOCK(); /* Destroy IP reassembly queue. */ ipreass_destroy(); /* Cleanup in_ifaddr hash table; should be empty. */ hashdestroy(V_in_ifaddrhashtbl, M_IFADDR, V_in_ifaddrhmask); } VNET_SYSUNINIT(ip, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, ip_destroy, NULL); #endif #ifdef RSS /* * IP direct input routine. * * This is called when reinjecting completed fragments where * all of the previous checking and book-keeping has been done. */ void ip_direct_input(struct mbuf *m) { struct ip *ip; int hlen; ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if (IPSEC_INPUT(ipv4, m, hlen, ip->ip_p) != 0) return; } #endif /* IPSEC */ IPSTAT_INC(ips_delivered); (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); return; } #endif /* * Ip input routine. Checksum and byte swap header. If fragmented * try to reassemble. Process options. Pass to next level. */ void ip_input(struct mbuf *m) { struct ip *ip = NULL; struct in_ifaddr *ia = NULL; struct ifaddr *ifa; struct ifnet *ifp; int checkif, hlen = 0; uint16_t sum, ip_len; int dchg = 0; /* dest changed after fw */ struct in_addr odst; /* original dst address */ M_ASSERTPKTHDR(m); if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; /* Set up some basics that will be used later. */ ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; ip_len = ntohs(ip->ip_len); goto ours; } IPSTAT_INC(ips_total); if (m->m_pkthdr.len < sizeof(struct ip)) goto tooshort; if (m->m_len < sizeof (struct ip) && (m = m_pullup(m, sizeof (struct ip))) == NULL) { IPSTAT_INC(ips_toosmall); return; } ip = mtod(m, struct ip *); if (ip->ip_v != IPVERSION) { IPSTAT_INC(ips_badvers); goto bad; } hlen = ip->ip_hl << 2; if (hlen < sizeof(struct ip)) { /* minimum header length */ IPSTAT_INC(ips_badhlen); goto bad; } if (hlen > m->m_len) { if ((m = m_pullup(m, hlen)) == NULL) { IPSTAT_INC(ips_badhlen); return; } ip = mtod(m, struct ip *); } IP_PROBE(receive, NULL, NULL, ip, m->m_pkthdr.rcvif, ip, NULL); /* 127/8 must not appear on wire - RFC1122 */ ifp = m->m_pkthdr.rcvif; if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { if ((ifp->if_flags & IFF_LOOPBACK) == 0) { IPSTAT_INC(ips_badaddr); goto bad; } } if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) { sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); } else { if (hlen == sizeof(struct ip)) { sum = in_cksum_hdr(ip); } else { sum = in_cksum(m, hlen); } } if (sum) { IPSTAT_INC(ips_badsum); goto bad; } #ifdef ALTQ if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) /* packet is dropped by traffic conditioner */ return; #endif ip_len = ntohs(ip->ip_len); if (ip_len < hlen) { IPSTAT_INC(ips_badlen); goto bad; } /* * Check that the amount of data in the buffers * is as at least much as the IP header would have us expect. * Trim mbufs if longer than we expect. * Drop packet if shorter than we expect. */ if (m->m_pkthdr.len < ip_len) { tooshort: IPSTAT_INC(ips_tooshort); goto bad; } if (m->m_pkthdr.len > ip_len) { if (m->m_len == m->m_pkthdr.len) { m->m_len = ip_len; m->m_pkthdr.len = ip_len; } else m_adj(m, ip_len - m->m_pkthdr.len); } /* * Try to forward the packet, but if we fail continue. * ip_tryforward() does inbound and outbound packet firewall * processing. If firewall has decided that destination becomes * our local address, it sets M_FASTFWD_OURS flag. In this * case skip another inbound firewall processing and update * ip pointer. */ if (V_ipforwarding != 0 #if defined(IPSEC) || defined(IPSEC_SUPPORT) && (!IPSEC_ENABLED(ipv4) || IPSEC_CAPS(ipv4, m, IPSEC_CAP_OPERABLE) == 0) #endif ) { if ((m = ip_tryforward(m)) == NULL) return; if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; ip = mtod(m, struct ip *); goto ours; } } #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * Bypass packet filtering for packets previously handled by IPsec. */ if (IPSEC_ENABLED(ipv4) && IPSEC_CAPS(ipv4, m, IPSEC_CAP_BYPASS_FILTER) != 0) goto passin; #endif /* * Run through list of hooks for input packets. * * NB: Beware of the destination address changing (e.g. * by NAT rewriting). When this happens, tell * ip_forward to do the right thing. */ /* Jump over all PFIL processing if hooks are not active. */ if (!PFIL_HOOKED(&V_inet_pfil_hook)) goto passin; odst = ip->ip_dst; if (pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_IN, 0, NULL) != 0) return; if (m == NULL) /* consumed by filter */ return; ip = mtod(m, struct ip *); dchg = (odst.s_addr != ip->ip_dst.s_addr); ifp = m->m_pkthdr.rcvif; if (m->m_flags & M_FASTFWD_OURS) { m->m_flags &= ~M_FASTFWD_OURS; goto ours; } if (m->m_flags & M_IP_NEXTHOP) { if (m_tag_find(m, PACKET_TAG_IPFORWARD, NULL) != NULL) { /* * Directly ship the packet on. This allows * forwarding packets originally destined to us * to some other directly connected host. */ ip_forward(m, 1); return; } } passin: /* * Process options and, if not destined for us, * ship it on. ip_dooptions returns 1 when an * error was detected (causing an icmp message * to be sent and the original packet to be freed). */ if (hlen > sizeof (struct ip) && ip_dooptions(m, 0)) return; /* greedy RSVP, snatches any PATH packet of the RSVP protocol and no * matter if it is destined to another node, or whether it is * a multicast one, RSVP wants it! and prevents it from being forwarded * anywhere else. Also checks if the rsvp daemon is running before * grabbing the packet. */ if (V_rsvp_on && ip->ip_p==IPPROTO_RSVP) goto ours; /* * Check our list of addresses, to see if the packet is for us. * If we don't have any addresses, assume any unicast packet * we receive might be for us (and let the upper layers deal * with it). */ if (CK_STAILQ_EMPTY(&V_in_ifaddrhead) && (m->m_flags & (M_MCAST|M_BCAST)) == 0) goto ours; /* * Enable a consistency check between the destination address * and the arrival interface for a unicast packet (the RFC 1122 * strong ES model) if IP forwarding is disabled and the packet * is not locally generated and the packet is not subject to * 'ipfw fwd'. * * XXX - Checking also should be disabled if the destination * address is ipnat'ed to a different interface. * * XXX - Checking is incompatible with IP aliases added * to the loopback interface instead of the interface where * the packets are received. * * XXX - This is the case for carp vhost IPs as well so we * insert a workaround. If the packet got here, we already * checked with carp_iamatch() and carp_forus(). */ checkif = V_ip_checkinterface && (V_ipforwarding == 0) && ifp != NULL && ((ifp->if_flags & IFF_LOOPBACK) == 0) && ifp->if_carp == NULL && (dchg == 0); /* * Check for exact addresses in the hash bucket. */ /* IN_IFADDR_RLOCK(); */ LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { /* * If the address matches, verify that the packet * arrived via the correct interface if checking is * enabled. */ if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr && (!checkif || ia->ia_ifp == ifp)) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); /* IN_IFADDR_RUNLOCK(); */ goto ours; } } /* IN_IFADDR_RUNLOCK(); */ /* * Check for broadcast addresses. * * Only accept broadcast packets that arrive via the matching * interface. Reception of forwarded directed broadcasts would * be handled via ip_forward() and ether_output() with the loopback * into the stack for SIMPLEX interfaces handled by ether_output(). */ if (ifp != NULL && ifp->if_flags & IFF_BROADCAST) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET) continue; ia = ifatoia(ifa); if (satosin(&ia->ia_broadaddr)->sin_addr.s_addr == ip->ip_dst.s_addr) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); IF_ADDR_RUNLOCK(ifp); goto ours; } #ifdef BOOTP_COMPAT if (IA_SIN(ia)->sin_addr.s_addr == INADDR_ANY) { counter_u64_add(ia->ia_ifa.ifa_ipackets, 1); counter_u64_add(ia->ia_ifa.ifa_ibytes, m->m_pkthdr.len); IF_ADDR_RUNLOCK(ifp); goto ours; } #endif } IF_ADDR_RUNLOCK(ifp); ia = NULL; } /* RFC 3927 2.7: Do not forward datagrams for 169.254.0.0/16. */ if (IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr))) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { if (V_ip_mrouter) { /* * If we are acting as a multicast router, all * incoming multicast packets are passed to the * kernel-level multicast forwarding function. * The packet is returned (relatively) intact; if * ip_mforward() returns a non-zero value, the packet * must be discarded, else it may be accepted below. */ if (ip_mforward && ip_mforward(ip, ifp, m, 0) != 0) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } /* * The process-level routing daemon needs to receive * all multicast IGMP packets, whether or not this * host belongs to their destination groups. */ if (ip->ip_p == IPPROTO_IGMP) goto ours; IPSTAT_INC(ips_forward); } /* * Assume the packet is for us, to avoid prematurely taking * a lock on the in_multi hash. Protocols must perform * their own filtering and update statistics accordingly. */ goto ours; } if (ip->ip_dst.s_addr == (u_long)INADDR_BROADCAST) goto ours; if (ip->ip_dst.s_addr == INADDR_ANY) goto ours; /* * Not for us; forward if possible and desirable. */ if (V_ipforwarding == 0) { IPSTAT_INC(ips_cantforward); m_freem(m); } else { ip_forward(m, dchg); } return; ours: #ifdef IPSTEALTH /* * IPSTEALTH: Process non-routing options only * if the packet is destined for us. */ if (V_ipstealth && hlen > sizeof (struct ip) && ip_dooptions(m, 1)) return; #endif /* IPSTEALTH */ /* * Attempt reassembly; if it succeeds, proceed. * ip_reass() will return a different mbuf. */ if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) { /* XXXGL: shouldn't we save & set m_flags? */ m = ip_reass(m); if (m == NULL) return; ip = mtod(m, struct ip *); /* Get the header length of the reassembled packet */ hlen = ip->ip_hl << 2; } #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if (IPSEC_INPUT(ipv4, m, hlen, ip->ip_p) != 0) return; } #endif /* IPSEC */ /* * Switch out to protocol's input routine. */ IPSTAT_INC(ips_delivered); (*inetsw[ip_protox[ip->ip_p]].pr_input)(&m, &hlen, ip->ip_p); return; bad: m_freem(m); } /* * IP timer processing; * if a timer expires on a reassembly * queue, discard it. */ void ip_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); ipreass_slowtimo(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } void ip_drain(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); ipreass_drain(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * The protocol to be inserted into ip_protox[] must be already registered * in inetsw[], either statically or through pf_proto_register(). */ int ipproto_register(short ipproto) { struct protosw *pr; /* Sanity checks. */ if (ipproto <= 0 || ipproto >= IPPROTO_MAX) return (EPROTONOSUPPORT); /* * The protocol slot must not be occupied by another protocol * already. An index pointing to IPPROTO_RAW is unused. */ pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) return (EPFNOSUPPORT); if (ip_protox[ipproto] != pr - inetsw) /* IPPROTO_RAW */ return (EEXIST); /* Find the protocol position in inetsw[] and set the index. */ for (pr = inetdomain.dom_protosw; pr < inetdomain.dom_protoswNPROTOSW; pr++) { if (pr->pr_domain->dom_family == PF_INET && pr->pr_protocol && pr->pr_protocol == ipproto) { ip_protox[pr->pr_protocol] = pr - inetsw; return (0); } } return (EPROTONOSUPPORT); } int ipproto_unregister(short ipproto) { struct protosw *pr; /* Sanity checks. */ if (ipproto <= 0 || ipproto >= IPPROTO_MAX) return (EPROTONOSUPPORT); /* Check if the protocol was indeed registered. */ pr = pffindproto(PF_INET, IPPROTO_RAW, SOCK_RAW); if (pr == NULL) return (EPFNOSUPPORT); if (ip_protox[ipproto] == pr - inetsw) /* IPPROTO_RAW */ return (ENOENT); /* Reset the protocol slot to IPPROTO_RAW. */ ip_protox[ipproto] = pr - inetsw; return (0); } u_char inetctlerrmap[PRC_NCMDS] = { 0, 0, 0, 0, 0, EMSGSIZE, EHOSTDOWN, EHOSTUNREACH, EHOSTUNREACH, EHOSTUNREACH, ECONNREFUSED, ECONNREFUSED, EMSGSIZE, EHOSTUNREACH, 0, 0, 0, 0, EHOSTUNREACH, 0, ENOPROTOOPT, ECONNREFUSED }; /* * Forward a packet. If some error occurs return the sender * an icmp packet. Note we can't always generate a meaningful * icmp message because icmp doesn't have a large enough repertoire * of codes and types. * * If not forwarding, just drop the packet. This could be confusing * if ipforwarding was zero but some routing protocol was advancing * us as a gateway to somewhere. However, we must let the routing * protocol deal with that. * * The srcrt parameter indicates whether the packet is being forwarded * via a source route. */ void ip_forward(struct mbuf *m, int srcrt) { struct ip *ip = mtod(m, struct ip *); struct in_ifaddr *ia; struct mbuf *mcopy; struct sockaddr_in *sin; struct in_addr dest; struct route ro; int error, type = 0, code = 0, mtu = 0; if (m->m_flags & (M_BCAST|M_MCAST) || in_canforward(ip->ip_dst) == 0) { IPSTAT_INC(ips_cantforward); m_freem(m); return; } if ( #ifdef IPSTEALTH V_ipstealth == 0 && #endif ip->ip_ttl <= IPTTLDEC) { icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0); return; } bzero(&ro, sizeof(ro)); sin = (struct sockaddr_in *)&ro.ro_dst; sin->sin_family = AF_INET; sin->sin_len = sizeof(*sin); sin->sin_addr = ip->ip_dst; #ifdef RADIX_MPATH rtalloc_mpath_fib(&ro, ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), M_GETFIB(m)); #else in_rtalloc_ign(&ro, 0, M_GETFIB(m)); #endif NET_EPOCH_ENTER(); if (ro.ro_rt != NULL) { ia = ifatoia(ro.ro_rt->rt_ifa); } else ia = NULL; /* * Save the IP header and at most 8 bytes of the payload, * in case we need to generate an ICMP message to the src. * * XXX this can be optimized a lot by saving the data in a local * buffer on the stack (72 bytes at most), and only allocating the * mbuf if really necessary. The vast majority of the packets * are forwarded without having to send an ICMP back (either * because unnecessary, or because rate limited), so we are * really we are wasting a lot of work here. * * We don't use m_copym() because it might return a reference * to a shared cluster. Both this function and ip_output() * assume exclusive access to the IP header in `m', so any * data in a cluster may change before we reach icmp_error(). */ mcopy = m_gethdr(M_NOWAIT, m->m_type); if (mcopy != NULL && !m_dup_pkthdr(mcopy, m, M_NOWAIT)) { /* * It's probably ok if the pkthdr dup fails (because * the deep copy of the tag chain failed), but for now * be conservative and just discard the copy since * code below may some day want the tags. */ m_free(mcopy); mcopy = NULL; } if (mcopy != NULL) { mcopy->m_len = min(ntohs(ip->ip_len), M_TRAILINGSPACE(mcopy)); mcopy->m_pkthdr.len = mcopy->m_len; m_copydata(m, 0, mcopy->m_len, mtod(mcopy, caddr_t)); } #ifdef IPSTEALTH if (V_ipstealth == 0) #endif ip->ip_ttl -= IPTTLDEC; #if defined(IPSEC) || defined(IPSEC_SUPPORT) if (IPSEC_ENABLED(ipv4)) { if ((error = IPSEC_FORWARD(ipv4, m)) != 0) { /* mbuf consumed by IPsec */ m_freem(mcopy); if (error != EINPROGRESS) IPSTAT_INC(ips_cantforward); goto out; } /* No IPsec processing required */ } #endif /* IPSEC */ /* * If forwarding packet using same interface that it came in on, * perhaps should send a redirect to sender to shortcut a hop. * Only send redirect if source is sending directly to us, * and if packet was not source routed (or has any options). * Also, don't send redirect if forwarding using a default route * or a route modified by a redirect. */ dest.s_addr = 0; if (!srcrt && V_ipsendredirects && ia != NULL && ia->ia_ifp == m->m_pkthdr.rcvif) { struct rtentry *rt; rt = ro.ro_rt; if (rt && (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 && satosin(rt_key(rt))->sin_addr.s_addr != 0) { #define RTA(rt) ((struct in_ifaddr *)(rt->rt_ifa)) u_long src = ntohl(ip->ip_src.s_addr); if (RTA(rt) && (src & RTA(rt)->ia_subnetmask) == RTA(rt)->ia_subnet) { if (rt->rt_flags & RTF_GATEWAY) dest.s_addr = satosin(rt->rt_gateway)->sin_addr.s_addr; else dest.s_addr = ip->ip_dst.s_addr; /* Router requirements says to only send host redirects */ type = ICMP_REDIRECT; code = ICMP_REDIRECT_HOST; } } } error = ip_output(m, NULL, &ro, IP_FORWARDING, NULL, NULL); if (error == EMSGSIZE && ro.ro_rt) mtu = ro.ro_rt->rt_mtu; RO_RTFREE(&ro); if (error) IPSTAT_INC(ips_cantforward); else { IPSTAT_INC(ips_forward); if (type) IPSTAT_INC(ips_redirectsent); else { if (mcopy) m_freem(mcopy); goto out; } } if (mcopy == NULL) goto out; switch (error) { case 0: /* forwarded, but need redirect */ /* type, code set above */ break; case ENETUNREACH: case EHOSTUNREACH: case ENETDOWN: case EHOSTDOWN: default: type = ICMP_UNREACH; code = ICMP_UNREACH_HOST; break; case EMSGSIZE: type = ICMP_UNREACH; code = ICMP_UNREACH_NEEDFRAG; /* * If the MTU was set before make sure we are below the * interface MTU. * If the MTU wasn't set before use the interface mtu or * fall back to the next smaller mtu step compared to the * current packet size. */ if (mtu != 0) { if (ia != NULL) mtu = min(mtu, ia->ia_ifp->if_mtu); } else { if (ia != NULL) mtu = ia->ia_ifp->if_mtu; else mtu = ip_next_mtu(ntohs(ip->ip_len), 0); } IPSTAT_INC(ips_cantfrag); break; case ENOBUFS: case EACCES: /* ipfw denied packet */ m_freem(mcopy); goto out; } icmp_error(mcopy, type, code, dest.s_addr, mtu); out: NET_EPOCH_EXIT(); } #define CHECK_SO_CT(sp, ct) \ (((sp->so_options & SO_TIMESTAMP) && (sp->so_ts_clock == ct)) ? 1 : 0) void ip_savecontrol(struct inpcb *inp, struct mbuf **mp, struct ip *ip, struct mbuf *m) { bool stamped; stamped = false; if ((inp->inp_socket->so_options & SO_BINTIME) || CHECK_SO_CT(inp->inp_socket, SO_TS_BINTIME)) { struct bintime boottimebin, bt; struct timespec ts1; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts1); timespec2bintime(&ts1, &bt); getboottimebin(&boottimebin); bintime_add(&bt, &boottimebin); } else { bintime(&bt); } *mp = sbcreatecontrol((caddr_t)&bt, sizeof(bt), SCM_BINTIME, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } if (CHECK_SO_CT(inp->inp_socket, SO_TS_REALTIME_MICRO)) { struct bintime boottimebin, bt1; struct timespec ts1;; struct timeval tv; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts1); timespec2bintime(&ts1, &bt1); getboottimebin(&boottimebin); bintime_add(&bt1, &boottimebin); bintime2timeval(&bt1, &tv); } else { microtime(&tv); } *mp = sbcreatecontrol((caddr_t)&tv, sizeof(tv), SCM_TIMESTAMP, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } else if (CHECK_SO_CT(inp->inp_socket, SO_TS_REALTIME)) { struct bintime boottimebin; struct timespec ts, ts1; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { mbuf_tstmp2timespec(m, &ts); getboottimebin(&boottimebin); bintime2timespec(&boottimebin, &ts1); timespecadd(&ts, &ts1); } else { nanotime(&ts); } *mp = sbcreatecontrol((caddr_t)&ts, sizeof(ts), SCM_REALTIME, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } else if (CHECK_SO_CT(inp->inp_socket, SO_TS_MONOTONIC)) { struct timespec ts; if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) mbuf_tstmp2timespec(m, &ts); else nanouptime(&ts); *mp = sbcreatecontrol((caddr_t)&ts, sizeof(ts), SCM_MONOTONIC, SOL_SOCKET); if (*mp != NULL) { mp = &(*mp)->m_next; stamped = true; } } if (stamped && (m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) { struct sock_timestamp_info sti; bzero(&sti, sizeof(sti)); sti.st_info_flags = ST_INFO_HW; if ((m->m_flags & M_TSTMP_HPREC) != 0) sti.st_info_flags |= ST_INFO_HW_HPREC; *mp = sbcreatecontrol((caddr_t)&sti, sizeof(sti), SCM_TIME_INFO, SOL_SOCKET); if (*mp != NULL) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVDSTADDR) { *mp = sbcreatecontrol((caddr_t)&ip->ip_dst, sizeof(struct in_addr), IP_RECVDSTADDR, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVTTL) { *mp = sbcreatecontrol((caddr_t)&ip->ip_ttl, sizeof(u_char), IP_RECVTTL, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #ifdef notyet /* XXX * Moving these out of udp_input() made them even more broken * than they already were. */ /* options were tossed already */ if (inp->inp_flags & INP_RECVOPTS) { *mp = sbcreatecontrol((caddr_t)opts_deleted_above, sizeof(struct in_addr), IP_RECVOPTS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } /* ip_srcroute doesn't do what we want here, need to fix */ if (inp->inp_flags & INP_RECVRETOPTS) { *mp = sbcreatecontrol((caddr_t)ip_srcroute(m), sizeof(struct in_addr), IP_RECVRETOPTS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #endif if (inp->inp_flags & INP_RECVIF) { struct ifnet *ifp; struct sdlbuf { struct sockaddr_dl sdl; u_char pad[32]; } sdlbuf; struct sockaddr_dl *sdp; struct sockaddr_dl *sdl2 = &sdlbuf.sdl; if ((ifp = m->m_pkthdr.rcvif) && ifp->if_index && ifp->if_index <= V_if_index) { sdp = (struct sockaddr_dl *)ifp->if_addr->ifa_addr; /* * Change our mind and don't try copy. */ if (sdp->sdl_family != AF_LINK || sdp->sdl_len > sizeof(sdlbuf)) { goto makedummy; } bcopy(sdp, sdl2, sdp->sdl_len); } else { makedummy: sdl2->sdl_len = offsetof(struct sockaddr_dl, sdl_data[0]); sdl2->sdl_family = AF_LINK; sdl2->sdl_index = 0; sdl2->sdl_nlen = sdl2->sdl_alen = sdl2->sdl_slen = 0; } *mp = sbcreatecontrol((caddr_t)sdl2, sdl2->sdl_len, IP_RECVIF, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags & INP_RECVTOS) { *mp = sbcreatecontrol((caddr_t)&ip->ip_tos, sizeof(u_char), IP_RECVTOS, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } if (inp->inp_flags2 & INP_RECVFLOWID) { uint32_t flowid, flow_type; flowid = m->m_pkthdr.flowid; flow_type = M_HASHTYPE_GET(m); /* * XXX should handle the failure of one or the * other - don't populate both? */ *mp = sbcreatecontrol((caddr_t) &flowid, sizeof(uint32_t), IP_FLOWID, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; *mp = sbcreatecontrol((caddr_t) &flow_type, sizeof(uint32_t), IP_FLOWTYPE, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } #ifdef RSS if (inp->inp_flags2 & INP_RECVRSSBUCKETID) { uint32_t flowid, flow_type; uint32_t rss_bucketid; flowid = m->m_pkthdr.flowid; flow_type = M_HASHTYPE_GET(m); if (rss_hash2bucket(flowid, flow_type, &rss_bucketid) == 0) { *mp = sbcreatecontrol((caddr_t) &rss_bucketid, sizeof(uint32_t), IP_RSSBUCKETID, IPPROTO_IP); if (*mp) mp = &(*mp)->m_next; } } #endif } /* * XXXRW: Multicast routing code in ip_mroute.c is generally MPSAFE, but the * ip_rsvp and ip_rsvp_on variables need to be interlocked with rsvp_on * locking. This code remains in ip_input.c as ip_mroute.c is optionally * compiled. */ -static VNET_DEFINE(int, ip_rsvp_on); +VNET_DEFINE_STATIC(int, ip_rsvp_on); VNET_DEFINE(struct socket *, ip_rsvpd); #define V_ip_rsvp_on VNET(ip_rsvp_on) int ip_rsvp_init(struct socket *so) { if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_RSVP) return EOPNOTSUPP; if (V_ip_rsvpd != NULL) return EADDRINUSE; V_ip_rsvpd = so; /* * This may seem silly, but we need to be sure we don't over-increment * the RSVP counter, in case something slips up. */ if (!V_ip_rsvp_on) { V_ip_rsvp_on = 1; V_rsvp_on++; } return 0; } int ip_rsvp_done(void) { V_ip_rsvpd = NULL; /* * This may seem silly, but we need to be sure we don't over-decrement * the RSVP counter, in case something slips up. */ if (V_ip_rsvp_on) { V_ip_rsvp_on = 0; V_rsvp_on--; } return 0; } int rsvp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m; m = *mp; *mp = NULL; if (rsvp_input_p) { /* call the real one if loaded */ *mp = m; rsvp_input_p(mp, offp, proto); return (IPPROTO_DONE); } /* Can still get packets with rsvp_on = 0 if there is a local member * of the group to which the RSVP packet is addressed. But in this * case we want to throw the packet away. */ if (!V_rsvp_on) { m_freem(m); return (IPPROTO_DONE); } if (V_ip_rsvpd != NULL) { *mp = m; rip_input(mp, offp, proto); return (IPPROTO_DONE); } /* Drop the packet */ m_freem(m); return (IPPROTO_DONE); } Index: head/sys/netinet/ip_mroute.c =================================================================== --- head/sys/netinet/ip_mroute.c (revision 336675) +++ head/sys/netinet/ip_mroute.c (revision 336676) @@ -1,2940 +1,2940 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989 Stephen Deering * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_mroute.c 8.2 (Berkeley) 11/15/93 */ /* * IP multicast forwarding procedures * * Written by David Waitzman, BBN Labs, August 1988. * Modified by Steve Deering, Stanford, February 1989. * Modified by Mark J. Steiglitz, Stanford, May, 1991 * Modified by Van Jacobson, LBL, January 1993 * Modified by Ajit Thyagarajan, PARC, August 1993 * Modified by Bill Fenner, PARC, April 1995 * Modified by Ahmed Helmy, SGI, June 1996 * Modified by George Edmond Eddy (Rusty), ISI, February 1998 * Modified by Pavlin Radoslavov, USC/ISI, May 1998, August 1999, October 2000 * Modified by Hitoshi Asaeda, WIDE, August 2000 * Modified by Pavlin Radoslavov, ICSI, October 2002 * * MROUTING Revision: 3.5 * and PIM-SMv2 and PIM-DM support, advanced API support, * bandwidth metering and signaling */ /* * TODO: Prefix functions with ipmf_. * TODO: Maintain a refcount on if_allmulti() in ifnet or in the protocol * domain attachment (if_afdata) so we can track consumers of that service. * TODO: Deprecate routing socket path for SIOCGETSGCNT and SIOCGETVIFCNT, * move it to socket options. * TODO: Cleanup LSRR removal further. * TODO: Push RSVP stubs into raw_ip.c. * TODO: Use bitstring.h for vif set. * TODO: Fix mrt6_ioctl dangling ref when dynamically loaded. * TODO: Sync ip6_mroute.c with this file. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_mrouting.h" #define _PIM_VT 1 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef KTR_IPMF #define KTR_IPMF KTR_INET #endif #define VIFI_INVALID ((vifi_t) -1) -static VNET_DEFINE(uint32_t, last_tv_sec); /* last time we processed this */ +VNET_DEFINE_STATIC(uint32_t, last_tv_sec); /* last time we processed this */ #define V_last_tv_sec VNET(last_tv_sec) static MALLOC_DEFINE(M_MRTABLE, "mroutetbl", "multicast forwarding cache"); /* * Locking. We use two locks: one for the virtual interface table and * one for the forwarding table. These locks may be nested in which case * the VIF lock must always be taken first. Note that each lock is used * to cover not only the specific data structure but also related data * structures. */ static struct mtx mrouter_mtx; #define MROUTER_LOCK() mtx_lock(&mrouter_mtx) #define MROUTER_UNLOCK() mtx_unlock(&mrouter_mtx) #define MROUTER_LOCK_ASSERT() mtx_assert(&mrouter_mtx, MA_OWNED) #define MROUTER_LOCK_INIT() \ mtx_init(&mrouter_mtx, "IPv4 multicast forwarding", NULL, MTX_DEF) #define MROUTER_LOCK_DESTROY() mtx_destroy(&mrouter_mtx) static int ip_mrouter_cnt; /* # of vnets with active mrouters */ static int ip_mrouter_unloading; /* Allow no more V_ip_mrouter sockets */ -static VNET_PCPUSTAT_DEFINE(struct mrtstat, mrtstat); +VNET_PCPUSTAT_DEFINE_STATIC(struct mrtstat, mrtstat); VNET_PCPUSTAT_SYSINIT(mrtstat); VNET_PCPUSTAT_SYSUNINIT(mrtstat); SYSCTL_VNET_PCPUSTAT(_net_inet_ip, OID_AUTO, mrtstat, struct mrtstat, mrtstat, "IPv4 Multicast Forwarding Statistics (struct mrtstat, " "netinet/ip_mroute.h)"); -static VNET_DEFINE(u_long, mfchash); +VNET_DEFINE_STATIC(u_long, mfchash); #define V_mfchash VNET(mfchash) #define MFCHASH(a, g) \ ((((a).s_addr >> 20) ^ ((a).s_addr >> 10) ^ (a).s_addr ^ \ ((g).s_addr >> 20) ^ ((g).s_addr >> 10) ^ (g).s_addr) & V_mfchash) #define MFCHASHSIZE 256 static u_long mfchashsize; /* Hash size */ -static VNET_DEFINE(u_char *, nexpire); /* 0..mfchashsize-1 */ +VNET_DEFINE_STATIC(u_char *, nexpire); /* 0..mfchashsize-1 */ #define V_nexpire VNET(nexpire) -static VNET_DEFINE(LIST_HEAD(mfchashhdr, mfc)*, mfchashtbl); +VNET_DEFINE_STATIC(LIST_HEAD(mfchashhdr, mfc)*, mfchashtbl); #define V_mfchashtbl VNET(mfchashtbl) static struct mtx mfc_mtx; #define MFC_LOCK() mtx_lock(&mfc_mtx) #define MFC_UNLOCK() mtx_unlock(&mfc_mtx) #define MFC_LOCK_ASSERT() mtx_assert(&mfc_mtx, MA_OWNED) #define MFC_LOCK_INIT() \ mtx_init(&mfc_mtx, "IPv4 multicast forwarding cache", NULL, MTX_DEF) #define MFC_LOCK_DESTROY() mtx_destroy(&mfc_mtx) -static VNET_DEFINE(vifi_t, numvifs); +VNET_DEFINE_STATIC(vifi_t, numvifs); #define V_numvifs VNET(numvifs) -static VNET_DEFINE(struct vif, viftable[MAXVIFS]); +VNET_DEFINE_STATIC(struct vif, viftable[MAXVIFS]); #define V_viftable VNET(viftable) SYSCTL_OPAQUE(_net_inet_ip, OID_AUTO, viftable, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(viftable), sizeof(V_viftable), "S,vif[MAXVIFS]", "IPv4 Multicast Interfaces (struct vif[MAXVIFS], netinet/ip_mroute.h)"); static struct mtx vif_mtx; #define VIF_LOCK() mtx_lock(&vif_mtx) #define VIF_UNLOCK() mtx_unlock(&vif_mtx) #define VIF_LOCK_ASSERT() mtx_assert(&vif_mtx, MA_OWNED) #define VIF_LOCK_INIT() \ mtx_init(&vif_mtx, "IPv4 multicast interfaces", NULL, MTX_DEF) #define VIF_LOCK_DESTROY() mtx_destroy(&vif_mtx) static eventhandler_tag if_detach_event_tag = NULL; -static VNET_DEFINE(struct callout, expire_upcalls_ch); +VNET_DEFINE_STATIC(struct callout, expire_upcalls_ch); #define V_expire_upcalls_ch VNET(expire_upcalls_ch) #define EXPIRE_TIMEOUT (hz / 4) /* 4x / second */ #define UPCALL_EXPIRE 6 /* number of timeouts */ /* * Bandwidth meter variables and constants */ static MALLOC_DEFINE(M_BWMETER, "bwmeter", "multicast upcall bw meters"); /* * Pending timeouts are stored in a hash table, the key being the * expiration time. Periodically, the entries are analysed and processed. */ #define BW_METER_BUCKETS 1024 -static VNET_DEFINE(struct bw_meter*, bw_meter_timers[BW_METER_BUCKETS]); +VNET_DEFINE_STATIC(struct bw_meter*, bw_meter_timers[BW_METER_BUCKETS]); #define V_bw_meter_timers VNET(bw_meter_timers) -static VNET_DEFINE(struct callout, bw_meter_ch); +VNET_DEFINE_STATIC(struct callout, bw_meter_ch); #define V_bw_meter_ch VNET(bw_meter_ch) #define BW_METER_PERIOD (hz) /* periodical handling of bw meters */ /* * Pending upcalls are stored in a vector which is flushed when * full, or periodically */ -static VNET_DEFINE(struct bw_upcall, bw_upcalls[BW_UPCALLS_MAX]); +VNET_DEFINE_STATIC(struct bw_upcall, bw_upcalls[BW_UPCALLS_MAX]); #define V_bw_upcalls VNET(bw_upcalls) -static VNET_DEFINE(u_int, bw_upcalls_n); /* # of pending upcalls */ +VNET_DEFINE_STATIC(u_int, bw_upcalls_n); /* # of pending upcalls */ #define V_bw_upcalls_n VNET(bw_upcalls_n) -static VNET_DEFINE(struct callout, bw_upcalls_ch); +VNET_DEFINE_STATIC(struct callout, bw_upcalls_ch); #define V_bw_upcalls_ch VNET(bw_upcalls_ch) #define BW_UPCALLS_PERIOD (hz) /* periodical flush of bw upcalls */ -static VNET_PCPUSTAT_DEFINE(struct pimstat, pimstat); +VNET_PCPUSTAT_DEFINE_STATIC(struct pimstat, pimstat); VNET_PCPUSTAT_SYSINIT(pimstat); VNET_PCPUSTAT_SYSUNINIT(pimstat); SYSCTL_NODE(_net_inet, IPPROTO_PIM, pim, CTLFLAG_RW, 0, "PIM"); SYSCTL_VNET_PCPUSTAT(_net_inet_pim, PIMCTL_STATS, stats, struct pimstat, pimstat, "PIM Statistics (struct pimstat, netinet/pim_var.h)"); static u_long pim_squelch_wholepkt = 0; SYSCTL_ULONG(_net_inet_pim, OID_AUTO, squelch_wholepkt, CTLFLAG_RW, &pim_squelch_wholepkt, 0, "Disable IGMP_WHOLEPKT notifications if rendezvous point is unspecified"); static const struct encaptab *pim_encap_cookie; static int pim_encapcheck(const struct mbuf *, int, int, void *); static int pim_input(struct mbuf *, int, int, void *); static const struct encap_config ipv4_encap_cfg = { .proto = IPPROTO_PIM, .min_length = sizeof(struct ip) + PIM_MINLEN, .exact_match = 8, .check = pim_encapcheck, .input = pim_input }; /* * Note: the PIM Register encapsulation adds the following in front of a * data packet: * * struct pim_encap_hdr { * struct ip ip; * struct pim_encap_pimhdr pim; * } * */ struct pim_encap_pimhdr { struct pim pim; uint32_t flags; }; #define PIM_ENCAP_TTL 64 static struct ip pim_encap_iphdr = { #if BYTE_ORDER == LITTLE_ENDIAN sizeof(struct ip) >> 2, IPVERSION, #else IPVERSION, sizeof(struct ip) >> 2, #endif 0, /* tos */ sizeof(struct ip), /* total length */ 0, /* id */ 0, /* frag offset */ PIM_ENCAP_TTL, IPPROTO_PIM, 0, /* checksum */ }; static struct pim_encap_pimhdr pim_encap_pimhdr = { { PIM_MAKE_VT(PIM_VERSION, PIM_REGISTER), /* PIM vers and message type */ 0, /* reserved */ 0, /* checksum */ }, 0 /* flags */ }; -static VNET_DEFINE(vifi_t, reg_vif_num) = VIFI_INVALID; +VNET_DEFINE_STATIC(vifi_t, reg_vif_num) = VIFI_INVALID; #define V_reg_vif_num VNET(reg_vif_num) -static VNET_DEFINE(struct ifnet, multicast_register_if); +VNET_DEFINE_STATIC(struct ifnet, multicast_register_if); #define V_multicast_register_if VNET(multicast_register_if) /* * Private variables. */ static u_long X_ip_mcast_src(int); static int X_ip_mforward(struct ip *, struct ifnet *, struct mbuf *, struct ip_moptions *); static int X_ip_mrouter_done(void); static int X_ip_mrouter_get(struct socket *, struct sockopt *); static int X_ip_mrouter_set(struct socket *, struct sockopt *); static int X_legal_vif_num(int); static int X_mrt_ioctl(u_long, caddr_t, int); static int add_bw_upcall(struct bw_upcall *); static int add_mfc(struct mfcctl2 *); static int add_vif(struct vifctl *); static void bw_meter_prepare_upcall(struct bw_meter *, struct timeval *); static void bw_meter_process(void); static void bw_meter_receive_packet(struct bw_meter *, int, struct timeval *); static void bw_upcalls_send(void); static int del_bw_upcall(struct bw_upcall *); static int del_mfc(struct mfcctl2 *); static int del_vif(vifi_t); static int del_vif_locked(vifi_t); static void expire_bw_meter_process(void *); static void expire_bw_upcalls_send(void *); static void expire_mfc(struct mfc *); static void expire_upcalls(void *); static void free_bw_list(struct bw_meter *); static int get_sg_cnt(struct sioc_sg_req *); static int get_vif_cnt(struct sioc_vif_req *); static void if_detached_event(void *, struct ifnet *); static int ip_mdq(struct mbuf *, struct ifnet *, struct mfc *, vifi_t); static int ip_mrouter_init(struct socket *, int); static __inline struct mfc * mfc_find(struct in_addr *, struct in_addr *); static void phyint_send(struct ip *, struct vif *, struct mbuf *); static struct mbuf * pim_register_prepare(struct ip *, struct mbuf *); static int pim_register_send(struct ip *, struct vif *, struct mbuf *, struct mfc *); static int pim_register_send_rp(struct ip *, struct vif *, struct mbuf *, struct mfc *); static int pim_register_send_upcall(struct ip *, struct vif *, struct mbuf *, struct mfc *); static void schedule_bw_meter(struct bw_meter *, struct timeval *); static void send_packet(struct vif *, struct mbuf *); static int set_api_config(uint32_t *); static int set_assert(int); static int socket_send(struct socket *, struct mbuf *, struct sockaddr_in *); static void unschedule_bw_meter(struct bw_meter *); /* * Kernel multicast forwarding API capabilities and setup. * If more API capabilities are added to the kernel, they should be * recorded in `mrt_api_support'. */ #define MRT_API_VERSION 0x0305 static const int mrt_api_version = MRT_API_VERSION; static const uint32_t mrt_api_support = (MRT_MFC_FLAGS_DISABLE_WRONGVIF | MRT_MFC_FLAGS_BORDER_VIF | MRT_MFC_RP | MRT_MFC_BW_UPCALL); -static VNET_DEFINE(uint32_t, mrt_api_config); +VNET_DEFINE_STATIC(uint32_t, mrt_api_config); #define V_mrt_api_config VNET(mrt_api_config) -static VNET_DEFINE(int, pim_assert_enabled); +VNET_DEFINE_STATIC(int, pim_assert_enabled); #define V_pim_assert_enabled VNET(pim_assert_enabled) static struct timeval pim_assert_interval = { 3, 0 }; /* Rate limit */ /* * Find a route for a given origin IP address and multicast group address. * Statistics must be updated by the caller. */ static __inline struct mfc * mfc_find(struct in_addr *o, struct in_addr *g) { struct mfc *rt; MFC_LOCK_ASSERT(); LIST_FOREACH(rt, &V_mfchashtbl[MFCHASH(*o, *g)], mfc_hash) { if (in_hosteq(rt->mfc_origin, *o) && in_hosteq(rt->mfc_mcastgrp, *g) && TAILQ_EMPTY(&rt->mfc_stall)) break; } return (rt); } /* * Handle MRT setsockopt commands to modify the multicast forwarding tables. */ static int X_ip_mrouter_set(struct socket *so, struct sockopt *sopt) { int error, optval; vifi_t vifi; struct vifctl vifc; struct mfcctl2 mfc; struct bw_upcall bw_upcall; uint32_t i; if (so != V_ip_mrouter && sopt->sopt_name != MRT_INIT) return EPERM; error = 0; switch (sopt->sopt_name) { case MRT_INIT: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; error = ip_mrouter_init(so, optval); break; case MRT_DONE: error = ip_mrouter_done(); break; case MRT_ADD_VIF: error = sooptcopyin(sopt, &vifc, sizeof vifc, sizeof vifc); if (error) break; error = add_vif(&vifc); break; case MRT_DEL_VIF: error = sooptcopyin(sopt, &vifi, sizeof vifi, sizeof vifi); if (error) break; error = del_vif(vifi); break; case MRT_ADD_MFC: case MRT_DEL_MFC: /* * select data size depending on API version. */ if (sopt->sopt_name == MRT_ADD_MFC && V_mrt_api_config & MRT_API_FLAGS_ALL) { error = sooptcopyin(sopt, &mfc, sizeof(struct mfcctl2), sizeof(struct mfcctl2)); } else { error = sooptcopyin(sopt, &mfc, sizeof(struct mfcctl), sizeof(struct mfcctl)); bzero((caddr_t)&mfc + sizeof(struct mfcctl), sizeof(mfc) - sizeof(struct mfcctl)); } if (error) break; if (sopt->sopt_name == MRT_ADD_MFC) error = add_mfc(&mfc); else error = del_mfc(&mfc); break; case MRT_ASSERT: error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval); if (error) break; set_assert(optval); break; case MRT_API_CONFIG: error = sooptcopyin(sopt, &i, sizeof i, sizeof i); if (!error) error = set_api_config(&i); if (!error) error = sooptcopyout(sopt, &i, sizeof i); break; case MRT_ADD_BW_UPCALL: case MRT_DEL_BW_UPCALL: error = sooptcopyin(sopt, &bw_upcall, sizeof bw_upcall, sizeof bw_upcall); if (error) break; if (sopt->sopt_name == MRT_ADD_BW_UPCALL) error = add_bw_upcall(&bw_upcall); else error = del_bw_upcall(&bw_upcall); break; default: error = EOPNOTSUPP; break; } return error; } /* * Handle MRT getsockopt commands */ static int X_ip_mrouter_get(struct socket *so, struct sockopt *sopt) { int error; switch (sopt->sopt_name) { case MRT_VERSION: error = sooptcopyout(sopt, &mrt_api_version, sizeof mrt_api_version); break; case MRT_ASSERT: error = sooptcopyout(sopt, &V_pim_assert_enabled, sizeof V_pim_assert_enabled); break; case MRT_API_SUPPORT: error = sooptcopyout(sopt, &mrt_api_support, sizeof mrt_api_support); break; case MRT_API_CONFIG: error = sooptcopyout(sopt, &V_mrt_api_config, sizeof V_mrt_api_config); break; default: error = EOPNOTSUPP; break; } return error; } /* * Handle ioctl commands to obtain information from the cache */ static int X_mrt_ioctl(u_long cmd, caddr_t data, int fibnum __unused) { int error = 0; /* * Currently the only function calling this ioctl routine is rtioctl_fib(). * Typically, only root can create the raw socket in order to execute * this ioctl method, however the request might be coming from a prison */ error = priv_check(curthread, PRIV_NETINET_MROUTE); if (error) return (error); switch (cmd) { case (SIOCGETVIFCNT): error = get_vif_cnt((struct sioc_vif_req *)data); break; case (SIOCGETSGCNT): error = get_sg_cnt((struct sioc_sg_req *)data); break; default: error = EINVAL; break; } return error; } /* * returns the packet, byte, rpf-failure count for the source group provided */ static int get_sg_cnt(struct sioc_sg_req *req) { struct mfc *rt; MFC_LOCK(); rt = mfc_find(&req->src, &req->grp); if (rt == NULL) { MFC_UNLOCK(); req->pktcnt = req->bytecnt = req->wrong_if = 0xffffffff; return EADDRNOTAVAIL; } req->pktcnt = rt->mfc_pkt_cnt; req->bytecnt = rt->mfc_byte_cnt; req->wrong_if = rt->mfc_wrong_if; MFC_UNLOCK(); return 0; } /* * returns the input and output packet and byte counts on the vif provided */ static int get_vif_cnt(struct sioc_vif_req *req) { vifi_t vifi = req->vifi; VIF_LOCK(); if (vifi >= V_numvifs) { VIF_UNLOCK(); return EINVAL; } req->icount = V_viftable[vifi].v_pkt_in; req->ocount = V_viftable[vifi].v_pkt_out; req->ibytes = V_viftable[vifi].v_bytes_in; req->obytes = V_viftable[vifi].v_bytes_out; VIF_UNLOCK(); return 0; } static void if_detached_event(void *arg __unused, struct ifnet *ifp) { vifi_t vifi; u_long i; MROUTER_LOCK(); if (V_ip_mrouter == NULL) { MROUTER_UNLOCK(); return; } VIF_LOCK(); MFC_LOCK(); /* * Tear down multicast forwarder state associated with this ifnet. * 1. Walk the vif list, matching vifs against this ifnet. * 2. Walk the multicast forwarding cache (mfc) looking for * inner matches with this vif's index. * 3. Expire any matching multicast forwarding cache entries. * 4. Free vif state. This should disable ALLMULTI on the interface. */ for (vifi = 0; vifi < V_numvifs; vifi++) { if (V_viftable[vifi].v_ifp != ifp) continue; for (i = 0; i < mfchashsize; i++) { struct mfc *rt, *nrt; LIST_FOREACH_SAFE(rt, &V_mfchashtbl[i], mfc_hash, nrt) { if (rt->mfc_parent == vifi) { expire_mfc(rt); } } } del_vif_locked(vifi); } MFC_UNLOCK(); VIF_UNLOCK(); MROUTER_UNLOCK(); } /* * Enable multicast forwarding. */ static int ip_mrouter_init(struct socket *so, int version) { CTR3(KTR_IPMF, "%s: so_type %d, pr_protocol %d", __func__, so->so_type, so->so_proto->pr_protocol); if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_IGMP) return EOPNOTSUPP; if (version != 1) return ENOPROTOOPT; MROUTER_LOCK(); if (ip_mrouter_unloading) { MROUTER_UNLOCK(); return ENOPROTOOPT; } if (V_ip_mrouter != NULL) { MROUTER_UNLOCK(); return EADDRINUSE; } V_mfchashtbl = hashinit_flags(mfchashsize, M_MRTABLE, &V_mfchash, HASH_NOWAIT); callout_reset(&V_expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, curvnet); callout_reset(&V_bw_upcalls_ch, BW_UPCALLS_PERIOD, expire_bw_upcalls_send, curvnet); callout_reset(&V_bw_meter_ch, BW_METER_PERIOD, expire_bw_meter_process, curvnet); V_ip_mrouter = so; ip_mrouter_cnt++; MROUTER_UNLOCK(); CTR1(KTR_IPMF, "%s: done", __func__); return 0; } /* * Disable multicast forwarding. */ static int X_ip_mrouter_done(void) { struct ifnet *ifp; u_long i; vifi_t vifi; MROUTER_LOCK(); if (V_ip_mrouter == NULL) { MROUTER_UNLOCK(); return EINVAL; } /* * Detach/disable hooks to the reset of the system. */ V_ip_mrouter = NULL; ip_mrouter_cnt--; V_mrt_api_config = 0; VIF_LOCK(); /* * For each phyint in use, disable promiscuous reception of all IP * multicasts. */ for (vifi = 0; vifi < V_numvifs; vifi++) { if (!in_nullhost(V_viftable[vifi].v_lcl_addr) && !(V_viftable[vifi].v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) { ifp = V_viftable[vifi].v_ifp; if_allmulti(ifp, 0); } } bzero((caddr_t)V_viftable, sizeof(V_viftable)); V_numvifs = 0; V_pim_assert_enabled = 0; VIF_UNLOCK(); callout_stop(&V_expire_upcalls_ch); callout_stop(&V_bw_upcalls_ch); callout_stop(&V_bw_meter_ch); MFC_LOCK(); /* * Free all multicast forwarding cache entries. * Do not use hashdestroy(), as we must perform other cleanup. */ for (i = 0; i < mfchashsize; i++) { struct mfc *rt, *nrt; LIST_FOREACH_SAFE(rt, &V_mfchashtbl[i], mfc_hash, nrt) { expire_mfc(rt); } } free(V_mfchashtbl, M_MRTABLE); V_mfchashtbl = NULL; bzero(V_nexpire, sizeof(V_nexpire[0]) * mfchashsize); V_bw_upcalls_n = 0; bzero(V_bw_meter_timers, sizeof(V_bw_meter_timers)); MFC_UNLOCK(); V_reg_vif_num = VIFI_INVALID; MROUTER_UNLOCK(); CTR1(KTR_IPMF, "%s: done", __func__); return 0; } /* * Set PIM assert processing global */ static int set_assert(int i) { if ((i != 1) && (i != 0)) return EINVAL; V_pim_assert_enabled = i; return 0; } /* * Configure API capabilities */ int set_api_config(uint32_t *apival) { u_long i; /* * We can set the API capabilities only if it is the first operation * after MRT_INIT. I.e.: * - there are no vifs installed * - pim_assert is not enabled * - the MFC table is empty */ if (V_numvifs > 0) { *apival = 0; return EPERM; } if (V_pim_assert_enabled) { *apival = 0; return EPERM; } MFC_LOCK(); for (i = 0; i < mfchashsize; i++) { if (LIST_FIRST(&V_mfchashtbl[i]) != NULL) { MFC_UNLOCK(); *apival = 0; return EPERM; } } MFC_UNLOCK(); V_mrt_api_config = *apival & mrt_api_support; *apival = V_mrt_api_config; return 0; } /* * Add a vif to the vif table */ static int add_vif(struct vifctl *vifcp) { struct vif *vifp = V_viftable + vifcp->vifc_vifi; struct sockaddr_in sin = {sizeof sin, AF_INET}; struct ifaddr *ifa; struct ifnet *ifp; int error; VIF_LOCK(); if (vifcp->vifc_vifi >= MAXVIFS) { VIF_UNLOCK(); return EINVAL; } /* rate limiting is no longer supported by this code */ if (vifcp->vifc_rate_limit != 0) { log(LOG_ERR, "rate limiting is no longer supported\n"); VIF_UNLOCK(); return EINVAL; } if (!in_nullhost(vifp->v_lcl_addr)) { VIF_UNLOCK(); return EADDRINUSE; } if (in_nullhost(vifcp->vifc_lcl_addr)) { VIF_UNLOCK(); return EADDRNOTAVAIL; } /* Find the interface with an address in AF_INET family */ if (vifcp->vifc_flags & VIFF_REGISTER) { /* * XXX: Because VIFF_REGISTER does not really need a valid * local interface (e.g. it could be 127.0.0.2), we don't * check its address. */ ifp = NULL; } else { sin.sin_addr = vifcp->vifc_lcl_addr; NET_EPOCH_ENTER(); ifa = ifa_ifwithaddr((struct sockaddr *)&sin); if (ifa == NULL) { NET_EPOCH_EXIT(); VIF_UNLOCK(); return EADDRNOTAVAIL; } ifp = ifa->ifa_ifp; NET_EPOCH_EXIT(); } if ((vifcp->vifc_flags & VIFF_TUNNEL) != 0) { CTR1(KTR_IPMF, "%s: tunnels are no longer supported", __func__); VIF_UNLOCK(); return EOPNOTSUPP; } else if (vifcp->vifc_flags & VIFF_REGISTER) { ifp = &V_multicast_register_if; CTR2(KTR_IPMF, "%s: add register vif for ifp %p", __func__, ifp); if (V_reg_vif_num == VIFI_INVALID) { if_initname(&V_multicast_register_if, "register_vif", 0); V_multicast_register_if.if_flags = IFF_LOOPBACK; V_reg_vif_num = vifcp->vifc_vifi; } } else { /* Make sure the interface supports multicast */ if ((ifp->if_flags & IFF_MULTICAST) == 0) { VIF_UNLOCK(); return EOPNOTSUPP; } /* Enable promiscuous reception of all IP multicasts from the if */ error = if_allmulti(ifp, 1); if (error) { VIF_UNLOCK(); return error; } } vifp->v_flags = vifcp->vifc_flags; vifp->v_threshold = vifcp->vifc_threshold; vifp->v_lcl_addr = vifcp->vifc_lcl_addr; vifp->v_rmt_addr = vifcp->vifc_rmt_addr; vifp->v_ifp = ifp; /* initialize per vif pkt counters */ vifp->v_pkt_in = 0; vifp->v_pkt_out = 0; vifp->v_bytes_in = 0; vifp->v_bytes_out = 0; /* Adjust numvifs up if the vifi is higher than numvifs */ if (V_numvifs <= vifcp->vifc_vifi) V_numvifs = vifcp->vifc_vifi + 1; VIF_UNLOCK(); CTR4(KTR_IPMF, "%s: add vif %d laddr 0x%08x thresh %x", __func__, (int)vifcp->vifc_vifi, ntohl(vifcp->vifc_lcl_addr.s_addr), (int)vifcp->vifc_threshold); return 0; } /* * Delete a vif from the vif table */ static int del_vif_locked(vifi_t vifi) { struct vif *vifp; VIF_LOCK_ASSERT(); if (vifi >= V_numvifs) { return EINVAL; } vifp = &V_viftable[vifi]; if (in_nullhost(vifp->v_lcl_addr)) { return EADDRNOTAVAIL; } if (!(vifp->v_flags & (VIFF_TUNNEL | VIFF_REGISTER))) if_allmulti(vifp->v_ifp, 0); if (vifp->v_flags & VIFF_REGISTER) V_reg_vif_num = VIFI_INVALID; bzero((caddr_t)vifp, sizeof (*vifp)); CTR2(KTR_IPMF, "%s: delete vif %d", __func__, (int)vifi); /* Adjust numvifs down */ for (vifi = V_numvifs; vifi > 0; vifi--) if (!in_nullhost(V_viftable[vifi-1].v_lcl_addr)) break; V_numvifs = vifi; return 0; } static int del_vif(vifi_t vifi) { int cc; VIF_LOCK(); cc = del_vif_locked(vifi); VIF_UNLOCK(); return cc; } /* * update an mfc entry without resetting counters and S,G addresses. */ static void update_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp) { int i; rt->mfc_parent = mfccp->mfcc_parent; for (i = 0; i < V_numvifs; i++) { rt->mfc_ttls[i] = mfccp->mfcc_ttls[i]; rt->mfc_flags[i] = mfccp->mfcc_flags[i] & V_mrt_api_config & MRT_MFC_FLAGS_ALL; } /* set the RP address */ if (V_mrt_api_config & MRT_MFC_RP) rt->mfc_rp = mfccp->mfcc_rp; else rt->mfc_rp.s_addr = INADDR_ANY; } /* * fully initialize an mfc entry from the parameter. */ static void init_mfc_params(struct mfc *rt, struct mfcctl2 *mfccp) { rt->mfc_origin = mfccp->mfcc_origin; rt->mfc_mcastgrp = mfccp->mfcc_mcastgrp; update_mfc_params(rt, mfccp); /* initialize pkt counters per src-grp */ rt->mfc_pkt_cnt = 0; rt->mfc_byte_cnt = 0; rt->mfc_wrong_if = 0; timevalclear(&rt->mfc_last_assert); } static void expire_mfc(struct mfc *rt) { struct rtdetq *rte, *nrte; MFC_LOCK_ASSERT(); free_bw_list(rt->mfc_bw_meter); TAILQ_FOREACH_SAFE(rte, &rt->mfc_stall, rte_link, nrte) { m_freem(rte->m); TAILQ_REMOVE(&rt->mfc_stall, rte, rte_link); free(rte, M_MRTABLE); } LIST_REMOVE(rt, mfc_hash); free(rt, M_MRTABLE); } /* * Add an mfc entry */ static int add_mfc(struct mfcctl2 *mfccp) { struct mfc *rt; struct rtdetq *rte, *nrte; u_long hash = 0; u_short nstl; VIF_LOCK(); MFC_LOCK(); rt = mfc_find(&mfccp->mfcc_origin, &mfccp->mfcc_mcastgrp); /* If an entry already exists, just update the fields */ if (rt) { CTR4(KTR_IPMF, "%s: update mfc orig 0x%08x group %lx parent %x", __func__, ntohl(mfccp->mfcc_origin.s_addr), (u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr), mfccp->mfcc_parent); update_mfc_params(rt, mfccp); MFC_UNLOCK(); VIF_UNLOCK(); return (0); } /* * Find the entry for which the upcall was made and update */ nstl = 0; hash = MFCHASH(mfccp->mfcc_origin, mfccp->mfcc_mcastgrp); LIST_FOREACH(rt, &V_mfchashtbl[hash], mfc_hash) { if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) && in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp) && !TAILQ_EMPTY(&rt->mfc_stall)) { CTR5(KTR_IPMF, "%s: add mfc orig 0x%08x group %lx parent %x qh %p", __func__, ntohl(mfccp->mfcc_origin.s_addr), (u_long)ntohl(mfccp->mfcc_mcastgrp.s_addr), mfccp->mfcc_parent, TAILQ_FIRST(&rt->mfc_stall)); if (nstl++) CTR1(KTR_IPMF, "%s: multiple matches", __func__); init_mfc_params(rt, mfccp); rt->mfc_expire = 0; /* Don't clean this guy up */ V_nexpire[hash]--; /* Free queued packets, but attempt to forward them first. */ TAILQ_FOREACH_SAFE(rte, &rt->mfc_stall, rte_link, nrte) { if (rte->ifp != NULL) ip_mdq(rte->m, rte->ifp, rt, -1); m_freem(rte->m); TAILQ_REMOVE(&rt->mfc_stall, rte, rte_link); rt->mfc_nstall--; free(rte, M_MRTABLE); } } } /* * It is possible that an entry is being inserted without an upcall */ if (nstl == 0) { CTR1(KTR_IPMF, "%s: adding mfc w/o upcall", __func__); LIST_FOREACH(rt, &V_mfchashtbl[hash], mfc_hash) { if (in_hosteq(rt->mfc_origin, mfccp->mfcc_origin) && in_hosteq(rt->mfc_mcastgrp, mfccp->mfcc_mcastgrp)) { init_mfc_params(rt, mfccp); if (rt->mfc_expire) V_nexpire[hash]--; rt->mfc_expire = 0; break; /* XXX */ } } if (rt == NULL) { /* no upcall, so make a new entry */ rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE, M_NOWAIT); if (rt == NULL) { MFC_UNLOCK(); VIF_UNLOCK(); return (ENOBUFS); } init_mfc_params(rt, mfccp); TAILQ_INIT(&rt->mfc_stall); rt->mfc_nstall = 0; rt->mfc_expire = 0; rt->mfc_bw_meter = NULL; /* insert new entry at head of hash chain */ LIST_INSERT_HEAD(&V_mfchashtbl[hash], rt, mfc_hash); } } MFC_UNLOCK(); VIF_UNLOCK(); return (0); } /* * Delete an mfc entry */ static int del_mfc(struct mfcctl2 *mfccp) { struct in_addr origin; struct in_addr mcastgrp; struct mfc *rt; origin = mfccp->mfcc_origin; mcastgrp = mfccp->mfcc_mcastgrp; CTR3(KTR_IPMF, "%s: delete mfc orig 0x%08x group %lx", __func__, ntohl(origin.s_addr), (u_long)ntohl(mcastgrp.s_addr)); MFC_LOCK(); rt = mfc_find(&origin, &mcastgrp); if (rt == NULL) { MFC_UNLOCK(); return EADDRNOTAVAIL; } /* * free the bw_meter entries */ free_bw_list(rt->mfc_bw_meter); rt->mfc_bw_meter = NULL; LIST_REMOVE(rt, mfc_hash); free(rt, M_MRTABLE); MFC_UNLOCK(); return (0); } /* * Send a message to the routing daemon on the multicast routing socket. */ static int socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in *src) { if (s) { SOCKBUF_LOCK(&s->so_rcv); if (sbappendaddr_locked(&s->so_rcv, (struct sockaddr *)src, mm, NULL) != 0) { sorwakeup_locked(s); return 0; } SOCKBUF_UNLOCK(&s->so_rcv); } m_freem(mm); return -1; } /* * IP multicast forwarding function. This function assumes that the packet * pointed to by "ip" has arrived on (or is about to be sent to) the interface * pointed to by "ifp", and the packet is to be relayed to other networks * that have members of the packet's destination IP multicast group. * * The packet is returned unscathed to the caller, unless it is * erroneous, in which case a non-zero return value tells the caller to * discard it. */ #define TUNNEL_LEN 12 /* # bytes of IP option for tunnel encapsulation */ static int X_ip_mforward(struct ip *ip, struct ifnet *ifp, struct mbuf *m, struct ip_moptions *imo) { struct mfc *rt; int error; vifi_t vifi; CTR3(KTR_IPMF, "ip_mforward: delete mfc orig 0x%08x group %lx ifp %p", ntohl(ip->ip_src.s_addr), (u_long)ntohl(ip->ip_dst.s_addr), ifp); if (ip->ip_hl < (sizeof(struct ip) + TUNNEL_LEN) >> 2 || ((u_char *)(ip + 1))[1] != IPOPT_LSRR ) { /* * Packet arrived via a physical interface or * an encapsulated tunnel or a register_vif. */ } else { /* * Packet arrived through a source-route tunnel. * Source-route tunnels are no longer supported. */ return (1); } VIF_LOCK(); MFC_LOCK(); if (imo && ((vifi = imo->imo_multicast_vif) < V_numvifs)) { if (ip->ip_ttl < MAXTTL) ip->ip_ttl++; /* compensate for -1 in *_send routines */ error = ip_mdq(m, ifp, NULL, vifi); MFC_UNLOCK(); VIF_UNLOCK(); return error; } /* * Don't forward a packet with time-to-live of zero or one, * or a packet destined to a local-only group. */ if (ip->ip_ttl <= 1 || IN_LOCAL_GROUP(ntohl(ip->ip_dst.s_addr))) { MFC_UNLOCK(); VIF_UNLOCK(); return 0; } /* * Determine forwarding vifs from the forwarding cache table */ MRTSTAT_INC(mrts_mfc_lookups); rt = mfc_find(&ip->ip_src, &ip->ip_dst); /* Entry exists, so forward if necessary */ if (rt != NULL) { error = ip_mdq(m, ifp, rt, -1); MFC_UNLOCK(); VIF_UNLOCK(); return error; } else { /* * If we don't have a route for packet's origin, * Make a copy of the packet & send message to routing daemon */ struct mbuf *mb0; struct rtdetq *rte; u_long hash; int hlen = ip->ip_hl << 2; MRTSTAT_INC(mrts_mfc_misses); MRTSTAT_INC(mrts_no_route); CTR2(KTR_IPMF, "ip_mforward: no mfc for (0x%08x,%lx)", ntohl(ip->ip_src.s_addr), (u_long)ntohl(ip->ip_dst.s_addr)); /* * Allocate mbufs early so that we don't do extra work if we are * just going to fail anyway. Make sure to pullup the header so * that other people can't step on it. */ rte = (struct rtdetq *)malloc((sizeof *rte), M_MRTABLE, M_NOWAIT|M_ZERO); if (rte == NULL) { MFC_UNLOCK(); VIF_UNLOCK(); return ENOBUFS; } mb0 = m_copypacket(m, M_NOWAIT); if (mb0 && (!M_WRITABLE(mb0) || mb0->m_len < hlen)) mb0 = m_pullup(mb0, hlen); if (mb0 == NULL) { free(rte, M_MRTABLE); MFC_UNLOCK(); VIF_UNLOCK(); return ENOBUFS; } /* is there an upcall waiting for this flow ? */ hash = MFCHASH(ip->ip_src, ip->ip_dst); LIST_FOREACH(rt, &V_mfchashtbl[hash], mfc_hash) { if (in_hosteq(ip->ip_src, rt->mfc_origin) && in_hosteq(ip->ip_dst, rt->mfc_mcastgrp) && !TAILQ_EMPTY(&rt->mfc_stall)) break; } if (rt == NULL) { int i; struct igmpmsg *im; struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET }; struct mbuf *mm; /* * Locate the vifi for the incoming interface for this packet. * If none found, drop packet. */ for (vifi = 0; vifi < V_numvifs && V_viftable[vifi].v_ifp != ifp; vifi++) ; if (vifi >= V_numvifs) /* vif not found, drop packet */ goto non_fatal; /* no upcall, so make a new entry */ rt = (struct mfc *)malloc(sizeof(*rt), M_MRTABLE, M_NOWAIT); if (rt == NULL) goto fail; /* Make a copy of the header to send to the user level process */ mm = m_copym(mb0, 0, hlen, M_NOWAIT); if (mm == NULL) goto fail1; /* * Send message to routing daemon to install * a route into the kernel table */ im = mtod(mm, struct igmpmsg *); im->im_msgtype = IGMPMSG_NOCACHE; im->im_mbz = 0; im->im_vif = vifi; MRTSTAT_INC(mrts_upcalls); k_igmpsrc.sin_addr = ip->ip_src; if (socket_send(V_ip_mrouter, mm, &k_igmpsrc) < 0) { CTR0(KTR_IPMF, "ip_mforward: socket queue full"); MRTSTAT_INC(mrts_upq_sockfull); fail1: free(rt, M_MRTABLE); fail: free(rte, M_MRTABLE); m_freem(mb0); MFC_UNLOCK(); VIF_UNLOCK(); return ENOBUFS; } /* insert new entry at head of hash chain */ rt->mfc_origin.s_addr = ip->ip_src.s_addr; rt->mfc_mcastgrp.s_addr = ip->ip_dst.s_addr; rt->mfc_expire = UPCALL_EXPIRE; V_nexpire[hash]++; for (i = 0; i < V_numvifs; i++) { rt->mfc_ttls[i] = 0; rt->mfc_flags[i] = 0; } rt->mfc_parent = -1; /* clear the RP address */ rt->mfc_rp.s_addr = INADDR_ANY; rt->mfc_bw_meter = NULL; /* initialize pkt counters per src-grp */ rt->mfc_pkt_cnt = 0; rt->mfc_byte_cnt = 0; rt->mfc_wrong_if = 0; timevalclear(&rt->mfc_last_assert); TAILQ_INIT(&rt->mfc_stall); rt->mfc_nstall = 0; /* link into table */ LIST_INSERT_HEAD(&V_mfchashtbl[hash], rt, mfc_hash); TAILQ_INSERT_HEAD(&rt->mfc_stall, rte, rte_link); rt->mfc_nstall++; } else { /* determine if queue has overflowed */ if (rt->mfc_nstall > MAX_UPQ) { MRTSTAT_INC(mrts_upq_ovflw); non_fatal: free(rte, M_MRTABLE); m_freem(mb0); MFC_UNLOCK(); VIF_UNLOCK(); return (0); } TAILQ_INSERT_TAIL(&rt->mfc_stall, rte, rte_link); rt->mfc_nstall++; } rte->m = mb0; rte->ifp = ifp; MFC_UNLOCK(); VIF_UNLOCK(); return 0; } } /* * Clean up the cache entry if upcall is not serviced */ static void expire_upcalls(void *arg) { u_long i; CURVNET_SET((struct vnet *) arg); MFC_LOCK(); for (i = 0; i < mfchashsize; i++) { struct mfc *rt, *nrt; if (V_nexpire[i] == 0) continue; LIST_FOREACH_SAFE(rt, &V_mfchashtbl[i], mfc_hash, nrt) { if (TAILQ_EMPTY(&rt->mfc_stall)) continue; if (rt->mfc_expire == 0 || --rt->mfc_expire > 0) continue; /* * free the bw_meter entries */ while (rt->mfc_bw_meter != NULL) { struct bw_meter *x = rt->mfc_bw_meter; rt->mfc_bw_meter = x->bm_mfc_next; free(x, M_BWMETER); } MRTSTAT_INC(mrts_cache_cleanups); CTR3(KTR_IPMF, "%s: expire (%lx, %lx)", __func__, (u_long)ntohl(rt->mfc_origin.s_addr), (u_long)ntohl(rt->mfc_mcastgrp.s_addr)); expire_mfc(rt); } } MFC_UNLOCK(); callout_reset(&V_expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, curvnet); CURVNET_RESTORE(); } /* * Packet forwarding routine once entry in the cache is made */ static int ip_mdq(struct mbuf *m, struct ifnet *ifp, struct mfc *rt, vifi_t xmt_vif) { struct ip *ip = mtod(m, struct ip *); vifi_t vifi; int plen = ntohs(ip->ip_len); VIF_LOCK_ASSERT(); /* * If xmt_vif is not -1, send on only the requested vif. * * (since vifi_t is u_short, -1 becomes MAXUSHORT, which > numvifs.) */ if (xmt_vif < V_numvifs) { if (V_viftable[xmt_vif].v_flags & VIFF_REGISTER) pim_register_send(ip, V_viftable + xmt_vif, m, rt); else phyint_send(ip, V_viftable + xmt_vif, m); return 1; } /* * Don't forward if it didn't arrive from the parent vif for its origin. */ vifi = rt->mfc_parent; if ((vifi >= V_numvifs) || (V_viftable[vifi].v_ifp != ifp)) { CTR4(KTR_IPMF, "%s: rx on wrong ifp %p (vifi %d, v_ifp %p)", __func__, ifp, (int)vifi, V_viftable[vifi].v_ifp); MRTSTAT_INC(mrts_wrong_if); ++rt->mfc_wrong_if; /* * If we are doing PIM assert processing, send a message * to the routing daemon. * * XXX: A PIM-SM router needs the WRONGVIF detection so it * can complete the SPT switch, regardless of the type * of the iif (broadcast media, GRE tunnel, etc). */ if (V_pim_assert_enabled && (vifi < V_numvifs) && V_viftable[vifi].v_ifp) { if (ifp == &V_multicast_register_if) PIMSTAT_INC(pims_rcv_registers_wrongiif); /* Get vifi for the incoming packet */ for (vifi = 0; vifi < V_numvifs && V_viftable[vifi].v_ifp != ifp; vifi++) ; if (vifi >= V_numvifs) return 0; /* The iif is not found: ignore the packet. */ if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_DISABLE_WRONGVIF) return 0; /* WRONGVIF disabled: ignore the packet */ if (ratecheck(&rt->mfc_last_assert, &pim_assert_interval)) { struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET }; struct igmpmsg *im; int hlen = ip->ip_hl << 2; struct mbuf *mm = m_copym(m, 0, hlen, M_NOWAIT); if (mm && (!M_WRITABLE(mm) || mm->m_len < hlen)) mm = m_pullup(mm, hlen); if (mm == NULL) return ENOBUFS; im = mtod(mm, struct igmpmsg *); im->im_msgtype = IGMPMSG_WRONGVIF; im->im_mbz = 0; im->im_vif = vifi; MRTSTAT_INC(mrts_upcalls); k_igmpsrc.sin_addr = im->im_src; if (socket_send(V_ip_mrouter, mm, &k_igmpsrc) < 0) { CTR1(KTR_IPMF, "%s: socket queue full", __func__); MRTSTAT_INC(mrts_upq_sockfull); return ENOBUFS; } } } return 0; } /* If I sourced this packet, it counts as output, else it was input. */ if (in_hosteq(ip->ip_src, V_viftable[vifi].v_lcl_addr)) { V_viftable[vifi].v_pkt_out++; V_viftable[vifi].v_bytes_out += plen; } else { V_viftable[vifi].v_pkt_in++; V_viftable[vifi].v_bytes_in += plen; } rt->mfc_pkt_cnt++; rt->mfc_byte_cnt += plen; /* * For each vif, decide if a copy of the packet should be forwarded. * Forward if: * - the ttl exceeds the vif's threshold * - there are group members downstream on interface */ for (vifi = 0; vifi < V_numvifs; vifi++) if ((rt->mfc_ttls[vifi] > 0) && (ip->ip_ttl > rt->mfc_ttls[vifi])) { V_viftable[vifi].v_pkt_out++; V_viftable[vifi].v_bytes_out += plen; if (V_viftable[vifi].v_flags & VIFF_REGISTER) pim_register_send(ip, V_viftable + vifi, m, rt); else phyint_send(ip, V_viftable + vifi, m); } /* * Perform upcall-related bw measuring. */ if (rt->mfc_bw_meter != NULL) { struct bw_meter *x; struct timeval now; microtime(&now); MFC_LOCK_ASSERT(); for (x = rt->mfc_bw_meter; x != NULL; x = x->bm_mfc_next) bw_meter_receive_packet(x, plen, &now); } return 0; } /* * Check if a vif number is legal/ok. This is used by in_mcast.c. */ static int X_legal_vif_num(int vif) { int ret; ret = 0; if (vif < 0) return (ret); VIF_LOCK(); if (vif < V_numvifs) ret = 1; VIF_UNLOCK(); return (ret); } /* * Return the local address used by this vif */ static u_long X_ip_mcast_src(int vifi) { in_addr_t addr; addr = INADDR_ANY; if (vifi < 0) return (addr); VIF_LOCK(); if (vifi < V_numvifs) addr = V_viftable[vifi].v_lcl_addr.s_addr; VIF_UNLOCK(); return (addr); } static void phyint_send(struct ip *ip, struct vif *vifp, struct mbuf *m) { struct mbuf *mb_copy; int hlen = ip->ip_hl << 2; VIF_LOCK_ASSERT(); /* * Make a new reference to the packet; make sure that * the IP header is actually copied, not just referenced, * so that ip_output() only scribbles on the copy. */ mb_copy = m_copypacket(m, M_NOWAIT); if (mb_copy && (!M_WRITABLE(mb_copy) || mb_copy->m_len < hlen)) mb_copy = m_pullup(mb_copy, hlen); if (mb_copy == NULL) return; send_packet(vifp, mb_copy); } static void send_packet(struct vif *vifp, struct mbuf *m) { struct ip_moptions imo; struct in_multi *imm[2]; int error __unused; VIF_LOCK_ASSERT(); imo.imo_multicast_ifp = vifp->v_ifp; imo.imo_multicast_ttl = mtod(m, struct ip *)->ip_ttl - 1; imo.imo_multicast_loop = 1; imo.imo_multicast_vif = -1; imo.imo_num_memberships = 0; imo.imo_max_memberships = 2; imo.imo_membership = &imm[0]; /* * Re-entrancy should not be a problem here, because * the packets that we send out and are looped back at us * should get rejected because they appear to come from * the loopback interface, thus preventing looping. */ error = ip_output(m, NULL, NULL, IP_FORWARDING, &imo, NULL); CTR3(KTR_IPMF, "%s: vif %td err %d", __func__, (ptrdiff_t)(vifp - V_viftable), error); } /* * Stubs for old RSVP socket shim implementation. */ static int X_ip_rsvp_vif(struct socket *so __unused, struct sockopt *sopt __unused) { return (EOPNOTSUPP); } static void X_ip_rsvp_force_done(struct socket *so __unused) { } static int X_rsvp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m; m = *mp; *mp = NULL; if (!V_rsvp_on) m_freem(m); return (IPPROTO_DONE); } /* * Code for bandwidth monitors */ /* * Define common interface for timeval-related methods */ #define BW_TIMEVALCMP(tvp, uvp, cmp) timevalcmp((tvp), (uvp), cmp) #define BW_TIMEVALDECR(vvp, uvp) timevalsub((vvp), (uvp)) #define BW_TIMEVALADD(vvp, uvp) timevaladd((vvp), (uvp)) static uint32_t compute_bw_meter_flags(struct bw_upcall *req) { uint32_t flags = 0; if (req->bu_flags & BW_UPCALL_UNIT_PACKETS) flags |= BW_METER_UNIT_PACKETS; if (req->bu_flags & BW_UPCALL_UNIT_BYTES) flags |= BW_METER_UNIT_BYTES; if (req->bu_flags & BW_UPCALL_GEQ) flags |= BW_METER_GEQ; if (req->bu_flags & BW_UPCALL_LEQ) flags |= BW_METER_LEQ; return flags; } /* * Add a bw_meter entry */ static int add_bw_upcall(struct bw_upcall *req) { struct mfc *mfc; struct timeval delta = { BW_UPCALL_THRESHOLD_INTERVAL_MIN_SEC, BW_UPCALL_THRESHOLD_INTERVAL_MIN_USEC }; struct timeval now; struct bw_meter *x; uint32_t flags; if (!(V_mrt_api_config & MRT_MFC_BW_UPCALL)) return EOPNOTSUPP; /* Test if the flags are valid */ if (!(req->bu_flags & (BW_UPCALL_UNIT_PACKETS | BW_UPCALL_UNIT_BYTES))) return EINVAL; if (!(req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ))) return EINVAL; if ((req->bu_flags & (BW_UPCALL_GEQ | BW_UPCALL_LEQ)) == (BW_UPCALL_GEQ | BW_UPCALL_LEQ)) return EINVAL; /* Test if the threshold time interval is valid */ if (BW_TIMEVALCMP(&req->bu_threshold.b_time, &delta, <)) return EINVAL; flags = compute_bw_meter_flags(req); /* * Find if we have already same bw_meter entry */ MFC_LOCK(); mfc = mfc_find(&req->bu_src, &req->bu_dst); if (mfc == NULL) { MFC_UNLOCK(); return EADDRNOTAVAIL; } for (x = mfc->mfc_bw_meter; x != NULL; x = x->bm_mfc_next) { if ((BW_TIMEVALCMP(&x->bm_threshold.b_time, &req->bu_threshold.b_time, ==)) && (x->bm_threshold.b_packets == req->bu_threshold.b_packets) && (x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) && (x->bm_flags & BW_METER_USER_FLAGS) == flags) { MFC_UNLOCK(); return 0; /* XXX Already installed */ } } /* Allocate the new bw_meter entry */ x = (struct bw_meter *)malloc(sizeof(*x), M_BWMETER, M_NOWAIT); if (x == NULL) { MFC_UNLOCK(); return ENOBUFS; } /* Set the new bw_meter entry */ x->bm_threshold.b_time = req->bu_threshold.b_time; microtime(&now); x->bm_start_time = now; x->bm_threshold.b_packets = req->bu_threshold.b_packets; x->bm_threshold.b_bytes = req->bu_threshold.b_bytes; x->bm_measured.b_packets = 0; x->bm_measured.b_bytes = 0; x->bm_flags = flags; x->bm_time_next = NULL; x->bm_time_hash = BW_METER_BUCKETS; /* Add the new bw_meter entry to the front of entries for this MFC */ x->bm_mfc = mfc; x->bm_mfc_next = mfc->mfc_bw_meter; mfc->mfc_bw_meter = x; schedule_bw_meter(x, &now); MFC_UNLOCK(); return 0; } static void free_bw_list(struct bw_meter *list) { while (list != NULL) { struct bw_meter *x = list; list = list->bm_mfc_next; unschedule_bw_meter(x); free(x, M_BWMETER); } } /* * Delete one or multiple bw_meter entries */ static int del_bw_upcall(struct bw_upcall *req) { struct mfc *mfc; struct bw_meter *x; if (!(V_mrt_api_config & MRT_MFC_BW_UPCALL)) return EOPNOTSUPP; MFC_LOCK(); /* Find the corresponding MFC entry */ mfc = mfc_find(&req->bu_src, &req->bu_dst); if (mfc == NULL) { MFC_UNLOCK(); return EADDRNOTAVAIL; } else if (req->bu_flags & BW_UPCALL_DELETE_ALL) { /* * Delete all bw_meter entries for this mfc */ struct bw_meter *list; list = mfc->mfc_bw_meter; mfc->mfc_bw_meter = NULL; free_bw_list(list); MFC_UNLOCK(); return 0; } else { /* Delete a single bw_meter entry */ struct bw_meter *prev; uint32_t flags = 0; flags = compute_bw_meter_flags(req); /* Find the bw_meter entry to delete */ for (prev = NULL, x = mfc->mfc_bw_meter; x != NULL; prev = x, x = x->bm_mfc_next) { if ((BW_TIMEVALCMP(&x->bm_threshold.b_time, &req->bu_threshold.b_time, ==)) && (x->bm_threshold.b_packets == req->bu_threshold.b_packets) && (x->bm_threshold.b_bytes == req->bu_threshold.b_bytes) && (x->bm_flags & BW_METER_USER_FLAGS) == flags) break; } if (x != NULL) { /* Delete entry from the list for this MFC */ if (prev != NULL) prev->bm_mfc_next = x->bm_mfc_next; /* remove from middle*/ else x->bm_mfc->mfc_bw_meter = x->bm_mfc_next;/* new head of list */ unschedule_bw_meter(x); MFC_UNLOCK(); /* Free the bw_meter entry */ free(x, M_BWMETER); return 0; } else { MFC_UNLOCK(); return EINVAL; } } /* NOTREACHED */ } /* * Perform bandwidth measurement processing that may result in an upcall */ static void bw_meter_receive_packet(struct bw_meter *x, int plen, struct timeval *nowp) { struct timeval delta; MFC_LOCK_ASSERT(); delta = *nowp; BW_TIMEVALDECR(&delta, &x->bm_start_time); if (x->bm_flags & BW_METER_GEQ) { /* * Processing for ">=" type of bw_meter entry */ if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) { /* Reset the bw_meter entry */ x->bm_start_time = *nowp; x->bm_measured.b_packets = 0; x->bm_measured.b_bytes = 0; x->bm_flags &= ~BW_METER_UPCALL_DELIVERED; } /* Record that a packet is received */ x->bm_measured.b_packets++; x->bm_measured.b_bytes += plen; /* * Test if we should deliver an upcall */ if (!(x->bm_flags & BW_METER_UPCALL_DELIVERED)) { if (((x->bm_flags & BW_METER_UNIT_PACKETS) && (x->bm_measured.b_packets >= x->bm_threshold.b_packets)) || ((x->bm_flags & BW_METER_UNIT_BYTES) && (x->bm_measured.b_bytes >= x->bm_threshold.b_bytes))) { /* Prepare an upcall for delivery */ bw_meter_prepare_upcall(x, nowp); x->bm_flags |= BW_METER_UPCALL_DELIVERED; } } } else if (x->bm_flags & BW_METER_LEQ) { /* * Processing for "<=" type of bw_meter entry */ if (BW_TIMEVALCMP(&delta, &x->bm_threshold.b_time, >)) { /* * We are behind time with the multicast forwarding table * scanning for "<=" type of bw_meter entries, so test now * if we should deliver an upcall. */ if (((x->bm_flags & BW_METER_UNIT_PACKETS) && (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) || ((x->bm_flags & BW_METER_UNIT_BYTES) && (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) { /* Prepare an upcall for delivery */ bw_meter_prepare_upcall(x, nowp); } /* Reschedule the bw_meter entry */ unschedule_bw_meter(x); schedule_bw_meter(x, nowp); } /* Record that a packet is received */ x->bm_measured.b_packets++; x->bm_measured.b_bytes += plen; /* * Test if we should restart the measuring interval */ if ((x->bm_flags & BW_METER_UNIT_PACKETS && x->bm_measured.b_packets <= x->bm_threshold.b_packets) || (x->bm_flags & BW_METER_UNIT_BYTES && x->bm_measured.b_bytes <= x->bm_threshold.b_bytes)) { /* Don't restart the measuring interval */ } else { /* Do restart the measuring interval */ /* * XXX: note that we don't unschedule and schedule, because this * might be too much overhead per packet. Instead, when we process * all entries for a given timer hash bin, we check whether it is * really a timeout. If not, we reschedule at that time. */ x->bm_start_time = *nowp; x->bm_measured.b_packets = 0; x->bm_measured.b_bytes = 0; x->bm_flags &= ~BW_METER_UPCALL_DELIVERED; } } } /* * Prepare a bandwidth-related upcall */ static void bw_meter_prepare_upcall(struct bw_meter *x, struct timeval *nowp) { struct timeval delta; struct bw_upcall *u; MFC_LOCK_ASSERT(); /* * Compute the measured time interval */ delta = *nowp; BW_TIMEVALDECR(&delta, &x->bm_start_time); /* * If there are too many pending upcalls, deliver them now */ if (V_bw_upcalls_n >= BW_UPCALLS_MAX) bw_upcalls_send(); /* * Set the bw_upcall entry */ u = &V_bw_upcalls[V_bw_upcalls_n++]; u->bu_src = x->bm_mfc->mfc_origin; u->bu_dst = x->bm_mfc->mfc_mcastgrp; u->bu_threshold.b_time = x->bm_threshold.b_time; u->bu_threshold.b_packets = x->bm_threshold.b_packets; u->bu_threshold.b_bytes = x->bm_threshold.b_bytes; u->bu_measured.b_time = delta; u->bu_measured.b_packets = x->bm_measured.b_packets; u->bu_measured.b_bytes = x->bm_measured.b_bytes; u->bu_flags = 0; if (x->bm_flags & BW_METER_UNIT_PACKETS) u->bu_flags |= BW_UPCALL_UNIT_PACKETS; if (x->bm_flags & BW_METER_UNIT_BYTES) u->bu_flags |= BW_UPCALL_UNIT_BYTES; if (x->bm_flags & BW_METER_GEQ) u->bu_flags |= BW_UPCALL_GEQ; if (x->bm_flags & BW_METER_LEQ) u->bu_flags |= BW_UPCALL_LEQ; } /* * Send the pending bandwidth-related upcalls */ static void bw_upcalls_send(void) { struct mbuf *m; int len = V_bw_upcalls_n * sizeof(V_bw_upcalls[0]); struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET }; static struct igmpmsg igmpmsg = { 0, /* unused1 */ 0, /* unused2 */ IGMPMSG_BW_UPCALL,/* im_msgtype */ 0, /* im_mbz */ 0, /* im_vif */ 0, /* unused3 */ { 0 }, /* im_src */ { 0 } }; /* im_dst */ MFC_LOCK_ASSERT(); if (V_bw_upcalls_n == 0) return; /* No pending upcalls */ V_bw_upcalls_n = 0; /* * Allocate a new mbuf, initialize it with the header and * the payload for the pending calls. */ m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { log(LOG_WARNING, "bw_upcalls_send: cannot allocate mbuf\n"); return; } m_copyback(m, 0, sizeof(struct igmpmsg), (caddr_t)&igmpmsg); m_copyback(m, sizeof(struct igmpmsg), len, (caddr_t)&V_bw_upcalls[0]); /* * Send the upcalls * XXX do we need to set the address in k_igmpsrc ? */ MRTSTAT_INC(mrts_upcalls); if (socket_send(V_ip_mrouter, m, &k_igmpsrc) < 0) { log(LOG_WARNING, "bw_upcalls_send: ip_mrouter socket queue full\n"); MRTSTAT_INC(mrts_upq_sockfull); } } /* * Compute the timeout hash value for the bw_meter entries */ #define BW_METER_TIMEHASH(bw_meter, hash) \ do { \ struct timeval next_timeval = (bw_meter)->bm_start_time; \ \ BW_TIMEVALADD(&next_timeval, &(bw_meter)->bm_threshold.b_time); \ (hash) = next_timeval.tv_sec; \ if (next_timeval.tv_usec) \ (hash)++; /* XXX: make sure we don't timeout early */ \ (hash) %= BW_METER_BUCKETS; \ } while (0) /* * Schedule a timer to process periodically bw_meter entry of type "<=" * by linking the entry in the proper hash bucket. */ static void schedule_bw_meter(struct bw_meter *x, struct timeval *nowp) { int time_hash; MFC_LOCK_ASSERT(); if (!(x->bm_flags & BW_METER_LEQ)) return; /* XXX: we schedule timers only for "<=" entries */ /* * Reset the bw_meter entry */ x->bm_start_time = *nowp; x->bm_measured.b_packets = 0; x->bm_measured.b_bytes = 0; x->bm_flags &= ~BW_METER_UPCALL_DELIVERED; /* * Compute the timeout hash value and insert the entry */ BW_METER_TIMEHASH(x, time_hash); x->bm_time_next = V_bw_meter_timers[time_hash]; V_bw_meter_timers[time_hash] = x; x->bm_time_hash = time_hash; } /* * Unschedule the periodic timer that processes bw_meter entry of type "<=" * by removing the entry from the proper hash bucket. */ static void unschedule_bw_meter(struct bw_meter *x) { int time_hash; struct bw_meter *prev, *tmp; MFC_LOCK_ASSERT(); if (!(x->bm_flags & BW_METER_LEQ)) return; /* XXX: we schedule timers only for "<=" entries */ /* * Compute the timeout hash value and delete the entry */ time_hash = x->bm_time_hash; if (time_hash >= BW_METER_BUCKETS) return; /* Entry was not scheduled */ for (prev = NULL, tmp = V_bw_meter_timers[time_hash]; tmp != NULL; prev = tmp, tmp = tmp->bm_time_next) if (tmp == x) break; if (tmp == NULL) panic("unschedule_bw_meter: bw_meter entry not found"); if (prev != NULL) prev->bm_time_next = x->bm_time_next; else V_bw_meter_timers[time_hash] = x->bm_time_next; x->bm_time_next = NULL; x->bm_time_hash = BW_METER_BUCKETS; } /* * Process all "<=" type of bw_meter that should be processed now, * and for each entry prepare an upcall if necessary. Each processed * entry is rescheduled again for the (periodic) processing. * * This is run periodically (once per second normally). On each round, * all the potentially matching entries are in the hash slot that we are * looking at. */ static void bw_meter_process() { uint32_t loops; int i; struct timeval now, process_endtime; microtime(&now); if (V_last_tv_sec == now.tv_sec) return; /* nothing to do */ loops = now.tv_sec - V_last_tv_sec; V_last_tv_sec = now.tv_sec; if (loops > BW_METER_BUCKETS) loops = BW_METER_BUCKETS; MFC_LOCK(); /* * Process all bins of bw_meter entries from the one after the last * processed to the current one. On entry, i points to the last bucket * visited, so we need to increment i at the beginning of the loop. */ for (i = (now.tv_sec - loops) % BW_METER_BUCKETS; loops > 0; loops--) { struct bw_meter *x, *tmp_list; if (++i >= BW_METER_BUCKETS) i = 0; /* Disconnect the list of bw_meter entries from the bin */ tmp_list = V_bw_meter_timers[i]; V_bw_meter_timers[i] = NULL; /* Process the list of bw_meter entries */ while (tmp_list != NULL) { x = tmp_list; tmp_list = tmp_list->bm_time_next; /* Test if the time interval is over */ process_endtime = x->bm_start_time; BW_TIMEVALADD(&process_endtime, &x->bm_threshold.b_time); if (BW_TIMEVALCMP(&process_endtime, &now, >)) { /* Not yet: reschedule, but don't reset */ int time_hash; BW_METER_TIMEHASH(x, time_hash); if (time_hash == i && process_endtime.tv_sec == now.tv_sec) { /* * XXX: somehow the bin processing is a bit ahead of time. * Put the entry in the next bin. */ if (++time_hash >= BW_METER_BUCKETS) time_hash = 0; } x->bm_time_next = V_bw_meter_timers[time_hash]; V_bw_meter_timers[time_hash] = x; x->bm_time_hash = time_hash; continue; } /* * Test if we should deliver an upcall */ if (((x->bm_flags & BW_METER_UNIT_PACKETS) && (x->bm_measured.b_packets <= x->bm_threshold.b_packets)) || ((x->bm_flags & BW_METER_UNIT_BYTES) && (x->bm_measured.b_bytes <= x->bm_threshold.b_bytes))) { /* Prepare an upcall for delivery */ bw_meter_prepare_upcall(x, &now); } /* * Reschedule for next processing */ schedule_bw_meter(x, &now); } } /* Send all upcalls that are pending delivery */ bw_upcalls_send(); MFC_UNLOCK(); } /* * A periodic function for sending all upcalls that are pending delivery */ static void expire_bw_upcalls_send(void *arg) { CURVNET_SET((struct vnet *) arg); MFC_LOCK(); bw_upcalls_send(); MFC_UNLOCK(); callout_reset(&V_bw_upcalls_ch, BW_UPCALLS_PERIOD, expire_bw_upcalls_send, curvnet); CURVNET_RESTORE(); } /* * A periodic function for periodic scanning of the multicast forwarding * table for processing all "<=" bw_meter entries. */ static void expire_bw_meter_process(void *arg) { CURVNET_SET((struct vnet *) arg); if (V_mrt_api_config & MRT_MFC_BW_UPCALL) bw_meter_process(); callout_reset(&V_bw_meter_ch, BW_METER_PERIOD, expire_bw_meter_process, curvnet); CURVNET_RESTORE(); } /* * End of bandwidth monitoring code */ /* * Send the packet up to the user daemon, or eventually do kernel encapsulation * */ static int pim_register_send(struct ip *ip, struct vif *vifp, struct mbuf *m, struct mfc *rt) { struct mbuf *mb_copy, *mm; /* * Do not send IGMP_WHOLEPKT notifications to userland, if the * rendezvous point was unspecified, and we were told not to. */ if (pim_squelch_wholepkt != 0 && (V_mrt_api_config & MRT_MFC_RP) && in_nullhost(rt->mfc_rp)) return 0; mb_copy = pim_register_prepare(ip, m); if (mb_copy == NULL) return ENOBUFS; /* * Send all the fragments. Note that the mbuf for each fragment * is freed by the sending machinery. */ for (mm = mb_copy; mm; mm = mb_copy) { mb_copy = mm->m_nextpkt; mm->m_nextpkt = 0; mm = m_pullup(mm, sizeof(struct ip)); if (mm != NULL) { ip = mtod(mm, struct ip *); if ((V_mrt_api_config & MRT_MFC_RP) && !in_nullhost(rt->mfc_rp)) { pim_register_send_rp(ip, vifp, mm, rt); } else { pim_register_send_upcall(ip, vifp, mm, rt); } } } return 0; } /* * Return a copy of the data packet that is ready for PIM Register * encapsulation. * XXX: Note that in the returned copy the IP header is a valid one. */ static struct mbuf * pim_register_prepare(struct ip *ip, struct mbuf *m) { struct mbuf *mb_copy = NULL; int mtu; /* Take care of delayed checksums */ if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { in_delayed_cksum(m); m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } /* * Copy the old packet & pullup its IP header into the * new mbuf so we can modify it. */ mb_copy = m_copypacket(m, M_NOWAIT); if (mb_copy == NULL) return NULL; mb_copy = m_pullup(mb_copy, ip->ip_hl << 2); if (mb_copy == NULL) return NULL; /* take care of the TTL */ ip = mtod(mb_copy, struct ip *); --ip->ip_ttl; /* Compute the MTU after the PIM Register encapsulation */ mtu = 0xffff - sizeof(pim_encap_iphdr) - sizeof(pim_encap_pimhdr); if (ntohs(ip->ip_len) <= mtu) { /* Turn the IP header into a valid one */ ip->ip_sum = 0; ip->ip_sum = in_cksum(mb_copy, ip->ip_hl << 2); } else { /* Fragment the packet */ mb_copy->m_pkthdr.csum_flags |= CSUM_IP; if (ip_fragment(ip, &mb_copy, mtu, 0) != 0) { m_freem(mb_copy); return NULL; } } return mb_copy; } /* * Send an upcall with the data packet to the user-level process. */ static int pim_register_send_upcall(struct ip *ip, struct vif *vifp, struct mbuf *mb_copy, struct mfc *rt) { struct mbuf *mb_first; int len = ntohs(ip->ip_len); struct igmpmsg *im; struct sockaddr_in k_igmpsrc = { sizeof k_igmpsrc, AF_INET }; VIF_LOCK_ASSERT(); /* * Add a new mbuf with an upcall header */ mb_first = m_gethdr(M_NOWAIT, MT_DATA); if (mb_first == NULL) { m_freem(mb_copy); return ENOBUFS; } mb_first->m_data += max_linkhdr; mb_first->m_pkthdr.len = len + sizeof(struct igmpmsg); mb_first->m_len = sizeof(struct igmpmsg); mb_first->m_next = mb_copy; /* Send message to routing daemon */ im = mtod(mb_first, struct igmpmsg *); im->im_msgtype = IGMPMSG_WHOLEPKT; im->im_mbz = 0; im->im_vif = vifp - V_viftable; im->im_src = ip->ip_src; im->im_dst = ip->ip_dst; k_igmpsrc.sin_addr = ip->ip_src; MRTSTAT_INC(mrts_upcalls); if (socket_send(V_ip_mrouter, mb_first, &k_igmpsrc) < 0) { CTR1(KTR_IPMF, "%s: socket queue full", __func__); MRTSTAT_INC(mrts_upq_sockfull); return ENOBUFS; } /* Keep statistics */ PIMSTAT_INC(pims_snd_registers_msgs); PIMSTAT_ADD(pims_snd_registers_bytes, len); return 0; } /* * Encapsulate the data packet in PIM Register message and send it to the RP. */ static int pim_register_send_rp(struct ip *ip, struct vif *vifp, struct mbuf *mb_copy, struct mfc *rt) { struct mbuf *mb_first; struct ip *ip_outer; struct pim_encap_pimhdr *pimhdr; int len = ntohs(ip->ip_len); vifi_t vifi = rt->mfc_parent; VIF_LOCK_ASSERT(); if ((vifi >= V_numvifs) || in_nullhost(V_viftable[vifi].v_lcl_addr)) { m_freem(mb_copy); return EADDRNOTAVAIL; /* The iif vif is invalid */ } /* * Add a new mbuf with the encapsulating header */ mb_first = m_gethdr(M_NOWAIT, MT_DATA); if (mb_first == NULL) { m_freem(mb_copy); return ENOBUFS; } mb_first->m_data += max_linkhdr; mb_first->m_len = sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr); mb_first->m_next = mb_copy; mb_first->m_pkthdr.len = len + mb_first->m_len; /* * Fill in the encapsulating IP and PIM header */ ip_outer = mtod(mb_first, struct ip *); *ip_outer = pim_encap_iphdr; ip_outer->ip_len = htons(len + sizeof(pim_encap_iphdr) + sizeof(pim_encap_pimhdr)); ip_outer->ip_src = V_viftable[vifi].v_lcl_addr; ip_outer->ip_dst = rt->mfc_rp; /* * Copy the inner header TOS to the outer header, and take care of the * IP_DF bit. */ ip_outer->ip_tos = ip->ip_tos; if (ip->ip_off & htons(IP_DF)) ip_outer->ip_off |= htons(IP_DF); ip_fillid(ip_outer); pimhdr = (struct pim_encap_pimhdr *)((caddr_t)ip_outer + sizeof(pim_encap_iphdr)); *pimhdr = pim_encap_pimhdr; /* If the iif crosses a border, set the Border-bit */ if (rt->mfc_flags[vifi] & MRT_MFC_FLAGS_BORDER_VIF & V_mrt_api_config) pimhdr->flags |= htonl(PIM_BORDER_REGISTER); mb_first->m_data += sizeof(pim_encap_iphdr); pimhdr->pim.pim_cksum = in_cksum(mb_first, sizeof(pim_encap_pimhdr)); mb_first->m_data -= sizeof(pim_encap_iphdr); send_packet(vifp, mb_first); /* Keep statistics */ PIMSTAT_INC(pims_snd_registers_msgs); PIMSTAT_ADD(pims_snd_registers_bytes, len); return 0; } /* * pim_encapcheck() is called by the encap4_input() path at runtime to * determine if a packet is for PIM; allowing PIM to be dynamically loaded * into the kernel. */ static int pim_encapcheck(const struct mbuf *m __unused, int off __unused, int proto __unused, void *arg __unused) { KASSERT(proto == IPPROTO_PIM, ("not for IPPROTO_PIM")); return (8); /* claim the datagram. */ } /* * PIM-SMv2 and PIM-DM messages processing. * Receives and verifies the PIM control messages, and passes them * up to the listening socket, using rip_input(). * The only message with special processing is the PIM_REGISTER message * (used by PIM-SM): the PIM header is stripped off, and the inner packet * is passed to if_simloop(). */ static int pim_input(struct mbuf *m, int off, int proto, void *arg __unused) { struct ip *ip = mtod(m, struct ip *); struct pim *pim; int iphlen = off; int minlen; int datalen = ntohs(ip->ip_len) - iphlen; int ip_tos; /* Keep statistics */ PIMSTAT_INC(pims_rcv_total_msgs); PIMSTAT_ADD(pims_rcv_total_bytes, datalen); /* * Validate lengths */ if (datalen < PIM_MINLEN) { PIMSTAT_INC(pims_rcv_tooshort); CTR3(KTR_IPMF, "%s: short packet (%d) from 0x%08x", __func__, datalen, ntohl(ip->ip_src.s_addr)); m_freem(m); return (IPPROTO_DONE); } /* * If the packet is at least as big as a REGISTER, go agead * and grab the PIM REGISTER header size, to avoid another * possible m_pullup() later. * * PIM_MINLEN == pimhdr + u_int32_t == 4 + 4 = 8 * PIM_REG_MINLEN == pimhdr + reghdr + encap_iphdr == 4 + 4 + 20 = 28 */ minlen = iphlen + (datalen >= PIM_REG_MINLEN ? PIM_REG_MINLEN : PIM_MINLEN); /* * Get the IP and PIM headers in contiguous memory, and * possibly the PIM REGISTER header. */ if (m->m_len < minlen && (m = m_pullup(m, minlen)) == NULL) { CTR1(KTR_IPMF, "%s: m_pullup() failed", __func__); return (IPPROTO_DONE); } /* m_pullup() may have given us a new mbuf so reset ip. */ ip = mtod(m, struct ip *); ip_tos = ip->ip_tos; /* adjust mbuf to point to the PIM header */ m->m_data += iphlen; m->m_len -= iphlen; pim = mtod(m, struct pim *); /* * Validate checksum. If PIM REGISTER, exclude the data packet. * * XXX: some older PIMv2 implementations don't make this distinction, * so for compatibility reason perform the checksum over part of the * message, and if error, then over the whole message. */ if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER && in_cksum(m, PIM_MINLEN) == 0) { /* do nothing, checksum okay */ } else if (in_cksum(m, datalen)) { PIMSTAT_INC(pims_rcv_badsum); CTR1(KTR_IPMF, "%s: invalid checksum", __func__); m_freem(m); return (IPPROTO_DONE); } /* PIM version check */ if (PIM_VT_V(pim->pim_vt) < PIM_VERSION) { PIMSTAT_INC(pims_rcv_badversion); CTR3(KTR_IPMF, "%s: bad version %d expect %d", __func__, (int)PIM_VT_V(pim->pim_vt), PIM_VERSION); m_freem(m); return (IPPROTO_DONE); } /* restore mbuf back to the outer IP */ m->m_data -= iphlen; m->m_len += iphlen; if (PIM_VT_T(pim->pim_vt) == PIM_REGISTER) { /* * Since this is a REGISTER, we'll make a copy of the register * headers ip + pim + u_int32 + encap_ip, to be passed up to the * routing daemon. */ struct sockaddr_in dst = { sizeof(dst), AF_INET }; struct mbuf *mcp; struct ip *encap_ip; u_int32_t *reghdr; struct ifnet *vifp; VIF_LOCK(); if ((V_reg_vif_num >= V_numvifs) || (V_reg_vif_num == VIFI_INVALID)) { VIF_UNLOCK(); CTR2(KTR_IPMF, "%s: register vif not set: %d", __func__, (int)V_reg_vif_num); m_freem(m); return (IPPROTO_DONE); } /* XXX need refcnt? */ vifp = V_viftable[V_reg_vif_num].v_ifp; VIF_UNLOCK(); /* * Validate length */ if (datalen < PIM_REG_MINLEN) { PIMSTAT_INC(pims_rcv_tooshort); PIMSTAT_INC(pims_rcv_badregisters); CTR1(KTR_IPMF, "%s: register packet size too small", __func__); m_freem(m); return (IPPROTO_DONE); } reghdr = (u_int32_t *)(pim + 1); encap_ip = (struct ip *)(reghdr + 1); CTR3(KTR_IPMF, "%s: register: encap ip src 0x%08x len %d", __func__, ntohl(encap_ip->ip_src.s_addr), ntohs(encap_ip->ip_len)); /* verify the version number of the inner packet */ if (encap_ip->ip_v != IPVERSION) { PIMSTAT_INC(pims_rcv_badregisters); CTR1(KTR_IPMF, "%s: bad encap ip version", __func__); m_freem(m); return (IPPROTO_DONE); } /* verify the inner packet is destined to a mcast group */ if (!IN_MULTICAST(ntohl(encap_ip->ip_dst.s_addr))) { PIMSTAT_INC(pims_rcv_badregisters); CTR2(KTR_IPMF, "%s: bad encap ip dest 0x%08x", __func__, ntohl(encap_ip->ip_dst.s_addr)); m_freem(m); return (IPPROTO_DONE); } /* If a NULL_REGISTER, pass it to the daemon */ if ((ntohl(*reghdr) & PIM_NULL_REGISTER)) goto pim_input_to_daemon; /* * Copy the TOS from the outer IP header to the inner IP header. */ if (encap_ip->ip_tos != ip_tos) { /* Outer TOS -> inner TOS */ encap_ip->ip_tos = ip_tos; /* Recompute the inner header checksum. Sigh... */ /* adjust mbuf to point to the inner IP header */ m->m_data += (iphlen + PIM_MINLEN); m->m_len -= (iphlen + PIM_MINLEN); encap_ip->ip_sum = 0; encap_ip->ip_sum = in_cksum(m, encap_ip->ip_hl << 2); /* restore mbuf to point back to the outer IP header */ m->m_data -= (iphlen + PIM_MINLEN); m->m_len += (iphlen + PIM_MINLEN); } /* * Decapsulate the inner IP packet and loopback to forward it * as a normal multicast packet. Also, make a copy of the * outer_iphdr + pimhdr + reghdr + encap_iphdr * to pass to the daemon later, so it can take the appropriate * actions (e.g., send back PIM_REGISTER_STOP). * XXX: here m->m_data points to the outer IP header. */ mcp = m_copym(m, 0, iphlen + PIM_REG_MINLEN, M_NOWAIT); if (mcp == NULL) { CTR1(KTR_IPMF, "%s: m_copym() failed", __func__); m_freem(m); return (IPPROTO_DONE); } /* Keep statistics */ /* XXX: registers_bytes include only the encap. mcast pkt */ PIMSTAT_INC(pims_rcv_registers_msgs); PIMSTAT_ADD(pims_rcv_registers_bytes, ntohs(encap_ip->ip_len)); /* * forward the inner ip packet; point m_data at the inner ip. */ m_adj(m, iphlen + PIM_MINLEN); CTR4(KTR_IPMF, "%s: forward decap'd REGISTER: src %lx dst %lx vif %d", __func__, (u_long)ntohl(encap_ip->ip_src.s_addr), (u_long)ntohl(encap_ip->ip_dst.s_addr), (int)V_reg_vif_num); /* NB: vifp was collected above; can it change on us? */ if_simloop(vifp, m, dst.sin_family, 0); /* prepare the register head to send to the mrouting daemon */ m = mcp; } pim_input_to_daemon: /* * Pass the PIM message up to the daemon; if it is a Register message, * pass the 'head' only up to the daemon. This includes the * outer IP header, PIM header, PIM-Register header and the * inner IP header. * XXX: the outer IP header pkt size of a Register is not adjust to * reflect the fact that the inner multicast data is truncated. */ return (rip_input(&m, &off, proto)); } static int sysctl_mfctable(SYSCTL_HANDLER_ARGS) { struct mfc *rt; int error, i; if (req->newptr) return (EPERM); if (V_mfchashtbl == NULL) /* XXX unlocked */ return (0); error = sysctl_wire_old_buffer(req, 0); if (error) return (error); MFC_LOCK(); for (i = 0; i < mfchashsize; i++) { LIST_FOREACH(rt, &V_mfchashtbl[i], mfc_hash) { error = SYSCTL_OUT(req, rt, sizeof(struct mfc)); if (error) goto out_locked; } } out_locked: MFC_UNLOCK(); return (error); } static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mfctable, CTLFLAG_RD, sysctl_mfctable, "IPv4 Multicast Forwarding Table " "(struct *mfc[mfchashsize], netinet/ip_mroute.h)"); static void vnet_mroute_init(const void *unused __unused) { V_nexpire = malloc(mfchashsize, M_MRTABLE, M_WAITOK|M_ZERO); bzero(V_bw_meter_timers, sizeof(V_bw_meter_timers)); callout_init(&V_expire_upcalls_ch, 1); callout_init(&V_bw_upcalls_ch, 1); callout_init(&V_bw_meter_ch, 1); } VNET_SYSINIT(vnet_mroute_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mroute_init, NULL); static void vnet_mroute_uninit(const void *unused __unused) { free(V_nexpire, M_MRTABLE); V_nexpire = NULL; } VNET_SYSUNINIT(vnet_mroute_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, vnet_mroute_uninit, NULL); static int ip_mroute_modevent(module_t mod, int type, void *unused) { switch (type) { case MOD_LOAD: MROUTER_LOCK_INIT(); if_detach_event_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, if_detached_event, NULL, EVENTHANDLER_PRI_ANY); if (if_detach_event_tag == NULL) { printf("ip_mroute: unable to register " "ifnet_departure_event handler\n"); MROUTER_LOCK_DESTROY(); return (EINVAL); } MFC_LOCK_INIT(); VIF_LOCK_INIT(); mfchashsize = MFCHASHSIZE; if (TUNABLE_ULONG_FETCH("net.inet.ip.mfchashsize", &mfchashsize) && !powerof2(mfchashsize)) { printf("WARNING: %s not a power of 2; using default\n", "net.inet.ip.mfchashsize"); mfchashsize = MFCHASHSIZE; } pim_squelch_wholepkt = 0; TUNABLE_ULONG_FETCH("net.inet.pim.squelch_wholepkt", &pim_squelch_wholepkt); pim_encap_cookie = ip_encap_attach(&ipv4_encap_cfg, NULL, M_WAITOK); if (pim_encap_cookie == NULL) { printf("ip_mroute: unable to attach pim encap\n"); VIF_LOCK_DESTROY(); MFC_LOCK_DESTROY(); MROUTER_LOCK_DESTROY(); return (EINVAL); } ip_mcast_src = X_ip_mcast_src; ip_mforward = X_ip_mforward; ip_mrouter_done = X_ip_mrouter_done; ip_mrouter_get = X_ip_mrouter_get; ip_mrouter_set = X_ip_mrouter_set; ip_rsvp_force_done = X_ip_rsvp_force_done; ip_rsvp_vif = X_ip_rsvp_vif; legal_vif_num = X_legal_vif_num; mrt_ioctl = X_mrt_ioctl; rsvp_input_p = X_rsvp_input; break; case MOD_UNLOAD: /* * Typically module unload happens after the user-level * process has shutdown the kernel services (the check * below insures someone can't just yank the module out * from under a running process). But if the module is * just loaded and then unloaded w/o starting up a user * process we still need to cleanup. */ MROUTER_LOCK(); if (ip_mrouter_cnt != 0) { MROUTER_UNLOCK(); return (EINVAL); } ip_mrouter_unloading = 1; MROUTER_UNLOCK(); EVENTHANDLER_DEREGISTER(ifnet_departure_event, if_detach_event_tag); if (pim_encap_cookie) { ip_encap_detach(pim_encap_cookie); pim_encap_cookie = NULL; } ip_mcast_src = NULL; ip_mforward = NULL; ip_mrouter_done = NULL; ip_mrouter_get = NULL; ip_mrouter_set = NULL; ip_rsvp_force_done = NULL; ip_rsvp_vif = NULL; legal_vif_num = NULL; mrt_ioctl = NULL; rsvp_input_p = NULL; VIF_LOCK_DESTROY(); MFC_LOCK_DESTROY(); MROUTER_LOCK_DESTROY(); break; default: return EOPNOTSUPP; } return 0; } static moduledata_t ip_mroutemod = { "ip_mroute", ip_mroute_modevent, 0 }; DECLARE_MODULE(ip_mroute, ip_mroutemod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); Index: head/sys/netinet/ip_options.c =================================================================== --- head/sys/netinet/ip_options.c (revision 336675) +++ head/sys/netinet/ip_options.c (revision 336676) @@ -1,763 +1,763 @@ /* * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. * Copyright (c) 2005 Andre Oppermann, Internet Business Solutions AG. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_ipstealth.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include -static VNET_DEFINE(int, ip_dosourceroute); +VNET_DEFINE_STATIC(int, ip_dosourceroute); SYSCTL_INT(_net_inet_ip, IPCTL_SOURCEROUTE, sourceroute, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_dosourceroute), 0, "Enable forwarding source routed IP packets"); #define V_ip_dosourceroute VNET(ip_dosourceroute) -static VNET_DEFINE(int, ip_acceptsourceroute); +VNET_DEFINE_STATIC(int, ip_acceptsourceroute); SYSCTL_INT(_net_inet_ip, IPCTL_ACCEPTSOURCEROUTE, accept_sourceroute, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_acceptsourceroute), 0, "Enable accepting source routed IP packets"); #define V_ip_acceptsourceroute VNET(ip_acceptsourceroute) VNET_DEFINE(int, ip_doopts) = 1; /* 0 = ignore, 1 = process, 2 = reject */ SYSCTL_INT(_net_inet_ip, OID_AUTO, process_options, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_doopts), 0, "Enable IP options processing ([LS]SRR, RR, TS)"); static void save_rte(struct mbuf *m, u_char *, struct in_addr); /* * Do option processing on a datagram, possibly discarding it if bad options * are encountered, or forwarding it if source-routed. * * The pass argument is used when operating in the IPSTEALTH mode to tell * what options to process: [LS]SRR (pass 0) or the others (pass 1). The * reason for as many as two passes is that when doing IPSTEALTH, non-routing * options should be processed only if the packet is for us. * * Returns 1 if packet has been forwarded/freed, 0 if the packet should be * processed further. */ int ip_dooptions(struct mbuf *m, int pass) { struct ip *ip = mtod(m, struct ip *); u_char *cp; struct in_ifaddr *ia; int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB, forward = 0; struct in_addr *sin, dst; uint32_t ntime; struct nhop4_extended nh_ext; struct sockaddr_in ipaddr = { sizeof(ipaddr), AF_INET }; NET_EPOCH_ENTER(); /* Ignore or reject packets with IP options. */ if (V_ip_doopts == 0) return 0; else if (V_ip_doopts == 2) { type = ICMP_UNREACH; code = ICMP_UNREACH_FILTER_PROHIB; goto bad; } dst = ip->ip_dst; cp = (u_char *)(ip + 1); cnt = (ip->ip_hl << 2) - sizeof (struct ip); for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[IPOPT_OPTVAL]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) optlen = 1; else { if (cnt < IPOPT_OLEN + sizeof(*cp)) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; } optlen = cp[IPOPT_OLEN]; if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; } } switch (opt) { default: break; /* * Source routing with record. Find interface with current * destination address. If none on this machine then drop if * strictly routed, or do nothing if loosely routed. Record * interface address and bring up next address component. If * strictly routed make sure next address is on directly * accessible net. */ case IPOPT_LSRR: case IPOPT_SSRR: #ifdef IPSTEALTH if (V_ipstealth && pass > 0) break; #endif if (optlen < IPOPT_OFFSET + sizeof(*cp)) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; } if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } ipaddr.sin_addr = ip->ip_dst; if (ifa_ifwithaddr_check((struct sockaddr *)&ipaddr) == 0) { if (opt == IPOPT_SSRR) { type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; goto bad; } if (!V_ip_dosourceroute) goto nosourcerouting; /* * Loose routing, and not at next destination * yet; nothing to do except forward. */ break; } off--; /* 0 origin */ if (off > optlen - (int)sizeof(struct in_addr)) { /* * End of source route. Should be for us. */ if (!V_ip_acceptsourceroute) goto nosourcerouting; save_rte(m, cp, ip->ip_src); break; } #ifdef IPSTEALTH if (V_ipstealth) goto dropit; #endif if (!V_ip_dosourceroute) { if (V_ipforwarding) { char srcbuf[INET_ADDRSTRLEN]; char dstbuf[INET_ADDRSTRLEN]; /* * Acting as a router, so generate * ICMP */ nosourcerouting: log(LOG_WARNING, "attempted source route from %s " "to %s\n", inet_ntoa_r(ip->ip_src, srcbuf), inet_ntoa_r(ip->ip_dst, dstbuf)); type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; goto bad; } else { /* * Not acting as a router, so * silently drop. */ #ifdef IPSTEALTH dropit: #endif IPSTAT_INC(ips_cantforward); m_freem(m); NET_EPOCH_EXIT(); return (1); } } /* * locate outgoing interface */ (void)memcpy(&ipaddr.sin_addr, cp + off, sizeof(ipaddr.sin_addr)); type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; if (opt == IPOPT_SSRR) { #define INA struct in_ifaddr * #define SA struct sockaddr * ia = (INA)ifa_ifwithdstaddr((SA)&ipaddr, RT_ALL_FIBS); if (ia == NULL) ia = (INA)ifa_ifwithnet((SA)&ipaddr, 0, RT_ALL_FIBS); if (ia == NULL) goto bad; memcpy(cp + off, &(IA_SIN(ia)->sin_addr), sizeof(struct in_addr)); } else { /* XXX MRT 0 for routing */ if (fib4_lookup_nh_ext(M_GETFIB(m), ipaddr.sin_addr, 0, 0, &nh_ext) != 0) goto bad; memcpy(cp + off, &nh_ext.nh_src, sizeof(struct in_addr)); } ip->ip_dst = ipaddr.sin_addr; cp[IPOPT_OFFSET] += sizeof(struct in_addr); /* * Let ip_intr's mcast routing check handle mcast pkts */ forward = !IN_MULTICAST(ntohl(ip->ip_dst.s_addr)); break; case IPOPT_RR: #ifdef IPSTEALTH if (V_ipstealth && pass == 0) break; #endif if (optlen < IPOPT_OFFSET + sizeof(*cp)) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } if ((off = cp[IPOPT_OFFSET]) < IPOPT_MINOFF) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } /* * If no space remains, ignore. */ off--; /* 0 origin */ if (off > optlen - (int)sizeof(struct in_addr)) break; (void)memcpy(&ipaddr.sin_addr, &ip->ip_dst, sizeof(ipaddr.sin_addr)); /* * Locate outgoing interface; if we're the * destination, use the incoming interface (should be * same). */ if ((ia = (INA)ifa_ifwithaddr((SA)&ipaddr)) != NULL) { memcpy(cp + off, &(IA_SIN(ia)->sin_addr), sizeof(struct in_addr)); } else if (fib4_lookup_nh_ext(M_GETFIB(m), ipaddr.sin_addr, 0, 0, &nh_ext) == 0) { memcpy(cp + off, &nh_ext.nh_src, sizeof(struct in_addr)); } else { type = ICMP_UNREACH; code = ICMP_UNREACH_HOST; goto bad; } cp[IPOPT_OFFSET] += sizeof(struct in_addr); break; case IPOPT_TS: #ifdef IPSTEALTH if (V_ipstealth && pass == 0) break; #endif code = cp - (u_char *)ip; if (optlen < 4 || optlen > 40) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; } if ((off = cp[IPOPT_OFFSET]) < 5) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; } if (off > optlen - (int)sizeof(int32_t)) { cp[IPOPT_OFFSET + 1] += (1 << 4); if ((cp[IPOPT_OFFSET + 1] & 0xf0) == 0) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } break; } off--; /* 0 origin */ sin = (struct in_addr *)(cp + off); switch (cp[IPOPT_OFFSET + 1] & 0x0f) { case IPOPT_TS_TSONLY: break; case IPOPT_TS_TSANDADDR: if (off + sizeof(uint32_t) + sizeof(struct in_addr) > optlen) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } ipaddr.sin_addr = dst; ia = (INA)ifaof_ifpforaddr((SA)&ipaddr, m->m_pkthdr.rcvif); if (ia == NULL) continue; (void)memcpy(sin, &IA_SIN(ia)->sin_addr, sizeof(struct in_addr)); cp[IPOPT_OFFSET] += sizeof(struct in_addr); off += sizeof(struct in_addr); break; case IPOPT_TS_PRESPEC: if (off + sizeof(uint32_t) + sizeof(struct in_addr) > optlen) { code = &cp[IPOPT_OFFSET] - (u_char *)ip; goto bad; } (void)memcpy(&ipaddr.sin_addr, sin, sizeof(struct in_addr)); if (ifa_ifwithaddr_check((SA)&ipaddr) == 0) continue; cp[IPOPT_OFFSET] += sizeof(struct in_addr); off += sizeof(struct in_addr); break; default: code = &cp[IPOPT_OFFSET + 1] - (u_char *)ip; goto bad; } ntime = iptime(); (void)memcpy(cp + off, &ntime, sizeof(uint32_t)); cp[IPOPT_OFFSET] += sizeof(uint32_t); } } NET_EPOCH_EXIT(); if (forward && V_ipforwarding) { ip_forward(m, 1); return (1); } return (0); bad: NET_EPOCH_EXIT(); icmp_error(m, type, code, 0, 0); IPSTAT_INC(ips_badoptions); return (1); } /* * Save incoming source route for use in replies, to be picked up later by * ip_srcroute if the receiver is interested. */ static void save_rte(struct mbuf *m, u_char *option, struct in_addr dst) { unsigned olen; struct ipopt_tag *opts; opts = (struct ipopt_tag *)m_tag_get(PACKET_TAG_IPOPTIONS, sizeof(struct ipopt_tag), M_NOWAIT); if (opts == NULL) return; olen = option[IPOPT_OLEN]; if (olen > sizeof(opts->ip_srcrt) - (1 + sizeof(dst))) { m_tag_free((struct m_tag *)opts); return; } bcopy(option, opts->ip_srcrt.srcopt, olen); opts->ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr); opts->ip_srcrt.dst = dst; m_tag_prepend(m, (struct m_tag *)opts); } /* * Retrieve incoming source route for use in replies, in the same form used * by setsockopt. The first hop is placed before the options, will be * removed later. */ struct mbuf * ip_srcroute(struct mbuf *m0) { struct in_addr *p, *q; struct mbuf *m; struct ipopt_tag *opts; opts = (struct ipopt_tag *)m_tag_find(m0, PACKET_TAG_IPOPTIONS, NULL); if (opts == NULL) return (NULL); if (opts->ip_nhops == 0) return (NULL); m = m_get(M_NOWAIT, MT_DATA); if (m == NULL) return (NULL); #define OPTSIZ (sizeof(opts->ip_srcrt.nop) + sizeof(opts->ip_srcrt.srcopt)) /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */ m->m_len = opts->ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) + OPTSIZ; /* * First, save first hop for return route. */ p = &(opts->ip_srcrt.route[opts->ip_nhops - 1]); *(mtod(m, struct in_addr *)) = *p--; /* * Copy option fields and padding (nop) to mbuf. */ opts->ip_srcrt.nop = IPOPT_NOP; opts->ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF; (void)memcpy(mtod(m, caddr_t) + sizeof(struct in_addr), &(opts->ip_srcrt.nop), OPTSIZ); q = (struct in_addr *)(mtod(m, caddr_t) + sizeof(struct in_addr) + OPTSIZ); #undef OPTSIZ /* * Record return path as an IP source route, reversing the path * (pointers are now aligned). */ while (p >= opts->ip_srcrt.route) { *q++ = *p--; } /* * Last hop goes to final destination. */ *q = opts->ip_srcrt.dst; m_tag_delete(m0, (struct m_tag *)opts); return (m); } /* * Strip out IP options, at higher level protocol in the kernel. */ void ip_stripoptions(struct mbuf *m) { struct ip *ip = mtod(m, struct ip *); int olen; olen = (ip->ip_hl << 2) - sizeof(struct ip); m->m_len -= olen; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len -= olen; ip->ip_len = htons(ntohs(ip->ip_len) - olen); ip->ip_hl = sizeof(struct ip) >> 2; bcopy((char *)ip + sizeof(struct ip) + olen, (ip + 1), (size_t )(m->m_len - sizeof(struct ip))); } /* * Insert IP options into preformed packet. Adjust IP destination as * required for IP source routing, as indicated by a non-zero in_addr at the * start of the options. * * XXX This routine assumes that the packet has no options in place. */ struct mbuf * ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen) { struct ipoption *p = mtod(opt, struct ipoption *); struct mbuf *n; struct ip *ip = mtod(m, struct ip *); unsigned optlen; optlen = opt->m_len - sizeof(p->ipopt_dst); if (optlen + ntohs(ip->ip_len) > IP_MAXPACKET) { *phlen = 0; return (m); /* XXX should fail */ } if (p->ipopt_dst.s_addr) ip->ip_dst = p->ipopt_dst; if (!M_WRITABLE(m) || M_LEADINGSPACE(m) < optlen) { n = m_gethdr(M_NOWAIT, MT_DATA); if (n == NULL) { *phlen = 0; return (m); } m_move_pkthdr(n, m); n->m_pkthdr.rcvif = NULL; n->m_pkthdr.len += optlen; m->m_len -= sizeof(struct ip); m->m_data += sizeof(struct ip); n->m_next = m; m = n; m->m_len = optlen + sizeof(struct ip); m->m_data += max_linkhdr; bcopy(ip, mtod(m, void *), sizeof(struct ip)); } else { m->m_data -= optlen; m->m_len += optlen; m->m_pkthdr.len += optlen; bcopy(ip, mtod(m, void *), sizeof(struct ip)); } ip = mtod(m, struct ip *); bcopy(p->ipopt_list, ip + 1, optlen); *phlen = sizeof(struct ip) + optlen; ip->ip_v = IPVERSION; ip->ip_hl = *phlen >> 2; ip->ip_len = htons(ntohs(ip->ip_len) + optlen); return (m); } /* * Copy options from ip to jp, omitting those not copied during * fragmentation. */ int ip_optcopy(struct ip *ip, struct ip *jp) { u_char *cp, *dp; int opt, optlen, cnt; cp = (u_char *)(ip + 1); dp = (u_char *)(jp + 1); cnt = (ip->ip_hl << 2) - sizeof (struct ip); for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) { /* Preserve for IP mcast tunnel's LSRR alignment. */ *dp++ = IPOPT_NOP; optlen = 1; continue; } KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp), ("ip_optcopy: malformed ipv4 option")); optlen = cp[IPOPT_OLEN]; KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt, ("ip_optcopy: malformed ipv4 option")); /* Bogus lengths should have been caught by ip_dooptions. */ if (optlen > cnt) optlen = cnt; if (IPOPT_COPIED(opt)) { bcopy(cp, dp, optlen); dp += optlen; } } for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++) *dp++ = IPOPT_EOL; return (optlen); } /* * Set up IP options in pcb for insertion in output packets. Store in mbuf * with pointer in pcbopt, adding pseudo-option with destination address if * source routed. */ int ip_pcbopts(struct inpcb *inp, int optname, struct mbuf *m) { int cnt, optlen; u_char *cp; struct mbuf **pcbopt; u_char opt; INP_WLOCK_ASSERT(inp); pcbopt = &inp->inp_options; /* turn off any old options */ if (*pcbopt) (void)m_free(*pcbopt); *pcbopt = NULL; if (m == NULL || m->m_len == 0) { /* * Only turning off any previous options. */ if (m != NULL) (void)m_free(m); return (0); } if (m->m_len % sizeof(int32_t)) goto bad; /* * IP first-hop destination address will be stored before actual * options; move other options back and clear it when none present. */ if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN]) goto bad; cnt = m->m_len; m->m_len += sizeof(struct in_addr); cp = mtod(m, u_char *) + sizeof(struct in_addr); bcopy(mtod(m, void *), cp, (unsigned)cnt); bzero(mtod(m, void *), sizeof(struct in_addr)); for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[IPOPT_OPTVAL]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) optlen = 1; else { if (cnt < IPOPT_OLEN + sizeof(*cp)) goto bad; optlen = cp[IPOPT_OLEN]; if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) goto bad; } switch (opt) { default: break; case IPOPT_LSRR: case IPOPT_SSRR: /* * User process specifies route as: * * ->A->B->C->D * * D must be our final destination (but we can't * check that since we may not have connected yet). * A is first hop destination, which doesn't appear * in actual IP option, but is stored before the * options. */ /* XXX-BZ PRIV_NETINET_SETHDROPTS? */ if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr)) goto bad; m->m_len -= sizeof(struct in_addr); cnt -= sizeof(struct in_addr); optlen -= sizeof(struct in_addr); cp[IPOPT_OLEN] = optlen; /* * Move first hop before start of options. */ bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t), sizeof(struct in_addr)); /* * Then copy rest of options back * to close up the deleted entry. */ bcopy((&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr)), &cp[IPOPT_OFFSET+1], (unsigned)cnt - (IPOPT_MINOFF - 1)); break; } } if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr)) goto bad; *pcbopt = m; return (0); bad: (void)m_free(m); return (EINVAL); } /* * Check for the presence of the IP Router Alert option [RFC2113] * in the header of an IPv4 datagram. * * This call is not intended for use from the forwarding path; it is here * so that protocol domains may check for the presence of the option. * Given how FreeBSD's IPv4 stack is currently structured, the Router Alert * option does not have much relevance to the implementation, though this * may change in future. * Router alert options SHOULD be passed if running in IPSTEALTH mode and * we are not the endpoint. * Length checks on individual options should already have been performed * by ip_dooptions() therefore they are folded under INVARIANTS here. * * Return zero if not present or options are invalid, non-zero if present. */ int ip_checkrouteralert(struct mbuf *m) { struct ip *ip = mtod(m, struct ip *); u_char *cp; int opt, optlen, cnt, found_ra; found_ra = 0; cp = (u_char *)(ip + 1); cnt = (ip->ip_hl << 2) - sizeof (struct ip); for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[IPOPT_OPTVAL]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) optlen = 1; else { #ifdef INVARIANTS if (cnt < IPOPT_OLEN + sizeof(*cp)) break; #endif optlen = cp[IPOPT_OLEN]; #ifdef INVARIANTS if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) break; #endif } switch (opt) { case IPOPT_RA: #ifdef INVARIANTS if (optlen != IPOPT_OFFSET + sizeof(uint16_t) || (*((uint16_t *)&cp[IPOPT_OFFSET]) != 0)) break; else #endif found_ra = 1; break; default: break; } } return (found_ra); } Index: head/sys/netinet/ip_reass.c =================================================================== --- head/sys/netinet/ip_reass.c (revision 336675) +++ head/sys/netinet/ip_reass.c (revision 336676) @@ -1,659 +1,659 @@ /*- * Copyright (c) 2015 Gleb Smirnoff * Copyright (c) 2015 Adrian Chadd * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef MAC #include #endif SYSCTL_DECL(_net_inet_ip); /* * Reassembly headers are stored in hash buckets. */ #define IPREASS_NHASH_LOG2 6 #define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2) #define IPREASS_HMASK (IPREASS_NHASH - 1) struct ipqbucket { TAILQ_HEAD(ipqhead, ipq) head; struct mtx lock; }; -static VNET_DEFINE(struct ipqbucket, ipq[IPREASS_NHASH]); +VNET_DEFINE_STATIC(struct ipqbucket, ipq[IPREASS_NHASH]); #define V_ipq VNET(ipq) -static VNET_DEFINE(uint32_t, ipq_hashseed); +VNET_DEFINE_STATIC(uint32_t, ipq_hashseed); #define V_ipq_hashseed VNET(ipq_hashseed) #define IPQ_LOCK(i) mtx_lock(&V_ipq[i].lock) #define IPQ_TRYLOCK(i) mtx_trylock(&V_ipq[i].lock) #define IPQ_UNLOCK(i) mtx_unlock(&V_ipq[i].lock) #define IPQ_LOCK_ASSERT(i) mtx_assert(&V_ipq[i].lock, MA_OWNED) void ipreass_init(void); void ipreass_drain(void); void ipreass_slowtimo(void); #ifdef VIMAGE void ipreass_destroy(void); #endif static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS); static void ipreass_zone_change(void *); static void ipreass_drain_tomax(void); static void ipq_free(struct ipqhead *, struct ipq *); static struct ipq * ipq_reuse(int); static inline void ipq_timeout(struct ipqhead *head, struct ipq *fp) { IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); ipq_free(head, fp); } static inline void ipq_drop(struct ipqhead *head, struct ipq *fp) { IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); ipq_free(head, fp); } -static VNET_DEFINE(uma_zone_t, ipq_zone); +VNET_DEFINE_STATIC(uma_zone_t, ipq_zone); #define V_ipq_zone VNET(ipq_zone) SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_maxfragpackets, "I", "Maximum number of IPv4 fragment reassembly queue entries"); SYSCTL_UMA_CUR(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET, &VNET_NAME(ipq_zone), "Current number of IPv4 fragment reassembly queue entries"); -static VNET_DEFINE(int, noreass); +VNET_DEFINE_STATIC(int, noreass); #define V_noreass VNET(noreass) -static VNET_DEFINE(int, maxfragsperpacket); +VNET_DEFINE_STATIC(int, maxfragsperpacket); #define V_maxfragsperpacket VNET(maxfragsperpacket) SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(maxfragsperpacket), 0, "Maximum number of IPv4 fragments allowed per packet"); /* * Take incoming datagram fragment and try to reassemble it into * whole datagram. If the argument is the first fragment or one * in between the function will return NULL and store the mbuf * in the fragment chain. If the argument is the last fragment * the packet will be reassembled and the pointer to the new * mbuf returned for further processing. Only m_tags attached * to the first packet/fragment are preserved. * The IP header is *NOT* adjusted out of iplen. */ #define M_IP_FRAG M_PROTO9 struct mbuf * ip_reass(struct mbuf *m) { struct ip *ip; struct mbuf *p, *q, *nq, *t; struct ipq *fp; struct ipqhead *head; int i, hlen, next; u_int8_t ecn, ecn0; uint32_t hash; #ifdef RSS uint32_t rss_hash, rss_type; #endif /* * If no reassembling or maxfragsperpacket are 0, * never accept fragments. */ if (V_noreass == 1 || V_maxfragsperpacket == 0) { IPSTAT_INC(ips_fragments); IPSTAT_INC(ips_fragdropped); m_freem(m); return (NULL); } ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; /* * Adjust ip_len to not reflect header, * convert offset of this to bytes. */ ip->ip_len = htons(ntohs(ip->ip_len) - hlen); if (ip->ip_off & htons(IP_MF)) { /* * Make sure that fragments have a data length * that's a non-zero multiple of 8 bytes. */ if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) { IPSTAT_INC(ips_toosmall); /* XXX */ IPSTAT_INC(ips_fragdropped); m_freem(m); return (NULL); } m->m_flags |= M_IP_FRAG; } else m->m_flags &= ~M_IP_FRAG; ip->ip_off = htons(ntohs(ip->ip_off) << 3); /* * Attempt reassembly; if it succeeds, proceed. * ip_reass() will return a different mbuf. */ IPSTAT_INC(ips_fragments); m->m_pkthdr.PH_loc.ptr = ip; /* * Presence of header sizes in mbufs * would confuse code below. */ m->m_data += hlen; m->m_len -= hlen; hash = ip->ip_src.s_addr ^ ip->ip_id; hash = jenkins_hash32(&hash, 1, V_ipq_hashseed) & IPREASS_HMASK; head = &V_ipq[hash].head; IPQ_LOCK(hash); /* * Look for queue of fragments * of this datagram. */ TAILQ_FOREACH(fp, head, ipq_list) if (ip->ip_id == fp->ipq_id && ip->ip_src.s_addr == fp->ipq_src.s_addr && ip->ip_dst.s_addr == fp->ipq_dst.s_addr && #ifdef MAC mac_ipq_match(m, fp) && #endif ip->ip_p == fp->ipq_p) break; /* * If first fragment to arrive, create a reassembly queue. */ if (fp == NULL) { fp = uma_zalloc(V_ipq_zone, M_NOWAIT); if (fp == NULL) fp = ipq_reuse(hash); #ifdef MAC if (mac_ipq_init(fp, M_NOWAIT) != 0) { uma_zfree(V_ipq_zone, fp); fp = NULL; goto dropfrag; } mac_ipq_create(m, fp); #endif TAILQ_INSERT_HEAD(head, fp, ipq_list); fp->ipq_nfrags = 1; fp->ipq_ttl = IPFRAGTTL; fp->ipq_p = ip->ip_p; fp->ipq_id = ip->ip_id; fp->ipq_src = ip->ip_src; fp->ipq_dst = ip->ip_dst; fp->ipq_frags = m; m->m_nextpkt = NULL; goto done; } else { fp->ipq_nfrags++; #ifdef MAC mac_ipq_update(m, fp); #endif } #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) /* * Handle ECN by comparing this segment with the first one; * if CE is set, do not lose CE. * drop if CE and not-ECT are mixed for the same packet. */ ecn = ip->ip_tos & IPTOS_ECN_MASK; ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; if (ecn == IPTOS_ECN_CE) { if (ecn0 == IPTOS_ECN_NOTECT) goto dropfrag; if (ecn0 != IPTOS_ECN_CE) GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; } if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) goto dropfrag; /* * Find a segment which begins after this one does. */ for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) break; /* * If there is a preceding segment, it may provide some of * our data already. If so, drop the data from the incoming * segment. If it provides all of our data, drop us, otherwise * stick new segment in the proper place. * * If some of the data is dropped from the preceding * segment, then it's checksum is invalidated. */ if (p) { i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - ntohs(ip->ip_off); if (i > 0) { if (i >= ntohs(ip->ip_len)) goto dropfrag; m_adj(m, i); m->m_pkthdr.csum_flags = 0; ip->ip_off = htons(ntohs(ip->ip_off) + i); ip->ip_len = htons(ntohs(ip->ip_len) - i); } m->m_nextpkt = p->m_nextpkt; p->m_nextpkt = m; } else { m->m_nextpkt = fp->ipq_frags; fp->ipq_frags = m; } /* * While we overlap succeeding segments trim them or, * if they are completely covered, dequeue them. */ for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > ntohs(GETIP(q)->ip_off); q = nq) { i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - ntohs(GETIP(q)->ip_off); if (i < ntohs(GETIP(q)->ip_len)) { GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); m_adj(q, i); q->m_pkthdr.csum_flags = 0; break; } nq = q->m_nextpkt; m->m_nextpkt = nq; IPSTAT_INC(ips_fragdropped); fp->ipq_nfrags--; m_freem(q); } /* * Check for complete reassembly and perform frag per packet * limiting. * * Frag limiting is performed here so that the nth frag has * a chance to complete the packet before we drop the packet. * As a result, n+1 frags are actually allowed per packet, but * only n will ever be stored. (n = maxfragsperpacket.) * */ next = 0; for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { if (ntohs(GETIP(q)->ip_off) != next) { if (fp->ipq_nfrags > V_maxfragsperpacket) ipq_drop(head, fp); goto done; } next += ntohs(GETIP(q)->ip_len); } /* Make sure the last packet didn't have the IP_MF flag */ if (p->m_flags & M_IP_FRAG) { if (fp->ipq_nfrags > V_maxfragsperpacket) ipq_drop(head, fp); goto done; } /* * Reassembly is complete. Make sure the packet is a sane size. */ q = fp->ipq_frags; ip = GETIP(q); if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { IPSTAT_INC(ips_toolong); ipq_drop(head, fp); goto done; } /* * Concatenate fragments. */ m = q; t = m->m_next; m->m_next = NULL; m_cat(m, t); nq = q->m_nextpkt; q->m_nextpkt = NULL; for (q = nq; q != NULL; q = nq) { nq = q->m_nextpkt; q->m_nextpkt = NULL; m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; m_demote_pkthdr(q); m_cat(m, q); } /* * In order to do checksumming faster we do 'end-around carry' here * (and not in for{} loop), though it implies we are not going to * reassemble more than 64k fragments. */ while (m->m_pkthdr.csum_data & 0xffff0000) m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); #ifdef MAC mac_ipq_reassemble(fp, m); mac_ipq_destroy(fp); #endif /* * Create header for new ip packet by modifying header of first * packet; dequeue and discard fragment reassembly header. * Make header visible. */ ip->ip_len = htons((ip->ip_hl << 2) + next); ip->ip_src = fp->ipq_src; ip->ip_dst = fp->ipq_dst; TAILQ_REMOVE(head, fp, ipq_list); uma_zfree(V_ipq_zone, fp); m->m_len += (ip->ip_hl << 2); m->m_data -= (ip->ip_hl << 2); /* some debugging cruft by sklower, below, will go away soon */ if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ m_fixhdr(m); IPSTAT_INC(ips_reassembled); IPQ_UNLOCK(hash); #ifdef RSS /* * Query the RSS layer for the flowid / flowtype for the * mbuf payload. * * For now, just assume we have to calculate a new one. * Later on we should check to see if the assigned flowid matches * what RSS wants for the given IP protocol and if so, just keep it. * * We then queue into the relevant netisr so it can be dispatched * to the correct CPU. * * Note - this may return 1, which means the flowid in the mbuf * is correct for the configured RSS hash types and can be used. */ if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { m->m_pkthdr.flowid = rss_hash; M_HASHTYPE_SET(m, rss_type); } /* * Queue/dispatch for reprocessing. * * Note: this is much slower than just handling the frame in the * current receive context. It's likely worth investigating * why this is. */ netisr_dispatch(NETISR_IP_DIRECT, m); return (NULL); #endif /* Handle in-line */ return (m); dropfrag: IPSTAT_INC(ips_fragdropped); if (fp != NULL) fp->ipq_nfrags--; m_freem(m); done: IPQ_UNLOCK(hash); return (NULL); #undef GETIP } /* * Initialize IP reassembly structures. */ void ipreass_init(void) { for (int i = 0; i < IPREASS_NHASH; i++) { TAILQ_INIT(&V_ipq[i].head); mtx_init(&V_ipq[i].lock, "IP reassembly", NULL, MTX_DEF | MTX_DUPOK); } V_ipq_hashseed = arc4random(); V_maxfragsperpacket = 16; V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_ipq_zone, nmbclusters / 32); if (IS_DEFAULT_VNET(curvnet)) EVENTHANDLER_REGISTER(nmbclusters_change, ipreass_zone_change, NULL, EVENTHANDLER_PRI_ANY); } /* * If a timer expires on a reassembly queue, discard it. */ void ipreass_slowtimo(void) { struct ipq *fp, *tmp; for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, tmp) if (--fp->ipq_ttl == 0) ipq_timeout(&V_ipq[i].head, fp); IPQ_UNLOCK(i); } } /* * Drain off all datagram fragments. */ void ipreass_drain(void) { for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); while(!TAILQ_EMPTY(&V_ipq[i].head)) ipq_drop(&V_ipq[i].head, TAILQ_FIRST(&V_ipq[i].head)); IPQ_UNLOCK(i); } } #ifdef VIMAGE /* * Destroy IP reassembly structures. */ void ipreass_destroy(void) { ipreass_drain(); uma_zdestroy(V_ipq_zone); for (int i = 0; i < IPREASS_NHASH; i++) mtx_destroy(&V_ipq[i].lock); } #endif /* * After maxnipq has been updated, propagate the change to UMA. The UMA zone * max has slightly different semantics than the sysctl, for historical * reasons. */ static void ipreass_drain_tomax(void) { int target; /* * If we are over the maximum number of fragments, * drain off enough to get down to the new limit, * stripping off last elements on queues. Every * run we strip the oldest element from each bucket. */ target = uma_zone_get_max(V_ipq_zone); while (uma_zone_get_cur(V_ipq_zone) > target) { struct ipq *fp; for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); fp = TAILQ_LAST(&V_ipq[i].head, ipqhead); if (fp != NULL) ipq_timeout(&V_ipq[i].head, fp); IPQ_UNLOCK(i); } } } static void ipreass_zone_change(void *tag) { uma_zone_set_max(V_ipq_zone, nmbclusters / 32); ipreass_drain_tomax(); } /* * Change the limit on the UMA zone, or disable the fragment allocation * at all. Since 0 and -1 is a special values here, we need our own handler, * instead of sysctl_handle_uma_zone_max(). */ static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS) { int error, max; if (V_noreass == 0) { max = uma_zone_get_max(V_ipq_zone); if (max == 0) max = -1; } else max = 0; error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); if (max > 0) { /* * XXXRW: Might be a good idea to sanity check the argument * and place an extreme upper bound. */ max = uma_zone_set_max(V_ipq_zone, max); ipreass_drain_tomax(); V_noreass = 0; } else if (max == 0) { V_noreass = 1; ipreass_drain(); } else if (max == -1) { V_noreass = 0; uma_zone_set_max(V_ipq_zone, 0); } else return (EINVAL); return (0); } /* * Seek for old fragment queue header that can be reused. Try to * reuse a header from currently locked hash bucket. */ static struct ipq * ipq_reuse(int start) { struct ipq *fp; int i; IPQ_LOCK_ASSERT(start); for (i = start;; i++) { if (i == IPREASS_NHASH) i = 0; if (i != start && IPQ_TRYLOCK(i) == 0) continue; fp = TAILQ_LAST(&V_ipq[i].head, ipqhead); if (fp) { struct mbuf *m; IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); while (fp->ipq_frags) { m = fp->ipq_frags; fp->ipq_frags = m->m_nextpkt; m_freem(m); } TAILQ_REMOVE(&V_ipq[i].head, fp, ipq_list); if (i != start) IPQ_UNLOCK(i); IPQ_LOCK_ASSERT(start); return (fp); } if (i != start) IPQ_UNLOCK(i); } } /* * Free a fragment reassembly header and all associated datagrams. */ static void ipq_free(struct ipqhead *fhp, struct ipq *fp) { struct mbuf *q; while (fp->ipq_frags) { q = fp->ipq_frags; fp->ipq_frags = q->m_nextpkt; m_freem(q); } TAILQ_REMOVE(fhp, fp, ipq_list); uma_zfree(V_ipq_zone, fp); } Index: head/sys/netinet/tcp_fastopen.c =================================================================== --- head/sys/netinet/tcp_fastopen.c (revision 336675) +++ head/sys/netinet/tcp_fastopen.c (revision 336676) @@ -1,1225 +1,1225 @@ /*- * Copyright (c) 2015-2017 Patrick Kelsey * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This is an implementation of TCP Fast Open (TFO) [RFC7413]. To include * this code, add the following line to your kernel config: * * options TCP_RFC7413 * * * The generated TFO cookies are the 64-bit output of * SipHash24(key=<16-byte-key>, msg=). Multiple concurrent valid * keys are supported so that time-based rolling cookie invalidation * policies can be implemented in the system. The default number of * concurrent keys is 2. This can be adjusted in the kernel config as * follows: * * options TCP_RFC7413_MAX_KEYS= * * * In addition to the facilities defined in RFC7413, this implementation * supports a pre-shared key (PSK) mode of operation in which the TFO server * requires the client to be in posession of a shared secret in order for * the client to be able to successfully open TFO connections with the * server. This is useful, for example, in environments where TFO servers * are exposed to both internal and external clients and only wish to allow * TFO connections from internal clients. * * In the PSK mode of operation, the server generates and sends TFO cookies * to requesting clients as usual. However, when validating cookies * received in TFO SYNs from clients, the server requires the * client-supplied cookie to equal SipHash24(key=<16-byte-psk>, * msg=). * * Multiple concurrent valid pre-shared keys are supported so that * time-based rolling PSK invalidation policies can be implemented in the * system. The default number of concurrent pre-shared keys is 2. This can * be adjusted in the kernel config as follows: * * options TCP_RFC7413_MAX_PSKS= * * * The following TFO-specific sysctls are defined: * * net.inet.tcp.fastopen.acceptany (RW, default 0) * When non-zero, all client-supplied TFO cookies will be considered to * be valid. * * net.inet.tcp.fastopen.autokey (RW, default 120) * When this and net.inet.tcp.fastopen.server_enable are non-zero, a new * key will be automatically generated after this many seconds. * * net.inet.tcp.fastopen.ccache_bucket_limit * (RWTUN, default TCP_FASTOPEN_CCACHE_BUCKET_LIMIT_DEFAULT) * The maximum number of entries in a client cookie cache bucket. * * net.inet.tcp.fastopen.ccache_buckets * (RDTUN, default TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT) * The number of client cookie cache buckets. * * net.inet.tcp.fastopen.ccache_list (RO) * Print the client cookie cache. * * net.inet.tcp.fastopen.client_enable (RW, default 0) * When zero, no new active (i.e., client) TFO connections can be * created. On the transition from enabled to disabled, the client * cookie cache is cleared and disabled. The transition from enabled to * disabled does not affect any active TFO connections in progress; it * only prevents new ones from being made. * * net.inet.tcp.fastopen.keylen (RD) * The key length in bytes. * * net.inet.tcp.fastopen.maxkeys (RD) * The maximum number of keys supported. * * net.inet.tcp.fastopen.maxpsks (RD) * The maximum number of pre-shared keys supported. * * net.inet.tcp.fastopen.numkeys (RD) * The current number of keys installed. * * net.inet.tcp.fastopen.numpsks (RD) * The current number of pre-shared keys installed. * * net.inet.tcp.fastopen.path_disable_time * (RW, default TCP_FASTOPEN_PATH_DISABLE_TIME_DEFAULT) * When a failure occurs while trying to create a new active (i.e., * client) TFO connection, new active connections on the same path, as * determined by the tuple {client_ip, server_ip, server_port}, will be * forced to be non-TFO for this many seconds. Note that the path * disable mechanism relies on state stored in client cookie cache * entries, so it is possible for the disable time for a given path to * be reduced if the corresponding client cookie cache entry is reused * due to resource pressure before the disable period has elapsed. * * net.inet.tcp.fastopen.psk_enable (RW, default 0) * When non-zero, pre-shared key (PSK) mode is enabled for all TFO * servers. On the transition from enabled to disabled, all installed * pre-shared keys are removed. * * net.inet.tcp.fastopen.server_enable (RW, default 0) * When zero, no new passive (i.e., server) TFO connections can be * created. On the transition from enabled to disabled, all installed * keys and pre-shared keys are removed. On the transition from * disabled to enabled, if net.inet.tcp.fastopen.autokey is non-zero and * there are no keys installed, a new key will be generated immediately. * The transition from enabled to disabled does not affect any passive * TFO connections in progress; it only prevents new ones from being * made. * * net.inet.tcp.fastopen.setkey (WR) * Install a new key by writing net.inet.tcp.fastopen.keylen bytes to * this sysctl. * * net.inet.tcp.fastopen.setpsk (WR) * Install a new pre-shared key by writing net.inet.tcp.fastopen.keylen * bytes to this sysctl. * * In order for TFO connections to be created via a listen socket, that * socket must have the TCP_FASTOPEN socket option set on it. This option * can be set on the socket either before or after the listen() is invoked. * Clearing this option on a listen socket after it has been set has no * effect on existing TFO connections or TFO connections in progress; it * only prevents new TFO connections from being made. * * For passively-created sockets, the TCP_FASTOPEN socket option can be * queried to determine whether the connection was established using TFO. * Note that connections that are established via a TFO SYN, but that fall * back to using a non-TFO SYN|ACK will have the TCP_FASTOPEN socket option * set. * * Per the RFC, this implementation limits the number of TFO connections * that can be in the SYN_RECEIVED state on a per listen-socket basis. * Whenever this limit is exceeded, requests for new TFO connections are * serviced as non-TFO requests. Without such a limit, given a valid TFO * cookie, an attacker could keep the listen queue in an overflow condition * using a TFO SYN flood. This implementation sets the limit at half the * configured listen backlog. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCP_FASTOPEN_KEY_LEN SIPHASH_KEY_LENGTH #if TCP_FASTOPEN_PSK_LEN != TCP_FASTOPEN_KEY_LEN #error TCP_FASTOPEN_PSK_LEN must be equal to TCP_FASTOPEN_KEY_LEN #endif /* * Because a PSK-mode setsockopt() uses tcpcb.t_tfo_cookie.client to hold * the PSK until the connect occurs. */ #if TCP_FASTOPEN_MAX_COOKIE_LEN < TCP_FASTOPEN_PSK_LEN #error TCP_FASTOPEN_MAX_COOKIE_LEN must be >= TCP_FASTOPEN_PSK_LEN #endif #define TCP_FASTOPEN_CCACHE_BUCKET_LIMIT_DEFAULT 16 #define TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT 2048 /* must be power of 2 */ #define TCP_FASTOPEN_PATH_DISABLE_TIME_DEFAULT 900 /* seconds */ #if !defined(TCP_RFC7413_MAX_KEYS) || (TCP_RFC7413_MAX_KEYS < 1) #define TCP_FASTOPEN_MAX_KEYS 2 #else #define TCP_FASTOPEN_MAX_KEYS TCP_RFC7413_MAX_KEYS #endif #if TCP_FASTOPEN_MAX_KEYS > 10 #undef TCP_FASTOPEN_MAX_KEYS #define TCP_FASTOPEN_MAX_KEYS 10 #endif #if !defined(TCP_RFC7413_MAX_PSKS) || (TCP_RFC7413_MAX_PSKS < 1) #define TCP_FASTOPEN_MAX_PSKS 2 #else #define TCP_FASTOPEN_MAX_PSKS TCP_RFC7413_MAX_PSKS #endif #if TCP_FASTOPEN_MAX_PSKS > 10 #undef TCP_FASTOPEN_MAX_PSKS #define TCP_FASTOPEN_MAX_PSKS 10 #endif struct tcp_fastopen_keylist { unsigned int newest; unsigned int newest_psk; uint8_t key[TCP_FASTOPEN_MAX_KEYS][TCP_FASTOPEN_KEY_LEN]; uint8_t psk[TCP_FASTOPEN_MAX_PSKS][TCP_FASTOPEN_KEY_LEN]; }; struct tcp_fastopen_callout { struct callout c; struct vnet *v; }; static struct tcp_fastopen_ccache_entry *tcp_fastopen_ccache_lookup( struct in_conninfo *, struct tcp_fastopen_ccache_bucket **); static struct tcp_fastopen_ccache_entry *tcp_fastopen_ccache_create( struct tcp_fastopen_ccache_bucket *, struct in_conninfo *, uint16_t, uint8_t, uint8_t *); static void tcp_fastopen_ccache_bucket_trim(struct tcp_fastopen_ccache_bucket *, unsigned int); static void tcp_fastopen_ccache_entry_drop(struct tcp_fastopen_ccache_entry *, struct tcp_fastopen_ccache_bucket *); SYSCTL_NODE(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW, 0, "TCP Fast Open"); -static VNET_DEFINE(int, tcp_fastopen_acceptany) = 0; +VNET_DEFINE_STATIC(int, tcp_fastopen_acceptany) = 0; #define V_tcp_fastopen_acceptany VNET(tcp_fastopen_acceptany) SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, acceptany, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_fastopen_acceptany), 0, "Accept any non-empty cookie"); -static VNET_DEFINE(unsigned int, tcp_fastopen_autokey) = 120; +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_autokey) = 120; #define V_tcp_fastopen_autokey VNET(tcp_fastopen_autokey) static int sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, autokey, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, &sysctl_net_inet_tcp_fastopen_autokey, "IU", "Number of seconds between auto-generation of a new key; zero disables"); static int sysctl_net_inet_tcp_fastopen_ccache_bucket_limit(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, ccache_bucket_limit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RWTUN, NULL, 0, &sysctl_net_inet_tcp_fastopen_ccache_bucket_limit, "IU", "Max entries per bucket in client cookie cache"); -static VNET_DEFINE(unsigned int, tcp_fastopen_ccache_buckets) = +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_ccache_buckets) = TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT; #define V_tcp_fastopen_ccache_buckets VNET(tcp_fastopen_ccache_buckets) SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, ccache_buckets, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_fastopen_ccache_buckets), 0, "Client cookie cache number of buckets (power of 2)"); VNET_DEFINE(unsigned int, tcp_fastopen_client_enable) = 1; static int sysctl_net_inet_tcp_fastopen_client_enable(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, client_enable, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, &sysctl_net_inet_tcp_fastopen_client_enable, "IU", "Enable/disable TCP Fast Open client functionality"); SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, keylen, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_KEY_LEN, "Key length in bytes"); SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, maxkeys, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_MAX_KEYS, "Maximum number of keys supported"); SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, maxpsks, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_MAX_PSKS, "Maximum number of pre-shared keys supported"); -static VNET_DEFINE(unsigned int, tcp_fastopen_numkeys) = 0; +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_numkeys) = 0; #define V_tcp_fastopen_numkeys VNET(tcp_fastopen_numkeys) SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, numkeys, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_fastopen_numkeys), 0, "Number of keys installed"); -static VNET_DEFINE(unsigned int, tcp_fastopen_numpsks) = 0; +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_numpsks) = 0; #define V_tcp_fastopen_numpsks VNET(tcp_fastopen_numpsks) SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, numpsks, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_fastopen_numpsks), 0, "Number of pre-shared keys installed"); -static VNET_DEFINE(unsigned int, tcp_fastopen_path_disable_time) = +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_path_disable_time) = TCP_FASTOPEN_PATH_DISABLE_TIME_DEFAULT; #define V_tcp_fastopen_path_disable_time VNET(tcp_fastopen_path_disable_time) SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, path_disable_time, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_fastopen_path_disable_time), 0, "Seconds a TFO failure disables a {client_ip, server_ip, server_port} path"); -static VNET_DEFINE(unsigned int, tcp_fastopen_psk_enable) = 0; +VNET_DEFINE_STATIC(unsigned int, tcp_fastopen_psk_enable) = 0; #define V_tcp_fastopen_psk_enable VNET(tcp_fastopen_psk_enable) static int sysctl_net_inet_tcp_fastopen_psk_enable(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, psk_enable, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, &sysctl_net_inet_tcp_fastopen_psk_enable, "IU", "Enable/disable TCP Fast Open server pre-shared key mode"); VNET_DEFINE(unsigned int, tcp_fastopen_server_enable) = 0; static int sysctl_net_inet_tcp_fastopen_server_enable(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, server_enable, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, &sysctl_net_inet_tcp_fastopen_server_enable, "IU", "Enable/disable TCP Fast Open server functionality"); static int sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, setkey, CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, &sysctl_net_inet_tcp_fastopen_setkey, "", "Install a new key"); static int sysctl_net_inet_tcp_fastopen_setpsk(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, setpsk, CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, &sysctl_net_inet_tcp_fastopen_setpsk, "", "Install a new pre-shared key"); static int sysctl_net_inet_tcp_fastopen_ccache_list(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, ccache_list, CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, NULL, 0, sysctl_net_inet_tcp_fastopen_ccache_list, "A", "List of all client cookie cache entries"); -static VNET_DEFINE(struct rmlock, tcp_fastopen_keylock); +VNET_DEFINE_STATIC(struct rmlock, tcp_fastopen_keylock); #define V_tcp_fastopen_keylock VNET(tcp_fastopen_keylock) #define TCP_FASTOPEN_KEYS_RLOCK(t) rm_rlock(&V_tcp_fastopen_keylock, (t)) #define TCP_FASTOPEN_KEYS_RUNLOCK(t) rm_runlock(&V_tcp_fastopen_keylock, (t)) #define TCP_FASTOPEN_KEYS_WLOCK() rm_wlock(&V_tcp_fastopen_keylock) #define TCP_FASTOPEN_KEYS_WUNLOCK() rm_wunlock(&V_tcp_fastopen_keylock) -static VNET_DEFINE(struct tcp_fastopen_keylist, tcp_fastopen_keys); +VNET_DEFINE_STATIC(struct tcp_fastopen_keylist, tcp_fastopen_keys); #define V_tcp_fastopen_keys VNET(tcp_fastopen_keys) -static VNET_DEFINE(struct tcp_fastopen_callout, tcp_fastopen_autokey_ctx); +VNET_DEFINE_STATIC(struct tcp_fastopen_callout, tcp_fastopen_autokey_ctx); #define V_tcp_fastopen_autokey_ctx VNET(tcp_fastopen_autokey_ctx) -static VNET_DEFINE(uma_zone_t, counter_zone); +VNET_DEFINE_STATIC(uma_zone_t, counter_zone); #define V_counter_zone VNET(counter_zone) static MALLOC_DEFINE(M_TCP_FASTOPEN_CCACHE, "tfo_ccache", "TFO client cookie cache buckets"); -static VNET_DEFINE(struct tcp_fastopen_ccache, tcp_fastopen_ccache); +VNET_DEFINE_STATIC(struct tcp_fastopen_ccache, tcp_fastopen_ccache); #define V_tcp_fastopen_ccache VNET(tcp_fastopen_ccache) #define CCB_LOCK(ccb) mtx_lock(&(ccb)->ccb_mtx) #define CCB_UNLOCK(ccb) mtx_unlock(&(ccb)->ccb_mtx) #define CCB_LOCK_ASSERT(ccb) mtx_assert(&(ccb)->ccb_mtx, MA_OWNED) void tcp_fastopen_init(void) { unsigned int i; V_counter_zone = uma_zcreate("tfo", sizeof(unsigned int), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); rm_init(&V_tcp_fastopen_keylock, "tfo_keylock"); callout_init_rm(&V_tcp_fastopen_autokey_ctx.c, &V_tcp_fastopen_keylock, 0); V_tcp_fastopen_autokey_ctx.v = curvnet; V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; V_tcp_fastopen_keys.newest_psk = TCP_FASTOPEN_MAX_PSKS - 1; /* May already be non-zero if kernel tunable was set */ if (V_tcp_fastopen_ccache.bucket_limit == 0) V_tcp_fastopen_ccache.bucket_limit = TCP_FASTOPEN_CCACHE_BUCKET_LIMIT_DEFAULT; /* May already be non-zero if kernel tunable was set */ if ((V_tcp_fastopen_ccache_buckets == 0) || !powerof2(V_tcp_fastopen_ccache_buckets)) V_tcp_fastopen_ccache.buckets = TCP_FASTOPEN_CCACHE_BUCKETS_DEFAULT; else V_tcp_fastopen_ccache.buckets = V_tcp_fastopen_ccache_buckets; V_tcp_fastopen_ccache.mask = V_tcp_fastopen_ccache.buckets - 1; V_tcp_fastopen_ccache.secret = arc4random(); V_tcp_fastopen_ccache.base = malloc(V_tcp_fastopen_ccache.buckets * sizeof(struct tcp_fastopen_ccache_bucket), M_TCP_FASTOPEN_CCACHE, M_WAITOK | M_ZERO); for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { TAILQ_INIT(&V_tcp_fastopen_ccache.base[i].ccb_entries); mtx_init(&V_tcp_fastopen_ccache.base[i].ccb_mtx, "tfo_ccache_bucket", NULL, MTX_DEF); if (V_tcp_fastopen_client_enable) { /* enable bucket */ V_tcp_fastopen_ccache.base[i].ccb_num_entries = 0; } else { /* disable bucket */ V_tcp_fastopen_ccache.base[i].ccb_num_entries = -1; } V_tcp_fastopen_ccache.base[i].ccb_ccache = &V_tcp_fastopen_ccache; } /* * Note that while the total number of entries in the cookie cache * is limited by the table management logic to * V_tcp_fastopen_ccache.buckets * * V_tcp_fastopen_ccache.bucket_limit, the total number of items in * this zone can exceed that amount by the number of CPUs in the * system times the maximum number of unallocated items that can be * present in each UMA per-CPU cache for this zone. */ V_tcp_fastopen_ccache.zone = uma_zcreate("tfo_ccache_entries", sizeof(struct tcp_fastopen_ccache_entry), NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, 0); } void tcp_fastopen_destroy(void) { struct tcp_fastopen_ccache_bucket *ccb; unsigned int i; for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; tcp_fastopen_ccache_bucket_trim(ccb, 0); mtx_destroy(&ccb->ccb_mtx); } KASSERT(uma_zone_get_cur(V_tcp_fastopen_ccache.zone) == 0, ("%s: TFO ccache zone allocation count not 0", __func__)); uma_zdestroy(V_tcp_fastopen_ccache.zone); free(V_tcp_fastopen_ccache.base, M_TCP_FASTOPEN_CCACHE); callout_drain(&V_tcp_fastopen_autokey_ctx.c); rm_destroy(&V_tcp_fastopen_keylock); uma_zdestroy(V_counter_zone); } unsigned int * tcp_fastopen_alloc_counter(void) { unsigned int *counter; counter = uma_zalloc(V_counter_zone, M_NOWAIT); if (counter) *counter = 1; return (counter); } void tcp_fastopen_decrement_counter(unsigned int *counter) { if (*counter == 1) uma_zfree(V_counter_zone, counter); else atomic_subtract_int(counter, 1); } static void tcp_fastopen_addkey_locked(uint8_t *key) { V_tcp_fastopen_keys.newest++; if (V_tcp_fastopen_keys.newest == TCP_FASTOPEN_MAX_KEYS) V_tcp_fastopen_keys.newest = 0; memcpy(V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], key, TCP_FASTOPEN_KEY_LEN); if (V_tcp_fastopen_numkeys < TCP_FASTOPEN_MAX_KEYS) V_tcp_fastopen_numkeys++; } static void tcp_fastopen_addpsk_locked(uint8_t *psk) { V_tcp_fastopen_keys.newest_psk++; if (V_tcp_fastopen_keys.newest_psk == TCP_FASTOPEN_MAX_PSKS) V_tcp_fastopen_keys.newest_psk = 0; memcpy(V_tcp_fastopen_keys.psk[V_tcp_fastopen_keys.newest_psk], psk, TCP_FASTOPEN_KEY_LEN); if (V_tcp_fastopen_numpsks < TCP_FASTOPEN_MAX_PSKS) V_tcp_fastopen_numpsks++; } static void tcp_fastopen_autokey_locked(void) { uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; arc4rand(newkey, TCP_FASTOPEN_KEY_LEN, 0); tcp_fastopen_addkey_locked(newkey); } static void tcp_fastopen_autokey_callout(void *arg) { struct tcp_fastopen_callout *ctx = arg; CURVNET_SET(ctx->v); tcp_fastopen_autokey_locked(); callout_reset(&ctx->c, V_tcp_fastopen_autokey * hz, tcp_fastopen_autokey_callout, ctx); CURVNET_RESTORE(); } static uint64_t tcp_fastopen_make_cookie(uint8_t key[SIPHASH_KEY_LENGTH], struct in_conninfo *inc) { SIPHASH_CTX ctx; uint64_t siphash; SipHash24_Init(&ctx); SipHash_SetKey(&ctx, key); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(inc->inc_faddr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(inc->inc6_faddr)); break; #endif } SipHash_Final((u_int8_t *)&siphash, &ctx); return (siphash); } static uint64_t tcp_fastopen_make_psk_cookie(uint8_t *psk, uint8_t *cookie, uint8_t cookie_len) { SIPHASH_CTX ctx; uint64_t psk_cookie; SipHash24_Init(&ctx); SipHash_SetKey(&ctx, psk); SipHash_Update(&ctx, cookie, cookie_len); SipHash_Final((u_int8_t *)&psk_cookie, &ctx); return (psk_cookie); } static int tcp_fastopen_find_cookie_match_locked(uint8_t *wire_cookie, uint64_t *cur_cookie) { unsigned int i, psk_index; uint64_t psk_cookie; if (V_tcp_fastopen_psk_enable) { psk_index = V_tcp_fastopen_keys.newest_psk; for (i = 0; i < V_tcp_fastopen_numpsks; i++) { psk_cookie = tcp_fastopen_make_psk_cookie( V_tcp_fastopen_keys.psk[psk_index], (uint8_t *)cur_cookie, TCP_FASTOPEN_COOKIE_LEN); if (memcmp(wire_cookie, &psk_cookie, TCP_FASTOPEN_COOKIE_LEN) == 0) return (1); if (psk_index == 0) psk_index = TCP_FASTOPEN_MAX_PSKS - 1; else psk_index--; } } else if (memcmp(wire_cookie, cur_cookie, TCP_FASTOPEN_COOKIE_LEN) == 0) return (1); return (0); } /* * Return values: * -1 the cookie is invalid and no valid cookie is available * 0 the cookie is invalid and the latest cookie has been returned * 1 the cookie is valid and the latest cookie has been returned */ int tcp_fastopen_check_cookie(struct in_conninfo *inc, uint8_t *cookie, unsigned int len, uint64_t *latest_cookie) { struct rm_priotracker tracker; unsigned int i, key_index; int rv; uint64_t cur_cookie; if (V_tcp_fastopen_acceptany) { *latest_cookie = 0; return (1); } TCP_FASTOPEN_KEYS_RLOCK(&tracker); if (len != TCP_FASTOPEN_COOKIE_LEN) { if (V_tcp_fastopen_numkeys > 0) { *latest_cookie = tcp_fastopen_make_cookie( V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], inc); rv = 0; } else rv = -1; goto out; } /* * Check against each available key, from newest to oldest. */ key_index = V_tcp_fastopen_keys.newest; for (i = 0; i < V_tcp_fastopen_numkeys; i++) { cur_cookie = tcp_fastopen_make_cookie(V_tcp_fastopen_keys.key[key_index], inc); if (i == 0) *latest_cookie = cur_cookie; rv = tcp_fastopen_find_cookie_match_locked(cookie, &cur_cookie); if (rv) goto out; if (key_index == 0) key_index = TCP_FASTOPEN_MAX_KEYS - 1; else key_index--; } rv = 0; out: TCP_FASTOPEN_KEYS_RUNLOCK(&tracker); return (rv); } static int sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS) { int error; unsigned int new; new = V_tcp_fastopen_autokey; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new > (INT_MAX / hz)) return (EINVAL); TCP_FASTOPEN_KEYS_WLOCK(); if (V_tcp_fastopen_server_enable) { if (V_tcp_fastopen_autokey && !new) callout_stop(&V_tcp_fastopen_autokey_ctx.c); else if (new) callout_reset(&V_tcp_fastopen_autokey_ctx.c, new * hz, tcp_fastopen_autokey_callout, &V_tcp_fastopen_autokey_ctx); } V_tcp_fastopen_autokey = new; TCP_FASTOPEN_KEYS_WUNLOCK(); } return (error); } static int sysctl_net_inet_tcp_fastopen_psk_enable(SYSCTL_HANDLER_ARGS) { int error; unsigned int new; new = V_tcp_fastopen_psk_enable; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (V_tcp_fastopen_psk_enable && !new) { /* enabled -> disabled */ TCP_FASTOPEN_KEYS_WLOCK(); V_tcp_fastopen_numpsks = 0; V_tcp_fastopen_keys.newest_psk = TCP_FASTOPEN_MAX_PSKS - 1; V_tcp_fastopen_psk_enable = 0; TCP_FASTOPEN_KEYS_WUNLOCK(); } else if (!V_tcp_fastopen_psk_enable && new) { /* disabled -> enabled */ TCP_FASTOPEN_KEYS_WLOCK(); V_tcp_fastopen_psk_enable = 1; TCP_FASTOPEN_KEYS_WUNLOCK(); } } return (error); } static int sysctl_net_inet_tcp_fastopen_server_enable(SYSCTL_HANDLER_ARGS) { int error; unsigned int new; new = V_tcp_fastopen_server_enable; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (V_tcp_fastopen_server_enable && !new) { /* enabled -> disabled */ TCP_FASTOPEN_KEYS_WLOCK(); V_tcp_fastopen_numkeys = 0; V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; if (V_tcp_fastopen_autokey) callout_stop(&V_tcp_fastopen_autokey_ctx.c); V_tcp_fastopen_numpsks = 0; V_tcp_fastopen_keys.newest_psk = TCP_FASTOPEN_MAX_PSKS - 1; V_tcp_fastopen_server_enable = 0; TCP_FASTOPEN_KEYS_WUNLOCK(); } else if (!V_tcp_fastopen_server_enable && new) { /* disabled -> enabled */ TCP_FASTOPEN_KEYS_WLOCK(); if (V_tcp_fastopen_autokey && (V_tcp_fastopen_numkeys == 0)) { tcp_fastopen_autokey_locked(); callout_reset(&V_tcp_fastopen_autokey_ctx.c, V_tcp_fastopen_autokey * hz, tcp_fastopen_autokey_callout, &V_tcp_fastopen_autokey_ctx); } V_tcp_fastopen_server_enable = 1; TCP_FASTOPEN_KEYS_WUNLOCK(); } } return (error); } static int sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS) { int error; uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen != sizeof(newkey)) return (EINVAL); error = SYSCTL_IN(req, newkey, sizeof(newkey)); if (error) return (error); TCP_FASTOPEN_KEYS_WLOCK(); tcp_fastopen_addkey_locked(newkey); TCP_FASTOPEN_KEYS_WUNLOCK(); return (0); } static int sysctl_net_inet_tcp_fastopen_setpsk(SYSCTL_HANDLER_ARGS) { int error; uint8_t newpsk[TCP_FASTOPEN_KEY_LEN]; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen != sizeof(newpsk)) return (EINVAL); error = SYSCTL_IN(req, newpsk, sizeof(newpsk)); if (error) return (error); TCP_FASTOPEN_KEYS_WLOCK(); tcp_fastopen_addpsk_locked(newpsk); TCP_FASTOPEN_KEYS_WUNLOCK(); return (0); } static int sysctl_net_inet_tcp_fastopen_ccache_bucket_limit(SYSCTL_HANDLER_ARGS) { struct tcp_fastopen_ccache_bucket *ccb; int error; unsigned int new; unsigned int i; new = V_tcp_fastopen_ccache.bucket_limit; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if ((new == 0) || (new > INT_MAX)) error = EINVAL; else { if (new < V_tcp_fastopen_ccache.bucket_limit) { for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; tcp_fastopen_ccache_bucket_trim(ccb, new); } } V_tcp_fastopen_ccache.bucket_limit = new; } } return (error); } static int sysctl_net_inet_tcp_fastopen_client_enable(SYSCTL_HANDLER_ARGS) { struct tcp_fastopen_ccache_bucket *ccb; int error; unsigned int new, i; new = V_tcp_fastopen_client_enable; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (V_tcp_fastopen_client_enable && !new) { /* enabled -> disabled */ for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; KASSERT(ccb->ccb_num_entries > -1, ("%s: ccb->ccb_num_entries %d is negative", __func__, ccb->ccb_num_entries)); tcp_fastopen_ccache_bucket_trim(ccb, 0); } V_tcp_fastopen_client_enable = 0; } else if (!V_tcp_fastopen_client_enable && new) { /* disabled -> enabled */ for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; CCB_LOCK(ccb); KASSERT(TAILQ_EMPTY(&ccb->ccb_entries), ("%s: ccb->ccb_entries not empty", __func__)); KASSERT(ccb->ccb_num_entries == -1, ("%s: ccb->ccb_num_entries %d not -1", __func__, ccb->ccb_num_entries)); ccb->ccb_num_entries = 0; /* enable bucket */ CCB_UNLOCK(ccb); } V_tcp_fastopen_client_enable = 1; } } return (error); } void tcp_fastopen_connect(struct tcpcb *tp) { struct inpcb *inp; struct tcp_fastopen_ccache_bucket *ccb; struct tcp_fastopen_ccache_entry *cce; sbintime_t now; uint16_t server_mss; uint64_t psk_cookie; psk_cookie = 0; inp = tp->t_inpcb; cce = tcp_fastopen_ccache_lookup(&inp->inp_inc, &ccb); if (cce) { if (cce->disable_time == 0) { if ((cce->cookie_len > 0) && (tp->t_tfo_client_cookie_len == TCP_FASTOPEN_PSK_LEN)) { psk_cookie = tcp_fastopen_make_psk_cookie( tp->t_tfo_cookie.client, cce->cookie, cce->cookie_len); } else { tp->t_tfo_client_cookie_len = cce->cookie_len; memcpy(tp->t_tfo_cookie.client, cce->cookie, cce->cookie_len); } server_mss = cce->server_mss; CCB_UNLOCK(ccb); if (tp->t_tfo_client_cookie_len == TCP_FASTOPEN_PSK_LEN && psk_cookie) { tp->t_tfo_client_cookie_len = TCP_FASTOPEN_COOKIE_LEN; memcpy(tp->t_tfo_cookie.client, &psk_cookie, TCP_FASTOPEN_COOKIE_LEN); } tcp_mss(tp, server_mss ? server_mss : -1); tp->snd_wnd = tp->t_maxseg; } else { /* * The path is disabled. Check the time and * possibly re-enable. */ now = getsbinuptime(); if (now - cce->disable_time > ((sbintime_t)V_tcp_fastopen_path_disable_time << 32)) { /* * Re-enable path. Force a TFO cookie * request. Forget the old MSS as it may be * bogus now, and we will rediscover it in * the SYN|ACK. */ cce->disable_time = 0; cce->server_mss = 0; cce->cookie_len = 0; /* * tp->t_tfo... cookie details are already * zero from the tcpcb init. */ } else { /* * Path is disabled, so disable TFO on this * connection. */ tp->t_flags &= ~TF_FASTOPEN; } CCB_UNLOCK(ccb); tcp_mss(tp, -1); /* * snd_wnd is irrelevant since we are either forcing * a TFO cookie request or disabling TFO - either * way, no data with the SYN. */ } } else { /* * A new entry for this path will be created when a SYN|ACK * comes back, or the attempt otherwise fails. */ CCB_UNLOCK(ccb); tcp_mss(tp, -1); /* * snd_wnd is irrelevant since we are forcing a TFO cookie * request. */ } } void tcp_fastopen_disable_path(struct tcpcb *tp) { struct in_conninfo *inc = &tp->t_inpcb->inp_inc; struct tcp_fastopen_ccache_bucket *ccb; struct tcp_fastopen_ccache_entry *cce; cce = tcp_fastopen_ccache_lookup(inc, &ccb); if (cce) { cce->server_mss = 0; cce->cookie_len = 0; /* * Preserve the existing disable time if it is already * disabled. */ if (cce->disable_time == 0) cce->disable_time = getsbinuptime(); } else /* use invalid cookie len to create disabled entry */ tcp_fastopen_ccache_create(ccb, inc, 0, TCP_FASTOPEN_MAX_COOKIE_LEN + 1, NULL); CCB_UNLOCK(ccb); tp->t_flags &= ~TF_FASTOPEN; } void tcp_fastopen_update_cache(struct tcpcb *tp, uint16_t mss, uint8_t cookie_len, uint8_t *cookie) { struct in_conninfo *inc = &tp->t_inpcb->inp_inc; struct tcp_fastopen_ccache_bucket *ccb; struct tcp_fastopen_ccache_entry *cce; cce = tcp_fastopen_ccache_lookup(inc, &ccb); if (cce) { if ((cookie_len >= TCP_FASTOPEN_MIN_COOKIE_LEN) && (cookie_len <= TCP_FASTOPEN_MAX_COOKIE_LEN) && ((cookie_len & 0x1) == 0)) { cce->server_mss = mss; cce->cookie_len = cookie_len; memcpy(cce->cookie, cookie, cookie_len); cce->disable_time = 0; } else { /* invalid cookie length, disable entry */ cce->server_mss = 0; cce->cookie_len = 0; /* * Preserve the existing disable time if it is * already disabled. */ if (cce->disable_time == 0) cce->disable_time = getsbinuptime(); } } else tcp_fastopen_ccache_create(ccb, inc, mss, cookie_len, cookie); CCB_UNLOCK(ccb); } static struct tcp_fastopen_ccache_entry * tcp_fastopen_ccache_lookup(struct in_conninfo *inc, struct tcp_fastopen_ccache_bucket **ccbp) { struct tcp_fastopen_ccache_bucket *ccb; struct tcp_fastopen_ccache_entry *cce; uint32_t last_word; uint32_t hash; hash = jenkins_hash32((uint32_t *)&inc->inc_ie.ie_dependladdr, 4, V_tcp_fastopen_ccache.secret); hash = jenkins_hash32((uint32_t *)&inc->inc_ie.ie_dependfaddr, 4, hash); last_word = inc->inc_fport; hash = jenkins_hash32(&last_word, 1, hash); ccb = &V_tcp_fastopen_ccache.base[hash & V_tcp_fastopen_ccache.mask]; *ccbp = ccb; CCB_LOCK(ccb); /* * Always returns with locked bucket. */ TAILQ_FOREACH(cce, &ccb->ccb_entries, cce_link) if ((!(cce->af == AF_INET6) == !(inc->inc_flags & INC_ISIPV6)) && (cce->server_port == inc->inc_ie.ie_fport) && (((cce->af == AF_INET) && (cce->cce_client_ip.v4.s_addr == inc->inc_laddr.s_addr) && (cce->cce_server_ip.v4.s_addr == inc->inc_faddr.s_addr)) || ((cce->af == AF_INET6) && IN6_ARE_ADDR_EQUAL(&cce->cce_client_ip.v6, &inc->inc6_laddr) && IN6_ARE_ADDR_EQUAL(&cce->cce_server_ip.v6, &inc->inc6_faddr)))) break; return (cce); } static struct tcp_fastopen_ccache_entry * tcp_fastopen_ccache_create(struct tcp_fastopen_ccache_bucket *ccb, struct in_conninfo *inc, uint16_t mss, uint8_t cookie_len, uint8_t *cookie) { struct tcp_fastopen_ccache_entry *cce; /* * 1. Create a new entry, or * 2. Reclaim an existing entry, or * 3. Fail */ CCB_LOCK_ASSERT(ccb); cce = NULL; if (ccb->ccb_num_entries < V_tcp_fastopen_ccache.bucket_limit) cce = uma_zalloc(V_tcp_fastopen_ccache.zone, M_NOWAIT); if (cce == NULL) { /* * At bucket limit, or out of memory - reclaim last * entry in bucket. */ cce = TAILQ_LAST(&ccb->ccb_entries, bucket_entries); if (cce == NULL) { /* XXX count this event */ return (NULL); } TAILQ_REMOVE(&ccb->ccb_entries, cce, cce_link); } else ccb->ccb_num_entries++; TAILQ_INSERT_HEAD(&ccb->ccb_entries, cce, cce_link); cce->af = (inc->inc_flags & INC_ISIPV6) ? AF_INET6 : AF_INET; if (cce->af == AF_INET) { cce->cce_client_ip.v4 = inc->inc_laddr; cce->cce_server_ip.v4 = inc->inc_faddr; } else { cce->cce_client_ip.v6 = inc->inc6_laddr; cce->cce_server_ip.v6 = inc->inc6_faddr; } cce->server_port = inc->inc_fport; if ((cookie_len >= TCP_FASTOPEN_MIN_COOKIE_LEN) && (cookie_len <= TCP_FASTOPEN_MAX_COOKIE_LEN) && ((cookie_len & 0x1) == 0)) { cce->server_mss = mss; cce->cookie_len = cookie_len; memcpy(cce->cookie, cookie, cookie_len); cce->disable_time = 0; } else { /* invalid cookie length, disable cce */ cce->server_mss = 0; cce->cookie_len = 0; cce->disable_time = getsbinuptime(); } return (cce); } static void tcp_fastopen_ccache_bucket_trim(struct tcp_fastopen_ccache_bucket *ccb, unsigned int limit) { struct tcp_fastopen_ccache_entry *cce, *cce_tmp; unsigned int entries; CCB_LOCK(ccb); entries = 0; TAILQ_FOREACH_SAFE(cce, &ccb->ccb_entries, cce_link, cce_tmp) { entries++; if (entries > limit) tcp_fastopen_ccache_entry_drop(cce, ccb); } KASSERT(ccb->ccb_num_entries <= (int)limit, ("%s: ccb->ccb_num_entries %d exceeds limit %d", __func__, ccb->ccb_num_entries, limit)); if (limit == 0) { KASSERT(TAILQ_EMPTY(&ccb->ccb_entries), ("%s: ccb->ccb_entries not empty", __func__)); ccb->ccb_num_entries = -1; /* disable bucket */ } CCB_UNLOCK(ccb); } static void tcp_fastopen_ccache_entry_drop(struct tcp_fastopen_ccache_entry *cce, struct tcp_fastopen_ccache_bucket *ccb) { CCB_LOCK_ASSERT(ccb); TAILQ_REMOVE(&ccb->ccb_entries, cce, cce_link); ccb->ccb_num_entries--; uma_zfree(V_tcp_fastopen_ccache.zone, cce); } static int sysctl_net_inet_tcp_fastopen_ccache_list(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct tcp_fastopen_ccache_bucket *ccb; struct tcp_fastopen_ccache_entry *cce; sbintime_t now, duration, limit; const int linesize = 128; int i, error, num_entries; unsigned int j; #ifdef INET6 char clt_buf[INET6_ADDRSTRLEN], srv_buf[INET6_ADDRSTRLEN]; #else char clt_buf[INET_ADDRSTRLEN], srv_buf[INET_ADDRSTRLEN]; #endif if (jailed_without_vnet(curthread->td_ucred) != 0) return (EPERM); /* Only allow root to read the client cookie cache */ if (curthread->td_ucred->cr_uid != 0) return (EPERM); num_entries = 0; for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; CCB_LOCK(ccb); if (ccb->ccb_num_entries > 0) num_entries += ccb->ccb_num_entries; CCB_UNLOCK(ccb); } sbuf_new(&sb, NULL, linesize * (num_entries + 1), SBUF_INCLUDENUL); sbuf_printf(&sb, "\nLocal IP address Remote IP address Port MSS" " Disabled Cookie\n"); now = getsbinuptime(); limit = (sbintime_t)V_tcp_fastopen_path_disable_time << 32; for (i = 0; i < V_tcp_fastopen_ccache.buckets; i++) { ccb = &V_tcp_fastopen_ccache.base[i]; CCB_LOCK(ccb); TAILQ_FOREACH(cce, &ccb->ccb_entries, cce_link) { if (cce->disable_time != 0) { duration = now - cce->disable_time; if (limit >= duration) duration = limit - duration; else duration = 0; } else duration = 0; sbuf_printf(&sb, "%-20s %-20s %5u %5u ", inet_ntop(cce->af, &cce->cce_client_ip, clt_buf, sizeof(clt_buf)), inet_ntop(cce->af, &cce->cce_server_ip, srv_buf, sizeof(srv_buf)), ntohs(cce->server_port), cce->server_mss); if (duration > 0) sbuf_printf(&sb, "%7ds ", sbintime_getsec(duration)); else sbuf_printf(&sb, "%8s ", "No"); for (j = 0; j < cce->cookie_len; j++) sbuf_printf(&sb, "%02x", cce->cookie[j]); sbuf_putc(&sb, '\n'); } CCB_UNLOCK(ccb); } error = sbuf_finish(&sb); if (error == 0) error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb)); sbuf_delete(&sb); return (error); } Index: head/sys/netinet/tcp_hostcache.c =================================================================== --- head/sys/netinet/tcp_hostcache.c (revision 336675) +++ head/sys/netinet/tcp_hostcache.c (revision 336676) @@ -1,748 +1,748 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2002 Andre Oppermann, Internet Business Solutions AG * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote * products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * The tcp_hostcache moves the tcp-specific cached metrics from the routing * table to a dedicated structure indexed by the remote IP address. It keeps * information on the measured TCP parameters of past TCP sessions to allow * better initial start values to be used with later connections to/from the * same source. Depending on the network parameters (delay, max MTU, * congestion window) between local and remote sites, this can lead to * significant speed-ups for new TCP connections after the first one. * * Due to the tcp_hostcache, all TCP-specific metrics information in the * routing table have been removed. The inpcb no longer keeps a pointer to * the routing entry, and protocol-initiated route cloning has been removed * as well. With these changes, the routing table has gone back to being * more lightwight and only carries information related to packet forwarding. * * tcp_hostcache is designed for multiple concurrent access in SMP * environments and high contention. All bucket rows have their own lock and * thus multiple lookups and modifies can be done at the same time as long as * they are in different bucket rows. If a request for insertion of a new * record can't be satisfied, it simply returns an empty structure. Nobody * and nothing outside of tcp_hostcache.c will ever point directly to any * entry in the tcp_hostcache. All communication is done in an * object-oriented way and only functions of tcp_hostcache will manipulate * hostcache entries. Otherwise, we are unable to achieve good behaviour in * concurrent access situations. Since tcp_hostcache is only caching * information, there are no fatal consequences if we either can't satisfy * any particular request or have to drop/overwrite an existing entry because * of bucket limit memory constrains. */ /* * Many thanks to jlemon for basic structure of tcp_syncache which is being * followed here. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #endif #include #include #include #ifdef INET6 #include #endif #include /* Arbitrary values */ #define TCP_HOSTCACHE_HASHSIZE 512 #define TCP_HOSTCACHE_BUCKETLIMIT 30 #define TCP_HOSTCACHE_EXPIRE 60*60 /* one hour */ #define TCP_HOSTCACHE_PRUNE 5*60 /* every 5 minutes */ -static VNET_DEFINE(struct tcp_hostcache, tcp_hostcache); +VNET_DEFINE_STATIC(struct tcp_hostcache, tcp_hostcache); #define V_tcp_hostcache VNET(tcp_hostcache) -static VNET_DEFINE(struct callout, tcp_hc_callout); +VNET_DEFINE_STATIC(struct callout, tcp_hc_callout); #define V_tcp_hc_callout VNET(tcp_hc_callout) static struct hc_metrics *tcp_hc_lookup(struct in_conninfo *); static struct hc_metrics *tcp_hc_insert(struct in_conninfo *); static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS); static int sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS); static void tcp_hc_purge_internal(int); static void tcp_hc_purge(void *); static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hostcache, CTLFLAG_RW, 0, "TCP Host cache"); VNET_DEFINE(int, tcp_use_hostcache) = 1; #define V_tcp_use_hostcache VNET(tcp_use_hostcache) SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_use_hostcache), 0, "Enable the TCP hostcache"); SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.cache_limit), 0, "Overall entry limit for hostcache"); SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.hashsize), 0, "Size of TCP hostcache hashtable"); SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, bucketlimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_hostcache.bucket_limit), 0, "Per-bucket hash limit for hostcache"); SYSCTL_UINT(_net_inet_tcp_hostcache, OID_AUTO, count, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_hostcache.cache_count), 0, "Current number of entries in hostcache"); SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, expire, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_hostcache.expire), 0, "Expire time of TCP hostcache entries"); SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, prune, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_hostcache.prune), 0, "Time between purge runs"); SYSCTL_INT(_net_inet_tcp_hostcache, OID_AUTO, purge, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_hostcache.purgeall), 0, "Expire all entires on next purge run"); SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, list, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_SKIP, 0, 0, sysctl_tcp_hc_list, "A", "List of all hostcache entries"); SYSCTL_PROC(_net_inet_tcp_hostcache, OID_AUTO, purgenow, CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_tcp_hc_purgenow, "I", "Immediately purge all entries"); static MALLOC_DEFINE(M_HOSTCACHE, "hostcache", "TCP hostcache"); #define HOSTCACHE_HASH(ip) \ (((ip)->s_addr ^ ((ip)->s_addr >> 7) ^ ((ip)->s_addr >> 17)) & \ V_tcp_hostcache.hashmask) /* XXX: What is the recommended hash to get good entropy for IPv6 addresses? */ #define HOSTCACHE_HASH6(ip6) \ (((ip6)->s6_addr32[0] ^ \ (ip6)->s6_addr32[1] ^ \ (ip6)->s6_addr32[2] ^ \ (ip6)->s6_addr32[3]) & \ V_tcp_hostcache.hashmask) #define THC_LOCK(lp) mtx_lock(lp) #define THC_UNLOCK(lp) mtx_unlock(lp) void tcp_hc_init(void) { u_int cache_limit; int i; /* * Initialize hostcache structures. */ V_tcp_hostcache.cache_count = 0; V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; V_tcp_hostcache.bucket_limit = TCP_HOSTCACHE_BUCKETLIMIT; V_tcp_hostcache.expire = TCP_HOSTCACHE_EXPIRE; V_tcp_hostcache.prune = TCP_HOSTCACHE_PRUNE; TUNABLE_INT_FETCH("net.inet.tcp.hostcache.hashsize", &V_tcp_hostcache.hashsize); if (!powerof2(V_tcp_hostcache.hashsize)) { printf("WARNING: hostcache hash size is not a power of 2.\n"); V_tcp_hostcache.hashsize = TCP_HOSTCACHE_HASHSIZE; /* default */ } V_tcp_hostcache.hashmask = V_tcp_hostcache.hashsize - 1; TUNABLE_INT_FETCH("net.inet.tcp.hostcache.bucketlimit", &V_tcp_hostcache.bucket_limit); cache_limit = V_tcp_hostcache.hashsize * V_tcp_hostcache.bucket_limit; V_tcp_hostcache.cache_limit = cache_limit; TUNABLE_INT_FETCH("net.inet.tcp.hostcache.cachelimit", &V_tcp_hostcache.cache_limit); if (V_tcp_hostcache.cache_limit > cache_limit) V_tcp_hostcache.cache_limit = cache_limit; /* * Allocate the hash table. */ V_tcp_hostcache.hashbase = (struct hc_head *) malloc(V_tcp_hostcache.hashsize * sizeof(struct hc_head), M_HOSTCACHE, M_WAITOK | M_ZERO); /* * Initialize the hash buckets. */ for (i = 0; i < V_tcp_hostcache.hashsize; i++) { TAILQ_INIT(&V_tcp_hostcache.hashbase[i].hch_bucket); V_tcp_hostcache.hashbase[i].hch_length = 0; mtx_init(&V_tcp_hostcache.hashbase[i].hch_mtx, "tcp_hc_entry", NULL, MTX_DEF); } /* * Allocate the hostcache entries. */ V_tcp_hostcache.zone = uma_zcreate("hostcache", sizeof(struct hc_metrics), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_tcp_hostcache.zone, V_tcp_hostcache.cache_limit); /* * Set up periodic cache cleanup. */ callout_init(&V_tcp_hc_callout, 1); callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, tcp_hc_purge, curvnet); } #ifdef VIMAGE void tcp_hc_destroy(void) { int i; callout_drain(&V_tcp_hc_callout); /* Purge all hc entries. */ tcp_hc_purge_internal(1); /* Free the uma zone and the allocated hash table. */ uma_zdestroy(V_tcp_hostcache.zone); for (i = 0; i < V_tcp_hostcache.hashsize; i++) mtx_destroy(&V_tcp_hostcache.hashbase[i].hch_mtx); free(V_tcp_hostcache.hashbase, M_HOSTCACHE); } #endif /* * Internal function: look up an entry in the hostcache or return NULL. * * If an entry has been returned, the caller becomes responsible for * unlocking the bucket row after he is done reading/modifying the entry. */ static struct hc_metrics * tcp_hc_lookup(struct in_conninfo *inc) { int hash; struct hc_head *hc_head; struct hc_metrics *hc_entry; if (!V_tcp_use_hostcache) return NULL; KASSERT(inc != NULL, ("tcp_hc_lookup with NULL in_conninfo pointer")); /* * Hash the foreign ip address. */ if (inc->inc_flags & INC_ISIPV6) hash = HOSTCACHE_HASH6(&inc->inc6_faddr); else hash = HOSTCACHE_HASH(&inc->inc_faddr); hc_head = &V_tcp_hostcache.hashbase[hash]; /* * Acquire lock for this bucket row; we release the lock if we don't * find an entry, otherwise the caller has to unlock after he is * done. */ THC_LOCK(&hc_head->hch_mtx); /* * Iterate through entries in bucket row looking for a match. */ TAILQ_FOREACH(hc_entry, &hc_head->hch_bucket, rmx_q) { if (inc->inc_flags & INC_ISIPV6) { /* XXX: check ip6_zoneid */ if (memcmp(&inc->inc6_faddr, &hc_entry->ip6, sizeof(inc->inc6_faddr)) == 0) return hc_entry; } else { if (memcmp(&inc->inc_faddr, &hc_entry->ip4, sizeof(inc->inc_faddr)) == 0) return hc_entry; } } /* * We were unsuccessful and didn't find anything. */ THC_UNLOCK(&hc_head->hch_mtx); return NULL; } /* * Internal function: insert an entry into the hostcache or return NULL if * unable to allocate a new one. * * If an entry has been returned, the caller becomes responsible for * unlocking the bucket row after he is done reading/modifying the entry. */ static struct hc_metrics * tcp_hc_insert(struct in_conninfo *inc) { int hash; struct hc_head *hc_head; struct hc_metrics *hc_entry; if (!V_tcp_use_hostcache) return NULL; KASSERT(inc != NULL, ("tcp_hc_insert with NULL in_conninfo pointer")); /* * Hash the foreign ip address. */ if (inc->inc_flags & INC_ISIPV6) hash = HOSTCACHE_HASH6(&inc->inc6_faddr); else hash = HOSTCACHE_HASH(&inc->inc_faddr); hc_head = &V_tcp_hostcache.hashbase[hash]; /* * Acquire lock for this bucket row; we release the lock if we don't * find an entry, otherwise the caller has to unlock after he is * done. */ THC_LOCK(&hc_head->hch_mtx); /* * If the bucket limit is reached, reuse the least-used element. */ if (hc_head->hch_length >= V_tcp_hostcache.bucket_limit || V_tcp_hostcache.cache_count >= V_tcp_hostcache.cache_limit) { hc_entry = TAILQ_LAST(&hc_head->hch_bucket, hc_qhead); /* * At first we were dropping the last element, just to * reacquire it in the next two lines again, which isn't very * efficient. Instead just reuse the least used element. * We may drop something that is still "in-use" but we can be * "lossy". * Just give up if this bucket row is empty and we don't have * anything to replace. */ if (hc_entry == NULL) { THC_UNLOCK(&hc_head->hch_mtx); return NULL; } TAILQ_REMOVE(&hc_head->hch_bucket, hc_entry, rmx_q); V_tcp_hostcache.hashbase[hash].hch_length--; V_tcp_hostcache.cache_count--; TCPSTAT_INC(tcps_hc_bucketoverflow); #if 0 uma_zfree(V_tcp_hostcache.zone, hc_entry); #endif } else { /* * Allocate a new entry, or balk if not possible. */ hc_entry = uma_zalloc(V_tcp_hostcache.zone, M_NOWAIT); if (hc_entry == NULL) { THC_UNLOCK(&hc_head->hch_mtx); return NULL; } } /* * Initialize basic information of hostcache entry. */ bzero(hc_entry, sizeof(*hc_entry)); if (inc->inc_flags & INC_ISIPV6) { hc_entry->ip6 = inc->inc6_faddr; hc_entry->ip6_zoneid = inc->inc6_zoneid; } else hc_entry->ip4 = inc->inc_faddr; hc_entry->rmx_head = hc_head; hc_entry->rmx_expire = V_tcp_hostcache.expire; /* * Put it upfront. */ TAILQ_INSERT_HEAD(&hc_head->hch_bucket, hc_entry, rmx_q); V_tcp_hostcache.hashbase[hash].hch_length++; V_tcp_hostcache.cache_count++; TCPSTAT_INC(tcps_hc_added); return hc_entry; } /* * External function: look up an entry in the hostcache and fill out the * supplied TCP metrics structure. Fills in NULL when no entry was found or * a value is not set. */ void tcp_hc_get(struct in_conninfo *inc, struct hc_metrics_lite *hc_metrics_lite) { struct hc_metrics *hc_entry; if (!V_tcp_use_hostcache) return; /* * Find the right bucket. */ hc_entry = tcp_hc_lookup(inc); /* * If we don't have an existing object. */ if (hc_entry == NULL) { bzero(hc_metrics_lite, sizeof(*hc_metrics_lite)); return; } hc_entry->rmx_hits++; hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ hc_metrics_lite->rmx_mtu = hc_entry->rmx_mtu; hc_metrics_lite->rmx_ssthresh = hc_entry->rmx_ssthresh; hc_metrics_lite->rmx_rtt = hc_entry->rmx_rtt; hc_metrics_lite->rmx_rttvar = hc_entry->rmx_rttvar; hc_metrics_lite->rmx_cwnd = hc_entry->rmx_cwnd; hc_metrics_lite->rmx_sendpipe = hc_entry->rmx_sendpipe; hc_metrics_lite->rmx_recvpipe = hc_entry->rmx_recvpipe; /* * Unlock bucket row. */ THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); } /* * External function: look up an entry in the hostcache and return the * discovered path MTU. Returns 0 if no entry is found or value is not * set. */ uint32_t tcp_hc_getmtu(struct in_conninfo *inc) { struct hc_metrics *hc_entry; uint32_t mtu; if (!V_tcp_use_hostcache) return 0; hc_entry = tcp_hc_lookup(inc); if (hc_entry == NULL) { return 0; } hc_entry->rmx_hits++; hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ mtu = hc_entry->rmx_mtu; THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); return mtu; } /* * External function: update the MTU value of an entry in the hostcache. * Creates a new entry if none was found. */ void tcp_hc_updatemtu(struct in_conninfo *inc, uint32_t mtu) { struct hc_metrics *hc_entry; if (!V_tcp_use_hostcache) return; /* * Find the right bucket. */ hc_entry = tcp_hc_lookup(inc); /* * If we don't have an existing object, try to insert a new one. */ if (hc_entry == NULL) { hc_entry = tcp_hc_insert(inc); if (hc_entry == NULL) return; } hc_entry->rmx_updates++; hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ hc_entry->rmx_mtu = mtu; /* * Put it upfront so we find it faster next time. */ TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); /* * Unlock bucket row. */ THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); } /* * External function: update the TCP metrics of an entry in the hostcache. * Creates a new entry if none was found. */ void tcp_hc_update(struct in_conninfo *inc, struct hc_metrics_lite *hcml) { struct hc_metrics *hc_entry; if (!V_tcp_use_hostcache) return; hc_entry = tcp_hc_lookup(inc); if (hc_entry == NULL) { hc_entry = tcp_hc_insert(inc); if (hc_entry == NULL) return; } hc_entry->rmx_updates++; hc_entry->rmx_expire = V_tcp_hostcache.expire; /* start over again */ if (hcml->rmx_rtt != 0) { if (hc_entry->rmx_rtt == 0) hc_entry->rmx_rtt = hcml->rmx_rtt; else hc_entry->rmx_rtt = ((uint64_t)hc_entry->rmx_rtt + (uint64_t)hcml->rmx_rtt) / 2; TCPSTAT_INC(tcps_cachedrtt); } if (hcml->rmx_rttvar != 0) { if (hc_entry->rmx_rttvar == 0) hc_entry->rmx_rttvar = hcml->rmx_rttvar; else hc_entry->rmx_rttvar = ((uint64_t)hc_entry->rmx_rttvar + (uint64_t)hcml->rmx_rttvar) / 2; TCPSTAT_INC(tcps_cachedrttvar); } if (hcml->rmx_ssthresh != 0) { if (hc_entry->rmx_ssthresh == 0) hc_entry->rmx_ssthresh = hcml->rmx_ssthresh; else hc_entry->rmx_ssthresh = (hc_entry->rmx_ssthresh + hcml->rmx_ssthresh) / 2; TCPSTAT_INC(tcps_cachedssthresh); } if (hcml->rmx_cwnd != 0) { if (hc_entry->rmx_cwnd == 0) hc_entry->rmx_cwnd = hcml->rmx_cwnd; else hc_entry->rmx_cwnd = ((uint64_t)hc_entry->rmx_cwnd + (uint64_t)hcml->rmx_cwnd) / 2; /* TCPSTAT_INC(tcps_cachedcwnd); */ } if (hcml->rmx_sendpipe != 0) { if (hc_entry->rmx_sendpipe == 0) hc_entry->rmx_sendpipe = hcml->rmx_sendpipe; else hc_entry->rmx_sendpipe = ((uint64_t)hc_entry->rmx_sendpipe + (uint64_t)hcml->rmx_sendpipe) /2; /* TCPSTAT_INC(tcps_cachedsendpipe); */ } if (hcml->rmx_recvpipe != 0) { if (hc_entry->rmx_recvpipe == 0) hc_entry->rmx_recvpipe = hcml->rmx_recvpipe; else hc_entry->rmx_recvpipe = ((uint64_t)hc_entry->rmx_recvpipe + (uint64_t)hcml->rmx_recvpipe) /2; /* TCPSTAT_INC(tcps_cachedrecvpipe); */ } TAILQ_REMOVE(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); TAILQ_INSERT_HEAD(&hc_entry->rmx_head->hch_bucket, hc_entry, rmx_q); THC_UNLOCK(&hc_entry->rmx_head->hch_mtx); } /* * Sysctl function: prints the list and values of all hostcache entries in * unsorted order. */ static int sysctl_tcp_hc_list(SYSCTL_HANDLER_ARGS) { const int linesize = 128; struct sbuf sb; int i, error; struct hc_metrics *hc_entry; char ip4buf[INET_ADDRSTRLEN]; #ifdef INET6 char ip6buf[INET6_ADDRSTRLEN]; #endif if (jailed_without_vnet(curthread->td_ucred) != 0) return (EPERM); sbuf_new(&sb, NULL, linesize * (V_tcp_hostcache.cache_count + 1), SBUF_INCLUDENUL); sbuf_printf(&sb, "\nIP address MTU SSTRESH RTT RTTVAR " " CWND SENDPIPE RECVPIPE HITS UPD EXP\n"); #define msec(u) (((u) + 500) / 1000) for (i = 0; i < V_tcp_hostcache.hashsize; i++) { THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); TAILQ_FOREACH(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q) { sbuf_printf(&sb, "%-15s %5u %8u %6lums %6lums %8u %8u %8u %4lu " "%4lu %4i\n", hc_entry->ip4.s_addr ? inet_ntoa_r(hc_entry->ip4, ip4buf) : #ifdef INET6 ip6_sprintf(ip6buf, &hc_entry->ip6), #else "IPv6?", #endif hc_entry->rmx_mtu, hc_entry->rmx_ssthresh, msec((u_long)hc_entry->rmx_rtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE))), msec((u_long)hc_entry->rmx_rttvar * (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE))), hc_entry->rmx_cwnd, hc_entry->rmx_sendpipe, hc_entry->rmx_recvpipe, hc_entry->rmx_hits, hc_entry->rmx_updates, hc_entry->rmx_expire); } THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); } #undef msec error = sbuf_finish(&sb); if (error == 0) error = SYSCTL_OUT(req, sbuf_data(&sb), sbuf_len(&sb)); sbuf_delete(&sb); return(error); } /* * Caller has to make sure the curvnet is set properly. */ static void tcp_hc_purge_internal(int all) { struct hc_metrics *hc_entry, *hc_next; int i; for (i = 0; i < V_tcp_hostcache.hashsize; i++) { THC_LOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); TAILQ_FOREACH_SAFE(hc_entry, &V_tcp_hostcache.hashbase[i].hch_bucket, rmx_q, hc_next) { if (all || hc_entry->rmx_expire <= 0) { TAILQ_REMOVE(&V_tcp_hostcache.hashbase[i].hch_bucket, hc_entry, rmx_q); uma_zfree(V_tcp_hostcache.zone, hc_entry); V_tcp_hostcache.hashbase[i].hch_length--; V_tcp_hostcache.cache_count--; } else hc_entry->rmx_expire -= V_tcp_hostcache.prune; } THC_UNLOCK(&V_tcp_hostcache.hashbase[i].hch_mtx); } } /* * Expire and purge (old|all) entries in the tcp_hostcache. Runs * periodically from the callout. */ static void tcp_hc_purge(void *arg) { CURVNET_SET((struct vnet *) arg); int all = 0; if (V_tcp_hostcache.purgeall) { all = 1; V_tcp_hostcache.purgeall = 0; } tcp_hc_purge_internal(all); callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, tcp_hc_purge, arg); CURVNET_RESTORE(); } /* * Expire and purge all entries in hostcache immediately. */ static int sysctl_tcp_hc_purgenow(SYSCTL_HANDLER_ARGS) { int error, val; val = 0; error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return (error); tcp_hc_purge_internal(1); callout_reset(&V_tcp_hc_callout, V_tcp_hostcache.prune * hz, tcp_hc_purge, curvnet); return (0); } Index: head/sys/netinet/tcp_subr.c =================================================================== --- head/sys/netinet/tcp_subr.c (revision 336675) +++ head/sys/netinet/tcp_subr.c (revision 336676) @@ -1,3235 +1,3235 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef TCP_HHOOK #include #endif #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #include #include #endif #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include #include VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; #ifdef INET6 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; #endif struct rwlock tcp_function_lock; static int sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", "Default TCP Maximum Segment Size"); #ifdef INET6 static int sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_v6mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_v6mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", "Default TCP Maximum Segment Size for IPv6"); #endif /* INET6 */ /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds * of packets instead of one. The effect scales with the available * bandwidth and quickly saturates the CPU and network interface * with packet generation and sending. Set to zero to disable MINMSS * checking. This setting prevents us from sending too small packets. */ VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_minmss), 0, "Minimum TCP Maximum Segment Size"); VNET_DEFINE(int, tcp_do_rfc1323) = 1; SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc1323), 0, "Enable rfc1323 (high performance TCP) extensions"); static int tcp_log_debug = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); static int tcp_tcbhashsize; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); static int do_tcpdrain = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); -static VNET_DEFINE(int, icmp_may_rst) = 1; +VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; #define V_icmp_may_rst VNET(icmp_may_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_may_rst), 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); -static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; +VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_isn_reseed_interval), 0, "Seconds between reseeding of ISN secret"); static int tcp_soreceive_stream; SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); VNET_DEFINE(uma_zone_t, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) #ifdef TCP_HHOOK VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); #endif static int tcp_default_fb_init(struct tcpcb *tp); static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); static int tcp_default_handoff_ok(struct tcpcb *tp); static struct inpcb *tcp_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); static void tcp_mtudisc(struct inpcb *, int); static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr); static struct tcp_function_block tcp_def_funcblk = { .tfb_tcp_block_name = "freebsd", .tfb_tcp_output = tcp_output, .tfb_tcp_do_segment = tcp_do_segment, .tfb_tcp_ctloutput = tcp_default_ctloutput, .tfb_tcp_handoff_ok = tcp_default_handoff_ok, .tfb_tcp_fb_init = tcp_default_fb_init, .tfb_tcp_fb_fini = tcp_default_fb_fini, }; int t_functions_inited = 0; static int tcp_fb_cnt = 0; struct tcp_funchead t_functions; static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; static void init_tcp_functions(void) { if (t_functions_inited == 0) { TAILQ_INIT(&t_functions); rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); t_functions_inited = 1; } } static struct tcp_function_block * find_tcp_functions_locked(struct tcp_function_set *fs) { struct tcp_function *f; struct tcp_function_block *blk=NULL; TAILQ_FOREACH(f, &t_functions, tf_next) { if (strcmp(f->tf_name, fs->function_set_name) == 0) { blk = f->tf_fb; break; } } return(blk); } static struct tcp_function_block * find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) { struct tcp_function_block *rblk=NULL; struct tcp_function *f; TAILQ_FOREACH(f, &t_functions, tf_next) { if (f->tf_fb == blk) { rblk = blk; if (s) { *s = f; } break; } } return (rblk); } struct tcp_function_block * find_and_ref_tcp_functions(struct tcp_function_set *fs) { struct tcp_function_block *blk; rw_rlock(&tcp_function_lock); blk = find_tcp_functions_locked(fs); if (blk) refcount_acquire(&blk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(blk); } struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *blk) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = find_tcp_fb_locked(blk, NULL); if (rblk) refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(rblk); } static struct tcp_function_block * find_and_ref_tcp_default_fb(void) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = tcp_func_set_ptr; refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return (rblk); } void tcp_switch_back_to_default(struct tcpcb *tp) { struct tcp_function_block *tfb; KASSERT(tp->t_fb != &tcp_def_funcblk, ("%s: called by the built-in default stack", __func__)); /* * Release the old stack. This function will either find a new one * or panic. */ if (tp->t_fb->tfb_tcp_fb_fini != NULL) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); /* * Now, we'll find a new function block to use. * Start by trying the current user-selected * default, unless this stack is the user-selected * default. */ tfb = find_and_ref_tcp_default_fb(); if (tfb == tp->t_fb) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Does the stack accept this connection? */ if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Try to use that stack. */ if (tfb != NULL) { /* Initialize the new stack. If it succeeds, we are done. */ tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init == NULL || (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) return; /* * Initialization failed. Release the reference count on * the stack. */ refcount_release(&tfb->tfb_refcnt); } /* * If that wasn't feasible, use the built-in default * stack which is not allowed to reject anyone. */ tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); if (tfb == NULL) { /* there always should be a default */ panic("Can't refer to tcp_def_funcblk"); } if (tfb->tfb_tcp_handoff_ok != NULL) { if ((*tfb->tfb_tcp_handoff_ok) (tp)) { /* The default stack cannot say no */ panic("Default stack rejects a new session?"); } } tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init != NULL && (*tp->t_fb->tfb_tcp_fb_init)(tp)) { /* The default stack cannot fail */ panic("Default stack initialization failed"); } } static int sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) { int error=ENOENT; struct tcp_function_set fs; struct tcp_function_block *blk; memset(&fs, 0, sizeof(fs)); rw_rlock(&tcp_function_lock); blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); if (blk) { /* Found him */ strcpy(fs.function_set_name, blk->tfb_tcp_block_name); fs.pcbcnt = blk->tfb_refcnt; } rw_runlock(&tcp_function_lock); error = sysctl_handle_string(oidp, fs.function_set_name, sizeof(fs.function_set_name), req); /* Check for error or no change */ if (error != 0 || req->newptr == NULL) return(error); rw_wlock(&tcp_function_lock); blk = find_tcp_functions_locked(&fs); if ((blk == NULL) || (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { error = ENOENT; goto done; } tcp_func_set_ptr = blk; done: rw_wunlock(&tcp_function_lock); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, CTLTYPE_STRING | CTLFLAG_RW, NULL, 0, sysctl_net_inet_default_tcp_functions, "A", "Set/get the default TCP functions"); static int sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) { int error, cnt, linesz; struct tcp_function *f; char *buffer, *cp; size_t bufsz, outsz; bool alias; cnt = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { cnt++; } rw_runlock(&tcp_function_lock); bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; buffer = malloc(bufsz, M_TEMP, M_WAITOK); error = 0; cp = buffer; linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', "Alias", "PCB count"); cp += linesz; bufsz -= linesz; outsz = linesz; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", f->tf_fb->tfb_tcp_block_name, (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', alias ? f->tf_name : "-", f->tf_fb->tfb_refcnt); if (linesz >= bufsz) { error = EOVERFLOW; break; } cp += linesz; bufsz -= linesz; outsz += linesz; } rw_runlock(&tcp_function_lock); if (error == 0) error = sysctl_handle_string(oidp, buffer, outsz + 1, req); free(buffer, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, CTLTYPE_STRING|CTLFLAG_RD, NULL, 0, sysctl_net_inet_list_available, "A", "list available TCP Function sets"); /* * Exports one (struct tcp_function_info) for each alias/name. */ static int sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) { int cnt, error; struct tcp_function *f; struct tcp_function_info tfi; /* * We don't allow writes. */ if (req->newptr != NULL) return (EINVAL); /* * Wire the old buffer so we can directly copy the functions to * user space without dropping the lock. */ if (req->oldptr != NULL) { error = sysctl_wire_old_buffer(req, 0); if (error) return (error); } /* * Walk the list and copy out matching entries. If INVARIANTS * is compiled in, also walk the list to verify the length of * the list matches what we have recorded. */ rw_rlock(&tcp_function_lock); cnt = 0; #ifndef INVARIANTS if (req->oldptr == NULL) { cnt = tcp_fb_cnt; goto skip_loop; } #endif TAILQ_FOREACH(f, &t_functions, tf_next) { #ifdef INVARIANTS cnt++; #endif if (req->oldptr != NULL) { tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; tfi.tfi_id = f->tf_fb->tfb_id; (void)strncpy(tfi.tfi_alias, f->tf_name, TCP_FUNCTION_NAME_LEN_MAX); tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; (void)strncpy(tfi.tfi_name, f->tf_fb->tfb_tcp_block_name, TCP_FUNCTION_NAME_LEN_MAX); tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); /* * Don't stop on error, as that is the * mechanism we use to accumulate length * information if the buffer was too short. */ } } KASSERT(cnt == tcp_fb_cnt, ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); #ifndef INVARIANTS skip_loop: #endif rw_runlock(&tcp_function_lock); if (req->oldptr == NULL) error = SYSCTL_OUT(req, NULL, (cnt + 1) * sizeof(struct tcp_function_info)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", "List TCP function block name-to-ID mappings"); /* * tfb_tcp_handoff_ok() function for the default stack. * Note that we'll basically try to take all comers. */ static int tcp_default_handoff_ok(struct tcpcb *tp) { return (0); } /* * tfb_tcp_fb_init() function for the default stack. * * This handles making sure we have appropriate timers set if you are * transitioning a socket that has some amount of setup done. * * The init() fuction from the default can *never* return non-zero i.e. * it is required to always succeed since it is the stack of last resort! */ static int tcp_default_fb_init(struct tcpcb *tp) { struct socket *so; INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, ("%s: connection %p in unexpected state %d", __func__, tp, tp->t_state)); /* * Nothing to do for ESTABLISHED or LISTEN states. And, we don't * know what to do for unexpected states (which includes TIME_WAIT). */ if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) return (0); /* * Make sure some kind of transmission timer is set if there is * outstanding data. */ so = tp->t_inpcb->inp_socket; if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || tcp_timer_active(tp, TT_PERSIST))) { /* * If the session has established and it looks like it should * be in the persist state, set the persist timer. Otherwise, * set the retransmit timer. */ if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && (int32_t)(tp->snd_nxt - tp->snd_una) < (int32_t)sbavail(&so->so_snd)) tcp_setpersist(tp); else tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); } /* All non-embryonic sessions get a keepalive timer. */ if (!tcp_timer_active(tp, TT_KEEP)) tcp_timer_activate(tp, TT_KEEP, TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : TP_KEEPINIT(tp)); return (0); } /* * tfb_tcp_fb_fini() function for the default stack. * * This changes state as necessary (or prudent) to prepare for another stack * to assume responsibility for the connection. */ static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) { INP_WLOCK_ASSERT(tp->t_inpcb); return; } /* * Target size of TCP PCB hash tables. Must be a power of two. * * Note that this can be overridden by the kernel environment * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE #define TCBHASHSIZE 0 #endif /* * XXX * Callouts should be moved into struct tcp directly. They are currently * separate because the tcpcb structure is exported to userland for sysctl * parsing purposes, which do not know about callouts. */ struct tcpcb_mem { struct tcpcb tcb; struct tcp_timer tt; struct cc_var ccv; #ifdef TCP_HHOOK struct osd osd; #endif }; -static VNET_DEFINE(uma_zone_t, tcpcb_zone); +VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); #define V_tcpcb_zone VNET(tcpcb_zone) MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); static struct mtx isn_mtx; #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) #define ISN_LOCK() mtx_lock(&isn_mtx) #define ISN_UNLOCK() mtx_unlock(&isn_mtx) /* * TCP initialization. */ static void tcp_zone_change(void *tag) { uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_tcpcb_zone, maxsockets); tcp_tw_zone_change(); } static int tcp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp = mem; INP_LOCK_INIT(inp, "inp", "tcpinp"); return (0); } /* * Take a value and get the next power of 2 that doesn't overflow. * Used to size the tcp_inpcb hash buckets. */ static int maketcp_hashsize(int size) { int hashsize; /* * auto tune. * get the next power of 2 higher than maxsockets. */ hashsize = 1 << fls(size); /* catch overflow, and just go one power of 2 smaller */ if (hashsize < size) { hashsize = 1 << (fls(size) - 1); } return (hashsize); } static volatile int next_tcp_stack_id = 1; /* * Register a TCP function block with the name provided in the names * array. (Note that this function does NOT automatically register * blk->tfb_tcp_block_name as a stack name. Therefore, you should * explicitly include blk->tfb_tcp_block_name in the list of names if * you wish to register the stack with that name.) * * Either all name registrations will succeed or all will fail. If * a name registration fails, the function will update the num_names * argument to point to the array index of the name that encountered * the failure. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, const char *names[], int *num_names) { struct tcp_function *n; struct tcp_function_set fs; int error, i; KASSERT(names != NULL && *num_names > 0, ("%s: Called with 0-length name list", __func__)); KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); if (t_functions_inited == 0) { init_tcp_functions(); } if ((blk->tfb_tcp_output == NULL) || (blk->tfb_tcp_do_segment == NULL) || (blk->tfb_tcp_ctloutput == NULL) || (strlen(blk->tfb_tcp_block_name) == 0)) { /* * These functions are required and you * need a name. */ *num_names = 0; return (EINVAL); } if (blk->tfb_tcp_timer_stop_all || blk->tfb_tcp_timer_activate || blk->tfb_tcp_timer_active || blk->tfb_tcp_timer_stop) { /* * If you define one timer function you * must have them all. */ if ((blk->tfb_tcp_timer_stop_all == NULL) || (blk->tfb_tcp_timer_activate == NULL) || (blk->tfb_tcp_timer_active == NULL) || (blk->tfb_tcp_timer_stop == NULL)) { *num_names = 0; return (EINVAL); } } refcount_init(&blk->tfb_refcnt, 0); blk->tfb_flags = 0; blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); for (i = 0; i < *num_names; i++) { n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); if (n == NULL) { error = ENOMEM; goto cleanup; } n->tf_fb = blk; (void)strncpy(fs.function_set_name, names[i], TCP_FUNCTION_NAME_LEN_MAX); fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; rw_wlock(&tcp_function_lock); if (find_tcp_functions_locked(&fs) != NULL) { /* Duplicate name space not allowed */ rw_wunlock(&tcp_function_lock); free(n, M_TCPFUNCTIONS); error = EALREADY; goto cleanup; } (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX); n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; TAILQ_INSERT_TAIL(&t_functions, n, tf_next); tcp_fb_cnt++; rw_wunlock(&tcp_function_lock); } return(0); cleanup: /* * Deregister the names we just added. Because registration failed * for names[i], we don't need to deregister that name. */ *num_names = i; rw_wlock(&tcp_function_lock); while (--i >= 0) { TAILQ_FOREACH(n, &t_functions, tf_next) { if (!strncmp(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX)) { TAILQ_REMOVE(&t_functions, n, tf_next); tcp_fb_cnt--; n->tf_fb = NULL; free(n, M_TCPFUNCTIONS); break; } } } rw_wunlock(&tcp_function_lock); return (error); } /* * Register a TCP function block using the name provided in the name * argument. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, int wait) { const char *name_list[1]; int num_names, rv; num_names = 1; if (name != NULL) name_list[0] = name; else name_list[0] = blk->tfb_tcp_block_name; rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); return (rv); } /* * Register a TCP function block using the name defined in * blk->tfb_tcp_block_name. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions(struct tcp_function_block *blk, int wait) { return (register_tcp_functions_as_name(blk, NULL, wait)); } /* * Deregister all names associated with a function block. This * functionally removes the function block from use within the system. * * When called with a true quiesce argument, mark the function block * as being removed so no more stacks will use it and determine * whether the removal would succeed. * * When called with a false quiesce argument, actually attempt the * removal. * * When called with a force argument, attempt to switch all TCBs to * use the default stack instead of returning EBUSY. * * Returns 0 on success (or if the removal would succeed, or an error * code on failure. */ int deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, bool force) { struct tcp_function *f; if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { /* You can't un-register the default */ return (EPERM); } rw_wlock(&tcp_function_lock); if (blk == tcp_func_set_ptr) { /* You can't free the current default */ rw_wunlock(&tcp_function_lock); return (EBUSY); } /* Mark the block so no more stacks can use it. */ blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; /* * If TCBs are still attached to the stack, attempt to switch them * to the default stack. */ if (force && blk->tfb_refcnt) { struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); rw_wunlock(&tcp_function_lock); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inp); if (inp->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inp); continue; } tp = intotcpcb(inp); if (tp == NULL || tp->t_fb != blk) { INP_WUNLOCK(inp); continue; } tcp_switch_back_to_default(tp); INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); rw_wlock(&tcp_function_lock); } if (blk->tfb_refcnt) { /* TCBs still attached. */ rw_wunlock(&tcp_function_lock); return (EBUSY); } if (quiesce) { /* Skip removal. */ rw_wunlock(&tcp_function_lock); return (0); } /* Remove any function names that map to this function block. */ while (find_tcp_fb_locked(blk, &f) != NULL) { TAILQ_REMOVE(&t_functions, f, tf_next); tcp_fb_cnt--; f->tf_fb = NULL; free(f, M_TCPFUNCTIONS); } rw_wunlock(&tcp_function_lock); return (0); } void tcp_init(void) { const char *tcbhash_tuneable; int hashsize; tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; #ifdef TCP_HHOOK if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); #endif hashsize = TCBHASHSIZE; TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); if (hashsize == 0) { /* * Auto tune the hash size based on maxsockets. * A perfect hash would have a 1:1 mapping * (hashsize = maxsockets) however it's been * suggested that O(2) average is better. */ hashsize = maketcp_hashsize(maxsockets / 4); /* * Our historical default is 512, * do not autotune lower than this. */ if (hashsize < 512) hashsize = 512; if (bootverbose && IS_DEFAULT_VNET(curvnet)) printf("%s: %s auto tuned to %d\n", __func__, tcbhash_tuneable, hashsize); } /* * We require a hashsize to be a power of two. * Previously if it was not a power of two we would just reset it * back to 512, which could be a nasty surprise if you did not notice * the error message. * Instead what we do is clip it to the closest power of two lower * than the specified hash value. */ if (!powerof2(hashsize)) { int oldhashsize = hashsize; hashsize = maketcp_hashsize(hashsize); /* prevent absurdly low value */ if (hashsize < 16) hashsize = 16; printf("%s: WARNING: TCB hash size not a power of 2, " "clipped from %d to %d.\n", __func__, oldhashsize, hashsize); } in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); /* * These have to be type stable for the benefit of the timers. */ V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_tcpcb_zone, maxsockets); uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); tcp_tw_init(); syncache_init(); tcp_hc_init(); TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); tcp_fastopen_init(); /* Skip initialization of globals for non-default instances. */ if (!IS_DEFAULT_VNET(curvnet)) return; tcp_reass_global_init(); /* XXX virtualize those bellow? */ tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; tcp_rexmit_min = TCPTV_MIN; if (tcp_rexmit_min < 1) tcp_rexmit_min = 1; tcp_persmin = TCPTV_PERSMIN; tcp_persmax = TCPTV_PERSMAX; tcp_rexmit_slop = TCPTV_CPU_VAR; tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; tcp_tcbhashsize = hashsize; /* Setup the tcp function block list */ init_tcp_functions(); register_tcp_functions(&tcp_def_funcblk, M_WAITOK); #ifdef TCP_BLACKBOX /* Initialize the TCP logging data. */ tcp_log_init(); #endif if (tcp_soreceive_stream) { #ifdef INET tcp_usrreqs.pru_soreceive = soreceive_stream; #endif #ifdef INET6 tcp6_usrreqs.pru_soreceive = soreceive_stream; #endif /* INET6 */ } #ifdef INET6 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ if (max_protohdr < TCP_MINPROTOHDR) max_protohdr = TCP_MINPROTOHDR; if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) panic("tcp_init"); #undef TCP_MINPROTOHDR ISN_LOCK_INIT(); EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, SHUTDOWN_PRI_DEFAULT); EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, EVENTHANDLER_PRI_ANY); #ifdef TCPPCAP tcp_pcap_init(); #endif } #ifdef VIMAGE static void tcp_destroy(void *unused __unused) { int n; #ifdef TCP_HHOOK int error; #endif /* * All our processes are gone, all our sockets should be cleaned * up, which means, we should be past the tcp_discardcb() calls. * Sleep to let all tcpcb timers really disappear and cleanup. */ for (;;) { INP_LIST_RLOCK(&V_tcbinfo); n = V_tcbinfo.ipi_count; INP_LIST_RUNLOCK(&V_tcbinfo); if (n == 0) break; pause("tcpdes", hz / 10); } tcp_hc_destroy(); syncache_destroy(); tcp_tw_destroy(); in_pcbinfo_destroy(&V_tcbinfo); /* tcp_discardcb() clears the sack_holes up. */ uma_zdestroy(V_sack_hole_zone); uma_zdestroy(V_tcpcb_zone); /* * Cannot free the zone until all tcpcbs are released as we attach * the allocations to them. */ tcp_fastopen_destroy(); #ifdef TCP_HHOOK error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); } error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); } #endif } VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); #endif void tcp_fini(void *xtp) { } /* * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. * tcp_template used to store this data in mbufs, but we now recopy it out * of the tcpcb each time to conserve mbufs. */ void tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) { struct tcphdr *th = (struct tcphdr *)tcp_ptr; INP_WLOCK_ASSERT(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { struct ip *ip; ip = (struct ip *)ip_ptr; ip->ip_v = IPVERSION; ip->ip_hl = 5; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = 0; ip->ip_id = 0; ip->ip_off = 0; ip->ip_ttl = inp->inp_ip_ttl; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } #endif /* INET */ th->th_sport = inp->inp_lport; th->th_dport = inp->inp_fport; th->th_seq = 0; th->th_ack = 0; th->th_x2 = 0; th->th_off = 5; th->th_flags = 0; th->th_win = 0; th->th_urp = 0; th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ } /* * Create template to be used to send tcp packets on a connection. * Allocates an mbuf and fills in a skeletal tcp/ip header. The only * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * tcpip_maketemplate(struct inpcb *inp) { struct tcptemp *t; t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); if (t == NULL) return (NULL); tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); return (t); } /* * Send a single message to the TCP at address specified by * the given TCP/IP header. If m == NULL, then we make a copy * of the tcpiphdr at th and send directly to the addressed host. * This is used to force keep alive messages out using the TCP * template for a connection. If flags are given then we send * a message back to the TCP which originated the segment th, * and discard the mbuf containing it and any other attached mbufs. * * In any case the ack and sequence number of the transmitted * segment are as specified by the parameters. * * NOTE: If m != NULL, then th must point to *inside* the mbuf. */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) { struct tcpopt to; struct inpcb *inp; struct ip *ip; struct mbuf *optm; struct tcphdr *nth; u_char *optp; #ifdef INET6 struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ int optlen, tlen, win; bool incl_opts; KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); #ifdef INET6 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); ip6 = ipgen; #endif /* INET6 */ ip = ipgen; if (tp != NULL) { inp = tp->t_inpcb; KASSERT(inp != NULL, ("tcp control block w/o inpcb")); INP_WLOCK_ASSERT(inp); } else inp = NULL; incl_opts = false; win = 0; if (tp != NULL) { if (!(flags & TH_RST)) { win = sbspace(&inp->inp_socket->so_rcv); if (win > TCP_MAXWIN << tp->rcv_scale) win = TCP_MAXWIN << tp->rcv_scale; } if ((tp->t_flags & TF_NOOPT) == 0) incl_opts = true; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; m->m_data += max_linkhdr; #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); flags = TH_ACK; } else if (!M_WRITABLE(m)) { struct mbuf *n; /* Can't reuse 'm', allocate a new mbuf. */ n = m_gethdr(M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return; } if (!m_dup_pkthdr(n, m, M_NOWAIT)) { m_freem(m); m_freem(n); return; } n->m_data += max_linkhdr; /* m_len is set later */ #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(n, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(n, struct ip6_hdr *); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); ip = mtod(n, struct ip *); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); xchg(nth->th_dport, nth->th_sport, uint16_t); th = nth; m_freem(m); m = n; } else { /* * reuse the mbuf. * XXX MRT We inherit the FIB, which is lucky. */ m_freem(m->m_next); m->m_next = NULL; m->m_data = (caddr_t)ipgen; /* m_len is set later */ #ifdef INET6 if (isipv6) { xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } if (th != nth) { /* * this is usually a case when an extension header * exists between the IPv6 header and the * TCP header. */ nth->th_sport = th->th_sport; nth->th_dport = th->th_dport; } xchg(nth->th_dport, nth->th_sport, uint16_t); #undef xchg } tlen = 0; #ifdef INET6 if (isipv6) tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tlen = sizeof (struct tcpiphdr); #endif #ifdef INVARIANTS m->m_len = 0; KASSERT(M_TRAILINGSPACE(m) >= tlen, ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", m, tlen, (long)M_TRAILINGSPACE(m))); #endif m->m_len = tlen; to.to_flags = 0; if (incl_opts) { /* Make sure we have room. */ if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { m->m_next = m_get(M_NOWAIT, MT_DATA); if (m->m_next) { optp = mtod(m->m_next, u_char *); optm = m->m_next; } else incl_opts = false; } else { optp = (u_char *) (nth + 1); optm = m; } } if (incl_opts) { /* Timestamps. */ if (tp->t_flags & TF_RCVD_TSTMP) { to.to_tsval = tcp_ts_getticks() + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* Add the options. */ tlen += optlen = tcp_addoptions(&to, optp); /* Update m_len in the correct mbuf. */ optm->m_len += optlen; } else optlen = 0; #ifdef INET6 if (isipv6) { ip6->ip6_flow = 0; ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(tlen - sizeof(*ip6)); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(tlen); ip->ip_ttl = V_ip_defttl; if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); } #endif m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef MAC if (inp != NULL) { /* * Packet is associated with a socket, so allow the * label of the response to reflect the socket label. */ INP_WLOCK_ASSERT(inp); mac_inpcb_create_mbuf(inp, m); } else { /* * Packet is not associated with a socket, so possibly * update the label in place. */ mac_netinet_tcp_reply(m); } #endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; nth->th_flags = flags; if (tp != NULL) nth->th_win = htons((u_short) (win >> tp->rcv_scale)); else nth->th_win = htons((u_short)win); nth->th_urp = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { m_freem(m); return; } } #endif m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #ifdef INET6 if (isipv6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; nth->th_sum = in6_cksum_pseudo(ip6, tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : NULL, NULL); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); } #endif /* INET */ #ifdef TCPDEBUG if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); #endif TCP_PROBE3(debug__output, tp, th, m); if (flags & TH_RST) TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); #ifdef INET6 if (isipv6) { TCP_PROBE5(send, NULL, tp, ip6, tp, nth); (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { TCP_PROBE5(send, NULL, tp, ip, tp, nth); (void)ip_output(m, NULL, NULL, 0, NULL, inp); } #endif } /* * Create a new TCP control block, making an * empty reassembly queue and hooking it to the argument * protocol control block. The `inp' parameter must have * come from the zone allocator set up in tcp_init(). */ struct tcpcb * tcp_newtcpcb(struct inpcb *inp) { struct tcpcb_mem *tm; struct tcpcb *tp; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) return (NULL); tp = &tm->tcb; /* Initialise cc_var struct for this tcpcb. */ tp->ccv = &tm->ccv; tp->ccv->type = IPPROTO_TCP; tp->ccv->ccvc.tcp = tp; rw_rlock(&tcp_function_lock); tp->t_fb = tcp_func_set_ptr; refcount_acquire(&tp->t_fb->tfb_refcnt); rw_runlock(&tcp_function_lock); /* * Use the current system default CC algorithm. */ CC_LIST_RLOCK(); KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); CC_ALGO(tp) = CC_DEFAULT(); CC_LIST_RUNLOCK(); if (CC_ALGO(tp)->cb_init != NULL) if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #ifdef TCP_HHOOK tp->osd = &tm->osd; if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #endif #ifdef VIMAGE tp->t_vnet = inp->inp_vnet; #endif tp->t_timers = &tm->tt; /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ tp->t_maxseg = #ifdef INET6 isipv6 ? V_tcp_v6mssdflt : #endif /* INET6 */ V_tcp_mssdflt; /* Set up our timeouts. */ callout_init(&tp->t_timers->tt_rexmt, 1); callout_init(&tp->t_timers->tt_persist, 1); callout_init(&tp->t_timers->tt_keep, 1); callout_init(&tp->t_timers->tt_2msl, 1); callout_init(&tp->t_timers->tt_delack, 1); if (V_tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); if (V_tcp_do_sack) tp->t_flags |= TF_SACK_PERMIT; TAILQ_INIT(&tp->snd_holes); /* * The tcpcb will hold a reference on its inpcb until tcp_discardcb() * is called. */ in_pcbref(inp); /* Reference for tcpcb */ tp->t_inpcb = inp; /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_rexmit_min; tp->t_rxtcur = TCPTV_RTOBASE; tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = ticks; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ inp->inp_ip_ttl = V_ip_defttl; inp->inp_ppcb = tp; #ifdef TCPPCAP /* * Init the TCP PCAP queues. */ tcp_pcap_tcpcb_init(tp); #endif #ifdef TCP_BLACKBOX /* Initialize the per-TCPCB log data. */ tcp_log_tcpcbinit(tp); #endif if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } return (tp); /* XXX */ } /* * Switch the congestion control algorithm back to NewReno for any active * control blocks using an algorithm which is about to go away. * This ensures the CC framework can allow the unload to proceed without leaving * any dangling pointers which would trigger a panic. * Returning non-zero would inform the CC framework that something went wrong * and it would be unsafe to allow the unload to proceed. However, there is no * way for this to occur with this implementation so we always return zero. */ int tcp_ccalgounload(struct cc_algo *unload_algo) { struct cc_algo *tmpalgo; struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); /* * Check all active control blocks across all network stacks and change * any that are using "unload_algo" back to NewReno. If "unload_algo" * requires cleanup code to be run, call it. */ VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); /* * New connections already part way through being initialised * with the CC algo we're removing will not race with this code * because the INP_INFO_WLOCK is held during initialisation. We * therefore don't enter the loop below until the connection * list has stabilised. */ CK_LIST_FOREACH(inp, &V_tcb, inp_list) { INP_WLOCK(inp); /* Important to skip tcptw structs. */ if (!(inp->inp_flags & INP_TIMEWAIT) && (tp = intotcpcb(inp)) != NULL) { /* * By holding INP_WLOCK here, we are assured * that the connection is not currently * executing inside the CC module's functions * i.e. it is safe to make the switch back to * NewReno. */ if (CC_ALGO(tp) == unload_algo) { tmpalgo = CC_ALGO(tp); if (tmpalgo->cb_destroy != NULL) tmpalgo->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; /* * NewReno may allocate memory on * demand for certain stateful * configuration as needed, but is * coded to never fail on memory * allocation failure so it is a safe * fallback. */ CC_ALGO(tp) = &newreno_cc_algo; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } /* * Drop a TCP connection, reporting * the specified error. If connection is synchronized, * then send a RST to peer. */ struct tcpcb * tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); if (TCPS_HAVERCVDSYN(tp->t_state)) { tcp_state_change(tp, TCPS_CLOSED); (void) tp->t_fb->tfb_tcp_output(tp); TCPSTAT_INC(tcps_drops); } else TCPSTAT_INC(tcps_conndrops); if (errno == ETIMEDOUT && tp->t_softerror) errno = tp->t_softerror; so->so_error = errno; return (tcp_close(tp)); } void tcp_discardcb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ int released __unused; INP_WLOCK_ASSERT(inp); /* * Make sure that all of our timers are stopped before we delete the * PCB. * * If stopping a timer fails, we schedule a discard function in same * callout, and the last discard function called will take care of * deleting the tcpcb. */ tp->t_timers->tt_draincnt = 0; tcp_timer_stop(tp, TT_REXMT); tcp_timer_stop(tp, TT_PERSIST); tcp_timer_stop(tp, TT_KEEP); tcp_timer_stop(tp, TT_2MSL); tcp_timer_stop(tp, TT_DELACK); if (tp->t_fb->tfb_tcp_timer_stop_all) { /* * Call the stop-all function of the methods, * this function should call the tcp_timer_stop() * method with each of the function specific timeouts. * That stop will be called via the tfb_tcp_timer_stop() * which should use the async drain function of the * callout system (see tcp_var.h). */ tp->t_fb->tfb_tcp_timer_stop_all(tp); } /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. * 'Enough' is arbitrarily defined as 4 rtt samples. * 4 samples is enough for the srtt filter to converge * to within enough % of the correct value; fewer samples * and we could save a bogus rtt. The danger is not high * as tcp quickly recovers from everything. * XXX: Works very well but needs some more statistics! */ if (tp->t_rttupdated >= 4) { struct hc_metrics_lite metrics; uint32_t ssthresh; bzero(&metrics, sizeof(metrics)); /* * Update the ssthresh always when the conditions below * are satisfied. This gives us better new start value * for the congestion avoidance for new connections. * ssthresh is only set if packet loss occurred on a session. * * XXXRW: 'so' may be NULL here, and/or socket buffer may be * being torn down. Ideally this code would not use 'so'. */ ssthresh = tp->snd_ssthresh; if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { /* * convert the limit from user data bytes to * packets then to packet data bytes. */ ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; if (ssthresh < 2) ssthresh = 2; ssthresh *= (tp->t_maxseg + #ifdef INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : #endif sizeof (struct tcpiphdr) #ifdef INET6 ) #endif ); } else ssthresh = 0; metrics.rmx_ssthresh = ssthresh; metrics.rmx_rtt = tp->t_srtt; metrics.rmx_rttvar = tp->t_rttvar; metrics.rmx_cwnd = tp->snd_cwnd; metrics.rmx_sendpipe = 0; metrics.rmx_recvpipe = 0; tcp_hc_update(&inp->inp_inc, &metrics); } /* free the reassembly queue, if any */ tcp_reass_flush(tp); #ifdef TCP_OFFLOAD /* Disconnect offload device, if any. */ if (tp->t_flags & TF_TOE) tcp_offload_detach(tp); #endif tcp_free_sackholes(tp); #ifdef TCPPCAP /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tp->t_inpkts)); tcp_pcap_drain(&(tp->t_outpkts)); #endif /* Allow the CC algorithm to clean up after itself. */ if (CC_ALGO(tp)->cb_destroy != NULL) CC_ALGO(tp)->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; #ifdef TCP_HHOOK khelp_destroy_osd(tp->osd); #endif CC_ALGO(tp) = NULL; inp->inp_ppcb = NULL; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); released = in_pcbrele_wlocked(inp); KASSERT(!released, ("%s: inp %p should not have been released " "here", __func__, inp)); } } void tcp_timer_discard(void *ptp) { struct inpcb *inp; struct tcpcb *tp; struct epoch_tracker et; tp = (struct tcpcb *)ptp; CURVNET_SET(tp->t_vnet); INP_INFO_RLOCK_ET(&V_tcbinfo, et); inp = tp->t_inpcb; KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp)); INP_WLOCK(inp); KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, ("%s: tcpcb has to be stopped here", __func__)); tp->t_timers->tt_draincnt--; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on this tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); if (in_pcbrele_wlocked(inp)) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); CURVNET_RESTORE(); return; } } INP_WUNLOCK(inp); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); CURVNET_RESTORE(); } /* * Attempt to close a TCP control block, marking it as dropped, and freeing * the socket if we hold the only reference. */ struct tcpcb * tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_state == TCPS_LISTEN) tcp_offload_listen_stop(tp); #endif /* * This releases the TFO pending counter resource for TFO listen * sockets as well as passively-created TFO sockets that transition * from SYN_RECEIVED to CLOSED. */ if (tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } in_pcbdrop(inp); TCPSTAT_INC(tcps_closed); if (tp->t_state != TCPS_CLOSED) tcp_state_change(tp, TCPS_CLOSED); KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); so = inp->inp_socket; soisdisconnected(so); if (inp->inp_flags & INP_SOCKREF) { KASSERT(so->so_state & SS_PROTOREF, ("tcp_close: !SS_PROTOREF")); inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); SOCK_LOCK(so); so->so_state &= ~SS_PROTOREF; sofree(so); return (NULL); } return (tp); } void tcp_drain(void) { VNET_ITERATOR_DECL(vnet_iter); if (!do_tcpdrain) return; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct inpcb *inpb; struct tcpcb *tcpb; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... * XXX: The "Net/3" implementation doesn't imply that the TCP * reassembly queue should be flushed, but in a situation * where we're really low on mbufs, this is potentially * useful. */ INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inpb); if (inpb->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inpb); continue; } if ((tcpb = intotcpcb(inpb)) != NULL) { tcp_reass_flush(tcpb); tcp_clean_sackreport(tcpb); #ifdef TCP_BLACKBOX tcp_log_drain(tcpb); #endif #ifdef TCPPCAP if (tcp_pcap_aggressive_free) { /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tcpb->t_inpkts)); tcp_pcap_drain(&(tcpb->t_outpkts)); } #endif } INP_WUNLOCK(inpb); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Notify a tcp user of an asynchronous error; * store error as soft error, but wake up user * (for now, won't do anything until can select for soft error). * * Do not wake up user since there currently is no mechanism for * reporting soft errors (yet - a kqueue filter may be added). */ static struct inpcb * tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, * and receives a second error, give up now. This is better * than waiting a long time to establish a connection that * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && (error == EHOSTUNREACH || error == ENETUNREACH || error == EHOSTDOWN)) { if (inp->inp_route.ro_rt) { RTFREE(inp->inp_route.ro_rt); inp->inp_route.ro_rt = (struct rtentry *)NULL; } return (inp); } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) { tp = tcp_drop(tp, error); if (tp != NULL) return (inp); else return (NULL); } else { tp->t_softerror = error; return (inp); } #if 0 wakeup( &so->so_timeo); sorwakeup(so); sowwakeup(so); #endif } static int tcp_pcblist(SYSCTL_HANDLER_ARGS) { int error, i, m, n, pcb_count; struct inpcb *inp, **inp_list; inp_gen_t gencnt; struct xinpgen xig; struct epoch_tracker et; /* * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ INP_LIST_RLOCK(&V_tcbinfo); gencnt = V_tcbinfo.ipi_gencnt; n = V_tcbinfo.ipi_count; INP_LIST_RUNLOCK(&V_tcbinfo); m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) + (n + m) * sizeof(struct xtcpcb)); if (error != 0) return (error); xig.xig_len = sizeof xig; xig.xig_count = n + m; xig.xig_gen = gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); error = syncache_pcblist(req, m, &pcb_count); if (error) return (error); inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); INP_INFO_WLOCK(&V_tcbinfo); for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) { INP_WLOCK(inp); if (inp->inp_gencnt <= gencnt) { /* * XXX: This use of cr_cansee(), introduced with * TCP state changes, is not quite right, but for * now, better than nothing. */ if (inp->inp_flags & INP_TIMEWAIT) { if (intotw(inp) != NULL) error = cr_cansee(req->td->td_ucred, intotw(inp)->tw_cred); else error = EINVAL; /* Skip this inp. */ } else error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) { in_pcbref(inp); inp_list[i++] = inp; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); n = i; error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (inp->inp_gencnt <= gencnt) { struct xtcpcb xt; tcp_inptoxtp(inp, &xt); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xt, sizeof xt); } else INP_RUNLOCK(inp); } INP_INFO_RLOCK_ET(&V_tcbinfo, et); for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); if (!error) { /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ INP_LIST_RLOCK(&V_tcbinfo); xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_tcbinfo.ipi_count + pcb_count; INP_LIST_RUNLOCK(&V_tcbinfo); error = SYSCTL_OUT(req, &xig, sizeof xig); } free(inp_list, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); #ifdef INET static int tcp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); #endif /* INET */ #ifdef INET6 static int tcp6_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in6 addrs[2]; struct inpcb *inp; int error; #ifdef INET int mapped = 0; #endif error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { return (error); } if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) mapped = 1; else #endif return (EINVAL); } #ifdef INET if (mapped == 1) inp = in_pcblookup(&V_tcbinfo, *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], addrs[1].sin6_port, *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); else #endif inp = in6_pcblookup(&V_tcbinfo, &addrs[1].sin6_addr, addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); #endif /* INET6 */ #ifdef INET void tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { struct ip *ip = vip; struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct icmp *icp; struct in_conninfo inc; struct epoch_tracker et; tcp_seq icmp_tcp_seq; int mtu; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip == NULL) { in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); return; } icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); INP_INFO_RLOCK_ET(&V_tcbinfo, et); inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } icmp_tcp_seq = th->th_seq; if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohs(icp->icmp_nextmtu); /* * If no alternative MTU was * proposed, try the next smaller * one. */ if (!mtu) mtu = ip_next_mtu( ntohs(ip->ip_len), 1); if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) mtu = V_tcp_minmss + sizeof(struct tcpiphdr); /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof(struct tcpiphdr)) { bzero(&inc, sizeof(inc)); inc.inc_faddr = faddr; inc.inc_fibnum = inp->inp_inc.inc_fibnum; tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); } } else inp = (*notify)(inp, inetctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fport = th->th_dport; inc.inc_lport = th->th_sport; inc.inc_faddr = faddr; inc.inc_laddr = ip->ip_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); } #endif /* INET */ #ifdef INET6 void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { struct in6_addr *dst; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; struct inpcb *inp; struct tcpcb *tp; struct icmp6_hdr *icmp6; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; struct in_conninfo inc; struct epoch_tracker et; struct tcp_ports { uint16_t th_sport; uint16_t th_dport; } t_ports; tcp_seq icmp_tcp_seq; unsigned int mtu; unsigned int off; if (sa->sa_family != AF_INET6 || sa->sa_len != sizeof(struct sockaddr_in6)) return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; icmp6 = ip6cp->ip6c_icmp6; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; dst = ip6cp->ip6c_finaldst; } else { m = NULL; ip6 = NULL; off = 0; /* fool gcc */ sa6_src = &sa6_any; dst = NULL; } if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip6 = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) return; if (ip6 == NULL) { in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); return; } /* Check if we can safely get the ports from the tcp hdr */ if (m == NULL || (m->m_pkthdr.len < (int32_t) (off + sizeof(struct tcp_ports)))) { return; } bzero(&t_ports, sizeof(struct tcp_ports)); m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); INP_INFO_RLOCK_ET(&V_tcbinfo, et); inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } off += sizeof(struct tcp_ports); if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { goto out; } m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohl(icmp6->icmp6_mtu); /* * If no alternative MTU was * proposed, or the proposed * MTU was too small, set to * the min. */ if (mtu < IPV6_MMTU) mtu = IPV6_MMTU - 8; bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL)) goto out; /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof (struct tcphdr) + sizeof (struct ip6_hdr)) { tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); ICMP6STAT_INC(icp6s_pmtuchg); } } else inp = (*notify)(inp, inet6ctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc_fport = t_ports.th_dport; inc.inc_lport = t_ports.th_sport; inc.inc6_faddr = *dst; inc.inc6_laddr = ip6->ip6_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); } #endif /* INET6 */ /* * Following is where TCP initial sequence number generation occurs. * * There are two places where we must use initial sequence numbers: * 1. In SYN-ACK packets. * 2. In SYN packets. * * All ISNs for SYN-ACK packets are generated by the syncache. See * tcp_syncache.c for details. * * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling * depends on this property. In addition, these ISNs should be * unguessable so as to prevent connection hijacking. To satisfy * the requirements of this situation, the algorithm outlined in * RFC 1948 is used, with only small modifications. * * Implementation details: * * Time is based off the system timer, and is corrected so that it * increases by one megabyte per second. This allows for proper * recycling on high speed LANs while still leaving over an hour * before rollover. * * As reading the *exact* system time is too expensive to be done * whenever setting up a TCP connection, we increment the time * offset in two ways. First, a small random positive increment * is added to isn_offset for each connection that is set up. * Second, the function tcp_isn_tick fires once per clock tick * and increments isn_offset as necessary so that sequence numbers * are incremented at approximately ISN_BYTES_PER_SECOND. The * random positive increments serve only to ensure that the same * exact sequence number is never sent out twice (as could otherwise * happen when a port is recycled in less than the system tick * interval.) * * net.inet.tcp.isn_reseed_interval controls the number of seconds * between seeding of isn_secret. This is normally set to zero, * as reseeding should not be necessary. * * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In * general, this means holding an exclusive (write) lock. */ #define ISN_BYTES_PER_SECOND 1048576 #define ISN_STATIC_INCREMENT 4096 #define ISN_RANDOM_INCREMENT (4096 - 1) -static VNET_DEFINE(u_char, isn_secret[32]); -static VNET_DEFINE(int, isn_last); -static VNET_DEFINE(int, isn_last_reseed); -static VNET_DEFINE(u_int32_t, isn_offset); -static VNET_DEFINE(u_int32_t, isn_offset_old); +VNET_DEFINE_STATIC(u_char, isn_secret[32]); +VNET_DEFINE_STATIC(int, isn_last); +VNET_DEFINE_STATIC(int, isn_last_reseed); +VNET_DEFINE_STATIC(u_int32_t, isn_offset); +VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); #define V_isn_secret VNET(isn_secret) #define V_isn_last VNET(isn_last) #define V_isn_last_reseed VNET(isn_last_reseed) #define V_isn_offset VNET(isn_offset) #define V_isn_offset_old VNET(isn_offset_old) tcp_seq tcp_new_isn(struct tcpcb *tp) { MD5_CTX isn_ctx; u_int32_t md5_buffer[4]; tcp_seq new_isn; u_int32_t projected_offset; INP_WLOCK_ASSERT(tp->t_inpcb); ISN_LOCK(); /* Seed if this is the first use, reseed if requested. */ if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) < (u_int)ticks))) { read_random(&V_isn_secret, sizeof(V_isn_secret)); V_isn_last_reseed = ticks; } /* Compute the md5 hash and return the ISN. */ MD5Init(&isn_ctx); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); #ifdef INET6 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, sizeof(struct in6_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, sizeof(struct in6_addr)); } else #endif { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, sizeof(struct in_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, sizeof(struct in_addr)); } MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); MD5Final((u_char *) &md5_buffer, &isn_ctx); new_isn = (tcp_seq) md5_buffer[0]; V_isn_offset += ISN_STATIC_INCREMENT + (arc4random() & ISN_RANDOM_INCREMENT); if (ticks != V_isn_last) { projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); if (SEQ_GT(projected_offset, V_isn_offset)) V_isn_offset = projected_offset; V_isn_offset_old = V_isn_offset; V_isn_last = ticks; } new_isn += V_isn_offset; ISN_UNLOCK(); return (new_isn); } /* * When a specific ICMP unreachable message is received and the * connection state is SYN-SENT, drop the connection. This behavior * is controlled by the icmp_may_rst sysctl. */ struct inpcb * tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp; INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); if (tp->t_state != TCPS_SYN_SENT) return (inp); if (IS_FASTOPEN(tp->t_flags)) tcp_fastopen_disable_path(tp); tp = tcp_drop(tp, errno); if (tp != NULL) return (inp); else return (NULL); } /* * When `need fragmentation' ICMP is received, update our idea of the MSS * based on the new value. Also nudge TCP to send something, since we * know the packet we just sent was dropped. * This duplicates some code in the tcp_mss() function in tcp_input.c. */ static struct inpcb * tcp_mtudisc_notify(struct inpcb *inp, int error) { tcp_mtudisc(inp, -1); return (inp); } static void tcp_mtudisc(struct inpcb *inp, int mtuoffer) { struct tcpcb *tp; struct socket *so; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return; tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); /* If the mss is larger than the socket buffer, decrease the mss. */ if (so->so_snd.sb_hiwat < tp->t_maxseg) tp->t_maxseg = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); TCPSTAT_INC(tcps_mturesent); tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; tcp_free_sackholes(tp); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_SACK_PERMIT) EXIT_FASTRECOVERY(tp->t_flags); tp->t_fb->tfb_tcp_output(tp); } #ifdef INET /* * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated, then return 0. This routine * is called by TCP routines that access the rmx structure and by * tcp_mss_update to get the peer/interface MTU. */ uint32_t tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop4_extended nh4; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); if (inc->inc_faddr.s_addr != INADDR_ANY) { if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, NHR_REF, 0, &nh4) != 0) return (0); ifp = nh4.nh_ifp; maxmtu = nh4.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO4 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib4_free_nh_ext(inc->inc_fibnum, &nh4); } return (maxmtu); } #endif /* INET */ #ifdef INET6 uint32_t tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop6_extended nh6; struct in6_addr dst6; uint32_t scopeid; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 0, &nh6) != 0) return (0); ifp = nh6.nh_ifp; maxmtu = nh6.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO6 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib6_free_nh_ext(inc->inc_fibnum, &nh6); } return (maxmtu); } #endif /* INET6 */ /* * Calculate effective SMSS per RFC5681 definition for a given TCP * connection at its current state, taking into account SACK and etc. */ u_int tcp_maxseg(const struct tcpcb *tp) { u_int optlen; if (tp->t_flags & TF_NOOPT) return (tp->t_maxseg); /* * Here we have a simplified code from tcp_addoptions(), * without a proper loop, and having most of paddings hardcoded. * We might make mistakes with padding here in some edge cases, * but this is harmless, since result of tcp_maxseg() is used * only in cwnd and ssthresh estimations. */ #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) if (TCPS_HAVEESTABLISHED(tp->t_state)) { if (tp->t_flags & TF_RCVD_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { optlen += TCPOLEN_SACKHDR; optlen += tp->rcv_numsacks * TCPOLEN_SACK; optlen = PAD(optlen); } } else { if (tp->t_flags & TF_REQ_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = PAD(TCPOLEN_MAXSEG); if (tp->t_flags & TF_REQ_SCALE) optlen += PAD(TCPOLEN_WINDOW); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if (tp->t_flags & TF_SACK_PERMIT) optlen += PAD(TCPOLEN_SACK_PERMITTED); } #undef PAD optlen = min(optlen, TCP_MAXOLEN); return (tp->t_maxseg - optlen); } static int sysctl_drop(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct tcpcb *tp; struct tcptw *tw; struct sockaddr_in *fin, *lin; struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } INP_INFO_RLOCK_ET(&V_tcbinfo, et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } if (inp != NULL) { if (inp->inp_flags & INP_TIMEWAIT) { /* * XXXRW: There currently exists a state where an * inpcb is present, but its timewait state has been * discarded. For now, don't allow dropping of this * type of inpcb. */ tw = intotw(inp); if (tw != NULL) tcp_twclose(tw, 0); else INP_WUNLOCK(inp); } else if (!(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { tp = intotcpcb(inp); tp = tcp_drop(tp, ECONNABORTED); if (tp != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } else error = ESRCH; INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 0, sysctl_drop, "", "Drop TCP connection"); /* * Generate a standardized TCP log line for use throughout the * tcp subsystem. Memory allocation is done with M_NOWAIT to * allow use in the interrupt context. * * NB: The caller MUST free(s, M_TCPLOG) the returned string. * NB: The function may return NULL if memory allocation failed. * * Due to header inclusion and ordering limitations the struct ip * and ip6_hdr pointers have to be passed as void pointers. */ char * tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_in_vain == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } char * tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_debug == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { char *s, *sp; size_t size; struct ip *ip; #ifdef INET6 const struct ip6_hdr *ip6; ip6 = (const struct ip6_hdr *)ip6hdr; #endif /* INET6 */ ip = (struct ip *)ip4hdr; /* * The log line looks like this: * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2" */ size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + sizeof(PRINT_TH_FLAGS) + 1 + #ifdef INET6 2 * INET6_ADDRSTRLEN; #else 2 * INET_ADDRSTRLEN; #endif /* INET6 */ s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); if (s == NULL) return (NULL); strcat(s, "TCP: ["); sp = s + strlen(s); if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { inet_ntoa_r(inc->inc_faddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); inet_ntoa_r(inc->inc_laddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); #ifdef INET6 } else if (inc) { ip6_sprintf(sp, &inc->inc6_faddr); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); ip6_sprintf(sp, &inc->inc6_laddr); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); } else if (ip6 && th) { ip6_sprintf(sp, &ip6->ip6_src); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); ip6_sprintf(sp, &ip6->ip6_dst); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET6 */ #ifdef INET } else if (ip && th) { inet_ntoa_r(ip->ip_src, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); inet_ntoa_r(ip->ip_dst, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET */ } else { free(s, M_TCPLOG); return (NULL); } sp = s + strlen(s); if (th) sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); if (*(s + size - 1) != '\0') panic("%s: string too long", __func__); return (s); } /* * A subroutine which makes it easy to track TCP state changes with DTrace. * This function shouldn't be called for t_state initializations that don't * correspond to actual TCP state transitions. */ void tcp_state_change(struct tcpcb *tp, int newstate) { #if defined(KDTRACE_HOOKS) int pstate = tp->t_state; #endif TCPSTATES_DEC(tp->t_state); TCPSTATES_INC(newstate); tp->t_state = newstate; TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); } /* * Create an external-format (``xtcpcb'') structure using the information in * the kernel-format tcpcb structure pointed to by tp. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) { struct tcpcb *tp = intotcpcb(inp); sbintime_t now; if (inp->inp_flags & INP_TIMEWAIT) { bzero(xt, sizeof(struct xtcpcb)); xt->t_state = TCPS_TIME_WAIT; } else { xt->t_state = tp->t_state; xt->t_logstate = tp->t_logstate; xt->t_flags = tp->t_flags; xt->t_sndzerowin = tp->t_sndzerowin; xt->t_sndrexmitpack = tp->t_sndrexmitpack; xt->t_rcvoopack = tp->t_rcvoopack; now = getsbinuptime(); #define COPYTIMER(ttt) do { \ if (callout_active(&tp->t_timers->ttt)) \ xt->ttt = (tp->t_timers->ttt.c_time - now) / \ SBT_1MS; \ else \ xt->ttt = 0; \ } while (0) COPYTIMER(tt_delack); COPYTIMER(tt_rexmt); COPYTIMER(tt_persist); COPYTIMER(tt_keep); COPYTIMER(tt_2msl); #undef COPYTIMER xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, TCP_FUNCTION_NAME_LEN_MAX); bzero(xt->xt_logid, TCP_LOG_ID_LEN); #ifdef TCP_BLACKBOX (void)tcp_log_get_id(tp, xt->xt_logid); #endif } xt->xt_len = sizeof(struct xtcpcb); in_pcbtoxinpcb(inp, &xt->xt_inp); if (inp->inp_socket == NULL) xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; } Index: head/sys/netinet/tcp_syncache.c =================================================================== --- head/sys/netinet/tcp_syncache.c (revision 336675) +++ head/sys/netinet/tcp_syncache.c (revision 336676) @@ -1,2247 +1,2247 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2001 McAfee, Inc. * Copyright (c) 2006,2013 Andre Oppermann, Internet Business Solutions AG * All rights reserved. * * This software was developed for the FreeBSD Project by Jonathan Lemon * and McAfee Research, the Security Research Division of McAfee, Inc. under * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the * DARPA CHATS research program. [2001 McAfee, Inc.] * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_pcbgroup.h" #include #include #include #include #include #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #endif #include #include #include #include #include #include #include #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include -static VNET_DEFINE(int, tcp_syncookies) = 1; +VNET_DEFINE_STATIC(int, tcp_syncookies) = 1; #define V_tcp_syncookies VNET(tcp_syncookies) SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_syncookies), 0, "Use TCP SYN cookies if the syncache overflows"); -static VNET_DEFINE(int, tcp_syncookiesonly) = 0; +VNET_DEFINE_STATIC(int, tcp_syncookiesonly) = 0; #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly) SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_syncookiesonly), 0, "Use only TCP SYN cookies"); -static VNET_DEFINE(int, functions_inherit_listen_socket_stack) = 1; +VNET_DEFINE_STATIC(int, functions_inherit_listen_socket_stack) = 1; #define V_functions_inherit_listen_socket_stack \ VNET(functions_inherit_listen_socket_stack) SYSCTL_INT(_net_inet_tcp, OID_AUTO, functions_inherit_listen_socket_stack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(functions_inherit_listen_socket_stack), 0, "Inherit listen socket's stack"); #ifdef TCP_OFFLOAD #define ADDED_BY_TOE(sc) ((sc)->sc_tod != NULL) #endif static void syncache_drop(struct syncache *, struct syncache_head *); static void syncache_free(struct syncache *); static void syncache_insert(struct syncache *, struct syncache_head *); static int syncache_respond(struct syncache *, struct syncache_head *, int, const struct mbuf *); static struct socket *syncache_socket(struct syncache *, struct socket *, struct mbuf *m); static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout); static void syncache_timer(void *); static uint32_t syncookie_mac(struct in_conninfo *, tcp_seq, uint8_t, uint8_t *, uintptr_t); static tcp_seq syncookie_generate(struct syncache_head *, struct syncache *); static struct syncache *syncookie_lookup(struct in_conninfo *, struct syncache_head *, struct syncache *, struct tcphdr *, struct tcpopt *, struct socket *); static void syncookie_reseed(void *); #ifdef INVARIANTS static int syncookie_cmp(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso); #endif /* * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. * 3 retransmits corresponds to a timeout of 3 * (1 + 2 + 4 + 8) == 45 seconds, * the odds are that the user has given up attempting to connect by then. */ #define SYNCACHE_MAXREXMTS 3 /* Arbitrary values */ #define TCP_SYNCACHE_HASHSIZE 512 #define TCP_SYNCACHE_BUCKETLIMIT 30 -static VNET_DEFINE(struct tcp_syncache, tcp_syncache); +VNET_DEFINE_STATIC(struct tcp_syncache, tcp_syncache); #define V_tcp_syncache VNET(tcp_syncache) static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.bucket_limit), 0, "Per-bucket hash limit for syncache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.cache_limit), 0, "Overall entry limit for syncache"); SYSCTL_UMA_CUR(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_VNET, &VNET_NAME(tcp_syncache.zone), "Current number of entries in syncache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.hashsize), 0, "Size of TCP syncache hashtable"); static int sysctl_net_inet_tcp_syncache_rexmtlimit_check(SYSCTL_HANDLER_ARGS) { int error; u_int new; new = V_tcp_syncache.rexmt_limit; error = sysctl_handle_int(oidp, &new, 0, req); if ((error == 0) && (req->newptr != NULL)) { if (new > TCP_MAXRXTSHIFT) error = EINVAL; else V_tcp_syncache.rexmt_limit = new; } return (error); } SYSCTL_PROC(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(tcp_syncache.rexmt_limit), 0, sysctl_net_inet_tcp_syncache_rexmtlimit_check, "UI", "Limit on SYN/ACK retransmissions"); VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1; SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0, "Send reset on socket allocation failure"); static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) /* * Requires the syncache entry to be already removed from the bucket list. */ static void syncache_free(struct syncache *sc) { if (sc->sc_ipopts) (void) m_free(sc->sc_ipopts); if (sc->sc_cred) crfree(sc->sc_cred); #ifdef MAC mac_syncache_destroy(&sc->sc_label); #endif uma_zfree(V_tcp_syncache.zone, sc); } void syncache_init(void) { int i; V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; V_tcp_syncache.hash_secret = arc4random(); TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", &V_tcp_syncache.hashsize); TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", &V_tcp_syncache.bucket_limit); if (!powerof2(V_tcp_syncache.hashsize) || V_tcp_syncache.hashsize == 0) { printf("WARNING: syncache hash size is not a power of 2.\n"); V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; } V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1; /* Set limits. */ V_tcp_syncache.cache_limit = V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit; TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", &V_tcp_syncache.cache_limit); /* Allocate the hash table. */ V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize * sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO); #ifdef VIMAGE V_tcp_syncache.vnet = curvnet; #endif /* Initialize the hash buckets. */ for (i = 0; i < V_tcp_syncache.hashsize; i++) { TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket); mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", NULL, MTX_DEF); callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer, &V_tcp_syncache.hashbase[i].sch_mtx, 0); V_tcp_syncache.hashbase[i].sch_length = 0; V_tcp_syncache.hashbase[i].sch_sc = &V_tcp_syncache; V_tcp_syncache.hashbase[i].sch_last_overflow = -(SYNCOOKIE_LIFETIME + 1); } /* Create the syncache entry zone. */ V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_tcp_syncache.cache_limit = uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit); /* Start the SYN cookie reseeder callout. */ callout_init(&V_tcp_syncache.secret.reseed, 1); arc4rand(V_tcp_syncache.secret.key[0], SYNCOOKIE_SECRET_SIZE, 0); arc4rand(V_tcp_syncache.secret.key[1], SYNCOOKIE_SECRET_SIZE, 0); callout_reset(&V_tcp_syncache.secret.reseed, SYNCOOKIE_LIFETIME * hz, syncookie_reseed, &V_tcp_syncache); } #ifdef VIMAGE void syncache_destroy(void) { struct syncache_head *sch; struct syncache *sc, *nsc; int i; /* * Stop the re-seed timer before freeing resources. No need to * possibly schedule it another time. */ callout_drain(&V_tcp_syncache.secret.reseed); /* Cleanup hash buckets: stop timers, free entries, destroy locks. */ for (i = 0; i < V_tcp_syncache.hashsize; i++) { sch = &V_tcp_syncache.hashbase[i]; callout_drain(&sch->sch_timer); SCH_LOCK(sch); TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) syncache_drop(sc, sch); SCH_UNLOCK(sch); KASSERT(TAILQ_EMPTY(&sch->sch_bucket), ("%s: sch->sch_bucket not empty", __func__)); KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0", __func__, sch->sch_length)); mtx_destroy(&sch->sch_mtx); } KASSERT(uma_zone_get_cur(V_tcp_syncache.zone) == 0, ("%s: cache_count not 0", __func__)); /* Free the allocated global resources. */ uma_zdestroy(V_tcp_syncache.zone); free(V_tcp_syncache.hashbase, M_SYNCACHE); } #endif /* * Inserts a syncache entry into the specified bucket row. * Locks and unlocks the syncache_head autonomously. */ static void syncache_insert(struct syncache *sc, struct syncache_head *sch) { struct syncache *sc2; SCH_LOCK(sch); /* * Make sure that we don't overflow the per-bucket limit. * If the bucket is full, toss the oldest element. */ if (sch->sch_length >= V_tcp_syncache.bucket_limit) { KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), ("sch->sch_length incorrect")); sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); sch->sch_last_overflow = time_uptime; syncache_drop(sc2, sch); TCPSTAT_INC(tcps_sc_bucketoverflow); } /* Put it into the bucket. */ TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); sch->sch_length++; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_added(tod, sc->sc_todctx); } #endif /* Reinitialize the bucket row's timer. */ if (sch->sch_length == 1) sch->sch_nextc = ticks + INT_MAX; syncache_timeout(sc, sch, 1); SCH_UNLOCK(sch); TCPSTATES_INC(TCPS_SYN_RECEIVED); TCPSTAT_INC(tcps_sc_added); } /* * Remove and free entry from syncache bucket row. * Expects locked syncache head. */ static void syncache_drop(struct syncache *sc, struct syncache_head *sch) { SCH_LOCK_ASSERT(sch); TCPSTATES_DEC(TCPS_SYN_RECEIVED); TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); sch->sch_length--; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_removed(tod, sc->sc_todctx); } #endif syncache_free(sc); } /* * Engage/reengage time on bucket row. */ static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) { int rexmt; if (sc->sc_rxmits == 0) rexmt = TCPTV_RTOBASE; else TCPT_RANGESET(rexmt, TCPTV_RTOBASE * tcp_syn_backoff[sc->sc_rxmits], tcp_rexmit_min, TCPTV_REXMTMAX); sc->sc_rxttime = ticks + rexmt; sc->sc_rxmits++; if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) { sch->sch_nextc = sc->sc_rxttime; if (docallout) callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, syncache_timer, (void *)sch); } } /* * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. * If we have retransmitted an entry the maximum number of times, expire it. * One separate timer for each bucket row. */ static void syncache_timer(void *xsch) { struct syncache_head *sch = (struct syncache_head *)xsch; struct syncache *sc, *nsc; int tick = ticks; char *s; CURVNET_SET(sch->sch_sc->vnet); /* NB: syncache_head has already been locked by the callout. */ SCH_LOCK_ASSERT(sch); /* * In the following cycle we may remove some entries and/or * advance some timeouts, so re-initialize the bucket timer. */ sch->sch_nextc = tick + INT_MAX; TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { /* * We do not check if the listen socket still exists * and accept the case where the listen socket may be * gone by the time we resend the SYN/ACK. We do * not expect this to happens often. If it does, * then the RST will be sent by the time the remote * host does the SYN/ACK->ACK. */ if (TSTMP_GT(sc->sc_rxttime, tick)) { if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) sch->sch_nextc = sc->sc_rxttime; continue; } if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) { if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " "giving up and removing syncache entry\n", s, __func__); free(s, M_TCPLOG); } syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_stale); continue; } if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Response timeout, " "retransmitting (%u) SYN|ACK\n", s, __func__, sc->sc_rxmits); free(s, M_TCPLOG); } syncache_respond(sc, sch, 1, NULL); TCPSTAT_INC(tcps_sc_retransmitted); syncache_timeout(sc, sch, 0); } if (!TAILQ_EMPTY(&(sch)->sch_bucket)) callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, syncache_timer, (void *)(sch)); CURVNET_RESTORE(); } /* * Find an entry in the syncache. * Returns always with locked syncache_head plus a matching entry or NULL. */ static struct syncache * syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) { struct syncache *sc; struct syncache_head *sch; uint32_t hash; /* * The hash is built on foreign port + local port + foreign address. * We rely on the fact that struct in_conninfo starts with 16 bits * of foreign port, then 16 bits of local port then followed by 128 * bits of foreign address. In case of IPv4 address, the first 3 * 32-bit words of the address always are zeroes. */ hash = jenkins_hash32((uint32_t *)&inc->inc_ie, 5, V_tcp_syncache.hash_secret) & V_tcp_syncache.hashmask; sch = &V_tcp_syncache.hashbase[hash]; *schp = sch; SCH_LOCK(sch); /* Circle through bucket row to find matching entry. */ TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) if (bcmp(&inc->inc_ie, &sc->sc_inc.inc_ie, sizeof(struct in_endpoints)) == 0) break; return (sc); /* Always returns with locked sch. */ } /* * This function is called when we get a RST for a * non-existent connection, so that we can see if the * connection is in the syn cache. If it is, zap it. */ void syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th) { struct syncache *sc; struct syncache_head *sch; char *s = NULL; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); /* * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. * See RFC 793 page 65, section SEGMENT ARRIVES. */ if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " "FIN flag set, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badrst); goto done; } /* * No corresponding connection was found in syncache. * If syncookies are enabled and possibly exclusively * used, or we are under memory pressure, a valid RST * may not find a syncache entry. In that case we're * done and no SYN|ACK retransmissions will happen. * Otherwise the RST was misdirected or spoofed. */ if (sc == NULL) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious RST without matching " "syncache entry (possibly syncookie only), " "segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badrst); goto done; } /* * If the RST bit is set, check the sequence number to see * if this is a valid reset segment. * RFC 793 page 37: * In all states except SYN-SENT, all reset (RST) segments * are validated by checking their SEQ-fields. A reset is * valid if its sequence number is in the window. * * The sequence number in the reset segment is normally an * echo of our outgoing acknowlegement numbers, but some hosts * send a reset with the sequence number at the rightmost edge * of our receive window, and we have to handle this case. */ if (SEQ_GEQ(th->th_seq, sc->sc_irs) && SEQ_LEQ(th->th_seq, sc->sc_irs + sc->sc_wnd)) { syncache_drop(sc, sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " "connection attempt aborted by remote endpoint\n", s, __func__); TCPSTAT_INC(tcps_sc_reset); } else { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != " "IRS %u (+WND %u), segment ignored\n", s, __func__, th->th_seq, sc->sc_irs, sc->sc_wnd); TCPSTAT_INC(tcps_badrst); } done: if (s != NULL) free(s, M_TCPLOG); SCH_UNLOCK(sch); } void syncache_badack(struct in_conninfo *inc) { struct syncache *sc; struct syncache_head *sch; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); if (sc != NULL) { syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_badack); } SCH_UNLOCK(sch); } void syncache_unreach(struct in_conninfo *inc, tcp_seq th_seq) { struct syncache *sc; struct syncache_head *sch; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); if (sc == NULL) goto done; /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ if (ntohl(th_seq) != sc->sc_iss) goto done; /* * If we've rertransmitted 3 times and this is our second error, * we remove the entry. Otherwise, we allow it to continue on. * This prevents us from incorrectly nuking an entry during a * spurious network outage. * * See tcp_notify(). */ if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { sc->sc_flags |= SCF_UNREACH; goto done; } syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_unreach); done: SCH_UNLOCK(sch); } /* * Build a new TCP socket structure from a syncache entry. * * On success return the newly created socket with its underlying inp locked. */ static struct socket * syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) { struct tcp_function_block *blk; struct inpcb *inp = NULL; struct socket *so; struct tcpcb *tp; int error; char *s; INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* * Ok, create the full blown connection, and set things up * as they would have been set up if we had created the * connection when the SYN arrived. If we can't create * the connection, abort it. */ so = sonewconn(lso, 0); if (so == NULL) { /* * Drop the connection; we will either send a RST or * have the peer retransmit its SYN again after its * RTO and try again. */ TCPSTAT_INC(tcps_listendrop); if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Socket create failed " "due to limits or memory shortage\n", s, __func__); free(s, M_TCPLOG); } goto abort2; } #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif inp = sotoinpcb(so); inp->inp_inc.inc_fibnum = so->so_fibnum; INP_WLOCK(inp); /* * Exclusive pcbinfo lock is not required in syncache socket case even * if two inpcb locks can be acquired simultaneously: * - the inpcb in LISTEN state, * - the newly created inp. * * In this case, an inp cannot be at same time in LISTEN state and * just created by an accept() call. */ INP_HASH_WLOCK(&V_tcbinfo); /* Insert new socket into PCB hash list. */ inp->inp_inc.inc_flags = sc->sc_inc.inc_flags; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { inp->inp_vflag &= ~INP_IPV4; inp->inp_vflag |= INP_IPV6; inp->in6p_laddr = sc->sc_inc.inc6_laddr; } else { inp->inp_vflag &= ~INP_IPV6; inp->inp_vflag |= INP_IPV4; #endif inp->inp_laddr = sc->sc_inc.inc_laddr; #ifdef INET6 } #endif /* * If there's an mbuf and it has a flowid, then let's initialise the * inp with that particular flowid. */ if (m != NULL && M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); } /* * Install in the reservation hash table for now, but don't yet * install a connection group since the full 4-tuple isn't yet * configured. */ inp->inp_lport = sc->sc_inc.inc_lport; if ((error = in_pcbinshash_nopcbgroup(inp)) != 0) { /* * Undo the assignments above if we failed to * put the PCB on the hash lists. */ #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) inp->in6p_laddr = in6addr_any; else #endif inp->inp_laddr.s_addr = INADDR_ANY; inp->inp_lport = 0; if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: in_pcbinshash failed " "with error %i\n", s, __func__, error); free(s, M_TCPLOG); } INP_HASH_WUNLOCK(&V_tcbinfo); goto abort; } #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { struct inpcb *oinp = sotoinpcb(lso); struct in6_addr laddr6; struct sockaddr_in6 sin6; /* * Inherit socket options from the listening socket. * Note that in6p_inputopts are not (and should not be) * copied, since it stores previously received options and is * used to detect if each new option is different than the * previous one and hence should be passed to a user. * If we copied in6p_inputopts, a user would not be able to * receive options just after calling the accept system call. */ inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; if (oinp->in6p_outputopts) inp->in6p_outputopts = ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(sin6); sin6.sin6_addr = sc->sc_inc.inc6_faddr; sin6.sin6_port = sc->sc_inc.inc_fport; sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; laddr6 = inp->in6p_laddr; if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) inp->in6p_laddr = sc->sc_inc.inc6_laddr; if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6, thread0.td_ucred, m)) != 0) { inp->in6p_laddr = laddr6; if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed " "with error %i\n", s, __func__, error); free(s, M_TCPLOG); } INP_HASH_WUNLOCK(&V_tcbinfo); goto abort; } /* Override flowlabel from in6_pcbconnect. */ inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; inp->inp_flow |= sc->sc_flowlabel; } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { struct in_addr laddr; struct sockaddr_in sin; inp->inp_options = (m) ? ip_srcroute(m) : NULL; if (inp->inp_options == NULL) { inp->inp_options = sc->sc_ipopts; sc->sc_ipopts = NULL; } sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); sin.sin_addr = sc->sc_inc.inc_faddr; sin.sin_port = sc->sc_inc.inc_fport; bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); laddr = inp->inp_laddr; if (inp->inp_laddr.s_addr == INADDR_ANY) inp->inp_laddr = sc->sc_inc.inc_laddr; if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin, thread0.td_ucred, m)) != 0) { inp->inp_laddr = laddr; if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: in_pcbconnect failed " "with error %i\n", s, __func__, error); free(s, M_TCPLOG); } INP_HASH_WUNLOCK(&V_tcbinfo); goto abort; } } #endif /* INET */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* Copy old policy into new socket's. */ if (ipsec_copy_pcbpolicy(sotoinpcb(lso), inp) != 0) printf("syncache_socket: could not copy policy\n"); #endif INP_HASH_WUNLOCK(&V_tcbinfo); tp = intotcpcb(inp); tcp_state_change(tp, TCPS_SYN_RECEIVED); tp->iss = sc->sc_iss; tp->irs = sc->sc_irs; tcp_rcvseqinit(tp); tcp_sendseqinit(tp); blk = sototcpcb(lso)->t_fb; if (V_functions_inherit_listen_socket_stack && blk != tp->t_fb) { /* * Our parents t_fb was not the default, * we need to release our ref on tp->t_fb and * pickup one on the new entry. */ struct tcp_function_block *rblk; rblk = find_and_ref_tcp_fb(blk); KASSERT(rblk != NULL, ("cannot find blk %p out of syncache?", blk)); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_fb = rblk; /* * XXXrrs this is quite dangerous, it is possible * for the new function to fail to init. We also * are not asking if the handoff_is_ok though at * the very start thats probalbly ok. */ if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } } tp->snd_wl1 = sc->sc_irs; tp->snd_max = tp->iss + 1; tp->snd_nxt = tp->iss + 1; tp->rcv_up = sc->sc_irs + 1; tp->rcv_wnd = sc->sc_wnd; tp->rcv_adv += tp->rcv_wnd; tp->last_ack_sent = tp->rcv_nxt; tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); if (sc->sc_flags & SCF_NOOPT) tp->t_flags |= TF_NOOPT; else { if (sc->sc_flags & SCF_WINSCALE) { tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; tp->snd_scale = sc->sc_requested_s_scale; tp->request_r_scale = sc->sc_requested_r_scale; } if (sc->sc_flags & SCF_TIMESTAMP) { tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; tp->ts_recent = sc->sc_tsreflect; tp->ts_recent_age = tcp_ts_getticks(); tp->ts_offset = sc->sc_tsoff; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) tp->t_flags |= TF_SIGNATURE; #endif if (sc->sc_flags & SCF_SACK) tp->t_flags |= TF_SACK_PERMIT; } if (sc->sc_flags & SCF_ECN) tp->t_flags |= TF_ECN_PERMIT; /* * Set up MSS and get cached values from tcp_hostcache. * This might overwrite some of the defaults we just set. */ tcp_mss(tp, sc->sc_peer_mss); /* * If the SYN,ACK was retransmitted, indicate that CWND to be * limited to one segment in cc_conn_init(). * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits. */ if (sc->sc_rxmits > 1) tp->snd_cwnd = 1; #ifdef TCP_OFFLOAD /* * Allow a TOE driver to install its hooks. Note that we hold the * pcbinfo lock too and that prevents tcp_usr_accept from accepting a * new connection before the TOE driver has done its thing. */ if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_offload_socket(tod, sc->sc_todctx, so); } #endif /* * Copy and activate timers. */ tp->t_keepinit = sototcpcb(lso)->t_keepinit; tp->t_keepidle = sototcpcb(lso)->t_keepidle; tp->t_keepintvl = sototcpcb(lso)->t_keepintvl; tp->t_keepcnt = sototcpcb(lso)->t_keepcnt; tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); TCPSTAT_INC(tcps_accepts); return (so); abort: INP_WUNLOCK(inp); abort2: if (so != NULL) soabort(so); return (NULL); } /* * This function gets called when we receive an ACK for a * socket in the LISTEN state. We look up the connection * in the syncache, and if its there, we pull it out of * the cache and turn it into a full-blown connection in * the SYN-RECEIVED state. * * On syncache_socket() success the newly created socket * has its underlying inp locked. */ int syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, struct socket **lsop, struct mbuf *m) { struct syncache *sc; struct syncache_head *sch; struct syncache scs; char *s; /* * Global TCP locks are held because we manipulate the PCB lists * and create a new socket. */ INP_INFO_RLOCK_ASSERT(&V_tcbinfo); KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, ("%s: can handle only ACK", __func__)); sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); #ifdef INVARIANTS /* * Test code for syncookies comparing the syncache stored * values with the reconstructed values from the cookie. */ if (sc != NULL) syncookie_cmp(inc, sch, sc, th, to, *lsop); #endif if (sc == NULL) { /* * There is no syncache entry, so see if this ACK is * a returning syncookie. To do this, first: * A. Check if syncookies are used in case of syncache * overflows * B. See if this socket has had a syncache entry dropped in * the recent past. We don't want to accept a bogus * syncookie if we've never received a SYN or accept it * twice. * C. check that the syncookie is valid. If it is, then * cobble up a fake syncache entry, and return. */ if (!V_tcp_syncookies) { SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious ACK, " "segment rejected (syncookies disabled)\n", s, __func__); goto failed; } if (!V_tcp_syncookiesonly && sch->sch_last_overflow < time_uptime - SYNCOOKIE_LIFETIME) { SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious ACK, " "segment rejected (no syncache entry)\n", s, __func__); goto failed; } bzero(&scs, sizeof(scs)); sc = syncookie_lookup(inc, sch, &scs, th, to, *lsop); SCH_UNLOCK(sch); if (sc == NULL) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Segment failed " "SYNCOOKIE authentication, segment rejected " "(probably spoofed)\n", s, __func__); goto failed; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* If received ACK has MD5 signature, check it. */ if ((to->to_flags & TOF_SIGNATURE) != 0 && (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0)) { /* Drop the ACK. */ if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment rejected, " "MD5 signature doesn't match.\n", s, __func__); free(s, M_TCPLOG); } TCPSTAT_INC(tcps_sig_err_sigopt); return (-1); /* Do not send RST */ } #endif /* TCP_SIGNATURE */ } else { #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, check that * received ACK has signature and it is correct. * If not, drop the ACK and leave sc entry in th cache, * because SYN was received with correct signature. */ if (sc->sc_flags & SCF_SIGNATURE) { if ((to->to_flags & TOF_SIGNATURE) == 0) { /* No signature */ TCPSTAT_INC(tcps_sig_err_nosigopt); SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment " "rejected, MD5 signature wasn't " "provided.\n", s, __func__); free(s, M_TCPLOG); } return (-1); /* Do not send RST */ } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0) { /* Doesn't match or no SA */ SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment " "rejected, MD5 signature doesn't " "match.\n", s, __func__); free(s, M_TCPLOG); } return (-1); /* Do not send RST */ } } #endif /* TCP_SIGNATURE */ /* * Pull out the entry to unlock the bucket row. * * NOTE: We must decrease TCPS_SYN_RECEIVED count here, not * tcp_state_change(). The tcpcb is not existent at this * moment. A new one will be allocated via syncache_socket-> * sonewconn->tcp_usr_attach in TCPS_CLOSED state, then * syncache_socket() will change it to TCPS_SYN_RECEIVED. */ TCPSTATES_DEC(TCPS_SYN_RECEIVED); TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); sch->sch_length--; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_removed(tod, sc->sc_todctx); } #endif SCH_UNLOCK(sch); } /* * Segment validation: * ACK must match our initial sequence number + 1 (the SYN|ACK). */ if (th->th_ack != sc->sc_iss + 1) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " "rejected\n", s, __func__, th->th_ack, sc->sc_iss); goto failed; } /* * The SEQ must fall in the window starting at the received * initial receive sequence number + 1 (the SYN). */ if (SEQ_LEQ(th->th_seq, sc->sc_irs) || SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " "rejected\n", s, __func__, th->th_seq, sc->sc_irs); goto failed; } /* * If timestamps were not negotiated during SYN/ACK they * must not appear on any segment during this session. */ if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "segment rejected\n", s, __func__); goto failed; } /* * If timestamps were negotiated during SYN/ACK they should * appear on every segment during this session. * XXXAO: This is only informal as there have been unverified * reports of non-compliants stacks. */ if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "no action\n", s, __func__); free(s, M_TCPLOG); s = NULL; } } *lsop = syncache_socket(sc, *lsop, m); if (*lsop == NULL) TCPSTAT_INC(tcps_sc_aborted); else TCPSTAT_INC(tcps_sc_completed); /* how do we find the inp for the new socket? */ if (sc != &scs) syncache_free(sc); return (1); failed: if (sc != NULL && sc != &scs) syncache_free(sc); if (s != NULL) free(s, M_TCPLOG); *lsop = NULL; return (0); } static void syncache_tfo_expand(struct syncache *sc, struct socket **lsop, struct mbuf *m, uint64_t response_cookie) { struct inpcb *inp; struct tcpcb *tp; unsigned int *pending_counter; /* * Global TCP locks are held because we manipulate the PCB lists * and create a new socket. */ INP_INFO_RLOCK_ASSERT(&V_tcbinfo); pending_counter = intotcpcb(sotoinpcb(*lsop))->t_tfo_pending; *lsop = syncache_socket(sc, *lsop, m); if (*lsop == NULL) { TCPSTAT_INC(tcps_sc_aborted); atomic_subtract_int(pending_counter, 1); } else { soisconnected(*lsop); inp = sotoinpcb(*lsop); tp = intotcpcb(inp); tp->t_flags |= TF_FASTOPEN; tp->t_tfo_cookie.server = response_cookie; tp->snd_max = tp->iss; tp->snd_nxt = tp->iss; tp->t_tfo_pending = pending_counter; TCPSTAT_INC(tcps_sc_completed); } } /* * Given a LISTEN socket and an inbound SYN request, add * this to the syn cache, and send back a segment: * * to the source. * * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. * Doing so would require that we hold onto the data and deliver it * to the application. However, if we are the target of a SYN-flood * DoS attack, an attacker could send data which would eventually * consume all available buffer space if it were ACKed. By not ACKing * the data, we avoid this DoS scenario. * * The exception to the above is when a SYN with a valid TCP Fast Open (TFO) * cookie is processed and a new socket is created. In this case, any data * accompanying the SYN will be queued to the socket by tcp_input() and will * be ACKed either when the application sends response data or the delayed * ACK timer expires, whichever comes first. */ int syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, struct inpcb *inp, struct socket **lsop, struct mbuf *m, void *tod, void *todctx) { struct tcpcb *tp; struct socket *so; struct syncache *sc = NULL; struct syncache_head *sch; struct mbuf *ipopts = NULL; u_int ltflags; int win, ip_ttl, ip_tos; char *s; int rv = 0; #ifdef INET6 int autoflowlabel = 0; #endif #ifdef MAC struct label *maclabel; #endif struct syncache scs; struct ucred *cred; uint64_t tfo_response_cookie; unsigned int *tfo_pending = NULL; int tfo_cookie_valid = 0; int tfo_response_cookie_valid = 0; INP_WLOCK_ASSERT(inp); /* listen socket */ KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, ("%s: unexpected tcp flags", __func__)); /* * Combine all so/tp operations very early to drop the INP lock as * soon as possible. */ so = *lsop; KASSERT(SOLISTENING(so), ("%s: %p not listening", __func__, so)); tp = sototcpcb(so); cred = crhold(so->so_cred); #ifdef INET6 if ((inc->inc_flags & INC_ISIPV6) && (inp->inp_flags & IN6P_AUTOFLOWLABEL)) autoflowlabel = 1; #endif ip_ttl = inp->inp_ip_ttl; ip_tos = inp->inp_ip_tos; win = so->sol_sbrcv_hiwat; ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE)); if (V_tcp_fastopen_server_enable && IS_FASTOPEN(tp->t_flags) && (tp->t_tfo_pending != NULL) && (to->to_flags & TOF_FASTOPEN)) { /* * Limit the number of pending TFO connections to * approximately half of the queue limit. This prevents TFO * SYN floods from starving the service by filling the * listen queue with bogus TFO connections. */ if (atomic_fetchadd_int(tp->t_tfo_pending, 1) <= (so->sol_qlimit / 2)) { int result; result = tcp_fastopen_check_cookie(inc, to->to_tfo_cookie, to->to_tfo_len, &tfo_response_cookie); tfo_cookie_valid = (result > 0); tfo_response_cookie_valid = (result >= 0); } /* * Remember the TFO pending counter as it will have to be * decremented below if we don't make it to syncache_tfo_expand(). */ tfo_pending = tp->t_tfo_pending; } /* By the time we drop the lock these should no longer be used. */ so = NULL; tp = NULL; #ifdef MAC if (mac_syncache_init(&maclabel) != 0) { INP_WUNLOCK(inp); goto done; } else mac_syncache_create(maclabel, inp); #endif if (!tfo_cookie_valid) INP_WUNLOCK(inp); /* * Remember the IP options, if any. */ #ifdef INET6 if (!(inc->inc_flags & INC_ISIPV6)) #endif #ifdef INET ipopts = (m) ? ip_srcroute(m) : NULL; #else ipopts = NULL; #endif #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, check that received * SYN has signature and it is correct. If signature doesn't match * or TCP_SIGNATURE support isn't enabled, drop the packet. */ if (ltflags & TF_SIGNATURE) { if ((to->to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); goto done; } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0) goto done; } #endif /* TCP_SIGNATURE */ /* * See if we already have an entry for this connection. * If we do, resend the SYN,ACK, and reset the retransmit timer. * * XXX: should the syncache be re-initialized with the contents * of the new SYN here (which may have different options?) * * XXX: We do not check the sequence number to see if this is a * real retransmit or a new connection attempt. The question is * how to handle such a case; either ignore it as spoofed, or * drop the current entry and create a new one? */ sc = syncache_lookup(inc, &sch); /* returns locked entry */ SCH_LOCK_ASSERT(sch); if (sc != NULL) { if (tfo_cookie_valid) INP_WUNLOCK(inp); TCPSTAT_INC(tcps_sc_dupsyn); if (ipopts) { /* * If we were remembering a previous source route, * forget it and use the new one we've been given. */ if (sc->sc_ipopts) (void) m_free(sc->sc_ipopts); sc->sc_ipopts = ipopts; } /* * Update timestamp if present. */ if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) sc->sc_tsreflect = to->to_tsval; else sc->sc_flags &= ~SCF_TIMESTAMP; #ifdef MAC /* * Since we have already unconditionally allocated label * storage, free it up. The syncache entry will already * have an initialized label we can use. */ mac_syncache_destroy(&maclabel); #endif /* Retransmit SYN|ACK and reset retransmit count. */ if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " "resetting timer and retransmitting SYN|ACK\n", s, __func__); free(s, M_TCPLOG); } if (syncache_respond(sc, sch, 1, m) == 0) { sc->sc_rxmits = 0; syncache_timeout(sc, sch, 1); TCPSTAT_INC(tcps_sndacks); TCPSTAT_INC(tcps_sndtotal); } SCH_UNLOCK(sch); goto done; } if (tfo_cookie_valid) { bzero(&scs, sizeof(scs)); sc = &scs; goto skip_alloc; } sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); if (sc == NULL) { /* * The zone allocator couldn't provide more entries. * Treat this as if the cache was full; drop the oldest * entry and insert the new one. */ TCPSTAT_INC(tcps_sc_zonefail); if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) { sch->sch_last_overflow = time_uptime; syncache_drop(sc, sch); } sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); if (sc == NULL) { if (V_tcp_syncookies) { bzero(&scs, sizeof(scs)); sc = &scs; } else { SCH_UNLOCK(sch); if (ipopts) (void) m_free(ipopts); goto done; } } } skip_alloc: if (!tfo_cookie_valid && tfo_response_cookie_valid) sc->sc_tfo_cookie = &tfo_response_cookie; /* * Fill in the syncache values. */ #ifdef MAC sc->sc_label = maclabel; #endif sc->sc_cred = cred; cred = NULL; sc->sc_ipopts = ipopts; bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); #ifdef INET6 if (!(inc->inc_flags & INC_ISIPV6)) #endif { sc->sc_ip_tos = ip_tos; sc->sc_ip_ttl = ip_ttl; } #ifdef TCP_OFFLOAD sc->sc_tod = tod; sc->sc_todctx = todctx; #endif sc->sc_irs = th->th_seq; sc->sc_iss = arc4random(); sc->sc_flags = 0; sc->sc_flowlabel = 0; /* * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. * win was derived from socket earlier in the function. */ win = imax(win, 0); win = imin(win, TCP_MAXWIN); sc->sc_wnd = win; if (V_tcp_do_rfc1323) { /* * A timestamp received in a SYN makes * it ok to send timestamp requests and replies. */ if (to->to_flags & TOF_TS) { sc->sc_tsreflect = to->to_tsval; sc->sc_flags |= SCF_TIMESTAMP; } if (to->to_flags & TOF_SCALE) { int wscale = 0; /* * Pick the smallest possible scaling factor that * will still allow us to scale up to sb_max, aka * kern.ipc.maxsockbuf. * * We do this because there are broken firewalls that * will corrupt the window scale option, leading to * the other endpoint believing that our advertised * window is unscaled. At scale factors larger than * 5 the unscaled window will drop below 1500 bytes, * leading to serious problems when traversing these * broken firewalls. * * With the default maxsockbuf of 256K, a scale factor * of 3 will be chosen by this algorithm. Those who * choose a larger maxsockbuf should watch out * for the compatibility problems mentioned above. * * RFC1323: The Window field in a SYN (i.e., a * or ) segment itself is never scaled. */ while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < sb_max) wscale++; sc->sc_requested_r_scale = wscale; sc->sc_requested_s_scale = to->to_wscale; sc->sc_flags |= SCF_WINSCALE; } } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, flag this in the * syncache so that syncache_respond() will do the right thing * with the SYN+ACK. */ if (ltflags & TF_SIGNATURE) sc->sc_flags |= SCF_SIGNATURE; #endif /* TCP_SIGNATURE */ if (to->to_flags & TOF_SACKPERM) sc->sc_flags |= SCF_SACK; if (to->to_flags & TOF_MSS) sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ if (ltflags & TF_NOOPT) sc->sc_flags |= SCF_NOOPT; if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn) sc->sc_flags |= SCF_ECN; if (V_tcp_syncookies) sc->sc_iss = syncookie_generate(sch, sc); #ifdef INET6 if (autoflowlabel) { if (V_tcp_syncookies) sc->sc_flowlabel = sc->sc_iss; else sc->sc_flowlabel = ip6_randomflowlabel(); sc->sc_flowlabel = htonl(sc->sc_flowlabel) & IPV6_FLOWLABEL_MASK; } #endif SCH_UNLOCK(sch); if (tfo_cookie_valid) { syncache_tfo_expand(sc, lsop, m, tfo_response_cookie); /* INP_WUNLOCK(inp) will be performed by the caller */ rv = 1; goto tfo_expanded; } /* * Do a standard 3-way handshake. */ if (syncache_respond(sc, sch, 0, m) == 0) { if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs) syncache_free(sc); else if (sc != &scs) syncache_insert(sc, sch); /* locks and unlocks sch */ TCPSTAT_INC(tcps_sndacks); TCPSTAT_INC(tcps_sndtotal); } else { if (sc != &scs) syncache_free(sc); TCPSTAT_INC(tcps_sc_dropped); } done: if (m) { *lsop = NULL; m_freem(m); } /* * If tfo_pending is not NULL here, then a TFO SYN that did not * result in a new socket was processed and the associated pending * counter has not yet been decremented. All such TFO processing paths * transit this point. */ if (tfo_pending != NULL) tcp_fastopen_decrement_counter(tfo_pending); tfo_expanded: if (cred != NULL) crfree(cred); #ifdef MAC if (sc == &scs) mac_syncache_destroy(&maclabel); #endif return (rv); } /* * Send SYN|ACK to the peer. Either in response to the peer's SYN, * i.e. m0 != NULL, or upon 3WHS ACK timeout, i.e. m0 == NULL. */ static int syncache_respond(struct syncache *sc, struct syncache_head *sch, int locked, const struct mbuf *m0) { struct ip *ip = NULL; struct mbuf *m; struct tcphdr *th = NULL; int optlen, error = 0; /* Make compiler happy */ u_int16_t hlen, tlen, mssopt; struct tcpopt to; #ifdef INET6 struct ip6_hdr *ip6 = NULL; #endif hlen = #ifdef INET6 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) : #endif sizeof(struct ip); tlen = hlen + sizeof(struct tcphdr); /* Determine MSS we advertize to other end of connection. */ mssopt = max(tcp_mssopt(&sc->sc_inc), V_tcp_minmss); /* XXX: Assume that the entire packet will fit in a header mbuf. */ KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, ("syncache: mbuf too small")); /* Create the IP+TCP header from scratch. */ m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); #ifdef MAC mac_syncache_create_mbuf(sc->sc_label, m); #endif m->m_data += max_linkhdr; m->m_len = tlen; m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_src = sc->sc_inc.inc6_laddr; ip6->ip6_dst = sc->sc_inc.inc6_faddr; ip6->ip6_plen = htons(tlen - hlen); /* ip6_hlim is set after checksum */ ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; ip6->ip6_flow |= sc->sc_flowlabel; th = (struct tcphdr *)(ip6 + 1); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = sizeof(struct ip) >> 2; ip->ip_len = htons(tlen); ip->ip_id = 0; ip->ip_off = 0; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = sc->sc_inc.inc_laddr; ip->ip_dst = sc->sc_inc.inc_faddr; ip->ip_ttl = sc->sc_ip_ttl; ip->ip_tos = sc->sc_ip_tos; /* * See if we should do MTU discovery. Route lookups are * expensive, so we will only unset the DF bit if: * * 1) path_mtu_discovery is disabled * 2) the SCF_UNREACH flag has been set */ if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) ip->ip_off |= htons(IP_DF); th = (struct tcphdr *)(ip + 1); } #endif /* INET */ th->th_sport = sc->sc_inc.inc_lport; th->th_dport = sc->sc_inc.inc_fport; th->th_seq = htonl(sc->sc_iss); th->th_ack = htonl(sc->sc_irs + 1); th->th_off = sizeof(struct tcphdr) >> 2; th->th_x2 = 0; th->th_flags = TH_SYN|TH_ACK; th->th_win = htons(sc->sc_wnd); th->th_urp = 0; if (sc->sc_flags & SCF_ECN) { th->th_flags |= TH_ECE; TCPSTAT_INC(tcps_ecn_shs); } /* Tack on the TCP options. */ if ((sc->sc_flags & SCF_NOOPT) == 0) { to.to_flags = 0; to.to_mss = mssopt; to.to_flags = TOF_MSS; if (sc->sc_flags & SCF_WINSCALE) { to.to_wscale = sc->sc_requested_r_scale; to.to_flags |= TOF_SCALE; } if (sc->sc_flags & SCF_TIMESTAMP) { to.to_tsval = sc->sc_tsoff + tcp_ts_getticks(); to.to_tsecr = sc->sc_tsreflect; to.to_flags |= TOF_TS; } if (sc->sc_flags & SCF_SACK) to.to_flags |= TOF_SACKPERM; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif if (sc->sc_tfo_cookie) { to.to_flags |= TOF_FASTOPEN; to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; to.to_tfo_cookie = sc->sc_tfo_cookie; /* don't send cookie again when retransmitting response */ sc->sc_tfo_cookie = NULL; } optlen = tcp_addoptions(&to, (u_char *)(th + 1)); /* Adjust headers by option size. */ th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; m->m_len += optlen; m->m_pkthdr.len += optlen; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); else #endif ip->ip_len = htons(ntohs(ip->ip_len) + optlen); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) { KASSERT(to.to_flags & TOF_SIGNATURE, ("tcp_addoptions() didn't set tcp_signature")); /* NOTE: to.to_signature is inside of mbuf */ if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, to.to_signature) != 0) { m_freem(m); return (EACCES); } } #endif } else optlen = 0; M_SETFIB(m, sc->sc_inc.inc_fibnum); m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); /* * If we have peer's SYN and it has a flowid, then let's assign it to * our SYN|ACK. ip6_output() and ip_output() will not assign flowid * to SYN|ACK due to lack of inp here. */ if (m0 != NULL && M_HASHTYPE_GET(m0) != M_HASHTYPE_NONE) { m->m_pkthdr.flowid = m0->m_pkthdr.flowid; M_HASHTYPE_SET(m, M_HASHTYPE_GET(m0)); } #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; th->th_sum = in6_cksum_pseudo(ip6, tlen + optlen - hlen, IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(NULL, NULL); #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; error = tod->tod_syncache_respond(tod, sc->sc_todctx, m); return (error); } #endif error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(tlen + optlen - hlen + IPPROTO_TCP)); #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; error = tod->tod_syncache_respond(tod, sc->sc_todctx, m); return (error); } #endif error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); } #endif return (error); } /* * The purpose of syncookies is to handle spoofed SYN flooding DoS attacks * that exceed the capacity of the syncache by avoiding the storage of any * of the SYNs we receive. Syncookies defend against blind SYN flooding * attacks where the attacker does not have access to our responses. * * Syncookies encode and include all necessary information about the * connection setup within the SYN|ACK that we send back. That way we * can avoid keeping any local state until the ACK to our SYN|ACK returns * (if ever). Normally the syncache and syncookies are running in parallel * with the latter taking over when the former is exhausted. When matching * syncache entry is found the syncookie is ignored. * * The only reliable information persisting the 3WHS is our initial sequence * number ISS of 32 bits. Syncookies embed a cryptographically sufficient * strong hash (MAC) value and a few bits of TCP SYN options in the ISS * of our SYN|ACK. The MAC can be recomputed when the ACK to our SYN|ACK * returns and signifies a legitimate connection if it matches the ACK. * * The available space of 32 bits to store the hash and to encode the SYN * option information is very tight and we should have at least 24 bits for * the MAC to keep the number of guesses by blind spoofing reasonably high. * * SYN option information we have to encode to fully restore a connection: * MSS: is imporant to chose an optimal segment size to avoid IP level * fragmentation along the path. The common MSS values can be encoded * in a 3-bit table. Uncommon values are captured by the next lower value * in the table leading to a slight increase in packetization overhead. * WSCALE: is necessary to allow large windows to be used for high delay- * bandwidth product links. Not scaling the window when it was initially * negotiated is bad for performance as lack of scaling further decreases * the apparent available send window. We only need to encode the WSCALE * we received from the remote end. Our end can be recalculated at any * time. The common WSCALE values can be encoded in a 3-bit table. * Uncommon values are captured by the next lower value in the table * making us under-estimate the available window size halving our * theoretically possible maximum throughput for that connection. * SACK: Greatly assists in packet loss recovery and requires 1 bit. * TIMESTAMP and SIGNATURE is not encoded because they are permanent options * that are included in all segments on a connection. We enable them when * the ACK has them. * * Security of syncookies and attack vectors: * * The MAC is computed over (faddr||laddr||fport||lport||irs||flags||secmod) * together with the gloabl secret to make it unique per connection attempt. * Thus any change of any of those parameters results in a different MAC output * in an unpredictable way unless a collision is encountered. 24 bits of the * MAC are embedded into the ISS. * * To prevent replay attacks two rotating global secrets are updated with a * new random value every 15 seconds. The life-time of a syncookie is thus * 15-30 seconds. * * Vector 1: Attacking the secret. This requires finding a weakness in the * MAC itself or the way it is used here. The attacker can do a chosen plain * text attack by varying and testing the all parameters under his control. * The strength depends on the size and randomness of the secret, and the * cryptographic security of the MAC function. Due to the constant updating * of the secret the attacker has at most 29.999 seconds to find the secret * and launch spoofed connections. After that he has to start all over again. * * Vector 2: Collision attack on the MAC of a single ACK. With a 24 bit MAC * size an average of 4,823 attempts are required for a 50% chance of success * to spoof a single syncookie (birthday collision paradox). However the * attacker is blind and doesn't know if one of his attempts succeeded unless * he has a side channel to interfere success from. A single connection setup * success average of 90% requires 8,790 packets, 99.99% requires 17,578 packets. * This many attempts are required for each one blind spoofed connection. For * every additional spoofed connection he has to launch another N attempts. * Thus for a sustained rate 100 spoofed connections per second approximately * 1,800,000 packets per second would have to be sent. * * NB: The MAC function should be fast so that it doesn't become a CPU * exhaustion attack vector itself. * * References: * RFC4987 TCP SYN Flooding Attacks and Common Mitigations * SYN cookies were first proposed by cryptographer Dan J. Bernstein in 1996 * http://cr.yp.to/syncookies.html (overview) * http://cr.yp.to/syncookies/archive (details) * * * Schematic construction of a syncookie enabled Initial Sequence Number: * 0 1 2 3 * 12345678901234567890123456789012 * |xxxxxxxxxxxxxxxxxxxxxxxxWWWMMMSP| * * x 24 MAC (truncated) * W 3 Send Window Scale index * M 3 MSS index * S 1 SACK permitted * P 1 Odd/even secret */ /* * Distribution and probability of certain MSS values. Those in between are * rounded down to the next lower one. * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011] * .2% .3% 5% 7% 7% 20% 15% 45% */ static int tcp_sc_msstab[] = { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 }; /* * Distribution and probability of certain WSCALE values. We have to map the * (send) window scale (shift) option with a range of 0-14 from 4 bits into 3 * bits based on prevalence of certain values. Where we don't have an exact * match for are rounded down to the next lower one letting us under-estimate * the true available window. At the moment this would happen only for the * very uncommon values 3, 5 and those above 8 (more than 16MB socket buffer * and window size). The absence of the WSCALE option (no scaling in either * direction) is encoded with index zero. * [WSCALE values histograms, Allman, 2012] * X 10 10 35 5 6 14 10% by host * X 11 4 5 5 18 49 3% by connections */ static int tcp_sc_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 }; /* * Compute the MAC for the SYN cookie. SIPHASH-2-4 is chosen for its speed * and good cryptographic properties. */ static uint32_t syncookie_mac(struct in_conninfo *inc, tcp_seq irs, uint8_t flags, uint8_t *secbits, uintptr_t secmod) { SIPHASH_CTX ctx; uint32_t siphash[2]; SipHash24_Init(&ctx); SipHash_SetKey(&ctx, secbits); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(inc->inc_faddr)); SipHash_Update(&ctx, &inc->inc_laddr, sizeof(inc->inc_laddr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(inc->inc6_faddr)); SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(inc->inc6_laddr)); break; #endif } SipHash_Update(&ctx, &inc->inc_fport, sizeof(inc->inc_fport)); SipHash_Update(&ctx, &inc->inc_lport, sizeof(inc->inc_lport)); SipHash_Update(&ctx, &irs, sizeof(irs)); SipHash_Update(&ctx, &flags, sizeof(flags)); SipHash_Update(&ctx, &secmod, sizeof(secmod)); SipHash_Final((u_int8_t *)&siphash, &ctx); return (siphash[0] ^ siphash[1]); } static tcp_seq syncookie_generate(struct syncache_head *sch, struct syncache *sc) { u_int i, secbit, wscale; uint32_t iss, hash; uint8_t *secbits; union syncookie cookie; SCH_LOCK_ASSERT(sch); cookie.cookie = 0; /* Map our computed MSS into the 3-bit index. */ for (i = nitems(tcp_sc_msstab) - 1; tcp_sc_msstab[i] > sc->sc_peer_mss && i > 0; i--) ; cookie.flags.mss_idx = i; /* * Map the send window scale into the 3-bit index but only if * the wscale option was received. */ if (sc->sc_flags & SCF_WINSCALE) { wscale = sc->sc_requested_s_scale; for (i = nitems(tcp_sc_wstab) - 1; tcp_sc_wstab[i] > wscale && i > 0; i--) ; cookie.flags.wscale_idx = i; } /* Can we do SACK? */ if (sc->sc_flags & SCF_SACK) cookie.flags.sack_ok = 1; /* Which of the two secrets to use. */ secbit = sch->sch_sc->secret.oddeven & 0x1; cookie.flags.odd_even = secbit; secbits = sch->sch_sc->secret.key[secbit]; hash = syncookie_mac(&sc->sc_inc, sc->sc_irs, cookie.cookie, secbits, (uintptr_t)sch); /* * Put the flags into the hash and XOR them to get better ISS number * variance. This doesn't enhance the cryptographic strength and is * done to prevent the 8 cookie bits from showing up directly on the * wire. */ iss = hash & ~0xff; iss |= cookie.cookie ^ (hash >> 24); /* Randomize the timestamp. */ if (sc->sc_flags & SCF_TIMESTAMP) { sc->sc_tsoff = arc4random() - tcp_ts_getticks(); } TCPSTAT_INC(tcps_sc_sendcookie); return (iss); } static struct syncache * syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso) { uint32_t hash; uint8_t *secbits; tcp_seq ack, seq; int wnd, wscale = 0; union syncookie cookie; SCH_LOCK_ASSERT(sch); /* * Pull information out of SYN-ACK/ACK and revert sequence number * advances. */ ack = th->th_ack - 1; seq = th->th_seq - 1; /* * Unpack the flags containing enough information to restore the * connection. */ cookie.cookie = (ack & 0xff) ^ (ack >> 24); /* Which of the two secrets to use. */ secbits = sch->sch_sc->secret.key[cookie.flags.odd_even]; hash = syncookie_mac(inc, seq, cookie.cookie, secbits, (uintptr_t)sch); /* The recomputed hash matches the ACK if this was a genuine cookie. */ if ((ack & ~0xff) != (hash & ~0xff)) return (NULL); /* Fill in the syncache values. */ sc->sc_flags = 0; bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); sc->sc_ipopts = NULL; sc->sc_irs = seq; sc->sc_iss = ack; switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: sc->sc_ip_ttl = sotoinpcb(lso)->inp_ip_ttl; sc->sc_ip_tos = sotoinpcb(lso)->inp_ip_tos; break; #endif #ifdef INET6 case INC_ISIPV6: if (sotoinpcb(lso)->inp_flags & IN6P_AUTOFLOWLABEL) sc->sc_flowlabel = sc->sc_iss & IPV6_FLOWLABEL_MASK; break; #endif } sc->sc_peer_mss = tcp_sc_msstab[cookie.flags.mss_idx]; /* We can simply recompute receive window scale we sent earlier. */ while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < sb_max) wscale++; /* Only use wscale if it was enabled in the orignal SYN. */ if (cookie.flags.wscale_idx > 0) { sc->sc_requested_r_scale = wscale; sc->sc_requested_s_scale = tcp_sc_wstab[cookie.flags.wscale_idx]; sc->sc_flags |= SCF_WINSCALE; } wnd = lso->sol_sbrcv_hiwat; wnd = imax(wnd, 0); wnd = imin(wnd, TCP_MAXWIN); sc->sc_wnd = wnd; if (cookie.flags.sack_ok) sc->sc_flags |= SCF_SACK; if (to->to_flags & TOF_TS) { sc->sc_flags |= SCF_TIMESTAMP; sc->sc_tsreflect = to->to_tsval; sc->sc_tsoff = to->to_tsecr - tcp_ts_getticks(); } if (to->to_flags & TOF_SIGNATURE) sc->sc_flags |= SCF_SIGNATURE; sc->sc_rxmits = 0; TCPSTAT_INC(tcps_sc_recvcookie); return (sc); } #ifdef INVARIANTS static int syncookie_cmp(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso) { struct syncache scs, *scx; char *s; bzero(&scs, sizeof(scs)); scx = syncookie_lookup(inc, sch, &scs, th, to, lso); if ((s = tcp_log_addrs(inc, th, NULL, NULL)) == NULL) return (0); if (scx != NULL) { if (sc->sc_peer_mss != scx->sc_peer_mss) log(LOG_DEBUG, "%s; %s: mss different %i vs %i\n", s, __func__, sc->sc_peer_mss, scx->sc_peer_mss); if (sc->sc_requested_r_scale != scx->sc_requested_r_scale) log(LOG_DEBUG, "%s; %s: rwscale different %i vs %i\n", s, __func__, sc->sc_requested_r_scale, scx->sc_requested_r_scale); if (sc->sc_requested_s_scale != scx->sc_requested_s_scale) log(LOG_DEBUG, "%s; %s: swscale different %i vs %i\n", s, __func__, sc->sc_requested_s_scale, scx->sc_requested_s_scale); if ((sc->sc_flags & SCF_SACK) != (scx->sc_flags & SCF_SACK)) log(LOG_DEBUG, "%s; %s: SACK different\n", s, __func__); } if (s != NULL) free(s, M_TCPLOG); return (0); } #endif /* INVARIANTS */ static void syncookie_reseed(void *arg) { struct tcp_syncache *sc = arg; uint8_t *secbits; int secbit; /* * Reseeding the secret doesn't have to be protected by a lock. * It only must be ensured that the new random values are visible * to all CPUs in a SMP environment. The atomic with release * semantics ensures that. */ secbit = (sc->secret.oddeven & 0x1) ? 0 : 1; secbits = sc->secret.key[secbit]; arc4rand(secbits, SYNCOOKIE_SECRET_SIZE, 0); atomic_add_rel_int(&sc->secret.oddeven, 1); /* Reschedule ourself. */ callout_schedule(&sc->secret.reseed, SYNCOOKIE_LIFETIME * hz); } /* * Exports the syncache entries to userland so that netstat can display * them alongside the other sockets. This function is intended to be * called only from tcp_pcblist. * * Due to concurrency on an active system, the number of pcbs exported * may have no relation to max_pcbs. max_pcbs merely indicates the * amount of space the caller allocated for this function to use. */ int syncache_pcblist(struct sysctl_req *req, int max_pcbs, int *pcbs_exported) { struct xtcpcb xt; struct syncache *sc; struct syncache_head *sch; int count, error, i; for (count = 0, error = 0, i = 0; i < V_tcp_syncache.hashsize; i++) { sch = &V_tcp_syncache.hashbase[i]; SCH_LOCK(sch); TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { if (count >= max_pcbs) { SCH_UNLOCK(sch); goto exit; } if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0) continue; bzero(&xt, sizeof(xt)); xt.xt_len = sizeof(xt); if (sc->sc_inc.inc_flags & INC_ISIPV6) xt.xt_inp.inp_vflag = INP_IPV6; else xt.xt_inp.inp_vflag = INP_IPV4; bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); xt.t_state = TCPS_SYN_RECEIVED; xt.xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; xt.xt_inp.xi_socket.xso_len = sizeof (struct xsocket); xt.xt_inp.xi_socket.so_type = SOCK_STREAM; xt.xt_inp.xi_socket.so_state = SS_ISCONNECTING; error = SYSCTL_OUT(req, &xt, sizeof xt); if (error) { SCH_UNLOCK(sch); goto exit; } count++; } SCH_UNLOCK(sch); } exit: *pcbs_exported = count; return error; } Index: head/sys/netinet/tcp_timewait.c =================================================================== --- head/sys/netinet/tcp_timewait.c (revision 336675) +++ head/sys/netinet/tcp_timewait.c (revision 336676) @@ -1,755 +1,755 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_tcpdebug.h" #include #include #include #include #include #include #include #include #include #include #include #ifndef INVARIANTS #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #endif #include #include #include #include #include #ifdef INET6 #include #endif #include #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #include #include -static VNET_DEFINE(uma_zone_t, tcptw_zone); +VNET_DEFINE_STATIC(uma_zone_t, tcptw_zone); #define V_tcptw_zone VNET(tcptw_zone) static int maxtcptw; /* * The timed wait queue contains references to each of the TCP sessions * currently in the TIME_WAIT state. The queue pointers, including the * queue pointers in each tcptw structure, are protected using the global * timewait lock, which must be held over queue iteration and modification. * * Rules on tcptw usage: * - a inpcb is always freed _after_ its tcptw * - a tcptw relies on its inpcb reference counting for memory stability * - a tcptw is dereferenceable only while its inpcb is locked */ -static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl); +VNET_DEFINE_STATIC(TAILQ_HEAD(, tcptw), twq_2msl); #define V_twq_2msl VNET(twq_2msl) /* Global timewait lock */ -static VNET_DEFINE(struct rwlock, tw_lock); +VNET_DEFINE_STATIC(struct rwlock, tw_lock); #define V_tw_lock VNET(tw_lock) #define TW_LOCK_INIT(tw, d) rw_init_flags(&(tw), (d), 0) #define TW_LOCK_DESTROY(tw) rw_destroy(&(tw)) #define TW_RLOCK(tw) rw_rlock(&(tw)) #define TW_WLOCK(tw) rw_wlock(&(tw)) #define TW_RUNLOCK(tw) rw_runlock(&(tw)) #define TW_WUNLOCK(tw) rw_wunlock(&(tw)) #define TW_LOCK_ASSERT(tw) rw_assert(&(tw), RA_LOCKED) #define TW_RLOCK_ASSERT(tw) rw_assert(&(tw), RA_RLOCKED) #define TW_WLOCK_ASSERT(tw) rw_assert(&(tw), RA_WLOCKED) #define TW_UNLOCK_ASSERT(tw) rw_assert(&(tw), RA_UNLOCKED) static void tcp_tw_2msl_reset(struct tcptw *, int); static void tcp_tw_2msl_stop(struct tcptw *, int); static int tcp_twrespond(struct tcptw *, int); static int tcptw_auto_size(void) { int halfrange; /* * Max out at half the ephemeral port range so that TIME_WAIT * sockets don't tie up too many ephemeral ports. */ if (V_ipport_lastauto > V_ipport_firstauto) halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2; else halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2; /* Protect against goofy port ranges smaller than 32. */ return (imin(imax(halfrange, 32), maxsockets / 5)); } static int sysctl_maxtcptw(SYSCTL_HANDLER_ARGS) { int error, new; if (maxtcptw == 0) new = tcptw_auto_size(); else new = maxtcptw; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) if (new >= 32) { maxtcptw = new; uma_zone_set_max(V_tcptw_zone, maxtcptw); } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW, &maxtcptw, 0, sysctl_maxtcptw, "IU", "Maximum number of compressed TCP TIME_WAIT entries"); -static VNET_DEFINE(int, nolocaltimewait) = 0; +VNET_DEFINE_STATIC(int, nolocaltimewait) = 0; #define V_nolocaltimewait VNET(nolocaltimewait) SYSCTL_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nolocaltimewait), 0, "Do not create compressed TCP TIME_WAIT entries for local connections"); void tcp_tw_zone_change(void) { if (maxtcptw == 0) uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); } void tcp_tw_init(void) { V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw); if (maxtcptw == 0) uma_zone_set_max(V_tcptw_zone, tcptw_auto_size()); else uma_zone_set_max(V_tcptw_zone, maxtcptw); TAILQ_INIT(&V_twq_2msl); TW_LOCK_INIT(V_tw_lock, "tcptw"); } #ifdef VIMAGE void tcp_tw_destroy(void) { struct tcptw *tw; struct epoch_tracker et; INP_INFO_RLOCK_ET(&V_tcbinfo, et); while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL) tcp_twclose(tw, 0); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); TW_LOCK_DESTROY(V_tw_lock); uma_zdestroy(V_tcptw_zone); } #endif /* * Move a TCP connection into TIME_WAIT state. * tcbinfo is locked. * inp is locked, and is unlocked before returning. */ void tcp_twstart(struct tcpcb *tp) { struct tcptw twlocal, *tw; struct inpcb *inp = tp->t_inpcb; struct socket *so; bool acknow, local; #ifdef INET6 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; #endif INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); /* A dropped inp should never transition to TIME_WAIT state. */ KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: " "(inp->inp_flags & INP_DROPPED) != 0")); if (V_nolocaltimewait) { #ifdef INET6 if (isipv6) local = in6_localaddr(&inp->in6p_faddr); else #endif #ifdef INET local = in_localip(inp->inp_faddr); #else local = false; #endif } else local = false; /* * For use only by DTrace. We do not reference the state * after this point so modifying it in place is not a problem. */ tcp_state_change(tp, TCPS_TIME_WAIT); if (local) tw = &twlocal; else tw = uma_zalloc(V_tcptw_zone, M_NOWAIT); if (tw == NULL) { /* * Reached limit on total number of TIMEWAIT connections * allowed. Remove a connection from TIMEWAIT queue in LRU * fashion to make room for this connection. * * XXX: Check if it possible to always have enough room * in advance based on guarantees provided by uma_zalloc(). */ tw = tcp_tw_2msl_scan(1); if (tw == NULL) { tp = tcp_close(tp); if (tp != NULL) INP_WUNLOCK(inp); return; } } /* * For !local case the tcptw will hold a reference on its inpcb * until tcp_twclose is called. */ tw->tw_inpcb = inp; /* * Recover last window size sent. */ if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale; else tw->last_win = 0; /* * Set t_recent if timestamps are used on the connection. */ if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) == (TF_REQ_TSTMP|TF_RCVD_TSTMP)) { tw->t_recent = tp->ts_recent; tw->ts_offset = tp->ts_offset; } else { tw->t_recent = 0; tw->ts_offset = 0; } tw->snd_nxt = tp->snd_nxt; tw->rcv_nxt = tp->rcv_nxt; tw->iss = tp->iss; tw->irs = tp->irs; tw->t_starttime = tp->t_starttime; tw->tw_time = 0; /* XXX * If this code will * be used for fin-wait-2 state also, then we may need * a ts_recent from the last segment. */ acknow = tp->t_flags & TF_ACKNOW; /* * First, discard tcpcb state, which includes stopping its timers and * freeing it. tcp_discardcb() used to also release the inpcb, but * that work is now done in the caller. * * Note: soisdisconnected() call used to be made in tcp_discardcb(), * and might not be needed here any longer. */ tcp_discardcb(tp); so = inp->inp_socket; soisdisconnected(so); tw->tw_so_options = so->so_options; inp->inp_flags |= INP_TIMEWAIT; if (acknow) tcp_twrespond(tw, TH_ACK); if (local) in_pcbdrop(inp); else { in_pcbref(inp); /* Reference from tw */ tw->tw_cred = crhold(so->so_cred); inp->inp_ppcb = tw; TCPSTATES_INC(TCPS_TIME_WAIT); tcp_tw_2msl_reset(tw, 0); } /* * If the inpcb owns the sole reference to the socket, then we can * detach and free the socket as it is not needed in time wait. */ if (inp->inp_flags & INP_SOCKREF) { KASSERT(so->so_state & SS_PROTOREF, ("tcp_twstart: !SS_PROTOREF")); inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); SOCK_LOCK(so); so->so_state &= ~SS_PROTOREF; sofree(so); } else INP_WUNLOCK(inp); } /* * Returns 1 if the TIME_WAIT state was killed and we should start over, * looking for a pcb in the listen state. Returns 0 otherwise. */ int tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th, struct mbuf *m, int tlen) { struct tcptw *tw; int thflags; tcp_seq seq; INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); /* * XXXRW: Time wait state for inpcb has been recycled, but inpcb is * still present. This is undesirable, but temporarily necessary * until we work out how to handle inpcb's who's timewait state has * been removed. */ tw = intotw(inp); if (tw == NULL) goto drop; thflags = th->th_flags; /* * NOTE: for FIN_WAIT_2 (to be added later), * must validate sequence number before accepting RST */ /* * If the segment contains RST: * Drop the segment - see Stevens, vol. 2, p. 964 and * RFC 1337. */ if (thflags & TH_RST) goto drop; #if 0 /* PAWS not needed at the moment */ /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { if ((thflags & TH_ACK) == 0) goto drop; goto ack; } /* * ts_recent is never updated because we never accept new segments. */ #endif /* * If a new connection request is received * while in TIME_WAIT, drop the old connection * and start over if the sequence numbers * are above the previous ones. */ if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { tcp_twclose(tw, 0); return (1); } /* * Drop the segment if it does not contain an ACK. */ if ((thflags & TH_ACK) == 0) goto drop; /* * Reset the 2MSL timer if this is a duplicate FIN. */ if (thflags & TH_FIN) { seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); if (seq + 1 == tw->rcv_nxt) tcp_tw_2msl_reset(tw, 1); } /* * Acknowledge the segment if it has data or is not a duplicate ACK. */ if (thflags != TH_ACK || tlen != 0 || th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) tcp_twrespond(tw, TH_ACK); drop: INP_WUNLOCK(inp); m_freem(m); return (0); } void tcp_twclose(struct tcptw *tw, int reuse) { struct socket *so; struct inpcb *inp; /* * At this point, we are in one of two situations: * * (1) We have no socket, just an inpcb<->twtcp pair. We can free * all state. * * (2) We have a socket -- if we own a reference, release it and * notify the socket layer. */ inp = tw->tw_inpcb; KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait")); KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw")); INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* in_pcbfree() */ INP_WLOCK_ASSERT(inp); tcp_tw_2msl_stop(tw, reuse); inp->inp_ppcb = NULL; in_pcbdrop(inp); so = inp->inp_socket; if (so != NULL) { /* * If there's a socket, handle two cases: first, we own a * strong reference, which we will now release, or we don't * in which case another reference exists (XXXRW: think * about this more), and we don't need to take action. */ if (inp->inp_flags & INP_SOCKREF) { inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); SOCK_LOCK(so); KASSERT(so->so_state & SS_PROTOREF, ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF")); so->so_state &= ~SS_PROTOREF; sofree(so); } else { /* * If we don't own the only reference, the socket and * inpcb need to be left around to be handled by * tcp_usr_detach() later. */ INP_WUNLOCK(inp); } } else { /* * The socket has been already cleaned-up for us, only free the * inpcb. */ in_pcbfree(inp); } TCPSTAT_INC(tcps_closed); } static int tcp_twrespond(struct tcptw *tw, int flags) { struct inpcb *inp = tw->tw_inpcb; #if defined(INET6) || defined(INET) struct tcphdr *th = NULL; #endif struct mbuf *m; #ifdef INET struct ip *ip = NULL; #endif u_int hdrlen, optlen; int error = 0; /* Keep compiler happy */ struct tcpopt to; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; #endif hdrlen = 0; /* Keep compiler happy */ INP_WLOCK_ASSERT(inp); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); m->m_data += max_linkhdr; #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif #ifdef INET6 if (isipv6) { hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)(ip6 + 1); tcpip_fillheaders(inp, ip6, th); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { hdrlen = sizeof(struct tcpiphdr); ip = mtod(m, struct ip *); th = (struct tcphdr *)(ip + 1); tcpip_fillheaders(inp, ip, th); } #endif to.to_flags = 0; /* * Send a timestamp and echo-reply if both our side and our peer * have sent timestamps in our SYN's and this is not a RST. */ if (tw->t_recent && flags == TH_ACK) { to.to_flags |= TOF_TS; to.to_tsval = tcp_ts_getticks() + tw->ts_offset; to.to_tsecr = tw->t_recent; } optlen = tcp_addoptions(&to, (u_char *)(th + 1)); m->m_len = hdrlen + optlen; m->m_pkthdr.len = m->m_len; KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small")); th->th_seq = htonl(tw->snd_nxt); th->th_ack = htonl(tw->rcv_nxt); th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; th->th_flags = flags; th->th_win = htons(tw->last_win); m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #ifdef INET6 if (isipv6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(inp, NULL); error = ip6_output(m, inp->in6p_outputopts, NULL, (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP)); ip->ip_len = htons(m->m_pkthdr.len); if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); error = ip_output(m, inp->inp_options, NULL, ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), NULL, inp); } #endif if (flags & TH_ACK) TCPSTAT_INC(tcps_sndacks); else TCPSTAT_INC(tcps_sndctrl); TCPSTAT_INC(tcps_sndtotal); return (error); } static void tcp_tw_2msl_reset(struct tcptw *tw, int rearm) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tw->tw_inpcb); TW_WLOCK(V_tw_lock); if (rearm) TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); tw->tw_time = ticks + 2 * tcp_msl; TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl); TW_WUNLOCK(V_tw_lock); } static void tcp_tw_2msl_stop(struct tcptw *tw, int reuse) { struct ucred *cred; struct inpcb *inp; int released __unused; INP_INFO_RLOCK_ASSERT(&V_tcbinfo); TW_WLOCK(V_tw_lock); inp = tw->tw_inpcb; tw->tw_inpcb = NULL; TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl); cred = tw->tw_cred; tw->tw_cred = NULL; TW_WUNLOCK(V_tw_lock); if (cred != NULL) crfree(cred); released = in_pcbrele_wlocked(inp); KASSERT(!released, ("%s: inp should not be released here", __func__)); if (!reuse) uma_zfree(V_tcptw_zone, tw); TCPSTATES_DEC(TCPS_TIME_WAIT); } struct tcptw * tcp_tw_2msl_scan(int reuse) { struct tcptw *tw; struct inpcb *inp; struct epoch_tracker et; #ifdef INVARIANTS if (reuse) { /* * Exclusive pcbinfo lock is not required in reuse case even if * two inpcb locks can be acquired simultaneously: * - the inpcb transitioning to TIME_WAIT state in * tcp_tw_start(), * - the inpcb closed by tcp_twclose(). * * It is because only inpcbs in FIN_WAIT2 or CLOSING states can * transition in TIME_WAIT state. Then a pcbcb cannot be in * TIME_WAIT list and transitioning to TIME_WAIT state at same * time. */ INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } #endif for (;;) { TW_RLOCK(V_tw_lock); tw = TAILQ_FIRST(&V_twq_2msl); if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) { TW_RUNLOCK(V_tw_lock); break; } KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL", __func__)); inp = tw->tw_inpcb; in_pcbref(inp); TW_RUNLOCK(V_tw_lock); INP_INFO_RLOCK_ET(&V_tcbinfo, et); INP_WLOCK(inp); tw = intotw(inp); if (in_pcbrele_wlocked(inp)) { if (__predict_true(tw == NULL)) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); continue; } else { /* This should not happen as in TIMEWAIT * state the inp should not be destroyed * before its tcptw. If INVARIANTS is * defined panic. */ #ifdef INVARIANTS panic("%s: Panic before an infinite " "loop: INP_TIMEWAIT && (INP_FREED " "|| inp last reference) && tw != " "NULL", __func__); #else log(LOG_ERR, "%s: Avoid an infinite " "loop: INP_TIMEWAIT && (INP_FREED " "|| inp last reference) && tw != " "NULL", __func__); #endif INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); break; } } if (tw == NULL) { /* tcp_twclose() has already been called */ INP_WUNLOCK(inp); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); continue; } tcp_twclose(tw, reuse); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); if (reuse) return tw; } return NULL; } Index: head/sys/netinet/udp_usrreq.c =================================================================== --- head/sys/netinet/udp_usrreq.c (revision 336675) +++ head/sys/netinet/udp_usrreq.c (revision 336676) @@ -1,1804 +1,1804 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. * Copyright (c) 2008 Robert N. M. Watson * Copyright (c) 2010-2011 Juniper Networks, Inc. * Copyright (c) 2014 Kevin Lo * All rights reserved. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include /* * UDP and UDP-Lite protocols implementation. * Per RFC 768, August, 1980. * Per RFC 3828, July, 2004. */ /* * BSD 4.2 defaulted the udp checksum to be off. Turning off udp checksums * removes the only data integrity mechanism for packets and malformed * packets that would otherwise be discarded due to bad checksums, and may * cause problems (especially for NFS data blocks). */ VNET_DEFINE(int, udp_cksum) = 1; SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(udp_cksum), 0, "compute udp checksum"); int udp_log_in_vain = 0; SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, &udp_log_in_vain, 0, "Log all incoming UDP packets"); VNET_DEFINE(int, udp_blackhole) = 0; SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(udp_blackhole), 0, "Do not send port unreachables for refused connects"); u_long udp_sendspace = 9216; /* really max datagram size */ SYSCTL_ULONG(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); u_long udp_recvspace = 40 * (1024 + #ifdef INET6 sizeof(struct sockaddr_in6) #else sizeof(struct sockaddr_in) #endif ); /* 40 1K datagrams */ SYSCTL_ULONG(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, &udp_recvspace, 0, "Maximum space for incoming UDP datagrams"); VNET_DEFINE(struct inpcbhead, udb); /* from udp_var.h */ VNET_DEFINE(struct inpcbinfo, udbinfo); VNET_DEFINE(struct inpcbhead, ulitecb); VNET_DEFINE(struct inpcbinfo, ulitecbinfo); -static VNET_DEFINE(uma_zone_t, udpcb_zone); +VNET_DEFINE_STATIC(uma_zone_t, udpcb_zone); #define V_udpcb_zone VNET(udpcb_zone) #ifndef UDBHASHSIZE #define UDBHASHSIZE 128 #endif VNET_PCPUSTAT_DEFINE(struct udpstat, udpstat); /* from udp_var.h */ VNET_PCPUSTAT_SYSINIT(udpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_udp, UDPCTL_STATS, stats, struct udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(udpstat); #endif /* VIMAGE */ #ifdef INET static void udp_detach(struct socket *so); static int udp_output(struct inpcb *, struct mbuf *, struct sockaddr *, struct mbuf *, struct thread *); #endif static void udp_zone_change(void *tag) { uma_zone_set_max(V_udbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_udpcb_zone, maxsockets); } static int udp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp; inp = mem; INP_LOCK_INIT(inp, "inp", "udpinp"); return (0); } static int udplite_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp; inp = mem; INP_LOCK_INIT(inp, "inp", "udpliteinp"); return (0); } void udp_init(void) { /* * For now default to 2-tuple UDP hashing - until the fragment * reassembly code can also update the flowid. * * Once we can calculate the flowid that way and re-establish * a 4-tuple, flip this to 4-tuple. */ in_pcbinfo_init(&V_udbinfo, "udp", &V_udb, UDBHASHSIZE, UDBHASHSIZE, "udp_inpcb", udp_inpcb_init, IPI_HASHFIELDS_2TUPLE); V_udpcb_zone = uma_zcreate("udpcb", sizeof(struct udpcb), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_udpcb_zone, maxsockets); uma_zone_set_warning(V_udpcb_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, udp_zone_change, NULL, EVENTHANDLER_PRI_ANY); } void udplite_init(void) { in_pcbinfo_init(&V_ulitecbinfo, "udplite", &V_ulitecb, UDBHASHSIZE, UDBHASHSIZE, "udplite_inpcb", udplite_inpcb_init, IPI_HASHFIELDS_2TUPLE); } /* * Kernel module interface for updating udpstat. The argument is an index * into udpstat treated as an array of u_long. While this encodes the * general layout of udpstat into the caller, it doesn't encode its location, * so that future changes to add, for example, per-CPU stats support won't * cause binary compatibility problems for kernel modules. */ void kmod_udpstat_inc(int statnum) { counter_u64_add(VNET(udpstat)[statnum], 1); } int udp_newudpcb(struct inpcb *inp) { struct udpcb *up; up = uma_zalloc(V_udpcb_zone, M_NOWAIT | M_ZERO); if (up == NULL) return (ENOBUFS); inp->inp_ppcb = up; return (0); } void udp_discardcb(struct udpcb *up) { uma_zfree(V_udpcb_zone, up); } #ifdef VIMAGE static void udp_destroy(void *unused __unused) { in_pcbinfo_destroy(&V_udbinfo); uma_zdestroy(V_udpcb_zone); } VNET_SYSUNINIT(udp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udp_destroy, NULL); static void udplite_destroy(void *unused __unused) { in_pcbinfo_destroy(&V_ulitecbinfo); } VNET_SYSUNINIT(udplite, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, udplite_destroy, NULL); #endif #ifdef INET /* * Subroutine of udp_input(), which appends the provided mbuf chain to the * passed pcb/socket. The caller must provide a sockaddr_in via udp_in that * contains the source address. If the socket ends up being an IPv6 socket, * udp_append() will convert to a sockaddr_in6 before passing the address * into the socket code. * * In the normal case udp_append() will return 0, indicating that you * must unlock the inp. However if a tunneling protocol is in place we increment * the inpcb refcnt and unlock the inp, on return from the tunneling protocol we * then decrement the reference count. If the inp_rele returns 1, indicating the * inp is gone, we return that to the caller to tell them *not* to unlock * the inp. In the case of multi-cast this will cause the distribution * to stop (though most tunneling protocols known currently do *not* use * multicast). */ static int udp_append(struct inpcb *inp, struct ip *ip, struct mbuf *n, int off, struct sockaddr_in *udp_in) { struct sockaddr *append_sa; struct socket *so; struct mbuf *tmpopts, *opts = NULL; #ifdef INET6 struct sockaddr_in6 udp_in6; #endif struct udpcb *up; INP_LOCK_ASSERT(inp); /* * Engage the tunneling protocol. */ up = intoudpcb(inp); if (up->u_tun_func != NULL) { in_pcbref(inp); INP_RUNLOCK(inp); (*up->u_tun_func)(n, off, inp, (struct sockaddr *)&udp_in[0], up->u_tun_ctx); INP_RLOCK(inp); return (in_pcbrele_rlocked(inp)); } off += sizeof(struct udphdr); #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* Check AH/ESP integrity. */ if (IPSEC_ENABLED(ipv4) && IPSEC_CHECK_POLICY(ipv4, n, inp) != 0) { m_freem(n); return (0); } if (up->u_flags & UF_ESPINUDP) {/* IPSec UDP encaps. */ if (IPSEC_ENABLED(ipv4) && UDPENCAP_INPUT(n, off, AF_INET) != 0) return (0); /* Consumed. */ } #endif /* IPSEC */ #ifdef MAC if (mac_inpcb_check_deliver(inp, n) != 0) { m_freem(n); return (0); } #endif /* MAC */ if (inp->inp_flags & INP_CONTROLOPTS || inp->inp_socket->so_options & (SO_TIMESTAMP | SO_BINTIME)) { #ifdef INET6 if (inp->inp_vflag & INP_IPV6) (void)ip6_savecontrol_v4(inp, n, &opts, NULL); else #endif /* INET6 */ ip_savecontrol(inp, &opts, ip, n); } if ((inp->inp_vflag & INP_IPV4) && (inp->inp_flags2 & INP_ORIGDSTADDR)) { tmpopts = sbcreatecontrol((caddr_t)&udp_in[1], sizeof(struct sockaddr_in), IP_ORIGDSTADDR, IPPROTO_IP); if (tmpopts) { if (opts) { tmpopts->m_next = opts; opts = tmpopts; } else opts = tmpopts; } } #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { bzero(&udp_in6, sizeof(udp_in6)); udp_in6.sin6_len = sizeof(udp_in6); udp_in6.sin6_family = AF_INET6; in6_sin_2_v4mapsin6(&udp_in[0], &udp_in6); append_sa = (struct sockaddr *)&udp_in6; } else #endif /* INET6 */ append_sa = (struct sockaddr *)&udp_in[0]; m_adj(n, off); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_rcv); if (sbappendaddr_locked(&so->so_rcv, append_sa, n, opts) == 0) { SOCKBUF_UNLOCK(&so->so_rcv); m_freem(n); if (opts) m_freem(opts); UDPSTAT_INC(udps_fullsock); } else sorwakeup_locked(so); return (0); } int udp_input(struct mbuf **mp, int *offp, int proto) { struct ip *ip; struct udphdr *uh; struct ifnet *ifp; struct inpcb *inp; uint16_t len, ip_len; struct inpcbinfo *pcbinfo; struct ip save_ip; struct sockaddr_in udp_in[2]; struct mbuf *m; struct m_tag *fwd_tag; struct epoch_tracker et; int cscov_partial, iphlen; m = *mp; iphlen = *offp; ifp = m->m_pkthdr.rcvif; *mp = NULL; UDPSTAT_INC(udps_ipackets); /* * Strip IP options, if any; should skip this, make available to * user, and use on returned packets, but we don't yet have a way to * check the checksum with options still present. */ if (iphlen > sizeof (struct ip)) { ip_stripoptions(m); iphlen = sizeof(struct ip); } /* * Get IP and UDP header together in first mbuf. */ ip = mtod(m, struct ip *); if (m->m_len < iphlen + sizeof(struct udphdr)) { if ((m = m_pullup(m, iphlen + sizeof(struct udphdr))) == NULL) { UDPSTAT_INC(udps_hdrops); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); } uh = (struct udphdr *)((caddr_t)ip + iphlen); cscov_partial = (proto == IPPROTO_UDPLITE) ? 1 : 0; /* * Destination port of 0 is illegal, based on RFC768. */ if (uh->uh_dport == 0) goto badunlocked; /* * Construct sockaddr format source address. Stuff source address * and datagram in user buffer. */ bzero(&udp_in[0], sizeof(struct sockaddr_in) * 2); udp_in[0].sin_len = sizeof(struct sockaddr_in); udp_in[0].sin_family = AF_INET; udp_in[0].sin_port = uh->uh_sport; udp_in[0].sin_addr = ip->ip_src; udp_in[1].sin_len = sizeof(struct sockaddr_in); udp_in[1].sin_family = AF_INET; udp_in[1].sin_port = uh->uh_dport; udp_in[1].sin_addr = ip->ip_dst; /* * Make mbuf data length reflect UDP length. If not enough data to * reflect UDP length, drop. */ len = ntohs((u_short)uh->uh_ulen); ip_len = ntohs(ip->ip_len) - iphlen; if (proto == IPPROTO_UDPLITE && (len == 0 || len == ip_len)) { /* Zero means checksum over the complete packet. */ if (len == 0) len = ip_len; cscov_partial = 0; } if (ip_len != len) { if (len > ip_len || len < sizeof(struct udphdr)) { UDPSTAT_INC(udps_badlen); goto badunlocked; } if (proto == IPPROTO_UDP) m_adj(m, len - ip_len); } /* * Save a copy of the IP header in case we want restore it for * sending an ICMP error message in response. */ if (!V_udp_blackhole) save_ip = *ip; else memset(&save_ip, 0, sizeof(save_ip)); /* * Checksum extended UDP header and data. */ if (uh->uh_sum) { u_short uh_sum; if ((m->m_pkthdr.csum_flags & CSUM_DATA_VALID) && !cscov_partial) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) uh_sum = m->m_pkthdr.csum_data; else uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + proto)); uh_sum ^= 0xffff; } else { char b[9]; bcopy(((struct ipovly *)ip)->ih_x1, b, 9); bzero(((struct ipovly *)ip)->ih_x1, 9); ((struct ipovly *)ip)->ih_len = (proto == IPPROTO_UDP) ? uh->uh_ulen : htons(ip_len); uh_sum = in_cksum(m, len + sizeof (struct ip)); bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); } if (uh_sum) { UDPSTAT_INC(udps_badsum); m_freem(m); return (IPPROTO_DONE); } } else { if (proto == IPPROTO_UDP) { UDPSTAT_INC(udps_nosum); } else { /* UDPLite requires a checksum */ /* XXX: What is the right UDPLite MIB counter here? */ m_freem(m); return (IPPROTO_DONE); } } pcbinfo = udp_get_inpcbinfo(proto); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || in_broadcast(ip->ip_dst, ifp)) { struct inpcb *last; struct inpcbhead *pcblist; struct ip_moptions *imo; INP_INFO_RLOCK_ET(pcbinfo, et); pcblist = udp_get_pcblist(proto); last = NULL; CK_LIST_FOREACH(inp, pcblist, inp_list) { if (inp->inp_lport != uh->uh_dport) continue; #ifdef INET6 if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif if (inp->inp_laddr.s_addr != INADDR_ANY && inp->inp_laddr.s_addr != ip->ip_dst.s_addr) continue; if (inp->inp_faddr.s_addr != INADDR_ANY && inp->inp_faddr.s_addr != ip->ip_src.s_addr) continue; if (inp->inp_fport != 0 && inp->inp_fport != uh->uh_sport) continue; INP_RLOCK(inp); /* * XXXRW: Because we weren't holding either the inpcb * or the hash lock when we checked for a match * before, we should probably recheck now that the * inpcb lock is held. */ /* * Handle socket delivery policy for any-source * and source-specific multicast. [RFC3678] */ imo = inp->inp_moptions; if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { struct sockaddr_in group; int blocked; if (imo == NULL) { INP_RUNLOCK(inp); continue; } bzero(&group, sizeof(struct sockaddr_in)); group.sin_len = sizeof(struct sockaddr_in); group.sin_family = AF_INET; group.sin_addr = ip->ip_dst; blocked = imo_multi_filter(imo, ifp, (struct sockaddr *)&group, (struct sockaddr *)&udp_in[0]); if (blocked != MCAST_PASS) { if (blocked == MCAST_NOTGMEMBER) IPSTAT_INC(ips_notmember); if (blocked == MCAST_NOTSMEMBER || blocked == MCAST_MUTED) UDPSTAT_INC(udps_filtermcast); INP_RUNLOCK(inp); continue; } } if (last != NULL) { struct mbuf *n; if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { UDP_PROBE(receive, NULL, last, ip, last, uh); if (udp_append(last, ip, n, iphlen, udp_in)) { goto inp_lost; } } INP_RUNLOCK(last); } last = inp; /* * Don't look for additional matches if this one does * not have either the SO_REUSEPORT or SO_REUSEADDR * socket options set. This heuristic avoids * searching through all pcbs in the common case of a * non-shared port. It assumes that an application * will never clear these options after setting them. */ if ((last->inp_socket->so_options & (SO_REUSEPORT|SO_REUSEPORT_LB|SO_REUSEADDR)) == 0) break; } if (last == NULL) { /* * No matching pcb found; discard datagram. (No need * to send an ICMP Port Unreachable for a broadcast * or multicast datgram.) */ UDPSTAT_INC(udps_noportbcast); if (inp) INP_RUNLOCK(inp); INP_INFO_RUNLOCK_ET(pcbinfo, et); goto badunlocked; } UDP_PROBE(receive, NULL, last, ip, last, uh); if (udp_append(last, ip, m, iphlen, udp_in) == 0) INP_RUNLOCK(last); inp_lost: INP_INFO_RUNLOCK_ET(pcbinfo, et); return (IPPROTO_DONE); } /* * Locate pcb for datagram. */ /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ((m->m_flags & M_IP_NEXTHOP) && (fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, ip->ip_dst, uh->uh_dport, INPLOOKUP_RLOCKPCB, ifp, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(pcbinfo, ip->ip_src, uh->uh_sport, next_hop->sin_addr, next_hop->sin_port ? htons(next_hop->sin_port) : uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, ifp); } /* Remove the tag from the packet. We don't need it anymore. */ m_tag_delete(m, fwd_tag); m->m_flags &= ~M_IP_NEXTHOP; } else inp = in_pcblookup_mbuf(pcbinfo, ip->ip_src, uh->uh_sport, ip->ip_dst, uh->uh_dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, ifp, m); if (inp == NULL) { if (udp_log_in_vain) { char src[INET_ADDRSTRLEN]; char dst[INET_ADDRSTRLEN]; log(LOG_INFO, "Connection attempt to UDP %s:%d from %s:%d\n", inet_ntoa_r(ip->ip_dst, dst), ntohs(uh->uh_dport), inet_ntoa_r(ip->ip_src, src), ntohs(uh->uh_sport)); } UDP_PROBE(receive, NULL, NULL, ip, NULL, uh); UDPSTAT_INC(udps_noport); if (m->m_flags & (M_BCAST | M_MCAST)) { UDPSTAT_INC(udps_noportbcast); goto badunlocked; } if (V_udp_blackhole) goto badunlocked; if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) goto badunlocked; *ip = save_ip; icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); return (IPPROTO_DONE); } /* * Check the minimum TTL for socket. */ INP_RLOCK_ASSERT(inp); if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl) { UDP_PROBE(receive, NULL, inp, ip, inp, uh); INP_RUNLOCK(inp); m_freem(m); return (IPPROTO_DONE); } if (cscov_partial) { struct udpcb *up; up = intoudpcb(inp); if (up->u_rxcslen == 0 || up->u_rxcslen > len) { INP_RUNLOCK(inp); m_freem(m); return (IPPROTO_DONE); } } UDP_PROBE(receive, NULL, inp, ip, inp, uh); if (udp_append(inp, ip, m, iphlen, udp_in) == 0) INP_RUNLOCK(inp); return (IPPROTO_DONE); badunlocked: m_freem(m); return (IPPROTO_DONE); } #endif /* INET */ /* * Notify a udp user of an asynchronous error; just wake up so that they can * collect error status. */ struct inpcb * udp_notify(struct inpcb *inp, int errno) { /* * While udp_ctlinput() always calls udp_notify() with a read lock * when invoking it directly, in_pcbnotifyall() currently uses write * locks due to sharing code with TCP. For now, accept either a read * or a write lock, but a read lock is sufficient. */ INP_LOCK_ASSERT(inp); if ((errno == EHOSTUNREACH || errno == ENETUNREACH || errno == EHOSTDOWN) && inp->inp_route.ro_rt) { RTFREE(inp->inp_route.ro_rt); inp->inp_route.ro_rt = (struct rtentry *)NULL; } inp->inp_socket->so_error = errno; sorwakeup(inp->inp_socket); sowwakeup(inp->inp_socket); return (inp); } #ifdef INET static void udp_common_ctlinput(int cmd, struct sockaddr *sa, void *vip, struct inpcbinfo *pcbinfo) { struct ip *ip = vip; struct udphdr *uh; struct in_addr faddr; struct inpcb *inp; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ in_pcbnotifyall(&V_udbinfo, faddr, EHOSTDOWN, udp_notify); return; } /* * Hostdead is ugly because it goes linearly through all PCBs. * * XXX: We never get this from ICMP, otherwise it makes an excellent * DoS attack on machines with many connections. */ if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip != NULL) { uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, ip->ip_src, uh->uh_sport, INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { INP_RLOCK_ASSERT(inp); if (inp->inp_socket != NULL) { udp_notify(inp, inetctlerrmap[cmd]); } INP_RUNLOCK(inp); } else { inp = in_pcblookup(pcbinfo, faddr, uh->uh_dport, ip->ip_src, uh->uh_sport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { struct udpcb *up; void *ctx; udp_tun_icmp_t func; up = intoudpcb(inp); ctx = up->u_tun_ctx; func = up->u_icmp_func; INP_RUNLOCK(inp); if (func != NULL) (*func)(cmd, sa, vip, ctx); } } } else in_pcbnotifyall(pcbinfo, faddr, inetctlerrmap[cmd], udp_notify); } void udp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { return (udp_common_ctlinput(cmd, sa, vip, &V_udbinfo)); } void udplite_ctlinput(int cmd, struct sockaddr *sa, void *vip) { return (udp_common_ctlinput(cmd, sa, vip, &V_ulitecbinfo)); } #endif /* INET */ static int udp_pcblist(SYSCTL_HANDLER_ARGS) { int error, i, n; struct inpcb *inp, **inp_list; inp_gen_t gencnt; struct xinpgen xig; struct epoch_tracker et; /* * The process of preparing the PCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == 0) { n = V_udbinfo.ipi_count; n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); return (0); } if (req->newptr != 0) return (EPERM); /* * OK, now we're committed to doing something. */ INP_INFO_RLOCK_ET(&V_udbinfo, et); gencnt = V_udbinfo.ipi_gencnt; n = V_udbinfo.ipi_count; INP_INFO_RUNLOCK_ET(&V_udbinfo, et); error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) + n * sizeof(struct xinpcb)); if (error != 0) return (error); xig.xig_len = sizeof xig; xig.xig_count = n; xig.xig_gen = gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == NULL) return (ENOMEM); INP_INFO_RLOCK_ET(&V_udbinfo, et); for (inp = CK_LIST_FIRST(V_udbinfo.ipi_listhead), i = 0; inp && i < n; inp = CK_LIST_NEXT(inp, inp_list)) { INP_WLOCK(inp); if (inp->inp_gencnt <= gencnt && cr_canseeinpcb(req->td->td_ucred, inp) == 0) { in_pcbref(inp); inp_list[i++] = inp; } INP_WUNLOCK(inp); } INP_INFO_RUNLOCK_ET(&V_udbinfo, et); n = i; error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (inp->inp_gencnt <= gencnt) { struct xinpcb xi; in_pcbtoxinpcb(inp, &xi); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xi, sizeof xi); } else INP_RUNLOCK(inp); } INP_INFO_WLOCK(&V_udbinfo); for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_WUNLOCK(&V_udbinfo); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ INP_INFO_RLOCK_ET(&V_udbinfo, et); xig.xig_gen = V_udbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_udbinfo.ipi_count; INP_INFO_RUNLOCK_ET(&V_udbinfo, et); error = SYSCTL_OUT(req, &xig, sizeof xig); } free(inp_list, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, udp_pcblist, "S,xinpcb", "List of active UDP sockets"); #ifdef INET static int udp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); inp = in_pcblookup(&V_udbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { INP_RLOCK_ASSERT(inp); if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, udp_getcred, "S,xucred", "Get the xucred of a UDP connection"); #endif /* INET */ int udp_ctloutput(struct socket *so, struct sockopt *sopt) { struct inpcb *inp; struct udpcb *up; int isudplite, error, optval; error = 0; isudplite = (so->so_proto->pr_protocol == IPPROTO_UDPLITE) ? 1 : 0; inp = sotoinpcb(so); KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); INP_WLOCK(inp); if (sopt->sopt_level != so->so_proto->pr_protocol) { #ifdef INET6 if (INP_CHECK_SOCKAF(so, AF_INET6)) { INP_WUNLOCK(inp); error = ip6_ctloutput(so, sopt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { INP_WUNLOCK(inp); error = ip_ctloutput(so, sopt); } #endif return (error); } switch (sopt->sopt_dir) { case SOPT_SET: switch (sopt->sopt_name) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET case UDP_ENCAP: if (!IPSEC_ENABLED(ipv4)) { INP_WUNLOCK(inp); return (ENOPROTOOPT); } error = UDPENCAP_PCBCTL(inp, sopt); break; #endif /* INET */ #endif /* IPSEC */ case UDPLITE_SEND_CSCOV: case UDPLITE_RECV_CSCOV: if (!isudplite) { INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } INP_WUNLOCK(inp); error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error != 0) break; inp = sotoinpcb(so); KASSERT(inp != NULL, ("%s: inp == NULL", __func__)); INP_WLOCK(inp); up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); if ((optval != 0 && optval < 8) || (optval > 65535)) { INP_WUNLOCK(inp); error = EINVAL; break; } if (sopt->sopt_name == UDPLITE_SEND_CSCOV) up->u_txcslen = optval; else up->u_rxcslen = optval; INP_WUNLOCK(inp); break; default: INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } break; case SOPT_GET: switch (sopt->sopt_name) { #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET case UDP_ENCAP: if (!IPSEC_ENABLED(ipv4)) { INP_WUNLOCK(inp); return (ENOPROTOOPT); } error = UDPENCAP_PCBCTL(inp, sopt); break; #endif /* INET */ #endif /* IPSEC */ case UDPLITE_SEND_CSCOV: case UDPLITE_RECV_CSCOV: if (!isudplite) { INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); if (sopt->sopt_name == UDPLITE_SEND_CSCOV) optval = up->u_txcslen; else optval = up->u_rxcslen; INP_WUNLOCK(inp); error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: INP_WUNLOCK(inp); error = ENOPROTOOPT; break; } break; } return (error); } #ifdef INET #define UH_WLOCKED 2 #define UH_RLOCKED 1 #define UH_UNLOCKED 0 static int udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { struct udpiphdr *ui; int len = m->m_pkthdr.len; struct in_addr faddr, laddr; struct cmsghdr *cm; struct inpcbinfo *pcbinfo; struct sockaddr_in *sin, src; struct epoch_tracker et; int cscov_partial = 0; int error = 0; int ipflags; u_short fport, lport; int unlock_udbinfo, unlock_inp; u_char tos; uint8_t pr; uint16_t cscov = 0; uint32_t flowid = 0; uint8_t flowtype = M_HASHTYPE_NONE; /* * udp_output() may need to temporarily bind or connect the current * inpcb. As such, we don't know up front whether we will need the * pcbinfo lock or not. Do any work to decide what is needed up * front before acquiring any locks. */ if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { if (control) m_freem(control); m_freem(m); return (EMSGSIZE); } src.sin_family = 0; sin = (struct sockaddr_in *)addr; if (sin == NULL || (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { INP_WLOCK(inp); unlock_inp = UH_WLOCKED; } else { INP_RLOCK(inp); unlock_inp = UH_RLOCKED; } tos = inp->inp_ip_tos; if (control != NULL) { /* * XXX: Currently, we assume all the optional information is * stored in a single mbuf. */ if (control->m_next) { if (unlock_inp == UH_WLOCKED) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); m_freem(control); m_freem(m); return (EINVAL); } for (; control->m_len > 0; control->m_data += CMSG_ALIGN(cm->cmsg_len), control->m_len -= CMSG_ALIGN(cm->cmsg_len)) { cm = mtod(control, struct cmsghdr *); if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0 || cm->cmsg_len > control->m_len) { error = EINVAL; break; } if (cm->cmsg_level != IPPROTO_IP) continue; switch (cm->cmsg_type) { case IP_SENDSRCADDR: if (cm->cmsg_len != CMSG_LEN(sizeof(struct in_addr))) { error = EINVAL; break; } bzero(&src, sizeof(src)); src.sin_family = AF_INET; src.sin_len = sizeof(src); src.sin_port = inp->inp_lport; src.sin_addr = *(struct in_addr *)CMSG_DATA(cm); break; case IP_TOS: if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) { error = EINVAL; break; } tos = *(u_char *)CMSG_DATA(cm); break; case IP_FLOWID: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } flowid = *(uint32_t *) CMSG_DATA(cm); break; case IP_FLOWTYPE: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } flowtype = *(uint32_t *) CMSG_DATA(cm); break; #ifdef RSS case IP_RSSBUCKETID: if (cm->cmsg_len != CMSG_LEN(sizeof(uint32_t))) { error = EINVAL; break; } /* This is just a placeholder for now */ break; #endif /* RSS */ default: error = ENOPROTOOPT; break; } if (error) break; } m_freem(control); } if (error) { if (unlock_inp == UH_WLOCKED) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); m_freem(m); return (error); } /* * Depending on whether or not the application has bound or connected * the socket, we may have to do varying levels of work. The optimal * case is for a connected UDP socket, as a global lock isn't * required at all. * * In order to decide which we need, we require stability of the * inpcb binding, which we ensure by acquiring a read lock on the * inpcb. This doesn't strictly follow the lock order, so we play * the trylock and retry game; note that we may end up with more * conservative locks than required the second time around, so later * assertions have to accept that. Further analysis of the number of * misses under contention is required. * * XXXRW: Check that hash locking update here is correct. */ pr = inp->inp_socket->so_proto->pr_protocol; pcbinfo = udp_get_inpcbinfo(pr); sin = (struct sockaddr_in *)addr; if (sin != NULL && (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) { INP_HASH_WLOCK(pcbinfo); unlock_udbinfo = UH_WLOCKED; } else if ((sin != NULL && ( (sin->sin_addr.s_addr == INADDR_ANY) || (sin->sin_addr.s_addr == INADDR_BROADCAST) || (inp->inp_laddr.s_addr == INADDR_ANY) || (inp->inp_lport == 0))) || (src.sin_family == AF_INET)) { INP_HASH_RLOCK_ET(pcbinfo, et); unlock_udbinfo = UH_RLOCKED; } else unlock_udbinfo = UH_UNLOCKED; /* * If the IP_SENDSRCADDR control message was specified, override the * source address for this datagram. Its use is invalidated if the * address thus specified is incomplete or clobbers other inpcbs. */ laddr = inp->inp_laddr; lport = inp->inp_lport; if (src.sin_family == AF_INET) { INP_HASH_LOCK_ASSERT(pcbinfo); if ((lport == 0) || (laddr.s_addr == INADDR_ANY && src.sin_addr.s_addr == INADDR_ANY)) { error = EINVAL; goto release; } error = in_pcbbind_setup(inp, (struct sockaddr *)&src, &laddr.s_addr, &lport, td->td_ucred); if (error) goto release; } /* * If a UDP socket has been connected, then a local address/port will * have been selected and bound. * * If a UDP socket has not been connected to, then an explicit * destination address must be used, in which case a local * address/port may not have been selected and bound. */ if (sin != NULL) { INP_LOCK_ASSERT(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { error = EISCONN; goto release; } /* * Jail may rewrite the destination address, so let it do * that before we use it. */ error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); if (error) goto release; /* * If a local address or port hasn't yet been selected, or if * the destination address needs to be rewritten due to using * a special INADDR_ constant, invoke in_pcbconnect_setup() * to do the heavy lifting. Once a port is selected, we * commit the binding back to the socket; we also commit the * binding of the address if in jail. * * If we already have a valid binding and we're not * requesting a destination address rewrite, use a fast path. */ if (inp->inp_laddr.s_addr == INADDR_ANY || inp->inp_lport == 0 || sin->sin_addr.s_addr == INADDR_ANY || sin->sin_addr.s_addr == INADDR_BROADCAST) { INP_HASH_LOCK_ASSERT(pcbinfo); error = in_pcbconnect_setup(inp, addr, &laddr.s_addr, &lport, &faddr.s_addr, &fport, NULL, td->td_ucred); if (error) goto release; /* * XXXRW: Why not commit the port if the address is * !INADDR_ANY? */ /* Commit the local port if newly assigned. */ if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) { INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); /* * Remember addr if jailed, to prevent * rebinding. */ if (prison_flag(td->td_ucred, PR_IP4)) inp->inp_laddr = laddr; inp->inp_lport = lport; if (in_pcbinshash(inp) != 0) { inp->inp_lport = 0; error = EAGAIN; goto release; } inp->inp_flags |= INP_ANONPORT; } } else { faddr = sin->sin_addr; fport = sin->sin_port; } } else { INP_LOCK_ASSERT(inp); faddr = inp->inp_faddr; fport = inp->inp_fport; if (faddr.s_addr == INADDR_ANY) { error = ENOTCONN; goto release; } } /* * Calculate data length and get a mbuf for UDP, IP, and possible * link-layer headers. Immediate slide the data pointer back forward * since we won't use that space at this layer. */ M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT); if (m == NULL) { error = ENOBUFS; goto release; } m->m_data += max_linkhdr; m->m_len -= max_linkhdr; m->m_pkthdr.len -= max_linkhdr; /* * Fill in mbuf with extended UDP header and addresses and length put * into network format. */ ui = mtod(m, struct udpiphdr *); bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */ ui->ui_v = IPVERSION << 4; ui->ui_pr = pr; ui->ui_src = laddr; ui->ui_dst = faddr; ui->ui_sport = lport; ui->ui_dport = fport; ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); if (pr == IPPROTO_UDPLITE) { struct udpcb *up; uint16_t plen; up = intoudpcb(inp); cscov = up->u_txcslen; plen = (u_short)len + sizeof(struct udphdr); if (cscov >= plen) cscov = 0; ui->ui_len = htons(plen); ui->ui_ulen = htons(cscov); /* * For UDP-Lite, checksum coverage length of zero means * the entire UDPLite packet is covered by the checksum. */ cscov_partial = (cscov == 0) ? 0 : 1; } /* * Set the Don't Fragment bit in the IP header. */ if (inp->inp_flags & INP_DONTFRAG) { struct ip *ip; ip = (struct ip *)&ui->ui_i; ip->ip_off |= htons(IP_DF); } ipflags = 0; if (inp->inp_socket->so_options & SO_DONTROUTE) ipflags |= IP_ROUTETOIF; if (inp->inp_socket->so_options & SO_BROADCAST) ipflags |= IP_ALLOWBROADCAST; if (inp->inp_flags & INP_ONESBCAST) ipflags |= IP_SENDONES; #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif /* * Set up checksum and output datagram. */ ui->ui_sum = 0; if (pr == IPPROTO_UDPLITE) { if (inp->inp_flags & INP_ONESBCAST) faddr.s_addr = INADDR_BROADCAST; if (cscov_partial) { if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0) ui->ui_sum = 0xffff; } else { if ((ui->ui_sum = in_cksum(m, sizeof(struct udpiphdr) + len)) == 0) ui->ui_sum = 0xffff; } } else if (V_udp_cksum) { if (inp->inp_flags & INP_ONESBCAST) faddr.s_addr = INADDR_BROADCAST; ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr, htons((u_short)len + sizeof(struct udphdr) + pr)); m->m_pkthdr.csum_flags = CSUM_UDP; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); } ((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len); ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ ((struct ip *)ui)->ip_tos = tos; /* XXX */ UDPSTAT_INC(udps_opackets); /* * Setup flowid / RSS information for outbound socket. * * Once the UDP code decides to set a flowid some other way, * this allows the flowid to be overridden by userland. */ if (flowtype != M_HASHTYPE_NONE) { m->m_pkthdr.flowid = flowid; M_HASHTYPE_SET(m, flowtype); #ifdef RSS } else { uint32_t hash_val, hash_type; /* * Calculate an appropriate RSS hash for UDP and * UDP Lite. * * The called function will take care of figuring out * whether a 2-tuple or 4-tuple hash is required based * on the currently configured scheme. * * Later later on connected socket values should be * cached in the inpcb and reused, rather than constantly * re-calculating it. * * UDP Lite is a different protocol number and will * likely end up being hashed as a 2-tuple until * RSS / NICs grow UDP Lite protocol awareness. */ if (rss_proto_software_hash_v4(faddr, laddr, fport, lport, pr, &hash_val, &hash_type) == 0) { m->m_pkthdr.flowid = hash_val; M_HASHTYPE_SET(m, hash_type); } #endif } #ifdef RSS /* * Don't override with the inp cached flowid value. * * Depending upon the kind of send being done, the inp * flowid/flowtype values may actually not be appropriate * for this particular socket send. * * We should either leave the flowid at zero (which is what is * currently done) or set it to some software generated * hash value based on the packet contents. */ ipflags |= IP_NODEFAULTFLOWID; #endif /* RSS */ if (unlock_udbinfo == UH_WLOCKED) INP_HASH_WUNLOCK(pcbinfo); else if (unlock_udbinfo == UH_RLOCKED) INP_HASH_RUNLOCK_ET(pcbinfo, et); UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u); error = ip_output(m, inp->inp_options, (unlock_inp == UH_WLOCKED ? &inp->inp_route : NULL), ipflags, inp->inp_moptions, inp); if (unlock_inp == UH_WLOCKED) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); return (error); release: if (unlock_udbinfo == UH_WLOCKED) { KASSERT(unlock_inp == UH_WLOCKED, ("%s: excl udbinfo lock, shared inp lock", __func__)); INP_HASH_WUNLOCK(pcbinfo); INP_WUNLOCK(inp); } else if (unlock_udbinfo == UH_RLOCKED) { KASSERT(unlock_inp == UH_RLOCKED, ("%s: shared udbinfo lock, excl inp lock", __func__)); INP_HASH_RUNLOCK_ET(pcbinfo, et); INP_RUNLOCK(inp); } else if (unlock_inp == UH_WLOCKED) INP_WUNLOCK(inp); else INP_RUNLOCK(inp); m_freem(m); return (error); } static void udp_abort(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_abort: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); soisdisconnected(so); } INP_WUNLOCK(inp); } static int udp_attach(struct socket *so, int proto, struct thread *td) { static uint32_t udp_flowid; struct inpcb *inp; struct inpcbinfo *pcbinfo; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp == NULL, ("udp_attach: inp != NULL")); error = soreserve(so, udp_sendspace, udp_recvspace); if (error) return (error); INP_INFO_WLOCK(pcbinfo); error = in_pcballoc(so, pcbinfo); if (error) { INP_INFO_WUNLOCK(pcbinfo); return (error); } inp = sotoinpcb(so); inp->inp_vflag |= INP_IPV4; inp->inp_ip_ttl = V_ip_defttl; inp->inp_flowid = atomic_fetchadd_int(&udp_flowid, 1); inp->inp_flowtype = M_HASHTYPE_OPAQUE; error = udp_newudpcb(inp); if (error) { in_pcbdetach(inp); in_pcbfree(inp); INP_INFO_WUNLOCK(pcbinfo); return (error); } INP_WUNLOCK(inp); INP_INFO_WUNLOCK(pcbinfo); return (0); } #endif /* INET */ int udp_set_kernel_tunneling(struct socket *so, udp_tun_func_t f, udp_tun_icmp_t i, void *ctx) { struct inpcb *inp; struct udpcb *up; KASSERT(so->so_type == SOCK_DGRAM, ("udp_set_kernel_tunneling: !dgram")); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_set_kernel_tunneling: inp == NULL")); INP_WLOCK(inp); up = intoudpcb(inp); if ((up->u_tun_func != NULL) || (up->u_icmp_func != NULL)) { INP_WUNLOCK(inp); return (EBUSY); } up->u_tun_func = f; up->u_icmp_func = i; up->u_tun_ctx = ctx; INP_WUNLOCK(inp); return (0); } #ifdef INET static int udp_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { struct inpcb *inp; struct inpcbinfo *pcbinfo; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_bind: inp == NULL")); INP_WLOCK(inp); INP_HASH_WLOCK(pcbinfo); error = in_pcbbind(inp, nam, td->td_ucred); INP_HASH_WUNLOCK(pcbinfo); INP_WUNLOCK(inp); return (error); } static void udp_close(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_close: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); soisdisconnected(so); } INP_WUNLOCK(inp); } static int udp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { struct inpcb *inp; struct inpcbinfo *pcbinfo; struct sockaddr_in *sin; int error; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_connect: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr != INADDR_ANY) { INP_WUNLOCK(inp); return (EISCONN); } sin = (struct sockaddr_in *)nam; error = prison_remote_ip4(td->td_ucred, &sin->sin_addr); if (error != 0) { INP_WUNLOCK(inp); return (error); } INP_HASH_WLOCK(pcbinfo); error = in_pcbconnect(inp, nam, td->td_ucred); INP_HASH_WUNLOCK(pcbinfo); if (error == 0) soisconnected(so); INP_WUNLOCK(inp); return (error); } static void udp_detach(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; struct udpcb *up; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_detach: inp == NULL")); KASSERT(inp->inp_faddr.s_addr == INADDR_ANY, ("udp_detach: not disconnected")); INP_INFO_WLOCK(pcbinfo); INP_WLOCK(inp); up = intoudpcb(inp); KASSERT(up != NULL, ("%s: up == NULL", __func__)); inp->inp_ppcb = NULL; in_pcbdetach(inp); in_pcbfree(inp); INP_INFO_WUNLOCK(pcbinfo); udp_discardcb(up); } static int udp_disconnect(struct socket *so) { struct inpcb *inp; struct inpcbinfo *pcbinfo; pcbinfo = udp_get_inpcbinfo(so->so_proto->pr_protocol); inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_disconnect: inp == NULL")); INP_WLOCK(inp); if (inp->inp_faddr.s_addr == INADDR_ANY) { INP_WUNLOCK(inp); return (ENOTCONN); } INP_HASH_WLOCK(pcbinfo); in_pcbdisconnect(inp); inp->inp_laddr.s_addr = INADDR_ANY; INP_HASH_WUNLOCK(pcbinfo); SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; /* XXX */ SOCK_UNLOCK(so); INP_WUNLOCK(inp); return (0); } static int udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_send: inp == NULL")); return (udp_output(inp, m, addr, control, td)); } #endif /* INET */ int udp_shutdown(struct socket *so) { struct inpcb *inp; inp = sotoinpcb(so); KASSERT(inp != NULL, ("udp_shutdown: inp == NULL")); INP_WLOCK(inp); socantsendmore(so); INP_WUNLOCK(inp); return (0); } #ifdef INET struct pr_usrreqs udp_usrreqs = { .pru_abort = udp_abort, .pru_attach = udp_attach, .pru_bind = udp_bind, .pru_connect = udp_connect, .pru_control = in_control, .pru_detach = udp_detach, .pru_disconnect = udp_disconnect, .pru_peeraddr = in_getpeeraddr, .pru_send = udp_send, .pru_soreceive = soreceive_dgram, .pru_sosend = sosend_dgram, .pru_shutdown = udp_shutdown, .pru_sockaddr = in_getsockaddr, .pru_sosetlabel = in_pcbsosetlabel, .pru_close = udp_close, }; #endif /* INET */ Index: head/sys/netinet6/frag6.c =================================================================== --- head/sys/netinet6/frag6.c (revision 336675) +++ head/sys/netinet6/frag6.c (revision 336676) @@ -1,832 +1,832 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for ECN definitions */ #include /* for ECN definitions */ #include static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *); static void frag6_deq(struct ip6asfrag *); static void frag6_insque(struct ip6q *, struct ip6q *); static void frag6_remque(struct ip6q *); static void frag6_freef(struct ip6q *); static struct mtx ip6qlock; /* * These fields all protected by ip6qlock. */ -static VNET_DEFINE(u_int, frag6_nfragpackets); -static VNET_DEFINE(u_int, frag6_nfrags); -static VNET_DEFINE(struct ip6q, ip6q); /* ip6 reassemble queue */ +VNET_DEFINE_STATIC(u_int, frag6_nfragpackets); +VNET_DEFINE_STATIC(u_int, frag6_nfrags); +VNET_DEFINE_STATIC(struct ip6q, ip6q); /* ip6 reassemble queue */ #define V_frag6_nfragpackets VNET(frag6_nfragpackets) #define V_frag6_nfrags VNET(frag6_nfrags) #define V_ip6q VNET(ip6q) #define IP6Q_LOCK_INIT() mtx_init(&ip6qlock, "ip6qlock", NULL, MTX_DEF); #define IP6Q_LOCK() mtx_lock(&ip6qlock) #define IP6Q_TRYLOCK() mtx_trylock(&ip6qlock) #define IP6Q_LOCK_ASSERT() mtx_assert(&ip6qlock, MA_OWNED) #define IP6Q_UNLOCK() mtx_unlock(&ip6qlock) static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header"); /* * Initialise reassembly queue and fragment identifier. */ static void frag6_change(void *tag) { V_ip6_maxfragpackets = nmbclusters / 4; V_ip6_maxfrags = nmbclusters / 4; } void frag6_init(void) { V_ip6_maxfragpackets = nmbclusters / 4; V_ip6_maxfrags = nmbclusters / 4; V_ip6q.ip6q_next = V_ip6q.ip6q_prev = &V_ip6q; if (!IS_DEFAULT_VNET(curvnet)) return; EVENTHANDLER_REGISTER(nmbclusters_change, frag6_change, NULL, EVENTHANDLER_PRI_ANY); IP6Q_LOCK_INIT(); } /* * In RFC2460, fragment and reassembly rule do not agree with each other, * in terms of next header field handling in fragment header. * While the sender will use the same value for all of the fragmented packets, * receiver is suggested not to check the consistency. * * fragment rule (p20): * (2) A Fragment header containing: * The Next Header value that identifies the first header of * the Fragmentable Part of the original packet. * -> next header field is same for all fragments * * reassembly rule (p21): * The Next Header field of the last header of the Unfragmentable * Part is obtained from the Next Header field of the first * fragment's Fragment header. * -> should grab it from the first fragment only * * The following note also contradicts with fragment rule - no one is going to * send different fragment with different next header field. * * additional note (p22): * The Next Header values in the Fragment headers of different * fragments of the same original packet may differ. Only the value * from the Offset zero fragment packet is used for reassembly. * -> should grab it from the first fragment only * * There is no explicit reason given in the RFC. Historical reason maybe? */ /* * Fragment input */ int frag6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp, *t; struct ip6_hdr *ip6; struct ip6_frag *ip6f; struct ip6q *q6; struct ip6asfrag *af6, *ip6af, *af6dwn; struct in6_ifaddr *ia; int offset = *offp, nxt, i, next; int first_frag = 0; int fragoff, frgpartlen; /* must be larger than u_int16_t */ struct ifnet *dstifp; u_int8_t ecn, ecn0; #ifdef RSS struct m_tag *mtag; struct ip6_direct_ctx *ip6dc; #endif #if 0 char ip6buf[INET6_ADDRSTRLEN]; #endif ip6 = mtod(m, struct ip6_hdr *); #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE); ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); #else IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f)); if (ip6f == NULL) return (IPPROTO_DONE); #endif dstifp = NULL; /* find the destination interface of the packet. */ ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia != NULL) { dstifp = ia->ia_ifp; ifa_free(&ia->ia_ifa); } /* jumbo payload can't contain a fragment header */ if (ip6->ip6_plen == 0) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); in6_ifstat_inc(dstifp, ifs6_reass_fail); return IPPROTO_DONE; } /* * check whether fragment packet's fragment length is * multiple of 8 octets. * sizeof(struct ip6_frag) == 8 * sizeof(struct ip6_hdr) = 40 */ if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offsetof(struct ip6_hdr, ip6_plen)); in6_ifstat_inc(dstifp, ifs6_reass_fail); return IPPROTO_DONE; } IP6STAT_INC(ip6s_fragments); in6_ifstat_inc(dstifp, ifs6_reass_reqd); /* offset now points to data portion */ offset += sizeof(struct ip6_frag); /* * RFC 6946: Handle "atomic" fragments (offset and m bit set to 0) * upfront, unrelated to any reassembly. Just skip the fragment header. */ if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { /* XXX-BZ we want dedicated counters for this. */ IP6STAT_INC(ip6s_reassembled); in6_ifstat_inc(dstifp, ifs6_reass_ok); *offp = offset; m->m_flags |= M_FRAGMENTED; return (ip6f->ip6f_nxt); } IP6Q_LOCK(); /* * Enforce upper bound on number of fragments. * If maxfrag is 0, never accept fragments. * If maxfrag is -1, accept all fragments without limitation. */ if (V_ip6_maxfrags < 0) ; else if (V_frag6_nfrags >= (u_int)V_ip6_maxfrags) goto dropfrag; for (q6 = V_ip6q.ip6q_next; q6 != &V_ip6q; q6 = q6->ip6q_next) if (ip6f->ip6f_ident == q6->ip6q_ident && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) #ifdef MAC && mac_ip6q_match(m, q6) #endif ) break; if (q6 == &V_ip6q) { /* * the first fragment to arrive, create a reassembly queue. */ first_frag = 1; /* * Enforce upper bound on number of fragmented packets * for which we attempt reassembly; * If maxfragpackets is 0, never accept fragments. * If maxfragpackets is -1, accept all fragments without * limitation. */ if (V_ip6_maxfragpackets < 0) ; else if (V_frag6_nfragpackets >= (u_int)V_ip6_maxfragpackets) goto dropfrag; V_frag6_nfragpackets++; q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE, M_NOWAIT); if (q6 == NULL) goto dropfrag; bzero(q6, sizeof(*q6)); #ifdef MAC if (mac_ip6q_init(q6, M_NOWAIT) != 0) { free(q6, M_FTABLE); goto dropfrag; } mac_ip6q_create(m, q6); #endif frag6_insque(q6, &V_ip6q); /* ip6q_nxt will be filled afterwards, from 1st fragment */ q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6; #ifdef notyet q6->ip6q_nxtp = (u_char *)nxtp; #endif q6->ip6q_ident = ip6f->ip6f_ident; q6->ip6q_ttl = IPV6_FRAGTTL; q6->ip6q_src = ip6->ip6_src; q6->ip6q_dst = ip6->ip6_dst; q6->ip6q_ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ q6->ip6q_nfrag = 0; } /* * If it's the 1st fragment, record the length of the * unfragmentable part and the next header of the fragment header. */ fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); if (fragoff == 0) { q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - sizeof(struct ip6_frag); q6->ip6q_nxt = ip6f->ip6f_nxt; } /* * Check that the reassembled packet would not exceed 65535 bytes * in size. * If it would exceed, discard the fragment and return an ICMP error. */ frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; if (q6->ip6q_unfrglen >= 0) { /* The 1st fragment has already arrived. */ if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); IP6Q_UNLOCK(); return (IPPROTO_DONE); } } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); IP6Q_UNLOCK(); return (IPPROTO_DONE); } /* * If it's the first fragment, do the above check for each * fragment already stored in the reassembly queue. */ if (fragoff == 0) { for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6dwn) { af6dwn = af6->ip6af_down; if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen > IPV6_MAXPACKET) { struct mbuf *merr = IP6_REASS_MBUF(af6); struct ip6_hdr *ip6err; int erroff = af6->ip6af_offset; /* dequeue the fragment. */ frag6_deq(af6); free(af6, M_FTABLE); /* adjust pointer. */ ip6err = mtod(merr, struct ip6_hdr *); /* * Restore source and destination addresses * in the erroneous IPv6 header. */ ip6err->ip6_src = q6->ip6q_src; ip6err->ip6_dst = q6->ip6q_dst; icmp6_error(merr, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, erroff - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); } } } ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE, M_NOWAIT); if (ip6af == NULL) goto dropfrag; bzero(ip6af, sizeof(*ip6af)); ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG; ip6af->ip6af_off = fragoff; ip6af->ip6af_frglen = frgpartlen; ip6af->ip6af_offset = offset; IP6_REASS_MBUF(ip6af) = m; if (first_frag) { af6 = (struct ip6asfrag *)q6; goto insert; } /* * Handle ECN by comparing this segment with the first one; * if CE is set, do not lose CE. * drop if CE and not-ECT are mixed for the same packet. */ ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; ecn0 = q6->ip6q_ecn; if (ecn == IPTOS_ECN_CE) { if (ecn0 == IPTOS_ECN_NOTECT) { free(ip6af, M_FTABLE); goto dropfrag; } if (ecn0 != IPTOS_ECN_CE) q6->ip6q_ecn = IPTOS_ECN_CE; } if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { free(ip6af, M_FTABLE); goto dropfrag; } /* * Find a segment which begins after this one does. */ for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6->ip6af_down) if (af6->ip6af_off > ip6af->ip6af_off) break; #if 0 /* * If there is a preceding segment, it may provide some of * our data already. If so, drop the data from the incoming * segment. If it provides all of our data, drop us. */ if (af6->ip6af_up != (struct ip6asfrag *)q6) { i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen - ip6af->ip6af_off; if (i > 0) { if (i >= ip6af->ip6af_frglen) goto dropfrag; m_adj(IP6_REASS_MBUF(ip6af), i); ip6af->ip6af_off += i; ip6af->ip6af_frglen -= i; } } /* * While we overlap succeeding segments trim them or, * if they are completely covered, dequeue them. */ while (af6 != (struct ip6asfrag *)q6 && ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) { i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; if (i < af6->ip6af_frglen) { af6->ip6af_frglen -= i; af6->ip6af_off += i; m_adj(IP6_REASS_MBUF(af6), i); break; } af6 = af6->ip6af_down; m_freem(IP6_REASS_MBUF(af6->ip6af_up)); frag6_deq(af6->ip6af_up); } #else /* * If the incoming framgent overlaps some existing fragments in * the reassembly queue, drop it, since it is dangerous to override * existing fragments from a security point of view. * We don't know which fragment is the bad guy - here we trust * fragment that came in earlier, with no real reason. * * Note: due to changes after disabling this part, mbuf passed to * m_adj() below now does not meet the requirement. */ if (af6->ip6af_up != (struct ip6asfrag *)q6) { i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen - ip6af->ip6af_off; if (i > 0) { #if 0 /* suppress the noisy log */ log(LOG_ERR, "%d bytes of a fragment from %s " "overlaps the previous fragment\n", i, ip6_sprintf(ip6buf, &q6->ip6q_src)); #endif free(ip6af, M_FTABLE); goto dropfrag; } } if (af6 != (struct ip6asfrag *)q6) { i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; if (i > 0) { #if 0 /* suppress the noisy log */ log(LOG_ERR, "%d bytes of a fragment from %s " "overlaps the succeeding fragment", i, ip6_sprintf(ip6buf, &q6->ip6q_src)); #endif free(ip6af, M_FTABLE); goto dropfrag; } } #endif insert: #ifdef MAC if (!first_frag) mac_ip6q_update(m, q6); #endif /* * Stick new segment in its place; * check for complete reassembly. * Move to front of packet queue, as we are * the most recently active fragmented packet. */ frag6_enq(ip6af, af6->ip6af_up); V_frag6_nfrags++; q6->ip6q_nfrag++; #if 0 /* xxx */ if (q6 != V_ip6q.ip6q_next) { frag6_remque(q6); frag6_insque(q6, &V_ip6q); } #endif next = 0; for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6->ip6af_down) { if (af6->ip6af_off != next) { IP6Q_UNLOCK(); return IPPROTO_DONE; } next += af6->ip6af_frglen; } if (af6->ip6af_up->ip6af_mff) { IP6Q_UNLOCK(); return IPPROTO_DONE; } /* * Reassembly is complete; concatenate fragments. */ ip6af = q6->ip6q_down; t = m = IP6_REASS_MBUF(ip6af); af6 = ip6af->ip6af_down; frag6_deq(ip6af); while (af6 != (struct ip6asfrag *)q6) { m->m_pkthdr.csum_flags &= IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags; m->m_pkthdr.csum_data += IP6_REASS_MBUF(af6)->m_pkthdr.csum_data; af6dwn = af6->ip6af_down; frag6_deq(af6); while (t->m_next) t = t->m_next; m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset); m_demote_pkthdr(IP6_REASS_MBUF(af6)); m_cat(t, IP6_REASS_MBUF(af6)); free(af6, M_FTABLE); af6 = af6dwn; } while (m->m_pkthdr.csum_data & 0xffff0000) m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); /* adjust offset to point where the original next header starts */ offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); free(ip6af, M_FTABLE); ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr)); if (q6->ip6q_ecn == IPTOS_ECN_CE) ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); nxt = q6->ip6q_nxt; #ifdef notyet *q6->ip6q_nxtp = (u_char)(nxt & 0xff); #endif if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) { frag6_remque(q6); V_frag6_nfrags -= q6->ip6q_nfrag; #ifdef MAC mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); V_frag6_nfragpackets--; goto dropfrag; } /* * Store NXT to the original. */ m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), (caddr_t)&nxt); frag6_remque(q6); V_frag6_nfrags -= q6->ip6q_nfrag; #ifdef MAC mac_ip6q_reassemble(q6, m); mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); V_frag6_nfragpackets--; if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ int plen = 0; for (t = m; t; t = t->m_next) plen += t->m_len; m->m_pkthdr.len = plen; } #ifdef RSS mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), M_NOWAIT); if (mtag == NULL) goto dropfrag; ip6dc = (struct ip6_direct_ctx *)(mtag + 1); ip6dc->ip6dc_nxt = nxt; ip6dc->ip6dc_off = offset; m_tag_prepend(m, mtag); #endif IP6Q_UNLOCK(); IP6STAT_INC(ip6s_reassembled); in6_ifstat_inc(dstifp, ifs6_reass_ok); #ifdef RSS /* * Queue/dispatch for reprocessing. */ netisr_dispatch(NETISR_IPV6_DIRECT, m); return IPPROTO_DONE; #endif /* * Tell launch routine the next header */ *mp = m; *offp = offset; return nxt; dropfrag: IP6Q_UNLOCK(); in6_ifstat_inc(dstifp, ifs6_reass_fail); IP6STAT_INC(ip6s_fragdropped); m_freem(m); return IPPROTO_DONE; } /* * Free a fragment reassembly header and all * associated datagrams. */ void frag6_freef(struct ip6q *q6) { struct ip6asfrag *af6, *down6; IP6Q_LOCK_ASSERT(); for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = down6) { struct mbuf *m = IP6_REASS_MBUF(af6); down6 = af6->ip6af_down; frag6_deq(af6); /* * Return ICMP time exceeded error for the 1st fragment. * Just free other fragments. */ if (af6->ip6af_off == 0) { struct ip6_hdr *ip6; /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); /* restore source and destination addresses */ ip6->ip6_src = q6->ip6q_src; ip6->ip6_dst = q6->ip6q_dst; icmp6_error(m, ICMP6_TIME_EXCEEDED, ICMP6_TIME_EXCEED_REASSEMBLY, 0); } else m_freem(m); free(af6, M_FTABLE); } frag6_remque(q6); V_frag6_nfrags -= q6->ip6q_nfrag; #ifdef MAC mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); V_frag6_nfragpackets--; } /* * Put an ip fragment on a reassembly chain. * Like insque, but pointers in middle of structure. */ void frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6) { IP6Q_LOCK_ASSERT(); af6->ip6af_up = up6; af6->ip6af_down = up6->ip6af_down; up6->ip6af_down->ip6af_up = af6; up6->ip6af_down = af6; } /* * To frag6_enq as remque is to insque. */ void frag6_deq(struct ip6asfrag *af6) { IP6Q_LOCK_ASSERT(); af6->ip6af_up->ip6af_down = af6->ip6af_down; af6->ip6af_down->ip6af_up = af6->ip6af_up; } void frag6_insque(struct ip6q *new, struct ip6q *old) { IP6Q_LOCK_ASSERT(); new->ip6q_prev = old; new->ip6q_next = old->ip6q_next; old->ip6q_next->ip6q_prev= new; old->ip6q_next = new; } void frag6_remque(struct ip6q *p6) { IP6Q_LOCK_ASSERT(); p6->ip6q_prev->ip6q_next = p6->ip6q_next; p6->ip6q_next->ip6q_prev = p6->ip6q_prev; } /* * IPv6 reassembling timer processing; * if a timer expires on a reassembly * queue, discard it. */ void frag6_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); struct ip6q *q6; VNET_LIST_RLOCK_NOSLEEP(); IP6Q_LOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); q6 = V_ip6q.ip6q_next; if (q6) while (q6 != &V_ip6q) { --q6->ip6q_ttl; q6 = q6->ip6q_next; if (q6->ip6q_prev->ip6q_ttl == 0) { IP6STAT_INC(ip6s_fragtimeout); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ frag6_freef(q6->ip6q_prev); } } /* * If we are over the maximum number of fragments * (due to the limit being lowered), drain off * enough to get down to the new limit. */ while (V_frag6_nfragpackets > (u_int)V_ip6_maxfragpackets && V_ip6q.ip6q_prev) { IP6STAT_INC(ip6s_fragoverflow); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ frag6_freef(V_ip6q.ip6q_prev); } CURVNET_RESTORE(); } IP6Q_UNLOCK(); VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Drain off all datagram fragments. */ void frag6_drain(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); if (IP6Q_TRYLOCK() == 0) { VNET_LIST_RUNLOCK_NOSLEEP(); return; } VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); while (V_ip6q.ip6q_next != &V_ip6q) { IP6STAT_INC(ip6s_fragdropped); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ frag6_freef(V_ip6q.ip6q_next); } CURVNET_RESTORE(); } IP6Q_UNLOCK(); VNET_LIST_RUNLOCK_NOSLEEP(); } int ip6_deletefraghdr(struct mbuf *m, int offset, int wait) { struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct mbuf *t; /* Delete frag6 header. */ if (m->m_len >= offset + sizeof(struct ip6_frag)) { /* This is the only possible case with !PULLDOWN_TEST. */ bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset); m->m_data += sizeof(struct ip6_frag); m->m_len -= sizeof(struct ip6_frag); } else { /* This comes with no copy if the boundary is on cluster. */ if ((t = m_split(m, offset, wait)) == NULL) return (ENOMEM); m_adj(t, sizeof(struct ip6_frag)); m_cat(m, t); } m->m_flags |= M_FRAGMENTED; return (0); } Index: head/sys/netinet6/icmp6.c =================================================================== --- head/sys/netinet6/icmp6.c (revision 336675) +++ head/sys/netinet6/icmp6.c (revision 336676) @@ -1,2819 +1,2819 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: icmp6.c,v 1.211 2001/04/04 05:56:20 itojun Exp $ */ /*- * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_icmp.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #define MBUF_PRIVATE /* XXXRW: Optimisation tries to avoid M_EXT mbufs */ #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include extern struct domain inet6domain; VNET_PCPUSTAT_DEFINE(struct icmp6stat, icmp6stat); VNET_PCPUSTAT_SYSINIT(icmp6stat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(icmp6stat); #endif /* VIMAGE */ VNET_DECLARE(struct inpcbinfo, ripcbinfo); VNET_DECLARE(struct inpcbhead, ripcb); VNET_DECLARE(int, icmp6errppslim); -static VNET_DEFINE(int, icmp6errpps_count) = 0; -static VNET_DEFINE(struct timeval, icmp6errppslim_last); +VNET_DEFINE_STATIC(int, icmp6errpps_count) = 0; +VNET_DEFINE_STATIC(struct timeval, icmp6errppslim_last); VNET_DECLARE(int, icmp6_nodeinfo); #define V_ripcbinfo VNET(ripcbinfo) #define V_ripcb VNET(ripcb) #define V_icmp6errppslim VNET(icmp6errppslim) #define V_icmp6errpps_count VNET(icmp6errpps_count) #define V_icmp6errppslim_last VNET(icmp6errppslim_last) #define V_icmp6_nodeinfo VNET(icmp6_nodeinfo) static void icmp6_errcount(int, int); static int icmp6_rip6_input(struct mbuf **, int); static int icmp6_ratelimit(const struct in6_addr *, const int, const int); static const char *icmp6_redirect_diag(struct in6_addr *, struct in6_addr *, struct in6_addr *); static struct mbuf *ni6_input(struct mbuf *, int); static struct mbuf *ni6_nametodns(const char *, int, int); static int ni6_dnsmatch(const char *, int, const char *, int); static int ni6_addrs(struct icmp6_nodeinfo *, struct mbuf *, struct ifnet **, struct in6_addr *); static int ni6_store_addrs(struct icmp6_nodeinfo *, struct icmp6_nodeinfo *, struct ifnet *, int); static int icmp6_notify_error(struct mbuf **, int, int, int); /* * Kernel module interface for updating icmp6stat. The argument is an index * into icmp6stat treated as an array of u_quad_t. While this encodes the * general layout of icmp6stat into the caller, it doesn't encode its * location, so that future changes to add, for example, per-CPU stats * support won't cause binary compatibility problems for kernel modules. */ void kmod_icmp6stat_inc(int statnum) { counter_u64_add(VNET(icmp6stat)[statnum], 1); } static void icmp6_errcount(int type, int code) { switch (type) { case ICMP6_DST_UNREACH: switch (code) { case ICMP6_DST_UNREACH_NOROUTE: ICMP6STAT_INC(icp6s_odst_unreach_noroute); return; case ICMP6_DST_UNREACH_ADMIN: ICMP6STAT_INC(icp6s_odst_unreach_admin); return; case ICMP6_DST_UNREACH_BEYONDSCOPE: ICMP6STAT_INC(icp6s_odst_unreach_beyondscope); return; case ICMP6_DST_UNREACH_ADDR: ICMP6STAT_INC(icp6s_odst_unreach_addr); return; case ICMP6_DST_UNREACH_NOPORT: ICMP6STAT_INC(icp6s_odst_unreach_noport); return; } break; case ICMP6_PACKET_TOO_BIG: ICMP6STAT_INC(icp6s_opacket_too_big); return; case ICMP6_TIME_EXCEEDED: switch (code) { case ICMP6_TIME_EXCEED_TRANSIT: ICMP6STAT_INC(icp6s_otime_exceed_transit); return; case ICMP6_TIME_EXCEED_REASSEMBLY: ICMP6STAT_INC(icp6s_otime_exceed_reassembly); return; } break; case ICMP6_PARAM_PROB: switch (code) { case ICMP6_PARAMPROB_HEADER: ICMP6STAT_INC(icp6s_oparamprob_header); return; case ICMP6_PARAMPROB_NEXTHEADER: ICMP6STAT_INC(icp6s_oparamprob_nextheader); return; case ICMP6_PARAMPROB_OPTION: ICMP6STAT_INC(icp6s_oparamprob_option); return; } break; case ND_REDIRECT: ICMP6STAT_INC(icp6s_oredirect); return; } ICMP6STAT_INC(icp6s_ounknown); } /* * A wrapper function for icmp6_error() necessary when the erroneous packet * may not contain enough scope zone information. */ void icmp6_error2(struct mbuf *m, int type, int code, int param, struct ifnet *ifp) { struct ip6_hdr *ip6; if (ifp == NULL) return; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, sizeof(struct ip6_hdr), ); #else if (m->m_len < sizeof(struct ip6_hdr)) { m = m_pullup(m, sizeof(struct ip6_hdr)); if (m == NULL) return; } #endif ip6 = mtod(m, struct ip6_hdr *); if (in6_setscope(&ip6->ip6_src, ifp, NULL) != 0) return; if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0) return; icmp6_error(m, type, code, param); } /* * Generate an error packet of type error in response to bad IP6 packet. */ void icmp6_error(struct mbuf *m, int type, int code, int param) { struct ip6_hdr *oip6, *nip6; struct icmp6_hdr *icmp6; u_int preplen; int off; int nxt; ICMP6STAT_INC(icp6s_error); /* count per-type-code statistics */ icmp6_errcount(type, code); #ifdef M_DECRYPTED /*not openbsd*/ if (m->m_flags & M_DECRYPTED) { ICMP6STAT_INC(icp6s_canterror); goto freeit; } #endif #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, sizeof(struct ip6_hdr), ); #else if (m->m_len < sizeof(struct ip6_hdr)) { m = m_pullup(m, sizeof(struct ip6_hdr)); if (m == NULL) return; } #endif oip6 = mtod(m, struct ip6_hdr *); /* * If the destination address of the erroneous packet is a multicast * address, or the packet was sent using link-layer multicast, * we should basically suppress sending an error (RFC 2463, Section * 2.4). * We have two exceptions (the item e.2 in that section): * - the Packet Too Big message can be sent for path MTU discovery. * - the Parameter Problem Message that can be allowed an icmp6 error * in the option type field. This check has been done in * ip6_unknown_opt(), so we can just check the type and code. */ if ((m->m_flags & (M_BCAST|M_MCAST) || IN6_IS_ADDR_MULTICAST(&oip6->ip6_dst)) && (type != ICMP6_PACKET_TOO_BIG && (type != ICMP6_PARAM_PROB || code != ICMP6_PARAMPROB_OPTION))) goto freeit; /* * RFC 2463, 2.4 (e.5): source address check. * XXX: the case of anycast source? */ if (IN6_IS_ADDR_UNSPECIFIED(&oip6->ip6_src) || IN6_IS_ADDR_MULTICAST(&oip6->ip6_src)) goto freeit; /* * If we are about to send ICMPv6 against ICMPv6 error/redirect, * don't do it. */ nxt = -1; off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt); if (off >= 0 && nxt == IPPROTO_ICMPV6) { struct icmp6_hdr *icp; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, off + sizeof(struct icmp6_hdr), ); icp = (struct icmp6_hdr *)(mtod(m, caddr_t) + off); #else IP6_EXTHDR_GET(icp, struct icmp6_hdr *, m, off, sizeof(*icp)); if (icp == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif if (icp->icmp6_type < ICMP6_ECHO_REQUEST || icp->icmp6_type == ND_REDIRECT) { /* * ICMPv6 error * Special case: for redirect (which is * informational) we must not send icmp6 error. */ ICMP6STAT_INC(icp6s_canterror); goto freeit; } else { /* ICMPv6 informational - send the error */ } } else { /* non-ICMPv6 - send the error */ } oip6 = mtod(m, struct ip6_hdr *); /* adjust pointer */ /* Finally, do rate limitation check. */ if (icmp6_ratelimit(&oip6->ip6_src, type, code)) { ICMP6STAT_INC(icp6s_toofreq); goto freeit; } /* * OK, ICMP6 can be generated. */ if (m->m_pkthdr.len >= ICMPV6_PLD_MAXLEN) m_adj(m, ICMPV6_PLD_MAXLEN - m->m_pkthdr.len); preplen = sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr); M_PREPEND(m, preplen, M_NOWAIT); /* FIB is also copied over. */ if (m == NULL) { nd6log((LOG_DEBUG, "ENOBUFS in icmp6_error %d\n", __LINE__)); return; } nip6 = mtod(m, struct ip6_hdr *); nip6->ip6_src = oip6->ip6_src; nip6->ip6_dst = oip6->ip6_dst; in6_clearscope(&oip6->ip6_src); in6_clearscope(&oip6->ip6_dst); icmp6 = (struct icmp6_hdr *)(nip6 + 1); icmp6->icmp6_type = type; icmp6->icmp6_code = code; icmp6->icmp6_pptr = htonl((u_int32_t)param); ICMP6STAT_INC(icp6s_outhist[type]); icmp6_reflect(m, sizeof(struct ip6_hdr)); /* header order: IPv6 - ICMPv6 */ return; freeit: /* * If we can't tell whether or not we can generate ICMP6, free it. */ m_freem(m); } /* * Process a received ICMP6 message. */ int icmp6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp, *n; struct ifnet *ifp; struct ip6_hdr *ip6, *nip6; struct icmp6_hdr *icmp6, *nicmp6; int off = *offp; int icmp6len = m->m_pkthdr.len - *offp; int code, sum, noff; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; int ip6len, error; ifp = m->m_pkthdr.rcvif; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, sizeof(struct icmp6_hdr), IPPROTO_DONE); /* m might change if M_LOOP. So, call mtod after this */ #endif /* * Locate icmp6 structure in mbuf, and check * that not corrupted and of at least minimum length */ ip6 = mtod(m, struct ip6_hdr *); ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); if (icmp6len < sizeof(struct icmp6_hdr)) { ICMP6STAT_INC(icp6s_tooshort); goto freeit; } /* * Check multicast group membership. * Note: SSM filters are not applied for ICMPv6 traffic. */ if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { struct in6_multi *inm; inm = in6m_lookup(ifp, &ip6->ip6_dst); if (inm == NULL) { IP6STAT_INC(ip6s_notmember); in6_ifstat_inc(m->m_pkthdr.rcvif, ifs6_in_discard); goto freeit; } } /* * calculate the checksum */ #ifndef PULLDOWN_TEST icmp6 = (struct icmp6_hdr *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, off, sizeof(*icmp6)); if (icmp6 == NULL) { ICMP6STAT_INC(icp6s_tooshort); return IPPROTO_DONE; } #endif code = icmp6->icmp6_code; if ((sum = in6_cksum(m, IPPROTO_ICMPV6, off, icmp6len)) != 0) { nd6log((LOG_ERR, "ICMP6 checksum error(%d|%x) %s\n", icmp6->icmp6_type, sum, ip6_sprintf(ip6bufs, &ip6->ip6_src))); ICMP6STAT_INC(icp6s_checksum); goto freeit; } ICMP6STAT_INC(icp6s_inhist[icmp6->icmp6_type]); icmp6_ifstat_inc(ifp, ifs6_in_msg); if (icmp6->icmp6_type < ICMP6_INFOMSG_MASK) icmp6_ifstat_inc(ifp, ifs6_in_error); switch (icmp6->icmp6_type) { case ICMP6_DST_UNREACH: icmp6_ifstat_inc(ifp, ifs6_in_dstunreach); switch (code) { case ICMP6_DST_UNREACH_NOROUTE: case ICMP6_DST_UNREACH_ADDR: /* PRC_HOSTDEAD is a DOS */ code = PRC_UNREACH_NET; break; case ICMP6_DST_UNREACH_ADMIN: icmp6_ifstat_inc(ifp, ifs6_in_adminprohib); code = PRC_UNREACH_ADMIN_PROHIB; break; case ICMP6_DST_UNREACH_BEYONDSCOPE: /* I mean "source address was incorrect." */ code = PRC_PARAMPROB; break; case ICMP6_DST_UNREACH_NOPORT: code = PRC_UNREACH_PORT; break; default: goto badcode; } goto deliver; break; case ICMP6_PACKET_TOO_BIG: icmp6_ifstat_inc(ifp, ifs6_in_pkttoobig); /* validation is made in icmp6_mtudisc_update */ code = PRC_MSGSIZE; /* * Updating the path MTU will be done after examining * intermediate extension headers. */ goto deliver; break; case ICMP6_TIME_EXCEEDED: icmp6_ifstat_inc(ifp, ifs6_in_timeexceed); switch (code) { case ICMP6_TIME_EXCEED_TRANSIT: code = PRC_TIMXCEED_INTRANS; break; case ICMP6_TIME_EXCEED_REASSEMBLY: code = PRC_TIMXCEED_REASS; break; default: goto badcode; } goto deliver; break; case ICMP6_PARAM_PROB: icmp6_ifstat_inc(ifp, ifs6_in_paramprob); switch (code) { case ICMP6_PARAMPROB_NEXTHEADER: code = PRC_UNREACH_PROTOCOL; break; case ICMP6_PARAMPROB_HEADER: case ICMP6_PARAMPROB_OPTION: code = PRC_PARAMPROB; break; default: goto badcode; } goto deliver; break; case ICMP6_ECHO_REQUEST: icmp6_ifstat_inc(ifp, ifs6_in_echo); if (code != 0) goto badcode; if ((n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL) { /* Give up remote */ break; } if (!M_WRITABLE(n) || n->m_len < off + sizeof(struct icmp6_hdr)) { struct mbuf *n0 = n; int n0len; CTASSERT(sizeof(*nip6) + sizeof(*nicmp6) <= MHLEN); n = m_gethdr(M_NOWAIT, n0->m_type); if (n == NULL) { /* Give up remote */ m_freem(n0); break; } m_move_pkthdr(n, n0); /* FIB copied. */ n0len = n0->m_pkthdr.len; /* save for use below */ /* * Copy IPv6 and ICMPv6 only. */ nip6 = mtod(n, struct ip6_hdr *); bcopy(ip6, nip6, sizeof(struct ip6_hdr)); nicmp6 = (struct icmp6_hdr *)(nip6 + 1); bcopy(icmp6, nicmp6, sizeof(struct icmp6_hdr)); noff = sizeof(struct ip6_hdr); /* new mbuf contains only ipv6+icmpv6 headers */ n->m_len = noff + sizeof(struct icmp6_hdr); /* * Adjust mbuf. ip6_plen will be adjusted in * ip6_output(). */ m_adj(n0, off + sizeof(struct icmp6_hdr)); /* recalculate complete packet size */ n->m_pkthdr.len = n0len + (noff - off); n->m_next = n0; } else { IP6_EXTHDR_GET(nicmp6, struct icmp6_hdr *, n, off, sizeof(*nicmp6)); noff = off; } if (n) { nicmp6->icmp6_type = ICMP6_ECHO_REPLY; nicmp6->icmp6_code = 0; ICMP6STAT_INC(icp6s_reflect); ICMP6STAT_INC(icp6s_outhist[ICMP6_ECHO_REPLY]); icmp6_reflect(n, noff); } break; case ICMP6_ECHO_REPLY: icmp6_ifstat_inc(ifp, ifs6_in_echoreply); if (code != 0) goto badcode; break; case MLD_LISTENER_QUERY: case MLD_LISTENER_REPORT: case MLD_LISTENER_DONE: case MLDV2_LISTENER_REPORT: /* * Drop MLD traffic which is not link-local, has a hop limit * of greater than 1 hop, or which does not have the * IPv6 HBH Router Alert option. * As IPv6 HBH options are stripped in ip6_input() we must * check an mbuf header flag. * XXX Should we also sanity check that these messages * were directed to a link-local multicast prefix? */ if ((ip6->ip6_hlim != 1) || (m->m_flags & M_RTALERT_MLD) == 0) goto freeit; if (mld_input(m, off, icmp6len) != 0) return (IPPROTO_DONE); /* m stays. */ break; case ICMP6_WRUREQUEST: /* ICMP6_FQDN_QUERY */ { enum { WRU, FQDN } mode; if (!V_icmp6_nodeinfo) break; if (icmp6len == sizeof(struct icmp6_hdr) + 4) mode = WRU; else if (icmp6len >= sizeof(struct icmp6_nodeinfo)) mode = FQDN; else goto badlen; if (mode == FQDN) { #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, sizeof(struct icmp6_nodeinfo), IPPROTO_DONE); #endif n = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (n) n = ni6_input(n, off); /* XXX meaningless if n == NULL */ noff = sizeof(struct ip6_hdr); } else { struct prison *pr; u_char *p; int maxhlen, hlen; /* * XXX: this combination of flags is pointless, * but should we keep this for compatibility? */ if ((V_icmp6_nodeinfo & (ICMP6_NODEINFO_FQDNOK | ICMP6_NODEINFO_TMPADDROK)) != (ICMP6_NODEINFO_FQDNOK | ICMP6_NODEINFO_TMPADDROK)) break; if (code != 0) goto badcode; CTASSERT(sizeof(*nip6) + sizeof(*nicmp6) + 4 <= MHLEN); n = m_gethdr(M_NOWAIT, m->m_type); if (n == NULL) { /* Give up remote */ break; } if (!m_dup_pkthdr(n, m, M_NOWAIT)) { /* * Previous code did a blind M_COPY_PKTHDR * and said "just for rcvif". If true, then * we could tolerate the dup failing (due to * the deep copy of the tag chain). For now * be conservative and just fail. */ m_free(n); n = NULL; break; } maxhlen = M_TRAILINGSPACE(n) - (sizeof(*nip6) + sizeof(*nicmp6) + 4); pr = curthread->td_ucred->cr_prison; mtx_lock(&pr->pr_mtx); hlen = strlen(pr->pr_hostname); if (maxhlen > hlen) maxhlen = hlen; /* * Copy IPv6 and ICMPv6 only. */ nip6 = mtod(n, struct ip6_hdr *); bcopy(ip6, nip6, sizeof(struct ip6_hdr)); nicmp6 = (struct icmp6_hdr *)(nip6 + 1); bcopy(icmp6, nicmp6, sizeof(struct icmp6_hdr)); p = (u_char *)(nicmp6 + 1); bzero(p, 4); /* meaningless TTL */ bcopy(pr->pr_hostname, p + 4, maxhlen); mtx_unlock(&pr->pr_mtx); noff = sizeof(struct ip6_hdr); n->m_pkthdr.len = n->m_len = sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr) + 4 + maxhlen; nicmp6->icmp6_type = ICMP6_WRUREPLY; nicmp6->icmp6_code = 0; } if (n) { ICMP6STAT_INC(icp6s_reflect); ICMP6STAT_INC(icp6s_outhist[ICMP6_WRUREPLY]); icmp6_reflect(n, noff); } break; } case ICMP6_WRUREPLY: if (code != 0) goto badcode; break; case ND_ROUTER_SOLICIT: icmp6_ifstat_inc(ifp, ifs6_in_routersolicit); if (code != 0) goto badcode; if (icmp6len < sizeof(struct nd_router_solicit)) goto badlen; if (send_sendso_input_hook != NULL) { IP6_EXTHDR_CHECK(m, off, icmp6len, IPPROTO_DONE); error = send_sendso_input_hook(m, ifp, SND_IN, ip6len); if (error == 0) { m = NULL; goto freeit; } } n = m_copym(m, 0, M_COPYALL, M_NOWAIT); nd6_rs_input(m, off, icmp6len); m = n; if (m == NULL) goto freeit; break; case ND_ROUTER_ADVERT: icmp6_ifstat_inc(ifp, ifs6_in_routeradvert); if (code != 0) goto badcode; if (icmp6len < sizeof(struct nd_router_advert)) goto badlen; if (send_sendso_input_hook != NULL) { error = send_sendso_input_hook(m, ifp, SND_IN, ip6len); if (error == 0) { m = NULL; goto freeit; } } n = m_copym(m, 0, M_COPYALL, M_NOWAIT); nd6_ra_input(m, off, icmp6len); m = n; if (m == NULL) goto freeit; break; case ND_NEIGHBOR_SOLICIT: icmp6_ifstat_inc(ifp, ifs6_in_neighborsolicit); if (code != 0) goto badcode; if (icmp6len < sizeof(struct nd_neighbor_solicit)) goto badlen; if (send_sendso_input_hook != NULL) { error = send_sendso_input_hook(m, ifp, SND_IN, ip6len); if (error == 0) { m = NULL; goto freeit; } } n = m_copym(m, 0, M_COPYALL, M_NOWAIT); nd6_ns_input(m, off, icmp6len); m = n; if (m == NULL) goto freeit; break; case ND_NEIGHBOR_ADVERT: icmp6_ifstat_inc(ifp, ifs6_in_neighboradvert); if (code != 0) goto badcode; if (icmp6len < sizeof(struct nd_neighbor_advert)) goto badlen; if (send_sendso_input_hook != NULL) { error = send_sendso_input_hook(m, ifp, SND_IN, ip6len); if (error == 0) { m = NULL; goto freeit; } } n = m_copym(m, 0, M_COPYALL, M_NOWAIT); nd6_na_input(m, off, icmp6len); m = n; if (m == NULL) goto freeit; break; case ND_REDIRECT: icmp6_ifstat_inc(ifp, ifs6_in_redirect); if (code != 0) goto badcode; if (icmp6len < sizeof(struct nd_redirect)) goto badlen; if (send_sendso_input_hook != NULL) { error = send_sendso_input_hook(m, ifp, SND_IN, ip6len); if (error == 0) { m = NULL; goto freeit; } } n = m_copym(m, 0, M_COPYALL, M_NOWAIT); icmp6_redirect_input(m, off); m = n; if (m == NULL) goto freeit; break; case ICMP6_ROUTER_RENUMBERING: if (code != ICMP6_ROUTER_RENUMBERING_COMMAND && code != ICMP6_ROUTER_RENUMBERING_RESULT) goto badcode; if (icmp6len < sizeof(struct icmp6_router_renum)) goto badlen; break; default: nd6log((LOG_DEBUG, "icmp6_input: unknown type %d(src=%s, dst=%s, ifid=%d)\n", icmp6->icmp6_type, ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), ifp ? ifp->if_index : 0)); if (icmp6->icmp6_type < ICMP6_ECHO_REQUEST) { /* ICMPv6 error: MUST deliver it by spec... */ code = PRC_NCMDS; /* deliver */ } else { /* ICMPv6 informational: MUST not deliver */ break; } deliver: if (icmp6_notify_error(&m, off, icmp6len, code) != 0) { /* In this case, m should've been freed. */ return (IPPROTO_DONE); } break; badcode: ICMP6STAT_INC(icp6s_badcode); break; badlen: ICMP6STAT_INC(icp6s_badlen); break; } /* deliver the packet to appropriate sockets */ icmp6_rip6_input(&m, *offp); return IPPROTO_DONE; freeit: m_freem(m); return IPPROTO_DONE; } static int icmp6_notify_error(struct mbuf **mp, int off, int icmp6len, int code) { struct mbuf *m = *mp; struct icmp6_hdr *icmp6; struct ip6_hdr *eip6; u_int32_t notifymtu; struct sockaddr_in6 icmp6src, icmp6dst; if (icmp6len < sizeof(struct icmp6_hdr) + sizeof(struct ip6_hdr)) { ICMP6STAT_INC(icp6s_tooshort); goto freeit; } #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, sizeof(struct icmp6_hdr) + sizeof(struct ip6_hdr), -1); icmp6 = (struct icmp6_hdr *)(mtod(m, caddr_t) + off); #else IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, off, sizeof(*icmp6) + sizeof(struct ip6_hdr)); if (icmp6 == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif eip6 = (struct ip6_hdr *)(icmp6 + 1); /* Detect the upper level protocol */ { void (*ctlfunc)(int, struct sockaddr *, void *); u_int8_t nxt = eip6->ip6_nxt; int eoff = off + sizeof(struct icmp6_hdr) + sizeof(struct ip6_hdr); struct ip6ctlparam ip6cp; struct in6_addr *finaldst = NULL; int icmp6type = icmp6->icmp6_type; struct ip6_frag *fh; struct ip6_rthdr *rth; struct ip6_rthdr0 *rth0; int rthlen; while (1) { /* XXX: should avoid infinite loop explicitly? */ struct ip6_ext *eh; switch (nxt) { case IPPROTO_HOPOPTS: case IPPROTO_DSTOPTS: case IPPROTO_AH: #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, eoff + sizeof(struct ip6_ext), -1); eh = (struct ip6_ext *)(mtod(m, caddr_t) + eoff); #else IP6_EXTHDR_GET(eh, struct ip6_ext *, m, eoff, sizeof(*eh)); if (eh == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif if (nxt == IPPROTO_AH) eoff += (eh->ip6e_len + 2) << 2; else eoff += (eh->ip6e_len + 1) << 3; nxt = eh->ip6e_nxt; break; case IPPROTO_ROUTING: /* * When the erroneous packet contains a * routing header, we should examine the * header to determine the final destination. * Otherwise, we can't properly update * information that depends on the final * destination (e.g. path MTU). */ #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, eoff + sizeof(*rth), -1); rth = (struct ip6_rthdr *) (mtod(m, caddr_t) + eoff); #else IP6_EXTHDR_GET(rth, struct ip6_rthdr *, m, eoff, sizeof(*rth)); if (rth == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif rthlen = (rth->ip6r_len + 1) << 3; /* * XXX: currently there is no * officially defined type other * than type-0. * Note that if the segment left field * is 0, all intermediate hops must * have been passed. */ if (rth->ip6r_segleft && rth->ip6r_type == IPV6_RTHDR_TYPE_0) { int hops; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, eoff + rthlen, -1); rth0 = (struct ip6_rthdr0 *) (mtod(m, caddr_t) + eoff); #else IP6_EXTHDR_GET(rth0, struct ip6_rthdr0 *, m, eoff, rthlen); if (rth0 == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif /* just ignore a bogus header */ if ((rth0->ip6r0_len % 2) == 0 && (hops = rth0->ip6r0_len/2)) finaldst = (struct in6_addr *)(rth0 + 1) + (hops - 1); } eoff += rthlen; nxt = rth->ip6r_nxt; break; case IPPROTO_FRAGMENT: #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, 0, eoff + sizeof(struct ip6_frag), -1); fh = (struct ip6_frag *)(mtod(m, caddr_t) + eoff); #else IP6_EXTHDR_GET(fh, struct ip6_frag *, m, eoff, sizeof(*fh)); if (fh == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif /* * Data after a fragment header is meaningless * unless it is the first fragment, but * we'll go to the notify label for path MTU * discovery. */ if (fh->ip6f_offlg & IP6F_OFF_MASK) goto notify; eoff += sizeof(struct ip6_frag); nxt = fh->ip6f_nxt; break; default: /* * This case includes ESP and the No Next * Header. In such cases going to the notify * label does not have any meaning * (i.e. ctlfunc will be NULL), but we go * anyway since we might have to update * path MTU information. */ goto notify; } } notify: #ifndef PULLDOWN_TEST icmp6 = (struct icmp6_hdr *)(mtod(m, caddr_t) + off); #else IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, off, sizeof(*icmp6) + sizeof(struct ip6_hdr)); if (icmp6 == NULL) { ICMP6STAT_INC(icp6s_tooshort); return (-1); } #endif /* * retrieve parameters from the inner IPv6 header, and convert * them into sockaddr structures. * XXX: there is no guarantee that the source or destination * addresses of the inner packet are in the same scope as * the addresses of the icmp packet. But there is no other * way to determine the zone. */ eip6 = (struct ip6_hdr *)(icmp6 + 1); bzero(&icmp6dst, sizeof(icmp6dst)); icmp6dst.sin6_len = sizeof(struct sockaddr_in6); icmp6dst.sin6_family = AF_INET6; if (finaldst == NULL) icmp6dst.sin6_addr = eip6->ip6_dst; else icmp6dst.sin6_addr = *finaldst; if (in6_setscope(&icmp6dst.sin6_addr, m->m_pkthdr.rcvif, NULL)) goto freeit; bzero(&icmp6src, sizeof(icmp6src)); icmp6src.sin6_len = sizeof(struct sockaddr_in6); icmp6src.sin6_family = AF_INET6; icmp6src.sin6_addr = eip6->ip6_src; if (in6_setscope(&icmp6src.sin6_addr, m->m_pkthdr.rcvif, NULL)) goto freeit; icmp6src.sin6_flowinfo = (eip6->ip6_flow & IPV6_FLOWLABEL_MASK); if (finaldst == NULL) finaldst = &eip6->ip6_dst; ip6cp.ip6c_m = m; ip6cp.ip6c_icmp6 = icmp6; ip6cp.ip6c_ip6 = (struct ip6_hdr *)(icmp6 + 1); ip6cp.ip6c_off = eoff; ip6cp.ip6c_finaldst = finaldst; ip6cp.ip6c_src = &icmp6src; ip6cp.ip6c_nxt = nxt; if (icmp6type == ICMP6_PACKET_TOO_BIG) { notifymtu = ntohl(icmp6->icmp6_mtu); ip6cp.ip6c_cmdarg = (void *)¬ifymtu; icmp6_mtudisc_update(&ip6cp, 1); /*XXX*/ } ctlfunc = (void (*)(int, struct sockaddr *, void *)) (inet6sw[ip6_protox[nxt]].pr_ctlinput); if (ctlfunc) { (void) (*ctlfunc)(code, (struct sockaddr *)&icmp6dst, &ip6cp); } } *mp = m; return (0); freeit: m_freem(m); return (-1); } void icmp6_mtudisc_update(struct ip6ctlparam *ip6cp, int validated) { struct in6_addr *dst = ip6cp->ip6c_finaldst; struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6; struct mbuf *m = ip6cp->ip6c_m; /* will be necessary for scope issue */ u_int mtu = ntohl(icmp6->icmp6_mtu); struct in_conninfo inc; #if 0 /* * RFC2460 section 5, last paragraph. * even though minimum link MTU for IPv6 is IPV6_MMTU, * we may see ICMPv6 too big with mtu < IPV6_MMTU * due to packet translator in the middle. * see ip6_output() and ip6_getpmtu() "alwaysfrag" case for * special handling. */ if (mtu < IPV6_MMTU) return; #endif /* * we reject ICMPv6 too big with abnormally small value. * XXX what is the good definition of "abnormally small"? */ if (mtu < sizeof(struct ip6_hdr) + sizeof(struct ip6_frag) + 8) return; if (!validated) return; /* * In case the suggested mtu is less than IPV6_MMTU, we * only need to remember that it was for above mentioned * "alwaysfrag" case. * Try to be as close to the spec as possible. */ if (mtu < IPV6_MMTU) mtu = IPV6_MMTU - 8; bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL)) return; if (mtu < tcp_maxmtu6(&inc, NULL)) { tcp_hc_updatemtu(&inc, mtu); ICMP6STAT_INC(icp6s_pmtuchg); } } /* * Process a Node Information Query packet, based on * draft-ietf-ipngwg-icmp-name-lookups-07. * * Spec incompatibilities: * - IPv6 Subject address handling * - IPv4 Subject address handling support missing * - Proxy reply (answer even if it's not for me) * - joins NI group address at in6_ifattach() time only, does not cope * with hostname changes by sethostname(3) */ static struct mbuf * ni6_input(struct mbuf *m, int off) { struct icmp6_nodeinfo *ni6, *nni6; struct mbuf *n = NULL; struct prison *pr; u_int16_t qtype; int subjlen; int replylen = sizeof(struct ip6_hdr) + sizeof(struct icmp6_nodeinfo); struct ni_reply_fqdn *fqdn; int addrs; /* for NI_QTYPE_NODEADDR */ struct ifnet *ifp = NULL; /* for NI_QTYPE_NODEADDR */ struct in6_addr in6_subj; /* subject address */ struct ip6_hdr *ip6; int oldfqdn = 0; /* if 1, return pascal string (03 draft) */ char *subj = NULL; struct in6_ifaddr *ia6 = NULL; ip6 = mtod(m, struct ip6_hdr *); #ifndef PULLDOWN_TEST ni6 = (struct icmp6_nodeinfo *)(mtod(m, caddr_t) + off); #else IP6_EXTHDR_GET(ni6, struct icmp6_nodeinfo *, m, off, sizeof(*ni6)); if (ni6 == NULL) { /* m is already reclaimed */ return (NULL); } #endif /* * Validate IPv6 source address. * The default configuration MUST be to refuse answering queries from * global-scope addresses according to RFC4602. * Notes: * - it's not very clear what "refuse" means; this implementation * simply drops it. * - it's not very easy to identify global-scope (unicast) addresses * since there are many prefixes for them. It should be safer * and in practice sufficient to check "all" but loopback and * link-local (note that site-local unicast was deprecated and * ULA is defined as global scope-wise) */ if ((V_icmp6_nodeinfo & ICMP6_NODEINFO_GLOBALOK) == 0 && !IN6_IS_ADDR_LOOPBACK(&ip6->ip6_src) && !IN6_IS_ADDR_LINKLOCAL(&ip6->ip6_src)) goto bad; /* * Validate IPv6 destination address. * * The Responder must discard the Query without further processing * unless it is one of the Responder's unicast or anycast addresses, or * a link-local scope multicast address which the Responder has joined. * [RFC4602, Section 5.] */ if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { if (!IN6_IS_ADDR_MC_LINKLOCAL(&ip6->ip6_dst)) goto bad; /* else it's a link-local multicast, fine */ } else { /* unicast or anycast */ ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia6 == NULL) goto bad; /* XXX impossible */ if ((ia6->ia6_flags & IN6_IFF_TEMPORARY) && !(V_icmp6_nodeinfo & ICMP6_NODEINFO_TMPADDROK)) { ifa_free(&ia6->ia_ifa); nd6log((LOG_DEBUG, "ni6_input: ignore node info to " "a temporary address in %s:%d", __FILE__, __LINE__)); goto bad; } ifa_free(&ia6->ia_ifa); } /* validate query Subject field. */ qtype = ntohs(ni6->ni_qtype); subjlen = m->m_pkthdr.len - off - sizeof(struct icmp6_nodeinfo); switch (qtype) { case NI_QTYPE_NOOP: case NI_QTYPE_SUPTYPES: /* 07 draft */ if (ni6->ni_code == ICMP6_NI_SUBJ_FQDN && subjlen == 0) break; /* FALLTHROUGH */ case NI_QTYPE_FQDN: case NI_QTYPE_NODEADDR: case NI_QTYPE_IPV4ADDR: switch (ni6->ni_code) { case ICMP6_NI_SUBJ_IPV6: #if ICMP6_NI_SUBJ_IPV6 != 0 case 0: #endif /* * backward compatibility - try to accept 03 draft * format, where no Subject is present. */ if (qtype == NI_QTYPE_FQDN && ni6->ni_code == 0 && subjlen == 0) { oldfqdn++; break; } #if ICMP6_NI_SUBJ_IPV6 != 0 if (ni6->ni_code != ICMP6_NI_SUBJ_IPV6) goto bad; #endif if (subjlen != sizeof(struct in6_addr)) goto bad; /* * Validate Subject address. * * Not sure what exactly "address belongs to the node" * means in the spec, is it just unicast, or what? * * At this moment we consider Subject address as * "belong to the node" if the Subject address equals * to the IPv6 destination address; validation for * IPv6 destination address should have done enough * check for us. * * We do not do proxy at this moment. */ /* m_pulldown instead of copy? */ m_copydata(m, off + sizeof(struct icmp6_nodeinfo), subjlen, (caddr_t)&in6_subj); if (in6_setscope(&in6_subj, m->m_pkthdr.rcvif, NULL)) goto bad; subj = (char *)&in6_subj; if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &in6_subj)) break; /* * XXX if we are to allow other cases, we should really * be careful about scope here. * basically, we should disallow queries toward IPv6 * destination X with subject Y, * if scope(X) > scope(Y). * if we allow scope(X) > scope(Y), it will result in * information leakage across scope boundary. */ goto bad; case ICMP6_NI_SUBJ_FQDN: /* * Validate Subject name with gethostname(3). * * The behavior may need some debate, since: * - we are not sure if the node has FQDN as * hostname (returned by gethostname(3)). * - the code does wildcard match for truncated names. * however, we are not sure if we want to perform * wildcard match, if gethostname(3) side has * truncated hostname. */ pr = curthread->td_ucred->cr_prison; mtx_lock(&pr->pr_mtx); n = ni6_nametodns(pr->pr_hostname, strlen(pr->pr_hostname), 0); mtx_unlock(&pr->pr_mtx); if (!n || n->m_next || n->m_len == 0) goto bad; IP6_EXTHDR_GET(subj, char *, m, off + sizeof(struct icmp6_nodeinfo), subjlen); if (subj == NULL) goto bad; if (!ni6_dnsmatch(subj, subjlen, mtod(n, const char *), n->m_len)) { goto bad; } m_freem(n); n = NULL; break; case ICMP6_NI_SUBJ_IPV4: /* XXX: to be implemented? */ default: goto bad; } break; } /* refuse based on configuration. XXX ICMP6_NI_REFUSED? */ switch (qtype) { case NI_QTYPE_FQDN: if ((V_icmp6_nodeinfo & ICMP6_NODEINFO_FQDNOK) == 0) goto bad; break; case NI_QTYPE_NODEADDR: case NI_QTYPE_IPV4ADDR: if ((V_icmp6_nodeinfo & ICMP6_NODEINFO_NODEADDROK) == 0) goto bad; break; } /* guess reply length */ switch (qtype) { case NI_QTYPE_NOOP: break; /* no reply data */ case NI_QTYPE_SUPTYPES: replylen += sizeof(u_int32_t); break; case NI_QTYPE_FQDN: /* XXX will append an mbuf */ replylen += offsetof(struct ni_reply_fqdn, ni_fqdn_namelen); break; case NI_QTYPE_NODEADDR: addrs = ni6_addrs(ni6, m, &ifp, (struct in6_addr *)subj); if ((replylen += addrs * (sizeof(struct in6_addr) + sizeof(u_int32_t))) > MCLBYTES) replylen = MCLBYTES; /* XXX: will truncate pkt later */ break; case NI_QTYPE_IPV4ADDR: /* unsupported - should respond with unknown Qtype? */ break; default: /* * XXX: We must return a reply with the ICMP6 code * `unknown Qtype' in this case. However we regard the case * as an FQDN query for backward compatibility. * Older versions set a random value to this field, * so it rarely varies in the defined qtypes. * But the mechanism is not reliable... * maybe we should obsolete older versions. */ qtype = NI_QTYPE_FQDN; /* XXX will append an mbuf */ replylen += offsetof(struct ni_reply_fqdn, ni_fqdn_namelen); oldfqdn++; break; } /* Allocate an mbuf to reply. */ if (replylen > MCLBYTES) { /* * XXX: should we try to allocate more? But MCLBYTES * is probably much larger than IPV6_MMTU... */ goto bad; } if (replylen > MHLEN) n = m_getcl(M_NOWAIT, m->m_type, M_PKTHDR); else n = m_gethdr(M_NOWAIT, m->m_type); if (n == NULL) { m_freem(m); return (NULL); } m_move_pkthdr(n, m); /* just for recvif and FIB */ n->m_pkthdr.len = n->m_len = replylen; /* copy mbuf header and IPv6 + Node Information base headers */ bcopy(mtod(m, caddr_t), mtod(n, caddr_t), sizeof(struct ip6_hdr)); nni6 = (struct icmp6_nodeinfo *)(mtod(n, struct ip6_hdr *) + 1); bcopy((caddr_t)ni6, (caddr_t)nni6, sizeof(struct icmp6_nodeinfo)); /* qtype dependent procedure */ switch (qtype) { case NI_QTYPE_NOOP: nni6->ni_code = ICMP6_NI_SUCCESS; nni6->ni_flags = 0; break; case NI_QTYPE_SUPTYPES: { u_int32_t v; nni6->ni_code = ICMP6_NI_SUCCESS; nni6->ni_flags = htons(0x0000); /* raw bitmap */ /* supports NOOP, SUPTYPES, FQDN, and NODEADDR */ v = (u_int32_t)htonl(0x0000000f); bcopy(&v, nni6 + 1, sizeof(u_int32_t)); break; } case NI_QTYPE_FQDN: nni6->ni_code = ICMP6_NI_SUCCESS; fqdn = (struct ni_reply_fqdn *)(mtod(n, caddr_t) + sizeof(struct ip6_hdr) + sizeof(struct icmp6_nodeinfo)); nni6->ni_flags = 0; /* XXX: meaningless TTL */ fqdn->ni_fqdn_ttl = 0; /* ditto. */ /* * XXX do we really have FQDN in hostname? */ pr = curthread->td_ucred->cr_prison; mtx_lock(&pr->pr_mtx); n->m_next = ni6_nametodns(pr->pr_hostname, strlen(pr->pr_hostname), oldfqdn); mtx_unlock(&pr->pr_mtx); if (n->m_next == NULL) goto bad; /* XXX we assume that n->m_next is not a chain */ if (n->m_next->m_next != NULL) goto bad; n->m_pkthdr.len += n->m_next->m_len; break; case NI_QTYPE_NODEADDR: { int lenlim, copied; nni6->ni_code = ICMP6_NI_SUCCESS; n->m_pkthdr.len = n->m_len = sizeof(struct ip6_hdr) + sizeof(struct icmp6_nodeinfo); lenlim = M_TRAILINGSPACE(n); copied = ni6_store_addrs(ni6, nni6, ifp, lenlim); /* XXX: reset mbuf length */ n->m_pkthdr.len = n->m_len = sizeof(struct ip6_hdr) + sizeof(struct icmp6_nodeinfo) + copied; break; } default: break; /* XXX impossible! */ } nni6->ni_type = ICMP6_NI_REPLY; m_freem(m); return (n); bad: m_freem(m); if (n) m_freem(n); return (NULL); } /* * make a mbuf with DNS-encoded string. no compression support. * * XXX names with less than 2 dots (like "foo" or "foo.section") will be * treated as truncated name (two \0 at the end). this is a wild guess. * * old - return pascal string if non-zero */ static struct mbuf * ni6_nametodns(const char *name, int namelen, int old) { struct mbuf *m; char *cp, *ep; const char *p, *q; int i, len, nterm; if (old) len = namelen + 1; else len = MCLBYTES; /* Because MAXHOSTNAMELEN is usually 256, we use cluster mbuf. */ if (len > MLEN) m = m_getcl(M_NOWAIT, MT_DATA, 0); else m = m_get(M_NOWAIT, MT_DATA); if (m == NULL) goto fail; if (old) { m->m_len = len; *mtod(m, char *) = namelen; bcopy(name, mtod(m, char *) + 1, namelen); return m; } else { m->m_len = 0; cp = mtod(m, char *); ep = mtod(m, char *) + M_TRAILINGSPACE(m); /* if not certain about my name, return empty buffer */ if (namelen == 0) return m; /* * guess if it looks like shortened hostname, or FQDN. * shortened hostname needs two trailing "\0". */ i = 0; for (p = name; p < name + namelen; p++) { if (*p && *p == '.') i++; } if (i < 2) nterm = 2; else nterm = 1; p = name; while (cp < ep && p < name + namelen) { i = 0; for (q = p; q < name + namelen && *q && *q != '.'; q++) i++; /* result does not fit into mbuf */ if (cp + i + 1 >= ep) goto fail; /* * DNS label length restriction, RFC1035 page 8. * "i == 0" case is included here to avoid returning * 0-length label on "foo..bar". */ if (i <= 0 || i >= 64) goto fail; *cp++ = i; bcopy(p, cp, i); cp += i; p = q; if (p < name + namelen && *p == '.') p++; } /* termination */ if (cp + nterm >= ep) goto fail; while (nterm-- > 0) *cp++ = '\0'; m->m_len = cp - mtod(m, char *); return m; } panic("should not reach here"); /* NOTREACHED */ fail: if (m) m_freem(m); return NULL; } /* * check if two DNS-encoded string matches. takes care of truncated * form (with \0\0 at the end). no compression support. * XXX upper/lowercase match (see RFC2065) */ static int ni6_dnsmatch(const char *a, int alen, const char *b, int blen) { const char *a0, *b0; int l; /* simplest case - need validation? */ if (alen == blen && bcmp(a, b, alen) == 0) return 1; a0 = a; b0 = b; /* termination is mandatory */ if (alen < 2 || blen < 2) return 0; if (a0[alen - 1] != '\0' || b0[blen - 1] != '\0') return 0; alen--; blen--; while (a - a0 < alen && b - b0 < blen) { if (a - a0 + 1 > alen || b - b0 + 1 > blen) return 0; if ((signed char)a[0] < 0 || (signed char)b[0] < 0) return 0; /* we don't support compression yet */ if (a[0] >= 64 || b[0] >= 64) return 0; /* truncated case */ if (a[0] == 0 && a - a0 == alen - 1) return 1; if (b[0] == 0 && b - b0 == blen - 1) return 1; if (a[0] == 0 || b[0] == 0) return 0; if (a[0] != b[0]) return 0; l = a[0]; if (a - a0 + 1 + l > alen || b - b0 + 1 + l > blen) return 0; if (bcmp(a + 1, b + 1, l) != 0) return 0; a += 1 + l; b += 1 + l; } if (a - a0 == alen && b - b0 == blen) return 1; else return 0; } /* * calculate the number of addresses to be returned in the node info reply. */ static int ni6_addrs(struct icmp6_nodeinfo *ni6, struct mbuf *m, struct ifnet **ifpp, struct in6_addr *subj) { struct ifnet *ifp; struct in6_ifaddr *ifa6; struct ifaddr *ifa; int addrs = 0, addrsofif, iffound = 0; int niflags = ni6->ni_flags; if ((niflags & NI_NODEADDR_FLAG_ALL) == 0) { switch (ni6->ni_code) { case ICMP6_NI_SUBJ_IPV6: if (subj == NULL) /* must be impossible... */ return (0); break; default: /* * XXX: we only support IPv6 subject address for * this Qtype. */ return (0); } } IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { addrsofif = 0; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; ifa6 = (struct in6_ifaddr *)ifa; if ((niflags & NI_NODEADDR_FLAG_ALL) == 0 && IN6_ARE_ADDR_EQUAL(subj, &ifa6->ia_addr.sin6_addr)) iffound = 1; /* * IPv4-mapped addresses can only be returned by a * Node Information proxy, since they represent * addresses of IPv4-only nodes, which perforce do * not implement this protocol. * [icmp-name-lookups-07, Section 5.4] * So we don't support NI_NODEADDR_FLAG_COMPAT in * this function at this moment. */ /* What do we have to do about ::1? */ switch (in6_addrscope(&ifa6->ia_addr.sin6_addr)) { case IPV6_ADDR_SCOPE_LINKLOCAL: if ((niflags & NI_NODEADDR_FLAG_LINKLOCAL) == 0) continue; break; case IPV6_ADDR_SCOPE_SITELOCAL: if ((niflags & NI_NODEADDR_FLAG_SITELOCAL) == 0) continue; break; case IPV6_ADDR_SCOPE_GLOBAL: if ((niflags & NI_NODEADDR_FLAG_GLOBAL) == 0) continue; break; default: continue; } /* * check if anycast is okay. * XXX: just experimental. not in the spec. */ if ((ifa6->ia6_flags & IN6_IFF_ANYCAST) != 0 && (niflags & NI_NODEADDR_FLAG_ANYCAST) == 0) continue; /* we need only unicast addresses */ if ((ifa6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && (V_icmp6_nodeinfo & ICMP6_NODEINFO_TMPADDROK) == 0) { continue; } addrsofif++; /* count the address */ } IF_ADDR_RUNLOCK(ifp); if (iffound) { *ifpp = ifp; IFNET_RUNLOCK_NOSLEEP(); return (addrsofif); } addrs += addrsofif; } IFNET_RUNLOCK_NOSLEEP(); return (addrs); } static int ni6_store_addrs(struct icmp6_nodeinfo *ni6, struct icmp6_nodeinfo *nni6, struct ifnet *ifp0, int resid) { struct ifnet *ifp; struct in6_ifaddr *ifa6; struct ifaddr *ifa; struct ifnet *ifp_dep = NULL; int copied = 0, allow_deprecated = 0; u_char *cp = (u_char *)(nni6 + 1); int niflags = ni6->ni_flags; u_int32_t ltime; if (ifp0 == NULL && !(niflags & NI_NODEADDR_FLAG_ALL)) return (0); /* needless to copy */ IFNET_RLOCK_NOSLEEP(); ifp = ifp0 ? ifp0 : CK_STAILQ_FIRST(&V_ifnet); again: for (; ifp; ifp = CK_STAILQ_NEXT(ifp, if_link)) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; ifa6 = (struct in6_ifaddr *)ifa; if ((ifa6->ia6_flags & IN6_IFF_DEPRECATED) != 0 && allow_deprecated == 0) { /* * prefererred address should be put before * deprecated addresses. */ /* record the interface for later search */ if (ifp_dep == NULL) ifp_dep = ifp; continue; } else if ((ifa6->ia6_flags & IN6_IFF_DEPRECATED) == 0 && allow_deprecated != 0) continue; /* we now collect deprecated addrs */ /* What do we have to do about ::1? */ switch (in6_addrscope(&ifa6->ia_addr.sin6_addr)) { case IPV6_ADDR_SCOPE_LINKLOCAL: if ((niflags & NI_NODEADDR_FLAG_LINKLOCAL) == 0) continue; break; case IPV6_ADDR_SCOPE_SITELOCAL: if ((niflags & NI_NODEADDR_FLAG_SITELOCAL) == 0) continue; break; case IPV6_ADDR_SCOPE_GLOBAL: if ((niflags & NI_NODEADDR_FLAG_GLOBAL) == 0) continue; break; default: continue; } /* * check if anycast is okay. * XXX: just experimental. not in the spec. */ if ((ifa6->ia6_flags & IN6_IFF_ANYCAST) != 0 && (niflags & NI_NODEADDR_FLAG_ANYCAST) == 0) continue; if ((ifa6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && (V_icmp6_nodeinfo & ICMP6_NODEINFO_TMPADDROK) == 0) { continue; } /* now we can copy the address */ if (resid < sizeof(struct in6_addr) + sizeof(u_int32_t)) { IF_ADDR_RUNLOCK(ifp); /* * We give up much more copy. * Set the truncate flag and return. */ nni6->ni_flags |= NI_NODEADDR_FLAG_TRUNCATE; IFNET_RUNLOCK_NOSLEEP(); return (copied); } /* * Set the TTL of the address. * The TTL value should be one of the following * according to the specification: * * 1. The remaining lifetime of a DHCP lease on the * address, or * 2. The remaining Valid Lifetime of a prefix from * which the address was derived through Stateless * Autoconfiguration. * * Note that we currently do not support stateful * address configuration by DHCPv6, so the former * case can't happen. */ if (ifa6->ia6_lifetime.ia6t_expire == 0) ltime = ND6_INFINITE_LIFETIME; else { if (ifa6->ia6_lifetime.ia6t_expire > time_uptime) ltime = htonl(ifa6->ia6_lifetime.ia6t_expire - time_uptime); else ltime = 0; } bcopy(<ime, cp, sizeof(u_int32_t)); cp += sizeof(u_int32_t); /* copy the address itself */ bcopy(&ifa6->ia_addr.sin6_addr, cp, sizeof(struct in6_addr)); in6_clearscope((struct in6_addr *)cp); /* XXX */ cp += sizeof(struct in6_addr); resid -= (sizeof(struct in6_addr) + sizeof(u_int32_t)); copied += (sizeof(struct in6_addr) + sizeof(u_int32_t)); } IF_ADDR_RUNLOCK(ifp); if (ifp0) /* we need search only on the specified IF */ break; } if (allow_deprecated == 0 && ifp_dep != NULL) { ifp = ifp_dep; allow_deprecated = 1; goto again; } IFNET_RUNLOCK_NOSLEEP(); return (copied); } /* * XXX almost dup'ed code with rip6_input. */ static int icmp6_rip6_input(struct mbuf **mp, int off) { struct mbuf *m = *mp; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct inpcb *in6p; struct inpcb *last = NULL; struct sockaddr_in6 fromsa; struct icmp6_hdr *icmp6; struct epoch_tracker et; struct mbuf *opts = NULL; #ifndef PULLDOWN_TEST /* this is assumed to be safe. */ icmp6 = (struct icmp6_hdr *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, off, sizeof(*icmp6)); if (icmp6 == NULL) { /* m is already reclaimed */ return (IPPROTO_DONE); } #endif /* * XXX: the address may have embedded scope zone ID, which should be * hidden from applications. */ bzero(&fromsa, sizeof(fromsa)); fromsa.sin6_family = AF_INET6; fromsa.sin6_len = sizeof(struct sockaddr_in6); fromsa.sin6_addr = ip6->ip6_src; if (sa6_recoverscope(&fromsa)) { m_freem(m); return (IPPROTO_DONE); } INP_INFO_RLOCK_ET(&V_ripcbinfo, et); CK_LIST_FOREACH(in6p, &V_ripcb, inp_list) { if ((in6p->inp_vflag & INP_IPV6) == 0) continue; if (in6p->inp_ip_p != IPPROTO_ICMPV6) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_laddr) && !IN6_ARE_ADDR_EQUAL(&in6p->in6p_laddr, &ip6->ip6_dst)) continue; if (!IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr) && !IN6_ARE_ADDR_EQUAL(&in6p->in6p_faddr, &ip6->ip6_src)) continue; INP_RLOCK(in6p); if (ICMP6_FILTER_WILLBLOCK(icmp6->icmp6_type, in6p->in6p_icmp6filt)) { INP_RUNLOCK(in6p); continue; } if (last != NULL) { struct mbuf *n = NULL; /* * Recent network drivers tend to allocate a single * mbuf cluster, rather than to make a couple of * mbufs without clusters. Also, since the IPv6 code * path tries to avoid m_pullup(), it is highly * probable that we still have an mbuf cluster here * even though the necessary length can be stored in an * mbuf's internal buffer. * Meanwhile, the default size of the receive socket * buffer for raw sockets is not so large. This means * the possibility of packet loss is relatively higher * than before. To avoid this scenario, we copy the * received data to a separate mbuf that does not use * a cluster, if possible. * XXX: it is better to copy the data after stripping * intermediate headers. */ if ((m->m_flags & M_EXT) && m->m_next == NULL && m->m_len <= MHLEN) { n = m_get(M_NOWAIT, m->m_type); if (n != NULL) { if (m_dup_pkthdr(n, m, M_NOWAIT)) { bcopy(m->m_data, n->m_data, m->m_len); n->m_len = m->m_len; } else { m_free(n); n = NULL; } } } if (n != NULL || (n = m_copym(m, 0, M_COPYALL, M_NOWAIT)) != NULL) { if (last->inp_flags & INP_CONTROLOPTS) ip6_savecontrol(last, n, &opts); /* strip intermediate headers */ m_adj(n, off); SOCKBUF_LOCK(&last->inp_socket->so_rcv); if (sbappendaddr_locked( &last->inp_socket->so_rcv, (struct sockaddr *)&fromsa, n, opts) == 0) { /* should notify about lost packet */ m_freem(n); if (opts) { m_freem(opts); } SOCKBUF_UNLOCK( &last->inp_socket->so_rcv); } else sorwakeup_locked(last->inp_socket); opts = NULL; } INP_RUNLOCK(last); } last = in6p; } INP_INFO_RUNLOCK_ET(&V_ripcbinfo, et); if (last != NULL) { if (last->inp_flags & INP_CONTROLOPTS) ip6_savecontrol(last, m, &opts); /* strip intermediate headers */ m_adj(m, off); /* avoid using mbuf clusters if possible (see above) */ if ((m->m_flags & M_EXT) && m->m_next == NULL && m->m_len <= MHLEN) { struct mbuf *n; n = m_get(M_NOWAIT, m->m_type); if (n != NULL) { if (m_dup_pkthdr(n, m, M_NOWAIT)) { bcopy(m->m_data, n->m_data, m->m_len); n->m_len = m->m_len; m_freem(m); m = n; } else { m_freem(n); n = NULL; } } } SOCKBUF_LOCK(&last->inp_socket->so_rcv); if (sbappendaddr_locked(&last->inp_socket->so_rcv, (struct sockaddr *)&fromsa, m, opts) == 0) { m_freem(m); if (opts) m_freem(opts); SOCKBUF_UNLOCK(&last->inp_socket->so_rcv); } else sorwakeup_locked(last->inp_socket); INP_RUNLOCK(last); } else { m_freem(m); IP6STAT_DEC(ip6s_delivered); } return IPPROTO_DONE; } /* * Reflect the ip6 packet back to the source. * OFF points to the icmp6 header, counted from the top of the mbuf. */ void icmp6_reflect(struct mbuf *m, size_t off) { struct in6_addr src6, *srcp; struct ip6_hdr *ip6; struct icmp6_hdr *icmp6; struct in6_ifaddr *ia = NULL; struct ifnet *outif = NULL; int plen; int type, code, hlim; /* too short to reflect */ if (off < sizeof(struct ip6_hdr)) { nd6log((LOG_DEBUG, "sanity fail: off=%lx, sizeof(ip6)=%lx in %s:%d\n", (u_long)off, (u_long)sizeof(struct ip6_hdr), __FILE__, __LINE__)); goto bad; } /* * If there are extra headers between IPv6 and ICMPv6, strip * off that header first. */ #ifdef DIAGNOSTIC if (sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr) > MHLEN) panic("assumption failed in icmp6_reflect"); #endif if (off > sizeof(struct ip6_hdr)) { size_t l; struct ip6_hdr nip6; l = off - sizeof(struct ip6_hdr); m_copydata(m, 0, sizeof(nip6), (caddr_t)&nip6); m_adj(m, l); l = sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr); if (m->m_len < l) { if ((m = m_pullup(m, l)) == NULL) return; } bcopy((caddr_t)&nip6, mtod(m, caddr_t), sizeof(nip6)); } else /* off == sizeof(struct ip6_hdr) */ { size_t l; l = sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr); if (m->m_len < l) { if ((m = m_pullup(m, l)) == NULL) return; } } plen = m->m_pkthdr.len - sizeof(struct ip6_hdr); ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_nxt = IPPROTO_ICMPV6; icmp6 = (struct icmp6_hdr *)(ip6 + 1); type = icmp6->icmp6_type; /* keep type for statistics */ code = icmp6->icmp6_code; /* ditto. */ hlim = 0; srcp = NULL; /* * If the incoming packet was addressed directly to us (i.e. unicast), * use dst as the src for the reply. * The IN6_IFF_NOTREADY case should be VERY rare, but is possible * (for example) when we encounter an error while forwarding procedure * destined to a duplicated address of ours. */ if (!IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia != NULL && !(ia->ia6_flags & (IN6_IFF_ANYCAST|IN6_IFF_NOTREADY))) { src6 = ia->ia_addr.sin6_addr; srcp = &src6; if (m->m_pkthdr.rcvif != NULL) { /* XXX: This may not be the outgoing interface */ hlim = ND_IFINFO(m->m_pkthdr.rcvif)->chlim; } else hlim = V_ip6_defhlim; } if (ia != NULL) ifa_free(&ia->ia_ifa); } if (srcp == NULL) { int error; struct in6_addr dst6; uint32_t scopeid; /* * This case matches to multicasts, our anycast, or unicasts * that we do not own. Select a source address based on the * source address of the erroneous packet. */ in6_splitscope(&ip6->ip6_src, &dst6, &scopeid); error = in6_selectsrc_addr(M_GETFIB(m), &dst6, scopeid, NULL, &src6, &hlim); if (error) { char ip6buf[INET6_ADDRSTRLEN]; nd6log((LOG_DEBUG, "icmp6_reflect: source can't be determined: " "dst=%s, error=%d\n", ip6_sprintf(ip6buf, &ip6->ip6_dst), error)); goto bad; } srcp = &src6; } /* * ip6_input() drops a packet if its src is multicast. * So, the src is never multicast. */ ip6->ip6_dst = ip6->ip6_src; ip6->ip6_src = *srcp; ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_hlim = hlim; icmp6->icmp6_cksum = 0; icmp6->icmp6_cksum = in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), plen); /* * XXX option handling */ m->m_flags &= ~(M_BCAST|M_MCAST); m->m_pkthdr.rcvif = NULL; ip6_output(m, NULL, NULL, 0, NULL, &outif, NULL); if (outif) icmp6_ifoutstat_inc(outif, type, code); return; bad: m_freem(m); return; } void icmp6_fasttimo(void) { mld_fasttimo(); } void icmp6_slowtimo(void) { mld_slowtimo(); } static const char * icmp6_redirect_diag(struct in6_addr *src6, struct in6_addr *dst6, struct in6_addr *tgt6) { static char buf[1024]; char ip6bufs[INET6_ADDRSTRLEN]; char ip6bufd[INET6_ADDRSTRLEN]; char ip6buft[INET6_ADDRSTRLEN]; snprintf(buf, sizeof(buf), "(src=%s dst=%s tgt=%s)", ip6_sprintf(ip6bufs, src6), ip6_sprintf(ip6bufd, dst6), ip6_sprintf(ip6buft, tgt6)); return buf; } void icmp6_redirect_input(struct mbuf *m, int off) { struct ifnet *ifp; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct nd_redirect *nd_rd; int icmp6len = ntohs(ip6->ip6_plen); char *lladdr = NULL; int lladdrlen = 0; int is_router; int is_onlink; struct in6_addr src6 = ip6->ip6_src; struct in6_addr redtgt6; struct in6_addr reddst6; union nd_opts ndopts; char ip6buf[INET6_ADDRSTRLEN]; M_ASSERTPKTHDR(m); KASSERT(m->m_pkthdr.rcvif != NULL, ("%s: no rcvif", __func__)); ifp = m->m_pkthdr.rcvif; /* XXX if we are router, we don't update route by icmp6 redirect */ if (V_ip6_forwarding) goto freeit; if (!V_icmp6_rediraccept) goto freeit; /* RFC 6980: Nodes MUST silently ignore fragments */ if(m->m_flags & M_FRAGMENTED) goto freeit; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, icmp6len,); nd_rd = (struct nd_redirect *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(nd_rd, struct nd_redirect *, m, off, icmp6len); if (nd_rd == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif redtgt6 = nd_rd->nd_rd_target; reddst6 = nd_rd->nd_rd_dst; if (in6_setscope(&redtgt6, m->m_pkthdr.rcvif, NULL) || in6_setscope(&reddst6, m->m_pkthdr.rcvif, NULL)) { goto freeit; } /* validation */ if (!IN6_IS_ADDR_LINKLOCAL(&src6)) { nd6log((LOG_ERR, "ICMP6 redirect sent from %s rejected; " "must be from linklocal\n", ip6_sprintf(ip6buf, &src6))); goto bad; } if (ip6->ip6_hlim != 255) { nd6log((LOG_ERR, "ICMP6 redirect sent from %s rejected; " "hlim=%d (must be 255)\n", ip6_sprintf(ip6buf, &src6), ip6->ip6_hlim)); goto bad; } { /* ip6->ip6_src must be equal to gw for icmp6->icmp6_reddst */ struct nhop6_basic nh6; struct in6_addr kdst; uint32_t scopeid; in6_splitscope(&reddst6, &kdst, &scopeid); if (fib6_lookup_nh_basic(ifp->if_fib, &kdst, scopeid, 0, 0,&nh6)==0){ if ((nh6.nh_flags & NHF_GATEWAY) == 0) { nd6log((LOG_ERR, "ICMP6 redirect rejected; no route " "with inet6 gateway found for redirect dst: %s\n", icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } /* * Embed scope zone id into next hop address, since * fib6_lookup_nh_basic() returns address without embedded * scope zone id. */ if (in6_setscope(&nh6.nh_addr, m->m_pkthdr.rcvif, NULL)) goto freeit; if (IN6_ARE_ADDR_EQUAL(&src6, &nh6.nh_addr) == 0) { nd6log((LOG_ERR, "ICMP6 redirect rejected; " "not equal to gw-for-src=%s (must be same): " "%s\n", ip6_sprintf(ip6buf, &nh6.nh_addr), icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } } else { nd6log((LOG_ERR, "ICMP6 redirect rejected; " "no route found for redirect dst: %s\n", icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } } if (IN6_IS_ADDR_MULTICAST(&reddst6)) { nd6log((LOG_ERR, "ICMP6 redirect rejected; " "redirect dst must be unicast: %s\n", icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } is_router = is_onlink = 0; if (IN6_IS_ADDR_LINKLOCAL(&redtgt6)) is_router = 1; /* router case */ if (bcmp(&redtgt6, &reddst6, sizeof(redtgt6)) == 0) is_onlink = 1; /* on-link destination case */ if (!is_router && !is_onlink) { nd6log((LOG_ERR, "ICMP6 redirect rejected; " "neither router case nor onlink case: %s\n", icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } icmp6len -= sizeof(*nd_rd); nd6_option_init(nd_rd + 1, icmp6len, &ndopts); if (nd6_options(&ndopts) < 0) { nd6log((LOG_INFO, "%s: invalid ND option, rejected: %s\n", __func__, icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); /* nd6_options have incremented stats */ goto freeit; } if (ndopts.nd_opts_tgt_lladdr) { lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1); lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3; } if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) { nd6log((LOG_INFO, "%s: lladdrlen mismatch for %s " "(if %d, icmp6 packet %d): %s\n", __func__, ip6_sprintf(ip6buf, &redtgt6), ifp->if_addrlen, lladdrlen - 2, icmp6_redirect_diag(&src6, &reddst6, &redtgt6))); goto bad; } /* Validation passed. */ /* RFC 2461 8.3 */ nd6_cache_lladdr(ifp, &redtgt6, lladdr, lladdrlen, ND_REDIRECT, is_onlink ? ND_REDIRECT_ONLINK : ND_REDIRECT_ROUTER); /* * Install a gateway route in the better-router case or an interface * route in the on-link-destination case. */ { struct sockaddr_in6 sdst; struct sockaddr_in6 sgw; struct sockaddr_in6 ssrc; struct sockaddr *gw; int rt_flags; u_int fibnum; bzero(&sdst, sizeof(sdst)); bzero(&ssrc, sizeof(ssrc)); sdst.sin6_family = ssrc.sin6_family = AF_INET6; sdst.sin6_len = ssrc.sin6_len = sizeof(struct sockaddr_in6); bcopy(&reddst6, &sdst.sin6_addr, sizeof(struct in6_addr)); bcopy(&src6, &ssrc.sin6_addr, sizeof(struct in6_addr)); rt_flags = RTF_HOST; if (is_router) { bzero(&sgw, sizeof(sgw)); sgw.sin6_family = AF_INET6; sgw.sin6_len = sizeof(struct sockaddr_in6); bcopy(&redtgt6, &sgw.sin6_addr, sizeof(struct in6_addr)); gw = (struct sockaddr *)&sgw; rt_flags |= RTF_GATEWAY; } else gw = ifp->if_addr->ifa_addr; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) in6_rtredirect((struct sockaddr *)&sdst, gw, (struct sockaddr *)NULL, rt_flags, (struct sockaddr *)&ssrc, fibnum); } /* finally update cached route in each socket via pfctlinput */ { struct sockaddr_in6 sdst; bzero(&sdst, sizeof(sdst)); sdst.sin6_family = AF_INET6; sdst.sin6_len = sizeof(struct sockaddr_in6); bcopy(&reddst6, &sdst.sin6_addr, sizeof(struct in6_addr)); pfctlinput(PRC_REDIRECT_HOST, (struct sockaddr *)&sdst); } freeit: m_freem(m); return; bad: ICMP6STAT_INC(icp6s_badredirect); m_freem(m); } void icmp6_redirect_output(struct mbuf *m0, struct rtentry *rt) { struct ifnet *ifp; /* my outgoing interface */ struct in6_addr *ifp_ll6; struct in6_addr *router_ll6; struct ip6_hdr *sip6; /* m0 as struct ip6_hdr */ struct mbuf *m = NULL; /* newly allocated one */ struct m_tag *mtag; struct ip6_hdr *ip6; /* m as struct ip6_hdr */ struct nd_redirect *nd_rd; struct llentry *ln = NULL; size_t maxlen; u_char *p; struct ifnet *outif = NULL; struct sockaddr_in6 src_sa; icmp6_errcount(ND_REDIRECT, 0); /* if we are not router, we don't send icmp6 redirect */ if (!V_ip6_forwarding) goto fail; /* sanity check */ if (!m0 || !rt || !(rt->rt_flags & RTF_UP) || !(ifp = rt->rt_ifp)) goto fail; /* * Address check: * the source address must identify a neighbor, and * the destination address must not be a multicast address * [RFC 2461, sec 8.2] */ sip6 = mtod(m0, struct ip6_hdr *); bzero(&src_sa, sizeof(src_sa)); src_sa.sin6_family = AF_INET6; src_sa.sin6_len = sizeof(src_sa); src_sa.sin6_addr = sip6->ip6_src; if (nd6_is_addr_neighbor(&src_sa, ifp) == 0) goto fail; if (IN6_IS_ADDR_MULTICAST(&sip6->ip6_dst)) goto fail; /* what should we do here? */ /* rate limit */ if (icmp6_ratelimit(&sip6->ip6_src, ND_REDIRECT, 0)) goto fail; /* * Since we are going to append up to 1280 bytes (= IPV6_MMTU), * we almost always ask for an mbuf cluster for simplicity. * (MHLEN < IPV6_MMTU is almost always true) */ #if IPV6_MMTU >= MCLBYTES # error assumption failed about IPV6_MMTU and MCLBYTES #endif m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) goto fail; M_SETFIB(m, rt->rt_fibnum); maxlen = M_TRAILINGSPACE(m); maxlen = min(IPV6_MMTU, maxlen); /* just for safety */ if (maxlen < sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr) + ((sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7)) { goto fail; } { /* get ip6 linklocal address for ifp(my outgoing interface). */ struct in6_ifaddr *ia; if ((ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY| IN6_IFF_ANYCAST)) == NULL) goto fail; ifp_ll6 = &ia->ia_addr.sin6_addr; /* XXXRW: reference released prematurely. */ ifa_free(&ia->ia_ifa); } /* get ip6 linklocal address for the router. */ if (rt->rt_gateway && (rt->rt_flags & RTF_GATEWAY)) { struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)rt->rt_gateway; router_ll6 = &sin6->sin6_addr; if (!IN6_IS_ADDR_LINKLOCAL(router_ll6)) router_ll6 = (struct in6_addr *)NULL; } else router_ll6 = (struct in6_addr *)NULL; /* ip6 */ ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; /* ip6->ip6_plen will be set later */ ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_hlim = 255; /* ip6->ip6_src must be linklocal addr for my outgoing if. */ bcopy(ifp_ll6, &ip6->ip6_src, sizeof(struct in6_addr)); bcopy(&sip6->ip6_src, &ip6->ip6_dst, sizeof(struct in6_addr)); /* ND Redirect */ nd_rd = (struct nd_redirect *)(ip6 + 1); nd_rd->nd_rd_type = ND_REDIRECT; nd_rd->nd_rd_code = 0; nd_rd->nd_rd_reserved = 0; if (rt->rt_flags & RTF_GATEWAY) { /* * nd_rd->nd_rd_target must be a link-local address in * better router cases. */ if (!router_ll6) goto fail; bcopy(router_ll6, &nd_rd->nd_rd_target, sizeof(nd_rd->nd_rd_target)); bcopy(&sip6->ip6_dst, &nd_rd->nd_rd_dst, sizeof(nd_rd->nd_rd_dst)); } else { /* make sure redtgt == reddst */ bcopy(&sip6->ip6_dst, &nd_rd->nd_rd_target, sizeof(nd_rd->nd_rd_target)); bcopy(&sip6->ip6_dst, &nd_rd->nd_rd_dst, sizeof(nd_rd->nd_rd_dst)); } p = (u_char *)(nd_rd + 1); if (!router_ll6) goto nolladdropt; { /* target lladdr option */ int len; struct nd_opt_hdr *nd_opt; char *lladdr; IF_AFDATA_RLOCK(ifp); ln = nd6_lookup(router_ll6, 0, ifp); IF_AFDATA_RUNLOCK(ifp); if (ln == NULL) goto nolladdropt; len = sizeof(*nd_opt) + ifp->if_addrlen; len = (len + 7) & ~7; /* round by 8 */ /* safety check */ if (len + (p - (u_char *)ip6) > maxlen) goto nolladdropt; if (ln->la_flags & LLE_VALID) { nd_opt = (struct nd_opt_hdr *)p; nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR; nd_opt->nd_opt_len = len >> 3; lladdr = (char *)(nd_opt + 1); bcopy(ln->ll_addr, lladdr, ifp->if_addrlen); p += len; } } nolladdropt: if (ln != NULL) LLE_RUNLOCK(ln); m->m_pkthdr.len = m->m_len = p - (u_char *)ip6; /* just to be safe */ #ifdef M_DECRYPTED /*not openbsd*/ if (m0->m_flags & M_DECRYPTED) goto noredhdropt; #endif if (p - (u_char *)ip6 > maxlen) goto noredhdropt; { /* redirected header option */ int len; struct nd_opt_rd_hdr *nd_opt_rh; /* * compute the maximum size for icmp6 redirect header option. * XXX room for auth header? */ len = maxlen - (p - (u_char *)ip6); len &= ~7; /* This is just for simplicity. */ if (m0->m_pkthdr.len != m0->m_len) { if (m0->m_next) { m_freem(m0->m_next); m0->m_next = NULL; } m0->m_pkthdr.len = m0->m_len; } /* * Redirected header option spec (RFC2461 4.6.3) talks nothing * about padding/truncate rule for the original IP packet. * From the discussion on IPv6imp in Feb 1999, * the consensus was: * - "attach as much as possible" is the goal * - pad if not aligned (original size can be guessed by * original ip6 header) * Following code adds the padding if it is simple enough, * and truncates if not. */ if (m0->m_next || m0->m_pkthdr.len != m0->m_len) panic("assumption failed in %s:%d", __FILE__, __LINE__); if (len - sizeof(*nd_opt_rh) < m0->m_pkthdr.len) { /* not enough room, truncate */ m0->m_pkthdr.len = m0->m_len = len - sizeof(*nd_opt_rh); } else { /* enough room, pad or truncate */ size_t extra; extra = m0->m_pkthdr.len % 8; if (extra) { /* pad if easy enough, truncate if not */ if (8 - extra <= M_TRAILINGSPACE(m0)) { /* pad */ m0->m_len += (8 - extra); m0->m_pkthdr.len += (8 - extra); } else { /* truncate */ m0->m_pkthdr.len -= extra; m0->m_len -= extra; } } len = m0->m_pkthdr.len + sizeof(*nd_opt_rh); m0->m_pkthdr.len = m0->m_len = len - sizeof(*nd_opt_rh); } nd_opt_rh = (struct nd_opt_rd_hdr *)p; bzero(nd_opt_rh, sizeof(*nd_opt_rh)); nd_opt_rh->nd_opt_rh_type = ND_OPT_REDIRECTED_HEADER; nd_opt_rh->nd_opt_rh_len = len >> 3; p += sizeof(*nd_opt_rh); m->m_pkthdr.len = m->m_len = p - (u_char *)ip6; /* connect m0 to m */ m_tag_delete_chain(m0, NULL); m0->m_flags &= ~M_PKTHDR; m->m_next = m0; m->m_pkthdr.len = m->m_len + m0->m_len; m0 = NULL; } noredhdropt:; if (m0) { m_freem(m0); m0 = NULL; } /* XXX: clear embedded link IDs in the inner header */ in6_clearscope(&sip6->ip6_src); in6_clearscope(&sip6->ip6_dst); in6_clearscope(&nd_rd->nd_rd_target); in6_clearscope(&nd_rd->nd_rd_dst); ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(struct ip6_hdr)); nd_rd->nd_rd_cksum = 0; nd_rd->nd_rd_cksum = in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), ntohs(ip6->ip6_plen)); if (send_sendso_input_hook != NULL) { mtag = m_tag_get(PACKET_TAG_ND_OUTGOING, sizeof(unsigned short), M_NOWAIT); if (mtag == NULL) goto fail; *(unsigned short *)(mtag + 1) = nd_rd->nd_rd_type; m_tag_prepend(m, mtag); } /* send the packet to outside... */ ip6_output(m, NULL, NULL, 0, NULL, &outif, NULL); if (outif) { icmp6_ifstat_inc(outif, ifs6_out_msg); icmp6_ifstat_inc(outif, ifs6_out_redirect); } ICMP6STAT_INC(icp6s_outhist[ND_REDIRECT]); return; fail: if (m) m_freem(m); if (m0) m_freem(m0); } /* * ICMPv6 socket option processing. */ int icmp6_ctloutput(struct socket *so, struct sockopt *sopt) { int error = 0; int optlen; struct inpcb *inp = sotoinpcb(so); int level, op, optname; if (sopt) { level = sopt->sopt_level; op = sopt->sopt_dir; optname = sopt->sopt_name; optlen = sopt->sopt_valsize; } else level = op = optname = optlen = 0; if (level != IPPROTO_ICMPV6) { return EINVAL; } switch (op) { case PRCO_SETOPT: switch (optname) { case ICMP6_FILTER: { struct icmp6_filter ic6f; if (optlen != sizeof(ic6f)) { error = EMSGSIZE; break; } error = sooptcopyin(sopt, &ic6f, optlen, optlen); if (error == 0) { INP_WLOCK(inp); *inp->in6p_icmp6filt = ic6f; INP_WUNLOCK(inp); } break; } default: error = ENOPROTOOPT; break; } break; case PRCO_GETOPT: switch (optname) { case ICMP6_FILTER: { struct icmp6_filter ic6f; INP_RLOCK(inp); ic6f = *inp->in6p_icmp6filt; INP_RUNLOCK(inp); error = sooptcopyout(sopt, &ic6f, sizeof(ic6f)); break; } default: error = ENOPROTOOPT; break; } break; } return (error); } /* * Perform rate limit check. * Returns 0 if it is okay to send the icmp6 packet. * Returns 1 if the router SHOULD NOT send this icmp6 packet due to rate * limitation. * * XXX per-destination/type check necessary? * * dst - not used at this moment * type - not used at this moment * code - not used at this moment */ static int icmp6_ratelimit(const struct in6_addr *dst, const int type, const int code) { int ret; ret = 0; /* okay to send */ /* PPS limit */ if (!ppsratecheck(&V_icmp6errppslim_last, &V_icmp6errpps_count, V_icmp6errppslim)) { /* The packet is subject to rate limit */ ret++; } return ret; } Index: head/sys/netinet6/in6_gif.c =================================================================== --- head/sys/netinet6/in6_gif.c (revision 336675) +++ head/sys/netinet6/in6_gif.c (revision 336676) @@ -1,428 +1,428 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * Copyright (c) 2018 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: in6_gif.c,v 1.49 2001/05/14 14:02:17 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif #include #include #include #include #include #include #include #include #define GIF_HLIM 30 -static VNET_DEFINE(int, ip6_gif_hlim) = GIF_HLIM; +VNET_DEFINE_STATIC(int, ip6_gif_hlim) = GIF_HLIM; #define V_ip6_gif_hlim VNET(ip6_gif_hlim) SYSCTL_DECL(_net_inet6_ip6); SYSCTL_INT(_net_inet6_ip6, IPV6CTL_GIF_HLIM, gifhlim, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_gif_hlim), 0, "Default hop limit for encapsulated packets"); /* * We keep interfaces in a hash table using src+dst as key. * Interfaces with GIF_IGNORE_SOURCE flag are linked into plain list. */ -static VNET_DEFINE(struct gif_list *, ipv6_hashtbl) = NULL; -static VNET_DEFINE(struct gif_list, ipv6_list) = CK_LIST_HEAD_INITIALIZER(); +VNET_DEFINE_STATIC(struct gif_list *, ipv6_hashtbl) = NULL; +VNET_DEFINE_STATIC(struct gif_list, ipv6_list) = CK_LIST_HEAD_INITIALIZER(); #define V_ipv6_hashtbl VNET(ipv6_hashtbl) #define V_ipv6_list VNET(ipv6_list) #define GIF_HASH(src, dst) (V_ipv6_hashtbl[\ in6_gif_hashval((src), (dst)) & (GIF_HASH_SIZE - 1)]) #define GIF_HASH_SC(sc) GIF_HASH(&(sc)->gif_ip6hdr->ip6_src,\ &(sc)->gif_ip6hdr->ip6_dst) static uint32_t in6_gif_hashval(const struct in6_addr *src, const struct in6_addr *dst) { uint32_t ret; ret = fnv_32_buf(src, sizeof(*src), FNV1_32_INIT); return (fnv_32_buf(dst, sizeof(*dst), ret)); } static int in6_gif_checkdup(const struct gif_softc *sc, const struct in6_addr *src, const struct in6_addr *dst) { struct gif_softc *tmp; if (sc->gif_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&sc->gif_ip6hdr->ip6_src, src) && IN6_ARE_ADDR_EQUAL(&sc->gif_ip6hdr->ip6_dst, dst)) return (EEXIST); CK_LIST_FOREACH(tmp, &GIF_HASH(src, dst), chain) { if (tmp == sc) continue; if (IN6_ARE_ADDR_EQUAL(&tmp->gif_ip6hdr->ip6_src, src) && IN6_ARE_ADDR_EQUAL(&tmp->gif_ip6hdr->ip6_dst, dst)) return (EADDRNOTAVAIL); } return (0); } static void in6_gif_attach(struct gif_softc *sc) { if (sc->gif_options & GIF_IGNORE_SOURCE) CK_LIST_INSERT_HEAD(&V_ipv6_list, sc, chain); else CK_LIST_INSERT_HEAD(&GIF_HASH_SC(sc), sc, chain); } int in6_gif_setopts(struct gif_softc *sc, u_int options) { /* NOTE: we are protected with gif_ioctl_sx lock */ MPASS(sc->gif_family == AF_INET6); MPASS(sc->gif_options != options); if ((options & GIF_IGNORE_SOURCE) != (sc->gif_options & GIF_IGNORE_SOURCE)) { CK_LIST_REMOVE(sc, chain); sc->gif_options = options; in6_gif_attach(sc); } return (0); } int in6_gif_ioctl(struct gif_softc *sc, u_long cmd, caddr_t data) { struct in6_ifreq *ifr = (struct in6_ifreq *)data; struct sockaddr_in6 *dst, *src; struct ip6_hdr *ip6; int error; /* NOTE: we are protected with gif_ioctl_sx lock */ error = EINVAL; switch (cmd) { case SIOCSIFPHYADDR_IN6: src = &((struct in6_aliasreq *)data)->ifra_addr; dst = &((struct in6_aliasreq *)data)->ifra_dstaddr; /* sanity checks */ if (src->sin6_family != dst->sin6_family || src->sin6_family != AF_INET6 || src->sin6_len != dst->sin6_len || src->sin6_len != sizeof(*src)) break; if (IN6_IS_ADDR_UNSPECIFIED(&src->sin6_addr) || IN6_IS_ADDR_UNSPECIFIED(&dst->sin6_addr)) { error = EADDRNOTAVAIL; break; } /* * Check validity of the scope zone ID of the * addresses, and convert it into the kernel * internal form if necessary. */ if ((error = sa6_embedscope(src, 0)) != 0 || (error = sa6_embedscope(dst, 0)) != 0) break; if (V_ipv6_hashtbl == NULL) V_ipv6_hashtbl = gif_hashinit(); error = in6_gif_checkdup(sc, &src->sin6_addr, &dst->sin6_addr); if (error == EADDRNOTAVAIL) break; if (error == EEXIST) { /* Addresses are the same. Just return. */ error = 0; break; } ip6 = malloc(sizeof(*ip6), M_GIF, M_WAITOK | M_ZERO); ip6->ip6_src = src->sin6_addr; ip6->ip6_dst = dst->sin6_addr; if (sc->gif_family != 0) { /* Detach existing tunnel first */ CK_LIST_REMOVE(sc, chain); GIF_WAIT(); free(sc->gif_hdr, M_GIF); /* XXX: should we notify about link state change? */ } sc->gif_family = AF_INET6; sc->gif_ip6hdr = ip6; in6_gif_attach(sc); break; case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: if (sc->gif_family != AF_INET6) { error = EADDRNOTAVAIL; break; } src = (struct sockaddr_in6 *)&ifr->ifr_addr; memset(src, 0, sizeof(*src)); src->sin6_family = AF_INET6; src->sin6_len = sizeof(*src); src->sin6_addr = (cmd == SIOCGIFPSRCADDR_IN6) ? sc->gif_ip6hdr->ip6_src: sc->gif_ip6hdr->ip6_dst; error = prison_if(curthread->td_ucred, (struct sockaddr *)src); if (error == 0) error = sa6_recoverscope(src); if (error != 0) memset(src, 0, sizeof(*src)); break; } return (error); } int in6_gif_output(struct ifnet *ifp, struct mbuf *m, int proto, uint8_t ecn) { struct gif_softc *sc = ifp->if_softc; struct ip6_hdr *ip6; int len; /* prepend new IP header */ MPASS(in_epoch(net_epoch_preempt)); len = sizeof(struct ip6_hdr); #ifndef __NO_STRICT_ALIGNMENT if (proto == IPPROTO_ETHERIP) len += ETHERIP_ALIGN; #endif M_PREPEND(m, len, M_NOWAIT); if (m == NULL) return (ENOBUFS); #ifndef __NO_STRICT_ALIGNMENT if (proto == IPPROTO_ETHERIP) { len = mtod(m, vm_offset_t) & 3; KASSERT(len == 0 || len == ETHERIP_ALIGN, ("in6_gif_output: unexpected misalignment")); m->m_data += len; m->m_len -= ETHERIP_ALIGN; } #endif ip6 = mtod(m, struct ip6_hdr *); MPASS(sc->gif_family == AF_INET6); bcopy(sc->gif_ip6hdr, ip6, sizeof(struct ip6_hdr)); ip6->ip6_flow |= htonl((uint32_t)ecn << 20); ip6->ip6_nxt = proto; ip6->ip6_hlim = V_ip6_gif_hlim; /* * force fragmentation to minimum MTU, to avoid path MTU discovery. * it is too painful to ask for resend of inner packet, to achieve * path MTU discovery for encapsulated packets. */ return (ip6_output(m, 0, NULL, IPV6_MINMTU, 0, NULL, NULL)); } static int in6_gif_input(struct mbuf *m, int off, int proto, void *arg) { struct gif_softc *sc = arg; struct ifnet *gifp; struct ip6_hdr *ip6; uint8_t ecn; MPASS(in_epoch(net_epoch_preempt)); if (sc == NULL) { m_freem(m); IP6STAT_INC(ip6s_nogif); return (IPPROTO_DONE); } gifp = GIF2IFP(sc); if ((gifp->if_flags & IFF_UP) != 0) { ip6 = mtod(m, struct ip6_hdr *); ecn = (ntohl(ip6->ip6_flow) >> 20) & 0xff; m_adj(m, off); gif_input(m, gifp, proto, ecn); } else { m_freem(m); IP6STAT_INC(ip6s_nogif); } return (IPPROTO_DONE); } static int in6_gif_lookup(const struct mbuf *m, int off, int proto, void **arg) { const struct ip6_hdr *ip6; struct gif_softc *sc; int ret; if (V_ipv6_hashtbl == NULL) return (0); MPASS(in_epoch(net_epoch_preempt)); /* * NOTE: it is safe to iterate without any locking here, because softc * can be reclaimed only when we are not within net_epoch_preempt * section, but ip_encap lookup+input are executed in epoch section. */ ip6 = mtod(m, const struct ip6_hdr *); ret = 0; CK_LIST_FOREACH(sc, &GIF_HASH(&ip6->ip6_dst, &ip6->ip6_src), chain) { /* * This is an inbound packet, its ip6_dst is source address * in softc. */ if (IN6_ARE_ADDR_EQUAL(&sc->gif_ip6hdr->ip6_src, &ip6->ip6_dst) && IN6_ARE_ADDR_EQUAL(&sc->gif_ip6hdr->ip6_dst, &ip6->ip6_src)) { ret = ENCAP_DRV_LOOKUP; goto done; } } /* * No exact match. * Check the list of interfaces with GIF_IGNORE_SOURCE flag. */ CK_LIST_FOREACH(sc, &V_ipv6_list, chain) { if (IN6_ARE_ADDR_EQUAL(&sc->gif_ip6hdr->ip6_src, &ip6->ip6_dst)) { ret = 128 + 8; /* src + proto */ goto done; } } return (0); done: if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0) return (0); /* ingress filters on outer source */ if ((GIF2IFP(sc)->if_flags & IFF_LINK2) == 0) { struct nhop6_basic nh6; if (fib6_lookup_nh_basic(sc->gif_fibnum, &ip6->ip6_src, ntohs(in6_getscope(&ip6->ip6_src)), 0, 0, &nh6) != 0) return (0); if (nh6.nh_ifp != m->m_pkthdr.rcvif) return (0); } *arg = sc; return (ret); } static struct { const struct encap_config encap; const struct encaptab *cookie; } ipv6_encap_cfg[] = { #ifdef INET { .encap = { .proto = IPPROTO_IPV4, .min_length = sizeof(struct ip6_hdr) + sizeof(struct ip), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in6_gif_lookup, .input = in6_gif_input }, }, #endif { .encap = { .proto = IPPROTO_IPV6, .min_length = 2 * sizeof(struct ip6_hdr), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in6_gif_lookup, .input = in6_gif_input }, }, { .encap = { .proto = IPPROTO_ETHERIP, .min_length = sizeof(struct ip6_hdr) + sizeof(struct etherip_header) + sizeof(struct ether_header), .exact_match = ENCAP_DRV_LOOKUP, .lookup = in6_gif_lookup, .input = in6_gif_input }, } }; void in6_gif_init(void) { int i; if (!IS_DEFAULT_VNET(curvnet)) return; for (i = 0; i < nitems(ipv6_encap_cfg); i++) ipv6_encap_cfg[i].cookie = ip6_encap_attach( &ipv6_encap_cfg[i].encap, NULL, M_WAITOK); } void in6_gif_uninit(void) { int i; if (IS_DEFAULT_VNET(curvnet)) { for (i = 0; i < nitems(ipv6_encap_cfg); i++) ip6_encap_detach(ipv6_encap_cfg[i].cookie); } if (V_ipv6_hashtbl != NULL) gif_hashdestroy(V_ipv6_hashtbl); } Index: head/sys/netinet6/in6_rmx.c =================================================================== --- head/sys/netinet6/in6_rmx.c (revision 336675) +++ head/sys/netinet6/in6_rmx.c (revision 336676) @@ -1,286 +1,286 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: in6_rmx.c,v 1.11 2001/07/26 06:53:16 jinmei Exp $ */ /*- * Copyright 1994, 1995 Massachusetts Institute of Technology * * Permission to use, copy, modify, and distribute this software and * its documentation for any purpose and without fee is hereby * granted, provided that both the above copyright notice and this * permission notice appear in all copies, that both the above * copyright notice and this permission notice appear in all * supporting documentation, and that the name of M.I.T. not be used * in advertising or publicity pertaining to distribution of the * software without specific, written prior permission. M.I.T. makes * no representations about the suitability of this software for any * purpose. It is provided "as is" without express or implied * warranty. * * THIS SOFTWARE IS PROVIDED BY M.I.T. ``AS IS''. M.I.T. DISCLAIMS * ALL EXPRESS OR IMPLIED WARRANTIES WITH REGARD TO THIS SOFTWARE, * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. IN NO EVENT * SHALL M.I.T. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF * USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include extern int in6_inithead(void **head, int off); #ifdef VIMAGE extern int in6_detachhead(void **head, int off); #endif /* * Do what we need to do when inserting a route. */ static struct radix_node * in6_addroute(void *v_arg, void *n_arg, struct radix_head *head, struct radix_node *treenodes) { struct rtentry *rt = (struct rtentry *)treenodes; struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)rt_key(rt); if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) rt->rt_flags |= RTF_MULTICAST; /* * A little bit of help for both IPv6 output and input: * For local addresses, we make sure that RTF_LOCAL is set, * with the thought that this might one day be used to speed up * ip_input(). * * We also mark routes to multicast addresses as such, because * it's easy to do and might be useful (but this is much more * dubious since it's so easy to inspect the address). (This * is done above.) * * XXX * should elaborate the code. */ if (rt->rt_flags & RTF_HOST) { if (IN6_ARE_ADDR_EQUAL(&satosin6(rt->rt_ifa->ifa_addr) ->sin6_addr, &sin6->sin6_addr)) { rt->rt_flags |= RTF_LOCAL; } } if (rt->rt_ifp != NULL) { /* * Check route MTU: * inherit interface MTU if not set or * check if MTU is too large. */ if (rt->rt_mtu == 0) { rt->rt_mtu = IN6_LINKMTU(rt->rt_ifp); } else if (rt->rt_mtu > IN6_LINKMTU(rt->rt_ifp)) rt->rt_mtu = IN6_LINKMTU(rt->rt_ifp); } return (rn_addroute(v_arg, n_arg, head, treenodes)); } /* * Age old PMTUs. */ struct mtuex_arg { struct rib_head *rnh; time_t nextstop; }; -static VNET_DEFINE(struct callout, rtq_mtutimer); +VNET_DEFINE_STATIC(struct callout, rtq_mtutimer); #define V_rtq_mtutimer VNET(rtq_mtutimer) static int in6_mtuexpire(struct rtentry *rt, void *rock) { struct mtuex_arg *ap = rock; if (rt->rt_expire && !(rt->rt_flags & RTF_PROBEMTU)) { if (rt->rt_expire <= time_uptime) { rt->rt_flags |= RTF_PROBEMTU; } else { ap->nextstop = lmin(ap->nextstop, rt->rt_expire); } } return (0); } #define MTUTIMO_DEFAULT (60*1) static void in6_mtutimo_setwa(struct rib_head *rnh, uint32_t fibum, int af, void *_arg) { struct mtuex_arg *arg; arg = (struct mtuex_arg *)_arg; arg->rnh = rnh; } static void in6_mtutimo(void *rock) { CURVNET_SET_QUIET((struct vnet *) rock); struct timeval atv; struct mtuex_arg arg; rt_foreach_fib_walk(AF_INET6, in6_mtutimo_setwa, in6_mtuexpire, &arg); atv.tv_sec = MTUTIMO_DEFAULT; atv.tv_usec = 0; callout_reset(&V_rtq_mtutimer, tvtohz(&atv), in6_mtutimo, rock); CURVNET_RESTORE(); } /* * Initialize our routing tree. */ -static VNET_DEFINE(int, _in6_rt_was_here); +VNET_DEFINE_STATIC(int, _in6_rt_was_here); #define V__in6_rt_was_here VNET(_in6_rt_was_here) int in6_inithead(void **head, int off) { struct rib_head *rh; rh = rt_table_init(offsetof(struct sockaddr_in6, sin6_addr) << 3); if (rh == NULL) return (0); rh->rnh_addaddr = in6_addroute; *head = (void *)rh; if (V__in6_rt_was_here == 0) { callout_init(&V_rtq_mtutimer, 1); in6_mtutimo(curvnet); /* kick off timeout first time */ V__in6_rt_was_here = 1; } return (1); } #ifdef VIMAGE int in6_detachhead(void **head, int off) { callout_drain(&V_rtq_mtutimer); rt_table_destroy((struct rib_head *)(*head)); return (1); } #endif /* * Extended API for IPv6 FIB support. */ void in6_rtredirect(struct sockaddr *dst, struct sockaddr *gw, struct sockaddr *nm, int flags, struct sockaddr *src, u_int fibnum) { rtredirect_fib(dst, gw, nm, flags, src, fibnum); } int in6_rtrequest(int req, struct sockaddr *dst, struct sockaddr *gw, struct sockaddr *mask, int flags, struct rtentry **ret_nrt, u_int fibnum) { return (rtrequest_fib(req, dst, gw, mask, flags, ret_nrt, fibnum)); } void in6_rtalloc(struct route_in6 *ro, u_int fibnum) { rtalloc_ign_fib((struct route *)ro, 0ul, fibnum); } void in6_rtalloc_ign(struct route_in6 *ro, u_long ignflags, u_int fibnum) { rtalloc_ign_fib((struct route *)ro, ignflags, fibnum); } struct rtentry * in6_rtalloc1(struct sockaddr *dst, int report, u_long ignflags, u_int fibnum) { return (rtalloc1_fib(dst, report, ignflags, fibnum)); } Index: head/sys/netinet6/in6_src.c =================================================================== --- head/sys/netinet6/in6_src.c (revision 336675) +++ head/sys/netinet6/in6_src.c (revision 336676) @@ -1,1243 +1,1243 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: in6_src.c,v 1.132 2003/08/26 04:42:27 keiichi Exp $ */ /*- * Copyright (c) 1982, 1986, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)in_pcb.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static struct mtx addrsel_lock; #define ADDRSEL_LOCK_INIT() mtx_init(&addrsel_lock, "addrsel_lock", NULL, MTX_DEF) #define ADDRSEL_LOCK() mtx_lock(&addrsel_lock) #define ADDRSEL_UNLOCK() mtx_unlock(&addrsel_lock) #define ADDRSEL_LOCK_ASSERT() mtx_assert(&addrsel_lock, MA_OWNED) static struct sx addrsel_sxlock; #define ADDRSEL_SXLOCK_INIT() sx_init(&addrsel_sxlock, "addrsel_sxlock") #define ADDRSEL_SLOCK() sx_slock(&addrsel_sxlock) #define ADDRSEL_SUNLOCK() sx_sunlock(&addrsel_sxlock) #define ADDRSEL_XLOCK() sx_xlock(&addrsel_sxlock) #define ADDRSEL_XUNLOCK() sx_xunlock(&addrsel_sxlock) #define ADDR_LABEL_NOTAPP (-1) -static VNET_DEFINE(struct in6_addrpolicy, defaultaddrpolicy); +VNET_DEFINE_STATIC(struct in6_addrpolicy, defaultaddrpolicy); #define V_defaultaddrpolicy VNET(defaultaddrpolicy) VNET_DEFINE(int, ip6_prefer_tempaddr) = 0; static int selectroute(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct route_in6 *, struct ifnet **, struct rtentry **, int, u_int); static int in6_selectif(struct sockaddr_in6 *, struct ip6_pktopts *, struct ip6_moptions *, struct ifnet **, struct ifnet *, u_int); static int in6_selectsrc(uint32_t, struct sockaddr_in6 *, struct ip6_pktopts *, struct inpcb *, struct ucred *, struct ifnet **, struct in6_addr *); static struct in6_addrpolicy *lookup_addrsel_policy(struct sockaddr_in6 *); static void init_policy_queue(void); static int add_addrsel_policyent(struct in6_addrpolicy *); static int delete_addrsel_policyent(struct in6_addrpolicy *); static int walk_addrsel_policy(int (*)(struct in6_addrpolicy *, void *), void *); static int dump_addrsel_policyent(struct in6_addrpolicy *, void *); static struct in6_addrpolicy *match_addrsel_policy(struct sockaddr_in6 *); /* * Return an IPv6 address, which is the most appropriate for a given * destination and user specified options. * If necessary, this function lookups the routing table and returns * an entry to the caller for later use. */ #define REPLACE(r) do {\ IP6STAT_INC(ip6s_sources_rule[(r)]); \ /* { \ char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \ printf("in6_selectsrc: replace %s with %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \ } */ \ goto replace; \ } while(0) #define NEXT(r) do {\ /* { \ char ip6buf[INET6_ADDRSTRLEN], ip6b[INET6_ADDRSTRLEN]; \ printf("in6_selectsrc: keep %s against %s by %d\n", ia_best ? ip6_sprintf(ip6buf, &ia_best->ia_addr.sin6_addr) : "none", ip6_sprintf(ip6b, &ia->ia_addr.sin6_addr), (r)); \ } */ \ goto next; /* XXX: we can't use 'continue' here */ \ } while(0) #define BREAK(r) do { \ IP6STAT_INC(ip6s_sources_rule[(r)]); \ goto out; /* XXX: we can't use 'break' here */ \ } while(0) static int in6_selectsrc(uint32_t fibnum, struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred, struct ifnet **ifpp, struct in6_addr *srcp) { struct rm_priotracker in6_ifa_tracker; struct in6_addr dst, tmp; struct ifnet *ifp = NULL, *oifp = NULL; struct in6_ifaddr *ia = NULL, *ia_best = NULL; struct in6_pktinfo *pi = NULL; int dst_scope = -1, best_scope = -1, best_matchlen = -1; struct in6_addrpolicy *dst_policy = NULL, *best_policy = NULL; u_int32_t odstzone; int prefer_tempaddr; int error; struct ip6_moptions *mopts; KASSERT(srcp != NULL, ("%s: srcp is NULL", __func__)); dst = dstsock->sin6_addr; /* make a copy for local operation */ if (ifpp) { /* * Save a possibly passed in ifp for in6_selectsrc. Only * neighbor discovery code should use this feature, where * we may know the interface but not the FIB number holding * the connected subnet in case someone deleted it from the * default FIB and we need to check the interface. */ if (*ifpp != NULL) oifp = *ifpp; *ifpp = NULL; } if (inp != NULL) { INP_LOCK_ASSERT(inp); mopts = inp->in6p_moptions; } else { mopts = NULL; } /* * If the source address is explicitly specified by the caller, * check if the requested source address is indeed a unicast address * assigned to the node, and can be used as the packet's source * address. If everything is okay, use the address as source. */ if (opts && (pi = opts->ip6po_pktinfo) && !IN6_IS_ADDR_UNSPECIFIED(&pi->ipi6_addr)) { /* get the outgoing interface */ if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp, fibnum)) != 0) return (error); /* * determine the appropriate zone id of the source based on * the zone of the destination and the outgoing interface. * If the specified address is ambiguous wrt the scope zone, * the interface must be specified; otherwise, ifa_ifwithaddr() * will fail matching the address. */ tmp = pi->ipi6_addr; if (ifp) { error = in6_setscope(&tmp, ifp, &odstzone); if (error) return (error); } if (cred != NULL && (error = prison_local_ip6(cred, &tmp, (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) return (error); /* * If IPV6_BINDANY socket option is set, we allow to specify * non local addresses as source address in IPV6_PKTINFO * ancillary data. */ if ((inp->inp_flags & INP_BINDANY) == 0) { ia = in6ifa_ifwithaddr(&tmp, 0 /* XXX */); if (ia == NULL || (ia->ia6_flags & (IN6_IFF_ANYCAST | IN6_IFF_NOTREADY))) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (EADDRNOTAVAIL); } bcopy(&ia->ia_addr.sin6_addr, srcp, sizeof(*srcp)); ifa_free(&ia->ia_ifa); } else bcopy(&tmp, srcp, sizeof(*srcp)); pi->ipi6_addr = tmp; /* XXX: this overrides pi */ if (ifpp) *ifpp = ifp; return (0); } /* * Otherwise, if the socket has already bound the source, just use it. */ if (inp != NULL && !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) { if (cred != NULL && (error = prison_local_ip6(cred, &inp->in6p_laddr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0))) != 0) return (error); bcopy(&inp->in6p_laddr, srcp, sizeof(*srcp)); return (0); } /* * Bypass source address selection and use the primary jail IP * if requested. */ if (cred != NULL && !prison_saddrsel_ip6(cred, srcp)) return (0); /* * If the address is not specified, choose the best one based on * the outgoing interface and the destination address. */ /* get the outgoing interface */ if ((error = in6_selectif(dstsock, opts, mopts, &ifp, oifp, (inp != NULL) ? inp->inp_inc.inc_fibnum : fibnum)) != 0) return (error); #ifdef DIAGNOSTIC if (ifp == NULL) /* this should not happen */ panic("in6_selectsrc: NULL ifp"); #endif error = in6_setscope(&dst, ifp, &odstzone); if (error) return (error); IN6_IFADDR_RLOCK(&in6_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { int new_scope = -1, new_matchlen = -1; struct in6_addrpolicy *new_policy = NULL; u_int32_t srczone, osrczone, dstzone; struct in6_addr src; struct ifnet *ifp1 = ia->ia_ifp; /* * We'll never take an address that breaks the scope zone * of the destination. We also skip an address if its zone * does not contain the outgoing interface. * XXX: we should probably use sin6_scope_id here. */ if (in6_setscope(&dst, ifp1, &dstzone) || odstzone != dstzone) { continue; } src = ia->ia_addr.sin6_addr; if (in6_setscope(&src, ifp, &osrczone) || in6_setscope(&src, ifp1, &srczone) || osrczone != srczone) { continue; } /* avoid unusable addresses */ if ((ia->ia6_flags & (IN6_IFF_NOTREADY | IN6_IFF_ANYCAST | IN6_IFF_DETACHED))) { continue; } if (!V_ip6_use_deprecated && IFA6_IS_DEPRECATED(ia)) continue; /* If jailed only take addresses of the jail into account. */ if (cred != NULL && prison_check_ip6(cred, &ia->ia_addr.sin6_addr) != 0) continue; /* Rule 1: Prefer same address */ if (IN6_ARE_ADDR_EQUAL(&dst, &ia->ia_addr.sin6_addr)) { ia_best = ia; BREAK(1); /* there should be no better candidate */ } if (ia_best == NULL) REPLACE(0); /* Rule 2: Prefer appropriate scope */ if (dst_scope < 0) dst_scope = in6_addrscope(&dst); new_scope = in6_addrscope(&ia->ia_addr.sin6_addr); if (IN6_ARE_SCOPE_CMP(best_scope, new_scope) < 0) { if (IN6_ARE_SCOPE_CMP(best_scope, dst_scope) < 0) REPLACE(2); NEXT(2); } else if (IN6_ARE_SCOPE_CMP(new_scope, best_scope) < 0) { if (IN6_ARE_SCOPE_CMP(new_scope, dst_scope) < 0) NEXT(2); REPLACE(2); } /* * Rule 3: Avoid deprecated addresses. Note that the case of * !ip6_use_deprecated is already rejected above. */ if (!IFA6_IS_DEPRECATED(ia_best) && IFA6_IS_DEPRECATED(ia)) NEXT(3); if (IFA6_IS_DEPRECATED(ia_best) && !IFA6_IS_DEPRECATED(ia)) REPLACE(3); /* Rule 4: Prefer home addresses */ /* * XXX: This is a TODO. We should probably merge the MIP6 * case above. */ /* Rule 5: Prefer outgoing interface */ if (!(ND_IFINFO(ifp)->flags & ND6_IFF_NO_PREFER_IFACE)) { if (ia_best->ia_ifp == ifp && ia->ia_ifp != ifp) NEXT(5); if (ia_best->ia_ifp != ifp && ia->ia_ifp == ifp) REPLACE(5); } /* * Rule 6: Prefer matching label * Note that best_policy should be non-NULL here. */ if (dst_policy == NULL) dst_policy = lookup_addrsel_policy(dstsock); if (dst_policy->label != ADDR_LABEL_NOTAPP) { new_policy = lookup_addrsel_policy(&ia->ia_addr); if (dst_policy->label == best_policy->label && dst_policy->label != new_policy->label) NEXT(6); if (dst_policy->label != best_policy->label && dst_policy->label == new_policy->label) REPLACE(6); } /* * Rule 7: Prefer public addresses. * We allow users to reverse the logic by configuring * a sysctl variable, so that privacy conscious users can * always prefer temporary addresses. */ if (opts == NULL || opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_SYSTEM) { prefer_tempaddr = V_ip6_prefer_tempaddr; } else if (opts->ip6po_prefer_tempaddr == IP6PO_TEMPADDR_NOTPREFER) { prefer_tempaddr = 0; } else prefer_tempaddr = 1; if (!(ia_best->ia6_flags & IN6_IFF_TEMPORARY) && (ia->ia6_flags & IN6_IFF_TEMPORARY)) { if (prefer_tempaddr) REPLACE(7); else NEXT(7); } if ((ia_best->ia6_flags & IN6_IFF_TEMPORARY) && !(ia->ia6_flags & IN6_IFF_TEMPORARY)) { if (prefer_tempaddr) NEXT(7); else REPLACE(7); } /* * Rule 8: prefer addresses on alive interfaces. * This is a KAME specific rule. */ if ((ia_best->ia_ifp->if_flags & IFF_UP) && !(ia->ia_ifp->if_flags & IFF_UP)) NEXT(8); if (!(ia_best->ia_ifp->if_flags & IFF_UP) && (ia->ia_ifp->if_flags & IFF_UP)) REPLACE(8); /* * Rule 9: prefer address with better virtual status. */ if (ifa_preferred(&ia_best->ia_ifa, &ia->ia_ifa)) REPLACE(9); if (ifa_preferred(&ia->ia_ifa, &ia_best->ia_ifa)) NEXT(9); /* * Rule 10: prefer address with `prefer_source' flag. */ if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0 && (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0) REPLACE(10); if ((ia_best->ia6_flags & IN6_IFF_PREFER_SOURCE) != 0 && (ia->ia6_flags & IN6_IFF_PREFER_SOURCE) == 0) NEXT(10); /* * Rule 14: Use longest matching prefix. * Note: in the address selection draft, this rule is * documented as "Rule 8". However, since it is also * documented that this rule can be overridden, we assign * a large number so that it is easy to assign smaller numbers * to more preferred rules. */ new_matchlen = in6_matchlen(&ia->ia_addr.sin6_addr, &dst); if (best_matchlen < new_matchlen) REPLACE(14); if (new_matchlen < best_matchlen) NEXT(14); /* Rule 15 is reserved. */ /* * Last resort: just keep the current candidate. * Or, do we need more rules? */ continue; replace: ia_best = ia; best_scope = (new_scope >= 0 ? new_scope : in6_addrscope(&ia_best->ia_addr.sin6_addr)); best_policy = (new_policy ? new_policy : lookup_addrsel_policy(&ia_best->ia_addr)); best_matchlen = (new_matchlen >= 0 ? new_matchlen : in6_matchlen(&ia_best->ia_addr.sin6_addr, &dst)); next: continue; out: break; } if ((ia = ia_best) == NULL) { IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); IP6STAT_INC(ip6s_sources_none); return (EADDRNOTAVAIL); } /* * At this point at least one of the addresses belonged to the jail * but it could still be, that we want to further restrict it, e.g. * theoratically IN6_IS_ADDR_LOOPBACK. * It must not be IN6_IS_ADDR_UNSPECIFIED anymore. * prison_local_ip6() will fix an IN6_IS_ADDR_LOOPBACK but should * let all others previously selected pass. * Use tmp to not change ::1 on lo0 to the primary jail address. */ tmp = ia->ia_addr.sin6_addr; if (cred != NULL && prison_local_ip6(cred, &tmp, (inp != NULL && (inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)) != 0) { IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); IP6STAT_INC(ip6s_sources_none); return (EADDRNOTAVAIL); } if (ifpp) *ifpp = ifp; bcopy(&tmp, srcp, sizeof(*srcp)); if (ia->ia_ifp == ifp) IP6STAT_INC(ip6s_sources_sameif[best_scope]); else IP6STAT_INC(ip6s_sources_otherif[best_scope]); if (dst_scope == best_scope) IP6STAT_INC(ip6s_sources_samescope[best_scope]); else IP6STAT_INC(ip6s_sources_otherscope[best_scope]); if (IFA6_IS_DEPRECATED(ia)) IP6STAT_INC(ip6s_sources_deprecated[best_scope]); IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); return (0); } /* * Select source address based on @inp, @dstsock and @opts. * Stores selected address to @srcp. If @scope_ambiguous is set, * embed scope from selected outgoing interface. If @hlim pointer * is provided, stores calculated hop limit there. * Returns 0 on success. */ int in6_selectsrc_socket(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct inpcb *inp, struct ucred *cred, int scope_ambiguous, struct in6_addr *srcp, int *hlim) { struct ifnet *retifp; uint32_t fibnum; int error; fibnum = inp->inp_inc.inc_fibnum; retifp = NULL; error = in6_selectsrc(fibnum, dstsock, opts, inp, cred, &retifp, srcp); if (error != 0) return (error); if (hlim != NULL) *hlim = in6_selecthlim(inp, retifp); if (retifp == NULL || scope_ambiguous == 0) return (0); /* * Application should provide a proper zone ID or the use of * default zone IDs should be enabled. Unfortunately, some * applications do not behave as it should, so we need a * workaround. Even if an appropriate ID is not determined * (when it's required), if we can determine the outgoing * interface. determine the zone ID based on the interface. */ error = in6_setscope(&dstsock->sin6_addr, retifp, NULL); return (error); } /* * Select source address based on @fibnum, @dst and @scopeid. * Stores selected address to @srcp. * Returns 0 on success. * * Used by non-socket based consumers (ND code mostly) */ int in6_selectsrc_addr(uint32_t fibnum, const struct in6_addr *dst, uint32_t scopeid, struct ifnet *ifp, struct in6_addr *srcp, int *hlim) { struct ifnet *retifp; struct sockaddr_in6 dst_sa; int error; retifp = ifp; bzero(&dst_sa, sizeof(dst_sa)); dst_sa.sin6_family = AF_INET6; dst_sa.sin6_len = sizeof(dst_sa); dst_sa.sin6_addr = *dst; dst_sa.sin6_scope_id = scopeid; sa6_embedscope(&dst_sa, 0); error = in6_selectsrc(fibnum, &dst_sa, NULL, NULL, NULL, &retifp, srcp); if (hlim != NULL) *hlim = in6_selecthlim(NULL, retifp); return (error); } /* * clone - meaningful only for bsdi and freebsd */ static int selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct route_in6 *ro, struct ifnet **retifp, struct rtentry **retrt, int norouteok, u_int fibnum) { int error = 0; struct ifnet *ifp = NULL; struct rtentry *rt = NULL; struct sockaddr_in6 *sin6_next; struct in6_pktinfo *pi = NULL; struct in6_addr *dst = &dstsock->sin6_addr; uint32_t zoneid; #if 0 char ip6buf[INET6_ADDRSTRLEN]; if (dstsock->sin6_addr.s6_addr32[0] == 0 && dstsock->sin6_addr.s6_addr32[1] == 0 && !IN6_IS_ADDR_LOOPBACK(&dstsock->sin6_addr)) { printf("in6_selectroute: strange destination %s\n", ip6_sprintf(ip6buf, &dstsock->sin6_addr)); } else { printf("in6_selectroute: destination = %s%%%d\n", ip6_sprintf(ip6buf, &dstsock->sin6_addr), dstsock->sin6_scope_id); /* for debug */ } #endif /* If the caller specify the outgoing interface explicitly, use it. */ if (opts && (pi = opts->ip6po_pktinfo) != NULL && pi->ipi6_ifindex) { /* XXX boundary check is assumed to be already done. */ ifp = ifnet_byindex(pi->ipi6_ifindex); if (ifp != NULL && (norouteok || retrt == NULL || IN6_IS_ADDR_MULTICAST(dst))) { /* * we do not have to check or get the route for * multicast. */ goto done; } else goto getroute; } /* * If the destination address is a multicast address and the outgoing * interface for the address is specified by the caller, use it. */ if (IN6_IS_ADDR_MULTICAST(dst) && mopts != NULL && (ifp = mopts->im6o_multicast_ifp) != NULL) { goto done; /* we do not need a route for multicast. */ } /* * If destination address is LLA or link- or node-local multicast, * use it's embedded scope zone id to determine outgoing interface. */ if (IN6_IS_ADDR_MC_LINKLOCAL(dst) || IN6_IS_ADDR_MC_NODELOCAL(dst)) { zoneid = ntohs(in6_getscope(dst)); if (zoneid > 0) { ifp = in6_getlinkifnet(zoneid); goto done; } } getroute: /* * If the next hop address for the packet is specified by the caller, * use it as the gateway. */ if (opts && opts->ip6po_nexthop) { struct route_in6 *ron; sin6_next = satosin6(opts->ip6po_nexthop); if (IN6_IS_ADDR_LINKLOCAL(&sin6_next->sin6_addr)) { /* * Next hop is LLA, thus it should be neighbor. * Determine outgoing interface by zone index. */ zoneid = ntohs(in6_getscope(&sin6_next->sin6_addr)); if (zoneid > 0) { ifp = in6_getlinkifnet(zoneid); goto done; } } ron = &opts->ip6po_nextroute; /* Use a cached route if it exists and is valid. */ if (ron->ro_rt != NULL && ( (ron->ro_rt->rt_flags & RTF_UP) == 0 || ron->ro_dst.sin6_family != AF_INET6 || !IN6_ARE_ADDR_EQUAL(&ron->ro_dst.sin6_addr, &sin6_next->sin6_addr))) RO_RTFREE(ron); if (ron->ro_rt == NULL) { ron->ro_dst = *sin6_next; in6_rtalloc(ron, fibnum); /* multi path case? */ } /* * The node identified by that address must be a * neighbor of the sending host. */ if (ron->ro_rt == NULL || (ron->ro_rt->rt_flags & RTF_GATEWAY) != 0) error = EHOSTUNREACH; goto done; } /* * Use a cached route if it exists and is valid, else try to allocate * a new one. Note that we should check the address family of the * cached destination, in case of sharing the cache with IPv4. */ if (ro) { if (ro->ro_rt && (!(ro->ro_rt->rt_flags & RTF_UP) || ((struct sockaddr *)(&ro->ro_dst))->sa_family != AF_INET6 || !IN6_ARE_ADDR_EQUAL(&satosin6(&ro->ro_dst)->sin6_addr, dst))) { RTFREE(ro->ro_rt); ro->ro_rt = (struct rtentry *)NULL; } if (ro->ro_rt == (struct rtentry *)NULL) { struct sockaddr_in6 *sa6; /* No route yet, so try to acquire one */ bzero(&ro->ro_dst, sizeof(struct sockaddr_in6)); sa6 = (struct sockaddr_in6 *)&ro->ro_dst; *sa6 = *dstsock; sa6->sin6_scope_id = 0; #ifdef RADIX_MPATH rtalloc_mpath_fib((struct route *)ro, ntohl(sa6->sin6_addr.s6_addr32[3]), fibnum); #else ro->ro_rt = in6_rtalloc1((struct sockaddr *) &ro->ro_dst, 0, 0UL, fibnum); if (ro->ro_rt) RT_UNLOCK(ro->ro_rt); #endif } /* * do not care about the result if we have the nexthop * explicitly specified. */ if (opts && opts->ip6po_nexthop) goto done; if (ro->ro_rt) { ifp = ro->ro_rt->rt_ifp; if (ifp == NULL) { /* can this really happen? */ RTFREE(ro->ro_rt); ro->ro_rt = NULL; } } if (ro->ro_rt == NULL) error = EHOSTUNREACH; rt = ro->ro_rt; /* * Check if the outgoing interface conflicts with * the interface specified by ipi6_ifindex (if specified). * Note that loopback interface is always okay. * (this may happen when we are sending a packet to one of * our own addresses.) */ if (ifp && opts && opts->ip6po_pktinfo && opts->ip6po_pktinfo->ipi6_ifindex) { if (!(ifp->if_flags & IFF_LOOPBACK) && ifp->if_index != opts->ip6po_pktinfo->ipi6_ifindex) { error = EHOSTUNREACH; goto done; } } } done: if (ifp == NULL && rt == NULL) { /* * This can happen if the caller did not pass a cached route * nor any other hints. We treat this case an error. */ error = EHOSTUNREACH; } if (error == EHOSTUNREACH) IP6STAT_INC(ip6s_noroute); if (retifp != NULL) { *retifp = ifp; /* * Adjust the "outgoing" interface. If we're going to loop * the packet back to ourselves, the ifp would be the loopback * interface. However, we'd rather know the interface associated * to the destination address (which should probably be one of * our own addresses.) */ if (rt) { if ((rt->rt_ifp->if_flags & IFF_LOOPBACK) && (rt->rt_gateway->sa_family == AF_LINK)) *retifp = ifnet_byindex(((struct sockaddr_dl *) rt->rt_gateway)->sdl_index); } } if (retrt != NULL) *retrt = rt; /* rt may be NULL */ return (error); } static int in6_selectif(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct ifnet **retifp, struct ifnet *oifp, u_int fibnum) { int error; struct route_in6 sro; struct rtentry *rt = NULL; int rt_flags; KASSERT(retifp != NULL, ("%s: retifp is NULL", __func__)); bzero(&sro, sizeof(sro)); rt_flags = 0; error = selectroute(dstsock, opts, mopts, &sro, retifp, &rt, 1, fibnum); if (rt) rt_flags = rt->rt_flags; if (rt && rt == sro.ro_rt) RTFREE(rt); if (error != 0) { /* Help ND. See oifp comment in in6_selectsrc(). */ if (oifp != NULL && fibnum == RT_DEFAULT_FIB) { *retifp = oifp; error = 0; } return (error); } /* * do not use a rejected or black hole route. * XXX: this check should be done in the L2 output routine. * However, if we skipped this check here, we'd see the following * scenario: * - install a rejected route for a scoped address prefix * (like fe80::/10) * - send a packet to a destination that matches the scoped prefix, * with ambiguity about the scope zone. * - pick the outgoing interface from the route, and disambiguate the * scope zone with the interface. * - ip6_output() would try to get another route with the "new" * destination, which may be valid. * - we'd see no error on output. * Although this may not be very harmful, it should still be confusing. * We thus reject the case here. */ if (rt_flags & (RTF_REJECT | RTF_BLACKHOLE)) { error = (rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); return (error); } return (0); } /* * Public wrapper function to selectroute(). * * XXX-BZ in6_selectroute() should and will grow the FIB argument. The * in6_selectroute_fib() function is only there for backward compat on stable. */ int in6_selectroute(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct route_in6 *ro, struct ifnet **retifp, struct rtentry **retrt) { return (selectroute(dstsock, opts, mopts, ro, retifp, retrt, 0, RT_DEFAULT_FIB)); } #ifndef BURN_BRIDGES int in6_selectroute_fib(struct sockaddr_in6 *dstsock, struct ip6_pktopts *opts, struct ip6_moptions *mopts, struct route_in6 *ro, struct ifnet **retifp, struct rtentry **retrt, u_int fibnum) { return (selectroute(dstsock, opts, mopts, ro, retifp, retrt, 0, fibnum)); } #endif /* * Default hop limit selection. The precedence is as follows: * 1. Hoplimit value specified via ioctl. * 2. (If the outgoing interface is detected) the current * hop limit of the interface specified by router advertisement. * 3. The system default hoplimit. */ int in6_selecthlim(struct inpcb *in6p, struct ifnet *ifp) { if (in6p && in6p->in6p_hops >= 0) return (in6p->in6p_hops); else if (ifp) return (ND_IFINFO(ifp)->chlim); else if (in6p && !IN6_IS_ADDR_UNSPECIFIED(&in6p->in6p_faddr)) { struct nhop6_basic nh6; struct in6_addr dst; uint32_t fibnum, scopeid; int hlim; fibnum = in6p->inp_inc.inc_fibnum; in6_splitscope(&in6p->in6p_faddr, &dst, &scopeid); if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6)==0){ hlim = ND_IFINFO(nh6.nh_ifp)->chlim; return (hlim); } } return (V_ip6_defhlim); } /* * XXX: this is borrowed from in6_pcbbind(). If possible, we should * share this function by all *bsd*... */ int in6_pcbsetport(struct in6_addr *laddr, struct inpcb *inp, struct ucred *cred) { struct socket *so = inp->inp_socket; u_int16_t lport = 0; int error, lookupflags = 0; #ifdef INVARIANTS struct inpcbinfo *pcbinfo = inp->inp_pcbinfo; #endif INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); error = prison_local_ip6(cred, laddr, ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0)); if (error) return(error); /* XXX: this is redundant when called from in6_pcbbind */ if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0) lookupflags = INPLOOKUP_WILDCARD; inp->inp_flags |= INP_ANONPORT; error = in_pcb_lport(inp, NULL, &lport, cred, lookupflags); if (error != 0) return (error); inp->inp_lport = lport; if (in_pcbinshash(inp) != 0) { inp->in6p_laddr = in6addr_any; inp->inp_lport = 0; return (EAGAIN); } return (0); } void addrsel_policy_init(void) { init_policy_queue(); /* initialize the "last resort" policy */ bzero(&V_defaultaddrpolicy, sizeof(V_defaultaddrpolicy)); V_defaultaddrpolicy.label = ADDR_LABEL_NOTAPP; if (!IS_DEFAULT_VNET(curvnet)) return; ADDRSEL_LOCK_INIT(); ADDRSEL_SXLOCK_INIT(); } static struct in6_addrpolicy * lookup_addrsel_policy(struct sockaddr_in6 *key) { struct in6_addrpolicy *match = NULL; ADDRSEL_LOCK(); match = match_addrsel_policy(key); if (match == NULL) match = &V_defaultaddrpolicy; else match->use++; ADDRSEL_UNLOCK(); return (match); } /* * Subroutines to manage the address selection policy table via sysctl. */ struct walkarg { struct sysctl_req *w_req; }; static int in6_src_sysctl(SYSCTL_HANDLER_ARGS); SYSCTL_DECL(_net_inet6_ip6); static SYSCTL_NODE(_net_inet6_ip6, IPV6CTL_ADDRCTLPOLICY, addrctlpolicy, CTLFLAG_RD, in6_src_sysctl, ""); static int in6_src_sysctl(SYSCTL_HANDLER_ARGS) { struct walkarg w; if (req->newptr) return EPERM; bzero(&w, sizeof(w)); w.w_req = req; return (walk_addrsel_policy(dump_addrsel_policyent, &w)); } int in6_src_ioctl(u_long cmd, caddr_t data) { struct in6_addrpolicy ent0; if (cmd != SIOCAADDRCTL_POLICY && cmd != SIOCDADDRCTL_POLICY) return (EOPNOTSUPP); /* check for safety */ ent0 = *(struct in6_addrpolicy *)data; if (ent0.label == ADDR_LABEL_NOTAPP) return (EINVAL); /* check if the prefix mask is consecutive. */ if (in6_mask2len(&ent0.addrmask.sin6_addr, NULL) < 0) return (EINVAL); /* clear trailing garbages (if any) of the prefix address. */ IN6_MASK_ADDR(&ent0.addr.sin6_addr, &ent0.addrmask.sin6_addr); ent0.use = 0; switch (cmd) { case SIOCAADDRCTL_POLICY: return (add_addrsel_policyent(&ent0)); case SIOCDADDRCTL_POLICY: return (delete_addrsel_policyent(&ent0)); } return (0); /* XXX: compromise compilers */ } /* * The followings are implementation of the policy table using a * simple tail queue. * XXX such details should be hidden. * XXX implementation using binary tree should be more efficient. */ struct addrsel_policyent { TAILQ_ENTRY(addrsel_policyent) ape_entry; struct in6_addrpolicy ape_policy; }; TAILQ_HEAD(addrsel_policyhead, addrsel_policyent); -static VNET_DEFINE(struct addrsel_policyhead, addrsel_policytab); +VNET_DEFINE_STATIC(struct addrsel_policyhead, addrsel_policytab); #define V_addrsel_policytab VNET(addrsel_policytab) static void init_policy_queue(void) { TAILQ_INIT(&V_addrsel_policytab); } static int add_addrsel_policyent(struct in6_addrpolicy *newpolicy) { struct addrsel_policyent *new, *pol; new = malloc(sizeof(*new), M_IFADDR, M_WAITOK); ADDRSEL_XLOCK(); ADDRSEL_LOCK(); /* duplication check */ TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if (IN6_ARE_ADDR_EQUAL(&newpolicy->addr.sin6_addr, &pol->ape_policy.addr.sin6_addr) && IN6_ARE_ADDR_EQUAL(&newpolicy->addrmask.sin6_addr, &pol->ape_policy.addrmask.sin6_addr)) { ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); free(new, M_IFADDR); return (EEXIST); /* or override it? */ } } bzero(new, sizeof(*new)); /* XXX: should validate entry */ new->ape_policy = *newpolicy; TAILQ_INSERT_TAIL(&V_addrsel_policytab, new, ape_entry); ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); return (0); } static int delete_addrsel_policyent(struct in6_addrpolicy *key) { struct addrsel_policyent *pol; ADDRSEL_XLOCK(); ADDRSEL_LOCK(); /* search for the entry in the table */ TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if (IN6_ARE_ADDR_EQUAL(&key->addr.sin6_addr, &pol->ape_policy.addr.sin6_addr) && IN6_ARE_ADDR_EQUAL(&key->addrmask.sin6_addr, &pol->ape_policy.addrmask.sin6_addr)) { break; } } if (pol == NULL) { ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); return (ESRCH); } TAILQ_REMOVE(&V_addrsel_policytab, pol, ape_entry); ADDRSEL_UNLOCK(); ADDRSEL_XUNLOCK(); free(pol, M_IFADDR); return (0); } static int walk_addrsel_policy(int (*callback)(struct in6_addrpolicy *, void *), void *w) { struct addrsel_policyent *pol; int error = 0; ADDRSEL_SLOCK(); TAILQ_FOREACH(pol, &V_addrsel_policytab, ape_entry) { if ((error = (*callback)(&pol->ape_policy, w)) != 0) { ADDRSEL_SUNLOCK(); return (error); } } ADDRSEL_SUNLOCK(); return (error); } static int dump_addrsel_policyent(struct in6_addrpolicy *pol, void *arg) { int error = 0; struct walkarg *w = arg; error = SYSCTL_OUT(w->w_req, pol, sizeof(*pol)); return (error); } static struct in6_addrpolicy * match_addrsel_policy(struct sockaddr_in6 *key) { struct addrsel_policyent *pent; struct in6_addrpolicy *bestpol = NULL, *pol; int matchlen, bestmatchlen = -1; u_char *mp, *ep, *k, *p, m; TAILQ_FOREACH(pent, &V_addrsel_policytab, ape_entry) { matchlen = 0; pol = &pent->ape_policy; mp = (u_char *)&pol->addrmask.sin6_addr; ep = mp + 16; /* XXX: scope field? */ k = (u_char *)&key->sin6_addr; p = (u_char *)&pol->addr.sin6_addr; for (; mp < ep && *mp; mp++, k++, p++) { m = *mp; if ((*k & m) != *p) goto next; /* not match */ if (m == 0xff) /* short cut for a typical case */ matchlen += 8; else { while (m >= 0x80) { matchlen++; m <<= 1; } } } /* matched. check if this is better than the current best. */ if (bestpol == NULL || matchlen > bestmatchlen) { bestpol = pol; bestmatchlen = matchlen; } next: continue; } return (bestpol); } Index: head/sys/netinet6/ip6_gre.c =================================================================== --- head/sys/netinet6/ip6_gre.c (revision 336675) +++ head/sys/netinet6/ip6_gre.c (revision 336676) @@ -1,288 +1,288 @@ /*- * Copyright (c) 2014, 2018 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET #include #include #endif #include #include #include #include #include #include VNET_DEFINE(int, ip6_gre_hlim) = IPV6_DEFHLIM; #define V_ip6_gre_hlim VNET(ip6_gre_hlim) SYSCTL_DECL(_net_inet6_ip6); SYSCTL_INT(_net_inet6_ip6, OID_AUTO, grehlim, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_gre_hlim), 0, "Default hop limit for encapsulated packets"); -static VNET_DEFINE(struct gre_list *, ipv6_hashtbl) = NULL; +VNET_DEFINE_STATIC(struct gre_list *, ipv6_hashtbl) = NULL; #define V_ipv6_hashtbl VNET(ipv6_hashtbl) #define GRE_HASH(src, dst) (V_ipv6_hashtbl[\ in6_gre_hashval((src), (dst)) & (GRE_HASH_SIZE - 1)]) #define GRE_HASH_SC(sc) GRE_HASH(&(sc)->gre_oip6.ip6_src,\ &(sc)->gre_oip6.ip6_dst) static uint32_t in6_gre_hashval(const struct in6_addr *src, const struct in6_addr *dst) { uint32_t ret; ret = fnv_32_buf(src, sizeof(*src), FNV1_32_INIT); return (fnv_32_buf(dst, sizeof(*dst), ret)); } static int in6_gre_checkdup(const struct gre_softc *sc, const struct in6_addr *src, const struct in6_addr *dst) { struct gre_softc *tmp; if (sc->gre_family == AF_INET6 && IN6_ARE_ADDR_EQUAL(&sc->gre_oip6.ip6_src, src) && IN6_ARE_ADDR_EQUAL(&sc->gre_oip6.ip6_dst, dst)) return (EEXIST); CK_LIST_FOREACH(tmp, &GRE_HASH(src, dst), chain) { if (tmp == sc) continue; if (IN6_ARE_ADDR_EQUAL(&tmp->gre_oip6.ip6_src, src) && IN6_ARE_ADDR_EQUAL(&tmp->gre_oip6.ip6_dst, dst)) return (EADDRNOTAVAIL); } return (0); } static int in6_gre_lookup(const struct mbuf *m, int off, int proto, void **arg) { const struct ip6_hdr *ip6; struct gre_softc *sc; if (V_ipv6_hashtbl == NULL) return (0); MPASS(in_epoch(net_epoch_preempt)); ip6 = mtod(m, const struct ip6_hdr *); CK_LIST_FOREACH(sc, &GRE_HASH(&ip6->ip6_dst, &ip6->ip6_src), chain) { /* * This is an inbound packet, its ip6_dst is source address * in softc. */ if (IN6_ARE_ADDR_EQUAL(&sc->gre_oip6.ip6_src, &ip6->ip6_dst) && IN6_ARE_ADDR_EQUAL(&sc->gre_oip6.ip6_dst, &ip6->ip6_src)) { if ((GRE2IFP(sc)->if_flags & IFF_UP) == 0) return (0); *arg = sc; return (ENCAP_DRV_LOOKUP); } } return (0); } static void in6_gre_attach(struct gre_softc *sc) { sc->gre_hlen = sizeof(struct greip6); sc->gre_oip6.ip6_vfc = IPV6_VERSION; sc->gre_oip6.ip6_nxt = IPPROTO_GRE; gre_updatehdr(sc, &sc->gre_gi6hdr->gi6_gre); CK_LIST_INSERT_HEAD(&GRE_HASH_SC(sc), sc, chain); } void in6_gre_setopts(struct gre_softc *sc, u_long cmd, uint32_t value) { MPASS(cmd == GRESKEY || cmd == GRESOPTS); /* NOTE: we are protected with gre_ioctl_sx lock */ MPASS(sc->gre_family == AF_INET6); CK_LIST_REMOVE(sc, chain); GRE_WAIT(); if (cmd == GRESKEY) sc->gre_key = value; else sc->gre_options = value; in6_gre_attach(sc); } int in6_gre_ioctl(struct gre_softc *sc, u_long cmd, caddr_t data) { struct in6_ifreq *ifr = (struct in6_ifreq *)data; struct sockaddr_in6 *dst, *src; struct ip6_hdr *ip6; int error; /* NOTE: we are protected with gre_ioctl_sx lock */ error = EINVAL; switch (cmd) { case SIOCSIFPHYADDR_IN6: src = &((struct in6_aliasreq *)data)->ifra_addr; dst = &((struct in6_aliasreq *)data)->ifra_dstaddr; /* sanity checks */ if (src->sin6_family != dst->sin6_family || src->sin6_family != AF_INET6 || src->sin6_len != dst->sin6_len || src->sin6_len != sizeof(*src)) break; if (IN6_IS_ADDR_UNSPECIFIED(&src->sin6_addr) || IN6_IS_ADDR_UNSPECIFIED(&dst->sin6_addr)) { error = EADDRNOTAVAIL; break; } /* * Check validity of the scope zone ID of the * addresses, and convert it into the kernel * internal form if necessary. */ if ((error = sa6_embedscope(src, 0)) != 0 || (error = sa6_embedscope(dst, 0)) != 0) break; if (V_ipv6_hashtbl == NULL) V_ipv6_hashtbl = gre_hashinit(); error = in6_gre_checkdup(sc, &src->sin6_addr, &dst->sin6_addr); if (error == EADDRNOTAVAIL) break; if (error == EEXIST) { /* Addresses are the same. Just return. */ error = 0; break; } ip6 = malloc(sizeof(struct greip6) + 3 * sizeof(uint32_t), M_GRE, M_WAITOK | M_ZERO); ip6->ip6_src = src->sin6_addr; ip6->ip6_dst = dst->sin6_addr; if (sc->gre_family != 0) { /* Detach existing tunnel first */ CK_LIST_REMOVE(sc, chain); GRE_WAIT(); free(sc->gre_hdr, M_GRE); /* XXX: should we notify about link state change? */ } sc->gre_family = AF_INET6; sc->gre_hdr = ip6; sc->gre_oseq = 0; sc->gre_iseq = UINT32_MAX; in6_gre_attach(sc); break; case SIOCGIFPSRCADDR_IN6: case SIOCGIFPDSTADDR_IN6: if (sc->gre_family != AF_INET6) { error = EADDRNOTAVAIL; break; } src = (struct sockaddr_in6 *)&ifr->ifr_addr; memset(src, 0, sizeof(*src)); src->sin6_family = AF_INET6; src->sin6_len = sizeof(*src); src->sin6_addr = (cmd == SIOCGIFPSRCADDR_IN6) ? sc->gre_oip6.ip6_src: sc->gre_oip6.ip6_dst; error = prison_if(curthread->td_ucred, (struct sockaddr *)src); if (error == 0) error = sa6_recoverscope(src); if (error != 0) memset(src, 0, sizeof(*src)); break; } return (error); } int in6_gre_output(struct mbuf *m, int af __unused, int hlen __unused) { struct greip6 *gi6; gi6 = mtod(m, struct greip6 *); gi6->gi6_ip6.ip6_hlim = V_ip6_gre_hlim; return (ip6_output(m, NULL, NULL, IPV6_MINMTU, NULL, NULL, NULL)); } static const struct encaptab *ecookie = NULL; static const struct encap_config ipv6_encap_cfg = { .proto = IPPROTO_GRE, .min_length = sizeof(struct greip6) + #ifdef INET sizeof(struct ip), #else sizeof(struct ip6_hdr), #endif .exact_match = ENCAP_DRV_LOOKUP, .lookup = in6_gre_lookup, .input = gre_input }; void in6_gre_init(void) { if (!IS_DEFAULT_VNET(curvnet)) return; ecookie = ip6_encap_attach(&ipv6_encap_cfg, NULL, M_WAITOK); } void in6_gre_uninit(void) { if (IS_DEFAULT_VNET(curvnet)) ip6_encap_detach(ecookie); if (V_ipv6_hashtbl != NULL) gre_hashdestroy(V_ipv6_hashtbl); } Index: head/sys/netinet6/ip6_mroute.c =================================================================== --- head/sys/netinet6/ip6_mroute.c (revision 336675) +++ head/sys/netinet6/ip6_mroute.c (revision 336676) @@ -1,1963 +1,1963 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: ip6_mroute.c,v 1.58 2001/12/18 02:36:31 itojun Exp $ */ /*- * Copyright (c) 1989 Stephen Deering * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_mroute.c 8.2 (Berkeley) 11/15/93 * BSDI ip_mroute.c,v 2.10 1996/11/14 00:29:52 jch Exp */ /* * IP multicast forwarding procedures * * Written by David Waitzman, BBN Labs, August 1988. * Modified by Steve Deering, Stanford, February 1989. * Modified by Mark J. Steiglitz, Stanford, May, 1991 * Modified by Van Jacobson, LBL, January 1993 * Modified by Ajit Thyagarajan, PARC, August 1993 * Modified by Bill Fenner, PARC, April 1994 * * MROUTING Revision: 3.5.1.2 + PIM-SMv2 (pimd) Support */ #include __FBSDID("$FreeBSD$"); #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_MRTABLE6, "mf6c", "multicast forwarding cache entry"); static int ip6_mdq(struct mbuf *, struct ifnet *, struct mf6c *); static void phyint_send(struct ip6_hdr *, struct mif6 *, struct mbuf *); static int register_send(struct ip6_hdr *, struct mif6 *, struct mbuf *); static int set_pim6(int *); static int socket_send(struct socket *, struct mbuf *, struct sockaddr_in6 *); extern int in6_mcast_loop; extern struct domain inet6domain; static const struct encaptab *pim6_encap_cookie; static int pim6_encapcheck(const struct mbuf *, int, int, void *); static int pim6_input(struct mbuf *, int, int, void *); static const struct encap_config ipv6_encap_cfg = { .proto = IPPROTO_PIM, .min_length = sizeof(struct ip6_hdr) + PIM_MINLEN, .exact_match = 8, .check = pim6_encapcheck, .input = pim6_input }; -static VNET_DEFINE(int, ip6_mrouter_ver) = 0; +VNET_DEFINE_STATIC(int, ip6_mrouter_ver) = 0; #define V_ip6_mrouter_ver VNET(ip6_mrouter_ver) SYSCTL_DECL(_net_inet6); SYSCTL_DECL(_net_inet6_ip6); static SYSCTL_NODE(_net_inet6, IPPROTO_PIM, pim, CTLFLAG_RW, 0, "PIM"); static struct mrt6stat mrt6stat; SYSCTL_STRUCT(_net_inet6_ip6, OID_AUTO, mrt6stat, CTLFLAG_RW, &mrt6stat, mrt6stat, "Multicast Routing Statistics (struct mrt6stat, netinet6/ip6_mroute.h)"); #define MRT6STAT_INC(name) mrt6stat.name += 1 #define NO_RTE_FOUND 0x1 #define RTE_FOUND 0x2 static struct mtx mrouter6_mtx; #define MROUTER6_LOCK() mtx_lock(&mrouter6_mtx) #define MROUTER6_UNLOCK() mtx_unlock(&mrouter6_mtx) #define MROUTER6_LOCK_ASSERT() do { \ mtx_assert(&mrouter6_mtx, MA_OWNED); \ NET_ASSERT_GIANT(); \ } while (0) #define MROUTER6_LOCK_INIT() \ mtx_init(&mrouter6_mtx, "IPv6 multicast forwarding", NULL, MTX_DEF) #define MROUTER6_LOCK_DESTROY() mtx_destroy(&mrouter6_mtx) static struct mf6c *mf6ctable[MF6CTBLSIZ]; SYSCTL_OPAQUE(_net_inet6_ip6, OID_AUTO, mf6ctable, CTLFLAG_RD, &mf6ctable, sizeof(mf6ctable), "S,*mf6ctable[MF6CTBLSIZ]", "IPv6 Multicast Forwarding Table (struct *mf6ctable[MF6CTBLSIZ], " "netinet6/ip6_mroute.h)"); static struct mtx mfc6_mtx; #define MFC6_LOCK() mtx_lock(&mfc6_mtx) #define MFC6_UNLOCK() mtx_unlock(&mfc6_mtx) #define MFC6_LOCK_ASSERT() do { \ mtx_assert(&mfc6_mtx, MA_OWNED); \ NET_ASSERT_GIANT(); \ } while (0) #define MFC6_LOCK_INIT() \ mtx_init(&mfc6_mtx, "IPv6 multicast forwarding cache", NULL, MTX_DEF) #define MFC6_LOCK_DESTROY() mtx_destroy(&mfc6_mtx) static u_char n6expire[MF6CTBLSIZ]; static struct mif6 mif6table[MAXMIFS]; static int sysctl_mif6table(SYSCTL_HANDLER_ARGS) { struct mif6_sctl *out; int error; out = malloc(sizeof(struct mif6_sctl) * MAXMIFS, M_TEMP, M_WAITOK); for (int i = 0; i < MAXMIFS; i++) { out[i].m6_flags = mif6table[i].m6_flags; out[i].m6_rate_limit = mif6table[i].m6_rate_limit; out[i].m6_lcl_addr = mif6table[i].m6_lcl_addr; if (mif6table[i].m6_ifp != NULL) out[i].m6_ifp = mif6table[i].m6_ifp->if_index; else out[i].m6_ifp = 0; out[i].m6_pkt_in = mif6table[i].m6_pkt_in; out[i].m6_pkt_out = mif6table[i].m6_pkt_out; out[i].m6_bytes_in = mif6table[i].m6_bytes_in; out[i].m6_bytes_out = mif6table[i].m6_bytes_out; } error = SYSCTL_OUT(req, out, sizeof(struct mif6_sctl) * MAXMIFS); free(out, M_TEMP); return (error); } SYSCTL_PROC(_net_inet6_ip6, OID_AUTO, mif6table, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, sysctl_mif6table, "S,mif6_sctl[MAXMIFS]", "IPv6 Multicast Interfaces (struct mif6_sctl[MAXMIFS], " "netinet6/ip6_mroute.h)"); static struct mtx mif6_mtx; #define MIF6_LOCK() mtx_lock(&mif6_mtx) #define MIF6_UNLOCK() mtx_unlock(&mif6_mtx) #define MIF6_LOCK_ASSERT() mtx_assert(&mif6_mtx, MA_OWNED) #define MIF6_LOCK_INIT() \ mtx_init(&mif6_mtx, "IPv6 multicast interfaces", NULL, MTX_DEF) #define MIF6_LOCK_DESTROY() mtx_destroy(&mif6_mtx) #ifdef MRT6DEBUG -static VNET_DEFINE(u_int, mrt6debug) = 0; /* debug level */ +VNET_DEFINE_STATIC(u_int, mrt6debug) = 0; /* debug level */ #define V_mrt6debug VNET(mrt6debug) #define DEBUG_MFC 0x02 #define DEBUG_FORWARD 0x04 #define DEBUG_EXPIRE 0x08 #define DEBUG_XMIT 0x10 #define DEBUG_REG 0x20 #define DEBUG_PIM 0x40 #define DEBUG_ERR 0x80 #define DEBUG_ANY 0x7f #define MRT6_DLOG(m, fmt, ...) \ if (V_mrt6debug & (m)) \ log(((m) & DEBUG_ERR) ? LOG_ERR: LOG_DEBUG, \ "%s: " fmt "\n", __func__, ##__VA_ARGS__) #else #define MRT6_DLOG(m, fmt, ...) #endif static void expire_upcalls(void *); #define EXPIRE_TIMEOUT (hz / 4) /* 4x / second */ #define UPCALL_EXPIRE 6 /* number of timeouts */ /* * XXX TODO: maintain a count to if_allmulti() calls in struct ifnet. */ /* * 'Interfaces' associated with decapsulator (so we can tell * packets that went through it from ones that get reflected * by a broken gateway). Different from IPv4 register_if, * these interfaces are linked into the system ifnet list, * because per-interface IPv6 statistics are maintained in * ifp->if_afdata. But it does not have any routes point * to them. I.e., packets can't be sent this way. They * only exist as a placeholder for multicast source * verification. */ static struct ifnet *multicast_register_if6; #define ENCAP_HOPS 64 /* * Private variables. */ static mifi_t nummifs = 0; static mifi_t reg_mif_num = (mifi_t)-1; static struct pim6stat pim6stat; SYSCTL_STRUCT(_net_inet6_pim, PIM6CTL_STATS, stats, CTLFLAG_RW, &pim6stat, pim6stat, "PIM Statistics (struct pim6stat, netinet6/pim6_var.h)"); #define PIM6STAT_INC(name) pim6stat.name += 1 -static VNET_DEFINE(int, pim6); +VNET_DEFINE_STATIC(int, pim6); #define V_pim6 VNET(pim6) /* * Hash function for a source, group entry */ #define MF6CHASH(a, g) MF6CHASHMOD((a).s6_addr32[0] ^ (a).s6_addr32[1] ^ \ (a).s6_addr32[2] ^ (a).s6_addr32[3] ^ \ (g).s6_addr32[0] ^ (g).s6_addr32[1] ^ \ (g).s6_addr32[2] ^ (g).s6_addr32[3]) /* * Find a route for a given origin IPv6 address and Multicast group address. */ #define MF6CFIND(o, g, rt) do { \ struct mf6c *_rt = mf6ctable[MF6CHASH(o,g)]; \ rt = NULL; \ while (_rt) { \ if (IN6_ARE_ADDR_EQUAL(&_rt->mf6c_origin.sin6_addr, &(o)) && \ IN6_ARE_ADDR_EQUAL(&_rt->mf6c_mcastgrp.sin6_addr, &(g)) && \ (_rt->mf6c_stall == NULL)) { \ rt = _rt; \ break; \ } \ _rt = _rt->mf6c_next; \ } \ if (rt == NULL) { \ MRT6STAT_INC(mrt6s_mfc_misses); \ } \ } while (/*CONSTCOND*/ 0) /* * Macros to compute elapsed time efficiently * Borrowed from Van Jacobson's scheduling code * XXX: replace with timersub() ? */ #define TV_DELTA(a, b, delta) do { \ int xxs; \ \ delta = (a).tv_usec - (b).tv_usec; \ if ((xxs = (a).tv_sec - (b).tv_sec)) { \ switch (xxs) { \ case 2: \ delta += 1000000; \ /* FALLTHROUGH */ \ case 1: \ delta += 1000000; \ break; \ default: \ delta += (1000000 * xxs); \ } \ } \ } while (/*CONSTCOND*/ 0) /* XXX: replace with timercmp(a, b, <) ? */ #define TV_LT(a, b) (((a).tv_usec < (b).tv_usec && \ (a).tv_sec <= (b).tv_sec) || (a).tv_sec < (b).tv_sec) #ifdef UPCALL_TIMING #define UPCALL_MAX 50 static u_long upcall_data[UPCALL_MAX + 1]; static void collate(); #endif /* UPCALL_TIMING */ static int ip6_mrouter_init(struct socket *, int, int); static int add_m6fc(struct mf6cctl *); static int add_m6if(struct mif6ctl *); static int del_m6fc(struct mf6cctl *); static int del_m6if(mifi_t *); static int del_m6if_locked(mifi_t *); static int get_mif6_cnt(struct sioc_mif_req6 *); static int get_sg_cnt(struct sioc_sg_req6 *); static struct callout expire_upcalls_ch; int X_ip6_mforward(struct ip6_hdr *, struct ifnet *, struct mbuf *); int X_ip6_mrouter_done(void); int X_ip6_mrouter_set(struct socket *, struct sockopt *); int X_ip6_mrouter_get(struct socket *, struct sockopt *); int X_mrt6_ioctl(u_long, caddr_t); /* * Handle MRT setsockopt commands to modify the multicast routing tables. */ int X_ip6_mrouter_set(struct socket *so, struct sockopt *sopt) { int error = 0; int optval; struct mif6ctl mifc; struct mf6cctl mfcc; mifi_t mifi; if (so != V_ip6_mrouter && sopt->sopt_name != MRT6_INIT) return (EPERM); switch (sopt->sopt_name) { case MRT6_INIT: #ifdef MRT6_OINIT case MRT6_OINIT: #endif error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; error = ip6_mrouter_init(so, optval, sopt->sopt_name); break; case MRT6_DONE: error = X_ip6_mrouter_done(); break; case MRT6_ADD_MIF: error = sooptcopyin(sopt, &mifc, sizeof(mifc), sizeof(mifc)); if (error) break; error = add_m6if(&mifc); break; case MRT6_ADD_MFC: error = sooptcopyin(sopt, &mfcc, sizeof(mfcc), sizeof(mfcc)); if (error) break; error = add_m6fc(&mfcc); break; case MRT6_DEL_MFC: error = sooptcopyin(sopt, &mfcc, sizeof(mfcc), sizeof(mfcc)); if (error) break; error = del_m6fc(&mfcc); break; case MRT6_DEL_MIF: error = sooptcopyin(sopt, &mifi, sizeof(mifi), sizeof(mifi)); if (error) break; error = del_m6if(&mifi); break; case MRT6_PIM: error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; error = set_pim6(&optval); break; default: error = EOPNOTSUPP; break; } return (error); } /* * Handle MRT getsockopt commands */ int X_ip6_mrouter_get(struct socket *so, struct sockopt *sopt) { int error = 0; if (so != V_ip6_mrouter) return (EACCES); switch (sopt->sopt_name) { case MRT6_PIM: error = sooptcopyout(sopt, &V_pim6, sizeof(V_pim6)); break; } return (error); } /* * Handle ioctl commands to obtain information from the cache */ int X_mrt6_ioctl(u_long cmd, caddr_t data) { int ret; ret = EINVAL; switch (cmd) { case SIOCGETSGCNT_IN6: ret = get_sg_cnt((struct sioc_sg_req6 *)data); break; case SIOCGETMIFCNT_IN6: ret = get_mif6_cnt((struct sioc_mif_req6 *)data); break; default: break; } return (ret); } /* * returns the packet, byte, rpf-failure count for the source group provided */ static int get_sg_cnt(struct sioc_sg_req6 *req) { struct mf6c *rt; int ret; ret = 0; MFC6_LOCK(); MF6CFIND(req->src.sin6_addr, req->grp.sin6_addr, rt); if (rt == NULL) { ret = ESRCH; } else { req->pktcnt = rt->mf6c_pkt_cnt; req->bytecnt = rt->mf6c_byte_cnt; req->wrong_if = rt->mf6c_wrong_if; } MFC6_UNLOCK(); return (ret); } /* * returns the input and output packet and byte counts on the mif provided */ static int get_mif6_cnt(struct sioc_mif_req6 *req) { mifi_t mifi; int ret; ret = 0; mifi = req->mifi; MIF6_LOCK(); if (mifi >= nummifs) { ret = EINVAL; } else { req->icount = mif6table[mifi].m6_pkt_in; req->ocount = mif6table[mifi].m6_pkt_out; req->ibytes = mif6table[mifi].m6_bytes_in; req->obytes = mif6table[mifi].m6_bytes_out; } MIF6_UNLOCK(); return (ret); } static int set_pim6(int *i) { if ((*i != 1) && (*i != 0)) return (EINVAL); V_pim6 = *i; return (0); } /* * Enable multicast routing */ static int ip6_mrouter_init(struct socket *so, int v, int cmd) { MRT6_DLOG(DEBUG_ANY, "so_type = %d, pr_protocol = %d", so->so_type, so->so_proto->pr_protocol); if (so->so_type != SOCK_RAW || so->so_proto->pr_protocol != IPPROTO_ICMPV6) return (EOPNOTSUPP); if (v != 1) return (ENOPROTOOPT); MROUTER6_LOCK(); if (V_ip6_mrouter != NULL) { MROUTER6_UNLOCK(); return (EADDRINUSE); } V_ip6_mrouter = so; V_ip6_mrouter_ver = cmd; bzero((caddr_t)mf6ctable, sizeof(mf6ctable)); bzero((caddr_t)n6expire, sizeof(n6expire)); V_pim6 = 0;/* used for stubbing out/in pim stuff */ callout_init(&expire_upcalls_ch, 0); callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, NULL); MROUTER6_UNLOCK(); MRT6_DLOG(DEBUG_ANY, "finished"); return (0); } /* * Disable IPv6 multicast forwarding. */ int X_ip6_mrouter_done(void) { mifi_t mifi; u_long i; struct mf6c *rt; struct rtdetq *rte; MROUTER6_LOCK(); if (V_ip6_mrouter == NULL) { MROUTER6_UNLOCK(); return (EINVAL); } /* * For each phyint in use, disable promiscuous reception of all IPv6 * multicasts. */ for (mifi = 0; mifi < nummifs; mifi++) { if (mif6table[mifi].m6_ifp && !(mif6table[mifi].m6_flags & MIFF_REGISTER)) { if_allmulti(mif6table[mifi].m6_ifp, 0); } } bzero((caddr_t)mif6table, sizeof(mif6table)); nummifs = 0; V_pim6 = 0; /* used to stub out/in pim specific code */ callout_stop(&expire_upcalls_ch); /* * Free all multicast forwarding cache entries. */ MFC6_LOCK(); for (i = 0; i < MF6CTBLSIZ; i++) { rt = mf6ctable[i]; while (rt) { struct mf6c *frt; for (rte = rt->mf6c_stall; rte != NULL; ) { struct rtdetq *n = rte->next; m_freem(rte->m); free(rte, M_MRTABLE6); rte = n; } frt = rt; rt = rt->mf6c_next; free(frt, M_MRTABLE6); } } bzero((caddr_t)mf6ctable, sizeof(mf6ctable)); MFC6_UNLOCK(); /* * Reset register interface */ if (reg_mif_num != (mifi_t)-1 && multicast_register_if6 != NULL) { if_detach(multicast_register_if6); if_free(multicast_register_if6); reg_mif_num = (mifi_t)-1; multicast_register_if6 = NULL; } V_ip6_mrouter = NULL; V_ip6_mrouter_ver = 0; MROUTER6_UNLOCK(); MRT6_DLOG(DEBUG_ANY, "finished"); return (0); } static struct sockaddr_in6 sin6 = { sizeof(sin6), AF_INET6 }; /* * Add a mif to the mif table */ static int add_m6if(struct mif6ctl *mifcp) { struct mif6 *mifp; struct ifnet *ifp; int error; MIF6_LOCK(); if (mifcp->mif6c_mifi >= MAXMIFS) { MIF6_UNLOCK(); return (EINVAL); } mifp = mif6table + mifcp->mif6c_mifi; if (mifp->m6_ifp != NULL) { MIF6_UNLOCK(); return (EADDRINUSE); /* XXX: is it appropriate? */ } if (mifcp->mif6c_pifi == 0 || mifcp->mif6c_pifi > V_if_index) { MIF6_UNLOCK(); return (ENXIO); } ifp = ifnet_byindex(mifcp->mif6c_pifi); if (mifcp->mif6c_flags & MIFF_REGISTER) { if (reg_mif_num == (mifi_t)-1) { ifp = if_alloc(IFT_OTHER); if_initname(ifp, "register_mif", 0); ifp->if_flags |= IFF_LOOPBACK; if_attach(ifp); multicast_register_if6 = ifp; reg_mif_num = mifcp->mif6c_mifi; /* * it is impossible to guess the ifindex of the * register interface. So mif6c_pifi is automatically * calculated. */ mifcp->mif6c_pifi = ifp->if_index; } else { ifp = multicast_register_if6; } } else { /* Make sure the interface supports multicast */ if ((ifp->if_flags & IFF_MULTICAST) == 0) { MIF6_UNLOCK(); return (EOPNOTSUPP); } error = if_allmulti(ifp, 1); if (error) { MIF6_UNLOCK(); return (error); } } mifp->m6_flags = mifcp->mif6c_flags; mifp->m6_ifp = ifp; /* initialize per mif pkt counters */ mifp->m6_pkt_in = 0; mifp->m6_pkt_out = 0; mifp->m6_bytes_in = 0; mifp->m6_bytes_out = 0; /* Adjust nummifs up if the mifi is higher than nummifs */ if (nummifs <= mifcp->mif6c_mifi) nummifs = mifcp->mif6c_mifi + 1; MIF6_UNLOCK(); MRT6_DLOG(DEBUG_ANY, "mif #%d, phyint %s", mifcp->mif6c_mifi, if_name(ifp)); return (0); } /* * Delete a mif from the mif table */ static int del_m6if_locked(mifi_t *mifip) { struct mif6 *mifp = mif6table + *mifip; mifi_t mifi; struct ifnet *ifp; MIF6_LOCK_ASSERT(); if (*mifip >= nummifs) return (EINVAL); if (mifp->m6_ifp == NULL) return (EINVAL); if (!(mifp->m6_flags & MIFF_REGISTER)) { /* XXX: TODO: Maintain an ALLMULTI refcount in struct ifnet. */ ifp = mifp->m6_ifp; if_allmulti(ifp, 0); } else { if (reg_mif_num != (mifi_t)-1 && multicast_register_if6 != NULL) { if_detach(multicast_register_if6); if_free(multicast_register_if6); reg_mif_num = (mifi_t)-1; multicast_register_if6 = NULL; } } bzero((caddr_t)mifp, sizeof(*mifp)); /* Adjust nummifs down */ for (mifi = nummifs; mifi > 0; mifi--) if (mif6table[mifi - 1].m6_ifp) break; nummifs = mifi; MRT6_DLOG(DEBUG_ANY, "mif %d, nummifs %d", *mifip, nummifs); return (0); } static int del_m6if(mifi_t *mifip) { int cc; MIF6_LOCK(); cc = del_m6if_locked(mifip); MIF6_UNLOCK(); return (cc); } /* * Add an mfc entry */ static int add_m6fc(struct mf6cctl *mfccp) { struct mf6c *rt; u_long hash; struct rtdetq *rte; u_short nstl; char ip6bufo[INET6_ADDRSTRLEN], ip6bufg[INET6_ADDRSTRLEN]; MFC6_LOCK(); MF6CFIND(mfccp->mf6cc_origin.sin6_addr, mfccp->mf6cc_mcastgrp.sin6_addr, rt); /* If an entry already exists, just update the fields */ if (rt) { MRT6_DLOG(DEBUG_MFC, "no upcall o %s g %s p %x", ip6_sprintf(ip6bufo, &mfccp->mf6cc_origin.sin6_addr), ip6_sprintf(ip6bufg, &mfccp->mf6cc_mcastgrp.sin6_addr), mfccp->mf6cc_parent); rt->mf6c_parent = mfccp->mf6cc_parent; rt->mf6c_ifset = mfccp->mf6cc_ifset; MFC6_UNLOCK(); return (0); } /* * Find the entry for which the upcall was made and update */ hash = MF6CHASH(mfccp->mf6cc_origin.sin6_addr, mfccp->mf6cc_mcastgrp.sin6_addr); for (rt = mf6ctable[hash], nstl = 0; rt; rt = rt->mf6c_next) { if (IN6_ARE_ADDR_EQUAL(&rt->mf6c_origin.sin6_addr, &mfccp->mf6cc_origin.sin6_addr) && IN6_ARE_ADDR_EQUAL(&rt->mf6c_mcastgrp.sin6_addr, &mfccp->mf6cc_mcastgrp.sin6_addr) && (rt->mf6c_stall != NULL)) { if (nstl++) log(LOG_ERR, "add_m6fc: %s o %s g %s p %x dbx %p\n", "multiple kernel entries", ip6_sprintf(ip6bufo, &mfccp->mf6cc_origin.sin6_addr), ip6_sprintf(ip6bufg, &mfccp->mf6cc_mcastgrp.sin6_addr), mfccp->mf6cc_parent, rt->mf6c_stall); MRT6_DLOG(DEBUG_MFC, "o %s g %s p %x dbg %p", ip6_sprintf(ip6bufo, &mfccp->mf6cc_origin.sin6_addr), ip6_sprintf(ip6bufg, &mfccp->mf6cc_mcastgrp.sin6_addr), mfccp->mf6cc_parent, rt->mf6c_stall); rt->mf6c_origin = mfccp->mf6cc_origin; rt->mf6c_mcastgrp = mfccp->mf6cc_mcastgrp; rt->mf6c_parent = mfccp->mf6cc_parent; rt->mf6c_ifset = mfccp->mf6cc_ifset; /* initialize pkt counters per src-grp */ rt->mf6c_pkt_cnt = 0; rt->mf6c_byte_cnt = 0; rt->mf6c_wrong_if = 0; rt->mf6c_expire = 0; /* Don't clean this guy up */ n6expire[hash]--; /* free packets Qed at the end of this entry */ for (rte = rt->mf6c_stall; rte != NULL; ) { struct rtdetq *n = rte->next; ip6_mdq(rte->m, rte->ifp, rt); m_freem(rte->m); #ifdef UPCALL_TIMING collate(&(rte->t)); #endif /* UPCALL_TIMING */ free(rte, M_MRTABLE6); rte = n; } rt->mf6c_stall = NULL; } } /* * It is possible that an entry is being inserted without an upcall */ if (nstl == 0) { MRT6_DLOG(DEBUG_MFC, "no upcall h %lu o %s g %s p %x", hash, ip6_sprintf(ip6bufo, &mfccp->mf6cc_origin.sin6_addr), ip6_sprintf(ip6bufg, &mfccp->mf6cc_mcastgrp.sin6_addr), mfccp->mf6cc_parent); for (rt = mf6ctable[hash]; rt; rt = rt->mf6c_next) { if (IN6_ARE_ADDR_EQUAL(&rt->mf6c_origin.sin6_addr, &mfccp->mf6cc_origin.sin6_addr)&& IN6_ARE_ADDR_EQUAL(&rt->mf6c_mcastgrp.sin6_addr, &mfccp->mf6cc_mcastgrp.sin6_addr)) { rt->mf6c_origin = mfccp->mf6cc_origin; rt->mf6c_mcastgrp = mfccp->mf6cc_mcastgrp; rt->mf6c_parent = mfccp->mf6cc_parent; rt->mf6c_ifset = mfccp->mf6cc_ifset; /* initialize pkt counters per src-grp */ rt->mf6c_pkt_cnt = 0; rt->mf6c_byte_cnt = 0; rt->mf6c_wrong_if = 0; if (rt->mf6c_expire) n6expire[hash]--; rt->mf6c_expire = 0; } } if (rt == NULL) { /* no upcall, so make a new entry */ rt = (struct mf6c *)malloc(sizeof(*rt), M_MRTABLE6, M_NOWAIT); if (rt == NULL) { MFC6_UNLOCK(); return (ENOBUFS); } /* insert new entry at head of hash chain */ rt->mf6c_origin = mfccp->mf6cc_origin; rt->mf6c_mcastgrp = mfccp->mf6cc_mcastgrp; rt->mf6c_parent = mfccp->mf6cc_parent; rt->mf6c_ifset = mfccp->mf6cc_ifset; /* initialize pkt counters per src-grp */ rt->mf6c_pkt_cnt = 0; rt->mf6c_byte_cnt = 0; rt->mf6c_wrong_if = 0; rt->mf6c_expire = 0; rt->mf6c_stall = NULL; /* link into table */ rt->mf6c_next = mf6ctable[hash]; mf6ctable[hash] = rt; } } MFC6_UNLOCK(); return (0); } #ifdef UPCALL_TIMING /* * collect delay statistics on the upcalls */ static void collate(struct timeval *t) { u_long d; struct timeval tp; u_long delta; GET_TIME(tp); if (TV_LT(*t, tp)) { TV_DELTA(tp, *t, delta); d = delta >> 10; if (d > UPCALL_MAX) d = UPCALL_MAX; ++upcall_data[d]; } } #endif /* UPCALL_TIMING */ /* * Delete an mfc entry */ static int del_m6fc(struct mf6cctl *mfccp) { #ifdef MRT6DEBUG char ip6bufo[INET6_ADDRSTRLEN], ip6bufg[INET6_ADDRSTRLEN]; #endif struct sockaddr_in6 origin; struct sockaddr_in6 mcastgrp; struct mf6c *rt; struct mf6c **nptr; u_long hash; origin = mfccp->mf6cc_origin; mcastgrp = mfccp->mf6cc_mcastgrp; hash = MF6CHASH(origin.sin6_addr, mcastgrp.sin6_addr); MRT6_DLOG(DEBUG_MFC, "orig %s mcastgrp %s", ip6_sprintf(ip6bufo, &origin.sin6_addr), ip6_sprintf(ip6bufg, &mcastgrp.sin6_addr)); MFC6_LOCK(); nptr = &mf6ctable[hash]; while ((rt = *nptr) != NULL) { if (IN6_ARE_ADDR_EQUAL(&origin.sin6_addr, &rt->mf6c_origin.sin6_addr) && IN6_ARE_ADDR_EQUAL(&mcastgrp.sin6_addr, &rt->mf6c_mcastgrp.sin6_addr) && rt->mf6c_stall == NULL) break; nptr = &rt->mf6c_next; } if (rt == NULL) { MFC6_UNLOCK(); return (EADDRNOTAVAIL); } *nptr = rt->mf6c_next; free(rt, M_MRTABLE6); MFC6_UNLOCK(); return (0); } static int socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in6 *src) { if (s) { if (sbappendaddr(&s->so_rcv, (struct sockaddr *)src, mm, (struct mbuf *)0) != 0) { sorwakeup(s); return (0); } } m_freem(mm); return (-1); } /* * IPv6 multicast forwarding function. This function assumes that the packet * pointed to by "ip6" has arrived on (or is about to be sent to) the interface * pointed to by "ifp", and the packet is to be relayed to other networks * that have members of the packet's destination IPv6 multicast group. * * The packet is returned unscathed to the caller, unless it is * erroneous, in which case a non-zero return value tells the caller to * discard it. * * NOTE: this implementation assumes that m->m_pkthdr.rcvif is NULL iff * this function is called in the originating context (i.e., not when * forwarding a packet from other node). ip6_output(), which is currently the * only function that calls this function is called in the originating context, * explicitly ensures this condition. It is caller's responsibility to ensure * that if this function is called from somewhere else in the originating * context in the future. */ int X_ip6_mforward(struct ip6_hdr *ip6, struct ifnet *ifp, struct mbuf *m) { struct rtdetq *rte; struct mbuf *mb0; struct mf6c *rt; struct mif6 *mifp; struct mbuf *mm; u_long hash; mifi_t mifi; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; #ifdef UPCALL_TIMING struct timeval tp; GET_TIME(tp); #endif /* UPCALL_TIMING */ MRT6_DLOG(DEBUG_FORWARD, "src %s, dst %s, ifindex %d", ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), ifp->if_index); /* * Don't forward a packet with Hop limit of zero or one, * or a packet destined to a local-only group. */ if (ip6->ip6_hlim <= 1 || IN6_IS_ADDR_MC_INTFACELOCAL(&ip6->ip6_dst) || IN6_IS_ADDR_MC_LINKLOCAL(&ip6->ip6_dst)) return (0); ip6->ip6_hlim--; /* * Source address check: do not forward packets with unspecified * source. It was discussed in July 2000, on ipngwg mailing list. * This is rather more serious than unicast cases, because some * MLD packets can be sent with the unspecified source address * (although such packets must normally set 1 to the hop limit field). */ if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { IP6STAT_INC(ip6s_cantforward); if (V_ip6_log_time + V_ip6_log_interval < time_uptime) { V_ip6_log_time = time_uptime; log(LOG_DEBUG, "cannot forward " "from %s to %s nxt %d received on %s\n", ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), ip6->ip6_nxt, if_name(m->m_pkthdr.rcvif)); } return (0); } MFC6_LOCK(); /* * Determine forwarding mifs from the forwarding cache table */ MF6CFIND(ip6->ip6_src, ip6->ip6_dst, rt); MRT6STAT_INC(mrt6s_mfc_lookups); /* Entry exists, so forward if necessary */ if (rt) { MFC6_UNLOCK(); return (ip6_mdq(m, ifp, rt)); } /* * If we don't have a route for packet's origin, * Make a copy of the packet & send message to routing daemon. */ MRT6STAT_INC(mrt6s_no_route); MRT6_DLOG(DEBUG_FORWARD | DEBUG_MFC, "no rte s %s g %s", ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst)); /* * Allocate mbufs early so that we don't do extra work if we * are just going to fail anyway. */ rte = (struct rtdetq *)malloc(sizeof(*rte), M_MRTABLE6, M_NOWAIT); if (rte == NULL) { MFC6_UNLOCK(); return (ENOBUFS); } mb0 = m_copym(m, 0, M_COPYALL, M_NOWAIT); /* * Pullup packet header if needed before storing it, * as other references may modify it in the meantime. */ if (mb0 && (!M_WRITABLE(mb0) || mb0->m_len < sizeof(struct ip6_hdr))) mb0 = m_pullup(mb0, sizeof(struct ip6_hdr)); if (mb0 == NULL) { free(rte, M_MRTABLE6); MFC6_UNLOCK(); return (ENOBUFS); } /* is there an upcall waiting for this packet? */ hash = MF6CHASH(ip6->ip6_src, ip6->ip6_dst); for (rt = mf6ctable[hash]; rt; rt = rt->mf6c_next) { if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &rt->mf6c_origin.sin6_addr) && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &rt->mf6c_mcastgrp.sin6_addr) && (rt->mf6c_stall != NULL)) break; } if (rt == NULL) { struct mrt6msg *im; #ifdef MRT6_OINIT struct omrt6msg *oim; #endif /* no upcall, so make a new entry */ rt = (struct mf6c *)malloc(sizeof(*rt), M_MRTABLE6, M_NOWAIT); if (rt == NULL) { free(rte, M_MRTABLE6); m_freem(mb0); MFC6_UNLOCK(); return (ENOBUFS); } /* * Make a copy of the header to send to the user * level process */ mm = m_copym(mb0, 0, sizeof(struct ip6_hdr), M_NOWAIT); if (mm == NULL) { free(rte, M_MRTABLE6); m_freem(mb0); free(rt, M_MRTABLE6); MFC6_UNLOCK(); return (ENOBUFS); } /* * Send message to routing daemon */ sin6.sin6_addr = ip6->ip6_src; im = NULL; #ifdef MRT6_OINIT oim = NULL; #endif switch (V_ip6_mrouter_ver) { #ifdef MRT6_OINIT case MRT6_OINIT: oim = mtod(mm, struct omrt6msg *); oim->im6_msgtype = MRT6MSG_NOCACHE; oim->im6_mbz = 0; break; #endif case MRT6_INIT: im = mtod(mm, struct mrt6msg *); im->im6_msgtype = MRT6MSG_NOCACHE; im->im6_mbz = 0; break; default: free(rte, M_MRTABLE6); m_freem(mb0); free(rt, M_MRTABLE6); MFC6_UNLOCK(); return (EINVAL); } MRT6_DLOG(DEBUG_FORWARD, "getting the iif info in the kernel"); for (mifp = mif6table, mifi = 0; mifi < nummifs && mifp->m6_ifp != ifp; mifp++, mifi++) ; switch (V_ip6_mrouter_ver) { #ifdef MRT6_OINIT case MRT6_OINIT: oim->im6_mif = mifi; break; #endif case MRT6_INIT: im->im6_mif = mifi; break; } if (socket_send(V_ip6_mrouter, mm, &sin6) < 0) { log(LOG_WARNING, "ip6_mforward: ip6_mrouter " "socket queue full\n"); MRT6STAT_INC(mrt6s_upq_sockfull); free(rte, M_MRTABLE6); m_freem(mb0); free(rt, M_MRTABLE6); MFC6_UNLOCK(); return (ENOBUFS); } MRT6STAT_INC(mrt6s_upcalls); /* insert new entry at head of hash chain */ bzero(rt, sizeof(*rt)); rt->mf6c_origin.sin6_family = AF_INET6; rt->mf6c_origin.sin6_len = sizeof(struct sockaddr_in6); rt->mf6c_origin.sin6_addr = ip6->ip6_src; rt->mf6c_mcastgrp.sin6_family = AF_INET6; rt->mf6c_mcastgrp.sin6_len = sizeof(struct sockaddr_in6); rt->mf6c_mcastgrp.sin6_addr = ip6->ip6_dst; rt->mf6c_expire = UPCALL_EXPIRE; n6expire[hash]++; rt->mf6c_parent = MF6C_INCOMPLETE_PARENT; /* link into table */ rt->mf6c_next = mf6ctable[hash]; mf6ctable[hash] = rt; /* Add this entry to the end of the queue */ rt->mf6c_stall = rte; } else { /* determine if q has overflowed */ struct rtdetq **p; int npkts = 0; for (p = &rt->mf6c_stall; *p != NULL; p = &(*p)->next) if (++npkts > MAX_UPQ6) { MRT6STAT_INC(mrt6s_upq_ovflw); free(rte, M_MRTABLE6); m_freem(mb0); MFC6_UNLOCK(); return (0); } /* Add this entry to the end of the queue */ *p = rte; } rte->next = NULL; rte->m = mb0; rte->ifp = ifp; #ifdef UPCALL_TIMING rte->t = tp; #endif /* UPCALL_TIMING */ MFC6_UNLOCK(); return (0); } /* * Clean up cache entries if upcalls are not serviced * Call from the Slow Timeout mechanism, every half second. */ static void expire_upcalls(void *unused) { #ifdef MRT6DEBUG char ip6bufo[INET6_ADDRSTRLEN], ip6bufg[INET6_ADDRSTRLEN]; #endif struct rtdetq *rte; struct mf6c *mfc, **nptr; u_long i; MFC6_LOCK(); for (i = 0; i < MF6CTBLSIZ; i++) { if (n6expire[i] == 0) continue; nptr = &mf6ctable[i]; while ((mfc = *nptr) != NULL) { rte = mfc->mf6c_stall; /* * Skip real cache entries * Make sure it wasn't marked to not expire (shouldn't happen) * If it expires now */ if (rte != NULL && mfc->mf6c_expire != 0 && --mfc->mf6c_expire == 0) { MRT6_DLOG(DEBUG_EXPIRE, "expiring (%s %s)", ip6_sprintf(ip6bufo, &mfc->mf6c_origin.sin6_addr), ip6_sprintf(ip6bufg, &mfc->mf6c_mcastgrp.sin6_addr)); /* * drop all the packets * free the mbuf with the pkt, if, timing info */ do { struct rtdetq *n = rte->next; m_freem(rte->m); free(rte, M_MRTABLE6); rte = n; } while (rte != NULL); MRT6STAT_INC(mrt6s_cache_cleanups); n6expire[i]--; *nptr = mfc->mf6c_next; free(mfc, M_MRTABLE6); } else { nptr = &mfc->mf6c_next; } } } MFC6_UNLOCK(); callout_reset(&expire_upcalls_ch, EXPIRE_TIMEOUT, expire_upcalls, NULL); } /* * Packet forwarding routine once entry in the cache is made */ static int ip6_mdq(struct mbuf *m, struct ifnet *ifp, struct mf6c *rt) { struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); mifi_t mifi, iif; struct mif6 *mifp; int plen = m->m_pkthdr.len; struct in6_addr src0, dst0; /* copies for local work */ u_int32_t iszone, idzone, oszone, odzone; int error = 0; /* * Macro to send packet on mif. Since RSVP packets don't get counted on * input, they shouldn't get counted on output, so statistics keeping is * separate. */ #define MC6_SEND(ip6, mifp, m) do { \ if ((mifp)->m6_flags & MIFF_REGISTER) \ register_send((ip6), (mifp), (m)); \ else \ phyint_send((ip6), (mifp), (m)); \ } while (/*CONSTCOND*/ 0) /* * Don't forward if it didn't arrive from the parent mif * for its origin. */ mifi = rt->mf6c_parent; if ((mifi >= nummifs) || (mif6table[mifi].m6_ifp != ifp)) { /* came in the wrong interface */ MRT6_DLOG(DEBUG_FORWARD, "wrong if: ifid %d mifi %d mififid %x", ifp->if_index, mifi, mif6table[mifi].m6_ifp->if_index); MRT6STAT_INC(mrt6s_wrong_if); rt->mf6c_wrong_if++; /* * If we are doing PIM processing, and we are forwarding * packets on this interface, send a message to the * routing daemon. */ /* have to make sure this is a valid mif */ if (mifi < nummifs && mif6table[mifi].m6_ifp) if (V_pim6 && (m->m_flags & M_LOOP) == 0) { /* * Check the M_LOOP flag to avoid an * unnecessary PIM assert. * XXX: M_LOOP is an ad-hoc hack... */ static struct sockaddr_in6 sin6 = { sizeof(sin6), AF_INET6 }; struct mbuf *mm; struct mrt6msg *im; #ifdef MRT6_OINIT struct omrt6msg *oim; #endif mm = m_copym(m, 0, sizeof(struct ip6_hdr), M_NOWAIT); if (mm && (!M_WRITABLE(mm) || mm->m_len < sizeof(struct ip6_hdr))) mm = m_pullup(mm, sizeof(struct ip6_hdr)); if (mm == NULL) return (ENOBUFS); #ifdef MRT6_OINIT oim = NULL; #endif im = NULL; switch (V_ip6_mrouter_ver) { #ifdef MRT6_OINIT case MRT6_OINIT: oim = mtod(mm, struct omrt6msg *); oim->im6_msgtype = MRT6MSG_WRONGMIF; oim->im6_mbz = 0; break; #endif case MRT6_INIT: im = mtod(mm, struct mrt6msg *); im->im6_msgtype = MRT6MSG_WRONGMIF; im->im6_mbz = 0; break; default: m_freem(mm); return (EINVAL); } for (mifp = mif6table, iif = 0; iif < nummifs && mifp && mifp->m6_ifp != ifp; mifp++, iif++) ; switch (V_ip6_mrouter_ver) { #ifdef MRT6_OINIT case MRT6_OINIT: oim->im6_mif = iif; sin6.sin6_addr = oim->im6_src; break; #endif case MRT6_INIT: im->im6_mif = iif; sin6.sin6_addr = im->im6_src; break; } MRT6STAT_INC(mrt6s_upcalls); if (socket_send(V_ip6_mrouter, mm, &sin6) < 0) { MRT6_DLOG(DEBUG_ANY, "ip6_mrouter socket queue full"); MRT6STAT_INC(mrt6s_upq_sockfull); return (ENOBUFS); } /* if socket Q full */ } /* if PIM */ return (0); } /* if wrong iif */ /* If I sourced this packet, it counts as output, else it was input. */ if (m->m_pkthdr.rcvif == NULL) { /* XXX: is rcvif really NULL when output?? */ mif6table[mifi].m6_pkt_out++; mif6table[mifi].m6_bytes_out += plen; } else { mif6table[mifi].m6_pkt_in++; mif6table[mifi].m6_bytes_in += plen; } rt->mf6c_pkt_cnt++; rt->mf6c_byte_cnt += plen; /* * For each mif, forward a copy of the packet if there are group * members downstream on the interface. */ src0 = ip6->ip6_src; dst0 = ip6->ip6_dst; if ((error = in6_setscope(&src0, ifp, &iszone)) != 0 || (error = in6_setscope(&dst0, ifp, &idzone)) != 0) { IP6STAT_INC(ip6s_badscope); return (error); } for (mifp = mif6table, mifi = 0; mifi < nummifs; mifp++, mifi++) { if (IF_ISSET(mifi, &rt->mf6c_ifset)) { /* * check if the outgoing packet is going to break * a scope boundary. * XXX For packets through PIM register tunnel * interface, we believe a routing daemon. */ if (!(mif6table[rt->mf6c_parent].m6_flags & MIFF_REGISTER) && !(mif6table[mifi].m6_flags & MIFF_REGISTER)) { if (in6_setscope(&src0, mif6table[mifi].m6_ifp, &oszone) || in6_setscope(&dst0, mif6table[mifi].m6_ifp, &odzone) || iszone != oszone || idzone != odzone) { IP6STAT_INC(ip6s_badscope); continue; } } mifp->m6_pkt_out++; mifp->m6_bytes_out += plen; MC6_SEND(ip6, mifp, m); } } return (0); } static void phyint_send(struct ip6_hdr *ip6, struct mif6 *mifp, struct mbuf *m) { #ifdef MRT6DEBUG char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; #endif struct mbuf *mb_copy; struct ifnet *ifp = mifp->m6_ifp; int error = 0; u_long linkmtu; /* * Make a new reference to the packet; make sure that * the IPv6 header is actually copied, not just referenced, * so that ip6_output() only scribbles on the copy. */ mb_copy = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (mb_copy && (!M_WRITABLE(mb_copy) || mb_copy->m_len < sizeof(struct ip6_hdr))) mb_copy = m_pullup(mb_copy, sizeof(struct ip6_hdr)); if (mb_copy == NULL) { return; } /* set MCAST flag to the outgoing packet */ mb_copy->m_flags |= M_MCAST; /* * If we sourced the packet, call ip6_output since we may devide * the packet into fragments when the packet is too big for the * outgoing interface. * Otherwise, we can simply send the packet to the interface * sending queue. */ if (m->m_pkthdr.rcvif == NULL) { struct ip6_moptions im6o; im6o.im6o_multicast_ifp = ifp; /* XXX: ip6_output will override ip6->ip6_hlim */ im6o.im6o_multicast_hlim = ip6->ip6_hlim; im6o.im6o_multicast_loop = 1; error = ip6_output(mb_copy, NULL, NULL, IPV6_FORWARDING, &im6o, NULL, NULL); MRT6_DLOG(DEBUG_XMIT, "mif %u err %d", (uint16_t)(mifp - mif6table), error); return; } /* * If configured to loop back multicasts by default, * loop back a copy now. */ if (in6_mcast_loop) ip6_mloopback(ifp, m); /* * Put the packet into the sending queue of the outgoing interface * if it would fit in the MTU of the interface. */ linkmtu = IN6_LINKMTU(ifp); if (mb_copy->m_pkthdr.len <= linkmtu || linkmtu < IPV6_MMTU) { struct sockaddr_in6 dst6; bzero(&dst6, sizeof(dst6)); dst6.sin6_len = sizeof(struct sockaddr_in6); dst6.sin6_family = AF_INET6; dst6.sin6_addr = ip6->ip6_dst; IP_PROBE(send, NULL, NULL, ip6, ifp, NULL, ip6); /* * We just call if_output instead of nd6_output here, since * we need no ND for a multicast forwarded packet...right? */ m_clrprotoflags(m); /* Avoid confusing lower layers. */ error = (*ifp->if_output)(ifp, mb_copy, (struct sockaddr *)&dst6, NULL); MRT6_DLOG(DEBUG_XMIT, "mif %u err %d", (uint16_t)(mifp - mif6table), error); } else { /* * pMTU discovery is intentionally disabled by default, since * various router may notify pMTU in multicast, which can be * a DDoS to a router */ if (V_ip6_mcast_pmtu) icmp6_error(mb_copy, ICMP6_PACKET_TOO_BIG, 0, linkmtu); else { MRT6_DLOG(DEBUG_XMIT, " packet too big on %s o %s " "g %s size %d (discarded)", if_name(ifp), ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), mb_copy->m_pkthdr.len); m_freem(mb_copy); /* simply discard the packet */ } } } static int register_send(struct ip6_hdr *ip6, struct mif6 *mif, struct mbuf *m) { #ifdef MRT6DEBUG char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; #endif struct mbuf *mm; int i, len = m->m_pkthdr.len; static struct sockaddr_in6 sin6 = { sizeof(sin6), AF_INET6 }; struct mrt6msg *im6; MRT6_DLOG(DEBUG_ANY, "src %s dst %s", ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst)); PIM6STAT_INC(pim6s_snd_registers); /* Make a copy of the packet to send to the user level process. */ mm = m_gethdr(M_NOWAIT, MT_DATA); if (mm == NULL) return (ENOBUFS); mm->m_data += max_linkhdr; mm->m_len = sizeof(struct ip6_hdr); if ((mm->m_next = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL) { m_freem(mm); return (ENOBUFS); } i = MHLEN - M_LEADINGSPACE(mm); if (i > len) i = len; mm = m_pullup(mm, i); if (mm == NULL) return (ENOBUFS); /* TODO: check it! */ mm->m_pkthdr.len = len + sizeof(struct ip6_hdr); /* * Send message to routing daemon */ sin6.sin6_addr = ip6->ip6_src; im6 = mtod(mm, struct mrt6msg *); im6->im6_msgtype = MRT6MSG_WHOLEPKT; im6->im6_mbz = 0; im6->im6_mif = mif - mif6table; /* iif info is not given for reg. encap.n */ MRT6STAT_INC(mrt6s_upcalls); if (socket_send(V_ip6_mrouter, mm, &sin6) < 0) { MRT6_DLOG(DEBUG_ANY, "ip6_mrouter socket queue full"); MRT6STAT_INC(mrt6s_upq_sockfull); return (ENOBUFS); } return (0); } /* * pim6_encapcheck() is called by the encap6_input() path at runtime to * determine if a packet is for PIM; allowing PIM to be dynamically loaded * into the kernel. */ static int pim6_encapcheck(const struct mbuf *m __unused, int off __unused, int proto __unused, void *arg __unused) { KASSERT(proto == IPPROTO_PIM, ("not for IPPROTO_PIM")); return (8); /* claim the datagram. */ } /* * PIM sparse mode hook * Receives the pim control messages, and passes them up to the listening * socket, using rip6_input. * The only message processed is the REGISTER pim message; the pim header * is stripped off, and the inner packet is passed to register_mforward. */ static int pim6_input(struct mbuf *m, int off, int proto, void *arg __unused) { struct pim *pim; /* pointer to a pim struct */ struct ip6_hdr *ip6; int pimlen; int minlen; PIM6STAT_INC(pim6s_rcv_total); ip6 = mtod(m, struct ip6_hdr *); pimlen = m->m_pkthdr.len - off; /* * Validate lengths */ if (pimlen < PIM_MINLEN) { PIM6STAT_INC(pim6s_rcv_tooshort); MRT6_DLOG(DEBUG_PIM, "PIM packet too short"); m_freem(m); return (IPPROTO_DONE); } /* * if the packet is at least as big as a REGISTER, go ahead * and grab the PIM REGISTER header size, to avoid another * possible m_pullup() later. * * PIM_MINLEN == pimhdr + u_int32 == 8 * PIM6_REG_MINLEN == pimhdr + reghdr + eip6hdr == 4 + 4 + 40 */ minlen = (pimlen >= PIM6_REG_MINLEN) ? PIM6_REG_MINLEN : PIM_MINLEN; /* * Make sure that the IP6 and PIM headers in contiguous memory, and * possibly the PIM REGISTER header */ #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, minlen, IPPROTO_DONE); /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); /* adjust mbuf to point to the PIM header */ pim = (struct pim *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(pim, struct pim *, m, off, minlen); if (pim == NULL) { PIM6STAT_INC(pim6s_rcv_tooshort); return (IPPROTO_DONE); } #endif #define PIM6_CHECKSUM #ifdef PIM6_CHECKSUM { int cksumlen; /* * Validate checksum. * If PIM REGISTER, exclude the data packet */ if (pim->pim_type == PIM_REGISTER) cksumlen = PIM_MINLEN; else cksumlen = pimlen; if (in6_cksum(m, IPPROTO_PIM, off, cksumlen)) { PIM6STAT_INC(pim6s_rcv_badsum); MRT6_DLOG(DEBUG_PIM, "invalid checksum"); m_freem(m); return (IPPROTO_DONE); } } #endif /* PIM_CHECKSUM */ /* PIM version check */ if (pim->pim_ver != PIM_VERSION) { PIM6STAT_INC(pim6s_rcv_badversion); MRT6_DLOG(DEBUG_ANY | DEBUG_ERR, "incorrect version %d, expecting %d", pim->pim_ver, PIM_VERSION); m_freem(m); return (IPPROTO_DONE); } if (pim->pim_type == PIM_REGISTER) { /* * since this is a REGISTER, we'll make a copy of the register * headers ip6+pim+u_int32_t+encap_ip6, to be passed up to the * routing daemon. */ static struct sockaddr_in6 dst = { sizeof(dst), AF_INET6 }; struct mbuf *mcp; struct ip6_hdr *eip6; u_int32_t *reghdr; int rc; #ifdef MRT6DEBUG char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; #endif PIM6STAT_INC(pim6s_rcv_registers); if ((reg_mif_num >= nummifs) || (reg_mif_num == (mifi_t) -1)) { MRT6_DLOG(DEBUG_PIM, "register mif not set: %d", reg_mif_num); m_freem(m); return (IPPROTO_DONE); } reghdr = (u_int32_t *)(pim + 1); if ((ntohl(*reghdr) & PIM_NULL_REGISTER)) goto pim6_input_to_daemon; /* * Validate length */ if (pimlen < PIM6_REG_MINLEN) { PIM6STAT_INC(pim6s_rcv_tooshort); PIM6STAT_INC(pim6s_rcv_badregisters); MRT6_DLOG(DEBUG_ANY | DEBUG_ERR, "register packet " "size too small %d from %s", pimlen, ip6_sprintf(ip6bufs, &ip6->ip6_src)); m_freem(m); return (IPPROTO_DONE); } eip6 = (struct ip6_hdr *) (reghdr + 1); MRT6_DLOG(DEBUG_PIM, "eip6: %s -> %s, eip6 plen %d", ip6_sprintf(ip6bufs, &eip6->ip6_src), ip6_sprintf(ip6bufd, &eip6->ip6_dst), ntohs(eip6->ip6_plen)); /* verify the version number of the inner packet */ if ((eip6->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION) { PIM6STAT_INC(pim6s_rcv_badregisters); MRT6_DLOG(DEBUG_ANY, "invalid IP version (%d) " "of the inner packet", (eip6->ip6_vfc & IPV6_VERSION)); m_freem(m); return (IPPROTO_DONE); } /* verify the inner packet is destined to a mcast group */ if (!IN6_IS_ADDR_MULTICAST(&eip6->ip6_dst)) { PIM6STAT_INC(pim6s_rcv_badregisters); MRT6_DLOG(DEBUG_PIM, "inner packet of register " "is not multicast %s", ip6_sprintf(ip6bufd, &eip6->ip6_dst)); m_freem(m); return (IPPROTO_DONE); } /* * make a copy of the whole header to pass to the daemon later. */ mcp = m_copym(m, 0, off + PIM6_REG_MINLEN, M_NOWAIT); if (mcp == NULL) { MRT6_DLOG(DEBUG_ANY | DEBUG_ERR, "pim register: " "could not copy register head"); m_freem(m); return (IPPROTO_DONE); } /* * forward the inner ip6 packet; point m_data at the inner ip6. */ m_adj(m, off + PIM_MINLEN); MRT6_DLOG(DEBUG_PIM, "forwarding decapsulated register: " "src %s, dst %s, mif %d", ip6_sprintf(ip6bufs, &eip6->ip6_src), ip6_sprintf(ip6bufd, &eip6->ip6_dst), reg_mif_num); rc = if_simloop(mif6table[reg_mif_num].m6_ifp, m, dst.sin6_family, 0); /* prepare the register head to send to the mrouting daemon */ m = mcp; } /* * Pass the PIM message up to the daemon; if it is a register message * pass the 'head' only up to the daemon. This includes the * encapsulator ip6 header, pim header, register header and the * encapsulated ip6 header. */ pim6_input_to_daemon: return (rip6_input(&m, &off, proto)); } static int ip6_mroute_modevent(module_t mod, int type, void *unused) { switch (type) { case MOD_LOAD: MROUTER6_LOCK_INIT(); MFC6_LOCK_INIT(); MIF6_LOCK_INIT(); pim6_encap_cookie = ip6_encap_attach(&ipv6_encap_cfg, NULL, M_WAITOK); if (pim6_encap_cookie == NULL) { printf("ip6_mroute: unable to attach pim6 encap\n"); MIF6_LOCK_DESTROY(); MFC6_LOCK_DESTROY(); MROUTER6_LOCK_DESTROY(); return (EINVAL); } ip6_mforward = X_ip6_mforward; ip6_mrouter_done = X_ip6_mrouter_done; ip6_mrouter_get = X_ip6_mrouter_get; ip6_mrouter_set = X_ip6_mrouter_set; mrt6_ioctl = X_mrt6_ioctl; break; case MOD_UNLOAD: if (V_ip6_mrouter != NULL) return EINVAL; if (pim6_encap_cookie) { ip6_encap_detach(pim6_encap_cookie); pim6_encap_cookie = NULL; } X_ip6_mrouter_done(); ip6_mforward = NULL; ip6_mrouter_done = NULL; ip6_mrouter_get = NULL; ip6_mrouter_set = NULL; mrt6_ioctl = NULL; MIF6_LOCK_DESTROY(); MFC6_LOCK_DESTROY(); MROUTER6_LOCK_DESTROY(); break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t ip6_mroutemod = { "ip6_mroute", ip6_mroute_modevent, 0 }; DECLARE_MODULE(ip6_mroute, ip6_mroutemod, SI_SUB_PROTO_MC, SI_ORDER_ANY); Index: head/sys/netinet6/mld6.c =================================================================== --- head/sys/netinet6/mld6.c (revision 336675) +++ head/sys/netinet6/mld6.c (revision 336676) @@ -1,3321 +1,3321 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2009 Bruce Simpson. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote * products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ */ /*- * Copyright (c) 1988 Stephen Deering. * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)igmp.c 8.1 (Berkeley) 7/19/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef KTR_MLD #define KTR_MLD KTR_INET6 #endif static struct mld_ifsoftc * mli_alloc_locked(struct ifnet *); static void mli_delete_locked(const struct ifnet *); static void mld_dispatch_packet(struct mbuf *); static void mld_dispatch_queue(struct mbufq *, int); static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *); static void mld_fasttimo_vnet(void); static int mld_handle_state_change(struct in6_multi *, struct mld_ifsoftc *); static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *, const int); #ifdef KTR static char * mld_rec_type_to_str(const int); #endif static void mld_set_version(struct mld_ifsoftc *, const int); static void mld_slowtimo_vnet(void); static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, /*const*/ struct mld_hdr *); static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, /*const*/ struct mld_hdr *); static void mld_v1_process_group_timer(struct in6_multi_head *, struct in6_multi *); static void mld_v1_process_querier_timers(struct mld_ifsoftc *); static int mld_v1_transmit_report(struct in6_multi *, const int); static void mld_v1_update_group(struct in6_multi *, const int); static void mld_v2_cancel_link_timers(struct mld_ifsoftc *); static void mld_v2_dispatch_general_query(struct mld_ifsoftc *); static struct mbuf * mld_v2_encap_report(struct ifnet *, struct mbuf *); static int mld_v2_enqueue_filter_change(struct mbufq *, struct in6_multi *); static int mld_v2_enqueue_group_record(struct mbufq *, struct in6_multi *, const int, const int, const int, const int); static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, struct mbuf *, const int, const int); static int mld_v2_merge_state_changes(struct in6_multi *, struct mbufq *); static void mld_v2_process_group_timers(struct in6_multi_head *, struct mbufq *, struct mbufq *, struct in6_multi *, const int); static int mld_v2_process_group_query(struct in6_multi *, struct mld_ifsoftc *mli, int, struct mbuf *, const int); static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); /* * Normative references: RFC 2710, RFC 3590, RFC 3810. * * Locking: * * The MLD subsystem lock ends up being system-wide for the moment, * but could be per-VIMAGE later on. * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. * Any may be taken independently; if any are held at the same * time, the above lock order must be followed. * * IN6_MULTI_LOCK covers in_multi. * * MLD_LOCK covers per-link state and any global variables in this file. * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of * per-link state iterators. * * XXX LOR PREVENTION * A special case for IPv6 is the in6_setscope() routine. ip6_output() * will not accept an ifp; it wants an embedded scope ID, unlike * ip_output(), which happily takes the ifp given to it. The embedded * scope ID is only used by MLD to select the outgoing interface. * * During interface attach and detach, MLD will take MLD_LOCK *after* * the IF_AFDATA_LOCK. * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call * it with MLD_LOCK held without triggering an LOR. A netisr with indirect * dispatch could work around this, but we'd rather not do that, as it * can introduce other races. * * As such, we exploit the fact that the scope ID is just the interface * index, and embed it in the IPv6 destination address accordingly. * This is potentially NOT VALID for MLDv1 reports, as they * are always sent to the multicast group itself; as MLDv2 * reports are always sent to ff02::16, this is not an issue * when MLDv2 is in use. * * This does not however eliminate the LOR when ip6_output() itself * calls in6_setscope() internally whilst MLD_LOCK is held. This will * trigger a LOR warning in WITNESS when the ifnet is detached. * * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given * how it's used across the network stack. Here we're simply exploiting * the fact that MLD runs at a similar layer in the stack to scope6.c. * * VIMAGE: * * Each in6_multi corresponds to an ifp, and each ifp corresponds * to a vnet in ifp->if_vnet. */ static struct mtx mld_mtx; static MALLOC_DEFINE(M_MLD, "mld", "mld state"); #define MLD_EMBEDSCOPE(pin6, zoneid) \ if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ /* * VIMAGE-wide globals. */ -static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; -static VNET_DEFINE(LIST_HEAD(, mld_ifsoftc), mli_head); -static VNET_DEFINE(int, interface_timers_running6); -static VNET_DEFINE(int, state_change_timers_running6); -static VNET_DEFINE(int, current_state_timers_running6); +VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0}; +VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head); +VNET_DEFINE_STATIC(int, interface_timers_running6); +VNET_DEFINE_STATIC(int, state_change_timers_running6); +VNET_DEFINE_STATIC(int, current_state_timers_running6); #define V_mld_gsrdelay VNET(mld_gsrdelay) #define V_mli_head VNET(mli_head) #define V_interface_timers_running6 VNET(interface_timers_running6) #define V_state_change_timers_running6 VNET(state_change_timers_running6) #define V_current_state_timers_running6 VNET(current_state_timers_running6) SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, "IPv6 Multicast Listener Discovery"); /* * Virtualized sysctls. */ SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", "Rate limit for MLDv2 Group-and-Source queries in seconds"); /* * Non-virtualized sysctls. */ static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, "Per-interface MLDv2 state"); static int mld_v1enable = 1; SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN, &mld_v1enable, 0, "Enable fallback to MLDv1"); static int mld_use_allow = 1; SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN, &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); /* * Packed Router Alert option structure declaration. */ struct mld_raopt { struct ip6_hbh hbh; struct ip6_opt pad; struct ip6_opt_router ra; } __packed; /* * Router Alert hop-by-hop option header. */ static struct mld_raopt mld_ra = { .hbh = { 0, 0 }, .pad = { .ip6o_type = IP6OPT_PADN, 0 }, .ra = { .ip6or_type = IP6OPT_ROUTER_ALERT, .ip6or_len = IP6OPT_RTALERT_LEN - 2, .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) } }; static struct ip6_pktopts mld_po; static __inline void mld_save_context(struct mbuf *m, struct ifnet *ifp) { #ifdef VIMAGE m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; #endif /* VIMAGE */ m->m_pkthdr.flowid = ifp->if_index; } static __inline void mld_scrub_context(struct mbuf *m) { m->m_pkthdr.PH_loc.ptr = NULL; m->m_pkthdr.flowid = 0; } /* * Restore context from a queued output chain. * Return saved ifindex. * * VIMAGE: The assertion is there to make sure that we * actually called CURVNET_SET() with what's in the mbuf chain. */ static __inline uint32_t mld_restore_context(struct mbuf *m) { #if defined(VIMAGE) && defined(INVARIANTS) KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, ("%s: called when curvnet was not restored: cuvnet %p m ptr %p", __func__, curvnet, m->m_pkthdr.PH_loc.ptr)); #endif return (m->m_pkthdr.flowid); } /* * Retrieve or set threshold between group-source queries in seconds. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by MLD lock. */ static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) { int error; int i; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); MLD_LOCK(); i = V_mld_gsrdelay.tv_sec; error = sysctl_handle_int(oidp, &i, 0, req); if (error || !req->newptr) goto out_locked; if (i < -1 || i >= 60) { error = EINVAL; goto out_locked; } CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", V_mld_gsrdelay.tv_sec, i); V_mld_gsrdelay.tv_sec = i; out_locked: MLD_UNLOCK(); return (error); } /* * Expose struct mld_ifsoftc to userland, keyed by ifindex. * For use by ifmcstat(8). * * SMPng: NOTE: Does an unlocked ifindex space read. * VIMAGE: Assume curvnet set by caller. The node handler itself * is not directly virtualized. */ static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) { int *name; int error; u_int namelen; struct ifnet *ifp; struct mld_ifsoftc *mli; name = (int *)arg1; namelen = arg2; if (req->newptr != NULL) return (EPERM); if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); if (error) return (error); IN6_MULTI_LOCK(); IN6_MULTI_LIST_LOCK(); MLD_LOCK(); if (name[0] <= 0 || name[0] > V_if_index) { error = ENOENT; goto out_locked; } error = ENOENT; ifp = ifnet_byindex(name[0]); if (ifp == NULL) goto out_locked; LIST_FOREACH(mli, &V_mli_head, mli_link) { if (ifp == mli->mli_ifp) { struct mld_ifinfo info; info.mli_version = mli->mli_version; info.mli_v1_timer = mli->mli_v1_timer; info.mli_v2_timer = mli->mli_v2_timer; info.mli_flags = mli->mli_flags; info.mli_rv = mli->mli_rv; info.mli_qi = mli->mli_qi; info.mli_qri = mli->mli_qri; info.mli_uri = mli->mli_uri; error = SYSCTL_OUT(req, &info, sizeof(info)); break; } } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); IN6_MULTI_UNLOCK(); return (error); } /* * Dispatch an entire queue of pending packet chains. * VIMAGE: Assumes the vnet pointer has been set. */ static void mld_dispatch_queue(struct mbufq *mq, int limit) { struct mbuf *m; while ((m = mbufq_dequeue(mq)) != NULL) { CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m); mld_dispatch_packet(m); if (--limit == 0) break; } } /* * Filter outgoing MLD report state by group. * * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) * and node-local addresses. However, kernel and socket consumers * always embed the KAME scope ID in the address provided, so strip it * when performing comparison. * Note: This is not the same as the *multicast* scope. * * Return zero if the given group is one for which MLD reports * should be suppressed, or non-zero if reports should be issued. */ static __inline int mld_is_addr_reported(const struct in6_addr *addr) { KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) return (0); if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { struct in6_addr tmp = *addr; in6_clearscope(&tmp); if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) return (0); } return (1); } /* * Attach MLD when PF_INET6 is attached to an interface. * * SMPng: Normally called with IF_AFDATA_LOCK held. */ struct mld_ifsoftc * mld_domifattach(struct ifnet *ifp) { struct mld_ifsoftc *mli; CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK(); mli = mli_alloc_locked(ifp); if (!(ifp->if_flags & IFF_MULTICAST)) mli->mli_flags |= MLIF_SILENT; if (mld_use_allow) mli->mli_flags |= MLIF_USEALLOW; MLD_UNLOCK(); return (mli); } /* * VIMAGE: assume curvnet set by caller. */ static struct mld_ifsoftc * mli_alloc_locked(/*const*/ struct ifnet *ifp) { struct mld_ifsoftc *mli; MLD_LOCK_ASSERT(); mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO); if (mli == NULL) goto out; mli->mli_ifp = ifp; mli->mli_version = MLD_VERSION_2; mli->mli_flags = 0; mli->mli_rv = MLD_RV_INIT; mli->mli_qi = MLD_QI_INIT; mli->mli_qri = MLD_QRI_INIT; mli->mli_uri = MLD_URI_INIT; mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)", ifp, if_name(ifp)); out: return (mli); } /* * Hook for ifdetach. * * NOTE: Some finalization tasks need to run before the protocol domain * is detached, but also before the link layer does its cleanup. * Run before link-layer cleanup; cleanup groups, but do not free MLD state. * * SMPng: Caller must hold IN6_MULTI_LOCK(). * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. * XXX This routine is also bitten by unlocked ifma_protospec access. */ void mld_ifdetach(struct ifnet *ifp) { struct mld_ifsoftc *mli; struct ifmultiaddr *ifma, *next; struct in6_multi *inm; struct in6_multi_head inmh; CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); SLIST_INIT(&inmh); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK(); mli = MLD_IFINFO(ifp); if (mli->mli_version == MLD_VERSION_2) { IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET6 || ifma->ifma_protospec == NULL) continue; inm = (struct in6_multi *)ifma->ifma_protospec; if (inm->in6m_state == MLD_LEAVING_MEMBER) { in6m_rele_locked(&inmh, inm); ifma->ifma_protospec = NULL; } in6m_clear_recorded(inm); if (__predict_false(ifma6_restart)) { ifma6_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); } MLD_UNLOCK(); in6m_release_list_deferred(&inmh); } /* * Hook for domifdetach. * Runs after link-layer cleanup; free MLD state. * * SMPng: Normally called with IF_AFDATA_LOCK held. */ void mld_domifdetach(struct ifnet *ifp) { CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK(); mli_delete_locked(ifp); MLD_UNLOCK(); } static void mli_delete_locked(const struct ifnet *ifp) { struct mld_ifsoftc *mli, *tmli; CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK_ASSERT(); LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { if (mli->mli_ifp == ifp) { /* * Free deferred General Query responses. */ mbufq_drain(&mli->mli_gq); LIST_REMOVE(mli, mli_link); free(mli, M_MLD); return; } } } /* * Process a received MLDv1 general or address-specific query. * Assumes that the query header has been pulled up to sizeof(mld_hdr). * * NOTE: Can't be fully const correct as we temporarily embed scope ID in * mld_addr. This is OK as we own the mbuf chain. */ static int mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld) { struct ifmultiaddr *ifma; struct mld_ifsoftc *mli; struct in6_multi *inm; int is_general_query; uint16_t timer; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif is_general_query = 0; if (!mld_v1enable) { CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); return (0); } /* * RFC3810 Section 6.2: MLD queries must originate from * a router's link-local address. */ if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (0); } /* * Do address field validation upfront before we accept * the query. */ if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { /* * MLDv1 General Query. * If this was not sent to the all-nodes group, ignore it. */ struct in6_addr dst; dst = ip6->ip6_dst; in6_clearscope(&dst); if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) return (EINVAL); is_general_query = 1; } else { /* * Embed scope ID of receiving interface in MLD query for * lookup whilst we don't hold other locks. */ in6_setscope(&mld->mld_addr, ifp, NULL); } IN6_MULTI_LIST_LOCK(); MLD_LOCK(); /* * Switch to MLDv1 host compatibility mode. */ mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); mld_set_version(mli, MLD_VERSION_1); timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; if (timer == 0) timer = 1; IF_ADDR_RLOCK(ifp); if (is_general_query) { /* * For each reporting group joined on this * interface, kick the report timer. */ CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", ifp, if_name(ifp)); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET6 || ifma->ifma_protospec == NULL) continue; inm = (struct in6_multi *)ifma->ifma_protospec; mld_v1_update_group(inm, timer); } } else { /* * MLDv1 Group-Specific Query. * If this is a group-specific MLDv1 query, we need only * look up the single group to process it. */ inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm != NULL) { CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); mld_v1_update_group(inm, timer); } /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); } IF_ADDR_RUNLOCK(ifp); MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); return (0); } /* * Update the report timer on a group in response to an MLDv1 query. * * If we are becoming the reporting member for this group, start the timer. * If we already are the reporting member for this group, and timer is * below the threshold, reset it. * * We may be updating the group for the first time since we switched * to MLDv2. If we are, then we must clear any recorded source lists, * and transition to REPORTING state; the group timer is overloaded * for group and group-source query responses. * * Unlike MLDv2, the delay per group should be jittered * to avoid bursts of MLDv1 reports. */ static void mld_v1_update_group(struct in6_multi *inm, const int timer) { #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp), timer); IN6_MULTI_LIST_LOCK_ASSERT(); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: break; case MLD_REPORTING_MEMBER: if (inm->in6m_timer != 0 && inm->in6m_timer <= timer) { CTR1(KTR_MLD, "%s: REPORTING and timer running, " "skipping.", __func__); break; } /* FALLTHROUGH */ case MLD_SG_QUERY_PENDING_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: CTR1(KTR_MLD, "%s: ->REPORTING", __func__); inm->in6m_state = MLD_REPORTING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; break; case MLD_SLEEPING_MEMBER: CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); inm->in6m_state = MLD_AWAKENING_MEMBER; break; case MLD_LEAVING_MEMBER: break; } } /* * Process a received MLDv2 general, group-specific or * group-and-source-specific query. * * Assumes that the query header has been pulled up to sizeof(mldv2_query). * * Return 0 if successful, otherwise an appropriate error code is returned. */ static int mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, struct mbuf *m, const int off, const int icmp6len) { struct mld_ifsoftc *mli; struct mldv2_query *mld; struct in6_multi *inm; uint32_t maxdelay, nsrc, qqi; int is_general_query; uint16_t timer; uint8_t qrv; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif is_general_query = 0; /* * RFC3810 Section 6.2: MLD queries must originate from * a router's link-local address. */ if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (0); } CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ if (maxdelay >= 32768) { maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << (MLD_MRC_EXP(maxdelay) + 3); } timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; if (timer == 0) timer = 1; qrv = MLD_QRV(mld->mld_misc); if (qrv < 2) { CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, qrv, MLD_RV_INIT); qrv = MLD_RV_INIT; } qqi = mld->mld_qqi; if (qqi >= 128) { qqi = MLD_QQIC_MANT(mld->mld_qqi) << (MLD_QQIC_EXP(mld->mld_qqi) + 3); } nsrc = ntohs(mld->mld_numsrc); if (nsrc > MLD_MAX_GS_SOURCES) return (EMSGSIZE); if (icmp6len < sizeof(struct mldv2_query) + (nsrc * sizeof(struct in6_addr))) return (EMSGSIZE); /* * Do further input validation upfront to avoid resetting timers * should we need to discard this query. */ if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { /* * A general query with a source list has undefined * behaviour; discard it. */ if (nsrc > 0) return (EINVAL); is_general_query = 1; } else { /* * Embed scope ID of receiving interface in MLD query for * lookup whilst we don't hold other locks (due to KAME * locking lameness). We own this mbuf chain just now. */ in6_setscope(&mld->mld_addr, ifp, NULL); } IN6_MULTI_LIST_LOCK(); MLD_LOCK(); mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); /* * Discard the v2 query if we're in Compatibility Mode. * The RFC is pretty clear that hosts need to stay in MLDv1 mode * until the Old Version Querier Present timer expires. */ if (mli->mli_version != MLD_VERSION_2) goto out_locked; mld_set_version(mli, MLD_VERSION_2); mli->mli_rv = qrv; mli->mli_qi = qqi; mli->mli_qri = maxdelay; CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, maxdelay); if (is_general_query) { /* * MLDv2 General Query. * * Schedule a current-state report on this ifp for * all groups, possibly containing source lists. * * If there is a pending General Query response * scheduled earlier than the selected delay, do * not schedule any other reports. * Otherwise, reset the interface timer. */ CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", ifp, if_name(ifp)); if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); V_interface_timers_running6 = 1; } } else { /* * MLDv2 Group-specific or Group-and-source-specific Query. * * Group-source-specific queries are throttled on * a per-group basis to defeat denial-of-service attempts. * Queries for groups we are not a member of on this * link are simply ignored. */ IF_ADDR_RLOCK(ifp); inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm == NULL) { IF_ADDR_RUNLOCK(ifp); goto out_locked; } if (nsrc > 0) { if (!ratecheck(&inm->in6m_lastgsrtv, &V_mld_gsrdelay)) { CTR1(KTR_MLD, "%s: GS query throttled.", __func__); IF_ADDR_RUNLOCK(ifp); goto out_locked; } } CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", ifp, if_name(ifp)); /* * If there is a pending General Query response * scheduled sooner than the selected delay, no * further report need be scheduled. * Otherwise, prepare to respond to the * group-specific or group-and-source query. */ if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) mld_v2_process_group_query(inm, mli, timer, m, off); /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); IF_ADDR_RUNLOCK(ifp); } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received MLDv2 group-specific or group-and-source-specific * query. * Return <0 if any error occurred. Currently this is ignored. */ static int mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli, int timer, struct mbuf *m0, const int off) { struct mldv2_query *mld; int retval; uint16_t nsrc; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); retval = 0; mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LEAVING_MEMBER: return (retval); break; case MLD_REPORTING_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: break; } nsrc = ntohs(mld->mld_numsrc); /* * Deal with group-specific queries upfront. * If any group query is already pending, purge any recorded * source-list state if it exists, and schedule a query response * for this group-specific query. */ if (nsrc == 0) { if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { in6m_clear_recorded(inm); timer = min(inm->in6m_timer, timer); } inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; return (retval); } /* * Deal with the case where a group-and-source-specific query has * been received but a group-specific query is already pending. */ if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { timer = min(inm->in6m_timer, timer); inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; return (retval); } /* * Finally, deal with the case where a group-and-source-specific * query has been received, where a response to a previous g-s-r * query exists, or none exists. * In this case, we need to parse the source-list which the Querier * has provided us with and check if we have any source list filter * entries at T1 for these sources. If we do not, there is no need * schedule a report and the query may be dropped. * If we do, we must record them and schedule a current-state * report for those sources. */ if (inm->in6m_nsrc > 0) { struct mbuf *m; uint8_t *sp; int i, nrecorded; int soff; m = m0; soff = off + sizeof(struct mldv2_query); nrecorded = 0; for (i = 0; i < nsrc; i++) { sp = mtod(m, uint8_t *) + soff; retval = in6m_record_source(inm, (const struct in6_addr *)sp); if (retval < 0) break; nrecorded += retval; soff += sizeof(struct in6_addr); if (soff >= m->m_len) { soff = soff - m->m_len; m = m->m_next; if (m == NULL) break; } } if (nrecorded > 0) { CTR1(KTR_MLD, "%s: schedule response to SG query", __func__); inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; } } return (retval); } /* * Process a received MLDv1 host membership report. * Assumes mld points to mld_hdr in pulled up mbuf chain. * * NOTE: Can't be fully const correct as we temporarily embed scope ID in * mld_addr. This is OK as we own the mbuf chain. */ static int mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld) { struct in6_addr src, dst; struct in6_ifaddr *ia; struct in6_multi *inm; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif if (!mld_v1enable) { CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); return (0); } if (ifp->if_flags & IFF_LOOPBACK) return (0); /* * MLDv1 reports must originate from a host's link-local address, * or the unspecified address (when booting). */ src = ip6->ip6_src; in6_clearscope(&src); if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (EINVAL); } /* * RFC2710 Section 4: MLDv1 reports must pertain to a multicast * group, and must be directed to the group itself. */ dst = ip6->ip6_dst; in6_clearscope(&dst); if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_dst), ifp, if_name(ifp)); return (EINVAL); } /* * Make sure we don't hear our own membership report, as fast * leave requires knowing that we are the only member of a * group. Assume we used the link-local address if available, * otherwise look for ::. * * XXX Note that scope ID comparison is needed for the address * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be * performed for the on-wire address. */ ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (0); } if (ia != NULL) ifa_free(&ia->ia_ifa); CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); /* * Embed scope ID of receiving interface in MLD query for lookup * whilst we don't hold other locks (due to KAME locking lameness). */ if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) in6_setscope(&mld->mld_addr, ifp, NULL); IN6_MULTI_LIST_LOCK(); MLD_LOCK(); IF_ADDR_RLOCK(ifp); /* * MLDv1 report suppression. * If we are a member of this group, and our membership should be * reported, and our group timer is pending or about to be reset, * stop our group timer by transitioning to the 'lazy' state. */ inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm != NULL) { struct mld_ifsoftc *mli; mli = inm->in6m_mli; KASSERT(mli != NULL, ("%s: no mli for ifp %p", __func__, ifp)); /* * If we are in MLDv2 host mode, do not allow the * other host's MLDv1 report to suppress our reports. */ if (mli->mli_version == MLD_VERSION_2) goto out_locked; inm->in6m_timer = 0; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_AWAKENING_MEMBER: CTR3(KTR_MLD, "report suppressed for %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); case MLD_LAZY_MEMBER: inm->in6m_state = MLD_LAZY_MEMBER; break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } out_locked: IF_ADDR_RUNLOCK(ifp); MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); return (0); } /* * MLD input path. * * Assume query messages which fit in a single ICMPv6 message header * have been pulled up. * Assume that userland will want to see the message, even if it * otherwise fails kernel input validation; do not free it. * Pullup may however free the mbuf chain m if it fails. * * Return IPPROTO_DONE if we freed m. Otherwise, return 0. */ int mld_input(struct mbuf *m, int off, int icmp6len) { struct ifnet *ifp; struct ip6_hdr *ip6; struct mld_hdr *mld; int mldlen; CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); ifp = m->m_pkthdr.rcvif; ip6 = mtod(m, struct ip6_hdr *); /* Pullup to appropriate size. */ mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); if (mld->mld_type == MLD_LISTENER_QUERY && icmp6len >= sizeof(struct mldv2_query)) { mldlen = sizeof(struct mldv2_query); } else { mldlen = sizeof(struct mld_hdr); } IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); if (mld == NULL) { ICMP6STAT_INC(icp6s_badlen); return (IPPROTO_DONE); } /* * Userland needs to see all of this traffic for implementing * the endpoint discovery portion of multicast routing. */ switch (mld->mld_type) { case MLD_LISTENER_QUERY: icmp6_ifstat_inc(ifp, ifs6_in_mldquery); if (icmp6len == sizeof(struct mld_hdr)) { if (mld_v1_input_query(ifp, ip6, mld) != 0) return (0); } else if (icmp6len >= sizeof(struct mldv2_query)) { if (mld_v2_input_query(ifp, ip6, m, off, icmp6len) != 0) return (0); } break; case MLD_LISTENER_REPORT: icmp6_ifstat_inc(ifp, ifs6_in_mldreport); if (mld_v1_input_report(ifp, ip6, mld) != 0) return (0); break; case MLDV2_LISTENER_REPORT: icmp6_ifstat_inc(ifp, ifs6_in_mldreport); break; case MLD_LISTENER_DONE: icmp6_ifstat_inc(ifp, ifs6_in_mlddone); break; default: break; } return (0); } /* * Fast timeout handler (global). * VIMAGE: Timeout handlers are expected to service all vimages. */ void mld_fasttimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); mld_fasttimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Fast timeout handler (per-vnet). * * VIMAGE: Assume caller has set up our curvnet. */ static void mld_fasttimo_vnet(void) { struct mbufq scq; /* State-change packets */ struct mbufq qrq; /* Query response packets */ struct ifnet *ifp; struct mld_ifsoftc *mli; struct ifmultiaddr *ifma, *next; struct in6_multi *inm, *tinm; struct in6_multi_head inmh; int uri_fasthz; uri_fasthz = 0; /* * Quick check to see if any work needs to be done, in order to * minimize the overhead of fasttimo processing. * SMPng: XXX Unlocked reads. */ if (!V_current_state_timers_running6 && !V_interface_timers_running6 && !V_state_change_timers_running6) return; SLIST_INIT(&inmh); IN6_MULTI_LIST_LOCK(); MLD_LOCK(); /* * MLDv2 General Query response timer processing. */ if (V_interface_timers_running6) { CTR1(KTR_MLD, "%s: interface timers running", __func__); V_interface_timers_running6 = 0; LIST_FOREACH(mli, &V_mli_head, mli_link) { if (mli->mli_v2_timer == 0) { /* Do nothing. */ } else if (--mli->mli_v2_timer == 0) { mld_v2_dispatch_general_query(mli); } else { V_interface_timers_running6 = 1; } } } if (!V_current_state_timers_running6 && !V_state_change_timers_running6) goto out_locked; V_current_state_timers_running6 = 0; V_state_change_timers_running6 = 0; CTR1(KTR_MLD, "%s: state change timers running", __func__); /* * MLD host report and state-change timer processing. * Note: Processing a v2 group timer may remove a node. */ LIST_FOREACH(mli, &V_mli_head, mli_link) { ifp = mli->mli_ifp; if (mli->mli_version == MLD_VERSION_2) { uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * PR_FASTHZ); mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS); mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS); } IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET6 || ifma->ifma_protospec == NULL) continue; inm = (struct in6_multi *)ifma->ifma_protospec; switch (mli->mli_version) { case MLD_VERSION_1: mld_v1_process_group_timer(&inmh, inm); break; case MLD_VERSION_2: mld_v2_process_group_timers(&inmh, &qrq, &scq, inm, uri_fasthz); break; } if (__predict_false(ifma6_restart)) { ifma6_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); switch (mli->mli_version) { case MLD_VERSION_1: /* * Transmit reports for this lifecycle. This * is done while not holding IF_ADDR_LOCK * since this can call * in6ifa_ifpforlinklocal() which locks * IF_ADDR_LOCK internally as well as * ip6_output() to transmit a packet. */ SLIST_FOREACH_SAFE(inm, &inmh, in6m_nrele, tinm) { SLIST_REMOVE_HEAD(&inmh, in6m_nrele); (void)mld_v1_transmit_report(inm, MLD_LISTENER_REPORT); } break; case MLD_VERSION_2: mld_dispatch_queue(&qrq, 0); mld_dispatch_queue(&scq, 0); /* * Free the in_multi reference(s) for * this lifecycle. */ in6m_release_list_deferred(&inmh); break; } } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); } /* * Update host report group timer. * Will update the global pending timer flags. */ static void mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm) { int report_timer_expired; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); if (inm->in6m_timer == 0) { report_timer_expired = 0; } else if (--inm->in6m_timer == 0) { report_timer_expired = 1; } else { V_current_state_timers_running6 = 1; return; } switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: break; case MLD_REPORTING_MEMBER: if (report_timer_expired) { inm->in6m_state = MLD_IDLE_MEMBER; in6m_rele_locked(inmh, inm); } break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } /* * Update a group's timers for MLDv2. * Will update the global pending timer flags. * Note: Unlocked read from mli. */ static void mld_v2_process_group_timers(struct in6_multi_head *inmh, struct mbufq *qrq, struct mbufq *scq, struct in6_multi *inm, const int uri_fasthz) { int query_response_timer_expired; int state_change_retransmit_timer_expired; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); query_response_timer_expired = 0; state_change_retransmit_timer_expired = 0; /* * During a transition from compatibility mode back to MLDv2, * a group record in REPORTING state may still have its group * timer active. This is a no-op in this function; it is easier * to deal with it here than to complicate the slow-timeout path. */ if (inm->in6m_timer == 0) { query_response_timer_expired = 0; } else if (--inm->in6m_timer == 0) { query_response_timer_expired = 1; } else { V_current_state_timers_running6 = 1; } if (inm->in6m_sctimer == 0) { state_change_retransmit_timer_expired = 0; } else if (--inm->in6m_sctimer == 0) { state_change_retransmit_timer_expired = 1; } else { V_state_change_timers_running6 = 1; } /* We are in fasttimo, so be quick about it. */ if (!state_change_retransmit_timer_expired && !query_response_timer_expired) return; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: case MLD_IDLE_MEMBER: break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: /* * Respond to a previously pending Group-Specific * or Group-and-Source-Specific query by enqueueing * the appropriate Current-State report for * immediate transmission. */ if (query_response_timer_expired) { int retval; retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 0); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); inm->in6m_state = MLD_REPORTING_MEMBER; in6m_clear_recorded(inm); } /* FALLTHROUGH */ case MLD_REPORTING_MEMBER: case MLD_LEAVING_MEMBER: if (state_change_retransmit_timer_expired) { /* * State-change retransmission timer fired. * If there are any further pending retransmissions, * set the global pending state-change flag, and * reset the timer. */ if (--inm->in6m_scrv > 0) { inm->in6m_sctimer = uri_fasthz; V_state_change_timers_running6 = 1; } /* * Retransmit the previously computed state-change * report. If there are no further pending * retransmissions, the mbuf queue will be consumed. * Update T0 state to T1 as we have now sent * a state-change. */ (void)mld_v2_merge_state_changes(inm, scq); in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); /* * If we are leaving the group for good, make sure * we release MLD's reference to it. * This release must be deferred using a SLIST, * as we are called from a loop which traverses * the in_ifmultiaddr TAILQ. */ if (inm->in6m_state == MLD_LEAVING_MEMBER && inm->in6m_scrv == 0) { inm->in6m_state = MLD_NOT_MEMBER; in6m_rele_locked(inmh, inm); } } break; } } /* * Switch to a different version on the given interface, * as per Section 9.12. */ static void mld_set_version(struct mld_ifsoftc *mli, const int version) { int old_version_timer; MLD_LOCK_ASSERT(); CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, version, mli->mli_ifp, if_name(mli->mli_ifp)); if (version == MLD_VERSION_1) { /* * Compute the "Older Version Querier Present" timer as per * Section 9.12. */ old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; old_version_timer *= PR_SLOWHZ; mli->mli_v1_timer = old_version_timer; } if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { mli->mli_version = MLD_VERSION_1; mld_v2_cancel_link_timers(mli); } } /* * Cancel pending MLDv2 timers for the given link and all groups * joined on it; state-change, general-query, and group-query timers. */ static void mld_v2_cancel_link_timers(struct mld_ifsoftc *mli) { struct ifmultiaddr *ifma, *next; struct ifnet *ifp; struct in6_multi *inm; struct in6_multi_head inmh; CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, mli->mli_ifp, if_name(mli->mli_ifp)); SLIST_INIT(&inmh); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); /* * Fast-track this potentially expensive operation * by checking all the global 'timer pending' flags. */ if (!V_interface_timers_running6 && !V_state_change_timers_running6 && !V_current_state_timers_running6) return; mli->mli_v2_timer = 0; ifp = mli->mli_ifp; IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET6 || ifma->ifma_protospec == NULL) continue; inm = (struct in6_multi *)ifma->ifma_protospec; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: break; case MLD_LEAVING_MEMBER: /* * If we are leaving the group and switching * version, we need to release the final * reference held for issuing the INCLUDE {}. */ in6m_rele_locked(&inmh, inm); ifma->ifma_protospec = NULL; /* FALLTHROUGH */ case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: in6m_clear_recorded(inm); /* FALLTHROUGH */ case MLD_REPORTING_MEMBER: inm->in6m_sctimer = 0; inm->in6m_timer = 0; inm->in6m_state = MLD_REPORTING_MEMBER; /* * Free any pending MLDv2 state-change records. */ mbufq_drain(&inm->in6m_scq); break; } if (__predict_false(ifma6_restart)) { ifma6_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); in6m_release_list_deferred(&inmh); } /* * Global slowtimo handler. * VIMAGE: Timeout handlers are expected to service all vimages. */ void mld_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); mld_slowtimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Per-vnet slowtimo handler. */ static void mld_slowtimo_vnet(void) { struct mld_ifsoftc *mli; MLD_LOCK(); LIST_FOREACH(mli, &V_mli_head, mli_link) { mld_v1_process_querier_timers(mli); } MLD_UNLOCK(); } /* * Update the Older Version Querier Present timers for a link. * See Section 9.12 of RFC 3810. */ static void mld_v1_process_querier_timers(struct mld_ifsoftc *mli) { MLD_LOCK_ASSERT(); if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { /* * MLDv1 Querier Present timer expired; revert to MLDv2. */ CTR5(KTR_MLD, "%s: transition from v%d -> v%d on %p(%s)", __func__, mli->mli_version, MLD_VERSION_2, mli->mli_ifp, if_name(mli->mli_ifp)); mli->mli_version = MLD_VERSION_2; } } /* * Transmit an MLDv1 report immediately. */ static int mld_v1_transmit_report(struct in6_multi *in6m, const int type) { struct ifnet *ifp; struct in6_ifaddr *ia; struct ip6_hdr *ip6; struct mbuf *mh, *md; struct mld_hdr *mld; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); ifp = in6m->in6m_ifp; ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); /* ia may be NULL if link-local address is tentative. */ mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (ENOMEM); } md = m_get(M_NOWAIT, MT_DATA); if (md == NULL) { m_free(mh); if (ia != NULL) ifa_free(&ia->ia_ifa); return (ENOMEM); } mh->m_next = md; /* * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so * that ether_output() does not need to allocate another mbuf * for the header in the most common case. */ M_ALIGN(mh, sizeof(struct ip6_hdr)); mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); mh->m_len = sizeof(struct ip6_hdr); ip6 = mtod(mh, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; ip6->ip6_dst = in6m->in6m_addr; md->m_len = sizeof(struct mld_hdr); mld = mtod(md, struct mld_hdr *); mld->mld_type = type; mld->mld_code = 0; mld->mld_cksum = 0; mld->mld_maxdelay = 0; mld->mld_reserved = 0; mld->mld_addr = in6m->in6m_addr; in6_clearscope(&mld->mld_addr); mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); mld_save_context(mh, ifp); mh->m_flags |= M_MLDV1; mld_dispatch_packet(mh); if (ia != NULL) ifa_free(&ia->ia_ifa); return (0); } /* * Process a state change from the upper layer for the given IPv6 group. * * Each socket holds a reference on the in_multi in its own ip_moptions. * The socket layer will have made the necessary updates to.the group * state, it is now up to MLD to issue a state change report if there * has been any change between T0 (when the last state-change was issued) * and T1 (now). * * We use the MLDv2 state machine at group level. The MLd module * however makes the decision as to which MLD protocol version to speak. * A state change *from* INCLUDE {} always means an initial join. * A state change *to* INCLUDE {} always means a final leave. * * If delay is non-zero, and the state change is an initial multicast * join, the state change report will be delayed by 'delay' ticks * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise * the initial MLDv2 state change report will be delayed by whichever * is sooner, a pending state-change timer or delay itself. * * VIMAGE: curvnet should have been set by caller, as this routine * is called from the socket option handlers. */ int mld_change_state(struct in6_multi *inm, const int delay) { struct mld_ifsoftc *mli; struct ifnet *ifp; int error; IN6_MULTI_LIST_LOCK_ASSERT(); error = 0; /* * Try to detect if the upper layer just asked us to change state * for an interface which has now gone away. */ KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); ifp = inm->in6m_ifma->ifma_ifp; if (ifp != NULL) { /* * Sanity check that netinet6's notion of ifp is the * same as net's. */ KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); } MLD_LOCK(); mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); /* * If we detect a state transition to or from MCAST_UNDEFINED * for this group, then we are starting or finishing an MLD * life cycle for this group. */ if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_MLD, "%s: initial join", __func__); error = mld_initial_join(inm, mli, delay); goto out_locked; } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_MLD, "%s: final leave", __func__); mld_final_leave(inm, mli); goto out_locked; } } else { CTR1(KTR_MLD, "%s: filter set change", __func__); } error = mld_handle_state_change(inm, mli); out_locked: MLD_UNLOCK(); return (error); } /* * Perform the initial join for an MLD group. * * When joining a group: * If the group should have its MLD traffic suppressed, do nothing. * MLDv1 starts sending MLDv1 host membership reports. * MLDv2 will schedule an MLDv2 state-change report containing the * initial state of the membership. * * If the delay argument is non-zero, then we must delay sending the * initial state change for delay ticks (in units of PR_FASTHZ). */ static int mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli, const int delay) { struct ifnet *ifp; struct mbufq *mq; int error, retval, syncstates; int odelay; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); error = 0; syncstates = 1; ifp = inm->in6m_ifp; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); /* * Groups joined on loopback or marked as 'not reported', * enter the MLD_SILENT_MEMBER state and * are never reported in any protocol exchanges. * All other groups enter the appropriate state machine * for the version in use on this link. * A link marked as MLIF_SILENT causes MLD to be completely * disabled for the link. */ if ((ifp->if_flags & IFF_LOOPBACK) || (mli->mli_flags & MLIF_SILENT) || !mld_is_addr_reported(&inm->in6m_addr)) { CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); inm->in6m_state = MLD_SILENT_MEMBER; inm->in6m_timer = 0; } else { /* * Deal with overlapping in_multi lifecycle. * If this group was LEAVING, then make sure * we drop the reference we picked up to keep the * group around for the final INCLUDE {} enqueue. */ if (mli->mli_version == MLD_VERSION_2 && inm->in6m_state == MLD_LEAVING_MEMBER) in6m_release_deferred(inm); inm->in6m_state = MLD_REPORTING_MEMBER; switch (mli->mli_version) { case MLD_VERSION_1: /* * If a delay was provided, only use it if * it is greater than the delay normally * used for an MLDv1 state change report, * and delay sending the initial MLDv1 report * by not transitioning to the IDLE state. */ odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); if (delay) { inm->in6m_timer = max(delay, odelay); V_current_state_timers_running6 = 1; } else { inm->in6m_state = MLD_IDLE_MEMBER; error = mld_v1_transmit_report(inm, MLD_LISTENER_REPORT); if (error == 0) { inm->in6m_timer = odelay; V_current_state_timers_running6 = 1; } } break; case MLD_VERSION_2: /* * Defer update of T0 to T1, until the first copy * of the state change has been transmitted. */ syncstates = 0; /* * Immediately enqueue a State-Change Report for * this interface, freeing any previous reports. * Don't kick the timers if there is nothing to do, * or if an error occurred. */ mq = &inm->in6m_scq; mbufq_drain(mq); retval = mld_v2_enqueue_group_record(mq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) { error = retval * -1; break; } /* * Schedule transmission of pending state-change * report up to RV times for this link. The timer * will fire at the next mld_fasttimo (~200ms), * giving us an opportunity to merge the reports. * * If a delay was provided to this function, only * use this delay if sooner than the existing one. */ KASSERT(mli->mli_rv > 1, ("%s: invalid robustness %d", __func__, mli->mli_rv)); inm->in6m_scrv = mli->mli_rv; if (delay) { if (inm->in6m_sctimer > 1) { inm->in6m_sctimer = min(inm->in6m_sctimer, delay); } else inm->in6m_sctimer = delay; } else inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; error = 0; break; } } /* * Only update the T0 state if state change is atomic, * i.e. we don't need to wait for a timer to fire before we * can consider the state change to have been communicated. */ if (syncstates) { in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); } return (error); } /* * Issue an intermediate state change during the life-cycle. */ static int mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli) { struct ifnet *ifp; int retval; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); ifp = inm->in6m_ifp; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); if ((ifp->if_flags & IFF_LOOPBACK) || (mli->mli_flags & MLIF_SILENT) || !mld_is_addr_reported(&inm->in6m_addr) || (mli->mli_version != MLD_VERSION_2)) { if (!mld_is_addr_reported(&inm->in6m_addr)) { CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); } CTR1(KTR_MLD, "%s: nothing to do", __func__); in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); return (0); } mbufq_drain(&inm->in6m_scq); retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) return (-retval); /* * If record(s) were enqueued, start the state-change * report timer for this group. */ inm->in6m_scrv = mli->mli_rv; inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; return (0); } /* * Perform the final leave for a multicast address. * * When leaving a group: * MLDv1 sends a DONE message, if and only if we are the reporter. * MLDv2 enqueues a state-change report containing a transition * to INCLUDE {} for immediate transmission. */ static void mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli) { int syncstates; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif syncstates = 1; CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_LEAVING_MEMBER: /* Already leaving or left; do nothing. */ CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: if (mli->mli_version == MLD_VERSION_1) { #ifdef INVARIANTS if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) panic("%s: MLDv2 state reached, not MLDv2 mode", __func__); #endif mld_v1_transmit_report(inm, MLD_LISTENER_DONE); inm->in6m_state = MLD_NOT_MEMBER; V_current_state_timers_running6 = 1; } else if (mli->mli_version == MLD_VERSION_2) { /* * Stop group timer and all pending reports. * Immediately enqueue a state-change report * TO_IN {} to be sent on the next fast timeout, * giving us an opportunity to merge reports. */ mbufq_drain(&inm->in6m_scq); inm->in6m_timer = 0; inm->in6m_scrv = mli->mli_rv; CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " "pending retransmissions.", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp), inm->in6m_scrv); if (inm->in6m_scrv == 0) { inm->in6m_state = MLD_NOT_MEMBER; inm->in6m_sctimer = 0; } else { int retval; in6m_acquire_locked(inm); retval = mld_v2_enqueue_group_record( &inm->in6m_scq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); KASSERT(retval != 0, ("%s: enqueue record = %d", __func__, retval)); inm->in6m_state = MLD_LEAVING_MEMBER; inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; syncstates = 0; } break; } break; case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: /* Our reports are suppressed; do nothing. */ break; } if (syncstates) { in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); } } /* * Enqueue an MLDv2 group record to the given output queue. * * If is_state_change is zero, a current-state record is appended. * If is_state_change is non-zero, a state-change report is appended. * * If is_group_query is non-zero, an mbuf packet chain is allocated. * If is_group_query is zero, and if there is a packet with free space * at the tail of the queue, it will be appended to providing there * is enough free space. * Otherwise a new mbuf packet chain is allocated. * * If is_source_query is non-zero, each source is checked to see if * it was recorded for a Group-Source query, and will be omitted if * it is not both in-mode and recorded. * * If use_block_allow is non-zero, state change reports for initial join * and final leave, on an inclusive mode group with a source list, will be * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. * * The function will attempt to allocate leading space in the packet * for the IPv6+ICMP headers to be prepended without fragmenting the chain. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm, const int is_state_change, const int is_group_query, const int is_source_query, const int use_block_allow) { struct mldv2_record mr; struct mldv2_record *pmr; struct ifnet *ifp; struct ip6_msource *ims, *nims; struct mbuf *m0, *m, *md; int is_filter_list_change; int minrec0len, m0srcs, msrcs, nbytes, off; int record_has_sources; int now; int type; uint8_t mode; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); ifp = inm->in6m_ifp; is_filter_list_change = 0; m = NULL; m0 = NULL; m0srcs = 0; msrcs = 0; nbytes = 0; nims = NULL; record_has_sources = 1; pmr = NULL; type = MLD_DO_NOTHING; mode = inm->in6m_st[1].iss_fmode; /* * If we did not transition out of ASM mode during t0->t1, * and there are no source nodes to process, we can skip * the generation of source records. */ if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && inm->in6m_nsrc == 0) record_has_sources = 0; if (is_state_change) { /* * Queue a state change record. * If the mode did not change, and there are non-ASM * listeners or source filters present, * we potentially need to issue two records for the group. * If there are ASM listeners, and there was no filter * mode transition of any kind, do nothing. * * If we are transitioning to MCAST_UNDEFINED, we need * not send any sources. A transition to/from this state is * considered inclusive with some special treatment. * * If we are rewriting initial joins/leaves to use * ALLOW/BLOCK, and the group's membership is inclusive, * we need to send sources in all cases. */ if (mode != inm->in6m_st[0].iss_fmode) { if (mode == MCAST_EXCLUDE) { CTR1(KTR_MLD, "%s: change to EXCLUDE", __func__); type = MLD_CHANGE_TO_EXCLUDE_MODE; } else { CTR1(KTR_MLD, "%s: change to INCLUDE", __func__); if (use_block_allow) { /* * XXX * Here we're interested in state * edges either direction between * MCAST_UNDEFINED and MCAST_INCLUDE. * Perhaps we should just check * the group state, rather than * the filter mode. */ if (mode == MCAST_UNDEFINED) { type = MLD_BLOCK_OLD_SOURCES; } else { type = MLD_ALLOW_NEW_SOURCES; } } else { type = MLD_CHANGE_TO_INCLUDE_MODE; if (mode == MCAST_UNDEFINED) record_has_sources = 0; } } } else { if (record_has_sources) { is_filter_list_change = 1; } else { type = MLD_DO_NOTHING; } } } else { /* * Queue a current state record. */ if (mode == MCAST_EXCLUDE) { type = MLD_MODE_IS_EXCLUDE; } else if (mode == MCAST_INCLUDE) { type = MLD_MODE_IS_INCLUDE; KASSERT(inm->in6m_st[1].iss_asm == 0, ("%s: inm %p is INCLUDE but ASM count is %d", __func__, inm, inm->in6m_st[1].iss_asm)); } } /* * Generate the filter list changes using a separate function. */ if (is_filter_list_change) return (mld_v2_enqueue_filter_change(mq, inm)); if (type == MLD_DO_NOTHING) { CTR3(KTR_MLD, "%s: nothing to do for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); return (0); } /* * If any sources are present, we must be able to fit at least * one in the trailing space of the tail packet's mbuf, * ideally more. */ minrec0len = sizeof(struct mldv2_record); if (record_has_sources) minrec0len += sizeof(struct in6_addr); CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, mld_rec_type_to_str(type), ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); /* * Check if we have a packet in the tail of the queue for this * group into which the first group record for this group will fit. * Otherwise allocate a new packet. * Always allocate leading space for IP6+RA+ICMPV6+REPORT. * Note: Group records for G/GSR query responses MUST be sent * in their own packet. */ m0 = mbufq_last(mq); if (!is_group_query && m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && (m0->m_pkthdr.len + minrec0len) < (ifp->if_mtu - MLD_MTUSPACE)) { m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); m = m0; CTR1(KTR_MLD, "%s: use existing packet", __func__); } else { if (mbufq_full(mq)) { CTR1(KTR_MLD, "%s: outbound queue full", __func__); return (-ENOMEM); } m = NULL; m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); if (!is_state_change && !is_group_query) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (-ENOMEM); mld_save_context(m, ifp); CTR1(KTR_MLD, "%s: allocated first packet", __func__); } /* * Append group record. * If we have sources, we don't know how many yet. */ mr.mr_type = type; mr.mr_datalen = 0; mr.mr_numsrc = 0; mr.mr_addr = inm->in6m_addr; in6_clearscope(&mr.mr_addr); if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct mldv2_record); /* * Append as many sources as will fit in the first packet. * If we are appending to a new packet, the chain allocation * may potentially use clusters; use m_getptr() in this case. * If we are appending to an existing packet, we need to obtain * a pointer to the group record after m_append(), in case a new * mbuf was allocated. * * Only append sources which are in-mode at t1. If we are * transitioning to MCAST_UNDEFINED state on the group, and * use_block_allow is zero, do not include source entries. * Otherwise, we need to include this source in the report. * * Only report recorded sources in our filter set when responding * to a group-source query. */ if (record_has_sources) { if (m == m0) { md = m_last(m); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + md->m_len - nbytes); } else { md = m_getptr(m, 0, &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); } msrcs = 0; RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); CTR2(KTR_MLD, "%s: node is %d", __func__, now); if ((now != mode) || (now == mode && (!use_block_allow && mode == MCAST_UNDEFINED))) { CTR1(KTR_MLD, "%s: skip node", __func__); continue; } if (is_source_query && ims->im6s_stp == 0) { CTR1(KTR_MLD, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_MLD, "%s: append node", __func__); if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct in6_addr); ++msrcs; if (msrcs == m0srcs) break; } CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, msrcs); pmr->mr_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(struct in6_addr)); } if (is_source_query && msrcs == 0) { CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); if (m != m0) m_freem(m); return (0); } /* * We are good to go with first packet. */ if (m != m0) { CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); m->m_pkthdr.PH_vt.vt_nrecs = 1; mbufq_enqueue(mq, m); } else m->m_pkthdr.PH_vt.vt_nrecs++; /* * No further work needed if no source list in packet(s). */ if (!record_has_sources) return (nbytes); /* * Whilst sources remain to be announced, we need to allocate * a new packet and fill out as many sources as will fit. * Always try for a cluster first. */ while (nims != NULL) { if (mbufq_full(mq)) { CTR1(KTR_MLD, "%s: outbound queue full", __func__); return (-ENOMEM); } m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (-ENOMEM); mld_save_context(m, ifp); md = m_getptr(m, 0, &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); CTR1(KTR_MLD, "%s: allocated next packet", __func__); if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 1; nbytes += sizeof(struct mldv2_record); m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); msrcs = 0; RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); if ((now != mode) || (now == mode && (!use_block_allow && mode == MCAST_UNDEFINED))) { CTR1(KTR_MLD, "%s: skip node", __func__); continue; } if (is_source_query && ims->im6s_stp == 0) { CTR1(KTR_MLD, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_MLD, "%s: append node", __func__); if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } ++msrcs; if (msrcs == m0srcs) break; } pmr->mr_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(struct in6_addr)); CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); mbufq_enqueue(mq, m); } return (nbytes); } /* * Type used to mark record pass completion. * We exploit the fact we can cast to this easily from the * current filter modes on each ip_msource node. */ typedef enum { REC_NONE = 0x00, /* MCAST_UNDEFINED */ REC_ALLOW = 0x01, /* MCAST_INCLUDE */ REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ REC_FULL = REC_ALLOW | REC_BLOCK } rectype_t; /* * Enqueue an MLDv2 filter list change to the given output queue. * * Source list filter state is held in an RB-tree. When the filter list * for a group is changed without changing its mode, we need to compute * the deltas between T0 and T1 for each source in the filter set, * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. * * As we may potentially queue two record types, and the entire R-B tree * needs to be walked at once, we break this out into its own function * so we can generate a tightly packed queue of packets. * * XXX This could be written to only use one tree walk, although that makes * serializing into the mbuf chains a bit harder. For now we do two walks * which makes things easier on us, and it may or may not be harder on * the L2 cache. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm) { static const int MINRECLEN = sizeof(struct mldv2_record) + sizeof(struct in6_addr); struct ifnet *ifp; struct mldv2_record mr; struct mldv2_record *pmr; struct ip6_msource *ims, *nims; struct mbuf *m, *m0, *md; int m0srcs, nbytes, npbytes, off, rsrcs, schanged; int nallow, nblock; uint8_t mode, now, then; rectype_t crt, drt, nrt; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); if (inm->in6m_nsrc == 0 || (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) return (0); ifp = inm->in6m_ifp; /* interface */ mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ crt = REC_NONE; /* current group record type */ drt = REC_NONE; /* mask of completed group record types */ nrt = REC_NONE; /* record type for current node */ m0srcs = 0; /* # source which will fit in current mbuf chain */ npbytes = 0; /* # of bytes appended this packet */ nbytes = 0; /* # of bytes appended to group's state-change queue */ rsrcs = 0; /* # sources encoded in current record */ schanged = 0; /* # nodes encoded in overall filter change */ nallow = 0; /* # of source entries in ALLOW_NEW */ nblock = 0; /* # of source entries in BLOCK_OLD */ nims = NULL; /* next tree node pointer */ /* * For each possible filter record mode. * The first kind of source we encounter tells us which * is the first kind of record we start appending. * If a node transitioned to UNDEFINED at t1, its mode is treated * as the inverse of the group's filter mode. */ while (drt != REC_FULL) { do { m0 = mbufq_last(mq); if (m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && (m0->m_pkthdr.len + MINRECLEN) < (ifp->if_mtu - MLD_MTUSPACE)) { m = m0; m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); CTR1(KTR_MLD, "%s: use previous packet", __func__); } else { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { CTR1(KTR_MLD, "%s: m_get*() failed", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 0; mld_save_context(m, ifp); m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); npbytes = 0; CTR1(KTR_MLD, "%s: allocated new packet", __func__); } /* * Append the MLD group record header to the * current packet's data area. * Recalculate pointer to free space for next * group record, in case m_append() allocated * a new mbuf or cluster. */ memset(&mr, 0, sizeof(mr)); mr.mr_addr = inm->in6m_addr; in6_clearscope(&mr.mr_addr); if (!m_append(m, sizeof(mr), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed", __func__); return (-ENOMEM); } npbytes += sizeof(struct mldv2_record); if (m != m0) { /* new packet; offset in chain */ md = m_getptr(m, npbytes - sizeof(struct mldv2_record), &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); } else { /* current packet; offset from last append */ md = m_last(m); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + md->m_len - sizeof(struct mldv2_record)); } /* * Begin walking the tree for this record type * pass, or continue from where we left off * previously if we had to allocate a new packet. * Only report deltas in-mode at t1. * We need not report included sources as allowed * if we are in inclusive mode on the group, * however the converse is not true. */ rsrcs = 0; if (nims == NULL) { nims = RB_MIN(ip6_msource_tree, &inm->in6m_srcs); } RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); then = im6s_get_mode(inm, ims, 0); CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", __func__, then, now); if (now == then) { CTR1(KTR_MLD, "%s: skip unchanged", __func__); continue; } if (mode == MCAST_EXCLUDE && now == MCAST_INCLUDE) { CTR1(KTR_MLD, "%s: skip IN src on EX group", __func__); continue; } nrt = (rectype_t)now; if (nrt == REC_NONE) nrt = (rectype_t)(~mode & REC_FULL); if (schanged++ == 0) { crt = nrt; } else if (crt != nrt) continue; if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed", __func__); return (-ENOMEM); } nallow += !!(crt == REC_ALLOW); nblock += !!(crt == REC_BLOCK); if (++rsrcs == m0srcs) break; } /* * If we did not append any tree nodes on this * pass, back out of allocations. */ if (rsrcs == 0) { npbytes -= sizeof(struct mldv2_record); if (m != m0) { CTR1(KTR_MLD, "%s: m_free(m)", __func__); m_freem(m); } else { CTR1(KTR_MLD, "%s: m_adj(m, -mr)", __func__); m_adj(m, -((int)sizeof( struct mldv2_record))); } continue; } npbytes += (rsrcs * sizeof(struct in6_addr)); if (crt == REC_ALLOW) pmr->mr_type = MLD_ALLOW_NEW_SOURCES; else if (crt == REC_BLOCK) pmr->mr_type = MLD_BLOCK_OLD_SOURCES; pmr->mr_numsrc = htons(rsrcs); /* * Count the new group record, and enqueue this * packet if it wasn't already queued. */ m->m_pkthdr.PH_vt.vt_nrecs++; if (m != m0) mbufq_enqueue(mq, m); nbytes += npbytes; } while (nims != NULL); drt |= crt; crt = (~crt & REC_FULL); } CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, nallow, nblock); return (nbytes); } static int mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq) { struct mbufq *gq; struct mbuf *m; /* pending state-change */ struct mbuf *m0; /* copy of pending state-change */ struct mbuf *mt; /* last state-change in packet */ int docopy, domerge; u_int recslen; docopy = 0; domerge = 0; recslen = 0; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); /* * If there are further pending retransmissions, make a writable * copy of each queued state-change message before merging. */ if (inm->in6m_scrv > 0) docopy = 1; gq = &inm->in6m_scq; #ifdef KTR if (mbufq_first(gq) == NULL) { CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", __func__, inm); } #endif m = mbufq_first(gq); while (m != NULL) { /* * Only merge the report into the current packet if * there is sufficient space to do so; an MLDv2 report * packet may only contain 65,535 group records. * Always use a simple mbuf chain concatentation to do this, * as large state changes for single groups may have * allocated clusters. */ domerge = 0; mt = mbufq_last(scq); if (mt != NULL) { recslen = m_length(m, NULL); if ((mt->m_pkthdr.PH_vt.vt_nrecs + m->m_pkthdr.PH_vt.vt_nrecs <= MLD_V2_REPORT_MAXRECS) && (mt->m_pkthdr.len + recslen <= (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) domerge = 1; } if (!domerge && mbufq_full(gq)) { CTR2(KTR_MLD, "%s: outbound queue full, skipping whole packet %p", __func__, m); mt = m->m_nextpkt; if (!docopy) m_freem(m); m = mt; continue; } if (!docopy) { CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); m0 = mbufq_dequeue(gq); m = m0->m_nextpkt; } else { CTR2(KTR_MLD, "%s: copying %p", __func__, m); m0 = m_dup(m, M_NOWAIT); if (m0 == NULL) return (ENOMEM); m0->m_nextpkt = NULL; m = m->m_nextpkt; } if (!domerge) { CTR3(KTR_MLD, "%s: queueing %p to scq %p)", __func__, m0, scq); mbufq_enqueue(scq, m0); } else { struct mbuf *mtl; /* last mbuf of packet mt */ CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", __func__, m0, mt); mtl = m_last(mt); m0->m_flags &= ~M_PKTHDR; mt->m_pkthdr.len += recslen; mt->m_pkthdr.PH_vt.vt_nrecs += m0->m_pkthdr.PH_vt.vt_nrecs; mtl->m_next = m0; } } return (0); } /* * Respond to a pending MLDv2 General Query. */ static void mld_v2_dispatch_general_query(struct mld_ifsoftc *mli) { struct ifmultiaddr *ifma; struct ifnet *ifp; struct in6_multi *inm; int retval; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli->mli_version == MLD_VERSION_2, ("%s: called when version %d", __func__, mli->mli_version)); /* * Check that there are some packets queued. If so, send them first. * For large number of groups the reply to general query can take * many packets, we should finish sending them before starting of * queuing the new reply. */ if (mbufq_len(&mli->mli_gq) != 0) goto send; ifp = mli->mli_ifp; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET6 || ifma->ifma_protospec == NULL) continue; inm = (struct in6_multi *)ifma->ifma_protospec; KASSERT(ifp == inm->in6m_ifp, ("%s: inconsistent ifp", __func__)); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: inm->in6m_state = MLD_REPORTING_MEMBER; retval = mld_v2_enqueue_group_record(&mli->mli_gq, inm, 0, 0, 0, 0); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } IF_ADDR_RUNLOCK(ifp); send: mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); /* * Slew transmission of bursts over 500ms intervals. */ if (mbufq_first(&mli->mli_gq) != NULL) { mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( MLD_RESPONSE_BURST_INTERVAL); V_interface_timers_running6 = 1; } } /* * Transmit the next pending message in the output queue. * * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. * MRT: Nothing needs to be done, as MLD traffic is always local to * a link and uses a link-scope multicast address. */ static void mld_dispatch_packet(struct mbuf *m) { struct ip6_moptions im6o; struct ifnet *ifp; struct ifnet *oifp; struct mbuf *m0; struct mbuf *md; struct ip6_hdr *ip6; struct mld_hdr *mld; int error; int off; int type; uint32_t ifindex; CTR2(KTR_MLD, "%s: transmit %p", __func__, m); /* * Set VNET image pointer from enqueued mbuf chain * before doing anything else. Whilst we use interface * indexes to guard against interface detach, they are * unique to each VIMAGE and must be retrieved. */ ifindex = mld_restore_context(m); /* * Check if the ifnet still exists. This limits the scope of * any race in the absence of a global ifp lock for low cost * (an array lookup). */ ifp = ifnet_byindex(ifindex); if (ifp == NULL) { CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", __func__, m, ifindex); m_freem(m); IP6STAT_INC(ip6s_noroute); goto out; } im6o.im6o_multicast_hlim = 1; im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); im6o.im6o_multicast_ifp = ifp; if (m->m_flags & M_MLDV1) { m0 = m; } else { m0 = mld_v2_encap_report(ifp, m); if (m0 == NULL) { CTR2(KTR_MLD, "%s: dropped %p", __func__, m); IP6STAT_INC(ip6s_odropped); goto out; } } mld_scrub_context(m0); m_clrprotoflags(m); m0->m_pkthdr.rcvif = V_loif; ip6 = mtod(m0, struct ip6_hdr *); #if 0 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ #else /* * XXX XXX Break some KPI rules to prevent an LOR which would * occur if we called in6_setscope() at transmission. * See comments at top of file. */ MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); #endif /* * Retrieve the ICMPv6 type before handoff to ip6_output(), * so we can bump the stats. */ md = m_getptr(m0, sizeof(struct ip6_hdr), &off); mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); type = mld->mld_type; error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, &oifp, NULL); if (error) { CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); goto out; } ICMP6STAT_INC(icp6s_outhist[type]); if (oifp != NULL) { icmp6_ifstat_inc(oifp, ifs6_out_msg); switch (type) { case MLD_LISTENER_REPORT: case MLDV2_LISTENER_REPORT: icmp6_ifstat_inc(oifp, ifs6_out_mldreport); break; case MLD_LISTENER_DONE: icmp6_ifstat_inc(oifp, ifs6_out_mlddone); break; } } out: return; } /* * Encapsulate an MLDv2 report. * * KAME IPv6 requires that hop-by-hop options be passed separately, * and that the IPv6 header be prepended in a separate mbuf. * * Returns a pointer to the new mbuf chain head, or NULL if the * allocation failed. */ static struct mbuf * mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) { struct mbuf *mh; struct mldv2_report *mld; struct ip6_hdr *ip6; struct in6_ifaddr *ia; int mldreclen; KASSERT(ifp != NULL, ("%s: null ifp", __func__)); KASSERT((m->m_flags & M_PKTHDR), ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); /* * RFC3590: OK to send as :: or tentative during DAD. */ ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); if (ia == NULL) CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { if (ia != NULL) ifa_free(&ia->ia_ifa); m_freem(m); return (NULL); } M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); mldreclen = m_length(m, NULL); CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report) + mldreclen; ip6 = mtod(mh, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; if (ia != NULL) ifa_free(&ia->ia_ifa); ip6->ip6_dst = in6addr_linklocal_allv2routers; /* scope ID will be set in netisr */ mld = (struct mldv2_report *)(ip6 + 1); mld->mld_type = MLDV2_LISTENER_REPORT; mld->mld_code = 0; mld->mld_cksum = 0; mld->mld_v2_reserved = 0; mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); m->m_pkthdr.PH_vt.vt_nrecs = 0; mh->m_next = m; mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); return (mh); } #ifdef KTR static char * mld_rec_type_to_str(const int type) { switch (type) { case MLD_CHANGE_TO_EXCLUDE_MODE: return "TO_EX"; break; case MLD_CHANGE_TO_INCLUDE_MODE: return "TO_IN"; break; case MLD_MODE_IS_EXCLUDE: return "MODE_EX"; break; case MLD_MODE_IS_INCLUDE: return "MODE_IN"; break; case MLD_ALLOW_NEW_SOURCES: return "ALLOW_NEW"; break; case MLD_BLOCK_OLD_SOURCES: return "BLOCK_OLD"; break; default: break; } return "unknown"; } #endif static void mld_init(void *unused __unused) { CTR1(KTR_MLD, "%s: initializing", __func__); MLD_LOCK_INIT(); ip6_initpktopts(&mld_po); mld_po.ip6po_hlim = 1; mld_po.ip6po_hbh = &mld_ra.hbh; mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; mld_po.ip6po_flags = IP6PO_DONTFRAG; } SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL); static void mld_uninit(void *unused __unused) { CTR1(KTR_MLD, "%s: tearing down", __func__); MLD_LOCK_DESTROY(); } SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL); static void vnet_mld_init(const void *unused __unused) { CTR1(KTR_MLD, "%s: initializing", __func__); LIST_INIT(&V_mli_head); } VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init, NULL); static void vnet_mld_uninit(const void *unused __unused) { /* This can happen if we shutdown the network stack. */ CTR1(KTR_MLD, "%s: tearing down", __func__); } VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit, NULL); static int mld_modevent(module_t mod, int type, void *unused __unused) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t mld_mod = { "mld", mld_modevent, 0 }; DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY); Index: head/sys/netinet6/nd6.c =================================================================== --- head/sys/netinet6/nd6.c (revision 336675) +++ head/sys/netinet6/nd6.c (revision 336676) @@ -1,2740 +1,2740 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ #define SIN6(s) ((const struct sockaddr_in6 *)(s)) MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery"); /* timer values */ VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */ VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */ VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */ VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */ VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for * local traffic */ VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage * collection timer */ /* preventing too many loops in ND option parsing */ -static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ +VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper * layer hints */ -static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved +VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved * ND entries */ #define V_nd6_maxndopt VNET(nd6_maxndopt) #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen) #ifdef ND6_DEBUG VNET_DEFINE(int, nd6_debug) = 1; #else VNET_DEFINE(int, nd6_debug) = 0; #endif static eventhandler_tag lle_event_eh, iflladdr_event_eh; VNET_DEFINE(struct nd_drhead, nd_defrouter); VNET_DEFINE(struct nd_prhead, nd_prefix); VNET_DEFINE(struct rwlock, nd6_lock); VNET_DEFINE(uint64_t, nd6_list_genid); VNET_DEFINE(struct mtx, nd6_onlink_mtx); VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL; #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int); static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *, struct ifnet *); static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); static void nd6_slowtimo(void *); static int regen_tmpaddr(struct in6_ifaddr *); static void nd6_free(struct llentry **, int); static void nd6_free_redirect(const struct llentry *); static void nd6_llinfo_timer(void *); static void nd6_llinfo_settimer_locked(struct llentry *, long); static void clear_llinfo_pqueue(struct llentry *); static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *); static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *, const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **); static int nd6_need_cache(struct ifnet *); -static VNET_DEFINE(struct callout, nd6_slowtimo_ch); +VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch); #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) VNET_DEFINE(struct callout, nd6_timer_ch); #define V_nd6_timer_ch VNET(nd6_timer_ch) static void nd6_lle_event(void *arg __unused, struct llentry *lle, int evt) { struct rt_addrinfo rtinfo; struct sockaddr_in6 dst; struct sockaddr_dl gw; struct ifnet *ifp; int type; int fibnum; LLE_WLOCK_ASSERT(lle); if (lltable_get_af(lle->lle_tbl) != AF_INET6) return; switch (evt) { case LLENTRY_RESOLVED: type = RTM_ADD; KASSERT(lle->la_flags & LLE_VALID, ("%s: %p resolved but not valid?", __func__, lle)); break; case LLENTRY_EXPIRED: type = RTM_DELETE; break; default: return; } ifp = lltable_get_ifp(lle->lle_tbl); bzero(&dst, sizeof(dst)); bzero(&gw, sizeof(gw)); bzero(&rtinfo, sizeof(rtinfo)); lltable_fill_sa_entry(lle, (struct sockaddr *)&dst); dst.sin6_scope_id = in6_getscopezone(ifp, in6_addrscope(&dst.sin6_addr)); gw.sdl_len = sizeof(struct sockaddr_dl); gw.sdl_family = AF_LINK; gw.sdl_alen = ifp->if_addrlen; gw.sdl_index = ifp->if_index; gw.sdl_type = ifp->if_type; if (evt == LLENTRY_RESOLVED) bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen); rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst; rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw; rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY; fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib; rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | ( type == RTM_ADD ? RTF_UP: 0), 0, fibnum); } /* * A handler for interface link layer address change event. */ static void nd6_iflladdr(void *arg __unused, struct ifnet *ifp) { lltable_update_ifaddr(LLTABLE6(ifp)); } void nd6_init(void) { mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF); rw_init(&V_nd6_lock, "nd6 list"); LIST_INIT(&V_nd_prefix); TAILQ_INIT(&V_nd_defrouter); /* Start timers. */ callout_init(&V_nd6_slowtimo_ch, 0); callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, nd6_slowtimo, curvnet); callout_init(&V_nd6_timer_ch, 0); callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet); nd6_dad_init(); if (IS_DEFAULT_VNET(curvnet)) { lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event, NULL, EVENTHANDLER_PRI_ANY); iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event, nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY); } } #ifdef VIMAGE void nd6_destroy() { callout_drain(&V_nd6_slowtimo_ch); callout_drain(&V_nd6_timer_ch); if (IS_DEFAULT_VNET(curvnet)) { EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh); EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh); } rw_destroy(&V_nd6_lock); mtx_destroy(&V_nd6_onlink_mtx); } #endif struct nd_ifinfo * nd6_ifattach(struct ifnet *ifp) { struct nd_ifinfo *nd; nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO); nd->initialized = 1; nd->chlim = IPV6_DEFHLIM; nd->basereachable = REACHABLE_TIME; nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); nd->retrans = RETRANS_TIMER; nd->flags = ND6_IFF_PERFORMNUD; /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by * default regardless of the V_ip6_auto_linklocal configuration to * give a reasonable default behavior. */ if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) || (ifp->if_flags & IFF_LOOPBACK)) nd->flags |= ND6_IFF_AUTO_LINKLOCAL; /* * A loopback interface does not need to accept RTADV. * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by * default regardless of the V_ip6_accept_rtadv configuration to * prevent the interface from accepting RA messages arrived * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV. */ if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK) && (ifp->if_type != IFT_BRIDGE)) nd->flags |= ND6_IFF_ACCEPT_RTADV; if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) nd->flags |= ND6_IFF_NO_RADR; /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ nd6_setmtu0(ifp, nd); return nd; } void nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd) { struct ifaddr *ifa, *next; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; /* stop DAD processing */ nd6_dad_stop(ifa); } IF_ADDR_RUNLOCK(ifp); free(nd, M_IP6NDP); } /* * Reset ND level link MTU. This function is called when the physical MTU * changes, which means we might have to adjust the ND level MTU. */ void nd6_setmtu(struct ifnet *ifp) { if (ifp->if_afdata[AF_INET6] == NULL) return; nd6_setmtu0(ifp, ND_IFINFO(ifp)); } /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ void nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) { u_int32_t omaxmtu; omaxmtu = ndi->maxmtu; ndi->maxmtu = ifp->if_mtu; /* * Decreasing the interface MTU under IPV6 minimum MTU may cause * undesirable situation. We thus notify the operator of the change * explicitly. The check for omaxmtu is necessary to restrict the * log to the case of changing the MTU, not initializing it. */ if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { log(LOG_NOTICE, "nd6_setmtu0: " "new link MTU on %s (%lu) is too small for IPv6\n", if_name(ifp), (unsigned long)ndi->maxmtu); } if (ndi->maxmtu > V_in6_maxmtu) in6_setmaxmtu(); /* check all interfaces just in case */ } void nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) { bzero(ndopts, sizeof(*ndopts)); ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; ndopts->nd_opts_last = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); if (icmp6len == 0) { ndopts->nd_opts_done = 1; ndopts->nd_opts_search = NULL; } } /* * Take one ND option. */ struct nd_opt_hdr * nd6_option(union nd_opts *ndopts) { struct nd_opt_hdr *nd_opt; int olen; KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", __func__)); if (ndopts->nd_opts_search == NULL) return NULL; if (ndopts->nd_opts_done) return NULL; nd_opt = ndopts->nd_opts_search; /* make sure nd_opt_len is inside the buffer */ if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { bzero(ndopts, sizeof(*ndopts)); return NULL; } olen = nd_opt->nd_opt_len << 3; if (olen == 0) { /* * Message validation requires that all included * options have a length that is greater than zero. */ bzero(ndopts, sizeof(*ndopts)); return NULL; } ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); if (ndopts->nd_opts_search > ndopts->nd_opts_last) { /* option overruns the end of buffer, invalid */ bzero(ndopts, sizeof(*ndopts)); return NULL; } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { /* reached the end of options chain */ ndopts->nd_opts_done = 1; ndopts->nd_opts_search = NULL; } return nd_opt; } /* * Parse multiple ND options. * This function is much easier to use, for ND routines that do not need * multiple options of the same type. */ int nd6_options(union nd_opts *ndopts) { struct nd_opt_hdr *nd_opt; int i = 0; KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", __func__)); if (ndopts->nd_opts_search == NULL) return 0; while (1) { nd_opt = nd6_option(ndopts); if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { /* * Message validation requires that all included * options have a length that is greater than zero. */ ICMP6STAT_INC(icp6s_nd_badopt); bzero(ndopts, sizeof(*ndopts)); return -1; } if (nd_opt == NULL) goto skip1; switch (nd_opt->nd_opt_type) { case ND_OPT_SOURCE_LINKADDR: case ND_OPT_TARGET_LINKADDR: case ND_OPT_MTU: case ND_OPT_REDIRECTED_HEADER: case ND_OPT_NONCE: if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { nd6log((LOG_INFO, "duplicated ND6 option found (type=%d)\n", nd_opt->nd_opt_type)); /* XXX bark? */ } else { ndopts->nd_opt_array[nd_opt->nd_opt_type] = nd_opt; } break; case ND_OPT_PREFIX_INFORMATION: if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { ndopts->nd_opt_array[nd_opt->nd_opt_type] = nd_opt; } ndopts->nd_opts_pi_end = (struct nd_opt_prefix_info *)nd_opt; break; /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ case ND_OPT_RDNSS: /* RFC 6106 */ case ND_OPT_DNSSL: /* RFC 6106 */ /* * Silently ignore options we know and do not care about * in the kernel. */ break; default: /* * Unknown options must be silently ignored, * to accommodate future extension to the protocol. */ nd6log((LOG_DEBUG, "nd6_options: unsupported option %d - " "option ignored\n", nd_opt->nd_opt_type)); } skip1: i++; if (i > V_nd6_maxndopt) { ICMP6STAT_INC(icp6s_nd_toomanyopt); nd6log((LOG_INFO, "too many loop in nd opt\n")); break; } if (ndopts->nd_opts_done) break; } return 0; } /* * ND6 timer routine to handle ND6 entries */ static void nd6_llinfo_settimer_locked(struct llentry *ln, long tick) { int canceled; LLE_WLOCK_ASSERT(ln); if (tick < 0) { ln->la_expire = 0; ln->ln_ntick = 0; canceled = callout_stop(&ln->lle_timer); } else { ln->la_expire = time_uptime + tick / hz; LLE_ADDREF(ln); if (tick > INT_MAX) { ln->ln_ntick = tick - INT_MAX; canceled = callout_reset(&ln->lle_timer, INT_MAX, nd6_llinfo_timer, ln); } else { ln->ln_ntick = 0; canceled = callout_reset(&ln->lle_timer, tick, nd6_llinfo_timer, ln); } } if (canceled > 0) LLE_REMREF(ln); } /* * Gets source address of the first packet in hold queue * and stores it in @src. * Returns pointer to @src (if hold queue is not empty) or NULL. * * Set noinline to be dtrace-friendly */ static __noinline struct in6_addr * nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src) { struct ip6_hdr hdr; struct mbuf *m; if (ln->la_hold == NULL) return (NULL); /* * assume every packet in la_hold has the same IP header */ m = ln->la_hold; if (sizeof(hdr) > m->m_len) return (NULL); m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr); *src = hdr.ip6_src; return (src); } /* * Checks if we need to switch from STALE state. * * RFC 4861 requires switching from STALE to DELAY state * on first packet matching entry, waiting V_nd6_delay and * transition to PROBE state (if upper layer confirmation was * not received). * * This code performs a bit differently: * On packet hit we don't change state (but desired state * can be guessed by control plane). However, after V_nd6_delay * seconds code will transition to PROBE state (so DELAY state * is kinda skipped in most situations). * * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so * we perform the following upon entering STALE state: * * 1) Arm timer to run each V_nd6_delay seconds to make sure that * if packet was transmitted at the start of given interval, we * would be able to switch to PROBE state in V_nd6_delay seconds * as user expects. * * 2) Reschedule timer until original V_nd6_gctimer expires keeping * lle in STALE state (remaining timer value stored in lle_remtime). * * 3) Reschedule timer if packet was transmitted less that V_nd6_delay * seconds ago. * * Returns non-zero value if the entry is still STALE (storing * the next timer interval in @pdelay). * * Returns zero value if original timer expired or we need to switch to * PROBE (store that in @do_switch variable). */ static int nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch) { int nd_delay, nd_gctimer, r_skip_req; time_t lle_hittime; long delay; *do_switch = 0; nd_gctimer = V_nd6_gctimer; nd_delay = V_nd6_delay; LLE_REQ_LOCK(lle); r_skip_req = lle->r_skip_req; lle_hittime = lle->lle_hittime; LLE_REQ_UNLOCK(lle); if (r_skip_req > 0) { /* * Nonzero r_skip_req value was set upon entering * STALE state. Since value was not changed, no * packets were passed using this lle. Ask for * timer reschedule and keep STALE state. */ delay = (long)(MIN(nd_gctimer, nd_delay)); delay *= hz; if (lle->lle_remtime > delay) lle->lle_remtime -= delay; else { delay = lle->lle_remtime; lle->lle_remtime = 0; } if (delay == 0) { /* * The original ng6_gctime timeout ended, * no more rescheduling. */ return (0); } *pdelay = delay; return (1); } /* * Packet received. Verify timestamp */ delay = (long)(time_uptime - lle_hittime); if (delay < nd_delay) { /* * V_nd6_delay still not passed since the first * hit in STALE state. * Reshedule timer and return. */ *pdelay = (long)(nd_delay - delay) * hz; return (1); } /* Request switching to probe */ *do_switch = 1; return (0); } /* * Switch @lle state to new state optionally arming timers. * * Set noinline to be dtrace-friendly */ __noinline void nd6_llinfo_setstate(struct llentry *lle, int newstate) { struct ifnet *ifp; int nd_gctimer, nd_delay; long delay, remtime; delay = 0; remtime = 0; switch (newstate) { case ND6_LLINFO_INCOMPLETE: ifp = lle->lle_tbl->llt_ifp; delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000; break; case ND6_LLINFO_REACHABLE: if (!ND6_LLINFO_PERMANENT(lle)) { ifp = lle->lle_tbl->llt_ifp; delay = (long)ND_IFINFO(ifp)->reachable * hz; } break; case ND6_LLINFO_STALE: /* * Notify fast path that we want to know if any packet * is transmitted by setting r_skip_req. */ LLE_REQ_LOCK(lle); lle->r_skip_req = 1; LLE_REQ_UNLOCK(lle); nd_delay = V_nd6_delay; nd_gctimer = V_nd6_gctimer; delay = (long)(MIN(nd_gctimer, nd_delay)) * hz; remtime = (long)nd_gctimer * hz - delay; break; case ND6_LLINFO_DELAY: lle->la_asked = 0; delay = (long)V_nd6_delay * hz; break; } if (delay > 0) nd6_llinfo_settimer_locked(lle, delay); lle->lle_remtime = remtime; lle->ln_state = newstate; } /* * Timer-dependent part of nd state machine. * * Set noinline to be dtrace-friendly */ static __noinline void nd6_llinfo_timer(void *arg) { struct llentry *ln; struct in6_addr *dst, *pdst, *psrc, src; struct ifnet *ifp; struct nd_ifinfo *ndi; int do_switch, send_ns; long delay; KASSERT(arg != NULL, ("%s: arg NULL", __func__)); ln = (struct llentry *)arg; ifp = lltable_get_ifp(ln->lle_tbl); CURVNET_SET(ifp->if_vnet); ND6_RLOCK(); LLE_WLOCK(ln); if (callout_pending(&ln->lle_timer)) { /* * Here we are a bit odd here in the treatment of * active/pending. If the pending bit is set, it got * rescheduled before I ran. The active * bit we ignore, since if it was stopped * in ll_tablefree() and was currently running * it would have return 0 so the code would * not have deleted it since the callout could * not be stopped so we want to go through * with the delete here now. If the callout * was restarted, the pending bit will be back on and * we just want to bail since the callout_reset would * return 1 and our reference would have been removed * by nd6_llinfo_settimer_locked above since canceled * would have been 1. */ LLE_WUNLOCK(ln); ND6_RUNLOCK(); CURVNET_RESTORE(); return; } ndi = ND_IFINFO(ifp); send_ns = 0; dst = &ln->r_l3addr.addr6; pdst = dst; if (ln->ln_ntick > 0) { if (ln->ln_ntick > INT_MAX) { ln->ln_ntick -= INT_MAX; nd6_llinfo_settimer_locked(ln, INT_MAX); } else { ln->ln_ntick = 0; nd6_llinfo_settimer_locked(ln, ln->ln_ntick); } goto done; } if (ln->la_flags & LLE_STATIC) { goto done; } if (ln->la_flags & LLE_DELETED) { nd6_free(&ln, 0); goto done; } switch (ln->ln_state) { case ND6_LLINFO_INCOMPLETE: if (ln->la_asked < V_nd6_mmaxtries) { ln->la_asked++; send_ns = 1; /* Send NS to multicast address */ pdst = NULL; } else { struct mbuf *m = ln->la_hold; if (m) { struct mbuf *m0; /* * assuming every packet in la_hold has the * same IP header. Send error after unlock. */ m0 = m->m_nextpkt; m->m_nextpkt = NULL; ln->la_hold = m0; clear_llinfo_pqueue(ln); } nd6_free(&ln, 0); if (m != NULL) icmp6_error2(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 0, ifp); } break; case ND6_LLINFO_REACHABLE: if (!ND6_LLINFO_PERMANENT(ln)) nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); break; case ND6_LLINFO_STALE: if (nd6_is_stale(ln, &delay, &do_switch) != 0) { /* * No packet has used this entry and GC timeout * has not been passed. Reshedule timer and * return. */ nd6_llinfo_settimer_locked(ln, delay); break; } if (do_switch == 0) { /* * GC timer has ended and entry hasn't been used. * Run Garbage collector (RFC 4861, 5.3) */ if (!ND6_LLINFO_PERMANENT(ln)) nd6_free(&ln, 1); break; } /* Entry has been used AND delay timer has ended. */ /* FALLTHROUGH */ case ND6_LLINFO_DELAY: if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { /* We need NUD */ ln->la_asked = 1; nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE); send_ns = 1; } else nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */ break; case ND6_LLINFO_PROBE: if (ln->la_asked < V_nd6_umaxtries) { ln->la_asked++; send_ns = 1; } else { nd6_free(&ln, 0); } break; default: panic("%s: paths in a dark night can be confusing: %d", __func__, ln->ln_state); } done: if (ln != NULL) ND6_RUNLOCK(); if (send_ns != 0) { nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); psrc = nd6_llinfo_get_holdsrc(ln, &src); LLE_FREE_LOCKED(ln); ln = NULL; nd6_ns_output(ifp, psrc, pdst, dst, NULL); } if (ln != NULL) LLE_FREE_LOCKED(ln); CURVNET_RESTORE(); } /* * ND6 timer routine to expire default route list and prefix list */ void nd6_timer(void *arg) { CURVNET_SET((struct vnet *) arg); struct nd_drhead drq; struct nd_prhead prl; struct nd_defrouter *dr, *ndr; struct nd_prefix *pr, *npr; struct in6_ifaddr *ia6, *nia6; uint64_t genid; TAILQ_INIT(&drq); LIST_INIT(&prl); ND6_WLOCK(); TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) if (dr->expire && dr->expire < time_uptime) defrouter_unlink(dr, &drq); ND6_WUNLOCK(); while ((dr = TAILQ_FIRST(&drq)) != NULL) { TAILQ_REMOVE(&drq, dr, dr_entry); defrouter_del(dr); } /* * expire interface addresses. * in the past the loop was inside prefix expiry processing. * However, from a stricter speci-confrmance standpoint, we should * rather separate address lifetimes and prefix lifetimes. * * XXXRW: in6_ifaddrhead locking. */ addrloop: CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { /* check address lifetime */ if (IFA6_IS_INVALID(ia6)) { int regen = 0; /* * If the expiring address is temporary, try * regenerating a new one. This would be useful when * we suspended a laptop PC, then turned it on after a * period that could invalidate all temporary * addresses. Although we may have to restart the * loop (see below), it must be after purging the * address. Otherwise, we'd see an infinite loop of * regeneration. */ if (V_ip6_use_tempaddr && (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { if (regen_tmpaddr(ia6) == 0) regen = 1; } in6_purgeaddr(&ia6->ia_ifa); if (regen) goto addrloop; /* XXX: see below */ } else if (IFA6_IS_DEPRECATED(ia6)) { int oldflags = ia6->ia6_flags; ia6->ia6_flags |= IN6_IFF_DEPRECATED; /* * If a temporary address has just become deprecated, * regenerate a new one if possible. */ if (V_ip6_use_tempaddr && (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && (oldflags & IN6_IFF_DEPRECATED) == 0) { if (regen_tmpaddr(ia6) == 0) { /* * A new temporary address is * generated. * XXX: this means the address chain * has changed while we are still in * the loop. Although the change * would not cause disaster (because * it's not a deletion, but an * addition,) we'd rather restart the * loop just for safety. Or does this * significantly reduce performance?? */ goto addrloop; } } } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) { /* * Schedule DAD for a tentative address. This happens * if the interface was down or not running * when the address was configured. */ int delay; delay = arc4random() % (MAX_RTR_SOLICITATION_DELAY * hz); nd6_dad_start((struct ifaddr *)ia6, delay); } else { /* * Check status of the interface. If it is down, * mark the address as tentative for future DAD. */ if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 || (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (ND_IFINFO(ia6->ia_ifp)->flags & ND6_IFF_IFDISABLED) != 0) { ia6->ia6_flags &= ~IN6_IFF_DUPLICATED; ia6->ia6_flags |= IN6_IFF_TENTATIVE; } /* * A new RA might have made a deprecated address * preferred. */ ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; } } ND6_WLOCK(); restart: LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { /* * Expire prefixes. Since the pltime is only used for * autoconfigured addresses, pltime processing for prefixes is * not necessary. * * Only unlink after all derived addresses have expired. This * may not occur until two hours after the prefix has expired * per RFC 4862. If the prefix expires before its derived * addresses, mark it off-link. This will be done automatically * after unlinking if no address references remain. */ if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME || time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime) continue; if (pr->ndpr_addrcnt == 0) { nd6_prefix_unlink(pr, &prl); continue; } if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) { genid = V_nd6_list_genid; nd6_prefix_ref(pr); ND6_WUNLOCK(); ND6_ONLINK_LOCK(); (void)nd6_prefix_offlink(pr); ND6_ONLINK_UNLOCK(); ND6_WLOCK(); nd6_prefix_rele(pr); if (genid != V_nd6_list_genid) goto restart; } } ND6_WUNLOCK(); while ((pr = LIST_FIRST(&prl)) != NULL) { LIST_REMOVE(pr, ndpr_entry); nd6_prefix_del(pr); } callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, nd6_timer, curvnet); CURVNET_RESTORE(); } /* * ia6 - deprecated/invalidated temporary address */ static int regen_tmpaddr(struct in6_ifaddr *ia6) { struct ifaddr *ifa; struct ifnet *ifp; struct in6_ifaddr *public_ifa6 = NULL; ifp = ia6->ia_ifa.ifa_ifp; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct in6_ifaddr *it6; if (ifa->ifa_addr->sa_family != AF_INET6) continue; it6 = (struct in6_ifaddr *)ifa; /* ignore no autoconf addresses. */ if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) continue; /* ignore autoconf addresses with different prefixes. */ if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) continue; /* * Now we are looking at an autoconf address with the same * prefix as ours. If the address is temporary and is still * preferred, do not create another one. It would be rare, but * could happen, for example, when we resume a laptop PC after * a long period. */ if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && !IFA6_IS_DEPRECATED(it6)) { public_ifa6 = NULL; break; } /* * This is a public autoconf address that has the same prefix * as ours. If it is preferred, keep it. We can't break the * loop here, because there may be a still-preferred temporary * address with the prefix. */ if (!IFA6_IS_DEPRECATED(it6)) public_ifa6 = it6; } if (public_ifa6 != NULL) ifa_ref(&public_ifa6->ia_ifa); IF_ADDR_RUNLOCK(ifp); if (public_ifa6 != NULL) { int e; if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { ifa_free(&public_ifa6->ia_ifa); log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" " tmp addr,errno=%d\n", e); return (-1); } ifa_free(&public_ifa6->ia_ifa); return (0); } return (-1); } /* * Remove prefix and default router list entries corresponding to ifp. Neighbor * cache entries are freed in in6_domifdetach(). */ void nd6_purge(struct ifnet *ifp) { struct nd_drhead drq; struct nd_prhead prl; struct nd_defrouter *dr, *ndr; struct nd_prefix *pr, *npr; TAILQ_INIT(&drq); LIST_INIT(&prl); /* * Nuke default router list entries toward ifp. * We defer removal of default router list entries that is installed * in the routing table, in order to keep additional side effects as * small as possible. */ ND6_WLOCK(); TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { if (dr->installed) continue; if (dr->ifp == ifp) defrouter_unlink(dr, &drq); } TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { if (!dr->installed) continue; if (dr->ifp == ifp) defrouter_unlink(dr, &drq); } /* * Remove prefixes on ifp. We should have already removed addresses on * this interface, so no addresses should be referencing these prefixes. */ LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { if (pr->ndpr_ifp == ifp) nd6_prefix_unlink(pr, &prl); } ND6_WUNLOCK(); /* Delete the unlinked router and prefix objects. */ while ((dr = TAILQ_FIRST(&drq)) != NULL) { TAILQ_REMOVE(&drq, dr, dr_entry); defrouter_del(dr); } while ((pr = LIST_FIRST(&prl)) != NULL) { LIST_REMOVE(pr, ndpr_entry); nd6_prefix_del(pr); } /* cancel default outgoing interface setting */ if (V_nd6_defifindex == ifp->if_index) nd6_setdefaultiface(0); if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { /* Refresh default router list. */ defrouter_select_fib(ifp->if_fib); } } /* * the caller acquires and releases the lock on the lltbls * Returns the llentry locked */ struct llentry * nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp) { struct sockaddr_in6 sin6; struct llentry *ln; bzero(&sin6, sizeof(sin6)); sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_family = AF_INET6; sin6.sin6_addr = *addr6; IF_AFDATA_LOCK_ASSERT(ifp); ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6); return (ln); } struct llentry * nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp) { struct sockaddr_in6 sin6; struct llentry *ln; bzero(&sin6, sizeof(sin6)); sin6.sin6_len = sizeof(struct sockaddr_in6); sin6.sin6_family = AF_INET6; sin6.sin6_addr = *addr6; ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6); if (ln != NULL) ln->ln_state = ND6_LLINFO_NOSTATE; return (ln); } /* * Test whether a given IPv6 address is a neighbor or not, ignoring * the actual neighbor cache. The neighbor cache is ignored in order * to not reenter the routing code from within itself. */ static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) { struct nd_prefix *pr; struct ifaddr *ifa; struct rt_addrinfo info; struct sockaddr_in6 rt_key; const struct sockaddr *dst6; uint64_t genid; int error, fibnum; /* * A link-local address is always a neighbor. * XXX: a link does not necessarily specify a single interface. */ if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { struct sockaddr_in6 sin6_copy; u_int32_t zone; /* * We need sin6_copy since sa6_recoverscope() may modify the * content (XXX). */ sin6_copy = *addr; if (sa6_recoverscope(&sin6_copy)) return (0); /* XXX: should be impossible */ if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) return (0); if (sin6_copy.sin6_scope_id == zone) return (1); else return (0); } bzero(&rt_key, sizeof(rt_key)); bzero(&info, sizeof(info)); info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key; /* * If the address matches one of our addresses, * it should be a neighbor. * If the address matches one of our on-link prefixes, it should be a * neighbor. */ ND6_RLOCK(); restart: LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { if (pr->ndpr_ifp != ifp) continue; if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) { dst6 = (const struct sockaddr *)&pr->ndpr_prefix; /* * We only need to check all FIBs if add_addr_allfibs * is unset. If set, checking any FIB will suffice. */ fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0; for (; fibnum < rt_numfibs; fibnum++) { genid = V_nd6_list_genid; ND6_RUNLOCK(); /* * Restore length field before * retrying lookup */ rt_key.sin6_len = sizeof(rt_key); error = rib_lookup_info(fibnum, dst6, 0, 0, &info); ND6_RLOCK(); if (genid != V_nd6_list_genid) goto restart; if (error == 0) break; } if (error != 0) continue; /* * This is the case where multiple interfaces * have the same prefix, but only one is installed * into the routing table and that prefix entry * is not the one being examined here. In the case * where RADIX_MPATH is enabled, multiple route * entries (of the same rt_key value) will be * installed because the interface addresses all * differ. */ if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, &rt_key.sin6_addr)) continue; } if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, &addr->sin6_addr, &pr->ndpr_mask)) { ND6_RUNLOCK(); return (1); } } ND6_RUNLOCK(); /* * If the address is assigned on the node of the other side of * a p2p interface, the address should be a neighbor. */ if (ifp->if_flags & IFF_POINTOPOINT) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != addr->sin6_family) continue; if (ifa->ifa_dstaddr != NULL && sa_equal(addr, ifa->ifa_dstaddr)) { IF_ADDR_RUNLOCK(ifp); return 1; } } IF_ADDR_RUNLOCK(ifp); } /* * If the default router list is empty, all addresses are regarded * as on-link, and thus, as a neighbor. */ if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && TAILQ_EMPTY(&V_nd_defrouter) && V_nd6_defifindex == ifp->if_index) { return (1); } return (0); } /* * Detect if a given IPv6 address identifies a neighbor on a given link. * XXX: should take care of the destination of a p2p link? */ int nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) { struct llentry *lle; int rc = 0; IF_AFDATA_UNLOCK_ASSERT(ifp); if (nd6_is_new_addr_neighbor(addr, ifp)) return (1); /* * Even if the address matches none of our addresses, it might be * in the neighbor cache. */ IF_AFDATA_RLOCK(ifp); if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { LLE_RUNLOCK(lle); rc = 1; } IF_AFDATA_RUNLOCK(ifp); return (rc); } /* * Free an nd6 llinfo entry. * Since the function would cause significant changes in the kernel, DO NOT * make it global, unless you have a strong reason for the change, and are sure * that the change is safe. * * Set noinline to be dtrace-friendly */ static __noinline void nd6_free(struct llentry **lnp, int gc) { struct ifnet *ifp; struct llentry *ln; struct nd_defrouter *dr; ln = *lnp; *lnp = NULL; LLE_WLOCK_ASSERT(ln); ND6_RLOCK_ASSERT(); ifp = lltable_get_ifp(ln->lle_tbl); if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0) dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp); else dr = NULL; ND6_RUNLOCK(); if ((ln->la_flags & LLE_DELETED) == 0) EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); /* * we used to have pfctlinput(PRC_HOSTDEAD) here. * even though it is not harmful, it was not really necessary. */ /* cancel timer */ nd6_llinfo_settimer_locked(ln, -1); if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { if (dr != NULL && dr->expire && ln->ln_state == ND6_LLINFO_STALE && gc) { /* * If the reason for the deletion is just garbage * collection, and the neighbor is an active default * router, do not delete it. Instead, reset the GC * timer using the router's lifetime. * Simply deleting the entry would affect default * router selection, which is not necessarily a good * thing, especially when we're using router preference * values. * XXX: the check for ln_state would be redundant, * but we intentionally keep it just in case. */ if (dr->expire > time_uptime) nd6_llinfo_settimer_locked(ln, (dr->expire - time_uptime) * hz); else nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); LLE_REMREF(ln); LLE_WUNLOCK(ln); defrouter_rele(dr); return; } if (dr) { /* * Unreachablity of a router might affect the default * router selection and on-link detection of advertised * prefixes. */ /* * Temporarily fake the state to choose a new default * router and to perform on-link determination of * prefixes correctly. * Below the state will be set correctly, * or the entry itself will be deleted. */ ln->ln_state = ND6_LLINFO_INCOMPLETE; } if (ln->ln_router || dr) { /* * We need to unlock to avoid a LOR with rt6_flush() with the * rnh and for the calls to pfxlist_onlink_check() and * defrouter_select_fib() in the block further down for calls * into nd6_lookup(). We still hold a ref. */ LLE_WUNLOCK(ln); /* * rt6_flush must be called whether or not the neighbor * is in the Default Router List. * See a corresponding comment in nd6_na_input(). */ rt6_flush(&ln->r_l3addr.addr6, ifp); } if (dr) { /* * Since defrouter_select_fib() does not affect the * on-link determination and MIP6 needs the check * before the default router selection, we perform * the check now. */ pfxlist_onlink_check(); /* * Refresh default router list. */ defrouter_select_fib(dr->ifp->if_fib); } /* * If this entry was added by an on-link redirect, remove the * corresponding host route. */ if (ln->la_flags & LLE_REDIRECT) nd6_free_redirect(ln); if (ln->ln_router || dr) LLE_WLOCK(ln); } /* * Save to unlock. We still hold an extra reference and will not * free(9) in llentry_free() if someone else holds one as well. */ LLE_WUNLOCK(ln); IF_AFDATA_LOCK(ifp); LLE_WLOCK(ln); /* Guard against race with other llentry_free(). */ if (ln->la_flags & LLE_LINKED) { /* Remove callout reference */ LLE_REMREF(ln); lltable_unlink_entry(ln->lle_tbl, ln); } IF_AFDATA_UNLOCK(ifp); llentry_free(ln); if (dr != NULL) defrouter_rele(dr); } static int nd6_isdynrte(const struct rtentry *rt, void *xap) { if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC)) return (1); return (0); } /* * Remove the rtentry for the given llentry, * both of which were installed by a redirect. */ static void nd6_free_redirect(const struct llentry *ln) { int fibnum; struct sockaddr_in6 sin6; struct rt_addrinfo info; lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6); memset(&info, 0, sizeof(info)); info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6; info.rti_filter = nd6_isdynrte; for (fibnum = 0; fibnum < rt_numfibs; fibnum++) rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum); } /* * Rejuvenate this function for routing operations related * processing. */ void nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) { struct sockaddr_in6 *gateway; struct nd_defrouter *dr; struct ifnet *ifp; gateway = (struct sockaddr_in6 *)rt->rt_gateway; ifp = rt->rt_ifp; switch (req) { case RTM_ADD: break; case RTM_DELETE: if (!ifp) return; /* * Only indirect routes are interesting. */ if ((rt->rt_flags & RTF_GATEWAY) == 0) return; /* * check for default route */ if (IN6_ARE_ADDR_EQUAL(&in6addr_any, &SIN6(rt_key(rt))->sin6_addr)) { dr = defrouter_lookup(&gateway->sin6_addr, ifp); if (dr != NULL) { dr->installed = 0; defrouter_rele(dr); } } break; } } int nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) { struct in6_ndireq *ndi = (struct in6_ndireq *)data; struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; int error = 0; if (ifp->if_afdata[AF_INET6] == NULL) return (EPFNOSUPPORT); switch (cmd) { case OSIOCGIFINFO_IN6: #define ND ndi->ndi /* XXX: old ndp(8) assumes a positive value for linkmtu. */ bzero(&ND, sizeof(ND)); ND.linkmtu = IN6_LINKMTU(ifp); ND.maxmtu = ND_IFINFO(ifp)->maxmtu; ND.basereachable = ND_IFINFO(ifp)->basereachable; ND.reachable = ND_IFINFO(ifp)->reachable; ND.retrans = ND_IFINFO(ifp)->retrans; ND.flags = ND_IFINFO(ifp)->flags; ND.recalctm = ND_IFINFO(ifp)->recalctm; ND.chlim = ND_IFINFO(ifp)->chlim; break; case SIOCGIFINFO_IN6: ND = *ND_IFINFO(ifp); break; case SIOCSIFINFO_IN6: /* * used to change host variables from userland. * intended for a use on router to reflect RA configurations. */ /* 0 means 'unspecified' */ if (ND.linkmtu != 0) { if (ND.linkmtu < IPV6_MMTU || ND.linkmtu > IN6_LINKMTU(ifp)) { error = EINVAL; break; } ND_IFINFO(ifp)->linkmtu = ND.linkmtu; } if (ND.basereachable != 0) { int obasereachable = ND_IFINFO(ifp)->basereachable; ND_IFINFO(ifp)->basereachable = ND.basereachable; if (ND.basereachable != obasereachable) ND_IFINFO(ifp)->reachable = ND_COMPUTE_RTIME(ND.basereachable); } if (ND.retrans != 0) ND_IFINFO(ifp)->retrans = ND.retrans; if (ND.chlim != 0) ND_IFINFO(ifp)->chlim = ND.chlim; /* FALLTHROUGH */ case SIOCSIFINFO_FLAGS: { struct ifaddr *ifa; struct in6_ifaddr *ia; if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && !(ND.flags & ND6_IFF_IFDISABLED)) { /* ifdisabled 1->0 transision */ /* * If the interface is marked as ND6_IFF_IFDISABLED and * has an link-local address with IN6_IFF_DUPLICATED, * do not clear ND6_IFF_IFDISABLED. * See RFC 4862, Section 5.4.5. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; ia = (struct in6_ifaddr *)ifa; if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) break; } IF_ADDR_RUNLOCK(ifp); if (ifa != NULL) { /* LLA is duplicated. */ ND.flags |= ND6_IFF_IFDISABLED; log(LOG_ERR, "Cannot enable an interface" " with a link-local address marked" " duplicate.\n"); } else { ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; if (ifp->if_flags & IFF_UP) in6_if_up(ifp); } } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && (ND.flags & ND6_IFF_IFDISABLED)) { /* ifdisabled 0->1 transision */ /* Mark all IPv6 address as tentative. */ ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; if (V_ip6_dad_count > 0 && (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) { IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; ia = (struct in6_ifaddr *)ifa; ia->ia6_flags |= IN6_IFF_TENTATIVE; } IF_ADDR_RUNLOCK(ifp); } } if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { /* auto_linklocal 0->1 transision */ /* If no link-local address on ifp, configure */ ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; in6_ifattach(ifp, NULL); } else if (!(ND.flags & ND6_IFF_IFDISABLED) && ifp->if_flags & IFF_UP) { /* * When the IF already has * ND6_IFF_AUTO_LINKLOCAL, no link-local * address is assigned, and IFF_UP, try to * assign one. */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_INET6) continue; ia = (struct in6_ifaddr *)ifa; if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) break; } IF_ADDR_RUNLOCK(ifp); if (ifa != NULL) /* No LLA is configured. */ in6_ifattach(ifp, NULL); } } } ND_IFINFO(ifp)->flags = ND.flags; break; #undef ND case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ /* sync kernel routing table with the default router list */ defrouter_reset(); defrouter_select(); break; case SIOCSPFXFLUSH_IN6: { /* flush all the prefix advertised by routers */ struct in6_ifaddr *ia, *ia_next; struct nd_prefix *pr, *next; struct nd_prhead prl; LIST_INIT(&prl); ND6_WLOCK(); LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) continue; /* XXX */ nd6_prefix_unlink(pr, &prl); } ND6_WUNLOCK(); while ((pr = LIST_FIRST(&prl)) != NULL) { LIST_REMOVE(pr, ndpr_entry); /* XXXRW: in6_ifaddrhead locking. */ CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, ia_next) { if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) continue; if (ia->ia6_ndpr == pr) in6_purgeaddr(&ia->ia_ifa); } nd6_prefix_del(pr); } break; } case SIOCSRTRFLUSH_IN6: { /* flush all the default routers */ struct nd_drhead drq; struct nd_defrouter *dr; TAILQ_INIT(&drq); defrouter_reset(); ND6_WLOCK(); while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL) defrouter_unlink(dr, &drq); ND6_WUNLOCK(); while ((dr = TAILQ_FIRST(&drq)) != NULL) { TAILQ_REMOVE(&drq, dr, dr_entry); defrouter_del(dr); } defrouter_select(); break; } case SIOCGNBRINFO_IN6: { struct llentry *ln; struct in6_addr nb_addr = nbi->addr; /* make local for safety */ if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) return (error); IF_AFDATA_RLOCK(ifp); ln = nd6_lookup(&nb_addr, 0, ifp); IF_AFDATA_RUNLOCK(ifp); if (ln == NULL) { error = EINVAL; break; } nbi->state = ln->ln_state; nbi->asked = ln->la_asked; nbi->isrouter = ln->ln_router; if (ln->la_expire == 0) nbi->expire = 0; else nbi->expire = ln->la_expire + ln->lle_remtime / hz + (time_second - time_uptime); LLE_RUNLOCK(ln); break; } case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ ndif->ifindex = V_nd6_defifindex; break; case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ return (nd6_setdefaultiface(ndif->ifindex)); } return (error); } /* * Calculates new isRouter value based on provided parameters and * returns it. */ static int nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr, int ln_router) { /* * ICMP6 type dependent behavior. * * NS: clear IsRouter if new entry * RS: clear IsRouter * RA: set IsRouter if there's lladdr * redir: clear IsRouter if new entry * * RA case, (1): * The spec says that we must set IsRouter in the following cases: * - If lladdr exist, set IsRouter. This means (1-5). * - If it is old entry (!newentry), set IsRouter. This means (7). * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. * A quetion arises for (1) case. (1) case has no lladdr in the * neighbor cache, this is similar to (6). * This case is rare but we figured that we MUST NOT set IsRouter. * * is_new old_addr new_addr NS RS RA redir * D R * 0 n n (1) c ? s * 0 y n (2) c s s * 0 n y (3) c s s * 0 y y (4) c s s * 0 y y (5) c s s * 1 -- n (6) c c c s * 1 -- y (7) c c s c s * * (c=clear s=set) */ switch (type & 0xff) { case ND_NEIGHBOR_SOLICIT: /* * New entry must have is_router flag cleared. */ if (is_new) /* (6-7) */ ln_router = 0; break; case ND_REDIRECT: /* * If the icmp is a redirect to a better router, always set the * is_router flag. Otherwise, if the entry is newly created, * clear the flag. [RFC 2461, sec 8.3] */ if (code == ND_REDIRECT_ROUTER) ln_router = 1; else { if (is_new) /* (6-7) */ ln_router = 0; } break; case ND_ROUTER_SOLICIT: /* * is_router flag must always be cleared. */ ln_router = 0; break; case ND_ROUTER_ADVERT: /* * Mark an entry with lladdr as a router. */ if ((!is_new && (old_addr || new_addr)) || /* (2-5) */ (is_new && new_addr)) { /* (7) */ ln_router = 1; } break; } return (ln_router); } /* * Create neighbor cache entry and cache link-layer address, * on reception of inbound ND6 packets. (RS/RA/NS/redirect) * * type - ICMP6 type * code - type dependent information * */ void nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, int lladdrlen, int type, int code) { struct llentry *ln = NULL, *ln_tmp; int is_newentry; int do_update; int olladdr; int llchange; int flags; uint16_t router = 0; struct sockaddr_in6 sin6; struct mbuf *chain = NULL; u_char linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; int lladdr_off; IF_AFDATA_UNLOCK_ASSERT(ifp); KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); KASSERT(from != NULL, ("%s: from == NULL", __func__)); /* nothing must be updated for unspecified address */ if (IN6_IS_ADDR_UNSPECIFIED(from)) return; /* * Validation about ifp->if_addrlen and lladdrlen must be done in * the caller. * * XXX If the link does not have link-layer adderss, what should * we do? (ifp->if_addrlen == 0) * Spec says nothing in sections for RA, RS and NA. There's small * description on it in NS section (RFC 2461 7.2.3). */ flags = lladdr ? LLE_EXCLUSIVE : 0; IF_AFDATA_RLOCK(ifp); ln = nd6_lookup(from, flags, ifp); IF_AFDATA_RUNLOCK(ifp); is_newentry = 0; if (ln == NULL) { flags |= LLE_EXCLUSIVE; ln = nd6_alloc(from, 0, ifp); if (ln == NULL) return; /* * Since we already know all the data for the new entry, * fill it before insertion. */ if (lladdr != NULL) { linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET6, lladdr, linkhdr, &linkhdrsize, &lladdr_off) != 0) return; lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize, lladdr_off); } IF_AFDATA_WLOCK(ifp); LLE_WLOCK(ln); /* Prefer any existing lle over newly-created one */ ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp); if (ln_tmp == NULL) lltable_link_entry(LLTABLE6(ifp), ln); IF_AFDATA_WUNLOCK(ifp); if (ln_tmp == NULL) { /* No existing lle, mark as new entry (6,7) */ is_newentry = 1; if (lladdr != NULL) { /* (7) */ nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); } } else { lltable_free_entry(LLTABLE6(ifp), ln); ln = ln_tmp; ln_tmp = NULL; } } /* do nothing if static ndp is set */ if ((ln->la_flags & LLE_STATIC)) { if (flags & LLE_EXCLUSIVE) LLE_WUNLOCK(ln); else LLE_RUNLOCK(ln); return; } olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; if (olladdr && lladdr) { llchange = bcmp(lladdr, ln->ll_addr, ifp->if_addrlen); } else if (!olladdr && lladdr) llchange = 1; else llchange = 0; /* * newentry olladdr lladdr llchange (*=record) * 0 n n -- (1) * 0 y n -- (2) * 0 n y y (3) * STALE * 0 y y n (4) * * 0 y y y (5) * STALE * 1 -- n -- (6) NOSTATE(= PASSIVE) * 1 -- y -- (7) * STALE */ do_update = 0; if (is_newentry == 0 && llchange != 0) { do_update = 1; /* (3,5) */ /* * Record source link-layer address * XXX is it dependent to ifp->if_type? */ linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET6, lladdr, linkhdr, &linkhdrsize, &lladdr_off) != 0) return; if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize, lladdr_off) == 0) { /* Entry was deleted */ return; } nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); if (ln->la_hold != NULL) nd6_grab_holdchain(ln, &chain, &sin6); } /* Calculates new router status */ router = nd6_is_router(type, code, is_newentry, olladdr, lladdr != NULL ? 1 : 0, ln->ln_router); ln->ln_router = router; /* Mark non-router redirects with special flag */ if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER) ln->la_flags |= LLE_REDIRECT; if (flags & LLE_EXCLUSIVE) LLE_WUNLOCK(ln); else LLE_RUNLOCK(ln); if (chain != NULL) nd6_flush_holdchain(ifp, chain, &sin6); /* * When the link-layer address of a router changes, select the * best router again. In particular, when the neighbor entry is newly * created, it might affect the selection policy. * Question: can we restrict the first condition to the "is_newentry" * case? * XXX: when we hear an RA from a new router with the link-layer * address option, defrouter_select_fib() is called twice, since * defrtrlist_update called the function as well. However, I believe * we can compromise the overhead, since it only happens the first * time. * XXX: although defrouter_select_fib() should not have a bad effect * for those are not autoconfigured hosts, we explicitly avoid such * cases for safety. */ if ((do_update || is_newentry) && router && ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { /* * guaranteed recursion */ defrouter_select_fib(ifp->if_fib); } } static void nd6_slowtimo(void *arg) { CURVNET_SET((struct vnet *) arg); struct nd_ifinfo *nd6if; struct ifnet *ifp; callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, nd6_slowtimo, curvnet); IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) { if (ifp->if_afdata[AF_INET6] == NULL) continue; nd6if = ND_IFINFO(ifp); if (nd6if->basereachable && /* already initialized */ (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { /* * Since reachable time rarely changes by router * advertisements, we SHOULD insure that a new random * value gets recomputed at least once every few hours. * (RFC 2461, 6.3.4) */ nd6if->recalctm = V_nd6_recalc_reachtm_interval; nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); } } IFNET_RUNLOCK_NOSLEEP(); CURVNET_RESTORE(); } void nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain, struct sockaddr_in6 *sin6) { LLE_WLOCK_ASSERT(ln); *chain = ln->la_hold; ln->la_hold = NULL; lltable_fill_sa_entry(ln, (struct sockaddr *)sin6); if (ln->ln_state == ND6_LLINFO_STALE) { /* * The first time we send a packet to a * neighbor whose entry is STALE, we have * to change the state to DELAY and a sets * a timer to expire in DELAY_FIRST_PROBE_TIME * seconds to ensure do neighbor unreachability * detection on expiration. * (RFC 2461 7.3.3) */ nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY); } } int nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, struct sockaddr_in6 *dst, struct route *ro) { int error; int ip6len; struct ip6_hdr *ip6; struct m_tag *mtag; #ifdef MAC mac_netinet6_nd6_send(ifp, m); #endif /* * If called from nd6_ns_output() (NS), nd6_na_output() (NA), * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA * as handled by rtsol and rtadvd), mbufs will be tagged for SeND * to be diverted to user space. When re-injected into the kernel, * send_output() will directly dispatch them to the outgoing interface. */ if (send_sendso_input_hook != NULL) { mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); if (mtag != NULL) { ip6 = mtod(m, struct ip6_hdr *); ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); /* Use the SEND socket */ error = send_sendso_input_hook(m, ifp, SND_OUT, ip6len); /* -1 == no app on SEND socket */ if (error == 0 || error != -1) return (error); } } m_clrprotoflags(m); /* Avoid confusing lower layers. */ IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, mtod(m, struct ip6_hdr *)); if ((ifp->if_flags & IFF_LOOPBACK) == 0) origifp = ifp; error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro); return (error); } /* * Lookup link headerfor @sa_dst address. Stores found * data in @desten buffer. Copy of lle ln_flags can be also * saved in @pflags if @pflags is non-NULL. * * If destination LLE does not exists or lle state modification * is required, call "slow" version. * * Return values: * - 0 on success (address copied to buffer). * - EWOULDBLOCK (no local error, but address is still unresolved) * - other errors (alloc failure, etc) */ int nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m, const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags, struct llentry **plle) { struct llentry *ln = NULL; const struct sockaddr_in6 *dst6; if (pflags != NULL) *pflags = 0; dst6 = (const struct sockaddr_in6 *)sa_dst; /* discard the packet if IPv6 operation is disabled on the interface */ if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { m_freem(m); return (ENETDOWN); /* better error? */ } if (m != NULL && m->m_flags & M_MCAST) { switch (ifp->if_type) { case IFT_ETHER: case IFT_L2VLAN: case IFT_BRIDGE: ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr, desten); return (0); default: m_freem(m); return (EAFNOSUPPORT); } } IF_AFDATA_RLOCK(ifp); ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED, ifp); if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) { /* Entry found, let's copy lle info */ bcopy(ln->r_linkdata, desten, ln->r_hdrlen); if (pflags != NULL) *pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR); /* Check if we have feedback request from nd6 timer */ if (ln->r_skip_req != 0) { LLE_REQ_LOCK(ln); ln->r_skip_req = 0; /* Notify that entry was used */ ln->lle_hittime = time_uptime; LLE_REQ_UNLOCK(ln); } if (plle) { LLE_ADDREF(ln); *plle = ln; LLE_WUNLOCK(ln); } IF_AFDATA_RUNLOCK(ifp); return (0); } else if (plle && ln) LLE_WUNLOCK(ln); IF_AFDATA_RUNLOCK(ifp); return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle)); } /* * Do L2 address resolution for @sa_dst address. Stores found * address in @desten buffer. Copy of lle ln_flags can be also * saved in @pflags if @pflags is non-NULL. * * Heavy version. * Function assume that destination LLE does not exist, * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired. * * Set noinline to be dtrace-friendly */ static __noinline int nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m, const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags, struct llentry **plle) { struct llentry *lle = NULL, *lle_tmp; struct in6_addr *psrc, src; int send_ns, ll_len; char *lladdr; /* * Address resolution or Neighbor Unreachability Detection * for the next hop. * At this point, the destination of the packet must be a unicast * or an anycast address(i.e. not a multicast). */ if (lle == NULL) { IF_AFDATA_RLOCK(ifp); lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); IF_AFDATA_RUNLOCK(ifp); if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) { /* * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), * the condition below is not very efficient. But we believe * it is tolerable, because this should be a rare case. */ lle = nd6_alloc(&dst->sin6_addr, 0, ifp); if (lle == NULL) { char ip6buf[INET6_ADDRSTRLEN]; log(LOG_DEBUG, "nd6_output: can't allocate llinfo for %s " "(ln=%p)\n", ip6_sprintf(ip6buf, &dst->sin6_addr), lle); m_freem(m); return (ENOBUFS); } IF_AFDATA_WLOCK(ifp); LLE_WLOCK(lle); /* Prefer any existing entry over newly-created one */ lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); if (lle_tmp == NULL) lltable_link_entry(LLTABLE6(ifp), lle); IF_AFDATA_WUNLOCK(ifp); if (lle_tmp != NULL) { lltable_free_entry(LLTABLE6(ifp), lle); lle = lle_tmp; lle_tmp = NULL; } } } if (lle == NULL) { if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { m_freem(m); return (ENOBUFS); } if (m != NULL) m_freem(m); return (ENOBUFS); } LLE_WLOCK_ASSERT(lle); /* * The first time we send a packet to a neighbor whose entry is * STALE, we have to change the state to DELAY and a sets a timer to * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do * neighbor unreachability detection on expiration. * (RFC 2461 7.3.3) */ if (lle->ln_state == ND6_LLINFO_STALE) nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY); /* * If the neighbor cache entry has a state other than INCOMPLETE * (i.e. its link-layer address is already resolved), just * send the packet. */ if (lle->ln_state > ND6_LLINFO_INCOMPLETE) { if (flags & LLE_ADDRONLY) { lladdr = lle->ll_addr; ll_len = ifp->if_addrlen; } else { lladdr = lle->r_linkdata; ll_len = lle->r_hdrlen; } bcopy(lladdr, desten, ll_len); if (pflags != NULL) *pflags = lle->la_flags; if (plle) { LLE_ADDREF(lle); *plle = lle; } LLE_WUNLOCK(lle); return (0); } /* * There is a neighbor cache entry, but no ethernet address * response yet. Append this latest packet to the end of the * packet queue in the mbuf. When it exceeds nd6_maxqueuelen, * the oldest packet in the queue will be removed. */ if (lle->la_hold != NULL) { struct mbuf *m_hold; int i; i = 0; for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){ i++; if (m_hold->m_nextpkt == NULL) { m_hold->m_nextpkt = m; break; } } while (i >= V_nd6_maxqueuelen) { m_hold = lle->la_hold; lle->la_hold = lle->la_hold->m_nextpkt; m_freem(m_hold); i--; } } else { lle->la_hold = m; } /* * If there has been no NS for the neighbor after entering the * INCOMPLETE state, send the first solicitation. * Note that for newly-created lle la_asked will be 0, * so we will transition from ND6_LLINFO_NOSTATE to * ND6_LLINFO_INCOMPLETE state here. */ psrc = NULL; send_ns = 0; if (lle->la_asked == 0) { lle->la_asked++; send_ns = 1; psrc = nd6_llinfo_get_holdsrc(lle, &src); nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE); } LLE_WUNLOCK(lle); if (send_ns != 0) nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL); return (EWOULDBLOCK); } /* * Do L2 address resolution for @sa_dst address. Stores found * address in @desten buffer. Copy of lle ln_flags can be also * saved in @pflags if @pflags is non-NULL. * * Return values: * - 0 on success (address copied to buffer). * - EWOULDBLOCK (no local error, but address is still unresolved) * - other errors (alloc failure, etc) */ int nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst, char *desten, uint32_t *pflags) { int error; flags |= LLE_ADDRONLY; error = nd6_resolve_slow(ifp, flags, NULL, (const struct sockaddr_in6 *)dst, desten, pflags, NULL); return (error); } int nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain, struct sockaddr_in6 *dst) { struct mbuf *m, *m_head; int error = 0; m_head = chain; while (m_head) { m = m_head; m_head = m_head->m_nextpkt; error = nd6_output_ifp(ifp, ifp, m, dst, NULL); } /* * XXX * note that intermediate errors are blindly ignored */ return (error); } static int nd6_need_cache(struct ifnet *ifp) { /* * XXX: we currently do not make neighbor cache on any interface * other than Ethernet and GIF. * * RFC2893 says: * - unidirectional tunnels needs no ND */ switch (ifp->if_type) { case IFT_ETHER: case IFT_IEEE1394: case IFT_L2VLAN: case IFT_INFINIBAND: case IFT_BRIDGE: case IFT_PROPVIRTUAL: return (1); default: return (0); } } /* * Add pernament ND6 link-layer record for given * interface address. * * Very similar to IPv4 arp_ifinit(), but: * 1) IPv6 DAD is performed in different place * 2) It is called by IPv6 protocol stack in contrast to * arp_ifinit() which is typically called in SIOCSIFADDR * driver ioctl handler. * */ int nd6_add_ifa_lle(struct in6_ifaddr *ia) { struct ifnet *ifp; struct llentry *ln, *ln_tmp; struct sockaddr *dst; ifp = ia->ia_ifa.ifa_ifp; if (nd6_need_cache(ifp) == 0) return (0); ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; dst = (struct sockaddr *)&ia->ia_addr; ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst); if (ln == NULL) return (ENOBUFS); IF_AFDATA_WLOCK(ifp); LLE_WLOCK(ln); /* Unlink any entry if exists */ ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst); if (ln_tmp != NULL) lltable_unlink_entry(LLTABLE6(ifp), ln_tmp); lltable_link_entry(LLTABLE6(ifp), ln); IF_AFDATA_WUNLOCK(ifp); if (ln_tmp != NULL) EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED); EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); LLE_WUNLOCK(ln); if (ln_tmp != NULL) llentry_free(ln_tmp); return (0); } /* * Removes either all lle entries for given @ia, or lle * corresponding to @ia address. */ void nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all) { struct sockaddr_in6 mask, addr; struct sockaddr *saddr, *smask; struct ifnet *ifp; ifp = ia->ia_ifa.ifa_ifp; memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr)); memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask)); saddr = (struct sockaddr *)&addr; smask = (struct sockaddr *)&mask; if (all != 0) lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC); else lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr); } static void clear_llinfo_pqueue(struct llentry *ln) { struct mbuf *m_hold, *m_hold_next; for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { m_hold_next = m_hold->m_nextpkt; m_freem(m_hold); } ln->la_hold = NULL; } static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); SYSCTL_DECL(_net_inet6_icmp6); SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter", "NDP default router list"); SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, nd6_sysctl_prlist, "S,in6_prefix", "NDP prefix list"); SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) { struct in6_defrouter d; struct nd_defrouter *dr; int error; if (req->newptr != NULL) return (EPERM); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); bzero(&d, sizeof(d)); d.rtaddr.sin6_family = AF_INET6; d.rtaddr.sin6_len = sizeof(d.rtaddr); ND6_RLOCK(); TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { d.rtaddr.sin6_addr = dr->rtaddr; error = sa6_recoverscope(&d.rtaddr); if (error != 0) break; d.flags = dr->raflags; d.rtlifetime = dr->rtlifetime; d.expire = dr->expire + (time_second - time_uptime); d.if_index = dr->ifp->if_index; error = SYSCTL_OUT(req, &d, sizeof(d)); if (error != 0) break; } ND6_RUNLOCK(); return (error); } static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) { struct in6_prefix p; struct sockaddr_in6 s6; struct nd_prefix *pr; struct nd_pfxrouter *pfr; time_t maxexpire; int error; char ip6buf[INET6_ADDRSTRLEN]; if (req->newptr) return (EPERM); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); bzero(&p, sizeof(p)); p.origin = PR_ORIG_RA; bzero(&s6, sizeof(s6)); s6.sin6_family = AF_INET6; s6.sin6_len = sizeof(s6); ND6_RLOCK(); LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { p.prefix = pr->ndpr_prefix; if (sa6_recoverscope(&p.prefix)) { log(LOG_ERR, "scope error in prefix list (%s)\n", ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); /* XXX: press on... */ } p.raflags = pr->ndpr_raf; p.prefixlen = pr->ndpr_plen; p.vltime = pr->ndpr_vltime; p.pltime = pr->ndpr_pltime; p.if_index = pr->ndpr_ifp->if_index; if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) p.expire = 0; else { /* XXX: we assume time_t is signed. */ maxexpire = (-1) & ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) p.expire = pr->ndpr_lastupdate + pr->ndpr_vltime + (time_second - time_uptime); else p.expire = maxexpire; } p.refcnt = pr->ndpr_addrcnt; p.flags = pr->ndpr_stateflags; p.advrtrs = 0; LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) p.advrtrs++; error = SYSCTL_OUT(req, &p, sizeof(p)); if (error != 0) break; LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { s6.sin6_addr = pfr->router->rtaddr; if (sa6_recoverscope(&s6)) log(LOG_ERR, "scope error in prefix list (%s)\n", ip6_sprintf(ip6buf, &pfr->router->rtaddr)); error = SYSCTL_OUT(req, &s6, sizeof(s6)); if (error != 0) goto out; } } out: ND6_RUNLOCK(); return (error); } Index: head/sys/netinet6/nd6_nbr.c =================================================================== --- head/sys/netinet6/nd6_nbr.c (revision 336675) +++ head/sys/netinet6/nd6_nbr.c (revision 336676) @@ -1,1547 +1,1547 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: nd6_nbr.c,v 1.86 2002/01/21 02:33:04 jinmei Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_mpath.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RADIX_MPATH #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #define SDL(s) ((struct sockaddr_dl *)s) struct dadq; static struct dadq *nd6_dad_find(struct ifaddr *, struct nd_opt_nonce *); static void nd6_dad_add(struct dadq *dp); static void nd6_dad_del(struct dadq *dp); static void nd6_dad_rele(struct dadq *); static void nd6_dad_starttimer(struct dadq *, int, int); static void nd6_dad_stoptimer(struct dadq *); static void nd6_dad_timer(struct dadq *); static void nd6_dad_duplicated(struct ifaddr *, struct dadq *); static void nd6_dad_ns_output(struct dadq *); static void nd6_dad_ns_input(struct ifaddr *, struct nd_opt_nonce *); static void nd6_dad_na_input(struct ifaddr *); static void nd6_na_output_fib(struct ifnet *, const struct in6_addr *, const struct in6_addr *, u_long, int, struct sockaddr *, u_int); static void nd6_ns_output_fib(struct ifnet *, const struct in6_addr *, const struct in6_addr *, const struct in6_addr *, uint8_t *, u_int); -static VNET_DEFINE(int, dad_enhanced) = 1; +VNET_DEFINE_STATIC(int, dad_enhanced) = 1; #define V_dad_enhanced VNET(dad_enhanced) SYSCTL_DECL(_net_inet6_ip6); SYSCTL_INT(_net_inet6_ip6, OID_AUTO, dad_enhanced, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dad_enhanced), 0, "Enable Enhanced DAD, which adds a random nonce to NS messages for DAD."); -static VNET_DEFINE(int, dad_maxtry) = 15; /* max # of *tries* to +VNET_DEFINE_STATIC(int, dad_maxtry) = 15; /* max # of *tries* to transmit DAD packet */ #define V_dad_maxtry VNET(dad_maxtry) /* * Input a Neighbor Solicitation Message. * * Based on RFC 2461 * Based on RFC 2462 (duplicate address detection) */ void nd6_ns_input(struct mbuf *m, int off, int icmp6len) { struct ifnet *ifp = m->m_pkthdr.rcvif; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct nd_neighbor_solicit *nd_ns; struct in6_addr saddr6 = ip6->ip6_src; struct in6_addr daddr6 = ip6->ip6_dst; struct in6_addr taddr6; struct in6_addr myaddr6; char *lladdr = NULL; struct ifaddr *ifa = NULL; int lladdrlen = 0; int anycast = 0, proxy = 0, tentative = 0; int tlladdr; int rflag; union nd_opts ndopts; struct sockaddr_dl proxydl; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; /* RFC 6980: Nodes MUST silently ignore fragments */ if(m->m_flags & M_FRAGMENTED) goto freeit; rflag = (V_ip6_forwarding) ? ND_NA_FLAG_ROUTER : 0; if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && V_ip6_norbit_raif) rflag = 0; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, icmp6len,); nd_ns = (struct nd_neighbor_solicit *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(nd_ns, struct nd_neighbor_solicit *, m, off, icmp6len); if (nd_ns == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif ip6 = mtod(m, struct ip6_hdr *); /* adjust pointer for safety */ taddr6 = nd_ns->nd_ns_target; if (in6_setscope(&taddr6, ifp, NULL) != 0) goto bad; if (ip6->ip6_hlim != 255) { nd6log((LOG_ERR, "nd6_ns_input: invalid hlim (%d) from %s to %s on %s\n", ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp))); goto bad; } if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) { /* dst has to be a solicited node multicast address. */ if (daddr6.s6_addr16[0] == IPV6_ADDR_INT16_MLL && /* don't check ifindex portion */ daddr6.s6_addr32[1] == 0 && daddr6.s6_addr32[2] == IPV6_ADDR_INT32_ONE && daddr6.s6_addr8[12] == 0xff) { ; /* good */ } else { nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet " "(wrong ip6 dst)\n")); goto bad; } } else if (!V_nd6_onlink_ns_rfc4861) { struct sockaddr_in6 src_sa6; /* * According to recent IETF discussions, it is not a good idea * to accept a NS from an address which would not be deemed * to be a neighbor otherwise. This point is expected to be * clarified in future revisions of the specification. */ bzero(&src_sa6, sizeof(src_sa6)); src_sa6.sin6_family = AF_INET6; src_sa6.sin6_len = sizeof(src_sa6); src_sa6.sin6_addr = saddr6; if (nd6_is_addr_neighbor(&src_sa6, ifp) == 0) { nd6log((LOG_INFO, "nd6_ns_input: " "NS packet from non-neighbor\n")); goto bad; } } if (IN6_IS_ADDR_MULTICAST(&taddr6)) { nd6log((LOG_INFO, "nd6_ns_input: bad NS target (multicast)\n")); goto bad; } icmp6len -= sizeof(*nd_ns); nd6_option_init(nd_ns + 1, icmp6len, &ndopts); if (nd6_options(&ndopts) < 0) { nd6log((LOG_INFO, "nd6_ns_input: invalid ND option, ignored\n")); /* nd6_options have incremented stats */ goto freeit; } if (ndopts.nd_opts_src_lladdr) { lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1); lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3; } if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src) && lladdr) { nd6log((LOG_INFO, "nd6_ns_input: bad DAD packet " "(link-layer address option)\n")); goto bad; } /* * Attaching target link-layer address to the NA? * (RFC 2461 7.2.4) * * NS IP dst is unicast/anycast MUST NOT add * NS IP dst is solicited-node multicast MUST add * * In implementation, we add target link-layer address by default. * We do not add one in MUST NOT cases. */ if (!IN6_IS_ADDR_MULTICAST(&daddr6)) tlladdr = 0; else tlladdr = 1; /* * Target address (taddr6) must be either: * (1) Valid unicast/anycast address for my receiving interface, * (2) Unicast address for which I'm offering proxy service, or * (3) "tentative" address on which DAD is being performed. */ /* (1) and (3) check. */ if (ifp->if_carp) ifa = (*carp_iamatch6_p)(ifp, &taddr6); else ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6); /* (2) check. */ if (ifa == NULL) { struct sockaddr_dl rt_gateway; struct rt_addrinfo info; struct sockaddr_in6 dst6; bzero(&dst6, sizeof(dst6)); dst6.sin6_len = sizeof(struct sockaddr_in6); dst6.sin6_family = AF_INET6; dst6.sin6_addr = taddr6; bzero(&rt_gateway, sizeof(rt_gateway)); rt_gateway.sdl_len = sizeof(rt_gateway); bzero(&info, sizeof(info)); info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&rt_gateway; if (rib_lookup_info(ifp->if_fib, (struct sockaddr *)&dst6, 0, 0, &info) == 0) { if ((info.rti_flags & RTF_ANNOUNCE) != 0 && rt_gateway.sdl_family == AF_LINK) { /* * proxy NDP for single entry */ proxydl = *SDL(&rt_gateway); ifa = (struct ifaddr *)in6ifa_ifpforlinklocal( ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); if (ifa) proxy = 1; } } } if (ifa == NULL) { /* * We've got an NS packet, and we don't have that adddress * assigned for us. We MUST silently ignore it. * See RFC2461 7.2.3. */ goto freeit; } myaddr6 = *IFA_IN6(ifa); anycast = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_ANYCAST; tentative = ((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE; if (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_DUPLICATED) goto freeit; if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) { nd6log((LOG_INFO, "nd6_ns_input: lladdrlen mismatch for %s " "(if %d, NS packet %d)\n", ip6_sprintf(ip6bufs, &taddr6), ifp->if_addrlen, lladdrlen - 2)); goto bad; } if (IN6_ARE_ADDR_EQUAL(&myaddr6, &saddr6)) { nd6log((LOG_INFO, "nd6_ns_input: duplicate IP6 address %s\n", ip6_sprintf(ip6bufs, &saddr6))); goto freeit; } /* * We have neighbor solicitation packet, with target address equals to * one of my tentative address. * * src addr how to process? * --- --- * multicast of course, invalid (rejected in ip6_input) * unicast somebody is doing address resolution -> ignore * unspec dup address detection * * The processing is defined in RFC 2462. */ if (tentative) { /* * If source address is unspecified address, it is for * duplicate address detection. * * If not, the packet is for addess resolution; * silently ignore it. */ if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) nd6_dad_ns_input(ifa, ndopts.nd_opts_nonce); goto freeit; } /* * If the source address is unspecified address, entries must not * be created or updated. * It looks that sender is performing DAD. Output NA toward * all-node multicast address, to tell the sender that I'm using * the address. * S bit ("solicited") must be zero. */ if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) { struct in6_addr in6_all; in6_all = in6addr_linklocal_allnodes; if (in6_setscope(&in6_all, ifp, NULL) != 0) goto bad; nd6_na_output_fib(ifp, &in6_all, &taddr6, ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) | rflag, tlladdr, proxy ? (struct sockaddr *)&proxydl : NULL, M_GETFIB(m)); goto freeit; } nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen, ND_NEIGHBOR_SOLICIT, 0); nd6_na_output_fib(ifp, &saddr6, &taddr6, ((anycast || proxy || !tlladdr) ? 0 : ND_NA_FLAG_OVERRIDE) | rflag | ND_NA_FLAG_SOLICITED, tlladdr, proxy ? (struct sockaddr *)&proxydl : NULL, M_GETFIB(m)); freeit: if (ifa != NULL) ifa_free(ifa); m_freem(m); return; bad: nd6log((LOG_ERR, "nd6_ns_input: src=%s\n", ip6_sprintf(ip6bufs, &saddr6))); nd6log((LOG_ERR, "nd6_ns_input: dst=%s\n", ip6_sprintf(ip6bufs, &daddr6))); nd6log((LOG_ERR, "nd6_ns_input: tgt=%s\n", ip6_sprintf(ip6bufs, &taddr6))); ICMP6STAT_INC(icp6s_badns); if (ifa != NULL) ifa_free(ifa); m_freem(m); } /* * Output a Neighbor Solicitation Message. Caller specifies: * - ICMP6 header source IP6 address * - ND6 header target IP6 address * - ND6 header source datalink address * * Based on RFC 2461 * Based on RFC 2462 (duplicate address detection) * * ln - for source address determination * nonce - If non-NULL, NS is used for duplicate address detection and * the value (length is ND_OPT_NONCE_LEN) is used as a random nonce. */ static void nd6_ns_output_fib(struct ifnet *ifp, const struct in6_addr *saddr6, const struct in6_addr *daddr6, const struct in6_addr *taddr6, uint8_t *nonce, u_int fibnum) { struct mbuf *m; struct m_tag *mtag; struct ip6_hdr *ip6; struct nd_neighbor_solicit *nd_ns; struct ip6_moptions im6o; int icmp6len; int maxlen; caddr_t mac; if (IN6_IS_ADDR_MULTICAST(taddr6)) return; /* estimate the size of message */ maxlen = sizeof(*ip6) + sizeof(*nd_ns); maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7; KASSERT(max_linkhdr + maxlen <= MCLBYTES, ( "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)", __func__, max_linkhdr, maxlen, MCLBYTES)); if (max_linkhdr + maxlen > MHLEN) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; M_SETFIB(m, fibnum); if (daddr6 == NULL || IN6_IS_ADDR_MULTICAST(daddr6)) { m->m_flags |= M_MCAST; im6o.im6o_multicast_ifp = ifp; im6o.im6o_multicast_hlim = 255; im6o.im6o_multicast_loop = 0; } icmp6len = sizeof(*nd_ns); m->m_pkthdr.len = m->m_len = sizeof(*ip6) + icmp6len; m->m_data += max_linkhdr; /* or M_ALIGN() equivalent? */ /* fill neighbor solicitation packet */ ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; /* ip6->ip6_plen will be set later */ ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_hlim = 255; if (daddr6) ip6->ip6_dst = *daddr6; else { ip6->ip6_dst.s6_addr16[0] = IPV6_ADDR_INT16_MLL; ip6->ip6_dst.s6_addr16[1] = 0; ip6->ip6_dst.s6_addr32[1] = 0; ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_ONE; ip6->ip6_dst.s6_addr32[3] = taddr6->s6_addr32[3]; ip6->ip6_dst.s6_addr8[12] = 0xff; if (in6_setscope(&ip6->ip6_dst, ifp, NULL) != 0) goto bad; } if (nonce == NULL) { struct ifaddr *ifa = NULL; /* * RFC2461 7.2.2: * "If the source address of the packet prompting the * solicitation is the same as one of the addresses assigned * to the outgoing interface, that address SHOULD be placed * in the IP Source Address of the outgoing solicitation. * Otherwise, any one of the addresses assigned to the * interface should be used." * * We use the source address for the prompting packet * (saddr6), if saddr6 belongs to the outgoing interface. * Otherwise, we perform the source address selection as usual. */ if (saddr6 != NULL) ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, saddr6); if (ifa != NULL) { /* ip6_src set already. */ ip6->ip6_src = *saddr6; ifa_free(ifa); } else { int error; struct in6_addr dst6, src6; uint32_t scopeid; in6_splitscope(&ip6->ip6_dst, &dst6, &scopeid); error = in6_selectsrc_addr(fibnum, &dst6, scopeid, ifp, &src6, NULL); if (error) { char ip6buf[INET6_ADDRSTRLEN]; nd6log((LOG_DEBUG, "%s: source can't be " "determined: dst=%s, error=%d\n", __func__, ip6_sprintf(ip6buf, &dst6), error)); goto bad; } ip6->ip6_src = src6; } } else { /* * Source address for DAD packet must always be IPv6 * unspecified address. (0::0) * We actually don't have to 0-clear the address (we did it * above), but we do so here explicitly to make the intention * clearer. */ bzero(&ip6->ip6_src, sizeof(ip6->ip6_src)); } nd_ns = (struct nd_neighbor_solicit *)(ip6 + 1); nd_ns->nd_ns_type = ND_NEIGHBOR_SOLICIT; nd_ns->nd_ns_code = 0; nd_ns->nd_ns_reserved = 0; nd_ns->nd_ns_target = *taddr6; in6_clearscope(&nd_ns->nd_ns_target); /* XXX */ /* * Add source link-layer address option. * * spec implementation * --- --- * DAD packet MUST NOT do not add the option * there's no link layer address: * impossible do not add the option * there's link layer address: * Multicast NS MUST add one add the option * Unicast NS SHOULD add one add the option */ if (nonce == NULL && (mac = nd6_ifptomac(ifp))) { int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen; struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1); /* 8 byte alignments... */ optlen = (optlen + 7) & ~7; m->m_pkthdr.len += optlen; m->m_len += optlen; icmp6len += optlen; bzero((caddr_t)nd_opt, optlen); nd_opt->nd_opt_type = ND_OPT_SOURCE_LINKADDR; nd_opt->nd_opt_len = optlen >> 3; bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen); } /* * Add a Nonce option (RFC 3971) to detect looped back NS messages. * This behavior is documented as Enhanced Duplicate Address * Detection in RFC 7527. * net.inet6.ip6.dad_enhanced=0 disables this. */ if (V_dad_enhanced != 0 && nonce != NULL) { int optlen = sizeof(struct nd_opt_hdr) + ND_OPT_NONCE_LEN; struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_ns + 1); /* 8-byte alignment is required. */ optlen = (optlen + 7) & ~7; m->m_pkthdr.len += optlen; m->m_len += optlen; icmp6len += optlen; bzero((caddr_t)nd_opt, optlen); nd_opt->nd_opt_type = ND_OPT_NONCE; nd_opt->nd_opt_len = optlen >> 3; bcopy(nonce, (caddr_t)(nd_opt + 1), ND_OPT_NONCE_LEN); } ip6->ip6_plen = htons((u_short)icmp6len); nd_ns->nd_ns_cksum = 0; nd_ns->nd_ns_cksum = in6_cksum(m, IPPROTO_ICMPV6, sizeof(*ip6), icmp6len); if (send_sendso_input_hook != NULL) { mtag = m_tag_get(PACKET_TAG_ND_OUTGOING, sizeof(unsigned short), M_NOWAIT); if (mtag == NULL) goto bad; *(unsigned short *)(mtag + 1) = nd_ns->nd_ns_type; m_tag_prepend(m, mtag); } ip6_output(m, NULL, NULL, (nonce != NULL) ? IPV6_UNSPECSRC : 0, &im6o, NULL, NULL); icmp6_ifstat_inc(ifp, ifs6_out_msg); icmp6_ifstat_inc(ifp, ifs6_out_neighborsolicit); ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_SOLICIT]); return; bad: m_freem(m); } #ifndef BURN_BRIDGES void nd6_ns_output(struct ifnet *ifp, const struct in6_addr *saddr6, const struct in6_addr *daddr6, const struct in6_addr *taddr6,uint8_t *nonce) { nd6_ns_output_fib(ifp, saddr6, daddr6, taddr6, nonce, RT_DEFAULT_FIB); } #endif /* * Neighbor advertisement input handling. * * Based on RFC 2461 * Based on RFC 2462 (duplicate address detection) * * the following items are not implemented yet: * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD) * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD) */ void nd6_na_input(struct mbuf *m, int off, int icmp6len) { struct ifnet *ifp = m->m_pkthdr.rcvif; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct nd_neighbor_advert *nd_na; struct in6_addr daddr6 = ip6->ip6_dst; struct in6_addr taddr6; int flags; int is_router; int is_solicited; int is_override; char *lladdr = NULL; int lladdrlen = 0; int checklink = 0; struct ifaddr *ifa; struct llentry *ln = NULL; union nd_opts ndopts; struct mbuf *chain = NULL; struct sockaddr_in6 sin6; u_char linkhdr[LLE_MAX_LINKHDR]; size_t linkhdrsize; int lladdr_off; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; /* RFC 6980: Nodes MUST silently ignore fragments */ if(m->m_flags & M_FRAGMENTED) goto freeit; if (ip6->ip6_hlim != 255) { nd6log((LOG_ERR, "nd6_na_input: invalid hlim (%d) from %s to %s on %s\n", ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp))); goto bad; } #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, icmp6len,); nd_na = (struct nd_neighbor_advert *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(nd_na, struct nd_neighbor_advert *, m, off, icmp6len); if (nd_na == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif flags = nd_na->nd_na_flags_reserved; is_router = ((flags & ND_NA_FLAG_ROUTER) != 0); is_solicited = ((flags & ND_NA_FLAG_SOLICITED) != 0); is_override = ((flags & ND_NA_FLAG_OVERRIDE) != 0); memset(&sin6, 0, sizeof(sin6)); taddr6 = nd_na->nd_na_target; if (in6_setscope(&taddr6, ifp, NULL)) goto bad; /* XXX: impossible */ if (IN6_IS_ADDR_MULTICAST(&taddr6)) { nd6log((LOG_ERR, "nd6_na_input: invalid target address %s\n", ip6_sprintf(ip6bufs, &taddr6))); goto bad; } if (IN6_IS_ADDR_MULTICAST(&daddr6)) if (is_solicited) { nd6log((LOG_ERR, "nd6_na_input: a solicited adv is multicasted\n")); goto bad; } icmp6len -= sizeof(*nd_na); nd6_option_init(nd_na + 1, icmp6len, &ndopts); if (nd6_options(&ndopts) < 0) { nd6log((LOG_INFO, "nd6_na_input: invalid ND option, ignored\n")); /* nd6_options have incremented stats */ goto freeit; } if (ndopts.nd_opts_tgt_lladdr) { lladdr = (char *)(ndopts.nd_opts_tgt_lladdr + 1); lladdrlen = ndopts.nd_opts_tgt_lladdr->nd_opt_len << 3; } /* * This effectively disables the DAD check on a non-master CARP * address. */ if (ifp->if_carp) ifa = (*carp_iamatch6_p)(ifp, &taddr6); else ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &taddr6); /* * Target address matches one of my interface address. * * If my address is tentative, this means that there's somebody * already using the same address as mine. This indicates DAD failure. * This is defined in RFC 2462. * * Otherwise, process as defined in RFC 2461. */ if (ifa && (((struct in6_ifaddr *)ifa)->ia6_flags & IN6_IFF_TENTATIVE)) { nd6_dad_na_input(ifa); ifa_free(ifa); goto freeit; } /* Just for safety, maybe unnecessary. */ if (ifa) { ifa_free(ifa); log(LOG_ERR, "nd6_na_input: duplicate IP6 address %s\n", ip6_sprintf(ip6bufs, &taddr6)); goto freeit; } if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) { nd6log((LOG_INFO, "nd6_na_input: lladdrlen mismatch for %s " "(if %d, NA packet %d)\n", ip6_sprintf(ip6bufs, &taddr6), ifp->if_addrlen, lladdrlen - 2)); goto bad; } /* * If no neighbor cache entry is found, NA SHOULD silently be * discarded. */ IF_AFDATA_RLOCK(ifp); ln = nd6_lookup(&taddr6, LLE_EXCLUSIVE, ifp); IF_AFDATA_RUNLOCK(ifp); if (ln == NULL) { goto freeit; } if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { /* * If the link-layer has address, and no lladdr option came, * discard the packet. */ if (ifp->if_addrlen && lladdr == NULL) { goto freeit; } /* * Record link-layer address, and update the state. */ linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET6, lladdr, linkhdr, &linkhdrsize, &lladdr_off) != 0) return; if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize, lladdr_off) == 0) { ln = NULL; goto freeit; } EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); if (is_solicited) nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE); else nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); if ((ln->ln_router = is_router) != 0) { /* * This means a router's state has changed from * non-reachable to probably reachable, and might * affect the status of associated prefixes.. */ checklink = 1; } } else { int llchange; /* * Check if the link-layer address has changed or not. */ if (lladdr == NULL) llchange = 0; else { if (ln->la_flags & LLE_VALID) { if (bcmp(lladdr, ln->ll_addr, ifp->if_addrlen)) llchange = 1; else llchange = 0; } else llchange = 1; } /* * This is VERY complex. Look at it with care. * * override solicit lladdr llchange action * (L: record lladdr) * * 0 0 n -- (2c) * 0 0 y n (2b) L * 0 0 y y (1) REACHABLE->STALE * 0 1 n -- (2c) *->REACHABLE * 0 1 y n (2b) L *->REACHABLE * 0 1 y y (1) REACHABLE->STALE * 1 0 n -- (2a) * 1 0 y n (2a) L * 1 0 y y (2a) L *->STALE * 1 1 n -- (2a) *->REACHABLE * 1 1 y n (2a) L *->REACHABLE * 1 1 y y (2a) L *->REACHABLE */ if (!is_override && (lladdr != NULL && llchange)) { /* (1) */ /* * If state is REACHABLE, make it STALE. * no other updates should be done. */ if (ln->ln_state == ND6_LLINFO_REACHABLE) nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); goto freeit; } else if (is_override /* (2a) */ || (!is_override && (lladdr != NULL && !llchange)) /* (2b) */ || lladdr == NULL) { /* (2c) */ /* * Update link-local address, if any. */ if (lladdr != NULL) { linkhdrsize = sizeof(linkhdr); if (lltable_calc_llheader(ifp, AF_INET6, lladdr, linkhdr, &linkhdrsize, &lladdr_off) != 0) goto freeit; if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize, lladdr_off) == 0) { ln = NULL; goto freeit; } EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); } /* * If solicited, make the state REACHABLE. * If not solicited and the link-layer address was * changed, make it STALE. */ if (is_solicited) nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE); else { if (lladdr != NULL && llchange) nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); } } if (ln->ln_router && !is_router) { /* * The peer dropped the router flag. * Remove the sender from the Default Router List and * update the Destination Cache entries. */ struct ifnet *nd6_ifp; nd6_ifp = lltable_get_ifp(ln->lle_tbl); if (!defrouter_remove(&ln->r_l3addr.addr6, nd6_ifp) && (ND_IFINFO(nd6_ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0) /* * Even if the neighbor is not in the default * router list, the neighbor may be used as a * next hop for some destinations (e.g. redirect * case). So we must call rt6_flush explicitly. */ rt6_flush(&ip6->ip6_src, ifp); } ln->ln_router = is_router; } /* XXX - QL * Does this matter? * rt->rt_flags &= ~RTF_REJECT; */ ln->la_asked = 0; if (ln->la_hold != NULL) nd6_grab_holdchain(ln, &chain, &sin6); freeit: if (ln != NULL) LLE_WUNLOCK(ln); if (chain != NULL) nd6_flush_holdchain(ifp, chain, &sin6); if (checklink) pfxlist_onlink_check(); m_freem(m); return; bad: if (ln != NULL) LLE_WUNLOCK(ln); ICMP6STAT_INC(icp6s_badna); m_freem(m); } /* * Neighbor advertisement output handling. * * Based on RFC 2461 * * the following items are not implemented yet: * - proxy advertisement delay rule (RFC2461 7.2.8, last paragraph, SHOULD) * - anycast advertisement delay rule (RFC2461 7.2.7, SHOULD) * * tlladdr - 1 if include target link-layer address * sdl0 - sockaddr_dl (= proxy NA) or NULL */ static void nd6_na_output_fib(struct ifnet *ifp, const struct in6_addr *daddr6_0, const struct in6_addr *taddr6, u_long flags, int tlladdr, struct sockaddr *sdl0, u_int fibnum) { struct mbuf *m; struct m_tag *mtag; struct ip6_hdr *ip6; struct nd_neighbor_advert *nd_na; struct ip6_moptions im6o; struct in6_addr daddr6, dst6, src6; uint32_t scopeid; int icmp6len, maxlen, error; caddr_t mac = NULL; daddr6 = *daddr6_0; /* make a local copy for modification */ /* estimate the size of message */ maxlen = sizeof(*ip6) + sizeof(*nd_na); maxlen += (sizeof(struct nd_opt_hdr) + ifp->if_addrlen + 7) & ~7; KASSERT(max_linkhdr + maxlen <= MCLBYTES, ( "%s: max_linkhdr + maxlen > MCLBYTES (%d + %d > %d)", __func__, max_linkhdr, maxlen, MCLBYTES)); if (max_linkhdr + maxlen > MHLEN) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; M_SETFIB(m, fibnum); if (IN6_IS_ADDR_MULTICAST(&daddr6)) { m->m_flags |= M_MCAST; im6o.im6o_multicast_ifp = ifp; im6o.im6o_multicast_hlim = 255; im6o.im6o_multicast_loop = 0; } icmp6len = sizeof(*nd_na); m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + icmp6len; m->m_data += max_linkhdr; /* or M_ALIGN() equivalent? */ /* fill neighbor advertisement packet */ ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_hlim = 255; if (IN6_IS_ADDR_UNSPECIFIED(&daddr6)) { /* reply to DAD */ daddr6.s6_addr16[0] = IPV6_ADDR_INT16_MLL; daddr6.s6_addr16[1] = 0; daddr6.s6_addr32[1] = 0; daddr6.s6_addr32[2] = 0; daddr6.s6_addr32[3] = IPV6_ADDR_INT32_ONE; if (in6_setscope(&daddr6, ifp, NULL)) goto bad; flags &= ~ND_NA_FLAG_SOLICITED; } ip6->ip6_dst = daddr6; /* * Select a source whose scope is the same as that of the dest. */ in6_splitscope(&daddr6, &dst6, &scopeid); error = in6_selectsrc_addr(fibnum, &dst6, scopeid, ifp, &src6, NULL); if (error) { char ip6buf[INET6_ADDRSTRLEN]; nd6log((LOG_DEBUG, "nd6_na_output: source can't be " "determined: dst=%s, error=%d\n", ip6_sprintf(ip6buf, &daddr6), error)); goto bad; } ip6->ip6_src = src6; nd_na = (struct nd_neighbor_advert *)(ip6 + 1); nd_na->nd_na_type = ND_NEIGHBOR_ADVERT; nd_na->nd_na_code = 0; nd_na->nd_na_target = *taddr6; in6_clearscope(&nd_na->nd_na_target); /* XXX */ /* * "tlladdr" indicates NS's condition for adding tlladdr or not. * see nd6_ns_input() for details. * Basically, if NS packet is sent to unicast/anycast addr, * target lladdr option SHOULD NOT be included. */ if (tlladdr) { /* * sdl0 != NULL indicates proxy NA. If we do proxy, use * lladdr in sdl0. If we are not proxying (sending NA for * my address) use lladdr configured for the interface. */ if (sdl0 == NULL) { if (ifp->if_carp) mac = (*carp_macmatch6_p)(ifp, m, taddr6); if (mac == NULL) mac = nd6_ifptomac(ifp); } else if (sdl0->sa_family == AF_LINK) { struct sockaddr_dl *sdl; sdl = (struct sockaddr_dl *)sdl0; if (sdl->sdl_alen == ifp->if_addrlen) mac = LLADDR(sdl); } } if (tlladdr && mac) { int optlen = sizeof(struct nd_opt_hdr) + ifp->if_addrlen; struct nd_opt_hdr *nd_opt = (struct nd_opt_hdr *)(nd_na + 1); /* roundup to 8 bytes alignment! */ optlen = (optlen + 7) & ~7; m->m_pkthdr.len += optlen; m->m_len += optlen; icmp6len += optlen; bzero((caddr_t)nd_opt, optlen); nd_opt->nd_opt_type = ND_OPT_TARGET_LINKADDR; nd_opt->nd_opt_len = optlen >> 3; bcopy(mac, (caddr_t)(nd_opt + 1), ifp->if_addrlen); } else flags &= ~ND_NA_FLAG_OVERRIDE; ip6->ip6_plen = htons((u_short)icmp6len); nd_na->nd_na_flags_reserved = flags; nd_na->nd_na_cksum = 0; nd_na->nd_na_cksum = in6_cksum(m, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), icmp6len); if (send_sendso_input_hook != NULL) { mtag = m_tag_get(PACKET_TAG_ND_OUTGOING, sizeof(unsigned short), M_NOWAIT); if (mtag == NULL) goto bad; *(unsigned short *)(mtag + 1) = nd_na->nd_na_type; m_tag_prepend(m, mtag); } ip6_output(m, NULL, NULL, 0, &im6o, NULL, NULL); icmp6_ifstat_inc(ifp, ifs6_out_msg); icmp6_ifstat_inc(ifp, ifs6_out_neighboradvert); ICMP6STAT_INC(icp6s_outhist[ND_NEIGHBOR_ADVERT]); return; bad: m_freem(m); } #ifndef BURN_BRIDGES void nd6_na_output(struct ifnet *ifp, const struct in6_addr *daddr6_0, const struct in6_addr *taddr6, u_long flags, int tlladdr, struct sockaddr *sdl0) { nd6_na_output_fib(ifp, daddr6_0, taddr6, flags, tlladdr, sdl0, RT_DEFAULT_FIB); } #endif caddr_t nd6_ifptomac(struct ifnet *ifp) { switch (ifp->if_type) { case IFT_ETHER: case IFT_IEEE1394: case IFT_L2VLAN: case IFT_INFINIBAND: case IFT_BRIDGE: return IF_LLADDR(ifp); default: return NULL; } } struct dadq { TAILQ_ENTRY(dadq) dad_list; struct ifaddr *dad_ifa; int dad_count; /* max NS to send */ int dad_ns_tcount; /* # of trials to send NS */ int dad_ns_ocount; /* NS sent so far */ int dad_ns_icount; int dad_na_icount; int dad_ns_lcount; /* looped back NS */ int dad_loopbackprobe; /* probing state for loopback detection */ struct callout dad_timer_ch; struct vnet *dad_vnet; u_int dad_refcnt; #define ND_OPT_NONCE_LEN32 \ ((ND_OPT_NONCE_LEN + sizeof(uint32_t) - 1)/sizeof(uint32_t)) uint32_t dad_nonce[ND_OPT_NONCE_LEN32]; bool dad_ondadq; /* on dadq? Protected by DADQ_WLOCK. */ }; -static VNET_DEFINE(TAILQ_HEAD(, dadq), dadq); -static VNET_DEFINE(struct rwlock, dad_rwlock); +VNET_DEFINE_STATIC(TAILQ_HEAD(, dadq), dadq); +VNET_DEFINE_STATIC(struct rwlock, dad_rwlock); #define V_dadq VNET(dadq) #define V_dad_rwlock VNET(dad_rwlock) #define DADQ_RLOCK() rw_rlock(&V_dad_rwlock) #define DADQ_RUNLOCK() rw_runlock(&V_dad_rwlock) #define DADQ_WLOCK() rw_wlock(&V_dad_rwlock) #define DADQ_WUNLOCK() rw_wunlock(&V_dad_rwlock) static void nd6_dad_add(struct dadq *dp) { DADQ_WLOCK(); TAILQ_INSERT_TAIL(&V_dadq, dp, dad_list); dp->dad_ondadq = true; DADQ_WUNLOCK(); } static void nd6_dad_del(struct dadq *dp) { DADQ_WLOCK(); if (dp->dad_ondadq) { /* * Remove dp from the dadq and release the dadq's * reference. */ TAILQ_REMOVE(&V_dadq, dp, dad_list); dp->dad_ondadq = false; DADQ_WUNLOCK(); nd6_dad_rele(dp); } else DADQ_WUNLOCK(); } static struct dadq * nd6_dad_find(struct ifaddr *ifa, struct nd_opt_nonce *n) { struct dadq *dp; DADQ_RLOCK(); TAILQ_FOREACH(dp, &V_dadq, dad_list) { if (dp->dad_ifa != ifa) continue; /* * Skip if the nonce matches the received one. * +2 in the length is required because of type and * length fields are included in a header. */ if (n != NULL && n->nd_opt_nonce_len == (ND_OPT_NONCE_LEN + 2) / 8 && memcmp(&n->nd_opt_nonce[0], &dp->dad_nonce[0], ND_OPT_NONCE_LEN) == 0) { dp->dad_ns_lcount++; continue; } refcount_acquire(&dp->dad_refcnt); break; } DADQ_RUNLOCK(); return (dp); } static void nd6_dad_starttimer(struct dadq *dp, int ticks, int send_ns) { if (send_ns != 0) nd6_dad_ns_output(dp); callout_reset(&dp->dad_timer_ch, ticks, (void (*)(void *))nd6_dad_timer, (void *)dp); } static void nd6_dad_stoptimer(struct dadq *dp) { callout_drain(&dp->dad_timer_ch); } static void nd6_dad_rele(struct dadq *dp) { if (refcount_release(&dp->dad_refcnt)) { ifa_free(dp->dad_ifa); free(dp, M_IP6NDP); } } void nd6_dad_init(void) { rw_init(&V_dad_rwlock, "nd6 DAD queue"); TAILQ_INIT(&V_dadq); } /* * Start Duplicate Address Detection (DAD) for specified interface address. */ void nd6_dad_start(struct ifaddr *ifa, int delay) { struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa; struct dadq *dp; char ip6buf[INET6_ADDRSTRLEN]; KASSERT((ia->ia6_flags & IN6_IFF_TENTATIVE) != 0, ("starting DAD on non-tentative address %p", ifa)); /* * If we don't need DAD, don't do it. * There are several cases: * - DAD is disabled globally or on the interface * - the interface address is anycast */ if ((ia->ia6_flags & IN6_IFF_ANYCAST) != 0 || V_ip6_dad_count == 0 || (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_NO_DAD) != 0) { ia->ia6_flags &= ~IN6_IFF_TENTATIVE; return; } if ((ifa->ifa_ifp->if_flags & IFF_UP) == 0 || (ifa->ifa_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || (ND_IFINFO(ifa->ifa_ifp)->flags & ND6_IFF_IFDISABLED) != 0) return; if ((dp = nd6_dad_find(ifa, NULL)) != NULL) { /* * DAD is already in progress. Let the existing entry * finish it. */ nd6_dad_rele(dp); return; } dp = malloc(sizeof(*dp), M_IP6NDP, M_NOWAIT | M_ZERO); if (dp == NULL) { log(LOG_ERR, "nd6_dad_start: memory allocation failed for " "%s(%s)\n", ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr), ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???"); return; } callout_init(&dp->dad_timer_ch, 0); #ifdef VIMAGE dp->dad_vnet = curvnet; #endif nd6log((LOG_DEBUG, "%s: starting DAD for %s\n", if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr))); /* * Send NS packet for DAD, ip6_dad_count times. * Note that we must delay the first transmission, if this is the * first packet to be sent from the interface after interface * (re)initialization. */ dp->dad_ifa = ifa; ifa_ref(dp->dad_ifa); dp->dad_count = V_ip6_dad_count; dp->dad_ns_icount = dp->dad_na_icount = 0; dp->dad_ns_ocount = dp->dad_ns_tcount = 0; dp->dad_ns_lcount = dp->dad_loopbackprobe = 0; /* Add this to the dadq and add a reference for the dadq. */ refcount_init(&dp->dad_refcnt, 1); nd6_dad_add(dp); nd6_dad_starttimer(dp, delay, 0); } /* * terminate DAD unconditionally. used for address removals. */ void nd6_dad_stop(struct ifaddr *ifa) { struct dadq *dp; dp = nd6_dad_find(ifa, NULL); if (!dp) { /* DAD wasn't started yet */ return; } nd6_dad_stoptimer(dp); nd6_dad_del(dp); /* Release this function's reference, acquired by nd6_dad_find(). */ nd6_dad_rele(dp); } static void nd6_dad_timer(struct dadq *dp) { CURVNET_SET(dp->dad_vnet); struct ifaddr *ifa = dp->dad_ifa; struct ifnet *ifp = dp->dad_ifa->ifa_ifp; struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa; char ip6buf[INET6_ADDRSTRLEN]; KASSERT(ia != NULL, ("DAD entry %p with no address", dp)); if (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) { /* Do not need DAD for ifdisabled interface. */ log(LOG_ERR, "nd6_dad_timer: cancel DAD on %s because of " "ND6_IFF_IFDISABLED.\n", ifp->if_xname); goto err; } if (ia->ia6_flags & IN6_IFF_DUPLICATED) { log(LOG_ERR, "nd6_dad_timer: called with duplicated address " "%s(%s)\n", ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr), ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???"); goto err; } if ((ia->ia6_flags & IN6_IFF_TENTATIVE) == 0) { log(LOG_ERR, "nd6_dad_timer: called with non-tentative address " "%s(%s)\n", ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr), ifa->ifa_ifp ? if_name(ifa->ifa_ifp) : "???"); goto err; } /* Stop DAD if the interface is down even after dad_maxtry attempts. */ if ((dp->dad_ns_tcount > V_dad_maxtry) && (((ifp->if_flags & IFF_UP) == 0) || ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))) { nd6log((LOG_INFO, "%s: could not run DAD " "because the interface was down or not running.\n", if_name(ifa->ifa_ifp))); goto err; } /* Need more checks? */ if (dp->dad_ns_ocount < dp->dad_count) { /* * We have more NS to go. Send NS packet for DAD. */ nd6_dad_starttimer(dp, (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000, 1); goto done; } else { /* * We have transmitted sufficient number of DAD packets. * See what we've got. */ if (dp->dad_ns_icount > 0 || dp->dad_na_icount > 0) /* We've seen NS or NA, means DAD has failed. */ nd6_dad_duplicated(ifa, dp); else if (V_dad_enhanced != 0 && dp->dad_ns_lcount > 0 && dp->dad_ns_lcount > dp->dad_loopbackprobe) { /* * Sec. 4.1 in RFC 7527 requires transmission of * additional probes until the loopback condition * becomes clear when a looped back probe is detected. */ log(LOG_ERR, "%s: a looped back NS message is " "detected during DAD for %s. " "Another DAD probes are being sent.\n", if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, IFA_IN6(ifa))); dp->dad_loopbackprobe = dp->dad_ns_lcount; /* * Send an NS immediately and increase dad_count by * V_nd6_mmaxtries - 1. */ dp->dad_count = dp->dad_ns_ocount + V_nd6_mmaxtries - 1; nd6_dad_starttimer(dp, (long)ND_IFINFO(ifa->ifa_ifp)->retrans * hz / 1000, 1); goto done; } else { /* * We are done with DAD. No NA came, no NS came. * No duplicate address found. Check IFDISABLED flag * again in case that it is changed between the * beginning of this function and here. */ if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) == 0) ia->ia6_flags &= ~IN6_IFF_TENTATIVE; nd6log((LOG_DEBUG, "%s: DAD complete for %s - no duplicates found\n", if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr))); if (dp->dad_ns_lcount > 0) log(LOG_ERR, "%s: DAD completed while " "a looped back NS message is detected " "during DAD for %s.\n", if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, IFA_IN6(ifa))); } } err: nd6_dad_del(dp); done: CURVNET_RESTORE(); } static void nd6_dad_duplicated(struct ifaddr *ifa, struct dadq *dp) { struct in6_ifaddr *ia = (struct in6_ifaddr *)ifa; struct ifnet *ifp; char ip6buf[INET6_ADDRSTRLEN]; log(LOG_ERR, "%s: DAD detected duplicate IPv6 address %s: " "NS in/out/loopback=%d/%d/%d, NA in=%d\n", if_name(ifa->ifa_ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr), dp->dad_ns_icount, dp->dad_ns_ocount, dp->dad_ns_lcount, dp->dad_na_icount); ia->ia6_flags &= ~IN6_IFF_TENTATIVE; ia->ia6_flags |= IN6_IFF_DUPLICATED; ifp = ifa->ifa_ifp; log(LOG_ERR, "%s: DAD complete for %s - duplicate found\n", if_name(ifp), ip6_sprintf(ip6buf, &ia->ia_addr.sin6_addr)); log(LOG_ERR, "%s: manual intervention required\n", if_name(ifp)); /* * If the address is a link-local address formed from an interface * identifier based on the hardware address which is supposed to be * uniquely assigned (e.g., EUI-64 for an Ethernet interface), IP * operation on the interface SHOULD be disabled. * [RFC 4862, Section 5.4.5] */ if (IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) { struct in6_addr in6; /* * To avoid over-reaction, we only apply this logic when we are * very sure that hardware addresses are supposed to be unique. */ switch (ifp->if_type) { case IFT_ETHER: case IFT_ATM: case IFT_IEEE1394: case IFT_INFINIBAND: in6 = ia->ia_addr.sin6_addr; if (in6_get_hw_ifid(ifp, &in6) == 0 && IN6_ARE_ADDR_EQUAL(&ia->ia_addr.sin6_addr, &in6)) { ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; log(LOG_ERR, "%s: possible hardware address " "duplication detected, disable IPv6\n", if_name(ifp)); } break; } } } static void nd6_dad_ns_output(struct dadq *dp) { struct in6_ifaddr *ia = (struct in6_ifaddr *)dp->dad_ifa; struct ifnet *ifp = dp->dad_ifa->ifa_ifp; int i; dp->dad_ns_tcount++; if ((ifp->if_flags & IFF_UP) == 0) { return; } if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { return; } dp->dad_ns_ocount++; if (V_dad_enhanced != 0) { for (i = 0; i < ND_OPT_NONCE_LEN32; i++) dp->dad_nonce[i] = arc4random(); /* * XXXHRS: Note that in the case that * DupAddrDetectTransmits > 1, multiple NS messages with * different nonces can be looped back in an unexpected * order. The current implementation recognizes only * the latest nonce on the sender side. Practically it * should work well in almost all cases. */ } nd6_ns_output(ifp, NULL, NULL, &ia->ia_addr.sin6_addr, (uint8_t *)&dp->dad_nonce[0]); } static void nd6_dad_ns_input(struct ifaddr *ifa, struct nd_opt_nonce *ndopt_nonce) { struct dadq *dp; if (ifa == NULL) panic("ifa == NULL in nd6_dad_ns_input"); /* Ignore Nonce option when Enhanced DAD is disabled. */ if (V_dad_enhanced == 0) ndopt_nonce = NULL; dp = nd6_dad_find(ifa, ndopt_nonce); if (dp == NULL) return; dp->dad_ns_icount++; nd6_dad_rele(dp); } static void nd6_dad_na_input(struct ifaddr *ifa) { struct dadq *dp; if (ifa == NULL) panic("ifa == NULL in nd6_dad_na_input"); dp = nd6_dad_find(ifa, NULL); if (dp != NULL) { dp->dad_na_icount++; nd6_dad_rele(dp); } } Index: head/sys/netinet6/nd6_rtr.c =================================================================== --- head/sys/netinet6/nd6_rtr.c (revision 336675) +++ head/sys/netinet6/nd6_rtr.c (revision 336676) @@ -1,2408 +1,2408 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: nd6_rtr.c,v 1.111 2001/04/27 01:37:15 jinmei Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int rtpref(struct nd_defrouter *); static struct nd_defrouter *defrtrlist_update(struct nd_defrouter *); static int prelist_update(struct nd_prefixctl *, struct nd_defrouter *, struct mbuf *, int); static struct in6_ifaddr *in6_ifadd(struct nd_prefixctl *, int); static struct nd_pfxrouter *pfxrtr_lookup(struct nd_prefix *, struct nd_defrouter *); static void pfxrtr_add(struct nd_prefix *, struct nd_defrouter *); static void pfxrtr_del(struct nd_pfxrouter *); static struct nd_pfxrouter *find_pfxlist_reachable_router(struct nd_prefix *); static void defrouter_delreq(struct nd_defrouter *); static void nd6_rtmsg(int, struct rtentry *); static int in6_init_prefix_ltimes(struct nd_prefix *); static void in6_init_address_ltimes(struct nd_prefix *, struct in6_addrlifetime *); static int rt6_deleteroute(const struct rtentry *, void *); VNET_DECLARE(int, nd6_recalc_reachtm_interval); #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) -static VNET_DEFINE(struct ifnet *, nd6_defifp); +VNET_DEFINE_STATIC(struct ifnet *, nd6_defifp); VNET_DEFINE(int, nd6_defifindex); #define V_nd6_defifp VNET(nd6_defifp) VNET_DEFINE(int, ip6_use_tempaddr) = 0; VNET_DEFINE(int, ip6_desync_factor); VNET_DEFINE(u_int32_t, ip6_temp_preferred_lifetime) = DEF_TEMP_PREFERRED_LIFETIME; VNET_DEFINE(u_int32_t, ip6_temp_valid_lifetime) = DEF_TEMP_VALID_LIFETIME; VNET_DEFINE(int, ip6_temp_regen_advance) = TEMPADDR_REGEN_ADVANCE; /* RTPREF_MEDIUM has to be 0! */ #define RTPREF_HIGH 1 #define RTPREF_MEDIUM 0 #define RTPREF_LOW (-1) #define RTPREF_RESERVED (-2) #define RTPREF_INVALID (-3) /* internal */ /* * Receive Router Solicitation Message - just for routers. * Router solicitation/advertisement is mostly managed by userland program * (rtadvd) so here we have no function like nd6_ra_output(). * * Based on RFC 2461 */ void nd6_rs_input(struct mbuf *m, int off, int icmp6len) { struct ifnet *ifp = m->m_pkthdr.rcvif; struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct nd_router_solicit *nd_rs; struct in6_addr saddr6 = ip6->ip6_src; char *lladdr = NULL; int lladdrlen = 0; union nd_opts ndopts; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; /* * Accept RS only when V_ip6_forwarding=1 and the interface has * no ND6_IFF_ACCEPT_RTADV. */ if (!V_ip6_forwarding || ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) goto freeit; /* RFC 6980: Nodes MUST silently ignore fragments */ if(m->m_flags & M_FRAGMENTED) goto freeit; /* Sanity checks */ if (ip6->ip6_hlim != 255) { nd6log((LOG_ERR, "nd6_rs_input: invalid hlim (%d) from %s to %s on %s\n", ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp))); goto bad; } /* * Don't update the neighbor cache, if src = ::. * This indicates that the src has no IP address assigned yet. */ if (IN6_IS_ADDR_UNSPECIFIED(&saddr6)) goto freeit; #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, icmp6len,); nd_rs = (struct nd_router_solicit *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(nd_rs, struct nd_router_solicit *, m, off, icmp6len); if (nd_rs == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif icmp6len -= sizeof(*nd_rs); nd6_option_init(nd_rs + 1, icmp6len, &ndopts); if (nd6_options(&ndopts) < 0) { nd6log((LOG_INFO, "nd6_rs_input: invalid ND option, ignored\n")); /* nd6_options have incremented stats */ goto freeit; } if (ndopts.nd_opts_src_lladdr) { lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1); lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3; } if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) { nd6log((LOG_INFO, "nd6_rs_input: lladdrlen mismatch for %s " "(if %d, RS packet %d)\n", ip6_sprintf(ip6bufs, &saddr6), ifp->if_addrlen, lladdrlen - 2)); goto bad; } nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen, ND_ROUTER_SOLICIT, 0); freeit: m_freem(m); return; bad: ICMP6STAT_INC(icp6s_badrs); m_freem(m); } /* * Receive Router Advertisement Message. * * Based on RFC 2461 * TODO: on-link bit on prefix information * TODO: ND_RA_FLAG_{OTHER,MANAGED} processing */ void nd6_ra_input(struct mbuf *m, int off, int icmp6len) { struct ifnet *ifp = m->m_pkthdr.rcvif; struct nd_ifinfo *ndi = ND_IFINFO(ifp); struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct nd_router_advert *nd_ra; struct in6_addr saddr6 = ip6->ip6_src; int mcast = 0; union nd_opts ndopts; struct nd_defrouter *dr; char ip6bufs[INET6_ADDRSTRLEN], ip6bufd[INET6_ADDRSTRLEN]; dr = NULL; /* * We only accept RAs only when the per-interface flag * ND6_IFF_ACCEPT_RTADV is on the receiving interface. */ if (!(ndi->flags & ND6_IFF_ACCEPT_RTADV)) goto freeit; /* RFC 6980: Nodes MUST silently ignore fragments */ if(m->m_flags & M_FRAGMENTED) goto freeit; if (ip6->ip6_hlim != 255) { nd6log((LOG_ERR, "nd6_ra_input: invalid hlim (%d) from %s to %s on %s\n", ip6->ip6_hlim, ip6_sprintf(ip6bufs, &ip6->ip6_src), ip6_sprintf(ip6bufd, &ip6->ip6_dst), if_name(ifp))); goto bad; } if (!IN6_IS_ADDR_LINKLOCAL(&saddr6)) { nd6log((LOG_ERR, "nd6_ra_input: src %s is not link-local\n", ip6_sprintf(ip6bufs, &saddr6))); goto bad; } #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, off, icmp6len,); nd_ra = (struct nd_router_advert *)((caddr_t)ip6 + off); #else IP6_EXTHDR_GET(nd_ra, struct nd_router_advert *, m, off, icmp6len); if (nd_ra == NULL) { ICMP6STAT_INC(icp6s_tooshort); return; } #endif icmp6len -= sizeof(*nd_ra); nd6_option_init(nd_ra + 1, icmp6len, &ndopts); if (nd6_options(&ndopts) < 0) { nd6log((LOG_INFO, "nd6_ra_input: invalid ND option, ignored\n")); /* nd6_options have incremented stats */ goto freeit; } { struct nd_defrouter dr0; u_int32_t advreachable = nd_ra->nd_ra_reachable; /* remember if this is a multicasted advertisement */ if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) mcast = 1; bzero(&dr0, sizeof(dr0)); dr0.rtaddr = saddr6; dr0.raflags = nd_ra->nd_ra_flags_reserved; /* * Effectively-disable routes from RA messages when * ND6_IFF_NO_RADR enabled on the receiving interface or * (ip6.forwarding == 1 && ip6.rfc6204w3 != 1). */ if (ndi->flags & ND6_IFF_NO_RADR) dr0.rtlifetime = 0; else if (V_ip6_forwarding && !V_ip6_rfc6204w3) dr0.rtlifetime = 0; else dr0.rtlifetime = ntohs(nd_ra->nd_ra_router_lifetime); dr0.expire = time_uptime + dr0.rtlifetime; dr0.ifp = ifp; /* unspecified or not? (RFC 2461 6.3.4) */ if (advreachable) { advreachable = ntohl(advreachable); if (advreachable <= MAX_REACHABLE_TIME && ndi->basereachable != advreachable) { ndi->basereachable = advreachable; ndi->reachable = ND_COMPUTE_RTIME(ndi->basereachable); ndi->recalctm = V_nd6_recalc_reachtm_interval; /* reset */ } } if (nd_ra->nd_ra_retransmit) ndi->retrans = ntohl(nd_ra->nd_ra_retransmit); if (nd_ra->nd_ra_curhoplimit) { if (ndi->chlim < nd_ra->nd_ra_curhoplimit) ndi->chlim = nd_ra->nd_ra_curhoplimit; else if (ndi->chlim != nd_ra->nd_ra_curhoplimit) { log(LOG_ERR, "RA with a lower CurHopLimit sent from " "%s on %s (current = %d, received = %d). " "Ignored.\n", ip6_sprintf(ip6bufs, &ip6->ip6_src), if_name(ifp), ndi->chlim, nd_ra->nd_ra_curhoplimit); } } dr = defrtrlist_update(&dr0); } /* * prefix */ if (ndopts.nd_opts_pi) { struct nd_opt_hdr *pt; struct nd_opt_prefix_info *pi = NULL; struct nd_prefixctl pr; for (pt = (struct nd_opt_hdr *)ndopts.nd_opts_pi; pt <= (struct nd_opt_hdr *)ndopts.nd_opts_pi_end; pt = (struct nd_opt_hdr *)((caddr_t)pt + (pt->nd_opt_len << 3))) { if (pt->nd_opt_type != ND_OPT_PREFIX_INFORMATION) continue; pi = (struct nd_opt_prefix_info *)pt; if (pi->nd_opt_pi_len != 4) { nd6log((LOG_INFO, "nd6_ra_input: invalid option " "len %d for prefix information option, " "ignored\n", pi->nd_opt_pi_len)); continue; } if (128 < pi->nd_opt_pi_prefix_len) { nd6log((LOG_INFO, "nd6_ra_input: invalid prefix " "len %d for prefix information option, " "ignored\n", pi->nd_opt_pi_prefix_len)); continue; } if (IN6_IS_ADDR_MULTICAST(&pi->nd_opt_pi_prefix) || IN6_IS_ADDR_LINKLOCAL(&pi->nd_opt_pi_prefix)) { nd6log((LOG_INFO, "nd6_ra_input: invalid prefix " "%s, ignored\n", ip6_sprintf(ip6bufs, &pi->nd_opt_pi_prefix))); continue; } bzero(&pr, sizeof(pr)); pr.ndpr_prefix.sin6_family = AF_INET6; pr.ndpr_prefix.sin6_len = sizeof(pr.ndpr_prefix); pr.ndpr_prefix.sin6_addr = pi->nd_opt_pi_prefix; pr.ndpr_ifp = (struct ifnet *)m->m_pkthdr.rcvif; pr.ndpr_raf_onlink = (pi->nd_opt_pi_flags_reserved & ND_OPT_PI_FLAG_ONLINK) ? 1 : 0; pr.ndpr_raf_auto = (pi->nd_opt_pi_flags_reserved & ND_OPT_PI_FLAG_AUTO) ? 1 : 0; pr.ndpr_plen = pi->nd_opt_pi_prefix_len; pr.ndpr_vltime = ntohl(pi->nd_opt_pi_valid_time); pr.ndpr_pltime = ntohl(pi->nd_opt_pi_preferred_time); (void)prelist_update(&pr, dr, m, mcast); } } if (dr != NULL) { defrouter_rele(dr); dr = NULL; } /* * MTU */ if (ndopts.nd_opts_mtu && ndopts.nd_opts_mtu->nd_opt_mtu_len == 1) { u_long mtu; u_long maxmtu; mtu = (u_long)ntohl(ndopts.nd_opts_mtu->nd_opt_mtu_mtu); /* lower bound */ if (mtu < IPV6_MMTU) { nd6log((LOG_INFO, "nd6_ra_input: bogus mtu option " "mtu=%lu sent from %s, ignoring\n", mtu, ip6_sprintf(ip6bufs, &ip6->ip6_src))); goto skip; } /* upper bound */ maxmtu = (ndi->maxmtu && ndi->maxmtu < ifp->if_mtu) ? ndi->maxmtu : ifp->if_mtu; if (mtu <= maxmtu) { int change = (ndi->linkmtu != mtu); ndi->linkmtu = mtu; if (change) { /* in6_maxmtu may change */ in6_setmaxmtu(); rt_updatemtu(ifp); } } else { nd6log((LOG_INFO, "nd6_ra_input: bogus mtu " "mtu=%lu sent from %s; " "exceeds maxmtu %lu, ignoring\n", mtu, ip6_sprintf(ip6bufs, &ip6->ip6_src), maxmtu)); } } skip: /* * Source link layer address */ { char *lladdr = NULL; int lladdrlen = 0; if (ndopts.nd_opts_src_lladdr) { lladdr = (char *)(ndopts.nd_opts_src_lladdr + 1); lladdrlen = ndopts.nd_opts_src_lladdr->nd_opt_len << 3; } if (lladdr && ((ifp->if_addrlen + 2 + 7) & ~7) != lladdrlen) { nd6log((LOG_INFO, "nd6_ra_input: lladdrlen mismatch for %s " "(if %d, RA packet %d)\n", ip6_sprintf(ip6bufs, &saddr6), ifp->if_addrlen, lladdrlen - 2)); goto bad; } nd6_cache_lladdr(ifp, &saddr6, lladdr, lladdrlen, ND_ROUTER_ADVERT, 0); /* * Installing a link-layer address might change the state of the * router's neighbor cache, which might also affect our on-link * detection of adveritsed prefixes. */ pfxlist_onlink_check(); } freeit: m_freem(m); return; bad: ICMP6STAT_INC(icp6s_badra); m_freem(m); } /* tell the change to user processes watching the routing socket. */ static void nd6_rtmsg(int cmd, struct rtentry *rt) { struct rt_addrinfo info; struct ifnet *ifp; struct ifaddr *ifa; bzero((caddr_t)&info, sizeof(info)); info.rti_info[RTAX_DST] = rt_key(rt); info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; info.rti_info[RTAX_NETMASK] = rt_mask(rt); ifp = rt->rt_ifp; if (ifp != NULL) { IF_ADDR_RLOCK(ifp); ifa = CK_STAILQ_FIRST(&ifp->if_addrhead); info.rti_info[RTAX_IFP] = ifa->ifa_addr; ifa_ref(ifa); IF_ADDR_RUNLOCK(ifp); info.rti_info[RTAX_IFA] = rt->rt_ifa->ifa_addr; } else ifa = NULL; rt_missmsg_fib(cmd, &info, rt->rt_flags, 0, rt->rt_fibnum); if (ifa != NULL) ifa_free(ifa); } /* * default router list processing sub routines */ static void defrouter_addreq(struct nd_defrouter *new) { struct sockaddr_in6 def, mask, gate; struct rtentry *newrt = NULL; int error; bzero(&def, sizeof(def)); bzero(&mask, sizeof(mask)); bzero(&gate, sizeof(gate)); def.sin6_len = mask.sin6_len = gate.sin6_len = sizeof(struct sockaddr_in6); def.sin6_family = gate.sin6_family = AF_INET6; gate.sin6_addr = new->rtaddr; error = in6_rtrequest(RTM_ADD, (struct sockaddr *)&def, (struct sockaddr *)&gate, (struct sockaddr *)&mask, RTF_GATEWAY, &newrt, new->ifp->if_fib); if (newrt) { nd6_rtmsg(RTM_ADD, newrt); /* tell user process */ RTFREE(newrt); } if (error == 0) new->installed = 1; } struct nd_defrouter * defrouter_lookup_locked(struct in6_addr *addr, struct ifnet *ifp) { struct nd_defrouter *dr; ND6_LOCK_ASSERT(); TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) if (dr->ifp == ifp && IN6_ARE_ADDR_EQUAL(addr, &dr->rtaddr)) { defrouter_ref(dr); return (dr); } return (NULL); } struct nd_defrouter * defrouter_lookup(struct in6_addr *addr, struct ifnet *ifp) { struct nd_defrouter *dr; ND6_RLOCK(); dr = defrouter_lookup_locked(addr, ifp); ND6_RUNLOCK(); return (dr); } void defrouter_ref(struct nd_defrouter *dr) { refcount_acquire(&dr->refcnt); } void defrouter_rele(struct nd_defrouter *dr) { if (refcount_release(&dr->refcnt)) free(dr, M_IP6NDP); } /* * Remove the default route for a given router. * This is just a subroutine function for defrouter_select_fib(), and * should not be called from anywhere else. */ static void defrouter_delreq(struct nd_defrouter *dr) { struct sockaddr_in6 def, mask, gate; struct rtentry *oldrt = NULL; bzero(&def, sizeof(def)); bzero(&mask, sizeof(mask)); bzero(&gate, sizeof(gate)); def.sin6_len = mask.sin6_len = gate.sin6_len = sizeof(struct sockaddr_in6); def.sin6_family = gate.sin6_family = AF_INET6; gate.sin6_addr = dr->rtaddr; in6_rtrequest(RTM_DELETE, (struct sockaddr *)&def, (struct sockaddr *)&gate, (struct sockaddr *)&mask, RTF_GATEWAY, &oldrt, dr->ifp->if_fib); if (oldrt) { nd6_rtmsg(RTM_DELETE, oldrt); RTFREE(oldrt); } dr->installed = 0; } /* * Remove all default routes from default router list. */ void defrouter_reset(void) { struct nd_defrouter *dr, **dra; int count, i; count = i = 0; /* * We can't delete routes with the ND lock held, so make a copy of the * current default router list and use that when deleting routes. */ ND6_RLOCK(); TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) count++; ND6_RUNLOCK(); dra = malloc(count * sizeof(*dra), M_TEMP, M_WAITOK | M_ZERO); ND6_RLOCK(); TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { if (i == count) break; defrouter_ref(dr); dra[i++] = dr; } ND6_RUNLOCK(); for (i = 0; i < count && dra[i] != NULL; i++) { defrouter_delreq(dra[i]); defrouter_rele(dra[i]); } free(dra, M_TEMP); /* * XXX should we also nuke any default routers in the kernel, by * going through them by rtalloc1()? */ } /* * Look up a matching default router list entry and remove it. Returns true if a * matching entry was found, false otherwise. */ bool defrouter_remove(struct in6_addr *addr, struct ifnet *ifp) { struct nd_defrouter *dr; ND6_WLOCK(); dr = defrouter_lookup_locked(addr, ifp); if (dr == NULL) { ND6_WUNLOCK(); return (false); } defrouter_unlink(dr, NULL); ND6_WUNLOCK(); defrouter_del(dr); defrouter_rele(dr); return (true); } /* * Remove a router from the global list and optionally stash it in a * caller-supplied queue. * * The ND lock must be held. */ void defrouter_unlink(struct nd_defrouter *dr, struct nd_drhead *drq) { ND6_WLOCK_ASSERT(); TAILQ_REMOVE(&V_nd_defrouter, dr, dr_entry); V_nd6_list_genid++; if (drq != NULL) TAILQ_INSERT_TAIL(drq, dr, dr_entry); } void defrouter_del(struct nd_defrouter *dr) { struct nd_defrouter *deldr = NULL; struct nd_prefix *pr; struct nd_pfxrouter *pfxrtr; ND6_UNLOCK_ASSERT(); /* * Flush all the routing table entries that use the router * as a next hop. */ if (ND_IFINFO(dr->ifp)->flags & ND6_IFF_ACCEPT_RTADV) rt6_flush(&dr->rtaddr, dr->ifp); if (dr->installed) { deldr = dr; defrouter_delreq(dr); } /* * Also delete all the pointers to the router in each prefix lists. */ ND6_WLOCK(); LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { if ((pfxrtr = pfxrtr_lookup(pr, dr)) != NULL) pfxrtr_del(pfxrtr); } ND6_WUNLOCK(); pfxlist_onlink_check(); /* * If the router is the primary one, choose a new one. * Note that defrouter_select_fib() will remove the current * gateway from the routing table. */ if (deldr) defrouter_select_fib(deldr->ifp->if_fib); /* * Release the list reference. */ defrouter_rele(dr); } /* * Default Router Selection according to Section 6.3.6 of RFC 2461 and * draft-ietf-ipngwg-router-selection: * 1) Routers that are reachable or probably reachable should be preferred. * If we have more than one (probably) reachable router, prefer ones * with the highest router preference. * 2) When no routers on the list are known to be reachable or * probably reachable, routers SHOULD be selected in a round-robin * fashion, regardless of router preference values. * 3) If the Default Router List is empty, assume that all * destinations are on-link. * * We assume nd_defrouter is sorted by router preference value. * Since the code below covers both with and without router preference cases, * we do not need to classify the cases by ifdef. * * At this moment, we do not try to install more than one default router, * even when the multipath routing is available, because we're not sure about * the benefits for stub hosts comparing to the risk of making the code * complicated and the possibility of introducing bugs. * * We maintain a single list of routers for multiple FIBs, only considering one * at a time based on the receiving interface's FIB. If @fibnum is RT_ALL_FIBS, * we do the whole thing multiple times. */ void defrouter_select_fib(int fibnum) { struct nd_defrouter *dr, *selected_dr, *installed_dr; struct llentry *ln = NULL; if (fibnum == RT_ALL_FIBS) { for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { defrouter_select_fib(fibnum); } } ND6_RLOCK(); /* * Let's handle easy case (3) first: * If default router list is empty, there's nothing to be done. */ if (TAILQ_EMPTY(&V_nd_defrouter)) { ND6_RUNLOCK(); return; } /* * Search for a (probably) reachable router from the list. * We just pick up the first reachable one (if any), assuming that * the ordering rule of the list described in defrtrlist_update(). */ selected_dr = installed_dr = NULL; TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { IF_AFDATA_RLOCK(dr->ifp); if (selected_dr == NULL && dr->ifp->if_fib == fibnum && (ln = nd6_lookup(&dr->rtaddr, 0, dr->ifp)) && ND6_IS_LLINFO_PROBREACH(ln)) { selected_dr = dr; defrouter_ref(selected_dr); } IF_AFDATA_RUNLOCK(dr->ifp); if (ln != NULL) { LLE_RUNLOCK(ln); ln = NULL; } if (dr->installed && dr->ifp->if_fib == fibnum) { if (installed_dr == NULL) { installed_dr = dr; defrouter_ref(installed_dr); } else { /* * this should not happen. * warn for diagnosis. */ log(LOG_ERR, "defrouter_select_fib: more than " "one router is installed\n"); } } } /* * If none of the default routers was found to be reachable, * round-robin the list regardless of preference. * Otherwise, if we have an installed router, check if the selected * (reachable) router should really be preferred to the installed one. * We only prefer the new router when the old one is not reachable * or when the new one has a really higher preference value. */ if (selected_dr == NULL) { if (installed_dr == NULL || TAILQ_NEXT(installed_dr, dr_entry) == NULL) dr = TAILQ_FIRST(&V_nd_defrouter); else dr = TAILQ_NEXT(installed_dr, dr_entry); /* Ensure we select a router for this FIB. */ TAILQ_FOREACH_FROM(dr, &V_nd_defrouter, dr_entry) { if (dr->ifp->if_fib == fibnum) { selected_dr = dr; defrouter_ref(selected_dr); break; } } } else if (installed_dr != NULL) { IF_AFDATA_RLOCK(installed_dr->ifp); if ((ln = nd6_lookup(&installed_dr->rtaddr, 0, installed_dr->ifp)) && ND6_IS_LLINFO_PROBREACH(ln) && installed_dr->ifp->if_fib == fibnum && rtpref(selected_dr) <= rtpref(installed_dr)) { defrouter_rele(selected_dr); selected_dr = installed_dr; } IF_AFDATA_RUNLOCK(installed_dr->ifp); if (ln != NULL) LLE_RUNLOCK(ln); } ND6_RUNLOCK(); /* * If we selected a router for this FIB and it's different * than the installed one, remove the installed router and * install the selected one in its place. */ if (installed_dr != selected_dr) { if (installed_dr != NULL) { defrouter_delreq(installed_dr); defrouter_rele(installed_dr); } if (selected_dr != NULL) defrouter_addreq(selected_dr); } if (selected_dr != NULL) defrouter_rele(selected_dr); } /* * Maintain old KPI for default router selection. * If unspecified, we can re-select routers for all FIBs. */ void defrouter_select(void) { defrouter_select_fib(RT_ALL_FIBS); } /* * for default router selection * regards router-preference field as a 2-bit signed integer */ static int rtpref(struct nd_defrouter *dr) { switch (dr->raflags & ND_RA_FLAG_RTPREF_MASK) { case ND_RA_FLAG_RTPREF_HIGH: return (RTPREF_HIGH); case ND_RA_FLAG_RTPREF_MEDIUM: case ND_RA_FLAG_RTPREF_RSV: return (RTPREF_MEDIUM); case ND_RA_FLAG_RTPREF_LOW: return (RTPREF_LOW); default: /* * This case should never happen. If it did, it would mean a * serious bug of kernel internal. We thus always bark here. * Or, can we even panic? */ log(LOG_ERR, "rtpref: impossible RA flag %x\n", dr->raflags); return (RTPREF_INVALID); } /* NOTREACHED */ } static struct nd_defrouter * defrtrlist_update(struct nd_defrouter *new) { struct nd_defrouter *dr, *n; uint64_t genid; int oldpref; bool writelocked; if (new->rtlifetime == 0) { defrouter_remove(&new->rtaddr, new->ifp); return (NULL); } ND6_RLOCK(); writelocked = false; restart: dr = defrouter_lookup_locked(&new->rtaddr, new->ifp); if (dr != NULL) { oldpref = rtpref(dr); /* override */ dr->raflags = new->raflags; /* XXX flag check */ dr->rtlifetime = new->rtlifetime; dr->expire = new->expire; /* * If the preference does not change, there's no need * to sort the entries. Also make sure the selected * router is still installed in the kernel. */ if (dr->installed && rtpref(new) == oldpref) { if (writelocked) ND6_WUNLOCK(); else ND6_RUNLOCK(); return (dr); } } /* * The router needs to be reinserted into the default router * list, so upgrade to a write lock. If that fails and the list * has potentially changed while the lock was dropped, we'll * redo the lookup with the write lock held. */ if (!writelocked) { writelocked = true; if (!ND6_TRY_UPGRADE()) { genid = V_nd6_list_genid; ND6_RUNLOCK(); ND6_WLOCK(); if (genid != V_nd6_list_genid) goto restart; } } if (dr != NULL) { /* * The preferred router may have changed, so relocate this * router. */ TAILQ_REMOVE(&V_nd_defrouter, dr, dr_entry); n = dr; } else { n = malloc(sizeof(*n), M_IP6NDP, M_NOWAIT | M_ZERO); if (n == NULL) { ND6_WUNLOCK(); return (NULL); } memcpy(n, new, sizeof(*n)); /* Initialize with an extra reference for the caller. */ refcount_init(&n->refcnt, 2); } /* * Insert the new router in the Default Router List; * The Default Router List should be in the descending order * of router-preferece. Routers with the same preference are * sorted in the arriving time order. */ /* insert at the end of the group */ TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { if (rtpref(n) > rtpref(dr)) break; } if (dr != NULL) TAILQ_INSERT_BEFORE(dr, n, dr_entry); else TAILQ_INSERT_TAIL(&V_nd_defrouter, n, dr_entry); V_nd6_list_genid++; ND6_WUNLOCK(); defrouter_select_fib(new->ifp->if_fib); return (n); } static struct nd_pfxrouter * pfxrtr_lookup(struct nd_prefix *pr, struct nd_defrouter *dr) { struct nd_pfxrouter *search; ND6_LOCK_ASSERT(); LIST_FOREACH(search, &pr->ndpr_advrtrs, pfr_entry) { if (search->router == dr) break; } return (search); } static void pfxrtr_add(struct nd_prefix *pr, struct nd_defrouter *dr) { struct nd_pfxrouter *new; bool update; ND6_UNLOCK_ASSERT(); ND6_RLOCK(); if (pfxrtr_lookup(pr, dr) != NULL) { ND6_RUNLOCK(); return; } ND6_RUNLOCK(); new = malloc(sizeof(*new), M_IP6NDP, M_NOWAIT | M_ZERO); if (new == NULL) return; defrouter_ref(dr); new->router = dr; ND6_WLOCK(); if (pfxrtr_lookup(pr, dr) == NULL) { LIST_INSERT_HEAD(&pr->ndpr_advrtrs, new, pfr_entry); update = true; } else { /* We lost a race to add the reference. */ defrouter_rele(dr); free(new, M_IP6NDP); update = false; } ND6_WUNLOCK(); if (update) pfxlist_onlink_check(); } static void pfxrtr_del(struct nd_pfxrouter *pfr) { ND6_WLOCK_ASSERT(); LIST_REMOVE(pfr, pfr_entry); defrouter_rele(pfr->router); free(pfr, M_IP6NDP); } static struct nd_prefix * nd6_prefix_lookup_locked(struct nd_prefixctl *key) { struct nd_prefix *search; ND6_LOCK_ASSERT(); LIST_FOREACH(search, &V_nd_prefix, ndpr_entry) { if (key->ndpr_ifp == search->ndpr_ifp && key->ndpr_plen == search->ndpr_plen && in6_are_prefix_equal(&key->ndpr_prefix.sin6_addr, &search->ndpr_prefix.sin6_addr, key->ndpr_plen)) { nd6_prefix_ref(search); break; } } return (search); } struct nd_prefix * nd6_prefix_lookup(struct nd_prefixctl *key) { struct nd_prefix *search; ND6_RLOCK(); search = nd6_prefix_lookup_locked(key); ND6_RUNLOCK(); return (search); } void nd6_prefix_ref(struct nd_prefix *pr) { refcount_acquire(&pr->ndpr_refcnt); } void nd6_prefix_rele(struct nd_prefix *pr) { if (refcount_release(&pr->ndpr_refcnt)) { KASSERT(LIST_EMPTY(&pr->ndpr_advrtrs), ("prefix %p has advertising routers", pr)); free(pr, M_IP6NDP); } } int nd6_prelist_add(struct nd_prefixctl *pr, struct nd_defrouter *dr, struct nd_prefix **newp) { struct nd_prefix *new; char ip6buf[INET6_ADDRSTRLEN]; int error; new = malloc(sizeof(*new), M_IP6NDP, M_NOWAIT | M_ZERO); if (new == NULL) return (ENOMEM); refcount_init(&new->ndpr_refcnt, newp != NULL ? 2 : 1); new->ndpr_ifp = pr->ndpr_ifp; new->ndpr_prefix = pr->ndpr_prefix; new->ndpr_plen = pr->ndpr_plen; new->ndpr_vltime = pr->ndpr_vltime; new->ndpr_pltime = pr->ndpr_pltime; new->ndpr_flags = pr->ndpr_flags; if ((error = in6_init_prefix_ltimes(new)) != 0) { free(new, M_IP6NDP); return (error); } new->ndpr_lastupdate = time_uptime; /* initialization */ LIST_INIT(&new->ndpr_advrtrs); in6_prefixlen2mask(&new->ndpr_mask, new->ndpr_plen); /* make prefix in the canonical form */ IN6_MASK_ADDR(&new->ndpr_prefix.sin6_addr, &new->ndpr_mask); ND6_WLOCK(); LIST_INSERT_HEAD(&V_nd_prefix, new, ndpr_entry); V_nd6_list_genid++; ND6_WUNLOCK(); /* ND_OPT_PI_FLAG_ONLINK processing */ if (new->ndpr_raf_onlink) { ND6_ONLINK_LOCK(); if ((error = nd6_prefix_onlink(new)) != 0) { nd6log((LOG_ERR, "nd6_prelist_add: failed to make " "the prefix %s/%d on-link on %s (errno=%d)\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, if_name(pr->ndpr_ifp), error)); /* proceed anyway. XXX: is it correct? */ } ND6_ONLINK_UNLOCK(); } if (dr != NULL) pfxrtr_add(new, dr); if (newp != NULL) *newp = new; return (0); } /* * Remove a prefix from the prefix list and optionally stash it in a * caller-provided list. * * The ND6 lock must be held. */ void nd6_prefix_unlink(struct nd_prefix *pr, struct nd_prhead *list) { ND6_WLOCK_ASSERT(); LIST_REMOVE(pr, ndpr_entry); V_nd6_list_genid++; if (list != NULL) LIST_INSERT_HEAD(list, pr, ndpr_entry); } /* * Free an unlinked prefix, first marking it off-link if necessary. */ void nd6_prefix_del(struct nd_prefix *pr) { struct nd_pfxrouter *pfr, *next; int e; char ip6buf[INET6_ADDRSTRLEN]; KASSERT(pr->ndpr_addrcnt == 0, ("prefix %p has referencing addresses", pr)); ND6_UNLOCK_ASSERT(); /* * Though these flags are now meaningless, we'd rather keep the value * of pr->ndpr_raf_onlink and pr->ndpr_raf_auto not to confuse users * when executing "ndp -p". */ if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) { ND6_ONLINK_LOCK(); if ((e = nd6_prefix_offlink(pr)) != 0) { nd6log((LOG_ERR, "nd6_prefix_del: failed to make %s/%d offlink " "on %s, errno=%d\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, if_name(pr->ndpr_ifp), e)); /* what should we do? */ } ND6_ONLINK_UNLOCK(); } /* Release references to routers that have advertised this prefix. */ ND6_WLOCK(); LIST_FOREACH_SAFE(pfr, &pr->ndpr_advrtrs, pfr_entry, next) pfxrtr_del(pfr); ND6_WUNLOCK(); nd6_prefix_rele(pr); pfxlist_onlink_check(); } static int prelist_update(struct nd_prefixctl *new, struct nd_defrouter *dr, struct mbuf *m, int mcast) { struct in6_ifaddr *ia6 = NULL, *ia6_match = NULL; struct ifaddr *ifa; struct ifnet *ifp = new->ndpr_ifp; struct nd_prefix *pr; int error = 0; int auth; struct in6_addrlifetime lt6_tmp; char ip6buf[INET6_ADDRSTRLEN]; auth = 0; if (m) { /* * Authenticity for NA consists authentication for * both IP header and IP datagrams, doesn't it ? */ #if defined(M_AUTHIPHDR) && defined(M_AUTHIPDGM) auth = ((m->m_flags & M_AUTHIPHDR) && (m->m_flags & M_AUTHIPDGM)); #endif } if ((pr = nd6_prefix_lookup(new)) != NULL) { /* * nd6_prefix_lookup() ensures that pr and new have the same * prefix on a same interface. */ /* * Update prefix information. Note that the on-link (L) bit * and the autonomous (A) bit should NOT be changed from 1 * to 0. */ if (new->ndpr_raf_onlink == 1) pr->ndpr_raf_onlink = 1; if (new->ndpr_raf_auto == 1) pr->ndpr_raf_auto = 1; if (new->ndpr_raf_onlink) { pr->ndpr_vltime = new->ndpr_vltime; pr->ndpr_pltime = new->ndpr_pltime; (void)in6_init_prefix_ltimes(pr); /* XXX error case? */ pr->ndpr_lastupdate = time_uptime; } if (new->ndpr_raf_onlink && (pr->ndpr_stateflags & NDPRF_ONLINK) == 0) { ND6_ONLINK_LOCK(); if ((error = nd6_prefix_onlink(pr)) != 0) { nd6log((LOG_ERR, "prelist_update: failed to make " "the prefix %s/%d on-link on %s " "(errno=%d)\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, if_name(pr->ndpr_ifp), error)); /* proceed anyway. XXX: is it correct? */ } ND6_ONLINK_UNLOCK(); } if (dr != NULL) pfxrtr_add(pr, dr); } else { if (new->ndpr_vltime == 0) goto end; if (new->ndpr_raf_onlink == 0 && new->ndpr_raf_auto == 0) goto end; error = nd6_prelist_add(new, dr, &pr); if (error != 0) { nd6log((LOG_NOTICE, "prelist_update: " "nd6_prelist_add failed for %s/%d on %s errno=%d\n", ip6_sprintf(ip6buf, &new->ndpr_prefix.sin6_addr), new->ndpr_plen, if_name(new->ndpr_ifp), error)); goto end; /* we should just give up in this case. */ } /* * XXX: from the ND point of view, we can ignore a prefix * with the on-link bit being zero. However, we need a * prefix structure for references from autoconfigured * addresses. Thus, we explicitly make sure that the prefix * itself expires now. */ if (pr->ndpr_raf_onlink == 0) { pr->ndpr_vltime = 0; pr->ndpr_pltime = 0; in6_init_prefix_ltimes(pr); } } /* * Address autoconfiguration based on Section 5.5.3 of RFC 2462. * Note that pr must be non NULL at this point. */ /* 5.5.3 (a). Ignore the prefix without the A bit set. */ if (!new->ndpr_raf_auto) goto end; /* * 5.5.3 (b). the link-local prefix should have been ignored in * nd6_ra_input. */ /* 5.5.3 (c). Consistency check on lifetimes: pltime <= vltime. */ if (new->ndpr_pltime > new->ndpr_vltime) { error = EINVAL; /* XXX: won't be used */ goto end; } /* * 5.5.3 (d). If the prefix advertised is not equal to the prefix of * an address configured by stateless autoconfiguration already in the * list of addresses associated with the interface, and the Valid * Lifetime is not 0, form an address. We first check if we have * a matching prefix. * Note: we apply a clarification in rfc2462bis-02 here. We only * consider autoconfigured addresses while RFC2462 simply said * "address". */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { struct in6_ifaddr *ifa6; u_int32_t remaininglifetime; if (ifa->ifa_addr->sa_family != AF_INET6) continue; ifa6 = (struct in6_ifaddr *)ifa; /* * We only consider autoconfigured addresses as per rfc2462bis. */ if (!(ifa6->ia6_flags & IN6_IFF_AUTOCONF)) continue; /* * Spec is not clear here, but I believe we should concentrate * on unicast (i.e. not anycast) addresses. * XXX: other ia6_flags? detached or duplicated? */ if ((ifa6->ia6_flags & IN6_IFF_ANYCAST) != 0) continue; /* * Ignore the address if it is not associated with a prefix * or is associated with a prefix that is different from this * one. (pr is never NULL here) */ if (ifa6->ia6_ndpr != pr) continue; if (ia6_match == NULL) /* remember the first one */ ia6_match = ifa6; /* * An already autoconfigured address matched. Now that we * are sure there is at least one matched address, we can * proceed to 5.5.3. (e): update the lifetimes according to the * "two hours" rule and the privacy extension. * We apply some clarifications in rfc2462bis: * - use remaininglifetime instead of storedlifetime as a * variable name * - remove the dead code in the "two-hour" rule */ #define TWOHOUR (120*60) lt6_tmp = ifa6->ia6_lifetime; if (lt6_tmp.ia6t_vltime == ND6_INFINITE_LIFETIME) remaininglifetime = ND6_INFINITE_LIFETIME; else if (time_uptime - ifa6->ia6_updatetime > lt6_tmp.ia6t_vltime) { /* * The case of "invalid" address. We should usually * not see this case. */ remaininglifetime = 0; } else remaininglifetime = lt6_tmp.ia6t_vltime - (time_uptime - ifa6->ia6_updatetime); /* when not updating, keep the current stored lifetime. */ lt6_tmp.ia6t_vltime = remaininglifetime; if (TWOHOUR < new->ndpr_vltime || remaininglifetime < new->ndpr_vltime) { lt6_tmp.ia6t_vltime = new->ndpr_vltime; } else if (remaininglifetime <= TWOHOUR) { if (auth) { lt6_tmp.ia6t_vltime = new->ndpr_vltime; } } else { /* * new->ndpr_vltime <= TWOHOUR && * TWOHOUR < remaininglifetime */ lt6_tmp.ia6t_vltime = TWOHOUR; } /* The 2 hour rule is not imposed for preferred lifetime. */ lt6_tmp.ia6t_pltime = new->ndpr_pltime; in6_init_address_ltimes(pr, <6_tmp); /* * We need to treat lifetimes for temporary addresses * differently, according to * draft-ietf-ipv6-privacy-addrs-v2-01.txt 3.3 (1); * we only update the lifetimes when they are in the maximum * intervals. */ if ((ifa6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { u_int32_t maxvltime, maxpltime; if (V_ip6_temp_valid_lifetime > (u_int32_t)((time_uptime - ifa6->ia6_createtime) + V_ip6_desync_factor)) { maxvltime = V_ip6_temp_valid_lifetime - (time_uptime - ifa6->ia6_createtime) - V_ip6_desync_factor; } else maxvltime = 0; if (V_ip6_temp_preferred_lifetime > (u_int32_t)((time_uptime - ifa6->ia6_createtime) + V_ip6_desync_factor)) { maxpltime = V_ip6_temp_preferred_lifetime - (time_uptime - ifa6->ia6_createtime) - V_ip6_desync_factor; } else maxpltime = 0; if (lt6_tmp.ia6t_vltime == ND6_INFINITE_LIFETIME || lt6_tmp.ia6t_vltime > maxvltime) { lt6_tmp.ia6t_vltime = maxvltime; } if (lt6_tmp.ia6t_pltime == ND6_INFINITE_LIFETIME || lt6_tmp.ia6t_pltime > maxpltime) { lt6_tmp.ia6t_pltime = maxpltime; } } ifa6->ia6_lifetime = lt6_tmp; ifa6->ia6_updatetime = time_uptime; } IF_ADDR_RUNLOCK(ifp); if (ia6_match == NULL && new->ndpr_vltime) { int ifidlen; /* * 5.5.3 (d) (continued) * No address matched and the valid lifetime is non-zero. * Create a new address. */ /* * Prefix Length check: * If the sum of the prefix length and interface identifier * length does not equal 128 bits, the Prefix Information * option MUST be ignored. The length of the interface * identifier is defined in a separate link-type specific * document. */ ifidlen = in6_if2idlen(ifp); if (ifidlen < 0) { /* this should not happen, so we always log it. */ log(LOG_ERR, "prelist_update: IFID undefined (%s)\n", if_name(ifp)); goto end; } if (ifidlen + pr->ndpr_plen != 128) { nd6log((LOG_INFO, "prelist_update: invalid prefixlen " "%d for %s, ignored\n", pr->ndpr_plen, if_name(ifp))); goto end; } if ((ia6 = in6_ifadd(new, mcast)) != NULL) { /* * note that we should use pr (not new) for reference. */ pr->ndpr_addrcnt++; ia6->ia6_ndpr = pr; /* * RFC 3041 3.3 (2). * When a new public address is created as described * in RFC2462, also create a new temporary address. * * RFC 3041 3.5. * When an interface connects to a new link, a new * randomized interface identifier should be generated * immediately together with a new set of temporary * addresses. Thus, we specifiy 1 as the 2nd arg of * in6_tmpifadd(). */ if (V_ip6_use_tempaddr) { int e; if ((e = in6_tmpifadd(ia6, 1, 1)) != 0) { nd6log((LOG_NOTICE, "prelist_update: " "failed to create a temporary " "address, errno=%d\n", e)); } } ifa_free(&ia6->ia_ifa); /* * A newly added address might affect the status * of other addresses, so we check and update it. * XXX: what if address duplication happens? */ pfxlist_onlink_check(); } else { /* just set an error. do not bark here. */ error = EADDRNOTAVAIL; /* XXX: might be unused. */ } } end: if (pr != NULL) nd6_prefix_rele(pr); return (error); } /* * A supplement function used in the on-link detection below; * detect if a given prefix has a (probably) reachable advertising router. * XXX: lengthy function name... */ static struct nd_pfxrouter * find_pfxlist_reachable_router(struct nd_prefix *pr) { struct nd_pfxrouter *pfxrtr; struct llentry *ln; int canreach; ND6_LOCK_ASSERT(); LIST_FOREACH(pfxrtr, &pr->ndpr_advrtrs, pfr_entry) { IF_AFDATA_RLOCK(pfxrtr->router->ifp); ln = nd6_lookup(&pfxrtr->router->rtaddr, 0, pfxrtr->router->ifp); IF_AFDATA_RUNLOCK(pfxrtr->router->ifp); if (ln == NULL) continue; canreach = ND6_IS_LLINFO_PROBREACH(ln); LLE_RUNLOCK(ln); if (canreach) break; } return (pfxrtr); } /* * Check if each prefix in the prefix list has at least one available router * that advertised the prefix (a router is "available" if its neighbor cache * entry is reachable or probably reachable). * If the check fails, the prefix may be off-link, because, for example, * we have moved from the network but the lifetime of the prefix has not * expired yet. So we should not use the prefix if there is another prefix * that has an available router. * But, if there is no prefix that has an available router, we still regard * all the prefixes as on-link. This is because we can't tell if all the * routers are simply dead or if we really moved from the network and there * is no router around us. */ void pfxlist_onlink_check(void) { struct nd_prefix *pr; struct in6_ifaddr *ifa; struct nd_defrouter *dr; struct nd_pfxrouter *pfxrtr = NULL; struct rm_priotracker in6_ifa_tracker; uint64_t genid; uint32_t flags; ND6_ONLINK_LOCK(); ND6_RLOCK(); /* * Check if there is a prefix that has a reachable advertising * router. */ LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { if (pr->ndpr_raf_onlink && find_pfxlist_reachable_router(pr)) break; } /* * If we have no such prefix, check whether we still have a router * that does not advertise any prefixes. */ if (pr == NULL) { TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { struct nd_prefix *pr0; LIST_FOREACH(pr0, &V_nd_prefix, ndpr_entry) { if ((pfxrtr = pfxrtr_lookup(pr0, dr)) != NULL) break; } if (pfxrtr != NULL) break; } } if (pr != NULL || (!TAILQ_EMPTY(&V_nd_defrouter) && pfxrtr == NULL)) { /* * There is at least one prefix that has a reachable router, * or at least a router which probably does not advertise * any prefixes. The latter would be the case when we move * to a new link where we have a router that does not provide * prefixes and we configure an address by hand. * Detach prefixes which have no reachable advertising * router, and attach other prefixes. */ LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { /* XXX: a link-local prefix should never be detached */ if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr) || pr->ndpr_raf_onlink == 0 || pr->ndpr_raf_auto == 0) continue; if ((pr->ndpr_stateflags & NDPRF_DETACHED) == 0 && find_pfxlist_reachable_router(pr) == NULL) pr->ndpr_stateflags |= NDPRF_DETACHED; else if ((pr->ndpr_stateflags & NDPRF_DETACHED) != 0 && find_pfxlist_reachable_router(pr) != NULL) pr->ndpr_stateflags &= ~NDPRF_DETACHED; } } else { /* there is no prefix that has a reachable router */ LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr) || pr->ndpr_raf_onlink == 0 || pr->ndpr_raf_auto == 0) continue; pr->ndpr_stateflags &= ~NDPRF_DETACHED; } } /* * Remove each interface route associated with a (just) detached * prefix, and reinstall the interface route for a (just) attached * prefix. Note that all attempt of reinstallation does not * necessarily success, when a same prefix is shared among multiple * interfaces. Such cases will be handled in nd6_prefix_onlink, * so we don't have to care about them. */ restart: LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { char ip6buf[INET6_ADDRSTRLEN]; int e; if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr) || pr->ndpr_raf_onlink == 0 || pr->ndpr_raf_auto == 0) continue; flags = pr->ndpr_stateflags & (NDPRF_DETACHED | NDPRF_ONLINK); if (flags == 0 || flags == (NDPRF_DETACHED | NDPRF_ONLINK)) { genid = V_nd6_list_genid; ND6_RUNLOCK(); if ((flags & NDPRF_ONLINK) != 0 && (e = nd6_prefix_offlink(pr)) != 0) { nd6log((LOG_ERR, "pfxlist_onlink_check: failed to " "make %s/%d offlink, errno=%d\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, e)); } else if ((flags & NDPRF_ONLINK) == 0 && (e = nd6_prefix_onlink(pr)) != 0) { nd6log((LOG_ERR, "pfxlist_onlink_check: failed to " "make %s/%d onlink, errno=%d\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, e)); } ND6_RLOCK(); if (genid != V_nd6_list_genid) goto restart; } } /* * Changes on the prefix status might affect address status as well. * Make sure that all addresses derived from an attached prefix are * attached, and that all addresses derived from a detached prefix are * detached. Note, however, that a manually configured address should * always be attached. * The precise detection logic is same as the one for prefixes. */ IN6_IFADDR_RLOCK(&in6_ifa_tracker); CK_STAILQ_FOREACH(ifa, &V_in6_ifaddrhead, ia_link) { if (!(ifa->ia6_flags & IN6_IFF_AUTOCONF)) continue; if (ifa->ia6_ndpr == NULL) { /* * This can happen when we first configure the address * (i.e. the address exists, but the prefix does not). * XXX: complicated relationships... */ continue; } if (find_pfxlist_reachable_router(ifa->ia6_ndpr)) break; } if (ifa) { CK_STAILQ_FOREACH(ifa, &V_in6_ifaddrhead, ia_link) { if ((ifa->ia6_flags & IN6_IFF_AUTOCONF) == 0) continue; if (ifa->ia6_ndpr == NULL) /* XXX: see above. */ continue; if (find_pfxlist_reachable_router(ifa->ia6_ndpr)) { if (ifa->ia6_flags & IN6_IFF_DETACHED) { ifa->ia6_flags &= ~IN6_IFF_DETACHED; ifa->ia6_flags |= IN6_IFF_TENTATIVE; nd6_dad_start((struct ifaddr *)ifa, 0); } } else { ifa->ia6_flags |= IN6_IFF_DETACHED; } } } else { CK_STAILQ_FOREACH(ifa, &V_in6_ifaddrhead, ia_link) { if ((ifa->ia6_flags & IN6_IFF_AUTOCONF) == 0) continue; if (ifa->ia6_flags & IN6_IFF_DETACHED) { ifa->ia6_flags &= ~IN6_IFF_DETACHED; ifa->ia6_flags |= IN6_IFF_TENTATIVE; /* Do we need a delay in this case? */ nd6_dad_start((struct ifaddr *)ifa, 0); } } } IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); ND6_RUNLOCK(); ND6_ONLINK_UNLOCK(); } static int nd6_prefix_onlink_rtrequest(struct nd_prefix *pr, struct ifaddr *ifa) { static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; struct rib_head *rnh; struct rtentry *rt; struct sockaddr_in6 mask6; u_long rtflags; int error, a_failure, fibnum, maxfib; /* * in6_ifinit() sets nd6_rtrequest to ifa_rtrequest for all ifaddrs. * ifa->ifa_rtrequest = nd6_rtrequest; */ bzero(&mask6, sizeof(mask6)); mask6.sin6_len = sizeof(mask6); mask6.sin6_addr = pr->ndpr_mask; rtflags = (ifa->ifa_flags & ~IFA_RTSELF) | RTF_UP; if(V_rt_add_addr_allfibs) { fibnum = 0; maxfib = rt_numfibs; } else { fibnum = ifa->ifa_ifp->if_fib; maxfib = fibnum + 1; } a_failure = 0; for (; fibnum < maxfib; fibnum++) { rt = NULL; error = in6_rtrequest(RTM_ADD, (struct sockaddr *)&pr->ndpr_prefix, ifa->ifa_addr, (struct sockaddr *)&mask6, rtflags, &rt, fibnum); if (error == 0) { KASSERT(rt != NULL, ("%s: in6_rtrequest return no " "error(%d) but rt is NULL, pr=%p, ifa=%p", __func__, error, pr, ifa)); rnh = rt_tables_get_rnh(rt->rt_fibnum, AF_INET6); /* XXX what if rhn == NULL? */ RIB_WLOCK(rnh); RT_LOCK(rt); if (rt_setgate(rt, rt_key(rt), (struct sockaddr *)&null_sdl) == 0) { struct sockaddr_dl *dl; dl = (struct sockaddr_dl *)rt->rt_gateway; dl->sdl_type = rt->rt_ifp->if_type; dl->sdl_index = rt->rt_ifp->if_index; } RIB_WUNLOCK(rnh); nd6_rtmsg(RTM_ADD, rt); RT_UNLOCK(rt); pr->ndpr_stateflags |= NDPRF_ONLINK; } else { char ip6buf[INET6_ADDRSTRLEN]; char ip6bufg[INET6_ADDRSTRLEN]; char ip6bufm[INET6_ADDRSTRLEN]; struct sockaddr_in6 *sin6; sin6 = (struct sockaddr_in6 *)ifa->ifa_addr; nd6log((LOG_ERR, "nd6_prefix_onlink: failed to add " "route for a prefix (%s/%d) on %s, gw=%s, mask=%s, " "flags=%lx errno = %d\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, if_name(pr->ndpr_ifp), ip6_sprintf(ip6bufg, &sin6->sin6_addr), ip6_sprintf(ip6bufm, &mask6.sin6_addr), rtflags, error)); /* Save last error to return, see rtinit(). */ a_failure = error; } if (rt != NULL) { RT_LOCK(rt); RT_REMREF(rt); RT_UNLOCK(rt); } } /* Return the last error we got. */ return (a_failure); } int nd6_prefix_onlink(struct nd_prefix *pr) { struct ifaddr *ifa; struct ifnet *ifp = pr->ndpr_ifp; struct nd_prefix *opr; char ip6buf[INET6_ADDRSTRLEN]; int error; ND6_ONLINK_LOCK_ASSERT(); ND6_UNLOCK_ASSERT(); if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) return (EEXIST); /* * Add the interface route associated with the prefix. Before * installing the route, check if there's the same prefix on another * interface, and the prefix has already installed the interface route. * Although such a configuration is expected to be rare, we explicitly * allow it. */ ND6_RLOCK(); LIST_FOREACH(opr, &V_nd_prefix, ndpr_entry) { if (opr == pr) continue; if ((opr->ndpr_stateflags & NDPRF_ONLINK) == 0) continue; if (!V_rt_add_addr_allfibs && opr->ndpr_ifp->if_fib != pr->ndpr_ifp->if_fib) continue; if (opr->ndpr_plen == pr->ndpr_plen && in6_are_prefix_equal(&pr->ndpr_prefix.sin6_addr, &opr->ndpr_prefix.sin6_addr, pr->ndpr_plen)) { ND6_RUNLOCK(); return (0); } } ND6_RUNLOCK(); /* * We prefer link-local addresses as the associated interface address. */ /* search for a link-local addr */ ifa = (struct ifaddr *)in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST); if (ifa == NULL) { /* XXX: freebsd does not have ifa_ifwithaf */ IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family == AF_INET6) { ifa_ref(ifa); break; } } IF_ADDR_RUNLOCK(ifp); /* should we care about ia6_flags? */ } if (ifa == NULL) { /* * This can still happen, when, for example, we receive an RA * containing a prefix with the L bit set and the A bit clear, * after removing all IPv6 addresses on the receiving * interface. This should, of course, be rare though. */ nd6log((LOG_NOTICE, "nd6_prefix_onlink: failed to find any ifaddr" " to add route for a prefix(%s/%d) on %s\n", ip6_sprintf(ip6buf, &pr->ndpr_prefix.sin6_addr), pr->ndpr_plen, if_name(ifp))); return (0); } error = nd6_prefix_onlink_rtrequest(pr, ifa); if (ifa != NULL) ifa_free(ifa); return (error); } int nd6_prefix_offlink(struct nd_prefix *pr) { int error = 0; struct ifnet *ifp = pr->ndpr_ifp; struct nd_prefix *opr; struct sockaddr_in6 sa6, mask6; struct rtentry *rt; char ip6buf[INET6_ADDRSTRLEN]; uint64_t genid; int fibnum, maxfib, a_failure; ND6_ONLINK_LOCK_ASSERT(); ND6_UNLOCK_ASSERT(); if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) return (EEXIST); bzero(&sa6, sizeof(sa6)); sa6.sin6_family = AF_INET6; sa6.sin6_len = sizeof(sa6); bcopy(&pr->ndpr_prefix.sin6_addr, &sa6.sin6_addr, sizeof(struct in6_addr)); bzero(&mask6, sizeof(mask6)); mask6.sin6_family = AF_INET6; mask6.sin6_len = sizeof(sa6); bcopy(&pr->ndpr_mask, &mask6.sin6_addr, sizeof(struct in6_addr)); if (V_rt_add_addr_allfibs) { fibnum = 0; maxfib = rt_numfibs; } else { fibnum = ifp->if_fib; maxfib = fibnum + 1; } a_failure = 0; for (; fibnum < maxfib; fibnum++) { rt = NULL; error = in6_rtrequest(RTM_DELETE, (struct sockaddr *)&sa6, NULL, (struct sockaddr *)&mask6, 0, &rt, fibnum); if (error == 0) { /* report the route deletion to the routing socket. */ if (rt != NULL) nd6_rtmsg(RTM_DELETE, rt); } else { /* Save last error to return, see rtinit(). */ a_failure = error; } if (rt != NULL) { RTFREE(rt); } } error = a_failure; a_failure = 1; if (error == 0) { pr->ndpr_stateflags &= ~NDPRF_ONLINK; /* * There might be the same prefix on another interface, * the prefix which could not be on-link just because we have * the interface route (see comments in nd6_prefix_onlink). * If there's one, try to make the prefix on-link on the * interface. */ ND6_RLOCK(); restart: LIST_FOREACH(opr, &V_nd_prefix, ndpr_entry) { /* * KAME specific: detached prefixes should not be * on-link. */ if (opr == pr || (opr->ndpr_stateflags & (NDPRF_ONLINK | NDPRF_DETACHED)) != 0) continue; if (opr->ndpr_plen == pr->ndpr_plen && in6_are_prefix_equal(&pr->ndpr_prefix.sin6_addr, &opr->ndpr_prefix.sin6_addr, pr->ndpr_plen)) { int e; genid = V_nd6_list_genid; ND6_RUNLOCK(); if ((e = nd6_prefix_onlink(opr)) != 0) { nd6log((LOG_ERR, "nd6_prefix_offlink: failed to " "recover a prefix %s/%d from %s " "to %s (errno = %d)\n", ip6_sprintf(ip6buf, &opr->ndpr_prefix.sin6_addr), opr->ndpr_plen, if_name(ifp), if_name(opr->ndpr_ifp), e)); } else a_failure = 0; ND6_RLOCK(); if (genid != V_nd6_list_genid) goto restart; } } ND6_RUNLOCK(); } else { /* XXX: can we still set the NDPRF_ONLINK flag? */ nd6log((LOG_ERR, "nd6_prefix_offlink: failed to delete route: " "%s/%d on %s (errno = %d)\n", ip6_sprintf(ip6buf, &sa6.sin6_addr), pr->ndpr_plen, if_name(ifp), error)); } if (a_failure) lltable_prefix_free(AF_INET6, (struct sockaddr *)&sa6, (struct sockaddr *)&mask6, LLE_STATIC); return (error); } static struct in6_ifaddr * in6_ifadd(struct nd_prefixctl *pr, int mcast) { struct ifnet *ifp = pr->ndpr_ifp; struct ifaddr *ifa; struct in6_aliasreq ifra; struct in6_ifaddr *ia, *ib; int error, plen0; struct in6_addr mask; int prefixlen = pr->ndpr_plen; int updateflags; char ip6buf[INET6_ADDRSTRLEN]; in6_prefixlen2mask(&mask, prefixlen); /* * find a link-local address (will be interface ID). * Is it really mandatory? Theoretically, a global or a site-local * address can be configured without a link-local address, if we * have a unique interface identifier... * * it is not mandatory to have a link-local address, we can generate * interface identifier on the fly. we do this because: * (1) it should be the easiest way to find interface identifier. * (2) RFC2462 5.4 suggesting the use of the same interface identifier * for multiple addresses on a single interface, and possible shortcut * of DAD. we omitted DAD for this reason in the past. * (3) a user can prevent autoconfiguration of global address * by removing link-local address by hand (this is partly because we * don't have other way to control the use of IPv6 on an interface. * this has been our design choice - cf. NRL's "ifconfig auto"). * (4) it is easier to manage when an interface has addresses * with the same interface identifier, than to have multiple addresses * with different interface identifiers. */ ifa = (struct ifaddr *)in6ifa_ifpforlinklocal(ifp, 0); /* 0 is OK? */ if (ifa) ib = (struct in6_ifaddr *)ifa; else return NULL; /* prefixlen + ifidlen must be equal to 128 */ plen0 = in6_mask2len(&ib->ia_prefixmask.sin6_addr, NULL); if (prefixlen != plen0) { ifa_free(ifa); nd6log((LOG_INFO, "in6_ifadd: wrong prefixlen for %s " "(prefix=%d ifid=%d)\n", if_name(ifp), prefixlen, 128 - plen0)); return NULL; } /* make ifaddr */ in6_prepare_ifra(&ifra, &pr->ndpr_prefix.sin6_addr, &mask); IN6_MASK_ADDR(&ifra.ifra_addr.sin6_addr, &mask); /* interface ID */ ifra.ifra_addr.sin6_addr.s6_addr32[0] |= (ib->ia_addr.sin6_addr.s6_addr32[0] & ~mask.s6_addr32[0]); ifra.ifra_addr.sin6_addr.s6_addr32[1] |= (ib->ia_addr.sin6_addr.s6_addr32[1] & ~mask.s6_addr32[1]); ifra.ifra_addr.sin6_addr.s6_addr32[2] |= (ib->ia_addr.sin6_addr.s6_addr32[2] & ~mask.s6_addr32[2]); ifra.ifra_addr.sin6_addr.s6_addr32[3] |= (ib->ia_addr.sin6_addr.s6_addr32[3] & ~mask.s6_addr32[3]); ifa_free(ifa); /* lifetimes. */ ifra.ifra_lifetime.ia6t_vltime = pr->ndpr_vltime; ifra.ifra_lifetime.ia6t_pltime = pr->ndpr_pltime; /* XXX: scope zone ID? */ ifra.ifra_flags |= IN6_IFF_AUTOCONF; /* obey autoconf */ /* * Make sure that we do not have this address already. This should * usually not happen, but we can still see this case, e.g., if we * have manually configured the exact address to be configured. */ ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, &ifra.ifra_addr.sin6_addr); if (ifa != NULL) { ifa_free(ifa); /* this should be rare enough to make an explicit log */ log(LOG_INFO, "in6_ifadd: %s is already configured\n", ip6_sprintf(ip6buf, &ifra.ifra_addr.sin6_addr)); return (NULL); } /* * Allocate ifaddr structure, link into chain, etc. * If we are going to create a new address upon receiving a multicasted * RA, we need to impose a random delay before starting DAD. * [draft-ietf-ipv6-rfc2462bis-02.txt, Section 5.4.2] */ updateflags = 0; if (mcast) updateflags |= IN6_IFAUPDATE_DADDELAY; if ((error = in6_update_ifa(ifp, &ifra, NULL, updateflags)) != 0) { nd6log((LOG_ERR, "in6_ifadd: failed to make ifaddr %s on %s (errno=%d)\n", ip6_sprintf(ip6buf, &ifra.ifra_addr.sin6_addr), if_name(ifp), error)); return (NULL); /* ifaddr must not have been allocated. */ } ia = in6ifa_ifpwithaddr(ifp, &ifra.ifra_addr.sin6_addr); /* * XXXRW: Assumption of non-NULLness here might not be true with * fine-grained locking -- should we validate it? Or just return * earlier ifa rather than looking it up again? */ return (ia); /* this is always non-NULL and referenced. */ } /* * ia0 - corresponding public address */ int in6_tmpifadd(const struct in6_ifaddr *ia0, int forcegen, int delay) { struct ifnet *ifp = ia0->ia_ifa.ifa_ifp; struct in6_ifaddr *newia; struct in6_aliasreq ifra; int error; int trylimit = 3; /* XXX: adhoc value */ int updateflags; u_int32_t randid[2]; time_t vltime0, pltime0; in6_prepare_ifra(&ifra, &ia0->ia_addr.sin6_addr, &ia0->ia_prefixmask.sin6_addr); ifra.ifra_addr = ia0->ia_addr; /* XXX: do we need this ? */ /* clear the old IFID */ IN6_MASK_ADDR(&ifra.ifra_addr.sin6_addr, &ifra.ifra_prefixmask.sin6_addr); again: if (in6_get_tmpifid(ifp, (u_int8_t *)randid, (const u_int8_t *)&ia0->ia_addr.sin6_addr.s6_addr[8], forcegen)) { nd6log((LOG_NOTICE, "in6_tmpifadd: failed to find a good " "random IFID\n")); return (EINVAL); } ifra.ifra_addr.sin6_addr.s6_addr32[2] |= (randid[0] & ~(ifra.ifra_prefixmask.sin6_addr.s6_addr32[2])); ifra.ifra_addr.sin6_addr.s6_addr32[3] |= (randid[1] & ~(ifra.ifra_prefixmask.sin6_addr.s6_addr32[3])); /* * in6_get_tmpifid() quite likely provided a unique interface ID. * However, we may still have a chance to see collision, because * there may be a time lag between generation of the ID and generation * of the address. So, we'll do one more sanity check. */ if (in6_localip(&ifra.ifra_addr.sin6_addr) != 0) { if (trylimit-- > 0) { forcegen = 1; goto again; } /* Give up. Something strange should have happened. */ nd6log((LOG_NOTICE, "in6_tmpifadd: failed to " "find a unique random IFID\n")); return (EEXIST); } /* * The Valid Lifetime is the lower of the Valid Lifetime of the * public address or TEMP_VALID_LIFETIME. * The Preferred Lifetime is the lower of the Preferred Lifetime * of the public address or TEMP_PREFERRED_LIFETIME - * DESYNC_FACTOR. */ if (ia0->ia6_lifetime.ia6t_vltime != ND6_INFINITE_LIFETIME) { vltime0 = IFA6_IS_INVALID(ia0) ? 0 : (ia0->ia6_lifetime.ia6t_vltime - (time_uptime - ia0->ia6_updatetime)); if (vltime0 > V_ip6_temp_valid_lifetime) vltime0 = V_ip6_temp_valid_lifetime; } else vltime0 = V_ip6_temp_valid_lifetime; if (ia0->ia6_lifetime.ia6t_pltime != ND6_INFINITE_LIFETIME) { pltime0 = IFA6_IS_DEPRECATED(ia0) ? 0 : (ia0->ia6_lifetime.ia6t_pltime - (time_uptime - ia0->ia6_updatetime)); if (pltime0 > V_ip6_temp_preferred_lifetime - V_ip6_desync_factor){ pltime0 = V_ip6_temp_preferred_lifetime - V_ip6_desync_factor; } } else pltime0 = V_ip6_temp_preferred_lifetime - V_ip6_desync_factor; ifra.ifra_lifetime.ia6t_vltime = vltime0; ifra.ifra_lifetime.ia6t_pltime = pltime0; /* * A temporary address is created only if this calculated Preferred * Lifetime is greater than REGEN_ADVANCE time units. */ if (ifra.ifra_lifetime.ia6t_pltime <= V_ip6_temp_regen_advance) return (0); /* XXX: scope zone ID? */ ifra.ifra_flags |= (IN6_IFF_AUTOCONF|IN6_IFF_TEMPORARY); /* allocate ifaddr structure, link into chain, etc. */ updateflags = 0; if (delay) updateflags |= IN6_IFAUPDATE_DADDELAY; if ((error = in6_update_ifa(ifp, &ifra, NULL, updateflags)) != 0) return (error); newia = in6ifa_ifpwithaddr(ifp, &ifra.ifra_addr.sin6_addr); if (newia == NULL) { /* XXX: can it happen? */ nd6log((LOG_ERR, "in6_tmpifadd: ifa update succeeded, but we got " "no ifaddr\n")); return (EINVAL); /* XXX */ } newia->ia6_ndpr = ia0->ia6_ndpr; newia->ia6_ndpr->ndpr_addrcnt++; ifa_free(&newia->ia_ifa); /* * A newly added address might affect the status of other addresses. * XXX: when the temporary address is generated with a new public * address, the onlink check is redundant. However, it would be safe * to do the check explicitly everywhere a new address is generated, * and, in fact, we surely need the check when we create a new * temporary address due to deprecation of an old temporary address. */ pfxlist_onlink_check(); return (0); } static int in6_init_prefix_ltimes(struct nd_prefix *ndpr) { if (ndpr->ndpr_pltime == ND6_INFINITE_LIFETIME) ndpr->ndpr_preferred = 0; else ndpr->ndpr_preferred = time_uptime + ndpr->ndpr_pltime; if (ndpr->ndpr_vltime == ND6_INFINITE_LIFETIME) ndpr->ndpr_expire = 0; else ndpr->ndpr_expire = time_uptime + ndpr->ndpr_vltime; return 0; } static void in6_init_address_ltimes(struct nd_prefix *new, struct in6_addrlifetime *lt6) { /* init ia6t_expire */ if (lt6->ia6t_vltime == ND6_INFINITE_LIFETIME) lt6->ia6t_expire = 0; else { lt6->ia6t_expire = time_uptime; lt6->ia6t_expire += lt6->ia6t_vltime; } /* init ia6t_preferred */ if (lt6->ia6t_pltime == ND6_INFINITE_LIFETIME) lt6->ia6t_preferred = 0; else { lt6->ia6t_preferred = time_uptime; lt6->ia6t_preferred += lt6->ia6t_pltime; } } /* * Delete all the routing table entries that use the specified gateway. * XXX: this function causes search through all entries of routing table, so * it shouldn't be called when acting as a router. */ void rt6_flush(struct in6_addr *gateway, struct ifnet *ifp) { /* We'll care only link-local addresses */ if (!IN6_IS_ADDR_LINKLOCAL(gateway)) return; /* XXX Do we really need to walk any but the default FIB? */ rt_foreach_fib_walk_del(AF_INET6, rt6_deleteroute, (void *)gateway); } static int rt6_deleteroute(const struct rtentry *rt, void *arg) { #define SIN6(s) ((struct sockaddr_in6 *)s) struct in6_addr *gate = (struct in6_addr *)arg; if (rt->rt_gateway == NULL || rt->rt_gateway->sa_family != AF_INET6) return (0); if (!IN6_ARE_ADDR_EQUAL(gate, &SIN6(rt->rt_gateway)->sin6_addr)) { return (0); } /* * Do not delete a static route. * XXX: this seems to be a bit ad-hoc. Should we consider the * 'cloned' bit instead? */ if ((rt->rt_flags & RTF_STATIC) != 0) return (0); /* * We delete only host route. This means, in particular, we don't * delete default route. */ if ((rt->rt_flags & RTF_HOST) == 0) return (0); return (1); #undef SIN6 } int nd6_setdefaultiface(int ifindex) { int error = 0; if (ifindex < 0 || V_if_index < ifindex) return (EINVAL); if (ifindex != 0 && !ifnet_byindex(ifindex)) return (EINVAL); if (V_nd6_defifindex != ifindex) { V_nd6_defifindex = ifindex; if (V_nd6_defifindex > 0) V_nd6_defifp = ifnet_byindex(V_nd6_defifindex); else V_nd6_defifp = NULL; /* * Our current implementation assumes one-to-one maping between * interfaces and links, so it would be natural to use the * default interface as the default link. */ scope6_setdefault(V_nd6_defifp); } return (error); } Index: head/sys/netinet6/scope6.c =================================================================== --- head/sys/netinet6/scope6.c (revision 336675) +++ head/sys/netinet6/scope6.c (revision 336676) @@ -1,574 +1,574 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 2000 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: scope6.c,v 1.10 2000/07/24 13:29:31 itojun Exp $ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef ENABLE_DEFAULT_SCOPE VNET_DEFINE(int, ip6_use_defzone) = 1; #else VNET_DEFINE(int, ip6_use_defzone) = 0; #endif VNET_DEFINE(int, deembed_scopeid) = 1; SYSCTL_DECL(_net_inet6_ip6); SYSCTL_INT(_net_inet6_ip6, OID_AUTO, deembed_scopeid, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(deembed_scopeid), 0, "Extract embedded zone ID and set it to sin6_scope_id in sockaddr_in6."); /* * The scope6_lock protects the global sid default stored in * sid_default below. */ static struct mtx scope6_lock; #define SCOPE6_LOCK_INIT() mtx_init(&scope6_lock, "scope6_lock", NULL, MTX_DEF) #define SCOPE6_LOCK() mtx_lock(&scope6_lock) #define SCOPE6_UNLOCK() mtx_unlock(&scope6_lock) #define SCOPE6_LOCK_ASSERT() mtx_assert(&scope6_lock, MA_OWNED) -static VNET_DEFINE(struct scope6_id, sid_default); +VNET_DEFINE_STATIC(struct scope6_id, sid_default); #define V_sid_default VNET(sid_default) #define SID(ifp) \ (((struct in6_ifextra *)(ifp)->if_afdata[AF_INET6])->scope6_id) static int scope6_get(struct ifnet *, struct scope6_id *); static int scope6_set(struct ifnet *, struct scope6_id *); void scope6_init(void) { bzero(&V_sid_default, sizeof(V_sid_default)); if (!IS_DEFAULT_VNET(curvnet)) return; SCOPE6_LOCK_INIT(); } struct scope6_id * scope6_ifattach(struct ifnet *ifp) { struct scope6_id *sid; sid = malloc(sizeof(*sid), M_IFADDR, M_WAITOK | M_ZERO); /* * XXX: IPV6_ADDR_SCOPE_xxx macros are not standard. * Should we rather hardcode here? */ sid->s6id_list[IPV6_ADDR_SCOPE_INTFACELOCAL] = ifp->if_index; sid->s6id_list[IPV6_ADDR_SCOPE_LINKLOCAL] = ifp->if_index; return (sid); } void scope6_ifdetach(struct scope6_id *sid) { free(sid, M_IFADDR); } int scope6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) { struct in6_ifreq *ifr; if (ifp->if_afdata[AF_INET6] == NULL) return (EPFNOSUPPORT); ifr = (struct in6_ifreq *)data; switch (cmd) { case SIOCSSCOPE6: return (scope6_set(ifp, (struct scope6_id *)ifr->ifr_ifru.ifru_scope_id)); case SIOCGSCOPE6: return (scope6_get(ifp, (struct scope6_id *)ifr->ifr_ifru.ifru_scope_id)); case SIOCGSCOPE6DEF: return (scope6_get_default( (struct scope6_id *)ifr->ifr_ifru.ifru_scope_id)); default: return (EOPNOTSUPP); } } static int scope6_set(struct ifnet *ifp, struct scope6_id *idlist) { int i; int error = 0; struct scope6_id *sid = NULL; IF_AFDATA_WLOCK(ifp); sid = SID(ifp); if (!sid) { /* paranoid? */ IF_AFDATA_WUNLOCK(ifp); return (EINVAL); } /* * XXX: We need more consistency checks of the relationship among * scopes (e.g. an organization should be larger than a site). */ /* * TODO(XXX): after setting, we should reflect the changes to * interface addresses, routing table entries, PCB entries... */ for (i = 0; i < 16; i++) { if (idlist->s6id_list[i] && idlist->s6id_list[i] != sid->s6id_list[i]) { /* * An interface zone ID must be the corresponding * interface index by definition. */ if (i == IPV6_ADDR_SCOPE_INTFACELOCAL && idlist->s6id_list[i] != ifp->if_index) { IF_AFDATA_WUNLOCK(ifp); return (EINVAL); } if (i == IPV6_ADDR_SCOPE_LINKLOCAL && idlist->s6id_list[i] > V_if_index) { /* * XXX: theoretically, there should be no * relationship between link IDs and interface * IDs, but we check the consistency for * safety in later use. */ IF_AFDATA_WUNLOCK(ifp); return (EINVAL); } /* * XXX: we must need lots of work in this case, * but we simply set the new value in this initial * implementation. */ sid->s6id_list[i] = idlist->s6id_list[i]; } } IF_AFDATA_WUNLOCK(ifp); return (error); } static int scope6_get(struct ifnet *ifp, struct scope6_id *idlist) { struct scope6_id *sid; /* We only need to lock the interface's afdata for SID() to work. */ IF_AFDATA_RLOCK(ifp); sid = SID(ifp); if (sid == NULL) { /* paranoid? */ IF_AFDATA_RUNLOCK(ifp); return (EINVAL); } *idlist = *sid; IF_AFDATA_RUNLOCK(ifp); return (0); } /* * Get a scope of the address. Node-local, link-local, site-local or global. */ int in6_addrscope(const struct in6_addr *addr) { if (IN6_IS_ADDR_MULTICAST(addr)) { /* * Addresses with reserved value F must be treated as * global multicast addresses. */ if (IPV6_ADDR_MC_SCOPE(addr) == 0x0f) return (IPV6_ADDR_SCOPE_GLOBAL); return (IPV6_ADDR_MC_SCOPE(addr)); } if (IN6_IS_ADDR_LINKLOCAL(addr) || IN6_IS_ADDR_LOOPBACK(addr)) return (IPV6_ADDR_SCOPE_LINKLOCAL); if (IN6_IS_ADDR_SITELOCAL(addr)) return (IPV6_ADDR_SCOPE_SITELOCAL); return (IPV6_ADDR_SCOPE_GLOBAL); } /* * ifp - note that this might be NULL */ void scope6_setdefault(struct ifnet *ifp) { /* * Currently, this function just sets the default "interfaces" * and "links" according to the given interface. * We might eventually have to separate the notion of "link" from * "interface" and provide a user interface to set the default. */ SCOPE6_LOCK(); if (ifp) { V_sid_default.s6id_list[IPV6_ADDR_SCOPE_INTFACELOCAL] = ifp->if_index; V_sid_default.s6id_list[IPV6_ADDR_SCOPE_LINKLOCAL] = ifp->if_index; } else { V_sid_default.s6id_list[IPV6_ADDR_SCOPE_INTFACELOCAL] = 0; V_sid_default.s6id_list[IPV6_ADDR_SCOPE_LINKLOCAL] = 0; } SCOPE6_UNLOCK(); } int scope6_get_default(struct scope6_id *idlist) { SCOPE6_LOCK(); *idlist = V_sid_default; SCOPE6_UNLOCK(); return (0); } u_int32_t scope6_addr2default(struct in6_addr *addr) { u_int32_t id; /* * special case: The loopback address should be considered as * link-local, but there's no ambiguity in the syntax. */ if (IN6_IS_ADDR_LOOPBACK(addr)) return (0); /* * XXX: 32-bit read is atomic on all our platforms, is it OK * not to lock here? */ SCOPE6_LOCK(); id = V_sid_default.s6id_list[in6_addrscope(addr)]; SCOPE6_UNLOCK(); return (id); } /* * Validate the specified scope zone ID in the sin6_scope_id field. If the ID * is unspecified (=0), needs to be specified, and the default zone ID can be * used, the default value will be used. * This routine then generates the kernel-internal form: if the address scope * of is interface-local or link-local, embed the interface index in the * address. */ int sa6_embedscope(struct sockaddr_in6 *sin6, int defaultok) { u_int32_t zoneid; if ((zoneid = sin6->sin6_scope_id) == 0 && defaultok) zoneid = scope6_addr2default(&sin6->sin6_addr); if (zoneid != 0 && (IN6_IS_SCOPE_LINKLOCAL(&sin6->sin6_addr) || IN6_IS_ADDR_MC_INTFACELOCAL(&sin6->sin6_addr))) { /* * At this moment, we only check interface-local and * link-local scope IDs, and use interface indices as the * zone IDs assuming a one-to-one mapping between interfaces * and links. */ if (V_if_index < zoneid || ifnet_byindex(zoneid) == NULL) return (ENXIO); /* XXX assignment to 16bit from 32bit variable */ sin6->sin6_addr.s6_addr16[1] = htons(zoneid & 0xffff); sin6->sin6_scope_id = 0; } return 0; } /* * generate standard sockaddr_in6 from embedded form. */ int sa6_recoverscope(struct sockaddr_in6 *sin6) { char ip6buf[INET6_ADDRSTRLEN]; u_int32_t zoneid; if (IN6_IS_SCOPE_LINKLOCAL(&sin6->sin6_addr) || IN6_IS_ADDR_MC_INTFACELOCAL(&sin6->sin6_addr)) { /* * KAME assumption: link id == interface id */ zoneid = ntohs(sin6->sin6_addr.s6_addr16[1]); if (zoneid) { /* sanity check */ if (V_if_index < zoneid) return (ENXIO); #if 0 /* XXX: Disabled due to possible deadlock. */ if (!ifnet_byindex(zoneid)) return (ENXIO); #endif if (sin6->sin6_scope_id != 0 && zoneid != sin6->sin6_scope_id) { log(LOG_NOTICE, "%s: embedded scope mismatch: %s%%%d. " "sin6_scope_id was overridden\n", __func__, ip6_sprintf(ip6buf, &sin6->sin6_addr), sin6->sin6_scope_id); } sin6->sin6_addr.s6_addr16[1] = 0; sin6->sin6_scope_id = zoneid; } } return 0; } /* * Determine the appropriate scope zone ID for in6 and ifp. If ret_id is * non NULL, it is set to the zone ID. If the zone ID needs to be embedded * in the in6_addr structure, in6 will be modified. * * ret_id - unnecessary? */ int in6_setscope(struct in6_addr *in6, struct ifnet *ifp, u_int32_t *ret_id) { int scope; u_int32_t zoneid = 0; struct scope6_id *sid; /* * special case: the loopback address can only belong to a loopback * interface. */ if (IN6_IS_ADDR_LOOPBACK(in6)) { if (!(ifp->if_flags & IFF_LOOPBACK)) return (EINVAL); } else { scope = in6_addrscope(in6); if (scope == IPV6_ADDR_SCOPE_INTFACELOCAL || scope == IPV6_ADDR_SCOPE_LINKLOCAL) { /* * Currently we use interface indices as the * zone IDs for interface-local and link-local * scopes. */ zoneid = ifp->if_index; in6->s6_addr16[1] = htons(zoneid & 0xffff); /* XXX */ } else if (scope != IPV6_ADDR_SCOPE_GLOBAL) { IF_AFDATA_RLOCK(ifp); sid = SID(ifp); zoneid = sid->s6id_list[scope]; IF_AFDATA_RUNLOCK(ifp); } } if (ret_id != NULL) *ret_id = zoneid; return (0); } /* * Just clear the embedded scope identifier. Return 0 if the original address * is intact; return non 0 if the address is modified. */ int in6_clearscope(struct in6_addr *in6) { int modified = 0; if (IN6_IS_SCOPE_LINKLOCAL(in6) || IN6_IS_ADDR_MC_INTFACELOCAL(in6)) { if (in6->s6_addr16[1] != 0) modified = 1; in6->s6_addr16[1] = 0; } return (modified); } /* * Return the scope identifier or zero. */ uint16_t in6_getscope(const struct in6_addr *in6) { if (IN6_IS_SCOPE_LINKLOCAL(in6) || IN6_IS_ADDR_MC_INTFACELOCAL(in6)) return (in6->s6_addr16[1]); return (0); } /* * Return pointer to ifnet structure, corresponding to the zone id of * link-local scope. */ struct ifnet* in6_getlinkifnet(uint32_t zoneid) { return (ifnet_byindex((u_short)zoneid)); } /* * Return zone id for the specified scope. */ uint32_t in6_getscopezone(const struct ifnet *ifp, int scope) { if (scope == IPV6_ADDR_SCOPE_INTFACELOCAL || scope == IPV6_ADDR_SCOPE_LINKLOCAL) return (ifp->if_index); if (scope >= 0 && scope < IPV6_ADDR_SCOPES_COUNT) return (SID(ifp)->s6id_list[scope]); return (0); } /* * Extracts scope from adddress @dst, stores cleared address * inside @dst and zone inside @scopeid */ void in6_splitscope(const struct in6_addr *src, struct in6_addr *dst, uint32_t *scopeid) { uint32_t zoneid; *dst = *src; zoneid = ntohs(in6_getscope(dst)); in6_clearscope(dst); *scopeid = zoneid; } /* * This function is for checking sockaddr_in6 structure passed * from the application level (usually). * * sin6_scope_id should be set for link-local unicast, link-local and * interface-local multicast addresses. * * If it is zero, then look into default zone ids. If default zone id is * not set or disabled, then return error. */ int sa6_checkzone(struct sockaddr_in6 *sa6) { int scope; scope = in6_addrscope(&sa6->sin6_addr); if (scope == IPV6_ADDR_SCOPE_GLOBAL) return (sa6->sin6_scope_id ? EINVAL: 0); if (IN6_IS_ADDR_MULTICAST(&sa6->sin6_addr) && scope != IPV6_ADDR_SCOPE_LINKLOCAL && scope != IPV6_ADDR_SCOPE_INTFACELOCAL) { if (sa6->sin6_scope_id == 0 && V_ip6_use_defzone != 0) sa6->sin6_scope_id = V_sid_default.s6id_list[scope]; return (0); } /* * Since ::1 address always configured on the lo0, we can * automatically set its zone id, when it is not specified. * Return error, when specified zone id doesn't match with * actual value. */ if (IN6_IS_ADDR_LOOPBACK(&sa6->sin6_addr)) { if (sa6->sin6_scope_id == 0) sa6->sin6_scope_id = in6_getscopezone(V_loif, scope); else if (sa6->sin6_scope_id != in6_getscopezone(V_loif, scope)) return (EADDRNOTAVAIL); } /* XXX: we can validate sin6_scope_id here */ if (sa6->sin6_scope_id != 0) return (0); if (V_ip6_use_defzone != 0) sa6->sin6_scope_id = V_sid_default.s6id_list[scope]; /* Return error if we can't determine zone id */ return (sa6->sin6_scope_id ? 0: EADDRNOTAVAIL); } /* * This function is similar to sa6_checkzone, but it uses given ifp * to initialize sin6_scope_id. */ int sa6_checkzone_ifp(struct ifnet *ifp, struct sockaddr_in6 *sa6) { int scope; scope = in6_addrscope(&sa6->sin6_addr); if (scope == IPV6_ADDR_SCOPE_LINKLOCAL || scope == IPV6_ADDR_SCOPE_INTFACELOCAL) { if (sa6->sin6_scope_id == 0) { sa6->sin6_scope_id = in6_getscopezone(ifp, scope); return (0); } else if (sa6->sin6_scope_id != in6_getscopezone(ifp, scope)) return (EADDRNOTAVAIL); } return (sa6_checkzone(sa6)); } Index: head/sys/netinet6/send.c =================================================================== --- head/sys/netinet6/send.c (revision 336675) +++ head/sys/netinet6/send.c (revision 336676) @@ -1,376 +1,376 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2009-2010 Ana Kukec * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_SEND, "send", "Secure Neighbour Discovery"); /* * The socket used to communicate with the SeND daemon. */ -static VNET_DEFINE(struct socket *, send_so); +VNET_DEFINE_STATIC(struct socket *, send_so); #define V_send_so VNET(send_so) u_long send_sendspace = 8 * (1024 + sizeof(struct sockaddr_send)); u_long send_recvspace = 9216; struct mtx send_mtx; #define SEND_LOCK_INIT() mtx_init(&send_mtx, "send_mtx", NULL, MTX_DEF) #define SEND_LOCK() mtx_lock(&send_mtx) #define SEND_UNLOCK() mtx_unlock(&send_mtx) #define SEND_LOCK_DESTROY() mtx_destroy(&send_mtx) static int send_attach(struct socket *so, int proto, struct thread *td) { int error; SEND_LOCK(); if (V_send_so != NULL) { SEND_UNLOCK(); return (EEXIST); } error = priv_check(td, PRIV_NETINET_RAW); if (error) { SEND_UNLOCK(); return(error); } if (proto != IPPROTO_SEND) { SEND_UNLOCK(); return (EPROTONOSUPPORT); } error = soreserve(so, send_sendspace, send_recvspace); if (error) { SEND_UNLOCK(); return(error); } V_send_so = so; SEND_UNLOCK(); return (0); } static int send_output(struct mbuf *m, struct ifnet *ifp, int direction) { struct ip6_hdr *ip6; struct sockaddr_in6 dst; struct icmp6_hdr *icmp6; int icmp6len; /* * Receive incoming (SeND-protected) or outgoing traffic * (SeND-validated) from the SeND user space application. */ switch (direction) { case SND_IN: if (m->m_len < (sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr))) { m = m_pullup(m, sizeof(struct ip6_hdr) + sizeof(struct icmp6_hdr)); if (!m) return (ENOBUFS); } /* Before passing off the mbuf record the proper interface. */ m->m_pkthdr.rcvif = ifp; if (m->m_flags & M_PKTHDR) icmp6len = m->m_pkthdr.len - sizeof(struct ip6_hdr); else panic("Doh! not the first mbuf."); ip6 = mtod(m, struct ip6_hdr *); icmp6 = (struct icmp6_hdr *)(ip6 + 1); /* * Output the packet as icmp6.c:icpm6_input() would do. * The mbuf is always consumed, so we do not have to * care about that. */ switch (icmp6->icmp6_type) { case ND_NEIGHBOR_SOLICIT: nd6_ns_input(m, sizeof(struct ip6_hdr), icmp6len); break; case ND_NEIGHBOR_ADVERT: nd6_na_input(m, sizeof(struct ip6_hdr), icmp6len); break; case ND_REDIRECT: icmp6_redirect_input(m, sizeof(struct ip6_hdr)); break; case ND_ROUTER_SOLICIT: nd6_rs_input(m, sizeof(struct ip6_hdr), icmp6len); break; case ND_ROUTER_ADVERT: nd6_ra_input(m, sizeof(struct ip6_hdr), icmp6len); break; default: m_freem(m); return (ENOSYS); } return (0); case SND_OUT: if (m->m_len < sizeof(struct ip6_hdr)) { m = m_pullup(m, sizeof(struct ip6_hdr)); if (!m) return (ENOBUFS); } ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) m->m_flags |= M_MCAST; bzero(&dst, sizeof(dst)); dst.sin6_family = AF_INET6; dst.sin6_len = sizeof(dst); dst.sin6_addr = ip6->ip6_dst; m_clrprotoflags(m); /* Avoid confusing lower layers. */ IP_PROBE(send, NULL, NULL, ip6, ifp, NULL, ip6); /* * Output the packet as nd6.c:nd6_output_lle() would do. * The mbuf is always consumed, so we do not have to care * about that. * XXX-BZ as we added data, what about fragmenting, * if now needed? */ int error; error = ((*ifp->if_output)(ifp, m, (struct sockaddr *)&dst, NULL)); if (error) error = ENOENT; return (error); default: panic("%s: direction %d neither SND_IN nor SND_OUT.", __func__, direction); } } /* * Receive a SeND message from user space to be either send out by the kernel * or, with SeND ICMPv6 options removed, to be further processed by the icmp6 * input path. */ static int send_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { struct sockaddr_send *sendsrc; struct ifnet *ifp; int error; KASSERT(V_send_so == so, ("%s: socket %p not send socket %p", __func__, so, V_send_so)); sendsrc = (struct sockaddr_send *)nam; ifp = ifnet_byindex_ref(sendsrc->send_ifidx); if (ifp == NULL) { error = ENETUNREACH; goto err; } error = send_output(m, ifp, sendsrc->send_direction); if_rele(ifp); m = NULL; err: if (m != NULL) m_freem(m); return (error); } static void send_close(struct socket *so) { SEND_LOCK(); if (V_send_so) V_send_so = NULL; SEND_UNLOCK(); } /* * Send a SeND message to user space, that was either received and has to be * validated or was about to be send out and has to be handled by the SEND * daemon adding SeND ICMPv6 options. */ static int send_input(struct mbuf *m, struct ifnet *ifp, int direction, int msglen __unused) { struct ip6_hdr *ip6; struct sockaddr_send sendsrc; SEND_LOCK(); if (V_send_so == NULL) { SEND_UNLOCK(); return (-1); } /* * Make sure to clear any possible internally embedded scope before * passing the packet to user space for SeND cryptographic signature * validation to succeed. */ ip6 = mtod(m, struct ip6_hdr *); in6_clearscope(&ip6->ip6_src); in6_clearscope(&ip6->ip6_dst); bzero(&sendsrc, sizeof(sendsrc)); sendsrc.send_len = sizeof(sendsrc); sendsrc.send_family = AF_INET6; sendsrc.send_direction = direction; sendsrc.send_ifidx = ifp->if_index; /* * Send incoming or outgoing traffic to user space either to be * protected (outgoing) or validated (incoming) according to rfc3971. */ SOCKBUF_LOCK(&V_send_so->so_rcv); if (sbappendaddr_locked(&V_send_so->so_rcv, (struct sockaddr *)&sendsrc, m, NULL) == 0) { SOCKBUF_UNLOCK(&V_send_so->so_rcv); /* XXX stats. */ m_freem(m); } else { sorwakeup_locked(V_send_so); } SEND_UNLOCK(); return (0); } struct pr_usrreqs send_usrreqs = { .pru_attach = send_attach, .pru_send = send_send, .pru_detach = send_close }; struct protosw send_protosw = { .pr_type = SOCK_RAW, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_protocol = IPPROTO_SEND, .pr_usrreqs = &send_usrreqs }; static int send_modevent(module_t mod, int type, void *unused) { #ifdef __notyet__ VNET_ITERATOR_DECL(vnet_iter); #endif int error; switch (type) { case MOD_LOAD: SEND_LOCK_INIT(); error = pf_proto_register(PF_INET6, &send_protosw); if (error != 0) { printf("%s:%d: MOD_LOAD pf_proto_register(): %d\n", __func__, __LINE__, error); SEND_LOCK_DESTROY(); break; } send_sendso_input_hook = send_input; break; case MOD_UNLOAD: /* Do not allow unloading w/o locking. */ return (EBUSY); #ifdef __notyet__ VNET_LIST_RLOCK_NOSLEEP(); SEND_LOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); if (V_send_so != NULL) { CURVNET_RESTORE(); SEND_UNLOCK(); VNET_LIST_RUNLOCK_NOSLEEP(); return (EBUSY); } CURVNET_RESTORE(); } SEND_UNLOCK(); VNET_LIST_RUNLOCK_NOSLEEP(); error = pf_proto_unregister(PF_INET6, IPPROTO_SEND, SOCK_RAW); if (error == 0) SEND_LOCK_DESTROY(); send_sendso_input_hook = NULL; break; #endif default: error = 0; break; } return (error); } static moduledata_t sendmod = { "send", send_modevent, 0 }; DECLARE_MODULE(send, sendmod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); Index: head/sys/netipsec/ipsec.c =================================================================== --- head/sys/netipsec/ipsec.c (revision 336675) +++ head/sys/netipsec/ipsec.c (revision 336676) @@ -1,1405 +1,1405 @@ /* $FreeBSD$ */ /* $KAME: ipsec.c,v 1.103 2001/05/24 07:14:18 sakane Exp $ */ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * IPsec controller part. */ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #ifdef INET6 #include #endif #include #include #ifdef INET6 #include #endif #include #include #include /*XXX*/ #include #include #include #include #include #include #include #include /* NB: name changed so netstat doesn't use it. */ VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec4stat); VNET_PCPUSTAT_SYSINIT(ipsec4stat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(ipsec4stat); #endif /* VIMAGE */ /* DF bit on encap. 0: clear 1: set 2: copy */ VNET_DEFINE(int, ip4_ipsec_dfbit) = 0; VNET_DEFINE(int, ip4_esp_trans_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip4_esp_net_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip4_ah_trans_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip4_ah_net_deflev) = IPSEC_LEVEL_USE; /* ECN ignore(-1)/forbidden(0)/allowed(1) */ VNET_DEFINE(int, ip4_ipsec_ecn) = 0; -static VNET_DEFINE(int, ip4_filtertunnel) = 0; +VNET_DEFINE_STATIC(int, ip4_filtertunnel) = 0; #define V_ip4_filtertunnel VNET(ip4_filtertunnel) -static VNET_DEFINE(int, check_policy_history) = 0; +VNET_DEFINE_STATIC(int, check_policy_history) = 0; #define V_check_policy_history VNET(check_policy_history) -static VNET_DEFINE(struct secpolicy *, def_policy) = NULL; +VNET_DEFINE_STATIC(struct secpolicy *, def_policy) = NULL; #define V_def_policy VNET(def_policy) static int sysctl_def_policy(SYSCTL_HANDLER_ARGS) { int error, value; value = V_def_policy->policy; error = sysctl_handle_int(oidp, &value, 0, req); if (error == 0) { if (value != IPSEC_POLICY_DISCARD && value != IPSEC_POLICY_NONE) return (EINVAL); V_def_policy->policy = value; } return (error); } /* * Crypto support requirements: * * 1 require hardware support * -1 require software support * 0 take anything */ VNET_DEFINE(int, crypto_support) = CRYPTOCAP_F_HARDWARE | CRYPTOCAP_F_SOFTWARE; /* * Use asynchronous mode to parallelize crypto jobs: * * 0 - disabled * 1 - enabled */ VNET_DEFINE(int, async_crypto) = 0; /* * TCP/UDP checksum handling policy for transport mode NAT-T (RFC3948) * * 0 - auto: incrementally recompute, when checksum delta is known; * if checksum delta isn't known, reset checksum to zero for UDP, * and mark csum_flags as valid for TCP. * 1 - fully recompute TCP/UDP checksum. */ VNET_DEFINE(int, natt_cksum_policy) = 0; FEATURE(ipsec, "Internet Protocol Security (IPsec)"); FEATURE(ipsec_natt, "UDP Encapsulation of IPsec ESP Packets ('NAT-T')"); SYSCTL_DECL(_net_inet_ipsec); /* net.inet.ipsec */ SYSCTL_PROC(_net_inet_ipsec, IPSECCTL_DEF_POLICY, def_policy, CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I", "IPsec default policy."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_trans_deflev), 0, "Default ESP transport mode level"); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_net_deflev), 0, "Default ESP tunnel mode level."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_trans_deflev), 0, "AH transfer mode default level."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_net_deflev), 0, "AH tunnel mode default level."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_AH_CLEARTOS, ah_cleartos, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ah_cleartos), 0, "If set, clear type-of-service field when doing AH computation."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DFBIT, dfbit, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_dfbit), 0, "Do not fragment bit on encap."); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_ECN, ecn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_ecn), 0, "Explicit Congestion Notification handling."); SYSCTL_INT(_net_inet_ipsec, OID_AUTO, crypto_support, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(crypto_support), 0, "Crypto driver selection."); SYSCTL_INT(_net_inet_ipsec, OID_AUTO, async_crypto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(async_crypto), 0, "Use asynchronous mode to parallelize crypto jobs."); SYSCTL_INT(_net_inet_ipsec, OID_AUTO, check_policy_history, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(check_policy_history), 0, "Use strict check of inbound packets to security policy compliance."); SYSCTL_INT(_net_inet_ipsec, OID_AUTO, natt_cksum_policy, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(natt_cksum_policy), 0, "Method to fix TCP/UDP checksum for transport mode IPsec after NAT."); SYSCTL_INT(_net_inet_ipsec, OID_AUTO, filtertunnel, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_filtertunnel), 0, "If set, filter packets from an IPsec tunnel."); SYSCTL_VNET_PCPUSTAT(_net_inet_ipsec, OID_AUTO, ipsecstats, struct ipsecstat, ipsec4stat, "IPsec IPv4 statistics."); #ifdef REGRESSION /* * When set to 1, IPsec will send packets with the same sequence number. * This allows to verify if the other side has proper replay attacks detection. */ VNET_DEFINE(int, ipsec_replay) = 0; SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0, "Emulate replay attack"); /* * When set 1, IPsec will send packets with corrupted HMAC. * This allows to verify if the other side properly detects modified packets. */ VNET_DEFINE(int, ipsec_integrity) = 0; SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0, "Emulate man-in-the-middle attack"); #endif #ifdef INET6 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat); VNET_PCPUSTAT_SYSINIT(ipsec6stat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(ipsec6stat); #endif /* VIMAGE */ VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE; VNET_DEFINE(int, ip6_ipsec_ecn) = 0; /* ECN ignore(-1)/forbidden(0)/allowed(1) */ -static VNET_DEFINE(int, ip6_filtertunnel) = 0; +VNET_DEFINE_STATIC(int, ip6_filtertunnel) = 0; #define V_ip6_filtertunnel VNET(ip6_filtertunnel) SYSCTL_DECL(_net_inet6_ipsec6); /* net.inet6.ipsec6 */ SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy, CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I", "IPsec default policy."); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0, "Default ESP transport mode level."); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0, "Default ESP tunnel mode level."); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0, "AH transfer mode default level."); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0, "AH tunnel mode default level."); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0, "Explicit Congestion Notification handling."); SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel), 0, "If set, filter packets from an IPsec tunnel."); SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats, struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics."); #endif /* INET6 */ static int ipsec_in_reject(struct secpolicy *, struct inpcb *, const struct mbuf *); #ifdef INET static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int); static void ipsec4_setspidx_ipaddr(const struct mbuf *, struct secpolicyindex *); #endif #ifdef INET6 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int); static void ipsec6_setspidx_ipaddr(const struct mbuf *, struct secpolicyindex *); #endif /* * Return a held reference to the default SP. */ static struct secpolicy * key_allocsp_default(void) { key_addref(V_def_policy); return (V_def_policy); } static void ipsec_invalidate_cache(struct inpcb *inp, u_int dir) { struct secpolicy *sp; INP_WLOCK_ASSERT(inp); if (dir == IPSEC_DIR_OUTBOUND) { if (inp->inp_sp->flags & INP_INBOUND_POLICY) return; sp = inp->inp_sp->sp_in; inp->inp_sp->sp_in = NULL; } else { if (inp->inp_sp->flags & INP_OUTBOUND_POLICY) return; sp = inp->inp_sp->sp_out; inp->inp_sp->sp_out = NULL; } if (sp != NULL) key_freesp(&sp); /* release extra reference */ } static void ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir) { uint32_t genid; int downgrade; INP_LOCK_ASSERT(inp); if (dir == IPSEC_DIR_OUTBOUND) { /* Do we have configured PCB policy? */ if (inp->inp_sp->flags & INP_OUTBOUND_POLICY) return; /* Another thread has already set cached policy */ if (inp->inp_sp->sp_out != NULL) return; /* * Do not cache OUTBOUND policy if PCB isn't connected, * i.e. foreign address is INADDR_ANY/UNSPECIFIED. */ #ifdef INET if ((inp->inp_vflag & INP_IPV4) != 0 && inp->inp_faddr.s_addr == INADDR_ANY) return; #endif #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0 && IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) return; #endif } else { /* Do we have configured PCB policy? */ if (inp->inp_sp->flags & INP_INBOUND_POLICY) return; /* Another thread has already set cached policy */ if (inp->inp_sp->sp_in != NULL) return; /* * Do not cache INBOUND policy for listen socket, * that is bound to INADDR_ANY/UNSPECIFIED address. */ #ifdef INET if ((inp->inp_vflag & INP_IPV4) != 0 && inp->inp_faddr.s_addr == INADDR_ANY) return; #endif #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0 && IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) return; #endif } downgrade = 0; if (!INP_WLOCKED(inp)) { if ((downgrade = INP_TRY_UPGRADE(inp)) == 0) return; } if (dir == IPSEC_DIR_OUTBOUND) inp->inp_sp->sp_out = sp; else inp->inp_sp->sp_in = sp; /* * SP is already referenced by the lookup code. * We take extra reference here to avoid race in the * ipsec_getpcbpolicy() function - SP will not be freed in the * time between we take SP pointer from the cache and key_addref() * call. */ key_addref(sp); genid = key_getspgen(); if (genid != inp->inp_sp->genid) { ipsec_invalidate_cache(inp, dir); inp->inp_sp->genid = genid; } KEYDBG(IPSEC_STAMP, printf("%s: PCB(%p): cached %s SP(%p)\n", __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND": "INBOUND", sp)); if (downgrade != 0) INP_DOWNGRADE(inp); } static struct secpolicy * ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error) { /* Save found OUTBOUND policy into PCB SP cache. */ if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL) ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND); switch (sp->policy) { default: printf("%s: invalid policy %u\n", __func__, sp->policy); /* FALLTHROUGH */ case IPSEC_POLICY_DISCARD: *error = -EINVAL; /* Packet is discarded by caller. */ /* FALLTHROUGH */ case IPSEC_POLICY_BYPASS: case IPSEC_POLICY_NONE: key_freesp(&sp); sp = NULL; /* NB: force NULL result. */ break; case IPSEC_POLICY_IPSEC: /* XXXAE: handle LARVAL SP */ break; } KEYDBG(IPSEC_DUMP, printf("%s: get SP(%p), error %d\n", __func__, sp, *error)); return (sp); } static struct secpolicy * ipsec_getpcbpolicy(struct inpcb *inp, u_int dir) { struct secpolicy *sp; int flags, downgrade; if (inp == NULL || inp->inp_sp == NULL) return (NULL); INP_LOCK_ASSERT(inp); flags = inp->inp_sp->flags; if (dir == IPSEC_DIR_OUTBOUND) { sp = inp->inp_sp->sp_out; flags &= INP_OUTBOUND_POLICY; } else { sp = inp->inp_sp->sp_in; flags &= INP_INBOUND_POLICY; } /* * Check flags. If we have PCB SP, just return it. * Otherwise we need to check that cached SP entry isn't stale. */ if (flags == 0) { if (sp == NULL) return (NULL); if (inp->inp_sp->genid != key_getspgen()) { /* Invalidate the cache. */ downgrade = 0; if (!INP_WLOCKED(inp)) { if ((downgrade = INP_TRY_UPGRADE(inp)) == 0) return (NULL); } ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND); ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND); if (downgrade != 0) INP_DOWNGRADE(inp); return (NULL); } KEYDBG(IPSEC_STAMP, printf("%s: PCB(%p): cache hit SP(%p)\n", __func__, inp, sp)); /* Return referenced cached policy */ } key_addref(sp); return (sp); } #ifdef INET static void ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx, int needport) { uint8_t nxt; int off; /* Sanity check. */ IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip), ("packet too short")); if (m->m_len >= sizeof (struct ip)) { const struct ip *ip = mtod(m, const struct ip *); if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) goto done; off = ip->ip_hl << 2; nxt = ip->ip_p; } else { struct ip ih; m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih); if (ih.ip_off & htons(IP_MF | IP_OFFMASK)) goto done; off = ih.ip_hl << 2; nxt = ih.ip_p; } while (off < m->m_pkthdr.len) { struct ip6_ext ip6e; struct tcphdr th; struct udphdr uh; switch (nxt) { case IPPROTO_TCP: spidx->ul_proto = nxt; if (!needport) goto done_proto; if (off + sizeof(struct tcphdr) > m->m_pkthdr.len) goto done; m_copydata(m, off, sizeof (th), (caddr_t) &th); spidx->src.sin.sin_port = th.th_sport; spidx->dst.sin.sin_port = th.th_dport; return; case IPPROTO_UDP: spidx->ul_proto = nxt; if (!needport) goto done_proto; if (off + sizeof(struct udphdr) > m->m_pkthdr.len) goto done; m_copydata(m, off, sizeof (uh), (caddr_t) &uh); spidx->src.sin.sin_port = uh.uh_sport; spidx->dst.sin.sin_port = uh.uh_dport; return; case IPPROTO_AH: if (off + sizeof(ip6e) > m->m_pkthdr.len) goto done; /* XXX Sigh, this works but is totally bogus. */ m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e); off += (ip6e.ip6e_len + 2) << 2; nxt = ip6e.ip6e_nxt; break; case IPPROTO_ICMP: default: /* XXX Intermediate headers??? */ spidx->ul_proto = nxt; goto done_proto; } } done: spidx->ul_proto = IPSEC_ULPROTO_ANY; done_proto: spidx->src.sin.sin_port = IPSEC_PORT_ANY; spidx->dst.sin.sin_port = IPSEC_PORT_ANY; KEYDBG(IPSEC_DUMP, printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL)); } static void ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx) { ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst); spidx->prefs = sizeof(struct in_addr) << 3; spidx->prefd = sizeof(struct in_addr) << 3; } static struct secpolicy * ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir, int needport) { struct secpolicyindex spidx; struct secpolicy *sp; sp = ipsec_getpcbpolicy(inp, dir); if (sp == NULL && key_havesp(dir)) { /* Make an index to look for a policy. */ ipsec4_setspidx_ipaddr(m, &spidx); ipsec4_get_ulp(m, &spidx, needport); spidx.dir = dir; sp = key_allocsp(&spidx, dir); } if (sp == NULL) /* No SP found, use system default. */ sp = key_allocsp_default(); return (sp); } /* * Check security policy for *OUTBOUND* IPv4 packet. */ struct secpolicy * ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error, int needport) { struct secpolicy *sp; *error = 0; sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport); if (sp != NULL) sp = ipsec_checkpolicy(sp, inp, error); if (sp == NULL) { switch (*error) { case 0: /* No IPsec required: BYPASS or NONE */ break; case -EINVAL: IPSECSTAT_INC(ips_out_polvio); break; default: IPSECSTAT_INC(ips_out_inval); } } KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p), error %d\n", __func__, sp, *error)); if (sp != NULL) KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); return (sp); } /* * Check IPv4 packet against *INBOUND* security policy. * This function is called from tcp_input(), udp_input(), * rip_input() and sctp_input(). */ int ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp) { struct secpolicy *sp; int result; sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0); result = ipsec_in_reject(sp, inp, m); key_freesp(&sp); if (result != 0) IPSECSTAT_INC(ips_in_polvio); return (result); } /* * IPSEC_CAP() method implementation for IPv4. */ int ipsec4_capability(struct mbuf *m, u_int cap) { switch (cap) { case IPSEC_CAP_BYPASS_FILTER: /* * Bypass packet filtering for packets previously handled * by IPsec. */ if (!V_ip4_filtertunnel && m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL) return (1); return (0); case IPSEC_CAP_OPERABLE: /* Do we have active security policies? */ if (key_havesp(IPSEC_DIR_INBOUND) != 0 || key_havesp(IPSEC_DIR_OUTBOUND) != 0) return (1); return (0); }; return (EOPNOTSUPP); } #endif /* INET */ #ifdef INET6 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx, int needport) { struct tcphdr th; struct udphdr uh; struct icmp6_hdr ih; int off, nxt; IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr), ("packet too short")); /* Set default. */ spidx->ul_proto = IPSEC_ULPROTO_ANY; spidx->src.sin6.sin6_port = IPSEC_PORT_ANY; spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY; nxt = -1; off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt); if (off < 0 || m->m_pkthdr.len < off) return; switch (nxt) { case IPPROTO_TCP: spidx->ul_proto = nxt; if (!needport) break; if (off + sizeof(struct tcphdr) > m->m_pkthdr.len) break; m_copydata(m, off, sizeof(th), (caddr_t)&th); spidx->src.sin6.sin6_port = th.th_sport; spidx->dst.sin6.sin6_port = th.th_dport; break; case IPPROTO_UDP: spidx->ul_proto = nxt; if (!needport) break; if (off + sizeof(struct udphdr) > m->m_pkthdr.len) break; m_copydata(m, off, sizeof(uh), (caddr_t)&uh); spidx->src.sin6.sin6_port = uh.uh_sport; spidx->dst.sin6.sin6_port = uh.uh_dport; break; case IPPROTO_ICMPV6: spidx->ul_proto = nxt; if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len) break; m_copydata(m, off, sizeof(ih), (caddr_t)&ih); spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type); spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code); break; default: /* XXX Intermediate headers??? */ spidx->ul_proto = nxt; break; } KEYDBG(IPSEC_DUMP, printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL)); } static void ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx) { ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst); spidx->prefs = sizeof(struct in6_addr) << 3; spidx->prefd = sizeof(struct in6_addr) << 3; } static struct secpolicy * ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir, int needport) { struct secpolicyindex spidx; struct secpolicy *sp; sp = ipsec_getpcbpolicy(inp, dir); if (sp == NULL && key_havesp(dir)) { /* Make an index to look for a policy. */ ipsec6_setspidx_ipaddr(m, &spidx); ipsec6_get_ulp(m, &spidx, needport); spidx.dir = dir; sp = key_allocsp(&spidx, dir); } if (sp == NULL) /* No SP found, use system default. */ sp = key_allocsp_default(); return (sp); } /* * Check security policy for *OUTBOUND* IPv6 packet. */ struct secpolicy * ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error, int needport) { struct secpolicy *sp; *error = 0; sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport); if (sp != NULL) sp = ipsec_checkpolicy(sp, inp, error); if (sp == NULL) { switch (*error) { case 0: /* No IPsec required: BYPASS or NONE */ break; case -EINVAL: IPSEC6STAT_INC(ips_out_polvio); break; default: IPSEC6STAT_INC(ips_out_inval); } } KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p), error %d\n", __func__, sp, *error)); if (sp != NULL) KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); return (sp); } /* * Check IPv6 packet against inbound security policy. * This function is called from tcp6_input(), udp6_input(), * rip6_input() and sctp_input(). */ int ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp) { struct secpolicy *sp; int result; sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0); result = ipsec_in_reject(sp, inp, m); key_freesp(&sp); if (result) IPSEC6STAT_INC(ips_in_polvio); return (result); } /* * IPSEC_CAP() method implementation for IPv6. */ int ipsec6_capability(struct mbuf *m, u_int cap) { switch (cap) { case IPSEC_CAP_BYPASS_FILTER: /* * Bypass packet filtering for packets previously handled * by IPsec. */ if (!V_ip6_filtertunnel && m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL) return (1); return (0); case IPSEC_CAP_OPERABLE: /* Do we have active security policies? */ if (key_havesp(IPSEC_DIR_INBOUND) != 0 || key_havesp(IPSEC_DIR_OUTBOUND) != 0) return (1); return (0); }; return (EOPNOTSUPP); } #endif /* INET6 */ int ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type) { int idx; switch (ctx->af) { #ifdef INET case AF_INET: idx = HHOOK_IPSEC_INET; break; #endif #ifdef INET6 case AF_INET6: idx = HHOOK_IPSEC_INET6; break; #endif default: return (EPFNOSUPPORT); } if (type == HHOOK_TYPE_IPSEC_IN) HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL); else HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL); if (*ctx->mp == NULL) return (EACCES); return (0); } /* * Return current level. * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned. */ u_int ipsec_get_reqlevel(struct secpolicy *sp, u_int idx) { struct ipsecrequest *isr; u_int esp_trans_deflev, esp_net_deflev; u_int ah_trans_deflev, ah_net_deflev; u_int level = 0; IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx)); /* XXX Note that we have ipseclog() expanded here - code sync issue. */ #define IPSEC_CHECK_DEFAULT(lev) \ (((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE && \ (lev) != IPSEC_LEVEL_UNIQUE) \ ? (V_ipsec_debug ? \ log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\ (lev), IPSEC_LEVEL_REQUIRE) : 0), \ (lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev)) /* * IPsec VTI uses unique security policy with fake spidx filled * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing * full level lookup for such policies. */ if (sp->state == IPSEC_SPSTATE_IFNET) { IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE, ("Wrong IPsec request level %d", sp->req[idx]->level)); return (IPSEC_LEVEL_REQUIRE); } /* Set default level. */ switch (sp->spidx.src.sa.sa_family) { #ifdef INET case AF_INET: esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev); esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev); ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev); ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev); break; #endif #ifdef INET6 case AF_INET6: esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev); esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev); ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev); ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev); break; #endif /* INET6 */ default: panic("%s: unknown af %u", __func__, sp->spidx.src.sa.sa_family); } #undef IPSEC_CHECK_DEFAULT isr = sp->req[idx]; /* Set level. */ switch (isr->level) { case IPSEC_LEVEL_DEFAULT: switch (isr->saidx.proto) { case IPPROTO_ESP: if (isr->saidx.mode == IPSEC_MODE_TUNNEL) level = esp_net_deflev; else level = esp_trans_deflev; break; case IPPROTO_AH: if (isr->saidx.mode == IPSEC_MODE_TUNNEL) level = ah_net_deflev; else level = ah_trans_deflev; break; case IPPROTO_IPCOMP: /* * We don't really care, as IPcomp document says that * we shouldn't compress small packets. */ level = IPSEC_LEVEL_USE; break; default: panic("%s: Illegal protocol defined %u\n", __func__, isr->saidx.proto); } break; case IPSEC_LEVEL_USE: case IPSEC_LEVEL_REQUIRE: level = isr->level; break; case IPSEC_LEVEL_UNIQUE: level = IPSEC_LEVEL_REQUIRE; break; default: panic("%s: Illegal IPsec level %u\n", __func__, isr->level); } return (level); } static int ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx) { struct xform_history *xh; struct m_tag *mtag; mtag = NULL; while ((mtag = m_tag_find(__DECONST(struct mbuf *, m), PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) { xh = (struct xform_history *)(mtag + 1); KEYDBG(IPSEC_DATA, char buf[IPSEC_ADDRSTRLEN]; printf("%s: mode %s proto %u dst %s\n", __func__, kdebug_secasindex_mode(xh->mode), xh->proto, ipsec_address(&xh->dst, buf, sizeof(buf)))); if (xh->proto != sp->req[idx]->saidx.proto) continue; /* If SA had IPSEC_MODE_ANY, consider this as match. */ if (xh->mode != sp->req[idx]->saidx.mode && xh->mode != IPSEC_MODE_ANY) continue; /* * For transport mode IPsec request doesn't contain * addresses. We need to use address from spidx. */ if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) { if (key_sockaddrcmp_withmask(&xh->dst.sa, &sp->spidx.dst.sa, sp->spidx.prefd) != 0) continue; } else { if (key_sockaddrcmp(&xh->dst.sa, &sp->req[idx]->saidx.dst.sa, 0) != 0) continue; } return (0); /* matched */ } return (1); } /* * Check security policy requirements against the actual * packet contents. Return one if the packet should be * reject as "invalid"; otherwiser return zero to have the * packet treated as "valid". * * OUT: * 0: valid * 1: invalid */ static int ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m) { int i; KEYDBG(IPSEC_STAMP, printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp)); KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL) ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND); /* Check policy. */ switch (sp->policy) { case IPSEC_POLICY_DISCARD: return (1); case IPSEC_POLICY_BYPASS: case IPSEC_POLICY_NONE: return (0); } IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC, ("invalid policy %u", sp->policy)); /* * ipsec[46]_common_input_cb after each transform adds * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode * and destination address from saidx. We can compare info from * these tags with requirements in SP. */ for (i = 0; i < sp->tcount; i++) { /* * Do not check IPcomp, since IPcomp document * says that we shouldn't compress small packets. * IPComp policy should always be treated as being * in "use" level. */ if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP || ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE) continue; if (V_check_policy_history != 0 && ipsec_check_history(m, sp, i) != 0) return (1); else switch (sp->req[i]->saidx.proto) { case IPPROTO_ESP: if ((m->m_flags & M_DECRYPTED) == 0) { KEYDBG(IPSEC_DUMP, printf("%s: ESP m_flags:%x\n", __func__, m->m_flags)); return (1); } break; case IPPROTO_AH: if ((m->m_flags & M_AUTHIPHDR) == 0) { KEYDBG(IPSEC_DUMP, printf("%s: AH m_flags:%x\n", __func__, m->m_flags)); return (1); } break; } } return (0); /* Valid. */ } /* * Compute the byte size to be occupied by IPsec header. * In case it is tunnelled, it includes the size of outer IP header. */ static size_t ipsec_hdrsiz_internal(struct secpolicy *sp) { size_t size; int i; KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp)); KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); switch (sp->policy) { case IPSEC_POLICY_DISCARD: case IPSEC_POLICY_BYPASS: case IPSEC_POLICY_NONE: return (0); } IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC, ("invalid policy %u", sp->policy)); /* * XXX: for each transform we need to lookup suitable SA * and use info from SA to calculate headers size. * XXX: for NAT-T we need to cosider UDP header size. */ size = 0; for (i = 0; i < sp->tcount; i++) { switch (sp->req[i]->saidx.proto) { case IPPROTO_ESP: size += esp_hdrsiz(NULL); break; case IPPROTO_AH: size += ah_hdrsiz(NULL); break; case IPPROTO_IPCOMP: size += sizeof(struct ipcomp); break; } if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) { switch (sp->req[i]->saidx.dst.sa.sa_family) { #ifdef INET case AF_INET: size += sizeof(struct ip); break; #endif #ifdef INET6 case AF_INET6: size += sizeof(struct ip6_hdr); break; #endif default: ipseclog((LOG_ERR, "%s: unknown AF %d in " "IPsec tunnel SA\n", __func__, sp->req[i]->saidx.dst.sa.sa_family)); break; } } } return (size); } /* * Compute ESP/AH header size for protocols with PCB, including * outer IP header. Currently only tcp_output() uses it. */ size_t ipsec_hdrsiz_inpcb(struct inpcb *inp) { struct secpolicyindex spidx; struct secpolicy *sp; size_t sz; sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND); if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) { ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND); sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND); } if (sp == NULL) sp = key_allocsp_default(); sz = ipsec_hdrsiz_internal(sp); key_freesp(&sp); return (sz); } /* * Check the variable replay window. * ipsec_chkreplay() performs replay check before ICV verification. * ipsec_updatereplay() updates replay bitmap. This must be called after * ICV verification (it also performs replay check, which is usually done * beforehand). * 0 (zero) is returned if packet disallowed, 1 if packet permitted. * * Based on RFC 6479. Blocks are 32 bits unsigned integers */ #define IPSEC_BITMAP_INDEX_MASK(w) (w - 1) #define IPSEC_REDUNDANT_BIT_SHIFTS 5 #define IPSEC_REDUNDANT_BITS (1 << IPSEC_REDUNDANT_BIT_SHIFTS) #define IPSEC_BITMAP_LOC_MASK (IPSEC_REDUNDANT_BITS - 1) int ipsec_chkreplay(uint32_t seq, struct secasvar *sav) { const struct secreplay *replay; uint32_t wsizeb; /* Constant: window size. */ int index, bit_location; IPSEC_ASSERT(sav != NULL, ("Null SA")); IPSEC_ASSERT(sav->replay != NULL, ("Null replay state")); replay = sav->replay; /* No need to check replay if disabled. */ if (replay->wsize == 0) return (1); /* Constant. */ wsizeb = replay->wsize << 3; /* Sequence number of 0 is invalid. */ if (seq == 0) return (0); /* First time is always okay. */ if (replay->count == 0) return (1); /* Larger sequences are okay. */ if (seq > replay->lastseq) return (1); /* Over range to check, i.e. too old or wrapped. */ if (replay->lastseq - seq >= wsizeb) return (0); /* The sequence is inside the sliding window * now check the bit in the bitmap * bit location only depends on the sequence number */ bit_location = seq & IPSEC_BITMAP_LOC_MASK; index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS) & IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size); /* This packet already seen? */ if ((replay->bitmap)[index] & (1 << bit_location)) return (0); return (1); } /* * Check replay counter whether to update or not. * OUT: 0: OK * 1: NG */ int ipsec_updatereplay(uint32_t seq, struct secasvar *sav) { char buf[128]; struct secreplay *replay; uint32_t wsizeb; /* Constant: window size. */ int diff, index, bit_location; IPSEC_ASSERT(sav != NULL, ("Null SA")); IPSEC_ASSERT(sav->replay != NULL, ("Null replay state")); replay = sav->replay; if (replay->wsize == 0) goto ok; /* No need to check replay. */ /* Constant. */ wsizeb = replay->wsize << 3; /* Sequence number of 0 is invalid. */ if (seq == 0) return (1); /* The packet is too old, no need to update */ if (wsizeb + seq < replay->lastseq) goto ok; /* Now update the bit */ index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS); /* First check if the sequence number is in the range */ if (seq > replay->lastseq) { int id; int index_cur = replay->lastseq >> IPSEC_REDUNDANT_BIT_SHIFTS; diff = index - index_cur; if (diff > replay->bitmap_size) { /* something unusual in this case */ diff = replay->bitmap_size; } for (id = 0; id < diff; ++id) { replay->bitmap[(id + index_cur + 1) & IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0; } replay->lastseq = seq; } index &= IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size); bit_location = seq & IPSEC_BITMAP_LOC_MASK; /* this packet has already been received */ if (replay->bitmap[index] & (1 << bit_location)) return (1); replay->bitmap[index] |= (1 << bit_location); ok: if (replay->count == ~0) { /* Set overflow flag. */ replay->overflow++; /* Don't increment, no more packets accepted. */ if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) { if (sav->sah->saidx.proto == IPPROTO_AH) AHSTAT_INC(ahs_wrap); else if (sav->sah->saidx.proto == IPPROTO_ESP) ESPSTAT_INC(esps_wrap); return (1); } ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n", __func__, replay->overflow, ipsec_sa2str(sav, buf, sizeof(buf)))); } return (0); } int ipsec_updateid(struct secasvar *sav, crypto_session_t *new, crypto_session_t *old) { crypto_session_t tmp; /* * tdb_cryptoid is initialized by xform_init(). * Then it can be changed only when some crypto error occurred or * when SA is deleted. We stored used cryptoid in the xform_data * structure. In case when crypto error occurred and crypto * subsystem has reinited the session, it returns new cryptoid * and EAGAIN error code. * * This function will be called when we got EAGAIN from crypto * subsystem. * *new is cryptoid that was returned by crypto subsystem in * the crp_sid. * *old is the original cryptoid that we stored in xform_data. * * For first failed request *old == sav->tdb_cryptoid, then * we update sav->tdb_cryptoid and redo crypto_dispatch(). * For next failed request *old != sav->tdb_cryptoid, then * we store cryptoid from first request into the *new variable * and crp_sid from this second session will be returned via * *old pointer, so caller can release second session. * * XXXAE: check this more carefully. */ KEYDBG(IPSEC_STAMP, printf("%s: SA(%p) moves cryptoid %p -> %p\n", __func__, sav, *old, *new)); KEYDBG(IPSEC_DATA, kdebug_secasv(sav)); SECASVAR_LOCK(sav); if (sav->tdb_cryptoid != *old) { /* cryptoid was already updated */ tmp = *new; *new = sav->tdb_cryptoid; *old = tmp; SECASVAR_UNLOCK(sav); return (1); } sav->tdb_cryptoid = *new; SECASVAR_UNLOCK(sav); return (0); } int ipsec_initialized(void) { return (V_def_policy != NULL); } static void def_policy_init(const void *unused __unused) { V_def_policy = key_newsp(); if (V_def_policy != NULL) { V_def_policy->policy = IPSEC_POLICY_NONE; /* Force INPCB SP cache invalidation */ key_bumpspgen(); } else printf("%s: failed to initialize default policy\n", __func__); } static void def_policy_uninit(const void *unused __unused) { if (V_def_policy != NULL) { key_freesp(&V_def_policy); key_bumpspgen(); } } VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, def_policy_init, NULL); VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, def_policy_uninit, NULL); Index: head/sys/netipsec/key.c =================================================================== --- head/sys/netipsec/key.c (revision 336675) +++ head/sys/netipsec/key.c (revision 336676) @@ -1,8686 +1,8686 @@ /* $FreeBSD$ */ /* $KAME: key.c,v 1.191 2001/06/27 10:46:49 sakane Exp $ */ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This code is referd to RFC 2367 */ #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #endif /* INET6 */ #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include /* randomness */ #include #define FULLMASK 0xff #define _BITS(bytes) ((bytes) << 3) /* * Note on SA reference counting: * - SAs that are not in DEAD state will have (total external reference + 1) * following value in reference count field. they cannot be freed and are * referenced from SA header. * - SAs that are in DEAD state will have (total external reference) * in reference count field. they are ready to be freed. reference from * SA header will be removed in key_delsav(), when the reference count * field hits 0 (= no external reference other than from SA header. */ VNET_DEFINE(u_int32_t, key_debug_level) = 0; -static VNET_DEFINE(u_int, key_spi_trycnt) = 1000; -static VNET_DEFINE(u_int32_t, key_spi_minval) = 0x100; -static VNET_DEFINE(u_int32_t, key_spi_maxval) = 0x0fffffff; /* XXX */ -static VNET_DEFINE(u_int32_t, policy_id) = 0; +VNET_DEFINE_STATIC(u_int, key_spi_trycnt) = 1000; +VNET_DEFINE_STATIC(u_int32_t, key_spi_minval) = 0x100; +VNET_DEFINE_STATIC(u_int32_t, key_spi_maxval) = 0x0fffffff; /* XXX */ +VNET_DEFINE_STATIC(u_int32_t, policy_id) = 0; /*interval to initialize randseed,1(m)*/ -static VNET_DEFINE(u_int, key_int_random) = 60; +VNET_DEFINE_STATIC(u_int, key_int_random) = 60; /* interval to expire acquiring, 30(s)*/ -static VNET_DEFINE(u_int, key_larval_lifetime) = 30; +VNET_DEFINE_STATIC(u_int, key_larval_lifetime) = 30; /* counter for blocking SADB_ACQUIRE.*/ -static VNET_DEFINE(int, key_blockacq_count) = 10; +VNET_DEFINE_STATIC(int, key_blockacq_count) = 10; /* lifetime for blocking SADB_ACQUIRE.*/ -static VNET_DEFINE(int, key_blockacq_lifetime) = 20; +VNET_DEFINE_STATIC(int, key_blockacq_lifetime) = 20; /* preferred old sa rather than new sa.*/ -static VNET_DEFINE(int, key_preferred_oldsa) = 1; +VNET_DEFINE_STATIC(int, key_preferred_oldsa) = 1; #define V_key_spi_trycnt VNET(key_spi_trycnt) #define V_key_spi_minval VNET(key_spi_minval) #define V_key_spi_maxval VNET(key_spi_maxval) #define V_policy_id VNET(policy_id) #define V_key_int_random VNET(key_int_random) #define V_key_larval_lifetime VNET(key_larval_lifetime) #define V_key_blockacq_count VNET(key_blockacq_count) #define V_key_blockacq_lifetime VNET(key_blockacq_lifetime) #define V_key_preferred_oldsa VNET(key_preferred_oldsa) -static VNET_DEFINE(u_int32_t, acq_seq) = 0; +VNET_DEFINE_STATIC(u_int32_t, acq_seq) = 0; #define V_acq_seq VNET(acq_seq) -static VNET_DEFINE(uint32_t, sp_genid) = 0; +VNET_DEFINE_STATIC(uint32_t, sp_genid) = 0; #define V_sp_genid VNET(sp_genid) /* SPD */ TAILQ_HEAD(secpolicy_queue, secpolicy); LIST_HEAD(secpolicy_list, secpolicy); -static VNET_DEFINE(struct secpolicy_queue, sptree[IPSEC_DIR_MAX]); -static VNET_DEFINE(struct secpolicy_queue, sptree_ifnet[IPSEC_DIR_MAX]); +VNET_DEFINE_STATIC(struct secpolicy_queue, sptree[IPSEC_DIR_MAX]); +VNET_DEFINE_STATIC(struct secpolicy_queue, sptree_ifnet[IPSEC_DIR_MAX]); static struct rmlock sptree_lock; #define V_sptree VNET(sptree) #define V_sptree_ifnet VNET(sptree_ifnet) #define SPTREE_LOCK_INIT() rm_init(&sptree_lock, "sptree") #define SPTREE_LOCK_DESTROY() rm_destroy(&sptree_lock) #define SPTREE_RLOCK_TRACKER struct rm_priotracker sptree_tracker #define SPTREE_RLOCK() rm_rlock(&sptree_lock, &sptree_tracker) #define SPTREE_RUNLOCK() rm_runlock(&sptree_lock, &sptree_tracker) #define SPTREE_RLOCK_ASSERT() rm_assert(&sptree_lock, RA_RLOCKED) #define SPTREE_WLOCK() rm_wlock(&sptree_lock) #define SPTREE_WUNLOCK() rm_wunlock(&sptree_lock) #define SPTREE_WLOCK_ASSERT() rm_assert(&sptree_lock, RA_WLOCKED) #define SPTREE_UNLOCK_ASSERT() rm_assert(&sptree_lock, RA_UNLOCKED) /* Hash table for lookup SP using unique id */ -static VNET_DEFINE(struct secpolicy_list *, sphashtbl); -static VNET_DEFINE(u_long, sphash_mask); +VNET_DEFINE_STATIC(struct secpolicy_list *, sphashtbl); +VNET_DEFINE_STATIC(u_long, sphash_mask); #define V_sphashtbl VNET(sphashtbl) #define V_sphash_mask VNET(sphash_mask) #define SPHASH_NHASH_LOG2 7 #define SPHASH_NHASH (1 << SPHASH_NHASH_LOG2) #define SPHASH_HASHVAL(id) (key_u32hash(id) & V_sphash_mask) #define SPHASH_HASH(id) &V_sphashtbl[SPHASH_HASHVAL(id)] /* SPD cache */ struct spdcache_entry { struct secpolicyindex spidx; /* secpolicyindex */ struct secpolicy *sp; /* cached policy to be used */ LIST_ENTRY(spdcache_entry) chain; }; LIST_HEAD(spdcache_entry_list, spdcache_entry); #define SPDCACHE_MAX_ENTRIES_PER_HASH 8 -static VNET_DEFINE(u_int, key_spdcache_maxentries) = 0; +VNET_DEFINE_STATIC(u_int, key_spdcache_maxentries) = 0; #define V_key_spdcache_maxentries VNET(key_spdcache_maxentries) -static VNET_DEFINE(u_int, key_spdcache_threshold) = 32; +VNET_DEFINE_STATIC(u_int, key_spdcache_threshold) = 32; #define V_key_spdcache_threshold VNET(key_spdcache_threshold) -static VNET_DEFINE(unsigned long, spd_size) = 0; +VNET_DEFINE_STATIC(unsigned long, spd_size) = 0; #define V_spd_size VNET(spd_size) #define SPDCACHE_ENABLED() (V_key_spdcache_maxentries != 0) #define SPDCACHE_ACTIVE() \ (SPDCACHE_ENABLED() && V_spd_size >= V_key_spdcache_threshold) -static VNET_DEFINE(struct spdcache_entry_list *, spdcachehashtbl); -static VNET_DEFINE(u_long, spdcachehash_mask); +VNET_DEFINE_STATIC(struct spdcache_entry_list *, spdcachehashtbl); +VNET_DEFINE_STATIC(u_long, spdcachehash_mask); #define V_spdcachehashtbl VNET(spdcachehashtbl) #define V_spdcachehash_mask VNET(spdcachehash_mask) #define SPDCACHE_HASHVAL(idx) \ (key_addrprotohash(&(idx)->src, &(idx)->dst, &(idx)->ul_proto) & \ V_spdcachehash_mask) /* Each cache line is protected by a mutex */ -static VNET_DEFINE(struct mtx *, spdcache_lock); +VNET_DEFINE_STATIC(struct mtx *, spdcache_lock); #define V_spdcache_lock VNET(spdcache_lock) #define SPDCACHE_LOCK_INIT(a) \ mtx_init(&V_spdcache_lock[a], "spdcache", \ "fast ipsec SPD cache", MTX_DEF|MTX_DUPOK) #define SPDCACHE_LOCK_DESTROY(a) mtx_destroy(&V_spdcache_lock[a]) #define SPDCACHE_LOCK(a) mtx_lock(&V_spdcache_lock[a]); #define SPDCACHE_UNLOCK(a) mtx_unlock(&V_spdcache_lock[a]); /* SAD */ TAILQ_HEAD(secashead_queue, secashead); LIST_HEAD(secashead_list, secashead); -static VNET_DEFINE(struct secashead_queue, sahtree); +VNET_DEFINE_STATIC(struct secashead_queue, sahtree); static struct rmlock sahtree_lock; #define V_sahtree VNET(sahtree) #define SAHTREE_LOCK_INIT() rm_init(&sahtree_lock, "sahtree") #define SAHTREE_LOCK_DESTROY() rm_destroy(&sahtree_lock) #define SAHTREE_RLOCK_TRACKER struct rm_priotracker sahtree_tracker #define SAHTREE_RLOCK() rm_rlock(&sahtree_lock, &sahtree_tracker) #define SAHTREE_RUNLOCK() rm_runlock(&sahtree_lock, &sahtree_tracker) #define SAHTREE_RLOCK_ASSERT() rm_assert(&sahtree_lock, RA_RLOCKED) #define SAHTREE_WLOCK() rm_wlock(&sahtree_lock) #define SAHTREE_WUNLOCK() rm_wunlock(&sahtree_lock) #define SAHTREE_WLOCK_ASSERT() rm_assert(&sahtree_lock, RA_WLOCKED) #define SAHTREE_UNLOCK_ASSERT() rm_assert(&sahtree_lock, RA_UNLOCKED) /* Hash table for lookup in SAD using SA addresses */ -static VNET_DEFINE(struct secashead_list *, sahaddrhashtbl); -static VNET_DEFINE(u_long, sahaddrhash_mask); +VNET_DEFINE_STATIC(struct secashead_list *, sahaddrhashtbl); +VNET_DEFINE_STATIC(u_long, sahaddrhash_mask); #define V_sahaddrhashtbl VNET(sahaddrhashtbl) #define V_sahaddrhash_mask VNET(sahaddrhash_mask) #define SAHHASH_NHASH_LOG2 7 #define SAHHASH_NHASH (1 << SAHHASH_NHASH_LOG2) #define SAHADDRHASH_HASHVAL(idx) \ (key_addrprotohash(&(idx)->src, &(idx)->dst, &(idx)->proto) & \ V_sahaddrhash_mask) #define SAHADDRHASH_HASH(saidx) \ &V_sahaddrhashtbl[SAHADDRHASH_HASHVAL(saidx)] /* Hash table for lookup in SAD using SPI */ LIST_HEAD(secasvar_list, secasvar); -static VNET_DEFINE(struct secasvar_list *, savhashtbl); -static VNET_DEFINE(u_long, savhash_mask); +VNET_DEFINE_STATIC(struct secasvar_list *, savhashtbl); +VNET_DEFINE_STATIC(u_long, savhash_mask); #define V_savhashtbl VNET(savhashtbl) #define V_savhash_mask VNET(savhash_mask) #define SAVHASH_NHASH_LOG2 7 #define SAVHASH_NHASH (1 << SAVHASH_NHASH_LOG2) #define SAVHASH_HASHVAL(spi) (key_u32hash(spi) & V_savhash_mask) #define SAVHASH_HASH(spi) &V_savhashtbl[SAVHASH_HASHVAL(spi)] static uint32_t key_addrprotohash(const union sockaddr_union *src, const union sockaddr_union *dst, const uint8_t *proto) { uint32_t hval; hval = fnv_32_buf(proto, sizeof(*proto), FNV1_32_INIT); switch (dst->sa.sa_family) { #ifdef INET case AF_INET: hval = fnv_32_buf(&src->sin.sin_addr, sizeof(in_addr_t), hval); hval = fnv_32_buf(&dst->sin.sin_addr, sizeof(in_addr_t), hval); break; #endif #ifdef INET6 case AF_INET6: hval = fnv_32_buf(&src->sin6.sin6_addr, sizeof(struct in6_addr), hval); hval = fnv_32_buf(&dst->sin6.sin6_addr, sizeof(struct in6_addr), hval); break; #endif default: hval = 0; ipseclog((LOG_DEBUG, "%s: unknown address family %d", __func__, dst->sa.sa_family)); } return (hval); } static uint32_t key_u32hash(uint32_t val) { return (fnv_32_buf(&val, sizeof(val), FNV1_32_INIT)); } /* registed list */ -static VNET_DEFINE(LIST_HEAD(_regtree, secreg), regtree[SADB_SATYPE_MAX + 1]); +VNET_DEFINE_STATIC(LIST_HEAD(_regtree, secreg), regtree[SADB_SATYPE_MAX + 1]); #define V_regtree VNET(regtree) static struct mtx regtree_lock; #define REGTREE_LOCK_INIT() \ mtx_init(®tree_lock, "regtree", "fast ipsec regtree", MTX_DEF) #define REGTREE_LOCK_DESTROY() mtx_destroy(®tree_lock) #define REGTREE_LOCK() mtx_lock(®tree_lock) #define REGTREE_UNLOCK() mtx_unlock(®tree_lock) #define REGTREE_LOCK_ASSERT() mtx_assert(®tree_lock, MA_OWNED) /* Acquiring list */ LIST_HEAD(secacq_list, secacq); -static VNET_DEFINE(struct secacq_list, acqtree); +VNET_DEFINE_STATIC(struct secacq_list, acqtree); #define V_acqtree VNET(acqtree) static struct mtx acq_lock; #define ACQ_LOCK_INIT() \ mtx_init(&acq_lock, "acqtree", "ipsec SA acquiring list", MTX_DEF) #define ACQ_LOCK_DESTROY() mtx_destroy(&acq_lock) #define ACQ_LOCK() mtx_lock(&acq_lock) #define ACQ_UNLOCK() mtx_unlock(&acq_lock) #define ACQ_LOCK_ASSERT() mtx_assert(&acq_lock, MA_OWNED) /* Hash table for lookup in ACQ list using SA addresses */ -static VNET_DEFINE(struct secacq_list *, acqaddrhashtbl); -static VNET_DEFINE(u_long, acqaddrhash_mask); +VNET_DEFINE_STATIC(struct secacq_list *, acqaddrhashtbl); +VNET_DEFINE_STATIC(u_long, acqaddrhash_mask); #define V_acqaddrhashtbl VNET(acqaddrhashtbl) #define V_acqaddrhash_mask VNET(acqaddrhash_mask) /* Hash table for lookup in ACQ list using SEQ number */ -static VNET_DEFINE(struct secacq_list *, acqseqhashtbl); -static VNET_DEFINE(u_long, acqseqhash_mask); +VNET_DEFINE_STATIC(struct secacq_list *, acqseqhashtbl); +VNET_DEFINE_STATIC(u_long, acqseqhash_mask); #define V_acqseqhashtbl VNET(acqseqhashtbl) #define V_acqseqhash_mask VNET(acqseqhash_mask) #define ACQHASH_NHASH_LOG2 7 #define ACQHASH_NHASH (1 << ACQHASH_NHASH_LOG2) #define ACQADDRHASH_HASHVAL(idx) \ (key_addrprotohash(&(idx)->src, &(idx)->dst, &(idx)->proto) & \ V_acqaddrhash_mask) #define ACQSEQHASH_HASHVAL(seq) \ (key_u32hash(seq) & V_acqseqhash_mask) #define ACQADDRHASH_HASH(saidx) \ &V_acqaddrhashtbl[ACQADDRHASH_HASHVAL(saidx)] #define ACQSEQHASH_HASH(seq) \ &V_acqseqhashtbl[ACQSEQHASH_HASHVAL(seq)] /* SP acquiring list */ -static VNET_DEFINE(LIST_HEAD(_spacqtree, secspacq), spacqtree); +VNET_DEFINE_STATIC(LIST_HEAD(_spacqtree, secspacq), spacqtree); #define V_spacqtree VNET(spacqtree) static struct mtx spacq_lock; #define SPACQ_LOCK_INIT() \ mtx_init(&spacq_lock, "spacqtree", \ "fast ipsec security policy acquire list", MTX_DEF) #define SPACQ_LOCK_DESTROY() mtx_destroy(&spacq_lock) #define SPACQ_LOCK() mtx_lock(&spacq_lock) #define SPACQ_UNLOCK() mtx_unlock(&spacq_lock) #define SPACQ_LOCK_ASSERT() mtx_assert(&spacq_lock, MA_OWNED) static const int minsize[] = { sizeof(struct sadb_msg), /* SADB_EXT_RESERVED */ sizeof(struct sadb_sa), /* SADB_EXT_SA */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_CURRENT */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_HARD */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_SOFT */ sizeof(struct sadb_address), /* SADB_EXT_ADDRESS_SRC */ sizeof(struct sadb_address), /* SADB_EXT_ADDRESS_DST */ sizeof(struct sadb_address), /* SADB_EXT_ADDRESS_PROXY */ sizeof(struct sadb_key), /* SADB_EXT_KEY_AUTH */ sizeof(struct sadb_key), /* SADB_EXT_KEY_ENCRYPT */ sizeof(struct sadb_ident), /* SADB_EXT_IDENTITY_SRC */ sizeof(struct sadb_ident), /* SADB_EXT_IDENTITY_DST */ sizeof(struct sadb_sens), /* SADB_EXT_SENSITIVITY */ sizeof(struct sadb_prop), /* SADB_EXT_PROPOSAL */ sizeof(struct sadb_supported), /* SADB_EXT_SUPPORTED_AUTH */ sizeof(struct sadb_supported), /* SADB_EXT_SUPPORTED_ENCRYPT */ sizeof(struct sadb_spirange), /* SADB_EXT_SPIRANGE */ 0, /* SADB_X_EXT_KMPRIVATE */ sizeof(struct sadb_x_policy), /* SADB_X_EXT_POLICY */ sizeof(struct sadb_x_sa2), /* SADB_X_SA2 */ sizeof(struct sadb_x_nat_t_type),/* SADB_X_EXT_NAT_T_TYPE */ sizeof(struct sadb_x_nat_t_port),/* SADB_X_EXT_NAT_T_SPORT */ sizeof(struct sadb_x_nat_t_port),/* SADB_X_EXT_NAT_T_DPORT */ sizeof(struct sadb_address), /* SADB_X_EXT_NAT_T_OAI */ sizeof(struct sadb_address), /* SADB_X_EXT_NAT_T_OAR */ sizeof(struct sadb_x_nat_t_frag),/* SADB_X_EXT_NAT_T_FRAG */ sizeof(struct sadb_x_sa_replay), /* SADB_X_EXT_SA_REPLAY */ sizeof(struct sadb_address), /* SADB_X_EXT_NEW_ADDRESS_SRC */ sizeof(struct sadb_address), /* SADB_X_EXT_NEW_ADDRESS_DST */ }; _Static_assert(sizeof(minsize)/sizeof(int) == SADB_EXT_MAX + 1, "minsize size mismatch"); static const int maxsize[] = { sizeof(struct sadb_msg), /* SADB_EXT_RESERVED */ sizeof(struct sadb_sa), /* SADB_EXT_SA */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_CURRENT */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_HARD */ sizeof(struct sadb_lifetime), /* SADB_EXT_LIFETIME_SOFT */ 0, /* SADB_EXT_ADDRESS_SRC */ 0, /* SADB_EXT_ADDRESS_DST */ 0, /* SADB_EXT_ADDRESS_PROXY */ 0, /* SADB_EXT_KEY_AUTH */ 0, /* SADB_EXT_KEY_ENCRYPT */ 0, /* SADB_EXT_IDENTITY_SRC */ 0, /* SADB_EXT_IDENTITY_DST */ 0, /* SADB_EXT_SENSITIVITY */ 0, /* SADB_EXT_PROPOSAL */ 0, /* SADB_EXT_SUPPORTED_AUTH */ 0, /* SADB_EXT_SUPPORTED_ENCRYPT */ sizeof(struct sadb_spirange), /* SADB_EXT_SPIRANGE */ 0, /* SADB_X_EXT_KMPRIVATE */ 0, /* SADB_X_EXT_POLICY */ sizeof(struct sadb_x_sa2), /* SADB_X_SA2 */ sizeof(struct sadb_x_nat_t_type),/* SADB_X_EXT_NAT_T_TYPE */ sizeof(struct sadb_x_nat_t_port),/* SADB_X_EXT_NAT_T_SPORT */ sizeof(struct sadb_x_nat_t_port),/* SADB_X_EXT_NAT_T_DPORT */ 0, /* SADB_X_EXT_NAT_T_OAI */ 0, /* SADB_X_EXT_NAT_T_OAR */ sizeof(struct sadb_x_nat_t_frag),/* SADB_X_EXT_NAT_T_FRAG */ sizeof(struct sadb_x_sa_replay), /* SADB_X_EXT_SA_REPLAY */ 0, /* SADB_X_EXT_NEW_ADDRESS_SRC */ 0, /* SADB_X_EXT_NEW_ADDRESS_DST */ }; _Static_assert(sizeof(maxsize)/sizeof(int) == SADB_EXT_MAX + 1, "minsize size mismatch"); /* * Internal values for SA flags: * SADB_X_EXT_F_CLONED means that SA was cloned by key_updateaddresses, * thus we will not free the most of SA content in key_delsav(). */ #define SADB_X_EXT_F_CLONED 0x80000000 #define SADB_CHECKLEN(_mhp, _ext) \ ((_mhp)->extlen[(_ext)] < minsize[(_ext)] || (maxsize[(_ext)] != 0 && \ ((_mhp)->extlen[(_ext)] > maxsize[(_ext)]))) #define SADB_CHECKHDR(_mhp, _ext) ((_mhp)->ext[(_ext)] == NULL) -static VNET_DEFINE(int, ipsec_esp_keymin) = 256; -static VNET_DEFINE(int, ipsec_esp_auth) = 0; -static VNET_DEFINE(int, ipsec_ah_keymin) = 128; +VNET_DEFINE_STATIC(int, ipsec_esp_keymin) = 256; +VNET_DEFINE_STATIC(int, ipsec_esp_auth) = 0; +VNET_DEFINE_STATIC(int, ipsec_ah_keymin) = 128; #define V_ipsec_esp_keymin VNET(ipsec_esp_keymin) #define V_ipsec_esp_auth VNET(ipsec_esp_auth) #define V_ipsec_ah_keymin VNET(ipsec_ah_keymin) #ifdef IPSEC_DEBUG VNET_DEFINE(int, ipsec_debug) = 1; #else VNET_DEFINE(int, ipsec_debug) = 0; #endif #ifdef INET SYSCTL_DECL(_net_inet_ipsec); SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEBUG, debug, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_debug), 0, "Enable IPsec debugging output when set."); #endif #ifdef INET6 SYSCTL_DECL(_net_inet6_ipsec6); SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEBUG, debug, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_debug), 0, "Enable IPsec debugging output when set."); #endif SYSCTL_DECL(_net_key); SYSCTL_INT(_net_key, KEYCTL_DEBUG_LEVEL, debug, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_debug_level), 0, ""); /* max count of trial for the decision of spi value */ SYSCTL_INT(_net_key, KEYCTL_SPI_TRY, spi_trycnt, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_spi_trycnt), 0, ""); /* minimum spi value to allocate automatically. */ SYSCTL_INT(_net_key, KEYCTL_SPI_MIN_VALUE, spi_minval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_spi_minval), 0, ""); /* maximun spi value to allocate automatically. */ SYSCTL_INT(_net_key, KEYCTL_SPI_MAX_VALUE, spi_maxval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_spi_maxval), 0, ""); /* interval to initialize randseed */ SYSCTL_INT(_net_key, KEYCTL_RANDOM_INT, int_random, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_int_random), 0, ""); /* lifetime for larval SA */ SYSCTL_INT(_net_key, KEYCTL_LARVAL_LIFETIME, larval_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_larval_lifetime), 0, ""); /* counter for blocking to send SADB_ACQUIRE to IKEd */ SYSCTL_INT(_net_key, KEYCTL_BLOCKACQ_COUNT, blockacq_count, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_blockacq_count), 0, ""); /* lifetime for blocking to send SADB_ACQUIRE to IKEd */ SYSCTL_INT(_net_key, KEYCTL_BLOCKACQ_LIFETIME, blockacq_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_blockacq_lifetime), 0, ""); /* ESP auth */ SYSCTL_INT(_net_key, KEYCTL_ESP_AUTH, esp_auth, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_esp_auth), 0, ""); /* minimum ESP key length */ SYSCTL_INT(_net_key, KEYCTL_ESP_KEYMIN, esp_keymin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_esp_keymin), 0, ""); /* minimum AH key length */ SYSCTL_INT(_net_key, KEYCTL_AH_KEYMIN, ah_keymin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_ah_keymin), 0, ""); /* perfered old SA rather than new SA */ SYSCTL_INT(_net_key, KEYCTL_PREFERED_OLDSA, preferred_oldsa, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(key_preferred_oldsa), 0, ""); static SYSCTL_NODE(_net_key, OID_AUTO, spdcache, CTLFLAG_RW, 0, "SPD cache"); SYSCTL_UINT(_net_key_spdcache, OID_AUTO, maxentries, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(key_spdcache_maxentries), 0, "Maximum number of entries in the SPD cache" " (power of 2, 0 to disable)"); SYSCTL_UINT(_net_key_spdcache, OID_AUTO, threshold, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(key_spdcache_threshold), 0, "Number of SPs that make the SPD cache active"); #define __LIST_CHAINED(elm) \ (!((elm)->chain.le_next == NULL && (elm)->chain.le_prev == NULL)) MALLOC_DEFINE(M_IPSEC_SA, "secasvar", "ipsec security association"); MALLOC_DEFINE(M_IPSEC_SAH, "sahead", "ipsec sa head"); MALLOC_DEFINE(M_IPSEC_SP, "ipsecpolicy", "ipsec security policy"); MALLOC_DEFINE(M_IPSEC_SR, "ipsecrequest", "ipsec security request"); MALLOC_DEFINE(M_IPSEC_MISC, "ipsec-misc", "ipsec miscellaneous"); MALLOC_DEFINE(M_IPSEC_SAQ, "ipsec-saq", "ipsec sa acquire"); MALLOC_DEFINE(M_IPSEC_SAR, "ipsec-reg", "ipsec sa acquire"); MALLOC_DEFINE(M_IPSEC_SPDCACHE, "ipsec-spdcache", "ipsec SPD cache"); -static VNET_DEFINE(uma_zone_t, key_lft_zone); +VNET_DEFINE_STATIC(uma_zone_t, key_lft_zone); #define V_key_lft_zone VNET(key_lft_zone) static LIST_HEAD(xforms_list, xformsw) xforms = LIST_HEAD_INITIALIZER(); static struct mtx xforms_lock; #define XFORMS_LOCK_INIT() \ mtx_init(&xforms_lock, "xforms_list", "IPsec transforms list", MTX_DEF) #define XFORMS_LOCK_DESTROY() mtx_destroy(&xforms_lock) #define XFORMS_LOCK() mtx_lock(&xforms_lock) #define XFORMS_UNLOCK() mtx_unlock(&xforms_lock) /* * set parameters into secpolicyindex buffer. * Must allocate secpolicyindex buffer passed to this function. */ #define KEY_SETSECSPIDX(_dir, s, d, ps, pd, ulp, idx) \ do { \ bzero((idx), sizeof(struct secpolicyindex)); \ (idx)->dir = (_dir); \ (idx)->prefs = (ps); \ (idx)->prefd = (pd); \ (idx)->ul_proto = (ulp); \ bcopy((s), &(idx)->src, ((const struct sockaddr *)(s))->sa_len); \ bcopy((d), &(idx)->dst, ((const struct sockaddr *)(d))->sa_len); \ } while (0) /* * set parameters into secasindex buffer. * Must allocate secasindex buffer before calling this function. */ #define KEY_SETSECASIDX(p, m, r, s, d, idx) \ do { \ bzero((idx), sizeof(struct secasindex)); \ (idx)->proto = (p); \ (idx)->mode = (m); \ (idx)->reqid = (r); \ bcopy((s), &(idx)->src, ((const struct sockaddr *)(s))->sa_len); \ bcopy((d), &(idx)->dst, ((const struct sockaddr *)(d))->sa_len); \ key_porttosaddr(&(idx)->src.sa, 0); \ key_porttosaddr(&(idx)->dst.sa, 0); \ } while (0) /* key statistics */ struct _keystat { u_long getspi_count; /* the avarage of count to try to get new SPI */ } keystat; struct sadb_msghdr { struct sadb_msg *msg; struct sadb_ext *ext[SADB_EXT_MAX + 1]; int extoff[SADB_EXT_MAX + 1]; int extlen[SADB_EXT_MAX + 1]; }; static struct supported_ealgs { int sadb_alg; const struct enc_xform *xform; } supported_ealgs[] = { { SADB_EALG_DESCBC, &enc_xform_des }, { SADB_EALG_3DESCBC, &enc_xform_3des }, { SADB_X_EALG_AES, &enc_xform_rijndael128 }, { SADB_X_EALG_BLOWFISHCBC, &enc_xform_blf }, { SADB_X_EALG_CAST128CBC, &enc_xform_cast5 }, { SADB_EALG_NULL, &enc_xform_null }, { SADB_X_EALG_CAMELLIACBC, &enc_xform_camellia }, { SADB_X_EALG_AESCTR, &enc_xform_aes_icm }, { SADB_X_EALG_AESGCM16, &enc_xform_aes_nist_gcm }, { SADB_X_EALG_AESGMAC, &enc_xform_aes_nist_gmac }, }; static struct supported_aalgs { int sadb_alg; const struct auth_hash *xform; } supported_aalgs[] = { { SADB_X_AALG_NULL, &auth_hash_null }, { SADB_AALG_MD5HMAC, &auth_hash_hmac_md5 }, { SADB_AALG_SHA1HMAC, &auth_hash_hmac_sha1 }, { SADB_X_AALG_RIPEMD160HMAC, &auth_hash_hmac_ripemd_160 }, { SADB_X_AALG_MD5, &auth_hash_key_md5 }, { SADB_X_AALG_SHA, &auth_hash_key_sha1 }, { SADB_X_AALG_SHA2_256, &auth_hash_hmac_sha2_256 }, { SADB_X_AALG_SHA2_384, &auth_hash_hmac_sha2_384 }, { SADB_X_AALG_SHA2_512, &auth_hash_hmac_sha2_512 }, { SADB_X_AALG_AES128GMAC, &auth_hash_nist_gmac_aes_128 }, { SADB_X_AALG_AES192GMAC, &auth_hash_nist_gmac_aes_192 }, { SADB_X_AALG_AES256GMAC, &auth_hash_nist_gmac_aes_256 }, }; static struct supported_calgs { int sadb_alg; const struct comp_algo *xform; } supported_calgs[] = { { SADB_X_CALG_DEFLATE, &comp_algo_deflate }, }; #ifndef IPSEC_DEBUG2 static struct callout key_timer; #endif static void key_unlink(struct secpolicy *); static struct secpolicy *key_do_allocsp(struct secpolicyindex *spidx, u_int dir); static struct secpolicy *key_getsp(struct secpolicyindex *); static struct secpolicy *key_getspbyid(u_int32_t); static struct mbuf *key_gather_mbuf(struct mbuf *, const struct sadb_msghdr *, int, int, ...); static int key_spdadd(struct socket *, struct mbuf *, const struct sadb_msghdr *); static uint32_t key_getnewspid(void); static int key_spddelete(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_spddelete2(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_spdget(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_spdflush(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_spddump(struct socket *, struct mbuf *, const struct sadb_msghdr *); static struct mbuf *key_setdumpsp(struct secpolicy *, u_int8_t, u_int32_t, u_int32_t); static struct mbuf *key_sp2mbuf(struct secpolicy *); static size_t key_getspreqmsglen(struct secpolicy *); static int key_spdexpire(struct secpolicy *); static struct secashead *key_newsah(struct secasindex *); static void key_freesah(struct secashead **); static void key_delsah(struct secashead *); static struct secasvar *key_newsav(const struct sadb_msghdr *, struct secasindex *, uint32_t, int *); static void key_delsav(struct secasvar *); static void key_unlinksav(struct secasvar *); static struct secashead *key_getsah(struct secasindex *); static int key_checkspidup(uint32_t); static struct secasvar *key_getsavbyspi(uint32_t); static int key_setnatt(struct secasvar *, const struct sadb_msghdr *); static int key_setsaval(struct secasvar *, const struct sadb_msghdr *); static int key_updatelifetimes(struct secasvar *, const struct sadb_msghdr *); static int key_updateaddresses(struct socket *, struct mbuf *, const struct sadb_msghdr *, struct secasvar *, struct secasindex *); static struct mbuf *key_setdumpsa(struct secasvar *, u_int8_t, u_int8_t, u_int32_t, u_int32_t); static struct mbuf *key_setsadbmsg(u_int8_t, u_int16_t, u_int8_t, u_int32_t, pid_t, u_int16_t); static struct mbuf *key_setsadbsa(struct secasvar *); static struct mbuf *key_setsadbaddr(u_int16_t, const struct sockaddr *, u_int8_t, u_int16_t); static struct mbuf *key_setsadbxport(u_int16_t, u_int16_t); static struct mbuf *key_setsadbxtype(u_int16_t); static struct mbuf *key_setsadbxsa2(u_int8_t, u_int32_t, u_int32_t); static struct mbuf *key_setsadbxsareplay(u_int32_t); static struct mbuf *key_setsadbxpolicy(u_int16_t, u_int8_t, u_int32_t, u_int32_t); static struct seckey *key_dup_keymsg(const struct sadb_key *, size_t, struct malloc_type *); static struct seclifetime *key_dup_lifemsg(const struct sadb_lifetime *src, struct malloc_type *); /* flags for key_cmpsaidx() */ #define CMP_HEAD 1 /* protocol, addresses. */ #define CMP_MODE_REQID 2 /* additionally HEAD, reqid, mode. */ #define CMP_REQID 3 /* additionally HEAD, reaid. */ #define CMP_EXACTLY 4 /* all elements. */ static int key_cmpsaidx(const struct secasindex *, const struct secasindex *, int); static int key_cmpspidx_exactly(struct secpolicyindex *, struct secpolicyindex *); static int key_cmpspidx_withmask(struct secpolicyindex *, struct secpolicyindex *); static int key_bbcmp(const void *, const void *, u_int); static uint8_t key_satype2proto(uint8_t); static uint8_t key_proto2satype(uint8_t); static int key_getspi(struct socket *, struct mbuf *, const struct sadb_msghdr *); static uint32_t key_do_getnewspi(struct sadb_spirange *, struct secasindex *); static int key_update(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_add(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_setident(struct secashead *, const struct sadb_msghdr *); static struct mbuf *key_getmsgbuf_x1(struct mbuf *, const struct sadb_msghdr *); static int key_delete(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_delete_all(struct socket *, struct mbuf *, const struct sadb_msghdr *, struct secasindex *); static void key_delete_xform(const struct xformsw *); static int key_get(struct socket *, struct mbuf *, const struct sadb_msghdr *); static void key_getcomb_setlifetime(struct sadb_comb *); static struct mbuf *key_getcomb_ealg(void); static struct mbuf *key_getcomb_ah(void); static struct mbuf *key_getcomb_ipcomp(void); static struct mbuf *key_getprop(const struct secasindex *); static int key_acquire(const struct secasindex *, struct secpolicy *); static uint32_t key_newacq(const struct secasindex *, int *); static uint32_t key_getacq(const struct secasindex *, int *); static int key_acqdone(const struct secasindex *, uint32_t); static int key_acqreset(uint32_t); static struct secspacq *key_newspacq(struct secpolicyindex *); static struct secspacq *key_getspacq(struct secpolicyindex *); static int key_acquire2(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_register(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_expire(struct secasvar *, int); static int key_flush(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_dump(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_promisc(struct socket *, struct mbuf *, const struct sadb_msghdr *); static int key_senderror(struct socket *, struct mbuf *, int); static int key_validate_ext(const struct sadb_ext *, int); static int key_align(struct mbuf *, struct sadb_msghdr *); static struct mbuf *key_setlifetime(struct seclifetime *, uint16_t); static struct mbuf *key_setkey(struct seckey *, uint16_t); static int xform_init(struct secasvar *, u_short); static void spdcache_init(void); static void spdcache_clear(void); static struct spdcache_entry *spdcache_entry_alloc( const struct secpolicyindex *spidx, struct secpolicy *policy); static void spdcache_entry_free(struct spdcache_entry *entry); #ifdef VIMAGE static void spdcache_destroy(void); #endif #define DBG_IPSEC_INITREF(t, p) do { \ refcount_init(&(p)->refcnt, 1); \ KEYDBG(KEY_STAMP, \ printf("%s: Initialize refcnt %s(%p) = %u\n", \ __func__, #t, (p), (p)->refcnt)); \ } while (0) #define DBG_IPSEC_ADDREF(t, p) do { \ refcount_acquire(&(p)->refcnt); \ KEYDBG(KEY_STAMP, \ printf("%s: Acquire refcnt %s(%p) -> %u\n", \ __func__, #t, (p), (p)->refcnt)); \ } while (0) #define DBG_IPSEC_DELREF(t, p) do { \ KEYDBG(KEY_STAMP, \ printf("%s: Release refcnt %s(%p) -> %u\n", \ __func__, #t, (p), (p)->refcnt - 1)); \ refcount_release(&(p)->refcnt); \ } while (0) #define IPSEC_INITREF(t, p) refcount_init(&(p)->refcnt, 1) #define IPSEC_ADDREF(t, p) refcount_acquire(&(p)->refcnt) #define IPSEC_DELREF(t, p) refcount_release(&(p)->refcnt) #define SP_INITREF(p) IPSEC_INITREF(SP, p) #define SP_ADDREF(p) IPSEC_ADDREF(SP, p) #define SP_DELREF(p) IPSEC_DELREF(SP, p) #define SAH_INITREF(p) IPSEC_INITREF(SAH, p) #define SAH_ADDREF(p) IPSEC_ADDREF(SAH, p) #define SAH_DELREF(p) IPSEC_DELREF(SAH, p) #define SAV_INITREF(p) IPSEC_INITREF(SAV, p) #define SAV_ADDREF(p) IPSEC_ADDREF(SAV, p) #define SAV_DELREF(p) IPSEC_DELREF(SAV, p) /* * Update the refcnt while holding the SPTREE lock. */ void key_addref(struct secpolicy *sp) { SP_ADDREF(sp); } /* * Return 0 when there are known to be no SP's for the specified * direction. Otherwise return 1. This is used by IPsec code * to optimize performance. */ int key_havesp(u_int dir) { return (dir == IPSEC_DIR_INBOUND || dir == IPSEC_DIR_OUTBOUND ? TAILQ_FIRST(&V_sptree[dir]) != NULL : 1); } /* %%% IPsec policy management */ /* * Return current SPDB generation. */ uint32_t key_getspgen(void) { return (V_sp_genid); } void key_bumpspgen(void) { V_sp_genid++; } static int key_checksockaddrs(struct sockaddr *src, struct sockaddr *dst) { /* family match */ if (src->sa_family != dst->sa_family) return (EINVAL); /* sa_len match */ if (src->sa_len != dst->sa_len) return (EINVAL); switch (src->sa_family) { #ifdef INET case AF_INET: if (src->sa_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif #ifdef INET6 case AF_INET6: if (src->sa_len != sizeof(struct sockaddr_in6)) return (EINVAL); break; #endif default: return (EAFNOSUPPORT); } return (0); } struct secpolicy * key_do_allocsp(struct secpolicyindex *spidx, u_int dir) { SPTREE_RLOCK_TRACKER; struct secpolicy *sp; IPSEC_ASSERT(spidx != NULL, ("null spidx")); IPSEC_ASSERT(dir == IPSEC_DIR_INBOUND || dir == IPSEC_DIR_OUTBOUND, ("invalid direction %u", dir)); SPTREE_RLOCK(); TAILQ_FOREACH(sp, &V_sptree[dir], chain) { if (key_cmpspidx_withmask(&sp->spidx, spidx)) { SP_ADDREF(sp); break; } } SPTREE_RUNLOCK(); return (sp); } /* * allocating a SP for OUTBOUND or INBOUND packet. * Must call key_freesp() later. * OUT: NULL: not found * others: found and return the pointer. */ struct secpolicy * key_allocsp(struct secpolicyindex *spidx, u_int dir) { struct spdcache_entry *entry, *lastentry, *tmpentry; struct secpolicy *sp; uint32_t hashv; int nb_entries; if (!SPDCACHE_ACTIVE()) { sp = key_do_allocsp(spidx, dir); goto out; } hashv = SPDCACHE_HASHVAL(spidx); SPDCACHE_LOCK(hashv); nb_entries = 0; LIST_FOREACH_SAFE(entry, &V_spdcachehashtbl[hashv], chain, tmpentry) { /* Removed outdated entries */ if (entry->sp != NULL && entry->sp->state == IPSEC_SPSTATE_DEAD) { LIST_REMOVE(entry, chain); spdcache_entry_free(entry); continue; } nb_entries++; if (!key_cmpspidx_exactly(&entry->spidx, spidx)) { lastentry = entry; continue; } sp = entry->sp; if (entry->sp != NULL) SP_ADDREF(sp); /* IPSECSTAT_INC(ips_spdcache_hits); */ SPDCACHE_UNLOCK(hashv); goto out; } /* IPSECSTAT_INC(ips_spdcache_misses); */ sp = key_do_allocsp(spidx, dir); entry = spdcache_entry_alloc(spidx, sp); if (entry != NULL) { if (nb_entries >= SPDCACHE_MAX_ENTRIES_PER_HASH) { LIST_REMOVE(lastentry, chain); spdcache_entry_free(lastentry); } LIST_INSERT_HEAD(&V_spdcachehashtbl[hashv], entry, chain); } SPDCACHE_UNLOCK(hashv); out: if (sp != NULL) { /* found a SPD entry */ sp->lastused = time_second; KEYDBG(IPSEC_STAMP, printf("%s: return SP(%p)\n", __func__, sp)); KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); } else { KEYDBG(IPSEC_DATA, printf("%s: lookup failed for ", __func__); kdebug_secpolicyindex(spidx, NULL)); } return (sp); } /* * Allocating an SA entry for an *INBOUND* or *OUTBOUND* TCP packet, signed * or should be signed by MD5 signature. * We don't use key_allocsa() for such lookups, because we don't know SPI. * Unlike ESP and AH protocols, SPI isn't transmitted in the TCP header with * signed packet. We use SADB only as storage for password. * OUT: positive: corresponding SA for given saidx found. * NULL: SA not found */ struct secasvar * key_allocsa_tcpmd5(struct secasindex *saidx) { SAHTREE_RLOCK_TRACKER; struct secashead *sah; struct secasvar *sav; IPSEC_ASSERT(saidx->proto == IPPROTO_TCP, ("unexpected security protocol %u", saidx->proto)); IPSEC_ASSERT(saidx->mode == IPSEC_MODE_TCPMD5, ("unexpected mode %u", saidx->mode)); SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(saidx), addrhash) { KEYDBG(IPSEC_DUMP, printf("%s: checking SAH\n", __func__); kdebug_secash(sah, " ")); if (sah->saidx.proto != IPPROTO_TCP) continue; if (!key_sockaddrcmp(&saidx->dst.sa, &sah->saidx.dst.sa, 0) && !key_sockaddrcmp(&saidx->src.sa, &sah->saidx.src.sa, 0)) break; } if (sah != NULL) { if (V_key_preferred_oldsa) sav = TAILQ_LAST(&sah->savtree_alive, secasvar_queue); else sav = TAILQ_FIRST(&sah->savtree_alive); if (sav != NULL) SAV_ADDREF(sav); } else sav = NULL; SAHTREE_RUNLOCK(); if (sav != NULL) { KEYDBG(IPSEC_STAMP, printf("%s: return SA(%p)\n", __func__, sav)); KEYDBG(IPSEC_DATA, kdebug_secasv(sav)); } else { KEYDBG(IPSEC_STAMP, printf("%s: SA not found\n", __func__)); KEYDBG(IPSEC_DATA, kdebug_secasindex(saidx, NULL)); } return (sav); } /* * Allocating an SA entry for an *OUTBOUND* packet. * OUT: positive: corresponding SA for given saidx found. * NULL: SA not found, but will be acquired, check *error * for acquiring status. */ struct secasvar * key_allocsa_policy(struct secpolicy *sp, const struct secasindex *saidx, int *error) { SAHTREE_RLOCK_TRACKER; struct secashead *sah; struct secasvar *sav; IPSEC_ASSERT(saidx != NULL, ("null saidx")); IPSEC_ASSERT(saidx->mode == IPSEC_MODE_TRANSPORT || saidx->mode == IPSEC_MODE_TUNNEL, ("unexpected policy %u", saidx->mode)); /* * We check new SA in the IPsec request because a different * SA may be involved each time this request is checked, either * because new SAs are being configured, or this request is * associated with an unconnected datagram socket, or this request * is associated with a system default policy. */ SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(saidx), addrhash) { KEYDBG(IPSEC_DUMP, printf("%s: checking SAH\n", __func__); kdebug_secash(sah, " ")); if (key_cmpsaidx(&sah->saidx, saidx, CMP_MODE_REQID)) break; } if (sah != NULL) { /* * Allocate the oldest SA available according to * draft-jenkins-ipsec-rekeying-03. */ if (V_key_preferred_oldsa) sav = TAILQ_LAST(&sah->savtree_alive, secasvar_queue); else sav = TAILQ_FIRST(&sah->savtree_alive); if (sav != NULL) SAV_ADDREF(sav); } else sav = NULL; SAHTREE_RUNLOCK(); if (sav != NULL) { *error = 0; KEYDBG(IPSEC_STAMP, printf("%s: chosen SA(%p) for SP(%p)\n", __func__, sav, sp)); KEYDBG(IPSEC_DATA, kdebug_secasv(sav)); return (sav); /* return referenced SA */ } /* there is no SA */ *error = key_acquire(saidx, sp); if ((*error) != 0) ipseclog((LOG_DEBUG, "%s: error %d returned from key_acquire()\n", __func__, *error)); KEYDBG(IPSEC_STAMP, printf("%s: acquire SA for SP(%p), error %d\n", __func__, sp, *error)); KEYDBG(IPSEC_DATA, kdebug_secasindex(saidx, NULL)); return (NULL); } /* * allocating a usable SA entry for a *INBOUND* packet. * Must call key_freesav() later. * OUT: positive: pointer to a usable sav (i.e. MATURE or DYING state). * NULL: not found, or error occurred. * * According to RFC 2401 SA is uniquely identified by a triple SPI, * destination address, and security protocol. But according to RFC 4301, * SPI by itself suffices to specify an SA. * * Note that, however, we do need to keep source address in IPsec SA. * IKE specification and PF_KEY specification do assume that we * keep source address in IPsec SA. We see a tricky situation here. */ struct secasvar * key_allocsa(union sockaddr_union *dst, uint8_t proto, uint32_t spi) { SAHTREE_RLOCK_TRACKER; struct secasvar *sav; IPSEC_ASSERT(proto == IPPROTO_ESP || proto == IPPROTO_AH || proto == IPPROTO_IPCOMP, ("unexpected security protocol %u", proto)); SAHTREE_RLOCK(); LIST_FOREACH(sav, SAVHASH_HASH(spi), spihash) { if (sav->spi == spi) break; } /* * We use single SPI namespace for all protocols, so it is * impossible to have SPI duplicates in the SAVHASH. */ if (sav != NULL) { if (sav->state != SADB_SASTATE_LARVAL && sav->sah->saidx.proto == proto && key_sockaddrcmp(&dst->sa, &sav->sah->saidx.dst.sa, 0) == 0) SAV_ADDREF(sav); else sav = NULL; } SAHTREE_RUNLOCK(); if (sav == NULL) { KEYDBG(IPSEC_STAMP, char buf[IPSEC_ADDRSTRLEN]; printf("%s: SA not found for spi %u proto %u dst %s\n", __func__, ntohl(spi), proto, ipsec_address(dst, buf, sizeof(buf)))); } else { KEYDBG(IPSEC_STAMP, printf("%s: return SA(%p)\n", __func__, sav)); KEYDBG(IPSEC_DATA, kdebug_secasv(sav)); } return (sav); } struct secasvar * key_allocsa_tunnel(union sockaddr_union *src, union sockaddr_union *dst, uint8_t proto) { SAHTREE_RLOCK_TRACKER; struct secasindex saidx; struct secashead *sah; struct secasvar *sav; IPSEC_ASSERT(src != NULL, ("null src address")); IPSEC_ASSERT(dst != NULL, ("null dst address")); KEY_SETSECASIDX(proto, IPSEC_MODE_TUNNEL, 0, &src->sa, &dst->sa, &saidx); sav = NULL; SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(&saidx), addrhash) { if (IPSEC_MODE_TUNNEL != sah->saidx.mode) continue; if (proto != sah->saidx.proto) continue; if (key_sockaddrcmp(&src->sa, &sah->saidx.src.sa, 0) != 0) continue; if (key_sockaddrcmp(&dst->sa, &sah->saidx.dst.sa, 0) != 0) continue; /* XXXAE: is key_preferred_oldsa reasonably?*/ if (V_key_preferred_oldsa) sav = TAILQ_LAST(&sah->savtree_alive, secasvar_queue); else sav = TAILQ_FIRST(&sah->savtree_alive); if (sav != NULL) { SAV_ADDREF(sav); break; } } SAHTREE_RUNLOCK(); KEYDBG(IPSEC_STAMP, printf("%s: return SA(%p)\n", __func__, sav)); if (sav != NULL) KEYDBG(IPSEC_DATA, kdebug_secasv(sav)); return (sav); } /* * Must be called after calling key_allocsp(). */ void key_freesp(struct secpolicy **spp) { struct secpolicy *sp = *spp; IPSEC_ASSERT(sp != NULL, ("null sp")); if (SP_DELREF(sp) == 0) return; KEYDBG(IPSEC_STAMP, printf("%s: last reference to SP(%p)\n", __func__, sp)); KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp)); *spp = NULL; while (sp->tcount > 0) ipsec_delisr(sp->req[--sp->tcount]); free(sp, M_IPSEC_SP); } static void key_unlink(struct secpolicy *sp) { IPSEC_ASSERT(sp->spidx.dir == IPSEC_DIR_INBOUND || sp->spidx.dir == IPSEC_DIR_OUTBOUND, ("invalid direction %u", sp->spidx.dir)); SPTREE_UNLOCK_ASSERT(); KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, sp)); SPTREE_WLOCK(); if (sp->state != IPSEC_SPSTATE_ALIVE) { /* SP is already unlinked */ SPTREE_WUNLOCK(); return; } sp->state = IPSEC_SPSTATE_DEAD; TAILQ_REMOVE(&V_sptree[sp->spidx.dir], sp, chain); V_spd_size--; LIST_REMOVE(sp, idhash); V_sp_genid++; SPTREE_WUNLOCK(); if (SPDCACHE_ENABLED()) spdcache_clear(); key_freesp(&sp); } /* * insert a secpolicy into the SP database. Lower priorities first */ static void key_insertsp(struct secpolicy *newsp) { struct secpolicy *sp; SPTREE_WLOCK_ASSERT(); TAILQ_FOREACH(sp, &V_sptree[newsp->spidx.dir], chain) { if (newsp->priority < sp->priority) { TAILQ_INSERT_BEFORE(sp, newsp, chain); goto done; } } TAILQ_INSERT_TAIL(&V_sptree[newsp->spidx.dir], newsp, chain); done: LIST_INSERT_HEAD(SPHASH_HASH(newsp->id), newsp, idhash); newsp->state = IPSEC_SPSTATE_ALIVE; V_spd_size++; V_sp_genid++; } /* * Insert a bunch of VTI secpolicies into the SPDB. * We keep VTI policies in the separate list due to following reasons: * 1) they should be immutable to user's or some deamon's attempts to * delete. The only way delete such policies - destroy or unconfigure * corresponding virtual inteface. * 2) such policies have traffic selector that matches all traffic per * address family. * Since all VTI policies have the same priority, we don't care about * policies order. */ int key_register_ifnet(struct secpolicy **spp, u_int count) { struct mbuf *m; u_int i; SPTREE_WLOCK(); /* * First of try to acquire id for each SP. */ for (i = 0; i < count; i++) { IPSEC_ASSERT(spp[i]->spidx.dir == IPSEC_DIR_INBOUND || spp[i]->spidx.dir == IPSEC_DIR_OUTBOUND, ("invalid direction %u", spp[i]->spidx.dir)); if ((spp[i]->id = key_getnewspid()) == 0) { SPTREE_WUNLOCK(); return (EAGAIN); } } for (i = 0; i < count; i++) { TAILQ_INSERT_TAIL(&V_sptree_ifnet[spp[i]->spidx.dir], spp[i], chain); /* * NOTE: despite the fact that we keep VTI SP in the * separate list, SPHASH contains policies from both * sources. Thus SADB_X_SPDGET will correctly return * SP by id, because it uses SPHASH for lookups. */ LIST_INSERT_HEAD(SPHASH_HASH(spp[i]->id), spp[i], idhash); spp[i]->state = IPSEC_SPSTATE_IFNET; } SPTREE_WUNLOCK(); /* * Notify user processes about new SP. */ for (i = 0; i < count; i++) { m = key_setdumpsp(spp[i], SADB_X_SPDADD, 0, 0); if (m != NULL) key_sendup_mbuf(NULL, m, KEY_SENDUP_ALL); } return (0); } void key_unregister_ifnet(struct secpolicy **spp, u_int count) { struct mbuf *m; u_int i; SPTREE_WLOCK(); for (i = 0; i < count; i++) { IPSEC_ASSERT(spp[i]->spidx.dir == IPSEC_DIR_INBOUND || spp[i]->spidx.dir == IPSEC_DIR_OUTBOUND, ("invalid direction %u", spp[i]->spidx.dir)); if (spp[i]->state != IPSEC_SPSTATE_IFNET) continue; spp[i]->state = IPSEC_SPSTATE_DEAD; TAILQ_REMOVE(&V_sptree_ifnet[spp[i]->spidx.dir], spp[i], chain); V_spd_size--; LIST_REMOVE(spp[i], idhash); } SPTREE_WUNLOCK(); if (SPDCACHE_ENABLED()) spdcache_clear(); for (i = 0; i < count; i++) { m = key_setdumpsp(spp[i], SADB_X_SPDDELETE, 0, 0); if (m != NULL) key_sendup_mbuf(NULL, m, KEY_SENDUP_ALL); } } /* * Must be called after calling key_allocsa(). * This function is called by key_freesp() to free some SA allocated * for a policy. */ void key_freesav(struct secasvar **psav) { struct secasvar *sav = *psav; IPSEC_ASSERT(sav != NULL, ("null sav")); if (SAV_DELREF(sav) == 0) return; KEYDBG(IPSEC_STAMP, printf("%s: last reference to SA(%p)\n", __func__, sav)); *psav = NULL; key_delsav(sav); } /* * Unlink SA from SAH and SPI hash under SAHTREE_WLOCK. * Expect that SA has extra reference due to lookup. * Release this references, also release SAH reference after unlink. */ static void key_unlinksav(struct secasvar *sav) { struct secashead *sah; KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); SAHTREE_UNLOCK_ASSERT(); SAHTREE_WLOCK(); if (sav->state == SADB_SASTATE_DEAD) { /* SA is already unlinked */ SAHTREE_WUNLOCK(); return; } /* Unlink from SAH */ if (sav->state == SADB_SASTATE_LARVAL) TAILQ_REMOVE(&sav->sah->savtree_larval, sav, chain); else TAILQ_REMOVE(&sav->sah->savtree_alive, sav, chain); /* Unlink from SPI hash */ LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; sah = sav->sah; SAHTREE_WUNLOCK(); key_freesav(&sav); /* Since we are unlinked, release reference to SAH */ key_freesah(&sah); } /* %%% SPD management */ /* * search SPD * OUT: NULL : not found * others : found, pointer to a SP. */ static struct secpolicy * key_getsp(struct secpolicyindex *spidx) { SPTREE_RLOCK_TRACKER; struct secpolicy *sp; IPSEC_ASSERT(spidx != NULL, ("null spidx")); SPTREE_RLOCK(); TAILQ_FOREACH(sp, &V_sptree[spidx->dir], chain) { if (key_cmpspidx_exactly(spidx, &sp->spidx)) { SP_ADDREF(sp); break; } } SPTREE_RUNLOCK(); return sp; } /* * get SP by index. * OUT: NULL : not found * others : found, pointer to referenced SP. */ static struct secpolicy * key_getspbyid(uint32_t id) { SPTREE_RLOCK_TRACKER; struct secpolicy *sp; SPTREE_RLOCK(); LIST_FOREACH(sp, SPHASH_HASH(id), idhash) { if (sp->id == id) { SP_ADDREF(sp); break; } } SPTREE_RUNLOCK(); return (sp); } struct secpolicy * key_newsp(void) { struct secpolicy *sp; sp = malloc(sizeof(*sp), M_IPSEC_SP, M_NOWAIT | M_ZERO); if (sp != NULL) SP_INITREF(sp); return (sp); } struct ipsecrequest * ipsec_newisr(void) { return (malloc(sizeof(struct ipsecrequest), M_IPSEC_SR, M_NOWAIT | M_ZERO)); } void ipsec_delisr(struct ipsecrequest *p) { free(p, M_IPSEC_SR); } /* * create secpolicy structure from sadb_x_policy structure. * NOTE: `state', `secpolicyindex' and 'id' in secpolicy structure * are not set, so must be set properly later. */ struct secpolicy * key_msg2sp(struct sadb_x_policy *xpl0, size_t len, int *error) { struct secpolicy *newsp; IPSEC_ASSERT(xpl0 != NULL, ("null xpl0")); IPSEC_ASSERT(len >= sizeof(*xpl0), ("policy too short: %zu", len)); if (len != PFKEY_EXTLEN(xpl0)) { ipseclog((LOG_DEBUG, "%s: Invalid msg length.\n", __func__)); *error = EINVAL; return NULL; } if ((newsp = key_newsp()) == NULL) { *error = ENOBUFS; return NULL; } newsp->spidx.dir = xpl0->sadb_x_policy_dir; newsp->policy = xpl0->sadb_x_policy_type; newsp->priority = xpl0->sadb_x_policy_priority; newsp->tcount = 0; /* check policy */ switch (xpl0->sadb_x_policy_type) { case IPSEC_POLICY_DISCARD: case IPSEC_POLICY_NONE: case IPSEC_POLICY_ENTRUST: case IPSEC_POLICY_BYPASS: break; case IPSEC_POLICY_IPSEC: { struct sadb_x_ipsecrequest *xisr; struct ipsecrequest *isr; int tlen; /* validity check */ if (PFKEY_EXTLEN(xpl0) < sizeof(*xpl0)) { ipseclog((LOG_DEBUG, "%s: Invalid msg length.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } tlen = PFKEY_EXTLEN(xpl0) - sizeof(*xpl0); xisr = (struct sadb_x_ipsecrequest *)(xpl0 + 1); while (tlen > 0) { /* length check */ if (xisr->sadb_x_ipsecrequest_len < sizeof(*xisr) || xisr->sadb_x_ipsecrequest_len > tlen) { ipseclog((LOG_DEBUG, "%s: invalid ipsecrequest " "length.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } if (newsp->tcount >= IPSEC_MAXREQ) { ipseclog((LOG_DEBUG, "%s: too many ipsecrequests.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return (NULL); } /* allocate request buffer */ /* NB: data structure is zero'd */ isr = ipsec_newisr(); if (isr == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); key_freesp(&newsp); *error = ENOBUFS; return NULL; } newsp->req[newsp->tcount++] = isr; /* set values */ switch (xisr->sadb_x_ipsecrequest_proto) { case IPPROTO_ESP: case IPPROTO_AH: case IPPROTO_IPCOMP: break; default: ipseclog((LOG_DEBUG, "%s: invalid proto type=%u\n", __func__, xisr->sadb_x_ipsecrequest_proto)); key_freesp(&newsp); *error = EPROTONOSUPPORT; return NULL; } isr->saidx.proto = (uint8_t)xisr->sadb_x_ipsecrequest_proto; switch (xisr->sadb_x_ipsecrequest_mode) { case IPSEC_MODE_TRANSPORT: case IPSEC_MODE_TUNNEL: break; case IPSEC_MODE_ANY: default: ipseclog((LOG_DEBUG, "%s: invalid mode=%u\n", __func__, xisr->sadb_x_ipsecrequest_mode)); key_freesp(&newsp); *error = EINVAL; return NULL; } isr->saidx.mode = xisr->sadb_x_ipsecrequest_mode; switch (xisr->sadb_x_ipsecrequest_level) { case IPSEC_LEVEL_DEFAULT: case IPSEC_LEVEL_USE: case IPSEC_LEVEL_REQUIRE: break; case IPSEC_LEVEL_UNIQUE: /* validity check */ /* * If range violation of reqid, kernel will * update it, don't refuse it. */ if (xisr->sadb_x_ipsecrequest_reqid > IPSEC_MANUAL_REQID_MAX) { ipseclog((LOG_DEBUG, "%s: reqid=%d range " "violation, updated by kernel.\n", __func__, xisr->sadb_x_ipsecrequest_reqid)); xisr->sadb_x_ipsecrequest_reqid = 0; } /* allocate new reqid id if reqid is zero. */ if (xisr->sadb_x_ipsecrequest_reqid == 0) { u_int32_t reqid; if ((reqid = key_newreqid()) == 0) { key_freesp(&newsp); *error = ENOBUFS; return NULL; } isr->saidx.reqid = reqid; xisr->sadb_x_ipsecrequest_reqid = reqid; } else { /* set it for manual keying. */ isr->saidx.reqid = xisr->sadb_x_ipsecrequest_reqid; } break; default: ipseclog((LOG_DEBUG, "%s: invalid level=%u\n", __func__, xisr->sadb_x_ipsecrequest_level)); key_freesp(&newsp); *error = EINVAL; return NULL; } isr->level = xisr->sadb_x_ipsecrequest_level; /* set IP addresses if there */ if (xisr->sadb_x_ipsecrequest_len > sizeof(*xisr)) { struct sockaddr *paddr; len = tlen - sizeof(*xisr); paddr = (struct sockaddr *)(xisr + 1); /* validity check */ if (len < sizeof(struct sockaddr) || len < 2 * paddr->sa_len || paddr->sa_len > sizeof(isr->saidx.src)) { ipseclog((LOG_DEBUG, "%s: invalid " "request address length.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } /* * Request length should be enough to keep * source and destination addresses. */ if (xisr->sadb_x_ipsecrequest_len < sizeof(*xisr) + 2 * paddr->sa_len) { ipseclog((LOG_DEBUG, "%s: invalid " "ipsecrequest length.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return (NULL); } bcopy(paddr, &isr->saidx.src, paddr->sa_len); paddr = (struct sockaddr *)((caddr_t)paddr + paddr->sa_len); /* validity check */ if (paddr->sa_len != isr->saidx.src.sa.sa_len) { ipseclog((LOG_DEBUG, "%s: invalid " "request address length.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } /* AF family should match */ if (paddr->sa_family != isr->saidx.src.sa.sa_family) { ipseclog((LOG_DEBUG, "%s: address " "family doesn't match.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return (NULL); } bcopy(paddr, &isr->saidx.dst, paddr->sa_len); } else { /* * Addresses for TUNNEL mode requests are * mandatory. */ if (isr->saidx.mode == IPSEC_MODE_TUNNEL) { ipseclog((LOG_DEBUG, "%s: missing " "request addresses.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return (NULL); } } tlen -= xisr->sadb_x_ipsecrequest_len; /* validity check */ if (tlen < 0) { ipseclog((LOG_DEBUG, "%s: becoming tlen < 0.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } xisr = (struct sadb_x_ipsecrequest *)((caddr_t)xisr + xisr->sadb_x_ipsecrequest_len); } /* XXXAE: LARVAL SP */ if (newsp->tcount < 1) { ipseclog((LOG_DEBUG, "%s: valid IPSEC transforms " "not found.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return (NULL); } } break; default: ipseclog((LOG_DEBUG, "%s: invalid policy type.\n", __func__)); key_freesp(&newsp); *error = EINVAL; return NULL; } *error = 0; return (newsp); } uint32_t key_newreqid(void) { static uint32_t auto_reqid = IPSEC_MANUAL_REQID_MAX + 1; if (auto_reqid == ~0) auto_reqid = IPSEC_MANUAL_REQID_MAX + 1; else auto_reqid++; /* XXX should be unique check */ return (auto_reqid); } /* * copy secpolicy struct to sadb_x_policy structure indicated. */ static struct mbuf * key_sp2mbuf(struct secpolicy *sp) { struct mbuf *m; size_t tlen; tlen = key_getspreqmsglen(sp); m = m_get2(tlen, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, tlen); m->m_len = tlen; if (key_sp2msg(sp, m->m_data, &tlen) != 0) { m_freem(m); return (NULL); } return (m); } int key_sp2msg(struct secpolicy *sp, void *request, size_t *len) { struct sadb_x_ipsecrequest *xisr; struct sadb_x_policy *xpl; struct ipsecrequest *isr; size_t xlen, ilen; caddr_t p; int error, i; IPSEC_ASSERT(sp != NULL, ("null policy")); xlen = sizeof(*xpl); if (*len < xlen) return (EINVAL); error = 0; bzero(request, *len); xpl = (struct sadb_x_policy *)request; xpl->sadb_x_policy_exttype = SADB_X_EXT_POLICY; xpl->sadb_x_policy_type = sp->policy; xpl->sadb_x_policy_dir = sp->spidx.dir; xpl->sadb_x_policy_id = sp->id; xpl->sadb_x_policy_priority = sp->priority; switch (sp->state) { case IPSEC_SPSTATE_IFNET: xpl->sadb_x_policy_scope = IPSEC_POLICYSCOPE_IFNET; break; case IPSEC_SPSTATE_PCB: xpl->sadb_x_policy_scope = IPSEC_POLICYSCOPE_PCB; break; default: xpl->sadb_x_policy_scope = IPSEC_POLICYSCOPE_GLOBAL; } /* if is the policy for ipsec ? */ if (sp->policy == IPSEC_POLICY_IPSEC) { p = (caddr_t)xpl + sizeof(*xpl); for (i = 0; i < sp->tcount; i++) { isr = sp->req[i]; ilen = PFKEY_ALIGN8(sizeof(*xisr) + isr->saidx.src.sa.sa_len + isr->saidx.dst.sa.sa_len); xlen += ilen; if (xlen > *len) { error = ENOBUFS; /* Calculate needed size */ continue; } xisr = (struct sadb_x_ipsecrequest *)p; xisr->sadb_x_ipsecrequest_len = ilen; xisr->sadb_x_ipsecrequest_proto = isr->saidx.proto; xisr->sadb_x_ipsecrequest_mode = isr->saidx.mode; xisr->sadb_x_ipsecrequest_level = isr->level; xisr->sadb_x_ipsecrequest_reqid = isr->saidx.reqid; p += sizeof(*xisr); bcopy(&isr->saidx.src, p, isr->saidx.src.sa.sa_len); p += isr->saidx.src.sa.sa_len; bcopy(&isr->saidx.dst, p, isr->saidx.dst.sa.sa_len); p += isr->saidx.dst.sa.sa_len; } } xpl->sadb_x_policy_len = PFKEY_UNIT64(xlen); if (error == 0) *len = xlen; else *len = sizeof(*xpl); return (error); } /* m will not be freed nor modified */ static struct mbuf * key_gather_mbuf(struct mbuf *m, const struct sadb_msghdr *mhp, int ndeep, int nitem, ...) { va_list ap; int idx; int i; struct mbuf *result = NULL, *n; int len; IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); va_start(ap, nitem); for (i = 0; i < nitem; i++) { idx = va_arg(ap, int); if (idx < 0 || idx > SADB_EXT_MAX) goto fail; /* don't attempt to pull empty extension */ if (idx == SADB_EXT_RESERVED && mhp->msg == NULL) continue; if (idx != SADB_EXT_RESERVED && (mhp->ext[idx] == NULL || mhp->extlen[idx] == 0)) continue; if (idx == SADB_EXT_RESERVED) { len = PFKEY_ALIGN8(sizeof(struct sadb_msg)); IPSEC_ASSERT(len <= MHLEN, ("header too big %u", len)); MGETHDR(n, M_NOWAIT, MT_DATA); if (!n) goto fail; n->m_len = len; n->m_next = NULL; m_copydata(m, 0, sizeof(struct sadb_msg), mtod(n, caddr_t)); } else if (i < ndeep) { len = mhp->extlen[idx]; n = m_get2(len, M_NOWAIT, MT_DATA, 0); if (n == NULL) goto fail; m_align(n, len); n->m_len = len; m_copydata(m, mhp->extoff[idx], mhp->extlen[idx], mtod(n, caddr_t)); } else { n = m_copym(m, mhp->extoff[idx], mhp->extlen[idx], M_NOWAIT); } if (n == NULL) goto fail; if (result) m_cat(result, n); else result = n; } va_end(ap); if ((result->m_flags & M_PKTHDR) != 0) { result->m_pkthdr.len = 0; for (n = result; n; n = n->m_next) result->m_pkthdr.len += n->m_len; } return result; fail: m_freem(result); va_end(ap); return NULL; } /* * SADB_X_SPDADD, SADB_X_SPDSETIDX or SADB_X_SPDUPDATE processing * add an entry to SP database, when received * * from the user(?). * Adding to SP database, * and send * * to the socket which was send. * * SPDADD set a unique policy entry. * SPDSETIDX like SPDADD without a part of policy requests. * SPDUPDATE replace a unique policy entry. * * XXXAE: serialize this in PF_KEY to avoid races. * m will always be freed. */ static int key_spdadd(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secpolicyindex spidx; struct sadb_address *src0, *dst0; struct sadb_x_policy *xpl0, *xpl; struct sadb_lifetime *lft = NULL; struct secpolicy *newsp; int error; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKHDR(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKLEN(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } if (!SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD)) { if (SADB_CHECKLEN(mhp, SADB_EXT_LIFETIME_HARD)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } lft = (struct sadb_lifetime *)mhp->ext[SADB_EXT_LIFETIME_HARD]; } src0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_SRC]; dst0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_DST]; xpl0 = (struct sadb_x_policy *)mhp->ext[SADB_X_EXT_POLICY]; /* check the direciton */ switch (xpl0->sadb_x_policy_dir) { case IPSEC_DIR_INBOUND: case IPSEC_DIR_OUTBOUND: break; default: ipseclog((LOG_DEBUG, "%s: invalid SP direction.\n", __func__)); return key_senderror(so, m, EINVAL); } /* key_spdadd() accepts DISCARD, NONE and IPSEC. */ if (xpl0->sadb_x_policy_type != IPSEC_POLICY_DISCARD && xpl0->sadb_x_policy_type != IPSEC_POLICY_NONE && xpl0->sadb_x_policy_type != IPSEC_POLICY_IPSEC) { ipseclog((LOG_DEBUG, "%s: invalid policy type.\n", __func__)); return key_senderror(so, m, EINVAL); } /* policy requests are mandatory when action is ipsec. */ if (xpl0->sadb_x_policy_type == IPSEC_POLICY_IPSEC && mhp->extlen[SADB_X_EXT_POLICY] <= sizeof(*xpl0)) { ipseclog((LOG_DEBUG, "%s: policy requests required.\n", __func__)); return key_senderror(so, m, EINVAL); } error = key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)); if (error != 0 || src0->sadb_address_proto != dst0->sadb_address_proto) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, error); } /* make secindex */ KEY_SETSECSPIDX(xpl0->sadb_x_policy_dir, src0 + 1, dst0 + 1, src0->sadb_address_prefixlen, dst0->sadb_address_prefixlen, src0->sadb_address_proto, &spidx); /* Checking there is SP already or not. */ newsp = key_getsp(&spidx); if (newsp != NULL) { if (mhp->msg->sadb_msg_type == SADB_X_SPDUPDATE) { KEYDBG(KEY_STAMP, printf("%s: unlink SP(%p) for SPDUPDATE\n", __func__, newsp)); KEYDBG(KEY_DATA, kdebug_secpolicy(newsp)); key_unlink(newsp); key_freesp(&newsp); } else { key_freesp(&newsp); ipseclog((LOG_DEBUG, "%s: a SP entry exists already.", __func__)); return (key_senderror(so, m, EEXIST)); } } /* allocate new SP entry */ if ((newsp = key_msg2sp(xpl0, PFKEY_EXTLEN(xpl0), &error)) == NULL) { return key_senderror(so, m, error); } newsp->lastused = newsp->created = time_second; newsp->lifetime = lft ? lft->sadb_lifetime_addtime : 0; newsp->validtime = lft ? lft->sadb_lifetime_usetime : 0; bcopy(&spidx, &newsp->spidx, sizeof(spidx)); /* XXXAE: there is race between key_getsp() and key_insertsp() */ SPTREE_WLOCK(); if ((newsp->id = key_getnewspid()) == 0) { SPTREE_WUNLOCK(); key_freesp(&newsp); return key_senderror(so, m, ENOBUFS); } key_insertsp(newsp); SPTREE_WUNLOCK(); if (SPDCACHE_ENABLED()) spdcache_clear(); KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, newsp)); KEYDBG(KEY_DATA, kdebug_secpolicy(newsp)); { struct mbuf *n, *mpolicy; struct sadb_msg *newmsg; int off; /* create new sadb_msg to reply. */ if (lft) { n = key_gather_mbuf(m, mhp, 2, 5, SADB_EXT_RESERVED, SADB_X_EXT_POLICY, SADB_EXT_LIFETIME_HARD, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); } else { n = key_gather_mbuf(m, mhp, 2, 4, SADB_EXT_RESERVED, SADB_X_EXT_POLICY, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); } if (!n) return key_senderror(so, m, ENOBUFS); if (n->m_len < sizeof(*newmsg)) { n = m_pullup(n, sizeof(*newmsg)); if (!n) return key_senderror(so, m, ENOBUFS); } newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); off = 0; mpolicy = m_pulldown(n, PFKEY_ALIGN8(sizeof(struct sadb_msg)), sizeof(*xpl), &off); if (mpolicy == NULL) { /* n is already freed */ return key_senderror(so, m, ENOBUFS); } xpl = (struct sadb_x_policy *)(mtod(mpolicy, caddr_t) + off); if (xpl->sadb_x_policy_exttype != SADB_X_EXT_POLICY) { m_freem(n); return key_senderror(so, m, EINVAL); } xpl->sadb_x_policy_id = newsp->id; m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * get new policy id. * OUT: * 0: failure. * others: success. */ static uint32_t key_getnewspid(void) { struct secpolicy *sp; uint32_t newid = 0; int count = V_key_spi_trycnt; /* XXX */ SPTREE_WLOCK_ASSERT(); while (count--) { if (V_policy_id == ~0) /* overflowed */ newid = V_policy_id = 1; else newid = ++V_policy_id; LIST_FOREACH(sp, SPHASH_HASH(newid), idhash) { if (sp->id == newid) break; } if (sp == NULL) break; } if (count == 0 || newid == 0) { ipseclog((LOG_DEBUG, "%s: failed to allocate policy id.\n", __func__)); return (0); } return (newid); } /* * SADB_SPDDELETE processing * receive * * from the user(?), and set SADB_SASTATE_DEAD, * and send, * * to the ikmpd. * policy(*) including direction of policy. * * m will always be freed. */ static int key_spddelete(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secpolicyindex spidx; struct sadb_address *src0, *dst0; struct sadb_x_policy *xpl0; struct secpolicy *sp; IPSEC_ASSERT(so != NULL, ("null so")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKHDR(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKLEN(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } src0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_SRC]; dst0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_DST]; xpl0 = (struct sadb_x_policy *)mhp->ext[SADB_X_EXT_POLICY]; /* check the direciton */ switch (xpl0->sadb_x_policy_dir) { case IPSEC_DIR_INBOUND: case IPSEC_DIR_OUTBOUND: break; default: ipseclog((LOG_DEBUG, "%s: invalid SP direction.\n", __func__)); return key_senderror(so, m, EINVAL); } /* Only DISCARD, NONE and IPSEC are allowed */ if (xpl0->sadb_x_policy_type != IPSEC_POLICY_DISCARD && xpl0->sadb_x_policy_type != IPSEC_POLICY_NONE && xpl0->sadb_x_policy_type != IPSEC_POLICY_IPSEC) { ipseclog((LOG_DEBUG, "%s: invalid policy type.\n", __func__)); return key_senderror(so, m, EINVAL); } if (key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)) != 0 || src0->sadb_address_proto != dst0->sadb_address_proto) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, EINVAL); } /* make secindex */ KEY_SETSECSPIDX(xpl0->sadb_x_policy_dir, src0 + 1, dst0 + 1, src0->sadb_address_prefixlen, dst0->sadb_address_prefixlen, src0->sadb_address_proto, &spidx); /* Is there SP in SPD ? */ if ((sp = key_getsp(&spidx)) == NULL) { ipseclog((LOG_DEBUG, "%s: no SP found.\n", __func__)); return key_senderror(so, m, EINVAL); } /* save policy id to buffer to be returned. */ xpl0->sadb_x_policy_id = sp->id; KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, sp)); KEYDBG(KEY_DATA, kdebug_secpolicy(sp)); key_unlink(sp); key_freesp(&sp); { struct mbuf *n; struct sadb_msg *newmsg; /* create new sadb_msg to reply. */ n = key_gather_mbuf(m, mhp, 1, 4, SADB_EXT_RESERVED, SADB_X_EXT_POLICY, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); if (!n) return key_senderror(so, m, ENOBUFS); newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * SADB_SPDDELETE2 processing * receive * * from the user(?), and set SADB_SASTATE_DEAD, * and send, * * to the ikmpd. * policy(*) including direction of policy. * * m will always be freed. */ static int key_spddelete2(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secpolicy *sp; uint32_t id; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (SADB_CHECKHDR(mhp, SADB_X_EXT_POLICY) || SADB_CHECKLEN(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } id = ((struct sadb_x_policy *) mhp->ext[SADB_X_EXT_POLICY])->sadb_x_policy_id; /* Is there SP in SPD ? */ if ((sp = key_getspbyid(id)) == NULL) { ipseclog((LOG_DEBUG, "%s: no SP found for id %u.\n", __func__, id)); return key_senderror(so, m, EINVAL); } KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, sp)); KEYDBG(KEY_DATA, kdebug_secpolicy(sp)); key_unlink(sp); if (sp->state != IPSEC_SPSTATE_DEAD) { ipseclog((LOG_DEBUG, "%s: failed to delete SP with id %u.\n", __func__, id)); key_freesp(&sp); return (key_senderror(so, m, EACCES)); } key_freesp(&sp); { struct mbuf *n, *nn; struct sadb_msg *newmsg; int off, len; /* create new sadb_msg to reply. */ len = PFKEY_ALIGN8(sizeof(struct sadb_msg)); MGETHDR(n, M_NOWAIT, MT_DATA); if (n && len > MHLEN) { if (!(MCLGET(n, M_NOWAIT))) { m_freem(n); n = NULL; } } if (!n) return key_senderror(so, m, ENOBUFS); n->m_len = len; n->m_next = NULL; off = 0; m_copydata(m, 0, sizeof(struct sadb_msg), mtod(n, caddr_t) + off); off += PFKEY_ALIGN8(sizeof(struct sadb_msg)); IPSEC_ASSERT(off == len, ("length inconsistency (off %u len %u)", off, len)); n->m_next = m_copym(m, mhp->extoff[SADB_X_EXT_POLICY], mhp->extlen[SADB_X_EXT_POLICY], M_NOWAIT); if (!n->m_next) { m_freem(n); return key_senderror(so, m, ENOBUFS); } n->m_pkthdr.len = 0; for (nn = n; nn; nn = nn->m_next) n->m_pkthdr.len += nn->m_len; newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * SADB_X_SPDGET processing * receive * * from the user(?), * and send, * * to the ikmpd. * policy(*) including direction of policy. * * m will always be freed. */ static int key_spdget(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secpolicy *sp; struct mbuf *n; uint32_t id; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (SADB_CHECKHDR(mhp, SADB_X_EXT_POLICY) || SADB_CHECKLEN(mhp, SADB_X_EXT_POLICY)) { ipseclog((LOG_DEBUG, "%s: invalid message is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } id = ((struct sadb_x_policy *) mhp->ext[SADB_X_EXT_POLICY])->sadb_x_policy_id; /* Is there SP in SPD ? */ if ((sp = key_getspbyid(id)) == NULL) { ipseclog((LOG_DEBUG, "%s: no SP found for id %u.\n", __func__, id)); return key_senderror(so, m, ENOENT); } n = key_setdumpsp(sp, SADB_X_SPDGET, mhp->msg->sadb_msg_seq, mhp->msg->sadb_msg_pid); key_freesp(&sp); if (n != NULL) { m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } else return key_senderror(so, m, ENOBUFS); } /* * SADB_X_SPDACQUIRE processing. * Acquire policy and SA(s) for a *OUTBOUND* packet. * send * * to KMD, and expect to receive * with SADB_X_SPDACQUIRE if error occurred, * or * * with SADB_X_SPDUPDATE from KMD by PF_KEY. * policy(*) is without policy requests. * * 0 : succeed * others: error number */ int key_spdacquire(struct secpolicy *sp) { struct mbuf *result = NULL, *m; struct secspacq *newspacq; IPSEC_ASSERT(sp != NULL, ("null secpolicy")); IPSEC_ASSERT(sp->req == NULL, ("policy exists")); IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC, ("policy not IPSEC %u", sp->policy)); /* Get an entry to check whether sent message or not. */ newspacq = key_getspacq(&sp->spidx); if (newspacq != NULL) { if (V_key_blockacq_count < newspacq->count) { /* reset counter and do send message. */ newspacq->count = 0; } else { /* increment counter and do nothing. */ newspacq->count++; SPACQ_UNLOCK(); return (0); } SPACQ_UNLOCK(); } else { /* make new entry for blocking to send SADB_ACQUIRE. */ newspacq = key_newspacq(&sp->spidx); if (newspacq == NULL) return ENOBUFS; } /* create new sadb_msg to reply. */ m = key_setsadbmsg(SADB_X_SPDACQUIRE, 0, 0, 0, 0, 0); if (!m) return ENOBUFS; result = m; result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); return key_sendup_mbuf(NULL, m, KEY_SENDUP_REGISTERED); } /* * SADB_SPDFLUSH processing * receive * * from the user, and free all entries in secpctree. * and send, * * to the user. * NOTE: what to do is only marking SADB_SASTATE_DEAD. * * m will always be freed. */ static int key_spdflush(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secpolicy_queue drainq; struct sadb_msg *newmsg; struct secpolicy *sp, *nextsp; u_int dir; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (m->m_len != PFKEY_ALIGN8(sizeof(struct sadb_msg))) return key_senderror(so, m, EINVAL); TAILQ_INIT(&drainq); SPTREE_WLOCK(); for (dir = 0; dir < IPSEC_DIR_MAX; dir++) { TAILQ_CONCAT(&drainq, &V_sptree[dir], chain); } /* * We need to set state to DEAD for each policy to be sure, * that another thread won't try to unlink it. * Also remove SP from sphash. */ TAILQ_FOREACH(sp, &drainq, chain) { sp->state = IPSEC_SPSTATE_DEAD; LIST_REMOVE(sp, idhash); } V_sp_genid++; V_spd_size = 0; SPTREE_WUNLOCK(); if (SPDCACHE_ENABLED()) spdcache_clear(); sp = TAILQ_FIRST(&drainq); while (sp != NULL) { nextsp = TAILQ_NEXT(sp, chain); key_freesp(&sp); sp = nextsp; } if (sizeof(struct sadb_msg) > m->m_len + M_TRAILINGSPACE(m)) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return key_senderror(so, m, ENOBUFS); } if (m->m_next) m_freem(m->m_next); m->m_next = NULL; m->m_pkthdr.len = m->m_len = PFKEY_ALIGN8(sizeof(struct sadb_msg)); newmsg = mtod(m, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(m->m_pkthdr.len); return key_sendup_mbuf(so, m, KEY_SENDUP_ALL); } static uint8_t key_satype2scopemask(uint8_t satype) { if (satype == IPSEC_POLICYSCOPE_ANY) return (0xff); return (satype); } /* * SADB_SPDDUMP processing * receive * * from the user, and dump all SP leaves and send, * ..... * to the ikmpd. * * NOTE: * sadb_msg_satype is considered as mask of policy scopes. * m will always be freed. */ static int key_spddump(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { SPTREE_RLOCK_TRACKER; struct secpolicy *sp; struct mbuf *n; int cnt; u_int dir, scope; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* search SPD entry and get buffer size. */ cnt = 0; scope = key_satype2scopemask(mhp->msg->sadb_msg_satype); SPTREE_RLOCK(); for (dir = 0; dir < IPSEC_DIR_MAX; dir++) { if (scope & IPSEC_POLICYSCOPE_GLOBAL) { TAILQ_FOREACH(sp, &V_sptree[dir], chain) cnt++; } if (scope & IPSEC_POLICYSCOPE_IFNET) { TAILQ_FOREACH(sp, &V_sptree_ifnet[dir], chain) cnt++; } } if (cnt == 0) { SPTREE_RUNLOCK(); return key_senderror(so, m, ENOENT); } for (dir = 0; dir < IPSEC_DIR_MAX; dir++) { if (scope & IPSEC_POLICYSCOPE_GLOBAL) { TAILQ_FOREACH(sp, &V_sptree[dir], chain) { --cnt; n = key_setdumpsp(sp, SADB_X_SPDDUMP, cnt, mhp->msg->sadb_msg_pid); if (n != NULL) key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } } if (scope & IPSEC_POLICYSCOPE_IFNET) { TAILQ_FOREACH(sp, &V_sptree_ifnet[dir], chain) { --cnt; n = key_setdumpsp(sp, SADB_X_SPDDUMP, cnt, mhp->msg->sadb_msg_pid); if (n != NULL) key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } } } SPTREE_RUNLOCK(); m_freem(m); return (0); } static struct mbuf * key_setdumpsp(struct secpolicy *sp, u_int8_t type, u_int32_t seq, u_int32_t pid) { struct mbuf *result = NULL, *m; struct seclifetime lt; m = key_setsadbmsg(type, 0, SADB_SATYPE_UNSPEC, seq, pid, sp->refcnt); if (!m) goto fail; result = m; m = key_setsadbaddr(SADB_EXT_ADDRESS_SRC, &sp->spidx.src.sa, sp->spidx.prefs, sp->spidx.ul_proto); if (!m) goto fail; m_cat(result, m); m = key_setsadbaddr(SADB_EXT_ADDRESS_DST, &sp->spidx.dst.sa, sp->spidx.prefd, sp->spidx.ul_proto); if (!m) goto fail; m_cat(result, m); m = key_sp2mbuf(sp); if (!m) goto fail; m_cat(result, m); if(sp->lifetime){ lt.addtime=sp->created; lt.usetime= sp->lastused; m = key_setlifetime(<, SADB_EXT_LIFETIME_CURRENT); if (!m) goto fail; m_cat(result, m); lt.addtime=sp->lifetime; lt.usetime= sp->validtime; m = key_setlifetime(<, SADB_EXT_LIFETIME_HARD); if (!m) goto fail; m_cat(result, m); } if ((result->m_flags & M_PKTHDR) == 0) goto fail; if (result->m_len < sizeof(struct sadb_msg)) { result = m_pullup(result, sizeof(struct sadb_msg)); if (result == NULL) goto fail; } result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); return result; fail: m_freem(result); return NULL; } /* * get PFKEY message length for security policy and request. */ static size_t key_getspreqmsglen(struct secpolicy *sp) { size_t tlen, len; int i; tlen = sizeof(struct sadb_x_policy); /* if is the policy for ipsec ? */ if (sp->policy != IPSEC_POLICY_IPSEC) return (tlen); /* get length of ipsec requests */ for (i = 0; i < sp->tcount; i++) { len = sizeof(struct sadb_x_ipsecrequest) + sp->req[i]->saidx.src.sa.sa_len + sp->req[i]->saidx.dst.sa.sa_len; tlen += PFKEY_ALIGN8(len); } return (tlen); } /* * SADB_SPDEXPIRE processing * send * * to KMD by PF_KEY. * * OUT: 0 : succeed * others : error number */ static int key_spdexpire(struct secpolicy *sp) { struct sadb_lifetime *lt; struct mbuf *result = NULL, *m; int len, error = -1; IPSEC_ASSERT(sp != NULL, ("null secpolicy")); KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, sp)); KEYDBG(KEY_DATA, kdebug_secpolicy(sp)); /* set msg header */ m = key_setsadbmsg(SADB_X_SPDEXPIRE, 0, 0, 0, 0, 0); if (!m) { error = ENOBUFS; goto fail; } result = m; /* create lifetime extension (current and hard) */ len = PFKEY_ALIGN8(sizeof(*lt)) * 2; m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) { error = ENOBUFS; goto fail; } m_align(m, len); m->m_len = len; bzero(mtod(m, caddr_t), len); lt = mtod(m, struct sadb_lifetime *); lt->sadb_lifetime_len = PFKEY_UNIT64(sizeof(struct sadb_lifetime)); lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT; lt->sadb_lifetime_allocations = 0; lt->sadb_lifetime_bytes = 0; lt->sadb_lifetime_addtime = sp->created; lt->sadb_lifetime_usetime = sp->lastused; lt = (struct sadb_lifetime *)(mtod(m, caddr_t) + len / 2); lt->sadb_lifetime_len = PFKEY_UNIT64(sizeof(struct sadb_lifetime)); lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD; lt->sadb_lifetime_allocations = 0; lt->sadb_lifetime_bytes = 0; lt->sadb_lifetime_addtime = sp->lifetime; lt->sadb_lifetime_usetime = sp->validtime; m_cat(result, m); /* set sadb_address for source */ m = key_setsadbaddr(SADB_EXT_ADDRESS_SRC, &sp->spidx.src.sa, sp->spidx.prefs, sp->spidx.ul_proto); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* set sadb_address for destination */ m = key_setsadbaddr(SADB_EXT_ADDRESS_DST, &sp->spidx.dst.sa, sp->spidx.prefd, sp->spidx.ul_proto); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* set secpolicy */ m = key_sp2mbuf(sp); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); if ((result->m_flags & M_PKTHDR) == 0) { error = EINVAL; goto fail; } if (result->m_len < sizeof(struct sadb_msg)) { result = m_pullup(result, sizeof(struct sadb_msg)); if (result == NULL) { error = ENOBUFS; goto fail; } } result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); return key_sendup_mbuf(NULL, result, KEY_SENDUP_REGISTERED); fail: if (result) m_freem(result); return error; } /* %%% SAD management */ /* * allocating and initialize new SA head. * OUT: NULL : failure due to the lack of memory. * others : pointer to new SA head. */ static struct secashead * key_newsah(struct secasindex *saidx) { struct secashead *sah; sah = malloc(sizeof(struct secashead), M_IPSEC_SAH, M_NOWAIT | M_ZERO); if (sah == NULL) { PFKEYSTAT_INC(in_nomem); return (NULL); } TAILQ_INIT(&sah->savtree_larval); TAILQ_INIT(&sah->savtree_alive); sah->saidx = *saidx; sah->state = SADB_SASTATE_DEAD; SAH_INITREF(sah); KEYDBG(KEY_STAMP, printf("%s: SAH(%p)\n", __func__, sah)); KEYDBG(KEY_DATA, kdebug_secash(sah, NULL)); return (sah); } static void key_freesah(struct secashead **psah) { struct secashead *sah = *psah; if (SAH_DELREF(sah) == 0) return; KEYDBG(KEY_STAMP, printf("%s: last reference to SAH(%p)\n", __func__, sah)); KEYDBG(KEY_DATA, kdebug_secash(sah, NULL)); *psah = NULL; key_delsah(sah); } static void key_delsah(struct secashead *sah) { IPSEC_ASSERT(sah != NULL, ("NULL sah")); IPSEC_ASSERT(sah->state == SADB_SASTATE_DEAD, ("Attempt to free non DEAD SAH %p", sah)); IPSEC_ASSERT(TAILQ_EMPTY(&sah->savtree_larval), ("Attempt to free SAH %p with LARVAL SA", sah)); IPSEC_ASSERT(TAILQ_EMPTY(&sah->savtree_alive), ("Attempt to free SAH %p with ALIVE SA", sah)); free(sah, M_IPSEC_SAH); } /* * allocating a new SA for key_add() and key_getspi() call, * and copy the values of mhp into new buffer. * When SAD message type is SADB_GETSPI set SA state to LARVAL. * For SADB_ADD create and initialize SA with MATURE state. * OUT: NULL : fail * others : pointer to new secasvar. */ static struct secasvar * key_newsav(const struct sadb_msghdr *mhp, struct secasindex *saidx, uint32_t spi, int *errp) { struct secashead *sah; struct secasvar *sav; int isnew; IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); IPSEC_ASSERT(mhp->msg->sadb_msg_type == SADB_GETSPI || mhp->msg->sadb_msg_type == SADB_ADD, ("wrong message type")); sav = NULL; sah = NULL; /* check SPI value */ switch (saidx->proto) { case IPPROTO_ESP: case IPPROTO_AH: /* * RFC 4302, 2.4. Security Parameters Index (SPI), SPI values * 1-255 reserved by IANA for future use, * 0 for implementation specific, local use. */ if (ntohl(spi) <= 255) { ipseclog((LOG_DEBUG, "%s: illegal range of SPI %u.\n", __func__, ntohl(spi))); *errp = EINVAL; goto done; } break; } sav = malloc(sizeof(struct secasvar), M_IPSEC_SA, M_NOWAIT | M_ZERO); if (sav == NULL) { *errp = ENOBUFS; goto done; } sav->lock = malloc(sizeof(struct mtx), M_IPSEC_MISC, M_NOWAIT | M_ZERO); if (sav->lock == NULL) { *errp = ENOBUFS; goto done; } mtx_init(sav->lock, "ipsec association", NULL, MTX_DEF); sav->lft_c = uma_zalloc_pcpu(V_key_lft_zone, M_NOWAIT); if (sav->lft_c == NULL) { *errp = ENOBUFS; goto done; } counter_u64_zero(sav->lft_c_allocations); counter_u64_zero(sav->lft_c_bytes); sav->spi = spi; sav->seq = mhp->msg->sadb_msg_seq; sav->state = SADB_SASTATE_LARVAL; sav->pid = (pid_t)mhp->msg->sadb_msg_pid; SAV_INITREF(sav); again: sah = key_getsah(saidx); if (sah == NULL) { /* create a new SA index */ sah = key_newsah(saidx); if (sah == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); *errp = ENOBUFS; goto done; } isnew = 1; } else isnew = 0; sav->sah = sah; if (mhp->msg->sadb_msg_type == SADB_GETSPI) { sav->created = time_second; } else if (sav->state == SADB_SASTATE_LARVAL) { /* * Do not call key_setsaval() second time in case * of `goto again`. We will have MATURE state. */ *errp = key_setsaval(sav, mhp); if (*errp != 0) goto done; sav->state = SADB_SASTATE_MATURE; } SAHTREE_WLOCK(); /* * Check that existing SAH wasn't unlinked. * Since we didn't hold the SAHTREE lock, it is possible, * that callout handler or key_flush() or key_delete() could * unlink this SAH. */ if (isnew == 0 && sah->state == SADB_SASTATE_DEAD) { SAHTREE_WUNLOCK(); key_freesah(&sah); /* reference from key_getsah() */ goto again; } if (isnew != 0) { /* * Add new SAH into SADB. * * XXXAE: we can serialize key_add and key_getspi calls, so * several threads will not fight in the race. * Otherwise we should check under SAHTREE lock, that this * SAH would not added twice. */ TAILQ_INSERT_HEAD(&V_sahtree, sah, chain); /* Add new SAH into hash by addresses */ LIST_INSERT_HEAD(SAHADDRHASH_HASH(saidx), sah, addrhash); /* Now we are linked in the chain */ sah->state = SADB_SASTATE_MATURE; /* * SAV references this new SAH. * In case of existing SAH we reuse reference * from key_getsah(). */ SAH_ADDREF(sah); } /* Link SAV with SAH */ if (sav->state == SADB_SASTATE_MATURE) TAILQ_INSERT_HEAD(&sah->savtree_alive, sav, chain); else TAILQ_INSERT_HEAD(&sah->savtree_larval, sav, chain); /* Add SAV into SPI hash */ LIST_INSERT_HEAD(SAVHASH_HASH(sav->spi), sav, spihash); SAHTREE_WUNLOCK(); *errp = 0; /* success */ done: if (*errp != 0) { if (sav != NULL) { if (sav->lock != NULL) { mtx_destroy(sav->lock); free(sav->lock, M_IPSEC_MISC); } if (sav->lft_c != NULL) uma_zfree_pcpu(V_key_lft_zone, sav->lft_c); free(sav, M_IPSEC_SA), sav = NULL; } if (sah != NULL) key_freesah(&sah); if (*errp == ENOBUFS) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); PFKEYSTAT_INC(in_nomem); } } return (sav); } /* * free() SA variable entry. */ static void key_cleansav(struct secasvar *sav) { if (sav->natt != NULL) { free(sav->natt, M_IPSEC_MISC); sav->natt = NULL; } if (sav->flags & SADB_X_EXT_F_CLONED) return; /* * Cleanup xform state. Note that zeroize'ing causes the * keys to be cleared; otherwise we must do it ourself. */ if (sav->tdb_xform != NULL) { sav->tdb_xform->xf_zeroize(sav); sav->tdb_xform = NULL; } else { if (sav->key_auth != NULL) bzero(sav->key_auth->key_data, _KEYLEN(sav->key_auth)); if (sav->key_enc != NULL) bzero(sav->key_enc->key_data, _KEYLEN(sav->key_enc)); } if (sav->key_auth != NULL) { if (sav->key_auth->key_data != NULL) free(sav->key_auth->key_data, M_IPSEC_MISC); free(sav->key_auth, M_IPSEC_MISC); sav->key_auth = NULL; } if (sav->key_enc != NULL) { if (sav->key_enc->key_data != NULL) free(sav->key_enc->key_data, M_IPSEC_MISC); free(sav->key_enc, M_IPSEC_MISC); sav->key_enc = NULL; } if (sav->replay != NULL) { if (sav->replay->bitmap != NULL) free(sav->replay->bitmap, M_IPSEC_MISC); free(sav->replay, M_IPSEC_MISC); sav->replay = NULL; } if (sav->lft_h != NULL) { free(sav->lft_h, M_IPSEC_MISC); sav->lft_h = NULL; } if (sav->lft_s != NULL) { free(sav->lft_s, M_IPSEC_MISC); sav->lft_s = NULL; } } /* * free() SA variable entry. */ static void key_delsav(struct secasvar *sav) { IPSEC_ASSERT(sav != NULL, ("null sav")); IPSEC_ASSERT(sav->state == SADB_SASTATE_DEAD, ("attempt to free non DEAD SA %p", sav)); IPSEC_ASSERT(sav->refcnt == 0, ("reference count %u > 0", sav->refcnt)); /* * SA must be unlinked from the chain and hashtbl. * If SA was cloned, we leave all fields untouched, * except NAT-T config. */ key_cleansav(sav); if ((sav->flags & SADB_X_EXT_F_CLONED) == 0) { mtx_destroy(sav->lock); free(sav->lock, M_IPSEC_MISC); uma_zfree(V_key_lft_zone, sav->lft_c); } free(sav, M_IPSEC_SA); } /* * search SAH. * OUT: * NULL : not found * others : found, referenced pointer to a SAH. */ static struct secashead * key_getsah(struct secasindex *saidx) { SAHTREE_RLOCK_TRACKER; struct secashead *sah; SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(saidx), addrhash) { if (key_cmpsaidx(&sah->saidx, saidx, CMP_MODE_REQID) != 0) { SAH_ADDREF(sah); break; } } SAHTREE_RUNLOCK(); return (sah); } /* * Check not to be duplicated SPI. * OUT: * 0 : not found * 1 : found SA with given SPI. */ static int key_checkspidup(uint32_t spi) { SAHTREE_RLOCK_TRACKER; struct secasvar *sav; /* Assume SPI is in network byte order */ SAHTREE_RLOCK(); LIST_FOREACH(sav, SAVHASH_HASH(spi), spihash) { if (sav->spi == spi) break; } SAHTREE_RUNLOCK(); return (sav != NULL); } /* * Search SA by SPI. * OUT: * NULL : not found * others : found, referenced pointer to a SA. */ static struct secasvar * key_getsavbyspi(uint32_t spi) { SAHTREE_RLOCK_TRACKER; struct secasvar *sav; /* Assume SPI is in network byte order */ SAHTREE_RLOCK(); LIST_FOREACH(sav, SAVHASH_HASH(spi), spihash) { if (sav->spi != spi) continue; SAV_ADDREF(sav); break; } SAHTREE_RUNLOCK(); return (sav); } static int key_updatelifetimes(struct secasvar *sav, const struct sadb_msghdr *mhp) { struct seclifetime *lft_h, *lft_s, *tmp; /* Lifetime extension is optional, check that it is present. */ if (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD) && SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT)) { /* * In case of SADB_UPDATE we may need to change * existing lifetimes. */ if (sav->state == SADB_SASTATE_MATURE) { lft_h = lft_s = NULL; goto reset; } return (0); } /* Both HARD and SOFT extensions must present */ if ((SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT)) || (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD))) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return (EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_LIFETIME_HARD) || SADB_CHECKLEN(mhp, SADB_EXT_LIFETIME_SOFT)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return (EINVAL); } lft_h = key_dup_lifemsg((const struct sadb_lifetime *) mhp->ext[SADB_EXT_LIFETIME_HARD], M_IPSEC_MISC); if (lft_h == NULL) { PFKEYSTAT_INC(in_nomem); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (ENOBUFS); } lft_s = key_dup_lifemsg((const struct sadb_lifetime *) mhp->ext[SADB_EXT_LIFETIME_SOFT], M_IPSEC_MISC); if (lft_s == NULL) { PFKEYSTAT_INC(in_nomem); free(lft_h, M_IPSEC_MISC); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (ENOBUFS); } reset: if (sav->state != SADB_SASTATE_LARVAL) { /* * key_update() holds reference to this SA, * so it won't be deleted in meanwhile. */ SECASVAR_LOCK(sav); tmp = sav->lft_h; sav->lft_h = lft_h; lft_h = tmp; tmp = sav->lft_s; sav->lft_s = lft_s; lft_s = tmp; SECASVAR_UNLOCK(sav); if (lft_h != NULL) free(lft_h, M_IPSEC_MISC); if (lft_s != NULL) free(lft_s, M_IPSEC_MISC); return (0); } /* We can update lifetime without holding a lock */ IPSEC_ASSERT(sav->lft_h == NULL, ("lft_h is already initialized\n")); IPSEC_ASSERT(sav->lft_s == NULL, ("lft_s is already initialized\n")); sav->lft_h = lft_h; sav->lft_s = lft_s; return (0); } /* * copy SA values from PF_KEY message except *SPI, SEQ, PID and TYPE*. * You must update these if need. Expects only LARVAL SAs. * OUT: 0: success. * !0: failure. */ static int key_setsaval(struct secasvar *sav, const struct sadb_msghdr *mhp) { const struct sadb_sa *sa0; const struct sadb_key *key0; uint32_t replay; size_t len; int error; IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); IPSEC_ASSERT(sav->state == SADB_SASTATE_LARVAL, ("Attempt to update non LARVAL SA")); /* XXX rewrite */ error = key_setident(sav->sah, mhp); if (error != 0) goto fail; /* SA */ if (!SADB_CHECKHDR(mhp, SADB_EXT_SA)) { if (SADB_CHECKLEN(mhp, SADB_EXT_SA)) { error = EINVAL; goto fail; } sa0 = (const struct sadb_sa *)mhp->ext[SADB_EXT_SA]; sav->alg_auth = sa0->sadb_sa_auth; sav->alg_enc = sa0->sadb_sa_encrypt; sav->flags = sa0->sadb_sa_flags; if ((sav->flags & SADB_KEY_FLAGS_MAX) != sav->flags) { ipseclog((LOG_DEBUG, "%s: invalid sa_flags 0x%08x.\n", __func__, sav->flags)); error = EINVAL; goto fail; } /* Optional replay window */ replay = 0; if ((sa0->sadb_sa_flags & SADB_X_EXT_OLD) == 0) replay = sa0->sadb_sa_replay; if (!SADB_CHECKHDR(mhp, SADB_X_EXT_SA_REPLAY)) { if (SADB_CHECKLEN(mhp, SADB_X_EXT_SA_REPLAY)) { error = EINVAL; goto fail; } replay = ((const struct sadb_x_sa_replay *) mhp->ext[SADB_X_EXT_SA_REPLAY])->sadb_x_sa_replay_replay; if (replay > UINT32_MAX - 32) { ipseclog((LOG_DEBUG, "%s: replay window too big.\n", __func__)); error = EINVAL; goto fail; } replay = (replay + 7) >> 3; } sav->replay = malloc(sizeof(struct secreplay), M_IPSEC_MISC, M_NOWAIT | M_ZERO); if (sav->replay == NULL) { PFKEYSTAT_INC(in_nomem); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); error = ENOBUFS; goto fail; } if (replay != 0) { /* number of 32b blocks to be allocated */ uint32_t bitmap_size; /* RFC 6479: * - the allocated replay window size must be * a power of two. * - use an extra 32b block as a redundant window. */ bitmap_size = 1; while (replay + 4 > bitmap_size) bitmap_size <<= 1; bitmap_size = bitmap_size / 4; sav->replay->bitmap = malloc( bitmap_size * sizeof(uint32_t), M_IPSEC_MISC, M_NOWAIT | M_ZERO); if (sav->replay->bitmap == NULL) { PFKEYSTAT_INC(in_nomem); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); error = ENOBUFS; goto fail; } sav->replay->bitmap_size = bitmap_size; sav->replay->wsize = replay; } } /* Authentication keys */ if (!SADB_CHECKHDR(mhp, SADB_EXT_KEY_AUTH)) { if (SADB_CHECKLEN(mhp, SADB_EXT_KEY_AUTH)) { error = EINVAL; goto fail; } error = 0; key0 = (const struct sadb_key *)mhp->ext[SADB_EXT_KEY_AUTH]; len = mhp->extlen[SADB_EXT_KEY_AUTH]; switch (mhp->msg->sadb_msg_satype) { case SADB_SATYPE_AH: case SADB_SATYPE_ESP: case SADB_X_SATYPE_TCPSIGNATURE: if (len == PFKEY_ALIGN8(sizeof(struct sadb_key)) && sav->alg_auth != SADB_X_AALG_NULL) error = EINVAL; break; case SADB_X_SATYPE_IPCOMP: default: error = EINVAL; break; } if (error) { ipseclog((LOG_DEBUG, "%s: invalid key_auth values.\n", __func__)); goto fail; } sav->key_auth = key_dup_keymsg(key0, len, M_IPSEC_MISC); if (sav->key_auth == NULL ) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); PFKEYSTAT_INC(in_nomem); error = ENOBUFS; goto fail; } } /* Encryption key */ if (!SADB_CHECKHDR(mhp, SADB_EXT_KEY_ENCRYPT)) { if (SADB_CHECKLEN(mhp, SADB_EXT_KEY_ENCRYPT)) { error = EINVAL; goto fail; } error = 0; key0 = (const struct sadb_key *)mhp->ext[SADB_EXT_KEY_ENCRYPT]; len = mhp->extlen[SADB_EXT_KEY_ENCRYPT]; switch (mhp->msg->sadb_msg_satype) { case SADB_SATYPE_ESP: if (len == PFKEY_ALIGN8(sizeof(struct sadb_key)) && sav->alg_enc != SADB_EALG_NULL) { error = EINVAL; break; } sav->key_enc = key_dup_keymsg(key0, len, M_IPSEC_MISC); if (sav->key_enc == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); PFKEYSTAT_INC(in_nomem); error = ENOBUFS; goto fail; } break; case SADB_X_SATYPE_IPCOMP: if (len != PFKEY_ALIGN8(sizeof(struct sadb_key))) error = EINVAL; sav->key_enc = NULL; /*just in case*/ break; case SADB_SATYPE_AH: case SADB_X_SATYPE_TCPSIGNATURE: default: error = EINVAL; break; } if (error) { ipseclog((LOG_DEBUG, "%s: invalid key_enc value.\n", __func__)); goto fail; } } /* set iv */ sav->ivlen = 0; switch (mhp->msg->sadb_msg_satype) { case SADB_SATYPE_AH: if (sav->flags & SADB_X_EXT_DERIV) { ipseclog((LOG_DEBUG, "%s: invalid flag (derived) " "given to AH SA.\n", __func__)); error = EINVAL; goto fail; } if (sav->alg_enc != SADB_EALG_NONE) { ipseclog((LOG_DEBUG, "%s: protocol and algorithm " "mismated.\n", __func__)); error = EINVAL; goto fail; } error = xform_init(sav, XF_AH); break; case SADB_SATYPE_ESP: if ((sav->flags & (SADB_X_EXT_OLD | SADB_X_EXT_DERIV)) == (SADB_X_EXT_OLD | SADB_X_EXT_DERIV)) { ipseclog((LOG_DEBUG, "%s: invalid flag (derived) " "given to old-esp.\n", __func__)); error = EINVAL; goto fail; } error = xform_init(sav, XF_ESP); break; case SADB_X_SATYPE_IPCOMP: if (sav->alg_auth != SADB_AALG_NONE) { ipseclog((LOG_DEBUG, "%s: protocol and algorithm " "mismated.\n", __func__)); error = EINVAL; goto fail; } if ((sav->flags & SADB_X_EXT_RAWCPI) == 0 && ntohl(sav->spi) >= 0x10000) { ipseclog((LOG_DEBUG, "%s: invalid cpi for IPComp.\n", __func__)); error = EINVAL; goto fail; } error = xform_init(sav, XF_IPCOMP); break; case SADB_X_SATYPE_TCPSIGNATURE: if (sav->alg_enc != SADB_EALG_NONE) { ipseclog((LOG_DEBUG, "%s: protocol and algorithm " "mismated.\n", __func__)); error = EINVAL; goto fail; } error = xform_init(sav, XF_TCPSIGNATURE); break; default: ipseclog((LOG_DEBUG, "%s: Invalid satype.\n", __func__)); error = EPROTONOSUPPORT; goto fail; } if (error) { ipseclog((LOG_DEBUG, "%s: unable to initialize SA type %u.\n", __func__, mhp->msg->sadb_msg_satype)); goto fail; } /* Handle NAT-T headers */ error = key_setnatt(sav, mhp); if (error != 0) goto fail; /* Initialize lifetime for CURRENT */ sav->firstused = 0; sav->created = time_second; /* lifetimes for HARD and SOFT */ error = key_updatelifetimes(sav, mhp); if (error == 0) return (0); fail: key_cleansav(sav); return (error); } /* * subroutine for SADB_GET and SADB_DUMP. */ static struct mbuf * key_setdumpsa(struct secasvar *sav, uint8_t type, uint8_t satype, uint32_t seq, uint32_t pid) { struct seclifetime lft_c; struct mbuf *result = NULL, *tres = NULL, *m; int i, dumporder[] = { SADB_EXT_SA, SADB_X_EXT_SA2, SADB_X_EXT_SA_REPLAY, SADB_EXT_LIFETIME_HARD, SADB_EXT_LIFETIME_SOFT, SADB_EXT_LIFETIME_CURRENT, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST, SADB_EXT_ADDRESS_PROXY, SADB_EXT_KEY_AUTH, SADB_EXT_KEY_ENCRYPT, SADB_EXT_IDENTITY_SRC, SADB_EXT_IDENTITY_DST, SADB_EXT_SENSITIVITY, SADB_X_EXT_NAT_T_TYPE, SADB_X_EXT_NAT_T_SPORT, SADB_X_EXT_NAT_T_DPORT, SADB_X_EXT_NAT_T_OAI, SADB_X_EXT_NAT_T_OAR, SADB_X_EXT_NAT_T_FRAG, }; uint32_t replay_count; m = key_setsadbmsg(type, 0, satype, seq, pid, sav->refcnt); if (m == NULL) goto fail; result = m; for (i = nitems(dumporder) - 1; i >= 0; i--) { m = NULL; switch (dumporder[i]) { case SADB_EXT_SA: m = key_setsadbsa(sav); if (!m) goto fail; break; case SADB_X_EXT_SA2: SECASVAR_LOCK(sav); replay_count = sav->replay ? sav->replay->count : 0; SECASVAR_UNLOCK(sav); m = key_setsadbxsa2(sav->sah->saidx.mode, replay_count, sav->sah->saidx.reqid); if (!m) goto fail; break; case SADB_X_EXT_SA_REPLAY: if (sav->replay == NULL || sav->replay->wsize <= UINT8_MAX) continue; m = key_setsadbxsareplay(sav->replay->wsize); if (!m) goto fail; break; case SADB_EXT_ADDRESS_SRC: m = key_setsadbaddr(SADB_EXT_ADDRESS_SRC, &sav->sah->saidx.src.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) goto fail; break; case SADB_EXT_ADDRESS_DST: m = key_setsadbaddr(SADB_EXT_ADDRESS_DST, &sav->sah->saidx.dst.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) goto fail; break; case SADB_EXT_KEY_AUTH: if (!sav->key_auth) continue; m = key_setkey(sav->key_auth, SADB_EXT_KEY_AUTH); if (!m) goto fail; break; case SADB_EXT_KEY_ENCRYPT: if (!sav->key_enc) continue; m = key_setkey(sav->key_enc, SADB_EXT_KEY_ENCRYPT); if (!m) goto fail; break; case SADB_EXT_LIFETIME_CURRENT: lft_c.addtime = sav->created; lft_c.allocations = (uint32_t)counter_u64_fetch( sav->lft_c_allocations); lft_c.bytes = counter_u64_fetch(sav->lft_c_bytes); lft_c.usetime = sav->firstused; m = key_setlifetime(&lft_c, SADB_EXT_LIFETIME_CURRENT); if (!m) goto fail; break; case SADB_EXT_LIFETIME_HARD: if (!sav->lft_h) continue; m = key_setlifetime(sav->lft_h, SADB_EXT_LIFETIME_HARD); if (!m) goto fail; break; case SADB_EXT_LIFETIME_SOFT: if (!sav->lft_s) continue; m = key_setlifetime(sav->lft_s, SADB_EXT_LIFETIME_SOFT); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_TYPE: if (sav->natt == NULL) continue; m = key_setsadbxtype(UDP_ENCAP_ESPINUDP); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_DPORT: if (sav->natt == NULL) continue; m = key_setsadbxport(sav->natt->dport, SADB_X_EXT_NAT_T_DPORT); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_SPORT: if (sav->natt == NULL) continue; m = key_setsadbxport(sav->natt->sport, SADB_X_EXT_NAT_T_SPORT); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_OAI: if (sav->natt == NULL || (sav->natt->flags & IPSEC_NATT_F_OAI) == 0) continue; m = key_setsadbaddr(SADB_X_EXT_NAT_T_OAI, &sav->natt->oai.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_OAR: if (sav->natt == NULL || (sav->natt->flags & IPSEC_NATT_F_OAR) == 0) continue; m = key_setsadbaddr(SADB_X_EXT_NAT_T_OAR, &sav->natt->oar.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) goto fail; break; case SADB_X_EXT_NAT_T_FRAG: /* We do not (yet) support those. */ continue; case SADB_EXT_ADDRESS_PROXY: case SADB_EXT_IDENTITY_SRC: case SADB_EXT_IDENTITY_DST: /* XXX: should we brought from SPD ? */ case SADB_EXT_SENSITIVITY: default: continue; } if (!m) goto fail; if (tres) m_cat(m, tres); tres = m; } m_cat(result, tres); tres = NULL; if (result->m_len < sizeof(struct sadb_msg)) { result = m_pullup(result, sizeof(struct sadb_msg)); if (result == NULL) goto fail; } result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); return result; fail: m_freem(result); m_freem(tres); return NULL; } /* * set data into sadb_msg. */ static struct mbuf * key_setsadbmsg(u_int8_t type, u_int16_t tlen, u_int8_t satype, u_int32_t seq, pid_t pid, u_int16_t reserved) { struct mbuf *m; struct sadb_msg *p; int len; len = PFKEY_ALIGN8(sizeof(struct sadb_msg)); if (len > MCLBYTES) return NULL; MGETHDR(m, M_NOWAIT, MT_DATA); if (m && len > MHLEN) { if (!(MCLGET(m, M_NOWAIT))) { m_freem(m); m = NULL; } } if (!m) return NULL; m->m_pkthdr.len = m->m_len = len; m->m_next = NULL; p = mtod(m, struct sadb_msg *); bzero(p, len); p->sadb_msg_version = PF_KEY_V2; p->sadb_msg_type = type; p->sadb_msg_errno = 0; p->sadb_msg_satype = satype; p->sadb_msg_len = PFKEY_UNIT64(tlen); p->sadb_msg_reserved = reserved; p->sadb_msg_seq = seq; p->sadb_msg_pid = (u_int32_t)pid; return m; } /* * copy secasvar data into sadb_address. */ static struct mbuf * key_setsadbsa(struct secasvar *sav) { struct mbuf *m; struct sadb_sa *p; int len; len = PFKEY_ALIGN8(sizeof(struct sadb_sa)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_sa *); bzero(p, len); p->sadb_sa_len = PFKEY_UNIT64(len); p->sadb_sa_exttype = SADB_EXT_SA; p->sadb_sa_spi = sav->spi; p->sadb_sa_replay = sav->replay ? (sav->replay->wsize > UINT8_MAX ? UINT8_MAX : sav->replay->wsize): 0; p->sadb_sa_state = sav->state; p->sadb_sa_auth = sav->alg_auth; p->sadb_sa_encrypt = sav->alg_enc; p->sadb_sa_flags = sav->flags & SADB_KEY_FLAGS_MAX; return (m); } /* * set data into sadb_address. */ static struct mbuf * key_setsadbaddr(u_int16_t exttype, const struct sockaddr *saddr, u_int8_t prefixlen, u_int16_t ul_proto) { struct mbuf *m; struct sadb_address *p; size_t len; len = PFKEY_ALIGN8(sizeof(struct sadb_address)) + PFKEY_ALIGN8(saddr->sa_len); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_address *); bzero(p, len); p->sadb_address_len = PFKEY_UNIT64(len); p->sadb_address_exttype = exttype; p->sadb_address_proto = ul_proto; if (prefixlen == FULLMASK) { switch (saddr->sa_family) { case AF_INET: prefixlen = sizeof(struct in_addr) << 3; break; case AF_INET6: prefixlen = sizeof(struct in6_addr) << 3; break; default: ; /*XXX*/ } } p->sadb_address_prefixlen = prefixlen; p->sadb_address_reserved = 0; bcopy(saddr, mtod(m, caddr_t) + PFKEY_ALIGN8(sizeof(struct sadb_address)), saddr->sa_len); return m; } /* * set data into sadb_x_sa2. */ static struct mbuf * key_setsadbxsa2(u_int8_t mode, u_int32_t seq, u_int32_t reqid) { struct mbuf *m; struct sadb_x_sa2 *p; size_t len; len = PFKEY_ALIGN8(sizeof(struct sadb_x_sa2)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_x_sa2 *); bzero(p, len); p->sadb_x_sa2_len = PFKEY_UNIT64(len); p->sadb_x_sa2_exttype = SADB_X_EXT_SA2; p->sadb_x_sa2_mode = mode; p->sadb_x_sa2_reserved1 = 0; p->sadb_x_sa2_reserved2 = 0; p->sadb_x_sa2_sequence = seq; p->sadb_x_sa2_reqid = reqid; return m; } /* * Set data into sadb_x_sa_replay. */ static struct mbuf * key_setsadbxsareplay(u_int32_t replay) { struct mbuf *m; struct sadb_x_sa_replay *p; size_t len; len = PFKEY_ALIGN8(sizeof(struct sadb_x_sa_replay)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_x_sa_replay *); bzero(p, len); p->sadb_x_sa_replay_len = PFKEY_UNIT64(len); p->sadb_x_sa_replay_exttype = SADB_X_EXT_SA_REPLAY; p->sadb_x_sa_replay_replay = (replay << 3); return m; } /* * Set a type in sadb_x_nat_t_type. */ static struct mbuf * key_setsadbxtype(u_int16_t type) { struct mbuf *m; size_t len; struct sadb_x_nat_t_type *p; len = PFKEY_ALIGN8(sizeof(struct sadb_x_nat_t_type)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_x_nat_t_type *); bzero(p, len); p->sadb_x_nat_t_type_len = PFKEY_UNIT64(len); p->sadb_x_nat_t_type_exttype = SADB_X_EXT_NAT_T_TYPE; p->sadb_x_nat_t_type_type = type; return (m); } /* * Set a port in sadb_x_nat_t_port. * In contrast to default RFC 2367 behaviour, port is in network byte order. */ static struct mbuf * key_setsadbxport(u_int16_t port, u_int16_t type) { struct mbuf *m; size_t len; struct sadb_x_nat_t_port *p; len = PFKEY_ALIGN8(sizeof(struct sadb_x_nat_t_port)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_x_nat_t_port *); bzero(p, len); p->sadb_x_nat_t_port_len = PFKEY_UNIT64(len); p->sadb_x_nat_t_port_exttype = type; p->sadb_x_nat_t_port_port = port; return (m); } /* * Get port from sockaddr. Port is in network byte order. */ uint16_t key_portfromsaddr(struct sockaddr *sa) { switch (sa->sa_family) { #ifdef INET case AF_INET: return ((struct sockaddr_in *)sa)->sin_port; #endif #ifdef INET6 case AF_INET6: return ((struct sockaddr_in6 *)sa)->sin6_port; #endif } return (0); } /* * Set port in struct sockaddr. Port is in network byte order. */ void key_porttosaddr(struct sockaddr *sa, uint16_t port) { switch (sa->sa_family) { #ifdef INET case AF_INET: ((struct sockaddr_in *)sa)->sin_port = port; break; #endif #ifdef INET6 case AF_INET6: ((struct sockaddr_in6 *)sa)->sin6_port = port; break; #endif default: ipseclog((LOG_DEBUG, "%s: unexpected address family %d.\n", __func__, sa->sa_family)); break; } } /* * set data into sadb_x_policy */ static struct mbuf * key_setsadbxpolicy(u_int16_t type, u_int8_t dir, u_int32_t id, u_int32_t priority) { struct mbuf *m; struct sadb_x_policy *p; size_t len; len = PFKEY_ALIGN8(sizeof(struct sadb_x_policy)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return (NULL); m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_x_policy *); bzero(p, len); p->sadb_x_policy_len = PFKEY_UNIT64(len); p->sadb_x_policy_exttype = SADB_X_EXT_POLICY; p->sadb_x_policy_type = type; p->sadb_x_policy_dir = dir; p->sadb_x_policy_id = id; p->sadb_x_policy_priority = priority; return m; } /* %%% utilities */ /* Take a key message (sadb_key) from the socket and turn it into one * of the kernel's key structures (seckey). * * IN: pointer to the src * OUT: NULL no more memory */ struct seckey * key_dup_keymsg(const struct sadb_key *src, size_t len, struct malloc_type *type) { struct seckey *dst; dst = malloc(sizeof(*dst), type, M_NOWAIT); if (dst != NULL) { dst->bits = src->sadb_key_bits; dst->key_data = malloc(len, type, M_NOWAIT); if (dst->key_data != NULL) { bcopy((const char *)(src + 1), dst->key_data, len); } else { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); free(dst, type); dst = NULL; } } else { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); } return (dst); } /* Take a lifetime message (sadb_lifetime) passed in on a socket and * turn it into one of the kernel's lifetime structures (seclifetime). * * IN: pointer to the destination, source and malloc type * OUT: NULL, no more memory */ static struct seclifetime * key_dup_lifemsg(const struct sadb_lifetime *src, struct malloc_type *type) { struct seclifetime *dst; dst = malloc(sizeof(*dst), type, M_NOWAIT); if (dst == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (NULL); } dst->allocations = src->sadb_lifetime_allocations; dst->bytes = src->sadb_lifetime_bytes; dst->addtime = src->sadb_lifetime_addtime; dst->usetime = src->sadb_lifetime_usetime; return (dst); } /* * compare two secasindex structure. * flag can specify to compare 2 saidxes. * compare two secasindex structure without both mode and reqid. * don't compare port. * IN: * saidx0: source, it can be in SAD. * saidx1: object. * OUT: * 1 : equal * 0 : not equal */ static int key_cmpsaidx(const struct secasindex *saidx0, const struct secasindex *saidx1, int flag) { /* sanity */ if (saidx0 == NULL && saidx1 == NULL) return 1; if (saidx0 == NULL || saidx1 == NULL) return 0; if (saidx0->proto != saidx1->proto) return 0; if (flag == CMP_EXACTLY) { if (saidx0->mode != saidx1->mode) return 0; if (saidx0->reqid != saidx1->reqid) return 0; if (bcmp(&saidx0->src, &saidx1->src, saidx0->src.sa.sa_len) != 0 || bcmp(&saidx0->dst, &saidx1->dst, saidx0->dst.sa.sa_len) != 0) return 0; } else { /* CMP_MODE_REQID, CMP_REQID, CMP_HEAD */ if (flag == CMP_MODE_REQID || flag == CMP_REQID) { /* * If reqid of SPD is non-zero, unique SA is required. * The result must be of same reqid in this case. */ if (saidx1->reqid != 0 && saidx0->reqid != saidx1->reqid) return 0; } if (flag == CMP_MODE_REQID) { if (saidx0->mode != IPSEC_MODE_ANY && saidx0->mode != saidx1->mode) return 0; } if (key_sockaddrcmp(&saidx0->src.sa, &saidx1->src.sa, 0) != 0) return 0; if (key_sockaddrcmp(&saidx0->dst.sa, &saidx1->dst.sa, 0) != 0) return 0; } return 1; } /* * compare two secindex structure exactly. * IN: * spidx0: source, it is often in SPD. * spidx1: object, it is often from PFKEY message. * OUT: * 1 : equal * 0 : not equal */ static int key_cmpspidx_exactly(struct secpolicyindex *spidx0, struct secpolicyindex *spidx1) { /* sanity */ if (spidx0 == NULL && spidx1 == NULL) return 1; if (spidx0 == NULL || spidx1 == NULL) return 0; if (spidx0->prefs != spidx1->prefs || spidx0->prefd != spidx1->prefd || spidx0->ul_proto != spidx1->ul_proto || spidx0->dir != spidx1->dir) return 0; return key_sockaddrcmp(&spidx0->src.sa, &spidx1->src.sa, 1) == 0 && key_sockaddrcmp(&spidx0->dst.sa, &spidx1->dst.sa, 1) == 0; } /* * compare two secindex structure with mask. * IN: * spidx0: source, it is often in SPD. * spidx1: object, it is often from IP header. * OUT: * 1 : equal * 0 : not equal */ static int key_cmpspidx_withmask(struct secpolicyindex *spidx0, struct secpolicyindex *spidx1) { /* sanity */ if (spidx0 == NULL && spidx1 == NULL) return 1; if (spidx0 == NULL || spidx1 == NULL) return 0; if (spidx0->src.sa.sa_family != spidx1->src.sa.sa_family || spidx0->dst.sa.sa_family != spidx1->dst.sa.sa_family || spidx0->src.sa.sa_len != spidx1->src.sa.sa_len || spidx0->dst.sa.sa_len != spidx1->dst.sa.sa_len) return 0; /* if spidx.ul_proto == IPSEC_ULPROTO_ANY, ignore. */ if (spidx0->ul_proto != (u_int16_t)IPSEC_ULPROTO_ANY && spidx0->ul_proto != spidx1->ul_proto) return 0; switch (spidx0->src.sa.sa_family) { case AF_INET: if (spidx0->src.sin.sin_port != IPSEC_PORT_ANY && spidx0->src.sin.sin_port != spidx1->src.sin.sin_port) return 0; if (!key_bbcmp(&spidx0->src.sin.sin_addr, &spidx1->src.sin.sin_addr, spidx0->prefs)) return 0; break; case AF_INET6: if (spidx0->src.sin6.sin6_port != IPSEC_PORT_ANY && spidx0->src.sin6.sin6_port != spidx1->src.sin6.sin6_port) return 0; /* * scope_id check. if sin6_scope_id is 0, we regard it * as a wildcard scope, which matches any scope zone ID. */ if (spidx0->src.sin6.sin6_scope_id && spidx1->src.sin6.sin6_scope_id && spidx0->src.sin6.sin6_scope_id != spidx1->src.sin6.sin6_scope_id) return 0; if (!key_bbcmp(&spidx0->src.sin6.sin6_addr, &spidx1->src.sin6.sin6_addr, spidx0->prefs)) return 0; break; default: /* XXX */ if (bcmp(&spidx0->src, &spidx1->src, spidx0->src.sa.sa_len) != 0) return 0; break; } switch (spidx0->dst.sa.sa_family) { case AF_INET: if (spidx0->dst.sin.sin_port != IPSEC_PORT_ANY && spidx0->dst.sin.sin_port != spidx1->dst.sin.sin_port) return 0; if (!key_bbcmp(&spidx0->dst.sin.sin_addr, &spidx1->dst.sin.sin_addr, spidx0->prefd)) return 0; break; case AF_INET6: if (spidx0->dst.sin6.sin6_port != IPSEC_PORT_ANY && spidx0->dst.sin6.sin6_port != spidx1->dst.sin6.sin6_port) return 0; /* * scope_id check. if sin6_scope_id is 0, we regard it * as a wildcard scope, which matches any scope zone ID. */ if (spidx0->dst.sin6.sin6_scope_id && spidx1->dst.sin6.sin6_scope_id && spidx0->dst.sin6.sin6_scope_id != spidx1->dst.sin6.sin6_scope_id) return 0; if (!key_bbcmp(&spidx0->dst.sin6.sin6_addr, &spidx1->dst.sin6.sin6_addr, spidx0->prefd)) return 0; break; default: /* XXX */ if (bcmp(&spidx0->dst, &spidx1->dst, spidx0->dst.sa.sa_len) != 0) return 0; break; } /* XXX Do we check other field ? e.g. flowinfo */ return 1; } #ifdef satosin #undef satosin #endif #define satosin(s) ((const struct sockaddr_in *)s) #ifdef satosin6 #undef satosin6 #endif #define satosin6(s) ((const struct sockaddr_in6 *)s) /* returns 0 on match */ int key_sockaddrcmp(const struct sockaddr *sa1, const struct sockaddr *sa2, int port) { if (sa1->sa_family != sa2->sa_family || sa1->sa_len != sa2->sa_len) return 1; switch (sa1->sa_family) { #ifdef INET case AF_INET: if (sa1->sa_len != sizeof(struct sockaddr_in)) return 1; if (satosin(sa1)->sin_addr.s_addr != satosin(sa2)->sin_addr.s_addr) { return 1; } if (port && satosin(sa1)->sin_port != satosin(sa2)->sin_port) return 1; break; #endif #ifdef INET6 case AF_INET6: if (sa1->sa_len != sizeof(struct sockaddr_in6)) return 1; /*EINVAL*/ if (satosin6(sa1)->sin6_scope_id != satosin6(sa2)->sin6_scope_id) { return 1; } if (!IN6_ARE_ADDR_EQUAL(&satosin6(sa1)->sin6_addr, &satosin6(sa2)->sin6_addr)) { return 1; } if (port && satosin6(sa1)->sin6_port != satosin6(sa2)->sin6_port) { return 1; } break; #endif default: if (bcmp(sa1, sa2, sa1->sa_len) != 0) return 1; break; } return 0; } /* returns 0 on match */ int key_sockaddrcmp_withmask(const struct sockaddr *sa1, const struct sockaddr *sa2, size_t mask) { if (sa1->sa_family != sa2->sa_family || sa1->sa_len != sa2->sa_len) return (1); switch (sa1->sa_family) { #ifdef INET case AF_INET: return (!key_bbcmp(&satosin(sa1)->sin_addr, &satosin(sa2)->sin_addr, mask)); #endif #ifdef INET6 case AF_INET6: if (satosin6(sa1)->sin6_scope_id != satosin6(sa2)->sin6_scope_id) return (1); return (!key_bbcmp(&satosin6(sa1)->sin6_addr, &satosin6(sa2)->sin6_addr, mask)); #endif } return (1); } #undef satosin #undef satosin6 /* * compare two buffers with mask. * IN: * addr1: source * addr2: object * bits: Number of bits to compare * OUT: * 1 : equal * 0 : not equal */ static int key_bbcmp(const void *a1, const void *a2, u_int bits) { const unsigned char *p1 = a1; const unsigned char *p2 = a2; /* XXX: This could be considerably faster if we compare a word * at a time, but it is complicated on LSB Endian machines */ /* Handle null pointers */ if (p1 == NULL || p2 == NULL) return (p1 == p2); while (bits >= 8) { if (*p1++ != *p2++) return 0; bits -= 8; } if (bits > 0) { u_int8_t mask = ~((1<<(8-bits))-1); if ((*p1 & mask) != (*p2 & mask)) return 0; } return 1; /* Match! */ } static void key_flush_spd(time_t now) { SPTREE_RLOCK_TRACKER; struct secpolicy_list drainq; struct secpolicy *sp, *nextsp; u_int dir; LIST_INIT(&drainq); SPTREE_RLOCK(); for (dir = 0; dir < IPSEC_DIR_MAX; dir++) { TAILQ_FOREACH(sp, &V_sptree[dir], chain) { if (sp->lifetime == 0 && sp->validtime == 0) continue; if ((sp->lifetime && now - sp->created > sp->lifetime) || (sp->validtime && now - sp->lastused > sp->validtime)) { /* Hold extra reference to send SPDEXPIRE */ SP_ADDREF(sp); LIST_INSERT_HEAD(&drainq, sp, drainq); } } } SPTREE_RUNLOCK(); if (LIST_EMPTY(&drainq)) return; SPTREE_WLOCK(); sp = LIST_FIRST(&drainq); while (sp != NULL) { nextsp = LIST_NEXT(sp, drainq); /* Check that SP is still linked */ if (sp->state != IPSEC_SPSTATE_ALIVE) { LIST_REMOVE(sp, drainq); key_freesp(&sp); /* release extra reference */ sp = nextsp; continue; } TAILQ_REMOVE(&V_sptree[sp->spidx.dir], sp, chain); V_spd_size--; LIST_REMOVE(sp, idhash); sp->state = IPSEC_SPSTATE_DEAD; sp = nextsp; } V_sp_genid++; SPTREE_WUNLOCK(); if (SPDCACHE_ENABLED()) spdcache_clear(); sp = LIST_FIRST(&drainq); while (sp != NULL) { nextsp = LIST_NEXT(sp, drainq); key_spdexpire(sp); key_freesp(&sp); /* release extra reference */ key_freesp(&sp); /* release last reference */ sp = nextsp; } } static void key_flush_sad(time_t now) { SAHTREE_RLOCK_TRACKER; struct secashead_list emptyq; struct secasvar_list drainq, hexpireq, sexpireq, freeq; struct secashead *sah, *nextsah; struct secasvar *sav, *nextsav; LIST_INIT(&drainq); LIST_INIT(&hexpireq); LIST_INIT(&sexpireq); LIST_INIT(&emptyq); SAHTREE_RLOCK(); TAILQ_FOREACH(sah, &V_sahtree, chain) { /* Check for empty SAH */ if (TAILQ_EMPTY(&sah->savtree_larval) && TAILQ_EMPTY(&sah->savtree_alive)) { SAH_ADDREF(sah); LIST_INSERT_HEAD(&emptyq, sah, drainq); continue; } /* Add all stale LARVAL SAs into drainq */ TAILQ_FOREACH(sav, &sah->savtree_larval, chain) { if (now - sav->created < V_key_larval_lifetime) continue; SAV_ADDREF(sav); LIST_INSERT_HEAD(&drainq, sav, drainq); } TAILQ_FOREACH(sav, &sah->savtree_alive, chain) { /* lifetimes aren't specified */ if (sav->lft_h == NULL) continue; SECASVAR_LOCK(sav); /* * Check again with lock held, because it may * be updated by SADB_UPDATE. */ if (sav->lft_h == NULL) { SECASVAR_UNLOCK(sav); continue; } /* * RFC 2367: * HARD lifetimes MUST take precedence over SOFT * lifetimes, meaning if the HARD and SOFT lifetimes * are the same, the HARD lifetime will appear on the * EXPIRE message. */ /* check HARD lifetime */ if ((sav->lft_h->addtime != 0 && now - sav->created > sav->lft_h->addtime) || (sav->lft_h->usetime != 0 && sav->firstused && now - sav->firstused > sav->lft_h->usetime) || (sav->lft_h->bytes != 0 && counter_u64_fetch( sav->lft_c_bytes) > sav->lft_h->bytes)) { SECASVAR_UNLOCK(sav); SAV_ADDREF(sav); LIST_INSERT_HEAD(&hexpireq, sav, drainq); continue; } /* check SOFT lifetime (only for MATURE SAs) */ if (sav->state == SADB_SASTATE_MATURE && ( (sav->lft_s->addtime != 0 && now - sav->created > sav->lft_s->addtime) || (sav->lft_s->usetime != 0 && sav->firstused && now - sav->firstused > sav->lft_s->usetime) || (sav->lft_s->bytes != 0 && counter_u64_fetch( sav->lft_c_bytes) > sav->lft_s->bytes))) { SECASVAR_UNLOCK(sav); SAV_ADDREF(sav); LIST_INSERT_HEAD(&sexpireq, sav, drainq); continue; } SECASVAR_UNLOCK(sav); } } SAHTREE_RUNLOCK(); if (LIST_EMPTY(&emptyq) && LIST_EMPTY(&drainq) && LIST_EMPTY(&hexpireq) && LIST_EMPTY(&sexpireq)) return; LIST_INIT(&freeq); SAHTREE_WLOCK(); /* Unlink stale LARVAL SAs */ sav = LIST_FIRST(&drainq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); /* Check that SA is still LARVAL */ if (sav->state != SADB_SASTATE_LARVAL) { LIST_REMOVE(sav, drainq); LIST_INSERT_HEAD(&freeq, sav, drainq); sav = nextsav; continue; } TAILQ_REMOVE(&sav->sah->savtree_larval, sav, chain); LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; sav = nextsav; } /* Unlink all SAs with expired HARD lifetime */ sav = LIST_FIRST(&hexpireq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); /* Check that SA is not unlinked */ if (sav->state == SADB_SASTATE_DEAD) { LIST_REMOVE(sav, drainq); LIST_INSERT_HEAD(&freeq, sav, drainq); sav = nextsav; continue; } TAILQ_REMOVE(&sav->sah->savtree_alive, sav, chain); LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; sav = nextsav; } /* Mark all SAs with expired SOFT lifetime as DYING */ sav = LIST_FIRST(&sexpireq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); /* Check that SA is not unlinked */ if (sav->state == SADB_SASTATE_DEAD) { LIST_REMOVE(sav, drainq); LIST_INSERT_HEAD(&freeq, sav, drainq); sav = nextsav; continue; } /* * NOTE: this doesn't change SA order in the chain. */ sav->state = SADB_SASTATE_DYING; sav = nextsav; } /* Unlink empty SAHs */ sah = LIST_FIRST(&emptyq); while (sah != NULL) { nextsah = LIST_NEXT(sah, drainq); /* Check that SAH is still empty and not unlinked */ if (sah->state == SADB_SASTATE_DEAD || !TAILQ_EMPTY(&sah->savtree_larval) || !TAILQ_EMPTY(&sah->savtree_alive)) { LIST_REMOVE(sah, drainq); key_freesah(&sah); /* release extra reference */ sah = nextsah; continue; } TAILQ_REMOVE(&V_sahtree, sah, chain); LIST_REMOVE(sah, addrhash); sah->state = SADB_SASTATE_DEAD; sah = nextsah; } SAHTREE_WUNLOCK(); /* Send SPDEXPIRE messages */ sav = LIST_FIRST(&hexpireq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); key_expire(sav, 1); key_freesah(&sav->sah); /* release reference from SAV */ key_freesav(&sav); /* release extra reference */ key_freesav(&sav); /* release last reference */ sav = nextsav; } sav = LIST_FIRST(&sexpireq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); key_expire(sav, 0); key_freesav(&sav); /* release extra reference */ sav = nextsav; } /* Free stale LARVAL SAs */ sav = LIST_FIRST(&drainq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); key_freesah(&sav->sah); /* release reference from SAV */ key_freesav(&sav); /* release extra reference */ key_freesav(&sav); /* release last reference */ sav = nextsav; } /* Free SAs that were unlinked/changed by someone else */ sav = LIST_FIRST(&freeq); while (sav != NULL) { nextsav = LIST_NEXT(sav, drainq); key_freesav(&sav); /* release extra reference */ sav = nextsav; } /* Free empty SAH */ sah = LIST_FIRST(&emptyq); while (sah != NULL) { nextsah = LIST_NEXT(sah, drainq); key_freesah(&sah); /* release extra reference */ key_freesah(&sah); /* release last reference */ sah = nextsah; } } static void key_flush_acq(time_t now) { struct secacq *acq, *nextacq; /* ACQ tree */ ACQ_LOCK(); acq = LIST_FIRST(&V_acqtree); while (acq != NULL) { nextacq = LIST_NEXT(acq, chain); if (now - acq->created > V_key_blockacq_lifetime) { LIST_REMOVE(acq, chain); LIST_REMOVE(acq, addrhash); LIST_REMOVE(acq, seqhash); free(acq, M_IPSEC_SAQ); } acq = nextacq; } ACQ_UNLOCK(); } static void key_flush_spacq(time_t now) { struct secspacq *acq, *nextacq; /* SP ACQ tree */ SPACQ_LOCK(); for (acq = LIST_FIRST(&V_spacqtree); acq != NULL; acq = nextacq) { nextacq = LIST_NEXT(acq, chain); if (now - acq->created > V_key_blockacq_lifetime && __LIST_CHAINED(acq)) { LIST_REMOVE(acq, chain); free(acq, M_IPSEC_SAQ); } } SPACQ_UNLOCK(); } /* * time handler. * scanning SPD and SAD to check status for each entries, * and do to remove or to expire. * XXX: year 2038 problem may remain. */ static void key_timehandler(void *arg) { VNET_ITERATOR_DECL(vnet_iter); time_t now = time_second; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); key_flush_spd(now); key_flush_sad(now); key_flush_acq(now); key_flush_spacq(now); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); #ifndef IPSEC_DEBUG2 /* do exchange to tick time !! */ callout_schedule(&key_timer, hz); #endif /* IPSEC_DEBUG2 */ } u_long key_random() { u_long value; key_randomfill(&value, sizeof(value)); return value; } void key_randomfill(void *p, size_t l) { size_t n; u_long v; static int warn = 1; n = 0; n = (size_t)read_random(p, (u_int)l); /* last resort */ while (n < l) { v = random(); bcopy(&v, (u_int8_t *)p + n, l - n < sizeof(v) ? l - n : sizeof(v)); n += sizeof(v); if (warn) { printf("WARNING: pseudo-random number generator " "used for IPsec processing\n"); warn = 0; } } } /* * map SADB_SATYPE_* to IPPROTO_*. * if satype == SADB_SATYPE then satype is mapped to ~0. * OUT: * 0: invalid satype. */ static uint8_t key_satype2proto(uint8_t satype) { switch (satype) { case SADB_SATYPE_UNSPEC: return IPSEC_PROTO_ANY; case SADB_SATYPE_AH: return IPPROTO_AH; case SADB_SATYPE_ESP: return IPPROTO_ESP; case SADB_X_SATYPE_IPCOMP: return IPPROTO_IPCOMP; case SADB_X_SATYPE_TCPSIGNATURE: return IPPROTO_TCP; default: return 0; } /* NOTREACHED */ } /* * map IPPROTO_* to SADB_SATYPE_* * OUT: * 0: invalid protocol type. */ static uint8_t key_proto2satype(uint8_t proto) { switch (proto) { case IPPROTO_AH: return SADB_SATYPE_AH; case IPPROTO_ESP: return SADB_SATYPE_ESP; case IPPROTO_IPCOMP: return SADB_X_SATYPE_IPCOMP; case IPPROTO_TCP: return SADB_X_SATYPE_TCPSIGNATURE; default: return 0; } /* NOTREACHED */ } /* %%% PF_KEY */ /* * SADB_GETSPI processing is to receive * * from the IKMPd, to assign a unique spi value, to hang on the INBOUND * tree with the status of LARVAL, and send * * to the IKMPd. * * IN: mhp: pointer to the pointer to each header. * OUT: NULL if fail. * other if success, return pointer to the message to send. */ static int key_getspi(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secasindex saidx; struct sadb_address *src0, *dst0; struct secasvar *sav; uint32_t reqid, spi; int error; uint8_t mode, proto; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); if (SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) #ifdef PFKEY_STRICT_CHECKS || SADB_CHECKHDR(mhp, SADB_EXT_SPIRANGE) #endif ) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); error = EINVAL; goto fail; } if (SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST) #ifdef PFKEY_STRICT_CHECKS || SADB_CHECKLEN(mhp, SADB_EXT_SPIRANGE) #endif ) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); error = EINVAL; goto fail; } if (SADB_CHECKHDR(mhp, SADB_X_EXT_SA2)) { mode = IPSEC_MODE_ANY; reqid = 0; } else { if (SADB_CHECKLEN(mhp, SADB_X_EXT_SA2)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); error = EINVAL; goto fail; } mode = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_mode; reqid = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_reqid; } src0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_SRC]); dst0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_DST]); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); error = EINVAL; goto fail; } error = key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)); if (error != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); error = EINVAL; goto fail; } KEY_SETSECASIDX(proto, mode, reqid, src0 + 1, dst0 + 1, &saidx); /* SPI allocation */ spi = key_do_getnewspi( (struct sadb_spirange *)mhp->ext[SADB_EXT_SPIRANGE], &saidx); if (spi == 0) { /* * Requested SPI or SPI range is not available or * already used. */ error = EEXIST; goto fail; } sav = key_newsav(mhp, &saidx, spi, &error); if (sav == NULL) goto fail; if (sav->seq != 0) { /* * RFC2367: * If the SADB_GETSPI message is in response to a * kernel-generated SADB_ACQUIRE, the sadb_msg_seq * MUST be the same as the SADB_ACQUIRE message. * * XXXAE: However it doesn't definethe behaviour how to * check this and what to do if it doesn't match. * Also what we should do if it matches? * * We can compare saidx used in SADB_ACQUIRE with saidx * used in SADB_GETSPI, but this probably can break * existing software. For now just warn if it doesn't match. * * XXXAE: anyway it looks useless. */ key_acqdone(&saidx, sav->seq); } KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); { struct mbuf *n, *nn; struct sadb_sa *m_sa; struct sadb_msg *newmsg; int off, len; /* create new sadb_msg to reply. */ len = PFKEY_ALIGN8(sizeof(struct sadb_msg)) + PFKEY_ALIGN8(sizeof(struct sadb_sa)); MGETHDR(n, M_NOWAIT, MT_DATA); if (len > MHLEN) { if (!(MCLGET(n, M_NOWAIT))) { m_freem(n); n = NULL; } } if (!n) { error = ENOBUFS; goto fail; } n->m_len = len; n->m_next = NULL; off = 0; m_copydata(m, 0, sizeof(struct sadb_msg), mtod(n, caddr_t) + off); off += PFKEY_ALIGN8(sizeof(struct sadb_msg)); m_sa = (struct sadb_sa *)(mtod(n, caddr_t) + off); m_sa->sadb_sa_len = PFKEY_UNIT64(sizeof(struct sadb_sa)); m_sa->sadb_sa_exttype = SADB_EXT_SA; m_sa->sadb_sa_spi = spi; /* SPI is already in network byte order */ off += PFKEY_ALIGN8(sizeof(struct sadb_sa)); IPSEC_ASSERT(off == len, ("length inconsistency (off %u len %u)", off, len)); n->m_next = key_gather_mbuf(m, mhp, 0, 2, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); if (!n->m_next) { m_freem(n); error = ENOBUFS; goto fail; } if (n->m_len < sizeof(struct sadb_msg)) { n = m_pullup(n, sizeof(struct sadb_msg)); if (n == NULL) return key_sendup_mbuf(so, m, KEY_SENDUP_ONE); } n->m_pkthdr.len = 0; for (nn = n; nn; nn = nn->m_next) n->m_pkthdr.len += nn->m_len; newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_seq = sav->seq; newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } fail: return (key_senderror(so, m, error)); } /* * allocating new SPI * called by key_getspi(). * OUT: * 0: failure. * others: success, SPI in network byte order. */ static uint32_t key_do_getnewspi(struct sadb_spirange *spirange, struct secasindex *saidx) { uint32_t min, max, newspi, t; int count = V_key_spi_trycnt; /* set spi range to allocate */ if (spirange != NULL) { min = spirange->sadb_spirange_min; max = spirange->sadb_spirange_max; } else { min = V_key_spi_minval; max = V_key_spi_maxval; } /* IPCOMP needs 2-byte SPI */ if (saidx->proto == IPPROTO_IPCOMP) { if (min >= 0x10000) min = 0xffff; if (max >= 0x10000) max = 0xffff; if (min > max) { t = min; min = max; max = t; } } if (min == max) { if (!key_checkspidup(htonl(min))) { ipseclog((LOG_DEBUG, "%s: SPI %u exists already.\n", __func__, min)); return 0; } count--; /* taking one cost. */ newspi = min; } else { /* init SPI */ newspi = 0; /* when requesting to allocate spi ranged */ while (count--) { /* generate pseudo-random SPI value ranged. */ newspi = min + (key_random() % (max - min + 1)); if (!key_checkspidup(htonl(newspi))) break; } if (count == 0 || newspi == 0) { ipseclog((LOG_DEBUG, "%s: failed to allocate SPI.\n", __func__)); return 0; } } /* statistics */ keystat.getspi_count = (keystat.getspi_count + V_key_spi_trycnt - count) / 2; return (htonl(newspi)); } /* * Find TCP-MD5 SA with corresponding secasindex. * If not found, return NULL and fill SPI with usable value if needed. */ static struct secasvar * key_getsav_tcpmd5(struct secasindex *saidx, uint32_t *spi) { SAHTREE_RLOCK_TRACKER; struct secashead *sah; struct secasvar *sav; IPSEC_ASSERT(saidx->proto == IPPROTO_TCP, ("wrong proto")); SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(saidx), addrhash) { if (sah->saidx.proto != IPPROTO_TCP) continue; if (!key_sockaddrcmp(&saidx->dst.sa, &sah->saidx.dst.sa, 0) && !key_sockaddrcmp(&saidx->src.sa, &sah->saidx.src.sa, 0)) break; } if (sah != NULL) { if (V_key_preferred_oldsa) sav = TAILQ_LAST(&sah->savtree_alive, secasvar_queue); else sav = TAILQ_FIRST(&sah->savtree_alive); if (sav != NULL) { SAV_ADDREF(sav); SAHTREE_RUNLOCK(); return (sav); } } if (spi == NULL) { /* No SPI required */ SAHTREE_RUNLOCK(); return (NULL); } /* Check that SPI is unique */ LIST_FOREACH(sav, SAVHASH_HASH(*spi), spihash) { if (sav->spi == *spi) break; } if (sav == NULL) { SAHTREE_RUNLOCK(); /* SPI is already unique */ return (NULL); } SAHTREE_RUNLOCK(); /* XXX: not optimal */ *spi = key_do_getnewspi(NULL, saidx); return (NULL); } static int key_updateaddresses(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp, struct secasvar *sav, struct secasindex *saidx) { struct sockaddr *newaddr; struct secashead *sah; struct secasvar *newsav, *tmp; struct mbuf *n; int error, isnew; /* Check that we need to change SAH */ if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_SRC)) { newaddr = (struct sockaddr *)( ((struct sadb_address *) mhp->ext[SADB_X_EXT_NEW_ADDRESS_SRC]) + 1); bcopy(newaddr, &saidx->src, newaddr->sa_len); key_porttosaddr(&saidx->src.sa, 0); } if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_DST)) { newaddr = (struct sockaddr *)( ((struct sadb_address *) mhp->ext[SADB_X_EXT_NEW_ADDRESS_DST]) + 1); bcopy(newaddr, &saidx->dst, newaddr->sa_len); key_porttosaddr(&saidx->dst.sa, 0); } if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_SRC) || !SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_DST)) { error = key_checksockaddrs(&saidx->src.sa, &saidx->dst.sa); if (error != 0) { ipseclog((LOG_DEBUG, "%s: invalid new sockaddr.\n", __func__)); return (error); } sah = key_getsah(saidx); if (sah == NULL) { /* create a new SA index */ sah = key_newsah(saidx); if (sah == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (ENOBUFS); } isnew = 2; /* SAH is new */ } else isnew = 1; /* existing SAH is referenced */ } else { /* * src and dst addresses are still the same. * Do we want to change NAT-T config? */ if (sav->sah->saidx.proto != IPPROTO_ESP || SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_TYPE) || SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_SPORT) || SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_DPORT)) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return (EINVAL); } /* We hold reference to SA, thus SAH will be referenced too. */ sah = sav->sah; isnew = 0; } newsav = malloc(sizeof(struct secasvar), M_IPSEC_SA, M_NOWAIT | M_ZERO); if (newsav == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); error = ENOBUFS; goto fail; } /* Clone SA's content into newsav */ SAV_INITREF(newsav); bcopy(sav, newsav, offsetof(struct secasvar, chain)); /* * We create new NAT-T config if it is needed. * Old NAT-T config will be freed by key_cleansav() when * last reference to SA will be released. */ newsav->natt = NULL; newsav->sah = sah; newsav->state = SADB_SASTATE_MATURE; error = key_setnatt(newsav, mhp); if (error != 0) goto fail; SAHTREE_WLOCK(); /* Check that SA is still alive */ if (sav->state == SADB_SASTATE_DEAD) { /* SA was unlinked */ SAHTREE_WUNLOCK(); error = ESRCH; goto fail; } /* Unlink SA from SAH and SPI hash */ IPSEC_ASSERT((sav->flags & SADB_X_EXT_F_CLONED) == 0, ("SA is already cloned")); IPSEC_ASSERT(sav->state == SADB_SASTATE_MATURE || sav->state == SADB_SASTATE_DYING, ("Wrong SA state %u\n", sav->state)); TAILQ_REMOVE(&sav->sah->savtree_alive, sav, chain); LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; /* * Link new SA with SAH. Keep SAs ordered by * create time (newer are first). */ TAILQ_FOREACH(tmp, &sah->savtree_alive, chain) { if (newsav->created > tmp->created) { TAILQ_INSERT_BEFORE(tmp, newsav, chain); break; } } if (tmp == NULL) TAILQ_INSERT_TAIL(&sah->savtree_alive, newsav, chain); /* Add new SA into SPI hash. */ LIST_INSERT_HEAD(SAVHASH_HASH(newsav->spi), newsav, spihash); /* Add new SAH into SADB. */ if (isnew == 2) { TAILQ_INSERT_HEAD(&V_sahtree, sah, chain); LIST_INSERT_HEAD(SAHADDRHASH_HASH(saidx), sah, addrhash); sah->state = SADB_SASTATE_MATURE; SAH_ADDREF(sah); /* newsav references new SAH */ } /* * isnew == 1 -> @sah was referenced by key_getsah(). * isnew == 0 -> we use the same @sah, that was used by @sav, * and we use its reference for @newsav. */ SECASVAR_LOCK(sav); /* XXX: replace cntr with pointer? */ newsav->cntr = sav->cntr; sav->flags |= SADB_X_EXT_F_CLONED; SECASVAR_UNLOCK(sav); SAHTREE_WUNLOCK(); KEYDBG(KEY_STAMP, printf("%s: SA(%p) cloned into SA(%p)\n", __func__, sav, newsav)); KEYDBG(KEY_DATA, kdebug_secasv(newsav)); key_freesav(&sav); /* release last reference */ /* set msg buf from mhp */ n = key_getmsgbuf_x1(m, mhp); if (n == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (ENOBUFS); } m_freem(m); key_sendup_mbuf(so, n, KEY_SENDUP_ALL); return (0); fail: if (isnew != 0) key_freesah(&sah); if (newsav != NULL) { if (newsav->natt != NULL) free(newsav->natt, M_IPSEC_MISC); free(newsav, M_IPSEC_SA); } return (error); } /* * SADB_UPDATE processing * receive * * from the ikmpd, and update a secasvar entry whose status is SADB_SASTATE_LARVAL. * and send * * to the ikmpd. * * m will always be freed. */ static int key_update(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secasindex saidx; struct sadb_address *src0, *dst0; struct sadb_sa *sa0; struct secasvar *sav; uint32_t reqid; int error; uint8_t mode, proto; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_EXT_SA) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT)) || (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD))) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_SA) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_X_EXT_SA2)) { mode = IPSEC_MODE_ANY; reqid = 0; } else { if (SADB_CHECKLEN(mhp, SADB_X_EXT_SA2)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } mode = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_mode; reqid = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_reqid; } sa0 = (struct sadb_sa *)mhp->ext[SADB_EXT_SA]; src0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_SRC]); dst0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_DST]); /* * Only SADB_SASTATE_MATURE SAs may be submitted in an * SADB_UPDATE message. */ if (sa0->sadb_sa_state != SADB_SASTATE_MATURE) { ipseclog((LOG_DEBUG, "%s: invalid state.\n", __func__)); #ifdef PFKEY_STRICT_CHECKS return key_senderror(so, m, EINVAL); #endif } error = key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)); if (error != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, error); } KEY_SETSECASIDX(proto, mode, reqid, src0 + 1, dst0 + 1, &saidx); sav = key_getsavbyspi(sa0->sadb_sa_spi); if (sav == NULL) { ipseclog((LOG_DEBUG, "%s: no SA found for SPI %u\n", __func__, ntohl(sa0->sadb_sa_spi))); return key_senderror(so, m, EINVAL); } /* * Check that SADB_UPDATE issued by the same process that did * SADB_GETSPI or SADB_ADD. */ if (sav->pid != mhp->msg->sadb_msg_pid) { ipseclog((LOG_DEBUG, "%s: pid mismatched (SPI %u, pid %u vs. %u)\n", __func__, ntohl(sav->spi), sav->pid, mhp->msg->sadb_msg_pid)); key_freesav(&sav); return key_senderror(so, m, EINVAL); } /* saidx should match with SA. */ if (key_cmpsaidx(&sav->sah->saidx, &saidx, CMP_MODE_REQID) == 0) { ipseclog((LOG_DEBUG, "%s: saidx mismatched for SPI %u", __func__, ntohl(sav->spi))); key_freesav(&sav); return key_senderror(so, m, ESRCH); } if (sav->state == SADB_SASTATE_LARVAL) { if ((mhp->msg->sadb_msg_satype == SADB_SATYPE_ESP && SADB_CHECKHDR(mhp, SADB_EXT_KEY_ENCRYPT)) || (mhp->msg->sadb_msg_satype == SADB_SATYPE_AH && SADB_CHECKHDR(mhp, SADB_EXT_KEY_AUTH))) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); key_freesav(&sav); return key_senderror(so, m, EINVAL); } /* * We can set any values except src, dst and SPI. */ error = key_setsaval(sav, mhp); if (error != 0) { key_freesav(&sav); return (key_senderror(so, m, error)); } /* Change SA state to MATURE */ SAHTREE_WLOCK(); if (sav->state != SADB_SASTATE_LARVAL) { /* SA was deleted or another thread made it MATURE. */ SAHTREE_WUNLOCK(); key_freesav(&sav); return (key_senderror(so, m, ESRCH)); } /* * NOTE: we keep SAs in savtree_alive ordered by created * time. When SA's state changed from LARVAL to MATURE, * we update its created time in key_setsaval() and move * it into head of savtree_alive. */ TAILQ_REMOVE(&sav->sah->savtree_larval, sav, chain); TAILQ_INSERT_HEAD(&sav->sah->savtree_alive, sav, chain); sav->state = SADB_SASTATE_MATURE; SAHTREE_WUNLOCK(); } else { /* * For DYING and MATURE SA we can change only state * and lifetimes. Report EINVAL if something else attempted * to change. */ if (!SADB_CHECKHDR(mhp, SADB_EXT_KEY_ENCRYPT) || !SADB_CHECKHDR(mhp, SADB_EXT_KEY_AUTH)) { key_freesav(&sav); return (key_senderror(so, m, EINVAL)); } error = key_updatelifetimes(sav, mhp); if (error != 0) { key_freesav(&sav); return (key_senderror(so, m, error)); } /* * This is FreeBSD extension to RFC2367. * IKEd can specify SADB_X_EXT_NEW_ADDRESS_SRC and/or * SADB_X_EXT_NEW_ADDRESS_DST when it wants to change * SA addresses (for example to implement MOBIKE protocol * as described in RFC4555). Also we allow to change * NAT-T config. */ if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_SRC) || !SADB_CHECKHDR(mhp, SADB_X_EXT_NEW_ADDRESS_DST) || !SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_TYPE) || sav->natt != NULL) { error = key_updateaddresses(so, m, mhp, sav, &saidx); key_freesav(&sav); if (error != 0) return (key_senderror(so, m, error)); return (0); } /* Check that SA is still alive */ SAHTREE_WLOCK(); if (sav->state == SADB_SASTATE_DEAD) { /* SA was unlinked */ SAHTREE_WUNLOCK(); key_freesav(&sav); return (key_senderror(so, m, ESRCH)); } /* * NOTE: there is possible state moving from DYING to MATURE, * but this doesn't change created time, so we won't reorder * this SA. */ sav->state = SADB_SASTATE_MATURE; SAHTREE_WUNLOCK(); } KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); key_freesav(&sav); { struct mbuf *n; /* set msg buf from mhp */ n = key_getmsgbuf_x1(m, mhp); if (n == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return key_senderror(so, m, ENOBUFS); } m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * SADB_ADD processing * add an entry to SA database, when received * * from the ikmpd, * and send * * to the ikmpd. * * IGNORE identity and sensitivity messages. * * m will always be freed. */ static int key_add(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secasindex saidx; struct sadb_address *src0, *dst0; struct sadb_sa *sa0; struct secasvar *sav; uint32_t reqid, spi; uint8_t mode, proto; int error; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_EXT_SA) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || (mhp->msg->sadb_msg_satype == SADB_SATYPE_ESP && ( SADB_CHECKHDR(mhp, SADB_EXT_KEY_ENCRYPT) || SADB_CHECKLEN(mhp, SADB_EXT_KEY_ENCRYPT))) || (mhp->msg->sadb_msg_satype == SADB_SATYPE_AH && ( SADB_CHECKHDR(mhp, SADB_EXT_KEY_AUTH) || SADB_CHECKLEN(mhp, SADB_EXT_KEY_AUTH))) || (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT)) || (SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_SOFT) && !SADB_CHECKHDR(mhp, SADB_EXT_LIFETIME_HARD))) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_SA) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_X_EXT_SA2)) { mode = IPSEC_MODE_ANY; reqid = 0; } else { if (SADB_CHECKLEN(mhp, SADB_X_EXT_SA2)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } mode = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_mode; reqid = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_reqid; } sa0 = (struct sadb_sa *)mhp->ext[SADB_EXT_SA]; src0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_SRC]; dst0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_DST]; /* * Only SADB_SASTATE_MATURE SAs may be submitted in an * SADB_ADD message. */ if (sa0->sadb_sa_state != SADB_SASTATE_MATURE) { ipseclog((LOG_DEBUG, "%s: invalid state.\n", __func__)); #ifdef PFKEY_STRICT_CHECKS return key_senderror(so, m, EINVAL); #endif } error = key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)); if (error != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, error); } KEY_SETSECASIDX(proto, mode, reqid, src0 + 1, dst0 + 1, &saidx); spi = sa0->sadb_sa_spi; /* * For TCP-MD5 SAs we don't use SPI. Check the uniqueness using * secasindex. * XXXAE: IPComp seems also doesn't use SPI. */ if (proto == IPPROTO_TCP) { sav = key_getsav_tcpmd5(&saidx, &spi); if (sav == NULL && spi == 0) { /* Failed to allocate SPI */ ipseclog((LOG_DEBUG, "%s: SA already exists.\n", __func__)); return key_senderror(so, m, EEXIST); } /* XXX: SPI that we report back can have another value */ } else { /* We can create new SA only if SPI is different. */ sav = key_getsavbyspi(spi); } if (sav != NULL) { key_freesav(&sav); ipseclog((LOG_DEBUG, "%s: SA already exists.\n", __func__)); return key_senderror(so, m, EEXIST); } sav = key_newsav(mhp, &saidx, spi, &error); if (sav == NULL) return key_senderror(so, m, error); KEYDBG(KEY_STAMP, printf("%s: return SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); /* * If SADB_ADD was in response to SADB_ACQUIRE, we need to schedule * ACQ for deletion. */ if (sav->seq != 0) key_acqdone(&saidx, sav->seq); { /* * Don't call key_freesav() on error here, as we would like to * keep the SA in the database. */ struct mbuf *n; /* set msg buf from mhp */ n = key_getmsgbuf_x1(m, mhp); if (n == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return key_senderror(so, m, ENOBUFS); } m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * NAT-T support. * IKEd may request the use ESP in UDP encapsulation when it detects the * presence of NAT. It uses NAT-T extension headers for such SAs to specify * parameters needed for encapsulation and decapsulation. These PF_KEY * extension headers are not standardized, so this comment addresses our * implementation. * SADB_X_EXT_NAT_T_TYPE specifies type of encapsulation, we support only * UDP_ENCAP_ESPINUDP as described in RFC3948. * SADB_X_EXT_NAT_T_SPORT/DPORT specifies source and destination ports for * UDP header. We use these ports in UDP encapsulation procedure, also we * can check them in UDP decapsulation procedure. * SADB_X_EXT_NAT_T_OA[IR] specifies original address of initiator or * responder. These addresses can be used for transport mode to adjust * checksum after decapsulation and decryption. Since original IP addresses * used by peer usually different (we detected presence of NAT), TCP/UDP * pseudo header checksum and IP header checksum was calculated using original * addresses. After decapsulation and decryption we need to adjust checksum * to have correct datagram. * * We expect presence of NAT-T extension headers only in SADB_ADD and * SADB_UPDATE messages. We report NAT-T extension headers in replies * to SADB_ADD, SADB_UPDATE, SADB_GET, and SADB_DUMP messages. */ static int key_setnatt(struct secasvar *sav, const struct sadb_msghdr *mhp) { struct sadb_x_nat_t_port *port; struct sadb_x_nat_t_type *type; struct sadb_address *oai, *oar; struct sockaddr *sa; uint32_t addr; uint16_t cksum; IPSEC_ASSERT(sav->natt == NULL, ("natt is already initialized")); /* * Ignore NAT-T headers if sproto isn't ESP. */ if (sav->sah->saidx.proto != IPPROTO_ESP) return (0); if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_TYPE) && !SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_SPORT) && !SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_DPORT)) { if (SADB_CHECKLEN(mhp, SADB_X_EXT_NAT_T_TYPE) || SADB_CHECKLEN(mhp, SADB_X_EXT_NAT_T_SPORT) || SADB_CHECKLEN(mhp, SADB_X_EXT_NAT_T_DPORT)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return (EINVAL); } } else return (0); type = (struct sadb_x_nat_t_type *)mhp->ext[SADB_X_EXT_NAT_T_TYPE]; if (type->sadb_x_nat_t_type_type != UDP_ENCAP_ESPINUDP) { ipseclog((LOG_DEBUG, "%s: unsupported NAT-T type %u.\n", __func__, type->sadb_x_nat_t_type_type)); return (EINVAL); } /* * Allocate storage for NAT-T config. * On error it will be released by key_cleansav(). */ sav->natt = malloc(sizeof(struct secnatt), M_IPSEC_MISC, M_NOWAIT | M_ZERO); if (sav->natt == NULL) { PFKEYSTAT_INC(in_nomem); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return (ENOBUFS); } port = (struct sadb_x_nat_t_port *)mhp->ext[SADB_X_EXT_NAT_T_SPORT]; if (port->sadb_x_nat_t_port_port == 0) { ipseclog((LOG_DEBUG, "%s: invalid NAT-T sport specified.\n", __func__)); return (EINVAL); } sav->natt->sport = port->sadb_x_nat_t_port_port; port = (struct sadb_x_nat_t_port *)mhp->ext[SADB_X_EXT_NAT_T_DPORT]; if (port->sadb_x_nat_t_port_port == 0) { ipseclog((LOG_DEBUG, "%s: invalid NAT-T dport specified.\n", __func__)); return (EINVAL); } sav->natt->dport = port->sadb_x_nat_t_port_port; /* * SADB_X_EXT_NAT_T_OAI and SADB_X_EXT_NAT_T_OAR are optional * and needed only for transport mode IPsec. * Usually NAT translates only one address, but it is possible, * that both addresses could be translated. * NOTE: Value of SADB_X_EXT_NAT_T_OAI is equal to SADB_X_EXT_NAT_T_OA. */ if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_OAI)) { if (SADB_CHECKLEN(mhp, SADB_X_EXT_NAT_T_OAI)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return (EINVAL); } oai = (struct sadb_address *)mhp->ext[SADB_X_EXT_NAT_T_OAI]; } else oai = NULL; if (!SADB_CHECKHDR(mhp, SADB_X_EXT_NAT_T_OAR)) { if (SADB_CHECKLEN(mhp, SADB_X_EXT_NAT_T_OAR)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return (EINVAL); } oar = (struct sadb_address *)mhp->ext[SADB_X_EXT_NAT_T_OAR]; } else oar = NULL; /* Initialize addresses only for transport mode */ if (sav->sah->saidx.mode != IPSEC_MODE_TUNNEL) { cksum = 0; if (oai != NULL) { /* Currently we support only AF_INET */ sa = (struct sockaddr *)(oai + 1); if (sa->sa_family != AF_INET || sa->sa_len != sizeof(struct sockaddr_in)) { ipseclog((LOG_DEBUG, "%s: wrong NAT-OAi header.\n", __func__)); return (EINVAL); } /* Ignore address if it the same */ if (((struct sockaddr_in *)sa)->sin_addr.s_addr != sav->sah->saidx.src.sin.sin_addr.s_addr) { bcopy(sa, &sav->natt->oai.sa, sa->sa_len); sav->natt->flags |= IPSEC_NATT_F_OAI; /* Calculate checksum delta */ addr = sav->sah->saidx.src.sin.sin_addr.s_addr; cksum = in_addword(cksum, ~addr >> 16); cksum = in_addword(cksum, ~addr & 0xffff); addr = sav->natt->oai.sin.sin_addr.s_addr; cksum = in_addword(cksum, addr >> 16); cksum = in_addword(cksum, addr & 0xffff); } } if (oar != NULL) { /* Currently we support only AF_INET */ sa = (struct sockaddr *)(oar + 1); if (sa->sa_family != AF_INET || sa->sa_len != sizeof(struct sockaddr_in)) { ipseclog((LOG_DEBUG, "%s: wrong NAT-OAr header.\n", __func__)); return (EINVAL); } /* Ignore address if it the same */ if (((struct sockaddr_in *)sa)->sin_addr.s_addr != sav->sah->saidx.dst.sin.sin_addr.s_addr) { bcopy(sa, &sav->natt->oar.sa, sa->sa_len); sav->natt->flags |= IPSEC_NATT_F_OAR; /* Calculate checksum delta */ addr = sav->sah->saidx.dst.sin.sin_addr.s_addr; cksum = in_addword(cksum, ~addr >> 16); cksum = in_addword(cksum, ~addr & 0xffff); addr = sav->natt->oar.sin.sin_addr.s_addr; cksum = in_addword(cksum, addr >> 16); cksum = in_addword(cksum, addr & 0xffff); } } sav->natt->cksum = cksum; } return (0); } static int key_setident(struct secashead *sah, const struct sadb_msghdr *mhp) { const struct sadb_ident *idsrc, *iddst; IPSEC_ASSERT(sah != NULL, ("null secashead")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* don't make buffer if not there */ if (SADB_CHECKHDR(mhp, SADB_EXT_IDENTITY_SRC) && SADB_CHECKHDR(mhp, SADB_EXT_IDENTITY_DST)) { sah->idents = NULL; sah->identd = NULL; return (0); } if (SADB_CHECKHDR(mhp, SADB_EXT_IDENTITY_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_IDENTITY_DST)) { ipseclog((LOG_DEBUG, "%s: invalid identity.\n", __func__)); return (EINVAL); } idsrc = (const struct sadb_ident *)mhp->ext[SADB_EXT_IDENTITY_SRC]; iddst = (const struct sadb_ident *)mhp->ext[SADB_EXT_IDENTITY_DST]; /* validity check */ if (idsrc->sadb_ident_type != iddst->sadb_ident_type) { ipseclog((LOG_DEBUG, "%s: ident type mismatch.\n", __func__)); return EINVAL; } switch (idsrc->sadb_ident_type) { case SADB_IDENTTYPE_PREFIX: case SADB_IDENTTYPE_FQDN: case SADB_IDENTTYPE_USERFQDN: default: /* XXX do nothing */ sah->idents = NULL; sah->identd = NULL; return 0; } /* make structure */ sah->idents = malloc(sizeof(struct secident), M_IPSEC_MISC, M_NOWAIT); if (sah->idents == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return ENOBUFS; } sah->identd = malloc(sizeof(struct secident), M_IPSEC_MISC, M_NOWAIT); if (sah->identd == NULL) { free(sah->idents, M_IPSEC_MISC); sah->idents = NULL; ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return ENOBUFS; } sah->idents->type = idsrc->sadb_ident_type; sah->idents->id = idsrc->sadb_ident_id; sah->identd->type = iddst->sadb_ident_type; sah->identd->id = iddst->sadb_ident_id; return 0; } /* * m will not be freed on return. * it is caller's responsibility to free the result. * * Called from SADB_ADD and SADB_UPDATE. Reply will contain headers * from the request in defined order. */ static struct mbuf * key_getmsgbuf_x1(struct mbuf *m, const struct sadb_msghdr *mhp) { struct mbuf *n; IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* create new sadb_msg to reply. */ n = key_gather_mbuf(m, mhp, 1, 16, SADB_EXT_RESERVED, SADB_EXT_SA, SADB_X_EXT_SA2, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST, SADB_EXT_LIFETIME_HARD, SADB_EXT_LIFETIME_SOFT, SADB_EXT_IDENTITY_SRC, SADB_EXT_IDENTITY_DST, SADB_X_EXT_NAT_T_TYPE, SADB_X_EXT_NAT_T_SPORT, SADB_X_EXT_NAT_T_DPORT, SADB_X_EXT_NAT_T_OAI, SADB_X_EXT_NAT_T_OAR, SADB_X_EXT_NEW_ADDRESS_SRC, SADB_X_EXT_NEW_ADDRESS_DST); if (!n) return NULL; if (n->m_len < sizeof(struct sadb_msg)) { n = m_pullup(n, sizeof(struct sadb_msg)); if (n == NULL) return NULL; } mtod(n, struct sadb_msg *)->sadb_msg_errno = 0; mtod(n, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); return n; } /* * SADB_DELETE processing * receive * * from the ikmpd, and set SADB_SASTATE_DEAD, * and send, * * to the ikmpd. * * m will always be freed. */ static int key_delete(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secasindex saidx; struct sadb_address *src0, *dst0; struct secasvar *sav; struct sadb_sa *sa0; uint8_t proto; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST)) { ipseclog((LOG_DEBUG, "%s: invalid message is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } src0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_SRC]); dst0 = (struct sadb_address *)(mhp->ext[SADB_EXT_ADDRESS_DST]); if (key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)) != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return (key_senderror(so, m, EINVAL)); } KEY_SETSECASIDX(proto, IPSEC_MODE_ANY, 0, src0 + 1, dst0 + 1, &saidx); if (SADB_CHECKHDR(mhp, SADB_EXT_SA)) { /* * Caller wants us to delete all non-LARVAL SAs * that match the src/dst. This is used during * IKE INITIAL-CONTACT. * XXXAE: this looks like some extension to RFC2367. */ ipseclog((LOG_DEBUG, "%s: doing delete all.\n", __func__)); return (key_delete_all(so, m, mhp, &saidx)); } if (SADB_CHECKLEN(mhp, SADB_EXT_SA)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return (key_senderror(so, m, EINVAL)); } sa0 = (struct sadb_sa *)mhp->ext[SADB_EXT_SA]; if (proto == IPPROTO_TCP) sav = key_getsav_tcpmd5(&saidx, NULL); else sav = key_getsavbyspi(sa0->sadb_sa_spi); if (sav == NULL) { ipseclog((LOG_DEBUG, "%s: no SA found for SPI %u.\n", __func__, ntohl(sa0->sadb_sa_spi))); return (key_senderror(so, m, ESRCH)); } if (key_cmpsaidx(&sav->sah->saidx, &saidx, CMP_HEAD) == 0) { ipseclog((LOG_DEBUG, "%s: saidx mismatched for SPI %u.\n", __func__, ntohl(sav->spi))); key_freesav(&sav); return (key_senderror(so, m, ESRCH)); } KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); key_unlinksav(sav); key_freesav(&sav); { struct mbuf *n; struct sadb_msg *newmsg; /* create new sadb_msg to reply. */ n = key_gather_mbuf(m, mhp, 1, 4, SADB_EXT_RESERVED, SADB_EXT_SA, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); if (!n) return key_senderror(so, m, ENOBUFS); if (n->m_len < sizeof(struct sadb_msg)) { n = m_pullup(n, sizeof(struct sadb_msg)); if (n == NULL) return key_senderror(so, m, ENOBUFS); } newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * delete all SAs for src/dst. Called from key_delete(). */ static int key_delete_all(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp, struct secasindex *saidx) { struct secasvar_queue drainq; struct secashead *sah; struct secasvar *sav, *nextsav; TAILQ_INIT(&drainq); SAHTREE_WLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(saidx), addrhash) { if (key_cmpsaidx(&sah->saidx, saidx, CMP_HEAD) == 0) continue; /* Move all ALIVE SAs into drainq */ TAILQ_CONCAT(&drainq, &sah->savtree_alive, chain); } /* Unlink all queued SAs from SPI hash */ TAILQ_FOREACH(sav, &drainq, chain) { sav->state = SADB_SASTATE_DEAD; LIST_REMOVE(sav, spihash); } SAHTREE_WUNLOCK(); /* Now we can release reference for all SAs in drainq */ sav = TAILQ_FIRST(&drainq); while (sav != NULL) { KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); nextsav = TAILQ_NEXT(sav, chain); key_freesah(&sav->sah); /* release reference from SAV */ key_freesav(&sav); /* release last reference */ sav = nextsav; } { struct mbuf *n; struct sadb_msg *newmsg; /* create new sadb_msg to reply. */ n = key_gather_mbuf(m, mhp, 1, 3, SADB_EXT_RESERVED, SADB_EXT_ADDRESS_SRC, SADB_EXT_ADDRESS_DST); if (!n) return key_senderror(so, m, ENOBUFS); if (n->m_len < sizeof(struct sadb_msg)) { n = m_pullup(n, sizeof(struct sadb_msg)); if (n == NULL) return key_senderror(so, m, ENOBUFS); } newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(n->m_pkthdr.len); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ALL); } } /* * Delete all alive SAs for corresponding xform. * Larval SAs have not initialized tdb_xform, so it is safe to leave them * here when xform disappears. */ static void key_delete_xform(const struct xformsw *xsp) { struct secasvar_queue drainq; struct secashead *sah; struct secasvar *sav, *nextsav; TAILQ_INIT(&drainq); SAHTREE_WLOCK(); TAILQ_FOREACH(sah, &V_sahtree, chain) { sav = TAILQ_FIRST(&sah->savtree_alive); if (sav == NULL) continue; if (sav->tdb_xform != xsp) continue; /* * It is supposed that all SAs in the chain are related to * one xform. */ TAILQ_CONCAT(&drainq, &sah->savtree_alive, chain); } /* Unlink all queued SAs from SPI hash */ TAILQ_FOREACH(sav, &drainq, chain) { sav->state = SADB_SASTATE_DEAD; LIST_REMOVE(sav, spihash); } SAHTREE_WUNLOCK(); /* Now we can release reference for all SAs in drainq */ sav = TAILQ_FIRST(&drainq); while (sav != NULL) { KEYDBG(KEY_STAMP, printf("%s: SA(%p)\n", __func__, sav)); KEYDBG(KEY_DATA, kdebug_secasv(sav)); nextsav = TAILQ_NEXT(sav, chain); key_freesah(&sav->sah); /* release reference from SAV */ key_freesav(&sav); /* release last reference */ sav = nextsav; } } /* * SADB_GET processing * receive * * from the ikmpd, and get a SP and a SA to respond, * and send, * * to the ikmpd. * * m will always be freed. */ static int key_get(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secasindex saidx; struct sadb_address *src0, *dst0; struct sadb_sa *sa0; struct secasvar *sav; uint8_t proto; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_EXT_SA) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST)) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_SA) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } sa0 = (struct sadb_sa *)mhp->ext[SADB_EXT_SA]; src0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_SRC]; dst0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_DST]; if (key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)) != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, EINVAL); } KEY_SETSECASIDX(proto, IPSEC_MODE_ANY, 0, src0 + 1, dst0 + 1, &saidx); if (proto == IPPROTO_TCP) sav = key_getsav_tcpmd5(&saidx, NULL); else sav = key_getsavbyspi(sa0->sadb_sa_spi); if (sav == NULL) { ipseclog((LOG_DEBUG, "%s: no SA found.\n", __func__)); return key_senderror(so, m, ESRCH); } if (key_cmpsaidx(&sav->sah->saidx, &saidx, CMP_HEAD) == 0) { ipseclog((LOG_DEBUG, "%s: saidx mismatched for SPI %u.\n", __func__, ntohl(sa0->sadb_sa_spi))); key_freesav(&sav); return (key_senderror(so, m, ESRCH)); } { struct mbuf *n; uint8_t satype; /* map proto to satype */ if ((satype = key_proto2satype(sav->sah->saidx.proto)) == 0) { ipseclog((LOG_DEBUG, "%s: there was invalid proto in SAD.\n", __func__)); key_freesav(&sav); return key_senderror(so, m, EINVAL); } /* create new sadb_msg to reply. */ n = key_setdumpsa(sav, SADB_GET, satype, mhp->msg->sadb_msg_seq, mhp->msg->sadb_msg_pid); key_freesav(&sav); if (!n) return key_senderror(so, m, ENOBUFS); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } } /* XXX make it sysctl-configurable? */ static void key_getcomb_setlifetime(struct sadb_comb *comb) { comb->sadb_comb_soft_allocations = 1; comb->sadb_comb_hard_allocations = 1; comb->sadb_comb_soft_bytes = 0; comb->sadb_comb_hard_bytes = 0; comb->sadb_comb_hard_addtime = 86400; /* 1 day */ comb->sadb_comb_soft_addtime = comb->sadb_comb_soft_addtime * 80 / 100; comb->sadb_comb_soft_usetime = 28800; /* 8 hours */ comb->sadb_comb_hard_usetime = comb->sadb_comb_hard_usetime * 80 / 100; } /* * XXX reorder combinations by preference * XXX no idea if the user wants ESP authentication or not */ static struct mbuf * key_getcomb_ealg(void) { struct sadb_comb *comb; const struct enc_xform *algo; struct mbuf *result = NULL, *m, *n; int encmin; int i, off, o; int totlen; const int l = PFKEY_ALIGN8(sizeof(struct sadb_comb)); m = NULL; for (i = 1; i <= SADB_EALG_MAX; i++) { algo = enc_algorithm_lookup(i); if (algo == NULL) continue; /* discard algorithms with key size smaller than system min */ if (_BITS(algo->maxkey) < V_ipsec_esp_keymin) continue; if (_BITS(algo->minkey) < V_ipsec_esp_keymin) encmin = V_ipsec_esp_keymin; else encmin = _BITS(algo->minkey); if (V_ipsec_esp_auth) m = key_getcomb_ah(); else { IPSEC_ASSERT(l <= MLEN, ("l=%u > MLEN=%lu", l, (u_long) MLEN)); MGET(m, M_NOWAIT, MT_DATA); if (m) { M_ALIGN(m, l); m->m_len = l; m->m_next = NULL; bzero(mtod(m, caddr_t), m->m_len); } } if (!m) goto fail; totlen = 0; for (n = m; n; n = n->m_next) totlen += n->m_len; IPSEC_ASSERT((totlen % l) == 0, ("totlen=%u, l=%u", totlen, l)); for (off = 0; off < totlen; off += l) { n = m_pulldown(m, off, l, &o); if (!n) { /* m is already freed */ goto fail; } comb = (struct sadb_comb *)(mtod(n, caddr_t) + o); bzero(comb, sizeof(*comb)); key_getcomb_setlifetime(comb); comb->sadb_comb_encrypt = i; comb->sadb_comb_encrypt_minbits = encmin; comb->sadb_comb_encrypt_maxbits = _BITS(algo->maxkey); } if (!result) result = m; else m_cat(result, m); } return result; fail: if (result) m_freem(result); return NULL; } static void key_getsizes_ah(const struct auth_hash *ah, int alg, u_int16_t* min, u_int16_t* max) { *min = *max = ah->hashsize; if (ah->keysize == 0) { /* * Transform takes arbitrary key size but algorithm * key size is restricted. Enforce this here. */ switch (alg) { case SADB_X_AALG_MD5: *min = *max = 16; break; case SADB_X_AALG_SHA: *min = *max = 20; break; case SADB_X_AALG_NULL: *min = 1; *max = 256; break; case SADB_X_AALG_SHA2_256: *min = *max = 32; break; case SADB_X_AALG_SHA2_384: *min = *max = 48; break; case SADB_X_AALG_SHA2_512: *min = *max = 64; break; default: DPRINTF(("%s: unknown AH algorithm %u\n", __func__, alg)); break; } } } /* * XXX reorder combinations by preference */ static struct mbuf * key_getcomb_ah() { const struct auth_hash *algo; struct sadb_comb *comb; struct mbuf *m; u_int16_t minkeysize, maxkeysize; int i; const int l = PFKEY_ALIGN8(sizeof(struct sadb_comb)); m = NULL; for (i = 1; i <= SADB_AALG_MAX; i++) { #if 1 /* we prefer HMAC algorithms, not old algorithms */ if (i != SADB_AALG_SHA1HMAC && i != SADB_AALG_MD5HMAC && i != SADB_X_AALG_SHA2_256 && i != SADB_X_AALG_SHA2_384 && i != SADB_X_AALG_SHA2_512) continue; #endif algo = auth_algorithm_lookup(i); if (!algo) continue; key_getsizes_ah(algo, i, &minkeysize, &maxkeysize); /* discard algorithms with key size smaller than system min */ if (_BITS(minkeysize) < V_ipsec_ah_keymin) continue; if (!m) { IPSEC_ASSERT(l <= MLEN, ("l=%u > MLEN=%lu", l, (u_long) MLEN)); MGET(m, M_NOWAIT, MT_DATA); if (m) { M_ALIGN(m, l); m->m_len = l; m->m_next = NULL; } } else M_PREPEND(m, l, M_NOWAIT); if (!m) return NULL; comb = mtod(m, struct sadb_comb *); bzero(comb, sizeof(*comb)); key_getcomb_setlifetime(comb); comb->sadb_comb_auth = i; comb->sadb_comb_auth_minbits = _BITS(minkeysize); comb->sadb_comb_auth_maxbits = _BITS(maxkeysize); } return m; } /* * not really an official behavior. discussed in pf_key@inner.net in Sep2000. * XXX reorder combinations by preference */ static struct mbuf * key_getcomb_ipcomp() { const struct comp_algo *algo; struct sadb_comb *comb; struct mbuf *m; int i; const int l = PFKEY_ALIGN8(sizeof(struct sadb_comb)); m = NULL; for (i = 1; i <= SADB_X_CALG_MAX; i++) { algo = comp_algorithm_lookup(i); if (!algo) continue; if (!m) { IPSEC_ASSERT(l <= MLEN, ("l=%u > MLEN=%lu", l, (u_long) MLEN)); MGET(m, M_NOWAIT, MT_DATA); if (m) { M_ALIGN(m, l); m->m_len = l; m->m_next = NULL; } } else M_PREPEND(m, l, M_NOWAIT); if (!m) return NULL; comb = mtod(m, struct sadb_comb *); bzero(comb, sizeof(*comb)); key_getcomb_setlifetime(comb); comb->sadb_comb_encrypt = i; /* what should we set into sadb_comb_*_{min,max}bits? */ } return m; } /* * XXX no way to pass mode (transport/tunnel) to userland * XXX replay checking? * XXX sysctl interface to ipsec_{ah,esp}_keymin */ static struct mbuf * key_getprop(const struct secasindex *saidx) { struct sadb_prop *prop; struct mbuf *m, *n; const int l = PFKEY_ALIGN8(sizeof(struct sadb_prop)); int totlen; switch (saidx->proto) { case IPPROTO_ESP: m = key_getcomb_ealg(); break; case IPPROTO_AH: m = key_getcomb_ah(); break; case IPPROTO_IPCOMP: m = key_getcomb_ipcomp(); break; default: return NULL; } if (!m) return NULL; M_PREPEND(m, l, M_NOWAIT); if (!m) return NULL; totlen = 0; for (n = m; n; n = n->m_next) totlen += n->m_len; prop = mtod(m, struct sadb_prop *); bzero(prop, sizeof(*prop)); prop->sadb_prop_len = PFKEY_UNIT64(totlen); prop->sadb_prop_exttype = SADB_EXT_PROPOSAL; prop->sadb_prop_replay = 32; /* XXX */ return m; } /* * SADB_ACQUIRE processing called by key_checkrequest() and key_acquire2(). * send * * to KMD, and expect to receive * with SADB_ACQUIRE if error occurred, * or * with SADB_GETSPI * from KMD by PF_KEY. * * XXX x_policy is outside of RFC2367 (KAME extension). * XXX sensitivity is not supported. * XXX for ipcomp, RFC2367 does not define how to fill in proposal. * see comment for key_getcomb_ipcomp(). * * OUT: * 0 : succeed * others: error number */ static int key_acquire(const struct secasindex *saidx, struct secpolicy *sp) { union sockaddr_union addr; struct mbuf *result, *m; uint32_t seq; int error; uint16_t ul_proto; uint8_t mask, satype; IPSEC_ASSERT(saidx != NULL, ("null saidx")); satype = key_proto2satype(saidx->proto); IPSEC_ASSERT(satype != 0, ("null satype, protocol %u", saidx->proto)); error = -1; result = NULL; ul_proto = IPSEC_ULPROTO_ANY; /* Get seq number to check whether sending message or not. */ seq = key_getacq(saidx, &error); if (seq == 0) return (error); m = key_setsadbmsg(SADB_ACQUIRE, 0, satype, seq, 0, 0); if (!m) { error = ENOBUFS; goto fail; } result = m; /* * set sadb_address for saidx's. * * Note that if sp is supplied, then we're being called from * key_allocsa_policy() and should supply port and protocol * information. * XXXAE: why only TCP and UDP? ICMP and SCTP looks applicable too. * XXXAE: probably we can handle this in the ipsec[46]_allocsa(). * XXXAE: it looks like we should save this info in the ACQ entry. */ if (sp != NULL && (sp->spidx.ul_proto == IPPROTO_TCP || sp->spidx.ul_proto == IPPROTO_UDP)) ul_proto = sp->spidx.ul_proto; addr = saidx->src; mask = FULLMASK; if (ul_proto != IPSEC_ULPROTO_ANY) { switch (sp->spidx.src.sa.sa_family) { case AF_INET: if (sp->spidx.src.sin.sin_port != IPSEC_PORT_ANY) { addr.sin.sin_port = sp->spidx.src.sin.sin_port; mask = sp->spidx.prefs; } break; case AF_INET6: if (sp->spidx.src.sin6.sin6_port != IPSEC_PORT_ANY) { addr.sin6.sin6_port = sp->spidx.src.sin6.sin6_port; mask = sp->spidx.prefs; } break; default: break; } } m = key_setsadbaddr(SADB_EXT_ADDRESS_SRC, &addr.sa, mask, ul_proto); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); addr = saidx->dst; mask = FULLMASK; if (ul_proto != IPSEC_ULPROTO_ANY) { switch (sp->spidx.dst.sa.sa_family) { case AF_INET: if (sp->spidx.dst.sin.sin_port != IPSEC_PORT_ANY) { addr.sin.sin_port = sp->spidx.dst.sin.sin_port; mask = sp->spidx.prefd; } break; case AF_INET6: if (sp->spidx.dst.sin6.sin6_port != IPSEC_PORT_ANY) { addr.sin6.sin6_port = sp->spidx.dst.sin6.sin6_port; mask = sp->spidx.prefd; } break; default: break; } } m = key_setsadbaddr(SADB_EXT_ADDRESS_DST, &addr.sa, mask, ul_proto); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* XXX proxy address (optional) */ /* set sadb_x_policy */ if (sp != NULL) { m = key_setsadbxpolicy(sp->policy, sp->spidx.dir, sp->id, sp->priority); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); } /* XXX identity (optional) */ #if 0 if (idexttype && fqdn) { /* create identity extension (FQDN) */ struct sadb_ident *id; int fqdnlen; fqdnlen = strlen(fqdn) + 1; /* +1 for terminating-NUL */ id = (struct sadb_ident *)p; bzero(id, sizeof(*id) + PFKEY_ALIGN8(fqdnlen)); id->sadb_ident_len = PFKEY_UNIT64(sizeof(*id) + PFKEY_ALIGN8(fqdnlen)); id->sadb_ident_exttype = idexttype; id->sadb_ident_type = SADB_IDENTTYPE_FQDN; bcopy(fqdn, id + 1, fqdnlen); p += sizeof(struct sadb_ident) + PFKEY_ALIGN8(fqdnlen); } if (idexttype) { /* create identity extension (USERFQDN) */ struct sadb_ident *id; int userfqdnlen; if (userfqdn) { /* +1 for terminating-NUL */ userfqdnlen = strlen(userfqdn) + 1; } else userfqdnlen = 0; id = (struct sadb_ident *)p; bzero(id, sizeof(*id) + PFKEY_ALIGN8(userfqdnlen)); id->sadb_ident_len = PFKEY_UNIT64(sizeof(*id) + PFKEY_ALIGN8(userfqdnlen)); id->sadb_ident_exttype = idexttype; id->sadb_ident_type = SADB_IDENTTYPE_USERFQDN; /* XXX is it correct? */ if (curproc && curproc->p_cred) id->sadb_ident_id = curproc->p_cred->p_ruid; if (userfqdn && userfqdnlen) bcopy(userfqdn, id + 1, userfqdnlen); p += sizeof(struct sadb_ident) + PFKEY_ALIGN8(userfqdnlen); } #endif /* XXX sensitivity (optional) */ /* create proposal/combination extension */ m = key_getprop(saidx); #if 0 /* * spec conformant: always attach proposal/combination extension, * the problem is that we have no way to attach it for ipcomp, * due to the way sadb_comb is declared in RFC2367. */ if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); #else /* * outside of spec; make proposal/combination extension optional. */ if (m) m_cat(result, m); #endif if ((result->m_flags & M_PKTHDR) == 0) { error = EINVAL; goto fail; } if (result->m_len < sizeof(struct sadb_msg)) { result = m_pullup(result, sizeof(struct sadb_msg)); if (result == NULL) { error = ENOBUFS; goto fail; } } result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); KEYDBG(KEY_STAMP, printf("%s: SP(%p)\n", __func__, sp)); KEYDBG(KEY_DATA, kdebug_secasindex(saidx, NULL)); return key_sendup_mbuf(NULL, result, KEY_SENDUP_REGISTERED); fail: if (result) m_freem(result); return error; } static uint32_t key_newacq(const struct secasindex *saidx, int *perror) { struct secacq *acq; uint32_t seq; acq = malloc(sizeof(*acq), M_IPSEC_SAQ, M_NOWAIT | M_ZERO); if (acq == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); *perror = ENOBUFS; return (0); } /* copy secindex */ bcopy(saidx, &acq->saidx, sizeof(acq->saidx)); acq->created = time_second; acq->count = 0; /* add to acqtree */ ACQ_LOCK(); seq = acq->seq = (V_acq_seq == ~0 ? 1 : ++V_acq_seq); LIST_INSERT_HEAD(&V_acqtree, acq, chain); LIST_INSERT_HEAD(ACQADDRHASH_HASH(saidx), acq, addrhash); LIST_INSERT_HEAD(ACQSEQHASH_HASH(seq), acq, seqhash); ACQ_UNLOCK(); *perror = 0; return (seq); } static uint32_t key_getacq(const struct secasindex *saidx, int *perror) { struct secacq *acq; uint32_t seq; ACQ_LOCK(); LIST_FOREACH(acq, ACQADDRHASH_HASH(saidx), addrhash) { if (key_cmpsaidx(&acq->saidx, saidx, CMP_EXACTLY)) { if (acq->count > V_key_blockacq_count) { /* * Reset counter and send message. * Also reset created time to keep ACQ for * this saidx. */ acq->created = time_second; acq->count = 0; seq = acq->seq; } else { /* * Increment counter and do nothing. * We send SADB_ACQUIRE message only * for each V_key_blockacq_count packet. */ acq->count++; seq = 0; } break; } } ACQ_UNLOCK(); if (acq != NULL) { *perror = 0; return (seq); } /* allocate new entry */ return (key_newacq(saidx, perror)); } static int key_acqreset(uint32_t seq) { struct secacq *acq; ACQ_LOCK(); LIST_FOREACH(acq, ACQSEQHASH_HASH(seq), seqhash) { if (acq->seq == seq) { acq->count = 0; acq->created = time_second; break; } } ACQ_UNLOCK(); if (acq == NULL) return (ESRCH); return (0); } /* * Mark ACQ entry as stale to remove it in key_flush_acq(). * Called after successful SADB_GETSPI message. */ static int key_acqdone(const struct secasindex *saidx, uint32_t seq) { struct secacq *acq; ACQ_LOCK(); LIST_FOREACH(acq, ACQSEQHASH_HASH(seq), seqhash) { if (acq->seq == seq) break; } if (acq != NULL) { if (key_cmpsaidx(&acq->saidx, saidx, CMP_EXACTLY) == 0) { ipseclog((LOG_DEBUG, "%s: Mismatched saidx for ACQ %u", __func__, seq)); acq = NULL; } else { acq->created = 0; } } else { ipseclog((LOG_DEBUG, "%s: ACQ %u is not found.", __func__, seq)); } ACQ_UNLOCK(); if (acq == NULL) return (ESRCH); return (0); } static struct secspacq * key_newspacq(struct secpolicyindex *spidx) { struct secspacq *acq; /* get new entry */ acq = malloc(sizeof(struct secspacq), M_IPSEC_SAQ, M_NOWAIT|M_ZERO); if (acq == NULL) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return NULL; } /* copy secindex */ bcopy(spidx, &acq->spidx, sizeof(acq->spidx)); acq->created = time_second; acq->count = 0; /* add to spacqtree */ SPACQ_LOCK(); LIST_INSERT_HEAD(&V_spacqtree, acq, chain); SPACQ_UNLOCK(); return acq; } static struct secspacq * key_getspacq(struct secpolicyindex *spidx) { struct secspacq *acq; SPACQ_LOCK(); LIST_FOREACH(acq, &V_spacqtree, chain) { if (key_cmpspidx_exactly(spidx, &acq->spidx)) { /* NB: return holding spacq_lock */ return acq; } } SPACQ_UNLOCK(); return NULL; } /* * SADB_ACQUIRE processing, * in first situation, is receiving * * from the ikmpd, and clear sequence of its secasvar entry. * * In second situation, is receiving * * from a user land process, and return * * to the socket. * * m will always be freed. */ static int key_acquire2(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { SAHTREE_RLOCK_TRACKER; struct sadb_address *src0, *dst0; struct secasindex saidx; struct secashead *sah; uint32_t reqid; int error; uint8_t mode, proto; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* * Error message from KMd. * We assume that if error was occurred in IKEd, the length of PFKEY * message is equal to the size of sadb_msg structure. * We do not raise error even if error occurred in this function. */ if (mhp->msg->sadb_msg_len == PFKEY_UNIT64(sizeof(struct sadb_msg))) { /* check sequence number */ if (mhp->msg->sadb_msg_seq == 0 || mhp->msg->sadb_msg_errno == 0) { ipseclog((LOG_DEBUG, "%s: must specify sequence " "number and errno.\n", __func__)); } else { /* * IKEd reported that error occurred. * XXXAE: what it expects from the kernel? * Probably we should send SADB_ACQUIRE again? * If so, reset ACQ's state. * XXXAE: it looks useless. */ key_acqreset(mhp->msg->sadb_msg_seq); } m_freem(m); return (0); } /* * This message is from user land. */ /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKHDR(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKHDR(mhp, SADB_EXT_PROPOSAL)) { ipseclog((LOG_DEBUG, "%s: invalid message: missing required header.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_SRC) || SADB_CHECKLEN(mhp, SADB_EXT_ADDRESS_DST) || SADB_CHECKLEN(mhp, SADB_EXT_PROPOSAL)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } if (SADB_CHECKHDR(mhp, SADB_X_EXT_SA2)) { mode = IPSEC_MODE_ANY; reqid = 0; } else { if (SADB_CHECKLEN(mhp, SADB_X_EXT_SA2)) { ipseclog((LOG_DEBUG, "%s: invalid message: wrong header size.\n", __func__)); return key_senderror(so, m, EINVAL); } mode = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_mode; reqid = ((struct sadb_x_sa2 *) mhp->ext[SADB_X_EXT_SA2])->sadb_x_sa2_reqid; } src0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_SRC]; dst0 = (struct sadb_address *)mhp->ext[SADB_EXT_ADDRESS_DST]; error = key_checksockaddrs((struct sockaddr *)(src0 + 1), (struct sockaddr *)(dst0 + 1)); if (error != 0) { ipseclog((LOG_DEBUG, "%s: invalid sockaddr.\n", __func__)); return key_senderror(so, m, EINVAL); } KEY_SETSECASIDX(proto, mode, reqid, src0 + 1, dst0 + 1, &saidx); /* get a SA index */ SAHTREE_RLOCK(); LIST_FOREACH(sah, SAHADDRHASH_HASH(&saidx), addrhash) { if (key_cmpsaidx(&sah->saidx, &saidx, CMP_MODE_REQID)) break; } SAHTREE_RUNLOCK(); if (sah != NULL) { ipseclog((LOG_DEBUG, "%s: a SA exists already.\n", __func__)); return key_senderror(so, m, EEXIST); } error = key_acquire(&saidx, NULL); if (error != 0) { ipseclog((LOG_DEBUG, "%s: error %d returned from key_acquire()\n", __func__, error)); return key_senderror(so, m, error); } m_freem(m); return (0); } /* * SADB_REGISTER processing. * If SATYPE_UNSPEC has been passed as satype, only return sabd_supported. * receive * * from the ikmpd, and register a socket to send PF_KEY messages, * and send * * to KMD by PF_KEY. * If socket is detached, must free from regnode. * * m will always be freed. */ static int key_register(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secreg *reg, *newreg = NULL; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* check for invalid register message */ if (mhp->msg->sadb_msg_satype >= sizeof(V_regtree)/sizeof(V_regtree[0])) return key_senderror(so, m, EINVAL); /* When SATYPE_UNSPEC is specified, only return sabd_supported. */ if (mhp->msg->sadb_msg_satype == SADB_SATYPE_UNSPEC) goto setmsg; /* check whether existing or not */ REGTREE_LOCK(); LIST_FOREACH(reg, &V_regtree[mhp->msg->sadb_msg_satype], chain) { if (reg->so == so) { REGTREE_UNLOCK(); ipseclog((LOG_DEBUG, "%s: socket exists already.\n", __func__)); return key_senderror(so, m, EEXIST); } } /* create regnode */ newreg = malloc(sizeof(struct secreg), M_IPSEC_SAR, M_NOWAIT|M_ZERO); if (newreg == NULL) { REGTREE_UNLOCK(); ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return key_senderror(so, m, ENOBUFS); } newreg->so = so; ((struct keycb *)sotorawcb(so))->kp_registered++; /* add regnode to regtree. */ LIST_INSERT_HEAD(&V_regtree[mhp->msg->sadb_msg_satype], newreg, chain); REGTREE_UNLOCK(); setmsg: { struct mbuf *n; struct sadb_msg *newmsg; struct sadb_supported *sup; u_int len, alen, elen; int off; int i; struct sadb_alg *alg; /* create new sadb_msg to reply. */ alen = 0; for (i = 1; i <= SADB_AALG_MAX; i++) { if (auth_algorithm_lookup(i)) alen += sizeof(struct sadb_alg); } if (alen) alen += sizeof(struct sadb_supported); elen = 0; for (i = 1; i <= SADB_EALG_MAX; i++) { if (enc_algorithm_lookup(i)) elen += sizeof(struct sadb_alg); } if (elen) elen += sizeof(struct sadb_supported); len = sizeof(struct sadb_msg) + alen + elen; if (len > MCLBYTES) return key_senderror(so, m, ENOBUFS); MGETHDR(n, M_NOWAIT, MT_DATA); if (len > MHLEN) { if (!(MCLGET(n, M_NOWAIT))) { m_freem(n); n = NULL; } } if (!n) return key_senderror(so, m, ENOBUFS); n->m_pkthdr.len = n->m_len = len; n->m_next = NULL; off = 0; m_copydata(m, 0, sizeof(struct sadb_msg), mtod(n, caddr_t) + off); newmsg = mtod(n, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(len); off += PFKEY_ALIGN8(sizeof(struct sadb_msg)); /* for authentication algorithm */ if (alen) { sup = (struct sadb_supported *)(mtod(n, caddr_t) + off); sup->sadb_supported_len = PFKEY_UNIT64(alen); sup->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH; off += PFKEY_ALIGN8(sizeof(*sup)); for (i = 1; i <= SADB_AALG_MAX; i++) { const struct auth_hash *aalgo; u_int16_t minkeysize, maxkeysize; aalgo = auth_algorithm_lookup(i); if (!aalgo) continue; alg = (struct sadb_alg *)(mtod(n, caddr_t) + off); alg->sadb_alg_id = i; alg->sadb_alg_ivlen = 0; key_getsizes_ah(aalgo, i, &minkeysize, &maxkeysize); alg->sadb_alg_minbits = _BITS(minkeysize); alg->sadb_alg_maxbits = _BITS(maxkeysize); off += PFKEY_ALIGN8(sizeof(*alg)); } } /* for encryption algorithm */ if (elen) { sup = (struct sadb_supported *)(mtod(n, caddr_t) + off); sup->sadb_supported_len = PFKEY_UNIT64(elen); sup->sadb_supported_exttype = SADB_EXT_SUPPORTED_ENCRYPT; off += PFKEY_ALIGN8(sizeof(*sup)); for (i = 1; i <= SADB_EALG_MAX; i++) { const struct enc_xform *ealgo; ealgo = enc_algorithm_lookup(i); if (!ealgo) continue; alg = (struct sadb_alg *)(mtod(n, caddr_t) + off); alg->sadb_alg_id = i; alg->sadb_alg_ivlen = ealgo->ivsize; alg->sadb_alg_minbits = _BITS(ealgo->minkey); alg->sadb_alg_maxbits = _BITS(ealgo->maxkey); off += PFKEY_ALIGN8(sizeof(struct sadb_alg)); } } IPSEC_ASSERT(off == len, ("length assumption failed (off %u len %u)", off, len)); m_freem(m); return key_sendup_mbuf(so, n, KEY_SENDUP_REGISTERED); } } /* * free secreg entry registered. * XXX: I want to do free a socket marked done SADB_RESIGER to socket. */ void key_freereg(struct socket *so) { struct secreg *reg; int i; IPSEC_ASSERT(so != NULL, ("NULL so")); /* * check whether existing or not. * check all type of SA, because there is a potential that * one socket is registered to multiple type of SA. */ REGTREE_LOCK(); for (i = 0; i <= SADB_SATYPE_MAX; i++) { LIST_FOREACH(reg, &V_regtree[i], chain) { if (reg->so == so && __LIST_CHAINED(reg)) { LIST_REMOVE(reg, chain); free(reg, M_IPSEC_SAR); break; } } } REGTREE_UNLOCK(); } /* * SADB_EXPIRE processing * send * * to KMD by PF_KEY. * NOTE: We send only soft lifetime extension. * * OUT: 0 : succeed * others : error number */ static int key_expire(struct secasvar *sav, int hard) { struct mbuf *result = NULL, *m; struct sadb_lifetime *lt; uint32_t replay_count; int error, len; uint8_t satype; IPSEC_ASSERT (sav != NULL, ("null sav")); IPSEC_ASSERT (sav->sah != NULL, ("null sa header")); KEYDBG(KEY_STAMP, printf("%s: SA(%p) expired %s lifetime\n", __func__, sav, hard ? "hard": "soft")); KEYDBG(KEY_DATA, kdebug_secasv(sav)); /* set msg header */ satype = key_proto2satype(sav->sah->saidx.proto); IPSEC_ASSERT(satype != 0, ("invalid proto, satype %u", satype)); m = key_setsadbmsg(SADB_EXPIRE, 0, satype, sav->seq, 0, sav->refcnt); if (!m) { error = ENOBUFS; goto fail; } result = m; /* create SA extension */ m = key_setsadbsa(sav); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* create SA extension */ SECASVAR_LOCK(sav); replay_count = sav->replay ? sav->replay->count : 0; SECASVAR_UNLOCK(sav); m = key_setsadbxsa2(sav->sah->saidx.mode, replay_count, sav->sah->saidx.reqid); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); if (sav->replay && sav->replay->wsize > UINT8_MAX) { m = key_setsadbxsareplay(sav->replay->wsize); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); } /* create lifetime extension (current and soft) */ len = PFKEY_ALIGN8(sizeof(*lt)) * 2; m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) { error = ENOBUFS; goto fail; } m_align(m, len); m->m_len = len; bzero(mtod(m, caddr_t), len); lt = mtod(m, struct sadb_lifetime *); lt->sadb_lifetime_len = PFKEY_UNIT64(sizeof(struct sadb_lifetime)); lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT; lt->sadb_lifetime_allocations = (uint32_t)counter_u64_fetch(sav->lft_c_allocations); lt->sadb_lifetime_bytes = counter_u64_fetch(sav->lft_c_bytes); lt->sadb_lifetime_addtime = sav->created; lt->sadb_lifetime_usetime = sav->firstused; lt = (struct sadb_lifetime *)(mtod(m, caddr_t) + len / 2); lt->sadb_lifetime_len = PFKEY_UNIT64(sizeof(struct sadb_lifetime)); if (hard) { lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD; lt->sadb_lifetime_allocations = sav->lft_h->allocations; lt->sadb_lifetime_bytes = sav->lft_h->bytes; lt->sadb_lifetime_addtime = sav->lft_h->addtime; lt->sadb_lifetime_usetime = sav->lft_h->usetime; } else { lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_SOFT; lt->sadb_lifetime_allocations = sav->lft_s->allocations; lt->sadb_lifetime_bytes = sav->lft_s->bytes; lt->sadb_lifetime_addtime = sav->lft_s->addtime; lt->sadb_lifetime_usetime = sav->lft_s->usetime; } m_cat(result, m); /* set sadb_address for source */ m = key_setsadbaddr(SADB_EXT_ADDRESS_SRC, &sav->sah->saidx.src.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* set sadb_address for destination */ m = key_setsadbaddr(SADB_EXT_ADDRESS_DST, &sav->sah->saidx.dst.sa, FULLMASK, IPSEC_ULPROTO_ANY); if (!m) { error = ENOBUFS; goto fail; } m_cat(result, m); /* * XXX-BZ Handle NAT-T extensions here. * XXXAE: it doesn't seem quite useful. IKEs should not depend on * this information, we report only significant SA fields. */ if ((result->m_flags & M_PKTHDR) == 0) { error = EINVAL; goto fail; } if (result->m_len < sizeof(struct sadb_msg)) { result = m_pullup(result, sizeof(struct sadb_msg)); if (result == NULL) { error = ENOBUFS; goto fail; } } result->m_pkthdr.len = 0; for (m = result; m; m = m->m_next) result->m_pkthdr.len += m->m_len; mtod(result, struct sadb_msg *)->sadb_msg_len = PFKEY_UNIT64(result->m_pkthdr.len); return key_sendup_mbuf(NULL, result, KEY_SENDUP_REGISTERED); fail: if (result) m_freem(result); return error; } static void key_freesah_flushed(struct secashead_queue *flushq) { struct secashead *sah, *nextsah; struct secasvar *sav, *nextsav; sah = TAILQ_FIRST(flushq); while (sah != NULL) { sav = TAILQ_FIRST(&sah->savtree_larval); while (sav != NULL) { nextsav = TAILQ_NEXT(sav, chain); TAILQ_REMOVE(&sah->savtree_larval, sav, chain); key_freesav(&sav); /* release last reference */ key_freesah(&sah); /* release reference from SAV */ sav = nextsav; } sav = TAILQ_FIRST(&sah->savtree_alive); while (sav != NULL) { nextsav = TAILQ_NEXT(sav, chain); TAILQ_REMOVE(&sah->savtree_alive, sav, chain); key_freesav(&sav); /* release last reference */ key_freesah(&sah); /* release reference from SAV */ sav = nextsav; } nextsah = TAILQ_NEXT(sah, chain); key_freesah(&sah); /* release last reference */ sah = nextsah; } } /* * SADB_FLUSH processing * receive * * from the ikmpd, and free all entries in secastree. * and send, * * to the ikmpd. * NOTE: to do is only marking SADB_SASTATE_DEAD. * * m will always be freed. */ static int key_flush(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { struct secashead_queue flushq; struct sadb_msg *newmsg; struct secashead *sah, *nextsah; struct secasvar *sav; uint8_t proto; int i; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } KEYDBG(KEY_STAMP, printf("%s: proto %u\n", __func__, proto)); TAILQ_INIT(&flushq); if (proto == IPSEC_PROTO_ANY) { /* no SATYPE specified, i.e. flushing all SA. */ SAHTREE_WLOCK(); /* Move all SAHs into flushq */ TAILQ_CONCAT(&flushq, &V_sahtree, chain); /* Flush all buckets in SPI hash */ for (i = 0; i < V_savhash_mask + 1; i++) LIST_INIT(&V_savhashtbl[i]); /* Flush all buckets in SAHADDRHASH */ for (i = 0; i < V_sahaddrhash_mask + 1; i++) LIST_INIT(&V_sahaddrhashtbl[i]); /* Mark all SAHs as unlinked */ TAILQ_FOREACH(sah, &flushq, chain) { sah->state = SADB_SASTATE_DEAD; /* * Callout handler makes its job using * RLOCK and drain queues. In case, when this * function will be called just before it * acquires WLOCK, we need to mark SAs as * unlinked to prevent second unlink. */ TAILQ_FOREACH(sav, &sah->savtree_larval, chain) { sav->state = SADB_SASTATE_DEAD; } TAILQ_FOREACH(sav, &sah->savtree_alive, chain) { sav->state = SADB_SASTATE_DEAD; } } SAHTREE_WUNLOCK(); } else { SAHTREE_WLOCK(); sah = TAILQ_FIRST(&V_sahtree); while (sah != NULL) { IPSEC_ASSERT(sah->state != SADB_SASTATE_DEAD, ("DEAD SAH %p in SADB_FLUSH", sah)); nextsah = TAILQ_NEXT(sah, chain); if (sah->saidx.proto != proto) { sah = nextsah; continue; } sah->state = SADB_SASTATE_DEAD; TAILQ_REMOVE(&V_sahtree, sah, chain); LIST_REMOVE(sah, addrhash); /* Unlink all SAs from SPI hash */ TAILQ_FOREACH(sav, &sah->savtree_larval, chain) { LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; } TAILQ_FOREACH(sav, &sah->savtree_alive, chain) { LIST_REMOVE(sav, spihash); sav->state = SADB_SASTATE_DEAD; } /* Add SAH into flushq */ TAILQ_INSERT_HEAD(&flushq, sah, chain); sah = nextsah; } SAHTREE_WUNLOCK(); } key_freesah_flushed(&flushq); /* Free all queued SAs and SAHs */ if (m->m_len < sizeof(struct sadb_msg) || sizeof(struct sadb_msg) > m->m_len + M_TRAILINGSPACE(m)) { ipseclog((LOG_DEBUG, "%s: No more memory.\n", __func__)); return key_senderror(so, m, ENOBUFS); } if (m->m_next) m_freem(m->m_next); m->m_next = NULL; m->m_pkthdr.len = m->m_len = sizeof(struct sadb_msg); newmsg = mtod(m, struct sadb_msg *); newmsg->sadb_msg_errno = 0; newmsg->sadb_msg_len = PFKEY_UNIT64(m->m_pkthdr.len); return key_sendup_mbuf(so, m, KEY_SENDUP_ALL); } /* * SADB_DUMP processing * dump all entries including status of DEAD in SAD. * receive * * from the ikmpd, and dump all secasvar leaves * and send, * ..... * to the ikmpd. * * m will always be freed. */ static int key_dump(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { SAHTREE_RLOCK_TRACKER; struct secashead *sah; struct secasvar *sav; struct mbuf *n; uint32_t cnt; uint8_t proto, satype; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); /* map satype to proto */ if ((proto = key_satype2proto(mhp->msg->sadb_msg_satype)) == 0) { ipseclog((LOG_DEBUG, "%s: invalid satype is passed.\n", __func__)); return key_senderror(so, m, EINVAL); } /* count sav entries to be sent to the userland. */ cnt = 0; SAHTREE_RLOCK(); TAILQ_FOREACH(sah, &V_sahtree, chain) { if (mhp->msg->sadb_msg_satype != SADB_SATYPE_UNSPEC && proto != sah->saidx.proto) continue; TAILQ_FOREACH(sav, &sah->savtree_larval, chain) cnt++; TAILQ_FOREACH(sav, &sah->savtree_alive, chain) cnt++; } if (cnt == 0) { SAHTREE_RUNLOCK(); return key_senderror(so, m, ENOENT); } /* send this to the userland, one at a time. */ TAILQ_FOREACH(sah, &V_sahtree, chain) { if (mhp->msg->sadb_msg_satype != SADB_SATYPE_UNSPEC && proto != sah->saidx.proto) continue; /* map proto to satype */ if ((satype = key_proto2satype(sah->saidx.proto)) == 0) { SAHTREE_RUNLOCK(); ipseclog((LOG_DEBUG, "%s: there was invalid proto in " "SAD.\n", __func__)); return key_senderror(so, m, EINVAL); } TAILQ_FOREACH(sav, &sah->savtree_larval, chain) { n = key_setdumpsa(sav, SADB_DUMP, satype, --cnt, mhp->msg->sadb_msg_pid); if (n == NULL) { SAHTREE_RUNLOCK(); return key_senderror(so, m, ENOBUFS); } key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } TAILQ_FOREACH(sav, &sah->savtree_alive, chain) { n = key_setdumpsa(sav, SADB_DUMP, satype, --cnt, mhp->msg->sadb_msg_pid); if (n == NULL) { SAHTREE_RUNLOCK(); return key_senderror(so, m, ENOBUFS); } key_sendup_mbuf(so, n, KEY_SENDUP_ONE); } } SAHTREE_RUNLOCK(); m_freem(m); return (0); } /* * SADB_X_PROMISC processing * * m will always be freed. */ static int key_promisc(struct socket *so, struct mbuf *m, const struct sadb_msghdr *mhp) { int olen; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(mhp->msg != NULL, ("null msg")); olen = PFKEY_UNUNIT64(mhp->msg->sadb_msg_len); if (olen < sizeof(struct sadb_msg)) { #if 1 return key_senderror(so, m, EINVAL); #else m_freem(m); return 0; #endif } else if (olen == sizeof(struct sadb_msg)) { /* enable/disable promisc mode */ struct keycb *kp; if ((kp = (struct keycb *)sotorawcb(so)) == NULL) return key_senderror(so, m, EINVAL); mhp->msg->sadb_msg_errno = 0; switch (mhp->msg->sadb_msg_satype) { case 0: case 1: kp->kp_promisc = mhp->msg->sadb_msg_satype; break; default: return key_senderror(so, m, EINVAL); } /* send the original message back to everyone */ mhp->msg->sadb_msg_errno = 0; return key_sendup_mbuf(so, m, KEY_SENDUP_ALL); } else { /* send packet as is */ m_adj(m, PFKEY_ALIGN8(sizeof(struct sadb_msg))); /* TODO: if sadb_msg_seq is specified, send to specific pid */ return key_sendup_mbuf(so, m, KEY_SENDUP_ALL); } } static int (*key_typesw[])(struct socket *, struct mbuf *, const struct sadb_msghdr *) = { NULL, /* SADB_RESERVED */ key_getspi, /* SADB_GETSPI */ key_update, /* SADB_UPDATE */ key_add, /* SADB_ADD */ key_delete, /* SADB_DELETE */ key_get, /* SADB_GET */ key_acquire2, /* SADB_ACQUIRE */ key_register, /* SADB_REGISTER */ NULL, /* SADB_EXPIRE */ key_flush, /* SADB_FLUSH */ key_dump, /* SADB_DUMP */ key_promisc, /* SADB_X_PROMISC */ NULL, /* SADB_X_PCHANGE */ key_spdadd, /* SADB_X_SPDUPDATE */ key_spdadd, /* SADB_X_SPDADD */ key_spddelete, /* SADB_X_SPDDELETE */ key_spdget, /* SADB_X_SPDGET */ NULL, /* SADB_X_SPDACQUIRE */ key_spddump, /* SADB_X_SPDDUMP */ key_spdflush, /* SADB_X_SPDFLUSH */ key_spdadd, /* SADB_X_SPDSETIDX */ NULL, /* SADB_X_SPDEXPIRE */ key_spddelete2, /* SADB_X_SPDDELETE2 */ }; /* * parse sadb_msg buffer to process PFKEYv2, * and create a data to response if needed. * I think to be dealed with mbuf directly. * IN: * msgp : pointer to pointer to a received buffer pulluped. * This is rewrited to response. * so : pointer to socket. * OUT: * length for buffer to send to user process. */ int key_parse(struct mbuf *m, struct socket *so) { struct sadb_msg *msg; struct sadb_msghdr mh; u_int orglen; int error; int target; IPSEC_ASSERT(so != NULL, ("null socket")); IPSEC_ASSERT(m != NULL, ("null mbuf")); if (m->m_len < sizeof(struct sadb_msg)) { m = m_pullup(m, sizeof(struct sadb_msg)); if (!m) return ENOBUFS; } msg = mtod(m, struct sadb_msg *); orglen = PFKEY_UNUNIT64(msg->sadb_msg_len); target = KEY_SENDUP_ONE; if ((m->m_flags & M_PKTHDR) == 0 || m->m_pkthdr.len != orglen) { ipseclog((LOG_DEBUG, "%s: invalid message length.\n",__func__)); PFKEYSTAT_INC(out_invlen); error = EINVAL; goto senderror; } if (msg->sadb_msg_version != PF_KEY_V2) { ipseclog((LOG_DEBUG, "%s: PF_KEY version %u is mismatched.\n", __func__, msg->sadb_msg_version)); PFKEYSTAT_INC(out_invver); error = EINVAL; goto senderror; } if (msg->sadb_msg_type > SADB_MAX) { ipseclog((LOG_DEBUG, "%s: invalid type %u is passed.\n", __func__, msg->sadb_msg_type)); PFKEYSTAT_INC(out_invmsgtype); error = EINVAL; goto senderror; } /* for old-fashioned code - should be nuked */ if (m->m_pkthdr.len > MCLBYTES) { m_freem(m); return ENOBUFS; } if (m->m_next) { struct mbuf *n; MGETHDR(n, M_NOWAIT, MT_DATA); if (n && m->m_pkthdr.len > MHLEN) { if (!(MCLGET(n, M_NOWAIT))) { m_free(n); n = NULL; } } if (!n) { m_freem(m); return ENOBUFS; } m_copydata(m, 0, m->m_pkthdr.len, mtod(n, caddr_t)); n->m_pkthdr.len = n->m_len = m->m_pkthdr.len; n->m_next = NULL; m_freem(m); m = n; } /* align the mbuf chain so that extensions are in contiguous region. */ error = key_align(m, &mh); if (error) return error; msg = mh.msg; /* We use satype as scope mask for spddump */ if (msg->sadb_msg_type == SADB_X_SPDDUMP) { switch (msg->sadb_msg_satype) { case IPSEC_POLICYSCOPE_ANY: case IPSEC_POLICYSCOPE_GLOBAL: case IPSEC_POLICYSCOPE_IFNET: case IPSEC_POLICYSCOPE_PCB: break; default: ipseclog((LOG_DEBUG, "%s: illegal satype=%u\n", __func__, msg->sadb_msg_type)); PFKEYSTAT_INC(out_invsatype); error = EINVAL; goto senderror; } } else { switch (msg->sadb_msg_satype) { /* check SA type */ case SADB_SATYPE_UNSPEC: switch (msg->sadb_msg_type) { case SADB_GETSPI: case SADB_UPDATE: case SADB_ADD: case SADB_DELETE: case SADB_GET: case SADB_ACQUIRE: case SADB_EXPIRE: ipseclog((LOG_DEBUG, "%s: must specify satype " "when msg type=%u.\n", __func__, msg->sadb_msg_type)); PFKEYSTAT_INC(out_invsatype); error = EINVAL; goto senderror; } break; case SADB_SATYPE_AH: case SADB_SATYPE_ESP: case SADB_X_SATYPE_IPCOMP: case SADB_X_SATYPE_TCPSIGNATURE: switch (msg->sadb_msg_type) { case SADB_X_SPDADD: case SADB_X_SPDDELETE: case SADB_X_SPDGET: case SADB_X_SPDFLUSH: case SADB_X_SPDSETIDX: case SADB_X_SPDUPDATE: case SADB_X_SPDDELETE2: ipseclog((LOG_DEBUG, "%s: illegal satype=%u\n", __func__, msg->sadb_msg_type)); PFKEYSTAT_INC(out_invsatype); error = EINVAL; goto senderror; } break; case SADB_SATYPE_RSVP: case SADB_SATYPE_OSPFV2: case SADB_SATYPE_RIPV2: case SADB_SATYPE_MIP: ipseclog((LOG_DEBUG, "%s: type %u isn't supported.\n", __func__, msg->sadb_msg_satype)); PFKEYSTAT_INC(out_invsatype); error = EOPNOTSUPP; goto senderror; case 1: /* XXX: What does it do? */ if (msg->sadb_msg_type == SADB_X_PROMISC) break; /*FALLTHROUGH*/ default: ipseclog((LOG_DEBUG, "%s: invalid type %u is passed.\n", __func__, msg->sadb_msg_satype)); PFKEYSTAT_INC(out_invsatype); error = EINVAL; goto senderror; } } /* check field of upper layer protocol and address family */ if (mh.ext[SADB_EXT_ADDRESS_SRC] != NULL && mh.ext[SADB_EXT_ADDRESS_DST] != NULL) { struct sadb_address *src0, *dst0; u_int plen; src0 = (struct sadb_address *)(mh.ext[SADB_EXT_ADDRESS_SRC]); dst0 = (struct sadb_address *)(mh.ext[SADB_EXT_ADDRESS_DST]); /* check upper layer protocol */ if (src0->sadb_address_proto != dst0->sadb_address_proto) { ipseclog((LOG_DEBUG, "%s: upper layer protocol " "mismatched.\n", __func__)); PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } /* check family */ if (PFKEY_ADDR_SADDR(src0)->sa_family != PFKEY_ADDR_SADDR(dst0)->sa_family) { ipseclog((LOG_DEBUG, "%s: address family mismatched.\n", __func__)); PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } if (PFKEY_ADDR_SADDR(src0)->sa_len != PFKEY_ADDR_SADDR(dst0)->sa_len) { ipseclog((LOG_DEBUG, "%s: address struct size " "mismatched.\n", __func__)); PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } switch (PFKEY_ADDR_SADDR(src0)->sa_family) { case AF_INET: if (PFKEY_ADDR_SADDR(src0)->sa_len != sizeof(struct sockaddr_in)) { PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } break; case AF_INET6: if (PFKEY_ADDR_SADDR(src0)->sa_len != sizeof(struct sockaddr_in6)) { PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } break; default: ipseclog((LOG_DEBUG, "%s: unsupported address family\n", __func__)); PFKEYSTAT_INC(out_invaddr); error = EAFNOSUPPORT; goto senderror; } switch (PFKEY_ADDR_SADDR(src0)->sa_family) { case AF_INET: plen = sizeof(struct in_addr) << 3; break; case AF_INET6: plen = sizeof(struct in6_addr) << 3; break; default: plen = 0; /*fool gcc*/ break; } /* check max prefix length */ if (src0->sadb_address_prefixlen > plen || dst0->sadb_address_prefixlen > plen) { ipseclog((LOG_DEBUG, "%s: illegal prefixlen.\n", __func__)); PFKEYSTAT_INC(out_invaddr); error = EINVAL; goto senderror; } /* * prefixlen == 0 is valid because there can be a case when * all addresses are matched. */ } if (msg->sadb_msg_type >= nitems(key_typesw) || key_typesw[msg->sadb_msg_type] == NULL) { PFKEYSTAT_INC(out_invmsgtype); error = EINVAL; goto senderror; } return (*key_typesw[msg->sadb_msg_type])(so, m, &mh); senderror: msg->sadb_msg_errno = error; return key_sendup_mbuf(so, m, target); } static int key_senderror(struct socket *so, struct mbuf *m, int code) { struct sadb_msg *msg; IPSEC_ASSERT(m->m_len >= sizeof(struct sadb_msg), ("mbuf too small, len %u", m->m_len)); msg = mtod(m, struct sadb_msg *); msg->sadb_msg_errno = code; return key_sendup_mbuf(so, m, KEY_SENDUP_ONE); } /* * set the pointer to each header into message buffer. * m will be freed on error. * XXX larger-than-MCLBYTES extension? */ static int key_align(struct mbuf *m, struct sadb_msghdr *mhp) { struct mbuf *n; struct sadb_ext *ext; size_t off, end; int extlen; int toff; IPSEC_ASSERT(m != NULL, ("null mbuf")); IPSEC_ASSERT(mhp != NULL, ("null msghdr")); IPSEC_ASSERT(m->m_len >= sizeof(struct sadb_msg), ("mbuf too small, len %u", m->m_len)); /* initialize */ bzero(mhp, sizeof(*mhp)); mhp->msg = mtod(m, struct sadb_msg *); mhp->ext[0] = (struct sadb_ext *)mhp->msg; /*XXX backward compat */ end = PFKEY_UNUNIT64(mhp->msg->sadb_msg_len); extlen = end; /*just in case extlen is not updated*/ for (off = sizeof(struct sadb_msg); off < end; off += extlen) { n = m_pulldown(m, off, sizeof(struct sadb_ext), &toff); if (!n) { /* m is already freed */ return ENOBUFS; } ext = (struct sadb_ext *)(mtod(n, caddr_t) + toff); /* set pointer */ switch (ext->sadb_ext_type) { case SADB_EXT_SA: case SADB_EXT_ADDRESS_SRC: case SADB_EXT_ADDRESS_DST: case SADB_EXT_ADDRESS_PROXY: case SADB_EXT_LIFETIME_CURRENT: case SADB_EXT_LIFETIME_HARD: case SADB_EXT_LIFETIME_SOFT: case SADB_EXT_KEY_AUTH: case SADB_EXT_KEY_ENCRYPT: case SADB_EXT_IDENTITY_SRC: case SADB_EXT_IDENTITY_DST: case SADB_EXT_SENSITIVITY: case SADB_EXT_PROPOSAL: case SADB_EXT_SUPPORTED_AUTH: case SADB_EXT_SUPPORTED_ENCRYPT: case SADB_EXT_SPIRANGE: case SADB_X_EXT_POLICY: case SADB_X_EXT_SA2: case SADB_X_EXT_NAT_T_TYPE: case SADB_X_EXT_NAT_T_SPORT: case SADB_X_EXT_NAT_T_DPORT: case SADB_X_EXT_NAT_T_OAI: case SADB_X_EXT_NAT_T_OAR: case SADB_X_EXT_NAT_T_FRAG: case SADB_X_EXT_SA_REPLAY: case SADB_X_EXT_NEW_ADDRESS_SRC: case SADB_X_EXT_NEW_ADDRESS_DST: /* duplicate check */ /* * XXX Are there duplication payloads of either * KEY_AUTH or KEY_ENCRYPT ? */ if (mhp->ext[ext->sadb_ext_type] != NULL) { ipseclog((LOG_DEBUG, "%s: duplicate ext_type " "%u\n", __func__, ext->sadb_ext_type)); m_freem(m); PFKEYSTAT_INC(out_dupext); return EINVAL; } break; default: ipseclog((LOG_DEBUG, "%s: invalid ext_type %u\n", __func__, ext->sadb_ext_type)); m_freem(m); PFKEYSTAT_INC(out_invexttype); return EINVAL; } extlen = PFKEY_UNUNIT64(ext->sadb_ext_len); if (key_validate_ext(ext, extlen)) { m_freem(m); PFKEYSTAT_INC(out_invlen); return EINVAL; } n = m_pulldown(m, off, extlen, &toff); if (!n) { /* m is already freed */ return ENOBUFS; } ext = (struct sadb_ext *)(mtod(n, caddr_t) + toff); mhp->ext[ext->sadb_ext_type] = ext; mhp->extoff[ext->sadb_ext_type] = off; mhp->extlen[ext->sadb_ext_type] = extlen; } if (off != end) { m_freem(m); PFKEYSTAT_INC(out_invlen); return EINVAL; } return 0; } static int key_validate_ext(const struct sadb_ext *ext, int len) { const struct sockaddr *sa; enum { NONE, ADDR } checktype = NONE; int baselen = 0; const int sal = offsetof(struct sockaddr, sa_len) + sizeof(sa->sa_len); if (len != PFKEY_UNUNIT64(ext->sadb_ext_len)) return EINVAL; /* if it does not match minimum/maximum length, bail */ if (ext->sadb_ext_type >= nitems(minsize) || ext->sadb_ext_type >= nitems(maxsize)) return EINVAL; if (!minsize[ext->sadb_ext_type] || len < minsize[ext->sadb_ext_type]) return EINVAL; if (maxsize[ext->sadb_ext_type] && len > maxsize[ext->sadb_ext_type]) return EINVAL; /* more checks based on sadb_ext_type XXX need more */ switch (ext->sadb_ext_type) { case SADB_EXT_ADDRESS_SRC: case SADB_EXT_ADDRESS_DST: case SADB_EXT_ADDRESS_PROXY: case SADB_X_EXT_NAT_T_OAI: case SADB_X_EXT_NAT_T_OAR: case SADB_X_EXT_NEW_ADDRESS_SRC: case SADB_X_EXT_NEW_ADDRESS_DST: baselen = PFKEY_ALIGN8(sizeof(struct sadb_address)); checktype = ADDR; break; case SADB_EXT_IDENTITY_SRC: case SADB_EXT_IDENTITY_DST: if (((const struct sadb_ident *)ext)->sadb_ident_type == SADB_X_IDENTTYPE_ADDR) { baselen = PFKEY_ALIGN8(sizeof(struct sadb_ident)); checktype = ADDR; } else checktype = NONE; break; default: checktype = NONE; break; } switch (checktype) { case NONE: break; case ADDR: sa = (const struct sockaddr *)(((const u_int8_t*)ext)+baselen); if (len < baselen + sal) return EINVAL; if (baselen + PFKEY_ALIGN8(sa->sa_len) != len) return EINVAL; break; } return 0; } void spdcache_init(void) { int i; TUNABLE_INT_FETCH("net.key.spdcache.maxentries", &V_key_spdcache_maxentries); TUNABLE_INT_FETCH("net.key.spdcache.threshold", &V_key_spdcache_threshold); if (V_key_spdcache_maxentries) { V_key_spdcache_maxentries = MAX(V_key_spdcache_maxentries, SPDCACHE_MAX_ENTRIES_PER_HASH); V_spdcachehashtbl = hashinit(V_key_spdcache_maxentries / SPDCACHE_MAX_ENTRIES_PER_HASH, M_IPSEC_SPDCACHE, &V_spdcachehash_mask); V_key_spdcache_maxentries = (V_spdcachehash_mask + 1) * SPDCACHE_MAX_ENTRIES_PER_HASH; V_spdcache_lock = malloc(sizeof(struct mtx) * (V_spdcachehash_mask + 1), M_IPSEC_SPDCACHE, M_WAITOK|M_ZERO); for (i = 0; i < V_spdcachehash_mask + 1; ++i) SPDCACHE_LOCK_INIT(i); } } struct spdcache_entry * spdcache_entry_alloc(const struct secpolicyindex *spidx, struct secpolicy *sp) { struct spdcache_entry *entry; entry = malloc(sizeof(struct spdcache_entry), M_IPSEC_SPDCACHE, M_NOWAIT|M_ZERO); if (entry == NULL) return NULL; if (sp != NULL) SP_ADDREF(sp); entry->spidx = *spidx; entry->sp = sp; return (entry); } void spdcache_entry_free(struct spdcache_entry *entry) { if (entry->sp != NULL) key_freesp(&entry->sp); free(entry, M_IPSEC_SPDCACHE); } void spdcache_clear(void) { struct spdcache_entry *entry; int i; for (i = 0; i < V_spdcachehash_mask + 1; ++i) { SPDCACHE_LOCK(i); while (!LIST_EMPTY(&V_spdcachehashtbl[i])) { entry = LIST_FIRST(&V_spdcachehashtbl[i]); LIST_REMOVE(entry, chain); spdcache_entry_free(entry); } SPDCACHE_UNLOCK(i); } } #ifdef VIMAGE void spdcache_destroy(void) { int i; if (SPDCACHE_ENABLED()) { spdcache_clear(); hashdestroy(V_spdcachehashtbl, M_IPSEC_SPDCACHE, V_spdcachehash_mask); for (i = 0; i < V_spdcachehash_mask + 1; ++i) SPDCACHE_LOCK_DESTROY(i); free(V_spdcache_lock, M_IPSEC_SPDCACHE); } } #endif void key_init(void) { int i; for (i = 0; i < IPSEC_DIR_MAX; i++) { TAILQ_INIT(&V_sptree[i]); TAILQ_INIT(&V_sptree_ifnet[i]); } V_key_lft_zone = uma_zcreate("IPsec SA lft_c", sizeof(uint64_t) * 2, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); TAILQ_INIT(&V_sahtree); V_sphashtbl = hashinit(SPHASH_NHASH, M_IPSEC_SP, &V_sphash_mask); V_savhashtbl = hashinit(SAVHASH_NHASH, M_IPSEC_SA, &V_savhash_mask); V_sahaddrhashtbl = hashinit(SAHHASH_NHASH, M_IPSEC_SAH, &V_sahaddrhash_mask); V_acqaddrhashtbl = hashinit(ACQHASH_NHASH, M_IPSEC_SAQ, &V_acqaddrhash_mask); V_acqseqhashtbl = hashinit(ACQHASH_NHASH, M_IPSEC_SAQ, &V_acqseqhash_mask); spdcache_init(); for (i = 0; i <= SADB_SATYPE_MAX; i++) LIST_INIT(&V_regtree[i]); LIST_INIT(&V_acqtree); LIST_INIT(&V_spacqtree); if (!IS_DEFAULT_VNET(curvnet)) return; XFORMS_LOCK_INIT(); SPTREE_LOCK_INIT(); REGTREE_LOCK_INIT(); SAHTREE_LOCK_INIT(); ACQ_LOCK_INIT(); SPACQ_LOCK_INIT(); #ifndef IPSEC_DEBUG2 callout_init(&key_timer, 1); callout_reset(&key_timer, hz, key_timehandler, NULL); #endif /*IPSEC_DEBUG2*/ /* initialize key statistics */ keystat.getspi_count = 1; if (bootverbose) printf("IPsec: Initialized Security Association Processing.\n"); } #ifdef VIMAGE void key_destroy(void) { struct secashead_queue sahdrainq; struct secpolicy_queue drainq; struct secpolicy *sp, *nextsp; struct secacq *acq, *nextacq; struct secspacq *spacq, *nextspacq; struct secashead *sah; struct secasvar *sav; struct secreg *reg; int i; /* * XXX: can we just call free() for each object without * walking through safe way with releasing references? */ TAILQ_INIT(&drainq); SPTREE_WLOCK(); for (i = 0; i < IPSEC_DIR_MAX; i++) { TAILQ_CONCAT(&drainq, &V_sptree[i], chain); TAILQ_CONCAT(&drainq, &V_sptree_ifnet[i], chain); } for (i = 0; i < V_sphash_mask + 1; i++) LIST_INIT(&V_sphashtbl[i]); SPTREE_WUNLOCK(); spdcache_destroy(); sp = TAILQ_FIRST(&drainq); while (sp != NULL) { nextsp = TAILQ_NEXT(sp, chain); key_freesp(&sp); sp = nextsp; } TAILQ_INIT(&sahdrainq); SAHTREE_WLOCK(); TAILQ_CONCAT(&sahdrainq, &V_sahtree, chain); for (i = 0; i < V_savhash_mask + 1; i++) LIST_INIT(&V_savhashtbl[i]); for (i = 0; i < V_sahaddrhash_mask + 1; i++) LIST_INIT(&V_sahaddrhashtbl[i]); TAILQ_FOREACH(sah, &sahdrainq, chain) { sah->state = SADB_SASTATE_DEAD; TAILQ_FOREACH(sav, &sah->savtree_larval, chain) { sav->state = SADB_SASTATE_DEAD; } TAILQ_FOREACH(sav, &sah->savtree_alive, chain) { sav->state = SADB_SASTATE_DEAD; } } SAHTREE_WUNLOCK(); key_freesah_flushed(&sahdrainq); hashdestroy(V_sphashtbl, M_IPSEC_SP, V_sphash_mask); hashdestroy(V_savhashtbl, M_IPSEC_SA, V_savhash_mask); hashdestroy(V_sahaddrhashtbl, M_IPSEC_SAH, V_sahaddrhash_mask); REGTREE_LOCK(); for (i = 0; i <= SADB_SATYPE_MAX; i++) { LIST_FOREACH(reg, &V_regtree[i], chain) { if (__LIST_CHAINED(reg)) { LIST_REMOVE(reg, chain); free(reg, M_IPSEC_SAR); break; } } } REGTREE_UNLOCK(); ACQ_LOCK(); acq = LIST_FIRST(&V_acqtree); while (acq != NULL) { nextacq = LIST_NEXT(acq, chain); LIST_REMOVE(acq, chain); free(acq, M_IPSEC_SAQ); acq = nextacq; } for (i = 0; i < V_acqaddrhash_mask + 1; i++) LIST_INIT(&V_acqaddrhashtbl[i]); for (i = 0; i < V_acqseqhash_mask + 1; i++) LIST_INIT(&V_acqseqhashtbl[i]); ACQ_UNLOCK(); SPACQ_LOCK(); for (spacq = LIST_FIRST(&V_spacqtree); spacq != NULL; spacq = nextspacq) { nextspacq = LIST_NEXT(spacq, chain); if (__LIST_CHAINED(spacq)) { LIST_REMOVE(spacq, chain); free(spacq, M_IPSEC_SAQ); } } SPACQ_UNLOCK(); hashdestroy(V_acqaddrhashtbl, M_IPSEC_SAQ, V_acqaddrhash_mask); hashdestroy(V_acqseqhashtbl, M_IPSEC_SAQ, V_acqseqhash_mask); uma_zdestroy(V_key_lft_zone); if (!IS_DEFAULT_VNET(curvnet)) return; #ifndef IPSEC_DEBUG2 callout_drain(&key_timer); #endif XFORMS_LOCK_DESTROY(); SPTREE_LOCK_DESTROY(); REGTREE_LOCK_DESTROY(); SAHTREE_LOCK_DESTROY(); ACQ_LOCK_DESTROY(); SPACQ_LOCK_DESTROY(); } #endif /* record data transfer on SA, and update timestamps */ void key_sa_recordxfer(struct secasvar *sav, struct mbuf *m) { IPSEC_ASSERT(sav != NULL, ("Null secasvar")); IPSEC_ASSERT(m != NULL, ("Null mbuf")); /* * XXX Currently, there is a difference of bytes size * between inbound and outbound processing. */ counter_u64_add(sav->lft_c_bytes, m->m_pkthdr.len); /* * We use the number of packets as the unit of * allocations. We increment the variable * whenever {esp,ah}_{in,out}put is called. */ counter_u64_add(sav->lft_c_allocations, 1); /* * NOTE: We record CURRENT usetime by using wall clock, * in seconds. HARD and SOFT lifetime are measured by the time * difference (again in seconds) from usetime. * * usetime * v expire expire * -----+-----+--------+---> t * <--------------> HARD * <-----> SOFT */ if (sav->firstused == 0) sav->firstused = time_second; } /* * Take one of the kernel's security keys and convert it into a PF_KEY * structure within an mbuf, suitable for sending up to a waiting * application in user land. * * IN: * src: A pointer to a kernel security key. * exttype: Which type of key this is. Refer to the PF_KEY data structures. * OUT: * a valid mbuf or NULL indicating an error * */ static struct mbuf * key_setkey(struct seckey *src, uint16_t exttype) { struct mbuf *m; struct sadb_key *p; int len; if (src == NULL) return NULL; len = PFKEY_ALIGN8(sizeof(struct sadb_key) + _KEYLEN(src)); m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return NULL; m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_key *); bzero(p, len); p->sadb_key_len = PFKEY_UNIT64(len); p->sadb_key_exttype = exttype; p->sadb_key_bits = src->bits; bcopy(src->key_data, _KEYBUF(p), _KEYLEN(src)); return m; } /* * Take one of the kernel's lifetime data structures and convert it * into a PF_KEY structure within an mbuf, suitable for sending up to * a waiting application in user land. * * IN: * src: A pointer to a kernel lifetime structure. * exttype: Which type of lifetime this is. Refer to the PF_KEY * data structures for more information. * OUT: * a valid mbuf or NULL indicating an error * */ static struct mbuf * key_setlifetime(struct seclifetime *src, uint16_t exttype) { struct mbuf *m = NULL; struct sadb_lifetime *p; int len = PFKEY_ALIGN8(sizeof(struct sadb_lifetime)); if (src == NULL) return NULL; m = m_get2(len, M_NOWAIT, MT_DATA, 0); if (m == NULL) return m; m_align(m, len); m->m_len = len; p = mtod(m, struct sadb_lifetime *); bzero(p, len); p->sadb_lifetime_len = PFKEY_UNIT64(len); p->sadb_lifetime_exttype = exttype; p->sadb_lifetime_allocations = src->allocations; p->sadb_lifetime_bytes = src->bytes; p->sadb_lifetime_addtime = src->addtime; p->sadb_lifetime_usetime = src->usetime; return m; } const struct enc_xform * enc_algorithm_lookup(int alg) { int i; for (i = 0; i < nitems(supported_ealgs); i++) if (alg == supported_ealgs[i].sadb_alg) return (supported_ealgs[i].xform); return (NULL); } const struct auth_hash * auth_algorithm_lookup(int alg) { int i; for (i = 0; i < nitems(supported_aalgs); i++) if (alg == supported_aalgs[i].sadb_alg) return (supported_aalgs[i].xform); return (NULL); } const struct comp_algo * comp_algorithm_lookup(int alg) { int i; for (i = 0; i < nitems(supported_calgs); i++) if (alg == supported_calgs[i].sadb_alg) return (supported_calgs[i].xform); return (NULL); } /* * Register a transform. */ static int xform_register(struct xformsw* xsp) { struct xformsw *entry; XFORMS_LOCK(); LIST_FOREACH(entry, &xforms, chain) { if (entry->xf_type == xsp->xf_type) { XFORMS_UNLOCK(); return (EEXIST); } } LIST_INSERT_HEAD(&xforms, xsp, chain); XFORMS_UNLOCK(); return (0); } void xform_attach(void *data) { struct xformsw *xsp = (struct xformsw *)data; if (xform_register(xsp) != 0) printf("%s: failed to register %s xform\n", __func__, xsp->xf_name); } void xform_detach(void *data) { struct xformsw *xsp = (struct xformsw *)data; XFORMS_LOCK(); LIST_REMOVE(xsp, chain); XFORMS_UNLOCK(); /* Delete all SAs related to this xform. */ key_delete_xform(xsp); } /* * Initialize transform support in an sav. */ static int xform_init(struct secasvar *sav, u_short xftype) { struct xformsw *entry; int ret; IPSEC_ASSERT(sav->tdb_xform == NULL, ("tdb_xform is already initialized")); ret = EINVAL; XFORMS_LOCK(); LIST_FOREACH(entry, &xforms, chain) { if (entry->xf_type == xftype) { ret = (*entry->xf_init)(sav, entry); break; } } XFORMS_UNLOCK(); return (ret); } Index: head/sys/netipsec/keysock.c =================================================================== --- head/sys/netipsec/keysock.c (revision 336675) +++ head/sys/netipsec/keysock.c (revision 336676) @@ -1,474 +1,474 @@ /* $FreeBSD$ */ /* $KAME: keysock.c,v 1.25 2001/08/13 20:07:41 itojun Exp $ */ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include "opt_ipsec.h" /* This code has derived from sys/net/rtsock.c on FreeBSD2.2.5 */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct key_cb { int key_count; int any_count; }; -static VNET_DEFINE(struct key_cb, key_cb); +VNET_DEFINE_STATIC(struct key_cb, key_cb); #define V_key_cb VNET(key_cb) static struct sockaddr key_src = { 2, PF_KEY, }; static int key_sendup0(struct rawcb *, struct mbuf *, int); VNET_PCPUSTAT_DEFINE(struct pfkeystat, pfkeystat); VNET_PCPUSTAT_SYSINIT(pfkeystat); #ifdef VIMAGE VNET_PCPUSTAT_SYSUNINIT(pfkeystat); #endif /* VIMAGE */ /* * key_output() */ int key_output(struct mbuf *m, struct socket *so, ...) { struct sadb_msg *msg; int len, error = 0; if (m == NULL) panic("%s: NULL pointer was passed.\n", __func__); PFKEYSTAT_INC(out_total); PFKEYSTAT_ADD(out_bytes, m->m_pkthdr.len); len = m->m_pkthdr.len; if (len < sizeof(struct sadb_msg)) { PFKEYSTAT_INC(out_tooshort); error = EINVAL; goto end; } if (m->m_len < sizeof(struct sadb_msg)) { if ((m = m_pullup(m, sizeof(struct sadb_msg))) == NULL) { PFKEYSTAT_INC(out_nomem); error = ENOBUFS; goto end; } } M_ASSERTPKTHDR(m); KEYDBG(KEY_DUMP, kdebug_mbuf(m)); msg = mtod(m, struct sadb_msg *); PFKEYSTAT_INC(out_msgtype[msg->sadb_msg_type]); if (len != PFKEY_UNUNIT64(msg->sadb_msg_len)) { PFKEYSTAT_INC(out_invlen); error = EINVAL; goto end; } error = key_parse(m, so); m = NULL; end: if (m) m_freem(m); return error; } /* * send message to the socket. */ static int key_sendup0(struct rawcb *rp, struct mbuf *m, int promisc) { int error; if (promisc) { struct sadb_msg *pmsg; M_PREPEND(m, sizeof(struct sadb_msg), M_NOWAIT); if (m == NULL) { PFKEYSTAT_INC(in_nomem); return (ENOBUFS); } pmsg = mtod(m, struct sadb_msg *); bzero(pmsg, sizeof(*pmsg)); pmsg->sadb_msg_version = PF_KEY_V2; pmsg->sadb_msg_type = SADB_X_PROMISC; pmsg->sadb_msg_len = PFKEY_UNIT64(m->m_pkthdr.len); /* pid and seq? */ PFKEYSTAT_INC(in_msgtype[pmsg->sadb_msg_type]); } if (!sbappendaddr(&rp->rcb_socket->so_rcv, (struct sockaddr *)&key_src, m, NULL)) { PFKEYSTAT_INC(in_nomem); m_freem(m); error = ENOBUFS; } else error = 0; sorwakeup(rp->rcb_socket); return error; } /* so can be NULL if target != KEY_SENDUP_ONE */ int key_sendup_mbuf(struct socket *so, struct mbuf *m, int target) { struct mbuf *n; struct keycb *kp; struct rawcb *rp; int error = 0; KASSERT(m != NULL, ("NULL mbuf pointer was passed.")); KASSERT(so != NULL || target != KEY_SENDUP_ONE, ("NULL socket pointer was passed.")); KASSERT(target == KEY_SENDUP_ONE || target == KEY_SENDUP_ALL || target == KEY_SENDUP_REGISTERED, ("Wrong target %d", target)); PFKEYSTAT_INC(in_total); PFKEYSTAT_ADD(in_bytes, m->m_pkthdr.len); if (m->m_len < sizeof(struct sadb_msg)) { m = m_pullup(m, sizeof(struct sadb_msg)); if (m == NULL) { PFKEYSTAT_INC(in_nomem); return ENOBUFS; } } if (m->m_len >= sizeof(struct sadb_msg)) { struct sadb_msg *msg; msg = mtod(m, struct sadb_msg *); PFKEYSTAT_INC(in_msgtype[msg->sadb_msg_type]); } mtx_lock(&rawcb_mtx); if (V_key_cb.any_count == 0) { mtx_unlock(&rawcb_mtx); m_freem(m); return (0); } LIST_FOREACH(rp, &V_rawcb_list, list) { if (rp->rcb_proto.sp_family != PF_KEY) continue; if (rp->rcb_proto.sp_protocol && rp->rcb_proto.sp_protocol != PF_KEY_V2) { continue; } /* * If you are in promiscuous mode, and when you get broadcasted * reply, you'll get two PF_KEY messages. * (based on pf_key@inner.net message on 14 Oct 1998) */ kp = (struct keycb *)rp; if (kp->kp_promisc) { n = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (n != NULL) key_sendup0(rp, n, 1); else PFKEYSTAT_INC(in_nomem); } /* the exact target will be processed later */ if (so && sotorawcb(so) == rp) continue; if (target == KEY_SENDUP_ONE || ( target == KEY_SENDUP_REGISTERED && kp->kp_registered == 0)) continue; /* KEY_SENDUP_ALL + KEY_SENDUP_REGISTERED */ n = m_copym(m, 0, M_COPYALL, M_NOWAIT); if (n == NULL) { PFKEYSTAT_INC(in_nomem); /* Try send to another socket */ continue; } if (key_sendup0(rp, n, 0) == 0) PFKEYSTAT_INC(in_msgtarget[target]); } if (so) { /* KEY_SENDUP_ONE */ error = key_sendup0(sotorawcb(so), m, 0); if (error == 0) PFKEYSTAT_INC(in_msgtarget[KEY_SENDUP_ONE]); } else { error = 0; m_freem(m); } mtx_unlock(&rawcb_mtx); return (error); } /* * key_abort() * derived from net/rtsock.c:rts_abort() */ static void key_abort(struct socket *so) { raw_usrreqs.pru_abort(so); } /* * key_attach() * derived from net/rtsock.c:rts_attach() */ static int key_attach(struct socket *so, int proto, struct thread *td) { struct keycb *kp; int error; KASSERT(so->so_pcb == NULL, ("key_attach: so_pcb != NULL")); if (td != NULL) { error = priv_check(td, PRIV_NET_RAW); if (error) return error; } /* XXX */ kp = malloc(sizeof *kp, M_PCB, M_WAITOK | M_ZERO); if (kp == NULL) return ENOBUFS; so->so_pcb = (caddr_t)kp; error = raw_attach(so, proto); kp = (struct keycb *)sotorawcb(so); if (error) { free(kp, M_PCB); so->so_pcb = (caddr_t) 0; return error; } kp->kp_promisc = kp->kp_registered = 0; if (kp->kp_raw.rcb_proto.sp_protocol == PF_KEY) /* XXX: AF_KEY */ V_key_cb.key_count++; V_key_cb.any_count++; soisconnected(so); so->so_options |= SO_USELOOPBACK; return 0; } /* * key_bind() * derived from net/rtsock.c:rts_bind() */ static int key_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return EINVAL; } /* * key_close() * derived from net/rtsock.c:rts_close(). */ static void key_close(struct socket *so) { raw_usrreqs.pru_close(so); } /* * key_connect() * derived from net/rtsock.c:rts_connect() */ static int key_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return EINVAL; } /* * key_detach() * derived from net/rtsock.c:rts_detach() */ static void key_detach(struct socket *so) { struct keycb *kp = (struct keycb *)sotorawcb(so); KASSERT(kp != NULL, ("key_detach: kp == NULL")); if (kp->kp_raw.rcb_proto.sp_protocol == PF_KEY) /* XXX: AF_KEY */ V_key_cb.key_count--; V_key_cb.any_count--; key_freereg(so); raw_usrreqs.pru_detach(so); } /* * key_disconnect() * derived from net/rtsock.c:key_disconnect() */ static int key_disconnect(struct socket *so) { return(raw_usrreqs.pru_disconnect(so)); } /* * key_peeraddr() * derived from net/rtsock.c:rts_peeraddr() */ static int key_peeraddr(struct socket *so, struct sockaddr **nam) { return(raw_usrreqs.pru_peeraddr(so, nam)); } /* * key_send() * derived from net/rtsock.c:rts_send() */ static int key_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { return(raw_usrreqs.pru_send(so, flags, m, nam, control, td)); } /* * key_shutdown() * derived from net/rtsock.c:rts_shutdown() */ static int key_shutdown(struct socket *so) { return(raw_usrreqs.pru_shutdown(so)); } /* * key_sockaddr() * derived from net/rtsock.c:rts_sockaddr() */ static int key_sockaddr(struct socket *so, struct sockaddr **nam) { return(raw_usrreqs.pru_sockaddr(so, nam)); } struct pr_usrreqs key_usrreqs = { .pru_abort = key_abort, .pru_attach = key_attach, .pru_bind = key_bind, .pru_connect = key_connect, .pru_detach = key_detach, .pru_disconnect = key_disconnect, .pru_peeraddr = key_peeraddr, .pru_send = key_send, .pru_shutdown = key_shutdown, .pru_sockaddr = key_sockaddr, .pru_close = key_close, }; /* sysctl */ SYSCTL_NODE(_net, PF_KEY, key, CTLFLAG_RW, 0, "Key Family"); /* * Definitions of protocols supported in the KEY domain. */ extern struct domain keydomain; struct protosw keysw[] = { { .pr_type = SOCK_RAW, .pr_domain = &keydomain, .pr_protocol = PF_KEY_V2, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_output = key_output, .pr_ctlinput = raw_ctlinput, .pr_init = raw_init, .pr_usrreqs = &key_usrreqs } }; static void key_init0(void) { bzero((caddr_t)&V_key_cb, sizeof(V_key_cb)); key_init(); } struct domain keydomain = { .dom_family = PF_KEY, .dom_name = "key", .dom_init = key_init0, #ifdef VIMAGE .dom_destroy = key_destroy, #endif .dom_protosw = keysw, .dom_protoswNPROTOSW = &keysw[nitems(keysw)] }; VNET_DOMAIN_SET(key); Index: head/sys/netpfil/ipfw/ip_fw2.c =================================================================== --- head/sys/netpfil/ipfw/ip_fw2.c (revision 336675) +++ head/sys/netpfil/ipfw/ip_fw2.c (revision 336676) @@ -1,3456 +1,3456 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2009 Luigi Rizzo, Universita` di Pisa * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * The FreeBSD IP packet firewall, main file */ #include "opt_ipfw.h" #include "opt_ipdivert.h" #include "opt_inet.h" #ifndef INET #error "IPFIREWALL requires INET" #endif /* INET */ #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for ETHERTYPE_IP */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #endif #include /* for struct grehdr */ #include #include /* XXX for in_cksum */ #ifdef MAC #include #endif /* * static variables followed by global ones. * All ipfw global variables are here. */ -static VNET_DEFINE(int, fw_deny_unknown_exthdrs); +VNET_DEFINE_STATIC(int, fw_deny_unknown_exthdrs); #define V_fw_deny_unknown_exthdrs VNET(fw_deny_unknown_exthdrs) -static VNET_DEFINE(int, fw_permit_single_frag6) = 1; +VNET_DEFINE_STATIC(int, fw_permit_single_frag6) = 1; #define V_fw_permit_single_frag6 VNET(fw_permit_single_frag6) #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT static int default_to_accept = 1; #else static int default_to_accept; #endif VNET_DEFINE(int, autoinc_step); VNET_DEFINE(int, fw_one_pass) = 1; VNET_DEFINE(unsigned int, fw_tables_max); VNET_DEFINE(unsigned int, fw_tables_sets) = 0; /* Don't use set-aware tables */ /* Use 128 tables by default */ static unsigned int default_fw_tables = IPFW_TABLES_DEFAULT; #ifndef LINEAR_SKIPTO static int jump_fast(struct ip_fw_chain *chain, struct ip_fw *f, int num, int tablearg, int jump_backwards); #define JUMP(ch, f, num, targ, back) jump_fast(ch, f, num, targ, back) #else static int jump_linear(struct ip_fw_chain *chain, struct ip_fw *f, int num, int tablearg, int jump_backwards); #define JUMP(ch, f, num, targ, back) jump_linear(ch, f, num, targ, back) #endif /* * Each rule belongs to one of 32 different sets (0..31). * The variable set_disable contains one bit per set. * If the bit is set, all rules in the corresponding set * are disabled. Set RESVD_SET(31) is reserved for the default rule * and rules that are not deleted by the flush command, * and CANNOT be disabled. * Rules in set RESVD_SET can only be deleted individually. */ VNET_DEFINE(u_int32_t, set_disable); #define V_set_disable VNET(set_disable) VNET_DEFINE(int, fw_verbose); /* counter for ipfw_log(NULL...) */ VNET_DEFINE(u_int64_t, norule_counter); VNET_DEFINE(int, verbose_limit); /* layer3_chain contains the list of rules for layer 3 */ VNET_DEFINE(struct ip_fw_chain, layer3_chain); /* ipfw_vnet_ready controls when we are open for business */ VNET_DEFINE(int, ipfw_vnet_ready) = 0; VNET_DEFINE(int, ipfw_nat_ready) = 0; ipfw_nat_t *ipfw_nat_ptr = NULL; struct cfg_nat *(*lookup_nat_ptr)(struct nat_list *, int); ipfw_nat_cfg_t *ipfw_nat_cfg_ptr; ipfw_nat_cfg_t *ipfw_nat_del_ptr; ipfw_nat_cfg_t *ipfw_nat_get_cfg_ptr; ipfw_nat_cfg_t *ipfw_nat_get_log_ptr; #ifdef SYSCTL_NODE uint32_t dummy_def = IPFW_DEFAULT_RULE; static int sysctl_ipfw_table_num(SYSCTL_HANDLER_ARGS); static int sysctl_ipfw_tables_sets(SYSCTL_HANDLER_ARGS); SYSBEGIN(f3) SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_one_pass), 0, "Only do a single pass through ipfw when using dummynet(4)"); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(autoinc_step), 0, "Rule number auto-increment step"); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_verbose), 0, "Log matches to ipfw rules"); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(verbose_limit), 0, "Set upper limit of matches of ipfw rules logged"); SYSCTL_UINT(_net_inet_ip_fw, OID_AUTO, default_rule, CTLFLAG_RD, &dummy_def, 0, "The default/max possible rule number."); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, tables_max, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 0, 0, sysctl_ipfw_table_num, "IU", "Maximum number of concurrently used tables"); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, tables_sets, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 0, 0, sysctl_ipfw_tables_sets, "IU", "Use per-set namespace for tables"); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, default_to_accept, CTLFLAG_RDTUN, &default_to_accept, 0, "Make the default rule accept all packets."); TUNABLE_INT("net.inet.ip.fw.tables_max", (int *)&default_fw_tables); SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(layer3_chain.n_rules), 0, "Number of static rules"); #ifdef INET6 SYSCTL_DECL(_net_inet6_ip6); SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall"); SYSCTL_INT(_net_inet6_ip6_fw, OID_AUTO, deny_unknown_exthdrs, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(fw_deny_unknown_exthdrs), 0, "Deny packets with unknown IPv6 Extension Headers"); SYSCTL_INT(_net_inet6_ip6_fw, OID_AUTO, permit_single_frag6, CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE, &VNET_NAME(fw_permit_single_frag6), 0, "Permit single packet IPv6 fragments"); #endif /* INET6 */ SYSEND #endif /* SYSCTL_NODE */ /* * Some macros used in the various matching options. * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T * Other macros just cast void * into the appropriate type */ #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl)) #define TCP(p) ((struct tcphdr *)(p)) #define SCTP(p) ((struct sctphdr *)(p)) #define UDP(p) ((struct udphdr *)(p)) #define ICMP(p) ((struct icmphdr *)(p)) #define ICMP6(p) ((struct icmp6_hdr *)(p)) static __inline int icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd) { int type = icmp->icmp_type; return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<icmp_type; return (type <= ICMP_MAXTYPE && (TT & (1<arg1 or cmd->d[0]. * * We scan options and store the bits we find set. We succeed if * * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear * * The code is sometimes optimized not to store additional variables. */ static int flags_match(ipfw_insn *cmd, u_int8_t bits) { u_char want_clear; bits = ~bits; if ( ((cmd->arg1 & 0xff) & bits) != 0) return 0; /* some bits we want set were clear */ want_clear = (cmd->arg1 >> 8) & 0xff; if ( (want_clear & bits) != want_clear) return 0; /* some bits we want clear were set */ return 1; } static int ipopts_match(struct ip *ip, ipfw_insn *cmd) { int optlen, bits = 0; u_char *cp = (u_char *)(ip + 1); int x = (ip->ip_hl << 2) - sizeof (struct ip); for (; x > 0; x -= optlen, cp += optlen) { int opt = cp[IPOPT_OPTVAL]; if (opt == IPOPT_EOL) break; if (opt == IPOPT_NOP) optlen = 1; else { optlen = cp[IPOPT_OLEN]; if (optlen <= 0 || optlen > x) return 0; /* invalid or truncated */ } switch (opt) { default: break; case IPOPT_LSRR: bits |= IP_FW_IPOPT_LSRR; break; case IPOPT_SSRR: bits |= IP_FW_IPOPT_SSRR; break; case IPOPT_RR: bits |= IP_FW_IPOPT_RR; break; case IPOPT_TS: bits |= IP_FW_IPOPT_TS; break; } } return (flags_match(cmd, bits)); } static int tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd) { int optlen, bits = 0; u_char *cp = (u_char *)(tcp + 1); int x = (tcp->th_off << 2) - sizeof(struct tcphdr); for (; x > 0; x -= optlen, cp += optlen) { int opt = cp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { optlen = cp[1]; if (optlen <= 0) break; } switch (opt) { default: break; case TCPOPT_MAXSEG: bits |= IP_FW_TCPOPT_MSS; break; case TCPOPT_WINDOW: bits |= IP_FW_TCPOPT_WINDOW; break; case TCPOPT_SACK_PERMITTED: case TCPOPT_SACK: bits |= IP_FW_TCPOPT_SACK; break; case TCPOPT_TIMESTAMP: bits |= IP_FW_TCPOPT_TS; break; } } return (flags_match(cmd, bits)); } static int iface_match(struct ifnet *ifp, ipfw_insn_if *cmd, struct ip_fw_chain *chain, uint32_t *tablearg) { if (ifp == NULL) /* no iface with this packet, match fails */ return (0); /* Check by name or by IP address */ if (cmd->name[0] != '\0') { /* match by name */ if (cmd->name[0] == '\1') /* use tablearg to match */ return ipfw_lookup_table(chain, cmd->p.kidx, 0, &ifp->if_index, tablearg); /* Check name */ if (cmd->p.glob) { if (fnmatch(cmd->name, ifp->if_xname, 0) == 0) return(1); } else { if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0) return(1); } } else { #if !defined(USERSPACE) && defined(__FreeBSD__) /* and OSX too ? */ struct ifaddr *ia; if_addr_rlock(ifp); CK_STAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { if (ia->ifa_addr->sa_family != AF_INET) continue; if (cmd->p.ip.s_addr == ((struct sockaddr_in *) (ia->ifa_addr))->sin_addr.s_addr) { if_addr_runlock(ifp); return(1); /* match */ } } if_addr_runlock(ifp); #endif /* __FreeBSD__ */ } return(0); /* no match, fail ... */ } /* * The verify_path function checks if a route to the src exists and * if it is reachable via ifp (when provided). * * The 'verrevpath' option checks that the interface that an IP packet * arrives on is the same interface that traffic destined for the * packet's source address would be routed out of. * The 'versrcreach' option just checks that the source address is * reachable via any route (except default) in the routing table. * These two are a measure to block forged packets. This is also * commonly known as "anti-spoofing" or Unicast Reverse Path * Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs * is purposely reminiscent of the Cisco IOS command, * * ip verify unicast reverse-path * ip verify unicast source reachable-via any * * which implements the same functionality. But note that the syntax * is misleading, and the check may be performed on all IP packets * whether unicast, multicast, or broadcast. */ static int verify_path(struct in_addr src, struct ifnet *ifp, u_int fib) { #if defined(USERSPACE) || !defined(__FreeBSD__) return 0; #else struct nhop4_basic nh4; if (fib4_lookup_nh_basic(fib, src, NHR_IFAIF, 0, &nh4) != 0) return (0); /* * If ifp is provided, check for equality with rtentry. * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp, * in order to pass packets injected back by if_simloop(): * routing entry (via lo0) for our own address * may exist, so we need to handle routing assymetry. */ if (ifp != NULL && ifp != nh4.nh_ifp) return (0); /* if no ifp provided, check if rtentry is not default route */ if (ifp == NULL && (nh4.nh_flags & NHF_DEFAULT) != 0) return (0); /* or if this is a blackhole/reject route */ if (ifp == NULL && (nh4.nh_flags & (NHF_REJECT|NHF_BLACKHOLE)) != 0) return (0); /* found valid route */ return 1; #endif /* __FreeBSD__ */ } /* * Generate an SCTP packet containing an ABORT chunk. The verification tag * is given by vtag. The T-bit is set in the ABORT chunk if and only if * reflected is not 0. */ static struct mbuf * ipfw_send_abort(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t vtag, int reflected) { struct mbuf *m; struct ip *ip; #ifdef INET6 struct ip6_hdr *ip6; #endif struct sctphdr *sctp; struct sctp_chunkhdr *chunk; u_int16_t hlen, plen, tlen; MGETHDR(m, M_NOWAIT, MT_DATA); if (m == NULL) return (NULL); M_SETFIB(m, id->fib); #ifdef MAC if (replyto != NULL) mac_netinet_firewall_reply(replyto, m); else mac_netinet_firewall_send(m); #else (void)replyto; /* don't warn about unused arg */ #endif switch (id->addr_type) { case 4: hlen = sizeof(struct ip); break; #ifdef INET6 case 6: hlen = sizeof(struct ip6_hdr); break; #endif default: /* XXX: log me?!? */ FREE_PKT(m); return (NULL); } plen = sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr); tlen = hlen + plen; m->m_data += max_linkhdr; m->m_flags |= M_SKIP_FIREWALL; m->m_pkthdr.len = m->m_len = tlen; m->m_pkthdr.rcvif = NULL; bzero(m->m_data, tlen); switch (id->addr_type) { case 4: ip = mtod(m, struct ip *); ip->ip_v = 4; ip->ip_hl = sizeof(struct ip) >> 2; ip->ip_tos = IPTOS_LOWDELAY; ip->ip_len = htons(tlen); ip->ip_id = htons(0); ip->ip_off = htons(0); ip->ip_ttl = V_ip_defttl; ip->ip_p = IPPROTO_SCTP; ip->ip_sum = 0; ip->ip_src.s_addr = htonl(id->dst_ip); ip->ip_dst.s_addr = htonl(id->src_ip); sctp = (struct sctphdr *)(ip + 1); break; #ifdef INET6 case 6: ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_plen = htons(plen); ip6->ip6_nxt = IPPROTO_SCTP; ip6->ip6_hlim = IPV6_DEFHLIM; ip6->ip6_src = id->dst_ip6; ip6->ip6_dst = id->src_ip6; sctp = (struct sctphdr *)(ip6 + 1); break; #endif } sctp->src_port = htons(id->dst_port); sctp->dest_port = htons(id->src_port); sctp->v_tag = htonl(vtag); sctp->checksum = htonl(0); chunk = (struct sctp_chunkhdr *)(sctp + 1); chunk->chunk_type = SCTP_ABORT_ASSOCIATION; chunk->chunk_flags = 0; if (reflected != 0) { chunk->chunk_flags |= SCTP_HAD_NO_TCB; } chunk->chunk_length = htons(sizeof(struct sctp_chunkhdr)); sctp->checksum = sctp_calculate_cksum(m, hlen); return (m); } /* * Generate a TCP packet, containing either a RST or a keepalive. * When flags & TH_RST, we are sending a RST packet, because of a * "reset" action matched the packet. * Otherwise we are sending a keepalive, and flags & TH_ * The 'replyto' mbuf is the mbuf being replied to, if any, and is required * so that MAC can label the reply appropriately. */ struct mbuf * ipfw_send_pkt(struct mbuf *replyto, struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags) { struct mbuf *m = NULL; /* stupid compiler */ struct ip *h = NULL; /* stupid compiler */ #ifdef INET6 struct ip6_hdr *h6 = NULL; #endif struct tcphdr *th = NULL; int len, dir; MGETHDR(m, M_NOWAIT, MT_DATA); if (m == NULL) return (NULL); M_SETFIB(m, id->fib); #ifdef MAC if (replyto != NULL) mac_netinet_firewall_reply(replyto, m); else mac_netinet_firewall_send(m); #else (void)replyto; /* don't warn about unused arg */ #endif switch (id->addr_type) { case 4: len = sizeof(struct ip) + sizeof(struct tcphdr); break; #ifdef INET6 case 6: len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); break; #endif default: /* XXX: log me?!? */ FREE_PKT(m); return (NULL); } dir = ((flags & (TH_SYN | TH_RST)) == TH_SYN); m->m_data += max_linkhdr; m->m_flags |= M_SKIP_FIREWALL; m->m_pkthdr.len = m->m_len = len; m->m_pkthdr.rcvif = NULL; bzero(m->m_data, len); switch (id->addr_type) { case 4: h = mtod(m, struct ip *); /* prepare for checksum */ h->ip_p = IPPROTO_TCP; h->ip_len = htons(sizeof(struct tcphdr)); if (dir) { h->ip_src.s_addr = htonl(id->src_ip); h->ip_dst.s_addr = htonl(id->dst_ip); } else { h->ip_src.s_addr = htonl(id->dst_ip); h->ip_dst.s_addr = htonl(id->src_ip); } th = (struct tcphdr *)(h + 1); break; #ifdef INET6 case 6: h6 = mtod(m, struct ip6_hdr *); /* prepare for checksum */ h6->ip6_nxt = IPPROTO_TCP; h6->ip6_plen = htons(sizeof(struct tcphdr)); if (dir) { h6->ip6_src = id->src_ip6; h6->ip6_dst = id->dst_ip6; } else { h6->ip6_src = id->dst_ip6; h6->ip6_dst = id->src_ip6; } th = (struct tcphdr *)(h6 + 1); break; #endif } if (dir) { th->th_sport = htons(id->src_port); th->th_dport = htons(id->dst_port); } else { th->th_sport = htons(id->dst_port); th->th_dport = htons(id->src_port); } th->th_off = sizeof(struct tcphdr) >> 2; if (flags & TH_RST) { if (flags & TH_ACK) { th->th_seq = htonl(ack); th->th_flags = TH_RST; } else { if (flags & TH_SYN) seq++; th->th_ack = htonl(seq); th->th_flags = TH_RST | TH_ACK; } } else { /* * Keepalive - use caller provided sequence numbers */ th->th_seq = htonl(seq); th->th_ack = htonl(ack); th->th_flags = TH_ACK; } switch (id->addr_type) { case 4: th->th_sum = in_cksum(m, len); /* finish the ip header */ h->ip_v = 4; h->ip_hl = sizeof(*h) >> 2; h->ip_tos = IPTOS_LOWDELAY; h->ip_off = htons(0); h->ip_len = htons(len); h->ip_ttl = V_ip_defttl; h->ip_sum = 0; break; #ifdef INET6 case 6: th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(*h6), sizeof(struct tcphdr)); /* finish the ip6 header */ h6->ip6_vfc |= IPV6_VERSION; h6->ip6_hlim = IPV6_DEFHLIM; break; #endif } return (m); } #ifdef INET6 /* * ipv6 specific rules here... */ static __inline int icmp6type_match (int type, ipfw_insn_u32 *cmd) { return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) ); } static int flow6id_match( int curr_flow, ipfw_insn_u32 *cmd ) { int i; for (i=0; i <= cmd->o.arg1; ++i ) if (curr_flow == cmd->d[i] ) return 1; return 0; } /* support for IP6_*_ME opcodes */ static const struct in6_addr lla_mask = {{{ 0xff, 0xff, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }}}; static int ipfw_localip6(struct in6_addr *in6) { struct rm_priotracker in6_ifa_tracker; struct in6_ifaddr *ia; if (IN6_IS_ADDR_MULTICAST(in6)) return (0); if (!IN6_IS_ADDR_LINKLOCAL(in6)) return (in6_localip(in6)); IN6_IFADDR_RLOCK(&in6_ifa_tracker); CK_STAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { if (!IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) continue; if (IN6_ARE_MASKED_ADDR_EQUAL(&ia->ia_addr.sin6_addr, in6, &lla_mask)) { IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); return (1); } } IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); return (0); } static int verify_path6(struct in6_addr *src, struct ifnet *ifp, u_int fib) { struct nhop6_basic nh6; if (IN6_IS_SCOPE_LINKLOCAL(src)) return (1); if (fib6_lookup_nh_basic(fib, src, 0, NHR_IFAIF, 0, &nh6) != 0) return (0); /* If ifp is provided, check for equality with route table. */ if (ifp != NULL && ifp != nh6.nh_ifp) return (0); /* if no ifp provided, check if rtentry is not default route */ if (ifp == NULL && (nh6.nh_flags & NHF_DEFAULT) != 0) return (0); /* or if this is a blackhole/reject route */ if (ifp == NULL && (nh6.nh_flags & (NHF_REJECT|NHF_BLACKHOLE)) != 0) return (0); /* found valid route */ return 1; } static int is_icmp6_query(int icmp6_type) { if ((icmp6_type <= ICMP6_MAXTYPE) && (icmp6_type == ICMP6_ECHO_REQUEST || icmp6_type == ICMP6_MEMBERSHIP_QUERY || icmp6_type == ICMP6_WRUREQUEST || icmp6_type == ICMP6_FQDN_QUERY || icmp6_type == ICMP6_NI_QUERY)) return (1); return (0); } static int map_icmp_unreach(int code) { /* RFC 7915 p4.2 */ switch (code) { case ICMP_UNREACH_NET: case ICMP_UNREACH_HOST: case ICMP_UNREACH_SRCFAIL: case ICMP_UNREACH_NET_UNKNOWN: case ICMP_UNREACH_HOST_UNKNOWN: case ICMP_UNREACH_TOSNET: case ICMP_UNREACH_TOSHOST: return (ICMP6_DST_UNREACH_NOROUTE); case ICMP_UNREACH_PORT: return (ICMP6_DST_UNREACH_NOPORT); default: /* * Map the rest of codes into admit prohibited. * XXX: unreach proto should be mapped into ICMPv6 * parameter problem, but we use only unreach type. */ return (ICMP6_DST_UNREACH_ADMIN); } } static void send_reject6(struct ip_fw_args *args, int code, u_int hlen, struct ip6_hdr *ip6) { struct mbuf *m; m = args->m; if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) { struct tcphdr *tcp; tcp = (struct tcphdr *)((char *)ip6 + hlen); if ((tcp->th_flags & TH_RST) == 0) { struct mbuf *m0; m0 = ipfw_send_pkt(args->m, &(args->f_id), ntohl(tcp->th_seq), ntohl(tcp->th_ack), tcp->th_flags | TH_RST); if (m0 != NULL) ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); } FREE_PKT(m); } else if (code == ICMP6_UNREACH_ABORT && args->f_id.proto == IPPROTO_SCTP) { struct mbuf *m0; struct sctphdr *sctp; u_int32_t v_tag; int reflected; sctp = (struct sctphdr *)((char *)ip6 + hlen); reflected = 1; v_tag = ntohl(sctp->v_tag); /* Investigate the first chunk header if available */ if (m->m_len >= hlen + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr)) { struct sctp_chunkhdr *chunk; chunk = (struct sctp_chunkhdr *)(sctp + 1); switch (chunk->chunk_type) { case SCTP_INITIATION: /* * Packets containing an INIT chunk MUST have * a zero v-tag. */ if (v_tag != 0) { v_tag = 0; break; } /* INIT chunk MUST NOT be bundled */ if (m->m_pkthdr.len > hlen + sizeof(struct sctphdr) + ntohs(chunk->chunk_length) + 3) { break; } /* Use the initiate tag if available */ if ((m->m_len >= hlen + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd))) { struct sctp_init *init; init = (struct sctp_init *)(chunk + 1); v_tag = ntohl(init->initiate_tag); reflected = 0; } break; case SCTP_ABORT_ASSOCIATION: /* * If the packet contains an ABORT chunk, don't * reply. * XXX: We should search through all chunks, * but don't do to avoid attacks. */ v_tag = 0; break; } } if (v_tag == 0) { m0 = NULL; } else { m0 = ipfw_send_abort(args->m, &(args->f_id), v_tag, reflected); } if (m0 != NULL) ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL); FREE_PKT(m); } else if (code != ICMP6_UNREACH_RST && code != ICMP6_UNREACH_ABORT) { /* Send an ICMPv6 unreach. */ #if 0 /* * Unlike above, the mbufs need to line up with the ip6 hdr, * as the contents are read. We need to m_adj() the * needed amount. * The mbuf will however be thrown away so we can adjust it. * Remember we did an m_pullup on it already so we * can make some assumptions about contiguousness. */ if (args->L3offset) m_adj(m, args->L3offset); #endif icmp6_error(m, ICMP6_DST_UNREACH, code, 0); } else FREE_PKT(m); args->m = NULL; } #endif /* INET6 */ /* * sends a reject message, consuming the mbuf passed as an argument. */ static void send_reject(struct ip_fw_args *args, int code, int iplen, struct ip *ip) { #if 0 /* XXX When ip is not guaranteed to be at mtod() we will * need to account for this */ * The mbuf will however be thrown away so we can adjust it. * Remember we did an m_pullup on it already so we * can make some assumptions about contiguousness. */ if (args->L3offset) m_adj(m, args->L3offset); #endif if (code != ICMP_REJECT_RST && code != ICMP_REJECT_ABORT) { /* Send an ICMP unreach */ icmp_error(args->m, ICMP_UNREACH, code, 0L, 0); } else if (code == ICMP_REJECT_RST && args->f_id.proto == IPPROTO_TCP) { struct tcphdr *const tcp = L3HDR(struct tcphdr, mtod(args->m, struct ip *)); if ( (tcp->th_flags & TH_RST) == 0) { struct mbuf *m; m = ipfw_send_pkt(args->m, &(args->f_id), ntohl(tcp->th_seq), ntohl(tcp->th_ack), tcp->th_flags | TH_RST); if (m != NULL) ip_output(m, NULL, NULL, 0, NULL, NULL); } FREE_PKT(args->m); } else if (code == ICMP_REJECT_ABORT && args->f_id.proto == IPPROTO_SCTP) { struct mbuf *m; struct sctphdr *sctp; struct sctp_chunkhdr *chunk; struct sctp_init *init; u_int32_t v_tag; int reflected; sctp = L3HDR(struct sctphdr, mtod(args->m, struct ip *)); reflected = 1; v_tag = ntohl(sctp->v_tag); if (iplen >= (ip->ip_hl << 2) + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr)) { /* Look at the first chunk header if available */ chunk = (struct sctp_chunkhdr *)(sctp + 1); switch (chunk->chunk_type) { case SCTP_INITIATION: /* * Packets containing an INIT chunk MUST have * a zero v-tag. */ if (v_tag != 0) { v_tag = 0; break; } /* INIT chunk MUST NOT be bundled */ if (iplen > (ip->ip_hl << 2) + sizeof(struct sctphdr) + ntohs(chunk->chunk_length) + 3) { break; } /* Use the initiate tag if available */ if ((iplen >= (ip->ip_hl << 2) + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd))) { init = (struct sctp_init *)(chunk + 1); v_tag = ntohl(init->initiate_tag); reflected = 0; } break; case SCTP_ABORT_ASSOCIATION: /* * If the packet contains an ABORT chunk, don't * reply. * XXX: We should search through all chunks, * but don't do to avoid attacks. */ v_tag = 0; break; } } if (v_tag == 0) { m = NULL; } else { m = ipfw_send_abort(args->m, &(args->f_id), v_tag, reflected); } if (m != NULL) ip_output(m, NULL, NULL, 0, NULL, NULL); FREE_PKT(args->m); } else FREE_PKT(args->m); args->m = NULL; } /* * Support for uid/gid/jail lookup. These tests are expensive * (because we may need to look into the list of active sockets) * so we cache the results. ugid_lookupp is 0 if we have not * yet done a lookup, 1 if we succeeded, and -1 if we tried * and failed. The function always returns the match value. * We could actually spare the variable and use *uc, setting * it to '(void *)check_uidgid if we have no info, NULL if * we tried and failed, or any other value if successful. */ static int check_uidgid(ipfw_insn_u32 *insn, struct ip_fw_args *args, int *ugid_lookupp, struct ucred **uc) { #if defined(USERSPACE) return 0; // not supported in userspace #else #ifndef __FreeBSD__ /* XXX */ return cred_check(insn, proto, oif, dst_ip, dst_port, src_ip, src_port, (struct bsd_ucred *)uc, ugid_lookupp, ((struct mbuf *)inp)->m_skb); #else /* FreeBSD */ struct in_addr src_ip, dst_ip; struct inpcbinfo *pi; struct ipfw_flow_id *id; struct inpcb *pcb, *inp; struct ifnet *oif; int lookupflags; int match; id = &args->f_id; inp = args->inp; oif = args->oif; /* * Check to see if the UDP or TCP stack supplied us with * the PCB. If so, rather then holding a lock and looking * up the PCB, we can use the one that was supplied. */ if (inp && *ugid_lookupp == 0) { INP_LOCK_ASSERT(inp); if (inp->inp_socket != NULL) { *uc = crhold(inp->inp_cred); *ugid_lookupp = 1; } else *ugid_lookupp = -1; } /* * If we have already been here and the packet has no * PCB entry associated with it, then we can safely * assume that this is a no match. */ if (*ugid_lookupp == -1) return (0); if (id->proto == IPPROTO_TCP) { lookupflags = 0; pi = &V_tcbinfo; } else if (id->proto == IPPROTO_UDP) { lookupflags = INPLOOKUP_WILDCARD; pi = &V_udbinfo; } else if (id->proto == IPPROTO_UDPLITE) { lookupflags = INPLOOKUP_WILDCARD; pi = &V_ulitecbinfo; } else return 0; lookupflags |= INPLOOKUP_RLOCKPCB; match = 0; if (*ugid_lookupp == 0) { if (id->addr_type == 6) { #ifdef INET6 if (oif == NULL) pcb = in6_pcblookup_mbuf(pi, &id->src_ip6, htons(id->src_port), &id->dst_ip6, htons(id->dst_port), lookupflags, oif, args->m); else pcb = in6_pcblookup_mbuf(pi, &id->dst_ip6, htons(id->dst_port), &id->src_ip6, htons(id->src_port), lookupflags, oif, args->m); #else *ugid_lookupp = -1; return (0); #endif } else { src_ip.s_addr = htonl(id->src_ip); dst_ip.s_addr = htonl(id->dst_ip); if (oif == NULL) pcb = in_pcblookup_mbuf(pi, src_ip, htons(id->src_port), dst_ip, htons(id->dst_port), lookupflags, oif, args->m); else pcb = in_pcblookup_mbuf(pi, dst_ip, htons(id->dst_port), src_ip, htons(id->src_port), lookupflags, oif, args->m); } if (pcb != NULL) { INP_RLOCK_ASSERT(pcb); *uc = crhold(pcb->inp_cred); *ugid_lookupp = 1; INP_RUNLOCK(pcb); } if (*ugid_lookupp == 0) { /* * We tried and failed, set the variable to -1 * so we will not try again on this packet. */ *ugid_lookupp = -1; return (0); } } if (insn->o.opcode == O_UID) match = ((*uc)->cr_uid == (uid_t)insn->d[0]); else if (insn->o.opcode == O_GID) match = groupmember((gid_t)insn->d[0], *uc); else if (insn->o.opcode == O_JAIL) match = ((*uc)->cr_prison->pr_id == (int)insn->d[0]); return (match); #endif /* __FreeBSD__ */ #endif /* not supported in userspace */ } /* * Helper function to set args with info on the rule after the matching * one. slot is precise, whereas we guess rule_id as they are * assigned sequentially. */ static inline void set_match(struct ip_fw_args *args, int slot, struct ip_fw_chain *chain) { args->rule.chain_id = chain->id; args->rule.slot = slot + 1; /* we use 0 as a marker */ args->rule.rule_id = 1 + chain->map[slot]->id; args->rule.rulenum = chain->map[slot]->rulenum; } #ifndef LINEAR_SKIPTO /* * Helper function to enable cached rule lookups using * cached_id and cached_pos fields in ipfw rule. */ static int jump_fast(struct ip_fw_chain *chain, struct ip_fw *f, int num, int tablearg, int jump_backwards) { int f_pos; /* If possible use cached f_pos (in f->cached_pos), * whose version is written in f->cached_id * (horrible hacks to avoid changing the ABI). */ if (num != IP_FW_TARG && f->cached_id == chain->id) f_pos = f->cached_pos; else { int i = IP_FW_ARG_TABLEARG(chain, num, skipto); /* make sure we do not jump backward */ if (jump_backwards == 0 && i <= f->rulenum) i = f->rulenum + 1; if (chain->idxmap != NULL) f_pos = chain->idxmap[i]; else f_pos = ipfw_find_rule(chain, i, 0); /* update the cache */ if (num != IP_FW_TARG) { f->cached_id = chain->id; f->cached_pos = f_pos; } } return (f_pos); } #else /* * Helper function to enable real fast rule lookups. */ static int jump_linear(struct ip_fw_chain *chain, struct ip_fw *f, int num, int tablearg, int jump_backwards) { int f_pos; num = IP_FW_ARG_TABLEARG(chain, num, skipto); /* make sure we do not jump backward */ if (jump_backwards == 0 && num <= f->rulenum) num = f->rulenum + 1; f_pos = chain->idxmap[num]; return (f_pos); } #endif #define TARG(k, f) IP_FW_ARG_TABLEARG(chain, k, f) /* * The main check routine for the firewall. * * All arguments are in args so we can modify them and return them * back to the caller. * * Parameters: * * args->m (in/out) The packet; we set to NULL when/if we nuke it. * Starts with the IP header. * args->eh (in) Mac header if present, NULL for layer3 packet. * args->L3offset Number of bytes bypassed if we came from L2. * e.g. often sizeof(eh) ** NOTYET ** * args->oif Outgoing interface, NULL if packet is incoming. * The incoming interface is in the mbuf. (in) * args->divert_rule (in/out) * Skip up to the first rule past this rule number; * upon return, non-zero port number for divert or tee. * * args->rule Pointer to the last matching rule (in/out) * args->next_hop Socket we are forwarding to (out). * args->next_hop6 IPv6 next hop we are forwarding to (out). * args->f_id Addresses grabbed from the packet (out) * args->rule.info a cookie depending on rule action * * Return value: * * IP_FW_PASS the packet must be accepted * IP_FW_DENY the packet must be dropped * IP_FW_DIVERT divert packet, port in m_tag * IP_FW_TEE tee packet, port in m_tag * IP_FW_DUMMYNET to dummynet, pipe in args->cookie * IP_FW_NETGRAPH into netgraph, cookie args->cookie * args->rule contains the matching rule, * args->rule.info has additional information. * */ int ipfw_chk(struct ip_fw_args *args) { /* * Local variables holding state while processing a packet: * * IMPORTANT NOTE: to speed up the processing of rules, there * are some assumption on the values of the variables, which * are documented here. Should you change them, please check * the implementation of the various instructions to make sure * that they still work. * * args->eh The MAC header. It is non-null for a layer2 * packet, it is NULL for a layer-3 packet. * **notyet** * args->L3offset Offset in the packet to the L3 (IP or equiv.) header. * * m | args->m Pointer to the mbuf, as received from the caller. * It may change if ipfw_chk() does an m_pullup, or if it * consumes the packet because it calls send_reject(). * XXX This has to change, so that ipfw_chk() never modifies * or consumes the buffer. * ip is the beginning of the ip(4 or 6) header. * Calculated by adding the L3offset to the start of data. * (Until we start using L3offset, the packet is * supposed to start with the ip header). */ struct mbuf *m = args->m; struct ip *ip = mtod(m, struct ip *); /* * For rules which contain uid/gid or jail constraints, cache * a copy of the users credentials after the pcb lookup has been * executed. This will speed up the processing of rules with * these types of constraints, as well as decrease contention * on pcb related locks. */ #ifndef __FreeBSD__ struct bsd_ucred ucred_cache; #else struct ucred *ucred_cache = NULL; #endif int ucred_lookup = 0; /* * oif | args->oif If NULL, ipfw_chk has been called on the * inbound path (ether_input, ip_input). * If non-NULL, ipfw_chk has been called on the outbound path * (ether_output, ip_output). */ struct ifnet *oif = args->oif; int f_pos = 0; /* index of current rule in the array */ int retval = 0; /* * hlen The length of the IP header. */ u_int hlen = 0; /* hlen >0 means we have an IP pkt */ /* * offset The offset of a fragment. offset != 0 means that * we have a fragment at this offset of an IPv4 packet. * offset == 0 means that (if this is an IPv4 packet) * this is the first or only fragment. * For IPv6 offset|ip6f_mf == 0 means there is no Fragment Header * or there is a single packet fragment (fragment header added * without needed). We will treat a single packet fragment as if * there was no fragment header (or log/block depending on the * V_fw_permit_single_frag6 sysctl setting). */ u_short offset = 0; u_short ip6f_mf = 0; /* * Local copies of addresses. They are only valid if we have * an IP packet. * * proto The protocol. Set to 0 for non-ip packets, * or to the protocol read from the packet otherwise. * proto != 0 means that we have an IPv4 packet. * * src_port, dst_port port numbers, in HOST format. Only * valid for TCP and UDP packets. * * src_ip, dst_ip ip addresses, in NETWORK format. * Only valid for IPv4 packets. */ uint8_t proto; uint16_t src_port = 0, dst_port = 0; /* NOTE: host format */ struct in_addr src_ip, dst_ip; /* NOTE: network format */ int iplen = 0; int pktlen; uint16_t etype = 0; /* Host order stored ether type */ /* * dyn_dir = MATCH_UNKNOWN when rules unchecked, * MATCH_NONE when checked and not matched (q = NULL), * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL) */ struct ipfw_dyn_info dyn_info; struct ip_fw *q = NULL; struct ip_fw_chain *chain = &V_layer3_chain; /* * We store in ulp a pointer to the upper layer protocol header. * In the ipv4 case this is easy to determine from the header, * but for ipv6 we might have some additional headers in the middle. * ulp is NULL if not found. */ void *ulp = NULL; /* upper layer protocol pointer. */ /* XXX ipv6 variables */ int is_ipv6 = 0; uint8_t icmp6_type = 0; uint16_t ext_hd = 0; /* bits vector for extension header filtering */ /* end of ipv6 variables */ int is_ipv4 = 0; int done = 0; /* flag to exit the outer loop */ if (m->m_flags & M_SKIP_FIREWALL || (! V_ipfw_vnet_ready)) return (IP_FW_PASS); /* accept */ dst_ip.s_addr = 0; /* make sure it is initialized */ src_ip.s_addr = 0; /* make sure it is initialized */ pktlen = m->m_pkthdr.len; args->f_id.fib = M_GETFIB(m); /* note mbuf not altered) */ proto = args->f_id.proto = 0; /* mark f_id invalid */ /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */ DYN_INFO_INIT(&dyn_info); /* * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous, * then it sets p to point at the offset "len" in the mbuf. WARNING: the * pointer might become stale after other pullups (but we never use it * this way). */ #define PULLUP_TO(_len, p, T) PULLUP_LEN(_len, p, sizeof(T)) #define PULLUP_LEN(_len, p, T) \ do { \ int x = (_len) + T; \ if ((m)->m_len < x) { \ args->m = m = m_pullup(m, x); \ if (m == NULL) \ goto pullup_failed; \ } \ p = (mtod(m, char *) + (_len)); \ } while (0) /* * if we have an ether header, */ if (args->eh) etype = ntohs(args->eh->ether_type); /* Identify IP packets and fill up variables. */ if (pktlen >= sizeof(struct ip6_hdr) && (args->eh == NULL || etype == ETHERTYPE_IPV6) && ip->ip_v == 6) { struct ip6_hdr *ip6 = (struct ip6_hdr *)ip; is_ipv6 = 1; args->f_id.addr_type = 6; hlen = sizeof(struct ip6_hdr); proto = ip6->ip6_nxt; /* Search extension headers to find upper layer protocols */ while (ulp == NULL && offset == 0) { switch (proto) { case IPPROTO_ICMPV6: PULLUP_TO(hlen, ulp, struct icmp6_hdr); icmp6_type = ICMP6(ulp)->icmp6_type; break; case IPPROTO_TCP: PULLUP_TO(hlen, ulp, struct tcphdr); dst_port = TCP(ulp)->th_dport; src_port = TCP(ulp)->th_sport; /* save flags for dynamic rules */ args->f_id._flags = TCP(ulp)->th_flags; break; case IPPROTO_SCTP: if (pktlen >= hlen + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd)) PULLUP_LEN(hlen, ulp, sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd)); else if (pktlen >= hlen + sizeof(struct sctphdr)) PULLUP_LEN(hlen, ulp, pktlen - hlen); else PULLUP_LEN(hlen, ulp, sizeof(struct sctphdr)); src_port = SCTP(ulp)->src_port; dst_port = SCTP(ulp)->dest_port; break; case IPPROTO_UDP: case IPPROTO_UDPLITE: PULLUP_TO(hlen, ulp, struct udphdr); dst_port = UDP(ulp)->uh_dport; src_port = UDP(ulp)->uh_sport; break; case IPPROTO_HOPOPTS: /* RFC 2460 */ PULLUP_TO(hlen, ulp, struct ip6_hbh); ext_hd |= EXT_HOPOPTS; hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3; proto = ((struct ip6_hbh *)ulp)->ip6h_nxt; ulp = NULL; break; case IPPROTO_ROUTING: /* RFC 2460 */ PULLUP_TO(hlen, ulp, struct ip6_rthdr); switch (((struct ip6_rthdr *)ulp)->ip6r_type) { case 0: ext_hd |= EXT_RTHDR0; break; case 2: ext_hd |= EXT_RTHDR2; break; default: if (V_fw_verbose) printf("IPFW2: IPV6 - Unknown " "Routing Header type(%d)\n", ((struct ip6_rthdr *) ulp)->ip6r_type); if (V_fw_deny_unknown_exthdrs) return (IP_FW_DENY); break; } ext_hd |= EXT_ROUTING; hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3; proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt; ulp = NULL; break; case IPPROTO_FRAGMENT: /* RFC 2460 */ PULLUP_TO(hlen, ulp, struct ip6_frag); ext_hd |= EXT_FRAGMENT; hlen += sizeof (struct ip6_frag); proto = ((struct ip6_frag *)ulp)->ip6f_nxt; offset = ((struct ip6_frag *)ulp)->ip6f_offlg & IP6F_OFF_MASK; ip6f_mf = ((struct ip6_frag *)ulp)->ip6f_offlg & IP6F_MORE_FRAG; if (V_fw_permit_single_frag6 == 0 && offset == 0 && ip6f_mf == 0) { if (V_fw_verbose) printf("IPFW2: IPV6 - Invalid " "Fragment Header\n"); if (V_fw_deny_unknown_exthdrs) return (IP_FW_DENY); break; } args->f_id.extra = ntohl(((struct ip6_frag *)ulp)->ip6f_ident); ulp = NULL; break; case IPPROTO_DSTOPTS: /* RFC 2460 */ PULLUP_TO(hlen, ulp, struct ip6_hbh); ext_hd |= EXT_DSTOPTS; hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3; proto = ((struct ip6_hbh *)ulp)->ip6h_nxt; ulp = NULL; break; case IPPROTO_AH: /* RFC 2402 */ PULLUP_TO(hlen, ulp, struct ip6_ext); ext_hd |= EXT_AH; hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2; proto = ((struct ip6_ext *)ulp)->ip6e_nxt; ulp = NULL; break; case IPPROTO_ESP: /* RFC 2406 */ PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */ /* Anything past Seq# is variable length and * data past this ext. header is encrypted. */ ext_hd |= EXT_ESP; break; case IPPROTO_NONE: /* RFC 2460 */ /* * Packet ends here, and IPv6 header has * already been pulled up. If ip6e_len!=0 * then octets must be ignored. */ ulp = ip; /* non-NULL to get out of loop. */ break; case IPPROTO_OSPFIGP: /* XXX OSPF header check? */ PULLUP_TO(hlen, ulp, struct ip6_ext); break; case IPPROTO_PIM: /* XXX PIM header check? */ PULLUP_TO(hlen, ulp, struct pim); break; case IPPROTO_GRE: /* RFC 1701 */ /* XXX GRE header check? */ PULLUP_TO(hlen, ulp, struct grehdr); break; case IPPROTO_CARP: PULLUP_TO(hlen, ulp, struct carp_header); if (((struct carp_header *)ulp)->carp_version != CARP_VERSION) return (IP_FW_DENY); if (((struct carp_header *)ulp)->carp_type != CARP_ADVERTISEMENT) return (IP_FW_DENY); break; case IPPROTO_IPV6: /* RFC 2893 */ PULLUP_TO(hlen, ulp, struct ip6_hdr); break; case IPPROTO_IPV4: /* RFC 2893 */ PULLUP_TO(hlen, ulp, struct ip); break; default: if (V_fw_verbose) printf("IPFW2: IPV6 - Unknown " "Extension Header(%d), ext_hd=%x\n", proto, ext_hd); if (V_fw_deny_unknown_exthdrs) return (IP_FW_DENY); PULLUP_TO(hlen, ulp, struct ip6_ext); break; } /*switch */ } ip = mtod(m, struct ip *); ip6 = (struct ip6_hdr *)ip; args->f_id.src_ip6 = ip6->ip6_src; args->f_id.dst_ip6 = ip6->ip6_dst; args->f_id.src_ip = 0; args->f_id.dst_ip = 0; args->f_id.flow_id6 = ntohl(ip6->ip6_flow); iplen = ntohs(ip6->ip6_plen) + sizeof(*ip6); } else if (pktlen >= sizeof(struct ip) && (args->eh == NULL || etype == ETHERTYPE_IP) && ip->ip_v == 4) { is_ipv4 = 1; hlen = ip->ip_hl << 2; args->f_id.addr_type = 4; /* * Collect parameters into local variables for faster matching. */ proto = ip->ip_p; src_ip = ip->ip_src; dst_ip = ip->ip_dst; offset = ntohs(ip->ip_off) & IP_OFFMASK; iplen = ntohs(ip->ip_len); if (offset == 0) { switch (proto) { case IPPROTO_TCP: PULLUP_TO(hlen, ulp, struct tcphdr); dst_port = TCP(ulp)->th_dport; src_port = TCP(ulp)->th_sport; /* save flags for dynamic rules */ args->f_id._flags = TCP(ulp)->th_flags; break; case IPPROTO_SCTP: if (pktlen >= hlen + sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd)) PULLUP_LEN(hlen, ulp, sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) + offsetof(struct sctp_init, a_rwnd)); else if (pktlen >= hlen + sizeof(struct sctphdr)) PULLUP_LEN(hlen, ulp, pktlen - hlen); else PULLUP_LEN(hlen, ulp, sizeof(struct sctphdr)); src_port = SCTP(ulp)->src_port; dst_port = SCTP(ulp)->dest_port; break; case IPPROTO_UDP: case IPPROTO_UDPLITE: PULLUP_TO(hlen, ulp, struct udphdr); dst_port = UDP(ulp)->uh_dport; src_port = UDP(ulp)->uh_sport; break; case IPPROTO_ICMP: PULLUP_TO(hlen, ulp, struct icmphdr); //args->f_id.flags = ICMP(ulp)->icmp_type; break; default: break; } } ip = mtod(m, struct ip *); args->f_id.src_ip = ntohl(src_ip.s_addr); args->f_id.dst_ip = ntohl(dst_ip.s_addr); } #undef PULLUP_TO pktlen = iplen < pktlen ? iplen: pktlen; if (proto) { /* we may have port numbers, store them */ args->f_id.proto = proto; args->f_id.src_port = src_port = ntohs(src_port); args->f_id.dst_port = dst_port = ntohs(dst_port); } IPFW_PF_RLOCK(chain); if (! V_ipfw_vnet_ready) { /* shutting down, leave NOW. */ IPFW_PF_RUNLOCK(chain); return (IP_FW_PASS); /* accept */ } if (args->rule.slot) { /* * Packet has already been tagged as a result of a previous * match on rule args->rule aka args->rule_id (PIPE, QUEUE, * REASS, NETGRAPH, DIVERT/TEE...) * Validate the slot and continue from the next one * if still present, otherwise do a lookup. */ f_pos = (args->rule.chain_id == chain->id) ? args->rule.slot : ipfw_find_rule(chain, args->rule.rulenum, args->rule.rule_id); } else { f_pos = 0; } /* * Now scan the rules, and parse microinstructions for each rule. * We have two nested loops and an inner switch. Sometimes we * need to break out of one or both loops, or re-enter one of * the loops with updated variables. Loop variables are: * * f_pos (outer loop) points to the current rule. * On output it points to the matching rule. * done (outer loop) is used as a flag to break the loop. * l (inner loop) residual length of current rule. * cmd points to the current microinstruction. * * We break the inner loop by setting l=0 and possibly * cmdlen=0 if we don't want to advance cmd. * We break the outer loop by setting done=1 * We can restart the inner loop by setting l>0 and f_pos, f, cmd * as needed. */ for (; f_pos < chain->n_rules; f_pos++) { ipfw_insn *cmd; uint32_t tablearg = 0; int l, cmdlen, skip_or; /* skip rest of OR block */ struct ip_fw *f; f = chain->map[f_pos]; if (V_set_disable & (1 << f->set) ) continue; skip_or = 0; for (l = f->cmd_len, cmd = f->cmd ; l > 0 ; l -= cmdlen, cmd += cmdlen) { int match; /* * check_body is a jump target used when we find a * CHECK_STATE, and need to jump to the body of * the target rule. */ /* check_body: */ cmdlen = F_LEN(cmd); /* * An OR block (insn_1 || .. || insn_n) has the * F_OR bit set in all but the last instruction. * The first match will set "skip_or", and cause * the following instructions to be skipped until * past the one with the F_OR bit clear. */ if (skip_or) { /* skip this instruction */ if ((cmd->len & F_OR) == 0) skip_or = 0; /* next one is good */ continue; } match = 0; /* set to 1 if we succeed */ switch (cmd->opcode) { /* * The first set of opcodes compares the packet's * fields with some pattern, setting 'match' if a * match is found. At the end of the loop there is * logic to deal with F_NOT and F_OR flags associated * with the opcode. */ case O_NOP: match = 1; break; case O_FORWARD_MAC: printf("ipfw: opcode %d unimplemented\n", cmd->opcode); break; case O_GID: case O_UID: case O_JAIL: /* * We only check offset == 0 && proto != 0, * as this ensures that we have a * packet with the ports info. */ if (offset != 0) break; if (proto == IPPROTO_TCP || proto == IPPROTO_UDP || proto == IPPROTO_UDPLITE) match = check_uidgid( (ipfw_insn_u32 *)cmd, args, &ucred_lookup, #ifdef __FreeBSD__ &ucred_cache); #else (void *)&ucred_cache); #endif break; case O_RECV: match = iface_match(m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd, chain, &tablearg); break; case O_XMIT: match = iface_match(oif, (ipfw_insn_if *)cmd, chain, &tablearg); break; case O_VIA: match = iface_match(oif ? oif : m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd, chain, &tablearg); break; case O_MACADDR2: if (args->eh != NULL) { /* have MAC header */ u_int32_t *want = (u_int32_t *) ((ipfw_insn_mac *)cmd)->addr; u_int32_t *mask = (u_int32_t *) ((ipfw_insn_mac *)cmd)->mask; u_int32_t *hdr = (u_int32_t *)args->eh; match = ( want[0] == (hdr[0] & mask[0]) && want[1] == (hdr[1] & mask[1]) && want[2] == (hdr[2] & mask[2]) ); } break; case O_MAC_TYPE: if (args->eh != NULL) { u_int16_t *p = ((ipfw_insn_u16 *)cmd)->ports; int i; for (i = cmdlen - 1; !match && i>0; i--, p += 2) match = (etype >= p[0] && etype <= p[1]); } break; case O_FRAG: match = (offset != 0); break; case O_IN: /* "out" is "not in" */ match = (oif == NULL); break; case O_LAYER2: match = (args->eh != NULL); break; case O_DIVERTED: { /* For diverted packets, args->rule.info * contains the divert port (in host format) * reason and direction. */ uint32_t i = args->rule.info; match = (i&IPFW_IS_MASK) == IPFW_IS_DIVERT && cmd->arg1 & ((i & IPFW_INFO_IN) ? 1 : 2); } break; case O_PROTO: /* * We do not allow an arg of 0 so the * check of "proto" only suffices. */ match = (proto == cmd->arg1); break; case O_IP_SRC: match = is_ipv4 && (((ipfw_insn_ip *)cmd)->addr.s_addr == src_ip.s_addr); break; case O_IP_DST_LOOKUP: { void *pkey; uint32_t vidx, key; uint16_t keylen; if (cmdlen > F_INSN_SIZE(ipfw_insn_u32)) { /* Determine lookup key type */ vidx = ((ipfw_insn_u32 *)cmd)->d[1]; if (vidx != 4 /* uid */ && vidx != 5 /* jail */ && is_ipv6 == 0 && is_ipv4 == 0) break; /* Determine key length */ if (vidx == 0 /* dst-ip */ || vidx == 1 /* src-ip */) keylen = is_ipv6 ? sizeof(struct in6_addr): sizeof(in_addr_t); else { keylen = sizeof(key); pkey = &key; } if (vidx == 0 /* dst-ip */) pkey = is_ipv4 ? (void *)&dst_ip: (void *)&args->f_id.dst_ip6; else if (vidx == 1 /* src-ip */) pkey = is_ipv4 ? (void *)&src_ip: (void *)&args->f_id.src_ip6; else if (vidx == 6 /* dscp */) { if (is_ipv4) key = ip->ip_tos >> 2; else { key = args->f_id.flow_id6; key = (key & 0x0f) << 2 | (key & 0xf000) >> 14; } key &= 0x3f; } else if (vidx == 2 /* dst-port */ || vidx == 3 /* src-port */) { /* Skip fragments */ if (offset != 0) break; /* Skip proto without ports */ if (proto != IPPROTO_TCP && proto != IPPROTO_UDP && proto != IPPROTO_UDPLITE && proto != IPPROTO_SCTP) break; if (vidx == 2 /* dst-port */) key = dst_port; else key = src_port; } #ifndef USERSPACE else if (vidx == 4 /* uid */ || vidx == 5 /* jail */) { check_uidgid( (ipfw_insn_u32 *)cmd, args, &ucred_lookup, #ifdef __FreeBSD__ &ucred_cache); if (vidx == 4 /* uid */) key = ucred_cache->cr_uid; else if (vidx == 5 /* jail */) key = ucred_cache->cr_prison->pr_id; #else /* !__FreeBSD__ */ (void *)&ucred_cache); if (vidx == 4 /* uid */) key = ucred_cache.uid; else if (vidx == 5 /* jail */) key = ucred_cache.xid; #endif /* !__FreeBSD__ */ } #endif /* !USERSPACE */ else break; match = ipfw_lookup_table(chain, cmd->arg1, keylen, pkey, &vidx); if (!match) break; tablearg = vidx; break; } /* cmdlen =< F_INSN_SIZE(ipfw_insn_u32) */ /* FALLTHROUGH */ } case O_IP_SRC_LOOKUP: { void *pkey; uint32_t vidx; uint16_t keylen; if (is_ipv4) { keylen = sizeof(in_addr_t); if (cmd->opcode == O_IP_DST_LOOKUP) pkey = &dst_ip; else pkey = &src_ip; } else if (is_ipv6) { keylen = sizeof(struct in6_addr); if (cmd->opcode == O_IP_DST_LOOKUP) pkey = &args->f_id.dst_ip6; else pkey = &args->f_id.src_ip6; } else break; match = ipfw_lookup_table(chain, cmd->arg1, keylen, pkey, &vidx); if (!match) break; if (cmdlen == F_INSN_SIZE(ipfw_insn_u32)) { match = ((ipfw_insn_u32 *)cmd)->d[0] == TARG_VAL(chain, vidx, tag); if (!match) break; } tablearg = vidx; break; } case O_IP_FLOW_LOOKUP: { uint32_t v = 0; match = ipfw_lookup_table(chain, cmd->arg1, 0, &args->f_id, &v); if (cmdlen == F_INSN_SIZE(ipfw_insn_u32)) match = ((ipfw_insn_u32 *)cmd)->d[0] == TARG_VAL(chain, v, tag); if (match) tablearg = v; } break; case O_IP_SRC_MASK: case O_IP_DST_MASK: if (is_ipv4) { uint32_t a = (cmd->opcode == O_IP_DST_MASK) ? dst_ip.s_addr : src_ip.s_addr; uint32_t *p = ((ipfw_insn_u32 *)cmd)->d; int i = cmdlen-1; for (; !match && i>0; i-= 2, p+= 2) match = (p[0] == (a & p[1])); } break; case O_IP_SRC_ME: if (is_ipv4) { match = in_localip(src_ip); break; } #ifdef INET6 /* FALLTHROUGH */ case O_IP6_SRC_ME: match= is_ipv6 && ipfw_localip6(&args->f_id.src_ip6); #endif break; case O_IP_DST_SET: case O_IP_SRC_SET: if (is_ipv4) { u_int32_t *d = (u_int32_t *)(cmd+1); u_int32_t addr = cmd->opcode == O_IP_DST_SET ? args->f_id.dst_ip : args->f_id.src_ip; if (addr < d[0]) break; addr -= d[0]; /* subtract base */ match = (addr < cmd->arg1) && ( d[ 1 + (addr>>5)] & (1<<(addr & 0x1f)) ); } break; case O_IP_DST: match = is_ipv4 && (((ipfw_insn_ip *)cmd)->addr.s_addr == dst_ip.s_addr); break; case O_IP_DST_ME: if (is_ipv4) { match = in_localip(dst_ip); break; } #ifdef INET6 /* FALLTHROUGH */ case O_IP6_DST_ME: match= is_ipv6 && ipfw_localip6(&args->f_id.dst_ip6); #endif break; case O_IP_SRCPORT: case O_IP_DSTPORT: /* * offset == 0 && proto != 0 is enough * to guarantee that we have a * packet with port info. */ if ((proto == IPPROTO_UDP || proto == IPPROTO_UDPLITE || proto == IPPROTO_TCP || proto == IPPROTO_SCTP) && offset == 0) { u_int16_t x = (cmd->opcode == O_IP_SRCPORT) ? src_port : dst_port ; u_int16_t *p = ((ipfw_insn_u16 *)cmd)->ports; int i; for (i = cmdlen - 1; !match && i>0; i--, p += 2) match = (x>=p[0] && x<=p[1]); } break; case O_ICMPTYPE: match = (offset == 0 && proto==IPPROTO_ICMP && icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) ); break; #ifdef INET6 case O_ICMP6TYPE: match = is_ipv6 && offset == 0 && proto==IPPROTO_ICMPV6 && icmp6type_match( ICMP6(ulp)->icmp6_type, (ipfw_insn_u32 *)cmd); break; #endif /* INET6 */ case O_IPOPT: match = (is_ipv4 && ipopts_match(ip, cmd) ); break; case O_IPVER: match = (is_ipv4 && cmd->arg1 == ip->ip_v); break; case O_IPID: case O_IPLEN: case O_IPTTL: if (is_ipv4) { /* only for IP packets */ uint16_t x; uint16_t *p; int i; if (cmd->opcode == O_IPLEN) x = iplen; else if (cmd->opcode == O_IPTTL) x = ip->ip_ttl; else /* must be IPID */ x = ntohs(ip->ip_id); if (cmdlen == 1) { match = (cmd->arg1 == x); break; } /* otherwise we have ranges */ p = ((ipfw_insn_u16 *)cmd)->ports; i = cmdlen - 1; for (; !match && i>0; i--, p += 2) match = (x >= p[0] && x <= p[1]); } break; case O_IPPRECEDENCE: match = (is_ipv4 && (cmd->arg1 == (ip->ip_tos & 0xe0)) ); break; case O_IPTOS: match = (is_ipv4 && flags_match(cmd, ip->ip_tos)); break; case O_DSCP: { uint32_t *p; uint16_t x; p = ((ipfw_insn_u32 *)cmd)->d; if (is_ipv4) x = ip->ip_tos >> 2; else if (is_ipv6) { uint8_t *v; v = &((struct ip6_hdr *)ip)->ip6_vfc; x = (*v & 0x0F) << 2; v++; x |= *v >> 6; } else break; /* DSCP bitmask is stored as low_u32 high_u32 */ if (x >= 32) match = *(p + 1) & (1 << (x - 32)); else match = *p & (1 << x); } break; case O_TCPDATALEN: if (proto == IPPROTO_TCP && offset == 0) { struct tcphdr *tcp; uint16_t x; uint16_t *p; int i; #ifdef INET6 if (is_ipv6) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip; if (ip6->ip6_plen == 0) { /* * Jumbo payload is not * supported by this * opcode. */ break; } x = iplen - hlen; } else #endif /* INET6 */ x = iplen - (ip->ip_hl << 2); tcp = TCP(ulp); x -= tcp->th_off << 2; if (cmdlen == 1) { match = (cmd->arg1 == x); break; } /* otherwise we have ranges */ p = ((ipfw_insn_u16 *)cmd)->ports; i = cmdlen - 1; for (; !match && i>0; i--, p += 2) match = (x >= p[0] && x <= p[1]); } break; case O_TCPFLAGS: match = (proto == IPPROTO_TCP && offset == 0 && flags_match(cmd, TCP(ulp)->th_flags)); break; case O_TCPOPTS: if (proto == IPPROTO_TCP && offset == 0 && ulp){ PULLUP_LEN(hlen, ulp, (TCP(ulp)->th_off << 2)); match = tcpopts_match(TCP(ulp), cmd); } break; case O_TCPSEQ: match = (proto == IPPROTO_TCP && offset == 0 && ((ipfw_insn_u32 *)cmd)->d[0] == TCP(ulp)->th_seq); break; case O_TCPACK: match = (proto == IPPROTO_TCP && offset == 0 && ((ipfw_insn_u32 *)cmd)->d[0] == TCP(ulp)->th_ack); break; case O_TCPWIN: if (proto == IPPROTO_TCP && offset == 0) { uint16_t x; uint16_t *p; int i; x = ntohs(TCP(ulp)->th_win); if (cmdlen == 1) { match = (cmd->arg1 == x); break; } /* Otherwise we have ranges. */ p = ((ipfw_insn_u16 *)cmd)->ports; i = cmdlen - 1; for (; !match && i > 0; i--, p += 2) match = (x >= p[0] && x <= p[1]); } break; case O_ESTAB: /* reject packets which have SYN only */ /* XXX should i also check for TH_ACK ? */ match = (proto == IPPROTO_TCP && offset == 0 && (TCP(ulp)->th_flags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN); break; case O_ALTQ: { struct pf_mtag *at; struct m_tag *mtag; ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd; /* * ALTQ uses mbuf tags from another * packet filtering system - pf(4). * We allocate a tag in its format * and fill it in, pretending to be pf(4). */ match = 1; at = pf_find_mtag(m); if (at != NULL && at->qid != 0) break; mtag = m_tag_get(PACKET_TAG_PF, sizeof(struct pf_mtag), M_NOWAIT | M_ZERO); if (mtag == NULL) { /* * Let the packet fall back to the * default ALTQ. */ break; } m_tag_prepend(m, mtag); at = (struct pf_mtag *)(mtag + 1); at->qid = altq->qid; at->hdr = ip; break; } case O_LOG: ipfw_log(chain, f, hlen, args, m, oif, offset | ip6f_mf, tablearg, ip); match = 1; break; case O_PROB: match = (random()<((ipfw_insn_u32 *)cmd)->d[0]); break; case O_VERREVPATH: /* Outgoing packets automatically pass/match */ match = ((oif != NULL) || (m->m_pkthdr.rcvif == NULL) || ( #ifdef INET6 is_ipv6 ? verify_path6(&(args->f_id.src_ip6), m->m_pkthdr.rcvif, args->f_id.fib) : #endif verify_path(src_ip, m->m_pkthdr.rcvif, args->f_id.fib))); break; case O_VERSRCREACH: /* Outgoing packets automatically pass/match */ match = (hlen > 0 && ((oif != NULL) || #ifdef INET6 is_ipv6 ? verify_path6(&(args->f_id.src_ip6), NULL, args->f_id.fib) : #endif verify_path(src_ip, NULL, args->f_id.fib))); break; case O_ANTISPOOF: /* Outgoing packets automatically pass/match */ if (oif == NULL && hlen > 0 && ( (is_ipv4 && in_localaddr(src_ip)) #ifdef INET6 || (is_ipv6 && in6_localaddr(&(args->f_id.src_ip6))) #endif )) match = #ifdef INET6 is_ipv6 ? verify_path6( &(args->f_id.src_ip6), m->m_pkthdr.rcvif, args->f_id.fib) : #endif verify_path(src_ip, m->m_pkthdr.rcvif, args->f_id.fib); else match = 1; break; case O_IPSEC: match = (m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL); /* otherwise no match */ break; #ifdef INET6 case O_IP6_SRC: match = is_ipv6 && IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6, &((ipfw_insn_ip6 *)cmd)->addr6); break; case O_IP6_DST: match = is_ipv6 && IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6, &((ipfw_insn_ip6 *)cmd)->addr6); break; case O_IP6_SRC_MASK: case O_IP6_DST_MASK: if (is_ipv6) { int i = cmdlen - 1; struct in6_addr p; struct in6_addr *d = &((ipfw_insn_ip6 *)cmd)->addr6; for (; !match && i > 0; d += 2, i -= F_INSN_SIZE(struct in6_addr) * 2) { p = (cmd->opcode == O_IP6_SRC_MASK) ? args->f_id.src_ip6: args->f_id.dst_ip6; APPLY_MASK(&p, &d[1]); match = IN6_ARE_ADDR_EQUAL(&d[0], &p); } } break; case O_FLOW6ID: match = is_ipv6 && flow6id_match(args->f_id.flow_id6, (ipfw_insn_u32 *) cmd); break; case O_EXT_HDR: match = is_ipv6 && (ext_hd & ((ipfw_insn *) cmd)->arg1); break; case O_IP6: match = is_ipv6; break; #endif case O_IP4: match = is_ipv4; break; case O_TAG: { struct m_tag *mtag; uint32_t tag = TARG(cmd->arg1, tag); /* Packet is already tagged with this tag? */ mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL); /* We have `untag' action when F_NOT flag is * present. And we must remove this mtag from * mbuf and reset `match' to zero (`match' will * be inversed later). * Otherwise we should allocate new mtag and * push it into mbuf. */ if (cmd->len & F_NOT) { /* `untag' action */ if (mtag != NULL) m_tag_delete(m, mtag); match = 0; } else { if (mtag == NULL) { mtag = m_tag_alloc( MTAG_IPFW, tag, 0, M_NOWAIT); if (mtag != NULL) m_tag_prepend(m, mtag); } match = 1; } break; } case O_FIB: /* try match the specified fib */ if (args->f_id.fib == cmd->arg1) match = 1; break; case O_SOCKARG: { #ifndef USERSPACE /* not supported in userspace */ struct inpcb *inp = args->inp; struct inpcbinfo *pi; if (is_ipv6) /* XXX can we remove this ? */ break; if (proto == IPPROTO_TCP) pi = &V_tcbinfo; else if (proto == IPPROTO_UDP) pi = &V_udbinfo; else if (proto == IPPROTO_UDPLITE) pi = &V_ulitecbinfo; else break; /* * XXXRW: so_user_cookie should almost * certainly be inp_user_cookie? */ /* For incoming packet, lookup up the inpcb using the src/dest ip/port tuple */ if (inp == NULL) { inp = in_pcblookup(pi, src_ip, htons(src_port), dst_ip, htons(dst_port), INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { tablearg = inp->inp_socket->so_user_cookie; if (tablearg) match = 1; INP_RUNLOCK(inp); } } else { if (inp->inp_socket) { tablearg = inp->inp_socket->so_user_cookie; if (tablearg) match = 1; } } #endif /* !USERSPACE */ break; } case O_TAGGED: { struct m_tag *mtag; uint32_t tag = TARG(cmd->arg1, tag); if (cmdlen == 1) { match = m_tag_locate(m, MTAG_IPFW, tag, NULL) != NULL; break; } /* we have ranges */ for (mtag = m_tag_first(m); mtag != NULL && !match; mtag = m_tag_next(m, mtag)) { uint16_t *p; int i; if (mtag->m_tag_cookie != MTAG_IPFW) continue; p = ((ipfw_insn_u16 *)cmd)->ports; i = cmdlen - 1; for(; !match && i > 0; i--, p += 2) match = mtag->m_tag_id >= p[0] && mtag->m_tag_id <= p[1]; } break; } /* * The second set of opcodes represents 'actions', * i.e. the terminal part of a rule once the packet * matches all previous patterns. * Typically there is only one action for each rule, * and the opcode is stored at the end of the rule * (but there are exceptions -- see below). * * In general, here we set retval and terminate the * outer loop (would be a 'break 3' in some language, * but we need to set l=0, done=1) * * Exceptions: * O_COUNT and O_SKIPTO actions: * instead of terminating, we jump to the next rule * (setting l=0), or to the SKIPTO target (setting * f/f_len, cmd and l as needed), respectively. * * O_TAG, O_LOG and O_ALTQ action parameters: * perform some action and set match = 1; * * O_LIMIT and O_KEEP_STATE: these opcodes are * not real 'actions', and are stored right * before the 'action' part of the rule (one * exception is O_SKIP_ACTION which could be * between these opcodes and 'action' one). * These opcodes try to install an entry in the * state tables; if successful, we continue with * the next opcode (match=1; break;), otherwise * the packet must be dropped (set retval, * break loops with l=0, done=1) * * O_PROBE_STATE and O_CHECK_STATE: these opcodes * cause a lookup of the state table, and a jump * to the 'action' part of the parent rule * if an entry is found, or * (CHECK_STATE only) a jump to the next rule if * the entry is not found. * The result of the lookup is cached so that * further instances of these opcodes become NOPs. * The jump to the next rule is done by setting * l=0, cmdlen=0. * * O_SKIP_ACTION: this opcode is not a real 'action' * either, and is stored right before the 'action' * part of the rule, right after the O_KEEP_STATE * opcode. It causes match failure so the real * 'action' could be executed only if the rule * is checked via dynamic rule from the state * table, as in such case execution starts * from the true 'action' opcode directly. * */ case O_LIMIT: case O_KEEP_STATE: if (ipfw_dyn_install_state(chain, f, (ipfw_insn_limit *)cmd, args, ulp, pktlen, &dyn_info, tablearg)) { /* error or limit violation */ retval = IP_FW_DENY; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ } match = 1; break; case O_PROBE_STATE: case O_CHECK_STATE: /* * dynamic rules are checked at the first * keep-state or check-state occurrence, * with the result being stored in dyn_info. * The compiler introduces a PROBE_STATE * instruction for us when we have a * KEEP_STATE (because PROBE_STATE needs * to be run first). */ if (DYN_LOOKUP_NEEDED(&dyn_info, cmd) && (q = ipfw_dyn_lookup_state(args, ulp, pktlen, cmd, &dyn_info)) != NULL) { /* * Found dynamic entry, jump to the * 'action' part of the parent rule * by setting f, cmd, l and clearing * cmdlen. */ f = q; f_pos = dyn_info.f_pos; cmd = ACTION_PTR(f); l = f->cmd_len - f->act_ofs; cmdlen = 0; match = 1; break; } /* * Dynamic entry not found. If CHECK_STATE, * skip to next rule, if PROBE_STATE just * ignore and continue with next opcode. */ if (cmd->opcode == O_CHECK_STATE) l = 0; /* exit inner loop */ match = 1; break; case O_SKIP_ACTION: match = 0; /* skip to the next rule */ l = 0; /* exit inner loop */ break; case O_ACCEPT: retval = 0; /* accept */ l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; case O_PIPE: case O_QUEUE: set_match(args, f_pos, chain); args->rule.info = TARG(cmd->arg1, pipe); if (cmd->opcode == O_PIPE) args->rule.info |= IPFW_IS_PIPE; if (V_fw_one_pass) args->rule.info |= IPFW_ONEPASS; retval = IP_FW_DUMMYNET; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; case O_DIVERT: case O_TEE: if (args->eh) /* not on layer 2 */ break; /* otherwise this is terminal */ l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ retval = (cmd->opcode == O_DIVERT) ? IP_FW_DIVERT : IP_FW_TEE; set_match(args, f_pos, chain); args->rule.info = TARG(cmd->arg1, divert); break; case O_COUNT: IPFW_INC_RULE_COUNTER(f, pktlen); l = 0; /* exit inner loop */ break; case O_SKIPTO: IPFW_INC_RULE_COUNTER(f, pktlen); f_pos = JUMP(chain, f, cmd->arg1, tablearg, 0); /* * Skip disabled rules, and re-enter * the inner loop with the correct * f_pos, f, l and cmd. * Also clear cmdlen and skip_or */ for (; f_pos < chain->n_rules - 1 && (V_set_disable & (1 << chain->map[f_pos]->set)); f_pos++) ; /* Re-enter the inner loop at the skipto rule. */ f = chain->map[f_pos]; l = f->cmd_len; cmd = f->cmd; match = 1; cmdlen = 0; skip_or = 0; continue; break; /* not reached */ case O_CALLRETURN: { /* * Implementation of `subroutine' call/return, * in the stack carried in an mbuf tag. This * is different from `skipto' in that any call * address is possible (`skipto' must prevent * backward jumps to avoid endless loops). * We have `return' action when F_NOT flag is * present. The `m_tag_id' field is used as * stack pointer. */ struct m_tag *mtag; uint16_t jmpto, *stack; #define IS_CALL ((cmd->len & F_NOT) == 0) #define IS_RETURN ((cmd->len & F_NOT) != 0) /* * Hand-rolled version of m_tag_locate() with * wildcard `type'. * If not already tagged, allocate new tag. */ mtag = m_tag_first(m); while (mtag != NULL) { if (mtag->m_tag_cookie == MTAG_IPFW_CALL) break; mtag = m_tag_next(m, mtag); } if (mtag == NULL && IS_CALL) { mtag = m_tag_alloc(MTAG_IPFW_CALL, 0, IPFW_CALLSTACK_SIZE * sizeof(uint16_t), M_NOWAIT); if (mtag != NULL) m_tag_prepend(m, mtag); } /* * On error both `call' and `return' just * continue with next rule. */ if (IS_RETURN && (mtag == NULL || mtag->m_tag_id == 0)) { l = 0; /* exit inner loop */ break; } if (IS_CALL && (mtag == NULL || mtag->m_tag_id >= IPFW_CALLSTACK_SIZE)) { printf("ipfw: call stack error, " "go to next rule\n"); l = 0; /* exit inner loop */ break; } IPFW_INC_RULE_COUNTER(f, pktlen); stack = (uint16_t *)(mtag + 1); /* * The `call' action may use cached f_pos * (in f->next_rule), whose version is written * in f->next_rule. * The `return' action, however, doesn't have * fixed jump address in cmd->arg1 and can't use * cache. */ if (IS_CALL) { stack[mtag->m_tag_id] = f->rulenum; mtag->m_tag_id++; f_pos = JUMP(chain, f, cmd->arg1, tablearg, 1); } else { /* `return' action */ mtag->m_tag_id--; jmpto = stack[mtag->m_tag_id] + 1; f_pos = ipfw_find_rule(chain, jmpto, 0); } /* * Skip disabled rules, and re-enter * the inner loop with the correct * f_pos, f, l and cmd. * Also clear cmdlen and skip_or */ for (; f_pos < chain->n_rules - 1 && (V_set_disable & (1 << chain->map[f_pos]->set)); f_pos++) ; /* Re-enter the inner loop at the dest rule. */ f = chain->map[f_pos]; l = f->cmd_len; cmd = f->cmd; cmdlen = 0; skip_or = 0; continue; break; /* NOTREACHED */ } #undef IS_CALL #undef IS_RETURN case O_REJECT: /* * Drop the packet and send a reject notice * if the packet is not ICMP (or is an ICMP * query), and it is not multicast/broadcast. */ if (hlen > 0 && is_ipv4 && offset == 0 && (proto != IPPROTO_ICMP || is_icmp_query(ICMP(ulp))) && !(m->m_flags & (M_BCAST|M_MCAST)) && !IN_MULTICAST(ntohl(dst_ip.s_addr))) { send_reject(args, cmd->arg1, iplen, ip); m = args->m; } /* FALLTHROUGH */ #ifdef INET6 case O_UNREACH6: if (hlen > 0 && is_ipv6 && ((offset & IP6F_OFF_MASK) == 0) && (proto != IPPROTO_ICMPV6 || (is_icmp6_query(icmp6_type) == 1)) && !(m->m_flags & (M_BCAST|M_MCAST)) && !IN6_IS_ADDR_MULTICAST( &args->f_id.dst_ip6)) { send_reject6(args, cmd->opcode == O_REJECT ? map_icmp_unreach(cmd->arg1): cmd->arg1, hlen, (struct ip6_hdr *)ip); m = args->m; } /* FALLTHROUGH */ #endif case O_DENY: retval = IP_FW_DENY; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; case O_FORWARD_IP: if (args->eh) /* not valid on layer2 pkts */ break; if (q != f || dyn_info.direction == MATCH_FORWARD) { struct sockaddr_in *sa; sa = &(((ipfw_insn_sa *)cmd)->sa); if (sa->sin_addr.s_addr == INADDR_ANY) { #ifdef INET6 /* * We use O_FORWARD_IP opcode for * fwd rule with tablearg, but tables * now support IPv6 addresses. And * when we are inspecting IPv6 packet, * we can use nh6 field from * table_value as next_hop6 address. */ if (is_ipv6) { struct sockaddr_in6 *sa6; sa6 = args->next_hop6 = &args->hopstore6; sa6->sin6_family = AF_INET6; sa6->sin6_len = sizeof(*sa6); sa6->sin6_addr = TARG_VAL( chain, tablearg, nh6); sa6->sin6_port = sa->sin_port; /* * Set sin6_scope_id only for * link-local unicast addresses. */ if (IN6_IS_ADDR_LINKLOCAL( &sa6->sin6_addr)) sa6->sin6_scope_id = TARG_VAL(chain, tablearg, zoneid); } else #endif { args->hopstore.sin_port = sa->sin_port; sa = args->next_hop = &args->hopstore; sa->sin_family = AF_INET; sa->sin_len = sizeof(*sa); sa->sin_addr.s_addr = htonl( TARG_VAL(chain, tablearg, nh4)); } } else { args->next_hop = sa; } } retval = IP_FW_PASS; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; #ifdef INET6 case O_FORWARD_IP6: if (args->eh) /* not valid on layer2 pkts */ break; if (q != f || dyn_info.direction == MATCH_FORWARD) { struct sockaddr_in6 *sin6; sin6 = &(((ipfw_insn_sa6 *)cmd)->sa); args->next_hop6 = sin6; } retval = IP_FW_PASS; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; #endif case O_NETGRAPH: case O_NGTEE: set_match(args, f_pos, chain); args->rule.info = TARG(cmd->arg1, netgraph); if (V_fw_one_pass) args->rule.info |= IPFW_ONEPASS; retval = (cmd->opcode == O_NETGRAPH) ? IP_FW_NETGRAPH : IP_FW_NGTEE; l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ break; case O_SETFIB: { uint32_t fib; IPFW_INC_RULE_COUNTER(f, pktlen); fib = TARG(cmd->arg1, fib) & 0x7FFF; if (fib >= rt_numfibs) fib = 0; M_SETFIB(m, fib); args->f_id.fib = fib; l = 0; /* exit inner loop */ break; } case O_SETDSCP: { uint16_t code; code = TARG(cmd->arg1, dscp) & 0x3F; l = 0; /* exit inner loop */ if (is_ipv4) { uint16_t old; old = *(uint16_t *)ip; ip->ip_tos = (code << 2) | (ip->ip_tos & 0x03); ip->ip_sum = cksum_adjust(ip->ip_sum, old, *(uint16_t *)ip); } else if (is_ipv6) { uint8_t *v; v = &((struct ip6_hdr *)ip)->ip6_vfc; *v = (*v & 0xF0) | (code >> 2); v++; *v = (*v & 0x3F) | ((code & 0x03) << 6); } else break; IPFW_INC_RULE_COUNTER(f, pktlen); break; } case O_NAT: l = 0; /* exit inner loop */ done = 1; /* exit outer loop */ /* * Ensure that we do not invoke NAT handler for * non IPv4 packets. Libalias expects only IPv4. */ if (!is_ipv4 || !IPFW_NAT_LOADED) { retval = IP_FW_DENY; break; } struct cfg_nat *t; int nat_id; set_match(args, f_pos, chain); /* Check if this is 'global' nat rule */ if (cmd->arg1 == IP_FW_NAT44_GLOBAL) { retval = ipfw_nat_ptr(args, NULL, m); break; } t = ((ipfw_insn_nat *)cmd)->nat; if (t == NULL) { nat_id = TARG(cmd->arg1, nat); t = (*lookup_nat_ptr)(&chain->nat, nat_id); if (t == NULL) { retval = IP_FW_DENY; break; } if (cmd->arg1 != IP_FW_TARG) ((ipfw_insn_nat *)cmd)->nat = t; } retval = ipfw_nat_ptr(args, t, m); break; case O_REASS: { int ip_off; l = 0; /* in any case exit inner loop */ if (is_ipv6) /* IPv6 is not supported yet */ break; IPFW_INC_RULE_COUNTER(f, pktlen); ip_off = ntohs(ip->ip_off); /* if not fragmented, go to next rule */ if ((ip_off & (IP_MF | IP_OFFMASK)) == 0) break; args->m = m = ip_reass(m); /* * do IP header checksum fixup. */ if (m == NULL) { /* fragment got swallowed */ retval = IP_FW_DENY; } else { /* good, packet complete */ int hlen; ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; ip->ip_sum = 0; if (hlen == sizeof(struct ip)) ip->ip_sum = in_cksum_hdr(ip); else ip->ip_sum = in_cksum(m, hlen); retval = IP_FW_REASS; set_match(args, f_pos, chain); } done = 1; /* exit outer loop */ break; } case O_EXTERNAL_ACTION: l = 0; /* in any case exit inner loop */ retval = ipfw_run_eaction(chain, args, cmd, &done); /* * If both @retval and @done are zero, * consider this as rule matching and * update counters. */ if (retval == 0 && done == 0) { IPFW_INC_RULE_COUNTER(f, pktlen); /* * Reset the result of the last * dynamic state lookup. * External action can change * @args content, and it may be * used for new state lookup later. */ DYN_INFO_INIT(&dyn_info); } break; default: panic("-- unknown opcode %d\n", cmd->opcode); } /* end of switch() on opcodes */ /* * if we get here with l=0, then match is irrelevant. */ if (cmd->len & F_NOT) match = !match; if (match) { if (cmd->len & F_OR) skip_or = 1; } else { if (!(cmd->len & F_OR)) /* not an OR block, */ break; /* try next rule */ } } /* end of inner loop, scan opcodes */ #undef PULLUP_LEN if (done) break; /* next_rule:; */ /* try next rule */ } /* end of outer for, scan rules */ if (done) { struct ip_fw *rule = chain->map[f_pos]; /* Update statistics */ IPFW_INC_RULE_COUNTER(rule, pktlen); } else { retval = IP_FW_DENY; printf("ipfw: ouch!, skip past end of rules, denying packet\n"); } IPFW_PF_RUNLOCK(chain); #ifdef __FreeBSD__ if (ucred_cache != NULL) crfree(ucred_cache); #endif return (retval); pullup_failed: if (V_fw_verbose) printf("ipfw: pullup failed\n"); return (IP_FW_DENY); } /* * Set maximum number of tables that can be used in given VNET ipfw instance. */ #ifdef SYSCTL_NODE static int sysctl_ipfw_table_num(SYSCTL_HANDLER_ARGS) { int error; unsigned int ntables; ntables = V_fw_tables_max; error = sysctl_handle_int(oidp, &ntables, 0, req); /* Read operation or some error */ if ((error != 0) || (req->newptr == NULL)) return (error); return (ipfw_resize_tables(&V_layer3_chain, ntables)); } /* * Switches table namespace between global and per-set. */ static int sysctl_ipfw_tables_sets(SYSCTL_HANDLER_ARGS) { int error; unsigned int sets; sets = V_fw_tables_sets; error = sysctl_handle_int(oidp, &sets, 0, req); /* Read operation or some error */ if ((error != 0) || (req->newptr == NULL)) return (error); return (ipfw_switch_tables_namespace(&V_layer3_chain, sets)); } #endif /* * Module and VNET glue */ /* * Stuff that must be initialised only on boot or module load */ static int ipfw_init(void) { int error = 0; /* * Only print out this stuff the first time around, * when called from the sysinit code. */ printf("ipfw2 " #ifdef INET6 "(+ipv6) " #endif "initialized, divert %s, nat %s, " "default to %s, logging ", #ifdef IPDIVERT "enabled", #else "loadable", #endif #ifdef IPFIREWALL_NAT "enabled", #else "loadable", #endif default_to_accept ? "accept" : "deny"); /* * Note: V_xxx variables can be accessed here but the vnet specific * initializer may not have been called yet for the VIMAGE case. * Tuneables will have been processed. We will print out values for * the default vnet. * XXX This should all be rationalized AFTER 8.0 */ if (V_fw_verbose == 0) printf("disabled\n"); else if (V_verbose_limit == 0) printf("unlimited\n"); else printf("limited to %d packets/entry by default\n", V_verbose_limit); /* Check user-supplied table count for validness */ if (default_fw_tables > IPFW_TABLES_MAX) default_fw_tables = IPFW_TABLES_MAX; ipfw_init_sopt_handler(); ipfw_init_obj_rewriter(); ipfw_iface_init(); return (error); } /* * Called for the removal of the last instance only on module unload. */ static void ipfw_destroy(void) { ipfw_iface_destroy(); ipfw_destroy_sopt_handler(); ipfw_destroy_obj_rewriter(); printf("IP firewall unloaded\n"); } /* * Stuff that must be initialized for every instance * (including the first of course). */ static int vnet_ipfw_init(const void *unused) { int error, first; struct ip_fw *rule = NULL; struct ip_fw_chain *chain; chain = &V_layer3_chain; first = IS_DEFAULT_VNET(curvnet) ? 1 : 0; /* First set up some values that are compile time options */ V_autoinc_step = 100; /* bounded to 1..1000 in add_rule() */ V_fw_deny_unknown_exthdrs = 1; #ifdef IPFIREWALL_VERBOSE V_fw_verbose = 1; #endif #ifdef IPFIREWALL_VERBOSE_LIMIT V_verbose_limit = IPFIREWALL_VERBOSE_LIMIT; #endif #ifdef IPFIREWALL_NAT LIST_INIT(&chain->nat); #endif /* Init shared services hash table */ ipfw_init_srv(chain); ipfw_init_counters(); /* Set initial number of tables */ V_fw_tables_max = default_fw_tables; error = ipfw_init_tables(chain, first); if (error) { printf("ipfw2: setting up tables failed\n"); free(chain->map, M_IPFW); free(rule, M_IPFW); return (ENOSPC); } IPFW_LOCK_INIT(chain); /* fill and insert the default rule */ rule = ipfw_alloc_rule(chain, sizeof(struct ip_fw)); rule->cmd_len = 1; rule->cmd[0].len = 1; rule->cmd[0].opcode = default_to_accept ? O_ACCEPT : O_DENY; chain->default_rule = rule; ipfw_add_protected_rule(chain, rule, 0); ipfw_dyn_init(chain); ipfw_eaction_init(chain, first); #ifdef LINEAR_SKIPTO ipfw_init_skipto_cache(chain); #endif ipfw_bpf_init(first); /* First set up some values that are compile time options */ V_ipfw_vnet_ready = 1; /* Open for business */ /* * Hook the sockopt handler and pfil hooks for ipv4 and ipv6. * Even if the latter two fail we still keep the module alive * because the sockopt and layer2 paths are still useful. * ipfw[6]_hook return 0 on success, ENOENT on failure, * so we can ignore the exact return value and just set a flag. * * Note that V_fw[6]_enable are manipulated by a SYSCTL_PROC so * changes in the underlying (per-vnet) variables trigger * immediate hook()/unhook() calls. * In layer2 we have the same behaviour, except that V_ether_ipfw * is checked on each packet because there are no pfil hooks. */ V_ip_fw_ctl_ptr = ipfw_ctl3; error = ipfw_attach_hooks(1); return (error); } /* * Called for the removal of each instance. */ static int vnet_ipfw_uninit(const void *unused) { struct ip_fw *reap; struct ip_fw_chain *chain = &V_layer3_chain; int i, last; V_ipfw_vnet_ready = 0; /* tell new callers to go away */ /* * disconnect from ipv4, ipv6, layer2 and sockopt. * Then grab, release and grab again the WLOCK so we make * sure the update is propagated and nobody will be in. */ (void)ipfw_attach_hooks(0 /* detach */); V_ip_fw_ctl_ptr = NULL; last = IS_DEFAULT_VNET(curvnet) ? 1 : 0; IPFW_UH_WLOCK(chain); IPFW_UH_WUNLOCK(chain); ipfw_dyn_uninit(0); /* run the callout_drain */ IPFW_UH_WLOCK(chain); reap = NULL; IPFW_WLOCK(chain); for (i = 0; i < chain->n_rules; i++) ipfw_reap_add(chain, &reap, chain->map[i]); free(chain->map, M_IPFW); #ifdef LINEAR_SKIPTO ipfw_destroy_skipto_cache(chain); #endif IPFW_WUNLOCK(chain); IPFW_UH_WUNLOCK(chain); ipfw_destroy_tables(chain, last); ipfw_eaction_uninit(chain, last); if (reap != NULL) ipfw_reap_rules(reap); vnet_ipfw_iface_destroy(chain); ipfw_destroy_srv(chain); IPFW_LOCK_DESTROY(chain); ipfw_dyn_uninit(1); /* free the remaining parts */ ipfw_destroy_counters(); ipfw_bpf_uninit(last); return (0); } /* * Module event handler. * In general we have the choice of handling most of these events by the * event handler or by the (VNET_)SYS(UN)INIT handlers. I have chosen to * use the SYSINIT handlers as they are more capable of expressing the * flow of control during module and vnet operations, so this is just * a skeleton. Note there is no SYSINIT equivalent of the module * SHUTDOWN handler, but we don't have anything to do in that case anyhow. */ static int ipfw_modevent(module_t mod, int type, void *unused) { int err = 0; switch (type) { case MOD_LOAD: /* Called once at module load or * system boot if compiled in. */ break; case MOD_QUIESCE: /* Called before unload. May veto unloading. */ break; case MOD_UNLOAD: /* Called during unload. */ break; case MOD_SHUTDOWN: /* Called during system shutdown. */ break; default: err = EOPNOTSUPP; break; } return err; } static moduledata_t ipfwmod = { "ipfw", ipfw_modevent, 0 }; /* Define startup order. */ #define IPFW_SI_SUB_FIREWALL SI_SUB_PROTO_FIREWALL #define IPFW_MODEVENT_ORDER (SI_ORDER_ANY - 255) /* On boot slot in here. */ #define IPFW_MODULE_ORDER (IPFW_MODEVENT_ORDER + 1) /* A little later. */ #define IPFW_VNET_ORDER (IPFW_MODEVENT_ORDER + 2) /* Later still. */ DECLARE_MODULE(ipfw, ipfwmod, IPFW_SI_SUB_FIREWALL, IPFW_MODEVENT_ORDER); FEATURE(ipfw_ctl3, "ipfw new sockopt calls"); MODULE_VERSION(ipfw, 3); /* should declare some dependencies here */ /* * Starting up. Done in order after ipfwmod() has been called. * VNET_SYSINIT is also called for each existing vnet and each new vnet. */ SYSINIT(ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER, ipfw_init, NULL); VNET_SYSINIT(vnet_ipfw_init, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER, vnet_ipfw_init, NULL); /* * Closing up shop. These are done in REVERSE ORDER, but still * after ipfwmod() has been called. Not called on reboot. * VNET_SYSUNINIT is also called for each exiting vnet as it exits. * or when the module is unloaded. */ SYSUNINIT(ipfw_destroy, IPFW_SI_SUB_FIREWALL, IPFW_MODULE_ORDER, ipfw_destroy, NULL); VNET_SYSUNINIT(vnet_ipfw_uninit, IPFW_SI_SUB_FIREWALL, IPFW_VNET_ORDER, vnet_ipfw_uninit, NULL); /* end of file */ Index: head/sys/netpfil/ipfw/ip_fw_bpf.c =================================================================== --- head/sys/netpfil/ipfw/ip_fw_bpf.c (revision 336675) +++ head/sys/netpfil/ipfw/ip_fw_bpf.c (revision 336676) @@ -1,209 +1,209 @@ /*- * Copyright (c) 2016 Yandex LLC * Copyright (c) 2016 Andrey V. Elsukov * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include -static VNET_DEFINE(struct ifnet *, log_if); -static VNET_DEFINE(struct ifnet *, pflog_if); -static VNET_DEFINE(struct if_clone *, ipfw_cloner); -static VNET_DEFINE(struct if_clone *, ipfwlog_cloner); +VNET_DEFINE_STATIC(struct ifnet *, log_if); +VNET_DEFINE_STATIC(struct ifnet *, pflog_if); +VNET_DEFINE_STATIC(struct if_clone *, ipfw_cloner); +VNET_DEFINE_STATIC(struct if_clone *, ipfwlog_cloner); #define V_ipfw_cloner VNET(ipfw_cloner) #define V_ipfwlog_cloner VNET(ipfwlog_cloner) #define V_log_if VNET(log_if) #define V_pflog_if VNET(pflog_if) static struct rmlock log_if_lock; #define LOGIF_LOCK_INIT(x) rm_init(&log_if_lock, "ipfw log_if lock") #define LOGIF_LOCK_DESTROY(x) rm_destroy(&log_if_lock) #define LOGIF_RLOCK_TRACKER struct rm_priotracker _log_tracker #define LOGIF_RLOCK(x) rm_rlock(&log_if_lock, &_log_tracker) #define LOGIF_RUNLOCK(x) rm_runlock(&log_if_lock, &_log_tracker) #define LOGIF_WLOCK(x) rm_wlock(&log_if_lock) #define LOGIF_WUNLOCK(x) rm_wunlock(&log_if_lock) static const char ipfwname[] = "ipfw"; static const char ipfwlogname[] = "ipfwlog"; static int ipfw_bpf_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr) { return (EINVAL); } static int ipfw_bpf_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *ro) { if (m != NULL) FREE_PKT(m); return (0); } static void ipfw_clone_destroy(struct ifnet *ifp) { LOGIF_WLOCK(); if (ifp->if_hdrlen == ETHER_HDR_LEN) V_log_if = NULL; else V_pflog_if = NULL; LOGIF_WUNLOCK(); bpfdetach(ifp); if_detach(ifp); if_free(ifp); } static int ipfw_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ifnet *ifp; ifp = if_alloc(IFT_PFLOG); if (ifp == NULL) return (ENOSPC); if_initname(ifp, ipfwname, unit); ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_mtu = 65536; ifp->if_ioctl = ipfw_bpf_ioctl; ifp->if_output = ipfw_bpf_output; ifp->if_hdrlen = ETHER_HDR_LEN; if_attach(ifp); bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); LOGIF_WLOCK(); if (V_log_if != NULL) { LOGIF_WUNLOCK(); bpfdetach(ifp); if_detach(ifp); if_free(ifp); return (EEXIST); } V_log_if = ifp; LOGIF_WUNLOCK(); return (0); } static int ipfwlog_clone_create(struct if_clone *ifc, int unit, caddr_t params) { struct ifnet *ifp; ifp = if_alloc(IFT_PFLOG); if (ifp == NULL) return (ENOSPC); if_initname(ifp, ipfwlogname, unit); ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_mtu = 65536; ifp->if_ioctl = ipfw_bpf_ioctl; ifp->if_output = ipfw_bpf_output; ifp->if_hdrlen = PFLOG_HDRLEN; if_attach(ifp); bpfattach(ifp, DLT_PFLOG, PFLOG_HDRLEN); LOGIF_WLOCK(); if (V_pflog_if != NULL) { LOGIF_WUNLOCK(); bpfdetach(ifp); if_detach(ifp); if_free(ifp); return (EEXIST); } V_pflog_if = ifp; LOGIF_WUNLOCK(); return (0); } void ipfw_bpf_mtap2(void *data, u_int dlen, struct mbuf *m) { LOGIF_RLOCK_TRACKER; LOGIF_RLOCK(); if (dlen == ETHER_HDR_LEN) { if (V_log_if == NULL) { LOGIF_RUNLOCK(); return; } BPF_MTAP2(V_log_if, data, dlen, m); } else if (dlen == PFLOG_HDRLEN) { if (V_pflog_if == NULL) { LOGIF_RUNLOCK(); return; } BPF_MTAP2(V_pflog_if, data, dlen, m); } LOGIF_RUNLOCK(); } void ipfw_bpf_init(int first) { if (first) { LOGIF_LOCK_INIT(); V_log_if = NULL; V_pflog_if = NULL; } V_ipfw_cloner = if_clone_simple(ipfwname, ipfw_clone_create, ipfw_clone_destroy, 0); V_ipfwlog_cloner = if_clone_simple(ipfwlogname, ipfwlog_clone_create, ipfw_clone_destroy, 0); } void ipfw_bpf_uninit(int last) { if_clone_detach(V_ipfw_cloner); if_clone_detach(V_ipfwlog_cloner); if (last) LOGIF_LOCK_DESTROY(); } Index: head/sys/netpfil/ipfw/ip_fw_dynamic.c =================================================================== --- head/sys/netpfil/ipfw/ip_fw_dynamic.c (revision 336675) +++ head/sys/netpfil/ipfw/ip_fw_dynamic.c (revision 336676) @@ -1,3091 +1,3091 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2017-2018 Yandex LLC * Copyright (c) 2017-2018 Andrey V. Elsukov * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipfw.h" #ifndef INET #error IPFIREWALL requires INET. #endif /* INET */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* IN6_ARE_ADDR_EQUAL */ #ifdef INET6 #include #include #include #endif #include #include /* XXX for in_cksum */ #ifdef MAC #include #endif /* * Description of dynamic states. * * Dynamic states are stored in lists accessed through a hash tables * whose size is curr_dyn_buckets. This value can be modified through * the sysctl variable dyn_buckets. * * Currently there are four tables: dyn_ipv4, dyn_ipv6, dyn_ipv4_parent, * and dyn_ipv6_parent. * * When a packet is received, its address fields hashed, then matched * against the entries in the corresponding list by addr_type. * Dynamic states can be used for different purposes: * + stateful rules; * + enforcing limits on the number of sessions; * + in-kernel NAT (not implemented yet) * * The lifetime of dynamic states is regulated by dyn_*_lifetime, * measured in seconds and depending on the flags. * * The total number of dynamic states is equal to UMA zone items count. * The max number of dynamic states is dyn_max. When we reach * the maximum number of rules we do not create anymore. This is * done to avoid consuming too much memory, but also too much * time when searching on each packet (ideally, we should try instead * to put a limit on the length of the list on each bucket...). * * Each state holds a pointer to the parent ipfw rule so we know what * action to perform. Dynamic rules are removed when the parent rule is * deleted. * * There are some limitations with dynamic rules -- we do not * obey the 'randomized match', and we do not do multiple * passes through the firewall. XXX check the latter!!! */ /* By default use jenkins hash function */ #define IPFIREWALL_JENKINSHASH #define DYN_COUNTER_INC(d, dir, pktlen) do { \ (d)->pcnt_ ## dir++; \ (d)->bcnt_ ## dir += pktlen; \ } while (0) struct dyn_data { void *parent; /* pointer to parent rule */ uint32_t chain_id; /* cached ruleset id */ uint32_t f_pos; /* cached rule index */ uint32_t hashval; /* hash value used for hash resize */ uint16_t fibnum; /* fib used to send keepalives */ uint8_t _pad[3]; uint8_t set; /* parent rule set number */ uint16_t rulenum; /* parent rule number */ uint32_t ruleid; /* parent rule id */ uint32_t state; /* TCP session state and flags */ uint32_t ack_fwd; /* most recent ACKs in forward */ uint32_t ack_rev; /* and reverse direction (used */ /* to generate keepalives) */ uint32_t sync; /* synchronization time */ uint32_t expire; /* expire time */ uint64_t pcnt_fwd; /* bytes counter in forward */ uint64_t bcnt_fwd; /* packets counter in forward */ uint64_t pcnt_rev; /* bytes counter in reverse */ uint64_t bcnt_rev; /* packets counter in reverse */ }; #define DPARENT_COUNT_DEC(p) do { \ MPASS(p->count > 0); \ ck_pr_dec_32(&(p)->count); \ } while (0) #define DPARENT_COUNT_INC(p) ck_pr_inc_32(&(p)->count) #define DPARENT_COUNT(p) ck_pr_load_32(&(p)->count) struct dyn_parent { void *parent; /* pointer to parent rule */ uint32_t count; /* number of linked states */ uint8_t _pad; uint8_t set; /* parent rule set number */ uint16_t rulenum; /* parent rule number */ uint32_t ruleid; /* parent rule id */ uint32_t hashval; /* hash value used for hash resize */ uint32_t expire; /* expire time */ }; struct dyn_ipv4_state { uint8_t type; /* State type */ uint8_t proto; /* UL Protocol */ uint16_t kidx; /* named object index */ uint16_t sport, dport; /* ULP source and destination ports */ in_addr_t src, dst; /* IPv4 source and destination */ union { struct dyn_data *data; struct dyn_parent *limit; }; CK_SLIST_ENTRY(dyn_ipv4_state) entry; SLIST_ENTRY(dyn_ipv4_state) expired; }; CK_SLIST_HEAD(dyn_ipv4ck_slist, dyn_ipv4_state); -static VNET_DEFINE(struct dyn_ipv4ck_slist *, dyn_ipv4); -static VNET_DEFINE(struct dyn_ipv4ck_slist *, dyn_ipv4_parent); +VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4); +VNET_DEFINE_STATIC(struct dyn_ipv4ck_slist *, dyn_ipv4_parent); SLIST_HEAD(dyn_ipv4_slist, dyn_ipv4_state); -static VNET_DEFINE(struct dyn_ipv4_slist, dyn_expired_ipv4); +VNET_DEFINE_STATIC(struct dyn_ipv4_slist, dyn_expired_ipv4); #define V_dyn_ipv4 VNET(dyn_ipv4) #define V_dyn_ipv4_parent VNET(dyn_ipv4_parent) #define V_dyn_expired_ipv4 VNET(dyn_expired_ipv4) #ifdef INET6 struct dyn_ipv6_state { uint8_t type; /* State type */ uint8_t proto; /* UL Protocol */ uint16_t kidx; /* named object index */ uint16_t sport, dport; /* ULP source and destination ports */ struct in6_addr src, dst; /* IPv6 source and destination */ uint32_t zoneid; /* IPv6 scope zone id */ union { struct dyn_data *data; struct dyn_parent *limit; }; CK_SLIST_ENTRY(dyn_ipv6_state) entry; SLIST_ENTRY(dyn_ipv6_state) expired; }; CK_SLIST_HEAD(dyn_ipv6ck_slist, dyn_ipv6_state); -static VNET_DEFINE(struct dyn_ipv6ck_slist *, dyn_ipv6); -static VNET_DEFINE(struct dyn_ipv6ck_slist *, dyn_ipv6_parent); +VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6); +VNET_DEFINE_STATIC(struct dyn_ipv6ck_slist *, dyn_ipv6_parent); SLIST_HEAD(dyn_ipv6_slist, dyn_ipv6_state); -static VNET_DEFINE(struct dyn_ipv6_slist, dyn_expired_ipv6); +VNET_DEFINE_STATIC(struct dyn_ipv6_slist, dyn_expired_ipv6); #define V_dyn_ipv6 VNET(dyn_ipv6) #define V_dyn_ipv6_parent VNET(dyn_ipv6_parent) #define V_dyn_expired_ipv6 VNET(dyn_expired_ipv6) #endif /* INET6 */ /* * Per-CPU pointer indicates that specified state is currently in use * and must not be reclaimed by expiration callout. */ static void **dyn_hp_cache; DPCPU_DEFINE_STATIC(void *, dyn_hp); #define DYNSTATE_GET(cpu) ck_pr_load_ptr(DPCPU_ID_PTR((cpu), dyn_hp)) #define DYNSTATE_PROTECT(v) ck_pr_store_ptr(DPCPU_PTR(dyn_hp), (v)) #define DYNSTATE_RELEASE() DYNSTATE_PROTECT(NULL) #define DYNSTATE_CRITICAL_ENTER() critical_enter() #define DYNSTATE_CRITICAL_EXIT() do { \ DYNSTATE_RELEASE(); \ critical_exit(); \ } while (0); /* * We keep two version numbers, one is updated when new entry added to * the list. Second is updated when an entry deleted from the list. * Versions are updated under bucket lock. * * Bucket "add" version number is used to know, that in the time between * state lookup (i.e. ipfw_dyn_lookup_state()) and the followed state * creation (i.e. ipfw_dyn_install_state()) another concurrent thread did * not install some state in this bucket. Using this info we can avoid * additional state lookup, because we are sure that we will not install * the state twice. * * Also doing the tracking of bucket "del" version during lookup we can * be sure, that state entry was not unlinked and freed in time between * we read the state pointer and protect it with hazard pointer. * * An entry unlinked from CK list keeps unchanged until it is freed. * Unlinked entries are linked into expired lists using "expired" field. */ /* * dyn_expire_lock is used to protect access to dyn_expired_xxx lists. * dyn_bucket_lock is used to get write access to lists in specific bucket. * Currently one dyn_bucket_lock is used for all ipv4, ipv4_parent, ipv6, * and ipv6_parent lists. */ -static VNET_DEFINE(struct mtx, dyn_expire_lock); -static VNET_DEFINE(struct mtx *, dyn_bucket_lock); +VNET_DEFINE_STATIC(struct mtx, dyn_expire_lock); +VNET_DEFINE_STATIC(struct mtx *, dyn_bucket_lock); #define V_dyn_expire_lock VNET(dyn_expire_lock) #define V_dyn_bucket_lock VNET(dyn_bucket_lock) /* * Bucket's add/delete generation versions. */ -static VNET_DEFINE(uint32_t *, dyn_ipv4_add); -static VNET_DEFINE(uint32_t *, dyn_ipv4_del); -static VNET_DEFINE(uint32_t *, dyn_ipv4_parent_add); -static VNET_DEFINE(uint32_t *, dyn_ipv4_parent_del); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_add); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_del); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_add); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv4_parent_del); #define V_dyn_ipv4_add VNET(dyn_ipv4_add) #define V_dyn_ipv4_del VNET(dyn_ipv4_del) #define V_dyn_ipv4_parent_add VNET(dyn_ipv4_parent_add) #define V_dyn_ipv4_parent_del VNET(dyn_ipv4_parent_del) #ifdef INET6 -static VNET_DEFINE(uint32_t *, dyn_ipv6_add); -static VNET_DEFINE(uint32_t *, dyn_ipv6_del); -static VNET_DEFINE(uint32_t *, dyn_ipv6_parent_add); -static VNET_DEFINE(uint32_t *, dyn_ipv6_parent_del); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_add); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_del); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_add); +VNET_DEFINE_STATIC(uint32_t *, dyn_ipv6_parent_del); #define V_dyn_ipv6_add VNET(dyn_ipv6_add) #define V_dyn_ipv6_del VNET(dyn_ipv6_del) #define V_dyn_ipv6_parent_add VNET(dyn_ipv6_parent_add) #define V_dyn_ipv6_parent_del VNET(dyn_ipv6_parent_del) #endif /* INET6 */ #define DYN_BUCKET(h, b) ((h) & (b - 1)) #define DYN_BUCKET_VERSION(b, v) ck_pr_load_32(&V_dyn_ ## v[(b)]) #define DYN_BUCKET_VERSION_BUMP(b, v) ck_pr_inc_32(&V_dyn_ ## v[(b)]) #define DYN_BUCKET_LOCK_INIT(lock, b) \ mtx_init(&lock[(b)], "IPFW dynamic bucket", NULL, MTX_DEF) #define DYN_BUCKET_LOCK_DESTROY(lock, b) mtx_destroy(&lock[(b)]) #define DYN_BUCKET_LOCK(b) mtx_lock(&V_dyn_bucket_lock[(b)]) #define DYN_BUCKET_UNLOCK(b) mtx_unlock(&V_dyn_bucket_lock[(b)]) #define DYN_BUCKET_ASSERT(b) mtx_assert(&V_dyn_bucket_lock[(b)], MA_OWNED) #define DYN_EXPIRED_LOCK_INIT() \ mtx_init(&V_dyn_expire_lock, "IPFW expired states list", NULL, MTX_DEF) #define DYN_EXPIRED_LOCK_DESTROY() mtx_destroy(&V_dyn_expire_lock) #define DYN_EXPIRED_LOCK() mtx_lock(&V_dyn_expire_lock) #define DYN_EXPIRED_UNLOCK() mtx_unlock(&V_dyn_expire_lock) -static VNET_DEFINE(uint32_t, dyn_buckets_max); -static VNET_DEFINE(uint32_t, curr_dyn_buckets); -static VNET_DEFINE(struct callout, dyn_timeout); +VNET_DEFINE_STATIC(uint32_t, dyn_buckets_max); +VNET_DEFINE_STATIC(uint32_t, curr_dyn_buckets); +VNET_DEFINE_STATIC(struct callout, dyn_timeout); #define V_dyn_buckets_max VNET(dyn_buckets_max) #define V_curr_dyn_buckets VNET(curr_dyn_buckets) #define V_dyn_timeout VNET(dyn_timeout) /* Maximum length of states chain in a bucket */ -static VNET_DEFINE(uint32_t, curr_max_length); +VNET_DEFINE_STATIC(uint32_t, curr_max_length); #define V_curr_max_length VNET(curr_max_length) -static VNET_DEFINE(uint32_t, dyn_keep_states); +VNET_DEFINE_STATIC(uint32_t, dyn_keep_states); #define V_dyn_keep_states VNET(dyn_keep_states) -static VNET_DEFINE(uma_zone_t, dyn_data_zone); -static VNET_DEFINE(uma_zone_t, dyn_parent_zone); -static VNET_DEFINE(uma_zone_t, dyn_ipv4_zone); +VNET_DEFINE_STATIC(uma_zone_t, dyn_data_zone); +VNET_DEFINE_STATIC(uma_zone_t, dyn_parent_zone); +VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv4_zone); #ifdef INET6 -static VNET_DEFINE(uma_zone_t, dyn_ipv6_zone); +VNET_DEFINE_STATIC(uma_zone_t, dyn_ipv6_zone); #define V_dyn_ipv6_zone VNET(dyn_ipv6_zone) #endif /* INET6 */ #define V_dyn_data_zone VNET(dyn_data_zone) #define V_dyn_parent_zone VNET(dyn_parent_zone) #define V_dyn_ipv4_zone VNET(dyn_ipv4_zone) /* * Timeouts for various events in handing dynamic rules. */ -static VNET_DEFINE(uint32_t, dyn_ack_lifetime); -static VNET_DEFINE(uint32_t, dyn_syn_lifetime); -static VNET_DEFINE(uint32_t, dyn_fin_lifetime); -static VNET_DEFINE(uint32_t, dyn_rst_lifetime); -static VNET_DEFINE(uint32_t, dyn_udp_lifetime); -static VNET_DEFINE(uint32_t, dyn_short_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_ack_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_syn_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_fin_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_rst_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_udp_lifetime); +VNET_DEFINE_STATIC(uint32_t, dyn_short_lifetime); #define V_dyn_ack_lifetime VNET(dyn_ack_lifetime) #define V_dyn_syn_lifetime VNET(dyn_syn_lifetime) #define V_dyn_fin_lifetime VNET(dyn_fin_lifetime) #define V_dyn_rst_lifetime VNET(dyn_rst_lifetime) #define V_dyn_udp_lifetime VNET(dyn_udp_lifetime) #define V_dyn_short_lifetime VNET(dyn_short_lifetime) /* * Keepalives are sent if dyn_keepalive is set. They are sent every * dyn_keepalive_period seconds, in the last dyn_keepalive_interval * seconds of lifetime of a rule. * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower * than dyn_keepalive_period. */ #define DYN_KEEPALIVE_MAXQ 512 -static VNET_DEFINE(uint32_t, dyn_keepalive_interval); -static VNET_DEFINE(uint32_t, dyn_keepalive_period); -static VNET_DEFINE(uint32_t, dyn_keepalive); -static VNET_DEFINE(time_t, dyn_keepalive_last); +VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_interval); +VNET_DEFINE_STATIC(uint32_t, dyn_keepalive_period); +VNET_DEFINE_STATIC(uint32_t, dyn_keepalive); +VNET_DEFINE_STATIC(time_t, dyn_keepalive_last); #define V_dyn_keepalive_interval VNET(dyn_keepalive_interval) #define V_dyn_keepalive_period VNET(dyn_keepalive_period) #define V_dyn_keepalive VNET(dyn_keepalive) #define V_dyn_keepalive_last VNET(dyn_keepalive_last) -static VNET_DEFINE(uint32_t, dyn_max); /* max # of dynamic states */ -static VNET_DEFINE(uint32_t, dyn_count); /* number of states */ -static VNET_DEFINE(uint32_t, dyn_parent_max); /* max # of parent states */ -static VNET_DEFINE(uint32_t, dyn_parent_count); /* number of parent states */ +VNET_DEFINE_STATIC(uint32_t, dyn_max); /* max # of dynamic states */ +VNET_DEFINE_STATIC(uint32_t, dyn_count); /* number of states */ +VNET_DEFINE_STATIC(uint32_t, dyn_parent_max); /* max # of parent states */ +VNET_DEFINE_STATIC(uint32_t, dyn_parent_count); /* number of parent states */ #define V_dyn_max VNET(dyn_max) #define V_dyn_count VNET(dyn_count) #define V_dyn_parent_max VNET(dyn_parent_max) #define V_dyn_parent_count VNET(dyn_parent_count) #define DYN_COUNT_DEC(name) do { \ MPASS((V_ ## name) > 0); \ ck_pr_dec_32(&(V_ ## name)); \ } while (0) #define DYN_COUNT_INC(name) ck_pr_inc_32(&(V_ ## name)) #define DYN_COUNT(name) ck_pr_load_32(&(V_ ## name)) static time_t last_log; /* Log ratelimiting */ /* * Get/set maximum number of dynamic states in given VNET instance. */ static int sysctl_dyn_max(SYSCTL_HANDLER_ARGS) { uint32_t nstates; int error; nstates = V_dyn_max; error = sysctl_handle_32(oidp, &nstates, 0, req); /* Read operation or some error */ if ((error != 0) || (req->newptr == NULL)) return (error); V_dyn_max = nstates; uma_zone_set_max(V_dyn_data_zone, V_dyn_max); return (0); } static int sysctl_dyn_parent_max(SYSCTL_HANDLER_ARGS) { uint32_t nstates; int error; nstates = V_dyn_parent_max; error = sysctl_handle_32(oidp, &nstates, 0, req); /* Read operation or some error */ if ((error != 0) || (req->newptr == NULL)) return (error); V_dyn_parent_max = nstates; uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max); return (0); } static int sysctl_dyn_buckets(SYSCTL_HANDLER_ARGS) { uint32_t nbuckets; int error; nbuckets = V_dyn_buckets_max; error = sysctl_handle_32(oidp, &nbuckets, 0, req); /* Read operation or some error */ if ((error != 0) || (req->newptr == NULL)) return (error); if (nbuckets > 256) V_dyn_buckets_max = 1 << fls(nbuckets - 1); else return (EINVAL); return (0); } SYSCTL_DECL(_net_inet_ip_fw); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_count), 0, "Current number of dynamic states."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_parent_count, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(dyn_parent_count), 0, "Current number of parent states. "); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_dyn_buckets), 0, "Current number of buckets for states hash table."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, curr_max_length, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(curr_max_length), 0, "Current maximum length of states chains in hash buckets."); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW, 0, 0, sysctl_dyn_buckets, "IU", "Max number of buckets for dynamic states hash table."); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW, 0, 0, sysctl_dyn_max, "IU", "Max number of dynamic states."); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, dyn_parent_max, CTLFLAG_VNET | CTLTYPE_U32 | CTLFLAG_RW, 0, 0, sysctl_dyn_parent_max, "IU", "Max number of parent dynamic states."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_ack_lifetime), 0, "Lifetime of dynamic states for TCP ACK."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_syn_lifetime), 0, "Lifetime of dynamic states for TCP SYN."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_fin_lifetime), 0, "Lifetime of dynamic states for TCP FIN."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_rst_lifetime), 0, "Lifetime of dynamic states for TCP RST."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_udp_lifetime), 0, "Lifetime of dynamic states for UDP."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_short_lifetime), 0, "Lifetime of dynamic states for other situations."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keepalive), 0, "Enable keepalives for dynamic states."); SYSCTL_U32(_net_inet_ip_fw, OID_AUTO, dyn_keep_states, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(dyn_keep_states), 0, "Do not flush dynamic states on rule deletion"); #ifdef IPFIREWALL_DYNDEBUG #define DYN_DEBUG(fmt, ...) do { \ printf("%s: " fmt "\n", __func__, __VA_ARGS__); \ } while (0) #else #define DYN_DEBUG(fmt, ...) #endif /* !IPFIREWALL_DYNDEBUG */ #ifdef INET6 /* Functions to work with IPv6 states */ static struct dyn_ipv6_state *dyn_lookup_ipv6_state( const struct ipfw_flow_id *, uint32_t, const void *, struct ipfw_dyn_info *, int); static int dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *, uint32_t, const void *, int, uint32_t, uint16_t); static struct dyn_ipv6_state *dyn_alloc_ipv6_state( const struct ipfw_flow_id *, uint32_t, uint16_t, uint8_t); static int dyn_add_ipv6_state(void *, uint32_t, uint16_t, uint8_t, const struct ipfw_flow_id *, uint32_t, const void *, int, uint32_t, struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t); static void dyn_export_ipv6_state(const struct dyn_ipv6_state *, ipfw_dyn_rule *); static uint32_t dyn_getscopeid(const struct ip_fw_args *); static void dyn_make_keepalive_ipv6(struct mbuf *, const struct in6_addr *, const struct in6_addr *, uint32_t, uint32_t, uint32_t, uint16_t, uint16_t); static void dyn_enqueue_keepalive_ipv6(struct mbufq *, const struct dyn_ipv6_state *); static void dyn_send_keepalive_ipv6(struct ip_fw_chain *); static struct dyn_ipv6_state *dyn_lookup_ipv6_parent( const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t, uint32_t); static struct dyn_ipv6_state *dyn_lookup_ipv6_parent_locked( const struct ipfw_flow_id *, uint32_t, const void *, uint32_t, uint16_t, uint32_t); static struct dyn_ipv6_state *dyn_add_ipv6_parent(void *, uint32_t, uint16_t, uint8_t, const struct ipfw_flow_id *, uint32_t, uint32_t, uint32_t, uint16_t); #endif /* INET6 */ /* Functions to work with limit states */ static void *dyn_get_parent_state(const struct ipfw_flow_id *, uint32_t, struct ip_fw *, uint32_t, uint32_t, uint16_t); static struct dyn_ipv4_state *dyn_lookup_ipv4_parent( const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t); static struct dyn_ipv4_state *dyn_lookup_ipv4_parent_locked( const struct ipfw_flow_id *, const void *, uint32_t, uint16_t, uint32_t); static struct dyn_parent *dyn_alloc_parent(void *, uint32_t, uint16_t, uint8_t, uint32_t); static struct dyn_ipv4_state *dyn_add_ipv4_parent(void *, uint32_t, uint16_t, uint8_t, const struct ipfw_flow_id *, uint32_t, uint32_t, uint16_t); static void dyn_tick(void *); static void dyn_expire_states(struct ip_fw_chain *, ipfw_range_tlv *); static void dyn_free_states(struct ip_fw_chain *); static void dyn_export_parent(const struct dyn_parent *, uint16_t, ipfw_dyn_rule *); static void dyn_export_data(const struct dyn_data *, uint16_t, uint8_t, ipfw_dyn_rule *); static uint32_t dyn_update_tcp_state(struct dyn_data *, const struct ipfw_flow_id *, const struct tcphdr *, int); static void dyn_update_proto_state(struct dyn_data *, const struct ipfw_flow_id *, const void *, int, int); /* Functions to work with IPv4 states */ struct dyn_ipv4_state *dyn_lookup_ipv4_state(const struct ipfw_flow_id *, const void *, struct ipfw_dyn_info *, int); static int dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *, const void *, int, uint32_t, uint16_t); static struct dyn_ipv4_state *dyn_alloc_ipv4_state( const struct ipfw_flow_id *, uint16_t, uint8_t); static int dyn_add_ipv4_state(void *, uint32_t, uint16_t, uint8_t, const struct ipfw_flow_id *, const void *, int, uint32_t, struct ipfw_dyn_info *, uint16_t, uint16_t, uint8_t); static void dyn_export_ipv4_state(const struct dyn_ipv4_state *, ipfw_dyn_rule *); /* * Named states support. */ static char *default_state_name = "default"; struct dyn_state_obj { struct named_object no; char name[64]; }; #define DYN_STATE_OBJ(ch, cmd) \ ((struct dyn_state_obj *)SRV_OBJECT(ch, (cmd)->arg1)) /* * Classifier callback. * Return 0 if opcode contains object that should be referenced * or rewritten. */ static int dyn_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) { DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); /* Don't rewrite "check-state any" */ if (cmd->arg1 == 0 && cmd->opcode == O_CHECK_STATE) return (1); *puidx = cmd->arg1; *ptype = 0; return (0); } static void dyn_update(ipfw_insn *cmd, uint16_t idx) { cmd->arg1 = idx; DYN_DEBUG("opcode %d, arg1 %d", cmd->opcode, cmd->arg1); } static int dyn_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, struct named_object **pno) { ipfw_obj_ntlv *ntlv; const char *name; DYN_DEBUG("uidx %d", ti->uidx); if (ti->uidx != 0) { if (ti->tlvs == NULL) return (EINVAL); /* Search ntlv in the buffer provided by user */ ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, IPFW_TLV_STATE_NAME); if (ntlv == NULL) return (EINVAL); name = ntlv->name; } else name = default_state_name; /* * Search named object with corresponding name. * Since states objects are global - ignore the set value * and use zero instead. */ *pno = ipfw_objhash_lookup_name_type(CHAIN_TO_SRV(ch), 0, IPFW_TLV_STATE_NAME, name); /* * We always return success here. * The caller will check *pno and mark object as unresolved, * then it will automatically create "default" object. */ return (0); } static struct named_object * dyn_findbykidx(struct ip_fw_chain *ch, uint16_t idx) { DYN_DEBUG("kidx %d", idx); return (ipfw_objhash_lookup_kidx(CHAIN_TO_SRV(ch), idx)); } static int dyn_create(struct ip_fw_chain *ch, struct tid_info *ti, uint16_t *pkidx) { struct namedobj_instance *ni; struct dyn_state_obj *obj; struct named_object *no; ipfw_obj_ntlv *ntlv; char *name; DYN_DEBUG("uidx %d", ti->uidx); if (ti->uidx != 0) { if (ti->tlvs == NULL) return (EINVAL); ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, IPFW_TLV_STATE_NAME); if (ntlv == NULL) return (EINVAL); name = ntlv->name; } else name = default_state_name; ni = CHAIN_TO_SRV(ch); obj = malloc(sizeof(*obj), M_IPFW, M_WAITOK | M_ZERO); obj->no.name = obj->name; obj->no.etlv = IPFW_TLV_STATE_NAME; strlcpy(obj->name, name, sizeof(obj->name)); IPFW_UH_WLOCK(ch); no = ipfw_objhash_lookup_name_type(ni, 0, IPFW_TLV_STATE_NAME, name); if (no != NULL) { /* * Object is already created. * Just return its kidx and bump refcount. */ *pkidx = no->kidx; no->refcnt++; IPFW_UH_WUNLOCK(ch); free(obj, M_IPFW); DYN_DEBUG("\tfound kidx %d", *pkidx); return (0); } if (ipfw_objhash_alloc_idx(ni, &obj->no.kidx) != 0) { DYN_DEBUG("\talloc_idx failed for %s", name); IPFW_UH_WUNLOCK(ch); free(obj, M_IPFW); return (ENOSPC); } ipfw_objhash_add(ni, &obj->no); SRV_OBJECT(ch, obj->no.kidx) = obj; obj->no.refcnt++; *pkidx = obj->no.kidx; IPFW_UH_WUNLOCK(ch); DYN_DEBUG("\tcreated kidx %d", *pkidx); return (0); } static void dyn_destroy(struct ip_fw_chain *ch, struct named_object *no) { struct dyn_state_obj *obj; IPFW_UH_WLOCK_ASSERT(ch); KASSERT(no->refcnt == 1, ("Destroying object '%s' (type %u, idx %u) with refcnt %u", no->name, no->etlv, no->kidx, no->refcnt)); DYN_DEBUG("kidx %d", no->kidx); obj = SRV_OBJECT(ch, no->kidx); SRV_OBJECT(ch, no->kidx) = NULL; ipfw_objhash_del(CHAIN_TO_SRV(ch), no); ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), no->kidx); free(obj, M_IPFW); } static struct opcode_obj_rewrite dyn_opcodes[] = { { O_KEEP_STATE, IPFW_TLV_STATE_NAME, dyn_classify, dyn_update, dyn_findbyname, dyn_findbykidx, dyn_create, dyn_destroy }, { O_CHECK_STATE, IPFW_TLV_STATE_NAME, dyn_classify, dyn_update, dyn_findbyname, dyn_findbykidx, dyn_create, dyn_destroy }, { O_PROBE_STATE, IPFW_TLV_STATE_NAME, dyn_classify, dyn_update, dyn_findbyname, dyn_findbykidx, dyn_create, dyn_destroy }, { O_LIMIT, IPFW_TLV_STATE_NAME, dyn_classify, dyn_update, dyn_findbyname, dyn_findbykidx, dyn_create, dyn_destroy }, }; /* * IMPORTANT: the hash function for dynamic rules must be commutative * in source and destination (ip,port), because rules are bidirectional * and we want to find both in the same bucket. */ #ifndef IPFIREWALL_JENKINSHASH static __inline uint32_t hash_packet(const struct ipfw_flow_id *id) { uint32_t i; #ifdef INET6 if (IS_IP6_FLOW_ID(id)) i = ntohl((id->dst_ip6.__u6_addr.__u6_addr32[2]) ^ (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^ (id->src_ip6.__u6_addr.__u6_addr32[2]) ^ (id->src_ip6.__u6_addr.__u6_addr32[3])); else #endif /* INET6 */ i = (id->dst_ip) ^ (id->src_ip); i ^= (id->dst_port) ^ (id->src_port); return (i); } static __inline uint32_t hash_parent(const struct ipfw_flow_id *id, const void *rule) { return (hash_packet(id) ^ ((uintptr_t)rule)); } #else /* IPFIREWALL_JENKINSHASH */ -static VNET_DEFINE(uint32_t, dyn_hashseed); +VNET_DEFINE_STATIC(uint32_t, dyn_hashseed); #define V_dyn_hashseed VNET(dyn_hashseed) static __inline int addrcmp4(const struct ipfw_flow_id *id) { if (id->src_ip < id->dst_ip) return (0); if (id->src_ip > id->dst_ip) return (1); if (id->src_port <= id->dst_port) return (0); return (1); } #ifdef INET6 static __inline int addrcmp6(const struct ipfw_flow_id *id) { int ret; ret = memcmp(&id->src_ip6, &id->dst_ip6, sizeof(struct in6_addr)); if (ret < 0) return (0); if (ret > 0) return (1); if (id->src_port <= id->dst_port) return (0); return (1); } static __inline uint32_t hash_packet6(const struct ipfw_flow_id *id) { struct tuple6 { struct in6_addr addr[2]; uint16_t port[2]; } t6; if (addrcmp6(id) == 0) { t6.addr[0] = id->src_ip6; t6.addr[1] = id->dst_ip6; t6.port[0] = id->src_port; t6.port[1] = id->dst_port; } else { t6.addr[0] = id->dst_ip6; t6.addr[1] = id->src_ip6; t6.port[0] = id->dst_port; t6.port[1] = id->src_port; } return (jenkins_hash32((const uint32_t *)&t6, sizeof(t6) / sizeof(uint32_t), V_dyn_hashseed)); } #endif static __inline uint32_t hash_packet(const struct ipfw_flow_id *id) { struct tuple4 { in_addr_t addr[2]; uint16_t port[2]; } t4; if (IS_IP4_FLOW_ID(id)) { /* All fields are in host byte order */ if (addrcmp4(id) == 0) { t4.addr[0] = id->src_ip; t4.addr[1] = id->dst_ip; t4.port[0] = id->src_port; t4.port[1] = id->dst_port; } else { t4.addr[0] = id->dst_ip; t4.addr[1] = id->src_ip; t4.port[0] = id->dst_port; t4.port[1] = id->src_port; } return (jenkins_hash32((const uint32_t *)&t4, sizeof(t4) / sizeof(uint32_t), V_dyn_hashseed)); } else #ifdef INET6 if (IS_IP6_FLOW_ID(id)) return (hash_packet6(id)); #endif return (0); } static __inline uint32_t hash_parent(const struct ipfw_flow_id *id, const void *rule) { return (jenkins_hash32((const uint32_t *)&rule, sizeof(rule) / sizeof(uint32_t), hash_packet(id))); } #endif /* IPFIREWALL_JENKINSHASH */ /* * Print customizable flow id description via log(9) facility. */ static void print_dyn_rule_flags(const struct ipfw_flow_id *id, int dyn_type, int log_flags, char *prefix, char *postfix) { struct in_addr da; #ifdef INET6 char src[INET6_ADDRSTRLEN], dst[INET6_ADDRSTRLEN]; #else char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; #endif #ifdef INET6 if (IS_IP6_FLOW_ID(id)) { ip6_sprintf(src, &id->src_ip6); ip6_sprintf(dst, &id->dst_ip6); } else #endif { da.s_addr = htonl(id->src_ip); inet_ntop(AF_INET, &da, src, sizeof(src)); da.s_addr = htonl(id->dst_ip); inet_ntop(AF_INET, &da, dst, sizeof(dst)); } log(log_flags, "ipfw: %s type %d %s %d -> %s %d, %d %s\n", prefix, dyn_type, src, id->src_port, dst, id->dst_port, V_dyn_count, postfix); } #define print_dyn_rule(id, dtype, prefix, postfix) \ print_dyn_rule_flags(id, dtype, LOG_DEBUG, prefix, postfix) #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0) #define TIME_LE(a,b) ((int)((a)-(b)) < 0) #define _SEQ_GE(a,b) ((int)((a)-(b)) >= 0) #define BOTH_SYN (TH_SYN | (TH_SYN << 8)) #define BOTH_FIN (TH_FIN | (TH_FIN << 8)) #define TCP_FLAGS (TH_FLAGS | (TH_FLAGS << 8)) #define ACK_FWD 0x00010000 /* fwd ack seen */ #define ACK_REV 0x00020000 /* rev ack seen */ #define ACK_BOTH (ACK_FWD | ACK_REV) static uint32_t dyn_update_tcp_state(struct dyn_data *data, const struct ipfw_flow_id *pkt, const struct tcphdr *tcp, int dir) { uint32_t ack, expire; uint32_t state, old; uint8_t th_flags; expire = data->expire; old = state = data->state; th_flags = pkt->_flags & (TH_FIN | TH_SYN | TH_RST); state |= (dir == MATCH_FORWARD) ? th_flags: (th_flags << 8); switch (state & TCP_FLAGS) { case TH_SYN: /* opening */ expire = time_uptime + V_dyn_syn_lifetime; break; case BOTH_SYN: /* move to established */ case BOTH_SYN | TH_FIN: /* one side tries to close */ case BOTH_SYN | (TH_FIN << 8): if (tcp == NULL) break; ack = ntohl(tcp->th_ack); if (dir == MATCH_FORWARD) { if (data->ack_fwd == 0 || _SEQ_GE(ack, data->ack_fwd)) { state |= ACK_FWD; if (data->ack_fwd != ack) ck_pr_store_32(&data->ack_fwd, ack); } } else { if (data->ack_rev == 0 || _SEQ_GE(ack, data->ack_rev)) { state |= ACK_REV; if (data->ack_rev != ack) ck_pr_store_32(&data->ack_rev, ack); } } if ((state & ACK_BOTH) == ACK_BOTH) { /* * Set expire time to V_dyn_ack_lifetime only if * we got ACKs for both directions. * We use XOR here to avoid possible state * overwriting in concurrent thread. */ expire = time_uptime + V_dyn_ack_lifetime; ck_pr_xor_32(&data->state, ACK_BOTH); } else if ((data->state & ACK_BOTH) != (state & ACK_BOTH)) ck_pr_or_32(&data->state, state & ACK_BOTH); break; case BOTH_SYN | BOTH_FIN: /* both sides closed */ if (V_dyn_fin_lifetime >= V_dyn_keepalive_period) V_dyn_fin_lifetime = V_dyn_keepalive_period - 1; expire = time_uptime + V_dyn_fin_lifetime; break; default: if (V_dyn_rst_lifetime >= V_dyn_keepalive_period) V_dyn_rst_lifetime = V_dyn_keepalive_period - 1; expire = time_uptime + V_dyn_rst_lifetime; } /* Save TCP state if it was changed */ if ((state & TCP_FLAGS) != (old & TCP_FLAGS)) ck_pr_or_32(&data->state, state & TCP_FLAGS); return (expire); } /* * Update ULP specific state. * For TCP we keep sequence numbers and flags. For other protocols * currently we update only expire time. Packets and bytes counters * are also updated here. */ static void dyn_update_proto_state(struct dyn_data *data, const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, int dir) { uint32_t expire; /* NOTE: we are in critical section here. */ switch (pkt->proto) { case IPPROTO_UDP: case IPPROTO_UDPLITE: expire = time_uptime + V_dyn_udp_lifetime; break; case IPPROTO_TCP: expire = dyn_update_tcp_state(data, pkt, ulp, dir); break; default: expire = time_uptime + V_dyn_short_lifetime; } /* * Expiration timer has the per-second granularity, no need to update * it every time when state is matched. */ if (data->expire != expire) ck_pr_store_32(&data->expire, expire); if (dir == MATCH_FORWARD) DYN_COUNTER_INC(data, fwd, pktlen); else DYN_COUNTER_INC(data, rev, pktlen); } /* * Lookup IPv4 state. * Must be called in critical section. */ struct dyn_ipv4_state * dyn_lookup_ipv4_state(const struct ipfw_flow_id *pkt, const void *ulp, struct ipfw_dyn_info *info, int pktlen) { struct dyn_ipv4_state *s; uint32_t version, bucket; bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets); info->version = DYN_BUCKET_VERSION(bucket, ipv4_add); restart: version = DYN_BUCKET_VERSION(bucket, ipv4_del); CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { DYNSTATE_PROTECT(s); if (version != DYN_BUCKET_VERSION(bucket, ipv4_del)) goto restart; if (s->proto != pkt->proto) continue; if (info->kidx != 0 && s->kidx != info->kidx) continue; if (s->sport == pkt->src_port && s->dport == pkt->dst_port && s->src == pkt->src_ip && s->dst == pkt->dst_ip) { info->direction = MATCH_FORWARD; break; } if (s->sport == pkt->dst_port && s->dport == pkt->src_port && s->src == pkt->dst_ip && s->dst == pkt->src_ip) { info->direction = MATCH_REVERSE; break; } } if (s != NULL) dyn_update_proto_state(s->data, pkt, ulp, pktlen, info->direction); return (s); } /* * Lookup IPv4 state. * Simplifed version is used to check that matching state doesn't exist. */ static int dyn_lookup_ipv4_state_locked(const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx) { struct dyn_ipv4_state *s; int dir; dir = MATCH_NONE; DYN_BUCKET_ASSERT(bucket); CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { if (s->proto != pkt->proto || s->kidx != kidx) continue; if (s->sport == pkt->src_port && s->dport == pkt->dst_port && s->src == pkt->src_ip && s->dst == pkt->dst_ip) { dir = MATCH_FORWARD; break; } if (s->sport == pkt->dst_port && s->dport == pkt->src_port && s->src == pkt->dst_ip && s->dst == pkt->src_ip) { dir = MATCH_REVERSE; break; } } if (s != NULL) dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir); return (s != NULL); } struct dyn_ipv4_state * dyn_lookup_ipv4_parent(const struct ipfw_flow_id *pkt, const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t hashval) { struct dyn_ipv4_state *s; uint32_t version, bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); restart: version = DYN_BUCKET_VERSION(bucket, ipv4_parent_del); CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) { DYNSTATE_PROTECT(s); if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_del)) goto restart; /* * NOTE: we do not need to check kidx, because parent rule * can not create states with different kidx. * And parent rule always created for forward direction. */ if (s->limit->parent == rule && s->limit->ruleid == ruleid && s->limit->rulenum == rulenum && s->proto == pkt->proto && s->sport == pkt->src_port && s->dport == pkt->dst_port && s->src == pkt->src_ip && s->dst == pkt->dst_ip) { if (s->limit->expire != time_uptime + V_dyn_short_lifetime) ck_pr_store_32(&s->limit->expire, time_uptime + V_dyn_short_lifetime); break; } } return (s); } static struct dyn_ipv4_state * dyn_lookup_ipv4_parent_locked(const struct ipfw_flow_id *pkt, const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket) { struct dyn_ipv4_state *s; DYN_BUCKET_ASSERT(bucket); CK_SLIST_FOREACH(s, &V_dyn_ipv4_parent[bucket], entry) { if (s->limit->parent == rule && s->limit->ruleid == ruleid && s->limit->rulenum == rulenum && s->proto == pkt->proto && s->sport == pkt->src_port && s->dport == pkt->dst_port && s->src == pkt->src_ip && s->dst == pkt->dst_ip) break; } return (s); } #ifdef INET6 static uint32_t dyn_getscopeid(const struct ip_fw_args *args) { /* * If source or destination address is an scopeid address, we need * determine the scope zone id to resolve address scope ambiguity. */ if (IN6_IS_ADDR_LINKLOCAL(&args->f_id.src_ip6) || IN6_IS_ADDR_LINKLOCAL(&args->f_id.dst_ip6)) { MPASS(args->oif != NULL || args->m->m_pkthdr.rcvif != NULL); return (in6_getscopezone(args->oif != NULL ? args->oif: args->m->m_pkthdr.rcvif, IPV6_ADDR_SCOPE_LINKLOCAL)); } return (0); } /* * Lookup IPv6 state. * Must be called in critical section. */ static struct dyn_ipv6_state * dyn_lookup_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *ulp, struct ipfw_dyn_info *info, int pktlen) { struct dyn_ipv6_state *s; uint32_t version, bucket; bucket = DYN_BUCKET(info->hashval, V_curr_dyn_buckets); info->version = DYN_BUCKET_VERSION(bucket, ipv6_add); restart: version = DYN_BUCKET_VERSION(bucket, ipv6_del); CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { DYNSTATE_PROTECT(s); if (version != DYN_BUCKET_VERSION(bucket, ipv6_del)) goto restart; if (s->proto != pkt->proto || s->zoneid != zoneid) continue; if (info->kidx != 0 && s->kidx != info->kidx) continue; if (s->sport == pkt->src_port && s->dport == pkt->dst_port && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { info->direction = MATCH_FORWARD; break; } if (s->sport == pkt->dst_port && s->dport == pkt->src_port && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) { info->direction = MATCH_REVERSE; break; } } if (s != NULL) dyn_update_proto_state(s->data, pkt, ulp, pktlen, info->direction); return (s); } /* * Lookup IPv6 state. * Simplifed version is used to check that matching state doesn't exist. */ static int dyn_lookup_ipv6_state_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *ulp, int pktlen, uint32_t bucket, uint16_t kidx) { struct dyn_ipv6_state *s; int dir; dir = MATCH_NONE; DYN_BUCKET_ASSERT(bucket); CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { if (s->proto != pkt->proto || s->kidx != kidx || s->zoneid != zoneid) continue; if (s->sport == pkt->src_port && s->dport == pkt->dst_port && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { dir = MATCH_FORWARD; break; } if (s->sport == pkt->dst_port && s->dport == pkt->src_port && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->dst_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->src_ip6)) { dir = MATCH_REVERSE; break; } } if (s != NULL) dyn_update_proto_state(s->data, pkt, ulp, pktlen, dir); return (s != NULL); } static struct dyn_ipv6_state * dyn_lookup_ipv6_parent(const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t hashval) { struct dyn_ipv6_state *s; uint32_t version, bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); restart: version = DYN_BUCKET_VERSION(bucket, ipv6_parent_del); CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) { DYNSTATE_PROTECT(s); if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_del)) goto restart; /* * NOTE: we do not need to check kidx, because parent rule * can not create states with different kidx. * Also parent rule always created for forward direction. */ if (s->limit->parent == rule && s->limit->ruleid == ruleid && s->limit->rulenum == rulenum && s->proto == pkt->proto && s->sport == pkt->src_port && s->dport == pkt->dst_port && s->zoneid == zoneid && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) { if (s->limit->expire != time_uptime + V_dyn_short_lifetime) ck_pr_store_32(&s->limit->expire, time_uptime + V_dyn_short_lifetime); break; } } return (s); } static struct dyn_ipv6_state * dyn_lookup_ipv6_parent_locked(const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *rule, uint32_t ruleid, uint16_t rulenum, uint32_t bucket) { struct dyn_ipv6_state *s; DYN_BUCKET_ASSERT(bucket); CK_SLIST_FOREACH(s, &V_dyn_ipv6_parent[bucket], entry) { if (s->limit->parent == rule && s->limit->ruleid == ruleid && s->limit->rulenum == rulenum && s->proto == pkt->proto && s->sport == pkt->src_port && s->dport == pkt->dst_port && s->zoneid == zoneid && IN6_ARE_ADDR_EQUAL(&s->src, &pkt->src_ip6) && IN6_ARE_ADDR_EQUAL(&s->dst, &pkt->dst_ip6)) break; } return (s); } #endif /* INET6 */ /* * Lookup dynamic state. * pkt - filled by ipfw_chk() ipfw_flow_id; * ulp - determined by ipfw_chk() upper level protocol header; * dyn_info - info about matched state to return back; * Returns pointer to state's parent rule and dyn_info. If there is * no state, NULL is returned. * On match ipfw_dyn_lookup() updates state's counters. */ struct ip_fw * ipfw_dyn_lookup_state(const struct ip_fw_args *args, const void *ulp, int pktlen, const ipfw_insn *cmd, struct ipfw_dyn_info *info) { struct dyn_data *data; struct ip_fw *rule; IPFW_RLOCK_ASSERT(&V_layer3_chain); data = NULL; rule = NULL; info->kidx = cmd->arg1; info->direction = MATCH_NONE; info->hashval = hash_packet(&args->f_id); DYNSTATE_CRITICAL_ENTER(); if (IS_IP4_FLOW_ID(&args->f_id)) { struct dyn_ipv4_state *s; s = dyn_lookup_ipv4_state(&args->f_id, ulp, info, pktlen); if (s != NULL) { /* * Dynamic states are created using the same 5-tuple, * so it is assumed, that parent rule for O_LIMIT * state has the same address family. */ data = s->data; if (s->type == O_LIMIT) { s = data->parent; rule = s->limit->parent; } else rule = data->parent; } } #ifdef INET6 else if (IS_IP6_FLOW_ID(&args->f_id)) { struct dyn_ipv6_state *s; s = dyn_lookup_ipv6_state(&args->f_id, dyn_getscopeid(args), ulp, info, pktlen); if (s != NULL) { data = s->data; if (s->type == O_LIMIT) { s = data->parent; rule = s->limit->parent; } else rule = data->parent; } } #endif if (data != NULL) { /* * If cached chain id is the same, we can avoid rule index * lookup. Otherwise do lookup and update chain_id and f_pos. * It is safe even if there is concurrent thread that want * update the same state, because chain->id can be changed * only under IPFW_WLOCK(). */ if (data->chain_id != V_layer3_chain.id) { data->f_pos = ipfw_find_rule(&V_layer3_chain, data->rulenum, data->ruleid); /* * Check that found state has not orphaned. * When chain->id being changed the parent * rule can be deleted. If found rule doesn't * match the parent pointer, consider this * result as MATCH_NONE and return NULL. * * This will lead to creation of new similar state * that will be added into head of this bucket. * And the state that we currently have matched * should be deleted by dyn_expire_states(). * * In case when dyn_keep_states is enabled, return * pointer to default rule and corresponding f_pos * value. * XXX: In this case we lose the cache efficiency, * since f_pos is not cached, because it seems * there is no easy way to atomically switch * all fields related to parent rule of given * state. */ if (V_layer3_chain.map[data->f_pos] == rule) { data->chain_id = V_layer3_chain.id; info->f_pos = data->f_pos; } else if (V_dyn_keep_states != 0) { rule = V_layer3_chain.default_rule; info->f_pos = V_layer3_chain.n_rules - 1; } else { rule = NULL; info->direction = MATCH_NONE; DYN_DEBUG("rule %p [%u, %u] is considered " "invalid in data %p", rule, data->ruleid, data->rulenum, data); /* info->f_pos doesn't matter here. */ } } else info->f_pos = data->f_pos; } DYNSTATE_CRITICAL_EXIT(); #if 0 /* * Return MATCH_NONE if parent rule is in disabled set. * This will lead to creation of new similar state that * will be added into head of this bucket. * * XXXAE: we need to be able update state's set when parent * rule set is changed. */ if (rule != NULL && (V_set_disable & (1 << rule->set))) { rule = NULL; info->direction = MATCH_NONE; } #endif return (rule); } static struct dyn_parent * dyn_alloc_parent(void *parent, uint32_t ruleid, uint16_t rulenum, uint8_t set, uint32_t hashval) { struct dyn_parent *limit; limit = uma_zalloc(V_dyn_parent_zone, M_NOWAIT | M_ZERO); if (limit == NULL) { if (last_log != time_uptime) { last_log = time_uptime; log(LOG_DEBUG, "ipfw: Cannot allocate parent dynamic state, " "consider increasing " "net.inet.ip.fw.dyn_parent_max\n"); } return (NULL); } limit->parent = parent; limit->ruleid = ruleid; limit->rulenum = rulenum; limit->set = set; limit->hashval = hashval; limit->expire = time_uptime + V_dyn_short_lifetime; return (limit); } static struct dyn_data * dyn_alloc_dyndata(void *parent, uint32_t ruleid, uint16_t rulenum, uint8_t set, const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, uint32_t hashval, uint16_t fibnum) { struct dyn_data *data; data = uma_zalloc(V_dyn_data_zone, M_NOWAIT | M_ZERO); if (data == NULL) { if (last_log != time_uptime) { last_log = time_uptime; log(LOG_DEBUG, "ipfw: Cannot allocate dynamic state, " "consider increasing net.inet.ip.fw.dyn_max\n"); } return (NULL); } data->parent = parent; data->ruleid = ruleid; data->rulenum = rulenum; data->set = set; data->fibnum = fibnum; data->hashval = hashval; data->expire = time_uptime + V_dyn_syn_lifetime; dyn_update_proto_state(data, pkt, ulp, pktlen, MATCH_FORWARD); return (data); } static struct dyn_ipv4_state * dyn_alloc_ipv4_state(const struct ipfw_flow_id *pkt, uint16_t kidx, uint8_t type) { struct dyn_ipv4_state *s; s = uma_zalloc(V_dyn_ipv4_zone, M_NOWAIT | M_ZERO); if (s == NULL) return (NULL); s->type = type; s->kidx = kidx; s->proto = pkt->proto; s->sport = pkt->src_port; s->dport = pkt->dst_port; s->src = pkt->src_ip; s->dst = pkt->dst_ip; return (s); } /* * Add IPv4 parent state. * Returns pointer to parent state. When it is not NULL we are in * critical section and pointer protected by hazard pointer. * When some error occurs, it returns NULL and exit from critical section * is not needed. */ static struct dyn_ipv4_state * dyn_add_ipv4_parent(void *rule, uint32_t ruleid, uint16_t rulenum, uint8_t set, const struct ipfw_flow_id *pkt, uint32_t hashval, uint32_t version, uint16_t kidx) { struct dyn_ipv4_state *s; struct dyn_parent *limit; uint32_t bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); DYN_BUCKET_LOCK(bucket); if (version != DYN_BUCKET_VERSION(bucket, ipv4_parent_add)) { /* * Bucket version has been changed since last lookup, * do lookup again to be sure that state does not exist. */ s = dyn_lookup_ipv4_parent_locked(pkt, rule, ruleid, rulenum, bucket); if (s != NULL) { /* * Simultaneous thread has already created this * state. Just return it. */ DYNSTATE_CRITICAL_ENTER(); DYNSTATE_PROTECT(s); DYN_BUCKET_UNLOCK(bucket); return (s); } } limit = dyn_alloc_parent(rule, ruleid, rulenum, set, hashval); if (limit == NULL) { DYN_BUCKET_UNLOCK(bucket); return (NULL); } s = dyn_alloc_ipv4_state(pkt, kidx, O_LIMIT_PARENT); if (s == NULL) { DYN_BUCKET_UNLOCK(bucket); uma_zfree(V_dyn_parent_zone, limit); return (NULL); } s->limit = limit; CK_SLIST_INSERT_HEAD(&V_dyn_ipv4_parent[bucket], s, entry); DYN_COUNT_INC(dyn_parent_count); DYN_BUCKET_VERSION_BUMP(bucket, ipv4_parent_add); DYNSTATE_CRITICAL_ENTER(); DYNSTATE_PROTECT(s); DYN_BUCKET_UNLOCK(bucket); return (s); } static int dyn_add_ipv4_state(void *parent, uint32_t ruleid, uint16_t rulenum, uint8_t set, const struct ipfw_flow_id *pkt, const void *ulp, int pktlen, uint32_t hashval, struct ipfw_dyn_info *info, uint16_t fibnum, uint16_t kidx, uint8_t type) { struct dyn_ipv4_state *s; void *data; uint32_t bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); DYN_BUCKET_LOCK(bucket); if (info->direction == MATCH_UNKNOWN || info->kidx != kidx || info->hashval != hashval || info->version != DYN_BUCKET_VERSION(bucket, ipv4_add)) { /* * Bucket version has been changed since last lookup, * do lookup again to be sure that state does not exist. */ if (dyn_lookup_ipv4_state_locked(pkt, ulp, pktlen, bucket, kidx) != 0) { DYN_BUCKET_UNLOCK(bucket); return (EEXIST); } } data = dyn_alloc_dyndata(parent, ruleid, rulenum, set, pkt, ulp, pktlen, hashval, fibnum); if (data == NULL) { DYN_BUCKET_UNLOCK(bucket); return (ENOMEM); } s = dyn_alloc_ipv4_state(pkt, kidx, type); if (s == NULL) { DYN_BUCKET_UNLOCK(bucket); uma_zfree(V_dyn_data_zone, data); return (ENOMEM); } s->data = data; CK_SLIST_INSERT_HEAD(&V_dyn_ipv4[bucket], s, entry); DYN_COUNT_INC(dyn_count); DYN_BUCKET_VERSION_BUMP(bucket, ipv4_add); DYN_BUCKET_UNLOCK(bucket); return (0); } #ifdef INET6 static struct dyn_ipv6_state * dyn_alloc_ipv6_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, uint16_t kidx, uint8_t type) { struct dyn_ipv6_state *s; s = uma_zalloc(V_dyn_ipv6_zone, M_NOWAIT | M_ZERO); if (s == NULL) return (NULL); s->type = type; s->kidx = kidx; s->zoneid = zoneid; s->proto = pkt->proto; s->sport = pkt->src_port; s->dport = pkt->dst_port; s->src = pkt->src_ip6; s->dst = pkt->dst_ip6; return (s); } /* * Add IPv6 parent state. * Returns pointer to parent state. When it is not NULL we are in * critical section and pointer protected by hazard pointer. * When some error occurs, it return NULL and exit from critical section * is not needed. */ static struct dyn_ipv6_state * dyn_add_ipv6_parent(void *rule, uint32_t ruleid, uint16_t rulenum, uint8_t set, const struct ipfw_flow_id *pkt, uint32_t zoneid, uint32_t hashval, uint32_t version, uint16_t kidx) { struct dyn_ipv6_state *s; struct dyn_parent *limit; uint32_t bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); DYN_BUCKET_LOCK(bucket); if (version != DYN_BUCKET_VERSION(bucket, ipv6_parent_add)) { /* * Bucket version has been changed since last lookup, * do lookup again to be sure that state does not exist. */ s = dyn_lookup_ipv6_parent_locked(pkt, zoneid, rule, ruleid, rulenum, bucket); if (s != NULL) { /* * Simultaneous thread has already created this * state. Just return it. */ DYNSTATE_CRITICAL_ENTER(); DYNSTATE_PROTECT(s); DYN_BUCKET_UNLOCK(bucket); return (s); } } limit = dyn_alloc_parent(rule, ruleid, rulenum, set, hashval); if (limit == NULL) { DYN_BUCKET_UNLOCK(bucket); return (NULL); } s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, O_LIMIT_PARENT); if (s == NULL) { DYN_BUCKET_UNLOCK(bucket); uma_zfree(V_dyn_parent_zone, limit); return (NULL); } s->limit = limit; CK_SLIST_INSERT_HEAD(&V_dyn_ipv6_parent[bucket], s, entry); DYN_COUNT_INC(dyn_parent_count); DYN_BUCKET_VERSION_BUMP(bucket, ipv6_parent_add); DYNSTATE_CRITICAL_ENTER(); DYNSTATE_PROTECT(s); DYN_BUCKET_UNLOCK(bucket); return (s); } static int dyn_add_ipv6_state(void *parent, uint32_t ruleid, uint16_t rulenum, uint8_t set, const struct ipfw_flow_id *pkt, uint32_t zoneid, const void *ulp, int pktlen, uint32_t hashval, struct ipfw_dyn_info *info, uint16_t fibnum, uint16_t kidx, uint8_t type) { struct dyn_ipv6_state *s; struct dyn_data *data; uint32_t bucket; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); DYN_BUCKET_LOCK(bucket); if (info->direction == MATCH_UNKNOWN || info->kidx != kidx || info->hashval != hashval || info->version != DYN_BUCKET_VERSION(bucket, ipv6_add)) { /* * Bucket version has been changed since last lookup, * do lookup again to be sure that state does not exist. */ if (dyn_lookup_ipv6_state_locked(pkt, zoneid, ulp, pktlen, bucket, kidx) != 0) { DYN_BUCKET_UNLOCK(bucket); return (EEXIST); } } data = dyn_alloc_dyndata(parent, ruleid, rulenum, set, pkt, ulp, pktlen, hashval, fibnum); if (data == NULL) { DYN_BUCKET_UNLOCK(bucket); return (ENOMEM); } s = dyn_alloc_ipv6_state(pkt, zoneid, kidx, type); if (s == NULL) { DYN_BUCKET_UNLOCK(bucket); uma_zfree(V_dyn_data_zone, data); return (ENOMEM); } s->data = data; CK_SLIST_INSERT_HEAD(&V_dyn_ipv6[bucket], s, entry); DYN_COUNT_INC(dyn_count); DYN_BUCKET_VERSION_BUMP(bucket, ipv6_add); DYN_BUCKET_UNLOCK(bucket); return (0); } #endif /* INET6 */ static void * dyn_get_parent_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, struct ip_fw *rule, uint32_t hashval, uint32_t limit, uint16_t kidx) { char sbuf[24]; struct dyn_parent *p; void *ret; uint32_t bucket, version; p = NULL; ret = NULL; bucket = DYN_BUCKET(hashval, V_curr_dyn_buckets); DYNSTATE_CRITICAL_ENTER(); if (IS_IP4_FLOW_ID(pkt)) { struct dyn_ipv4_state *s; version = DYN_BUCKET_VERSION(bucket, ipv4_parent_add); s = dyn_lookup_ipv4_parent(pkt, rule, rule->id, rule->rulenum, bucket); if (s == NULL) { /* * Exit from critical section because dyn_add_parent() * will acquire bucket lock. */ DYNSTATE_CRITICAL_EXIT(); s = dyn_add_ipv4_parent(rule, rule->id, rule->rulenum, rule->set, pkt, hashval, version, kidx); if (s == NULL) return (NULL); /* Now we are in critical section again. */ } ret = s; p = s->limit; } #ifdef INET6 else if (IS_IP6_FLOW_ID(pkt)) { struct dyn_ipv6_state *s; version = DYN_BUCKET_VERSION(bucket, ipv6_parent_add); s = dyn_lookup_ipv6_parent(pkt, zoneid, rule, rule->id, rule->rulenum, bucket); if (s == NULL) { /* * Exit from critical section because dyn_add_parent() * can acquire bucket mutex. */ DYNSTATE_CRITICAL_EXIT(); s = dyn_add_ipv6_parent(rule, rule->id, rule->rulenum, rule->set, pkt, zoneid, hashval, version, kidx); if (s == NULL) return (NULL); /* Now we are in critical section again. */ } ret = s; p = s->limit; } #endif else { DYNSTATE_CRITICAL_EXIT(); return (NULL); } /* Check the limit */ if (DPARENT_COUNT(p) >= limit) { DYNSTATE_CRITICAL_EXIT(); if (V_fw_verbose && last_log != time_uptime) { last_log = time_uptime; snprintf(sbuf, sizeof(sbuf), "%u drop session", rule->rulenum); print_dyn_rule_flags(pkt, O_LIMIT, LOG_SECURITY | LOG_DEBUG, sbuf, "too many entries"); } return (NULL); } /* Take new session into account. */ DPARENT_COUNT_INC(p); /* * We must exit from critical section because the following code * can acquire bucket mutex. * We rely on the the 'count' field. The state will not expire * until it has some child states, i.e. 'count' field is not zero. * Return state pointer, it will be used by child states as parent. */ DYNSTATE_CRITICAL_EXIT(); return (ret); } static int dyn_install_state(const struct ipfw_flow_id *pkt, uint32_t zoneid, uint16_t fibnum, const void *ulp, int pktlen, void *rule, uint32_t ruleid, uint16_t rulenum, uint8_t set, struct ipfw_dyn_info *info, uint32_t limit, uint16_t limit_mask, uint16_t kidx, uint8_t type) { struct ipfw_flow_id id; uint32_t hashval, parent_hashval; int ret; MPASS(type == O_LIMIT || type == O_KEEP_STATE); if (type == O_LIMIT) { /* Create masked flow id and calculate bucket */ id.addr_type = pkt->addr_type; id.proto = pkt->proto; id.fib = fibnum; /* unused */ id.src_port = (limit_mask & DYN_SRC_PORT) ? pkt->src_port: 0; id.dst_port = (limit_mask & DYN_DST_PORT) ? pkt->dst_port: 0; if (IS_IP4_FLOW_ID(pkt)) { id.src_ip = (limit_mask & DYN_SRC_ADDR) ? pkt->src_ip: 0; id.dst_ip = (limit_mask & DYN_DST_ADDR) ? pkt->dst_ip: 0; } #ifdef INET6 else if (IS_IP6_FLOW_ID(pkt)) { if (limit_mask & DYN_SRC_ADDR) id.src_ip6 = pkt->src_ip6; else memset(&id.src_ip6, 0, sizeof(id.src_ip6)); if (limit_mask & DYN_DST_ADDR) id.dst_ip6 = pkt->dst_ip6; else memset(&id.dst_ip6, 0, sizeof(id.dst_ip6)); } #endif else return (EAFNOSUPPORT); parent_hashval = hash_parent(&id, rule); rule = dyn_get_parent_state(&id, zoneid, rule, parent_hashval, limit, kidx); if (rule == NULL) { #if 0 if (V_fw_verbose && last_log != time_uptime) { last_log = time_uptime; snprintf(sbuf, sizeof(sbuf), "%u drop session", rule->rulenum); print_dyn_rule_flags(pkt, O_LIMIT, LOG_SECURITY | LOG_DEBUG, sbuf, "too many entries"); } #endif return (EACCES); } /* * Limit is not reached, create new state. * Now rule points to parent state. */ } hashval = hash_packet(pkt); if (IS_IP4_FLOW_ID(pkt)) ret = dyn_add_ipv4_state(rule, ruleid, rulenum, set, pkt, ulp, pktlen, hashval, info, fibnum, kidx, type); #ifdef INET6 else if (IS_IP6_FLOW_ID(pkt)) ret = dyn_add_ipv6_state(rule, ruleid, rulenum, set, pkt, zoneid, ulp, pktlen, hashval, info, fibnum, kidx, type); #endif /* INET6 */ else ret = EAFNOSUPPORT; if (type == O_LIMIT) { if (ret != 0) { /* * We failed to create child state for O_LIMIT * opcode. Since we already counted it in the parent, * we must revert counter back. The 'rule' points to * parent state, use it to get dyn_parent. * * XXXAE: it should be safe to use 'rule' pointer * without extra lookup, parent state is referenced * and should not be freed. */ if (IS_IP4_FLOW_ID(&id)) DPARENT_COUNT_DEC( ((struct dyn_ipv4_state *)rule)->limit); #ifdef INET6 else if (IS_IP6_FLOW_ID(&id)) DPARENT_COUNT_DEC( ((struct dyn_ipv6_state *)rule)->limit); #endif } } /* * EEXIST means that simultaneous thread has created this * state. Consider this as success. * * XXXAE: should we invalidate 'info' content here? */ if (ret == EEXIST) return (0); return (ret); } /* * Install dynamic state. * chain - ipfw's instance; * rule - the parent rule that installs the state; * cmd - opcode that installs the state; * args - ipfw arguments; * ulp - upper level protocol header; * pktlen - packet length; * info - dynamic state lookup info; * tablearg - tablearg id. * * Returns non-zero value (failure) if state is not installed because * of errors or because session limitations are enforced. */ int ipfw_dyn_install_state(struct ip_fw_chain *chain, struct ip_fw *rule, const ipfw_insn_limit *cmd, const struct ip_fw_args *args, const void *ulp, int pktlen, struct ipfw_dyn_info *info, uint32_t tablearg) { uint32_t limit; uint16_t limit_mask; if (cmd->o.opcode == O_LIMIT) { limit = IP_FW_ARG_TABLEARG(chain, cmd->conn_limit, limit); limit_mask = cmd->limit_mask; } else { limit = 0; limit_mask = 0; } return (dyn_install_state(&args->f_id, #ifdef INET6 IS_IP6_FLOW_ID(&args->f_id) ? dyn_getscopeid(args): #endif 0, M_GETFIB(args->m), ulp, pktlen, rule, rule->id, rule->rulenum, rule->set, info, limit, limit_mask, cmd->o.arg1, cmd->o.opcode)); } /* * Free safe to remove state entries from expired lists. */ static void dyn_free_states(struct ip_fw_chain *chain) { struct dyn_ipv4_state *s4, *s4n; #ifdef INET6 struct dyn_ipv6_state *s6, *s6n; #endif int cached_count, i; /* * We keep pointers to objects that are in use on each CPU * in the per-cpu dyn_hp pointer. When object is going to be * removed, first of it is unlinked from the corresponding * list. This leads to changing of dyn_bucket_xxx_delver version. * Unlinked objects is placed into corresponding dyn_expired_xxx * list. Reader that is going to dereference object pointer checks * dyn_bucket_xxx_delver version before and after storing pointer * into dyn_hp. If version is the same, the object is protected * from freeing and it is safe to dereference. Othervise reader * tries to iterate list again from the beginning, but this object * now unlinked and thus will not be accessible. * * Copy dyn_hp pointers for each CPU into dyn_hp_cache array. * It does not matter that some pointer can be changed in * time while we are copying. We need to check, that objects * removed in the previous pass are not in use. And if dyn_hp * pointer does not contain it in the time when we are copying, * it will not appear there, because it is already unlinked. * And for new pointers we will not free objects that will be * unlinked in this pass. */ cached_count = 0; CPU_FOREACH(i) { dyn_hp_cache[cached_count] = DYNSTATE_GET(i); if (dyn_hp_cache[cached_count] != NULL) cached_count++; } /* * Free expired states that are safe to free. * Check each entry from previous pass in the dyn_expired_xxx * list, if pointer to the object is in the dyn_hp_cache array, * keep it until next pass. Otherwise it is safe to free the * object. * * XXXAE: optimize this to use SLIST_REMOVE_AFTER. */ #define DYN_FREE_STATES(s, next, name) do { \ s = SLIST_FIRST(&V_dyn_expired_ ## name); \ while (s != NULL) { \ next = SLIST_NEXT(s, expired); \ for (i = 0; i < cached_count; i++) \ if (dyn_hp_cache[i] == s) \ break; \ if (i == cached_count) { \ if (s->type == O_LIMIT_PARENT && \ s->limit->count != 0) { \ s = next; \ continue; \ } \ SLIST_REMOVE(&V_dyn_expired_ ## name, \ s, dyn_ ## name ## _state, expired); \ if (s->type == O_LIMIT_PARENT) \ uma_zfree(V_dyn_parent_zone, s->limit); \ else \ uma_zfree(V_dyn_data_zone, s->data); \ uma_zfree(V_dyn_ ## name ## _zone, s); \ } \ s = next; \ } \ } while (0) /* * Protect access to expired lists with DYN_EXPIRED_LOCK. * Userland can invoke ipfw_expire_dyn_states() to delete * specific states, this will lead to modification of expired * lists. * * XXXAE: do we need DYN_EXPIRED_LOCK? We can just use * IPFW_UH_WLOCK to protect access to these lists. */ DYN_EXPIRED_LOCK(); DYN_FREE_STATES(s4, s4n, ipv4); #ifdef INET6 DYN_FREE_STATES(s6, s6n, ipv6); #endif DYN_EXPIRED_UNLOCK(); #undef DYN_FREE_STATES } /* * Returns 1 when state is matched by specified range, otherwise returns 0. */ static int dyn_match_range(uint16_t rulenum, uint8_t set, const ipfw_range_tlv *rt) { MPASS(rt != NULL); /* flush all states */ if (rt->flags & IPFW_RCFLAG_ALL) return (1); if ((rt->flags & IPFW_RCFLAG_SET) != 0 && set != rt->set) return (0); if ((rt->flags & IPFW_RCFLAG_RANGE) != 0 && (rulenum < rt->start_rule || rulenum > rt->end_rule)) return (0); return (1); } static int dyn_match_ipv4_state(struct dyn_ipv4_state *s, const ipfw_range_tlv *rt) { if (s->type == O_LIMIT_PARENT) return (dyn_match_range(s->limit->rulenum, s->limit->set, rt)); if (s->type == O_LIMIT) return (dyn_match_range(s->data->rulenum, s->data->set, rt)); if (V_dyn_keep_states == 0 && dyn_match_range(s->data->rulenum, s->data->set, rt)) return (1); return (0); } #ifdef INET6 static int dyn_match_ipv6_state(struct dyn_ipv6_state *s, const ipfw_range_tlv *rt) { if (s->type == O_LIMIT_PARENT) return (dyn_match_range(s->limit->rulenum, s->limit->set, rt)); if (s->type == O_LIMIT) return (dyn_match_range(s->data->rulenum, s->data->set, rt)); if (V_dyn_keep_states == 0 && dyn_match_range(s->data->rulenum, s->data->set, rt)) return (1); return (0); } #endif /* * Unlink expired entries from states lists. * @rt can be used to specify the range of states for deletion. */ static void dyn_expire_states(struct ip_fw_chain *chain, ipfw_range_tlv *rt) { struct dyn_ipv4_slist expired_ipv4; #ifdef INET6 struct dyn_ipv6_slist expired_ipv6; struct dyn_ipv6_state *s6, *s6n, *s6p; #endif struct dyn_ipv4_state *s4, *s4n, *s4p; int bucket, removed, length, max_length; /* * Unlink expired states from each bucket. * With acquired bucket lock iterate entries of each lists: * ipv4, ipv4_parent, ipv6, and ipv6_parent. Check expired time * and unlink entry from the list, link entry into temporary * expired_xxx lists then bump "del" bucket version. * * When an entry is removed, corresponding states counter is * decremented. If entry has O_LIMIT type, parent's reference * counter is decremented. * * NOTE: this function can be called from userspace context * when user deletes rules. In this case all matched states * will be forcedly unlinked. O_LIMIT_PARENT states will be kept * in the expired lists until reference counter become zero. */ #define DYN_UNLINK_STATES(s, prev, next, exp, af, name, extra) do { \ length = 0; \ removed = 0; \ prev = NULL; \ s = CK_SLIST_FIRST(&V_dyn_ ## name [bucket]); \ while (s != NULL) { \ next = CK_SLIST_NEXT(s, entry); \ if ((TIME_LEQ((s)->exp, time_uptime) && extra) || \ (rt != NULL && dyn_match_ ## af ## _state(s, rt))) {\ if (prev != NULL) \ CK_SLIST_REMOVE_AFTER(prev, entry); \ else \ CK_SLIST_REMOVE_HEAD( \ &V_dyn_ ## name [bucket], entry); \ removed++; \ SLIST_INSERT_HEAD(&expired_ ## af, s, expired); \ if (s->type == O_LIMIT_PARENT) \ DYN_COUNT_DEC(dyn_parent_count); \ else { \ DYN_COUNT_DEC(dyn_count); \ if (s->type == O_LIMIT) { \ s = s->data->parent; \ DPARENT_COUNT_DEC(s->limit); \ } \ } \ } else { \ prev = s; \ length++; \ } \ s = next; \ } \ if (removed != 0) \ DYN_BUCKET_VERSION_BUMP(bucket, name ## _del); \ if (length > max_length) \ max_length = length; \ } while (0) SLIST_INIT(&expired_ipv4); #ifdef INET6 SLIST_INIT(&expired_ipv6); #endif max_length = 0; for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { DYN_BUCKET_LOCK(bucket); DYN_UNLINK_STATES(s4, s4p, s4n, data->expire, ipv4, ipv4, 1); DYN_UNLINK_STATES(s4, s4p, s4n, limit->expire, ipv4, ipv4_parent, (s4->limit->count == 0)); #ifdef INET6 DYN_UNLINK_STATES(s6, s6p, s6n, data->expire, ipv6, ipv6, 1); DYN_UNLINK_STATES(s6, s6p, s6n, limit->expire, ipv6, ipv6_parent, (s6->limit->count == 0)); #endif DYN_BUCKET_UNLOCK(bucket); } /* Update curr_max_length for statistics. */ V_curr_max_length = max_length; /* * Concatenate temporary lists with global expired lists. */ DYN_EXPIRED_LOCK(); SLIST_CONCAT(&V_dyn_expired_ipv4, &expired_ipv4, dyn_ipv4_state, expired); #ifdef INET6 SLIST_CONCAT(&V_dyn_expired_ipv6, &expired_ipv6, dyn_ipv6_state, expired); #endif DYN_EXPIRED_UNLOCK(); #undef DYN_UNLINK_STATES #undef DYN_UNREF_STATES } static struct mbuf * dyn_mgethdr(int len, uint16_t fibnum) { struct mbuf *m; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (NULL); #ifdef MAC mac_netinet_firewall_send(m); #endif M_SETFIB(m, fibnum); m->m_data += max_linkhdr; m->m_flags |= M_SKIP_FIREWALL; m->m_len = m->m_pkthdr.len = len; bzero(m->m_data, len); return (m); } static void dyn_make_keepalive_ipv4(struct mbuf *m, in_addr_t src, in_addr_t dst, uint32_t seq, uint32_t ack, uint16_t sport, uint16_t dport) { struct tcphdr *tcp; struct ip *ip; ip = mtod(m, struct ip *); ip->ip_v = 4; ip->ip_hl = sizeof(*ip) >> 2; ip->ip_tos = IPTOS_LOWDELAY; ip->ip_len = htons(m->m_len); ip->ip_off |= htons(IP_DF); ip->ip_ttl = V_ip_defttl; ip->ip_p = IPPROTO_TCP; ip->ip_src.s_addr = htonl(src); ip->ip_dst.s_addr = htonl(dst); tcp = mtodo(m, sizeof(struct ip)); tcp->th_sport = htons(sport); tcp->th_dport = htons(dport); tcp->th_off = sizeof(struct tcphdr) >> 2; tcp->th_seq = htonl(seq); tcp->th_ack = htonl(ack); tcp->th_flags = TH_ACK; tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(sizeof(struct tcphdr) + IPPROTO_TCP)); m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); m->m_pkthdr.csum_flags = CSUM_TCP; } static void dyn_enqueue_keepalive_ipv4(struct mbufq *q, const struct dyn_ipv4_state *s) { struct mbuf *m; if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) { m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr), s->data->fibnum); if (m != NULL) { dyn_make_keepalive_ipv4(m, s->dst, s->src, s->data->ack_fwd - 1, s->data->ack_rev, s->dport, s->sport); if (mbufq_enqueue(q, m)) { m_freem(m); log(LOG_DEBUG, "ipfw: limit for IPv4 " "keepalive queue is reached.\n"); return; } } } if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) { m = dyn_mgethdr(sizeof(struct ip) + sizeof(struct tcphdr), s->data->fibnum); if (m != NULL) { dyn_make_keepalive_ipv4(m, s->src, s->dst, s->data->ack_rev - 1, s->data->ack_fwd, s->sport, s->dport); if (mbufq_enqueue(q, m)) { m_freem(m); log(LOG_DEBUG, "ipfw: limit for IPv4 " "keepalive queue is reached.\n"); return; } } } } /* * Prepare and send keep-alive packets. */ static void dyn_send_keepalive_ipv4(struct ip_fw_chain *chain) { struct mbufq q; struct mbuf *m; struct dyn_ipv4_state *s; uint32_t bucket; mbufq_init(&q, DYN_KEEPALIVE_MAXQ); IPFW_UH_RLOCK(chain); /* * It is safe to not use hazard pointer and just do lockless * access to the lists, because states entries can not be deleted * while we hold IPFW_UH_RLOCK. */ for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { CK_SLIST_FOREACH(s, &V_dyn_ipv4[bucket], entry) { /* * Only established TCP connections that will * become expired withing dyn_keepalive_interval. */ if (s->proto != IPPROTO_TCP || (s->data->state & BOTH_SYN) != BOTH_SYN || TIME_LEQ(time_uptime + V_dyn_keepalive_interval, s->data->expire)) continue; dyn_enqueue_keepalive_ipv4(&q, s); } } IPFW_UH_RUNLOCK(chain); while ((m = mbufq_dequeue(&q)) != NULL) ip_output(m, NULL, NULL, 0, NULL, NULL); } #ifdef INET6 static void dyn_make_keepalive_ipv6(struct mbuf *m, const struct in6_addr *src, const struct in6_addr *dst, uint32_t zoneid, uint32_t seq, uint32_t ack, uint16_t sport, uint16_t dport) { struct tcphdr *tcp; struct ip6_hdr *ip6; ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_hlim = IPV6_DEFHLIM; ip6->ip6_src = *src; if (IN6_IS_ADDR_LINKLOCAL(src)) ip6->ip6_src.s6_addr16[1] = htons(zoneid & 0xffff); ip6->ip6_dst = *dst; if (IN6_IS_ADDR_LINKLOCAL(dst)) ip6->ip6_dst.s6_addr16[1] = htons(zoneid & 0xffff); tcp = mtodo(m, sizeof(struct ip6_hdr)); tcp->th_sport = htons(sport); tcp->th_dport = htons(dport); tcp->th_off = sizeof(struct tcphdr) >> 2; tcp->th_seq = htonl(seq); tcp->th_ack = htonl(ack); tcp->th_flags = TH_ACK; tcp->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr), IPPROTO_TCP, 0); m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; } static void dyn_enqueue_keepalive_ipv6(struct mbufq *q, const struct dyn_ipv6_state *s) { struct mbuf *m; if ((s->data->state & ACK_FWD) == 0 && s->data->ack_fwd > 0) { m = dyn_mgethdr(sizeof(struct ip6_hdr) + sizeof(struct tcphdr), s->data->fibnum); if (m != NULL) { dyn_make_keepalive_ipv6(m, &s->dst, &s->src, s->zoneid, s->data->ack_fwd - 1, s->data->ack_rev, s->dport, s->sport); if (mbufq_enqueue(q, m)) { m_freem(m); log(LOG_DEBUG, "ipfw: limit for IPv6 " "keepalive queue is reached.\n"); return; } } } if ((s->data->state & ACK_REV) == 0 && s->data->ack_rev > 0) { m = dyn_mgethdr(sizeof(struct ip6_hdr) + sizeof(struct tcphdr), s->data->fibnum); if (m != NULL) { dyn_make_keepalive_ipv6(m, &s->src, &s->dst, s->zoneid, s->data->ack_rev - 1, s->data->ack_fwd, s->sport, s->dport); if (mbufq_enqueue(q, m)) { m_freem(m); log(LOG_DEBUG, "ipfw: limit for IPv6 " "keepalive queue is reached.\n"); return; } } } } static void dyn_send_keepalive_ipv6(struct ip_fw_chain *chain) { struct mbufq q; struct mbuf *m; struct dyn_ipv6_state *s; uint32_t bucket; mbufq_init(&q, DYN_KEEPALIVE_MAXQ); IPFW_UH_RLOCK(chain); /* * It is safe to not use hazard pointer and just do lockless * access to the lists, because states entries can not be deleted * while we hold IPFW_UH_RLOCK. */ for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { CK_SLIST_FOREACH(s, &V_dyn_ipv6[bucket], entry) { /* * Only established TCP connections that will * become expired withing dyn_keepalive_interval. */ if (s->proto != IPPROTO_TCP || (s->data->state & BOTH_SYN) != BOTH_SYN || TIME_LEQ(time_uptime + V_dyn_keepalive_interval, s->data->expire)) continue; dyn_enqueue_keepalive_ipv6(&q, s); } } IPFW_UH_RUNLOCK(chain); while ((m = mbufq_dequeue(&q)) != NULL) ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); } #endif /* INET6 */ static void dyn_grow_hashtable(struct ip_fw_chain *chain, uint32_t new) { #ifdef INET6 struct dyn_ipv6ck_slist *ipv6, *ipv6_parent; uint32_t *ipv6_add, *ipv6_del, *ipv6_parent_add, *ipv6_parent_del; struct dyn_ipv6_state *s6; #endif struct dyn_ipv4ck_slist *ipv4, *ipv4_parent; uint32_t *ipv4_add, *ipv4_del, *ipv4_parent_add, *ipv4_parent_del; struct dyn_ipv4_state *s4; struct mtx *bucket_lock; void *tmp; uint32_t bucket; MPASS(powerof2(new)); DYN_DEBUG("grow hash size %u -> %u", V_curr_dyn_buckets, new); /* * Allocate and initialize new lists. * XXXAE: on memory pressure this can disable callout timer. */ bucket_lock = malloc(new * sizeof(struct mtx), M_IPFW, M_WAITOK | M_ZERO); ipv4 = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW, M_WAITOK | M_ZERO); ipv4_parent = malloc(new * sizeof(struct dyn_ipv4ck_slist), M_IPFW, M_WAITOK | M_ZERO); ipv4_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv4_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv4_parent_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv4_parent_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); #ifdef INET6 ipv6 = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW, M_WAITOK | M_ZERO); ipv6_parent = malloc(new * sizeof(struct dyn_ipv6ck_slist), M_IPFW, M_WAITOK | M_ZERO); ipv6_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv6_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv6_parent_add = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); ipv6_parent_del = malloc(new * sizeof(uint32_t), M_IPFW, M_WAITOK | M_ZERO); #endif for (bucket = 0; bucket < new; bucket++) { DYN_BUCKET_LOCK_INIT(bucket_lock, bucket); CK_SLIST_INIT(&ipv4[bucket]); CK_SLIST_INIT(&ipv4_parent[bucket]); #ifdef INET6 CK_SLIST_INIT(&ipv6[bucket]); CK_SLIST_INIT(&ipv6_parent[bucket]); #endif } #define DYN_RELINK_STATES(s, hval, i, head, ohead) do { \ while ((s = CK_SLIST_FIRST(&V_dyn_ ## ohead[i])) != NULL) { \ CK_SLIST_REMOVE_HEAD(&V_dyn_ ## ohead[i], entry); \ CK_SLIST_INSERT_HEAD(&head[DYN_BUCKET(s->hval, new)], \ s, entry); \ } \ } while (0) /* * Prevent rules changing from userland. */ IPFW_UH_WLOCK(chain); /* * Hold traffic processing until we finish resize to * prevent access to states lists. */ IPFW_WLOCK(chain); /* Re-link all dynamic states */ for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { DYN_RELINK_STATES(s4, data->hashval, bucket, ipv4, ipv4); DYN_RELINK_STATES(s4, limit->hashval, bucket, ipv4_parent, ipv4_parent); #ifdef INET6 DYN_RELINK_STATES(s6, data->hashval, bucket, ipv6, ipv6); DYN_RELINK_STATES(s6, limit->hashval, bucket, ipv6_parent, ipv6_parent); #endif } #define DYN_SWAP_PTR(old, new, tmp) do { \ tmp = old; \ old = new; \ new = tmp; \ } while (0) /* Swap pointers */ DYN_SWAP_PTR(V_dyn_bucket_lock, bucket_lock, tmp); DYN_SWAP_PTR(V_dyn_ipv4, ipv4, tmp); DYN_SWAP_PTR(V_dyn_ipv4_parent, ipv4_parent, tmp); DYN_SWAP_PTR(V_dyn_ipv4_add, ipv4_add, tmp); DYN_SWAP_PTR(V_dyn_ipv4_parent_add, ipv4_parent_add, tmp); DYN_SWAP_PTR(V_dyn_ipv4_del, ipv4_del, tmp); DYN_SWAP_PTR(V_dyn_ipv4_parent_del, ipv4_parent_del, tmp); #ifdef INET6 DYN_SWAP_PTR(V_dyn_ipv6, ipv6, tmp); DYN_SWAP_PTR(V_dyn_ipv6_parent, ipv6_parent, tmp); DYN_SWAP_PTR(V_dyn_ipv6_add, ipv6_add, tmp); DYN_SWAP_PTR(V_dyn_ipv6_parent_add, ipv6_parent_add, tmp); DYN_SWAP_PTR(V_dyn_ipv6_del, ipv6_del, tmp); DYN_SWAP_PTR(V_dyn_ipv6_parent_del, ipv6_parent_del, tmp); #endif bucket = V_curr_dyn_buckets; V_curr_dyn_buckets = new; IPFW_WUNLOCK(chain); IPFW_UH_WUNLOCK(chain); /* Release old resources */ while (bucket-- != 0) DYN_BUCKET_LOCK_DESTROY(bucket_lock, bucket); free(bucket_lock, M_IPFW); free(ipv4, M_IPFW); free(ipv4_parent, M_IPFW); free(ipv4_add, M_IPFW); free(ipv4_parent_add, M_IPFW); free(ipv4_del, M_IPFW); free(ipv4_parent_del, M_IPFW); #ifdef INET6 free(ipv6, M_IPFW); free(ipv6_parent, M_IPFW); free(ipv6_add, M_IPFW); free(ipv6_parent_add, M_IPFW); free(ipv6_del, M_IPFW); free(ipv6_parent_del, M_IPFW); #endif } /* * This function is used to perform various maintenance * on dynamic hash lists. Currently it is called every second. */ static void dyn_tick(void *vnetx) { uint32_t buckets; CURVNET_SET((struct vnet *)vnetx); /* * First free states unlinked in previous passes. */ dyn_free_states(&V_layer3_chain); /* * Now unlink others expired states. * We use IPFW_UH_WLOCK to avoid concurrent call of * dyn_expire_states(). It is the only function that does * deletion of state entries from states lists. */ IPFW_UH_WLOCK(&V_layer3_chain); dyn_expire_states(&V_layer3_chain, NULL); IPFW_UH_WUNLOCK(&V_layer3_chain); /* * Send keepalives if they are enabled and the time has come. */ if (V_dyn_keepalive != 0 && V_dyn_keepalive_last + V_dyn_keepalive_period <= time_uptime) { V_dyn_keepalive_last = time_uptime; dyn_send_keepalive_ipv4(&V_layer3_chain); #ifdef INET6 dyn_send_keepalive_ipv6(&V_layer3_chain); #endif } /* * Check if we need to resize the hash: * if current number of states exceeds number of buckets in hash, * and dyn_buckets_max permits to grow the number of buckets, then * do it. Grow hash size to the minimum power of 2 which is bigger * than current states count. */ if (V_curr_dyn_buckets < V_dyn_buckets_max && (V_curr_dyn_buckets < V_dyn_count / 2 || ( V_curr_dyn_buckets < V_dyn_count && V_curr_max_length > 8))) { buckets = 1 << fls(V_dyn_count); if (buckets > V_dyn_buckets_max) buckets = V_dyn_buckets_max; dyn_grow_hashtable(&V_layer3_chain, buckets); } callout_reset_on(&V_dyn_timeout, hz, dyn_tick, vnetx, 0); CURVNET_RESTORE(); } void ipfw_expire_dyn_states(struct ip_fw_chain *chain, ipfw_range_tlv *rt) { /* * Do not perform any checks if we currently have no dynamic states */ if (V_dyn_count == 0) return; IPFW_UH_WLOCK_ASSERT(chain); dyn_expire_states(chain, rt); } /* * Returns size of dynamic states in legacy format */ int ipfw_dyn_len(void) { return ((V_dyn_count + V_dyn_parent_count) * sizeof(ipfw_dyn_rule)); } /* * Returns number of dynamic states. * Used by dump format v1 (current). */ uint32_t ipfw_dyn_get_count(void) { return (V_dyn_count + V_dyn_parent_count); } /* * Check if rule contains at least one dynamic opcode. * * Returns 1 if such opcode is found, 0 otherwise. */ int ipfw_is_dyn_rule(struct ip_fw *rule) { int cmdlen, l; ipfw_insn *cmd; l = rule->cmd_len; cmd = rule->cmd; cmdlen = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); switch (cmd->opcode) { case O_LIMIT: case O_KEEP_STATE: case O_PROBE_STATE: case O_CHECK_STATE: return (1); } } return (0); } static void dyn_export_parent(const struct dyn_parent *p, uint16_t kidx, ipfw_dyn_rule *dst) { dst->dyn_type = O_LIMIT_PARENT; dst->kidx = kidx; dst->count = (uint16_t)DPARENT_COUNT(p); dst->expire = TIME_LEQ(p->expire, time_uptime) ? 0: p->expire - time_uptime; /* 'rule' is used to pass up the rule number and set */ memcpy(&dst->rule, &p->rulenum, sizeof(p->rulenum)); /* store set number into high word of dst->rule pointer. */ memcpy((char *)&dst->rule + sizeof(p->rulenum), &p->set, sizeof(p->set)); /* unused fields */ dst->pcnt = 0; dst->bcnt = 0; dst->parent = NULL; dst->state = 0; dst->ack_fwd = 0; dst->ack_rev = 0; dst->bucket = p->hashval; /* * The legacy userland code will interpret a NULL here as a marker * for the last dynamic rule. */ dst->next = (ipfw_dyn_rule *)1; } static void dyn_export_data(const struct dyn_data *data, uint16_t kidx, uint8_t type, ipfw_dyn_rule *dst) { dst->dyn_type = type; dst->kidx = kidx; dst->pcnt = data->pcnt_fwd + data->pcnt_rev; dst->bcnt = data->bcnt_fwd + data->bcnt_rev; dst->expire = TIME_LEQ(data->expire, time_uptime) ? 0: data->expire - time_uptime; /* 'rule' is used to pass up the rule number and set */ memcpy(&dst->rule, &data->rulenum, sizeof(data->rulenum)); /* store set number into high word of dst->rule pointer. */ memcpy((char *)&dst->rule + sizeof(data->rulenum), &data->set, sizeof(data->set)); /* unused fields */ dst->parent = NULL; dst->state = data->state; dst->ack_fwd = data->ack_fwd; dst->ack_rev = data->ack_rev; dst->count = 0; dst->bucket = data->hashval; /* * The legacy userland code will interpret a NULL here as a marker * for the last dynamic rule. */ dst->next = (ipfw_dyn_rule *)1; } static void dyn_export_ipv4_state(const struct dyn_ipv4_state *s, ipfw_dyn_rule *dst) { switch (s->type) { case O_LIMIT_PARENT: dyn_export_parent(s->limit, s->kidx, dst); break; default: dyn_export_data(s->data, s->kidx, s->type, dst); } dst->id.dst_ip = s->dst; dst->id.src_ip = s->src; dst->id.dst_port = s->dport; dst->id.src_port = s->sport; dst->id.fib = s->data->fibnum; dst->id.proto = s->proto; dst->id._flags = 0; dst->id.addr_type = 4; memset(&dst->id.dst_ip6, 0, sizeof(dst->id.dst_ip6)); memset(&dst->id.src_ip6, 0, sizeof(dst->id.src_ip6)); dst->id.flow_id6 = dst->id.extra = 0; } #ifdef INET6 static void dyn_export_ipv6_state(const struct dyn_ipv6_state *s, ipfw_dyn_rule *dst) { switch (s->type) { case O_LIMIT_PARENT: dyn_export_parent(s->limit, s->kidx, dst); break; default: dyn_export_data(s->data, s->kidx, s->type, dst); } dst->id.src_ip6 = s->src; dst->id.dst_ip6 = s->dst; dst->id.dst_port = s->dport; dst->id.src_port = s->sport; dst->id.fib = s->data->fibnum; dst->id.proto = s->proto; dst->id._flags = 0; dst->id.addr_type = 6; dst->id.dst_ip = dst->id.src_ip = 0; dst->id.flow_id6 = dst->id.extra = 0; } #endif /* INET6 */ /* * Fills the buffer given by @sd with dynamic states. * Used by dump format v1 (current). * * Returns 0 on success. */ int ipfw_dump_states(struct ip_fw_chain *chain, struct sockopt_data *sd) { #ifdef INET6 struct dyn_ipv6_state *s6; #endif struct dyn_ipv4_state *s4; ipfw_obj_dyntlv *dst, *last; ipfw_obj_ctlv *ctlv; uint32_t bucket; if (V_dyn_count == 0) return (0); /* * IPFW_UH_RLOCK garantees that another userland request * and callout thread will not delete entries from states * lists. */ IPFW_UH_RLOCK_ASSERT(chain); ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); if (ctlv == NULL) return (ENOMEM); ctlv->head.type = IPFW_TLV_DYNSTATE_LIST; ctlv->objsize = sizeof(ipfw_obj_dyntlv); last = NULL; #define DYN_EXPORT_STATES(s, af, h, b) \ CK_SLIST_FOREACH(s, &V_dyn_ ## h[b], entry) { \ dst = (ipfw_obj_dyntlv *)ipfw_get_sopt_space(sd, \ sizeof(ipfw_obj_dyntlv)); \ if (dst == NULL) \ return (ENOMEM); \ dyn_export_ ## af ## _state(s, &dst->state); \ dst->head.length = sizeof(ipfw_obj_dyntlv); \ dst->head.type = IPFW_TLV_DYN_ENT; \ last = dst; \ } for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket); DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket); #ifdef INET6 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket); DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket); #endif /* INET6 */ } /* mark last dynamic rule */ if (last != NULL) last->head.flags = IPFW_DF_LAST; /* XXX: unused */ return (0); #undef DYN_EXPORT_STATES } /* * Fill given buffer with dynamic states (legacy format). * IPFW_UH_RLOCK has to be held while calling. */ void ipfw_get_dynamic(struct ip_fw_chain *chain, char **pbp, const char *ep) { #ifdef INET6 struct dyn_ipv6_state *s6; #endif struct dyn_ipv4_state *s4; ipfw_dyn_rule *p, *last = NULL; char *bp; uint32_t bucket; if (V_dyn_count == 0) return; bp = *pbp; IPFW_UH_RLOCK_ASSERT(chain); #define DYN_EXPORT_STATES(s, af, head, b) \ CK_SLIST_FOREACH(s, &V_dyn_ ## head[b], entry) { \ if (bp + sizeof(*p) > ep) \ break; \ p = (ipfw_dyn_rule *)bp; \ dyn_export_ ## af ## _state(s, p); \ last = p; \ bp += sizeof(*p); \ } for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { DYN_EXPORT_STATES(s4, ipv4, ipv4_parent, bucket); DYN_EXPORT_STATES(s4, ipv4, ipv4, bucket); #ifdef INET6 DYN_EXPORT_STATES(s6, ipv6, ipv6_parent, bucket); DYN_EXPORT_STATES(s6, ipv6, ipv6, bucket); #endif /* INET6 */ } if (last != NULL) /* mark last dynamic rule */ last->next = NULL; *pbp = bp; #undef DYN_EXPORT_STATES } void ipfw_dyn_init(struct ip_fw_chain *chain) { #ifdef IPFIREWALL_JENKINSHASH V_dyn_hashseed = arc4random(); #endif V_dyn_max = 16384; /* max # of states */ V_dyn_parent_max = 4096; /* max # of parent states */ V_dyn_buckets_max = 8192; /* must be power of 2 */ V_dyn_ack_lifetime = 300; V_dyn_syn_lifetime = 20; V_dyn_fin_lifetime = 1; V_dyn_rst_lifetime = 1; V_dyn_udp_lifetime = 10; V_dyn_short_lifetime = 5; V_dyn_keepalive_interval = 20; V_dyn_keepalive_period = 5; V_dyn_keepalive = 1; /* send keepalives */ V_dyn_keepalive_last = time_uptime; V_dyn_data_zone = uma_zcreate("IPFW dynamic states data", sizeof(struct dyn_data), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_dyn_data_zone, V_dyn_max); V_dyn_parent_zone = uma_zcreate("IPFW parent dynamic states", sizeof(struct dyn_parent), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_dyn_parent_zone, V_dyn_parent_max); SLIST_INIT(&V_dyn_expired_ipv4); V_dyn_ipv4 = NULL; V_dyn_ipv4_parent = NULL; V_dyn_ipv4_zone = uma_zcreate("IPFW IPv4 dynamic states", sizeof(struct dyn_ipv4_state), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); #ifdef INET6 SLIST_INIT(&V_dyn_expired_ipv6); V_dyn_ipv6 = NULL; V_dyn_ipv6_parent = NULL; V_dyn_ipv6_zone = uma_zcreate("IPFW IPv6 dynamic states", sizeof(struct dyn_ipv6_state), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); #endif /* Initialize buckets. */ V_curr_dyn_buckets = 0; V_dyn_bucket_lock = NULL; dyn_grow_hashtable(chain, 256); if (IS_DEFAULT_VNET(curvnet)) dyn_hp_cache = malloc(mp_ncpus * sizeof(void *), M_IPFW, M_WAITOK | M_ZERO); DYN_EXPIRED_LOCK_INIT(); callout_init(&V_dyn_timeout, 1); callout_reset(&V_dyn_timeout, hz, dyn_tick, curvnet); IPFW_ADD_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); } void ipfw_dyn_uninit(int pass) { #ifdef INET6 struct dyn_ipv6_state *s6; #endif struct dyn_ipv4_state *s4; int bucket; if (pass == 0) { callout_drain(&V_dyn_timeout); return; } IPFW_DEL_OBJ_REWRITER(IS_DEFAULT_VNET(curvnet), dyn_opcodes); DYN_EXPIRED_LOCK_DESTROY(); #define DYN_FREE_STATES_FORCED(CK, s, af, name, en) do { \ while ((s = CK ## SLIST_FIRST(&V_dyn_ ## name)) != NULL) { \ CK ## SLIST_REMOVE_HEAD(&V_dyn_ ## name, en); \ if (s->type == O_LIMIT_PARENT) \ uma_zfree(V_dyn_parent_zone, s->limit); \ else \ uma_zfree(V_dyn_data_zone, s->data); \ uma_zfree(V_dyn_ ## af ## _zone, s); \ } \ } while (0) for (bucket = 0; bucket < V_curr_dyn_buckets; bucket++) { DYN_BUCKET_LOCK_DESTROY(V_dyn_bucket_lock, bucket); DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4[bucket], entry); DYN_FREE_STATES_FORCED(CK_, s4, ipv4, ipv4_parent[bucket], entry); #ifdef INET6 DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6[bucket], entry); DYN_FREE_STATES_FORCED(CK_, s6, ipv6, ipv6_parent[bucket], entry); #endif /* INET6 */ } DYN_FREE_STATES_FORCED(, s4, ipv4, expired_ipv4, expired); #ifdef INET6 DYN_FREE_STATES_FORCED(, s6, ipv6, expired_ipv6, expired); #endif #undef DYN_FREE_STATES_FORCED uma_zdestroy(V_dyn_ipv4_zone); uma_zdestroy(V_dyn_data_zone); uma_zdestroy(V_dyn_parent_zone); #ifdef INET6 uma_zdestroy(V_dyn_ipv6_zone); free(V_dyn_ipv6, M_IPFW); free(V_dyn_ipv6_parent, M_IPFW); free(V_dyn_ipv6_add, M_IPFW); free(V_dyn_ipv6_parent_add, M_IPFW); free(V_dyn_ipv6_del, M_IPFW); free(V_dyn_ipv6_parent_del, M_IPFW); #endif free(V_dyn_bucket_lock, M_IPFW); free(V_dyn_ipv4, M_IPFW); free(V_dyn_ipv4_parent, M_IPFW); free(V_dyn_ipv4_add, M_IPFW); free(V_dyn_ipv4_parent_add, M_IPFW); free(V_dyn_ipv4_del, M_IPFW); free(V_dyn_ipv4_parent_del, M_IPFW); if (IS_DEFAULT_VNET(curvnet)) free(dyn_hp_cache, M_IPFW); } Index: head/sys/netpfil/ipfw/ip_fw_pfil.c =================================================================== --- head/sys/netpfil/ipfw/ip_fw_pfil.c (revision 336675) +++ head/sys/netpfil/ipfw/ip_fw_pfil.c (revision 336676) @@ -1,582 +1,582 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2004 Andre Oppermann, Internet Business Solutions AG * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_ipfw.h" #include "opt_inet.h" #include "opt_inet6.h" #ifndef INET #error IPFIREWALL requires INET. #endif /* INET */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #endif #include #include #include -static VNET_DEFINE(int, fw_enable) = 1; +VNET_DEFINE_STATIC(int, fw_enable) = 1; #define V_fw_enable VNET(fw_enable) #ifdef INET6 -static VNET_DEFINE(int, fw6_enable) = 1; +VNET_DEFINE_STATIC(int, fw6_enable) = 1; #define V_fw6_enable VNET(fw6_enable) #endif -static VNET_DEFINE(int, fwlink_enable) = 0; +VNET_DEFINE_STATIC(int, fwlink_enable) = 0; #define V_fwlink_enable VNET(fwlink_enable) int ipfw_chg_hook(SYSCTL_HANDLER_ARGS); /* Forward declarations. */ static int ipfw_divert(struct mbuf **, int, struct ipfw_rule_ref *, int); int ipfw_check_packet(void *, struct mbuf **, struct ifnet *, int, struct inpcb *); int ipfw_check_frame(void *, struct mbuf **, struct ifnet *, int, struct inpcb *); #ifdef SYSCTL_NODE SYSBEGIN(f1) SYSCTL_DECL(_net_inet_ip_fw); SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw_enable), 0, ipfw_chg_hook, "I", "Enable ipfw"); #ifdef INET6 SYSCTL_DECL(_net_inet6_ip6_fw); SYSCTL_PROC(_net_inet6_ip6_fw, OID_AUTO, enable, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fw6_enable), 0, ipfw_chg_hook, "I", "Enable ipfw+6"); #endif /* INET6 */ SYSCTL_DECL(_net_link_ether); SYSCTL_PROC(_net_link_ether, OID_AUTO, ipfw, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &VNET_NAME(fwlink_enable), 0, ipfw_chg_hook, "I", "Pass ether pkts through firewall"); SYSEND #endif /* SYSCTL_NODE */ /* * The pfilter hook to pass packets to ipfw_chk and then to * dummynet, divert, netgraph or other modules. * The packet may be consumed. */ int ipfw_check_packet(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir, struct inpcb *inp) { struct ip_fw_args args; struct m_tag *tag; int ipfw; int ret; /* convert dir to IPFW values */ dir = (dir == PFIL_IN) ? DIR_IN : DIR_OUT; bzero(&args, sizeof(args)); again: /* * extract and remove the tag if present. If we are left * with onepass, optimize the outgoing path. */ tag = m_tag_locate(*m0, MTAG_IPFW_RULE, 0, NULL); if (tag != NULL) { args.rule = *((struct ipfw_rule_ref *)(tag+1)); m_tag_delete(*m0, tag); if (args.rule.info & IPFW_ONEPASS) return (0); } args.m = *m0; args.oif = dir == DIR_OUT ? ifp : NULL; args.inp = inp; ipfw = ipfw_chk(&args); *m0 = args.m; KASSERT(*m0 != NULL || ipfw == IP_FW_DENY, ("%s: m0 is NULL", __func__)); /* breaking out of the switch means drop */ ret = 0; /* default return value for pass */ switch (ipfw) { case IP_FW_PASS: /* next_hop may be set by ipfw_chk */ if (args.next_hop == NULL && args.next_hop6 == NULL) break; /* pass */ #if (!defined(INET6) && !defined(INET)) ret = EACCES; #else { struct m_tag *fwd_tag; size_t len; KASSERT(args.next_hop == NULL || args.next_hop6 == NULL, ("%s: both next_hop=%p and next_hop6=%p not NULL", __func__, args.next_hop, args.next_hop6)); #ifdef INET6 if (args.next_hop6 != NULL) len = sizeof(struct sockaddr_in6); #endif #ifdef INET if (args.next_hop != NULL) len = sizeof(struct sockaddr_in); #endif /* Incoming packets should not be tagged so we do not * m_tag_find. Outgoing packets may be tagged, so we * reuse the tag if present. */ fwd_tag = (dir == DIR_IN) ? NULL : m_tag_find(*m0, PACKET_TAG_IPFORWARD, NULL); if (fwd_tag != NULL) { m_tag_unlink(*m0, fwd_tag); } else { fwd_tag = m_tag_get(PACKET_TAG_IPFORWARD, len, M_NOWAIT); if (fwd_tag == NULL) { ret = EACCES; break; /* i.e. drop */ } } #ifdef INET6 if (args.next_hop6 != NULL) { struct sockaddr_in6 *sa6; sa6 = (struct sockaddr_in6 *)(fwd_tag + 1); bcopy(args.next_hop6, sa6, len); /* * If nh6 address is link-local we should convert * it to kernel internal form before doing any * comparisons. */ if (sa6_embedscope(sa6, V_ip6_use_defzone) != 0) { ret = EACCES; break; } if (in6_localip(&sa6->sin6_addr)) (*m0)->m_flags |= M_FASTFWD_OURS; (*m0)->m_flags |= M_IP6_NEXTHOP; } #endif #ifdef INET if (args.next_hop != NULL) { bcopy(args.next_hop, (fwd_tag+1), len); if (in_localip(args.next_hop->sin_addr)) (*m0)->m_flags |= M_FASTFWD_OURS; (*m0)->m_flags |= M_IP_NEXTHOP; } #endif m_tag_prepend(*m0, fwd_tag); } #endif /* INET || INET6 */ break; case IP_FW_DENY: ret = EACCES; break; /* i.e. drop */ case IP_FW_DUMMYNET: ret = EACCES; if (ip_dn_io_ptr == NULL) break; /* i.e. drop */ if (mtod(*m0, struct ip *)->ip_v == 4) ret = ip_dn_io_ptr(m0, dir, &args); else if (mtod(*m0, struct ip *)->ip_v == 6) ret = ip_dn_io_ptr(m0, dir | PROTO_IPV6, &args); else break; /* drop it */ /* * XXX should read the return value. * dummynet normally eats the packet and sets *m0=NULL * unless the packet can be sent immediately. In this * case args is updated and we should re-run the * check without clearing args. */ if (*m0 != NULL) goto again; break; case IP_FW_TEE: case IP_FW_DIVERT: if (ip_divert_ptr == NULL) { ret = EACCES; break; /* i.e. drop */ } ret = ipfw_divert(m0, dir, &args.rule, (ipfw == IP_FW_TEE) ? 1 : 0); /* continue processing for the original packet (tee). */ if (*m0) goto again; break; case IP_FW_NGTEE: case IP_FW_NETGRAPH: if (ng_ipfw_input_p == NULL) { ret = EACCES; break; /* i.e. drop */ } ret = ng_ipfw_input_p(m0, dir, &args, (ipfw == IP_FW_NGTEE) ? 1 : 0); if (ipfw == IP_FW_NGTEE) /* ignore errors for NGTEE */ goto again; /* continue with packet */ break; case IP_FW_NAT: /* honor one-pass in case of successful nat */ if (V_fw_one_pass) break; /* ret is already 0 */ goto again; case IP_FW_REASS: goto again; /* continue with packet */ default: KASSERT(0, ("%s: unknown retval", __func__)); } if (ret != 0) { if (*m0) FREE_PKT(*m0); *m0 = NULL; } return ret; } /* * ipfw processing for ethernet packets (in and out). */ int ipfw_check_frame(void *arg, struct mbuf **m0, struct ifnet *ifp, int dir, struct inpcb *inp) { struct ether_header *eh; struct ether_header save_eh; struct mbuf *m; int i, ret; struct ip_fw_args args; struct m_tag *mtag; /* fetch start point from rule, if any. remove the tag if present. */ mtag = m_tag_locate(*m0, MTAG_IPFW_RULE, 0, NULL); if (mtag == NULL) { args.rule.slot = 0; } else { args.rule = *((struct ipfw_rule_ref *)(mtag+1)); m_tag_delete(*m0, mtag); if (args.rule.info & IPFW_ONEPASS) return (0); } /* I need some amt of data to be contiguous */ m = *m0; i = min(m->m_pkthdr.len, max_protohdr); if (m->m_len < i) { m = m_pullup(m, i); if (m == NULL) { *m0 = m; return (0); } } eh = mtod(m, struct ether_header *); save_eh = *eh; /* save copy for restore below */ m_adj(m, ETHER_HDR_LEN); /* strip ethernet header */ args.m = m; /* the packet we are looking at */ args.oif = dir == PFIL_OUT ? ifp: NULL; /* destination, if any */ args.next_hop = NULL; /* we do not support forward yet */ args.next_hop6 = NULL; /* we do not support forward yet */ args.eh = &save_eh; /* MAC header for bridged/MAC packets */ args.inp = NULL; /* used by ipfw uid/gid/jail rules */ i = ipfw_chk(&args); m = args.m; if (m != NULL) { /* * Restore Ethernet header, as needed, in case the * mbuf chain was replaced by ipfw. */ M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT); if (m == NULL) { *m0 = NULL; return (0); } if (eh != mtod(m, struct ether_header *)) bcopy(&save_eh, mtod(m, struct ether_header *), ETHER_HDR_LEN); } *m0 = m; ret = 0; /* Check result of ipfw_chk() */ switch (i) { case IP_FW_PASS: break; case IP_FW_DENY: ret = EACCES; break; /* i.e. drop */ case IP_FW_DUMMYNET: ret = EACCES; if (ip_dn_io_ptr == NULL) break; /* i.e. drop */ *m0 = NULL; dir = (dir == PFIL_IN) ? DIR_IN : DIR_OUT; ip_dn_io_ptr(&m, dir | PROTO_LAYER2, &args); return 0; default: KASSERT(0, ("%s: unknown retval", __func__)); } if (ret != 0) { if (*m0) FREE_PKT(*m0); *m0 = NULL; } return ret; } /* do the divert, return 1 on error 0 on success */ static int ipfw_divert(struct mbuf **m0, int incoming, struct ipfw_rule_ref *rule, int tee) { /* * ipfw_chk() has already tagged the packet with the divert tag. * If tee is set, copy packet and return original. * If not tee, consume packet and send it to divert socket. */ struct mbuf *clone; struct ip *ip = mtod(*m0, struct ip *); struct m_tag *tag; /* Cloning needed for tee? */ if (tee == 0) { clone = *m0; /* use the original mbuf */ *m0 = NULL; } else { clone = m_dup(*m0, M_NOWAIT); /* If we cannot duplicate the mbuf, we sacrifice the divert * chain and continue with the tee-ed packet. */ if (clone == NULL) return 1; } /* * Divert listeners can normally handle non-fragmented packets, * but we can only reass in the non-tee case. * This means that listeners on a tee rule may get fragments, * and have to live with that. * Note that we now have the 'reass' ipfw option so if we care * we can do it before a 'tee'. */ if (!tee) switch (ip->ip_v) { case IPVERSION: if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) { int hlen; struct mbuf *reass; reass = ip_reass(clone); /* Reassemble packet. */ if (reass == NULL) return 0; /* not an error */ /* if reass = NULL then it was consumed by ip_reass */ /* * IP header checksum fixup after reassembly and leave header * in network byte order. */ ip = mtod(reass, struct ip *); hlen = ip->ip_hl << 2; ip->ip_sum = 0; if (hlen == sizeof(struct ip)) ip->ip_sum = in_cksum_hdr(ip); else ip->ip_sum = in_cksum(reass, hlen); clone = reass; } break; #ifdef INET6 case IPV6_VERSION >> 4: { struct ip6_hdr *const ip6 = mtod(clone, struct ip6_hdr *); if (ip6->ip6_nxt == IPPROTO_FRAGMENT) { int nxt, off; off = sizeof(struct ip6_hdr); nxt = frag6_input(&clone, &off, 0); if (nxt == IPPROTO_DONE) return (0); } break; } #endif } /* attach a tag to the packet with the reinject info */ tag = m_tag_alloc(MTAG_IPFW_RULE, 0, sizeof(struct ipfw_rule_ref), M_NOWAIT); if (tag == NULL) { FREE_PKT(clone); return 1; } *((struct ipfw_rule_ref *)(tag+1)) = *rule; m_tag_prepend(clone, tag); /* Do the dirty job... */ ip_divert_ptr(clone, incoming); return 0; } /* * attach or detach hooks for a given protocol family */ static int ipfw_hook(int onoff, int pf) { struct pfil_head *pfh; pfil_func_t hook_func; pfh = pfil_head_get(PFIL_TYPE_AF, pf); if (pfh == NULL) return ENOENT; hook_func = (pf == AF_LINK) ? ipfw_check_frame : ipfw_check_packet; (void) (onoff ? pfil_add_hook : pfil_remove_hook) (hook_func, NULL, PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh); return 0; } int ipfw_attach_hooks(int arg) { int error = 0; if (arg == 0) /* detach */ ipfw_hook(0, AF_INET); else if (V_fw_enable && ipfw_hook(1, AF_INET) != 0) { error = ENOENT; /* see ip_fw_pfil.c::ipfw_hook() */ printf("ipfw_hook() error\n"); } #ifdef INET6 if (arg == 0) /* detach */ ipfw_hook(0, AF_INET6); else if (V_fw6_enable && ipfw_hook(1, AF_INET6) != 0) { error = ENOENT; printf("ipfw6_hook() error\n"); } #endif if (arg == 0) /* detach */ ipfw_hook(0, AF_LINK); else if (V_fwlink_enable && ipfw_hook(1, AF_LINK) != 0) { error = ENOENT; printf("ipfw_link_hook() error\n"); } return error; } int ipfw_chg_hook(SYSCTL_HANDLER_ARGS) { int newval; int error; int af; if (arg1 == &V_fw_enable) af = AF_INET; #ifdef INET6 else if (arg1 == &V_fw6_enable) af = AF_INET6; #endif else if (arg1 == &V_fwlink_enable) af = AF_LINK; else return (EINVAL); newval = *(int *)arg1; /* Handle sysctl change */ error = sysctl_handle_int(oidp, &newval, 0, req); if (error) return (error); /* Formalize new value */ newval = (newval) ? 1 : 0; if (*(int *)arg1 == newval) return (0); error = ipfw_hook(newval, af); if (error) return (error); *(int *)arg1 = newval; return (0); } /* end of file */ Index: head/sys/netpfil/ipfw/ip_fw_sockopt.c =================================================================== --- head/sys/netpfil/ipfw/ip_fw_sockopt.c (revision 336675) +++ head/sys/netpfil/ipfw/ip_fw_sockopt.c (revision 336676) @@ -1,4652 +1,4652 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2009 Luigi Rizzo, Universita` di Pisa * Copyright (c) 2014 Yandex LLC * Copyright (c) 2014 Alexander V. Chernikov * * Supported by: Valeria Paoli * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * Control socket and rule management routines for ipfw. * Control is currently implemented via IP_FW3 setsockopt() code. */ #include "opt_ipfw.h" #include "opt_inet.h" #ifndef INET #error IPFIREWALL requires INET. #endif /* INET */ #include "opt_inet6.h" #include #include #include #include /* struct m_tag used by nested headers */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* hooks */ #include #include #include #ifdef MAC #include #endif static int ipfw_ctl(struct sockopt *sopt); static int check_ipfw_rule_body(ipfw_insn *cmd, int cmd_len, struct rule_check_info *ci); static int check_ipfw_rule1(struct ip_fw_rule *rule, int size, struct rule_check_info *ci); static int check_ipfw_rule0(struct ip_fw_rule0 *rule, int size, struct rule_check_info *ci); static int rewrite_rule_uidx(struct ip_fw_chain *chain, struct rule_check_info *ci); #define NAMEDOBJ_HASH_SIZE 32 struct namedobj_instance { struct namedobjects_head *names; struct namedobjects_head *values; uint32_t nn_size; /* names hash size */ uint32_t nv_size; /* number hash size */ u_long *idx_mask; /* used items bitmask */ uint32_t max_blocks; /* number of "long" blocks in bitmask */ uint32_t count; /* number of items */ uint16_t free_off[IPFW_MAX_SETS]; /* first possible free offset */ objhash_hash_f *hash_f; objhash_cmp_f *cmp_f; }; #define BLOCK_ITEMS (8 * sizeof(u_long)) /* Number of items for ffsl() */ static uint32_t objhash_hash_name(struct namedobj_instance *ni, const void *key, uint32_t kopt); static uint32_t objhash_hash_idx(struct namedobj_instance *ni, uint32_t val); static int objhash_cmp_name(struct named_object *no, const void *name, uint32_t set); MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's"); static int dump_config(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int add_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int del_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int clear_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int move_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int manage_sets(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int dump_soptcodes(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); static int dump_srvobjects(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd); /* ctl3 handler data */ struct mtx ctl3_lock; #define CTL3_LOCK_INIT() mtx_init(&ctl3_lock, "ctl3_lock", NULL, MTX_DEF) #define CTL3_LOCK_DESTROY() mtx_destroy(&ctl3_lock) #define CTL3_LOCK() mtx_lock(&ctl3_lock) #define CTL3_UNLOCK() mtx_unlock(&ctl3_lock) static struct ipfw_sopt_handler *ctl3_handlers; static size_t ctl3_hsize; static uint64_t ctl3_refct, ctl3_gencnt; #define CTL3_SMALLBUF 4096 /* small page-size write buffer */ #define CTL3_LARGEBUF 16 * 1024 * 1024 /* handle large rulesets */ static int ipfw_flush_sopt_data(struct sockopt_data *sd); static struct ipfw_sopt_handler scodes[] = { { IP_FW_XGET, 0, HDIR_GET, dump_config }, { IP_FW_XADD, 0, HDIR_BOTH, add_rules }, { IP_FW_XDEL, 0, HDIR_BOTH, del_rules }, { IP_FW_XZERO, 0, HDIR_SET, clear_rules }, { IP_FW_XRESETLOG, 0, HDIR_SET, clear_rules }, { IP_FW_XMOVE, 0, HDIR_SET, move_rules }, { IP_FW_SET_SWAP, 0, HDIR_SET, manage_sets }, { IP_FW_SET_MOVE, 0, HDIR_SET, manage_sets }, { IP_FW_SET_ENABLE, 0, HDIR_SET, manage_sets }, { IP_FW_DUMP_SOPTCODES, 0, HDIR_GET, dump_soptcodes }, { IP_FW_DUMP_SRVOBJECTS,0, HDIR_GET, dump_srvobjects }, }; static int set_legacy_obj_kidx(struct ip_fw_chain *ch, struct ip_fw_rule0 *rule); static struct opcode_obj_rewrite *find_op_rw(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype); static int mark_object_kidx(struct ip_fw_chain *ch, struct ip_fw *rule, uint32_t *bmask); static int ref_rule_objects(struct ip_fw_chain *ch, struct ip_fw *rule, struct rule_check_info *ci, struct obj_idx *oib, struct tid_info *ti); static int ref_opcode_object(struct ip_fw_chain *ch, ipfw_insn *cmd, struct tid_info *ti, struct obj_idx *pidx, int *unresolved); static void unref_rule_objects(struct ip_fw_chain *chain, struct ip_fw *rule); static void unref_oib_objects(struct ip_fw_chain *ch, ipfw_insn *cmd, struct obj_idx *oib, struct obj_idx *end); static int export_objhash_ntlv(struct namedobj_instance *ni, uint16_t kidx, struct sockopt_data *sd); /* * Opcode object rewriter variables */ struct opcode_obj_rewrite *ctl3_rewriters; static size_t ctl3_rsize; /* * static variables followed by global ones */ -static VNET_DEFINE(uma_zone_t, ipfw_cntr_zone); +VNET_DEFINE_STATIC(uma_zone_t, ipfw_cntr_zone); #define V_ipfw_cntr_zone VNET(ipfw_cntr_zone) void ipfw_init_counters() { V_ipfw_cntr_zone = uma_zcreate("IPFW counters", IPFW_RULE_CNTR_SIZE, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); } void ipfw_destroy_counters() { uma_zdestroy(V_ipfw_cntr_zone); } struct ip_fw * ipfw_alloc_rule(struct ip_fw_chain *chain, size_t rulesize) { struct ip_fw *rule; rule = malloc(rulesize, M_IPFW, M_WAITOK | M_ZERO); rule->cntr = uma_zalloc_pcpu(V_ipfw_cntr_zone, M_WAITOK | M_ZERO); return (rule); } static void free_rule(struct ip_fw *rule) { uma_zfree_pcpu(V_ipfw_cntr_zone, rule->cntr); free(rule, M_IPFW); } /* * Find the smallest rule >= key, id. * We could use bsearch but it is so simple that we code it directly */ int ipfw_find_rule(struct ip_fw_chain *chain, uint32_t key, uint32_t id) { int i, lo, hi; struct ip_fw *r; for (lo = 0, hi = chain->n_rules - 1; lo < hi;) { i = (lo + hi) / 2; r = chain->map[i]; if (r->rulenum < key) lo = i + 1; /* continue from the next one */ else if (r->rulenum > key) hi = i; /* this might be good */ else if (r->id < id) lo = i + 1; /* continue from the next one */ else /* r->id >= id */ hi = i; /* this might be good */ } return hi; } /* * Builds skipto cache on rule set @map. */ static void update_skipto_cache(struct ip_fw_chain *chain, struct ip_fw **map) { int *smap, rulenum; int i, mi; IPFW_UH_WLOCK_ASSERT(chain); mi = 0; rulenum = map[mi]->rulenum; smap = chain->idxmap_back; if (smap == NULL) return; for (i = 0; i < 65536; i++) { smap[i] = mi; /* Use the same rule index until i < rulenum */ if (i != rulenum || i == 65535) continue; /* Find next rule with num > i */ rulenum = map[++mi]->rulenum; while (rulenum == i) rulenum = map[++mi]->rulenum; } } /* * Swaps prepared (backup) index with current one. */ static void swap_skipto_cache(struct ip_fw_chain *chain) { int *map; IPFW_UH_WLOCK_ASSERT(chain); IPFW_WLOCK_ASSERT(chain); map = chain->idxmap; chain->idxmap = chain->idxmap_back; chain->idxmap_back = map; } /* * Allocate and initialize skipto cache. */ void ipfw_init_skipto_cache(struct ip_fw_chain *chain) { int *idxmap, *idxmap_back; idxmap = malloc(65536 * sizeof(int), M_IPFW, M_WAITOK | M_ZERO); idxmap_back = malloc(65536 * sizeof(int), M_IPFW, M_WAITOK); /* * Note we may be called at any time after initialization, * for example, on first skipto rule, so we need to * provide valid chain->idxmap on return */ IPFW_UH_WLOCK(chain); if (chain->idxmap != NULL) { IPFW_UH_WUNLOCK(chain); free(idxmap, M_IPFW); free(idxmap_back, M_IPFW); return; } /* Set backup pointer first to permit building cache */ chain->idxmap_back = idxmap_back; update_skipto_cache(chain, chain->map); IPFW_WLOCK(chain); /* It is now safe to set chain->idxmap ptr */ chain->idxmap = idxmap; swap_skipto_cache(chain); IPFW_WUNLOCK(chain); IPFW_UH_WUNLOCK(chain); } /* * Destroys skipto cache. */ void ipfw_destroy_skipto_cache(struct ip_fw_chain *chain) { if (chain->idxmap != NULL) free(chain->idxmap, M_IPFW); if (chain->idxmap != NULL) free(chain->idxmap_back, M_IPFW); } /* * allocate a new map, returns the chain locked. extra is the number * of entries to add or delete. */ static struct ip_fw ** get_map(struct ip_fw_chain *chain, int extra, int locked) { for (;;) { struct ip_fw **map; u_int i, mflags; mflags = M_ZERO | ((locked != 0) ? M_NOWAIT : M_WAITOK); i = chain->n_rules + extra; map = malloc(i * sizeof(struct ip_fw *), M_IPFW, mflags); if (map == NULL) { printf("%s: cannot allocate map\n", __FUNCTION__); return NULL; } if (!locked) IPFW_UH_WLOCK(chain); if (i >= chain->n_rules + extra) /* good */ return map; /* otherwise we lost the race, free and retry */ if (!locked) IPFW_UH_WUNLOCK(chain); free(map, M_IPFW); } } /* * swap the maps. It is supposed to be called with IPFW_UH_WLOCK */ static struct ip_fw ** swap_map(struct ip_fw_chain *chain, struct ip_fw **new_map, int new_len) { struct ip_fw **old_map; IPFW_WLOCK(chain); chain->id++; chain->n_rules = new_len; old_map = chain->map; chain->map = new_map; swap_skipto_cache(chain); IPFW_WUNLOCK(chain); return old_map; } static void export_cntr1_base(struct ip_fw *krule, struct ip_fw_bcounter *cntr) { struct timeval boottime; cntr->size = sizeof(*cntr); if (krule->cntr != NULL) { cntr->pcnt = counter_u64_fetch(krule->cntr); cntr->bcnt = counter_u64_fetch(krule->cntr + 1); cntr->timestamp = krule->timestamp; } if (cntr->timestamp > 0) { getboottime(&boottime); cntr->timestamp += boottime.tv_sec; } } static void export_cntr0_base(struct ip_fw *krule, struct ip_fw_bcounter0 *cntr) { struct timeval boottime; if (krule->cntr != NULL) { cntr->pcnt = counter_u64_fetch(krule->cntr); cntr->bcnt = counter_u64_fetch(krule->cntr + 1); cntr->timestamp = krule->timestamp; } if (cntr->timestamp > 0) { getboottime(&boottime); cntr->timestamp += boottime.tv_sec; } } /* * Copies rule @urule from v1 userland format (current). * to kernel @krule. * Assume @krule is zeroed. */ static void import_rule1(struct rule_check_info *ci) { struct ip_fw_rule *urule; struct ip_fw *krule; urule = (struct ip_fw_rule *)ci->urule; krule = (struct ip_fw *)ci->krule; /* copy header */ krule->act_ofs = urule->act_ofs; krule->cmd_len = urule->cmd_len; krule->rulenum = urule->rulenum; krule->set = urule->set; krule->flags = urule->flags; /* Save rulenum offset */ ci->urule_numoff = offsetof(struct ip_fw_rule, rulenum); /* Copy opcodes */ memcpy(krule->cmd, urule->cmd, krule->cmd_len * sizeof(uint32_t)); } /* * Export rule into v1 format (Current). * Layout: * [ ipfw_obj_tlv(IPFW_TLV_RULE_ENT) * [ ip_fw_rule ] OR * [ ip_fw_bcounter ip_fw_rule] (depends on rcntrs). * ] * Assume @data is zeroed. */ static void export_rule1(struct ip_fw *krule, caddr_t data, int len, int rcntrs) { struct ip_fw_bcounter *cntr; struct ip_fw_rule *urule; ipfw_obj_tlv *tlv; /* Fill in TLV header */ tlv = (ipfw_obj_tlv *)data; tlv->type = IPFW_TLV_RULE_ENT; tlv->length = len; if (rcntrs != 0) { /* Copy counters */ cntr = (struct ip_fw_bcounter *)(tlv + 1); urule = (struct ip_fw_rule *)(cntr + 1); export_cntr1_base(krule, cntr); } else urule = (struct ip_fw_rule *)(tlv + 1); /* copy header */ urule->act_ofs = krule->act_ofs; urule->cmd_len = krule->cmd_len; urule->rulenum = krule->rulenum; urule->set = krule->set; urule->flags = krule->flags; urule->id = krule->id; /* Copy opcodes */ memcpy(urule->cmd, krule->cmd, krule->cmd_len * sizeof(uint32_t)); } /* * Copies rule @urule from FreeBSD8 userland format (v0) * to kernel @krule. * Assume @krule is zeroed. */ static void import_rule0(struct rule_check_info *ci) { struct ip_fw_rule0 *urule; struct ip_fw *krule; int cmdlen, l; ipfw_insn *cmd; ipfw_insn_limit *lcmd; ipfw_insn_if *cmdif; urule = (struct ip_fw_rule0 *)ci->urule; krule = (struct ip_fw *)ci->krule; /* copy header */ krule->act_ofs = urule->act_ofs; krule->cmd_len = urule->cmd_len; krule->rulenum = urule->rulenum; krule->set = urule->set; if ((urule->_pad & 1) != 0) krule->flags |= IPFW_RULE_NOOPT; /* Save rulenum offset */ ci->urule_numoff = offsetof(struct ip_fw_rule0, rulenum); /* Copy opcodes */ memcpy(krule->cmd, urule->cmd, krule->cmd_len * sizeof(uint32_t)); /* * Alter opcodes: * 1) convert tablearg value from 65535 to 0 * 2) Add high bit to O_SETFIB/O_SETDSCP values (to make room * for targ). * 3) convert table number in iface opcodes to u16 * 4) convert old `nat global` into new 65535 */ l = krule->cmd_len; cmd = krule->cmd; cmdlen = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); switch (cmd->opcode) { /* Opcodes supporting tablearg */ case O_TAG: case O_TAGGED: case O_PIPE: case O_QUEUE: case O_DIVERT: case O_TEE: case O_SKIPTO: case O_CALLRETURN: case O_NETGRAPH: case O_NGTEE: case O_NAT: if (cmd->arg1 == IP_FW_TABLEARG) cmd->arg1 = IP_FW_TARG; else if (cmd->arg1 == 0) cmd->arg1 = IP_FW_NAT44_GLOBAL; break; case O_SETFIB: case O_SETDSCP: if (cmd->arg1 == IP_FW_TABLEARG) cmd->arg1 = IP_FW_TARG; else cmd->arg1 |= 0x8000; break; case O_LIMIT: lcmd = (ipfw_insn_limit *)cmd; if (lcmd->conn_limit == IP_FW_TABLEARG) lcmd->conn_limit = IP_FW_TARG; break; /* Interface tables */ case O_XMIT: case O_RECV: case O_VIA: /* Interface table, possibly */ cmdif = (ipfw_insn_if *)cmd; if (cmdif->name[0] != '\1') break; cmdif->p.kidx = (uint16_t)cmdif->p.glob; break; } } } /* * Copies rule @krule from kernel to FreeBSD8 userland format (v0) */ static void export_rule0(struct ip_fw *krule, struct ip_fw_rule0 *urule, int len) { int cmdlen, l; ipfw_insn *cmd; ipfw_insn_limit *lcmd; ipfw_insn_if *cmdif; /* copy header */ memset(urule, 0, len); urule->act_ofs = krule->act_ofs; urule->cmd_len = krule->cmd_len; urule->rulenum = krule->rulenum; urule->set = krule->set; if ((krule->flags & IPFW_RULE_NOOPT) != 0) urule->_pad |= 1; /* Copy opcodes */ memcpy(urule->cmd, krule->cmd, krule->cmd_len * sizeof(uint32_t)); /* Export counters */ export_cntr0_base(krule, (struct ip_fw_bcounter0 *)&urule->pcnt); /* * Alter opcodes: * 1) convert tablearg value from 0 to 65535 * 2) Remove highest bit from O_SETFIB/O_SETDSCP values. * 3) convert table number in iface opcodes to int */ l = urule->cmd_len; cmd = urule->cmd; cmdlen = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); switch (cmd->opcode) { /* Opcodes supporting tablearg */ case O_TAG: case O_TAGGED: case O_PIPE: case O_QUEUE: case O_DIVERT: case O_TEE: case O_SKIPTO: case O_CALLRETURN: case O_NETGRAPH: case O_NGTEE: case O_NAT: if (cmd->arg1 == IP_FW_TARG) cmd->arg1 = IP_FW_TABLEARG; else if (cmd->arg1 == IP_FW_NAT44_GLOBAL) cmd->arg1 = 0; break; case O_SETFIB: case O_SETDSCP: if (cmd->arg1 == IP_FW_TARG) cmd->arg1 = IP_FW_TABLEARG; else cmd->arg1 &= ~0x8000; break; case O_LIMIT: lcmd = (ipfw_insn_limit *)cmd; if (lcmd->conn_limit == IP_FW_TARG) lcmd->conn_limit = IP_FW_TABLEARG; break; /* Interface tables */ case O_XMIT: case O_RECV: case O_VIA: /* Interface table, possibly */ cmdif = (ipfw_insn_if *)cmd; if (cmdif->name[0] != '\1') break; cmdif->p.glob = cmdif->p.kidx; break; } } } /* * Add new rule(s) to the list possibly creating rule number for each. * Update the rule_number in the input struct so the caller knows it as well. * Must be called without IPFW_UH held */ static int commit_rules(struct ip_fw_chain *chain, struct rule_check_info *rci, int count) { int error, i, insert_before, tcount; uint16_t rulenum, *pnum; struct rule_check_info *ci; struct ip_fw *krule; struct ip_fw **map; /* the new array of pointers */ /* Check if we need to do table/obj index remap */ tcount = 0; for (ci = rci, i = 0; i < count; ci++, i++) { if (ci->object_opcodes == 0) continue; /* * Rule has some object opcodes. * We need to find (and create non-existing) * kernel objects, and reference existing ones. */ error = rewrite_rule_uidx(chain, ci); if (error != 0) { /* * rewrite failed, state for current rule * has been reverted. Check if we need to * revert more. */ if (tcount > 0) { /* * We have some more table rules * we need to rollback. */ IPFW_UH_WLOCK(chain); while (ci != rci) { ci--; if (ci->object_opcodes == 0) continue; unref_rule_objects(chain,ci->krule); } IPFW_UH_WUNLOCK(chain); } return (error); } tcount++; } /* get_map returns with IPFW_UH_WLOCK if successful */ map = get_map(chain, count, 0 /* not locked */); if (map == NULL) { if (tcount > 0) { /* Unbind tables */ IPFW_UH_WLOCK(chain); for (ci = rci, i = 0; i < count; ci++, i++) { if (ci->object_opcodes == 0) continue; unref_rule_objects(chain, ci->krule); } IPFW_UH_WUNLOCK(chain); } return (ENOSPC); } if (V_autoinc_step < 1) V_autoinc_step = 1; else if (V_autoinc_step > 1000) V_autoinc_step = 1000; /* FIXME: Handle count > 1 */ ci = rci; krule = ci->krule; rulenum = krule->rulenum; /* find the insertion point, we will insert before */ insert_before = rulenum ? rulenum + 1 : IPFW_DEFAULT_RULE; i = ipfw_find_rule(chain, insert_before, 0); /* duplicate first part */ if (i > 0) bcopy(chain->map, map, i * sizeof(struct ip_fw *)); map[i] = krule; /* duplicate remaining part, we always have the default rule */ bcopy(chain->map + i, map + i + 1, sizeof(struct ip_fw *) *(chain->n_rules - i)); if (rulenum == 0) { /* Compute rule number and write it back */ rulenum = i > 0 ? map[i-1]->rulenum : 0; if (rulenum < IPFW_DEFAULT_RULE - V_autoinc_step) rulenum += V_autoinc_step; krule->rulenum = rulenum; /* Save number to userland rule */ pnum = (uint16_t *)((caddr_t)ci->urule + ci->urule_numoff); *pnum = rulenum; } krule->id = chain->id + 1; update_skipto_cache(chain, map); map = swap_map(chain, map, chain->n_rules + 1); chain->static_len += RULEUSIZE0(krule); IPFW_UH_WUNLOCK(chain); if (map) free(map, M_IPFW); return (0); } int ipfw_add_protected_rule(struct ip_fw_chain *chain, struct ip_fw *rule, int locked) { struct ip_fw **map; map = get_map(chain, 1, locked); if (map == NULL) return (ENOMEM); if (chain->n_rules > 0) bcopy(chain->map, map, chain->n_rules * sizeof(struct ip_fw *)); map[chain->n_rules] = rule; rule->rulenum = IPFW_DEFAULT_RULE; rule->set = RESVD_SET; rule->id = chain->id + 1; /* We add rule in the end of chain, no need to update skipto cache */ map = swap_map(chain, map, chain->n_rules + 1); chain->static_len += RULEUSIZE0(rule); IPFW_UH_WUNLOCK(chain); free(map, M_IPFW); return (0); } /* * Adds @rule to the list of rules to reap */ void ipfw_reap_add(struct ip_fw_chain *chain, struct ip_fw **head, struct ip_fw *rule) { IPFW_UH_WLOCK_ASSERT(chain); /* Unlink rule from everywhere */ unref_rule_objects(chain, rule); *((struct ip_fw **)rule) = *head; *head = rule; } /* * Reclaim storage associated with a list of rules. This is * typically the list created using remove_rule. * A NULL pointer on input is handled correctly. */ void ipfw_reap_rules(struct ip_fw *head) { struct ip_fw *rule; while ((rule = head) != NULL) { head = *((struct ip_fw **)head); free_rule(rule); } } /* * Rules to keep are * (default || reserved || !match_set || !match_number) * where * default ::= (rule->rulenum == IPFW_DEFAULT_RULE) * // the default rule is always protected * * reserved ::= (cmd == 0 && n == 0 && rule->set == RESVD_SET) * // RESVD_SET is protected only if cmd == 0 and n == 0 ("ipfw flush") * * match_set ::= (cmd == 0 || rule->set == set) * // set number is ignored for cmd == 0 * * match_number ::= (cmd == 1 || n == 0 || n == rule->rulenum) * // number is ignored for cmd == 1 or n == 0 * */ int ipfw_match_range(struct ip_fw *rule, ipfw_range_tlv *rt) { /* Don't match default rule for modification queries */ if (rule->rulenum == IPFW_DEFAULT_RULE && (rt->flags & IPFW_RCFLAG_DEFAULT) == 0) return (0); /* Don't match rules in reserved set for flush requests */ if ((rt->flags & IPFW_RCFLAG_ALL) != 0 && rule->set == RESVD_SET) return (0); /* If we're filtering by set, don't match other sets */ if ((rt->flags & IPFW_RCFLAG_SET) != 0 && rule->set != rt->set) return (0); if ((rt->flags & IPFW_RCFLAG_RANGE) != 0 && (rule->rulenum < rt->start_rule || rule->rulenum > rt->end_rule)) return (0); return (1); } struct manage_sets_args { uint16_t set; uint8_t new_set; }; static int swap_sets_cb(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct manage_sets_args *args; args = (struct manage_sets_args *)arg; if (no->set == (uint8_t)args->set) no->set = args->new_set; else if (no->set == args->new_set) no->set = (uint8_t)args->set; return (0); } static int move_sets_cb(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct manage_sets_args *args; args = (struct manage_sets_args *)arg; if (no->set == (uint8_t)args->set) no->set = args->new_set; return (0); } static int test_sets_cb(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct manage_sets_args *args; args = (struct manage_sets_args *)arg; if (no->set != (uint8_t)args->set) return (0); if (ipfw_objhash_lookup_name_type(ni, args->new_set, no->etlv, no->name) != NULL) return (EEXIST); return (0); } /* * Generic function to handler moving and swapping sets. */ int ipfw_obj_manage_sets(struct namedobj_instance *ni, uint16_t type, uint16_t set, uint8_t new_set, enum ipfw_sets_cmd cmd) { struct manage_sets_args args; struct named_object *no; args.set = set; args.new_set = new_set; switch (cmd) { case SWAP_ALL: return (ipfw_objhash_foreach_type(ni, swap_sets_cb, &args, type)); case TEST_ALL: return (ipfw_objhash_foreach_type(ni, test_sets_cb, &args, type)); case MOVE_ALL: return (ipfw_objhash_foreach_type(ni, move_sets_cb, &args, type)); case COUNT_ONE: /* * @set used to pass kidx. * When @new_set is zero - reset object counter, * otherwise increment it. */ no = ipfw_objhash_lookup_kidx(ni, set); if (new_set != 0) no->ocnt++; else no->ocnt = 0; return (0); case TEST_ONE: /* @set used to pass kidx */ no = ipfw_objhash_lookup_kidx(ni, set); /* * First check number of references: * when it differs, this mean other rules are holding * reference to given object, so it is not possible to * change its set. Note that refcnt may account references * to some going-to-be-added rules. Since we don't know * their numbers (and even if they will be added) it is * perfectly OK to return error here. */ if (no->ocnt != no->refcnt) return (EBUSY); if (ipfw_objhash_lookup_name_type(ni, new_set, type, no->name) != NULL) return (EEXIST); return (0); case MOVE_ONE: /* @set used to pass kidx */ no = ipfw_objhash_lookup_kidx(ni, set); no->set = new_set; return (0); } return (EINVAL); } /* * Delete rules matching range @rt. * Saves number of deleted rules in @ndel. * * Returns 0 on success. */ static int delete_range(struct ip_fw_chain *chain, ipfw_range_tlv *rt, int *ndel) { struct ip_fw *reap, *rule, **map; int end, start; int i, n, ndyn, ofs; reap = NULL; IPFW_UH_WLOCK(chain); /* arbitrate writers */ /* * Stage 1: Determine range to inspect. * Range is half-inclusive, e.g [start, end). */ start = 0; end = chain->n_rules - 1; if ((rt->flags & IPFW_RCFLAG_RANGE) != 0) { start = ipfw_find_rule(chain, rt->start_rule, 0); if (rt->end_rule >= IPFW_DEFAULT_RULE) rt->end_rule = IPFW_DEFAULT_RULE - 1; end = ipfw_find_rule(chain, rt->end_rule, UINT32_MAX); } /* Allocate new map of the same size */ map = get_map(chain, 0, 1 /* locked */); if (map == NULL) { IPFW_UH_WUNLOCK(chain); return (ENOMEM); } n = 0; ndyn = 0; ofs = start; /* 1. bcopy the initial part of the map */ if (start > 0) bcopy(chain->map, map, start * sizeof(struct ip_fw *)); /* 2. copy active rules between start and end */ for (i = start; i < end; i++) { rule = chain->map[i]; if (ipfw_match_range(rule, rt) == 0) { map[ofs++] = rule; continue; } n++; if (ipfw_is_dyn_rule(rule) != 0) ndyn++; } /* 3. copy the final part of the map */ bcopy(chain->map + end, map + ofs, (chain->n_rules - end) * sizeof(struct ip_fw *)); /* 4. recalculate skipto cache */ update_skipto_cache(chain, map); /* 5. swap the maps (under UH_WLOCK + WHLOCK) */ map = swap_map(chain, map, chain->n_rules - n); /* 6. Remove all dynamic states originated by deleted rules */ if (ndyn > 0) ipfw_expire_dyn_states(chain, rt); /* 7. now remove the rules deleted from the old map */ for (i = start; i < end; i++) { rule = map[i]; if (ipfw_match_range(rule, rt) == 0) continue; chain->static_len -= RULEUSIZE0(rule); ipfw_reap_add(chain, &reap, rule); } IPFW_UH_WUNLOCK(chain); ipfw_reap_rules(reap); if (map != NULL) free(map, M_IPFW); *ndel = n; return (0); } static int move_objects(struct ip_fw_chain *ch, ipfw_range_tlv *rt) { struct opcode_obj_rewrite *rw; struct ip_fw *rule; ipfw_insn *cmd; int cmdlen, i, l, c; uint16_t kidx; IPFW_UH_WLOCK_ASSERT(ch); /* Stage 1: count number of references by given rules */ for (c = 0, i = 0; i < ch->n_rules - 1; i++) { rule = ch->map[i]; if (ipfw_match_range(rule, rt) == 0) continue; if (rule->set == rt->new_set) /* nothing to do */ continue; /* Search opcodes with named objects */ for (l = rule->cmd_len, cmdlen = 0, cmd = rule->cmd; l > 0; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); rw = find_op_rw(cmd, &kidx, NULL); if (rw == NULL || rw->manage_sets == NULL) continue; /* * When manage_sets() returns non-zero value to * COUNT_ONE command, consider this as an object * doesn't support sets (e.g. disabled with sysctl). * So, skip checks for this object. */ if (rw->manage_sets(ch, kidx, 1, COUNT_ONE) != 0) continue; c++; } } if (c == 0) /* No objects found */ return (0); /* Stage 2: verify "ownership" */ for (c = 0, i = 0; (i < ch->n_rules - 1) && c == 0; i++) { rule = ch->map[i]; if (ipfw_match_range(rule, rt) == 0) continue; if (rule->set == rt->new_set) /* nothing to do */ continue; /* Search opcodes with named objects */ for (l = rule->cmd_len, cmdlen = 0, cmd = rule->cmd; l > 0 && c == 0; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); rw = find_op_rw(cmd, &kidx, NULL); if (rw == NULL || rw->manage_sets == NULL) continue; /* Test for ownership and conflicting names */ c = rw->manage_sets(ch, kidx, (uint8_t)rt->new_set, TEST_ONE); } } /* Stage 3: change set and cleanup */ for (i = 0; i < ch->n_rules - 1; i++) { rule = ch->map[i]; if (ipfw_match_range(rule, rt) == 0) continue; if (rule->set == rt->new_set) /* nothing to do */ continue; /* Search opcodes with named objects */ for (l = rule->cmd_len, cmdlen = 0, cmd = rule->cmd; l > 0; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); rw = find_op_rw(cmd, &kidx, NULL); if (rw == NULL || rw->manage_sets == NULL) continue; /* cleanup object counter */ rw->manage_sets(ch, kidx, 0 /* reset counter */, COUNT_ONE); if (c != 0) continue; /* change set */ rw->manage_sets(ch, kidx, (uint8_t)rt->new_set, MOVE_ONE); } } return (c); }/* * Changes set of given rule rannge @rt * with each other. * * Returns 0 on success. */ static int move_range(struct ip_fw_chain *chain, ipfw_range_tlv *rt) { struct ip_fw *rule; int i; IPFW_UH_WLOCK(chain); /* * Move rules with matching paramenerts to a new set. * This one is much more complex. We have to ensure * that all referenced tables (if any) are referenced * by given rule subset only. Otherwise, we can't move * them to new set and have to return error. */ if ((i = move_objects(chain, rt)) != 0) { IPFW_UH_WUNLOCK(chain); return (i); } /* XXX: We have to do swap holding WLOCK */ for (i = 0; i < chain->n_rules; i++) { rule = chain->map[i]; if (ipfw_match_range(rule, rt) == 0) continue; rule->set = rt->new_set; } IPFW_UH_WUNLOCK(chain); return (0); } /* * Clear counters for a specific rule. * Normally run under IPFW_UH_RLOCK, but these are idempotent ops * so we only care that rules do not disappear. */ static void clear_counters(struct ip_fw *rule, int log_only) { ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule); if (log_only == 0) IPFW_ZERO_RULE_COUNTER(rule); if (l->o.opcode == O_LOG) l->log_left = l->max_log; } /* * Flushes rules counters and/or log values on matching range. * * Returns number of items cleared. */ static int clear_range(struct ip_fw_chain *chain, ipfw_range_tlv *rt, int log_only) { struct ip_fw *rule; int num; int i; num = 0; rt->flags |= IPFW_RCFLAG_DEFAULT; IPFW_UH_WLOCK(chain); /* arbitrate writers */ for (i = 0; i < chain->n_rules; i++) { rule = chain->map[i]; if (ipfw_match_range(rule, rt) == 0) continue; clear_counters(rule, log_only); num++; } IPFW_UH_WUNLOCK(chain); return (num); } static int check_range_tlv(ipfw_range_tlv *rt) { if (rt->head.length != sizeof(*rt)) return (1); if (rt->start_rule > rt->end_rule) return (1); if (rt->set >= IPFW_MAX_SETS || rt->new_set >= IPFW_MAX_SETS) return (1); if ((rt->flags & IPFW_RCFLAG_USER) != rt->flags) return (1); return (0); } /* * Delete rules matching specified parameters * Data layout (v0)(current): * Request: [ ipfw_obj_header ipfw_range_tlv ] * Reply: [ ipfw_obj_header ipfw_range_tlv ] * * Saves number of deleted rules in ipfw_range_tlv->new_set. * * Returns 0 on success. */ static int del_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_range_header *rh; int error, ndel; if (sd->valsize != sizeof(*rh)) return (EINVAL); rh = (ipfw_range_header *)ipfw_get_sopt_space(sd, sd->valsize); if (check_range_tlv(&rh->range) != 0) return (EINVAL); ndel = 0; if ((error = delete_range(chain, &rh->range, &ndel)) != 0) return (error); /* Save number of rules deleted */ rh->range.new_set = ndel; return (0); } /* * Move rules/sets matching specified parameters * Data layout (v0)(current): * Request: [ ipfw_obj_header ipfw_range_tlv ] * * Returns 0 on success. */ static int move_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_range_header *rh; if (sd->valsize != sizeof(*rh)) return (EINVAL); rh = (ipfw_range_header *)ipfw_get_sopt_space(sd, sd->valsize); if (check_range_tlv(&rh->range) != 0) return (EINVAL); return (move_range(chain, &rh->range)); } /* * Clear rule accounting data matching specified parameters * Data layout (v0)(current): * Request: [ ipfw_obj_header ipfw_range_tlv ] * Reply: [ ipfw_obj_header ipfw_range_tlv ] * * Saves number of cleared rules in ipfw_range_tlv->new_set. * * Returns 0 on success. */ static int clear_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_range_header *rh; int log_only, num; char *msg; if (sd->valsize != sizeof(*rh)) return (EINVAL); rh = (ipfw_range_header *)ipfw_get_sopt_space(sd, sd->valsize); if (check_range_tlv(&rh->range) != 0) return (EINVAL); log_only = (op3->opcode == IP_FW_XRESETLOG); num = clear_range(chain, &rh->range, log_only); if (rh->range.flags & IPFW_RCFLAG_ALL) msg = log_only ? "All logging counts reset" : "Accounting cleared"; else msg = log_only ? "logging count reset" : "cleared"; if (V_fw_verbose) { int lev = LOG_SECURITY | LOG_NOTICE; log(lev, "ipfw: %s.\n", msg); } /* Save number of rules cleared */ rh->range.new_set = num; return (0); } static void enable_sets(struct ip_fw_chain *chain, ipfw_range_tlv *rt) { uint32_t v_set; IPFW_UH_WLOCK_ASSERT(chain); /* Change enabled/disabled sets mask */ v_set = (V_set_disable | rt->set) & ~rt->new_set; v_set &= ~(1 << RESVD_SET); /* set RESVD_SET always enabled */ IPFW_WLOCK(chain); V_set_disable = v_set; IPFW_WUNLOCK(chain); } static int swap_sets(struct ip_fw_chain *chain, ipfw_range_tlv *rt, int mv) { struct opcode_obj_rewrite *rw; struct ip_fw *rule; int i; IPFW_UH_WLOCK_ASSERT(chain); if (rt->set == rt->new_set) /* nothing to do */ return (0); if (mv != 0) { /* * Berfore moving the rules we need to check that * there aren't any conflicting named objects. */ for (rw = ctl3_rewriters; rw < ctl3_rewriters + ctl3_rsize; rw++) { if (rw->manage_sets == NULL) continue; i = rw->manage_sets(chain, (uint8_t)rt->set, (uint8_t)rt->new_set, TEST_ALL); if (i != 0) return (EEXIST); } } /* Swap or move two sets */ for (i = 0; i < chain->n_rules - 1; i++) { rule = chain->map[i]; if (rule->set == (uint8_t)rt->set) rule->set = (uint8_t)rt->new_set; else if (rule->set == (uint8_t)rt->new_set && mv == 0) rule->set = (uint8_t)rt->set; } for (rw = ctl3_rewriters; rw < ctl3_rewriters + ctl3_rsize; rw++) { if (rw->manage_sets == NULL) continue; rw->manage_sets(chain, (uint8_t)rt->set, (uint8_t)rt->new_set, mv != 0 ? MOVE_ALL: SWAP_ALL); } return (0); } /* * Swaps or moves set * Data layout (v0)(current): * Request: [ ipfw_obj_header ipfw_range_tlv ] * * Returns 0 on success. */ static int manage_sets(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_range_header *rh; int ret; if (sd->valsize != sizeof(*rh)) return (EINVAL); rh = (ipfw_range_header *)ipfw_get_sopt_space(sd, sd->valsize); if (rh->range.head.length != sizeof(ipfw_range_tlv)) return (1); /* enable_sets() expects bitmasks. */ if (op3->opcode != IP_FW_SET_ENABLE && (rh->range.set >= IPFW_MAX_SETS || rh->range.new_set >= IPFW_MAX_SETS)) return (EINVAL); ret = 0; IPFW_UH_WLOCK(chain); switch (op3->opcode) { case IP_FW_SET_SWAP: case IP_FW_SET_MOVE: ret = swap_sets(chain, &rh->range, op3->opcode == IP_FW_SET_MOVE); break; case IP_FW_SET_ENABLE: enable_sets(chain, &rh->range); break; } IPFW_UH_WUNLOCK(chain); return (ret); } /** * Remove all rules with given number, or do set manipulation. * Assumes chain != NULL && *chain != NULL. * * The argument is an uint32_t. The low 16 bit are the rule or set number; * the next 8 bits are the new set; the top 8 bits indicate the command: * * 0 delete rules numbered "rulenum" * 1 delete rules in set "rulenum" * 2 move rules "rulenum" to set "new_set" * 3 move rules from set "rulenum" to set "new_set" * 4 swap sets "rulenum" and "new_set" * 5 delete rules "rulenum" and set "new_set" */ static int del_entry(struct ip_fw_chain *chain, uint32_t arg) { uint32_t num; /* rule number or old_set */ uint8_t cmd, new_set; int do_del, ndel; int error = 0; ipfw_range_tlv rt; num = arg & 0xffff; cmd = (arg >> 24) & 0xff; new_set = (arg >> 16) & 0xff; if (cmd > 5 || new_set > RESVD_SET) return EINVAL; if (cmd == 0 || cmd == 2 || cmd == 5) { if (num >= IPFW_DEFAULT_RULE) return EINVAL; } else { if (num > RESVD_SET) /* old_set */ return EINVAL; } /* Convert old requests into new representation */ memset(&rt, 0, sizeof(rt)); rt.start_rule = num; rt.end_rule = num; rt.set = num; rt.new_set = new_set; do_del = 0; switch (cmd) { case 0: /* delete rules numbered "rulenum" */ if (num == 0) rt.flags |= IPFW_RCFLAG_ALL; else rt.flags |= IPFW_RCFLAG_RANGE; do_del = 1; break; case 1: /* delete rules in set "rulenum" */ rt.flags |= IPFW_RCFLAG_SET; do_del = 1; break; case 5: /* delete rules "rulenum" and set "new_set" */ rt.flags |= IPFW_RCFLAG_RANGE | IPFW_RCFLAG_SET; rt.set = new_set; rt.new_set = 0; do_del = 1; break; case 2: /* move rules "rulenum" to set "new_set" */ rt.flags |= IPFW_RCFLAG_RANGE; break; case 3: /* move rules from set "rulenum" to set "new_set" */ IPFW_UH_WLOCK(chain); error = swap_sets(chain, &rt, 1); IPFW_UH_WUNLOCK(chain); return (error); case 4: /* swap sets "rulenum" and "new_set" */ IPFW_UH_WLOCK(chain); error = swap_sets(chain, &rt, 0); IPFW_UH_WUNLOCK(chain); return (error); default: return (ENOTSUP); } if (do_del != 0) { if ((error = delete_range(chain, &rt, &ndel)) != 0) return (error); if (ndel == 0 && (cmd != 1 && num != 0)) return (EINVAL); return (0); } return (move_range(chain, &rt)); } /** * Reset some or all counters on firewall rules. * The argument `arg' is an u_int32_t. The low 16 bit are the rule number, * the next 8 bits are the set number, the top 8 bits are the command: * 0 work with rules from all set's; * 1 work with rules only from specified set. * Specified rule number is zero if we want to clear all entries. * log_only is 1 if we only want to reset logs, zero otherwise. */ static int zero_entry(struct ip_fw_chain *chain, u_int32_t arg, int log_only) { struct ip_fw *rule; char *msg; int i; uint16_t rulenum = arg & 0xffff; uint8_t set = (arg >> 16) & 0xff; uint8_t cmd = (arg >> 24) & 0xff; if (cmd > 1) return (EINVAL); if (cmd == 1 && set > RESVD_SET) return (EINVAL); IPFW_UH_RLOCK(chain); if (rulenum == 0) { V_norule_counter = 0; for (i = 0; i < chain->n_rules; i++) { rule = chain->map[i]; /* Skip rules not in our set. */ if (cmd == 1 && rule->set != set) continue; clear_counters(rule, log_only); } msg = log_only ? "All logging counts reset" : "Accounting cleared"; } else { int cleared = 0; for (i = 0; i < chain->n_rules; i++) { rule = chain->map[i]; if (rule->rulenum == rulenum) { if (cmd == 0 || rule->set == set) clear_counters(rule, log_only); cleared = 1; } if (rule->rulenum > rulenum) break; } if (!cleared) { /* we did not find any matching rules */ IPFW_UH_RUNLOCK(chain); return (EINVAL); } msg = log_only ? "logging count reset" : "cleared"; } IPFW_UH_RUNLOCK(chain); if (V_fw_verbose) { int lev = LOG_SECURITY | LOG_NOTICE; if (rulenum) log(lev, "ipfw: Entry %d %s.\n", rulenum, msg); else log(lev, "ipfw: %s.\n", msg); } return (0); } /* * Check rule head in FreeBSD11 format * */ static int check_ipfw_rule1(struct ip_fw_rule *rule, int size, struct rule_check_info *ci) { int l; if (size < sizeof(*rule)) { printf("ipfw: rule too short\n"); return (EINVAL); } /* Check for valid cmd_len */ l = roundup2(RULESIZE(rule), sizeof(uint64_t)); if (l != size) { printf("ipfw: size mismatch (have %d want %d)\n", size, l); return (EINVAL); } if (rule->act_ofs >= rule->cmd_len) { printf("ipfw: bogus action offset (%u > %u)\n", rule->act_ofs, rule->cmd_len - 1); return (EINVAL); } if (rule->rulenum > IPFW_DEFAULT_RULE - 1) return (EINVAL); return (check_ipfw_rule_body(rule->cmd, rule->cmd_len, ci)); } /* * Check rule head in FreeBSD8 format * */ static int check_ipfw_rule0(struct ip_fw_rule0 *rule, int size, struct rule_check_info *ci) { int l; if (size < sizeof(*rule)) { printf("ipfw: rule too short\n"); return (EINVAL); } /* Check for valid cmd_len */ l = sizeof(*rule) + rule->cmd_len * 4 - 4; if (l != size) { printf("ipfw: size mismatch (have %d want %d)\n", size, l); return (EINVAL); } if (rule->act_ofs >= rule->cmd_len) { printf("ipfw: bogus action offset (%u > %u)\n", rule->act_ofs, rule->cmd_len - 1); return (EINVAL); } if (rule->rulenum > IPFW_DEFAULT_RULE - 1) return (EINVAL); return (check_ipfw_rule_body(rule->cmd, rule->cmd_len, ci)); } static int check_ipfw_rule_body(ipfw_insn *cmd, int cmd_len, struct rule_check_info *ci) { int cmdlen, l; int have_action; have_action = 0; /* * Now go for the individual checks. Very simple ones, basically only * instruction sizes. */ for (l = cmd_len; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); if (cmdlen > l) { printf("ipfw: opcode %d size truncated\n", cmd->opcode); return EINVAL; } switch (cmd->opcode) { case O_PROBE_STATE: case O_KEEP_STATE: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; ci->object_opcodes++; break; case O_PROTO: case O_IP_SRC_ME: case O_IP_DST_ME: case O_LAYER2: case O_IN: case O_FRAG: case O_DIVERTED: case O_IPOPT: case O_IPTOS: case O_IPPRECEDENCE: case O_IPVER: case O_SOCKARG: case O_TCPFLAGS: case O_TCPOPTS: case O_ESTAB: case O_VERREVPATH: case O_VERSRCREACH: case O_ANTISPOOF: case O_IPSEC: #ifdef INET6 case O_IP6_SRC_ME: case O_IP6_DST_ME: case O_EXT_HDR: case O_IP6: #endif case O_IP4: case O_TAG: case O_SKIP_ACTION: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; break; case O_EXTERNAL_ACTION: if (cmd->arg1 == 0 || cmdlen != F_INSN_SIZE(ipfw_insn)) { printf("ipfw: invalid external " "action opcode\n"); return (EINVAL); } ci->object_opcodes++; /* * Do we have O_EXTERNAL_INSTANCE or O_EXTERNAL_DATA * opcode? */ if (l != cmdlen) { l -= cmdlen; cmd += cmdlen; cmdlen = F_LEN(cmd); if (cmd->opcode == O_EXTERNAL_DATA) goto check_action; if (cmd->opcode != O_EXTERNAL_INSTANCE) { printf("ipfw: invalid opcode " "next to external action %u\n", cmd->opcode); return (EINVAL); } if (cmd->arg1 == 0 || cmdlen != F_INSN_SIZE(ipfw_insn)) { printf("ipfw: invalid external " "action instance opcode\n"); return (EINVAL); } ci->object_opcodes++; } goto check_action; case O_FIB: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; if (cmd->arg1 >= rt_numfibs) { printf("ipfw: invalid fib number %d\n", cmd->arg1); return EINVAL; } break; case O_SETFIB: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; if ((cmd->arg1 != IP_FW_TARG) && ((cmd->arg1 & 0x7FFF) >= rt_numfibs)) { printf("ipfw: invalid fib number %d\n", cmd->arg1 & 0x7FFF); return EINVAL; } goto check_action; case O_UID: case O_GID: case O_JAIL: case O_IP_SRC: case O_IP_DST: case O_TCPSEQ: case O_TCPACK: case O_PROB: case O_ICMPTYPE: if (cmdlen != F_INSN_SIZE(ipfw_insn_u32)) goto bad_size; break; case O_LIMIT: if (cmdlen != F_INSN_SIZE(ipfw_insn_limit)) goto bad_size; ci->object_opcodes++; break; case O_LOG: if (cmdlen != F_INSN_SIZE(ipfw_insn_log)) goto bad_size; ((ipfw_insn_log *)cmd)->log_left = ((ipfw_insn_log *)cmd)->max_log; break; case O_IP_SRC_MASK: case O_IP_DST_MASK: /* only odd command lengths */ if ((cmdlen & 1) == 0) goto bad_size; break; case O_IP_SRC_SET: case O_IP_DST_SET: if (cmd->arg1 == 0 || cmd->arg1 > 256) { printf("ipfw: invalid set size %d\n", cmd->arg1); return EINVAL; } if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + (cmd->arg1+31)/32 ) goto bad_size; break; case O_IP_SRC_LOOKUP: if (cmdlen > F_INSN_SIZE(ipfw_insn_u32)) goto bad_size; case O_IP_DST_LOOKUP: if (cmd->arg1 >= V_fw_tables_max) { printf("ipfw: invalid table number %d\n", cmd->arg1); return (EINVAL); } if (cmdlen != F_INSN_SIZE(ipfw_insn) && cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 1 && cmdlen != F_INSN_SIZE(ipfw_insn_u32)) goto bad_size; ci->object_opcodes++; break; case O_IP_FLOW_LOOKUP: if (cmd->arg1 >= V_fw_tables_max) { printf("ipfw: invalid table number %d\n", cmd->arg1); return (EINVAL); } if (cmdlen != F_INSN_SIZE(ipfw_insn) && cmdlen != F_INSN_SIZE(ipfw_insn_u32)) goto bad_size; ci->object_opcodes++; break; case O_MACADDR2: if (cmdlen != F_INSN_SIZE(ipfw_insn_mac)) goto bad_size; break; case O_NOP: case O_IPID: case O_IPTTL: case O_IPLEN: case O_TCPDATALEN: case O_TCPWIN: case O_TAGGED: if (cmdlen < 1 || cmdlen > 31) goto bad_size; break; case O_DSCP: if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + 1) goto bad_size; break; case O_MAC_TYPE: case O_IP_SRCPORT: case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */ if (cmdlen < 2 || cmdlen > 31) goto bad_size; break; case O_RECV: case O_XMIT: case O_VIA: if (cmdlen != F_INSN_SIZE(ipfw_insn_if)) goto bad_size; ci->object_opcodes++; break; case O_ALTQ: if (cmdlen != F_INSN_SIZE(ipfw_insn_altq)) goto bad_size; break; case O_PIPE: case O_QUEUE: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; goto check_action; case O_FORWARD_IP: if (cmdlen != F_INSN_SIZE(ipfw_insn_sa)) goto bad_size; goto check_action; #ifdef INET6 case O_FORWARD_IP6: if (cmdlen != F_INSN_SIZE(ipfw_insn_sa6)) goto bad_size; goto check_action; #endif /* INET6 */ case O_DIVERT: case O_TEE: if (ip_divert_ptr == NULL) return EINVAL; else goto check_size; case O_NETGRAPH: case O_NGTEE: if (ng_ipfw_input_p == NULL) return EINVAL; else goto check_size; case O_NAT: if (!IPFW_NAT_LOADED) return EINVAL; if (cmdlen != F_INSN_SIZE(ipfw_insn_nat)) goto bad_size; goto check_action; case O_CHECK_STATE: ci->object_opcodes++; /* FALLTHROUGH */ case O_FORWARD_MAC: /* XXX not implemented yet */ case O_COUNT: case O_ACCEPT: case O_DENY: case O_REJECT: case O_SETDSCP: #ifdef INET6 case O_UNREACH6: #endif case O_SKIPTO: case O_REASS: case O_CALLRETURN: check_size: if (cmdlen != F_INSN_SIZE(ipfw_insn)) goto bad_size; check_action: if (have_action) { printf("ipfw: opcode %d, multiple actions" " not allowed\n", cmd->opcode); return (EINVAL); } have_action = 1; if (l != cmdlen) { printf("ipfw: opcode %d, action must be" " last opcode\n", cmd->opcode); return (EINVAL); } break; #ifdef INET6 case O_IP6_SRC: case O_IP6_DST: if (cmdlen != F_INSN_SIZE(struct in6_addr) + F_INSN_SIZE(ipfw_insn)) goto bad_size; break; case O_FLOW6ID: if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) + ((ipfw_insn_u32 *)cmd)->o.arg1) goto bad_size; break; case O_IP6_SRC_MASK: case O_IP6_DST_MASK: if ( !(cmdlen & 1) || cmdlen > 127) goto bad_size; break; case O_ICMP6TYPE: if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) ) goto bad_size; break; #endif default: switch (cmd->opcode) { #ifndef INET6 case O_IP6_SRC_ME: case O_IP6_DST_ME: case O_EXT_HDR: case O_IP6: case O_UNREACH6: case O_IP6_SRC: case O_IP6_DST: case O_FLOW6ID: case O_IP6_SRC_MASK: case O_IP6_DST_MASK: case O_ICMP6TYPE: printf("ipfw: no IPv6 support in kernel\n"); return (EPROTONOSUPPORT); #endif default: printf("ipfw: opcode %d, unknown opcode\n", cmd->opcode); return (EINVAL); } } } if (have_action == 0) { printf("ipfw: missing action\n"); return (EINVAL); } return 0; bad_size: printf("ipfw: opcode %d size %d wrong\n", cmd->opcode, cmdlen); return (EINVAL); } /* * Translation of requests for compatibility with FreeBSD 7.2/8. * a static variable tells us if we have an old client from userland, * and if necessary we translate requests and responses between the * two formats. */ static int is7 = 0; struct ip_fw7 { struct ip_fw7 *next; /* linked list of rules */ struct ip_fw7 *next_rule; /* ptr to next [skipto] rule */ /* 'next_rule' is used to pass up 'set_disable' status */ uint16_t act_ofs; /* offset of action in 32-bit units */ uint16_t cmd_len; /* # of 32-bit words in cmd */ uint16_t rulenum; /* rule number */ uint8_t set; /* rule set (0..31) */ // #define RESVD_SET 31 /* set for default and persistent rules */ uint8_t _pad; /* padding */ // uint32_t id; /* rule id, only in v.8 */ /* These fields are present in all rules. */ uint64_t pcnt; /* Packet counter */ uint64_t bcnt; /* Byte counter */ uint32_t timestamp; /* tv_sec of last match */ ipfw_insn cmd[1]; /* storage for commands */ }; static int convert_rule_to_7(struct ip_fw_rule0 *rule); static int convert_rule_to_8(struct ip_fw_rule0 *rule); #ifndef RULESIZE7 #define RULESIZE7(rule) (sizeof(struct ip_fw7) + \ ((struct ip_fw7 *)(rule))->cmd_len * 4 - 4) #endif /* * Copy the static and dynamic rules to the supplied buffer * and return the amount of space actually used. * Must be run under IPFW_UH_RLOCK */ static size_t ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space) { char *bp = buf; char *ep = bp + space; struct ip_fw *rule; struct ip_fw_rule0 *dst; struct timeval boottime; int error, i, l, warnflag; time_t boot_seconds; warnflag = 0; getboottime(&boottime); boot_seconds = boottime.tv_sec; for (i = 0; i < chain->n_rules; i++) { rule = chain->map[i]; if (is7) { /* Convert rule to FreeBSd 7.2 format */ l = RULESIZE7(rule); if (bp + l + sizeof(uint32_t) <= ep) { bcopy(rule, bp, l + sizeof(uint32_t)); error = set_legacy_obj_kidx(chain, (struct ip_fw_rule0 *)bp); if (error != 0) return (0); error = convert_rule_to_7((struct ip_fw_rule0 *) bp); if (error) return 0; /*XXX correct? */ /* * XXX HACK. Store the disable mask in the "next" * pointer in a wild attempt to keep the ABI the same. * Why do we do this on EVERY rule? */ bcopy(&V_set_disable, &(((struct ip_fw7 *)bp)->next_rule), sizeof(V_set_disable)); if (((struct ip_fw7 *)bp)->timestamp) ((struct ip_fw7 *)bp)->timestamp += boot_seconds; bp += l; } continue; /* go to next rule */ } l = RULEUSIZE0(rule); if (bp + l > ep) { /* should not happen */ printf("overflow dumping static rules\n"); break; } dst = (struct ip_fw_rule0 *)bp; export_rule0(rule, dst, l); error = set_legacy_obj_kidx(chain, dst); /* * XXX HACK. Store the disable mask in the "next" * pointer in a wild attempt to keep the ABI the same. * Why do we do this on EVERY rule? * * XXX: "ipfw set show" (ab)uses IP_FW_GET to read disabled mask * so we need to fail _after_ saving at least one mask. */ bcopy(&V_set_disable, &dst->next_rule, sizeof(V_set_disable)); if (dst->timestamp) dst->timestamp += boot_seconds; bp += l; if (error != 0) { if (error == 2) { /* Non-fatal table rewrite error. */ warnflag = 1; continue; } printf("Stop on rule %d. Fail to convert table\n", rule->rulenum); break; } } if (warnflag != 0) printf("ipfw: process %s is using legacy interfaces," " consider rebuilding\n", ""); ipfw_get_dynamic(chain, &bp, ep); /* protected by the dynamic lock */ return (bp - (char *)buf); } struct dump_args { uint32_t b; /* start rule */ uint32_t e; /* end rule */ uint32_t rcount; /* number of rules */ uint32_t rsize; /* rules size */ uint32_t tcount; /* number of tables */ int rcounters; /* counters */ }; void ipfw_export_obj_ntlv(struct named_object *no, ipfw_obj_ntlv *ntlv) { ntlv->head.type = no->etlv; ntlv->head.length = sizeof(*ntlv); ntlv->idx = no->kidx; strlcpy(ntlv->name, no->name, sizeof(ntlv->name)); } /* * Export named object info in instance @ni, identified by @kidx * to ipfw_obj_ntlv. TLV is allocated from @sd space. * * Returns 0 on success. */ static int export_objhash_ntlv(struct namedobj_instance *ni, uint16_t kidx, struct sockopt_data *sd) { struct named_object *no; ipfw_obj_ntlv *ntlv; no = ipfw_objhash_lookup_kidx(ni, kidx); KASSERT(no != NULL, ("invalid object kernel index passed")); ntlv = (ipfw_obj_ntlv *)ipfw_get_sopt_space(sd, sizeof(*ntlv)); if (ntlv == NULL) return (ENOMEM); ipfw_export_obj_ntlv(no, ntlv); return (0); } /* * Dumps static rules with table TLVs in buffer @sd. * * Returns 0 on success. */ static int dump_static_rules(struct ip_fw_chain *chain, struct dump_args *da, uint32_t *bmask, struct sockopt_data *sd) { int error; int i, l; uint32_t tcount; ipfw_obj_ctlv *ctlv; struct ip_fw *krule; struct namedobj_instance *ni; caddr_t dst; /* Dump table names first (if any) */ if (da->tcount > 0) { /* Header first */ ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); if (ctlv == NULL) return (ENOMEM); ctlv->head.type = IPFW_TLV_TBLNAME_LIST; ctlv->head.length = da->tcount * sizeof(ipfw_obj_ntlv) + sizeof(*ctlv); ctlv->count = da->tcount; ctlv->objsize = sizeof(ipfw_obj_ntlv); } i = 0; tcount = da->tcount; ni = ipfw_get_table_objhash(chain); while (tcount > 0) { if ((bmask[i / 32] & (1 << (i % 32))) == 0) { i++; continue; } /* Jump to shared named object bitmask */ if (i >= IPFW_TABLES_MAX) { ni = CHAIN_TO_SRV(chain); i -= IPFW_TABLES_MAX; bmask += IPFW_TABLES_MAX / 32; } if ((error = export_objhash_ntlv(ni, i, sd)) != 0) return (error); i++; tcount--; } /* Dump rules */ ctlv = (ipfw_obj_ctlv *)ipfw_get_sopt_space(sd, sizeof(*ctlv)); if (ctlv == NULL) return (ENOMEM); ctlv->head.type = IPFW_TLV_RULE_LIST; ctlv->head.length = da->rsize + sizeof(*ctlv); ctlv->count = da->rcount; for (i = da->b; i < da->e; i++) { krule = chain->map[i]; l = RULEUSIZE1(krule) + sizeof(ipfw_obj_tlv); if (da->rcounters != 0) l += sizeof(struct ip_fw_bcounter); dst = (caddr_t)ipfw_get_sopt_space(sd, l); if (dst == NULL) return (ENOMEM); export_rule1(krule, dst, l, da->rcounters); } return (0); } /* * Marks every object index used in @rule with bit in @bmask. * Used to generate bitmask of referenced tables/objects for given ruleset * or its part. * * Returns number of newly-referenced objects. */ static int mark_object_kidx(struct ip_fw_chain *ch, struct ip_fw *rule, uint32_t *bmask) { struct opcode_obj_rewrite *rw; ipfw_insn *cmd; int bidx, cmdlen, l, count; uint16_t kidx; uint8_t subtype; l = rule->cmd_len; cmd = rule->cmd; cmdlen = 0; count = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); rw = find_op_rw(cmd, &kidx, &subtype); if (rw == NULL) continue; bidx = kidx / 32; /* * Maintain separate bitmasks for table and * non-table objects. */ if (rw->etlv != IPFW_TLV_TBL_NAME) bidx += IPFW_TABLES_MAX / 32; if ((bmask[bidx] & (1 << (kidx % 32))) == 0) count++; bmask[bidx] |= 1 << (kidx % 32); } return (count); } /* * Dumps requested objects data * Data layout (version 0)(current): * Request: [ ipfw_cfg_lheader ] + IPFW_CFG_GET_* flags * size = ipfw_cfg_lheader.size * Reply: [ ipfw_cfg_lheader * [ ipfw_obj_ctlv(IPFW_TLV_TBL_LIST) ipfw_obj_ntlv x N ] (optional) * [ ipfw_obj_ctlv(IPFW_TLV_RULE_LIST) * ipfw_obj_tlv(IPFW_TLV_RULE_ENT) [ ip_fw_bcounter (optional) ip_fw_rule ] * ] (optional) * [ ipfw_obj_ctlv(IPFW_TLV_STATE_LIST) ipfw_obj_dyntlv x N ] (optional) * ] * * NOTE IPFW_TLV_STATE_LIST has the single valid field: objsize. * The rest (size, count) are set to zero and needs to be ignored. * * Returns 0 on success. */ static int dump_config(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_cfg_lheader *hdr; struct ip_fw *rule; size_t sz, rnum; uint32_t hdr_flags; int error, i; struct dump_args da; uint32_t *bmask; hdr = (ipfw_cfg_lheader *)ipfw_get_sopt_header(sd, sizeof(*hdr)); if (hdr == NULL) return (EINVAL); error = 0; bmask = NULL; /* Allocate needed state. Note we allocate 2xspace mask, for table&srv */ if (hdr->flags & IPFW_CFG_GET_STATIC) bmask = malloc(IPFW_TABLES_MAX / 4, M_TEMP, M_WAITOK | M_ZERO); IPFW_UH_RLOCK(chain); /* * STAGE 1: Determine size/count for objects in range. * Prepare used tables bitmask. */ sz = sizeof(ipfw_cfg_lheader); memset(&da, 0, sizeof(da)); da.b = 0; da.e = chain->n_rules; if (hdr->end_rule != 0) { /* Handle custom range */ if ((rnum = hdr->start_rule) > IPFW_DEFAULT_RULE) rnum = IPFW_DEFAULT_RULE; da.b = ipfw_find_rule(chain, rnum, 0); rnum = (hdr->end_rule < IPFW_DEFAULT_RULE) ? hdr->end_rule + 1: IPFW_DEFAULT_RULE; da.e = ipfw_find_rule(chain, rnum, UINT32_MAX) + 1; } if (hdr->flags & IPFW_CFG_GET_STATIC) { for (i = da.b; i < da.e; i++) { rule = chain->map[i]; da.rsize += RULEUSIZE1(rule) + sizeof(ipfw_obj_tlv); da.rcount++; /* Update bitmask of used objects for given range */ da.tcount += mark_object_kidx(chain, rule, bmask); } /* Add counters if requested */ if (hdr->flags & IPFW_CFG_GET_COUNTERS) { da.rsize += sizeof(struct ip_fw_bcounter) * da.rcount; da.rcounters = 1; } if (da.tcount > 0) sz += da.tcount * sizeof(ipfw_obj_ntlv) + sizeof(ipfw_obj_ctlv); sz += da.rsize + sizeof(ipfw_obj_ctlv); } if (hdr->flags & IPFW_CFG_GET_STATES) sz += ipfw_dyn_get_count() * sizeof(ipfw_obj_dyntlv) + sizeof(ipfw_obj_ctlv); /* * Fill header anyway. * Note we have to save header fields to stable storage * buffer inside @sd can be flushed after dumping rules */ hdr->size = sz; hdr->set_mask = ~V_set_disable; hdr_flags = hdr->flags; hdr = NULL; if (sd->valsize < sz) { error = ENOMEM; goto cleanup; } /* STAGE2: Store actual data */ if (hdr_flags & IPFW_CFG_GET_STATIC) { error = dump_static_rules(chain, &da, bmask, sd); if (error != 0) goto cleanup; } if (hdr_flags & IPFW_CFG_GET_STATES) error = ipfw_dump_states(chain, sd); cleanup: IPFW_UH_RUNLOCK(chain); if (bmask != NULL) free(bmask, M_TEMP); return (error); } int ipfw_check_object_name_generic(const char *name) { int nsize; nsize = sizeof(((ipfw_obj_ntlv *)0)->name); if (strnlen(name, nsize) == nsize) return (EINVAL); if (name[0] == '\0') return (EINVAL); return (0); } /* * Creates non-existent objects referenced by rule. * * Return 0 on success. */ int create_objects_compat(struct ip_fw_chain *ch, ipfw_insn *cmd, struct obj_idx *oib, struct obj_idx *pidx, struct tid_info *ti) { struct opcode_obj_rewrite *rw; struct obj_idx *p; uint16_t kidx; int error; /* * Compatibility stuff: do actual creation for non-existing, * but referenced objects. */ for (p = oib; p < pidx; p++) { if (p->kidx != 0) continue; ti->uidx = p->uidx; ti->type = p->type; ti->atype = 0; rw = find_op_rw(cmd + p->off, NULL, NULL); KASSERT(rw != NULL, ("Unable to find handler for op %d", (cmd + p->off)->opcode)); if (rw->create_object == NULL) error = EOPNOTSUPP; else error = rw->create_object(ch, ti, &kidx); if (error == 0) { p->kidx = kidx; continue; } /* * Error happened. We have to rollback everything. * Drop all already acquired references. */ IPFW_UH_WLOCK(ch); unref_oib_objects(ch, cmd, oib, pidx); IPFW_UH_WUNLOCK(ch); return (error); } return (0); } /* * Compatibility function for old ipfw(8) binaries. * Rewrites table/nat kernel indices with userland ones. * Convert tables matching '/^\d+$/' to their atoi() value. * Use number 65535 for other tables. * * Returns 0 on success. */ static int set_legacy_obj_kidx(struct ip_fw_chain *ch, struct ip_fw_rule0 *rule) { struct opcode_obj_rewrite *rw; struct named_object *no; ipfw_insn *cmd; char *end; long val; int cmdlen, error, l; uint16_t kidx, uidx; uint8_t subtype; error = 0; l = rule->cmd_len; cmd = rule->cmd; cmdlen = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); /* Check if is index in given opcode */ rw = find_op_rw(cmd, &kidx, &subtype); if (rw == NULL) continue; /* Try to find referenced kernel object */ no = rw->find_bykidx(ch, kidx); if (no == NULL) continue; val = strtol(no->name, &end, 10); if (*end == '\0' && val < 65535) { uidx = val; } else { /* * We are called via legacy opcode. * Save error and show table as fake number * not to make ipfw(8) hang. */ uidx = 65535; error = 2; } rw->update(cmd, uidx); } return (error); } /* * Unreferences all already-referenced objects in given @cmd rule, * using information in @oib. * * Used to rollback partially converted rule on error. */ static void unref_oib_objects(struct ip_fw_chain *ch, ipfw_insn *cmd, struct obj_idx *oib, struct obj_idx *end) { struct opcode_obj_rewrite *rw; struct named_object *no; struct obj_idx *p; IPFW_UH_WLOCK_ASSERT(ch); for (p = oib; p < end; p++) { if (p->kidx == 0) continue; rw = find_op_rw(cmd + p->off, NULL, NULL); KASSERT(rw != NULL, ("Unable to find handler for op %d", (cmd + p->off)->opcode)); /* Find & unref by existing idx */ no = rw->find_bykidx(ch, p->kidx); KASSERT(no != NULL, ("Ref'd object %d disappeared", p->kidx)); no->refcnt--; } } /* * Remove references from every object used in @rule. * Used at rule removal code. */ static void unref_rule_objects(struct ip_fw_chain *ch, struct ip_fw *rule) { struct opcode_obj_rewrite *rw; struct named_object *no; ipfw_insn *cmd; int cmdlen, l; uint16_t kidx; uint8_t subtype; IPFW_UH_WLOCK_ASSERT(ch); l = rule->cmd_len; cmd = rule->cmd; cmdlen = 0; for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); rw = find_op_rw(cmd, &kidx, &subtype); if (rw == NULL) continue; no = rw->find_bykidx(ch, kidx); KASSERT(no != NULL, ("object id %d not found", kidx)); KASSERT(no->subtype == subtype, ("wrong type %d (%d) for object id %d", no->subtype, subtype, kidx)); KASSERT(no->refcnt > 0, ("refcount for object %d is %d", kidx, no->refcnt)); if (no->refcnt == 1 && rw->destroy_object != NULL) rw->destroy_object(ch, no); else no->refcnt--; } } /* * Find and reference object (if any) stored in instruction @cmd. * * Saves object info in @pidx, sets * - @unresolved to 1 if object should exists but not found * * Returns non-zero value in case of error. */ static int ref_opcode_object(struct ip_fw_chain *ch, ipfw_insn *cmd, struct tid_info *ti, struct obj_idx *pidx, int *unresolved) { struct named_object *no; struct opcode_obj_rewrite *rw; int error; /* Check if this opcode is candidate for rewrite */ rw = find_op_rw(cmd, &ti->uidx, &ti->type); if (rw == NULL) return (0); /* Need to rewrite. Save necessary fields */ pidx->uidx = ti->uidx; pidx->type = ti->type; /* Try to find referenced kernel object */ error = rw->find_byname(ch, ti, &no); if (error != 0) return (error); if (no == NULL) { /* * Report about unresolved object for automaic * creation. */ *unresolved = 1; return (0); } /* * Object is already exist. * Its subtype should match with expected value. */ if (ti->type != no->subtype) return (EINVAL); /* Bump refcount and update kidx. */ no->refcnt++; rw->update(cmd, no->kidx); return (0); } /* * Finds and bumps refcount for objects referenced by given @rule. * Auto-creates non-existing tables. * Fills in @oib array with userland/kernel indexes. * * Returns 0 on success. */ static int ref_rule_objects(struct ip_fw_chain *ch, struct ip_fw *rule, struct rule_check_info *ci, struct obj_idx *oib, struct tid_info *ti) { struct obj_idx *pidx; ipfw_insn *cmd; int cmdlen, error, l, unresolved; pidx = oib; l = rule->cmd_len; cmd = rule->cmd; cmdlen = 0; error = 0; IPFW_UH_WLOCK(ch); /* Increase refcount on each existing referenced table. */ for ( ; l > 0 ; l -= cmdlen, cmd += cmdlen) { cmdlen = F_LEN(cmd); unresolved = 0; error = ref_opcode_object(ch, cmd, ti, pidx, &unresolved); if (error != 0) break; /* * Compatibility stuff for old clients: * prepare to automaitcally create non-existing objects. */ if (unresolved != 0) { pidx->off = rule->cmd_len - l; pidx++; } } if (error != 0) { /* Unref everything we have already done */ unref_oib_objects(ch, rule->cmd, oib, pidx); IPFW_UH_WUNLOCK(ch); return (error); } IPFW_UH_WUNLOCK(ch); /* Perform auto-creation for non-existing objects */ if (pidx != oib) error = create_objects_compat(ch, rule->cmd, oib, pidx, ti); /* Calculate real number of dynamic objects */ ci->object_opcodes = (uint16_t)(pidx - oib); return (error); } /* * Checks is opcode is referencing table of appropriate type. * Adds reference count for found table if true. * Rewrites user-supplied opcode values with kernel ones. * * Returns 0 on success and appropriate error code otherwise. */ static int rewrite_rule_uidx(struct ip_fw_chain *chain, struct rule_check_info *ci) { int error; ipfw_insn *cmd; uint8_t type; struct obj_idx *p, *pidx_first, *pidx_last; struct tid_info ti; /* * Prepare an array for storing opcode indices. * Use stack allocation by default. */ if (ci->object_opcodes <= (sizeof(ci->obuf)/sizeof(ci->obuf[0]))) { /* Stack */ pidx_first = ci->obuf; } else pidx_first = malloc( ci->object_opcodes * sizeof(struct obj_idx), M_IPFW, M_WAITOK | M_ZERO); error = 0; type = 0; memset(&ti, 0, sizeof(ti)); /* Use set rule is assigned to. */ ti.set = ci->krule->set; if (ci->ctlv != NULL) { ti.tlvs = (void *)(ci->ctlv + 1); ti.tlen = ci->ctlv->head.length - sizeof(ipfw_obj_ctlv); } /* Reference all used tables and other objects */ error = ref_rule_objects(chain, ci->krule, ci, pidx_first, &ti); if (error != 0) goto free; /* * Note that ref_rule_objects() might have updated ci->object_opcodes * to reflect actual number of object opcodes. */ /* Perform rewrite of remaining opcodes */ p = pidx_first; pidx_last = pidx_first + ci->object_opcodes; for (p = pidx_first; p < pidx_last; p++) { cmd = ci->krule->cmd + p->off; update_opcode_kidx(cmd, p->kidx); } free: if (pidx_first != ci->obuf) free(pidx_first, M_IPFW); return (error); } /* * Adds one or more rules to ipfw @chain. * Data layout (version 0)(current): * Request: * [ * ip_fw3_opheader * [ ipfw_obj_ctlv(IPFW_TLV_TBL_LIST) ipfw_obj_ntlv x N ] (optional *1) * [ ipfw_obj_ctlv(IPFW_TLV_RULE_LIST) ip_fw x N ] (*2) (*3) * ] * Reply: * [ * ip_fw3_opheader * [ ipfw_obj_ctlv(IPFW_TLV_TBL_LIST) ipfw_obj_ntlv x N ] (optional) * [ ipfw_obj_ctlv(IPFW_TLV_RULE_LIST) ip_fw x N ] * ] * * Rules in reply are modified to store their actual ruleset number. * * (*1) TLVs inside IPFW_TLV_TBL_LIST needs to be sorted ascending * according to their idx field and there has to be no duplicates. * (*2) Numbered rules inside IPFW_TLV_RULE_LIST needs to be sorted ascending. * (*3) Each ip_fw structure needs to be aligned to u64 boundary. * * Returns 0 on success. */ static int add_rules(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_obj_ctlv *ctlv, *rtlv, *tstate; ipfw_obj_ntlv *ntlv; int clen, error, idx; uint32_t count, read; struct ip_fw_rule *r; struct rule_check_info rci, *ci, *cbuf; int i, rsize; op3 = (ip_fw3_opheader *)ipfw_get_sopt_space(sd, sd->valsize); ctlv = (ipfw_obj_ctlv *)(op3 + 1); read = sizeof(ip_fw3_opheader); rtlv = NULL; tstate = NULL; cbuf = NULL; memset(&rci, 0, sizeof(struct rule_check_info)); if (read + sizeof(*ctlv) > sd->valsize) return (EINVAL); if (ctlv->head.type == IPFW_TLV_TBLNAME_LIST) { clen = ctlv->head.length; /* Check size and alignment */ if (clen > sd->valsize || clen < sizeof(*ctlv)) return (EINVAL); if ((clen % sizeof(uint64_t)) != 0) return (EINVAL); /* * Some table names or other named objects. * Check for validness. */ count = (ctlv->head.length - sizeof(*ctlv)) / sizeof(*ntlv); if (ctlv->count != count || ctlv->objsize != sizeof(*ntlv)) return (EINVAL); /* * Check each TLV. * Ensure TLVs are sorted ascending and * there are no duplicates. */ idx = -1; ntlv = (ipfw_obj_ntlv *)(ctlv + 1); while (count > 0) { if (ntlv->head.length != sizeof(ipfw_obj_ntlv)) return (EINVAL); error = ipfw_check_object_name_generic(ntlv->name); if (error != 0) return (error); if (ntlv->idx <= idx) return (EINVAL); idx = ntlv->idx; count--; ntlv++; } tstate = ctlv; read += ctlv->head.length; ctlv = (ipfw_obj_ctlv *)((caddr_t)ctlv + ctlv->head.length); } if (read + sizeof(*ctlv) > sd->valsize) return (EINVAL); if (ctlv->head.type == IPFW_TLV_RULE_LIST) { clen = ctlv->head.length; if (clen + read > sd->valsize || clen < sizeof(*ctlv)) return (EINVAL); if ((clen % sizeof(uint64_t)) != 0) return (EINVAL); /* * TODO: Permit adding multiple rules at once */ if (ctlv->count != 1) return (ENOTSUP); clen -= sizeof(*ctlv); if (ctlv->count > clen / sizeof(struct ip_fw_rule)) return (EINVAL); /* Allocate state for each rule or use stack */ if (ctlv->count == 1) { memset(&rci, 0, sizeof(struct rule_check_info)); cbuf = &rci; } else cbuf = malloc(ctlv->count * sizeof(*ci), M_TEMP, M_WAITOK | M_ZERO); ci = cbuf; /* * Check each rule for validness. * Ensure numbered rules are sorted ascending * and properly aligned */ idx = 0; r = (struct ip_fw_rule *)(ctlv + 1); count = 0; error = 0; while (clen > 0) { rsize = roundup2(RULESIZE(r), sizeof(uint64_t)); if (rsize > clen || ctlv->count <= count) { error = EINVAL; break; } ci->ctlv = tstate; error = check_ipfw_rule1(r, rsize, ci); if (error != 0) break; /* Check sorting */ if (r->rulenum != 0 && r->rulenum < idx) { printf("rulenum %d idx %d\n", r->rulenum, idx); error = EINVAL; break; } idx = r->rulenum; ci->urule = (caddr_t)r; rsize = roundup2(rsize, sizeof(uint64_t)); clen -= rsize; r = (struct ip_fw_rule *)((caddr_t)r + rsize); count++; ci++; } if (ctlv->count != count || error != 0) { if (cbuf != &rci) free(cbuf, M_TEMP); return (EINVAL); } rtlv = ctlv; read += ctlv->head.length; ctlv = (ipfw_obj_ctlv *)((caddr_t)ctlv + ctlv->head.length); } if (read != sd->valsize || rtlv == NULL || rtlv->count == 0) { if (cbuf != NULL && cbuf != &rci) free(cbuf, M_TEMP); return (EINVAL); } /* * Passed rules seems to be valid. * Allocate storage and try to add them to chain. */ for (i = 0, ci = cbuf; i < rtlv->count; i++, ci++) { clen = RULEKSIZE1((struct ip_fw_rule *)ci->urule); ci->krule = ipfw_alloc_rule(chain, clen); import_rule1(ci); } if ((error = commit_rules(chain, cbuf, rtlv->count)) != 0) { /* Free allocate krules */ for (i = 0, ci = cbuf; i < rtlv->count; i++, ci++) free_rule(ci->krule); } if (cbuf != NULL && cbuf != &rci) free(cbuf, M_TEMP); return (error); } /* * Lists all sopts currently registered. * Data layout (v0)(current): * Request: [ ipfw_obj_lheader ], size = ipfw_obj_lheader.size * Reply: [ ipfw_obj_lheader ipfw_sopt_info x N ] * * Returns 0 on success */ static int dump_soptcodes(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { struct _ipfw_obj_lheader *olh; ipfw_sopt_info *i; struct ipfw_sopt_handler *sh; uint32_t count, n, size; olh = (struct _ipfw_obj_lheader *)ipfw_get_sopt_header(sd,sizeof(*olh)); if (olh == NULL) return (EINVAL); if (sd->valsize < olh->size) return (EINVAL); CTL3_LOCK(); count = ctl3_hsize; size = count * sizeof(ipfw_sopt_info) + sizeof(ipfw_obj_lheader); /* Fill in header regadless of buffer size */ olh->count = count; olh->objsize = sizeof(ipfw_sopt_info); if (size > olh->size) { olh->size = size; CTL3_UNLOCK(); return (ENOMEM); } olh->size = size; for (n = 1; n <= count; n++) { i = (ipfw_sopt_info *)ipfw_get_sopt_space(sd, sizeof(*i)); KASSERT(i != NULL, ("previously checked buffer is not enough")); sh = &ctl3_handlers[n]; i->opcode = sh->opcode; i->version = sh->version; i->refcnt = sh->refcnt; } CTL3_UNLOCK(); return (0); } /* * Compares two opcodes. * Used both in qsort() and bsearch(). * * Returns 0 if match is found. */ static int compare_opcodes(const void *_a, const void *_b) { const struct opcode_obj_rewrite *a, *b; a = (const struct opcode_obj_rewrite *)_a; b = (const struct opcode_obj_rewrite *)_b; if (a->opcode < b->opcode) return (-1); else if (a->opcode > b->opcode) return (1); return (0); } /* * XXX: Rewrite bsearch() */ static int find_op_rw_range(uint16_t op, struct opcode_obj_rewrite **plo, struct opcode_obj_rewrite **phi) { struct opcode_obj_rewrite *ctl3_max, *lo, *hi, h, *rw; memset(&h, 0, sizeof(h)); h.opcode = op; rw = (struct opcode_obj_rewrite *)bsearch(&h, ctl3_rewriters, ctl3_rsize, sizeof(h), compare_opcodes); if (rw == NULL) return (1); /* Find the first element matching the same opcode */ lo = rw; for ( ; lo > ctl3_rewriters && (lo - 1)->opcode == op; lo--) ; /* Find the last element matching the same opcode */ hi = rw; ctl3_max = ctl3_rewriters + ctl3_rsize; for ( ; (hi + 1) < ctl3_max && (hi + 1)->opcode == op; hi++) ; *plo = lo; *phi = hi; return (0); } /* * Finds opcode object rewriter based on @code. * * Returns pointer to handler or NULL. */ static struct opcode_obj_rewrite * find_op_rw(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) { struct opcode_obj_rewrite *rw, *lo, *hi; uint16_t uidx; uint8_t subtype; if (find_op_rw_range(cmd->opcode, &lo, &hi) != 0) return (NULL); for (rw = lo; rw <= hi; rw++) { if (rw->classifier(cmd, &uidx, &subtype) == 0) { if (puidx != NULL) *puidx = uidx; if (ptype != NULL) *ptype = subtype; return (rw); } } return (NULL); } int classify_opcode_kidx(ipfw_insn *cmd, uint16_t *puidx) { if (find_op_rw(cmd, puidx, NULL) == NULL) return (1); return (0); } void update_opcode_kidx(ipfw_insn *cmd, uint16_t idx) { struct opcode_obj_rewrite *rw; rw = find_op_rw(cmd, NULL, NULL); KASSERT(rw != NULL, ("No handler to update opcode %d", cmd->opcode)); rw->update(cmd, idx); } void ipfw_init_obj_rewriter() { ctl3_rewriters = NULL; ctl3_rsize = 0; } void ipfw_destroy_obj_rewriter() { if (ctl3_rewriters != NULL) free(ctl3_rewriters, M_IPFW); ctl3_rewriters = NULL; ctl3_rsize = 0; } /* * Adds one or more opcode object rewrite handlers to the global array. * Function may sleep. */ void ipfw_add_obj_rewriter(struct opcode_obj_rewrite *rw, size_t count) { size_t sz; struct opcode_obj_rewrite *tmp; CTL3_LOCK(); for (;;) { sz = ctl3_rsize + count; CTL3_UNLOCK(); tmp = malloc(sizeof(*rw) * sz, M_IPFW, M_WAITOK | M_ZERO); CTL3_LOCK(); if (ctl3_rsize + count <= sz) break; /* Retry */ free(tmp, M_IPFW); } /* Merge old & new arrays */ sz = ctl3_rsize + count; memcpy(tmp, ctl3_rewriters, ctl3_rsize * sizeof(*rw)); memcpy(&tmp[ctl3_rsize], rw, count * sizeof(*rw)); qsort(tmp, sz, sizeof(*rw), compare_opcodes); /* Switch new and free old */ if (ctl3_rewriters != NULL) free(ctl3_rewriters, M_IPFW); ctl3_rewriters = tmp; ctl3_rsize = sz; CTL3_UNLOCK(); } /* * Removes one or more object rewrite handlers from the global array. */ int ipfw_del_obj_rewriter(struct opcode_obj_rewrite *rw, size_t count) { size_t sz; struct opcode_obj_rewrite *ctl3_max, *ktmp, *lo, *hi; int i; CTL3_LOCK(); for (i = 0; i < count; i++) { if (find_op_rw_range(rw[i].opcode, &lo, &hi) != 0) continue; for (ktmp = lo; ktmp <= hi; ktmp++) { if (ktmp->classifier != rw[i].classifier) continue; ctl3_max = ctl3_rewriters + ctl3_rsize; sz = (ctl3_max - (ktmp + 1)) * sizeof(*ktmp); memmove(ktmp, ktmp + 1, sz); ctl3_rsize--; break; } } if (ctl3_rsize == 0) { if (ctl3_rewriters != NULL) free(ctl3_rewriters, M_IPFW); ctl3_rewriters = NULL; } CTL3_UNLOCK(); return (0); } static int export_objhash_ntlv_internal(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct sockopt_data *sd; ipfw_obj_ntlv *ntlv; sd = (struct sockopt_data *)arg; ntlv = (ipfw_obj_ntlv *)ipfw_get_sopt_space(sd, sizeof(*ntlv)); if (ntlv == NULL) return (ENOMEM); ipfw_export_obj_ntlv(no, ntlv); return (0); } /* * Lists all service objects. * Data layout (v0)(current): * Request: [ ipfw_obj_lheader ] size = ipfw_obj_lheader.size * Reply: [ ipfw_obj_lheader [ ipfw_obj_ntlv x N ] (optional) ] * Returns 0 on success */ static int dump_srvobjects(struct ip_fw_chain *chain, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_obj_lheader *hdr; int count; hdr = (ipfw_obj_lheader *)ipfw_get_sopt_header(sd, sizeof(*hdr)); if (hdr == NULL) return (EINVAL); IPFW_UH_RLOCK(chain); count = ipfw_objhash_count(CHAIN_TO_SRV(chain)); hdr->size = sizeof(ipfw_obj_lheader) + count * sizeof(ipfw_obj_ntlv); if (sd->valsize < hdr->size) { IPFW_UH_RUNLOCK(chain); return (ENOMEM); } hdr->count = count; hdr->objsize = sizeof(ipfw_obj_ntlv); if (count > 0) ipfw_objhash_foreach(CHAIN_TO_SRV(chain), export_objhash_ntlv_internal, sd); IPFW_UH_RUNLOCK(chain); return (0); } /* * Compares two sopt handlers (code, version and handler ptr). * Used both as qsort() and bsearch(). * Does not compare handler for latter case. * * Returns 0 if match is found. */ static int compare_sh(const void *_a, const void *_b) { const struct ipfw_sopt_handler *a, *b; a = (const struct ipfw_sopt_handler *)_a; b = (const struct ipfw_sopt_handler *)_b; if (a->opcode < b->opcode) return (-1); else if (a->opcode > b->opcode) return (1); if (a->version < b->version) return (-1); else if (a->version > b->version) return (1); /* bsearch helper */ if (a->handler == NULL) return (0); if ((uintptr_t)a->handler < (uintptr_t)b->handler) return (-1); else if ((uintptr_t)a->handler > (uintptr_t)b->handler) return (1); return (0); } /* * Finds sopt handler based on @code and @version. * * Returns pointer to handler or NULL. */ static struct ipfw_sopt_handler * find_sh(uint16_t code, uint8_t version, sopt_handler_f *handler) { struct ipfw_sopt_handler *sh, h; memset(&h, 0, sizeof(h)); h.opcode = code; h.version = version; h.handler = handler; sh = (struct ipfw_sopt_handler *)bsearch(&h, ctl3_handlers, ctl3_hsize, sizeof(h), compare_sh); return (sh); } static int find_ref_sh(uint16_t opcode, uint8_t version, struct ipfw_sopt_handler *psh) { struct ipfw_sopt_handler *sh; CTL3_LOCK(); if ((sh = find_sh(opcode, version, NULL)) == NULL) { CTL3_UNLOCK(); printf("ipfw: ipfw_ctl3 invalid option %d""v""%d\n", opcode, version); return (EINVAL); } sh->refcnt++; ctl3_refct++; /* Copy handler data to requested buffer */ *psh = *sh; CTL3_UNLOCK(); return (0); } static void find_unref_sh(struct ipfw_sopt_handler *psh) { struct ipfw_sopt_handler *sh; CTL3_LOCK(); sh = find_sh(psh->opcode, psh->version, NULL); KASSERT(sh != NULL, ("ctl3 handler disappeared")); sh->refcnt--; ctl3_refct--; CTL3_UNLOCK(); } void ipfw_init_sopt_handler() { CTL3_LOCK_INIT(); IPFW_ADD_SOPT_HANDLER(1, scodes); } void ipfw_destroy_sopt_handler() { IPFW_DEL_SOPT_HANDLER(1, scodes); CTL3_LOCK_DESTROY(); } /* * Adds one or more sockopt handlers to the global array. * Function may sleep. */ void ipfw_add_sopt_handler(struct ipfw_sopt_handler *sh, size_t count) { size_t sz; struct ipfw_sopt_handler *tmp; CTL3_LOCK(); for (;;) { sz = ctl3_hsize + count; CTL3_UNLOCK(); tmp = malloc(sizeof(*sh) * sz, M_IPFW, M_WAITOK | M_ZERO); CTL3_LOCK(); if (ctl3_hsize + count <= sz) break; /* Retry */ free(tmp, M_IPFW); } /* Merge old & new arrays */ sz = ctl3_hsize + count; memcpy(tmp, ctl3_handlers, ctl3_hsize * sizeof(*sh)); memcpy(&tmp[ctl3_hsize], sh, count * sizeof(*sh)); qsort(tmp, sz, sizeof(*sh), compare_sh); /* Switch new and free old */ if (ctl3_handlers != NULL) free(ctl3_handlers, M_IPFW); ctl3_handlers = tmp; ctl3_hsize = sz; ctl3_gencnt++; CTL3_UNLOCK(); } /* * Removes one or more sockopt handlers from the global array. */ int ipfw_del_sopt_handler(struct ipfw_sopt_handler *sh, size_t count) { size_t sz; struct ipfw_sopt_handler *tmp, *h; int i; CTL3_LOCK(); for (i = 0; i < count; i++) { tmp = &sh[i]; h = find_sh(tmp->opcode, tmp->version, tmp->handler); if (h == NULL) continue; sz = (ctl3_handlers + ctl3_hsize - (h + 1)) * sizeof(*h); memmove(h, h + 1, sz); ctl3_hsize--; } if (ctl3_hsize == 0) { if (ctl3_handlers != NULL) free(ctl3_handlers, M_IPFW); ctl3_handlers = NULL; } ctl3_gencnt++; CTL3_UNLOCK(); return (0); } /* * Writes data accumulated in @sd to sockopt buffer. * Zeroes internal @sd buffer. */ static int ipfw_flush_sopt_data(struct sockopt_data *sd) { struct sockopt *sopt; int error; size_t sz; sz = sd->koff; if (sz == 0) return (0); sopt = sd->sopt; if (sopt->sopt_dir == SOPT_GET) { error = copyout(sd->kbuf, sopt->sopt_val, sz); if (error != 0) return (error); } memset(sd->kbuf, 0, sd->ksize); sd->ktotal += sz; sd->koff = 0; if (sd->ktotal + sd->ksize < sd->valsize) sd->kavail = sd->ksize; else sd->kavail = sd->valsize - sd->ktotal; /* Update sopt buffer data */ sopt->sopt_valsize = sd->ktotal; sopt->sopt_val = sd->sopt_val + sd->ktotal; return (0); } /* * Ensures that @sd buffer has contiguous @neeeded number of * bytes. * * Returns pointer to requested space or NULL. */ caddr_t ipfw_get_sopt_space(struct sockopt_data *sd, size_t needed) { int error; caddr_t addr; if (sd->kavail < needed) { /* * Flush data and try another time. */ error = ipfw_flush_sopt_data(sd); if (sd->kavail < needed || error != 0) return (NULL); } addr = sd->kbuf + sd->koff; sd->koff += needed; sd->kavail -= needed; return (addr); } /* * Requests @needed contiguous bytes from @sd buffer. * Function is used to notify subsystem that we are * interesed in first @needed bytes (request header) * and the rest buffer can be safely zeroed. * * Returns pointer to requested space or NULL. */ caddr_t ipfw_get_sopt_header(struct sockopt_data *sd, size_t needed) { caddr_t addr; if ((addr = ipfw_get_sopt_space(sd, needed)) == NULL) return (NULL); if (sd->kavail > 0) memset(sd->kbuf + sd->koff, 0, sd->kavail); return (addr); } /* * New sockopt handler. */ int ipfw_ctl3(struct sockopt *sopt) { int error, locked; size_t size, valsize; struct ip_fw_chain *chain; char xbuf[256]; struct sockopt_data sdata; struct ipfw_sopt_handler h; ip_fw3_opheader *op3 = NULL; error = priv_check(sopt->sopt_td, PRIV_NETINET_IPFW); if (error != 0) return (error); if (sopt->sopt_name != IP_FW3) return (ipfw_ctl(sopt)); chain = &V_layer3_chain; error = 0; /* Save original valsize before it is altered via sooptcopyin() */ valsize = sopt->sopt_valsize; memset(&sdata, 0, sizeof(sdata)); /* Read op3 header first to determine actual operation */ op3 = (ip_fw3_opheader *)xbuf; error = sooptcopyin(sopt, op3, sizeof(*op3), sizeof(*op3)); if (error != 0) return (error); sopt->sopt_valsize = valsize; /* * Find and reference command. */ error = find_ref_sh(op3->opcode, op3->version, &h); if (error != 0) return (error); /* * Disallow modifications in really-really secure mode, but still allow * the logging counters to be reset. */ if ((h.dir & HDIR_SET) != 0 && h.opcode != IP_FW_XRESETLOG) { error = securelevel_ge(sopt->sopt_td->td_ucred, 3); if (error != 0) { find_unref_sh(&h); return (error); } } /* * Fill in sockopt_data structure that may be useful for * IP_FW3 get requests. */ locked = 0; if (valsize <= sizeof(xbuf)) { /* use on-stack buffer */ sdata.kbuf = xbuf; sdata.ksize = sizeof(xbuf); sdata.kavail = valsize; } else { /* * Determine opcode type/buffer size: * allocate sliding-window buf for data export or * contiguous buffer for special ops. */ if ((h.dir & HDIR_SET) != 0) { /* Set request. Allocate contigous buffer. */ if (valsize > CTL3_LARGEBUF) { find_unref_sh(&h); return (EFBIG); } size = valsize; } else { /* Get request. Allocate sliding window buffer */ size = (valsizesopt_val, valsize); if (error != 0) return (error); locked = 1; } } sdata.kbuf = malloc(size, M_TEMP, M_WAITOK | M_ZERO); sdata.ksize = size; sdata.kavail = size; } sdata.sopt = sopt; sdata.sopt_val = sopt->sopt_val; sdata.valsize = valsize; /* * Copy either all request (if valsize < bsize_max) * or first bsize_max bytes to guarantee most consumers * that all necessary data has been copied). * Anyway, copy not less than sizeof(ip_fw3_opheader). */ if ((error = sooptcopyin(sopt, sdata.kbuf, sdata.ksize, sizeof(ip_fw3_opheader))) != 0) return (error); op3 = (ip_fw3_opheader *)sdata.kbuf; /* Finally, run handler */ error = h.handler(chain, op3, &sdata); find_unref_sh(&h); /* Flush state and free buffers */ if (error == 0) error = ipfw_flush_sopt_data(&sdata); else ipfw_flush_sopt_data(&sdata); if (locked != 0) vsunlock(sdata.sopt_val, valsize); /* Restore original pointer and set number of bytes written */ sopt->sopt_val = sdata.sopt_val; sopt->sopt_valsize = sdata.ktotal; if (sdata.kbuf != xbuf) free(sdata.kbuf, M_TEMP); return (error); } /** * {set|get}sockopt parser. */ int ipfw_ctl(struct sockopt *sopt) { #define RULE_MAXSIZE (512*sizeof(u_int32_t)) int error; size_t size, valsize; struct ip_fw *buf; struct ip_fw_rule0 *rule; struct ip_fw_chain *chain; u_int32_t rulenum[2]; uint32_t opt; struct rule_check_info ci; IPFW_RLOCK_TRACKER; chain = &V_layer3_chain; error = 0; /* Save original valsize before it is altered via sooptcopyin() */ valsize = sopt->sopt_valsize; opt = sopt->sopt_name; /* * Disallow modifications in really-really secure mode, but still allow * the logging counters to be reset. */ if (opt == IP_FW_ADD || (sopt->sopt_dir == SOPT_SET && opt != IP_FW_RESETLOG)) { error = securelevel_ge(sopt->sopt_td->td_ucred, 3); if (error != 0) return (error); } switch (opt) { case IP_FW_GET: /* * pass up a copy of the current rules. Static rules * come first (the last of which has number IPFW_DEFAULT_RULE), * followed by a possibly empty list of dynamic rule. * The last dynamic rule has NULL in the "next" field. * * Note that the calculated size is used to bound the * amount of data returned to the user. The rule set may * change between calculating the size and returning the * data in which case we'll just return what fits. */ for (;;) { int len = 0, want; size = chain->static_len; size += ipfw_dyn_len(); if (size >= sopt->sopt_valsize) break; buf = malloc(size, M_TEMP, M_WAITOK | M_ZERO); IPFW_UH_RLOCK(chain); /* check again how much space we need */ want = chain->static_len + ipfw_dyn_len(); if (size >= want) len = ipfw_getrules(chain, buf, size); IPFW_UH_RUNLOCK(chain); if (size >= want) error = sooptcopyout(sopt, buf, len); free(buf, M_TEMP); if (size >= want) break; } break; case IP_FW_FLUSH: /* locking is done within del_entry() */ error = del_entry(chain, 0); /* special case, rule=0, cmd=0 means all */ break; case IP_FW_ADD: rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK); error = sooptcopyin(sopt, rule, RULE_MAXSIZE, sizeof(struct ip_fw7) ); memset(&ci, 0, sizeof(struct rule_check_info)); /* * If the size of commands equals RULESIZE7 then we assume * a FreeBSD7.2 binary is talking to us (set is7=1). * is7 is persistent so the next 'ipfw list' command * will use this format. * NOTE: If wrong version is guessed (this can happen if * the first ipfw command is 'ipfw [pipe] list') * the ipfw binary may crash or loop infinitly... */ size = sopt->sopt_valsize; if (size == RULESIZE7(rule)) { is7 = 1; error = convert_rule_to_8(rule); if (error) { free(rule, M_TEMP); return error; } size = RULESIZE(rule); } else is7 = 0; if (error == 0) error = check_ipfw_rule0(rule, size, &ci); if (error == 0) { /* locking is done within add_rule() */ struct ip_fw *krule; krule = ipfw_alloc_rule(chain, RULEKSIZE0(rule)); ci.urule = (caddr_t)rule; ci.krule = krule; import_rule0(&ci); error = commit_rules(chain, &ci, 1); if (error != 0) free_rule(ci.krule); else if (sopt->sopt_dir == SOPT_GET) { if (is7) { error = convert_rule_to_7(rule); size = RULESIZE7(rule); if (error) { free(rule, M_TEMP); return error; } } error = sooptcopyout(sopt, rule, size); } } free(rule, M_TEMP); break; case IP_FW_DEL: /* * IP_FW_DEL is used for deleting single rules or sets, * and (ab)used to atomically manipulate sets. Argument size * is used to distinguish between the two: * sizeof(u_int32_t) * delete single rule or set of rules, * or reassign rules (or sets) to a different set. * 2*sizeof(u_int32_t) * atomic disable/enable sets. * first u_int32_t contains sets to be disabled, * second u_int32_t contains sets to be enabled. */ error = sooptcopyin(sopt, rulenum, 2*sizeof(u_int32_t), sizeof(u_int32_t)); if (error) break; size = sopt->sopt_valsize; if (size == sizeof(u_int32_t) && rulenum[0] != 0) { /* delete or reassign, locking done in del_entry() */ error = del_entry(chain, rulenum[0]); } else if (size == 2*sizeof(u_int32_t)) { /* set enable/disable */ IPFW_UH_WLOCK(chain); V_set_disable = (V_set_disable | rulenum[0]) & ~rulenum[1] & ~(1<sopt_val != 0) { error = sooptcopyin(sopt, rulenum, sizeof(u_int32_t), sizeof(u_int32_t)); if (error) break; } error = zero_entry(chain, rulenum[0], sopt->sopt_name == IP_FW_RESETLOG); break; /*--- TABLE opcodes ---*/ case IP_FW_TABLE_ADD: case IP_FW_TABLE_DEL: { ipfw_table_entry ent; struct tentry_info tei; struct tid_info ti; struct table_value v; error = sooptcopyin(sopt, &ent, sizeof(ent), sizeof(ent)); if (error) break; memset(&tei, 0, sizeof(tei)); tei.paddr = &ent.addr; tei.subtype = AF_INET; tei.masklen = ent.masklen; ipfw_import_table_value_legacy(ent.value, &v); tei.pvalue = &v; memset(&ti, 0, sizeof(ti)); ti.uidx = ent.tbl; ti.type = IPFW_TABLE_CIDR; error = (opt == IP_FW_TABLE_ADD) ? add_table_entry(chain, &ti, &tei, 0, 1) : del_table_entry(chain, &ti, &tei, 0, 1); } break; case IP_FW_TABLE_FLUSH: { u_int16_t tbl; struct tid_info ti; error = sooptcopyin(sopt, &tbl, sizeof(tbl), sizeof(tbl)); if (error) break; memset(&ti, 0, sizeof(ti)); ti.uidx = tbl; error = flush_table(chain, &ti); } break; case IP_FW_TABLE_GETSIZE: { u_int32_t tbl, cnt; struct tid_info ti; if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl), sizeof(tbl)))) break; memset(&ti, 0, sizeof(ti)); ti.uidx = tbl; IPFW_RLOCK(chain); error = ipfw_count_table(chain, &ti, &cnt); IPFW_RUNLOCK(chain); if (error) break; error = sooptcopyout(sopt, &cnt, sizeof(cnt)); } break; case IP_FW_TABLE_LIST: { ipfw_table *tbl; struct tid_info ti; if (sopt->sopt_valsize < sizeof(*tbl)) { error = EINVAL; break; } size = sopt->sopt_valsize; tbl = malloc(size, M_TEMP, M_WAITOK); error = sooptcopyin(sopt, tbl, size, sizeof(*tbl)); if (error) { free(tbl, M_TEMP); break; } tbl->size = (size - sizeof(*tbl)) / sizeof(ipfw_table_entry); memset(&ti, 0, sizeof(ti)); ti.uidx = tbl->tbl; IPFW_RLOCK(chain); error = ipfw_dump_table_legacy(chain, &ti, tbl); IPFW_RUNLOCK(chain); if (error) { free(tbl, M_TEMP); break; } error = sooptcopyout(sopt, tbl, size); free(tbl, M_TEMP); } break; /*--- NAT operations are protected by the IPFW_LOCK ---*/ case IP_FW_NAT_CFG: if (IPFW_NAT_LOADED) error = ipfw_nat_cfg_ptr(sopt); else { printf("IP_FW_NAT_CFG: %s\n", "ipfw_nat not present, please load it"); error = EINVAL; } break; case IP_FW_NAT_DEL: if (IPFW_NAT_LOADED) error = ipfw_nat_del_ptr(sopt); else { printf("IP_FW_NAT_DEL: %s\n", "ipfw_nat not present, please load it"); error = EINVAL; } break; case IP_FW_NAT_GET_CONFIG: if (IPFW_NAT_LOADED) error = ipfw_nat_get_cfg_ptr(sopt); else { printf("IP_FW_NAT_GET_CFG: %s\n", "ipfw_nat not present, please load it"); error = EINVAL; } break; case IP_FW_NAT_GET_LOG: if (IPFW_NAT_LOADED) error = ipfw_nat_get_log_ptr(sopt); else { printf("IP_FW_NAT_GET_LOG: %s\n", "ipfw_nat not present, please load it"); error = EINVAL; } break; default: printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name); error = EINVAL; } return (error); #undef RULE_MAXSIZE } #define RULE_MAXSIZE (256*sizeof(u_int32_t)) /* Functions to convert rules 7.2 <==> 8.0 */ static int convert_rule_to_7(struct ip_fw_rule0 *rule) { /* Used to modify original rule */ struct ip_fw7 *rule7 = (struct ip_fw7 *)rule; /* copy of original rule, version 8 */ struct ip_fw_rule0 *tmp; /* Used to copy commands */ ipfw_insn *ccmd, *dst; int ll = 0, ccmdlen = 0; tmp = malloc(RULE_MAXSIZE, M_TEMP, M_NOWAIT | M_ZERO); if (tmp == NULL) { return 1; //XXX error } bcopy(rule, tmp, RULE_MAXSIZE); /* Copy fields */ //rule7->_pad = tmp->_pad; rule7->set = tmp->set; rule7->rulenum = tmp->rulenum; rule7->cmd_len = tmp->cmd_len; rule7->act_ofs = tmp->act_ofs; rule7->next_rule = (struct ip_fw7 *)tmp->next_rule; rule7->cmd_len = tmp->cmd_len; rule7->pcnt = tmp->pcnt; rule7->bcnt = tmp->bcnt; rule7->timestamp = tmp->timestamp; /* Copy commands */ for (ll = tmp->cmd_len, ccmd = tmp->cmd, dst = rule7->cmd ; ll > 0 ; ll -= ccmdlen, ccmd += ccmdlen, dst += ccmdlen) { ccmdlen = F_LEN(ccmd); bcopy(ccmd, dst, F_LEN(ccmd)*sizeof(uint32_t)); if (dst->opcode > O_NAT) /* O_REASS doesn't exists in 7.2 version, so * decrement opcode if it is after O_REASS */ dst->opcode--; if (ccmdlen > ll) { printf("ipfw: opcode %d size truncated\n", ccmd->opcode); return EINVAL; } } free(tmp, M_TEMP); return 0; } static int convert_rule_to_8(struct ip_fw_rule0 *rule) { /* Used to modify original rule */ struct ip_fw7 *rule7 = (struct ip_fw7 *) rule; /* Used to copy commands */ ipfw_insn *ccmd, *dst; int ll = 0, ccmdlen = 0; /* Copy of original rule */ struct ip_fw7 *tmp = malloc(RULE_MAXSIZE, M_TEMP, M_NOWAIT | M_ZERO); if (tmp == NULL) { return 1; //XXX error } bcopy(rule7, tmp, RULE_MAXSIZE); for (ll = tmp->cmd_len, ccmd = tmp->cmd, dst = rule->cmd ; ll > 0 ; ll -= ccmdlen, ccmd += ccmdlen, dst += ccmdlen) { ccmdlen = F_LEN(ccmd); bcopy(ccmd, dst, F_LEN(ccmd)*sizeof(uint32_t)); if (dst->opcode > O_NAT) /* O_REASS doesn't exists in 7.2 version, so * increment opcode if it is after O_REASS */ dst->opcode++; if (ccmdlen > ll) { printf("ipfw: opcode %d size truncated\n", ccmd->opcode); return EINVAL; } } rule->_pad = tmp->_pad; rule->set = tmp->set; rule->rulenum = tmp->rulenum; rule->cmd_len = tmp->cmd_len; rule->act_ofs = tmp->act_ofs; rule->next_rule = (struct ip_fw *)tmp->next_rule; rule->cmd_len = tmp->cmd_len; rule->id = 0; /* XXX see if is ok = 0 */ rule->pcnt = tmp->pcnt; rule->bcnt = tmp->bcnt; rule->timestamp = tmp->timestamp; free (tmp, M_TEMP); return 0; } /* * Named object api * */ void ipfw_init_srv(struct ip_fw_chain *ch) { ch->srvmap = ipfw_objhash_create(IPFW_OBJECTS_DEFAULT); ch->srvstate = malloc(sizeof(void *) * IPFW_OBJECTS_DEFAULT, M_IPFW, M_WAITOK | M_ZERO); } void ipfw_destroy_srv(struct ip_fw_chain *ch) { free(ch->srvstate, M_IPFW); ipfw_objhash_destroy(ch->srvmap); } /* * Allocate new bitmask which can be used to enlarge/shrink * named instance index. */ void ipfw_objhash_bitmap_alloc(uint32_t items, void **idx, int *pblocks) { size_t size; int max_blocks; u_long *idx_mask; KASSERT((items % BLOCK_ITEMS) == 0, ("bitmask size needs to power of 2 and greater or equal to %zu", BLOCK_ITEMS)); max_blocks = items / BLOCK_ITEMS; size = items / 8; idx_mask = malloc(size * IPFW_MAX_SETS, M_IPFW, M_WAITOK); /* Mark all as free */ memset(idx_mask, 0xFF, size * IPFW_MAX_SETS); *idx_mask &= ~(u_long)1; /* Skip index 0 */ *idx = idx_mask; *pblocks = max_blocks; } /* * Copy current bitmask index to new one. */ void ipfw_objhash_bitmap_merge(struct namedobj_instance *ni, void **idx, int *blocks) { int old_blocks, new_blocks; u_long *old_idx, *new_idx; int i; old_idx = ni->idx_mask; old_blocks = ni->max_blocks; new_idx = *idx; new_blocks = *blocks; for (i = 0; i < IPFW_MAX_SETS; i++) { memcpy(&new_idx[new_blocks * i], &old_idx[old_blocks * i], old_blocks * sizeof(u_long)); } } /* * Swaps current @ni index with new one. */ void ipfw_objhash_bitmap_swap(struct namedobj_instance *ni, void **idx, int *blocks) { int old_blocks; u_long *old_idx; old_idx = ni->idx_mask; old_blocks = ni->max_blocks; ni->idx_mask = *idx; ni->max_blocks = *blocks; /* Save old values */ *idx = old_idx; *blocks = old_blocks; } void ipfw_objhash_bitmap_free(void *idx, int blocks) { free(idx, M_IPFW); } /* * Creates named hash instance. * Must be called without holding any locks. * Return pointer to new instance. */ struct namedobj_instance * ipfw_objhash_create(uint32_t items) { struct namedobj_instance *ni; int i; size_t size; size = sizeof(struct namedobj_instance) + sizeof(struct namedobjects_head) * NAMEDOBJ_HASH_SIZE + sizeof(struct namedobjects_head) * NAMEDOBJ_HASH_SIZE; ni = malloc(size, M_IPFW, M_WAITOK | M_ZERO); ni->nn_size = NAMEDOBJ_HASH_SIZE; ni->nv_size = NAMEDOBJ_HASH_SIZE; ni->names = (struct namedobjects_head *)(ni +1); ni->values = &ni->names[ni->nn_size]; for (i = 0; i < ni->nn_size; i++) TAILQ_INIT(&ni->names[i]); for (i = 0; i < ni->nv_size; i++) TAILQ_INIT(&ni->values[i]); /* Set default hashing/comparison functions */ ni->hash_f = objhash_hash_name; ni->cmp_f = objhash_cmp_name; /* Allocate bitmask separately due to possible resize */ ipfw_objhash_bitmap_alloc(items, (void*)&ni->idx_mask, &ni->max_blocks); return (ni); } void ipfw_objhash_destroy(struct namedobj_instance *ni) { free(ni->idx_mask, M_IPFW); free(ni, M_IPFW); } void ipfw_objhash_set_funcs(struct namedobj_instance *ni, objhash_hash_f *hash_f, objhash_cmp_f *cmp_f) { ni->hash_f = hash_f; ni->cmp_f = cmp_f; } static uint32_t objhash_hash_name(struct namedobj_instance *ni, const void *name, uint32_t set) { return (fnv_32_str((const char *)name, FNV1_32_INIT)); } static int objhash_cmp_name(struct named_object *no, const void *name, uint32_t set) { if ((strcmp(no->name, (const char *)name) == 0) && (no->set == set)) return (0); return (1); } static uint32_t objhash_hash_idx(struct namedobj_instance *ni, uint32_t val) { uint32_t v; v = val % (ni->nv_size - 1); return (v); } struct named_object * ipfw_objhash_lookup_name(struct namedobj_instance *ni, uint32_t set, char *name) { struct named_object *no; uint32_t hash; hash = ni->hash_f(ni, name, set) % ni->nn_size; TAILQ_FOREACH(no, &ni->names[hash], nn_next) { if (ni->cmp_f(no, name, set) == 0) return (no); } return (NULL); } /* * Find named object by @uid. * Check @tlvs for valid data inside. * * Returns pointer to found TLV or NULL. */ ipfw_obj_ntlv * ipfw_find_name_tlv_type(void *tlvs, int len, uint16_t uidx, uint32_t etlv) { ipfw_obj_ntlv *ntlv; uintptr_t pa, pe; int l; pa = (uintptr_t)tlvs; pe = pa + len; l = 0; for (; pa < pe; pa += l) { ntlv = (ipfw_obj_ntlv *)pa; l = ntlv->head.length; if (l != sizeof(*ntlv)) return (NULL); if (ntlv->idx != uidx) continue; /* * When userland has specified zero TLV type, do * not compare it with eltv. In some cases userland * doesn't know what type should it have. Use only * uidx and name for search named_object. */ if (ntlv->head.type != 0 && ntlv->head.type != (uint16_t)etlv) continue; if (ipfw_check_object_name_generic(ntlv->name) != 0) return (NULL); return (ntlv); } return (NULL); } /* * Finds object config based on either legacy index * or name in ntlv. * Note @ti structure contains unchecked data from userland. * * Returns 0 in success and fills in @pno with found config */ int ipfw_objhash_find_type(struct namedobj_instance *ni, struct tid_info *ti, uint32_t etlv, struct named_object **pno) { char *name; ipfw_obj_ntlv *ntlv; uint32_t set; if (ti->tlvs == NULL) return (EINVAL); ntlv = ipfw_find_name_tlv_type(ti->tlvs, ti->tlen, ti->uidx, etlv); if (ntlv == NULL) return (EINVAL); name = ntlv->name; /* * Use set provided by @ti instead of @ntlv one. * This is needed due to different sets behavior * controlled by V_fw_tables_sets. */ set = ti->set; *pno = ipfw_objhash_lookup_name(ni, set, name); if (*pno == NULL) return (ESRCH); return (0); } /* * Find named object by name, considering also its TLV type. */ struct named_object * ipfw_objhash_lookup_name_type(struct namedobj_instance *ni, uint32_t set, uint32_t type, const char *name) { struct named_object *no; uint32_t hash; hash = ni->hash_f(ni, name, set) % ni->nn_size; TAILQ_FOREACH(no, &ni->names[hash], nn_next) { if (ni->cmp_f(no, name, set) == 0 && no->etlv == (uint16_t)type) return (no); } return (NULL); } struct named_object * ipfw_objhash_lookup_kidx(struct namedobj_instance *ni, uint16_t kidx) { struct named_object *no; uint32_t hash; hash = objhash_hash_idx(ni, kidx); TAILQ_FOREACH(no, &ni->values[hash], nv_next) { if (no->kidx == kidx) return (no); } return (NULL); } int ipfw_objhash_same_name(struct namedobj_instance *ni, struct named_object *a, struct named_object *b) { if ((strcmp(a->name, b->name) == 0) && a->set == b->set) return (1); return (0); } void ipfw_objhash_add(struct namedobj_instance *ni, struct named_object *no) { uint32_t hash; hash = ni->hash_f(ni, no->name, no->set) % ni->nn_size; TAILQ_INSERT_HEAD(&ni->names[hash], no, nn_next); hash = objhash_hash_idx(ni, no->kidx); TAILQ_INSERT_HEAD(&ni->values[hash], no, nv_next); ni->count++; } void ipfw_objhash_del(struct namedobj_instance *ni, struct named_object *no) { uint32_t hash; hash = ni->hash_f(ni, no->name, no->set) % ni->nn_size; TAILQ_REMOVE(&ni->names[hash], no, nn_next); hash = objhash_hash_idx(ni, no->kidx); TAILQ_REMOVE(&ni->values[hash], no, nv_next); ni->count--; } uint32_t ipfw_objhash_count(struct namedobj_instance *ni) { return (ni->count); } uint32_t ipfw_objhash_count_type(struct namedobj_instance *ni, uint16_t type) { struct named_object *no; uint32_t count; int i; count = 0; for (i = 0; i < ni->nn_size; i++) { TAILQ_FOREACH(no, &ni->names[i], nn_next) { if (no->etlv == type) count++; } } return (count); } /* * Runs @func for each found named object. * It is safe to delete objects from callback */ int ipfw_objhash_foreach(struct namedobj_instance *ni, objhash_cb_t *f, void *arg) { struct named_object *no, *no_tmp; int i, ret; for (i = 0; i < ni->nn_size; i++) { TAILQ_FOREACH_SAFE(no, &ni->names[i], nn_next, no_tmp) { ret = f(ni, no, arg); if (ret != 0) return (ret); } } return (0); } /* * Runs @f for each found named object with type @type. * It is safe to delete objects from callback */ int ipfw_objhash_foreach_type(struct namedobj_instance *ni, objhash_cb_t *f, void *arg, uint16_t type) { struct named_object *no, *no_tmp; int i, ret; for (i = 0; i < ni->nn_size; i++) { TAILQ_FOREACH_SAFE(no, &ni->names[i], nn_next, no_tmp) { if (no->etlv != type) continue; ret = f(ni, no, arg); if (ret != 0) return (ret); } } return (0); } /* * Removes index from given set. * Returns 0 on success. */ int ipfw_objhash_free_idx(struct namedobj_instance *ni, uint16_t idx) { u_long *mask; int i, v; i = idx / BLOCK_ITEMS; v = idx % BLOCK_ITEMS; if (i >= ni->max_blocks) return (1); mask = &ni->idx_mask[i]; if ((*mask & ((u_long)1 << v)) != 0) return (1); /* Mark as free */ *mask |= (u_long)1 << v; /* Update free offset */ if (ni->free_off[0] > i) ni->free_off[0] = i; return (0); } /* * Allocate new index in given instance and stores in in @pidx. * Returns 0 on success. */ int ipfw_objhash_alloc_idx(void *n, uint16_t *pidx) { struct namedobj_instance *ni; u_long *mask; int i, off, v; ni = (struct namedobj_instance *)n; off = ni->free_off[0]; mask = &ni->idx_mask[off]; for (i = off; i < ni->max_blocks; i++, mask++) { if ((v = ffsl(*mask)) == 0) continue; /* Mark as busy */ *mask &= ~ ((u_long)1 << (v - 1)); ni->free_off[0] = i; v = BLOCK_ITEMS * i + v - 1; *pidx = v; return (0); } return (1); } /* end of file */ Index: head/sys/netpfil/ipfw/nptv6/nptv6.c =================================================================== --- head/sys/netpfil/ipfw/nptv6/nptv6.c (revision 336675) +++ head/sys/netpfil/ipfw/nptv6/nptv6.c (revision 336676) @@ -1,889 +1,889 @@ /*- * Copyright (c) 2016 Yandex LLC * Copyright (c) 2016 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include -static VNET_DEFINE(uint16_t, nptv6_eid) = 0; +VNET_DEFINE_STATIC(uint16_t, nptv6_eid) = 0; #define V_nptv6_eid VNET(nptv6_eid) #define IPFW_TLV_NPTV6_NAME IPFW_TLV_EACTION_NAME(V_nptv6_eid) static struct nptv6_cfg *nptv6_alloc_config(const char *name, uint8_t set); static void nptv6_free_config(struct nptv6_cfg *cfg); static struct nptv6_cfg *nptv6_find(struct namedobj_instance *ni, const char *name, uint8_t set); static int nptv6_rewrite_internal(struct nptv6_cfg *cfg, struct mbuf **mp, int offset); static int nptv6_rewrite_external(struct nptv6_cfg *cfg, struct mbuf **mp, int offset); #define NPTV6_LOOKUP(chain, cmd) \ (struct nptv6_cfg *)SRV_OBJECT((chain), (cmd)->arg1) #ifndef IN6_MASK_ADDR #define IN6_MASK_ADDR(a, m) do { \ (a)->s6_addr32[0] &= (m)->s6_addr32[0]; \ (a)->s6_addr32[1] &= (m)->s6_addr32[1]; \ (a)->s6_addr32[2] &= (m)->s6_addr32[2]; \ (a)->s6_addr32[3] &= (m)->s6_addr32[3]; \ } while (0) #endif #ifndef IN6_ARE_MASKED_ADDR_EQUAL #define IN6_ARE_MASKED_ADDR_EQUAL(d, a, m) ( \ (((d)->s6_addr32[0] ^ (a)->s6_addr32[0]) & (m)->s6_addr32[0]) == 0 && \ (((d)->s6_addr32[1] ^ (a)->s6_addr32[1]) & (m)->s6_addr32[1]) == 0 && \ (((d)->s6_addr32[2] ^ (a)->s6_addr32[2]) & (m)->s6_addr32[2]) == 0 && \ (((d)->s6_addr32[3] ^ (a)->s6_addr32[3]) & (m)->s6_addr32[3]) == 0 ) #endif #if 0 #define NPTV6_DEBUG(fmt, ...) do { \ printf("%s: " fmt "\n", __func__, ## __VA_ARGS__); \ } while (0) #define NPTV6_IPDEBUG(fmt, ...) do { \ char _s[INET6_ADDRSTRLEN], _d[INET6_ADDRSTRLEN]; \ printf("%s: " fmt "\n", __func__, ## __VA_ARGS__); \ } while (0) #else #define NPTV6_DEBUG(fmt, ...) #define NPTV6_IPDEBUG(fmt, ...) #endif static int nptv6_getlasthdr(struct nptv6_cfg *cfg, struct mbuf *m, int *offset) { struct ip6_hdr *ip6; struct ip6_hbh *hbh; int proto, hlen; hlen = (offset == NULL) ? 0: *offset; if (m->m_len < hlen) return (-1); ip6 = mtodo(m, hlen); hlen += sizeof(*ip6); proto = ip6->ip6_nxt; while (proto == IPPROTO_HOPOPTS || proto == IPPROTO_ROUTING || proto == IPPROTO_DSTOPTS) { hbh = mtodo(m, hlen); if (m->m_len < hlen) return (-1); proto = hbh->ip6h_nxt; hlen += (hbh->ip6h_len + 1) << 3; } if (offset != NULL) *offset = hlen; return (proto); } static int nptv6_translate_icmpv6(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) { struct icmp6_hdr *icmp6; struct ip6_hdr *ip6; struct mbuf *m; m = *mp; if (offset > m->m_len) return (-1); icmp6 = mtodo(m, offset); NPTV6_DEBUG("ICMPv6 type %d", icmp6->icmp6_type); switch (icmp6->icmp6_type) { case ICMP6_DST_UNREACH: case ICMP6_PACKET_TOO_BIG: case ICMP6_TIME_EXCEEDED: case ICMP6_PARAM_PROB: break; case ICMP6_ECHO_REQUEST: case ICMP6_ECHO_REPLY: /* nothing to translate */ return (0); default: /* * XXX: We can add some checks to not translate NDP and MLD * messages. Currently user must explicitly allow these message * types, otherwise packets will be dropped. */ return (-1); } offset += sizeof(*icmp6); if (offset + sizeof(*ip6) > m->m_pkthdr.len) return (-1); if (offset + sizeof(*ip6) > m->m_len) *mp = m = m_pullup(m, offset + sizeof(*ip6)); if (m == NULL) return (-1); ip6 = mtodo(m, offset); NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), ip6->ip6_nxt); if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_src, &cfg->external, &cfg->mask)) return (nptv6_rewrite_external(cfg, mp, offset)); else if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, &cfg->internal, &cfg->mask)) return (nptv6_rewrite_internal(cfg, mp, offset)); /* * Addresses in the inner IPv6 header doesn't matched to * our prefixes. */ return (-1); } static int nptv6_search_index(struct nptv6_cfg *cfg, struct in6_addr *a) { int idx; if (cfg->flags & NPTV6_48PLEN) return (3); /* Search suitable word index for adjustment */ for (idx = 4; idx < 8; idx++) if (a->s6_addr16[idx] != 0xffff) break; /* * RFC 6296 p3.7: If an NPTv6 Translator discovers a datagram with * an IID of all-zeros while performing address mapping, that * datagram MUST be dropped, and an ICMPv6 Parameter Problem error * SHOULD be generated. */ if (idx == 8 || (a->s6_addr32[2] == 0 && a->s6_addr32[3] == 0)) return (-1); return (idx); } static void nptv6_copy_addr(struct in6_addr *src, struct in6_addr *dst, struct in6_addr *mask) { int i; for (i = 0; i < 8 && mask->s6_addr8[i] != 0; i++) { dst->s6_addr8[i] &= ~mask->s6_addr8[i]; dst->s6_addr8[i] |= src->s6_addr8[i] & mask->s6_addr8[i]; } } static int nptv6_rewrite_internal(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) { struct in6_addr *addr; struct ip6_hdr *ip6; int idx, proto; uint16_t adj; ip6 = mtodo(*mp, offset); NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), ip6->ip6_nxt); if (offset == 0) addr = &ip6->ip6_src; else { /* * When we rewriting inner IPv6 header, we need to rewrite * destination address back to external prefix. The datagram in * the ICMPv6 payload should looks like it was send from * external prefix. */ addr = &ip6->ip6_dst; } idx = nptv6_search_index(cfg, addr); if (idx < 0) { /* * Do not send ICMPv6 error when offset isn't zero. * This means we are rewriting inner IPv6 header in the * ICMPv6 error message. */ if (offset == 0) { icmp6_error2(*mp, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 0, (*mp)->m_pkthdr.rcvif); *mp = NULL; } return (IP_FW_DENY); } adj = addr->s6_addr16[idx]; nptv6_copy_addr(&cfg->external, addr, &cfg->mask); adj = cksum_add(adj, cfg->adjustment); if (adj == 0xffff) adj = 0; addr->s6_addr16[idx] = adj; if (offset == 0) { /* * We may need to translate addresses in the inner IPv6 * header for ICMPv6 error messages. */ proto = nptv6_getlasthdr(cfg, *mp, &offset); if (proto < 0 || (proto == IPPROTO_ICMPV6 && nptv6_translate_icmpv6(cfg, mp, offset) != 0)) return (IP_FW_DENY); NPTV6STAT_INC(cfg, in2ex); } return (0); } static int nptv6_rewrite_external(struct nptv6_cfg *cfg, struct mbuf **mp, int offset) { struct in6_addr *addr; struct ip6_hdr *ip6; int idx, proto; uint16_t adj; ip6 = mtodo(*mp, offset); NPTV6_IPDEBUG("offset %d, %s -> %s %d", offset, inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), ip6->ip6_nxt); if (offset == 0) addr = &ip6->ip6_dst; else { /* * When we rewriting inner IPv6 header, we need to rewrite * source address back to internal prefix. The datagram in * the ICMPv6 payload should looks like it was send from * internal prefix. */ addr = &ip6->ip6_src; } idx = nptv6_search_index(cfg, addr); if (idx < 0) { /* * Do not send ICMPv6 error when offset isn't zero. * This means we are rewriting inner IPv6 header in the * ICMPv6 error message. */ if (offset == 0) { icmp6_error2(*mp, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 0, (*mp)->m_pkthdr.rcvif); *mp = NULL; } return (IP_FW_DENY); } adj = addr->s6_addr16[idx]; nptv6_copy_addr(&cfg->internal, addr, &cfg->mask); adj = cksum_add(adj, ~cfg->adjustment); if (adj == 0xffff) adj = 0; addr->s6_addr16[idx] = adj; if (offset == 0) { /* * We may need to translate addresses in the inner IPv6 * header for ICMPv6 error messages. */ proto = nptv6_getlasthdr(cfg, *mp, &offset); if (proto < 0 || (proto == IPPROTO_ICMPV6 && nptv6_translate_icmpv6(cfg, mp, offset) != 0)) return (IP_FW_DENY); NPTV6STAT_INC(cfg, ex2in); } return (0); } /* * ipfw external action handler. */ static int ipfw_nptv6(struct ip_fw_chain *chain, struct ip_fw_args *args, ipfw_insn *cmd, int *done) { struct ip6_hdr *ip6; struct nptv6_cfg *cfg; ipfw_insn *icmd; int ret; *done = 0; /* try next rule if not matched */ ret = IP_FW_DENY; icmd = cmd + 1; if (cmd->opcode != O_EXTERNAL_ACTION || cmd->arg1 != V_nptv6_eid || icmd->opcode != O_EXTERNAL_INSTANCE || (cfg = NPTV6_LOOKUP(chain, icmd)) == NULL) return (ret); /* * We need act as router, so when forwarding is disabled - * do nothing. */ if (V_ip6_forwarding == 0 || args->f_id.addr_type != 6) return (ret); /* * NOTE: we expect ipfw_chk() did m_pullup() up to upper level * protocol's headers. Also we skip some checks, that ip6_input(), * ip6_forward(), ip6_fastfwd() and ipfw_chk() already did. */ ip6 = mtod(args->m, struct ip6_hdr *); NPTV6_IPDEBUG("eid %u, oid %u, %s -> %s %d", cmd->arg1, icmd->arg1, inet_ntop(AF_INET6, &ip6->ip6_src, _s, sizeof(_s)), inet_ntop(AF_INET6, &ip6->ip6_dst, _d, sizeof(_d)), ip6->ip6_nxt); if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_src, &cfg->internal, &cfg->mask)) { /* * XXX: Do not translate packets when both src and dst * are from internal prefix. */ if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, &cfg->internal, &cfg->mask)) return (ret); ret = nptv6_rewrite_internal(cfg, &args->m, 0); } else if (IN6_ARE_MASKED_ADDR_EQUAL(&ip6->ip6_dst, &cfg->external, &cfg->mask)) ret = nptv6_rewrite_external(cfg, &args->m, 0); else return (ret); /* * If address wasn't rewrited - free mbuf and terminate the search. */ if (ret != 0) { if (args->m != NULL) { m_freem(args->m); args->m = NULL; /* mark mbuf as consumed */ } NPTV6STAT_INC(cfg, dropped); *done = 1; } else { /* Terminate the search if one_pass is set */ *done = V_fw_one_pass; /* Update args->f_id when one_pass is off */ if (*done == 0) { ip6 = mtod(args->m, struct ip6_hdr *); args->f_id.src_ip6 = ip6->ip6_src; args->f_id.dst_ip6 = ip6->ip6_dst; } } return (ret); } static struct nptv6_cfg * nptv6_alloc_config(const char *name, uint8_t set) { struct nptv6_cfg *cfg; cfg = malloc(sizeof(struct nptv6_cfg), M_IPFW, M_WAITOK | M_ZERO); COUNTER_ARRAY_ALLOC(cfg->stats, NPTV6STATS, M_WAITOK); cfg->no.name = cfg->name; cfg->no.etlv = IPFW_TLV_NPTV6_NAME; cfg->no.set = set; strlcpy(cfg->name, name, sizeof(cfg->name)); return (cfg); } static void nptv6_free_config(struct nptv6_cfg *cfg) { COUNTER_ARRAY_FREE(cfg->stats, NPTV6STATS); free(cfg, M_IPFW); } static void nptv6_export_config(struct ip_fw_chain *ch, struct nptv6_cfg *cfg, ipfw_nptv6_cfg *uc) { uc->internal = cfg->internal; uc->external = cfg->external; uc->plen = cfg->plen; uc->flags = cfg->flags & NPTV6_FLAGSMASK; uc->set = cfg->no.set; strlcpy(uc->name, cfg->no.name, sizeof(uc->name)); } struct nptv6_dump_arg { struct ip_fw_chain *ch; struct sockopt_data *sd; }; static int export_config_cb(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct nptv6_dump_arg *da = (struct nptv6_dump_arg *)arg; ipfw_nptv6_cfg *uc; uc = (ipfw_nptv6_cfg *)ipfw_get_sopt_space(da->sd, sizeof(*uc)); nptv6_export_config(da->ch, (struct nptv6_cfg *)no, uc); return (0); } static struct nptv6_cfg * nptv6_find(struct namedobj_instance *ni, const char *name, uint8_t set) { struct nptv6_cfg *cfg; cfg = (struct nptv6_cfg *)ipfw_objhash_lookup_name_type(ni, set, IPFW_TLV_NPTV6_NAME, name); return (cfg); } static void nptv6_calculate_adjustment(struct nptv6_cfg *cfg) { uint16_t i, e; uint16_t *p; /* Calculate checksum of internal prefix */ for (i = 0, p = (uint16_t *)&cfg->internal; p < (uint16_t *)(&cfg->internal + 1); p++) i = cksum_add(i, *p); /* Calculate checksum of external prefix */ for (e = 0, p = (uint16_t *)&cfg->external; p < (uint16_t *)(&cfg->external + 1); p++) e = cksum_add(e, *p); /* Adjustment value for Int->Ext direction */ cfg->adjustment = cksum_add(~e, i); } /* * Creates new NPTv6 instance. * Data layout (v0)(current): * Request: [ ipfw_obj_lheader ipfw_nptv6_cfg ] * * Returns 0 on success */ static int nptv6_create(struct ip_fw_chain *ch, ip_fw3_opheader *op3, struct sockopt_data *sd) { struct in6_addr mask; ipfw_obj_lheader *olh; ipfw_nptv6_cfg *uc; struct namedobj_instance *ni; struct nptv6_cfg *cfg; if (sd->valsize != sizeof(*olh) + sizeof(*uc)) return (EINVAL); olh = (ipfw_obj_lheader *)sd->kbuf; uc = (ipfw_nptv6_cfg *)(olh + 1); if (ipfw_check_object_name_generic(uc->name) != 0) return (EINVAL); if (uc->plen < 8 || uc->plen > 64 || uc->set >= IPFW_MAX_SETS) return (EINVAL); if (IN6_IS_ADDR_MULTICAST(&uc->internal) || IN6_IS_ADDR_MULTICAST(&uc->external) || IN6_IS_ADDR_UNSPECIFIED(&uc->internal) || IN6_IS_ADDR_UNSPECIFIED(&uc->external) || IN6_IS_ADDR_LINKLOCAL(&uc->internal) || IN6_IS_ADDR_LINKLOCAL(&uc->external)) return (EINVAL); in6_prefixlen2mask(&mask, uc->plen); if (IN6_ARE_MASKED_ADDR_EQUAL(&uc->internal, &uc->external, &mask)) return (EINVAL); ni = CHAIN_TO_SRV(ch); IPFW_UH_RLOCK(ch); if (nptv6_find(ni, uc->name, uc->set) != NULL) { IPFW_UH_RUNLOCK(ch); return (EEXIST); } IPFW_UH_RUNLOCK(ch); cfg = nptv6_alloc_config(uc->name, uc->set); cfg->plen = uc->plen; if (cfg->plen <= 48) cfg->flags |= NPTV6_48PLEN; cfg->internal = uc->internal; cfg->external = uc->external; cfg->mask = mask; IN6_MASK_ADDR(&cfg->internal, &mask); IN6_MASK_ADDR(&cfg->external, &mask); nptv6_calculate_adjustment(cfg); IPFW_UH_WLOCK(ch); if (ipfw_objhash_alloc_idx(ni, &cfg->no.kidx) != 0) { IPFW_UH_WUNLOCK(ch); nptv6_free_config(cfg); return (ENOSPC); } ipfw_objhash_add(ni, &cfg->no); SRV_OBJECT(ch, cfg->no.kidx) = cfg; IPFW_UH_WUNLOCK(ch); return (0); } /* * Destroys NPTv6 instance. * Data layout (v0)(current): * Request: [ ipfw_obj_header ] * * Returns 0 on success */ static int nptv6_destroy(struct ip_fw_chain *ch, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_obj_header *oh; struct nptv6_cfg *cfg; if (sd->valsize != sizeof(*oh)) return (EINVAL); oh = (ipfw_obj_header *)sd->kbuf; if (ipfw_check_object_name_generic(oh->ntlv.name) != 0) return (EINVAL); IPFW_UH_WLOCK(ch); cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); if (cfg == NULL) { IPFW_UH_WUNLOCK(ch); return (ESRCH); } if (cfg->no.refcnt > 0) { IPFW_UH_WUNLOCK(ch); return (EBUSY); } SRV_OBJECT(ch, cfg->no.kidx) = NULL; ipfw_objhash_del(CHAIN_TO_SRV(ch), &cfg->no); ipfw_objhash_free_idx(CHAIN_TO_SRV(ch), cfg->no.kidx); IPFW_UH_WUNLOCK(ch); nptv6_free_config(cfg); return (0); } /* * Get or change nptv6 instance config. * Request: [ ipfw_obj_header [ ipfw_nptv6_cfg ] ] */ static int nptv6_config(struct ip_fw_chain *chain, ip_fw3_opheader *op, struct sockopt_data *sd) { return (EOPNOTSUPP); } /* * Lists all NPTv6 instances currently available in kernel. * Data layout (v0)(current): * Request: [ ipfw_obj_lheader ] * Reply: [ ipfw_obj_lheader ipfw_nptv6_cfg x N ] * * Returns 0 on success */ static int nptv6_list(struct ip_fw_chain *ch, ip_fw3_opheader *op3, struct sockopt_data *sd) { ipfw_obj_lheader *olh; struct nptv6_dump_arg da; /* Check minimum header size */ if (sd->valsize < sizeof(ipfw_obj_lheader)) return (EINVAL); olh = (ipfw_obj_lheader *)ipfw_get_sopt_header(sd, sizeof(*olh)); IPFW_UH_RLOCK(ch); olh->count = ipfw_objhash_count_type(CHAIN_TO_SRV(ch), IPFW_TLV_NPTV6_NAME); olh->objsize = sizeof(ipfw_nptv6_cfg); olh->size = sizeof(*olh) + olh->count * olh->objsize; if (sd->valsize < olh->size) { IPFW_UH_RUNLOCK(ch); return (ENOMEM); } memset(&da, 0, sizeof(da)); da.ch = ch; da.sd = sd; ipfw_objhash_foreach_type(CHAIN_TO_SRV(ch), export_config_cb, &da, IPFW_TLV_NPTV6_NAME); IPFW_UH_RUNLOCK(ch); return (0); } #define __COPY_STAT_FIELD(_cfg, _stats, _field) \ (_stats)->_field = NPTV6STAT_FETCH(_cfg, _field) static void export_stats(struct ip_fw_chain *ch, struct nptv6_cfg *cfg, struct ipfw_nptv6_stats *stats) { __COPY_STAT_FIELD(cfg, stats, in2ex); __COPY_STAT_FIELD(cfg, stats, ex2in); __COPY_STAT_FIELD(cfg, stats, dropped); } /* * Get NPTv6 statistics. * Data layout (v0)(current): * Request: [ ipfw_obj_header ] * Reply: [ ipfw_obj_header ipfw_obj_ctlv [ uint64_t x N ]] * * Returns 0 on success */ static int nptv6_stats(struct ip_fw_chain *ch, ip_fw3_opheader *op, struct sockopt_data *sd) { struct ipfw_nptv6_stats stats; struct nptv6_cfg *cfg; ipfw_obj_header *oh; ipfw_obj_ctlv *ctlv; size_t sz; sz = sizeof(ipfw_obj_header) + sizeof(ipfw_obj_ctlv) + sizeof(stats); if (sd->valsize % sizeof(uint64_t)) return (EINVAL); if (sd->valsize < sz) return (ENOMEM); oh = (ipfw_obj_header *)ipfw_get_sopt_header(sd, sz); if (oh == NULL) return (EINVAL); if (ipfw_check_object_name_generic(oh->ntlv.name) != 0 || oh->ntlv.set >= IPFW_MAX_SETS) return (EINVAL); memset(&stats, 0, sizeof(stats)); IPFW_UH_RLOCK(ch); cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); if (cfg == NULL) { IPFW_UH_RUNLOCK(ch); return (ESRCH); } export_stats(ch, cfg, &stats); IPFW_UH_RUNLOCK(ch); ctlv = (ipfw_obj_ctlv *)(oh + 1); memset(ctlv, 0, sizeof(*ctlv)); ctlv->head.type = IPFW_TLV_COUNTERS; ctlv->head.length = sz - sizeof(ipfw_obj_header); ctlv->count = sizeof(stats) / sizeof(uint64_t); ctlv->objsize = sizeof(uint64_t); ctlv->version = 1; memcpy(ctlv + 1, &stats, sizeof(stats)); return (0); } /* * Reset NPTv6 statistics. * Data layout (v0)(current): * Request: [ ipfw_obj_header ] * * Returns 0 on success */ static int nptv6_reset_stats(struct ip_fw_chain *ch, ip_fw3_opheader *op, struct sockopt_data *sd) { struct nptv6_cfg *cfg; ipfw_obj_header *oh; if (sd->valsize != sizeof(*oh)) return (EINVAL); oh = (ipfw_obj_header *)sd->kbuf; if (ipfw_check_object_name_generic(oh->ntlv.name) != 0 || oh->ntlv.set >= IPFW_MAX_SETS) return (EINVAL); IPFW_UH_WLOCK(ch); cfg = nptv6_find(CHAIN_TO_SRV(ch), oh->ntlv.name, oh->ntlv.set); if (cfg == NULL) { IPFW_UH_WUNLOCK(ch); return (ESRCH); } COUNTER_ARRAY_ZERO(cfg->stats, NPTV6STATS); IPFW_UH_WUNLOCK(ch); return (0); } static struct ipfw_sopt_handler scodes[] = { { IP_FW_NPTV6_CREATE, 0, HDIR_SET, nptv6_create }, { IP_FW_NPTV6_DESTROY,0, HDIR_SET, nptv6_destroy }, { IP_FW_NPTV6_CONFIG, 0, HDIR_BOTH, nptv6_config }, { IP_FW_NPTV6_LIST, 0, HDIR_GET, nptv6_list }, { IP_FW_NPTV6_STATS, 0, HDIR_GET, nptv6_stats }, { IP_FW_NPTV6_RESET_STATS,0, HDIR_SET, nptv6_reset_stats }, }; static int nptv6_classify(ipfw_insn *cmd, uint16_t *puidx, uint8_t *ptype) { ipfw_insn *icmd; icmd = cmd - 1; NPTV6_DEBUG("opcode %d, arg1 %d, opcode0 %d, arg1 %d", cmd->opcode, cmd->arg1, icmd->opcode, icmd->arg1); if (icmd->opcode != O_EXTERNAL_ACTION || icmd->arg1 != V_nptv6_eid) return (1); *puidx = cmd->arg1; *ptype = 0; return (0); } static void nptv6_update_arg1(ipfw_insn *cmd, uint16_t idx) { cmd->arg1 = idx; NPTV6_DEBUG("opcode %d, arg1 -> %d", cmd->opcode, cmd->arg1); } static int nptv6_findbyname(struct ip_fw_chain *ch, struct tid_info *ti, struct named_object **pno) { int err; err = ipfw_objhash_find_type(CHAIN_TO_SRV(ch), ti, IPFW_TLV_NPTV6_NAME, pno); NPTV6_DEBUG("uidx %u, type %u, err %d", ti->uidx, ti->type, err); return (err); } static struct named_object * nptv6_findbykidx(struct ip_fw_chain *ch, uint16_t idx) { struct namedobj_instance *ni; struct named_object *no; IPFW_UH_WLOCK_ASSERT(ch); ni = CHAIN_TO_SRV(ch); no = ipfw_objhash_lookup_kidx(ni, idx); KASSERT(no != NULL, ("NPT with index %d not found", idx)); NPTV6_DEBUG("kidx %u -> %s", idx, no->name); return (no); } static int nptv6_manage_sets(struct ip_fw_chain *ch, uint16_t set, uint8_t new_set, enum ipfw_sets_cmd cmd) { return (ipfw_obj_manage_sets(CHAIN_TO_SRV(ch), IPFW_TLV_NPTV6_NAME, set, new_set, cmd)); } static struct opcode_obj_rewrite opcodes[] = { { .opcode = O_EXTERNAL_INSTANCE, .etlv = IPFW_TLV_EACTION /* just show it isn't table */, .classifier = nptv6_classify, .update = nptv6_update_arg1, .find_byname = nptv6_findbyname, .find_bykidx = nptv6_findbykidx, .manage_sets = nptv6_manage_sets, }, }; static int destroy_config_cb(struct namedobj_instance *ni, struct named_object *no, void *arg) { struct nptv6_cfg *cfg; struct ip_fw_chain *ch; ch = (struct ip_fw_chain *)arg; IPFW_UH_WLOCK_ASSERT(ch); cfg = (struct nptv6_cfg *)SRV_OBJECT(ch, no->kidx); SRV_OBJECT(ch, no->kidx) = NULL; ipfw_objhash_del(ni, &cfg->no); ipfw_objhash_free_idx(ni, cfg->no.kidx); nptv6_free_config(cfg); return (0); } int nptv6_init(struct ip_fw_chain *ch, int first) { V_nptv6_eid = ipfw_add_eaction(ch, ipfw_nptv6, "nptv6"); if (V_nptv6_eid == 0) return (ENXIO); IPFW_ADD_SOPT_HANDLER(first, scodes); IPFW_ADD_OBJ_REWRITER(first, opcodes); return (0); } void nptv6_uninit(struct ip_fw_chain *ch, int last) { IPFW_DEL_OBJ_REWRITER(last, opcodes); IPFW_DEL_SOPT_HANDLER(last, scodes); ipfw_del_eaction(ch, V_nptv6_eid); /* * Since we already have deregistered external action, * our named objects become unaccessible via rules, because * all rules were truncated by ipfw_del_eaction(). * So, we can unlink and destroy our named objects without holding * IPFW_WLOCK(). */ IPFW_UH_WLOCK(ch); ipfw_objhash_foreach_type(CHAIN_TO_SRV(ch), destroy_config_cb, ch, IPFW_TLV_NPTV6_NAME); V_nptv6_eid = 0; IPFW_UH_WUNLOCK(ch); } Index: head/sys/netpfil/ipfw/pmod/tcpmod.c =================================================================== --- head/sys/netpfil/ipfw/pmod/tcpmod.c (revision 336675) +++ head/sys/netpfil/ipfw/pmod/tcpmod.c (revision 336676) @@ -1,246 +1,246 @@ /*- * Copyright (c) 2017 Yandex LLC * Copyright (c) 2017 Andrey V. Elsukov * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "opt_inet.h" #include "opt_inet6.h" #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include -static VNET_DEFINE(uint16_t, tcpmod_setmss_eid) = 0; +VNET_DEFINE_STATIC(uint16_t, tcpmod_setmss_eid) = 0; #define V_tcpmod_setmss_eid VNET(tcpmod_setmss_eid) static int tcpmod_setmss(struct mbuf **mp, struct tcphdr *tcp, int tlen, uint16_t mss) { struct mbuf *m; u_char *cp; int optlen, ret; uint16_t oldmss, csum; m = *mp; ret = IP_FW_DENY; if (m->m_len < m->m_pkthdr.len) { /* * We shouldn't have any data, IP packet contains only * TCP header with options. */ *mp = m = m_pullup(m, m->m_pkthdr.len); if (m == NULL) return (ret); } /* Parse TCP options. */ for (tlen -= sizeof(struct tcphdr), cp = (u_char *)(tcp + 1); tlen > 0; tlen -= optlen, cp += optlen) { if (cp[0] == TCPOPT_EOL) break; if (cp[0] == TCPOPT_NOP) { optlen = 1; continue; } if (tlen < 2) break; optlen = cp[1]; if (optlen < 2 || optlen > tlen) break; if (cp[0] == TCPOPT_MAXSEG) { if (optlen != TCPOLEN_MAXSEG) break; ret = 0; /* report success */ bcopy(cp + 2, &oldmss, sizeof(oldmss)); /* Do not update lower MSS value */ if (oldmss <= mss) break; bcopy(&mss, cp + 2, sizeof(mss)); /* Update checksum if it is not delayed. */ if ((m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_TCP_IPV6)) == 0) { bcopy(&tcp->th_sum, &csum, sizeof(csum)); csum = cksum_adjust(csum, oldmss, mss); bcopy(&csum, &tcp->th_sum, sizeof(csum)); } break; } } return (ret); } #ifdef INET6 static int tcpmod_ipv6_setmss(struct mbuf **mp, uint16_t mss) { struct ip6_hdr *ip6; struct ip6_hbh *hbh; struct tcphdr *tcp; int hlen, plen, proto; ip6 = mtod(*mp, struct ip6_hdr *); hlen = sizeof(*ip6); proto = ip6->ip6_nxt; /* * Skip IPv6 extension headers and get the TCP header. * ipfw_chk() has already done this work. So we are sure that * we will not do an access to the out of bounds. For this * reason we skip some checks here. */ while (proto == IPPROTO_HOPOPTS || proto == IPPROTO_ROUTING || proto == IPPROTO_DSTOPTS) { hbh = mtodo(*mp, hlen); proto = hbh->ip6h_nxt; hlen += (hbh->ip6h_len + 1) << 3; } tcp = mtodo(*mp, hlen); plen = (*mp)->m_pkthdr.len - hlen; hlen = tcp->th_off << 2; /* We must have TCP options and enough data in a packet. */ if (hlen <= sizeof(struct tcphdr) || hlen > plen) return (IP_FW_DENY); return (tcpmod_setmss(mp, tcp, hlen, mss)); } #endif /* INET6 */ #ifdef INET static int tcpmod_ipv4_setmss(struct mbuf **mp, uint16_t mss) { struct tcphdr *tcp; struct ip *ip; int hlen, plen; ip = mtod(*mp, struct ip *); hlen = ip->ip_hl << 2; tcp = mtodo(*mp, hlen); plen = (*mp)->m_pkthdr.len - hlen; hlen = tcp->th_off << 2; /* We must have TCP options and enough data in a packet. */ if (hlen <= sizeof(struct tcphdr) || hlen > plen) return (IP_FW_DENY); return (tcpmod_setmss(mp, tcp, hlen, mss)); } #endif /* INET */ /* * ipfw external action handler. */ static int ipfw_tcpmod(struct ip_fw_chain *chain, struct ip_fw_args *args, ipfw_insn *cmd, int *done) { ipfw_insn *icmd; int ret; *done = 0; /* try next rule if not matched */ ret = IP_FW_DENY; icmd = cmd + 1; if (cmd->opcode != O_EXTERNAL_ACTION || cmd->arg1 != V_tcpmod_setmss_eid || icmd->opcode != O_EXTERNAL_DATA || icmd->len != F_INSN_SIZE(ipfw_insn)) return (ret); /* * NOTE: ipfw_chk() can set f_id.proto from IPv6 fragment header, * but f_id._flags can be filled only from real TCP header. * * NOTE: ipfw_chk() drops very short packets in the PULLUP_TO() * macro. But we need to check that mbuf is contiguous more than * IP+IP_options/IP_extensions+tcphdr length, because TCP header * must have TCP options, and ipfw_chk() does PULLUP_TO() size of * struct tcphdr. * * NOTE: we require only the presence of SYN flag. User should * properly configure the rule to select the direction of packets, * that should be modified. */ if (args->f_id.proto != IPPROTO_TCP || (args->f_id._flags & TH_SYN) == 0) return (ret); switch (args->f_id.addr_type) { #ifdef INET case 4: ret = tcpmod_ipv4_setmss(&args->m, htons(icmd->arg1)); break; #endif #ifdef INET6 case 6: ret = tcpmod_ipv6_setmss(&args->m, htons(icmd->arg1)); break; #endif } /* * We return zero in both @ret and @done on success, and ipfw_chk() * will update rule counters. Otherwise a packet will not be matched * by rule. */ return (ret); } int tcpmod_init(struct ip_fw_chain *ch, int first) { V_tcpmod_setmss_eid = ipfw_add_eaction(ch, ipfw_tcpmod, "tcp-setmss"); if (V_tcpmod_setmss_eid == 0) return (ENXIO); return (0); } void tcpmod_uninit(struct ip_fw_chain *ch, int last) { ipfw_del_eaction(ch, V_tcpmod_setmss_eid); V_tcpmod_setmss_eid = 0; } Index: head/sys/netpfil/pf/if_pflog.c =================================================================== --- head/sys/netpfil/pf/if_pflog.c (revision 336675) +++ head/sys/netpfil/pf/if_pflog.c (revision 336676) @@ -1,320 +1,320 @@ /*- * SPDX-License-Identifier: ISC * * The authors of this code are John Ioannidis (ji@tla.org), * Angelos D. Keromytis (kermit@csd.uch.gr) and * Niels Provos (provos@physnet.uni-hamburg.de). * * This code was written by John Ioannidis for BSD/OS in Athens, Greece, * in November 1995. * * Ported to OpenBSD and NetBSD, with additional transforms, in December 1996, * by Angelos D. Keromytis. * * Additional transforms and features in 1997 and 1998 by Angelos D. Keromytis * and Niels Provos. * * Copyright (C) 1995, 1996, 1997, 1998 by John Ioannidis, Angelos D. Keromytis * and Niels Provos. * Copyright (c) 2001, Angelos D. Keromytis, Niels Provos. * * Permission to use, copy, and modify this software with or without fee * is hereby granted, provided that this entire notice is included in * all copies of any software which is or includes a copy or * modification of this software. * You may use this code under the GNU public license if you so wish. Please * contribute changes back to the authors under this freer than GPL license * so that we may further the use of strong encryption without limitations to * all. * * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR * IMPLIED WARRANTY. IN PARTICULAR, NONE OF THE AUTHORS MAKES ANY * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE * MERCHANTABILITY OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR * PURPOSE. * * $OpenBSD: if_pflog.c,v 1.26 2007/10/18 21:58:18 mpf Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_bpf.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #endif #ifdef INET #include #include #endif #ifdef INET6 #include #include #endif /* INET6 */ #ifdef INET #include #endif /* INET */ #define PFLOGMTU (32768 + MHLEN + MLEN) #ifdef PFLOGDEBUG #define DPRINTF(x) do { if (pflogdebug) printf x ; } while (0) #else #define DPRINTF(x) #endif static int pflogoutput(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static void pflogattach(int); static int pflogioctl(struct ifnet *, u_long, caddr_t); static void pflogstart(struct ifnet *); static int pflog_clone_create(struct if_clone *, int, caddr_t); static void pflog_clone_destroy(struct ifnet *); static const char pflogname[] = "pflog"; -static VNET_DEFINE(struct if_clone *, pflog_cloner); +VNET_DEFINE_STATIC(struct if_clone *, pflog_cloner); #define V_pflog_cloner VNET(pflog_cloner) VNET_DEFINE(struct ifnet *, pflogifs[PFLOGIFS_MAX]); /* for fast access */ #define V_pflogifs VNET(pflogifs) static void pflogattach(int npflog __unused) { int i; for (i = 0; i < PFLOGIFS_MAX; i++) V_pflogifs[i] = NULL; V_pflog_cloner = if_clone_simple(pflogname, pflog_clone_create, pflog_clone_destroy, 1); } static int pflog_clone_create(struct if_clone *ifc, int unit, caddr_t param) { struct ifnet *ifp; if (unit >= PFLOGIFS_MAX) return (EINVAL); ifp = if_alloc(IFT_PFLOG); if (ifp == NULL) { return (ENOSPC); } if_initname(ifp, pflogname, unit); ifp->if_mtu = PFLOGMTU; ifp->if_ioctl = pflogioctl; ifp->if_output = pflogoutput; ifp->if_start = pflogstart; ifp->if_snd.ifq_maxlen = ifqmaxlen; ifp->if_hdrlen = PFLOG_HDRLEN; if_attach(ifp); bpfattach(ifp, DLT_PFLOG, PFLOG_HDRLEN); V_pflogifs[unit] = ifp; return (0); } static void pflog_clone_destroy(struct ifnet *ifp) { int i; for (i = 0; i < PFLOGIFS_MAX; i++) if (V_pflogifs[i] == ifp) V_pflogifs[i] = NULL; bpfdetach(ifp); if_detach(ifp); if_free(ifp); } /* * Start output on the pflog interface. */ static void pflogstart(struct ifnet *ifp) { struct mbuf *m; for (;;) { IF_LOCK(&ifp->if_snd); _IF_DEQUEUE(&ifp->if_snd, m); IF_UNLOCK(&ifp->if_snd); if (m == NULL) return; else m_freem(m); } } static int pflogoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *rt) { m_freem(m); return (0); } /* ARGSUSED */ static int pflogioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { switch (cmd) { case SIOCSIFFLAGS: if (ifp->if_flags & IFF_UP) ifp->if_drv_flags |= IFF_DRV_RUNNING; else ifp->if_drv_flags &= ~IFF_DRV_RUNNING; break; default: return (ENOTTY); } return (0); } static int pflog_packet(struct pfi_kif *kif, struct mbuf *m, sa_family_t af, u_int8_t dir, u_int8_t reason, struct pf_rule *rm, struct pf_rule *am, struct pf_ruleset *ruleset, struct pf_pdesc *pd, int lookupsafe) { struct ifnet *ifn; struct pfloghdr hdr; if (kif == NULL || m == NULL || rm == NULL || pd == NULL) return ( 1); if ((ifn = V_pflogifs[rm->logif]) == NULL || !ifn->if_bpf) return (0); bzero(&hdr, sizeof(hdr)); hdr.length = PFLOG_REAL_HDRLEN; hdr.af = af; hdr.action = rm->action; hdr.reason = reason; memcpy(hdr.ifname, kif->pfik_name, sizeof(hdr.ifname)); if (am == NULL) { hdr.rulenr = htonl(rm->nr); hdr.subrulenr = -1; } else { hdr.rulenr = htonl(am->nr); hdr.subrulenr = htonl(rm->nr); if (ruleset != NULL && ruleset->anchor != NULL) strlcpy(hdr.ruleset, ruleset->anchor->name, sizeof(hdr.ruleset)); } /* * XXXGL: we avoid pf_socket_lookup() when we are holding * state lock, since this leads to unsafe LOR. * These conditions are very very rare, however. */ if (rm->log & PF_LOG_SOCKET_LOOKUP && !pd->lookup.done && lookupsafe) pd->lookup.done = pf_socket_lookup(dir, pd, m); if (pd->lookup.done > 0) hdr.uid = pd->lookup.uid; else hdr.uid = UID_MAX; hdr.pid = NO_PID; hdr.rule_uid = rm->cuid; hdr.rule_pid = rm->cpid; hdr.dir = dir; #ifdef INET if (af == AF_INET && dir == PF_OUT) { struct ip *ip; ip = mtod(m, struct ip *); ip->ip_sum = 0; ip->ip_sum = in_cksum(m, ip->ip_hl << 2); } #endif /* INET */ if_inc_counter(ifn, IFCOUNTER_OPACKETS, 1); if_inc_counter(ifn, IFCOUNTER_OBYTES, m->m_pkthdr.len); BPF_MTAP2(ifn, &hdr, PFLOG_HDRLEN, m); return (0); } static void vnet_pflog_init(const void *unused __unused) { pflogattach(1); } VNET_SYSINIT(vnet_pflog_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY, vnet_pflog_init, NULL); static void vnet_pflog_uninit(const void *unused __unused) { if_clone_detach(V_pflog_cloner); } /* * Detach after pf is gone; otherwise we might touch pflog memory * from within pf after freeing pflog. */ VNET_SYSUNINIT(vnet_pflog_uninit, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_pflog_uninit, NULL); static int pflog_modevent(module_t mod, int type, void *data) { int error = 0; switch (type) { case MOD_LOAD: PF_RULES_WLOCK(); pflog_packet_ptr = pflog_packet; PF_RULES_WUNLOCK(); break; case MOD_UNLOAD: PF_RULES_WLOCK(); pflog_packet_ptr = NULL; PF_RULES_WUNLOCK(); break; default: error = EOPNOTSUPP; break; } return error; } static moduledata_t pflog_mod = { pflogname, pflog_modevent, 0 }; #define PFLOG_MODVER 1 /* Do not run before pf is initialized as we depend on its locks. */ DECLARE_MODULE(pflog, pflog_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); MODULE_VERSION(pflog, PFLOG_MODVER); MODULE_DEPEND(pflog, pf, PF_MODVER, PF_MODVER, PF_MODVER); Index: head/sys/netpfil/pf/if_pfsync.c =================================================================== --- head/sys/netpfil/pf/if_pfsync.c (revision 336675) +++ head/sys/netpfil/pf/if_pfsync.c (revision 336676) @@ -1,2437 +1,2437 @@ /*- * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND ISC) * * Copyright (c) 2002 Michael Shalayeff * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ /*- * Copyright (c) 2009 David Gwynne * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ /* * $OpenBSD: if_pfsync.c,v 1.110 2009/02/24 05:39:19 dlg Exp $ * * Revisions picked from OpenBSD after revision 1.110 import: * 1.119 - don't m_copydata() beyond the len of mbuf in pfsync_input() * 1.118, 1.124, 1.148, 1.149, 1.151, 1.171 - fixes to bulk updates * 1.120, 1.175 - use monotonic time_uptime * 1.122 - reduce number of updates for non-TCP sessions * 1.125, 1.127 - rewrite merge or stale processing * 1.128 - cleanups * 1.146 - bzero() mbuf before sparsely filling it with data * 1.170 - SIOCSIFMTU checks * 1.126, 1.142 - deferred packets processing * 1.173 - correct expire time processing */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define PFSYNC_MINPKT ( \ sizeof(struct ip) + \ sizeof(struct pfsync_header) + \ sizeof(struct pfsync_subheader) ) struct pfsync_pkt { struct ip *ip; struct in_addr src; u_int8_t flags; }; static int pfsync_upd_tcp(struct pf_state *, struct pfsync_state_peer *, struct pfsync_state_peer *); static int pfsync_in_clr(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_ins(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_iack(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_upd(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_upd_c(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_ureq(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_del(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_del_c(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_bus(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_tdb(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_eof(struct pfsync_pkt *, struct mbuf *, int, int); static int pfsync_in_error(struct pfsync_pkt *, struct mbuf *, int, int); static int (*pfsync_acts[])(struct pfsync_pkt *, struct mbuf *, int, int) = { pfsync_in_clr, /* PFSYNC_ACT_CLR */ pfsync_in_ins, /* PFSYNC_ACT_INS */ pfsync_in_iack, /* PFSYNC_ACT_INS_ACK */ pfsync_in_upd, /* PFSYNC_ACT_UPD */ pfsync_in_upd_c, /* PFSYNC_ACT_UPD_C */ pfsync_in_ureq, /* PFSYNC_ACT_UPD_REQ */ pfsync_in_del, /* PFSYNC_ACT_DEL */ pfsync_in_del_c, /* PFSYNC_ACT_DEL_C */ pfsync_in_error, /* PFSYNC_ACT_INS_F */ pfsync_in_error, /* PFSYNC_ACT_DEL_F */ pfsync_in_bus, /* PFSYNC_ACT_BUS */ pfsync_in_tdb, /* PFSYNC_ACT_TDB */ pfsync_in_eof /* PFSYNC_ACT_EOF */ }; struct pfsync_q { void (*write)(struct pf_state *, void *); size_t len; u_int8_t action; }; /* we have one of these for every PFSYNC_S_ */ static void pfsync_out_state(struct pf_state *, void *); static void pfsync_out_iack(struct pf_state *, void *); static void pfsync_out_upd_c(struct pf_state *, void *); static void pfsync_out_del(struct pf_state *, void *); static struct pfsync_q pfsync_qs[] = { { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_INS }, { pfsync_out_iack, sizeof(struct pfsync_ins_ack), PFSYNC_ACT_INS_ACK }, { pfsync_out_state, sizeof(struct pfsync_state), PFSYNC_ACT_UPD }, { pfsync_out_upd_c, sizeof(struct pfsync_upd_c), PFSYNC_ACT_UPD_C }, { pfsync_out_del, sizeof(struct pfsync_del_c), PFSYNC_ACT_DEL_C } }; static void pfsync_q_ins(struct pf_state *, int, bool); static void pfsync_q_del(struct pf_state *, bool); static void pfsync_update_state(struct pf_state *); struct pfsync_upd_req_item { TAILQ_ENTRY(pfsync_upd_req_item) ur_entry; struct pfsync_upd_req ur_msg; }; struct pfsync_deferral { struct pfsync_softc *pd_sc; TAILQ_ENTRY(pfsync_deferral) pd_entry; u_int pd_refs; struct callout pd_tmo; struct pf_state *pd_st; struct mbuf *pd_m; }; struct pfsync_softc { /* Configuration */ struct ifnet *sc_ifp; struct ifnet *sc_sync_if; struct ip_moptions sc_imo; struct in_addr sc_sync_peer; uint32_t sc_flags; #define PFSYNCF_OK 0x00000001 #define PFSYNCF_DEFER 0x00000002 #define PFSYNCF_PUSH 0x00000004 uint8_t sc_maxupdates; struct ip sc_template; struct callout sc_tmo; struct mtx sc_mtx; /* Queued data */ size_t sc_len; TAILQ_HEAD(, pf_state) sc_qs[PFSYNC_S_COUNT]; TAILQ_HEAD(, pfsync_upd_req_item) sc_upd_req_list; TAILQ_HEAD(, pfsync_deferral) sc_deferrals; u_int sc_deferred; void *sc_plus; size_t sc_pluslen; /* Bulk update info */ struct mtx sc_bulk_mtx; uint32_t sc_ureq_sent; int sc_bulk_tries; uint32_t sc_ureq_received; int sc_bulk_hashid; uint64_t sc_bulk_stateid; uint32_t sc_bulk_creatorid; struct callout sc_bulk_tmo; struct callout sc_bulkfail_tmo; }; #define PFSYNC_LOCK(sc) mtx_lock(&(sc)->sc_mtx) #define PFSYNC_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) #define PFSYNC_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) #define PFSYNC_BLOCK(sc) mtx_lock(&(sc)->sc_bulk_mtx) #define PFSYNC_BUNLOCK(sc) mtx_unlock(&(sc)->sc_bulk_mtx) #define PFSYNC_BLOCK_ASSERT(sc) mtx_assert(&(sc)->sc_bulk_mtx, MA_OWNED) static const char pfsyncname[] = "pfsync"; static MALLOC_DEFINE(M_PFSYNC, pfsyncname, "pfsync(4) data"); -static VNET_DEFINE(struct pfsync_softc *, pfsyncif) = NULL; +VNET_DEFINE_STATIC(struct pfsync_softc *, pfsyncif) = NULL; #define V_pfsyncif VNET(pfsyncif) -static VNET_DEFINE(void *, pfsync_swi_cookie) = NULL; +VNET_DEFINE_STATIC(void *, pfsync_swi_cookie) = NULL; #define V_pfsync_swi_cookie VNET(pfsync_swi_cookie) -static VNET_DEFINE(struct pfsyncstats, pfsyncstats); +VNET_DEFINE_STATIC(struct pfsyncstats, pfsyncstats); #define V_pfsyncstats VNET(pfsyncstats) -static VNET_DEFINE(int, pfsync_carp_adj) = CARP_MAXSKEW; +VNET_DEFINE_STATIC(int, pfsync_carp_adj) = CARP_MAXSKEW; #define V_pfsync_carp_adj VNET(pfsync_carp_adj) static void pfsync_timeout(void *); static void pfsync_push(struct pfsync_softc *); static void pfsyncintr(void *); static int pfsync_multicast_setup(struct pfsync_softc *, struct ifnet *, void *); static void pfsync_multicast_cleanup(struct pfsync_softc *); static void pfsync_pointers_init(void); static void pfsync_pointers_uninit(void); static int pfsync_init(void); static void pfsync_uninit(void); SYSCTL_NODE(_net, OID_AUTO, pfsync, CTLFLAG_RW, 0, "PFSYNC"); SYSCTL_STRUCT(_net_pfsync, OID_AUTO, stats, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(pfsyncstats), pfsyncstats, "PFSYNC statistics (struct pfsyncstats, net/if_pfsync.h)"); SYSCTL_INT(_net_pfsync, OID_AUTO, carp_demotion_factor, CTLFLAG_RW, &VNET_NAME(pfsync_carp_adj), 0, "pfsync's CARP demotion factor adjustment"); static int pfsync_clone_create(struct if_clone *, int, caddr_t); static void pfsync_clone_destroy(struct ifnet *); static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *, struct pf_state_peer *); static int pfsyncoutput(struct ifnet *, struct mbuf *, const struct sockaddr *, struct route *); static int pfsyncioctl(struct ifnet *, u_long, caddr_t); static int pfsync_defer(struct pf_state *, struct mbuf *); static void pfsync_undefer(struct pfsync_deferral *, int); static void pfsync_undefer_state(struct pf_state *, int); static void pfsync_defer_tmo(void *); static void pfsync_request_update(u_int32_t, u_int64_t); static void pfsync_update_state_req(struct pf_state *); static void pfsync_drop(struct pfsync_softc *); static void pfsync_sendout(int); static void pfsync_send_plus(void *, size_t); static void pfsync_bulk_start(void); static void pfsync_bulk_status(u_int8_t); static void pfsync_bulk_update(void *); static void pfsync_bulk_fail(void *); #ifdef IPSEC static void pfsync_update_net_tdb(struct pfsync_tdb *); #endif #define PFSYNC_MAX_BULKTRIES 12 VNET_DEFINE(struct if_clone *, pfsync_cloner); #define V_pfsync_cloner VNET(pfsync_cloner) static int pfsync_clone_create(struct if_clone *ifc, int unit, caddr_t param) { struct pfsync_softc *sc; struct ifnet *ifp; int q; if (unit != 0) return (EINVAL); sc = malloc(sizeof(struct pfsync_softc), M_PFSYNC, M_WAITOK | M_ZERO); sc->sc_flags |= PFSYNCF_OK; for (q = 0; q < PFSYNC_S_COUNT; q++) TAILQ_INIT(&sc->sc_qs[q]); TAILQ_INIT(&sc->sc_upd_req_list); TAILQ_INIT(&sc->sc_deferrals); sc->sc_len = PFSYNC_MINPKT; sc->sc_maxupdates = 128; ifp = sc->sc_ifp = if_alloc(IFT_PFSYNC); if (ifp == NULL) { free(sc, M_PFSYNC); return (ENOSPC); } if_initname(ifp, pfsyncname, unit); ifp->if_softc = sc; ifp->if_ioctl = pfsyncioctl; ifp->if_output = pfsyncoutput; ifp->if_type = IFT_PFSYNC; ifp->if_snd.ifq_maxlen = ifqmaxlen; ifp->if_hdrlen = sizeof(struct pfsync_header); ifp->if_mtu = ETHERMTU; mtx_init(&sc->sc_mtx, pfsyncname, NULL, MTX_DEF); mtx_init(&sc->sc_bulk_mtx, "pfsync bulk", NULL, MTX_DEF); callout_init(&sc->sc_tmo, 1); callout_init_mtx(&sc->sc_bulk_tmo, &sc->sc_bulk_mtx, 0); callout_init_mtx(&sc->sc_bulkfail_tmo, &sc->sc_bulk_mtx, 0); if_attach(ifp); bpfattach(ifp, DLT_PFSYNC, PFSYNC_HDRLEN); V_pfsyncif = sc; return (0); } static void pfsync_clone_destroy(struct ifnet *ifp) { struct pfsync_softc *sc = ifp->if_softc; /* * At this stage, everything should have already been * cleared by pfsync_uninit(), and we have only to * drain callouts. */ while (sc->sc_deferred > 0) { struct pfsync_deferral *pd = TAILQ_FIRST(&sc->sc_deferrals); TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); sc->sc_deferred--; if (callout_stop(&pd->pd_tmo) > 0) { pf_release_state(pd->pd_st); m_freem(pd->pd_m); free(pd, M_PFSYNC); } else { pd->pd_refs++; callout_drain(&pd->pd_tmo); free(pd, M_PFSYNC); } } callout_drain(&sc->sc_tmo); callout_drain(&sc->sc_bulkfail_tmo); callout_drain(&sc->sc_bulk_tmo); if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync destroy"); bpfdetach(ifp); if_detach(ifp); pfsync_drop(sc); if_free(ifp); if (sc->sc_imo.imo_membership) pfsync_multicast_cleanup(sc); mtx_destroy(&sc->sc_mtx); mtx_destroy(&sc->sc_bulk_mtx); free(sc, M_PFSYNC); V_pfsyncif = NULL; } static int pfsync_alloc_scrub_memory(struct pfsync_state_peer *s, struct pf_state_peer *d) { if (s->scrub.scrub_flag && d->scrub == NULL) { d->scrub = uma_zalloc(V_pf_state_scrub_z, M_NOWAIT | M_ZERO); if (d->scrub == NULL) return (ENOMEM); } return (0); } static int pfsync_state_import(struct pfsync_state *sp, u_int8_t flags) { struct pfsync_softc *sc = V_pfsyncif; #ifndef __NO_STRICT_ALIGNMENT struct pfsync_state_key key[2]; #endif struct pfsync_state_key *kw, *ks; struct pf_state *st = NULL; struct pf_state_key *skw = NULL, *sks = NULL; struct pf_rule *r = NULL; struct pfi_kif *kif; int error; PF_RULES_RASSERT(); if (sp->creatorid == 0) { if (V_pf_status.debug >= PF_DEBUG_MISC) printf("%s: invalid creator id: %08x\n", __func__, ntohl(sp->creatorid)); return (EINVAL); } if ((kif = pfi_kif_find(sp->ifname)) == NULL) { if (V_pf_status.debug >= PF_DEBUG_MISC) printf("%s: unknown interface: %s\n", __func__, sp->ifname); if (flags & PFSYNC_SI_IOCTL) return (EINVAL); return (0); /* skip this state */ } /* * If the ruleset checksums match or the state is coming from the ioctl, * it's safe to associate the state with the rule of that number. */ if (sp->rule != htonl(-1) && sp->anchor == htonl(-1) && (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->rule) < pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) r = pf_main_ruleset.rules[ PF_RULESET_FILTER].active.ptr_array[ntohl(sp->rule)]; else r = &V_pf_default_rule; if ((r->max_states && counter_u64_fetch(r->states_cur) >= r->max_states)) goto cleanup; /* * XXXGL: consider M_WAITOK in ioctl path after. */ if ((st = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO)) == NULL) goto cleanup; if ((skw = uma_zalloc(V_pf_state_key_z, M_NOWAIT)) == NULL) goto cleanup; #ifndef __NO_STRICT_ALIGNMENT bcopy(&sp->key, key, sizeof(struct pfsync_state_key) * 2); kw = &key[PF_SK_WIRE]; ks = &key[PF_SK_STACK]; #else kw = &sp->key[PF_SK_WIRE]; ks = &sp->key[PF_SK_STACK]; #endif if (PF_ANEQ(&kw->addr[0], &ks->addr[0], sp->af) || PF_ANEQ(&kw->addr[1], &ks->addr[1], sp->af) || kw->port[0] != ks->port[0] || kw->port[1] != ks->port[1]) { sks = uma_zalloc(V_pf_state_key_z, M_NOWAIT); if (sks == NULL) goto cleanup; } else sks = skw; /* allocate memory for scrub info */ if (pfsync_alloc_scrub_memory(&sp->src, &st->src) || pfsync_alloc_scrub_memory(&sp->dst, &st->dst)) goto cleanup; /* Copy to state key(s). */ skw->addr[0] = kw->addr[0]; skw->addr[1] = kw->addr[1]; skw->port[0] = kw->port[0]; skw->port[1] = kw->port[1]; skw->proto = sp->proto; skw->af = sp->af; if (sks != skw) { sks->addr[0] = ks->addr[0]; sks->addr[1] = ks->addr[1]; sks->port[0] = ks->port[0]; sks->port[1] = ks->port[1]; sks->proto = sp->proto; sks->af = sp->af; } /* copy to state */ bcopy(&sp->rt_addr, &st->rt_addr, sizeof(st->rt_addr)); st->creation = time_uptime - ntohl(sp->creation); st->expire = time_uptime; if (sp->expire) { uint32_t timeout; timeout = r->timeout[sp->timeout]; if (!timeout) timeout = V_pf_default_rule.timeout[sp->timeout]; /* sp->expire may have been adaptively scaled by export. */ st->expire -= timeout - ntohl(sp->expire); } st->direction = sp->direction; st->log = sp->log; st->timeout = sp->timeout; st->state_flags = sp->state_flags; st->id = sp->id; st->creatorid = sp->creatorid; pf_state_peer_ntoh(&sp->src, &st->src); pf_state_peer_ntoh(&sp->dst, &st->dst); st->rule.ptr = r; st->nat_rule.ptr = NULL; st->anchor.ptr = NULL; st->rt_kif = NULL; st->pfsync_time = time_uptime; st->sync_state = PFSYNC_S_NONE; if (!(flags & PFSYNC_SI_IOCTL)) st->state_flags |= PFSTATE_NOSYNC; if ((error = pf_state_insert(kif, skw, sks, st)) != 0) goto cleanup_state; /* XXX when we have nat_rule/anchors, use STATE_INC_COUNTERS */ counter_u64_add(r->states_cur, 1); counter_u64_add(r->states_tot, 1); if (!(flags & PFSYNC_SI_IOCTL)) { st->state_flags &= ~PFSTATE_NOSYNC; if (st->state_flags & PFSTATE_ACK) { pfsync_q_ins(st, PFSYNC_S_IACK, true); pfsync_push(sc); } } st->state_flags &= ~PFSTATE_ACK; PF_STATE_UNLOCK(st); return (0); cleanup: error = ENOMEM; if (skw == sks) sks = NULL; if (skw != NULL) uma_zfree(V_pf_state_key_z, skw); if (sks != NULL) uma_zfree(V_pf_state_key_z, sks); cleanup_state: /* pf_state_insert() frees the state keys. */ if (st) { if (st->dst.scrub) uma_zfree(V_pf_state_scrub_z, st->dst.scrub); if (st->src.scrub) uma_zfree(V_pf_state_scrub_z, st->src.scrub); uma_zfree(V_pf_state_z, st); } return (error); } static int pfsync_input(struct mbuf **mp, int *offp __unused, int proto __unused) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_pkt pkt; struct mbuf *m = *mp; struct ip *ip = mtod(m, struct ip *); struct pfsync_header *ph; struct pfsync_subheader subh; int offset, len; int rv; uint16_t count; PF_RULES_RLOCK_TRACKER; *mp = NULL; V_pfsyncstats.pfsyncs_ipackets++; /* Verify that we have a sync interface configured. */ if (!sc || !sc->sc_sync_if || !V_pf_status.running || (sc->sc_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) goto done; /* verify that the packet came in on the right interface */ if (sc->sc_sync_if != m->m_pkthdr.rcvif) { V_pfsyncstats.pfsyncs_badif++; goto done; } if_inc_counter(sc->sc_ifp, IFCOUNTER_IPACKETS, 1); if_inc_counter(sc->sc_ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); /* verify that the IP TTL is 255. */ if (ip->ip_ttl != PFSYNC_DFLTTL) { V_pfsyncstats.pfsyncs_badttl++; goto done; } offset = ip->ip_hl << 2; if (m->m_pkthdr.len < offset + sizeof(*ph)) { V_pfsyncstats.pfsyncs_hdrops++; goto done; } if (offset + sizeof(*ph) > m->m_len) { if (m_pullup(m, offset + sizeof(*ph)) == NULL) { V_pfsyncstats.pfsyncs_hdrops++; return (IPPROTO_DONE); } ip = mtod(m, struct ip *); } ph = (struct pfsync_header *)((char *)ip + offset); /* verify the version */ if (ph->version != PFSYNC_VERSION) { V_pfsyncstats.pfsyncs_badver++; goto done; } len = ntohs(ph->len) + offset; if (m->m_pkthdr.len < len) { V_pfsyncstats.pfsyncs_badlen++; goto done; } /* Cheaper to grab this now than having to mess with mbufs later */ pkt.ip = ip; pkt.src = ip->ip_src; pkt.flags = 0; /* * Trusting pf_chksum during packet processing, as well as seeking * in interface name tree, require holding PF_RULES_RLOCK(). */ PF_RULES_RLOCK(); if (!bcmp(&ph->pfcksum, &V_pf_status.pf_chksum, PF_MD5_DIGEST_LENGTH)) pkt.flags |= PFSYNC_SI_CKSUM; offset += sizeof(*ph); while (offset <= len - sizeof(subh)) { m_copydata(m, offset, sizeof(subh), (caddr_t)&subh); offset += sizeof(subh); if (subh.action >= PFSYNC_ACT_MAX) { V_pfsyncstats.pfsyncs_badact++; PF_RULES_RUNLOCK(); goto done; } count = ntohs(subh.count); V_pfsyncstats.pfsyncs_iacts[subh.action] += count; rv = (*pfsync_acts[subh.action])(&pkt, m, offset, count); if (rv == -1) { PF_RULES_RUNLOCK(); return (IPPROTO_DONE); } offset += rv; } PF_RULES_RUNLOCK(); done: m_freem(m); return (IPPROTO_DONE); } static int pfsync_in_clr(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_clr *clr; struct mbuf *mp; int len = sizeof(*clr) * count; int i, offp; u_int32_t creatorid; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } clr = (struct pfsync_clr *)(mp->m_data + offp); for (i = 0; i < count; i++) { creatorid = clr[i].creatorid; if (clr[i].ifname[0] != '\0' && pfi_kif_find(clr[i].ifname) == NULL) continue; for (int i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state *s; relock: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (s->creatorid == creatorid) { s->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(s, PF_ENTER_LOCKED); goto relock; } } PF_HASHROW_UNLOCK(ih); } } return (len); } static int pfsync_in_ins(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct mbuf *mp; struct pfsync_state *sa, *sp; int len = sizeof(*sp) * count; int i, offp; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } sa = (struct pfsync_state *)(mp->m_data + offp); for (i = 0; i < count; i++) { sp = &sa[i]; /* Check for invalid values. */ if (sp->timeout >= PFTM_MAX || sp->src.state > PF_TCPS_PROXY_DST || sp->dst.state > PF_TCPS_PROXY_DST || sp->direction > PF_OUT || (sp->af != AF_INET && sp->af != AF_INET6)) { if (V_pf_status.debug >= PF_DEBUG_MISC) printf("%s: invalid value\n", __func__); V_pfsyncstats.pfsyncs_badval++; continue; } if (pfsync_state_import(sp, pkt->flags) == ENOMEM) /* Drop out, but process the rest of the actions. */ break; } return (len); } static int pfsync_in_iack(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_ins_ack *ia, *iaa; struct pf_state *st; struct mbuf *mp; int len = count * sizeof(*ia); int offp, i; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } iaa = (struct pfsync_ins_ack *)(mp->m_data + offp); for (i = 0; i < count; i++) { ia = &iaa[i]; st = pf_find_state_byid(ia->id, ia->creatorid); if (st == NULL) continue; if (st->state_flags & PFSTATE_ACK) { PFSYNC_LOCK(V_pfsyncif); pfsync_undefer_state(st, 0); PFSYNC_UNLOCK(V_pfsyncif); } PF_STATE_UNLOCK(st); } /* * XXX this is not yet implemented, but we know the size of the * message so we can skip it. */ return (count * sizeof(struct pfsync_ins_ack)); } static int pfsync_upd_tcp(struct pf_state *st, struct pfsync_state_peer *src, struct pfsync_state_peer *dst) { int sync = 0; PF_STATE_LOCK_ASSERT(st); /* * The state should never go backwards except * for syn-proxy states. Neither should the * sequence window slide backwards. */ if ((st->src.state > src->state && (st->src.state < PF_TCPS_PROXY_SRC || src->state >= PF_TCPS_PROXY_SRC)) || (st->src.state == src->state && SEQ_GT(st->src.seqlo, ntohl(src->seqlo)))) sync++; else pf_state_peer_ntoh(src, &st->src); if ((st->dst.state > dst->state) || (st->dst.state >= TCPS_SYN_SENT && SEQ_GT(st->dst.seqlo, ntohl(dst->seqlo)))) sync++; else pf_state_peer_ntoh(dst, &st->dst); return (sync); } static int pfsync_in_upd(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_state *sa, *sp; struct pf_state *st; int sync; struct mbuf *mp; int len = count * sizeof(*sp); int offp, i; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } sa = (struct pfsync_state *)(mp->m_data + offp); for (i = 0; i < count; i++) { sp = &sa[i]; /* check for invalid values */ if (sp->timeout >= PFTM_MAX || sp->src.state > PF_TCPS_PROXY_DST || sp->dst.state > PF_TCPS_PROXY_DST) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pfsync_input: PFSYNC_ACT_UPD: " "invalid value\n"); } V_pfsyncstats.pfsyncs_badval++; continue; } st = pf_find_state_byid(sp->id, sp->creatorid); if (st == NULL) { /* insert the update */ if (pfsync_state_import(sp, pkt->flags)) V_pfsyncstats.pfsyncs_badstate++; continue; } if (st->state_flags & PFSTATE_ACK) { PFSYNC_LOCK(sc); pfsync_undefer_state(st, 1); PFSYNC_UNLOCK(sc); } if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) sync = pfsync_upd_tcp(st, &sp->src, &sp->dst); else { sync = 0; /* * Non-TCP protocol state machine always go * forwards */ if (st->src.state > sp->src.state) sync++; else pf_state_peer_ntoh(&sp->src, &st->src); if (st->dst.state > sp->dst.state) sync++; else pf_state_peer_ntoh(&sp->dst, &st->dst); } if (sync < 2) { pfsync_alloc_scrub_memory(&sp->dst, &st->dst); pf_state_peer_ntoh(&sp->dst, &st->dst); st->expire = time_uptime; st->timeout = sp->timeout; } st->pfsync_time = time_uptime; if (sync) { V_pfsyncstats.pfsyncs_stale++; pfsync_update_state(st); PF_STATE_UNLOCK(st); PFSYNC_LOCK(sc); pfsync_push(sc); PFSYNC_UNLOCK(sc); continue; } PF_STATE_UNLOCK(st); } return (len); } static int pfsync_in_upd_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_upd_c *ua, *up; struct pf_state *st; int len = count * sizeof(*up); int sync; struct mbuf *mp; int offp, i; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } ua = (struct pfsync_upd_c *)(mp->m_data + offp); for (i = 0; i < count; i++) { up = &ua[i]; /* check for invalid values */ if (up->timeout >= PFTM_MAX || up->src.state > PF_TCPS_PROXY_DST || up->dst.state > PF_TCPS_PROXY_DST) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pfsync_input: " "PFSYNC_ACT_UPD_C: " "invalid value\n"); } V_pfsyncstats.pfsyncs_badval++; continue; } st = pf_find_state_byid(up->id, up->creatorid); if (st == NULL) { /* We don't have this state. Ask for it. */ PFSYNC_LOCK(sc); pfsync_request_update(up->creatorid, up->id); PFSYNC_UNLOCK(sc); continue; } if (st->state_flags & PFSTATE_ACK) { PFSYNC_LOCK(sc); pfsync_undefer_state(st, 1); PFSYNC_UNLOCK(sc); } if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) sync = pfsync_upd_tcp(st, &up->src, &up->dst); else { sync = 0; /* * Non-TCP protocol state machine always go * forwards */ if (st->src.state > up->src.state) sync++; else pf_state_peer_ntoh(&up->src, &st->src); if (st->dst.state > up->dst.state) sync++; else pf_state_peer_ntoh(&up->dst, &st->dst); } if (sync < 2) { pfsync_alloc_scrub_memory(&up->dst, &st->dst); pf_state_peer_ntoh(&up->dst, &st->dst); st->expire = time_uptime; st->timeout = up->timeout; } st->pfsync_time = time_uptime; if (sync) { V_pfsyncstats.pfsyncs_stale++; pfsync_update_state(st); PF_STATE_UNLOCK(st); PFSYNC_LOCK(sc); pfsync_push(sc); PFSYNC_UNLOCK(sc); continue; } PF_STATE_UNLOCK(st); } return (len); } static int pfsync_in_ureq(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_upd_req *ur, *ura; struct mbuf *mp; int len = count * sizeof(*ur); int i, offp; struct pf_state *st; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } ura = (struct pfsync_upd_req *)(mp->m_data + offp); for (i = 0; i < count; i++) { ur = &ura[i]; if (ur->id == 0 && ur->creatorid == 0) pfsync_bulk_start(); else { st = pf_find_state_byid(ur->id, ur->creatorid); if (st == NULL) { V_pfsyncstats.pfsyncs_badstate++; continue; } if (st->state_flags & PFSTATE_NOSYNC) { PF_STATE_UNLOCK(st); continue; } pfsync_update_state_req(st); PF_STATE_UNLOCK(st); } } return (len); } static int pfsync_in_del(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct mbuf *mp; struct pfsync_state *sa, *sp; struct pf_state *st; int len = count * sizeof(*sp); int offp, i; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } sa = (struct pfsync_state *)(mp->m_data + offp); for (i = 0; i < count; i++) { sp = &sa[i]; st = pf_find_state_byid(sp->id, sp->creatorid); if (st == NULL) { V_pfsyncstats.pfsyncs_badstate++; continue; } st->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(st, PF_ENTER_LOCKED); } return (len); } static int pfsync_in_del_c(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct mbuf *mp; struct pfsync_del_c *sa, *sp; struct pf_state *st; int len = count * sizeof(*sp); int offp, i; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } sa = (struct pfsync_del_c *)(mp->m_data + offp); for (i = 0; i < count; i++) { sp = &sa[i]; st = pf_find_state_byid(sp->id, sp->creatorid); if (st == NULL) { V_pfsyncstats.pfsyncs_badstate++; continue; } st->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(st, PF_ENTER_LOCKED); } return (len); } static int pfsync_in_bus(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_bus *bus; struct mbuf *mp; int len = count * sizeof(*bus); int offp; PFSYNC_BLOCK(sc); /* If we're not waiting for a bulk update, who cares. */ if (sc->sc_ureq_sent == 0) { PFSYNC_BUNLOCK(sc); return (len); } mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { PFSYNC_BUNLOCK(sc); V_pfsyncstats.pfsyncs_badlen++; return (-1); } bus = (struct pfsync_bus *)(mp->m_data + offp); switch (bus->status) { case PFSYNC_BUS_START: callout_reset(&sc->sc_bulkfail_tmo, 4 * hz + V_pf_limits[PF_LIMIT_STATES].limit / ((sc->sc_ifp->if_mtu - PFSYNC_MINPKT) / sizeof(struct pfsync_state)), pfsync_bulk_fail, sc); if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: received bulk update start\n"); break; case PFSYNC_BUS_END: if (time_uptime - ntohl(bus->endtime) >= sc->sc_ureq_sent) { /* that's it, we're happy */ sc->sc_ureq_sent = 0; sc->sc_bulk_tries = 0; callout_stop(&sc->sc_bulkfail_tmo); if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync bulk done"); sc->sc_flags |= PFSYNCF_OK; if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: received valid " "bulk update end\n"); } else { if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: received invalid " "bulk update end: bad timestamp\n"); } break; } PFSYNC_BUNLOCK(sc); return (len); } static int pfsync_in_tdb(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { int len = count * sizeof(struct pfsync_tdb); #if defined(IPSEC) struct pfsync_tdb *tp; struct mbuf *mp; int offp; int i; int s; mp = m_pulldown(m, offset, len, &offp); if (mp == NULL) { V_pfsyncstats.pfsyncs_badlen++; return (-1); } tp = (struct pfsync_tdb *)(mp->m_data + offp); for (i = 0; i < count; i++) pfsync_update_net_tdb(&tp[i]); #endif return (len); } #if defined(IPSEC) /* Update an in-kernel tdb. Silently fail if no tdb is found. */ static void pfsync_update_net_tdb(struct pfsync_tdb *pt) { struct tdb *tdb; int s; /* check for invalid values */ if (ntohl(pt->spi) <= SPI_RESERVED_MAX || (pt->dst.sa.sa_family != AF_INET && pt->dst.sa.sa_family != AF_INET6)) goto bad; tdb = gettdb(pt->spi, &pt->dst, pt->sproto); if (tdb) { pt->rpl = ntohl(pt->rpl); pt->cur_bytes = (unsigned long long)be64toh(pt->cur_bytes); /* Neither replay nor byte counter should ever decrease. */ if (pt->rpl < tdb->tdb_rpl || pt->cur_bytes < tdb->tdb_cur_bytes) { goto bad; } tdb->tdb_rpl = pt->rpl; tdb->tdb_cur_bytes = pt->cur_bytes; } return; bad: if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync_insert: PFSYNC_ACT_TDB_UPD: " "invalid value\n"); V_pfsyncstats.pfsyncs_badstate++; return; } #endif static int pfsync_in_eof(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { /* check if we are at the right place in the packet */ if (offset != m->m_pkthdr.len) V_pfsyncstats.pfsyncs_badlen++; /* we're done. free and let the caller return */ m_freem(m); return (-1); } static int pfsync_in_error(struct pfsync_pkt *pkt, struct mbuf *m, int offset, int count) { V_pfsyncstats.pfsyncs_badact++; m_freem(m); return (-1); } static int pfsyncoutput(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst, struct route *rt) { m_freem(m); return (0); } /* ARGSUSED */ static int pfsyncioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct pfsync_softc *sc = ifp->if_softc; struct ifreq *ifr = (struct ifreq *)data; struct pfsyncreq pfsyncr; int error; switch (cmd) { case SIOCSIFFLAGS: PFSYNC_LOCK(sc); if (ifp->if_flags & IFF_UP) { ifp->if_drv_flags |= IFF_DRV_RUNNING; PFSYNC_UNLOCK(sc); pfsync_pointers_init(); } else { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; PFSYNC_UNLOCK(sc); pfsync_pointers_uninit(); } break; case SIOCSIFMTU: if (!sc->sc_sync_if || ifr->ifr_mtu <= PFSYNC_MINPKT || ifr->ifr_mtu > sc->sc_sync_if->if_mtu) return (EINVAL); if (ifr->ifr_mtu < ifp->if_mtu) { PFSYNC_LOCK(sc); if (sc->sc_len > PFSYNC_MINPKT) pfsync_sendout(1); PFSYNC_UNLOCK(sc); } ifp->if_mtu = ifr->ifr_mtu; break; case SIOCGETPFSYNC: bzero(&pfsyncr, sizeof(pfsyncr)); PFSYNC_LOCK(sc); if (sc->sc_sync_if) { strlcpy(pfsyncr.pfsyncr_syncdev, sc->sc_sync_if->if_xname, IFNAMSIZ); } pfsyncr.pfsyncr_syncpeer = sc->sc_sync_peer; pfsyncr.pfsyncr_maxupdates = sc->sc_maxupdates; pfsyncr.pfsyncr_defer = (PFSYNCF_DEFER == (sc->sc_flags & PFSYNCF_DEFER)); PFSYNC_UNLOCK(sc); return (copyout(&pfsyncr, ifr_data_get_ptr(ifr), sizeof(pfsyncr))); case SIOCSETPFSYNC: { struct ip_moptions *imo = &sc->sc_imo; struct ifnet *sifp; struct ip *ip; void *mship = NULL; if ((error = priv_check(curthread, PRIV_NETINET_PF)) != 0) return (error); if ((error = copyin(ifr_data_get_ptr(ifr), &pfsyncr, sizeof(pfsyncr)))) return (error); if (pfsyncr.pfsyncr_maxupdates > 255) return (EINVAL); if (pfsyncr.pfsyncr_syncdev[0] == 0) sifp = NULL; else if ((sifp = ifunit_ref(pfsyncr.pfsyncr_syncdev)) == NULL) return (EINVAL); if (sifp != NULL && ( pfsyncr.pfsyncr_syncpeer.s_addr == 0 || pfsyncr.pfsyncr_syncpeer.s_addr == htonl(INADDR_PFSYNC_GROUP))) mship = malloc((sizeof(struct in_multi *) * IP_MIN_MEMBERSHIPS), M_PFSYNC, M_WAITOK | M_ZERO); PFSYNC_LOCK(sc); if (pfsyncr.pfsyncr_syncpeer.s_addr == 0) sc->sc_sync_peer.s_addr = htonl(INADDR_PFSYNC_GROUP); else sc->sc_sync_peer.s_addr = pfsyncr.pfsyncr_syncpeer.s_addr; sc->sc_maxupdates = pfsyncr.pfsyncr_maxupdates; if (pfsyncr.pfsyncr_defer) { sc->sc_flags |= PFSYNCF_DEFER; pfsync_defer_ptr = pfsync_defer; } else { sc->sc_flags &= ~PFSYNCF_DEFER; pfsync_defer_ptr = NULL; } if (sifp == NULL) { if (sc->sc_sync_if) if_rele(sc->sc_sync_if); sc->sc_sync_if = NULL; if (imo->imo_membership) pfsync_multicast_cleanup(sc); PFSYNC_UNLOCK(sc); break; } if (sc->sc_len > PFSYNC_MINPKT && (sifp->if_mtu < sc->sc_ifp->if_mtu || (sc->sc_sync_if != NULL && sifp->if_mtu < sc->sc_sync_if->if_mtu) || sifp->if_mtu < MCLBYTES - sizeof(struct ip))) pfsync_sendout(1); if (imo->imo_membership) pfsync_multicast_cleanup(sc); if (sc->sc_sync_peer.s_addr == htonl(INADDR_PFSYNC_GROUP)) { error = pfsync_multicast_setup(sc, sifp, mship); if (error) { if_rele(sifp); free(mship, M_PFSYNC); return (error); } } if (sc->sc_sync_if) if_rele(sc->sc_sync_if); sc->sc_sync_if = sifp; ip = &sc->sc_template; bzero(ip, sizeof(*ip)); ip->ip_v = IPVERSION; ip->ip_hl = sizeof(sc->sc_template) >> 2; ip->ip_tos = IPTOS_LOWDELAY; /* len and id are set later. */ ip->ip_off = htons(IP_DF); ip->ip_ttl = PFSYNC_DFLTTL; ip->ip_p = IPPROTO_PFSYNC; ip->ip_src.s_addr = INADDR_ANY; ip->ip_dst.s_addr = sc->sc_sync_peer.s_addr; /* Request a full state table update. */ if ((sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) (*carp_demote_adj_p)(V_pfsync_carp_adj, "pfsync bulk start"); sc->sc_flags &= ~PFSYNCF_OK; if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: requesting bulk update\n"); pfsync_request_update(0, 0); PFSYNC_UNLOCK(sc); PFSYNC_BLOCK(sc); sc->sc_ureq_sent = time_uptime; callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, sc); PFSYNC_BUNLOCK(sc); break; } default: return (ENOTTY); } return (0); } static void pfsync_out_state(struct pf_state *st, void *buf) { struct pfsync_state *sp = buf; pfsync_state_export(sp, st); } static void pfsync_out_iack(struct pf_state *st, void *buf) { struct pfsync_ins_ack *iack = buf; iack->id = st->id; iack->creatorid = st->creatorid; } static void pfsync_out_upd_c(struct pf_state *st, void *buf) { struct pfsync_upd_c *up = buf; bzero(up, sizeof(*up)); up->id = st->id; pf_state_peer_hton(&st->src, &up->src); pf_state_peer_hton(&st->dst, &up->dst); up->creatorid = st->creatorid; up->timeout = st->timeout; } static void pfsync_out_del(struct pf_state *st, void *buf) { struct pfsync_del_c *dp = buf; dp->id = st->id; dp->creatorid = st->creatorid; st->state_flags |= PFSTATE_NOSYNC; } static void pfsync_drop(struct pfsync_softc *sc) { struct pf_state *st, *next; struct pfsync_upd_req_item *ur; int q; for (q = 0; q < PFSYNC_S_COUNT; q++) { if (TAILQ_EMPTY(&sc->sc_qs[q])) continue; TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, next) { KASSERT(st->sync_state == q, ("%s: st->sync_state == q", __func__)); st->sync_state = PFSYNC_S_NONE; pf_release_state(st); } TAILQ_INIT(&sc->sc_qs[q]); } while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); free(ur, M_PFSYNC); } sc->sc_plus = NULL; sc->sc_len = PFSYNC_MINPKT; } static void pfsync_sendout(int schedswi) { struct pfsync_softc *sc = V_pfsyncif; struct ifnet *ifp = sc->sc_ifp; struct mbuf *m; struct ip *ip; struct pfsync_header *ph; struct pfsync_subheader *subh; struct pf_state *st, *st_next; struct pfsync_upd_req_item *ur; int offset; int q, count = 0; KASSERT(sc != NULL, ("%s: null sc", __func__)); KASSERT(sc->sc_len > PFSYNC_MINPKT, ("%s: sc_len %zu", __func__, sc->sc_len)); PFSYNC_LOCK_ASSERT(sc); if (ifp->if_bpf == NULL && sc->sc_sync_if == NULL) { pfsync_drop(sc); return; } m = m_get2(max_linkhdr + sc->sc_len, M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) { if_inc_counter(sc->sc_ifp, IFCOUNTER_OERRORS, 1); V_pfsyncstats.pfsyncs_onomem++; return; } m->m_data += max_linkhdr; m->m_len = m->m_pkthdr.len = sc->sc_len; /* build the ip header */ ip = (struct ip *)m->m_data; bcopy(&sc->sc_template, ip, sizeof(*ip)); offset = sizeof(*ip); ip->ip_len = htons(m->m_pkthdr.len); ip_fillid(ip); /* build the pfsync header */ ph = (struct pfsync_header *)(m->m_data + offset); bzero(ph, sizeof(*ph)); offset += sizeof(*ph); ph->version = PFSYNC_VERSION; ph->len = htons(sc->sc_len - sizeof(*ip)); bcopy(V_pf_status.pf_chksum, ph->pfcksum, PF_MD5_DIGEST_LENGTH); /* walk the queues */ for (q = 0; q < PFSYNC_S_COUNT; q++) { if (TAILQ_EMPTY(&sc->sc_qs[q])) continue; subh = (struct pfsync_subheader *)(m->m_data + offset); offset += sizeof(*subh); count = 0; TAILQ_FOREACH_SAFE(st, &sc->sc_qs[q], sync_list, st_next) { KASSERT(st->sync_state == q, ("%s: st->sync_state == q", __func__)); /* * XXXGL: some of write methods do unlocked reads * of state data :( */ pfsync_qs[q].write(st, m->m_data + offset); offset += pfsync_qs[q].len; st->sync_state = PFSYNC_S_NONE; pf_release_state(st); count++; } TAILQ_INIT(&sc->sc_qs[q]); bzero(subh, sizeof(*subh)); subh->action = pfsync_qs[q].action; subh->count = htons(count); V_pfsyncstats.pfsyncs_oacts[pfsync_qs[q].action] += count; } if (!TAILQ_EMPTY(&sc->sc_upd_req_list)) { subh = (struct pfsync_subheader *)(m->m_data + offset); offset += sizeof(*subh); count = 0; while ((ur = TAILQ_FIRST(&sc->sc_upd_req_list)) != NULL) { TAILQ_REMOVE(&sc->sc_upd_req_list, ur, ur_entry); bcopy(&ur->ur_msg, m->m_data + offset, sizeof(ur->ur_msg)); offset += sizeof(ur->ur_msg); free(ur, M_PFSYNC); count++; } bzero(subh, sizeof(*subh)); subh->action = PFSYNC_ACT_UPD_REQ; subh->count = htons(count); V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_UPD_REQ] += count; } /* has someone built a custom region for us to add? */ if (sc->sc_plus != NULL) { bcopy(sc->sc_plus, m->m_data + offset, sc->sc_pluslen); offset += sc->sc_pluslen; sc->sc_plus = NULL; } subh = (struct pfsync_subheader *)(m->m_data + offset); offset += sizeof(*subh); bzero(subh, sizeof(*subh)); subh->action = PFSYNC_ACT_EOF; subh->count = htons(1); V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_EOF]++; /* we're done, let's put it on the wire */ if (ifp->if_bpf) { m->m_data += sizeof(*ip); m->m_len = m->m_pkthdr.len = sc->sc_len - sizeof(*ip); BPF_MTAP(ifp, m); m->m_data -= sizeof(*ip); m->m_len = m->m_pkthdr.len = sc->sc_len; } if (sc->sc_sync_if == NULL) { sc->sc_len = PFSYNC_MINPKT; m_freem(m); return; } if_inc_counter(sc->sc_ifp, IFCOUNTER_OPACKETS, 1); if_inc_counter(sc->sc_ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len); sc->sc_len = PFSYNC_MINPKT; if (!_IF_QFULL(&sc->sc_ifp->if_snd)) _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); else { m_freem(m); if_inc_counter(sc->sc_ifp, IFCOUNTER_OQDROPS, 1); } if (schedswi) swi_sched(V_pfsync_swi_cookie, 0); } static void pfsync_insert_state(struct pf_state *st) { struct pfsync_softc *sc = V_pfsyncif; if (st->state_flags & PFSTATE_NOSYNC) return; if ((st->rule.ptr->rule_flag & PFRULE_NOSYNC) || st->key[PF_SK_WIRE]->proto == IPPROTO_PFSYNC) { st->state_flags |= PFSTATE_NOSYNC; return; } KASSERT(st->sync_state == PFSYNC_S_NONE, ("%s: st->sync_state %u", __func__, st->sync_state)); PFSYNC_LOCK(sc); if (sc->sc_len == PFSYNC_MINPKT) callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); pfsync_q_ins(st, PFSYNC_S_INS, true); PFSYNC_UNLOCK(sc); st->sync_updates = 0; } static int pfsync_defer(struct pf_state *st, struct mbuf *m) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_deferral *pd; if (m->m_flags & (M_BCAST|M_MCAST)) return (0); PFSYNC_LOCK(sc); if (sc == NULL || !(sc->sc_ifp->if_flags & IFF_DRV_RUNNING) || !(sc->sc_flags & PFSYNCF_DEFER)) { PFSYNC_UNLOCK(sc); return (0); } if (sc->sc_deferred >= 128) pfsync_undefer(TAILQ_FIRST(&sc->sc_deferrals), 0); pd = malloc(sizeof(*pd), M_PFSYNC, M_NOWAIT); if (pd == NULL) return (0); sc->sc_deferred++; m->m_flags |= M_SKIP_FIREWALL; st->state_flags |= PFSTATE_ACK; pd->pd_sc = sc; pd->pd_refs = 0; pd->pd_st = st; pf_ref_state(st); pd->pd_m = m; TAILQ_INSERT_TAIL(&sc->sc_deferrals, pd, pd_entry); callout_init_mtx(&pd->pd_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); callout_reset(&pd->pd_tmo, 10, pfsync_defer_tmo, pd); pfsync_push(sc); return (1); } static void pfsync_undefer(struct pfsync_deferral *pd, int drop) { struct pfsync_softc *sc = pd->pd_sc; struct mbuf *m = pd->pd_m; struct pf_state *st = pd->pd_st; PFSYNC_LOCK_ASSERT(sc); TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); sc->sc_deferred--; pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ free(pd, M_PFSYNC); pf_release_state(st); if (drop) m_freem(m); else { _IF_ENQUEUE(&sc->sc_ifp->if_snd, m); pfsync_push(sc); } } static void pfsync_defer_tmo(void *arg) { struct pfsync_deferral *pd = arg; struct pfsync_softc *sc = pd->pd_sc; struct mbuf *m = pd->pd_m; struct pf_state *st = pd->pd_st; PFSYNC_LOCK_ASSERT(sc); CURVNET_SET(m->m_pkthdr.rcvif->if_vnet); TAILQ_REMOVE(&sc->sc_deferrals, pd, pd_entry); sc->sc_deferred--; pd->pd_st->state_flags &= ~PFSTATE_ACK; /* XXX: locking! */ if (pd->pd_refs == 0) free(pd, M_PFSYNC); PFSYNC_UNLOCK(sc); ip_output(m, NULL, NULL, 0, NULL, NULL); pf_release_state(st); CURVNET_RESTORE(); } static void pfsync_undefer_state(struct pf_state *st, int drop) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_deferral *pd; PFSYNC_LOCK_ASSERT(sc); TAILQ_FOREACH(pd, &sc->sc_deferrals, pd_entry) { if (pd->pd_st == st) { if (callout_stop(&pd->pd_tmo) > 0) pfsync_undefer(pd, drop); return; } } panic("%s: unable to find deferred state", __func__); } static void pfsync_update_state(struct pf_state *st) { struct pfsync_softc *sc = V_pfsyncif; bool sync = false, ref = true; PF_STATE_LOCK_ASSERT(st); PFSYNC_LOCK(sc); if (st->state_flags & PFSTATE_ACK) pfsync_undefer_state(st, 0); if (st->state_flags & PFSTATE_NOSYNC) { if (st->sync_state != PFSYNC_S_NONE) pfsync_q_del(st, true); PFSYNC_UNLOCK(sc); return; } if (sc->sc_len == PFSYNC_MINPKT) callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); switch (st->sync_state) { case PFSYNC_S_UPD_C: case PFSYNC_S_UPD: case PFSYNC_S_INS: /* we're already handling it */ if (st->key[PF_SK_WIRE]->proto == IPPROTO_TCP) { st->sync_updates++; if (st->sync_updates >= sc->sc_maxupdates) sync = true; } break; case PFSYNC_S_IACK: pfsync_q_del(st, false); ref = false; /* FALLTHROUGH */ case PFSYNC_S_NONE: pfsync_q_ins(st, PFSYNC_S_UPD_C, ref); st->sync_updates = 0; break; default: panic("%s: unexpected sync state %d", __func__, st->sync_state); } if (sync || (time_uptime - st->pfsync_time) < 2) pfsync_push(sc); PFSYNC_UNLOCK(sc); } static void pfsync_request_update(u_int32_t creatorid, u_int64_t id) { struct pfsync_softc *sc = V_pfsyncif; struct pfsync_upd_req_item *item; size_t nlen = sizeof(struct pfsync_upd_req); PFSYNC_LOCK_ASSERT(sc); /* * This code does a bit to prevent multiple update requests for the * same state being generated. It searches current subheader queue, * but it doesn't lookup into queue of already packed datagrams. */ TAILQ_FOREACH(item, &sc->sc_upd_req_list, ur_entry) if (item->ur_msg.id == id && item->ur_msg.creatorid == creatorid) return; item = malloc(sizeof(*item), M_PFSYNC, M_NOWAIT); if (item == NULL) return; /* XXX stats */ item->ur_msg.id = id; item->ur_msg.creatorid = creatorid; if (TAILQ_EMPTY(&sc->sc_upd_req_list)) nlen += sizeof(struct pfsync_subheader); if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { pfsync_sendout(1); nlen = sizeof(struct pfsync_subheader) + sizeof(struct pfsync_upd_req); } TAILQ_INSERT_TAIL(&sc->sc_upd_req_list, item, ur_entry); sc->sc_len += nlen; } static void pfsync_update_state_req(struct pf_state *st) { struct pfsync_softc *sc = V_pfsyncif; bool ref = true; PF_STATE_LOCK_ASSERT(st); PFSYNC_LOCK(sc); if (st->state_flags & PFSTATE_NOSYNC) { if (st->sync_state != PFSYNC_S_NONE) pfsync_q_del(st, true); PFSYNC_UNLOCK(sc); return; } switch (st->sync_state) { case PFSYNC_S_UPD_C: case PFSYNC_S_IACK: pfsync_q_del(st, false); ref = false; /* FALLTHROUGH */ case PFSYNC_S_NONE: pfsync_q_ins(st, PFSYNC_S_UPD, ref); pfsync_push(sc); break; case PFSYNC_S_INS: case PFSYNC_S_UPD: case PFSYNC_S_DEL: /* we're already handling it */ break; default: panic("%s: unexpected sync state %d", __func__, st->sync_state); } PFSYNC_UNLOCK(sc); } static void pfsync_delete_state(struct pf_state *st) { struct pfsync_softc *sc = V_pfsyncif; bool ref = true; PFSYNC_LOCK(sc); if (st->state_flags & PFSTATE_ACK) pfsync_undefer_state(st, 1); if (st->state_flags & PFSTATE_NOSYNC) { if (st->sync_state != PFSYNC_S_NONE) pfsync_q_del(st, true); PFSYNC_UNLOCK(sc); return; } if (sc->sc_len == PFSYNC_MINPKT) callout_reset(&sc->sc_tmo, 1 * hz, pfsync_timeout, V_pfsyncif); switch (st->sync_state) { case PFSYNC_S_INS: /* We never got to tell the world so just forget about it. */ pfsync_q_del(st, true); break; case PFSYNC_S_UPD_C: case PFSYNC_S_UPD: case PFSYNC_S_IACK: pfsync_q_del(st, false); ref = false; /* FALLTHROUGH */ case PFSYNC_S_NONE: pfsync_q_ins(st, PFSYNC_S_DEL, ref); break; default: panic("%s: unexpected sync state %d", __func__, st->sync_state); } PFSYNC_UNLOCK(sc); } static void pfsync_clear_states(u_int32_t creatorid, const char *ifname) { struct pfsync_softc *sc = V_pfsyncif; struct { struct pfsync_subheader subh; struct pfsync_clr clr; } __packed r; bzero(&r, sizeof(r)); r.subh.action = PFSYNC_ACT_CLR; r.subh.count = htons(1); V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_CLR]++; strlcpy(r.clr.ifname, ifname, sizeof(r.clr.ifname)); r.clr.creatorid = creatorid; PFSYNC_LOCK(sc); pfsync_send_plus(&r, sizeof(r)); PFSYNC_UNLOCK(sc); } static void pfsync_q_ins(struct pf_state *st, int q, bool ref) { struct pfsync_softc *sc = V_pfsyncif; size_t nlen = pfsync_qs[q].len; PFSYNC_LOCK_ASSERT(sc); KASSERT(st->sync_state == PFSYNC_S_NONE, ("%s: st->sync_state %u", __func__, st->sync_state)); KASSERT(sc->sc_len >= PFSYNC_MINPKT, ("pfsync pkt len is too low %zu", sc->sc_len)); if (TAILQ_EMPTY(&sc->sc_qs[q])) nlen += sizeof(struct pfsync_subheader); if (sc->sc_len + nlen > sc->sc_ifp->if_mtu) { pfsync_sendout(1); nlen = sizeof(struct pfsync_subheader) + pfsync_qs[q].len; } sc->sc_len += nlen; TAILQ_INSERT_TAIL(&sc->sc_qs[q], st, sync_list); st->sync_state = q; if (ref) pf_ref_state(st); } static void pfsync_q_del(struct pf_state *st, bool unref) { struct pfsync_softc *sc = V_pfsyncif; int q = st->sync_state; PFSYNC_LOCK_ASSERT(sc); KASSERT(st->sync_state != PFSYNC_S_NONE, ("%s: st->sync_state != PFSYNC_S_NONE", __func__)); sc->sc_len -= pfsync_qs[q].len; TAILQ_REMOVE(&sc->sc_qs[q], st, sync_list); st->sync_state = PFSYNC_S_NONE; if (unref) pf_release_state(st); if (TAILQ_EMPTY(&sc->sc_qs[q])) sc->sc_len -= sizeof(struct pfsync_subheader); } static void pfsync_bulk_start(void) { struct pfsync_softc *sc = V_pfsyncif; if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: received bulk update request\n"); PFSYNC_BLOCK(sc); sc->sc_ureq_received = time_uptime; sc->sc_bulk_hashid = 0; sc->sc_bulk_stateid = 0; pfsync_bulk_status(PFSYNC_BUS_START); callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); PFSYNC_BUNLOCK(sc); } static void pfsync_bulk_update(void *arg) { struct pfsync_softc *sc = arg; struct pf_state *s; int i, sent = 0; PFSYNC_BLOCK_ASSERT(sc); CURVNET_SET(sc->sc_ifp->if_vnet); /* * Start with last state from previous invocation. * It may had gone, in this case start from the * hash slot. */ s = pf_find_state_byid(sc->sc_bulk_stateid, sc->sc_bulk_creatorid); if (s != NULL) i = PF_IDHASH(s); else i = sc->sc_bulk_hashid; for (; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; if (s != NULL) PF_HASHROW_ASSERT(ih); else { PF_HASHROW_LOCK(ih); s = LIST_FIRST(&ih->states); } for (; s; s = LIST_NEXT(s, entry)) { if (sent > 1 && (sc->sc_ifp->if_mtu - sc->sc_len) < sizeof(struct pfsync_state)) { /* We've filled a packet. */ sc->sc_bulk_hashid = i; sc->sc_bulk_stateid = s->id; sc->sc_bulk_creatorid = s->creatorid; PF_HASHROW_UNLOCK(ih); callout_reset(&sc->sc_bulk_tmo, 1, pfsync_bulk_update, sc); goto full; } if (s->sync_state == PFSYNC_S_NONE && s->timeout < PFTM_MAX && s->pfsync_time <= sc->sc_ureq_received) { pfsync_update_state_req(s); sent++; } } PF_HASHROW_UNLOCK(ih); } /* We're done. */ pfsync_bulk_status(PFSYNC_BUS_END); full: CURVNET_RESTORE(); } static void pfsync_bulk_status(u_int8_t status) { struct { struct pfsync_subheader subh; struct pfsync_bus bus; } __packed r; struct pfsync_softc *sc = V_pfsyncif; bzero(&r, sizeof(r)); r.subh.action = PFSYNC_ACT_BUS; r.subh.count = htons(1); V_pfsyncstats.pfsyncs_oacts[PFSYNC_ACT_BUS]++; r.bus.creatorid = V_pf_status.hostid; r.bus.endtime = htonl(time_uptime - sc->sc_ureq_received); r.bus.status = status; PFSYNC_LOCK(sc); pfsync_send_plus(&r, sizeof(r)); PFSYNC_UNLOCK(sc); } static void pfsync_bulk_fail(void *arg) { struct pfsync_softc *sc = arg; CURVNET_SET(sc->sc_ifp->if_vnet); PFSYNC_BLOCK_ASSERT(sc); if (sc->sc_bulk_tries++ < PFSYNC_MAX_BULKTRIES) { /* Try again */ callout_reset(&sc->sc_bulkfail_tmo, 5 * hz, pfsync_bulk_fail, V_pfsyncif); PFSYNC_LOCK(sc); pfsync_request_update(0, 0); PFSYNC_UNLOCK(sc); } else { /* Pretend like the transfer was ok. */ sc->sc_ureq_sent = 0; sc->sc_bulk_tries = 0; PFSYNC_LOCK(sc); if (!(sc->sc_flags & PFSYNCF_OK) && carp_demote_adj_p) (*carp_demote_adj_p)(-V_pfsync_carp_adj, "pfsync bulk fail"); sc->sc_flags |= PFSYNCF_OK; PFSYNC_UNLOCK(sc); if (V_pf_status.debug >= PF_DEBUG_MISC) printf("pfsync: failed to receive bulk update\n"); } CURVNET_RESTORE(); } static void pfsync_send_plus(void *plus, size_t pluslen) { struct pfsync_softc *sc = V_pfsyncif; PFSYNC_LOCK_ASSERT(sc); if (sc->sc_len + pluslen > sc->sc_ifp->if_mtu) pfsync_sendout(1); sc->sc_plus = plus; sc->sc_len += (sc->sc_pluslen = pluslen); pfsync_sendout(1); } static void pfsync_timeout(void *arg) { struct pfsync_softc *sc = arg; CURVNET_SET(sc->sc_ifp->if_vnet); PFSYNC_LOCK(sc); pfsync_push(sc); PFSYNC_UNLOCK(sc); CURVNET_RESTORE(); } static void pfsync_push(struct pfsync_softc *sc) { PFSYNC_LOCK_ASSERT(sc); sc->sc_flags |= PFSYNCF_PUSH; swi_sched(V_pfsync_swi_cookie, 0); } static void pfsyncintr(void *arg) { struct pfsync_softc *sc = arg; struct mbuf *m, *n; CURVNET_SET(sc->sc_ifp->if_vnet); PFSYNC_LOCK(sc); if ((sc->sc_flags & PFSYNCF_PUSH) && sc->sc_len > PFSYNC_MINPKT) { pfsync_sendout(0); sc->sc_flags &= ~PFSYNCF_PUSH; } _IF_DEQUEUE_ALL(&sc->sc_ifp->if_snd, m); PFSYNC_UNLOCK(sc); for (; m != NULL; m = n) { n = m->m_nextpkt; m->m_nextpkt = NULL; /* * We distinguish between a deferral packet and our * own pfsync packet based on M_SKIP_FIREWALL * flag. This is XXX. */ if (m->m_flags & M_SKIP_FIREWALL) ip_output(m, NULL, NULL, 0, NULL, NULL); else if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL) == 0) V_pfsyncstats.pfsyncs_opackets++; else V_pfsyncstats.pfsyncs_oerrors++; } CURVNET_RESTORE(); } static int pfsync_multicast_setup(struct pfsync_softc *sc, struct ifnet *ifp, void *mship) { struct ip_moptions *imo = &sc->sc_imo; int error; if (!(ifp->if_flags & IFF_MULTICAST)) return (EADDRNOTAVAIL); imo->imo_membership = (struct in_multi **)mship; imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; imo->imo_multicast_vif = -1; if ((error = in_joingroup(ifp, &sc->sc_sync_peer, NULL, &imo->imo_membership[0])) != 0) { imo->imo_membership = NULL; return (error); } imo->imo_num_memberships++; imo->imo_multicast_ifp = ifp; imo->imo_multicast_ttl = PFSYNC_DFLTTL; imo->imo_multicast_loop = 0; return (0); } static void pfsync_multicast_cleanup(struct pfsync_softc *sc) { struct ip_moptions *imo = &sc->sc_imo; in_leavegroup(imo->imo_membership[0], NULL); free(imo->imo_membership, M_PFSYNC); imo->imo_membership = NULL; imo->imo_multicast_ifp = NULL; } #ifdef INET extern struct domain inetdomain; static struct protosw in_pfsync_protosw = { .pr_type = SOCK_RAW, .pr_domain = &inetdomain, .pr_protocol = IPPROTO_PFSYNC, .pr_flags = PR_ATOMIC|PR_ADDR, .pr_input = pfsync_input, .pr_output = rip_output, .pr_ctloutput = rip_ctloutput, .pr_usrreqs = &rip_usrreqs }; #endif static void pfsync_pointers_init() { PF_RULES_WLOCK(); pfsync_state_import_ptr = pfsync_state_import; pfsync_insert_state_ptr = pfsync_insert_state; pfsync_update_state_ptr = pfsync_update_state; pfsync_delete_state_ptr = pfsync_delete_state; pfsync_clear_states_ptr = pfsync_clear_states; pfsync_defer_ptr = pfsync_defer; PF_RULES_WUNLOCK(); } static void pfsync_pointers_uninit() { PF_RULES_WLOCK(); pfsync_state_import_ptr = NULL; pfsync_insert_state_ptr = NULL; pfsync_update_state_ptr = NULL; pfsync_delete_state_ptr = NULL; pfsync_clear_states_ptr = NULL; pfsync_defer_ptr = NULL; PF_RULES_WUNLOCK(); } static void vnet_pfsync_init(const void *unused __unused) { int error; V_pfsync_cloner = if_clone_simple(pfsyncname, pfsync_clone_create, pfsync_clone_destroy, 1); error = swi_add(NULL, pfsyncname, pfsyncintr, V_pfsyncif, SWI_NET, INTR_MPSAFE, &V_pfsync_swi_cookie); if (error) { if_clone_detach(V_pfsync_cloner); log(LOG_INFO, "swi_add() failed in %s\n", __func__); } } VNET_SYSINIT(vnet_pfsync_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY, vnet_pfsync_init, NULL); static void vnet_pfsync_uninit(const void *unused __unused) { if_clone_detach(V_pfsync_cloner); swi_remove(V_pfsync_swi_cookie); } /* * Detach after pf is gone; otherwise we might touch pfsync memory * from within pf after freeing pfsync. */ VNET_SYSUNINIT(vnet_pfsync_uninit, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_pfsync_uninit, NULL); static int pfsync_init() { #ifdef INET int error; error = pf_proto_register(PF_INET, &in_pfsync_protosw); if (error) return (error); error = ipproto_register(IPPROTO_PFSYNC); if (error) { pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); return (error); } #endif pfsync_pointers_init(); return (0); } static void pfsync_uninit() { pfsync_pointers_uninit(); #ifdef INET ipproto_unregister(IPPROTO_PFSYNC); pf_proto_unregister(PF_INET, IPPROTO_PFSYNC, SOCK_RAW); #endif } static int pfsync_modevent(module_t mod, int type, void *data) { int error = 0; switch (type) { case MOD_LOAD: error = pfsync_init(); break; case MOD_QUIESCE: /* * Module should not be unloaded due to race conditions. */ error = EBUSY; break; case MOD_UNLOAD: pfsync_uninit(); break; default: error = EINVAL; break; } return (error); } static moduledata_t pfsync_mod = { pfsyncname, pfsync_modevent, 0 }; #define PFSYNC_MODVER 1 /* Stay on FIREWALL as we depend on pf being initialized and on inetdomain. */ DECLARE_MODULE(pfsync, pfsync_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); MODULE_VERSION(pfsync, PFSYNC_MODVER); MODULE_DEPEND(pfsync, pf, PF_MODVER, PF_MODVER, PF_MODVER); Index: head/sys/netpfil/pf/pf.c =================================================================== --- head/sys/netpfil/pf/pf.c (revision 336675) +++ head/sys/netpfil/pf/pf.c (revision 336676) @@ -1,6672 +1,6672 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * Copyright (c) 2002 - 2008 Henning Brauer * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * Effort sponsored in part by the Defense Advanced Research Projects * Agency (DARPA) and Air Force Research Laboratory, Air Force * Materiel Command, USAF, under agreement number F30602-01-2-0537. * * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_bpf.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* XXX: only for DIR_IN/DIR_OUT */ #ifdef INET6 #include #include #include #include #include #include #include #endif /* INET6 */ #include #include #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x /* * Global variables */ /* state tables */ VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]); VNET_DEFINE(struct pf_palist, pf_pabuf); VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); VNET_DEFINE(struct pf_kstatus, pf_status); VNET_DEFINE(u_int32_t, ticket_altqs_active); VNET_DEFINE(u_int32_t, ticket_altqs_inactive); VNET_DEFINE(int, altqs_inactive_open); VNET_DEFINE(u_int32_t, ticket_pabuf); VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) VNET_DEFINE(u_char, pf_tcp_secret[16]); #define V_pf_tcp_secret VNET(pf_tcp_secret) VNET_DEFINE(int, pf_tcp_secret_init); #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) VNET_DEFINE(int, pf_tcp_iss_off); #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) VNET_DECLARE(int, pf_vnet_active); #define V_pf_vnet_active VNET(pf_vnet_active) -static VNET_DEFINE(uint32_t, pf_purge_idx); +VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); #define V_pf_purge_idx VNET(pf_purge_idx) /* * Queue for pf_intr() sends. */ static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); struct pf_send_entry { STAILQ_ENTRY(pf_send_entry) pfse_next; struct mbuf *pfse_m; enum { PFSE_IP, PFSE_IP6, PFSE_ICMP, PFSE_ICMP6, } pfse_type; struct { int type; int code; int mtu; } icmpopts; }; STAILQ_HEAD(pf_send_head, pf_send_entry); -static VNET_DEFINE(struct pf_send_head, pf_sendqueue); +VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); #define V_pf_sendqueue VNET(pf_sendqueue) static struct mtx pf_sendqueue_mtx; MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) /* * Queue for pf_overload_task() tasks. */ struct pf_overload_entry { SLIST_ENTRY(pf_overload_entry) next; struct pf_addr addr; sa_family_t af; uint8_t dir; struct pf_rule *rule; }; SLIST_HEAD(pf_overload_head, pf_overload_entry); -static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue); +VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); #define V_pf_overloadqueue VNET(pf_overloadqueue) -static VNET_DEFINE(struct task, pf_overloadtask); +VNET_DEFINE_STATIC(struct task, pf_overloadtask); #define V_pf_overloadtask VNET(pf_overloadtask) static struct mtx pf_overloadqueue_mtx; MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, "pf overload/flush queue", MTX_DEF); #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules); struct mtx pf_unlnkdrules_mtx; MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", MTX_DEF); -static VNET_DEFINE(uma_zone_t, pf_sources_z); +VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); #define V_pf_sources_z VNET(pf_sources_z) uma_zone_t pf_mtag_z; VNET_DEFINE(uma_zone_t, pf_state_z); VNET_DEFINE(uma_zone_t, pf_state_key_z); VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); #define PFID_CPUBITS 8 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) #define PFID_MAXID (~PFID_CPUMASK) CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); static void pf_src_tree_remove_state(struct pf_state *); static void pf_init_threshold(struct pf_threshold *, u_int32_t, u_int32_t); static void pf_add_threshold(struct pf_threshold *); static int pf_check_threshold(struct pf_threshold *); static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, u_int16_t *, u_int16_t *, struct pf_addr *, u_int16_t, u_int8_t, sa_family_t); static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, struct tcphdr *, struct pf_state_peer *); static void pf_change_icmp(struct pf_addr *, u_int16_t *, struct pf_addr *, struct pf_addr *, u_int16_t, u_int16_t *, u_int16_t *, u_int16_t *, u_int16_t *, u_int8_t, sa_family_t); static void pf_send_tcp(struct mbuf *, const struct pf_rule *, sa_family_t, const struct pf_addr *, const struct pf_addr *, u_int16_t, u_int16_t, u_int32_t, u_int32_t, u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, u_int16_t, struct ifnet *); static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, sa_family_t, struct pf_rule *); static void pf_detach_state(struct pf_state *); static int pf_state_key_attach(struct pf_state_key *, struct pf_state_key *, struct pf_state *); static void pf_state_key_detach(struct pf_state *, int); static int pf_state_key_ctor(void *, int, void *, int); static u_int32_t pf_tcp_iss(struct pf_pdesc *); static int pf_test_rule(struct pf_rule **, struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, struct pf_pdesc *, struct pf_rule **, struct pf_ruleset **, struct inpcb *); static int pf_create_state(struct pf_rule *, struct pf_rule *, struct pf_rule *, struct pf_pdesc *, struct pf_src_node *, struct pf_state_key *, struct pf_state_key *, struct mbuf *, int, u_int16_t, u_int16_t, int *, struct pfi_kif *, struct pf_state **, int, u_int16_t, u_int16_t, int); static int pf_test_fragment(struct pf_rule **, int, struct pfi_kif *, struct mbuf *, void *, struct pf_pdesc *, struct pf_rule **, struct pf_ruleset **); static int pf_tcp_track_full(struct pf_state_peer *, struct pf_state_peer *, struct pf_state **, struct pfi_kif *, struct mbuf *, int, struct pf_pdesc *, u_short *, int *); static int pf_tcp_track_sloppy(struct pf_state_peer *, struct pf_state_peer *, struct pf_state **, struct pf_pdesc *, u_short *); static int pf_test_state_tcp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *, u_short *); static int pf_test_state_udp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *); static int pf_test_state_icmp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *, u_short *); static int pf_test_state_other(struct pf_state **, int, struct pfi_kif *, struct mbuf *, struct pf_pdesc *); static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, sa_family_t); static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, sa_family_t); static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, int, u_int16_t); static int pf_check_proto_cksum(struct mbuf *, int, int, u_int8_t, sa_family_t); static void pf_print_state_parts(struct pf_state *, struct pf_state_key *, struct pf_state_key *); static int pf_addr_wrap_neq(struct pf_addr_wrap *, struct pf_addr_wrap *); static struct pf_state *pf_find_state(struct pfi_kif *, struct pf_state_key_cmp *, u_int); static int pf_src_connlimit(struct pf_state **); static void pf_overload_task(void *v, int pending); static int pf_insert_src_node(struct pf_src_node **, struct pf_rule *, struct pf_addr *, sa_family_t); static u_int pf_purge_expired_states(u_int, int); static void pf_purge_unlinked_rules(void); static int pf_mtag_uminit(void *, int, int); static void pf_mtag_free(struct m_tag *); #ifdef INET static void pf_route(struct mbuf **, struct pf_rule *, int, struct ifnet *, struct pf_state *, struct pf_pdesc *, struct inpcb *); #endif /* INET */ #ifdef INET6 static void pf_change_a6(struct pf_addr *, u_int16_t *, struct pf_addr *, u_int8_t); static void pf_route6(struct mbuf **, struct pf_rule *, int, struct ifnet *, struct pf_state *, struct pf_pdesc *, struct inpcb *); #endif /* INET6 */ int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); extern int pf_end_threads; extern struct proc *pf_purge_proc; VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ (pd)->pf_mtag->flags & PF_PACKET_LOOPED) #define STATE_LOOKUP(i, k, d, s, pd) \ do { \ (s) = pf_find_state((i), (k), (d)); \ if ((s) == NULL) \ return (PF_DROP); \ if (PACKET_LOOPED(pd)) \ return (PF_PASS); \ if ((d) == PF_OUT && \ (((s)->rule.ptr->rt == PF_ROUTETO && \ (s)->rule.ptr->direction == PF_OUT) || \ ((s)->rule.ptr->rt == PF_REPLYTO && \ (s)->rule.ptr->direction == PF_IN)) && \ (s)->rt_kif != NULL && \ (s)->rt_kif != (i)) \ return (PF_PASS); \ } while (0) #define BOUND_IFACE(r, k) \ ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all #define STATE_INC_COUNTERS(s) \ do { \ counter_u64_add(s->rule.ptr->states_cur, 1); \ counter_u64_add(s->rule.ptr->states_tot, 1); \ if (s->anchor.ptr != NULL) { \ counter_u64_add(s->anchor.ptr->states_cur, 1); \ counter_u64_add(s->anchor.ptr->states_tot, 1); \ } \ if (s->nat_rule.ptr != NULL) { \ counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ } \ } while (0) #define STATE_DEC_COUNTERS(s) \ do { \ if (s->nat_rule.ptr != NULL) \ counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ if (s->anchor.ptr != NULL) \ counter_u64_add(s->anchor.ptr->states_cur, -1); \ counter_u64_add(s->rule.ptr->states_cur, -1); \ } while (0) static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); VNET_DEFINE(struct pf_keyhash *, pf_keyhash); VNET_DEFINE(struct pf_idhash *, pf_idhash); VNET_DEFINE(struct pf_srchash *, pf_srchash); SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)"); u_long pf_hashmask; u_long pf_srchashmask; static u_long pf_hashsize; static u_long pf_srchashsize; u_long pf_ioctl_maxcount = 65535; SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, &pf_hashsize, 0, "Size of pf(4) states hashtable"); SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RDTUN, &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); VNET_DEFINE(void *, pf_swi_cookie); VNET_DEFINE(uint32_t, pf_hashseed); #define V_pf_hashseed VNET(pf_hashseed) int pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: if (a->addr32[0] > b->addr32[0]) return (1); if (a->addr32[0] < b->addr32[0]) return (-1); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (a->addr32[3] > b->addr32[3]) return (1); if (a->addr32[3] < b->addr32[3]) return (-1); if (a->addr32[2] > b->addr32[2]) return (1); if (a->addr32[2] < b->addr32[2]) return (-1); if (a->addr32[1] > b->addr32[1]) return (1); if (a->addr32[1] < b->addr32[1]) return (-1); if (a->addr32[0] > b->addr32[0]) return (1); if (a->addr32[0] < b->addr32[0]) return (-1); break; #endif /* INET6 */ default: panic("%s: unknown address family %u", __func__, af); } return (0); } static __inline uint32_t pf_hashkey(struct pf_state_key *sk) { uint32_t h; h = murmur3_32_hash32((uint32_t *)sk, sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), V_pf_hashseed); return (h & pf_hashmask); } static __inline uint32_t pf_hashsrc(struct pf_addr *addr, sa_family_t af) { uint32_t h; switch (af) { case AF_INET: h = murmur3_32_hash32((uint32_t *)&addr->v4, sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); break; case AF_INET6: h = murmur3_32_hash32((uint32_t *)&addr->v6, sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); break; default: panic("%s: unknown address family %u", __func__, af); } return (h & pf_srchashmask); } #ifdef ALTQ static int pf_state_hash(struct pf_state *s) { u_int32_t hv = (intptr_t)s / sizeof(*s); hv ^= crc32(&s->src, sizeof(s->src)); hv ^= crc32(&s->dst, sizeof(s->dst)); if (hv == 0) hv = 1; return (hv); } #endif #ifdef INET6 void pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: dst->addr32[0] = src->addr32[0]; break; #endif /* INET */ case AF_INET6: dst->addr32[0] = src->addr32[0]; dst->addr32[1] = src->addr32[1]; dst->addr32[2] = src->addr32[2]; dst->addr32[3] = src->addr32[3]; break; } } #endif /* INET6 */ static void pf_init_threshold(struct pf_threshold *threshold, u_int32_t limit, u_int32_t seconds) { threshold->limit = limit * PF_THRESHOLD_MULT; threshold->seconds = seconds; threshold->count = 0; threshold->last = time_uptime; } static void pf_add_threshold(struct pf_threshold *threshold) { u_int32_t t = time_uptime, diff = t - threshold->last; if (diff >= threshold->seconds) threshold->count = 0; else threshold->count -= threshold->count * diff / threshold->seconds; threshold->count += PF_THRESHOLD_MULT; threshold->last = t; } static int pf_check_threshold(struct pf_threshold *threshold) { return (threshold->count > threshold->limit); } static int pf_src_connlimit(struct pf_state **state) { struct pf_overload_entry *pfoe; int bad = 0; PF_STATE_LOCK_ASSERT(*state); (*state)->src_node->conn++; (*state)->src.tcp_est = 1; pf_add_threshold(&(*state)->src_node->conn_rate); if ((*state)->rule.ptr->max_src_conn && (*state)->rule.ptr->max_src_conn < (*state)->src_node->conn) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); bad++; } if ((*state)->rule.ptr->max_src_conn_rate.limit && pf_check_threshold(&(*state)->src_node->conn_rate)) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); bad++; } if (!bad) return (0); /* Kill this state. */ (*state)->timeout = PFTM_PURGE; (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; if ((*state)->rule.ptr->overload_tbl == NULL) return (1); /* Schedule overloading and flushing task. */ pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); if (pfoe == NULL) return (1); /* too bad :( */ bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); pfoe->af = (*state)->key[PF_SK_WIRE]->af; pfoe->rule = (*state)->rule.ptr; pfoe->dir = (*state)->direction; PF_OVERLOADQ_LOCK(); SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); PF_OVERLOADQ_UNLOCK(); taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); return (1); } static void pf_overload_task(void *v, int pending) { struct pf_overload_head queue; struct pfr_addr p; struct pf_overload_entry *pfoe, *pfoe1; uint32_t killed = 0; CURVNET_SET((struct vnet *)v); PF_OVERLOADQ_LOCK(); queue = V_pf_overloadqueue; SLIST_INIT(&V_pf_overloadqueue); PF_OVERLOADQ_UNLOCK(); bzero(&p, sizeof(p)); SLIST_FOREACH(pfoe, &queue, next) { counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("%s: blocking address ", __func__); pf_print_host(&pfoe->addr, 0, pfoe->af); printf("\n"); } p.pfra_af = pfoe->af; switch (pfoe->af) { #ifdef INET case AF_INET: p.pfra_net = 32; p.pfra_ip4addr = pfoe->addr.v4; break; #endif #ifdef INET6 case AF_INET6: p.pfra_net = 128; p.pfra_ip6addr = pfoe->addr.v6; break; #endif } PF_RULES_WLOCK(); pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); PF_RULES_WUNLOCK(); } /* * Remove those entries, that don't need flushing. */ SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) if (pfoe->rule->flush == 0) { SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); free(pfoe, M_PFTEMP); } else counter_u64_add( V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); /* If nothing to flush, return. */ if (SLIST_EMPTY(&queue)) { CURVNET_RESTORE(); return; } for (int i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state_key *sk; struct pf_state *s; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { sk = s->key[PF_SK_WIRE]; SLIST_FOREACH(pfoe, &queue, next) if (sk->af == pfoe->af && ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || pfoe->rule == s->rule.ptr) && ((pfoe->dir == PF_OUT && PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || (pfoe->dir == PF_IN && PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { s->timeout = PFTM_PURGE; s->src.state = s->dst.state = TCPS_CLOSED; killed++; } } PF_HASHROW_UNLOCK(ih); } SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) free(pfoe, M_PFTEMP); if (V_pf_status.debug >= PF_DEBUG_MISC) printf("%s: %u states killed", __func__, killed); CURVNET_RESTORE(); } /* * Can return locked on failure, so that we can consistently * allocate and insert a new one. */ struct pf_src_node * pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af, int returnlocked) { struct pf_srchash *sh; struct pf_src_node *n; counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); sh = &V_pf_srchash[pf_hashsrc(src, af)]; PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) if (n->rule.ptr == rule && n->af == af && ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) break; if (n != NULL) { n->states++; PF_HASHROW_UNLOCK(sh); } else if (returnlocked == 0) PF_HASHROW_UNLOCK(sh); return (n); } static int pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, struct pf_addr *src, sa_family_t af) { KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK || rule->rpool.opts & PF_POOL_STICKYADDR), ("%s for non-tracking rule %p", __func__, rule)); if (*sn == NULL) *sn = pf_find_src_node(src, rule, af, 1); if (*sn == NULL) { struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; PF_HASHROW_ASSERT(sh); if (!rule->max_src_nodes || counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); else counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); if ((*sn) == NULL) { PF_HASHROW_UNLOCK(sh); return (-1); } pf_init_threshold(&(*sn)->conn_rate, rule->max_src_conn_rate.limit, rule->max_src_conn_rate.seconds); (*sn)->af = af; (*sn)->rule.ptr = rule; PF_ACPY(&(*sn)->addr, src, af); LIST_INSERT_HEAD(&sh->nodes, *sn, entry); (*sn)->creation = time_uptime; (*sn)->ruletype = rule->action; (*sn)->states = 1; if ((*sn)->rule.ptr != NULL) counter_u64_add((*sn)->rule.ptr->src_nodes, 1); PF_HASHROW_UNLOCK(sh); counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); } else { if (rule->max_src_states && (*sn)->states >= rule->max_src_states) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1); return (-1); } } return (0); } void pf_unlink_src_node(struct pf_src_node *src) { PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); LIST_REMOVE(src, entry); if (src->rule.ptr) counter_u64_add(src->rule.ptr->src_nodes, -1); } u_int pf_free_src_nodes(struct pf_src_node_list *head) { struct pf_src_node *sn, *tmp; u_int count = 0; LIST_FOREACH_SAFE(sn, head, entry, tmp) { uma_zfree(V_pf_sources_z, sn); count++; } counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); return (count); } void pf_mtag_initialize() { pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, UMA_ALIGN_PTR, 0); } /* Per-vnet data storage structures initialization. */ void pf_initialize() { struct pf_keyhash *kh; struct pf_idhash *ih; struct pf_srchash *sh; u_int i; if (pf_hashsize == 0 || !powerof2(pf_hashsize)) pf_hashsize = PF_HASHSIZ; if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) pf_srchashsize = PF_SRCHASHSIZ; V_pf_hashseed = arc4random(); /* States and state keys storage. */ V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); V_pf_state_key_z = uma_zcreate("pf state keys", sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_NOWAIT | M_ZERO); V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { printf("pf: Unable to allocate memory for " "state_hashsize %lu.\n", pf_hashsize); free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); pf_hashsize = PF_HASHSIZ; V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); } pf_hashmask = pf_hashsize - 1; for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; i++, kh++, ih++) { mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); } /* Source nodes. */ V_pf_sources_z = uma_zcreate("pf source nodes", sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); V_pf_srchash = mallocarray(pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_srchash == NULL) { printf("pf: Unable to allocate memory for " "source_hashsize %lu.\n", pf_srchashsize); pf_srchashsize = PF_SRCHASHSIZ; V_pf_srchash = mallocarray(pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); } pf_srchashmask = pf_srchashsize - 1; for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); /* ALTQ */ TAILQ_INIT(&V_pf_altqs[0]); TAILQ_INIT(&V_pf_altqs[1]); TAILQ_INIT(&V_pf_pabuf); V_pf_altqs_active = &V_pf_altqs[0]; V_pf_altqs_inactive = &V_pf_altqs[1]; /* Send & overload+flush queues. */ STAILQ_INIT(&V_pf_sendqueue); SLIST_INIT(&V_pf_overloadqueue); TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); /* Unlinked, but may be referenced rules. */ TAILQ_INIT(&V_pf_unlinked_rules); } void pf_mtag_cleanup() { uma_zdestroy(pf_mtag_z); } void pf_cleanup() { struct pf_keyhash *kh; struct pf_idhash *ih; struct pf_srchash *sh; struct pf_send_entry *pfse, *next; u_int i; for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; i++, kh++, ih++) { KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", __func__)); KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", __func__)); mtx_destroy(&kh->lock); mtx_destroy(&ih->lock); } free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { KASSERT(LIST_EMPTY(&sh->nodes), ("%s: source node hash not empty", __func__)); mtx_destroy(&sh->lock); } free(V_pf_srchash, M_PFHASH); STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { m_freem(pfse->pfse_m); free(pfse, M_PFTEMP); } uma_zdestroy(V_pf_sources_z); uma_zdestroy(V_pf_state_z); uma_zdestroy(V_pf_state_key_z); } static int pf_mtag_uminit(void *mem, int size, int how) { struct m_tag *t; t = (struct m_tag *)mem; t->m_tag_cookie = MTAG_ABI_COMPAT; t->m_tag_id = PACKET_TAG_PF; t->m_tag_len = sizeof(struct pf_mtag); t->m_tag_free = pf_mtag_free; return (0); } static void pf_mtag_free(struct m_tag *t) { uma_zfree(pf_mtag_z, t); } struct pf_mtag * pf_get_mtag(struct mbuf *m) { struct m_tag *mtag; if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) return ((struct pf_mtag *)(mtag + 1)); mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); if (mtag == NULL) return (NULL); bzero(mtag + 1, sizeof(struct pf_mtag)); m_tag_prepend(m, mtag); return ((struct pf_mtag *)(mtag + 1)); } static int pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, struct pf_state *s) { struct pf_keyhash *khs, *khw, *kh; struct pf_state_key *sk, *cur; struct pf_state *si, *olds = NULL; int idx; KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); /* * We need to lock hash slots of both keys. To avoid deadlock * we always lock the slot with lower address first. Unlock order * isn't important. * * We also need to lock ID hash slot before dropping key * locks. On success we return with ID hash slot locked. */ if (skw == sks) { khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; PF_HASHROW_LOCK(khs); } else { khs = &V_pf_keyhash[pf_hashkey(sks)]; khw = &V_pf_keyhash[pf_hashkey(skw)]; if (khs == khw) { PF_HASHROW_LOCK(khs); } else if (khs < khw) { PF_HASHROW_LOCK(khs); PF_HASHROW_LOCK(khw); } else { PF_HASHROW_LOCK(khw); PF_HASHROW_LOCK(khs); } } #define KEYS_UNLOCK() do { \ if (khs != khw) { \ PF_HASHROW_UNLOCK(khs); \ PF_HASHROW_UNLOCK(khw); \ } else \ PF_HASHROW_UNLOCK(khs); \ } while (0) /* * First run: start with wire key. */ sk = skw; kh = khw; idx = PF_SK_WIRE; keyattach: LIST_FOREACH(cur, &kh->keys, entry) if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) break; if (cur != NULL) { /* Key exists. Check for same kif, if none, add to key. */ TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; PF_HASHROW_LOCK(ih); if (si->kif == s->kif && si->direction == s->direction) { if (sk->proto == IPPROTO_TCP && si->src.state >= TCPS_FIN_WAIT_2 && si->dst.state >= TCPS_FIN_WAIT_2) { /* * New state matches an old >FIN_WAIT_2 * state. We can't drop key hash locks, * thus we can't unlink it properly. * * As a workaround we drop it into * TCPS_CLOSED state, schedule purge * ASAP and push it into the very end * of the slot TAILQ, so that it won't * conflict with our new state. */ si->src.state = si->dst.state = TCPS_CLOSED; si->timeout = PFTM_PURGE; olds = si; } else { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: %s key attach " "failed on %s: ", (idx == PF_SK_WIRE) ? "wire" : "stack", s->kif->pfik_name); pf_print_state_parts(s, (idx == PF_SK_WIRE) ? sk : NULL, (idx == PF_SK_STACK) ? sk : NULL); printf(", existing: "); pf_print_state_parts(si, (idx == PF_SK_WIRE) ? sk : NULL, (idx == PF_SK_STACK) ? sk : NULL); printf("\n"); } PF_HASHROW_UNLOCK(ih); KEYS_UNLOCK(); uma_zfree(V_pf_state_key_z, sk); if (idx == PF_SK_STACK) pf_detach_state(s); return (EEXIST); /* collision! */ } } PF_HASHROW_UNLOCK(ih); } uma_zfree(V_pf_state_key_z, sk); s->key[idx] = cur; } else { LIST_INSERT_HEAD(&kh->keys, sk, entry); s->key[idx] = sk; } stateattach: /* List is sorted, if-bound states before floating. */ if (s->kif == V_pfi_all) TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); else TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); if (olds) { TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, key_list[idx]); olds = NULL; } /* * Attach done. See how should we (or should not?) * attach a second key. */ if (sks == skw) { s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; idx = PF_SK_STACK; sks = NULL; goto stateattach; } else if (sks != NULL) { /* * Continue attaching with stack key. */ sk = sks; kh = khs; idx = PF_SK_STACK; sks = NULL; goto keyattach; } PF_STATE_LOCK(s); KEYS_UNLOCK(); KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, ("%s failure", __func__)); return (0); #undef KEYS_UNLOCK } static void pf_detach_state(struct pf_state *s) { struct pf_state_key *sks = s->key[PF_SK_STACK]; struct pf_keyhash *kh; if (sks != NULL) { kh = &V_pf_keyhash[pf_hashkey(sks)]; PF_HASHROW_LOCK(kh); if (s->key[PF_SK_STACK] != NULL) pf_state_key_detach(s, PF_SK_STACK); /* * If both point to same key, then we are done. */ if (sks == s->key[PF_SK_WIRE]) { pf_state_key_detach(s, PF_SK_WIRE); PF_HASHROW_UNLOCK(kh); return; } PF_HASHROW_UNLOCK(kh); } if (s->key[PF_SK_WIRE] != NULL) { kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; PF_HASHROW_LOCK(kh); if (s->key[PF_SK_WIRE] != NULL) pf_state_key_detach(s, PF_SK_WIRE); PF_HASHROW_UNLOCK(kh); } } static void pf_state_key_detach(struct pf_state *s, int idx) { struct pf_state_key *sk = s->key[idx]; #ifdef INVARIANTS struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; PF_HASHROW_ASSERT(kh); #endif TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); s->key[idx] = NULL; if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { LIST_REMOVE(sk, entry); uma_zfree(V_pf_state_key_z, sk); } } static int pf_state_key_ctor(void *mem, int size, void *arg, int flags) { struct pf_state_key *sk = mem; bzero(sk, sizeof(struct pf_state_key_cmp)); TAILQ_INIT(&sk->states[PF_SK_WIRE]); TAILQ_INIT(&sk->states[PF_SK_STACK]); return (0); } struct pf_state_key * pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) { struct pf_state_key *sk; sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); if (sk == NULL) return (NULL); PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); sk->port[pd->sidx] = sport; sk->port[pd->didx] = dport; sk->proto = pd->proto; sk->af = pd->af; return (sk); } struct pf_state_key * pf_state_key_clone(struct pf_state_key *orig) { struct pf_state_key *sk; sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); if (sk == NULL) return (NULL); bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); return (sk); } int pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, struct pf_state_key *sks, struct pf_state *s) { struct pf_idhash *ih; struct pf_state *cur; int error; KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), ("%s: sks not pristine", __func__)); KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), ("%s: skw not pristine", __func__)); KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); s->kif = kif; if (s->id == 0 && s->creatorid == 0) { /* XXX: should be atomic, but probability of collision low */ if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) V_pf_stateid[curcpu] = 1; s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; s->id = htobe64(s->id); s->creatorid = V_pf_status.hostid; } /* Returns with ID locked on success. */ if ((error = pf_state_key_attach(skw, sks, s)) != 0) return (error); ih = &V_pf_idhash[PF_IDHASH(s)]; PF_HASHROW_ASSERT(ih); LIST_FOREACH(cur, &ih->states, entry) if (cur->id == s->id && cur->creatorid == s->creatorid) break; if (cur != NULL) { PF_HASHROW_UNLOCK(ih); if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: state ID collision: " "id: %016llx creatorid: %08x\n", (unsigned long long)be64toh(s->id), ntohl(s->creatorid)); } pf_detach_state(s); return (EEXIST); } LIST_INSERT_HEAD(&ih->states, s, entry); /* One for keys, one for ID hash. */ refcount_init(&s->refs, 2); counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1); if (pfsync_insert_state_ptr != NULL) pfsync_insert_state_ptr(s); /* Returns locked. */ return (0); } /* * Find state by ID: returns with locked row on success. */ struct pf_state * pf_find_state_byid(uint64_t id, uint32_t creatorid) { struct pf_idhash *ih; struct pf_state *s; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) if (s->id == id && s->creatorid == creatorid) break; if (s == NULL) PF_HASHROW_UNLOCK(ih); return (s); } /* * Find state by key. * Returns with ID hash slot locked on success. */ static struct pf_state * pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir) { struct pf_keyhash *kh; struct pf_state_key *sk; struct pf_state *s; int idx; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; PF_HASHROW_LOCK(kh); LIST_FOREACH(sk, &kh->keys, entry) if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) break; if (sk == NULL) { PF_HASHROW_UNLOCK(kh); return (NULL); } idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); /* List is sorted, if-bound states before floating ones. */ TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) if (s->kif == V_pfi_all || s->kif == kif) { PF_STATE_LOCK(s); PF_HASHROW_UNLOCK(kh); if (s->timeout >= PFTM_MAX) { /* * State is either being processed by * pf_unlink_state() in an other thread, or * is scheduled for immediate expiry. */ PF_STATE_UNLOCK(s); return (NULL); } return (s); } PF_HASHROW_UNLOCK(kh); return (NULL); } struct pf_state * pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) { struct pf_keyhash *kh; struct pf_state_key *sk; struct pf_state *s, *ret = NULL; int idx, inout = 0; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; PF_HASHROW_LOCK(kh); LIST_FOREACH(sk, &kh->keys, entry) if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) break; if (sk == NULL) { PF_HASHROW_UNLOCK(kh); return (NULL); } switch (dir) { case PF_IN: idx = PF_SK_WIRE; break; case PF_OUT: idx = PF_SK_STACK; break; case PF_INOUT: idx = PF_SK_WIRE; inout = 1; break; default: panic("%s: dir %u", __func__, dir); } second_run: TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { if (more == NULL) { PF_HASHROW_UNLOCK(kh); return (s); } if (ret) (*more)++; else ret = s; } if (inout == 1) { inout = 0; idx = PF_SK_STACK; goto second_run; } PF_HASHROW_UNLOCK(kh); return (ret); } /* END state table stuff */ static void pf_send(struct pf_send_entry *pfse) { PF_SENDQ_LOCK(); STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); PF_SENDQ_UNLOCK(); swi_sched(V_pf_swi_cookie, 0); } void pf_intr(void *v) { struct pf_send_head queue; struct pf_send_entry *pfse, *next; CURVNET_SET((struct vnet *)v); PF_SENDQ_LOCK(); queue = V_pf_sendqueue; STAILQ_INIT(&V_pf_sendqueue); PF_SENDQ_UNLOCK(); STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { switch (pfse->pfse_type) { #ifdef INET case PFSE_IP: ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); break; case PFSE_ICMP: icmp_error(pfse->pfse_m, pfse->icmpopts.type, pfse->icmpopts.code, 0, pfse->icmpopts.mtu); break; #endif /* INET */ #ifdef INET6 case PFSE_IP6: ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, NULL); break; case PFSE_ICMP6: icmp6_error(pfse->pfse_m, pfse->icmpopts.type, pfse->icmpopts.code, pfse->icmpopts.mtu); break; #endif /* INET6 */ default: panic("%s: unknown type", __func__); } free(pfse, M_PFTEMP); } CURVNET_RESTORE(); } void pf_purge_thread(void *unused __unused) { VNET_ITERATOR_DECL(vnet_iter); sx_xlock(&pf_end_lock); while (pf_end_threads == 0) { sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); /* Wait until V_pf_default_rule is initialized. */ if (V_pf_vnet_active == 0) { CURVNET_RESTORE(); continue; } /* * Process 1/interval fraction of the state * table every run. */ V_pf_purge_idx = pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); /* * Purge other expired types every * PFTM_INTERVAL seconds. */ if (V_pf_purge_idx == 0) { /* * Order is important: * - states and src nodes reference rules * - states and rules reference kifs */ pf_purge_expired_fragments(); pf_purge_expired_src_nodes(); pf_purge_unlinked_rules(); pfi_kif_purge(); } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); } pf_end_threads++; sx_xunlock(&pf_end_lock); kproc_exit(0); } void pf_unload_vnet_purge(void) { /* * To cleanse up all kifs and rules we need * two runs: first one clears reference flags, * then pf_purge_expired_states() doesn't * raise them, and then second run frees. */ pf_purge_unlinked_rules(); pfi_kif_purge(); /* * Now purge everything. */ pf_purge_expired_states(0, pf_hashmask); pf_purge_fragments(UINT_MAX); pf_purge_expired_src_nodes(); /* * Now all kifs & rules should be unreferenced, * thus should be successfully freed. */ pf_purge_unlinked_rules(); pfi_kif_purge(); } u_int32_t pf_state_expires(const struct pf_state *state) { u_int32_t timeout; u_int32_t start; u_int32_t end; u_int32_t states; /* handle all PFTM_* > PFTM_MAX here */ if (state->timeout == PFTM_PURGE) return (time_uptime); KASSERT(state->timeout != PFTM_UNLINKED, ("pf_state_expires: timeout == PFTM_UNLINKED")); KASSERT((state->timeout < PFTM_MAX), ("pf_state_expires: timeout > PFTM_MAX")); timeout = state->rule.ptr->timeout[state->timeout]; if (!timeout) timeout = V_pf_default_rule.timeout[state->timeout]; start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; if (start) { end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; states = counter_u64_fetch(state->rule.ptr->states_cur); } else { start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; states = V_pf_status.states; } if (end && states > start && start < end) { if (states < end) return (state->expire + timeout * (end - states) / (end - start)); else return (time_uptime); } return (state->expire + timeout); } void pf_purge_expired_src_nodes() { struct pf_src_node_list freelist; struct pf_srchash *sh; struct pf_src_node *cur, *next; int i; LIST_INIT(&freelist); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) if (cur->states == 0 && cur->expire <= time_uptime) { pf_unlink_src_node(cur); LIST_INSERT_HEAD(&freelist, cur, entry); } else if (cur->rule.ptr != NULL) cur->rule.ptr->rule_flag |= PFRULE_REFS; PF_HASHROW_UNLOCK(sh); } pf_free_src_nodes(&freelist); V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); } static void pf_src_tree_remove_state(struct pf_state *s) { struct pf_src_node *sn; struct pf_srchash *sh; uint32_t timeout; timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? s->rule.ptr->timeout[PFTM_SRC_NODE] : V_pf_default_rule.timeout[PFTM_SRC_NODE]; if (s->src_node != NULL) { sn = s->src_node; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (s->src.tcp_est) --sn->conn; if (--sn->states == 0) sn->expire = time_uptime + timeout; PF_HASHROW_UNLOCK(sh); } if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { sn = s->nat_src_node; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (--sn->states == 0) sn->expire = time_uptime + timeout; PF_HASHROW_UNLOCK(sh); } s->src_node = s->nat_src_node = NULL; } /* * Unlink and potentilly free a state. Function may be * called with ID hash row locked, but always returns * unlocked, since it needs to go through key hash locking. */ int pf_unlink_state(struct pf_state *s, u_int flags) { struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; if ((flags & PF_ENTER_LOCKED) == 0) PF_HASHROW_LOCK(ih); else PF_HASHROW_ASSERT(ih); if (s->timeout == PFTM_UNLINKED) { /* * State is being processed * by pf_unlink_state() in * an other thread. */ PF_HASHROW_UNLOCK(ih); return (0); /* XXXGL: undefined actually */ } if (s->src.state == PF_TCPS_PROXY_DST) { /* XXX wire key the right one? */ pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, &s->key[PF_SK_WIRE]->addr[1], &s->key[PF_SK_WIRE]->addr[0], s->key[PF_SK_WIRE]->port[1], s->key[PF_SK_WIRE]->port[0], s->src.seqhi, s->src.seqlo + 1, TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); } LIST_REMOVE(s, entry); pf_src_tree_remove_state(s); if (pfsync_delete_state_ptr != NULL) pfsync_delete_state_ptr(s); STATE_DEC_COUNTERS(s); s->timeout = PFTM_UNLINKED; PF_HASHROW_UNLOCK(ih); pf_detach_state(s); /* pf_state_insert() initialises refs to 2, so we can never release the * last reference here, only in pf_release_state(). */ (void)refcount_release(&s->refs); return (pf_release_state(s)); } void pf_free_state(struct pf_state *cur) { KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, cur->timeout)); pf_normalize_tcp_cleanup(cur); uma_zfree(V_pf_state_z, cur); counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); } /* * Called only from pf_purge_thread(), thus serialized. */ static u_int pf_purge_expired_states(u_int i, int maxcheck) { struct pf_idhash *ih; struct pf_state *s; V_pf_status.states = uma_zone_get_cur(V_pf_state_z); /* * Go through hash and unlink states that expire now. */ while (maxcheck > 0) { ih = &V_pf_idhash[i]; relock: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (pf_state_expires(s) <= time_uptime) { V_pf_status.states -= pf_unlink_state(s, PF_ENTER_LOCKED); goto relock; } s->rule.ptr->rule_flag |= PFRULE_REFS; if (s->nat_rule.ptr != NULL) s->nat_rule.ptr->rule_flag |= PFRULE_REFS; if (s->anchor.ptr != NULL) s->anchor.ptr->rule_flag |= PFRULE_REFS; s->kif->pfik_flags |= PFI_IFLAG_REFS; if (s->rt_kif) s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; } PF_HASHROW_UNLOCK(ih); /* Return when we hit end of hash. */ if (++i > pf_hashmask) { V_pf_status.states = uma_zone_get_cur(V_pf_state_z); return (0); } maxcheck--; } V_pf_status.states = uma_zone_get_cur(V_pf_state_z); return (i); } static void pf_purge_unlinked_rules() { struct pf_rulequeue tmpq; struct pf_rule *r, *r1; /* * If we have overloading task pending, then we'd * better skip purging this time. There is a tiny * probability that overloading task references * an already unlinked rule. */ PF_OVERLOADQ_LOCK(); if (!SLIST_EMPTY(&V_pf_overloadqueue)) { PF_OVERLOADQ_UNLOCK(); return; } PF_OVERLOADQ_UNLOCK(); /* * Do naive mark-and-sweep garbage collecting of old rules. * Reference flag is raised by pf_purge_expired_states() * and pf_purge_expired_src_nodes(). * * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, * use a temporary queue. */ TAILQ_INIT(&tmpq); PF_UNLNKDRULES_LOCK(); TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { if (!(r->rule_flag & PFRULE_REFS)) { TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); TAILQ_INSERT_TAIL(&tmpq, r, entries); } else r->rule_flag &= ~PFRULE_REFS; } PF_UNLNKDRULES_UNLOCK(); if (!TAILQ_EMPTY(&tmpq)) { PF_RULES_WLOCK(); TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { TAILQ_REMOVE(&tmpq, r, entries); pf_free_rule(r); } PF_RULES_WUNLOCK(); } } void pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: { u_int32_t a = ntohl(addr->addr32[0]); printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, (a>>8)&255, a&255); if (p) { p = ntohs(p); printf(":%u", p); } break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { u_int16_t b; u_int8_t i, curstart, curend, maxstart, maxend; curstart = curend = maxstart = maxend = 255; for (i = 0; i < 8; i++) { if (!addr->addr16[i]) { if (curstart == 255) curstart = i; curend = i; } else { if ((curend - curstart) > (maxend - maxstart)) { maxstart = curstart; maxend = curend; } curstart = curend = 255; } } if ((curend - curstart) > (maxend - maxstart)) { maxstart = curstart; maxend = curend; } for (i = 0; i < 8; i++) { if (i >= maxstart && i <= maxend) { if (i == 0) printf(":"); if (i == maxend) printf(":"); } else { b = ntohs(addr->addr16[i]); printf("%x", b); if (i < 7) printf(":"); } } if (p) { p = ntohs(p); printf("[%u]", p); } break; } #endif /* INET6 */ } } void pf_print_state(struct pf_state *s) { pf_print_state_parts(s, NULL, NULL); } static void pf_print_state_parts(struct pf_state *s, struct pf_state_key *skwp, struct pf_state_key *sksp) { struct pf_state_key *skw, *sks; u_int8_t proto, dir; /* Do our best to fill these, but they're skipped if NULL */ skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); proto = skw ? skw->proto : (sks ? sks->proto : 0); dir = s ? s->direction : 0; switch (proto) { case IPPROTO_IPV4: printf("IPv4"); break; case IPPROTO_IPV6: printf("IPv6"); break; case IPPROTO_TCP: printf("TCP"); break; case IPPROTO_UDP: printf("UDP"); break; case IPPROTO_ICMP: printf("ICMP"); break; case IPPROTO_ICMPV6: printf("ICMPv6"); break; default: printf("%u", proto); break; } switch (dir) { case PF_IN: printf(" in"); break; case PF_OUT: printf(" out"); break; } if (skw) { printf(" wire: "); pf_print_host(&skw->addr[0], skw->port[0], skw->af); printf(" "); pf_print_host(&skw->addr[1], skw->port[1], skw->af); } if (sks) { printf(" stack: "); if (sks != skw) { pf_print_host(&sks->addr[0], sks->port[0], sks->af); printf(" "); pf_print_host(&sks->addr[1], sks->port[1], sks->af); } else printf("-"); } if (s) { if (proto == IPPROTO_TCP) { printf(" [lo=%u high=%u win=%u modulator=%u", s->src.seqlo, s->src.seqhi, s->src.max_win, s->src.seqdiff); if (s->src.wscale && s->dst.wscale) printf(" wscale=%u", s->src.wscale & PF_WSCALE_MASK); printf("]"); printf(" [lo=%u high=%u win=%u modulator=%u", s->dst.seqlo, s->dst.seqhi, s->dst.max_win, s->dst.seqdiff); if (s->src.wscale && s->dst.wscale) printf(" wscale=%u", s->dst.wscale & PF_WSCALE_MASK); printf("]"); } printf(" %u:%u", s->src.state, s->dst.state); } } void pf_print_flags(u_int8_t f) { if (f) printf(" "); if (f & TH_FIN) printf("F"); if (f & TH_SYN) printf("S"); if (f & TH_RST) printf("R"); if (f & TH_PUSH) printf("P"); if (f & TH_ACK) printf("A"); if (f & TH_URG) printf("U"); if (f & TH_ECE) printf("E"); if (f & TH_CWR) printf("W"); } #define PF_SET_SKIP_STEPS(i) \ do { \ while (head[i] != cur) { \ head[i]->skip[i].ptr = cur; \ head[i] = TAILQ_NEXT(head[i], entries); \ } \ } while (0) void pf_calc_skip_steps(struct pf_rulequeue *rules) { struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; int i; cur = TAILQ_FIRST(rules); prev = cur; for (i = 0; i < PF_SKIP_COUNT; ++i) head[i] = cur; while (cur != NULL) { if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) PF_SET_SKIP_STEPS(PF_SKIP_IFP); if (cur->direction != prev->direction) PF_SET_SKIP_STEPS(PF_SKIP_DIR); if (cur->af != prev->af) PF_SET_SKIP_STEPS(PF_SKIP_AF); if (cur->proto != prev->proto) PF_SET_SKIP_STEPS(PF_SKIP_PROTO); if (cur->src.neg != prev->src.neg || pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); if (cur->src.port[0] != prev->src.port[0] || cur->src.port[1] != prev->src.port[1] || cur->src.port_op != prev->src.port_op) PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); if (cur->dst.neg != prev->dst.neg || pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); if (cur->dst.port[0] != prev->dst.port[0] || cur->dst.port[1] != prev->dst.port[1] || cur->dst.port_op != prev->dst.port_op) PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); prev = cur; cur = TAILQ_NEXT(cur, entries); } for (i = 0; i < PF_SKIP_COUNT; ++i) PF_SET_SKIP_STEPS(i); } static int pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) { if (aw1->type != aw2->type) return (1); switch (aw1->type) { case PF_ADDR_ADDRMASK: case PF_ADDR_RANGE: if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) return (1); if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) return (1); return (0); case PF_ADDR_DYNIFTL: return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); case PF_ADDR_NOROUTE: case PF_ADDR_URPFFAILED: return (0); case PF_ADDR_TABLE: return (aw1->p.tbl != aw2->p.tbl); default: printf("invalid address type: %d\n", aw1->type); return (1); } } /** * Checksum updates are a little complicated because the checksum in the TCP/UDP * header isn't always a full checksum. In some cases (i.e. output) it's a * pseudo-header checksum, which is a partial checksum over src/dst IP * addresses, protocol number and length. * * That means we have the following cases: * * Input or forwarding: we don't have TSO, the checksum fields are full * checksums, we need to update the checksum whenever we change anything. * * Output (i.e. the checksum is a pseudo-header checksum): * x The field being updated is src/dst address or affects the length of * the packet. We need to update the pseudo-header checksum (note that this * checksum is not ones' complement). * x Some other field is being modified (e.g. src/dst port numbers): We * don't have to update anything. **/ u_int16_t pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) { u_int32_t l; if (udp && !cksum) return (0x0000); l = cksum + old - new; l = (l >> 16) + (l & 65535); l = l & 65535; if (udp && !l) return (0xFFFF); return (l); } u_int16_t pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) { if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) return (cksum); return (pf_cksum_fixup(cksum, old, new, udp)); } static void pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) { struct pf_addr ao; u_int16_t po = *p; PF_ACPY(&ao, a, af); PF_ACPY(a, an, af); if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) *pc = ~*pc; *p = pn; switch (af) { #ifdef INET case AF_INET: *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, ao.addr16[0], an->addr16[0], 0), ao.addr16[1], an->addr16[1], 0); *p = pn; *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u); *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); break; #endif /* INET */ #ifdef INET6 case AF_INET6: *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*pc, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u), ao.addr16[2], an->addr16[2], u), ao.addr16[3], an->addr16[3], u), ao.addr16[4], an->addr16[4], u), ao.addr16[5], an->addr16[5], u), ao.addr16[6], an->addr16[6], u), ao.addr16[7], an->addr16[7], u); *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); break; #endif /* INET6 */ } if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) { *pc = ~*pc; if (! *pc) *pc = 0xffff; } } /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ void pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) { u_int32_t ao; memcpy(&ao, a, sizeof(ao)); memcpy(a, &an, sizeof(u_int32_t)); *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), ao % 65536, an % 65536, u); } void pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) { u_int32_t ao; memcpy(&ao, a, sizeof(ao)); memcpy(a, &an, sizeof(u_int32_t)); *c = pf_proto_cksum_fixup(m, pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), ao % 65536, an % 65536, udp); } #ifdef INET6 static void pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) { struct pf_addr ao; PF_ACPY(&ao, a, AF_INET6); PF_ACPY(a, an, AF_INET6); *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*c, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u), ao.addr16[2], an->addr16[2], u), ao.addr16[3], an->addr16[3], u), ao.addr16[4], an->addr16[4], u), ao.addr16[5], an->addr16[5], u), ao.addr16[6], an->addr16[6], u), ao.addr16[7], an->addr16[7], u); } #endif /* INET6 */ static void pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) { struct pf_addr oia, ooa; PF_ACPY(&oia, ia, af); if (oa) PF_ACPY(&ooa, oa, af); /* Change inner protocol port, fix inner protocol checksum. */ if (ip != NULL) { u_int16_t oip = *ip; u_int32_t opc; if (pc != NULL) opc = *pc; *ip = np; if (pc != NULL) *pc = pf_cksum_fixup(*pc, oip, *ip, u); *ic = pf_cksum_fixup(*ic, oip, *ip, 0); if (pc != NULL) *ic = pf_cksum_fixup(*ic, opc, *pc, 0); } /* Change inner ip address, fix inner ip and icmp checksums. */ PF_ACPY(ia, na, af); switch (af) { #ifdef INET case AF_INET: { u_int32_t oh2c = *h2c; *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, oia.addr16[0], ia->addr16[0], 0), oia.addr16[1], ia->addr16[1], 0); *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, oia.addr16[0], ia->addr16[0], 0), oia.addr16[1], ia->addr16[1], 0); *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); break; } #endif /* INET */ #ifdef INET6 case AF_INET6: *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*ic, oia.addr16[0], ia->addr16[0], u), oia.addr16[1], ia->addr16[1], u), oia.addr16[2], ia->addr16[2], u), oia.addr16[3], ia->addr16[3], u), oia.addr16[4], ia->addr16[4], u), oia.addr16[5], ia->addr16[5], u), oia.addr16[6], ia->addr16[6], u), oia.addr16[7], ia->addr16[7], u); break; #endif /* INET6 */ } /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ if (oa) { PF_ACPY(oa, na, af); switch (af) { #ifdef INET case AF_INET: *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, ooa.addr16[0], oa->addr16[0], 0), ooa.addr16[1], oa->addr16[1], 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*ic, ooa.addr16[0], oa->addr16[0], u), ooa.addr16[1], oa->addr16[1], u), ooa.addr16[2], oa->addr16[2], u), ooa.addr16[3], oa->addr16[3], u), ooa.addr16[4], oa->addr16[4], u), ooa.addr16[5], oa->addr16[5], u), ooa.addr16[6], oa->addr16[6], u), ooa.addr16[7], oa->addr16[7], u); break; #endif /* INET6 */ } } } /* * Need to modulate the sequence numbers in the TCP SACK option * (credits to Krzysztof Pfaff for report and patch) */ static int pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, struct tcphdr *th, struct pf_state_peer *dst) { int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; u_int8_t opts[TCP_MAXOLEN], *opt = opts; int copyback = 0, i, olen; struct sackblk sack; #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) if (hlen < TCPOLEN_SACKLEN || !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) return 0; while (hlen >= TCPOLEN_SACKLEN) { olen = opt[1]; switch (*opt) { case TCPOPT_EOL: /* FALLTHROUGH */ case TCPOPT_NOP: opt++; hlen--; break; case TCPOPT_SACK: if (olen > hlen) olen = hlen; if (olen >= TCPOLEN_SACKLEN) { for (i = 2; i + TCPOLEN_SACK <= olen; i += TCPOLEN_SACK) { memcpy(&sack, &opt[i], sizeof(sack)); pf_change_proto_a(m, &sack.start, &th->th_sum, htonl(ntohl(sack.start) - dst->seqdiff), 0); pf_change_proto_a(m, &sack.end, &th->th_sum, htonl(ntohl(sack.end) - dst->seqdiff), 0); memcpy(&opt[i], &sack, sizeof(sack)); } copyback = 1; } /* FALLTHROUGH */ default: if (olen < 2) olen = 2; hlen -= olen; opt += olen; } } if (copyback) m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); return (copyback); } static void pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af, const struct pf_addr *saddr, const struct pf_addr *daddr, u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, u_int16_t rtag, struct ifnet *ifp) { struct pf_send_entry *pfse; struct mbuf *m; int len, tlen; #ifdef INET struct ip *h = NULL; #endif /* INET */ #ifdef INET6 struct ip6_hdr *h6 = NULL; #endif /* INET6 */ struct tcphdr *th; char *opt; struct pf_mtag *pf_mtag; len = 0; th = NULL; /* maximum segment size tcp option */ tlen = sizeof(struct tcphdr); if (mss) tlen += 4; switch (af) { #ifdef INET case AF_INET: len = sizeof(struct ip) + tlen; break; #endif /* INET */ #ifdef INET6 case AF_INET6: len = sizeof(struct ip6_hdr) + tlen; break; #endif /* INET6 */ default: panic("%s: unsupported af %d", __func__, af); } /* Allocate outgoing queue entry, mbuf and mbuf tag. */ pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); if (pfse == NULL) return; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { free(pfse, M_PFTEMP); return; } #ifdef MAC mac_netinet_firewall_send(m); #endif if ((pf_mtag = pf_get_mtag(m)) == NULL) { free(pfse, M_PFTEMP); m_freem(m); return; } if (tag) m->m_flags |= M_SKIP_FIREWALL; pf_mtag->tag = rtag; if (r != NULL && r->rtableid >= 0) M_SETFIB(m, r->rtableid); #ifdef ALTQ if (r != NULL && r->qid) { pf_mtag->qid = r->qid; /* add hints for ecn */ pf_mtag->hdr = mtod(m, struct ip *); } #endif /* ALTQ */ m->m_data += max_linkhdr; m->m_pkthdr.len = m->m_len = len; m->m_pkthdr.rcvif = NULL; bzero(m->m_data, len); switch (af) { #ifdef INET case AF_INET: h = mtod(m, struct ip *); /* IP header fields included in the TCP checksum */ h->ip_p = IPPROTO_TCP; h->ip_len = htons(tlen); h->ip_src.s_addr = saddr->v4.s_addr; h->ip_dst.s_addr = daddr->v4.s_addr; th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); break; #endif /* INET */ #ifdef INET6 case AF_INET6: h6 = mtod(m, struct ip6_hdr *); /* IP header fields included in the TCP checksum */ h6->ip6_nxt = IPPROTO_TCP; h6->ip6_plen = htons(tlen); memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); break; #endif /* INET6 */ } /* TCP header */ th->th_sport = sport; th->th_dport = dport; th->th_seq = htonl(seq); th->th_ack = htonl(ack); th->th_off = tlen >> 2; th->th_flags = flags; th->th_win = htons(win); if (mss) { opt = (char *)(th + 1); opt[0] = TCPOPT_MAXSEG; opt[1] = 4; HTONS(mss); bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); } switch (af) { #ifdef INET case AF_INET: /* TCP checksum */ th->th_sum = in_cksum(m, len); /* Finish the IP header */ h->ip_v = 4; h->ip_hl = sizeof(*h) >> 2; h->ip_tos = IPTOS_LOWDELAY; h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); h->ip_len = htons(len); h->ip_ttl = ttl ? ttl : V_ip_defttl; h->ip_sum = 0; pfse->pfse_type = PFSE_IP; break; #endif /* INET */ #ifdef INET6 case AF_INET6: /* TCP checksum */ th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen); h6->ip6_vfc |= IPV6_VERSION; h6->ip6_hlim = IPV6_DEFHLIM; pfse->pfse_type = PFSE_IP6; break; #endif /* INET6 */ } pfse->pfse_m = m; pf_send(pfse); } static void pf_return(struct pf_rule *r, struct pf_rule *nr, struct pf_pdesc *pd, struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, struct pfi_kif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, u_short *reason) { struct pf_addr * const saddr = pd->src; struct pf_addr * const daddr = pd->dst; sa_family_t af = pd->af; /* undo NAT changes, if they have taken place */ if (nr != NULL) { PF_ACPY(saddr, &sk->addr[pd->sidx], af); PF_ACPY(daddr, &sk->addr[pd->didx], af); if (pd->sport) *pd->sport = sk->port[pd->sidx]; if (pd->dport) *pd->dport = sk->port[pd->didx]; if (pd->proto_sum) *pd->proto_sum = bproto_sum; if (pd->ip_sum) *pd->ip_sum = bip_sum; m_copyback(m, off, hdrlen, pd->hdr.any); } if (pd->proto == IPPROTO_TCP && ((r->rule_flag & PFRULE_RETURNRST) || (r->rule_flag & PFRULE_RETURN)) && !(th->th_flags & TH_RST)) { u_int32_t ack = ntohl(th->th_seq) + pd->p_len; int len = 0; #ifdef INET struct ip *h4; #endif #ifdef INET6 struct ip6_hdr *h6; #endif switch (af) { #ifdef INET case AF_INET: h4 = mtod(m, struct ip *); len = ntohs(h4->ip_len) - off; break; #endif #ifdef INET6 case AF_INET6: h6 = mtod(m, struct ip6_hdr *); len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); break; #endif } if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) REASON_SET(reason, PFRES_PROTCKSUM); else { if (th->th_flags & TH_SYN) ack++; if (th->th_flags & TH_FIN) ack++; pf_send_tcp(m, r, af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, r->return_ttl, 1, 0, kif->pfik_ifp); } } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && r->return_icmp) pf_send_icmp(m, r->return_icmp >> 8, r->return_icmp & 255, af, r); else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && r->return_icmp6) pf_send_icmp(m, r->return_icmp6 >> 8, r->return_icmp6 & 255, af, r); } static int pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio) { struct m_tag *mtag; KASSERT(prio <= PF_PRIO_MAX, ("%s with invalid pcp", __func__)); mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL); if (mtag == NULL) { mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT, sizeof(uint8_t), M_NOWAIT); if (mtag == NULL) return (ENOMEM); m_tag_prepend(m, mtag); } *(uint8_t *)(mtag + 1) = prio; return (0); } static int pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) { struct m_tag *mtag; u_int8_t mpcp; mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); if (mtag == NULL) return (0); if (prio == PF_PRIO_ZERO) prio = 0; mpcp = *(uint8_t *)(mtag + 1); return (mpcp == prio); } static void pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, struct pf_rule *r) { struct pf_send_entry *pfse; struct mbuf *m0; struct pf_mtag *pf_mtag; /* Allocate outgoing queue entry, mbuf and mbuf tag. */ pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); if (pfse == NULL) return; if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { free(pfse, M_PFTEMP); return; } if ((pf_mtag = pf_get_mtag(m0)) == NULL) { free(pfse, M_PFTEMP); return; } /* XXX: revisit */ m0->m_flags |= M_SKIP_FIREWALL; if (r->rtableid >= 0) M_SETFIB(m0, r->rtableid); #ifdef ALTQ if (r->qid) { pf_mtag->qid = r->qid; /* add hints for ecn */ pf_mtag->hdr = mtod(m0, struct ip *); } #endif /* ALTQ */ switch (af) { #ifdef INET case AF_INET: pfse->pfse_type = PFSE_ICMP; break; #endif /* INET */ #ifdef INET6 case AF_INET6: pfse->pfse_type = PFSE_ICMP6; break; #endif /* INET6 */ } pfse->pfse_m = m0; pfse->icmpopts.type = type; pfse->icmpopts.code = code; pf_send(pfse); } /* * Return 1 if the addresses a and b match (with mask m), otherwise return 0. * If n is 0, they match if they are equal. If n is != 0, they match if they * are different. */ int pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, struct pf_addr *b, sa_family_t af) { int match = 0; switch (af) { #ifdef INET case AF_INET: if ((a->addr32[0] & m->addr32[0]) == (b->addr32[0] & m->addr32[0])) match++; break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (((a->addr32[0] & m->addr32[0]) == (b->addr32[0] & m->addr32[0])) && ((a->addr32[1] & m->addr32[1]) == (b->addr32[1] & m->addr32[1])) && ((a->addr32[2] & m->addr32[2]) == (b->addr32[2] & m->addr32[2])) && ((a->addr32[3] & m->addr32[3]) == (b->addr32[3] & m->addr32[3]))) match++; break; #endif /* INET6 */ } if (match) { if (n) return (0); else return (1); } else { if (n) return (1); else return (0); } } /* * Return 1 if b <= a <= e, otherwise return 0. */ int pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, struct pf_addr *a, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) return (0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: { int i; /* check a >= b */ for (i = 0; i < 4; ++i) if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) break; else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) return (0); /* check a <= e */ for (i = 0; i < 4; ++i) if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) break; else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) return (0); break; } #endif /* INET6 */ } return (1); } static int pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) { switch (op) { case PF_OP_IRG: return ((p > a1) && (p < a2)); case PF_OP_XRG: return ((p < a1) || (p > a2)); case PF_OP_RRG: return ((p >= a1) && (p <= a2)); case PF_OP_EQ: return (p == a1); case PF_OP_NE: return (p != a1); case PF_OP_LT: return (p < a1); case PF_OP_LE: return (p <= a1); case PF_OP_GT: return (p > a1); case PF_OP_GE: return (p >= a1); } return (0); /* never reached */ } int pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) { NTOHS(a1); NTOHS(a2); NTOHS(p); return (pf_match(op, a1, a2, p)); } static int pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) { if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) return (0); return (pf_match(op, a1, a2, u)); } static int pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) { if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) return (0); return (pf_match(op, a1, a2, g)); } int pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag) { if (*tag == -1) *tag = mtag; return ((!r->match_tag_not && r->match_tag == *tag) || (r->match_tag_not && r->match_tag != *tag)); } int pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) { KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) return (ENOMEM); pd->pf_mtag->tag = tag; return (0); } #define PF_ANCHOR_STACKSIZE 32 struct pf_anchor_stackframe { struct pf_ruleset *rs; struct pf_rule *r; /* XXX: + match bit */ struct pf_anchor *child; }; /* * XXX: We rely on malloc(9) returning pointer aligned addresses. */ #define PF_ANCHORSTACK_MATCH 0x00000001 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) #define PF_ANCHOR_RULE(f) (struct pf_rule *) \ ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ } while (0) void pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth, struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, int *match) { struct pf_anchor_stackframe *f; PF_RULES_RASSERT(); if (match) *match = 0; if (*depth >= PF_ANCHOR_STACKSIZE) { printf("%s: anchor stack overflow on %s\n", __func__, (*r)->anchor->name); *r = TAILQ_NEXT(*r, entries); return; } else if (*depth == 0 && a != NULL) *a = *r; f = stack + (*depth)++; f->rs = *rs; f->r = *r; if ((*r)->anchor_wildcard) { struct pf_anchor_node *parent = &(*r)->anchor->children; if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) { *r = NULL; return; } *rs = &f->child->ruleset; } else { f->child = NULL; *rs = &(*r)->anchor->ruleset; } *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); } int pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth, struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, int *match) { struct pf_anchor_stackframe *f; struct pf_rule *fr; int quick = 0; PF_RULES_RASSERT(); do { if (*depth <= 0) break; f = stack + *depth - 1; fr = PF_ANCHOR_RULE(f); if (f->child != NULL) { struct pf_anchor_node *parent; /* * This block traverses through * a wildcard anchor. */ parent = &fr->anchor->children; if (match != NULL && *match) { /* * If any of "*" matched, then * "foo/ *" matched, mark frame * appropriately. */ PF_ANCHOR_SET_MATCH(f); *match = 0; } f->child = RB_NEXT(pf_anchor_node, parent, f->child); if (f->child != NULL) { *rs = &f->child->ruleset; *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); if (*r == NULL) continue; else break; } } (*depth)--; if (*depth == 0 && a != NULL) *a = NULL; *rs = f->rs; if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) quick = fr->quick; *r = TAILQ_NEXT(fr, entries); } while (*r == NULL); return (quick); } #ifdef INET6 void pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); break; #endif /* INET */ case AF_INET6: naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); break; } } void pf_addr_inc(struct pf_addr *addr, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); break; #endif /* INET */ case AF_INET6: if (addr->addr32[3] == 0xffffffff) { addr->addr32[3] = 0; if (addr->addr32[2] == 0xffffffff) { addr->addr32[2] = 0; if (addr->addr32[1] == 0xffffffff) { addr->addr32[1] = 0; addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); } else addr->addr32[1] = htonl(ntohl(addr->addr32[1]) + 1); } else addr->addr32[2] = htonl(ntohl(addr->addr32[2]) + 1); } else addr->addr32[3] = htonl(ntohl(addr->addr32[3]) + 1); break; } } #endif /* INET6 */ int pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) { struct pf_addr *saddr, *daddr; u_int16_t sport, dport; struct inpcbinfo *pi; struct inpcb *inp; pd->lookup.uid = UID_MAX; pd->lookup.gid = GID_MAX; switch (pd->proto) { case IPPROTO_TCP: if (pd->hdr.tcp == NULL) return (-1); sport = pd->hdr.tcp->th_sport; dport = pd->hdr.tcp->th_dport; pi = &V_tcbinfo; break; case IPPROTO_UDP: if (pd->hdr.udp == NULL) return (-1); sport = pd->hdr.udp->uh_sport; dport = pd->hdr.udp->uh_dport; pi = &V_udbinfo; break; default: return (-1); } if (direction == PF_IN) { saddr = pd->src; daddr = pd->dst; } else { u_int16_t p; p = sport; sport = dport; dport = p; saddr = pd->dst; daddr = pd->src; } switch (pd->af) { #ifdef INET case AF_INET: inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, dport, INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) { inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) return (-1); } break; #endif /* INET */ #ifdef INET6 case AF_INET6: inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, dport, INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) { inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) return (-1); } break; #endif /* INET6 */ default: return (-1); } INP_RLOCK_ASSERT(inp); pd->lookup.uid = inp->inp_cred->cr_uid; pd->lookup.gid = inp->inp_cred->cr_groups[0]; INP_RUNLOCK(inp); return (1); } static u_int8_t pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) { int hlen; u_int8_t hdr[60]; u_int8_t *opt, optlen; u_int8_t wscale = 0; hlen = th_off << 2; /* hlen <= sizeof(hdr) */ if (hlen <= sizeof(struct tcphdr)) return (0); if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) return (0); opt = hdr + sizeof(struct tcphdr); hlen -= sizeof(struct tcphdr); while (hlen >= 3) { switch (*opt) { case TCPOPT_EOL: case TCPOPT_NOP: ++opt; --hlen; break; case TCPOPT_WINDOW: wscale = opt[2]; if (wscale > TCP_MAX_WINSHIFT) wscale = TCP_MAX_WINSHIFT; wscale |= PF_WSCALE_FLAG; /* FALLTHROUGH */ default: optlen = opt[1]; if (optlen < 2) optlen = 2; hlen -= optlen; opt += optlen; break; } } return (wscale); } static u_int16_t pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) { int hlen; u_int8_t hdr[60]; u_int8_t *opt, optlen; u_int16_t mss = V_tcp_mssdflt; hlen = th_off << 2; /* hlen <= sizeof(hdr) */ if (hlen <= sizeof(struct tcphdr)) return (0); if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) return (0); opt = hdr + sizeof(struct tcphdr); hlen -= sizeof(struct tcphdr); while (hlen >= TCPOLEN_MAXSEG) { switch (*opt) { case TCPOPT_EOL: case TCPOPT_NOP: ++opt; --hlen; break; case TCPOPT_MAXSEG: bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); NTOHS(mss); /* FALLTHROUGH */ default: optlen = opt[1]; if (optlen < 2) optlen = 2; hlen -= optlen; opt += optlen; break; } } return (mss); } static u_int16_t pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) { #ifdef INET struct nhop4_basic nh4; #endif /* INET */ #ifdef INET6 struct nhop6_basic nh6; struct in6_addr dst6; uint32_t scopeid; #endif /* INET6 */ int hlen = 0; uint16_t mss = 0; switch (af) { #ifdef INET case AF_INET: hlen = sizeof(struct ip); if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0) mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr); break; #endif /* INET */ #ifdef INET6 case AF_INET6: hlen = sizeof(struct ip6_hdr); in6_splitscope(&addr->v6, &dst6, &scopeid); if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0) mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr); break; #endif /* INET6 */ } mss = max(V_tcp_mssdflt, mss); mss = min(mss, offer); mss = max(mss, 64); /* sanity - at least max opt space */ return (mss); } static u_int32_t pf_tcp_iss(struct pf_pdesc *pd) { MD5_CTX ctx; u_int32_t digest[4]; if (V_pf_tcp_secret_init == 0) { read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); MD5Init(&V_pf_tcp_secret_ctx); MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); V_pf_tcp_secret_init = 1; } ctx = V_pf_tcp_secret_ctx; MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); if (pd->af == AF_INET6) { MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); } else { MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); } MD5Final((u_char *)digest, &ctx); V_pf_tcp_iss_off += 4096; #define ISN_RANDOM_INCREMENT (4096 - 1) return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + V_pf_tcp_iss_off); #undef ISN_RANDOM_INCREMENT } static int pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp) { struct pf_rule *nr = NULL; struct pf_addr * const saddr = pd->src; struct pf_addr * const daddr = pd->dst; sa_family_t af = pd->af; struct pf_rule *r, *a = NULL; struct pf_ruleset *ruleset = NULL; struct pf_src_node *nsn = NULL; struct tcphdr *th = pd->hdr.tcp; struct pf_state_key *sk = NULL, *nk = NULL; u_short reason; int rewrite = 0, hdrlen = 0; int tag = -1, rtableid = -1; int asd = 0; int match = 0; int state_icmp = 0; u_int16_t sport = 0, dport = 0; u_int16_t bproto_sum = 0, bip_sum = 0; u_int8_t icmptype = 0, icmpcode = 0; struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; PF_RULES_RASSERT(); if (inp != NULL) { INP_LOCK_ASSERT(inp); pd->lookup.uid = inp->inp_cred->cr_uid; pd->lookup.gid = inp->inp_cred->cr_groups[0]; pd->lookup.done = 1; } switch (pd->proto) { case IPPROTO_TCP: sport = th->th_sport; dport = th->th_dport; hdrlen = sizeof(*th); break; case IPPROTO_UDP: sport = pd->hdr.udp->uh_sport; dport = pd->hdr.udp->uh_dport; hdrlen = sizeof(*pd->hdr.udp); break; #ifdef INET case IPPROTO_ICMP: if (pd->af != AF_INET) break; sport = dport = pd->hdr.icmp->icmp_id; hdrlen = sizeof(*pd->hdr.icmp); icmptype = pd->hdr.icmp->icmp_type; icmpcode = pd->hdr.icmp->icmp_code; if (icmptype == ICMP_UNREACH || icmptype == ICMP_SOURCEQUENCH || icmptype == ICMP_REDIRECT || icmptype == ICMP_TIMXCEED || icmptype == ICMP_PARAMPROB) state_icmp++; break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: if (af != AF_INET6) break; sport = dport = pd->hdr.icmp6->icmp6_id; hdrlen = sizeof(*pd->hdr.icmp6); icmptype = pd->hdr.icmp6->icmp6_type; icmpcode = pd->hdr.icmp6->icmp6_code; if (icmptype == ICMP6_DST_UNREACH || icmptype == ICMP6_PACKET_TOO_BIG || icmptype == ICMP6_TIME_EXCEEDED || icmptype == ICMP6_PARAM_PROB) state_icmp++; break; #endif /* INET6 */ default: sport = dport = hdrlen = 0; break; } r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); /* check packet for BINAT/NAT/RDR */ if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { KASSERT(sk != NULL, ("%s: null sk", __func__)); KASSERT(nk != NULL, ("%s: null nk", __func__)); if (pd->ip_sum) bip_sum = *pd->ip_sum; switch (pd->proto) { case IPPROTO_TCP: bproto_sum = th->th_sum; pd->proto_sum = &th->th_sum; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || nk->port[pd->sidx] != sport) { pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 0, af); pd->sport = &th->th_sport; sport = th->th_sport; } if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || nk->port[pd->didx] != dport) { pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], nk->port[pd->didx], 0, af); dport = th->th_dport; pd->dport = &th->th_dport; } rewrite++; break; case IPPROTO_UDP: bproto_sum = pd->hdr.udp->uh_sum; pd->proto_sum = &pd->hdr.udp->uh_sum; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || nk->port[pd->sidx] != sport) { pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport, pd->ip_sum, &pd->hdr.udp->uh_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 1, af); sport = pd->hdr.udp->uh_sport; pd->sport = &pd->hdr.udp->uh_sport; } if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || nk->port[pd->didx] != dport) { pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport, pd->ip_sum, &pd->hdr.udp->uh_sum, &nk->addr[pd->didx], nk->port[pd->didx], 1, af); dport = pd->hdr.udp->uh_dport; pd->dport = &pd->hdr.udp->uh_dport; } rewrite++; break; #ifdef INET case IPPROTO_ICMP: nk->port[0] = nk->port[1]; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); if (nk->port[1] != pd->hdr.icmp->icmp_id) { pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( pd->hdr.icmp->icmp_cksum, sport, nk->port[1], 0); pd->hdr.icmp->icmp_id = nk->port[1]; pd->sport = &pd->hdr.icmp->icmp_id; } m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: nk->port[0] = nk->port[1]; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->sidx], 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->didx], 0); rewrite++; break; #endif /* INET */ default: switch (af) { #ifdef INET case AF_INET: if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) PF_ACPY(saddr, &nk->addr[pd->sidx], af); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) PF_ACPY(saddr, &nk->addr[pd->didx], af); break; #endif /* INET */ } break; } if (nr->natpass) r = NULL; pd->nat_rule = nr; } while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != direction) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != af) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != pd->proto) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, saddr, af, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; /* tcp/udp only. port_op always 0 in other cases */ else if (r->src.port_op && !pf_match_port(r->src.port_op, r->src.port[0], r->src.port[1], sport)) r = r->skip[PF_SKIP_SRC_PORT].ptr; else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; /* tcp/udp only. port_op always 0 in other cases */ else if (r->dst.port_op && !pf_match_port(r->dst.port_op, r->dst.port[0], r->dst.port[1], dport)) r = r->skip[PF_SKIP_DST_PORT].ptr; /* icmp only. type always 0 in other cases */ else if (r->type && r->type != icmptype + 1) r = TAILQ_NEXT(r, entries); /* icmp only. type always 0 in other cases */ else if (r->code && r->code != icmpcode + 1) r = TAILQ_NEXT(r, entries); else if (r->tos && !(r->tos == pd->tos)) r = TAILQ_NEXT(r, entries); else if (r->rule_flag & PFRULE_FRAGMENT) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_TCP && (r->flagset & th->th_flags) != r->flags) r = TAILQ_NEXT(r, entries); /* tcp/udp only. uid.op always 0 in other cases */ else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = pf_socket_lookup(direction, pd, m), 1)) && !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], pd->lookup.uid)) r = TAILQ_NEXT(r, entries); /* tcp/udp only. gid.op always 0 in other cases */ else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = pf_socket_lookup(direction, pd, m), 1)) && !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], pd->lookup.gid)) r = TAILQ_NEXT(r, entries); else if (r->prio && !pf_match_ieee8021q_pcp(r->prio, m)) r = TAILQ_NEXT(r, entries); else if (r->prob && r->prob <= arc4random()) r = TAILQ_NEXT(r, entries); else if (r->match_tag && !pf_match_tag(m, r, &tag, pd->pf_mtag ? pd->pf_mtag->tag : 0)) r = TAILQ_NEXT(r, entries); else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto != IPPROTO_TCP || !pf_osfp_match( pf_osfp_fingerprint(pd, m, off, th), r->os_fingerprint))) r = TAILQ_NEXT(r, entries); else { if (r->tag) tag = r->tag; if (r->rtableid >= 0) rtableid = r->rtableid; if (r->anchor == NULL) { match = 1; *rm = r; *am = a; *rsm = ruleset; if ((*rm)->quick) break; r = TAILQ_NEXT(r, entries); } else pf_step_into_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match); } if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match)) break; } r = *rm; a = *am; ruleset = *rsm; REASON_SET(&reason, PFRES_MATCH); if (r->log || (nr != NULL && nr->log)) { if (rewrite) m_copyback(m, off, hdrlen, pd->hdr.any); PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, ruleset, pd, 1); } if ((r->action == PF_DROP) && ((r->rule_flag & PFRULE_RETURNRST) || (r->rule_flag & PFRULE_RETURNICMP) || (r->rule_flag & PFRULE_RETURN))) { pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, bip_sum, hdrlen, &reason); } if (r->action == PF_DROP) goto cleanup; if (tag > 0 && pf_tag_packet(m, pd, tag)) { REASON_SET(&reason, PFRES_MEMORY); goto cleanup; } if (rtableid >= 0) M_SETFIB(m, rtableid); if (!state_icmp && (r->keep_state || nr != NULL || (pd->flags & PFDESC_TCP_NORM))) { int action; action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, hdrlen); if (action != PF_PASS) { if (action == PF_DROP && (r->rule_flag & PFRULE_RETURN)) pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, bip_sum, hdrlen, &reason); return (action); } } else { if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); } /* copy back packet headers if we performed NAT operations */ if (rewrite) m_copyback(m, off, hdrlen, pd->hdr.any); if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && direction == PF_OUT && pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m)) /* * We want the state created, but we dont * want to send this in case a partner * firewall has to know about it to allow * replies through it. */ return (PF_DEFER); return (PF_PASS); cleanup: if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); return (PF_DROP); } static int pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk, struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) { struct pf_state *s = NULL; struct pf_src_node *sn = NULL; struct tcphdr *th = pd->hdr.tcp; u_int16_t mss = V_tcp_mssdflt; u_short reason; /* check maximums */ if (r->max_states && (counter_u64_fetch(r->states_cur) >= r->max_states)) { counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); REASON_SET(&reason, PFRES_MAXSTATES); goto csfailed; } /* src node for filter rule */ if ((r->rule_flag & PFRULE_SRCTRACK || r->rpool.opts & PF_POOL_STICKYADDR) && pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { REASON_SET(&reason, PFRES_SRCLIMIT); goto csfailed; } /* src node for translation rule */ if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { REASON_SET(&reason, PFRES_SRCLIMIT); goto csfailed; } s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); if (s == NULL) { REASON_SET(&reason, PFRES_MEMORY); goto csfailed; } s->rule.ptr = r; s->nat_rule.ptr = nr; s->anchor.ptr = a; STATE_INC_COUNTERS(s); if (r->allow_opts) s->state_flags |= PFSTATE_ALLOWOPTS; if (r->rule_flag & PFRULE_STATESLOPPY) s->state_flags |= PFSTATE_SLOPPY; s->log = r->log & PF_LOG_ALL; s->sync_state = PFSYNC_S_NONE; if (nr != NULL) s->log |= nr->log & PF_LOG_ALL; switch (pd->proto) { case IPPROTO_TCP: s->src.seqlo = ntohl(th->th_seq); s->src.seqhi = s->src.seqlo + pd->p_len + 1; if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && r->keep_state == PF_STATE_MODULATE) { /* Generate sequence number modulator */ if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 0) s->src.seqdiff = 1; pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(s->src.seqlo + s->src.seqdiff), 0); *rewrite = 1; } else s->src.seqdiff = 0; if (th->th_flags & TH_SYN) { s->src.seqhi++; s->src.wscale = pf_get_wscale(m, off, th->th_off, pd->af); } s->src.max_win = MAX(ntohs(th->th_win), 1); if (s->src.wscale & PF_WSCALE_MASK) { /* Remove scale factor from initial window */ int win = s->src.max_win; win += 1 << (s->src.wscale & PF_WSCALE_MASK); s->src.max_win = (win - 1) >> (s->src.wscale & PF_WSCALE_MASK); } if (th->th_flags & TH_FIN) s->src.seqhi++; s->dst.seqhi = 1; s->dst.max_win = 1; s->src.state = TCPS_SYN_SENT; s->dst.state = TCPS_CLOSED; s->timeout = PFTM_TCP_FIRST_PACKET; break; case IPPROTO_UDP: s->src.state = PFUDPS_SINGLE; s->dst.state = PFUDPS_NO_TRAFFIC; s->timeout = PFTM_UDP_FIRST_PACKET; break; case IPPROTO_ICMP: #ifdef INET6 case IPPROTO_ICMPV6: #endif s->timeout = PFTM_ICMP_FIRST_PACKET; break; default: s->src.state = PFOTHERS_SINGLE; s->dst.state = PFOTHERS_NO_TRAFFIC; s->timeout = PFTM_OTHER_FIRST_PACKET; } if (r->rt) { if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { REASON_SET(&reason, PFRES_MAPFAILED); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); goto csfailed; } s->rt_kif = r->rpool.cur->kif; } s->creation = time_uptime; s->expire = time_uptime; if (sn != NULL) s->src_node = sn; if (nsn != NULL) { /* XXX We only modify one side for now. */ PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); s->nat_src_node = nsn; } if (pd->proto == IPPROTO_TCP) { if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { REASON_SET(&reason, PFRES_MEMORY); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, &s->src, &s->dst, rewrite)) { /* This really shouldn't happen!!! */ DPFPRINTF(PF_DEBUG_URGENT, ("pf_normalize_tcp_stateful failed on first pkt")); pf_normalize_tcp_cleanup(s); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } } s->direction = pd->dir; /* * sk/nk could already been setup by pf_get_translation(). */ if (nr == NULL) { KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", __func__, nr, sk, nk)); sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); if (sk == NULL) goto csfailed; nk = sk; } else KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", __func__, nr, sk, nk)); /* Swap sk/nk for PF_OUT. */ if (pf_state_insert(BOUND_IFACE(r, kif), (pd->dir == PF_IN) ? sk : nk, (pd->dir == PF_IN) ? nk : sk, s)) { if (pd->proto == IPPROTO_TCP) pf_normalize_tcp_cleanup(s); REASON_SET(&reason, PFRES_STATEINS); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } else *sm = s; if (tag > 0) s->tag = tag; if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { s->src.state = PF_TCPS_PROXY_SRC; /* undo NAT changes, if they have taken place */ if (nr != NULL) { struct pf_state_key *skt = s->key[PF_SK_WIRE]; if (pd->dir == PF_OUT) skt = s->key[PF_SK_STACK]; PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); if (pd->sport) *pd->sport = skt->port[pd->sidx]; if (pd->dport) *pd->dport = skt->port[pd->didx]; if (pd->proto_sum) *pd->proto_sum = bproto_sum; if (pd->ip_sum) *pd->ip_sum = bip_sum; m_copyback(m, off, hdrlen, pd->hdr.any); } s->src.seqhi = htonl(arc4random()); /* Find mss option */ int rtid = M_GETFIB(m); mss = pf_get_mss(m, off, th->th_off, pd->af); mss = pf_calc_mss(pd->src, pd->af, rtid, mss); mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); s->src.mss = mss; pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); REASON_SET(&reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } return (PF_PASS); csfailed: if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); if (sn != NULL) { struct pf_srchash *sh; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (--sn->states == 0 && sn->expire == 0) { pf_unlink_src_node(sn); uma_zfree(V_pf_sources_z, sn); counter_u64_add( V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); } PF_HASHROW_UNLOCK(sh); } if (nsn != sn && nsn != NULL) { struct pf_srchash *sh; sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; PF_HASHROW_LOCK(sh); if (--nsn->states == 0 && nsn->expire == 0) { pf_unlink_src_node(nsn); uma_zfree(V_pf_sources_z, nsn); counter_u64_add( V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); } PF_HASHROW_UNLOCK(sh); } return (PF_DROP); } static int pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm) { struct pf_rule *r, *a = NULL; struct pf_ruleset *ruleset = NULL; sa_family_t af = pd->af; u_short reason; int tag = -1; int asd = 0; int match = 0; struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; PF_RULES_RASSERT(); r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != direction) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != af) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != pd->proto) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; else if (r->tos && !(r->tos == pd->tos)) r = TAILQ_NEXT(r, entries); else if (r->os_fingerprint != PF_OSFP_ANY) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_UDP && (r->src.port_op || r->dst.port_op)) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_TCP && (r->src.port_op || r->dst.port_op || r->flagset)) r = TAILQ_NEXT(r, entries); else if ((pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) && (r->type || r->code)) r = TAILQ_NEXT(r, entries); else if (r->prio && !pf_match_ieee8021q_pcp(r->prio, m)) r = TAILQ_NEXT(r, entries); else if (r->prob && r->prob <= (arc4random() % (UINT_MAX - 1) + 1)) r = TAILQ_NEXT(r, entries); else if (r->match_tag && !pf_match_tag(m, r, &tag, pd->pf_mtag ? pd->pf_mtag->tag : 0)) r = TAILQ_NEXT(r, entries); else { if (r->anchor == NULL) { match = 1; *rm = r; *am = a; *rsm = ruleset; if ((*rm)->quick) break; r = TAILQ_NEXT(r, entries); } else pf_step_into_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match); } if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match)) break; } r = *rm; a = *am; ruleset = *rsm; REASON_SET(&reason, PFRES_MATCH); if (r->log) PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 1); if (r->action != PF_PASS) return (PF_DROP); if (tag > 0 && pf_tag_packet(m, pd, tag)) { REASON_SET(&reason, PFRES_MEMORY); return (PF_DROP); } return (PF_PASS); } static int pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, int *copyback) { struct tcphdr *th = pd->hdr.tcp; u_int16_t win = ntohs(th->th_win); u_int32_t ack, end, seq, orig_seq; u_int8_t sws, dws; int ackskew; if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { sws = src->wscale & PF_WSCALE_MASK; dws = dst->wscale & PF_WSCALE_MASK; } else sws = dws = 0; /* * Sequence tracking algorithm from Guido van Rooij's paper: * http://www.madison-gurkha.com/publications/tcp_filtering/ * tcp_filtering.ps */ orig_seq = seq = ntohl(th->th_seq); if (src->seqlo == 0) { /* First packet from this end. Set its state */ if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && src->scrub == NULL) { if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { REASON_SET(reason, PFRES_MEMORY); return (PF_DROP); } } /* Deferred generation of sequence number modulator */ if (dst->seqdiff && !src->seqdiff) { /* use random iss for the TCP server */ while ((src->seqdiff = arc4random() - seq) == 0) ; ack = ntohl(th->th_ack) - dst->seqdiff; pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + src->seqdiff), 0); pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); *copyback = 1; } else { ack = ntohl(th->th_ack); } end = seq + pd->p_len; if (th->th_flags & TH_SYN) { end++; if (dst->wscale & PF_WSCALE_FLAG) { src->wscale = pf_get_wscale(m, off, th->th_off, pd->af); if (src->wscale & PF_WSCALE_FLAG) { /* Remove scale factor from initial * window */ sws = src->wscale & PF_WSCALE_MASK; win = ((u_int32_t)win + (1 << sws) - 1) >> sws; dws = dst->wscale & PF_WSCALE_MASK; } else { /* fixup other window */ dst->max_win <<= dst->wscale & PF_WSCALE_MASK; /* in case of a retrans SYN|ACK */ dst->wscale = 0; } } } if (th->th_flags & TH_FIN) end++; src->seqlo = seq; if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; /* * May need to slide the window (seqhi may have been set by * the crappy stack check or if we picked up the connection * after establishment) */ if (src->seqhi == 1 || SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) src->seqhi = end + MAX(1, dst->max_win << dws); if (win > src->max_win) src->max_win = win; } else { ack = ntohl(th->th_ack) - dst->seqdiff; if (src->seqdiff) { /* Modulate sequence numbers */ pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + src->seqdiff), 0); pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); *copyback = 1; } end = seq + pd->p_len; if (th->th_flags & TH_SYN) end++; if (th->th_flags & TH_FIN) end++; } if ((th->th_flags & TH_ACK) == 0) { /* Let it pass through the ack skew check */ ack = dst->seqlo; } else if ((ack == 0 && (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || /* broken tcp stacks do not set ack */ (dst->state < TCPS_SYN_SENT)) { /* * Many stacks (ours included) will set the ACK number in an * FIN|ACK if the SYN times out -- no sequence to ACK. */ ack = dst->seqlo; } if (seq == end) { /* Ease sequencing restrictions on no data packets */ seq = src->seqlo; end = seq; } ackskew = dst->seqlo - ack; /* * Need to demodulate the sequence numbers in any TCP SACK options * (Selective ACK). We could optionally validate the SACK values * against the current ACK window, either forwards or backwards, but * I'm not confident that SACK has been implemented properly * everywhere. It wouldn't surprise me if several stacks accidentally * SACK too far backwards of previously ACKed data. There really aren't * any security implications of bad SACKing unless the target stack * doesn't validate the option length correctly. Someone trying to * spoof into a TCP connection won't bother blindly sending SACK * options anyway. */ if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { if (pf_modulate_sack(m, off, pd, th, dst)) *copyback = 1; } #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ if (SEQ_GEQ(src->seqhi, end) && /* Last octet inside other's window space */ SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && /* Retrans: not more than one window back */ (ackskew >= -MAXACKWINDOW) && /* Acking not more than one reassembled fragment backwards */ (ackskew <= (MAXACKWINDOW << sws)) && /* Acking not more than one window forward */ ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || (pd->flags & PFDESC_IP_REAS) == 0)) { /* Require an exact/+1 sequence match on resets when possible */ if (dst->scrub || src->scrub) { if (pf_normalize_tcp_stateful(m, off, pd, reason, th, *state, src, dst, copyback)) return (PF_DROP); } /* update max window */ if (src->max_win < win) src->max_win = win; /* synchronize sequencing */ if (SEQ_GT(end, src->seqlo)) src->seqlo = end; /* slide the window of what the other end can send */ if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) dst->seqhi = ack + MAX((win << sws), 1); /* update states */ if (th->th_flags & TH_SYN) if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_ACK) { if (dst->state == TCPS_SYN_SENT) { dst->state = TCPS_ESTABLISHED; if (src->state == TCPS_ESTABLISHED && (*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (dst->state == TCPS_CLOSING) dst->state = TCPS_FIN_WAIT_2; } if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* update expire time */ (*state)->expire = time_uptime; if (src->state >= TCPS_FIN_WAIT_2 && dst->state >= TCPS_FIN_WAIT_2) (*state)->timeout = PFTM_TCP_CLOSED; else if (src->state >= TCPS_CLOSING && dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_FIN_WAIT; else if (src->state < TCPS_ESTABLISHED || dst->state < TCPS_ESTABLISHED) (*state)->timeout = PFTM_TCP_OPENING; else if (src->state >= TCPS_CLOSING || dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_CLOSING; else (*state)->timeout = PFTM_TCP_ESTABLISHED; /* Fall through to PASS packet */ } else if ((dst->state < TCPS_SYN_SENT || dst->state >= TCPS_FIN_WAIT_2 || src->state >= TCPS_FIN_WAIT_2) && SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && /* Within a window forward of the originating packet */ SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { /* Within a window backward of the originating packet */ /* * This currently handles three situations: * 1) Stupid stacks will shotgun SYNs before their peer * replies. * 2) When PF catches an already established stream (the * firewall rebooted, the state table was flushed, routes * changed...) * 3) Packets get funky immediately after the connection * closes (this should catch Solaris spurious ACK|FINs * that web servers like to spew after a close) * * This must be a little more careful than the above code * since packet floods will also be caught here. We don't * update the TTL here to mitigate the damage of a packet * flood and so the same code can handle awkward establishment * and a loosened connection close. * In the establishment case, a correct peer response will * validate the connection, go through the normal state code * and keep updating the state TTL. */ if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: loose state match: "); pf_print_state(*state); pf_print_flags(th->th_flags); printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], (unsigned long long)(*state)->packets[1], pd->dir == PF_IN ? "in" : "out", pd->dir == (*state)->direction ? "fwd" : "rev"); } if (dst->scrub || src->scrub) { if (pf_normalize_tcp_stateful(m, off, pd, reason, th, *state, src, dst, copyback)) return (PF_DROP); } /* update max window */ if (src->max_win < win) src->max_win = win; /* synchronize sequencing */ if (SEQ_GT(end, src->seqlo)) src->seqlo = end; /* slide the window of what the other end can send */ if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) dst->seqhi = ack + MAX((win << sws), 1); /* * Cannot set dst->seqhi here since this could be a shotgunned * SYN and not an already established connection. */ if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* Fall through to PASS packet */ } else { if ((*state)->dst.state == TCPS_SYN_SENT && (*state)->src.state == TCPS_SYN_SENT) { /* Send RST for state mismatches during handshake */ if (!(th->th_flags & TH_RST)) pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), 0, TH_RST, 0, 0, (*state)->rule.ptr->return_ttl, 1, 0, kif->pfik_ifp); src->seqlo = 0; src->seqhi = 1; src->max_win = 1; } else if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: BAD state: "); pf_print_state(*state); pf_print_flags(th->th_flags); printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], (unsigned long long)(*state)->packets[1], pd->dir == PF_IN ? "in" : "out", pd->dir == (*state)->direction ? "fwd" : "rev"); printf("pf: State failure on: %c %c %c %c | %c %c\n", SEQ_GEQ(src->seqhi, end) ? ' ' : '1', SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? ' ': '2', (ackskew >= -MAXACKWINDOW) ? ' ' : '3', (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); } REASON_SET(reason, PFRES_BADSTATE); return (PF_DROP); } return (PF_PASS); } static int pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, struct pf_state **state, struct pf_pdesc *pd, u_short *reason) { struct tcphdr *th = pd->hdr.tcp; if (th->th_flags & TH_SYN) if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_ACK) { if (dst->state == TCPS_SYN_SENT) { dst->state = TCPS_ESTABLISHED; if (src->state == TCPS_ESTABLISHED && (*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (dst->state == TCPS_CLOSING) { dst->state = TCPS_FIN_WAIT_2; } else if (src->state == TCPS_SYN_SENT && dst->state < TCPS_SYN_SENT) { /* * Handle a special sloppy case where we only see one * half of the connection. If there is a ACK after * the initial SYN without ever seeing a packet from * the destination, set the connection to established. */ dst->state = src->state = TCPS_ESTABLISHED; if ((*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (src->state == TCPS_CLOSING && dst->state == TCPS_ESTABLISHED && dst->seqlo == 0) { /* * Handle the closing of half connections where we * don't see the full bidirectional FIN/ACK+ACK * handshake. */ dst->state = TCPS_CLOSING; } } if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* update expire time */ (*state)->expire = time_uptime; if (src->state >= TCPS_FIN_WAIT_2 && dst->state >= TCPS_FIN_WAIT_2) (*state)->timeout = PFTM_TCP_CLOSED; else if (src->state >= TCPS_CLOSING && dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_FIN_WAIT; else if (src->state < TCPS_ESTABLISHED || dst->state < TCPS_ESTABLISHED) (*state)->timeout = PFTM_TCP_OPENING; else if (src->state >= TCPS_CLOSING || dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_CLOSING; else (*state)->timeout = PFTM_TCP_ESTABLISHED; return (PF_PASS); } static int pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) { struct pf_state_key_cmp key; struct tcphdr *th = pd->hdr.tcp; int copyback = 0; struct pf_state_peer *src, *dst; struct pf_state_key *sk; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = IPPROTO_TCP; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = th->th_sport; key.port[1] = th->th_dport; } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = th->th_sport; key.port[0] = th->th_dport; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } sk = (*state)->key[pd->didx]; if ((*state)->src.state == PF_TCPS_PROXY_SRC) { if (direction != (*state)->direction) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } if (th->th_flags & TH_SYN) { if (ntohl(th->th_seq) != (*state)->src.seqlo) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, (*state)->src.seqhi, ntohl(th->th_seq) + 1, TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } else if (!(th->th_flags & TH_ACK) || (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } else if ((*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } else (*state)->src.state = PF_TCPS_PROXY_DST; } if ((*state)->src.state == PF_TCPS_PROXY_DST) { if (direction == (*state)->direction) { if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } (*state)->src.max_win = MAX(ntohs(th->th_win), 1); if ((*state)->dst.seqhi == 1) (*state)->dst.seqhi = htonl(arc4random()); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, &sk->addr[pd->sidx], &sk->addr[pd->didx], sk->port[pd->sidx], sk->port[pd->didx], (*state)->dst.seqhi, 0, TH_SYN, 0, (*state)->src.mss, 0, 0, (*state)->tag, NULL); REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } else if (((th->th_flags & (TH_SYN|TH_ACK)) != (TH_SYN|TH_ACK)) || (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } else { (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); (*state)->dst.seqlo = ntohl(th->th_seq); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), ntohl(th->th_seq) + 1, TH_ACK, (*state)->src.max_win, 0, 0, 0, (*state)->tag, NULL); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, &sk->addr[pd->sidx], &sk->addr[pd->didx], sk->port[pd->sidx], sk->port[pd->didx], (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); (*state)->src.seqdiff = (*state)->dst.seqhi - (*state)->src.seqlo; (*state)->dst.seqdiff = (*state)->src.seqhi - (*state)->dst.seqlo; (*state)->src.seqhi = (*state)->src.seqlo + (*state)->dst.max_win; (*state)->dst.seqhi = (*state)->dst.seqlo + (*state)->src.max_win; (*state)->src.wscale = (*state)->dst.wscale = 0; (*state)->src.state = (*state)->dst.state = TCPS_ESTABLISHED; REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } } if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && dst->state >= TCPS_FIN_WAIT_2 && src->state >= TCPS_FIN_WAIT_2) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: state reuse "); pf_print_state(*state); pf_print_flags(th->th_flags); printf("\n"); } /* XXX make sure it's the same direction ?? */ (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; pf_unlink_state(*state, PF_ENTER_LOCKED); *state = NULL; return (PF_DROP); } if ((*state)->state_flags & PFSTATE_SLOPPY) { if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) return (PF_DROP); } else { if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, ©back) == PF_DROP) return (PF_DROP); } /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || nk->port[pd->sidx] != th->th_sport) pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 0, pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || nk->port[pd->didx] != th->th_dport) pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], nk->port[pd->didx], 0, pd->af); copyback = 1; } /* Copyback sequence modulation or stateful scrub changes if needed */ if (copyback) m_copyback(m, off, sizeof(*th), (caddr_t)th); return (PF_PASS); } static int pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd) { struct pf_state_peer *src, *dst; struct pf_state_key_cmp key; struct udphdr *uh = pd->hdr.udp; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = IPPROTO_UDP; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = uh->uh_sport; key.port[1] = uh->uh_dport; } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = uh->uh_sport; key.port[0] = uh->uh_dport; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } /* update states */ if (src->state < PFUDPS_SINGLE) src->state = PFUDPS_SINGLE; if (dst->state == PFUDPS_SINGLE) dst->state = PFUDPS_MULTIPLE; /* update expire time */ (*state)->expire = time_uptime; if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) (*state)->timeout = PFTM_UDP_MULTIPLE; else (*state)->timeout = PFTM_UDP_SINGLE; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || nk->port[pd->sidx] != uh->uh_sport) pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, &uh->uh_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 1, pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || nk->port[pd->didx] != uh->uh_dport) pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, &uh->uh_sum, &nk->addr[pd->didx], nk->port[pd->didx], 1, pd->af); m_copyback(m, off, sizeof(*uh), (caddr_t)uh); } return (PF_PASS); } static int pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) { struct pf_addr *saddr = pd->src, *daddr = pd->dst; u_int16_t icmpid = 0, *icmpsum; u_int8_t icmptype; int state_icmp = 0; struct pf_state_key_cmp key; bzero(&key, sizeof(key)); switch (pd->proto) { #ifdef INET case IPPROTO_ICMP: icmptype = pd->hdr.icmp->icmp_type; icmpid = pd->hdr.icmp->icmp_id; icmpsum = &pd->hdr.icmp->icmp_cksum; if (icmptype == ICMP_UNREACH || icmptype == ICMP_SOURCEQUENCH || icmptype == ICMP_REDIRECT || icmptype == ICMP_TIMXCEED || icmptype == ICMP_PARAMPROB) state_icmp++; break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: icmptype = pd->hdr.icmp6->icmp6_type; icmpid = pd->hdr.icmp6->icmp6_id; icmpsum = &pd->hdr.icmp6->icmp6_cksum; if (icmptype == ICMP6_DST_UNREACH || icmptype == ICMP6_PACKET_TOO_BIG || icmptype == ICMP6_TIME_EXCEEDED || icmptype == ICMP6_PARAM_PROB) state_icmp++; break; #endif /* INET6 */ } if (!state_icmp) { /* * ICMP query/reply message not related to a TCP/UDP packet. * Search for an ICMP state. */ key.af = pd->af; key.proto = pd->proto; key.port[0] = key.port[1] = icmpid; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); } STATE_LOOKUP(kif, &key, direction, *state, pd); (*state)->expire = time_uptime; (*state)->timeout = PFTM_ICMP_ERROR_REPLY; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; switch (pd->af) { #ifdef INET case AF_INET: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); if (nk->port[0] != pd->hdr.icmp->icmp_id) { pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( pd->hdr.icmp->icmp_cksum, icmpid, nk->port[pd->sidx], 0); pd->hdr.icmp->icmp_id = nk->port[pd->sidx]; } m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->sidx], 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->didx], 0); m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); break; #endif /* INET6 */ } } return (PF_PASS); } else { /* * ICMP error message in response to a TCP/UDP packet. * Extract the inner TCP/UDP header and search for that state. */ struct pf_pdesc pd2; bzero(&pd2, sizeof pd2); #ifdef INET struct ip h2; #endif /* INET */ #ifdef INET6 struct ip6_hdr h2_6; int terminal = 0; #endif /* INET6 */ int ipoff2 = 0; int off2 = 0; pd2.af = pd->af; /* Payload packet is from the opposite direction. */ pd2.sidx = (direction == PF_IN) ? 1 : 0; pd2.didx = (direction == PF_IN) ? 0 : 1; switch (pd->af) { #ifdef INET case AF_INET: /* offset of h2 in mbuf chain */ ipoff2 = off + ICMP_MINLEN; if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(ip)\n")); return (PF_DROP); } /* * ICMP error messages don't refer to non-first * fragments */ if (h2.ip_off & htons(IP_OFFMASK)) { REASON_SET(reason, PFRES_FRAG); return (PF_DROP); } /* offset of protocol header that follows h2 */ off2 = ipoff2 + (h2.ip_hl << 2); pd2.proto = h2.ip_p; pd2.src = (struct pf_addr *)&h2.ip_src; pd2.dst = (struct pf_addr *)&h2.ip_dst; pd2.ip_sum = &h2.ip_sum; break; #endif /* INET */ #ifdef INET6 case AF_INET6: ipoff2 = off + sizeof(struct icmp6_hdr); if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(ip6)\n")); return (PF_DROP); } pd2.proto = h2_6.ip6_nxt; pd2.src = (struct pf_addr *)&h2_6.ip6_src; pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; pd2.ip_sum = NULL; off2 = ipoff2 + sizeof(h2_6); do { switch (pd2.proto) { case IPPROTO_FRAGMENT: /* * ICMPv6 error messages for * non-first fragments */ REASON_SET(reason, PFRES_FRAG); return (PF_DROP); case IPPROTO_AH: case IPPROTO_HOPOPTS: case IPPROTO_ROUTING: case IPPROTO_DSTOPTS: { /* get next header and header length */ struct ip6_ext opt6; if (!pf_pull_hdr(m, off2, &opt6, sizeof(opt6), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMPv6 short opt\n")); return (PF_DROP); } if (pd2.proto == IPPROTO_AH) off2 += (opt6.ip6e_len + 2) * 4; else off2 += (opt6.ip6e_len + 1) * 8; pd2.proto = opt6.ip6e_nxt; /* goto the next header */ break; } default: terminal++; break; } } while (!terminal); break; #endif /* INET6 */ } switch (pd2.proto) { case IPPROTO_TCP: { struct tcphdr th; u_int32_t seq; struct pf_state_peer *src, *dst; u_int8_t dws; int copyback = 0; /* * Only the first 8 bytes of the TCP header can be * expected. Don't access any TCP header fields after * th_seq, an ackskew test is not possible. */ if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(tcp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_TCP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[pd2.sidx] = th.th_sport; key.port[pd2.didx] = th.th_dport; STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->dst; dst = &(*state)->src; } else { src = &(*state)->src; dst = &(*state)->dst; } if (src->wscale && dst->wscale) dws = dst->wscale & PF_WSCALE_MASK; else dws = 0; /* Demodulate sequence number */ seq = ntohl(th.th_seq) - src->seqdiff; if (src->seqdiff) { pf_change_a(&th.th_seq, icmpsum, htonl(seq), 0); copyback = 1; } if (!((*state)->state_flags & PFSTATE_SLOPPY) && (!SEQ_GEQ(src->seqhi, seq) || !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: BAD ICMP %d:%d ", icmptype, pd->hdr.icmp->icmp_code); pf_print_host(pd->src, 0, pd->af); printf(" -> "); pf_print_host(pd->dst, 0, pd->af); printf(" state: "); pf_print_state(*state); printf(" seq=%u\n", seq); } REASON_SET(reason, PFRES_BADSTATE); return (PF_DROP); } else { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: OK ICMP %d:%d ", icmptype, pd->hdr.icmp->icmp_code); pf_print_host(pd->src, 0, pd->af); printf(" -> "); pf_print_host(pd->dst, 0, pd->af); printf(" state: "); pf_print_state(*state); printf(" seq=%u\n", seq); } } /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != th.th_sport) pf_change_icmp(pd2.src, &th.th_sport, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != th.th_dport) pf_change_icmp(pd2.dst, &th.th_dport, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); copyback = 1; } if (copyback) { switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t )&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } m_copyback(m, off2, 8, (caddr_t)&th); } return (PF_PASS); break; } case IPPROTO_UDP: { struct udphdr uh; if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(udp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_UDP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[pd2.sidx] = uh.uh_sport; key.port[pd2.didx] = uh.uh_dport; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != uh.uh_sport) pf_change_icmp(pd2.src, &uh.uh_sport, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], &uh.uh_sum, pd2.ip_sum, icmpsum, pd->ip_sum, 1, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != uh.uh_dport) pf_change_icmp(pd2.dst, &uh.uh_dport, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], &uh.uh_sum, pd2.ip_sum, icmpsum, pd->ip_sum, 1, pd2.af); switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); } return (PF_PASS); break; } #ifdef INET case IPPROTO_ICMP: { struct icmp iih; if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short i" "(icmp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_ICMP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = iih.icmp_id; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != iih.icmp_id) pf_change_icmp(pd2.src, &iih.icmp_id, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != iih.icmp_id) pf_change_icmp(pd2.dst, &iih.icmp_id, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET); m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); } return (PF_PASS); break; } #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: { struct icmp6_hdr iih; if (!pf_pull_hdr(m, off2, &iih, sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(icmp6)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_ICMPV6; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = iih.icmp6_id; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != iih.icmp6_id) pf_change_icmp(pd2.src, &iih.icmp6_id, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET6); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != iih.icmp6_id) pf_change_icmp(pd2.dst, &iih.icmp6_id, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET6); m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t)pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); m_copyback(m, off2, sizeof(struct icmp6_hdr), (caddr_t)&iih); } return (PF_PASS); break; } #endif /* INET6 */ default: { key.af = pd2.af; key.proto = pd2.proto; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = 0; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af)) pf_change_icmp(pd2.src, NULL, daddr, &nk->addr[pd2.sidx], 0, NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af)) pf_change_icmp(pd2.dst, NULL, saddr, &nk->addr[pd2.didx], 0, NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } } return (PF_PASS); break; } } } } static int pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, struct pf_pdesc *pd) { struct pf_state_peer *src, *dst; struct pf_state_key_cmp key; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = pd->proto; if (direction == PF_IN) { PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = key.port[1] = 0; } else { PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = key.port[0] = 0; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } /* update states */ if (src->state < PFOTHERS_SINGLE) src->state = PFOTHERS_SINGLE; if (dst->state == PFOTHERS_SINGLE) dst->state = PFOTHERS_MULTIPLE; /* update expire time */ (*state)->expire = time_uptime; if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) (*state)->timeout = PFTM_OTHER_MULTIPLE; else (*state)->timeout = PFTM_OTHER_SINGLE; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; KASSERT(nk, ("%s: nk is null", __func__)); KASSERT(pd, ("%s: pd is null", __func__)); KASSERT(pd->src, ("%s: pd->src is null", __func__)); KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); switch (pd->af) { #ifdef INET case AF_INET: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); #endif /* INET6 */ } } return (PF_PASS); } /* * ipoff and off are measured from the start of the mbuf chain. * h must be at "ipoff" on the mbuf chain. */ void * pf_pull_hdr(struct mbuf *m, int off, void *p, int len, u_short *actionp, u_short *reasonp, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: { struct ip *h = mtod(m, struct ip *); u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; if (fragoff) { if (fragoff >= len) ACTION_SET(actionp, PF_PASS); else { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_FRAG); } return (NULL); } if (m->m_pkthdr.len < off + len || ntohs(h->ip_len) < off + len) { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_SHORT); return (NULL); } break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { struct ip6_hdr *h = mtod(m, struct ip6_hdr *); if (m->m_pkthdr.len < off + len || (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < (unsigned)(off + len)) { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_SHORT); return (NULL); } break; } #endif /* INET6 */ } m_copydata(m, off, len, p); return (p); } #ifdef RADIX_MPATH static int pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, int rtableid) { struct radix_node_head *rnh; struct sockaddr_in *dst; int ret = 1; int check_mpath; #ifdef INET6 struct sockaddr_in6 *dst6; struct route_in6 ro; #else struct route ro; #endif struct radix_node *rn; struct rtentry *rt; struct ifnet *ifp; check_mpath = 0; /* XXX: stick to table 0 for now */ rnh = rt_tables_get_rnh(0, af); if (rnh != NULL && rn_mpath_capable(rnh)) check_mpath = 1; bzero(&ro, sizeof(ro)); switch (af) { case AF_INET: dst = satosin(&ro.ro_dst); dst->sin_family = AF_INET; dst->sin_len = sizeof(*dst); dst->sin_addr = addr->v4; break; #ifdef INET6 case AF_INET6: /* * Skip check for addresses with embedded interface scope, * as they would always match anyway. */ if (IN6_IS_SCOPE_EMBED(&addr->v6)) goto out; dst6 = (struct sockaddr_in6 *)&ro.ro_dst; dst6->sin6_family = AF_INET6; dst6->sin6_len = sizeof(*dst6); dst6->sin6_addr = addr->v6; break; #endif /* INET6 */ default: return (0); } /* Skip checks for ipsec interfaces */ if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) goto out; switch (af) { #ifdef INET6 case AF_INET6: in6_rtalloc_ign(&ro, 0, rtableid); break; #endif #ifdef INET case AF_INET: in_rtalloc_ign((struct route *)&ro, 0, rtableid); break; #endif } if (ro.ro_rt != NULL) { /* No interface given, this is a no-route check */ if (kif == NULL) goto out; if (kif->pfik_ifp == NULL) { ret = 0; goto out; } /* Perform uRPF check if passed input interface */ ret = 0; rn = (struct radix_node *)ro.ro_rt; do { rt = (struct rtentry *)rn; ifp = rt->rt_ifp; if (kif->pfik_ifp == ifp) ret = 1; rn = rn_mpath_next(rn); } while (check_mpath == 1 && rn != NULL && ret == 0); } else ret = 0; out: if (ro.ro_rt != NULL) RTFREE(ro.ro_rt); return (ret); } #endif int pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, int rtableid) { #ifdef INET struct nhop4_basic nh4; #endif #ifdef INET6 struct nhop6_basic nh6; #endif struct ifnet *ifp; #ifdef RADIX_MPATH struct radix_node_head *rnh; /* XXX: stick to table 0 for now */ rnh = rt_tables_get_rnh(0, af); if (rnh != NULL && rn_mpath_capable(rnh)) return (pf_routable_oldmpath(addr, af, kif, rtableid)); #endif /* * Skip check for addresses with embedded interface scope, * as they would always match anyway. */ if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) return (1); if (af != AF_INET && af != AF_INET6) return (0); /* Skip checks for ipsec interfaces */ if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) return (1); ifp = NULL; switch (af) { #ifdef INET6 case AF_INET6: if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0) return (0); ifp = nh6.nh_ifp; break; #endif #ifdef INET case AF_INET: if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0) return (0); ifp = nh4.nh_ifp; break; #endif } /* No interface given, this is a no-route check */ if (kif == NULL) return (1); if (kif->pfik_ifp == NULL) return (0); /* Perform uRPF check if passed input interface */ if (kif->pfik_ifp == ifp) return (1); return (0); } #ifdef INET static void pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) { struct mbuf *m0, *m1; struct sockaddr_in dst; struct ip *ip; struct ifnet *ifp = NULL; struct pf_addr naddr; struct pf_src_node *sn = NULL; int error = 0; uint16_t ip_len, ip_off; KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", __func__)); if ((pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || pd->pf_mtag->routed++ > 3) { m0 = *m; *m = NULL; goto bad_locked; } if (r->rt == PF_DUPTO) { if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { if (s) PF_STATE_UNLOCK(s); return; } } else { if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { if (s) PF_STATE_UNLOCK(s); return; } m0 = *m; } ip = mtod(m0, struct ip *); bzero(&dst, sizeof(dst)); dst.sin_family = AF_INET; dst.sin_len = sizeof(dst); dst.sin_addr = ip->ip_dst; if (TAILQ_EMPTY(&r->rpool.list)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); goto bad_locked; } if (s == NULL) { pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, &naddr, NULL, &sn); if (!PF_AZERO(&naddr, AF_INET)) dst.sin_addr.s_addr = naddr.v4.s_addr; ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; } else { if (!PF_AZERO(&s->rt_addr, AF_INET)) dst.sin_addr.s_addr = s->rt_addr.v4.s_addr; ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; PF_STATE_UNLOCK(s); } if (ifp == NULL) goto bad; if (oifp != ifp) { if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) goto bad; else if (m0 == NULL) goto done; if (m0->m_len < sizeof(struct ip)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); goto bad; } ip = mtod(m0, struct ip *); } if (ifp->if_flags & IFF_LOOPBACK) m0->m_flags |= M_SKIP_FIREWALL; ip_len = ntohs(ip->ip_len); ip_off = ntohs(ip->ip_off); /* Copied from FreeBSD 10.0-CURRENT ip_output. */ m0->m_pkthdr.csum_flags |= CSUM_IP; if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { in_delayed_cksum(m0); m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } #ifdef SCTP if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; } #endif /* * If small enough for interface, or the interface will take * care of the fragmentation for us, we can just send directly. */ if (ip_len <= ifp->if_mtu || (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { ip->ip_sum = 0; if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); m0->m_pkthdr.csum_flags &= ~CSUM_IP; } m_clrprotoflags(m0); /* Avoid confusing lower layers. */ error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); goto done; } /* Balk when DF bit is set or the interface didn't support TSO. */ if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { error = EMSGSIZE; KMOD_IPSTAT_INC(ips_cantfrag); if (r->rt != PF_DUPTO) { icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, ifp->if_mtu); goto done; } else goto bad; } error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); if (error) goto bad; for (; m0; m0 = m1) { m1 = m0->m_nextpkt; m0->m_nextpkt = NULL; if (error == 0) { m_clrprotoflags(m0); error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); } else m_freem(m0); } if (error == 0) KMOD_IPSTAT_INC(ips_fragmented); done: if (r->rt != PF_DUPTO) *m = NULL; return; bad_locked: if (s) PF_STATE_UNLOCK(s); bad: m_freem(m0); goto done; } #endif /* INET */ #ifdef INET6 static void pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) { struct mbuf *m0; struct sockaddr_in6 dst; struct ip6_hdr *ip6; struct ifnet *ifp = NULL; struct pf_addr naddr; struct pf_src_node *sn = NULL; KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", __func__)); if ((pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || pd->pf_mtag->routed++ > 3) { m0 = *m; *m = NULL; goto bad_locked; } if (r->rt == PF_DUPTO) { if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { if (s) PF_STATE_UNLOCK(s); return; } } else { if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { if (s) PF_STATE_UNLOCK(s); return; } m0 = *m; } ip6 = mtod(m0, struct ip6_hdr *); bzero(&dst, sizeof(dst)); dst.sin6_family = AF_INET6; dst.sin6_len = sizeof(dst); dst.sin6_addr = ip6->ip6_dst; if (TAILQ_EMPTY(&r->rpool.list)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); goto bad_locked; } if (s == NULL) { pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, &naddr, NULL, &sn); if (!PF_AZERO(&naddr, AF_INET6)) PF_ACPY((struct pf_addr *)&dst.sin6_addr, &naddr, AF_INET6); ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; } else { if (!PF_AZERO(&s->rt_addr, AF_INET6)) PF_ACPY((struct pf_addr *)&dst.sin6_addr, &s->rt_addr, AF_INET6); ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; } if (s) PF_STATE_UNLOCK(s); if (ifp == NULL) goto bad; if (oifp != ifp) { if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS) goto bad; else if (m0 == NULL) goto done; if (m0->m_len < sizeof(struct ip6_hdr)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", __func__)); goto bad; } ip6 = mtod(m0, struct ip6_hdr *); } if (ifp->if_flags & IFF_LOOPBACK) m0->m_flags |= M_SKIP_FIREWALL; if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & ~ifp->if_hwassist) { uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; } /* * If the packet is too large for the outgoing interface, * send back an icmp6 error. */ if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) nd6_output_ifp(ifp, ifp, m0, &dst, NULL); else { in6_ifstat_inc(ifp, ifs6_in_toobig); if (r->rt != PF_DUPTO) icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); else goto bad; } done: if (r->rt != PF_DUPTO) *m = NULL; return; bad_locked: if (s) PF_STATE_UNLOCK(s); bad: m_freem(m0); goto done; } #endif /* INET6 */ /* * FreeBSD supports cksum offloads for the following drivers. * em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4) * * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : * network driver performed cksum including pseudo header, need to verify * csum_data * CSUM_DATA_VALID : * network driver performed cksum, needs to additional pseudo header * cksum computation with partial csum_data(i.e. lack of H/W support for * pseudo header, for instance hme(4), sk(4) and possibly gem(4)) * * After validating the cksum of packet, set both flag CSUM_DATA_VALID and * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper * TCP/UDP layer. * Also, set csum_data to 0xffff to force cksum validation. */ static int pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) { u_int16_t sum = 0; int hw_assist = 0; struct ip *ip; if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) return (1); if (m->m_pkthdr.len < off + len) return (1); switch (p) { case IPPROTO_TCP: if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { sum = m->m_pkthdr.csum_data; } else { ip = mtod(m, struct ip *); sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + IPPROTO_TCP)); } sum ^= 0xffff; ++hw_assist; } break; case IPPROTO_UDP: if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { sum = m->m_pkthdr.csum_data; } else { ip = mtod(m, struct ip *); sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + IPPROTO_UDP)); } sum ^= 0xffff; ++hw_assist; } break; case IPPROTO_ICMP: #ifdef INET6 case IPPROTO_ICMPV6: #endif /* INET6 */ break; default: return (1); } if (!hw_assist) { switch (af) { case AF_INET: if (p == IPPROTO_ICMP) { if (m->m_len < off) return (1); m->m_data += off; m->m_len -= off; sum = in_cksum(m, len); m->m_data -= off; m->m_len += off; } else { if (m->m_len < sizeof(struct ip)) return (1); sum = in4_cksum(m, p, off, len); } break; #ifdef INET6 case AF_INET6: if (m->m_len < sizeof(struct ip6_hdr)) return (1); sum = in6_cksum(m, p, off, len); break; #endif /* INET6 */ default: return (1); } } if (sum) { switch (p) { case IPPROTO_TCP: { KMOD_TCPSTAT_INC(tcps_rcvbadsum); break; } case IPPROTO_UDP: { KMOD_UDPSTAT_INC(udps_badsum); break; } #ifdef INET case IPPROTO_ICMP: { KMOD_ICMPSTAT_INC(icps_checksum); break; } #endif #ifdef INET6 case IPPROTO_ICMPV6: { KMOD_ICMP6STAT_INC(icp6s_checksum); break; } #endif /* INET6 */ } return (1); } else { if (p == IPPROTO_TCP || p == IPPROTO_UDP) { m->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m->m_pkthdr.csum_data = 0xffff; } } return (0); } #ifdef INET int pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) { struct pfi_kif *kif; u_short action, reason = 0, log = 0; struct mbuf *m = *m0; struct ip *h = NULL; struct m_tag *ipfwtag; struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; struct pf_state *s = NULL; struct pf_ruleset *ruleset = NULL; struct pf_pdesc pd; int off, dirndx, pqid = 0; PF_RULES_RLOCK_TRACKER; M_ASSERTPKTHDR(m); if (!V_pf_status.running) return (PF_PASS); memset(&pd, 0, sizeof(pd)); kif = (struct pfi_kif *)ifp->if_pf_kif; if (kif == NULL) { DPFPRINTF(PF_DEBUG_URGENT, ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); return (PF_DROP); } if (kif->pfik_flags & PFI_IFLAG_SKIP) return (PF_PASS); if (m->m_flags & M_SKIP_FIREWALL) return (PF_PASS); pd.pf_mtag = pf_find_mtag(m); PF_RULES_RLOCK(); if (ip_divert_ptr != NULL && ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; goto done; } pd.pf_mtag->flags |= PF_PACKET_LOOPED; m_tag_delete(m, ipfwtag); } if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { m->m_flags |= M_FASTFWD_OURS; pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; } } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { /* We do IP header normalization and packet reassembly here */ action = PF_DROP; goto done; } m = *m0; /* pf_normalize messes with m0 */ h = mtod(m, struct ip *); off = h->ip_hl << 2; if (off < (int)sizeof(struct ip)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); log = 1; goto done; } pd.src = (struct pf_addr *)&h->ip_src; pd.dst = (struct pf_addr *)&h->ip_dst; pd.sport = pd.dport = NULL; pd.ip_sum = &h->ip_sum; pd.proto_sum = NULL; pd.proto = h->ip_p; pd.dir = dir; pd.sidx = (dir == PF_IN) ? 0 : 1; pd.didx = (dir == PF_IN) ? 1 : 0; pd.af = AF_INET; pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; pd.tot_len = ntohs(h->ip_len); /* handle fragments that didn't get reassembled by normalization */ if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { action = pf_test_fragment(&r, dir, kif, m, h, &pd, &a, &ruleset); goto done; } switch (h->ip_p) { case IPPROTO_TCP: { struct tcphdr th; pd.hdr.tcp = &th; if (!pf_pull_hdr(m, off, &th, sizeof(th), &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } pd.p_len = pd.tot_len - off - (th.th_off << 2); if ((th.th_flags & TH_ACK) && pd.p_len == 0) pqid = 1; action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); if (action == PF_DROP) goto done; action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_UDP: { struct udphdr uh; pd.hdr.udp = &uh; if (!pf_pull_hdr(m, off, &uh, sizeof(uh), &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } if (uh.uh_dport == 0 || ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); goto done; } action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_ICMP: { struct icmp ih; pd.hdr.icmp = &ih; if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } #ifdef INET6 case IPPROTO_ICMPV6: { action = PF_DROP; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping IPv4 packet with ICMPv6 payload\n")); goto done; } #endif default: action = pf_test_state_other(&s, dir, kif, m, &pd); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } done: PF_RULES_RUNLOCK(); if (action == PF_PASS && h->ip_hl > 5 && !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = r->log; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping packet with ip options\n")); } if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } if (r->rtableid >= 0) M_SETFIB(m, r->rtableid); if (r->scrub_flags & PFSTATE_SETPRIO) { if (pd.tos & IPTOS_LOWDELAY) pqid = 1; if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate 802.1q mtag\n")); } } #ifdef ALTQ if (action == PF_PASS && r->qid) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } else { if (s != NULL) pd.pf_mtag->qid_hash = pf_state_hash(s); if (pqid || (pd.tos & IPTOS_LOWDELAY)) pd.pf_mtag->qid = r->pqid; else pd.pf_mtag->qid = r->qid; /* Add hints for ecn. */ pd.pf_mtag->hdr = h; } } #endif /* ALTQ */ /* * connections redirected to loopback should not match sockets * bound specifically to loopback due to security implications, * see tcp_input() and in_pcblookup_listen(). */ if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && (s->nat_rule.ptr->action == PF_RDR || s->nat_rule.ptr->action == PF_BINAT) && (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) m->m_flags |= M_SKIP_FIREWALL; if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && !PACKET_LOOPED(&pd)) { ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); if (ipfwtag != NULL) { ((struct ipfw_rule_ref *)(ipfwtag+1))->info = ntohs(r->divert.port); ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; if (s) PF_STATE_UNLOCK(s); m_tag_prepend(m, ipfwtag); if (m->m_flags & M_FASTFWD_OURS) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate tag\n")); } else { pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT; m->m_flags &= ~M_FASTFWD_OURS; } } ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT); *m0 = NULL; return (action); } else { /* XXX: ipfw has the same behaviour! */ action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate divert tag\n")); } } if (log) { struct pf_rule *lr; if (s != NULL && s->nat_rule.ptr != NULL && s->nat_rule.ptr->log & PF_LOG_ALL) lr = s->nat_rule.ptr; else lr = r; PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, (s == NULL)); } kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; if (action == PF_PASS || r->action == PF_DROP) { dirndx = (dir == PF_OUT); r->packets[dirndx]++; r->bytes[dirndx] += pd.tot_len; if (a != NULL) { a->packets[dirndx]++; a->bytes[dirndx] += pd.tot_len; } if (s != NULL) { if (s->nat_rule.ptr != NULL) { s->nat_rule.ptr->packets[dirndx]++; s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; } if (s->src_node != NULL) { s->src_node->packets[dirndx]++; s->src_node->bytes[dirndx] += pd.tot_len; } if (s->nat_src_node != NULL) { s->nat_src_node->packets[dirndx]++; s->nat_src_node->bytes[dirndx] += pd.tot_len; } dirndx = (dir == s->direction) ? 0 : 1; s->packets[dirndx]++; s->bytes[dirndx] += pd.tot_len; } tr = r; nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; if (nr != NULL && r == &V_pf_default_rule) tr = nr; if (tr->src.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->src.addr.p.tbl, (s == NULL) ? pd.src : &s->key[(s->direction == PF_IN)]-> addr[(s->direction == PF_OUT)], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->src.neg); if (tr->dst.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->dst.addr.p.tbl, (s == NULL) ? pd.dst : &s->key[(s->direction == PF_IN)]-> addr[(s->direction == PF_IN)], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->dst.neg); } switch (action) { case PF_SYNPROXY_DROP: m_freem(*m0); case PF_DEFER: *m0 = NULL; action = PF_PASS; break; case PF_DROP: m_freem(*m0); *m0 = NULL; break; default: /* pf_route() returns unlocked. */ if (r->rt) { pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); return (action); } break; } if (s) PF_STATE_UNLOCK(s); return (action); } #endif /* INET */ #ifdef INET6 int pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) { struct pfi_kif *kif; u_short action, reason = 0, log = 0; struct mbuf *m = *m0, *n = NULL; struct m_tag *mtag; struct ip6_hdr *h = NULL; struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; struct pf_state *s = NULL; struct pf_ruleset *ruleset = NULL; struct pf_pdesc pd; int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; PF_RULES_RLOCK_TRACKER; M_ASSERTPKTHDR(m); if (!V_pf_status.running) return (PF_PASS); memset(&pd, 0, sizeof(pd)); pd.pf_mtag = pf_find_mtag(m); if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) return (PF_PASS); kif = (struct pfi_kif *)ifp->if_pf_kif; if (kif == NULL) { DPFPRINTF(PF_DEBUG_URGENT, ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); return (PF_DROP); } if (kif->pfik_flags & PFI_IFLAG_SKIP) return (PF_PASS); if (m->m_flags & M_SKIP_FIREWALL) return (PF_PASS); PF_RULES_RLOCK(); /* We do IP header normalization and packet reassembly here */ if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { action = PF_DROP; goto done; } m = *m0; /* pf_normalize messes with m0 */ h = mtod(m, struct ip6_hdr *); #if 1 /* * we do not support jumbogram yet. if we keep going, zero ip6_plen * will do something bad, so drop the packet for now. */ if (htons(h->ip6_plen) == 0) { action = PF_DROP; REASON_SET(&reason, PFRES_NORM); /*XXX*/ goto done; } #endif pd.src = (struct pf_addr *)&h->ip6_src; pd.dst = (struct pf_addr *)&h->ip6_dst; pd.sport = pd.dport = NULL; pd.ip_sum = NULL; pd.proto_sum = NULL; pd.dir = dir; pd.sidx = (dir == PF_IN) ? 0 : 1; pd.didx = (dir == PF_IN) ? 1 : 0; pd.af = AF_INET6; pd.tos = 0; pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); pd.proto = h->ip6_nxt; do { switch (pd.proto) { case IPPROTO_FRAGMENT: action = pf_test_fragment(&r, dir, kif, m, h, &pd, &a, &ruleset); if (action == PF_DROP) REASON_SET(&reason, PFRES_FRAG); goto done; case IPPROTO_ROUTING: { struct ip6_rthdr rthdr; if (rh_cnt++) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 more than one rthdr\n")); action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = 1; goto done; } if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, &reason, pd.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 short rthdr\n")); action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); log = 1; goto done; } if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 rthdr0\n")); action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = 1; goto done; } /* FALLTHROUGH */ } case IPPROTO_AH: case IPPROTO_HOPOPTS: case IPPROTO_DSTOPTS: { /* get next header and header length */ struct ip6_ext opt6; if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), NULL, &reason, pd.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 short opt\n")); action = PF_DROP; log = 1; goto done; } if (pd.proto == IPPROTO_AH) off += (opt6.ip6e_len + 2) * 4; else off += (opt6.ip6e_len + 1) * 8; pd.proto = opt6.ip6e_nxt; /* goto the next header */ break; } default: terminal++; break; } } while (!terminal); /* if there's no routing header, use unmodified mbuf for checksumming */ if (!n) n = m; switch (pd.proto) { case IPPROTO_TCP: { struct tcphdr th; pd.hdr.tcp = &th; if (!pf_pull_hdr(m, off, &th, sizeof(th), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } pd.p_len = pd.tot_len - off - (th.th_off << 2); action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); if (action == PF_DROP) goto done; action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_UDP: { struct udphdr uh; pd.hdr.udp = &uh; if (!pf_pull_hdr(m, off, &uh, sizeof(uh), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } if (uh.uh_dport == 0 || ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); goto done; } action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_ICMP: { action = PF_DROP; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping IPv6 packet with ICMPv4 payload\n")); goto done; } case IPPROTO_ICMPV6: { struct icmp6_hdr ih; pd.hdr.icmp6 = &ih; if (!pf_pull_hdr(m, off, &ih, sizeof(ih), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } default: action = pf_test_state_other(&s, dir, kif, m, &pd); if (action == PF_PASS) { if (pfsync_update_state_ptr != NULL) pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } done: PF_RULES_RUNLOCK(); if (n != m) { m_freem(n); n = NULL; } /* handle dangerous IPv6 extension headers. */ if (action == PF_PASS && rh_cnt && !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = r->log; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping packet with dangerous v6 headers\n")); } if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } if (r->rtableid >= 0) M_SETFIB(m, r->rtableid); if (r->scrub_flags & PFSTATE_SETPRIO) { if (pd.tos & IPTOS_LOWDELAY) pqid = 1; if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate 802.1q mtag\n")); } } #ifdef ALTQ if (action == PF_PASS && r->qid) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } else { if (s != NULL) pd.pf_mtag->qid_hash = pf_state_hash(s); if (pd.tos & IPTOS_LOWDELAY) pd.pf_mtag->qid = r->pqid; else pd.pf_mtag->qid = r->qid; /* Add hints for ecn. */ pd.pf_mtag->hdr = h; } } #endif /* ALTQ */ if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && (s->nat_rule.ptr->action == PF_RDR || s->nat_rule.ptr->action == PF_BINAT) && IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) m->m_flags |= M_SKIP_FIREWALL; /* XXX: Anybody working on it?! */ if (r->divert.port) printf("pf: divert(9) is not supported for IPv6\n"); if (log) { struct pf_rule *lr; if (s != NULL && s->nat_rule.ptr != NULL && s->nat_rule.ptr->log & PF_LOG_ALL) lr = s->nat_rule.ptr; else lr = r; PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, &pd, (s == NULL)); } kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; if (action == PF_PASS || r->action == PF_DROP) { dirndx = (dir == PF_OUT); r->packets[dirndx]++; r->bytes[dirndx] += pd.tot_len; if (a != NULL) { a->packets[dirndx]++; a->bytes[dirndx] += pd.tot_len; } if (s != NULL) { if (s->nat_rule.ptr != NULL) { s->nat_rule.ptr->packets[dirndx]++; s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; } if (s->src_node != NULL) { s->src_node->packets[dirndx]++; s->src_node->bytes[dirndx] += pd.tot_len; } if (s->nat_src_node != NULL) { s->nat_src_node->packets[dirndx]++; s->nat_src_node->bytes[dirndx] += pd.tot_len; } dirndx = (dir == s->direction) ? 0 : 1; s->packets[dirndx]++; s->bytes[dirndx] += pd.tot_len; } tr = r; nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; if (nr != NULL && r == &V_pf_default_rule) tr = nr; if (tr->src.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->src.addr.p.tbl, (s == NULL) ? pd.src : &s->key[(s->direction == PF_IN)]->addr[0], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->src.neg); if (tr->dst.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->dst.addr.p.tbl, (s == NULL) ? pd.dst : &s->key[(s->direction == PF_IN)]->addr[1], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->dst.neg); } switch (action) { case PF_SYNPROXY_DROP: m_freem(*m0); case PF_DEFER: *m0 = NULL; action = PF_PASS; break; case PF_DROP: m_freem(*m0); *m0 = NULL; break; default: /* pf_route6() returns unlocked. */ if (r->rt) { pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); return (action); } break; } if (s) PF_STATE_UNLOCK(s); /* If reassembled packet passed, create new fragments. */ if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) && (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) action = pf_refragment6(ifp, m0, mtag); return (action); } #endif /* INET6 */ Index: head/sys/netpfil/pf/pf_if.c =================================================================== --- head/sys/netpfil/pf/pf_if.c (revision 336675) +++ head/sys/netpfil/pf/pf_if.c (revision 336676) @@ -1,905 +1,905 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * Copyright (c) 2003 Cedric Berger * Copyright (c) 2005 Henning Brauer * Copyright (c) 2005 Ryan McBride * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * $OpenBSD: pf_if.c,v 1.54 2008/06/14 16:55:28 mk Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include VNET_DEFINE(struct pfi_kif *, pfi_all); -static VNET_DEFINE(long, pfi_update); +VNET_DEFINE_STATIC(long, pfi_update); #define V_pfi_update VNET(pfi_update) #define PFI_BUFFER_MAX 0x10000 VNET_DECLARE(int, pf_vnet_active); #define V_pf_vnet_active VNET(pf_vnet_active) -static VNET_DEFINE(struct pfr_addr *, pfi_buffer); -static VNET_DEFINE(int, pfi_buffer_cnt); -static VNET_DEFINE(int, pfi_buffer_max); +VNET_DEFINE_STATIC(struct pfr_addr *, pfi_buffer); +VNET_DEFINE_STATIC(int, pfi_buffer_cnt); +VNET_DEFINE_STATIC(int, pfi_buffer_max); #define V_pfi_buffer VNET(pfi_buffer) #define V_pfi_buffer_cnt VNET(pfi_buffer_cnt) #define V_pfi_buffer_max VNET(pfi_buffer_max) eventhandler_tag pfi_attach_cookie; eventhandler_tag pfi_detach_cookie; eventhandler_tag pfi_attach_group_cookie; eventhandler_tag pfi_change_group_cookie; eventhandler_tag pfi_detach_group_cookie; eventhandler_tag pfi_ifaddr_event_cookie; static void pfi_attach_ifnet(struct ifnet *); static void pfi_attach_ifgroup(struct ifg_group *); static void pfi_kif_update(struct pfi_kif *); static void pfi_dynaddr_update(struct pfi_dynaddr *dyn); static void pfi_table_update(struct pfr_ktable *, struct pfi_kif *, int, int); static void pfi_instance_add(struct ifnet *, int, int); static void pfi_address_add(struct sockaddr *, int, int); static int pfi_if_compare(struct pfi_kif *, struct pfi_kif *); static int pfi_skip_if(const char *, struct pfi_kif *); static int pfi_unmask(void *); static void pfi_attach_ifnet_event(void * __unused, struct ifnet *); static void pfi_detach_ifnet_event(void * __unused, struct ifnet *); static void pfi_attach_group_event(void * __unused, struct ifg_group *); static void pfi_change_group_event(void * __unused, char *); static void pfi_detach_group_event(void * __unused, struct ifg_group *); static void pfi_ifaddr_event(void * __unused, struct ifnet *); RB_HEAD(pfi_ifhead, pfi_kif); static RB_PROTOTYPE(pfi_ifhead, pfi_kif, pfik_tree, pfi_if_compare); static RB_GENERATE(pfi_ifhead, pfi_kif, pfik_tree, pfi_if_compare); -static VNET_DEFINE(struct pfi_ifhead, pfi_ifs); +VNET_DEFINE_STATIC(struct pfi_ifhead, pfi_ifs); #define V_pfi_ifs VNET(pfi_ifs) #define PFI_BUFFER_MAX 0x10000 MALLOC_DEFINE(PFI_MTYPE, "pf_ifnet", "pf(4) interface database"); LIST_HEAD(pfi_list, pfi_kif); -static VNET_DEFINE(struct pfi_list, pfi_unlinked_kifs); +VNET_DEFINE_STATIC(struct pfi_list, pfi_unlinked_kifs); #define V_pfi_unlinked_kifs VNET(pfi_unlinked_kifs) static struct mtx pfi_unlnkdkifs_mtx; MTX_SYSINIT(pfi_unlnkdkifs_mtx, &pfi_unlnkdkifs_mtx, "pf unlinked interfaces", MTX_DEF); void pfi_initialize_vnet(void) { struct ifg_group *ifg; struct ifnet *ifp; struct pfi_kif *kif; V_pfi_buffer_max = 64; V_pfi_buffer = malloc(V_pfi_buffer_max * sizeof(*V_pfi_buffer), PFI_MTYPE, M_WAITOK); kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); V_pfi_all = pfi_kif_attach(kif, IFG_ALL); PF_RULES_WUNLOCK(); IFNET_RLOCK(); CK_STAILQ_FOREACH(ifg, &V_ifg_head, ifg_next) pfi_attach_ifgroup(ifg); CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) pfi_attach_ifnet(ifp); IFNET_RUNLOCK(); } void pfi_initialize(void) { pfi_attach_cookie = EVENTHANDLER_REGISTER(ifnet_arrival_event, pfi_attach_ifnet_event, NULL, EVENTHANDLER_PRI_ANY); pfi_detach_cookie = EVENTHANDLER_REGISTER(ifnet_departure_event, pfi_detach_ifnet_event, NULL, EVENTHANDLER_PRI_ANY); pfi_attach_group_cookie = EVENTHANDLER_REGISTER(group_attach_event, pfi_attach_group_event, NULL, EVENTHANDLER_PRI_ANY); pfi_change_group_cookie = EVENTHANDLER_REGISTER(group_change_event, pfi_change_group_event, NULL, EVENTHANDLER_PRI_ANY); pfi_detach_group_cookie = EVENTHANDLER_REGISTER(group_detach_event, pfi_detach_group_event, NULL, EVENTHANDLER_PRI_ANY); pfi_ifaddr_event_cookie = EVENTHANDLER_REGISTER(ifaddr_event, pfi_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY); } void pfi_cleanup_vnet(void) { struct pfi_kif *kif; PF_RULES_WASSERT(); V_pfi_all = NULL; while ((kif = RB_MIN(pfi_ifhead, &V_pfi_ifs))) { RB_REMOVE(pfi_ifhead, &V_pfi_ifs, kif); if (kif->pfik_group) kif->pfik_group->ifg_pf_kif = NULL; if (kif->pfik_ifp) kif->pfik_ifp->if_pf_kif = NULL; free(kif, PFI_MTYPE); } mtx_lock(&pfi_unlnkdkifs_mtx); while ((kif = LIST_FIRST(&V_pfi_unlinked_kifs))) { LIST_REMOVE(kif, pfik_list); free(kif, PFI_MTYPE); } mtx_unlock(&pfi_unlnkdkifs_mtx); free(V_pfi_buffer, PFI_MTYPE); } void pfi_cleanup(void) { EVENTHANDLER_DEREGISTER(ifnet_arrival_event, pfi_attach_cookie); EVENTHANDLER_DEREGISTER(ifnet_departure_event, pfi_detach_cookie); EVENTHANDLER_DEREGISTER(group_attach_event, pfi_attach_group_cookie); EVENTHANDLER_DEREGISTER(group_change_event, pfi_change_group_cookie); EVENTHANDLER_DEREGISTER(group_detach_event, pfi_detach_group_cookie); EVENTHANDLER_DEREGISTER(ifaddr_event, pfi_ifaddr_event_cookie); } struct pfi_kif * pfi_kif_find(const char *kif_name) { struct pfi_kif_cmp s; PF_RULES_ASSERT(); bzero(&s, sizeof(s)); strlcpy(s.pfik_name, kif_name, sizeof(s.pfik_name)); return (RB_FIND(pfi_ifhead, &V_pfi_ifs, (struct pfi_kif *)&s)); } struct pfi_kif * pfi_kif_attach(struct pfi_kif *kif, const char *kif_name) { struct pfi_kif *kif1; PF_RULES_WASSERT(); KASSERT(kif != NULL, ("%s: null kif", __func__)); kif1 = pfi_kif_find(kif_name); if (kif1 != NULL) { free(kif, PFI_MTYPE); return (kif1); } bzero(kif, sizeof(*kif)); strlcpy(kif->pfik_name, kif_name, sizeof(kif->pfik_name)); /* * It seems that the value of time_second is in unintialzied state * when pf sets interface statistics clear time in boot phase if pf * was statically linked to kernel. Instead of setting the bogus * time value have pfi_get_ifaces handle this case. In * pfi_get_ifaces it uses time_second if it sees the time is 0. */ kif->pfik_tzero = time_second > 1 ? time_second : 0; TAILQ_INIT(&kif->pfik_dynaddrs); RB_INSERT(pfi_ifhead, &V_pfi_ifs, kif); return (kif); } void pfi_kif_ref(struct pfi_kif *kif) { PF_RULES_WASSERT(); kif->pfik_rulerefs++; } void pfi_kif_unref(struct pfi_kif *kif) { PF_RULES_WASSERT(); KASSERT(kif->pfik_rulerefs > 0, ("%s: %p has zero refs", __func__, kif)); kif->pfik_rulerefs--; if (kif->pfik_rulerefs > 0) return; /* kif referencing an existing ifnet or group should exist. */ if (kif->pfik_ifp != NULL || kif->pfik_group != NULL || kif == V_pfi_all) return; RB_REMOVE(pfi_ifhead, &V_pfi_ifs, kif); kif->pfik_flags |= PFI_IFLAG_REFS; mtx_lock(&pfi_unlnkdkifs_mtx); LIST_INSERT_HEAD(&V_pfi_unlinked_kifs, kif, pfik_list); mtx_unlock(&pfi_unlnkdkifs_mtx); } void pfi_kif_purge(void) { struct pfi_kif *kif, *kif1; /* * Do naive mark-and-sweep garbage collecting of old kifs. * Reference flag is raised by pf_purge_expired_states(). */ mtx_lock(&pfi_unlnkdkifs_mtx); LIST_FOREACH_SAFE(kif, &V_pfi_unlinked_kifs, pfik_list, kif1) { if (!(kif->pfik_flags & PFI_IFLAG_REFS)) { LIST_REMOVE(kif, pfik_list); free(kif, PFI_MTYPE); } else kif->pfik_flags &= ~PFI_IFLAG_REFS; } mtx_unlock(&pfi_unlnkdkifs_mtx); } int pfi_kif_match(struct pfi_kif *rule_kif, struct pfi_kif *packet_kif) { struct ifg_list *p; if (rule_kif == NULL || rule_kif == packet_kif) return (1); if (rule_kif->pfik_group != NULL) /* XXXGL: locking? */ CK_STAILQ_FOREACH(p, &packet_kif->pfik_ifp->if_groups, ifgl_next) if (p->ifgl_group == rule_kif->pfik_group) return (1); return (0); } static void pfi_attach_ifnet(struct ifnet *ifp) { struct pfi_kif *kif; kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); V_pfi_update++; kif = pfi_kif_attach(kif, ifp->if_xname); kif->pfik_ifp = ifp; ifp->if_pf_kif = kif; pfi_kif_update(kif); PF_RULES_WUNLOCK(); } static void pfi_attach_ifgroup(struct ifg_group *ifg) { struct pfi_kif *kif; kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); V_pfi_update++; kif = pfi_kif_attach(kif, ifg->ifg_group); kif->pfik_group = ifg; ifg->ifg_pf_kif = kif; PF_RULES_WUNLOCK(); } int pfi_match_addr(struct pfi_dynaddr *dyn, struct pf_addr *a, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: switch (dyn->pfid_acnt4) { case 0: return (0); case 1: return (PF_MATCHA(0, &dyn->pfid_addr4, &dyn->pfid_mask4, a, AF_INET)); default: return (pfr_match_addr(dyn->pfid_kt, a, AF_INET)); } break; #endif /* INET */ #ifdef INET6 case AF_INET6: switch (dyn->pfid_acnt6) { case 0: return (0); case 1: return (PF_MATCHA(0, &dyn->pfid_addr6, &dyn->pfid_mask6, a, AF_INET6)); default: return (pfr_match_addr(dyn->pfid_kt, a, AF_INET6)); } break; #endif /* INET6 */ default: return (0); } } int pfi_dynaddr_setup(struct pf_addr_wrap *aw, sa_family_t af) { struct pfi_dynaddr *dyn; char tblname[PF_TABLE_NAME_SIZE]; struct pf_ruleset *ruleset = NULL; struct pfi_kif *kif; int rv = 0; PF_RULES_WASSERT(); KASSERT(aw->type == PF_ADDR_DYNIFTL, ("%s: type %u", __func__, aw->type)); KASSERT(aw->p.dyn == NULL, ("%s: dyn is %p", __func__, aw->p.dyn)); if ((dyn = malloc(sizeof(*dyn), PFI_MTYPE, M_NOWAIT | M_ZERO)) == NULL) return (ENOMEM); if ((kif = malloc(sizeof(*kif), PFI_MTYPE, M_NOWAIT)) == NULL) { free(dyn, PFI_MTYPE); return (ENOMEM); } if (!strcmp(aw->v.ifname, "self")) dyn->pfid_kif = pfi_kif_attach(kif, IFG_ALL); else dyn->pfid_kif = pfi_kif_attach(kif, aw->v.ifname); pfi_kif_ref(dyn->pfid_kif); dyn->pfid_net = pfi_unmask(&aw->v.a.mask); if (af == AF_INET && dyn->pfid_net == 32) dyn->pfid_net = 128; strlcpy(tblname, aw->v.ifname, sizeof(tblname)); if (aw->iflags & PFI_AFLAG_NETWORK) strlcat(tblname, ":network", sizeof(tblname)); if (aw->iflags & PFI_AFLAG_BROADCAST) strlcat(tblname, ":broadcast", sizeof(tblname)); if (aw->iflags & PFI_AFLAG_PEER) strlcat(tblname, ":peer", sizeof(tblname)); if (aw->iflags & PFI_AFLAG_NOALIAS) strlcat(tblname, ":0", sizeof(tblname)); if (dyn->pfid_net != 128) snprintf(tblname + strlen(tblname), sizeof(tblname) - strlen(tblname), "/%d", dyn->pfid_net); if ((ruleset = pf_find_or_create_ruleset(PF_RESERVED_ANCHOR)) == NULL) { rv = ENOMEM; goto _bad; } if ((dyn->pfid_kt = pfr_attach_table(ruleset, tblname)) == NULL) { rv = ENOMEM; goto _bad; } dyn->pfid_kt->pfrkt_flags |= PFR_TFLAG_ACTIVE; dyn->pfid_iflags = aw->iflags; dyn->pfid_af = af; TAILQ_INSERT_TAIL(&dyn->pfid_kif->pfik_dynaddrs, dyn, entry); aw->p.dyn = dyn; pfi_kif_update(dyn->pfid_kif); return (0); _bad: if (dyn->pfid_kt != NULL) pfr_detach_table(dyn->pfid_kt); if (ruleset != NULL) pf_remove_if_empty_ruleset(ruleset); if (dyn->pfid_kif != NULL) pfi_kif_unref(dyn->pfid_kif); free(dyn, PFI_MTYPE); return (rv); } static void pfi_kif_update(struct pfi_kif *kif) { struct ifg_list *ifgl; struct pfi_dynaddr *p; PF_RULES_WASSERT(); /* update all dynaddr */ TAILQ_FOREACH(p, &kif->pfik_dynaddrs, entry) pfi_dynaddr_update(p); /* again for all groups kif is member of */ if (kif->pfik_ifp != NULL) { IF_ADDR_RLOCK(kif->pfik_ifp); CK_STAILQ_FOREACH(ifgl, &kif->pfik_ifp->if_groups, ifgl_next) pfi_kif_update((struct pfi_kif *) ifgl->ifgl_group->ifg_pf_kif); IF_ADDR_RUNLOCK(kif->pfik_ifp); } } static void pfi_dynaddr_update(struct pfi_dynaddr *dyn) { struct pfi_kif *kif; struct pfr_ktable *kt; PF_RULES_WASSERT(); KASSERT(dyn && dyn->pfid_kif && dyn->pfid_kt, ("%s: bad argument", __func__)); kif = dyn->pfid_kif; kt = dyn->pfid_kt; if (kt->pfrkt_larg != V_pfi_update) { /* this table needs to be brought up-to-date */ pfi_table_update(kt, kif, dyn->pfid_net, dyn->pfid_iflags); kt->pfrkt_larg = V_pfi_update; } pfr_dynaddr_update(kt, dyn); } static void pfi_table_update(struct pfr_ktable *kt, struct pfi_kif *kif, int net, int flags) { int e, size2 = 0; struct ifg_member *ifgm; V_pfi_buffer_cnt = 0; if (kif->pfik_ifp != NULL) pfi_instance_add(kif->pfik_ifp, net, flags); else if (kif->pfik_group != NULL) { IFNET_RLOCK_NOSLEEP(); CK_STAILQ_FOREACH(ifgm, &kif->pfik_group->ifg_members, ifgm_next) pfi_instance_add(ifgm->ifgm_ifp, net, flags); IFNET_RUNLOCK_NOSLEEP(); } if ((e = pfr_set_addrs(&kt->pfrkt_t, V_pfi_buffer, V_pfi_buffer_cnt, &size2, NULL, NULL, NULL, 0, PFR_TFLAG_ALLMASK))) printf("%s: cannot set %d new addresses into table %s: %d\n", __func__, V_pfi_buffer_cnt, kt->pfrkt_name, e); } static void pfi_instance_add(struct ifnet *ifp, int net, int flags) { struct ifaddr *ia; int got4 = 0, got6 = 0; int net2, af; IF_ADDR_RLOCK(ifp); CK_STAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) { if (ia->ifa_addr == NULL) continue; af = ia->ifa_addr->sa_family; if (af != AF_INET && af != AF_INET6) continue; /* * XXX: For point-to-point interfaces, (ifname:0) and IPv4, * jump over addresses without a proper route to work * around a problem with ppp not fully removing the * address used during IPCP. */ if ((ifp->if_flags & IFF_POINTOPOINT) && !(ia->ifa_flags & IFA_ROUTE) && (flags & PFI_AFLAG_NOALIAS) && (af == AF_INET)) continue; if ((flags & PFI_AFLAG_BROADCAST) && af == AF_INET6) continue; if ((flags & PFI_AFLAG_BROADCAST) && !(ifp->if_flags & IFF_BROADCAST)) continue; if ((flags & PFI_AFLAG_PEER) && !(ifp->if_flags & IFF_POINTOPOINT)) continue; if ((flags & PFI_AFLAG_NETWORK) && af == AF_INET6 && IN6_IS_ADDR_LINKLOCAL( &((struct sockaddr_in6 *)ia->ifa_addr)->sin6_addr)) continue; if (flags & PFI_AFLAG_NOALIAS) { if (af == AF_INET && got4) continue; if (af == AF_INET6 && got6) continue; } if (af == AF_INET) got4 = 1; else if (af == AF_INET6) got6 = 1; net2 = net; if (net2 == 128 && (flags & PFI_AFLAG_NETWORK)) { if (af == AF_INET) net2 = pfi_unmask(&((struct sockaddr_in *) ia->ifa_netmask)->sin_addr); else if (af == AF_INET6) net2 = pfi_unmask(&((struct sockaddr_in6 *) ia->ifa_netmask)->sin6_addr); } if (af == AF_INET && net2 > 32) net2 = 32; if (flags & PFI_AFLAG_BROADCAST) pfi_address_add(ia->ifa_broadaddr, af, net2); else if (flags & PFI_AFLAG_PEER) pfi_address_add(ia->ifa_dstaddr, af, net2); else pfi_address_add(ia->ifa_addr, af, net2); } IF_ADDR_RUNLOCK(ifp); } static void pfi_address_add(struct sockaddr *sa, int af, int net) { struct pfr_addr *p; int i; if (V_pfi_buffer_cnt >= V_pfi_buffer_max) { int new_max = V_pfi_buffer_max * 2; if (new_max > PFI_BUFFER_MAX) { printf("%s: address buffer full (%d/%d)\n", __func__, V_pfi_buffer_cnt, PFI_BUFFER_MAX); return; } p = malloc(new_max * sizeof(*V_pfi_buffer), PFI_MTYPE, M_NOWAIT); if (p == NULL) { printf("%s: no memory to grow buffer (%d/%d)\n", __func__, V_pfi_buffer_cnt, PFI_BUFFER_MAX); return; } memcpy(p, V_pfi_buffer, V_pfi_buffer_max * sizeof(*V_pfi_buffer)); /* no need to zero buffer */ free(V_pfi_buffer, PFI_MTYPE); V_pfi_buffer = p; V_pfi_buffer_max = new_max; } if (af == AF_INET && net > 32) net = 128; p = V_pfi_buffer + V_pfi_buffer_cnt++; bzero(p, sizeof(*p)); p->pfra_af = af; p->pfra_net = net; if (af == AF_INET) p->pfra_ip4addr = ((struct sockaddr_in *)sa)->sin_addr; else if (af == AF_INET6) { p->pfra_ip6addr = ((struct sockaddr_in6 *)sa)->sin6_addr; if (IN6_IS_SCOPE_EMBED(&p->pfra_ip6addr)) p->pfra_ip6addr.s6_addr16[1] = 0; } /* mask network address bits */ if (net < 128) ((caddr_t)p)[p->pfra_net/8] &= ~(0xFF >> (p->pfra_net%8)); for (i = (p->pfra_net+7)/8; i < sizeof(p->pfra_u); i++) ((caddr_t)p)[i] = 0; } void pfi_dynaddr_remove(struct pfi_dynaddr *dyn) { KASSERT(dyn->pfid_kif != NULL, ("%s: null pfid_kif", __func__)); KASSERT(dyn->pfid_kt != NULL, ("%s: null pfid_kt", __func__)); TAILQ_REMOVE(&dyn->pfid_kif->pfik_dynaddrs, dyn, entry); pfi_kif_unref(dyn->pfid_kif); pfr_detach_table(dyn->pfid_kt); free(dyn, PFI_MTYPE); } void pfi_dynaddr_copyout(struct pf_addr_wrap *aw) { KASSERT(aw->type == PF_ADDR_DYNIFTL, ("%s: type %u", __func__, aw->type)); if (aw->p.dyn == NULL || aw->p.dyn->pfid_kif == NULL) return; aw->p.dyncnt = aw->p.dyn->pfid_acnt4 + aw->p.dyn->pfid_acnt6; } static int pfi_if_compare(struct pfi_kif *p, struct pfi_kif *q) { return (strncmp(p->pfik_name, q->pfik_name, IFNAMSIZ)); } void pfi_update_status(const char *name, struct pf_status *pfs) { struct pfi_kif *p; struct pfi_kif_cmp key; struct ifg_member p_member, *ifgm; CK_STAILQ_HEAD(, ifg_member) ifg_members; int i, j, k; strlcpy(key.pfik_name, name, sizeof(key.pfik_name)); p = RB_FIND(pfi_ifhead, &V_pfi_ifs, (struct pfi_kif *)&key); if (p == NULL) return; if (p->pfik_group != NULL) { bcopy(&p->pfik_group->ifg_members, &ifg_members, sizeof(ifg_members)); } else { /* build a temporary list for p only */ bzero(&p_member, sizeof(p_member)); p_member.ifgm_ifp = p->pfik_ifp; CK_STAILQ_INIT(&ifg_members); CK_STAILQ_INSERT_TAIL(&ifg_members, &p_member, ifgm_next); } if (pfs) { bzero(pfs->pcounters, sizeof(pfs->pcounters)); bzero(pfs->bcounters, sizeof(pfs->bcounters)); } CK_STAILQ_FOREACH(ifgm, &ifg_members, ifgm_next) { if (ifgm->ifgm_ifp == NULL || ifgm->ifgm_ifp->if_pf_kif == NULL) continue; p = (struct pfi_kif *)ifgm->ifgm_ifp->if_pf_kif; /* just clear statistics */ if (pfs == NULL) { bzero(p->pfik_packets, sizeof(p->pfik_packets)); bzero(p->pfik_bytes, sizeof(p->pfik_bytes)); p->pfik_tzero = time_second; continue; } for (i = 0; i < 2; i++) for (j = 0; j < 2; j++) for (k = 0; k < 2; k++) { pfs->pcounters[i][j][k] += p->pfik_packets[i][j][k]; pfs->bcounters[i][j] += p->pfik_bytes[i][j][k]; } } } void pfi_get_ifaces(const char *name, struct pfi_kif *buf, int *size) { struct pfi_kif *p, *nextp; int n = 0; for (p = RB_MIN(pfi_ifhead, &V_pfi_ifs); p; p = nextp) { nextp = RB_NEXT(pfi_ifhead, &V_pfi_ifs, p); if (pfi_skip_if(name, p)) continue; if (*size <= n++) break; if (!p->pfik_tzero) p->pfik_tzero = time_second; bcopy(p, buf++, sizeof(*buf)); nextp = RB_NEXT(pfi_ifhead, &V_pfi_ifs, p); } *size = n; } static int pfi_skip_if(const char *filter, struct pfi_kif *p) { int n; if (filter == NULL || !*filter) return (0); if (!strcmp(p->pfik_name, filter)) return (0); /* exact match */ n = strlen(filter); if (n < 1 || n >= IFNAMSIZ) return (1); /* sanity check */ if (filter[n-1] >= '0' && filter[n-1] <= '9') return (1); /* only do exact match in that case */ if (strncmp(p->pfik_name, filter, n)) return (1); /* prefix doesn't match */ return (p->pfik_name[n] < '0' || p->pfik_name[n] > '9'); } int pfi_set_flags(const char *name, int flags) { struct pfi_kif *p; RB_FOREACH(p, pfi_ifhead, &V_pfi_ifs) { if (pfi_skip_if(name, p)) continue; p->pfik_flags |= flags; } return (0); } int pfi_clear_flags(const char *name, int flags) { struct pfi_kif *p; RB_FOREACH(p, pfi_ifhead, &V_pfi_ifs) { if (pfi_skip_if(name, p)) continue; p->pfik_flags &= ~flags; } return (0); } /* from pf_print_state.c */ static int pfi_unmask(void *addr) { struct pf_addr *m = addr; int i = 31, j = 0, b = 0; u_int32_t tmp; while (j < 4 && m->addr32[j] == 0xffffffff) { b += 32; j++; } if (j < 4) { tmp = ntohl(m->addr32[j]); for (i = 31; tmp & (1 << i); --i) b++; } return (b); } static void pfi_attach_ifnet_event(void *arg __unused, struct ifnet *ifp) { if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } pfi_attach_ifnet(ifp); #ifdef ALTQ PF_RULES_WLOCK(); pf_altq_ifnet_event(ifp, 0); PF_RULES_WUNLOCK(); #endif } static void pfi_detach_ifnet_event(void *arg __unused, struct ifnet *ifp) { struct pfi_kif *kif = (struct pfi_kif *)ifp->if_pf_kif; if (kif == NULL) return; if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } PF_RULES_WLOCK(); V_pfi_update++; pfi_kif_update(kif); kif->pfik_ifp = NULL; ifp->if_pf_kif = NULL; #ifdef ALTQ pf_altq_ifnet_event(ifp, 1); #endif PF_RULES_WUNLOCK(); } static void pfi_attach_group_event(void *arg __unused, struct ifg_group *ifg) { if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } pfi_attach_ifgroup(ifg); } static void pfi_change_group_event(void *arg __unused, char *gname) { struct pfi_kif *kif; if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); V_pfi_update++; kif = pfi_kif_attach(kif, gname); pfi_kif_update(kif); PF_RULES_WUNLOCK(); } static void pfi_detach_group_event(void *arg __unused, struct ifg_group *ifg) { struct pfi_kif *kif = (struct pfi_kif *)ifg->ifg_pf_kif; if (kif == NULL) return; if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } PF_RULES_WLOCK(); V_pfi_update++; kif->pfik_group = NULL; ifg->ifg_pf_kif = NULL; PF_RULES_WUNLOCK(); } static void pfi_ifaddr_event(void *arg __unused, struct ifnet *ifp) { if (ifp->if_pf_kif == NULL) return; if (V_pf_vnet_active == 0) { /* Avoid teardown race in the least expensive way. */ return; } PF_RULES_WLOCK(); if (ifp && ifp->if_pf_kif) { V_pfi_update++; pfi_kif_update(ifp->if_pf_kif); } PF_RULES_WUNLOCK(); } Index: head/sys/netpfil/pf/pf_ioctl.c =================================================================== --- head/sys/netpfil/pf/pf_ioctl.c (revision 336675) +++ head/sys/netpfil/pf/pf_ioctl.c (revision 336676) @@ -1,4054 +1,4054 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * Copyright (c) 2002,2003 Henning Brauer * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * Effort sponsored in part by the Defense Advanced Research Projects * Agency (DARPA) and Air Force Research Laboratory, Air Force * Materiel Command, USAF, under agreement number F30602-01-2-0537. * * $OpenBSD: pf_ioctl.c,v 1.213 2009/02/15 21:46:12 mbalmer Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_bpf.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif /* INET6 */ #ifdef ALTQ #include #endif static struct pf_pool *pf_get_pool(char *, u_int32_t, u_int8_t, u_int32_t, u_int8_t, u_int8_t, u_int8_t); static void pf_mv_pool(struct pf_palist *, struct pf_palist *); static void pf_empty_pool(struct pf_palist *); static int pfioctl(struct cdev *, u_long, caddr_t, int, struct thread *); #ifdef ALTQ static int pf_begin_altq(u_int32_t *); static int pf_rollback_altq(u_int32_t); static int pf_commit_altq(u_int32_t); static int pf_enable_altq(struct pf_altq *); static int pf_disable_altq(struct pf_altq *); static u_int32_t pf_qname2qid(char *); static void pf_qid_unref(u_int32_t); #endif /* ALTQ */ static int pf_begin_rules(u_int32_t *, int, const char *); static int pf_rollback_rules(u_int32_t, int, char *); static int pf_setup_pfsync_matching(struct pf_ruleset *); static void pf_hash_rule(MD5_CTX *, struct pf_rule *); static void pf_hash_rule_addr(MD5_CTX *, struct pf_rule_addr *); static int pf_commit_rules(u_int32_t, int, char *); static int pf_addr_setup(struct pf_ruleset *, struct pf_addr_wrap *, sa_family_t); static void pf_addr_copyout(struct pf_addr_wrap *); VNET_DEFINE(struct pf_rule, pf_default_rule); #ifdef ALTQ -static VNET_DEFINE(int, pf_altq_running); +VNET_DEFINE_STATIC(int, pf_altq_running); #define V_pf_altq_running VNET(pf_altq_running) #endif #define TAGID_MAX 50000 struct pf_tagname { TAILQ_ENTRY(pf_tagname) entries; char name[PF_TAG_NAME_SIZE]; uint16_t tag; int ref; }; TAILQ_HEAD(pf_tags, pf_tagname); #define V_pf_tags VNET(pf_tags) VNET_DEFINE(struct pf_tags, pf_tags); #define V_pf_qids VNET(pf_qids) VNET_DEFINE(struct pf_tags, pf_qids); static MALLOC_DEFINE(M_PFTAG, "pf_tag", "pf(4) tag names"); static MALLOC_DEFINE(M_PFALTQ, "pf_altq", "pf(4) altq configuration db"); static MALLOC_DEFINE(M_PFRULE, "pf_rule", "pf(4) rules"); #if (PF_QNAME_SIZE != PF_TAG_NAME_SIZE) #error PF_QNAME_SIZE must be equal to PF_TAG_NAME_SIZE #endif static u_int16_t tagname2tag(struct pf_tags *, char *); static u_int16_t pf_tagname2tag(char *); static void tag_unref(struct pf_tags *, u_int16_t); #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x struct cdev *pf_dev; /* * XXX - These are new and need to be checked when moveing to a new version */ static void pf_clear_states(void); static int pf_clear_tables(void); static void pf_clear_srcnodes(struct pf_src_node *); static void pf_kill_srcnodes(struct pfioc_src_node_kill *); static void pf_tbladdr_copyout(struct pf_addr_wrap *); /* * Wrapper functions for pfil(9) hooks */ #ifdef INET static int pf_check_in(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp); static int pf_check_out(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp); #endif #ifdef INET6 static int pf_check6_in(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp); static int pf_check6_out(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp); #endif static int hook_pf(void); static int dehook_pf(void); static int shutdown_pf(void); static int pf_load(void); static void pf_unload(void); static struct cdevsw pf_cdevsw = { .d_ioctl = pfioctl, .d_name = PF_NAME, .d_version = D_VERSION, }; -static volatile VNET_DEFINE(int, pf_pfil_hooked); +volatile VNET_DEFINE_STATIC(int, pf_pfil_hooked); #define V_pf_pfil_hooked VNET(pf_pfil_hooked) /* * We need a flag that is neither hooked nor running to know when * the VNET is "valid". We primarily need this to control (global) * external event, e.g., eventhandlers. */ VNET_DEFINE(int, pf_vnet_active); #define V_pf_vnet_active VNET(pf_vnet_active) int pf_end_threads; struct proc *pf_purge_proc; struct rmlock pf_rules_lock; struct sx pf_ioctl_lock; struct sx pf_end_lock; /* pfsync */ pfsync_state_import_t *pfsync_state_import_ptr = NULL; pfsync_insert_state_t *pfsync_insert_state_ptr = NULL; pfsync_update_state_t *pfsync_update_state_ptr = NULL; pfsync_delete_state_t *pfsync_delete_state_ptr = NULL; pfsync_clear_states_t *pfsync_clear_states_ptr = NULL; pfsync_defer_t *pfsync_defer_ptr = NULL; /* pflog */ pflog_packet_t *pflog_packet_ptr = NULL; extern u_long pf_ioctl_maxcount; static void pfattach_vnet(void) { u_int32_t *my_timeout = V_pf_default_rule.timeout; pf_initialize(); pfr_initialize(); pfi_initialize_vnet(); pf_normalize_init(); V_pf_limits[PF_LIMIT_STATES].limit = PFSTATE_HIWAT; V_pf_limits[PF_LIMIT_SRC_NODES].limit = PFSNODE_HIWAT; RB_INIT(&V_pf_anchors); pf_init_ruleset(&pf_main_ruleset); /* default rule should never be garbage collected */ V_pf_default_rule.entries.tqe_prev = &V_pf_default_rule.entries.tqe_next; #ifdef PF_DEFAULT_TO_DROP V_pf_default_rule.action = PF_DROP; #else V_pf_default_rule.action = PF_PASS; #endif V_pf_default_rule.nr = -1; V_pf_default_rule.rtableid = -1; V_pf_default_rule.states_cur = counter_u64_alloc(M_WAITOK); V_pf_default_rule.states_tot = counter_u64_alloc(M_WAITOK); V_pf_default_rule.src_nodes = counter_u64_alloc(M_WAITOK); /* initialize default timeouts */ my_timeout[PFTM_TCP_FIRST_PACKET] = PFTM_TCP_FIRST_PACKET_VAL; my_timeout[PFTM_TCP_OPENING] = PFTM_TCP_OPENING_VAL; my_timeout[PFTM_TCP_ESTABLISHED] = PFTM_TCP_ESTABLISHED_VAL; my_timeout[PFTM_TCP_CLOSING] = PFTM_TCP_CLOSING_VAL; my_timeout[PFTM_TCP_FIN_WAIT] = PFTM_TCP_FIN_WAIT_VAL; my_timeout[PFTM_TCP_CLOSED] = PFTM_TCP_CLOSED_VAL; my_timeout[PFTM_UDP_FIRST_PACKET] = PFTM_UDP_FIRST_PACKET_VAL; my_timeout[PFTM_UDP_SINGLE] = PFTM_UDP_SINGLE_VAL; my_timeout[PFTM_UDP_MULTIPLE] = PFTM_UDP_MULTIPLE_VAL; my_timeout[PFTM_ICMP_FIRST_PACKET] = PFTM_ICMP_FIRST_PACKET_VAL; my_timeout[PFTM_ICMP_ERROR_REPLY] = PFTM_ICMP_ERROR_REPLY_VAL; my_timeout[PFTM_OTHER_FIRST_PACKET] = PFTM_OTHER_FIRST_PACKET_VAL; my_timeout[PFTM_OTHER_SINGLE] = PFTM_OTHER_SINGLE_VAL; my_timeout[PFTM_OTHER_MULTIPLE] = PFTM_OTHER_MULTIPLE_VAL; my_timeout[PFTM_FRAG] = PFTM_FRAG_VAL; my_timeout[PFTM_INTERVAL] = PFTM_INTERVAL_VAL; my_timeout[PFTM_SRC_NODE] = PFTM_SRC_NODE_VAL; my_timeout[PFTM_TS_DIFF] = PFTM_TS_DIFF_VAL; my_timeout[PFTM_ADAPTIVE_START] = PFSTATE_ADAPT_START; my_timeout[PFTM_ADAPTIVE_END] = PFSTATE_ADAPT_END; bzero(&V_pf_status, sizeof(V_pf_status)); V_pf_status.debug = PF_DEBUG_URGENT; V_pf_pfil_hooked = 0; /* XXX do our best to avoid a conflict */ V_pf_status.hostid = arc4random(); for (int i = 0; i < PFRES_MAX; i++) V_pf_status.counters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < LCNT_MAX; i++) V_pf_status.lcounters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < FCNT_MAX; i++) V_pf_status.fcounters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < SCNT_MAX; i++) V_pf_status.scounters[i] = counter_u64_alloc(M_WAITOK); if (swi_add(NULL, "pf send", pf_intr, curvnet, SWI_NET, INTR_MPSAFE, &V_pf_swi_cookie) != 0) /* XXXGL: leaked all above. */ return; } static struct pf_pool * pf_get_pool(char *anchor, u_int32_t ticket, u_int8_t rule_action, u_int32_t rule_number, u_int8_t r_last, u_int8_t active, u_int8_t check_ticket) { struct pf_ruleset *ruleset; struct pf_rule *rule; int rs_num; ruleset = pf_find_ruleset(anchor); if (ruleset == NULL) return (NULL); rs_num = pf_get_ruleset_number(rule_action); if (rs_num >= PF_RULESET_MAX) return (NULL); if (active) { if (check_ticket && ticket != ruleset->rules[rs_num].active.ticket) return (NULL); if (r_last) rule = TAILQ_LAST(ruleset->rules[rs_num].active.ptr, pf_rulequeue); else rule = TAILQ_FIRST(ruleset->rules[rs_num].active.ptr); } else { if (check_ticket && ticket != ruleset->rules[rs_num].inactive.ticket) return (NULL); if (r_last) rule = TAILQ_LAST(ruleset->rules[rs_num].inactive.ptr, pf_rulequeue); else rule = TAILQ_FIRST(ruleset->rules[rs_num].inactive.ptr); } if (!r_last) { while ((rule != NULL) && (rule->nr != rule_number)) rule = TAILQ_NEXT(rule, entries); } if (rule == NULL) return (NULL); return (&rule->rpool); } static void pf_mv_pool(struct pf_palist *poola, struct pf_palist *poolb) { struct pf_pooladdr *mv_pool_pa; while ((mv_pool_pa = TAILQ_FIRST(poola)) != NULL) { TAILQ_REMOVE(poola, mv_pool_pa, entries); TAILQ_INSERT_TAIL(poolb, mv_pool_pa, entries); } } static void pf_empty_pool(struct pf_palist *poola) { struct pf_pooladdr *pa; while ((pa = TAILQ_FIRST(poola)) != NULL) { switch (pa->addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(pa->addr.p.dyn); break; case PF_ADDR_TABLE: /* XXX: this could be unfinished pooladdr on pabuf */ if (pa->addr.p.tbl != NULL) pfr_detach_table(pa->addr.p.tbl); break; } if (pa->kif) pfi_kif_unref(pa->kif); TAILQ_REMOVE(poola, pa, entries); free(pa, M_PFRULE); } } static void pf_unlink_rule(struct pf_rulequeue *rulequeue, struct pf_rule *rule) { PF_RULES_WASSERT(); TAILQ_REMOVE(rulequeue, rule, entries); PF_UNLNKDRULES_LOCK(); rule->rule_flag |= PFRULE_REFS; TAILQ_INSERT_TAIL(&V_pf_unlinked_rules, rule, entries); PF_UNLNKDRULES_UNLOCK(); } void pf_free_rule(struct pf_rule *rule) { PF_RULES_WASSERT(); if (rule->tag) tag_unref(&V_pf_tags, rule->tag); if (rule->match_tag) tag_unref(&V_pf_tags, rule->match_tag); #ifdef ALTQ if (rule->pqid != rule->qid) pf_qid_unref(rule->pqid); pf_qid_unref(rule->qid); #endif switch (rule->src.addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(rule->src.addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(rule->src.addr.p.tbl); break; } switch (rule->dst.addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(rule->dst.addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(rule->dst.addr.p.tbl); break; } if (rule->overload_tbl) pfr_detach_table(rule->overload_tbl); if (rule->kif) pfi_kif_unref(rule->kif); pf_anchor_remove(rule); pf_empty_pool(&rule->rpool.list); counter_u64_free(rule->states_cur); counter_u64_free(rule->states_tot); counter_u64_free(rule->src_nodes); free(rule, M_PFRULE); } static u_int16_t tagname2tag(struct pf_tags *head, char *tagname) { struct pf_tagname *tag, *p = NULL; u_int16_t new_tagid = 1; PF_RULES_WASSERT(); TAILQ_FOREACH(tag, head, entries) if (strcmp(tagname, tag->name) == 0) { tag->ref++; return (tag->tag); } /* * to avoid fragmentation, we do a linear search from the beginning * and take the first free slot we find. if there is none or the list * is empty, append a new entry at the end. */ /* new entry */ if (!TAILQ_EMPTY(head)) for (p = TAILQ_FIRST(head); p != NULL && p->tag == new_tagid; p = TAILQ_NEXT(p, entries)) new_tagid = p->tag + 1; if (new_tagid > TAGID_MAX) return (0); /* allocate and fill new struct pf_tagname */ tag = malloc(sizeof(*tag), M_PFTAG, M_NOWAIT|M_ZERO); if (tag == NULL) return (0); strlcpy(tag->name, tagname, sizeof(tag->name)); tag->tag = new_tagid; tag->ref++; if (p != NULL) /* insert new entry before p */ TAILQ_INSERT_BEFORE(p, tag, entries); else /* either list empty or no free slot in between */ TAILQ_INSERT_TAIL(head, tag, entries); return (tag->tag); } static void tag_unref(struct pf_tags *head, u_int16_t tag) { struct pf_tagname *p, *next; PF_RULES_WASSERT(); for (p = TAILQ_FIRST(head); p != NULL; p = next) { next = TAILQ_NEXT(p, entries); if (tag == p->tag) { if (--p->ref == 0) { TAILQ_REMOVE(head, p, entries); free(p, M_PFTAG); } break; } } } static u_int16_t pf_tagname2tag(char *tagname) { return (tagname2tag(&V_pf_tags, tagname)); } #ifdef ALTQ static u_int32_t pf_qname2qid(char *qname) { return ((u_int32_t)tagname2tag(&V_pf_qids, qname)); } static void pf_qid_unref(u_int32_t qid) { tag_unref(&V_pf_qids, (u_int16_t)qid); } static int pf_begin_altq(u_int32_t *ticket) { struct pf_altq *altq; int error = 0; PF_RULES_WASSERT(); /* Purge the old altq list */ while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ error = altq_remove(altq); } else pf_qid_unref(altq->qid); free(altq, M_PFALTQ); } if (error) return (error); *ticket = ++V_ticket_altqs_inactive; V_altqs_inactive_open = 1; return (0); } static int pf_rollback_altq(u_int32_t ticket) { struct pf_altq *altq; int error = 0; PF_RULES_WASSERT(); if (!V_altqs_inactive_open || ticket != V_ticket_altqs_inactive) return (0); /* Purge the old altq list */ while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ error = altq_remove(altq); } else pf_qid_unref(altq->qid); free(altq, M_PFALTQ); } V_altqs_inactive_open = 0; return (error); } static int pf_commit_altq(u_int32_t ticket) { struct pf_altqqueue *old_altqs; struct pf_altq *altq; int err, error = 0; PF_RULES_WASSERT(); if (!V_altqs_inactive_open || ticket != V_ticket_altqs_inactive) return (EBUSY); /* swap altqs, keep the old. */ old_altqs = V_pf_altqs_active; V_pf_altqs_active = V_pf_altqs_inactive; V_pf_altqs_inactive = old_altqs; V_ticket_altqs_active = V_ticket_altqs_inactive; /* Attach new disciplines */ TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* attach the discipline */ error = altq_pfattach(altq); if (error == 0 && V_pf_altq_running) error = pf_enable_altq(altq); if (error != 0) return (error); } } /* Purge the old altq list */ while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ if (V_pf_altq_running) error = pf_disable_altq(altq); err = altq_pfdetach(altq); if (err != 0 && error == 0) error = err; err = altq_remove(altq); if (err != 0 && error == 0) error = err; } else pf_qid_unref(altq->qid); free(altq, M_PFALTQ); } V_altqs_inactive_open = 0; return (error); } static int pf_enable_altq(struct pf_altq *altq) { struct ifnet *ifp; struct tb_profile tb; int error = 0; if ((ifp = ifunit(altq->ifname)) == NULL) return (EINVAL); if (ifp->if_snd.altq_type != ALTQT_NONE) error = altq_enable(&ifp->if_snd); /* set tokenbucket regulator */ if (error == 0 && ifp != NULL && ALTQ_IS_ENABLED(&ifp->if_snd)) { tb.rate = altq->ifbandwidth; tb.depth = altq->tbrsize; error = tbr_set(&ifp->if_snd, &tb); } return (error); } static int pf_disable_altq(struct pf_altq *altq) { struct ifnet *ifp; struct tb_profile tb; int error; if ((ifp = ifunit(altq->ifname)) == NULL) return (EINVAL); /* * when the discipline is no longer referenced, it was overridden * by a new one. if so, just return. */ if (altq->altq_disc != ifp->if_snd.altq_disc) return (0); error = altq_disable(&ifp->if_snd); if (error == 0) { /* clear tokenbucket regulator */ tb.rate = 0; error = tbr_set(&ifp->if_snd, &tb); } return (error); } void pf_altq_ifnet_event(struct ifnet *ifp, int remove) { struct ifnet *ifp1; struct pf_altq *a1, *a2, *a3; u_int32_t ticket; int error = 0; /* Interrupt userland queue modifications */ if (V_altqs_inactive_open) pf_rollback_altq(V_ticket_altqs_inactive); /* Start new altq ruleset */ if (pf_begin_altq(&ticket)) return; /* Copy the current active set */ TAILQ_FOREACH(a1, V_pf_altqs_active, entries) { a2 = malloc(sizeof(*a2), M_PFALTQ, M_NOWAIT); if (a2 == NULL) { error = ENOMEM; break; } bcopy(a1, a2, sizeof(struct pf_altq)); if (a2->qname[0] != 0) { if ((a2->qid = pf_qname2qid(a2->qname)) == 0) { error = EBUSY; free(a2, M_PFALTQ); break; } a2->altq_disc = NULL; TAILQ_FOREACH(a3, V_pf_altqs_inactive, entries) { if (strncmp(a3->ifname, a2->ifname, IFNAMSIZ) == 0 && a3->qname[0] == 0) { a2->altq_disc = a3->altq_disc; break; } } } /* Deactivate the interface in question */ a2->local_flags &= ~PFALTQ_FLAG_IF_REMOVED; if ((ifp1 = ifunit(a2->ifname)) == NULL || (remove && ifp1 == ifp)) { a2->local_flags |= PFALTQ_FLAG_IF_REMOVED; } else { error = altq_add(a2); if (ticket != V_ticket_altqs_inactive) error = EBUSY; if (error) { free(a2, M_PFALTQ); break; } } TAILQ_INSERT_TAIL(V_pf_altqs_inactive, a2, entries); } if (error != 0) pf_rollback_altq(ticket); else pf_commit_altq(ticket); } #endif /* ALTQ */ static int pf_begin_rules(u_int32_t *ticket, int rs_num, const char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_or_create_ruleset(anchor); if (rs == NULL) return (EINVAL); while ((rule = TAILQ_FIRST(rs->rules[rs_num].inactive.ptr)) != NULL) { pf_unlink_rule(rs->rules[rs_num].inactive.ptr, rule); rs->rules[rs_num].inactive.rcount--; } *ticket = ++rs->rules[rs_num].inactive.ticket; rs->rules[rs_num].inactive.open = 1; return (0); } static int pf_rollback_rules(u_int32_t ticket, int rs_num, char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_ruleset(anchor); if (rs == NULL || !rs->rules[rs_num].inactive.open || rs->rules[rs_num].inactive.ticket != ticket) return (0); while ((rule = TAILQ_FIRST(rs->rules[rs_num].inactive.ptr)) != NULL) { pf_unlink_rule(rs->rules[rs_num].inactive.ptr, rule); rs->rules[rs_num].inactive.rcount--; } rs->rules[rs_num].inactive.open = 0; return (0); } #define PF_MD5_UPD(st, elm) \ MD5Update(ctx, (u_int8_t *) &(st)->elm, sizeof((st)->elm)) #define PF_MD5_UPD_STR(st, elm) \ MD5Update(ctx, (u_int8_t *) (st)->elm, strlen((st)->elm)) #define PF_MD5_UPD_HTONL(st, elm, stor) do { \ (stor) = htonl((st)->elm); \ MD5Update(ctx, (u_int8_t *) &(stor), sizeof(u_int32_t));\ } while (0) #define PF_MD5_UPD_HTONS(st, elm, stor) do { \ (stor) = htons((st)->elm); \ MD5Update(ctx, (u_int8_t *) &(stor), sizeof(u_int16_t));\ } while (0) static void pf_hash_rule_addr(MD5_CTX *ctx, struct pf_rule_addr *pfr) { PF_MD5_UPD(pfr, addr.type); switch (pfr->addr.type) { case PF_ADDR_DYNIFTL: PF_MD5_UPD(pfr, addr.v.ifname); PF_MD5_UPD(pfr, addr.iflags); break; case PF_ADDR_TABLE: PF_MD5_UPD(pfr, addr.v.tblname); break; case PF_ADDR_ADDRMASK: /* XXX ignore af? */ PF_MD5_UPD(pfr, addr.v.a.addr.addr32); PF_MD5_UPD(pfr, addr.v.a.mask.addr32); break; } PF_MD5_UPD(pfr, port[0]); PF_MD5_UPD(pfr, port[1]); PF_MD5_UPD(pfr, neg); PF_MD5_UPD(pfr, port_op); } static void pf_hash_rule(MD5_CTX *ctx, struct pf_rule *rule) { u_int16_t x; u_int32_t y; pf_hash_rule_addr(ctx, &rule->src); pf_hash_rule_addr(ctx, &rule->dst); PF_MD5_UPD_STR(rule, label); PF_MD5_UPD_STR(rule, ifname); PF_MD5_UPD_STR(rule, match_tagname); PF_MD5_UPD_HTONS(rule, match_tag, x); /* dup? */ PF_MD5_UPD_HTONL(rule, os_fingerprint, y); PF_MD5_UPD_HTONL(rule, prob, y); PF_MD5_UPD_HTONL(rule, uid.uid[0], y); PF_MD5_UPD_HTONL(rule, uid.uid[1], y); PF_MD5_UPD(rule, uid.op); PF_MD5_UPD_HTONL(rule, gid.gid[0], y); PF_MD5_UPD_HTONL(rule, gid.gid[1], y); PF_MD5_UPD(rule, gid.op); PF_MD5_UPD_HTONL(rule, rule_flag, y); PF_MD5_UPD(rule, action); PF_MD5_UPD(rule, direction); PF_MD5_UPD(rule, af); PF_MD5_UPD(rule, quick); PF_MD5_UPD(rule, ifnot); PF_MD5_UPD(rule, match_tag_not); PF_MD5_UPD(rule, natpass); PF_MD5_UPD(rule, keep_state); PF_MD5_UPD(rule, proto); PF_MD5_UPD(rule, type); PF_MD5_UPD(rule, code); PF_MD5_UPD(rule, flags); PF_MD5_UPD(rule, flagset); PF_MD5_UPD(rule, allow_opts); PF_MD5_UPD(rule, rt); PF_MD5_UPD(rule, tos); } static int pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule, **old_array; struct pf_rulequeue *old_rules; int error; u_int32_t old_rcount; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_ruleset(anchor); if (rs == NULL || !rs->rules[rs_num].inactive.open || ticket != rs->rules[rs_num].inactive.ticket) return (EBUSY); /* Calculate checksum for the main ruleset */ if (rs == &pf_main_ruleset) { error = pf_setup_pfsync_matching(rs); if (error != 0) return (error); } /* Swap rules, keep the old. */ old_rules = rs->rules[rs_num].active.ptr; old_rcount = rs->rules[rs_num].active.rcount; old_array = rs->rules[rs_num].active.ptr_array; rs->rules[rs_num].active.ptr = rs->rules[rs_num].inactive.ptr; rs->rules[rs_num].active.ptr_array = rs->rules[rs_num].inactive.ptr_array; rs->rules[rs_num].active.rcount = rs->rules[rs_num].inactive.rcount; rs->rules[rs_num].inactive.ptr = old_rules; rs->rules[rs_num].inactive.ptr_array = old_array; rs->rules[rs_num].inactive.rcount = old_rcount; rs->rules[rs_num].active.ticket = rs->rules[rs_num].inactive.ticket; pf_calc_skip_steps(rs->rules[rs_num].active.ptr); /* Purge the old rule list. */ while ((rule = TAILQ_FIRST(old_rules)) != NULL) pf_unlink_rule(old_rules, rule); if (rs->rules[rs_num].inactive.ptr_array) free(rs->rules[rs_num].inactive.ptr_array, M_TEMP); rs->rules[rs_num].inactive.ptr_array = NULL; rs->rules[rs_num].inactive.rcount = 0; rs->rules[rs_num].inactive.open = 0; pf_remove_if_empty_ruleset(rs); return (0); } static int pf_setup_pfsync_matching(struct pf_ruleset *rs) { MD5_CTX ctx; struct pf_rule *rule; int rs_cnt; u_int8_t digest[PF_MD5_DIGEST_LENGTH]; MD5Init(&ctx); for (rs_cnt = 0; rs_cnt < PF_RULESET_MAX; rs_cnt++) { /* XXX PF_RULESET_SCRUB as well? */ if (rs_cnt == PF_RULESET_SCRUB) continue; if (rs->rules[rs_cnt].inactive.ptr_array) free(rs->rules[rs_cnt].inactive.ptr_array, M_TEMP); rs->rules[rs_cnt].inactive.ptr_array = NULL; if (rs->rules[rs_cnt].inactive.rcount) { rs->rules[rs_cnt].inactive.ptr_array = malloc(sizeof(caddr_t) * rs->rules[rs_cnt].inactive.rcount, M_TEMP, M_NOWAIT); if (!rs->rules[rs_cnt].inactive.ptr_array) return (ENOMEM); } TAILQ_FOREACH(rule, rs->rules[rs_cnt].inactive.ptr, entries) { pf_hash_rule(&ctx, rule); (rs->rules[rs_cnt].inactive.ptr_array)[rule->nr] = rule; } } MD5Final(digest, &ctx); memcpy(V_pf_status.pf_chksum, digest, sizeof(V_pf_status.pf_chksum)); return (0); } static int pf_addr_setup(struct pf_ruleset *ruleset, struct pf_addr_wrap *addr, sa_family_t af) { int error = 0; switch (addr->type) { case PF_ADDR_TABLE: addr->p.tbl = pfr_attach_table(ruleset, addr->v.tblname); if (addr->p.tbl == NULL) error = ENOMEM; break; case PF_ADDR_DYNIFTL: error = pfi_dynaddr_setup(addr, af); break; } return (error); } static void pf_addr_copyout(struct pf_addr_wrap *addr) { switch (addr->type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_copyout(addr); break; case PF_ADDR_TABLE: pf_tbladdr_copyout(addr); break; } } static int pfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) { int error = 0; PF_RULES_RLOCK_TRACKER; /* XXX keep in sync with switch() below */ if (securelevel_gt(td->td_ucred, 2)) switch (cmd) { case DIOCGETRULES: case DIOCGETRULE: case DIOCGETADDRS: case DIOCGETADDR: case DIOCGETSTATE: case DIOCSETSTATUSIF: case DIOCGETSTATUS: case DIOCCLRSTATUS: case DIOCNATLOOK: case DIOCSETDEBUG: case DIOCGETSTATES: case DIOCGETTIMEOUT: case DIOCCLRRULECTRS: case DIOCGETLIMIT: case DIOCGETALTQS: case DIOCGETALTQ: case DIOCGETQSTATS: case DIOCGETRULESETS: case DIOCGETRULESET: case DIOCRGETTABLES: case DIOCRGETTSTATS: case DIOCRCLRTSTATS: case DIOCRCLRADDRS: case DIOCRADDADDRS: case DIOCRDELADDRS: case DIOCRSETADDRS: case DIOCRGETADDRS: case DIOCRGETASTATS: case DIOCRCLRASTATS: case DIOCRTSTADDRS: case DIOCOSFPGET: case DIOCGETSRCNODES: case DIOCCLRSRCNODES: case DIOCIGETIFACES: case DIOCGIFSPEED: case DIOCSETIFFLAG: case DIOCCLRIFFLAG: break; case DIOCRCLRTABLES: case DIOCRADDTABLES: case DIOCRDELTABLES: case DIOCRSETTFLAGS: if (((struct pfioc_table *)addr)->pfrio_flags & PFR_FLAG_DUMMY) break; /* dummy operation ok */ return (EPERM); default: return (EPERM); } if (!(flags & FWRITE)) switch (cmd) { case DIOCGETRULES: case DIOCGETADDRS: case DIOCGETADDR: case DIOCGETSTATE: case DIOCGETSTATUS: case DIOCGETSTATES: case DIOCGETTIMEOUT: case DIOCGETLIMIT: case DIOCGETALTQS: case DIOCGETALTQ: case DIOCGETQSTATS: case DIOCGETRULESETS: case DIOCGETRULESET: case DIOCNATLOOK: case DIOCRGETTABLES: case DIOCRGETTSTATS: case DIOCRGETADDRS: case DIOCRGETASTATS: case DIOCRTSTADDRS: case DIOCOSFPGET: case DIOCGETSRCNODES: case DIOCIGETIFACES: case DIOCGIFSPEED: break; case DIOCRCLRTABLES: case DIOCRADDTABLES: case DIOCRDELTABLES: case DIOCRCLRTSTATS: case DIOCRCLRADDRS: case DIOCRADDADDRS: case DIOCRDELADDRS: case DIOCRSETADDRS: case DIOCRSETTFLAGS: if (((struct pfioc_table *)addr)->pfrio_flags & PFR_FLAG_DUMMY) { flags |= FWRITE; /* need write lock for dummy */ break; /* dummy operation ok */ } return (EACCES); case DIOCGETRULE: if (((struct pfioc_rule *)addr)->action == PF_GET_CLR_CNTR) return (EACCES); break; default: return (EACCES); } CURVNET_SET(TD_TO_VNET(td)); switch (cmd) { case DIOCSTART: sx_xlock(&pf_ioctl_lock); if (V_pf_status.running) error = EEXIST; else { int cpu; error = hook_pf(); if (error) { DPFPRINTF(PF_DEBUG_MISC, ("pf: pfil registration failed\n")); break; } V_pf_status.running = 1; V_pf_status.since = time_second; CPU_FOREACH(cpu) V_pf_stateid[cpu] = time_second; DPFPRINTF(PF_DEBUG_MISC, ("pf: started\n")); } break; case DIOCSTOP: sx_xlock(&pf_ioctl_lock); if (!V_pf_status.running) error = ENOENT; else { V_pf_status.running = 0; error = dehook_pf(); if (error) { V_pf_status.running = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: pfil unregistration failed\n")); } V_pf_status.since = time_second; DPFPRINTF(PF_DEBUG_MISC, ("pf: stopped\n")); } break; case DIOCADDRULE: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *rule, *tail; struct pf_pooladdr *pa; struct pfi_kif *kif = NULL; int rs_num; if (pr->rule.return_icmp >> 8 > ICMP_MAXTYPE) { error = EINVAL; break; } #ifndef INET if (pr->rule.af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pr->rule.af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ rule = malloc(sizeof(*rule), M_PFRULE, M_WAITOK); bcopy(&pr->rule, rule, sizeof(struct pf_rule)); if (rule->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); rule->states_cur = counter_u64_alloc(M_WAITOK); rule->states_tot = counter_u64_alloc(M_WAITOK); rule->src_nodes = counter_u64_alloc(M_WAITOK); rule->cuid = td->td_ucred->cr_ruid; rule->cpid = td->td_proc ? td->td_proc->p_pid : 0; TAILQ_INIT(&rule->rpool.list); #define ERROUT(x) { error = (x); goto DIOCADDRULE_error; } PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) ERROUT(EINVAL); rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) ERROUT(EINVAL); if (pr->ticket != ruleset->rules[rs_num].inactive.ticket) { DPFPRINTF(PF_DEBUG_MISC, ("ticket: %d != [%d]%d\n", pr->ticket, rs_num, ruleset->rules[rs_num].inactive.ticket)); ERROUT(EBUSY); } if (pr->pool_ticket != V_ticket_pabuf) { DPFPRINTF(PF_DEBUG_MISC, ("pool_ticket: %d != %d\n", pr->pool_ticket, V_ticket_pabuf)); ERROUT(EBUSY); } tail = TAILQ_LAST(ruleset->rules[rs_num].inactive.ptr, pf_rulequeue); if (tail) rule->nr = tail->nr + 1; else rule->nr = 0; if (rule->ifname[0]) { rule->kif = pfi_kif_attach(kif, rule->ifname); pfi_kif_ref(rule->kif); } else rule->kif = NULL; if (rule->rtableid > 0 && rule->rtableid >= rt_numfibs) error = EBUSY; #ifdef ALTQ /* set queue IDs */ if (rule->qname[0] != 0) { if ((rule->qid = pf_qname2qid(rule->qname)) == 0) error = EBUSY; else if (rule->pqname[0] != 0) { if ((rule->pqid = pf_qname2qid(rule->pqname)) == 0) error = EBUSY; } else rule->pqid = rule->qid; } #endif if (rule->tagname[0]) if ((rule->tag = pf_tagname2tag(rule->tagname)) == 0) error = EBUSY; if (rule->match_tagname[0]) if ((rule->match_tag = pf_tagname2tag(rule->match_tagname)) == 0) error = EBUSY; if (rule->rt && !rule->direction) error = EINVAL; if (!rule->log) rule->logif = 0; if (rule->logif >= PFLOGIFS_MAX) error = EINVAL; if (pf_addr_setup(ruleset, &rule->src.addr, rule->af)) error = ENOMEM; if (pf_addr_setup(ruleset, &rule->dst.addr, rule->af)) error = ENOMEM; if (pf_anchor_setup(rule, ruleset, pr->anchor_call)) error = EINVAL; if (rule->scrub_flags & PFSTATE_SETPRIO && (rule->set_prio[0] > PF_PRIO_MAX || rule->set_prio[1] > PF_PRIO_MAX)) error = EINVAL; TAILQ_FOREACH(pa, &V_pf_pabuf, entries) if (pa->addr.type == PF_ADDR_TABLE) { pa->addr.p.tbl = pfr_attach_table(ruleset, pa->addr.v.tblname); if (pa->addr.p.tbl == NULL) error = ENOMEM; } rule->overload_tbl = NULL; if (rule->overload_tblname[0]) { if ((rule->overload_tbl = pfr_attach_table(ruleset, rule->overload_tblname)) == NULL) error = EINVAL; else rule->overload_tbl->pfrkt_flags |= PFR_TFLAG_ACTIVE; } pf_mv_pool(&V_pf_pabuf, &rule->rpool.list); if (((((rule->action == PF_NAT) || (rule->action == PF_RDR) || (rule->action == PF_BINAT)) && rule->anchor == NULL) || (rule->rt > PF_NOPFROUTE)) && (TAILQ_FIRST(&rule->rpool.list) == NULL)) error = EINVAL; if (error) { pf_free_rule(rule); PF_RULES_WUNLOCK(); break; } rule->rpool.cur = TAILQ_FIRST(&rule->rpool.list); rule->evaluations = rule->packets[0] = rule->packets[1] = rule->bytes[0] = rule->bytes[1] = 0; TAILQ_INSERT_TAIL(ruleset->rules[rs_num].inactive.ptr, rule, entries); ruleset->rules[rs_num].inactive.rcount++; PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCADDRULE_error: PF_RULES_WUNLOCK(); counter_u64_free(rule->states_cur); counter_u64_free(rule->states_tot); counter_u64_free(rule->src_nodes); free(rule, M_PFRULE); if (kif) free(kif, PFI_MTYPE); break; } case DIOCGETRULES: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *tail; int rs_num; PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); error = EINVAL; break; } tail = TAILQ_LAST(ruleset->rules[rs_num].active.ptr, pf_rulequeue); if (tail) pr->nr = tail->nr + 1; else pr->nr = 0; pr->ticket = ruleset->rules[rs_num].active.ticket; PF_RULES_WUNLOCK(); break; } case DIOCGETRULE: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *rule; int rs_num, i; PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); error = EINVAL; break; } if (pr->ticket != ruleset->rules[rs_num].active.ticket) { PF_RULES_WUNLOCK(); error = EBUSY; break; } rule = TAILQ_FIRST(ruleset->rules[rs_num].active.ptr); while ((rule != NULL) && (rule->nr != pr->nr)) rule = TAILQ_NEXT(rule, entries); if (rule == NULL) { PF_RULES_WUNLOCK(); error = EBUSY; break; } bcopy(rule, &pr->rule, sizeof(struct pf_rule)); pr->rule.u_states_cur = counter_u64_fetch(rule->states_cur); pr->rule.u_states_tot = counter_u64_fetch(rule->states_tot); pr->rule.u_src_nodes = counter_u64_fetch(rule->src_nodes); if (pf_anchor_copyout(ruleset, rule, pr)) { PF_RULES_WUNLOCK(); error = EBUSY; break; } pf_addr_copyout(&pr->rule.src.addr); pf_addr_copyout(&pr->rule.dst.addr); for (i = 0; i < PF_SKIP_COUNT; ++i) if (rule->skip[i].ptr == NULL) pr->rule.skip[i].nr = -1; else pr->rule.skip[i].nr = rule->skip[i].ptr->nr; if (pr->action == PF_GET_CLR_CNTR) { rule->evaluations = 0; rule->packets[0] = rule->packets[1] = 0; rule->bytes[0] = rule->bytes[1] = 0; counter_u64_zero(rule->states_tot); } PF_RULES_WUNLOCK(); break; } case DIOCCHANGERULE: { struct pfioc_rule *pcr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *oldrule = NULL, *newrule = NULL; struct pfi_kif *kif = NULL; struct pf_pooladdr *pa; u_int32_t nr = 0; int rs_num; if (pcr->action < PF_CHANGE_ADD_HEAD || pcr->action > PF_CHANGE_GET_TICKET) { error = EINVAL; break; } if (pcr->rule.return_icmp >> 8 > ICMP_MAXTYPE) { error = EINVAL; break; } if (pcr->action != PF_CHANGE_REMOVE) { #ifndef INET if (pcr->rule.af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pcr->rule.af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ newrule = malloc(sizeof(*newrule), M_PFRULE, M_WAITOK); bcopy(&pcr->rule, newrule, sizeof(struct pf_rule)); if (newrule->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); newrule->states_cur = counter_u64_alloc(M_WAITOK); newrule->states_tot = counter_u64_alloc(M_WAITOK); newrule->src_nodes = counter_u64_alloc(M_WAITOK); newrule->cuid = td->td_ucred->cr_ruid; newrule->cpid = td->td_proc ? td->td_proc->p_pid : 0; TAILQ_INIT(&newrule->rpool.list); } #define ERROUT(x) { error = (x); goto DIOCCHANGERULE_error; } PF_RULES_WLOCK(); if (!(pcr->action == PF_CHANGE_REMOVE || pcr->action == PF_CHANGE_GET_TICKET) && pcr->pool_ticket != V_ticket_pabuf) ERROUT(EBUSY); ruleset = pf_find_ruleset(pcr->anchor); if (ruleset == NULL) ERROUT(EINVAL); rs_num = pf_get_ruleset_number(pcr->rule.action); if (rs_num >= PF_RULESET_MAX) ERROUT(EINVAL); if (pcr->action == PF_CHANGE_GET_TICKET) { pcr->ticket = ++ruleset->rules[rs_num].active.ticket; ERROUT(0); } else if (pcr->ticket != ruleset->rules[rs_num].active.ticket) ERROUT(EINVAL); if (pcr->action != PF_CHANGE_REMOVE) { if (newrule->ifname[0]) { newrule->kif = pfi_kif_attach(kif, newrule->ifname); pfi_kif_ref(newrule->kif); } else newrule->kif = NULL; if (newrule->rtableid > 0 && newrule->rtableid >= rt_numfibs) error = EBUSY; #ifdef ALTQ /* set queue IDs */ if (newrule->qname[0] != 0) { if ((newrule->qid = pf_qname2qid(newrule->qname)) == 0) error = EBUSY; else if (newrule->pqname[0] != 0) { if ((newrule->pqid = pf_qname2qid(newrule->pqname)) == 0) error = EBUSY; } else newrule->pqid = newrule->qid; } #endif /* ALTQ */ if (newrule->tagname[0]) if ((newrule->tag = pf_tagname2tag(newrule->tagname)) == 0) error = EBUSY; if (newrule->match_tagname[0]) if ((newrule->match_tag = pf_tagname2tag( newrule->match_tagname)) == 0) error = EBUSY; if (newrule->rt && !newrule->direction) error = EINVAL; if (!newrule->log) newrule->logif = 0; if (newrule->logif >= PFLOGIFS_MAX) error = EINVAL; if (pf_addr_setup(ruleset, &newrule->src.addr, newrule->af)) error = ENOMEM; if (pf_addr_setup(ruleset, &newrule->dst.addr, newrule->af)) error = ENOMEM; if (pf_anchor_setup(newrule, ruleset, pcr->anchor_call)) error = EINVAL; TAILQ_FOREACH(pa, &V_pf_pabuf, entries) if (pa->addr.type == PF_ADDR_TABLE) { pa->addr.p.tbl = pfr_attach_table(ruleset, pa->addr.v.tblname); if (pa->addr.p.tbl == NULL) error = ENOMEM; } newrule->overload_tbl = NULL; if (newrule->overload_tblname[0]) { if ((newrule->overload_tbl = pfr_attach_table( ruleset, newrule->overload_tblname)) == NULL) error = EINVAL; else newrule->overload_tbl->pfrkt_flags |= PFR_TFLAG_ACTIVE; } pf_mv_pool(&V_pf_pabuf, &newrule->rpool.list); if (((((newrule->action == PF_NAT) || (newrule->action == PF_RDR) || (newrule->action == PF_BINAT) || (newrule->rt > PF_NOPFROUTE)) && !newrule->anchor)) && (TAILQ_FIRST(&newrule->rpool.list) == NULL)) error = EINVAL; if (error) { pf_free_rule(newrule); PF_RULES_WUNLOCK(); break; } newrule->rpool.cur = TAILQ_FIRST(&newrule->rpool.list); newrule->evaluations = 0; newrule->packets[0] = newrule->packets[1] = 0; newrule->bytes[0] = newrule->bytes[1] = 0; } pf_empty_pool(&V_pf_pabuf); if (pcr->action == PF_CHANGE_ADD_HEAD) oldrule = TAILQ_FIRST( ruleset->rules[rs_num].active.ptr); else if (pcr->action == PF_CHANGE_ADD_TAIL) oldrule = TAILQ_LAST( ruleset->rules[rs_num].active.ptr, pf_rulequeue); else { oldrule = TAILQ_FIRST( ruleset->rules[rs_num].active.ptr); while ((oldrule != NULL) && (oldrule->nr != pcr->nr)) oldrule = TAILQ_NEXT(oldrule, entries); if (oldrule == NULL) { if (newrule != NULL) pf_free_rule(newrule); PF_RULES_WUNLOCK(); error = EINVAL; break; } } if (pcr->action == PF_CHANGE_REMOVE) { pf_unlink_rule(ruleset->rules[rs_num].active.ptr, oldrule); ruleset->rules[rs_num].active.rcount--; } else { if (oldrule == NULL) TAILQ_INSERT_TAIL( ruleset->rules[rs_num].active.ptr, newrule, entries); else if (pcr->action == PF_CHANGE_ADD_HEAD || pcr->action == PF_CHANGE_ADD_BEFORE) TAILQ_INSERT_BEFORE(oldrule, newrule, entries); else TAILQ_INSERT_AFTER( ruleset->rules[rs_num].active.ptr, oldrule, newrule, entries); ruleset->rules[rs_num].active.rcount++; } nr = 0; TAILQ_FOREACH(oldrule, ruleset->rules[rs_num].active.ptr, entries) oldrule->nr = nr++; ruleset->rules[rs_num].active.ticket++; pf_calc_skip_steps(ruleset->rules[rs_num].active.ptr); pf_remove_if_empty_ruleset(ruleset); PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCCHANGERULE_error: PF_RULES_WUNLOCK(); if (newrule != NULL) { counter_u64_free(newrule->states_cur); counter_u64_free(newrule->states_tot); counter_u64_free(newrule->src_nodes); free(newrule, M_PFRULE); } if (kif != NULL) free(kif, PFI_MTYPE); break; } case DIOCCLRSTATES: { struct pf_state *s; struct pfioc_state_kill *psk = (struct pfioc_state_kill *)addr; u_int i, killed = 0; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock_DIOCCLRSTATES: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) if (!psk->psk_ifname[0] || !strcmp(psk->psk_ifname, s->kif->pfik_name)) { /* * Don't send out individual * delete messages. */ s->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(s, PF_ENTER_LOCKED); killed++; goto relock_DIOCCLRSTATES; } PF_HASHROW_UNLOCK(ih); } psk->psk_killed = killed; if (pfsync_clear_states_ptr != NULL) pfsync_clear_states_ptr(V_pf_status.hostid, psk->psk_ifname); break; } case DIOCKILLSTATES: { struct pf_state *s; struct pf_state_key *sk; struct pf_addr *srcaddr, *dstaddr; u_int16_t srcport, dstport; struct pfioc_state_kill *psk = (struct pfioc_state_kill *)addr; u_int i, killed = 0; if (psk->psk_pfcmp.id) { if (psk->psk_pfcmp.creatorid == 0) psk->psk_pfcmp.creatorid = V_pf_status.hostid; if ((s = pf_find_state_byid(psk->psk_pfcmp.id, psk->psk_pfcmp.creatorid))) { pf_unlink_state(s, PF_ENTER_LOCKED); psk->psk_killed = 1; } break; } for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock_DIOCKILLSTATES: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { sk = s->key[PF_SK_WIRE]; if (s->direction == PF_OUT) { srcaddr = &sk->addr[1]; dstaddr = &sk->addr[0]; srcport = sk->port[1]; dstport = sk->port[0]; } else { srcaddr = &sk->addr[0]; dstaddr = &sk->addr[1]; srcport = sk->port[0]; dstport = sk->port[1]; } if ((!psk->psk_af || sk->af == psk->psk_af) && (!psk->psk_proto || psk->psk_proto == sk->proto) && PF_MATCHA(psk->psk_src.neg, &psk->psk_src.addr.v.a.addr, &psk->psk_src.addr.v.a.mask, srcaddr, sk->af) && PF_MATCHA(psk->psk_dst.neg, &psk->psk_dst.addr.v.a.addr, &psk->psk_dst.addr.v.a.mask, dstaddr, sk->af) && (psk->psk_src.port_op == 0 || pf_match_port(psk->psk_src.port_op, psk->psk_src.port[0], psk->psk_src.port[1], srcport)) && (psk->psk_dst.port_op == 0 || pf_match_port(psk->psk_dst.port_op, psk->psk_dst.port[0], psk->psk_dst.port[1], dstport)) && (!psk->psk_label[0] || (s->rule.ptr->label[0] && !strcmp(psk->psk_label, s->rule.ptr->label))) && (!psk->psk_ifname[0] || !strcmp(psk->psk_ifname, s->kif->pfik_name))) { pf_unlink_state(s, PF_ENTER_LOCKED); killed++; goto relock_DIOCKILLSTATES; } } PF_HASHROW_UNLOCK(ih); } psk->psk_killed = killed; break; } case DIOCADDSTATE: { struct pfioc_state *ps = (struct pfioc_state *)addr; struct pfsync_state *sp = &ps->state; if (sp->timeout >= PFTM_MAX) { error = EINVAL; break; } if (pfsync_state_import_ptr != NULL) { PF_RULES_RLOCK(); error = pfsync_state_import_ptr(sp, PFSYNC_SI_IOCTL); PF_RULES_RUNLOCK(); } else error = EOPNOTSUPP; break; } case DIOCGETSTATE: { struct pfioc_state *ps = (struct pfioc_state *)addr; struct pf_state *s; s = pf_find_state_byid(ps->state.id, ps->state.creatorid); if (s == NULL) { error = ENOENT; break; } pfsync_state_export(&ps->state, s); PF_STATE_UNLOCK(s); break; } case DIOCGETSTATES: { struct pfioc_states *ps = (struct pfioc_states *)addr; struct pf_state *s; struct pfsync_state *pstore, *p; int i, nr; if (ps->ps_len == 0) { nr = uma_zone_get_cur(V_pf_state_z); ps->ps_len = sizeof(struct pfsync_state) * nr; break; } p = pstore = malloc(ps->ps_len, M_TEMP, M_WAITOK); nr = 0; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (s->timeout == PFTM_UNLINKED) continue; if ((nr+1) * sizeof(*p) > ps->ps_len) { PF_HASHROW_UNLOCK(ih); goto DIOCGETSTATES_full; } pfsync_state_export(p, s); p++; nr++; } PF_HASHROW_UNLOCK(ih); } DIOCGETSTATES_full: error = copyout(pstore, ps->ps_states, sizeof(struct pfsync_state) * nr); if (error) { free(pstore, M_TEMP); break; } ps->ps_len = sizeof(struct pfsync_state) * nr; free(pstore, M_TEMP); break; } case DIOCGETSTATUS: { struct pf_status *s = (struct pf_status *)addr; PF_RULES_RLOCK(); s->running = V_pf_status.running; s->since = V_pf_status.since; s->debug = V_pf_status.debug; s->hostid = V_pf_status.hostid; s->states = V_pf_status.states; s->src_nodes = V_pf_status.src_nodes; for (int i = 0; i < PFRES_MAX; i++) s->counters[i] = counter_u64_fetch(V_pf_status.counters[i]); for (int i = 0; i < LCNT_MAX; i++) s->lcounters[i] = counter_u64_fetch(V_pf_status.lcounters[i]); for (int i = 0; i < FCNT_MAX; i++) s->fcounters[i] = counter_u64_fetch(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) s->scounters[i] = counter_u64_fetch(V_pf_status.scounters[i]); bcopy(V_pf_status.ifname, s->ifname, IFNAMSIZ); bcopy(V_pf_status.pf_chksum, s->pf_chksum, PF_MD5_DIGEST_LENGTH); pfi_update_status(s->ifname, s); PF_RULES_RUNLOCK(); break; } case DIOCSETSTATUSIF: { struct pfioc_if *pi = (struct pfioc_if *)addr; if (pi->ifname[0] == 0) { bzero(V_pf_status.ifname, IFNAMSIZ); break; } PF_RULES_WLOCK(); strlcpy(V_pf_status.ifname, pi->ifname, IFNAMSIZ); PF_RULES_WUNLOCK(); break; } case DIOCCLRSTATUS: { PF_RULES_WLOCK(); for (int i = 0; i < PFRES_MAX; i++) counter_u64_zero(V_pf_status.counters[i]); for (int i = 0; i < FCNT_MAX; i++) counter_u64_zero(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) counter_u64_zero(V_pf_status.scounters[i]); for (int i = 0; i < LCNT_MAX; i++) counter_u64_zero(V_pf_status.lcounters[i]); V_pf_status.since = time_second; if (*V_pf_status.ifname) pfi_update_status(V_pf_status.ifname, NULL); PF_RULES_WUNLOCK(); break; } case DIOCNATLOOK: { struct pfioc_natlook *pnl = (struct pfioc_natlook *)addr; struct pf_state_key *sk; struct pf_state *state; struct pf_state_key_cmp key; int m = 0, direction = pnl->direction; int sidx, didx; /* NATLOOK src and dst are reversed, so reverse sidx/didx */ sidx = (direction == PF_IN) ? 1 : 0; didx = (direction == PF_IN) ? 0 : 1; if (!pnl->proto || PF_AZERO(&pnl->saddr, pnl->af) || PF_AZERO(&pnl->daddr, pnl->af) || ((pnl->proto == IPPROTO_TCP || pnl->proto == IPPROTO_UDP) && (!pnl->dport || !pnl->sport))) error = EINVAL; else { bzero(&key, sizeof(key)); key.af = pnl->af; key.proto = pnl->proto; PF_ACPY(&key.addr[sidx], &pnl->saddr, pnl->af); key.port[sidx] = pnl->sport; PF_ACPY(&key.addr[didx], &pnl->daddr, pnl->af); key.port[didx] = pnl->dport; state = pf_find_state_all(&key, direction, &m); if (m > 1) error = E2BIG; /* more than one state */ else if (state != NULL) { /* XXXGL: not locked read */ sk = state->key[sidx]; PF_ACPY(&pnl->rsaddr, &sk->addr[sidx], sk->af); pnl->rsport = sk->port[sidx]; PF_ACPY(&pnl->rdaddr, &sk->addr[didx], sk->af); pnl->rdport = sk->port[didx]; } else error = ENOENT; } break; } case DIOCSETTIMEOUT: { struct pfioc_tm *pt = (struct pfioc_tm *)addr; int old; if (pt->timeout < 0 || pt->timeout >= PFTM_MAX || pt->seconds < 0) { error = EINVAL; break; } PF_RULES_WLOCK(); old = V_pf_default_rule.timeout[pt->timeout]; if (pt->timeout == PFTM_INTERVAL && pt->seconds == 0) pt->seconds = 1; V_pf_default_rule.timeout[pt->timeout] = pt->seconds; if (pt->timeout == PFTM_INTERVAL && pt->seconds < old) wakeup(pf_purge_thread); pt->seconds = old; PF_RULES_WUNLOCK(); break; } case DIOCGETTIMEOUT: { struct pfioc_tm *pt = (struct pfioc_tm *)addr; if (pt->timeout < 0 || pt->timeout >= PFTM_MAX) { error = EINVAL; break; } PF_RULES_RLOCK(); pt->seconds = V_pf_default_rule.timeout[pt->timeout]; PF_RULES_RUNLOCK(); break; } case DIOCGETLIMIT: { struct pfioc_limit *pl = (struct pfioc_limit *)addr; if (pl->index < 0 || pl->index >= PF_LIMIT_MAX) { error = EINVAL; break; } PF_RULES_RLOCK(); pl->limit = V_pf_limits[pl->index].limit; PF_RULES_RUNLOCK(); break; } case DIOCSETLIMIT: { struct pfioc_limit *pl = (struct pfioc_limit *)addr; int old_limit; PF_RULES_WLOCK(); if (pl->index < 0 || pl->index >= PF_LIMIT_MAX || V_pf_limits[pl->index].zone == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } uma_zone_set_max(V_pf_limits[pl->index].zone, pl->limit); old_limit = V_pf_limits[pl->index].limit; V_pf_limits[pl->index].limit = pl->limit; pl->limit = old_limit; PF_RULES_WUNLOCK(); break; } case DIOCSETDEBUG: { u_int32_t *level = (u_int32_t *)addr; PF_RULES_WLOCK(); V_pf_status.debug = *level; PF_RULES_WUNLOCK(); break; } case DIOCCLRRULECTRS: { /* obsoleted by DIOCGETRULE with action=PF_GET_CLR_CNTR */ struct pf_ruleset *ruleset = &pf_main_ruleset; struct pf_rule *rule; PF_RULES_WLOCK(); TAILQ_FOREACH(rule, ruleset->rules[PF_RULESET_FILTER].active.ptr, entries) { rule->evaluations = 0; rule->packets[0] = rule->packets[1] = 0; rule->bytes[0] = rule->bytes[1] = 0; } PF_RULES_WUNLOCK(); break; } case DIOCGIFSPEED: { struct pf_ifspeed *psp = (struct pf_ifspeed *)addr; struct pf_ifspeed ps; struct ifnet *ifp; if (psp->ifname[0] != 0) { /* Can we completely trust user-land? */ strlcpy(ps.ifname, psp->ifname, IFNAMSIZ); ifp = ifunit(ps.ifname); if (ifp != NULL) psp->baudrate = ifp->if_baudrate; else error = EINVAL; } else error = EINVAL; break; } #ifdef ALTQ case DIOCSTARTALTQ: { struct pf_altq *altq; PF_RULES_WLOCK(); /* enable all altq interfaces on active list */ TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { error = pf_enable_altq(altq); if (error != 0) break; } } if (error == 0) V_pf_altq_running = 1; PF_RULES_WUNLOCK(); DPFPRINTF(PF_DEBUG_MISC, ("altq: started\n")); break; } case DIOCSTOPALTQ: { struct pf_altq *altq; PF_RULES_WLOCK(); /* disable all altq interfaces on active list */ TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { if (altq->qname[0] == 0 && (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { error = pf_disable_altq(altq); if (error != 0) break; } } if (error == 0) V_pf_altq_running = 0; PF_RULES_WUNLOCK(); DPFPRINTF(PF_DEBUG_MISC, ("altq: stopped\n")); break; } case DIOCADDALTQ: { struct pfioc_altq *pa = (struct pfioc_altq *)addr; struct pf_altq *altq, *a; struct ifnet *ifp; altq = malloc(sizeof(*altq), M_PFALTQ, M_WAITOK); bcopy(&pa->altq, altq, sizeof(struct pf_altq)); altq->local_flags = 0; PF_RULES_WLOCK(); if (pa->ticket != V_ticket_altqs_inactive) { PF_RULES_WUNLOCK(); free(altq, M_PFALTQ); error = EBUSY; break; } /* * if this is for a queue, find the discipline and * copy the necessary fields */ if (altq->qname[0] != 0) { if ((altq->qid = pf_qname2qid(altq->qname)) == 0) { PF_RULES_WUNLOCK(); error = EBUSY; free(altq, M_PFALTQ); break; } altq->altq_disc = NULL; TAILQ_FOREACH(a, V_pf_altqs_inactive, entries) { if (strncmp(a->ifname, altq->ifname, IFNAMSIZ) == 0 && a->qname[0] == 0) { altq->altq_disc = a->altq_disc; break; } } } if ((ifp = ifunit(altq->ifname)) == NULL) altq->local_flags |= PFALTQ_FLAG_IF_REMOVED; else error = altq_add(altq); if (error) { PF_RULES_WUNLOCK(); free(altq, M_PFALTQ); break; } TAILQ_INSERT_TAIL(V_pf_altqs_inactive, altq, entries); bcopy(altq, &pa->altq, sizeof(struct pf_altq)); PF_RULES_WUNLOCK(); break; } case DIOCGETALTQS: { struct pfioc_altq *pa = (struct pfioc_altq *)addr; struct pf_altq *altq; PF_RULES_RLOCK(); pa->nr = 0; TAILQ_FOREACH(altq, V_pf_altqs_active, entries) pa->nr++; pa->ticket = V_ticket_altqs_active; PF_RULES_RUNLOCK(); break; } case DIOCGETALTQ: { struct pfioc_altq *pa = (struct pfioc_altq *)addr; struct pf_altq *altq; u_int32_t nr; PF_RULES_RLOCK(); if (pa->ticket != V_ticket_altqs_active) { PF_RULES_RUNLOCK(); error = EBUSY; break; } nr = 0; altq = TAILQ_FIRST(V_pf_altqs_active); while ((altq != NULL) && (nr < pa->nr)) { altq = TAILQ_NEXT(altq, entries); nr++; } if (altq == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } bcopy(altq, &pa->altq, sizeof(struct pf_altq)); PF_RULES_RUNLOCK(); break; } case DIOCCHANGEALTQ: /* CHANGEALTQ not supported yet! */ error = ENODEV; break; case DIOCGETQSTATS: { struct pfioc_qstats *pq = (struct pfioc_qstats *)addr; struct pf_altq *altq; u_int32_t nr; int nbytes; PF_RULES_RLOCK(); if (pq->ticket != V_ticket_altqs_active) { PF_RULES_RUNLOCK(); error = EBUSY; break; } nbytes = pq->nbytes; nr = 0; altq = TAILQ_FIRST(V_pf_altqs_active); while ((altq != NULL) && (nr < pq->nr)) { altq = TAILQ_NEXT(altq, entries); nr++; } if (altq == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) != 0) { PF_RULES_RUNLOCK(); error = ENXIO; break; } PF_RULES_RUNLOCK(); error = altq_getqstats(altq, pq->buf, &nbytes); if (error == 0) { pq->scheduler = altq->scheduler; pq->nbytes = nbytes; } break; } #endif /* ALTQ */ case DIOCBEGINADDRS: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; PF_RULES_WLOCK(); pf_empty_pool(&V_pf_pabuf); pp->ticket = ++V_ticket_pabuf; PF_RULES_WUNLOCK(); break; } case DIOCADDADDR: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pooladdr *pa; struct pfi_kif *kif = NULL; #ifndef INET if (pp->af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pp->af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ if (pp->addr.addr.type != PF_ADDR_ADDRMASK && pp->addr.addr.type != PF_ADDR_DYNIFTL && pp->addr.addr.type != PF_ADDR_TABLE) { error = EINVAL; break; } pa = malloc(sizeof(*pa), M_PFRULE, M_WAITOK); bcopy(&pp->addr, pa, sizeof(struct pf_pooladdr)); if (pa->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); if (pp->ticket != V_ticket_pabuf) { PF_RULES_WUNLOCK(); if (pa->ifname[0]) free(kif, PFI_MTYPE); free(pa, M_PFRULE); error = EBUSY; break; } if (pa->ifname[0]) { pa->kif = pfi_kif_attach(kif, pa->ifname); pfi_kif_ref(pa->kif); } else pa->kif = NULL; if (pa->addr.type == PF_ADDR_DYNIFTL && ((error = pfi_dynaddr_setup(&pa->addr, pp->af)) != 0)) { if (pa->ifname[0]) pfi_kif_unref(pa->kif); PF_RULES_WUNLOCK(); free(pa, M_PFRULE); break; } TAILQ_INSERT_TAIL(&V_pf_pabuf, pa, entries); PF_RULES_WUNLOCK(); break; } case DIOCGETADDRS: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *pa; PF_RULES_RLOCK(); pp->nr = 0; pool = pf_get_pool(pp->anchor, pp->ticket, pp->r_action, pp->r_num, 0, 1, 0); if (pool == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } TAILQ_FOREACH(pa, &pool->list, entries) pp->nr++; PF_RULES_RUNLOCK(); break; } case DIOCGETADDR: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *pa; u_int32_t nr = 0; PF_RULES_RLOCK(); pool = pf_get_pool(pp->anchor, pp->ticket, pp->r_action, pp->r_num, 0, 1, 1); if (pool == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } pa = TAILQ_FIRST(&pool->list); while ((pa != NULL) && (nr < pp->nr)) { pa = TAILQ_NEXT(pa, entries); nr++; } if (pa == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } bcopy(pa, &pp->addr, sizeof(struct pf_pooladdr)); pf_addr_copyout(&pp->addr.addr); PF_RULES_RUNLOCK(); break; } case DIOCCHANGEADDR: { struct pfioc_pooladdr *pca = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *oldpa = NULL, *newpa = NULL; struct pf_ruleset *ruleset; struct pfi_kif *kif = NULL; if (pca->action < PF_CHANGE_ADD_HEAD || pca->action > PF_CHANGE_REMOVE) { error = EINVAL; break; } if (pca->addr.addr.type != PF_ADDR_ADDRMASK && pca->addr.addr.type != PF_ADDR_DYNIFTL && pca->addr.addr.type != PF_ADDR_TABLE) { error = EINVAL; break; } if (pca->action != PF_CHANGE_REMOVE) { #ifndef INET if (pca->af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pca->af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ newpa = malloc(sizeof(*newpa), M_PFRULE, M_WAITOK); bcopy(&pca->addr, newpa, sizeof(struct pf_pooladdr)); if (newpa->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); newpa->kif = NULL; } #define ERROUT(x) { error = (x); goto DIOCCHANGEADDR_error; } PF_RULES_WLOCK(); ruleset = pf_find_ruleset(pca->anchor); if (ruleset == NULL) ERROUT(EBUSY); pool = pf_get_pool(pca->anchor, pca->ticket, pca->r_action, pca->r_num, pca->r_last, 1, 1); if (pool == NULL) ERROUT(EBUSY); if (pca->action != PF_CHANGE_REMOVE) { if (newpa->ifname[0]) { newpa->kif = pfi_kif_attach(kif, newpa->ifname); pfi_kif_ref(newpa->kif); kif = NULL; } switch (newpa->addr.type) { case PF_ADDR_DYNIFTL: error = pfi_dynaddr_setup(&newpa->addr, pca->af); break; case PF_ADDR_TABLE: newpa->addr.p.tbl = pfr_attach_table(ruleset, newpa->addr.v.tblname); if (newpa->addr.p.tbl == NULL) error = ENOMEM; break; } if (error) goto DIOCCHANGEADDR_error; } switch (pca->action) { case PF_CHANGE_ADD_HEAD: oldpa = TAILQ_FIRST(&pool->list); break; case PF_CHANGE_ADD_TAIL: oldpa = TAILQ_LAST(&pool->list, pf_palist); break; default: oldpa = TAILQ_FIRST(&pool->list); for (int i = 0; oldpa && i < pca->nr; i++) oldpa = TAILQ_NEXT(oldpa, entries); if (oldpa == NULL) ERROUT(EINVAL); } if (pca->action == PF_CHANGE_REMOVE) { TAILQ_REMOVE(&pool->list, oldpa, entries); switch (oldpa->addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(oldpa->addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(oldpa->addr.p.tbl); break; } if (oldpa->kif) pfi_kif_unref(oldpa->kif); free(oldpa, M_PFRULE); } else { if (oldpa == NULL) TAILQ_INSERT_TAIL(&pool->list, newpa, entries); else if (pca->action == PF_CHANGE_ADD_HEAD || pca->action == PF_CHANGE_ADD_BEFORE) TAILQ_INSERT_BEFORE(oldpa, newpa, entries); else TAILQ_INSERT_AFTER(&pool->list, oldpa, newpa, entries); } pool->cur = TAILQ_FIRST(&pool->list); PF_ACPY(&pool->counter, &pool->cur->addr.v.a.addr, pca->af); PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCCHANGEADDR_error: if (newpa != NULL) { if (newpa->kif) pfi_kif_unref(newpa->kif); free(newpa, M_PFRULE); } PF_RULES_WUNLOCK(); if (kif != NULL) free(kif, PFI_MTYPE); break; } case DIOCGETRULESETS: { struct pfioc_ruleset *pr = (struct pfioc_ruleset *)addr; struct pf_ruleset *ruleset; struct pf_anchor *anchor; PF_RULES_RLOCK(); pr->path[sizeof(pr->path) - 1] = 0; if ((ruleset = pf_find_ruleset(pr->path)) == NULL) { PF_RULES_RUNLOCK(); error = ENOENT; break; } pr->nr = 0; if (ruleset->anchor == NULL) { /* XXX kludge for pf_main_ruleset */ RB_FOREACH(anchor, pf_anchor_global, &V_pf_anchors) if (anchor->parent == NULL) pr->nr++; } else { RB_FOREACH(anchor, pf_anchor_node, &ruleset->anchor->children) pr->nr++; } PF_RULES_RUNLOCK(); break; } case DIOCGETRULESET: { struct pfioc_ruleset *pr = (struct pfioc_ruleset *)addr; struct pf_ruleset *ruleset; struct pf_anchor *anchor; u_int32_t nr = 0; PF_RULES_RLOCK(); pr->path[sizeof(pr->path) - 1] = 0; if ((ruleset = pf_find_ruleset(pr->path)) == NULL) { PF_RULES_RUNLOCK(); error = ENOENT; break; } pr->name[0] = 0; if (ruleset->anchor == NULL) { /* XXX kludge for pf_main_ruleset */ RB_FOREACH(anchor, pf_anchor_global, &V_pf_anchors) if (anchor->parent == NULL && nr++ == pr->nr) { strlcpy(pr->name, anchor->name, sizeof(pr->name)); break; } } else { RB_FOREACH(anchor, pf_anchor_node, &ruleset->anchor->children) if (nr++ == pr->nr) { strlcpy(pr->name, anchor->name, sizeof(pr->name)); break; } } if (!pr->name[0]) error = EBUSY; PF_RULES_RUNLOCK(); break; } case DIOCRCLRTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; if (io->pfrio_esize != 0) { error = ENODEV; break; } PF_RULES_WLOCK(); error = pfr_clr_tables(&io->pfrio_table, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); break; } case DIOCRADDTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_table))) { error = ENOMEM; break; } totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_WAITOK); error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_add_tables(pfrts, io->pfrio_size, &io->pfrio_nadd, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRDELTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_table))) { error = ENOMEM; break; } totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_WAITOK); error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_del_tables(pfrts, io->pfrio_size, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRGETTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_RLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_RUNLOCK(); break; } error = pfr_get_tables(&io->pfrio_table, pfrts, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfrts, io->pfrio_buffer, totlen); free(pfrts, M_TEMP); break; } case DIOCRGETTSTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_tstats *pfrtstats; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_tstats)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_tstats); pfrtstats = mallocarray(io->pfrio_size, sizeof(struct pfr_tstats), M_TEMP, M_NOWAIT); if (pfrtstats == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = pfr_get_tstats(&io->pfrio_table, pfrtstats, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0) error = copyout(pfrtstats, io->pfrio_buffer, totlen); free(pfrtstats, M_TEMP); break; } case DIOCRCLRTSTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); PF_RULES_WUNLOCK(); break; } error = pfr_clr_tstats(pfrts, io->pfrio_size, &io->pfrio_nzero, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRSETTFLAGS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); PF_RULES_WUNLOCK(); break; } error = pfr_set_tflags(pfrts, io->pfrio_size, io->pfrio_setflag, io->pfrio_clrflag, &io->pfrio_nchange, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRCLRADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; if (io->pfrio_esize != 0) { error = ENODEV; break; } PF_RULES_WLOCK(); error = pfr_clr_addrs(&io->pfrio_table, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); break; } case DIOCRADDADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_add_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nadd, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRDELADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_del_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRSETADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen, count; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size2 < 0) { error = EINVAL; break; } count = max(io->pfrio_size, io->pfrio_size2); if (count > pf_ioctl_maxcount || WOULD_OVERFLOW(count, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = count * sizeof(struct pfr_addr); pfras = mallocarray(count, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_set_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_size2, &io->pfrio_nadd, &io->pfrio_ndel, &io->pfrio_nchange, io->pfrio_flags | PFR_FLAG_USERIOCTL, 0); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRGETADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } PF_RULES_RLOCK(); error = pfr_get_addrs(&io->pfrio_table, pfras, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRGETASTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_astats *pfrastats; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_astats)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_astats))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_astats); pfrastats = mallocarray(io->pfrio_size, sizeof(struct pfr_astats), M_TEMP, M_NOWAIT); if (! pfrastats) { error = ENOMEM; break; } PF_RULES_RLOCK(); error = pfr_get_astats(&io->pfrio_table, pfrastats, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfrastats, io->pfrio_buffer, totlen); free(pfrastats, M_TEMP); break; } case DIOCRCLRASTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_clr_astats(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nzero, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRTSTADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_RLOCK(); error = pfr_tst_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nmatch, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRINADEFINE: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_ina_define(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nadd, &io->pfrio_naddr, io->pfrio_ticket, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfras, M_TEMP); break; } case DIOCOSFPADD: { struct pf_osfp_ioctl *io = (struct pf_osfp_ioctl *)addr; PF_RULES_WLOCK(); error = pf_osfp_add(io); PF_RULES_WUNLOCK(); break; } case DIOCOSFPGET: { struct pf_osfp_ioctl *io = (struct pf_osfp_ioctl *)addr; PF_RULES_RLOCK(); error = pf_osfp_get(io); PF_RULES_RUNLOCK(); break; } case DIOCXBEGIN: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioes, *ioe; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (! ioes) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if ((error = pf_begin_altq(&ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_begin(&table, &ioe->ticket, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; } default: if ((error = pf_begin_rules(&ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; } } PF_RULES_WUNLOCK(); error = copyout(ioes, io->array, totlen); free(ioes, M_TEMP); break; } case DIOCXROLLBACK: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioe, *ioes; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (! ioes) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if ((error = pf_rollback_altq(ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_rollback(&table, ioe->ticket, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } default: if ((error = pf_rollback_rules(ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } } PF_RULES_WUNLOCK(); free(ioes, M_TEMP); break; } case DIOCXCOMMIT: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioe, *ioes; struct pf_ruleset *rs; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (ioes == NULL) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); /* First makes sure everything will succeed. */ for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if (!V_altqs_inactive_open || ioe->ticket != V_ticket_altqs_inactive) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; #endif /* ALTQ */ case PF_RULESET_TABLE: rs = pf_find_ruleset(ioe->anchor); if (rs == NULL || !rs->topen || ioe->ticket != rs->tticket) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; default: if (ioe->rs_num < 0 || ioe->rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } rs = pf_find_ruleset(ioe->anchor); if (rs == NULL || !rs->rules[ioe->rs_num].inactive.open || rs->rules[ioe->rs_num].inactive.ticket != ioe->ticket) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; } } /* Now do the commit - no errors should happen here. */ for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if ((error = pf_commit_altq(ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_commit(&table, ioe->ticket, NULL, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } default: if ((error = pf_commit_rules(ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } } PF_RULES_WUNLOCK(); free(ioes, M_TEMP); break; } case DIOCGETSRCNODES: { struct pfioc_src_nodes *psn = (struct pfioc_src_nodes *)addr; struct pf_srchash *sh; struct pf_src_node *n, *p, *pstore; uint32_t i, nr = 0; if (psn->psn_len == 0) { for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) nr++; PF_HASHROW_UNLOCK(sh); } psn->psn_len = sizeof(struct pf_src_node) * nr; break; } p = pstore = malloc(psn->psn_len, M_TEMP, M_WAITOK); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { int secs = time_uptime, diff; if ((nr + 1) * sizeof(*p) > (unsigned)psn->psn_len) break; bcopy(n, p, sizeof(struct pf_src_node)); if (n->rule.ptr != NULL) p->rule.nr = n->rule.ptr->nr; p->creation = secs - p->creation; if (p->expire > secs) p->expire -= secs; else p->expire = 0; /* Adjust the connection rate estimate. */ diff = secs - n->conn_rate.last; if (diff >= n->conn_rate.seconds) p->conn_rate.count = 0; else p->conn_rate.count -= n->conn_rate.count * diff / n->conn_rate.seconds; p++; nr++; } PF_HASHROW_UNLOCK(sh); } error = copyout(pstore, psn->psn_src_nodes, sizeof(struct pf_src_node) * nr); if (error) { free(pstore, M_TEMP); break; } psn->psn_len = sizeof(struct pf_src_node) * nr; free(pstore, M_TEMP); break; } case DIOCCLRSRCNODES: { pf_clear_srcnodes(NULL); pf_purge_expired_src_nodes(); break; } case DIOCKILLSRCNODES: pf_kill_srcnodes((struct pfioc_src_node_kill *)addr); break; case DIOCSETHOSTID: { u_int32_t *hostid = (u_int32_t *)addr; PF_RULES_WLOCK(); if (*hostid == 0) V_pf_status.hostid = arc4random(); else V_pf_status.hostid = *hostid; PF_RULES_WUNLOCK(); break; } case DIOCOSFPFLUSH: PF_RULES_WLOCK(); pf_osfp_flush(); PF_RULES_WUNLOCK(); break; case DIOCIGETIFACES: { struct pfioc_iface *io = (struct pfioc_iface *)addr; struct pfi_kif *ifstore; size_t bufsiz; if (io->pfiio_esize != sizeof(struct pfi_kif)) { error = ENODEV; break; } if (io->pfiio_size < 0 || io->pfiio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfiio_size, sizeof(struct pfi_kif))) { error = EINVAL; break; } bufsiz = io->pfiio_size * sizeof(struct pfi_kif); ifstore = mallocarray(io->pfiio_size, sizeof(struct pfi_kif), M_TEMP, M_NOWAIT); if (ifstore == NULL) { error = ENOMEM; break; } PF_RULES_RLOCK(); pfi_get_ifaces(io->pfiio_name, ifstore, &io->pfiio_size); PF_RULES_RUNLOCK(); error = copyout(ifstore, io->pfiio_buffer, bufsiz); free(ifstore, M_TEMP); break; } case DIOCSETIFFLAG: { struct pfioc_iface *io = (struct pfioc_iface *)addr; PF_RULES_WLOCK(); error = pfi_set_flags(io->pfiio_name, io->pfiio_flags); PF_RULES_WUNLOCK(); break; } case DIOCCLRIFFLAG: { struct pfioc_iface *io = (struct pfioc_iface *)addr; PF_RULES_WLOCK(); error = pfi_clear_flags(io->pfiio_name, io->pfiio_flags); PF_RULES_WUNLOCK(); break; } default: error = ENODEV; break; } fail: if (sx_xlocked(&pf_ioctl_lock)) sx_xunlock(&pf_ioctl_lock); CURVNET_RESTORE(); return (error); } void pfsync_state_export(struct pfsync_state *sp, struct pf_state *st) { bzero(sp, sizeof(struct pfsync_state)); /* copy from state key */ sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; sp->proto = st->key[PF_SK_WIRE]->proto; sp->af = st->key[PF_SK_WIRE]->af; /* copy from state */ strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr)); sp->creation = htonl(time_uptime - st->creation); sp->expire = pf_state_expires(st); if (sp->expire <= time_uptime) sp->expire = htonl(0); else sp->expire = htonl(sp->expire - time_uptime); sp->direction = st->direction; sp->log = st->log; sp->timeout = st->timeout; sp->state_flags = st->state_flags; if (st->src_node) sp->sync_flags |= PFSYNC_FLAG_SRCNODE; if (st->nat_src_node) sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE; sp->id = st->id; sp->creatorid = st->creatorid; pf_state_peer_hton(&st->src, &sp->src); pf_state_peer_hton(&st->dst, &sp->dst); if (st->rule.ptr == NULL) sp->rule = htonl(-1); else sp->rule = htonl(st->rule.ptr->nr); if (st->anchor.ptr == NULL) sp->anchor = htonl(-1); else sp->anchor = htonl(st->anchor.ptr->nr); if (st->nat_rule.ptr == NULL) sp->nat_rule = htonl(-1); else sp->nat_rule = htonl(st->nat_rule.ptr->nr); pf_state_counter_hton(st->packets[0], sp->packets[0]); pf_state_counter_hton(st->packets[1], sp->packets[1]); pf_state_counter_hton(st->bytes[0], sp->bytes[0]); pf_state_counter_hton(st->bytes[1], sp->bytes[1]); } static void pf_tbladdr_copyout(struct pf_addr_wrap *aw) { struct pfr_ktable *kt; KASSERT(aw->type == PF_ADDR_TABLE, ("%s: type %u", __func__, aw->type)); kt = aw->p.tbl; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) kt = kt->pfrkt_root; aw->p.tbl = NULL; aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? kt->pfrkt_cnt : -1; } /* * XXX - Check for version missmatch!!! */ static void pf_clear_states(void) { struct pf_state *s; u_int i; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { s->timeout = PFTM_PURGE; /* Don't send out individual delete messages. */ s->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(s, PF_ENTER_LOCKED); goto relock; } PF_HASHROW_UNLOCK(ih); } } static int pf_clear_tables(void) { struct pfioc_table io; int error; bzero(&io, sizeof(io)); error = pfr_clr_tables(&io.pfrio_table, &io.pfrio_ndel, io.pfrio_flags); return (error); } static void pf_clear_srcnodes(struct pf_src_node *n) { struct pf_state *s; int i; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (n == NULL || n == s->src_node) s->src_node = NULL; if (n == NULL || n == s->nat_src_node) s->nat_src_node = NULL; } PF_HASHROW_UNLOCK(ih); } if (n == NULL) { struct pf_srchash *sh; for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { n->expire = 1; n->states = 0; } PF_HASHROW_UNLOCK(sh); } } else { /* XXX: hash slot should already be locked here. */ n->expire = 1; n->states = 0; } } static void pf_kill_srcnodes(struct pfioc_src_node_kill *psnk) { struct pf_src_node_list kill; LIST_INIT(&kill); for (int i = 0; i <= pf_srchashmask; i++) { struct pf_srchash *sh = &V_pf_srchash[i]; struct pf_src_node *sn, *tmp; PF_HASHROW_LOCK(sh); LIST_FOREACH_SAFE(sn, &sh->nodes, entry, tmp) if (PF_MATCHA(psnk->psnk_src.neg, &psnk->psnk_src.addr.v.a.addr, &psnk->psnk_src.addr.v.a.mask, &sn->addr, sn->af) && PF_MATCHA(psnk->psnk_dst.neg, &psnk->psnk_dst.addr.v.a.addr, &psnk->psnk_dst.addr.v.a.mask, &sn->raddr, sn->af)) { pf_unlink_src_node(sn); LIST_INSERT_HEAD(&kill, sn, entry); sn->expire = 1; } PF_HASHROW_UNLOCK(sh); } for (int i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state *s; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (s->src_node && s->src_node->expire == 1) s->src_node = NULL; if (s->nat_src_node && s->nat_src_node->expire == 1) s->nat_src_node = NULL; } PF_HASHROW_UNLOCK(ih); } psnk->psnk_killed = pf_free_src_nodes(&kill); } /* * XXX - Check for version missmatch!!! */ /* * Duplicate pfctl -Fa operation to get rid of as much as we can. */ static int shutdown_pf(void) { int error = 0; u_int32_t t[5]; char nn = '\0'; do { if ((error = pf_begin_rules(&t[0], PF_RULESET_SCRUB, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: SCRUB\n")); break; } if ((error = pf_begin_rules(&t[1], PF_RULESET_FILTER, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: FILTER\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[2], PF_RULESET_NAT, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: NAT\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[3], PF_RULESET_BINAT, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: BINAT\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[4], PF_RULESET_RDR, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: RDR\n")); break; /* XXX: rollback? */ } /* XXX: these should always succeed here */ pf_commit_rules(t[0], PF_RULESET_SCRUB, &nn); pf_commit_rules(t[1], PF_RULESET_FILTER, &nn); pf_commit_rules(t[2], PF_RULESET_NAT, &nn); pf_commit_rules(t[3], PF_RULESET_BINAT, &nn); pf_commit_rules(t[4], PF_RULESET_RDR, &nn); if ((error = pf_clear_tables()) != 0) break; #ifdef ALTQ if ((error = pf_begin_altq(&t[0])) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: ALTQ\n")); break; } pf_commit_altq(t[0]); #endif pf_clear_states(); pf_clear_srcnodes(NULL); /* status does not use malloced mem so no need to cleanup */ /* fingerprints and interfaces have their own cleanup code */ /* Free counters last as we updated them during shutdown. */ counter_u64_free(V_pf_default_rule.states_cur); counter_u64_free(V_pf_default_rule.states_tot); counter_u64_free(V_pf_default_rule.src_nodes); for (int i = 0; i < PFRES_MAX; i++) counter_u64_free(V_pf_status.counters[i]); for (int i = 0; i < LCNT_MAX; i++) counter_u64_free(V_pf_status.lcounters[i]); for (int i = 0; i < FCNT_MAX; i++) counter_u64_free(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) counter_u64_free(V_pf_status.scounters[i]); } while(0); return (error); } #ifdef INET static int pf_check_in(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp) { int chk; chk = pf_test(PF_IN, flags, ifp, m, inp); if (chk && *m) { m_freem(*m); *m = NULL; } if (chk != PF_PASS) return (EACCES); return (0); } static int pf_check_out(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp) { int chk; chk = pf_test(PF_OUT, flags, ifp, m, inp); if (chk && *m) { m_freem(*m); *m = NULL; } if (chk != PF_PASS) return (EACCES); return (0); } #endif #ifdef INET6 static int pf_check6_in(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp) { int chk; /* * In case of loopback traffic IPv6 uses the real interface in * order to support scoped addresses. In order to support stateful * filtering we have change this to lo0 as it is the case in IPv4. */ CURVNET_SET(ifp->if_vnet); chk = pf_test6(PF_IN, flags, (*m)->m_flags & M_LOOP ? V_loif : ifp, m, inp); CURVNET_RESTORE(); if (chk && *m) { m_freem(*m); *m = NULL; } if (chk != PF_PASS) return (EACCES); return (0); } static int pf_check6_out(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, int flags, struct inpcb *inp) { int chk; CURVNET_SET(ifp->if_vnet); chk = pf_test6(PF_OUT, flags, ifp, m, inp); CURVNET_RESTORE(); if (chk && *m) { m_freem(*m); *m = NULL; } if (chk != PF_PASS) return (EACCES); return (0); } #endif /* INET6 */ static int hook_pf(void) { #ifdef INET struct pfil_head *pfh_inet; #endif #ifdef INET6 struct pfil_head *pfh_inet6; #endif if (V_pf_pfil_hooked) return (0); #ifdef INET pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET); if (pfh_inet == NULL) return (ESRCH); /* XXX */ pfil_add_hook_flags(pf_check_in, NULL, PFIL_IN | PFIL_WAITOK, pfh_inet); pfil_add_hook_flags(pf_check_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh_inet); #endif #ifdef INET6 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6); if (pfh_inet6 == NULL) { #ifdef INET pfil_remove_hook_flags(pf_check_in, NULL, PFIL_IN | PFIL_WAITOK, pfh_inet); pfil_remove_hook_flags(pf_check_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh_inet); #endif return (ESRCH); /* XXX */ } pfil_add_hook_flags(pf_check6_in, NULL, PFIL_IN | PFIL_WAITOK, pfh_inet6); pfil_add_hook_flags(pf_check6_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh_inet6); #endif V_pf_pfil_hooked = 1; return (0); } static int dehook_pf(void) { #ifdef INET struct pfil_head *pfh_inet; #endif #ifdef INET6 struct pfil_head *pfh_inet6; #endif if (V_pf_pfil_hooked == 0) return (0); #ifdef INET pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET); if (pfh_inet == NULL) return (ESRCH); /* XXX */ pfil_remove_hook_flags(pf_check_in, NULL, PFIL_IN | PFIL_WAITOK, pfh_inet); pfil_remove_hook_flags(pf_check_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh_inet); #endif #ifdef INET6 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6); if (pfh_inet6 == NULL) return (ESRCH); /* XXX */ pfil_remove_hook_flags(pf_check6_in, NULL, PFIL_IN | PFIL_WAITOK, pfh_inet6); pfil_remove_hook_flags(pf_check6_out, NULL, PFIL_OUT | PFIL_WAITOK, pfh_inet6); #endif V_pf_pfil_hooked = 0; return (0); } static void pf_load_vnet(void) { TAILQ_INIT(&V_pf_tags); TAILQ_INIT(&V_pf_qids); pfattach_vnet(); V_pf_vnet_active = 1; } static int pf_load(void) { int error; rm_init(&pf_rules_lock, "pf rulesets"); sx_init(&pf_ioctl_lock, "pf ioctl"); sx_init(&pf_end_lock, "pf end thread"); pf_mtag_initialize(); pf_dev = make_dev(&pf_cdevsw, 0, 0, 0, 0600, PF_NAME); if (pf_dev == NULL) return (ENOMEM); pf_end_threads = 0; error = kproc_create(pf_purge_thread, NULL, &pf_purge_proc, 0, 0, "pf purge"); if (error != 0) return (error); pfi_initialize(); return (0); } static void pf_unload_vnet(void) { int error; V_pf_vnet_active = 0; V_pf_status.running = 0; error = dehook_pf(); if (error) { /* * Should not happen! * XXX Due to error code ESRCH, kldunload will show * a message like 'No such process'. */ printf("%s : pfil unregisteration fail\n", __FUNCTION__); return; } PF_RULES_WLOCK(); shutdown_pf(); PF_RULES_WUNLOCK(); swi_remove(V_pf_swi_cookie); pf_unload_vnet_purge(); pf_normalize_cleanup(); PF_RULES_WLOCK(); pfi_cleanup_vnet(); PF_RULES_WUNLOCK(); pfr_cleanup(); pf_osfp_flush(); pf_cleanup(); if (IS_DEFAULT_VNET(curvnet)) pf_mtag_cleanup(); } static void pf_unload(void) { sx_xlock(&pf_end_lock); pf_end_threads = 1; while (pf_end_threads < 2) { wakeup_one(pf_purge_thread); sx_sleep(pf_purge_proc, &pf_end_lock, 0, "pftmo", 0); } sx_xunlock(&pf_end_lock); if (pf_dev != NULL) destroy_dev(pf_dev); pfi_cleanup(); rm_destroy(&pf_rules_lock); sx_destroy(&pf_ioctl_lock); sx_destroy(&pf_end_lock); } static void vnet_pf_init(void *unused __unused) { pf_load_vnet(); } VNET_SYSINIT(vnet_pf_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_THIRD, vnet_pf_init, NULL); static void vnet_pf_uninit(const void *unused __unused) { pf_unload_vnet(); } SYSUNINIT(pf_unload, SI_SUB_PROTO_FIREWALL, SI_ORDER_SECOND, pf_unload, NULL); VNET_SYSUNINIT(vnet_pf_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_THIRD, vnet_pf_uninit, NULL); static int pf_modevent(module_t mod, int type, void *data) { int error = 0; switch(type) { case MOD_LOAD: error = pf_load(); break; case MOD_UNLOAD: /* Handled in SYSUNINIT(pf_unload) to ensure it's done after * the vnet_pf_uninit()s */ break; default: error = EINVAL; break; } return (error); } static moduledata_t pf_mod = { "pf", pf_modevent, 0 }; DECLARE_MODULE(pf, pf_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_SECOND); MODULE_VERSION(pf, PF_MODVER); Index: head/sys/netpfil/pf/pf_norm.c =================================================================== --- head/sys/netpfil/pf/pf_norm.c (revision 336675) +++ head/sys/netpfil/pf/pf_norm.c (revision 336676) @@ -1,1853 +1,1853 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright 2001 Niels Provos * Copyright 2011 Alexander Bluhm * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $OpenBSD: pf_norm.c,v 1.114 2009/01/29 14:11:45 henning Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif /* INET6 */ struct pf_frent { TAILQ_ENTRY(pf_frent) fr_next; struct mbuf *fe_m; uint16_t fe_hdrlen; /* ipv4 header length with ip options ipv6, extension, fragment header */ uint16_t fe_extoff; /* last extension header offset or 0 */ uint16_t fe_len; /* fragment length */ uint16_t fe_off; /* fragment offset */ uint16_t fe_mff; /* more fragment flag */ }; struct pf_fragment_cmp { struct pf_addr frc_src; struct pf_addr frc_dst; uint32_t frc_id; sa_family_t frc_af; uint8_t frc_proto; }; struct pf_fragment { struct pf_fragment_cmp fr_key; #define fr_src fr_key.frc_src #define fr_dst fr_key.frc_dst #define fr_id fr_key.frc_id #define fr_af fr_key.frc_af #define fr_proto fr_key.frc_proto RB_ENTRY(pf_fragment) fr_entry; TAILQ_ENTRY(pf_fragment) frag_next; uint32_t fr_timeout; uint16_t fr_maxlen; /* maximum length of single fragment */ TAILQ_HEAD(pf_fragq, pf_frent) fr_queue; }; struct pf_fragment_tag { uint16_t ft_hdrlen; /* header length of reassembled pkt */ uint16_t ft_extoff; /* last extension header offset or 0 */ uint16_t ft_maxlen; /* maximum fragment payload length */ uint32_t ft_id; /* fragment id */ }; static struct mtx pf_frag_mtx; MTX_SYSINIT(pf_frag_mtx, &pf_frag_mtx, "pf fragments", MTX_DEF); #define PF_FRAG_LOCK() mtx_lock(&pf_frag_mtx) #define PF_FRAG_UNLOCK() mtx_unlock(&pf_frag_mtx) #define PF_FRAG_ASSERT() mtx_assert(&pf_frag_mtx, MA_OWNED) VNET_DEFINE(uma_zone_t, pf_state_scrub_z); /* XXX: shared with pfsync */ -static VNET_DEFINE(uma_zone_t, pf_frent_z); +VNET_DEFINE_STATIC(uma_zone_t, pf_frent_z); #define V_pf_frent_z VNET(pf_frent_z) -static VNET_DEFINE(uma_zone_t, pf_frag_z); +VNET_DEFINE_STATIC(uma_zone_t, pf_frag_z); #define V_pf_frag_z VNET(pf_frag_z) TAILQ_HEAD(pf_fragqueue, pf_fragment); TAILQ_HEAD(pf_cachequeue, pf_fragment); -static VNET_DEFINE(struct pf_fragqueue, pf_fragqueue); +VNET_DEFINE_STATIC(struct pf_fragqueue, pf_fragqueue); #define V_pf_fragqueue VNET(pf_fragqueue) RB_HEAD(pf_frag_tree, pf_fragment); -static VNET_DEFINE(struct pf_frag_tree, pf_frag_tree); +VNET_DEFINE_STATIC(struct pf_frag_tree, pf_frag_tree); #define V_pf_frag_tree VNET(pf_frag_tree) static int pf_frag_compare(struct pf_fragment *, struct pf_fragment *); static RB_PROTOTYPE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); static RB_GENERATE(pf_frag_tree, pf_fragment, fr_entry, pf_frag_compare); static void pf_flush_fragments(void); static void pf_free_fragment(struct pf_fragment *); static void pf_remove_fragment(struct pf_fragment *); static int pf_normalize_tcpopt(struct pf_rule *, struct mbuf *, struct tcphdr *, int, sa_family_t); static struct pf_frent *pf_create_fragment(u_short *); static struct pf_fragment *pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree); static struct pf_fragment *pf_fillup_fragment(struct pf_fragment_cmp *, struct pf_frent *, u_short *); static int pf_isfull_fragment(struct pf_fragment *); static struct mbuf *pf_join_fragment(struct pf_fragment *); #ifdef INET static void pf_scrub_ip(struct mbuf **, uint32_t, uint8_t, uint8_t); static int pf_reassemble(struct mbuf **, struct ip *, int, u_short *); #endif /* INET */ #ifdef INET6 static int pf_reassemble6(struct mbuf **, struct ip6_hdr *, struct ip6_frag *, uint16_t, uint16_t, u_short *); static void pf_scrub_ip6(struct mbuf **, uint8_t); #endif /* INET6 */ #define DPFPRINTF(x) do { \ if (V_pf_status.debug >= PF_DEBUG_MISC) { \ printf("%s: ", __func__); \ printf x ; \ } \ } while(0) #ifdef INET static void pf_ip2key(struct ip *ip, int dir, struct pf_fragment_cmp *key) { key->frc_src.v4 = ip->ip_src; key->frc_dst.v4 = ip->ip_dst; key->frc_af = AF_INET; key->frc_proto = ip->ip_p; key->frc_id = ip->ip_id; } #endif /* INET */ void pf_normalize_init(void) { V_pf_frag_z = uma_zcreate("pf frags", sizeof(struct pf_fragment), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_frent_z = uma_zcreate("pf frag entries", sizeof(struct pf_frent), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_state_scrub_z = uma_zcreate("pf state scrubs", sizeof(struct pf_state_scrub), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_FRAGS].zone = V_pf_frent_z; V_pf_limits[PF_LIMIT_FRAGS].limit = PFFRAG_FRENT_HIWAT; uma_zone_set_max(V_pf_frent_z, PFFRAG_FRENT_HIWAT); uma_zone_set_warning(V_pf_frent_z, "PF frag entries limit reached"); TAILQ_INIT(&V_pf_fragqueue); } void pf_normalize_cleanup(void) { uma_zdestroy(V_pf_state_scrub_z); uma_zdestroy(V_pf_frent_z); uma_zdestroy(V_pf_frag_z); } static int pf_frag_compare(struct pf_fragment *a, struct pf_fragment *b) { int diff; if ((diff = a->fr_id - b->fr_id) != 0) return (diff); if ((diff = a->fr_proto - b->fr_proto) != 0) return (diff); if ((diff = a->fr_af - b->fr_af) != 0) return (diff); if ((diff = pf_addr_cmp(&a->fr_src, &b->fr_src, a->fr_af)) != 0) return (diff); if ((diff = pf_addr_cmp(&a->fr_dst, &b->fr_dst, a->fr_af)) != 0) return (diff); return (0); } void pf_purge_expired_fragments(void) { u_int32_t expire = time_uptime - V_pf_default_rule.timeout[PFTM_FRAG]; pf_purge_fragments(expire); } void pf_purge_fragments(uint32_t expire) { struct pf_fragment *frag; PF_FRAG_LOCK(); while ((frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue)) != NULL) { if (frag->fr_timeout > expire) break; DPFPRINTF(("expiring %d(%p)\n", frag->fr_id, frag)); pf_free_fragment(frag); } PF_FRAG_UNLOCK(); } /* * Try to flush old fragments to make space for new ones */ static void pf_flush_fragments(void) { struct pf_fragment *frag; int goal; PF_FRAG_ASSERT(); goal = uma_zone_get_cur(V_pf_frent_z) * 9 / 10; DPFPRINTF(("trying to free %d frag entriess\n", goal)); while (goal < uma_zone_get_cur(V_pf_frent_z)) { frag = TAILQ_LAST(&V_pf_fragqueue, pf_fragqueue); if (frag) pf_free_fragment(frag); else break; } } /* Frees the fragments and all associated entries */ static void pf_free_fragment(struct pf_fragment *frag) { struct pf_frent *frent; PF_FRAG_ASSERT(); /* Free all fragments */ for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = TAILQ_FIRST(&frag->fr_queue)) { TAILQ_REMOVE(&frag->fr_queue, frent, fr_next); m_freem(frent->fe_m); uma_zfree(V_pf_frent_z, frent); } pf_remove_fragment(frag); } static struct pf_fragment * pf_find_fragment(struct pf_fragment_cmp *key, struct pf_frag_tree *tree) { struct pf_fragment *frag; PF_FRAG_ASSERT(); frag = RB_FIND(pf_frag_tree, tree, (struct pf_fragment *)key); if (frag != NULL) { /* XXX Are we sure we want to update the timeout? */ frag->fr_timeout = time_uptime; TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); } return (frag); } /* Removes a fragment from the fragment queue and frees the fragment */ static void pf_remove_fragment(struct pf_fragment *frag) { PF_FRAG_ASSERT(); RB_REMOVE(pf_frag_tree, &V_pf_frag_tree, frag); TAILQ_REMOVE(&V_pf_fragqueue, frag, frag_next); uma_zfree(V_pf_frag_z, frag); } static struct pf_frent * pf_create_fragment(u_short *reason) { struct pf_frent *frent; PF_FRAG_ASSERT(); frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); if (frent == NULL) { pf_flush_fragments(); frent = uma_zalloc(V_pf_frent_z, M_NOWAIT); if (frent == NULL) { REASON_SET(reason, PFRES_MEMORY); return (NULL); } } return (frent); } static struct pf_fragment * pf_fillup_fragment(struct pf_fragment_cmp *key, struct pf_frent *frent, u_short *reason) { struct pf_frent *after, *next, *prev; struct pf_fragment *frag; uint16_t total; PF_FRAG_ASSERT(); /* No empty fragments. */ if (frent->fe_len == 0) { DPFPRINTF(("bad fragment: len 0")); goto bad_fragment; } /* All fragments are 8 byte aligned. */ if (frent->fe_mff && (frent->fe_len & 0x7)) { DPFPRINTF(("bad fragment: mff and len %d", frent->fe_len)); goto bad_fragment; } /* Respect maximum length, IP_MAXPACKET == IPV6_MAXPACKET. */ if (frent->fe_off + frent->fe_len > IP_MAXPACKET) { DPFPRINTF(("bad fragment: max packet %d", frent->fe_off + frent->fe_len)); goto bad_fragment; } DPFPRINTF((key->frc_af == AF_INET ? "reass frag %d @ %d-%d" : "reass frag %#08x @ %d-%d", key->frc_id, frent->fe_off, frent->fe_off + frent->fe_len)); /* Fully buffer all of the fragments in this fragment queue. */ frag = pf_find_fragment(key, &V_pf_frag_tree); /* Create a new reassembly queue for this packet. */ if (frag == NULL) { frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); if (frag == NULL) { pf_flush_fragments(); frag = uma_zalloc(V_pf_frag_z, M_NOWAIT); if (frag == NULL) { REASON_SET(reason, PFRES_MEMORY); goto drop_fragment; } } *(struct pf_fragment_cmp *)frag = *key; frag->fr_timeout = time_uptime; frag->fr_maxlen = frent->fe_len; TAILQ_INIT(&frag->fr_queue); RB_INSERT(pf_frag_tree, &V_pf_frag_tree, frag); TAILQ_INSERT_HEAD(&V_pf_fragqueue, frag, frag_next); /* We do not have a previous fragment. */ TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); return (frag); } KASSERT(!TAILQ_EMPTY(&frag->fr_queue), ("!TAILQ_EMPTY()->fr_queue")); /* Remember maximum fragment len for refragmentation. */ if (frent->fe_len > frag->fr_maxlen) frag->fr_maxlen = frent->fe_len; /* Maximum data we have seen already. */ total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; /* Non terminal fragments must have more fragments flag. */ if (frent->fe_off + frent->fe_len < total && !frent->fe_mff) goto bad_fragment; /* Check if we saw the last fragment already. */ if (!TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) { if (frent->fe_off + frent->fe_len > total || (frent->fe_off + frent->fe_len == total && frent->fe_mff)) goto bad_fragment; } else { if (frent->fe_off + frent->fe_len == total && !frent->fe_mff) goto bad_fragment; } /* Find a fragment after the current one. */ prev = NULL; TAILQ_FOREACH(after, &frag->fr_queue, fr_next) { if (after->fe_off > frent->fe_off) break; prev = after; } KASSERT(prev != NULL || after != NULL, ("prev != NULL || after != NULL")); if (prev != NULL && prev->fe_off + prev->fe_len > frent->fe_off) { uint16_t precut; precut = prev->fe_off + prev->fe_len - frent->fe_off; if (precut >= frent->fe_len) goto bad_fragment; DPFPRINTF(("overlap -%d", precut)); m_adj(frent->fe_m, precut); frent->fe_off += precut; frent->fe_len -= precut; } for (; after != NULL && frent->fe_off + frent->fe_len > after->fe_off; after = next) { uint16_t aftercut; aftercut = frent->fe_off + frent->fe_len - after->fe_off; DPFPRINTF(("adjust overlap %d", aftercut)); if (aftercut < after->fe_len) { m_adj(after->fe_m, aftercut); after->fe_off += aftercut; after->fe_len -= aftercut; break; } /* This fragment is completely overlapped, lose it. */ next = TAILQ_NEXT(after, fr_next); m_freem(after->fe_m); TAILQ_REMOVE(&frag->fr_queue, after, fr_next); uma_zfree(V_pf_frent_z, after); } if (prev == NULL) TAILQ_INSERT_HEAD(&frag->fr_queue, frent, fr_next); else TAILQ_INSERT_AFTER(&frag->fr_queue, prev, frent, fr_next); return (frag); bad_fragment: REASON_SET(reason, PFRES_FRAG); drop_fragment: uma_zfree(V_pf_frent_z, frent); return (NULL); } static int pf_isfull_fragment(struct pf_fragment *frag) { struct pf_frent *frent, *next; uint16_t off, total; /* Check if we are completely reassembled */ if (TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_mff) return (0); /* Maximum data we have seen already */ total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; /* Check if we have all the data */ off = 0; for (frent = TAILQ_FIRST(&frag->fr_queue); frent; frent = next) { next = TAILQ_NEXT(frent, fr_next); off += frent->fe_len; if (off < total && (next == NULL || next->fe_off != off)) { DPFPRINTF(("missing fragment at %d, next %d, total %d", off, next == NULL ? -1 : next->fe_off, total)); return (0); } } DPFPRINTF(("%d < %d?", off, total)); if (off < total) return (0); KASSERT(off == total, ("off == total")); return (1); } static struct mbuf * pf_join_fragment(struct pf_fragment *frag) { struct mbuf *m, *m2; struct pf_frent *frent, *next; frent = TAILQ_FIRST(&frag->fr_queue); next = TAILQ_NEXT(frent, fr_next); m = frent->fe_m; m_adj(m, (frent->fe_hdrlen + frent->fe_len) - m->m_pkthdr.len); uma_zfree(V_pf_frent_z, frent); for (frent = next; frent != NULL; frent = next) { next = TAILQ_NEXT(frent, fr_next); m2 = frent->fe_m; /* Strip off ip header. */ m_adj(m2, frent->fe_hdrlen); /* Strip off any trailing bytes. */ m_adj(m2, frent->fe_len - m2->m_pkthdr.len); uma_zfree(V_pf_frent_z, frent); m_cat(m, m2); } /* Remove from fragment queue. */ pf_remove_fragment(frag); return (m); } #ifdef INET static int pf_reassemble(struct mbuf **m0, struct ip *ip, int dir, u_short *reason) { struct mbuf *m = *m0; struct pf_frent *frent; struct pf_fragment *frag; struct pf_fragment_cmp key; uint16_t total, hdrlen; /* Get an entry for the fragment queue */ if ((frent = pf_create_fragment(reason)) == NULL) return (PF_DROP); frent->fe_m = m; frent->fe_hdrlen = ip->ip_hl << 2; frent->fe_extoff = 0; frent->fe_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); frent->fe_off = (ntohs(ip->ip_off) & IP_OFFMASK) << 3; frent->fe_mff = ntohs(ip->ip_off) & IP_MF; pf_ip2key(ip, dir, &key); if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) return (PF_DROP); /* The mbuf is part of the fragment entry, no direct free or access */ m = *m0 = NULL; if (!pf_isfull_fragment(frag)) return (PF_PASS); /* drop because *m0 is NULL, no error */ /* We have all the data */ frent = TAILQ_FIRST(&frag->fr_queue); KASSERT(frent != NULL, ("frent != NULL")); total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; hdrlen = frent->fe_hdrlen; m = *m0 = pf_join_fragment(frag); frag = NULL; if (m->m_flags & M_PKTHDR) { int plen = 0; for (m = *m0; m; m = m->m_next) plen += m->m_len; m = *m0; m->m_pkthdr.len = plen; } ip = mtod(m, struct ip *); ip->ip_len = htons(hdrlen + total); ip->ip_off &= ~(IP_MF|IP_OFFMASK); if (hdrlen + total > IP_MAXPACKET) { DPFPRINTF(("drop: too big: %d", total)); ip->ip_len = 0; REASON_SET(reason, PFRES_SHORT); /* PF_DROP requires a valid mbuf *m0 in pf_test() */ return (PF_DROP); } DPFPRINTF(("complete: %p(%d)\n", m, ntohs(ip->ip_len))); return (PF_PASS); } #endif /* INET */ #ifdef INET6 static int pf_reassemble6(struct mbuf **m0, struct ip6_hdr *ip6, struct ip6_frag *fraghdr, uint16_t hdrlen, uint16_t extoff, u_short *reason) { struct mbuf *m = *m0; struct pf_frent *frent; struct pf_fragment *frag; struct pf_fragment_cmp key; struct m_tag *mtag; struct pf_fragment_tag *ftag; int off; uint32_t frag_id; uint16_t total, maxlen; uint8_t proto; PF_FRAG_LOCK(); /* Get an entry for the fragment queue. */ if ((frent = pf_create_fragment(reason)) == NULL) { PF_FRAG_UNLOCK(); return (PF_DROP); } frent->fe_m = m; frent->fe_hdrlen = hdrlen; frent->fe_extoff = extoff; frent->fe_len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - hdrlen; frent->fe_off = ntohs(fraghdr->ip6f_offlg & IP6F_OFF_MASK); frent->fe_mff = fraghdr->ip6f_offlg & IP6F_MORE_FRAG; key.frc_src.v6 = ip6->ip6_src; key.frc_dst.v6 = ip6->ip6_dst; key.frc_af = AF_INET6; /* Only the first fragment's protocol is relevant. */ key.frc_proto = 0; key.frc_id = fraghdr->ip6f_ident; if ((frag = pf_fillup_fragment(&key, frent, reason)) == NULL) { PF_FRAG_UNLOCK(); return (PF_DROP); } /* The mbuf is part of the fragment entry, no direct free or access. */ m = *m0 = NULL; if (!pf_isfull_fragment(frag)) { PF_FRAG_UNLOCK(); return (PF_PASS); /* Drop because *m0 is NULL, no error. */ } /* We have all the data. */ extoff = frent->fe_extoff; maxlen = frag->fr_maxlen; frag_id = frag->fr_id; frent = TAILQ_FIRST(&frag->fr_queue); KASSERT(frent != NULL, ("frent != NULL")); total = TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_off + TAILQ_LAST(&frag->fr_queue, pf_fragq)->fe_len; hdrlen = frent->fe_hdrlen - sizeof(struct ip6_frag); m = *m0 = pf_join_fragment(frag); frag = NULL; PF_FRAG_UNLOCK(); /* Take protocol from first fragment header. */ m = m_getptr(m, hdrlen + offsetof(struct ip6_frag, ip6f_nxt), &off); KASSERT(m, ("%s: short mbuf chain", __func__)); proto = *(mtod(m, caddr_t) + off); m = *m0; /* Delete frag6 header */ if (ip6_deletefraghdr(m, hdrlen, M_NOWAIT) != 0) goto fail; if (m->m_flags & M_PKTHDR) { int plen = 0; for (m = *m0; m; m = m->m_next) plen += m->m_len; m = *m0; m->m_pkthdr.len = plen; } if ((mtag = m_tag_get(PF_REASSEMBLED, sizeof(struct pf_fragment_tag), M_NOWAIT)) == NULL) goto fail; ftag = (struct pf_fragment_tag *)(mtag + 1); ftag->ft_hdrlen = hdrlen; ftag->ft_extoff = extoff; ftag->ft_maxlen = maxlen; ftag->ft_id = frag_id; m_tag_prepend(m, mtag); ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_plen = htons(hdrlen - sizeof(struct ip6_hdr) + total); if (extoff) { /* Write protocol into next field of last extension header. */ m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), &off); KASSERT(m, ("%s: short mbuf chain", __func__)); *(mtod(m, char *) + off) = proto; m = *m0; } else ip6->ip6_nxt = proto; if (hdrlen - sizeof(struct ip6_hdr) + total > IPV6_MAXPACKET) { DPFPRINTF(("drop: too big: %d", total)); ip6->ip6_plen = 0; REASON_SET(reason, PFRES_SHORT); /* PF_DROP requires a valid mbuf *m0 in pf_test6(). */ return (PF_DROP); } DPFPRINTF(("complete: %p(%d)", m, ntohs(ip6->ip6_plen))); return (PF_PASS); fail: REASON_SET(reason, PFRES_MEMORY); /* PF_DROP requires a valid mbuf *m0 in pf_test6(), will free later. */ return (PF_DROP); } #endif /* INET6 */ #ifdef INET6 int pf_refragment6(struct ifnet *ifp, struct mbuf **m0, struct m_tag *mtag) { struct mbuf *m = *m0, *t; struct pf_fragment_tag *ftag = (struct pf_fragment_tag *)(mtag + 1); struct pf_pdesc pd; uint32_t frag_id; uint16_t hdrlen, extoff, maxlen; uint8_t proto; int error, action; hdrlen = ftag->ft_hdrlen; extoff = ftag->ft_extoff; maxlen = ftag->ft_maxlen; frag_id = ftag->ft_id; m_tag_delete(m, mtag); mtag = NULL; ftag = NULL; if (extoff) { int off; /* Use protocol from next field of last extension header */ m = m_getptr(m, extoff + offsetof(struct ip6_ext, ip6e_nxt), &off); KASSERT((m != NULL), ("pf_refragment6: short mbuf chain")); proto = *(mtod(m, caddr_t) + off); *(mtod(m, char *) + off) = IPPROTO_FRAGMENT; m = *m0; } else { struct ip6_hdr *hdr; hdr = mtod(m, struct ip6_hdr *); proto = hdr->ip6_nxt; hdr->ip6_nxt = IPPROTO_FRAGMENT; } /* The MTU must be a multiple of 8 bytes, or we risk doing the * fragmentation wrong. */ maxlen = maxlen & ~7; /* * Maxlen may be less than 8 if there was only a single * fragment. As it was fragmented before, add a fragment * header also for a single fragment. If total or maxlen * is less than 8, ip6_fragment() will return EMSGSIZE and * we drop the packet. */ error = ip6_fragment(ifp, m, hdrlen, proto, maxlen, frag_id); m = (*m0)->m_nextpkt; (*m0)->m_nextpkt = NULL; if (error == 0) { /* The first mbuf contains the unfragmented packet. */ m_freem(*m0); *m0 = NULL; action = PF_PASS; } else { /* Drop expects an mbuf to free. */ DPFPRINTF(("refragment error %d", error)); action = PF_DROP; } for (t = m; m; m = t) { t = m->m_nextpkt; m->m_nextpkt = NULL; m->m_flags |= M_SKIP_FIREWALL; memset(&pd, 0, sizeof(pd)); pd.pf_mtag = pf_find_mtag(m); if (error == 0) ip6_forward(m, 0); else m_freem(m); } return (action); } #endif /* INET6 */ #ifdef INET int pf_normalize_ip(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, struct pf_pdesc *pd) { struct mbuf *m = *m0; struct pf_rule *r; struct ip *h = mtod(m, struct ip *); int mff = (ntohs(h->ip_off) & IP_MF); int hlen = h->ip_hl << 2; u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; u_int16_t max; int ip_len; int ip_off; int tag = -1; int verdict; PF_RULES_RASSERT(); r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != dir) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != AF_INET) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != h->ip_p) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, (struct pf_addr *)&h->ip_src.s_addr, AF_INET, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; else if (PF_MISMATCHAW(&r->dst.addr, (struct pf_addr *)&h->ip_dst.s_addr, AF_INET, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; else if (r->match_tag && !pf_match_tag(m, r, &tag, pd->pf_mtag ? pd->pf_mtag->tag : 0)) r = TAILQ_NEXT(r, entries); else break; } if (r == NULL || r->action == PF_NOSCRUB) return (PF_PASS); else { r->packets[dir == PF_OUT]++; r->bytes[dir == PF_OUT] += pd->tot_len; } /* Check for illegal packets */ if (hlen < (int)sizeof(struct ip)) { REASON_SET(reason, PFRES_NORM); goto drop; } if (hlen > ntohs(h->ip_len)) { REASON_SET(reason, PFRES_NORM); goto drop; } /* Clear IP_DF if the rule uses the no-df option */ if (r->rule_flag & PFRULE_NODF && h->ip_off & htons(IP_DF)) { u_int16_t ip_off = h->ip_off; h->ip_off &= htons(~IP_DF); h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); } /* We will need other tests here */ if (!fragoff && !mff) goto no_fragment; /* We're dealing with a fragment now. Don't allow fragments * with IP_DF to enter the cache. If the flag was cleared by * no-df above, fine. Otherwise drop it. */ if (h->ip_off & htons(IP_DF)) { DPFPRINTF(("IP_DF\n")); goto bad; } ip_len = ntohs(h->ip_len) - hlen; ip_off = (ntohs(h->ip_off) & IP_OFFMASK) << 3; /* All fragments are 8 byte aligned */ if (mff && (ip_len & 0x7)) { DPFPRINTF(("mff and %d\n", ip_len)); goto bad; } /* Respect maximum length */ if (fragoff + ip_len > IP_MAXPACKET) { DPFPRINTF(("max packet %d\n", fragoff + ip_len)); goto bad; } max = fragoff + ip_len; /* Fully buffer all of the fragments * Might return a completely reassembled mbuf, or NULL */ PF_FRAG_LOCK(); DPFPRINTF(("reass frag %d @ %d-%d\n", h->ip_id, fragoff, max)); verdict = pf_reassemble(m0, h, dir, reason); PF_FRAG_UNLOCK(); if (verdict != PF_PASS) return (PF_DROP); m = *m0; if (m == NULL) return (PF_DROP); h = mtod(m, struct ip *); no_fragment: /* At this point, only IP_DF is allowed in ip_off */ if (h->ip_off & ~htons(IP_DF)) { u_int16_t ip_off = h->ip_off; h->ip_off &= htons(IP_DF); h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); } pf_scrub_ip(&m, r->rule_flag, r->min_ttl, r->set_tos); return (PF_PASS); bad: DPFPRINTF(("dropping bad fragment\n")); REASON_SET(reason, PFRES_FRAG); drop: if (r != NULL && r->log) PFLOG_PACKET(kif, m, AF_INET, dir, *reason, r, NULL, NULL, pd, 1); return (PF_DROP); } #endif #ifdef INET6 int pf_normalize_ip6(struct mbuf **m0, int dir, struct pfi_kif *kif, u_short *reason, struct pf_pdesc *pd) { struct mbuf *m = *m0; struct pf_rule *r; struct ip6_hdr *h = mtod(m, struct ip6_hdr *); int extoff; int off; struct ip6_ext ext; struct ip6_opt opt; struct ip6_opt_jumbo jumbo; struct ip6_frag frag; u_int32_t jumbolen = 0, plen; int optend; int ooff; u_int8_t proto; int terminal; PF_RULES_RASSERT(); r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != dir) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != AF_INET6) r = r->skip[PF_SKIP_AF].ptr; #if 0 /* header chain! */ else if (r->proto && r->proto != h->ip6_nxt) r = r->skip[PF_SKIP_PROTO].ptr; #endif else if (PF_MISMATCHAW(&r->src.addr, (struct pf_addr *)&h->ip6_src, AF_INET6, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; else if (PF_MISMATCHAW(&r->dst.addr, (struct pf_addr *)&h->ip6_dst, AF_INET6, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; else break; } if (r == NULL || r->action == PF_NOSCRUB) return (PF_PASS); else { r->packets[dir == PF_OUT]++; r->bytes[dir == PF_OUT] += pd->tot_len; } /* Check for illegal packets */ if (sizeof(struct ip6_hdr) + IPV6_MAXPACKET < m->m_pkthdr.len) goto drop; extoff = 0; off = sizeof(struct ip6_hdr); proto = h->ip6_nxt; terminal = 0; do { switch (proto) { case IPPROTO_FRAGMENT: goto fragment; break; case IPPROTO_AH: case IPPROTO_ROUTING: case IPPROTO_DSTOPTS: if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, NULL, AF_INET6)) goto shortpkt; extoff = off; if (proto == IPPROTO_AH) off += (ext.ip6e_len + 2) * 4; else off += (ext.ip6e_len + 1) * 8; proto = ext.ip6e_nxt; break; case IPPROTO_HOPOPTS: if (!pf_pull_hdr(m, off, &ext, sizeof(ext), NULL, NULL, AF_INET6)) goto shortpkt; extoff = off; optend = off + (ext.ip6e_len + 1) * 8; ooff = off + sizeof(ext); do { if (!pf_pull_hdr(m, ooff, &opt.ip6o_type, sizeof(opt.ip6o_type), NULL, NULL, AF_INET6)) goto shortpkt; if (opt.ip6o_type == IP6OPT_PAD1) { ooff++; continue; } if (!pf_pull_hdr(m, ooff, &opt, sizeof(opt), NULL, NULL, AF_INET6)) goto shortpkt; if (ooff + sizeof(opt) + opt.ip6o_len > optend) goto drop; switch (opt.ip6o_type) { case IP6OPT_JUMBO: if (h->ip6_plen != 0) goto drop; if (!pf_pull_hdr(m, ooff, &jumbo, sizeof(jumbo), NULL, NULL, AF_INET6)) goto shortpkt; memcpy(&jumbolen, jumbo.ip6oj_jumbo_len, sizeof(jumbolen)); jumbolen = ntohl(jumbolen); if (jumbolen <= IPV6_MAXPACKET) goto drop; if (sizeof(struct ip6_hdr) + jumbolen != m->m_pkthdr.len) goto drop; break; default: break; } ooff += sizeof(opt) + opt.ip6o_len; } while (ooff < optend); off = optend; proto = ext.ip6e_nxt; break; default: terminal = 1; break; } } while (!terminal); /* jumbo payload option must be present, or plen > 0 */ if (ntohs(h->ip6_plen) == 0) plen = jumbolen; else plen = ntohs(h->ip6_plen); if (plen == 0) goto drop; if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) goto shortpkt; pf_scrub_ip6(&m, r->min_ttl); return (PF_PASS); fragment: /* Jumbo payload packets cannot be fragmented. */ plen = ntohs(h->ip6_plen); if (plen == 0 || jumbolen) goto drop; if (sizeof(struct ip6_hdr) + plen > m->m_pkthdr.len) goto shortpkt; if (!pf_pull_hdr(m, off, &frag, sizeof(frag), NULL, NULL, AF_INET6)) goto shortpkt; /* Offset now points to data portion. */ off += sizeof(frag); /* Returns PF_DROP or *m0 is NULL or completely reassembled mbuf. */ if (pf_reassemble6(m0, h, &frag, off, extoff, reason) != PF_PASS) return (PF_DROP); m = *m0; if (m == NULL) return (PF_DROP); pd->flags |= PFDESC_IP_REAS; return (PF_PASS); shortpkt: REASON_SET(reason, PFRES_SHORT); if (r != NULL && r->log) PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1); return (PF_DROP); drop: REASON_SET(reason, PFRES_NORM); if (r != NULL && r->log) PFLOG_PACKET(kif, m, AF_INET6, dir, *reason, r, NULL, NULL, pd, 1); return (PF_DROP); } #endif /* INET6 */ int pf_normalize_tcp(int dir, struct pfi_kif *kif, struct mbuf *m, int ipoff, int off, void *h, struct pf_pdesc *pd) { struct pf_rule *r, *rm = NULL; struct tcphdr *th = pd->hdr.tcp; int rewrite = 0; u_short reason; u_int8_t flags; sa_family_t af = pd->af; PF_RULES_RASSERT(); r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_SCRUB].active.ptr); while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != dir) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != af) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != pd->proto) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; else if (r->src.port_op && !pf_match_port(r->src.port_op, r->src.port[0], r->src.port[1], th->th_sport)) r = r->skip[PF_SKIP_SRC_PORT].ptr; else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; else if (r->dst.port_op && !pf_match_port(r->dst.port_op, r->dst.port[0], r->dst.port[1], th->th_dport)) r = r->skip[PF_SKIP_DST_PORT].ptr; else if (r->os_fingerprint != PF_OSFP_ANY && !pf_osfp_match( pf_osfp_fingerprint(pd, m, off, th), r->os_fingerprint)) r = TAILQ_NEXT(r, entries); else { rm = r; break; } } if (rm == NULL || rm->action == PF_NOSCRUB) return (PF_PASS); else { r->packets[dir == PF_OUT]++; r->bytes[dir == PF_OUT] += pd->tot_len; } if (rm->rule_flag & PFRULE_REASSEMBLE_TCP) pd->flags |= PFDESC_TCP_NORM; flags = th->th_flags; if (flags & TH_SYN) { /* Illegal packet */ if (flags & TH_RST) goto tcp_drop; if (flags & TH_FIN) goto tcp_drop; } else { /* Illegal packet */ if (!(flags & (TH_ACK|TH_RST))) goto tcp_drop; } if (!(flags & TH_ACK)) { /* These flags are only valid if ACK is set */ if ((flags & TH_FIN) || (flags & TH_PUSH) || (flags & TH_URG)) goto tcp_drop; } /* Check for illegal header length */ if (th->th_off < (sizeof(struct tcphdr) >> 2)) goto tcp_drop; /* If flags changed, or reserved data set, then adjust */ if (flags != th->th_flags || th->th_x2 != 0) { u_int16_t ov, nv; ov = *(u_int16_t *)(&th->th_ack + 1); th->th_flags = flags; th->th_x2 = 0; nv = *(u_int16_t *)(&th->th_ack + 1); th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, ov, nv, 0); rewrite = 1; } /* Remove urgent pointer, if TH_URG is not set */ if (!(flags & TH_URG) && th->th_urp) { th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, th->th_urp, 0, 0); th->th_urp = 0; rewrite = 1; } /* Process options */ if (r->max_mss && pf_normalize_tcpopt(r, m, th, off, pd->af)) rewrite = 1; /* copy back packet headers if we sanitized */ if (rewrite) m_copyback(m, off, sizeof(*th), (caddr_t)th); return (PF_PASS); tcp_drop: REASON_SET(&reason, PFRES_NORM); if (rm != NULL && r->log) PFLOG_PACKET(kif, m, AF_INET, dir, reason, r, NULL, NULL, pd, 1); return (PF_DROP); } int pf_normalize_tcp_init(struct mbuf *m, int off, struct pf_pdesc *pd, struct tcphdr *th, struct pf_state_peer *src, struct pf_state_peer *dst) { u_int32_t tsval, tsecr; u_int8_t hdr[60]; u_int8_t *opt; KASSERT((src->scrub == NULL), ("pf_normalize_tcp_init: src->scrub != NULL")); src->scrub = uma_zalloc(V_pf_state_scrub_z, M_ZERO | M_NOWAIT); if (src->scrub == NULL) return (1); switch (pd->af) { #ifdef INET case AF_INET: { struct ip *h = mtod(m, struct ip *); src->scrub->pfss_ttl = h->ip_ttl; break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { struct ip6_hdr *h = mtod(m, struct ip6_hdr *); src->scrub->pfss_ttl = h->ip6_hlim; break; } #endif /* INET6 */ } /* * All normalizations below are only begun if we see the start of * the connections. They must all set an enabled bit in pfss_flags */ if ((th->th_flags & TH_SYN) == 0) return (0); if (th->th_off > (sizeof(struct tcphdr) >> 2) && src->scrub && pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { /* Diddle with TCP options */ int hlen; opt = hdr + sizeof(struct tcphdr); hlen = (th->th_off << 2) - sizeof(struct tcphdr); while (hlen >= TCPOLEN_TIMESTAMP) { switch (*opt) { case TCPOPT_EOL: /* FALLTHROUGH */ case TCPOPT_NOP: opt++; hlen--; break; case TCPOPT_TIMESTAMP: if (opt[1] >= TCPOLEN_TIMESTAMP) { src->scrub->pfss_flags |= PFSS_TIMESTAMP; src->scrub->pfss_ts_mod = htonl(arc4random()); /* note PFSS_PAWS not set yet */ memcpy(&tsval, &opt[2], sizeof(u_int32_t)); memcpy(&tsecr, &opt[6], sizeof(u_int32_t)); src->scrub->pfss_tsval0 = ntohl(tsval); src->scrub->pfss_tsval = ntohl(tsval); src->scrub->pfss_tsecr = ntohl(tsecr); getmicrouptime(&src->scrub->pfss_last); } /* FALLTHROUGH */ default: hlen -= MAX(opt[1], 2); opt += MAX(opt[1], 2); break; } } } return (0); } void pf_normalize_tcp_cleanup(struct pf_state *state) { if (state->src.scrub) uma_zfree(V_pf_state_scrub_z, state->src.scrub); if (state->dst.scrub) uma_zfree(V_pf_state_scrub_z, state->dst.scrub); /* Someday... flush the TCP segment reassembly descriptors. */ } int pf_normalize_tcp_stateful(struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, struct tcphdr *th, struct pf_state *state, struct pf_state_peer *src, struct pf_state_peer *dst, int *writeback) { struct timeval uptime; u_int32_t tsval, tsecr; u_int tsval_from_last; u_int8_t hdr[60]; u_int8_t *opt; int copyback = 0; int got_ts = 0; KASSERT((src->scrub || dst->scrub), ("%s: src->scrub && dst->scrub!", __func__)); /* * Enforce the minimum TTL seen for this connection. Negate a common * technique to evade an intrusion detection system and confuse * firewall state code. */ switch (pd->af) { #ifdef INET case AF_INET: { if (src->scrub) { struct ip *h = mtod(m, struct ip *); if (h->ip_ttl > src->scrub->pfss_ttl) src->scrub->pfss_ttl = h->ip_ttl; h->ip_ttl = src->scrub->pfss_ttl; } break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { if (src->scrub) { struct ip6_hdr *h = mtod(m, struct ip6_hdr *); if (h->ip6_hlim > src->scrub->pfss_ttl) src->scrub->pfss_ttl = h->ip6_hlim; h->ip6_hlim = src->scrub->pfss_ttl; } break; } #endif /* INET6 */ } if (th->th_off > (sizeof(struct tcphdr) >> 2) && ((src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) || (dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP))) && pf_pull_hdr(m, off, hdr, th->th_off << 2, NULL, NULL, pd->af)) { /* Diddle with TCP options */ int hlen; opt = hdr + sizeof(struct tcphdr); hlen = (th->th_off << 2) - sizeof(struct tcphdr); while (hlen >= TCPOLEN_TIMESTAMP) { switch (*opt) { case TCPOPT_EOL: /* FALLTHROUGH */ case TCPOPT_NOP: opt++; hlen--; break; case TCPOPT_TIMESTAMP: /* Modulate the timestamps. Can be used for * NAT detection, OS uptime determination or * reboot detection. */ if (got_ts) { /* Huh? Multiple timestamps!? */ if (V_pf_status.debug >= PF_DEBUG_MISC) { DPFPRINTF(("multiple TS??")); pf_print_state(state); printf("\n"); } REASON_SET(reason, PFRES_TS); return (PF_DROP); } if (opt[1] >= TCPOLEN_TIMESTAMP) { memcpy(&tsval, &opt[2], sizeof(u_int32_t)); if (tsval && src->scrub && (src->scrub->pfss_flags & PFSS_TIMESTAMP)) { tsval = ntohl(tsval); pf_change_proto_a(m, &opt[2], &th->th_sum, htonl(tsval + src->scrub->pfss_ts_mod), 0); copyback = 1; } /* Modulate TS reply iff valid (!0) */ memcpy(&tsecr, &opt[6], sizeof(u_int32_t)); if (tsecr && dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { tsecr = ntohl(tsecr) - dst->scrub->pfss_ts_mod; pf_change_proto_a(m, &opt[6], &th->th_sum, htonl(tsecr), 0); copyback = 1; } got_ts = 1; } /* FALLTHROUGH */ default: hlen -= MAX(opt[1], 2); opt += MAX(opt[1], 2); break; } } if (copyback) { /* Copyback the options, caller copys back header */ *writeback = 1; m_copyback(m, off + sizeof(struct tcphdr), (th->th_off << 2) - sizeof(struct tcphdr), hdr + sizeof(struct tcphdr)); } } /* * Must invalidate PAWS checks on connections idle for too long. * The fastest allowed timestamp clock is 1ms. That turns out to * be about 24 days before it wraps. XXX Right now our lowerbound * TS echo check only works for the first 12 days of a connection * when the TS has exhausted half its 32bit space */ #define TS_MAX_IDLE (24*24*60*60) #define TS_MAX_CONN (12*24*60*60) /* XXX remove when better tsecr check */ getmicrouptime(&uptime); if (src->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && (uptime.tv_sec - src->scrub->pfss_last.tv_sec > TS_MAX_IDLE || time_uptime - state->creation > TS_MAX_CONN)) { if (V_pf_status.debug >= PF_DEBUG_MISC) { DPFPRINTF(("src idled out of PAWS\n")); pf_print_state(state); printf("\n"); } src->scrub->pfss_flags = (src->scrub->pfss_flags & ~PFSS_PAWS) | PFSS_PAWS_IDLED; } if (dst->scrub && (dst->scrub->pfss_flags & PFSS_PAWS) && uptime.tv_sec - dst->scrub->pfss_last.tv_sec > TS_MAX_IDLE) { if (V_pf_status.debug >= PF_DEBUG_MISC) { DPFPRINTF(("dst idled out of PAWS\n")); pf_print_state(state); printf("\n"); } dst->scrub->pfss_flags = (dst->scrub->pfss_flags & ~PFSS_PAWS) | PFSS_PAWS_IDLED; } if (got_ts && src->scrub && dst->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && (dst->scrub->pfss_flags & PFSS_PAWS)) { /* Validate that the timestamps are "in-window". * RFC1323 describes TCP Timestamp options that allow * measurement of RTT (round trip time) and PAWS * (protection against wrapped sequence numbers). PAWS * gives us a set of rules for rejecting packets on * long fat pipes (packets that were somehow delayed * in transit longer than the time it took to send the * full TCP sequence space of 4Gb). We can use these * rules and infer a few others that will let us treat * the 32bit timestamp and the 32bit echoed timestamp * as sequence numbers to prevent a blind attacker from * inserting packets into a connection. * * RFC1323 tells us: * - The timestamp on this packet must be greater than * or equal to the last value echoed by the other * endpoint. The RFC says those will be discarded * since it is a dup that has already been acked. * This gives us a lowerbound on the timestamp. * timestamp >= other last echoed timestamp * - The timestamp will be less than or equal to * the last timestamp plus the time between the * last packet and now. The RFC defines the max * clock rate as 1ms. We will allow clocks to be * up to 10% fast and will allow a total difference * or 30 seconds due to a route change. And this * gives us an upperbound on the timestamp. * timestamp <= last timestamp + max ticks * We have to be careful here. Windows will send an * initial timestamp of zero and then initialize it * to a random value after the 3whs; presumably to * avoid a DoS by having to call an expensive RNG * during a SYN flood. Proof MS has at least one * good security geek. * * - The TCP timestamp option must also echo the other * endpoints timestamp. The timestamp echoed is the * one carried on the earliest unacknowledged segment * on the left edge of the sequence window. The RFC * states that the host will reject any echoed * timestamps that were larger than any ever sent. * This gives us an upperbound on the TS echo. * tescr <= largest_tsval * - The lowerbound on the TS echo is a little more * tricky to determine. The other endpoint's echoed * values will not decrease. But there may be * network conditions that re-order packets and * cause our view of them to decrease. For now the * only lowerbound we can safely determine is that * the TS echo will never be less than the original * TS. XXX There is probably a better lowerbound. * Remove TS_MAX_CONN with better lowerbound check. * tescr >= other original TS * * It is also important to note that the fastest * timestamp clock of 1ms will wrap its 32bit space in * 24 days. So we just disable TS checking after 24 * days of idle time. We actually must use a 12d * connection limit until we can come up with a better * lowerbound to the TS echo check. */ struct timeval delta_ts; int ts_fudge; /* * PFTM_TS_DIFF is how many seconds of leeway to allow * a host's timestamp. This can happen if the previous * packet got delayed in transit for much longer than * this packet. */ if ((ts_fudge = state->rule.ptr->timeout[PFTM_TS_DIFF]) == 0) ts_fudge = V_pf_default_rule.timeout[PFTM_TS_DIFF]; /* Calculate max ticks since the last timestamp */ #define TS_MAXFREQ 1100 /* RFC max TS freq of 1Khz + 10% skew */ #define TS_MICROSECS 1000000 /* microseconds per second */ delta_ts = uptime; timevalsub(&delta_ts, &src->scrub->pfss_last); tsval_from_last = (delta_ts.tv_sec + ts_fudge) * TS_MAXFREQ; tsval_from_last += delta_ts.tv_usec / (TS_MICROSECS/TS_MAXFREQ); if ((src->state >= TCPS_ESTABLISHED && dst->state >= TCPS_ESTABLISHED) && (SEQ_LT(tsval, dst->scrub->pfss_tsecr) || SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) || (tsecr && (SEQ_GT(tsecr, dst->scrub->pfss_tsval) || SEQ_LT(tsecr, dst->scrub->pfss_tsval0))))) { /* Bad RFC1323 implementation or an insertion attack. * * - Solaris 2.6 and 2.7 are known to send another ACK * after the FIN,FIN|ACK,ACK closing that carries * an old timestamp. */ DPFPRINTF(("Timestamp failed %c%c%c%c\n", SEQ_LT(tsval, dst->scrub->pfss_tsecr) ? '0' : ' ', SEQ_GT(tsval, src->scrub->pfss_tsval + tsval_from_last) ? '1' : ' ', SEQ_GT(tsecr, dst->scrub->pfss_tsval) ? '2' : ' ', SEQ_LT(tsecr, dst->scrub->pfss_tsval0)? '3' : ' ')); DPFPRINTF((" tsval: %u tsecr: %u +ticks: %u " "idle: %jus %lums\n", tsval, tsecr, tsval_from_last, (uintmax_t)delta_ts.tv_sec, delta_ts.tv_usec / 1000)); DPFPRINTF((" src->tsval: %u tsecr: %u\n", src->scrub->pfss_tsval, src->scrub->pfss_tsecr)); DPFPRINTF((" dst->tsval: %u tsecr: %u tsval0: %u" "\n", dst->scrub->pfss_tsval, dst->scrub->pfss_tsecr, dst->scrub->pfss_tsval0)); if (V_pf_status.debug >= PF_DEBUG_MISC) { pf_print_state(state); pf_print_flags(th->th_flags); printf("\n"); } REASON_SET(reason, PFRES_TS); return (PF_DROP); } /* XXX I'd really like to require tsecr but it's optional */ } else if (!got_ts && (th->th_flags & TH_RST) == 0 && ((src->state == TCPS_ESTABLISHED && dst->state == TCPS_ESTABLISHED) || pd->p_len > 0 || (th->th_flags & TH_SYN)) && src->scrub && dst->scrub && (src->scrub->pfss_flags & PFSS_PAWS) && (dst->scrub->pfss_flags & PFSS_PAWS)) { /* Didn't send a timestamp. Timestamps aren't really useful * when: * - connection opening or closing (often not even sent). * but we must not let an attacker to put a FIN on a * data packet to sneak it through our ESTABLISHED check. * - on a TCP reset. RFC suggests not even looking at TS. * - on an empty ACK. The TS will not be echoed so it will * probably not help keep the RTT calculation in sync and * there isn't as much danger when the sequence numbers * got wrapped. So some stacks don't include TS on empty * ACKs :-( * * To minimize the disruption to mostly RFC1323 conformant * stacks, we will only require timestamps on data packets. * * And what do ya know, we cannot require timestamps on data * packets. There appear to be devices that do legitimate * TCP connection hijacking. There are HTTP devices that allow * a 3whs (with timestamps) and then buffer the HTTP request. * If the intermediate device has the HTTP response cache, it * will spoof the response but not bother timestamping its * packets. So we can look for the presence of a timestamp in * the first data packet and if there, require it in all future * packets. */ if (pd->p_len > 0 && (src->scrub->pfss_flags & PFSS_DATA_TS)) { /* * Hey! Someone tried to sneak a packet in. Or the * stack changed its RFC1323 behavior?!?! */ if (V_pf_status.debug >= PF_DEBUG_MISC) { DPFPRINTF(("Did not receive expected RFC1323 " "timestamp\n")); pf_print_state(state); pf_print_flags(th->th_flags); printf("\n"); } REASON_SET(reason, PFRES_TS); return (PF_DROP); } } /* * We will note if a host sends his data packets with or without * timestamps. And require all data packets to contain a timestamp * if the first does. PAWS implicitly requires that all data packets be * timestamped. But I think there are middle-man devices that hijack * TCP streams immediately after the 3whs and don't timestamp their * packets (seen in a WWW accelerator or cache). */ if (pd->p_len > 0 && src->scrub && (src->scrub->pfss_flags & (PFSS_TIMESTAMP|PFSS_DATA_TS|PFSS_DATA_NOTS)) == PFSS_TIMESTAMP) { if (got_ts) src->scrub->pfss_flags |= PFSS_DATA_TS; else { src->scrub->pfss_flags |= PFSS_DATA_NOTS; if (V_pf_status.debug >= PF_DEBUG_MISC && dst->scrub && (dst->scrub->pfss_flags & PFSS_TIMESTAMP)) { /* Don't warn if other host rejected RFC1323 */ DPFPRINTF(("Broken RFC1323 stack did not " "timestamp data packet. Disabled PAWS " "security.\n")); pf_print_state(state); pf_print_flags(th->th_flags); printf("\n"); } } } /* * Update PAWS values */ if (got_ts && src->scrub && PFSS_TIMESTAMP == (src->scrub->pfss_flags & (PFSS_PAWS_IDLED|PFSS_TIMESTAMP))) { getmicrouptime(&src->scrub->pfss_last); if (SEQ_GEQ(tsval, src->scrub->pfss_tsval) || (src->scrub->pfss_flags & PFSS_PAWS) == 0) src->scrub->pfss_tsval = tsval; if (tsecr) { if (SEQ_GEQ(tsecr, src->scrub->pfss_tsecr) || (src->scrub->pfss_flags & PFSS_PAWS) == 0) src->scrub->pfss_tsecr = tsecr; if ((src->scrub->pfss_flags & PFSS_PAWS) == 0 && (SEQ_LT(tsval, src->scrub->pfss_tsval0) || src->scrub->pfss_tsval0 == 0)) { /* tsval0 MUST be the lowest timestamp */ src->scrub->pfss_tsval0 = tsval; } /* Only fully initialized after a TS gets echoed */ if ((src->scrub->pfss_flags & PFSS_PAWS) == 0) src->scrub->pfss_flags |= PFSS_PAWS; } } /* I have a dream.... TCP segment reassembly.... */ return (0); } static int pf_normalize_tcpopt(struct pf_rule *r, struct mbuf *m, struct tcphdr *th, int off, sa_family_t af) { u_int16_t *mss; int thoff; int opt, cnt, optlen = 0; int rewrite = 0; u_char opts[TCP_MAXOLEN]; u_char *optp = opts; thoff = th->th_off << 2; cnt = thoff - sizeof(struct tcphdr); if (cnt > 0 && !pf_pull_hdr(m, off + sizeof(*th), opts, cnt, NULL, NULL, af)) return (rewrite); for (; cnt > 0; cnt -= optlen, optp += optlen) { opt = optp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { if (cnt < 2) break; optlen = optp[1]; if (optlen < 2 || optlen > cnt) break; } switch (opt) { case TCPOPT_MAXSEG: mss = (u_int16_t *)(optp + 2); if ((ntohs(*mss)) > r->max_mss) { th->th_sum = pf_proto_cksum_fixup(m, th->th_sum, *mss, htons(r->max_mss), 0); *mss = htons(r->max_mss); rewrite = 1; } break; default: break; } } if (rewrite) m_copyback(m, off + sizeof(*th), thoff - sizeof(*th), opts); return (rewrite); } #ifdef INET static void pf_scrub_ip(struct mbuf **m0, u_int32_t flags, u_int8_t min_ttl, u_int8_t tos) { struct mbuf *m = *m0; struct ip *h = mtod(m, struct ip *); /* Clear IP_DF if no-df was requested */ if (flags & PFRULE_NODF && h->ip_off & htons(IP_DF)) { u_int16_t ip_off = h->ip_off; h->ip_off &= htons(~IP_DF); h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_off, h->ip_off, 0); } /* Enforce a minimum ttl, may cause endless packet loops */ if (min_ttl && h->ip_ttl < min_ttl) { u_int16_t ip_ttl = h->ip_ttl; h->ip_ttl = min_ttl; h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_ttl, h->ip_ttl, 0); } /* Enforce tos */ if (flags & PFRULE_SET_TOS) { u_int16_t ov, nv; ov = *(u_int16_t *)h; h->ip_tos = tos | (h->ip_tos & IPTOS_ECN_MASK); nv = *(u_int16_t *)h; h->ip_sum = pf_cksum_fixup(h->ip_sum, ov, nv, 0); } /* random-id, but not for fragments */ if (flags & PFRULE_RANDOMID && !(h->ip_off & ~htons(IP_DF))) { uint16_t ip_id = h->ip_id; ip_fillid(h); h->ip_sum = pf_cksum_fixup(h->ip_sum, ip_id, h->ip_id, 0); } } #endif /* INET */ #ifdef INET6 static void pf_scrub_ip6(struct mbuf **m0, u_int8_t min_ttl) { struct mbuf *m = *m0; struct ip6_hdr *h = mtod(m, struct ip6_hdr *); /* Enforce a minimum ttl, may cause endless packet loops */ if (min_ttl && h->ip6_hlim < min_ttl) h->ip6_hlim = min_ttl; } #endif Index: head/sys/netpfil/pf/pf_osfp.c =================================================================== --- head/sys/netpfil/pf/pf_osfp.c (revision 336675) +++ head/sys/netpfil/pf/pf_osfp.c (revision 336676) @@ -1,536 +1,536 @@ /*- * SPDX-License-Identifier: ISC * * Copyright (c) 2003 Mike Frantzen * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. * * $OpenBSD: pf_osfp.c,v 1.14 2008/06/12 18:17:01 henning Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif static MALLOC_DEFINE(M_PFOSFP, "pf_osfp", "pf(4) operating system fingerprints"); #define DPFPRINTF(format, x...) \ if (V_pf_status.debug >= PF_DEBUG_NOISY) \ printf(format , ##x) SLIST_HEAD(pf_osfp_list, pf_os_fingerprint); -static VNET_DEFINE(struct pf_osfp_list, pf_osfp_list) = +VNET_DEFINE_STATIC(struct pf_osfp_list, pf_osfp_list) = SLIST_HEAD_INITIALIZER(); #define V_pf_osfp_list VNET(pf_osfp_list) static struct pf_osfp_enlist *pf_osfp_fingerprint_hdr(const struct ip *, const struct ip6_hdr *, const struct tcphdr *); static struct pf_os_fingerprint *pf_osfp_find(struct pf_osfp_list *, struct pf_os_fingerprint *, u_int8_t); static struct pf_os_fingerprint *pf_osfp_find_exact(struct pf_osfp_list *, struct pf_os_fingerprint *); static void pf_osfp_insert(struct pf_osfp_list *, struct pf_os_fingerprint *); #ifdef PFDEBUG static struct pf_os_fingerprint *pf_osfp_validate(void); #endif /* * Passively fingerprint the OS of the host (IPv4 TCP SYN packets only) * Returns the list of possible OSes. */ struct pf_osfp_enlist * pf_osfp_fingerprint(struct pf_pdesc *pd, struct mbuf *m, int off, const struct tcphdr *tcp) { struct ip *ip; struct ip6_hdr *ip6; char hdr[60]; if ((pd->af != PF_INET && pd->af != PF_INET6) || pd->proto != IPPROTO_TCP || (tcp->th_off << 2) < sizeof(*tcp)) return (NULL); if (pd->af == PF_INET) { ip = mtod(m, struct ip *); ip6 = (struct ip6_hdr *)NULL; } else { ip = (struct ip *)NULL; ip6 = mtod(m, struct ip6_hdr *); } if (!pf_pull_hdr(m, off, hdr, tcp->th_off << 2, NULL, NULL, pd->af)) return (NULL); return (pf_osfp_fingerprint_hdr(ip, ip6, (struct tcphdr *)hdr)); } static struct pf_osfp_enlist * pf_osfp_fingerprint_hdr(const struct ip *ip, const struct ip6_hdr *ip6, const struct tcphdr *tcp) { struct pf_os_fingerprint fp, *fpresult; int cnt, optlen = 0; const u_int8_t *optp; #ifdef INET6 char srcname[INET6_ADDRSTRLEN]; #else char srcname[INET_ADDRSTRLEN]; #endif if ((tcp->th_flags & (TH_SYN|TH_ACK)) != TH_SYN) return (NULL); if (ip) { if ((ip->ip_off & htons(IP_OFFMASK)) != 0) return (NULL); } memset(&fp, 0, sizeof(fp)); if (ip) { fp.fp_psize = ntohs(ip->ip_len); fp.fp_ttl = ip->ip_ttl; if (ip->ip_off & htons(IP_DF)) fp.fp_flags |= PF_OSFP_DF; inet_ntoa_r(ip->ip_src, srcname); } #ifdef INET6 else if (ip6) { /* jumbo payload? */ fp.fp_psize = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); fp.fp_ttl = ip6->ip6_hlim; fp.fp_flags |= PF_OSFP_DF; fp.fp_flags |= PF_OSFP_INET6; ip6_sprintf(srcname, (const struct in6_addr *)&ip6->ip6_src); } #endif else return (NULL); fp.fp_wsize = ntohs(tcp->th_win); cnt = (tcp->th_off << 2) - sizeof(*tcp); optp = (const u_int8_t *)((const char *)tcp + sizeof(*tcp)); for (; cnt > 0; cnt -= optlen, optp += optlen) { if (*optp == TCPOPT_EOL) break; fp.fp_optcnt++; if (*optp == TCPOPT_NOP) { fp.fp_tcpopts = (fp.fp_tcpopts << PF_OSFP_TCPOPT_BITS) | PF_OSFP_TCPOPT_NOP; optlen = 1; } else { if (cnt < 2) return (NULL); optlen = optp[1]; if (optlen > cnt || optlen < 2) return (NULL); switch (*optp) { case TCPOPT_MAXSEG: if (optlen >= TCPOLEN_MAXSEG) memcpy(&fp.fp_mss, &optp[2], sizeof(fp.fp_mss)); fp.fp_tcpopts = (fp.fp_tcpopts << PF_OSFP_TCPOPT_BITS) | PF_OSFP_TCPOPT_MSS; NTOHS(fp.fp_mss); break; case TCPOPT_WINDOW: if (optlen >= TCPOLEN_WINDOW) memcpy(&fp.fp_wscale, &optp[2], sizeof(fp.fp_wscale)); NTOHS(fp.fp_wscale); fp.fp_tcpopts = (fp.fp_tcpopts << PF_OSFP_TCPOPT_BITS) | PF_OSFP_TCPOPT_WSCALE; break; case TCPOPT_SACK_PERMITTED: fp.fp_tcpopts = (fp.fp_tcpopts << PF_OSFP_TCPOPT_BITS) | PF_OSFP_TCPOPT_SACK; break; case TCPOPT_TIMESTAMP: if (optlen >= TCPOLEN_TIMESTAMP) { u_int32_t ts; memcpy(&ts, &optp[2], sizeof(ts)); if (ts == 0) fp.fp_flags |= PF_OSFP_TS0; } fp.fp_tcpopts = (fp.fp_tcpopts << PF_OSFP_TCPOPT_BITS) | PF_OSFP_TCPOPT_TS; break; default: return (NULL); } } optlen = MAX(optlen, 1); /* paranoia */ } DPFPRINTF("fingerprinted %s:%d %d:%d:%d:%d:%llx (%d) " "(TS=%s,M=%s%d,W=%s%d)\n", srcname, ntohs(tcp->th_sport), fp.fp_wsize, fp.fp_ttl, (fp.fp_flags & PF_OSFP_DF) != 0, fp.fp_psize, (long long int)fp.fp_tcpopts, fp.fp_optcnt, (fp.fp_flags & PF_OSFP_TS0) ? "0" : "", (fp.fp_flags & PF_OSFP_MSS_MOD) ? "%" : (fp.fp_flags & PF_OSFP_MSS_DC) ? "*" : "", fp.fp_mss, (fp.fp_flags & PF_OSFP_WSCALE_MOD) ? "%" : (fp.fp_flags & PF_OSFP_WSCALE_DC) ? "*" : "", fp.fp_wscale); if ((fpresult = pf_osfp_find(&V_pf_osfp_list, &fp, PF_OSFP_MAXTTL_OFFSET))) return (&fpresult->fp_oses); return (NULL); } /* Match a fingerprint ID against a list of OSes */ int pf_osfp_match(struct pf_osfp_enlist *list, pf_osfp_t os) { struct pf_osfp_entry *entry; int os_class, os_version, os_subtype; int en_class, en_version, en_subtype; if (os == PF_OSFP_ANY) return (1); if (list == NULL) { DPFPRINTF("osfp no match against %x\n", os); return (os == PF_OSFP_UNKNOWN); } PF_OSFP_UNPACK(os, os_class, os_version, os_subtype); SLIST_FOREACH(entry, list, fp_entry) { PF_OSFP_UNPACK(entry->fp_os, en_class, en_version, en_subtype); if ((os_class == PF_OSFP_ANY || en_class == os_class) && (os_version == PF_OSFP_ANY || en_version == os_version) && (os_subtype == PF_OSFP_ANY || en_subtype == os_subtype)) { DPFPRINTF("osfp matched %s %s %s %x==%x\n", entry->fp_class_nm, entry->fp_version_nm, entry->fp_subtype_nm, os, entry->fp_os); return (1); } } DPFPRINTF("fingerprint 0x%x didn't match\n", os); return (0); } /* Flush the fingerprint list */ void pf_osfp_flush(void) { struct pf_os_fingerprint *fp; struct pf_osfp_entry *entry; while ((fp = SLIST_FIRST(&V_pf_osfp_list))) { SLIST_REMOVE_HEAD(&V_pf_osfp_list, fp_next); while ((entry = SLIST_FIRST(&fp->fp_oses))) { SLIST_REMOVE_HEAD(&fp->fp_oses, fp_entry); free(entry, M_PFOSFP); } free(fp, M_PFOSFP); } } /* Add a fingerprint */ int pf_osfp_add(struct pf_osfp_ioctl *fpioc) { struct pf_os_fingerprint *fp, fpadd; struct pf_osfp_entry *entry; PF_RULES_WASSERT(); memset(&fpadd, 0, sizeof(fpadd)); fpadd.fp_tcpopts = fpioc->fp_tcpopts; fpadd.fp_wsize = fpioc->fp_wsize; fpadd.fp_psize = fpioc->fp_psize; fpadd.fp_mss = fpioc->fp_mss; fpadd.fp_flags = fpioc->fp_flags; fpadd.fp_optcnt = fpioc->fp_optcnt; fpadd.fp_wscale = fpioc->fp_wscale; fpadd.fp_ttl = fpioc->fp_ttl; #if 0 /* XXX RYAN wants to fix logging */ DPFPRINTF("adding osfp %s %s %s = %s%d:%d:%d:%s%d:0x%llx %d " "(TS=%s,M=%s%d,W=%s%d) %x\n", fpioc->fp_os.fp_class_nm, fpioc->fp_os.fp_version_nm, fpioc->fp_os.fp_subtype_nm, (fpadd.fp_flags & PF_OSFP_WSIZE_MOD) ? "%" : (fpadd.fp_flags & PF_OSFP_WSIZE_MSS) ? "S" : (fpadd.fp_flags & PF_OSFP_WSIZE_MTU) ? "T" : (fpadd.fp_flags & PF_OSFP_WSIZE_DC) ? "*" : "", fpadd.fp_wsize, fpadd.fp_ttl, (fpadd.fp_flags & PF_OSFP_DF) ? 1 : 0, (fpadd.fp_flags & PF_OSFP_PSIZE_MOD) ? "%" : (fpadd.fp_flags & PF_OSFP_PSIZE_DC) ? "*" : "", fpadd.fp_psize, (long long int)fpadd.fp_tcpopts, fpadd.fp_optcnt, (fpadd.fp_flags & PF_OSFP_TS0) ? "0" : "", (fpadd.fp_flags & PF_OSFP_MSS_MOD) ? "%" : (fpadd.fp_flags & PF_OSFP_MSS_DC) ? "*" : "", fpadd.fp_mss, (fpadd.fp_flags & PF_OSFP_WSCALE_MOD) ? "%" : (fpadd.fp_flags & PF_OSFP_WSCALE_DC) ? "*" : "", fpadd.fp_wscale, fpioc->fp_os.fp_os); #endif if ((fp = pf_osfp_find_exact(&V_pf_osfp_list, &fpadd))) { SLIST_FOREACH(entry, &fp->fp_oses, fp_entry) { if (PF_OSFP_ENTRY_EQ(entry, &fpioc->fp_os)) return (EEXIST); } if ((entry = malloc(sizeof(*entry), M_PFOSFP, M_NOWAIT)) == NULL) return (ENOMEM); } else { if ((fp = malloc(sizeof(*fp), M_PFOSFP, M_ZERO | M_NOWAIT)) == NULL) return (ENOMEM); fp->fp_tcpopts = fpioc->fp_tcpopts; fp->fp_wsize = fpioc->fp_wsize; fp->fp_psize = fpioc->fp_psize; fp->fp_mss = fpioc->fp_mss; fp->fp_flags = fpioc->fp_flags; fp->fp_optcnt = fpioc->fp_optcnt; fp->fp_wscale = fpioc->fp_wscale; fp->fp_ttl = fpioc->fp_ttl; SLIST_INIT(&fp->fp_oses); if ((entry = malloc(sizeof(*entry), M_PFOSFP, M_NOWAIT)) == NULL) { free(fp, M_PFOSFP); return (ENOMEM); } pf_osfp_insert(&V_pf_osfp_list, fp); } memcpy(entry, &fpioc->fp_os, sizeof(*entry)); /* Make sure the strings are NUL terminated */ entry->fp_class_nm[sizeof(entry->fp_class_nm)-1] = '\0'; entry->fp_version_nm[sizeof(entry->fp_version_nm)-1] = '\0'; entry->fp_subtype_nm[sizeof(entry->fp_subtype_nm)-1] = '\0'; SLIST_INSERT_HEAD(&fp->fp_oses, entry, fp_entry); #ifdef PFDEBUG if ((fp = pf_osfp_validate())) printf("Invalid fingerprint list\n"); #endif /* PFDEBUG */ return (0); } /* Find a fingerprint in the list */ static struct pf_os_fingerprint * pf_osfp_find(struct pf_osfp_list *list, struct pf_os_fingerprint *find, u_int8_t ttldiff) { struct pf_os_fingerprint *f; #define MATCH_INT(_MOD, _DC, _field) \ if ((f->fp_flags & _DC) == 0) { \ if ((f->fp_flags & _MOD) == 0) { \ if (f->_field != find->_field) \ continue; \ } else { \ if (f->_field == 0 || find->_field % f->_field) \ continue; \ } \ } SLIST_FOREACH(f, list, fp_next) { if (f->fp_tcpopts != find->fp_tcpopts || f->fp_optcnt != find->fp_optcnt || f->fp_ttl < find->fp_ttl || f->fp_ttl - find->fp_ttl > ttldiff || (f->fp_flags & (PF_OSFP_DF|PF_OSFP_TS0)) != (find->fp_flags & (PF_OSFP_DF|PF_OSFP_TS0))) continue; MATCH_INT(PF_OSFP_PSIZE_MOD, PF_OSFP_PSIZE_DC, fp_psize) MATCH_INT(PF_OSFP_MSS_MOD, PF_OSFP_MSS_DC, fp_mss) MATCH_INT(PF_OSFP_WSCALE_MOD, PF_OSFP_WSCALE_DC, fp_wscale) if ((f->fp_flags & PF_OSFP_WSIZE_DC) == 0) { if (f->fp_flags & PF_OSFP_WSIZE_MSS) { if (find->fp_mss == 0) continue; /* * Some "smart" NAT devices and DSL routers will tweak the MSS size and * will set it to whatever is suitable for the link type. */ #define SMART_MSS 1460 if ((find->fp_wsize % find->fp_mss || find->fp_wsize / find->fp_mss != f->fp_wsize) && (find->fp_wsize % SMART_MSS || find->fp_wsize / SMART_MSS != f->fp_wsize)) continue; } else if (f->fp_flags & PF_OSFP_WSIZE_MTU) { if (find->fp_mss == 0) continue; #define MTUOFF (sizeof(struct ip) + sizeof(struct tcphdr)) #define SMART_MTU (SMART_MSS + MTUOFF) if ((find->fp_wsize % (find->fp_mss + MTUOFF) || find->fp_wsize / (find->fp_mss + MTUOFF) != f->fp_wsize) && (find->fp_wsize % SMART_MTU || find->fp_wsize / SMART_MTU != f->fp_wsize)) continue; } else if (f->fp_flags & PF_OSFP_WSIZE_MOD) { if (f->fp_wsize == 0 || find->fp_wsize % f->fp_wsize) continue; } else { if (f->fp_wsize != find->fp_wsize) continue; } } return (f); } return (NULL); } /* Find an exact fingerprint in the list */ static struct pf_os_fingerprint * pf_osfp_find_exact(struct pf_osfp_list *list, struct pf_os_fingerprint *find) { struct pf_os_fingerprint *f; SLIST_FOREACH(f, list, fp_next) { if (f->fp_tcpopts == find->fp_tcpopts && f->fp_wsize == find->fp_wsize && f->fp_psize == find->fp_psize && f->fp_mss == find->fp_mss && f->fp_flags == find->fp_flags && f->fp_optcnt == find->fp_optcnt && f->fp_wscale == find->fp_wscale && f->fp_ttl == find->fp_ttl) return (f); } return (NULL); } /* Insert a fingerprint into the list */ static void pf_osfp_insert(struct pf_osfp_list *list, struct pf_os_fingerprint *ins) { struct pf_os_fingerprint *f, *prev = NULL; /* XXX need to go semi tree based. can key on tcp options */ SLIST_FOREACH(f, list, fp_next) prev = f; if (prev) SLIST_INSERT_AFTER(prev, ins, fp_next); else SLIST_INSERT_HEAD(list, ins, fp_next); } /* Fill a fingerprint by its number (from an ioctl) */ int pf_osfp_get(struct pf_osfp_ioctl *fpioc) { struct pf_os_fingerprint *fp; struct pf_osfp_entry *entry; int num = fpioc->fp_getnum; int i = 0; memset(fpioc, 0, sizeof(*fpioc)); SLIST_FOREACH(fp, &V_pf_osfp_list, fp_next) { SLIST_FOREACH(entry, &fp->fp_oses, fp_entry) { if (i++ == num) { fpioc->fp_mss = fp->fp_mss; fpioc->fp_wsize = fp->fp_wsize; fpioc->fp_flags = fp->fp_flags; fpioc->fp_psize = fp->fp_psize; fpioc->fp_ttl = fp->fp_ttl; fpioc->fp_wscale = fp->fp_wscale; fpioc->fp_getnum = num; memcpy(&fpioc->fp_os, entry, sizeof(fpioc->fp_os)); return (0); } } } return (EBUSY); } #ifdef PFDEBUG /* Validate that each signature is reachable */ static struct pf_os_fingerprint * pf_osfp_validate(void) { struct pf_os_fingerprint *f, *f2, find; SLIST_FOREACH(f, &V_pf_osfp_list, fp_next) { memcpy(&find, f, sizeof(find)); /* We do a few MSS/th_win percolations to make things unique */ if (find.fp_mss == 0) find.fp_mss = 128; if (f->fp_flags & PF_OSFP_WSIZE_MSS) find.fp_wsize *= find.fp_mss; else if (f->fp_flags & PF_OSFP_WSIZE_MTU) find.fp_wsize *= (find.fp_mss + 40); else if (f->fp_flags & PF_OSFP_WSIZE_MOD) find.fp_wsize *= 2; if (f != (f2 = pf_osfp_find(&V_pf_osfp_list, &find, 0))) { if (f2) printf("Found \"%s %s %s\" instead of " "\"%s %s %s\"\n", SLIST_FIRST(&f2->fp_oses)->fp_class_nm, SLIST_FIRST(&f2->fp_oses)->fp_version_nm, SLIST_FIRST(&f2->fp_oses)->fp_subtype_nm, SLIST_FIRST(&f->fp_oses)->fp_class_nm, SLIST_FIRST(&f->fp_oses)->fp_version_nm, SLIST_FIRST(&f->fp_oses)->fp_subtype_nm); else printf("Couldn't find \"%s %s %s\"\n", SLIST_FIRST(&f->fp_oses)->fp_class_nm, SLIST_FIRST(&f->fp_oses)->fp_version_nm, SLIST_FIRST(&f->fp_oses)->fp_subtype_nm); return (f); } } return (NULL); } #endif /* PFDEBUG */ Index: head/sys/netpfil/pf/pf_table.c =================================================================== --- head/sys/netpfil/pf/pf_table.c (revision 336675) +++ head/sys/netpfil/pf/pf_table.c (revision 336676) @@ -1,2200 +1,2200 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002 Cedric Berger * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * $OpenBSD: pf_table.c,v 1.79 2008/10/08 06:24:50 mcbride Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #define ACCEPT_FLAGS(flags, oklist) \ do { \ if ((flags & ~(oklist)) & \ PFR_FLAG_ALLMASK) \ return (EINVAL); \ } while (0) #define FILLIN_SIN(sin, addr) \ do { \ (sin).sin_len = sizeof(sin); \ (sin).sin_family = AF_INET; \ (sin).sin_addr = (addr); \ } while (0) #define FILLIN_SIN6(sin6, addr) \ do { \ (sin6).sin6_len = sizeof(sin6); \ (sin6).sin6_family = AF_INET6; \ (sin6).sin6_addr = (addr); \ } while (0) #define SWAP(type, a1, a2) \ do { \ type tmp = a1; \ a1 = a2; \ a2 = tmp; \ } while (0) #define SUNION2PF(su, af) (((af)==AF_INET) ? \ (struct pf_addr *)&(su)->sin.sin_addr : \ (struct pf_addr *)&(su)->sin6.sin6_addr) #define AF_BITS(af) (((af)==AF_INET)?32:128) #define ADDR_NETWORK(ad) ((ad)->pfra_net < AF_BITS((ad)->pfra_af)) #define KENTRY_NETWORK(ke) ((ke)->pfrke_net < AF_BITS((ke)->pfrke_af)) #define KENTRY_RNF_ROOT(ke) \ ((((struct radix_node *)(ke))->rn_flags & RNF_ROOT) != 0) #define NO_ADDRESSES (-1) #define ENQUEUE_UNMARKED_ONLY (1) #define INVERT_NEG_FLAG (1) struct pfr_walktree { enum pfrw_op { PFRW_MARK, PFRW_SWEEP, PFRW_ENQUEUE, PFRW_GET_ADDRS, PFRW_GET_ASTATS, PFRW_POOL_GET, PFRW_DYNADDR_UPDATE } pfrw_op; union { struct pfr_addr *pfrw1_addr; struct pfr_astats *pfrw1_astats; struct pfr_kentryworkq *pfrw1_workq; struct pfr_kentry *pfrw1_kentry; struct pfi_dynaddr *pfrw1_dyn; } pfrw_1; int pfrw_free; }; #define pfrw_addr pfrw_1.pfrw1_addr #define pfrw_astats pfrw_1.pfrw1_astats #define pfrw_workq pfrw_1.pfrw1_workq #define pfrw_kentry pfrw_1.pfrw1_kentry #define pfrw_dyn pfrw_1.pfrw1_dyn #define pfrw_cnt pfrw_free #define senderr(e) do { rv = (e); goto _bad; } while (0) static MALLOC_DEFINE(M_PFTABLE, "pf_table", "pf(4) tables structures"); -static VNET_DEFINE(uma_zone_t, pfr_kentry_z); +VNET_DEFINE_STATIC(uma_zone_t, pfr_kentry_z); #define V_pfr_kentry_z VNET(pfr_kentry_z) -static VNET_DEFINE(uma_zone_t, pfr_kcounters_z); +VNET_DEFINE_STATIC(uma_zone_t, pfr_kcounters_z); #define V_pfr_kcounters_z VNET(pfr_kcounters_z) static struct pf_addr pfr_ffaddr = { .addr32 = { 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff } }; static void pfr_copyout_addr(struct pfr_addr *, struct pfr_kentry *ke); static int pfr_validate_addr(struct pfr_addr *); static void pfr_enqueue_addrs(struct pfr_ktable *, struct pfr_kentryworkq *, int *, int); static void pfr_mark_addrs(struct pfr_ktable *); static struct pfr_kentry *pfr_lookup_addr(struct pfr_ktable *, struct pfr_addr *, int); static struct pfr_kentry *pfr_create_kentry(struct pfr_addr *); static void pfr_destroy_kentries(struct pfr_kentryworkq *); static void pfr_destroy_kentry(struct pfr_kentry *); static void pfr_insert_kentries(struct pfr_ktable *, struct pfr_kentryworkq *, long); static void pfr_remove_kentries(struct pfr_ktable *, struct pfr_kentryworkq *); static void pfr_clstats_kentries(struct pfr_kentryworkq *, long, int); static void pfr_reset_feedback(struct pfr_addr *, int); static void pfr_prepare_network(union sockaddr_union *, int, int); static int pfr_route_kentry(struct pfr_ktable *, struct pfr_kentry *); static int pfr_unroute_kentry(struct pfr_ktable *, struct pfr_kentry *); static int pfr_walktree(struct radix_node *, void *); static int pfr_validate_table(struct pfr_table *, int, int); static int pfr_fix_anchor(char *); static void pfr_commit_ktable(struct pfr_ktable *, long); static void pfr_insert_ktables(struct pfr_ktableworkq *); static void pfr_insert_ktable(struct pfr_ktable *); static void pfr_setflags_ktables(struct pfr_ktableworkq *); static void pfr_setflags_ktable(struct pfr_ktable *, int); static void pfr_clstats_ktables(struct pfr_ktableworkq *, long, int); static void pfr_clstats_ktable(struct pfr_ktable *, long, int); static struct pfr_ktable *pfr_create_ktable(struct pfr_table *, long, int); static void pfr_destroy_ktables(struct pfr_ktableworkq *, int); static void pfr_destroy_ktable(struct pfr_ktable *, int); static int pfr_ktable_compare(struct pfr_ktable *, struct pfr_ktable *); static struct pfr_ktable *pfr_lookup_table(struct pfr_table *); static void pfr_clean_node_mask(struct pfr_ktable *, struct pfr_kentryworkq *); static int pfr_skip_table(struct pfr_table *, struct pfr_ktable *, int); static struct pfr_kentry *pfr_kentry_byidx(struct pfr_ktable *, int, int); static RB_PROTOTYPE(pfr_ktablehead, pfr_ktable, pfrkt_tree, pfr_ktable_compare); static RB_GENERATE(pfr_ktablehead, pfr_ktable, pfrkt_tree, pfr_ktable_compare); -static VNET_DEFINE(struct pfr_ktablehead, pfr_ktables); +VNET_DEFINE_STATIC(struct pfr_ktablehead, pfr_ktables); #define V_pfr_ktables VNET(pfr_ktables) -static VNET_DEFINE(struct pfr_table, pfr_nulltable); +VNET_DEFINE_STATIC(struct pfr_table, pfr_nulltable); #define V_pfr_nulltable VNET(pfr_nulltable) -static VNET_DEFINE(int, pfr_ktable_cnt); +VNET_DEFINE_STATIC(int, pfr_ktable_cnt); #define V_pfr_ktable_cnt VNET(pfr_ktable_cnt) void pfr_initialize(void) { V_pfr_kentry_z = uma_zcreate("pf table entries", sizeof(struct pfr_kentry), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pfr_kcounters_z = uma_zcreate("pf table counters", sizeof(struct pfr_kcounters), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_TABLE_ENTRIES].zone = V_pfr_kentry_z; V_pf_limits[PF_LIMIT_TABLE_ENTRIES].limit = PFR_KENTRY_HIWAT; } void pfr_cleanup(void) { uma_zdestroy(V_pfr_kentry_z); uma_zdestroy(V_pfr_kcounters_z); } int pfr_clr_addrs(struct pfr_table *tbl, int *ndel, int flags) { struct pfr_ktable *kt; struct pfr_kentryworkq workq; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); if (pfr_validate_table(tbl, 0, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_flags & PFR_TFLAG_CONST) return (EPERM); pfr_enqueue_addrs(kt, &workq, ndel, 0); if (!(flags & PFR_FLAG_DUMMY)) { pfr_remove_kentries(kt, &workq); KASSERT(kt->pfrkt_cnt == 0, ("%s: non-null pfrkt_cnt", __func__)); } return (0); } int pfr_add_addrs(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *nadd, int flags) { struct pfr_ktable *kt, *tmpkt; struct pfr_kentryworkq workq; struct pfr_kentry *p, *q; struct pfr_addr *ad; int i, rv, xadd = 0; long tzero = time_second; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_FEEDBACK); if (pfr_validate_table(tbl, 0, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_flags & PFR_TFLAG_CONST) return (EPERM); tmpkt = pfr_create_ktable(&V_pfr_nulltable, 0, 0); if (tmpkt == NULL) return (ENOMEM); SLIST_INIT(&workq); for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) senderr(EINVAL); p = pfr_lookup_addr(kt, ad, 1); q = pfr_lookup_addr(tmpkt, ad, 1); if (flags & PFR_FLAG_FEEDBACK) { if (q != NULL) ad->pfra_fback = PFR_FB_DUPLICATE; else if (p == NULL) ad->pfra_fback = PFR_FB_ADDED; else if (p->pfrke_not != ad->pfra_not) ad->pfra_fback = PFR_FB_CONFLICT; else ad->pfra_fback = PFR_FB_NONE; } if (p == NULL && q == NULL) { p = pfr_create_kentry(ad); if (p == NULL) senderr(ENOMEM); if (pfr_route_kentry(tmpkt, p)) { pfr_destroy_kentry(p); ad->pfra_fback = PFR_FB_NONE; } else { SLIST_INSERT_HEAD(&workq, p, pfrke_workq); xadd++; } } } pfr_clean_node_mask(tmpkt, &workq); if (!(flags & PFR_FLAG_DUMMY)) pfr_insert_kentries(kt, &workq, tzero); else pfr_destroy_kentries(&workq); if (nadd != NULL) *nadd = xadd; pfr_destroy_ktable(tmpkt, 0); return (0); _bad: pfr_clean_node_mask(tmpkt, &workq); pfr_destroy_kentries(&workq); if (flags & PFR_FLAG_FEEDBACK) pfr_reset_feedback(addr, size); pfr_destroy_ktable(tmpkt, 0); return (rv); } int pfr_del_addrs(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *ndel, int flags) { struct pfr_ktable *kt; struct pfr_kentryworkq workq; struct pfr_kentry *p; struct pfr_addr *ad; int i, rv, xdel = 0, log = 1; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_FEEDBACK); if (pfr_validate_table(tbl, 0, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_flags & PFR_TFLAG_CONST) return (EPERM); /* * there are two algorithms to choose from here. * with: * n: number of addresses to delete * N: number of addresses in the table * * one is O(N) and is better for large 'n' * one is O(n*LOG(N)) and is better for small 'n' * * following code try to decide which one is best. */ for (i = kt->pfrkt_cnt; i > 0; i >>= 1) log++; if (size > kt->pfrkt_cnt/log) { /* full table scan */ pfr_mark_addrs(kt); } else { /* iterate over addresses to delete */ for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) return (EINVAL); p = pfr_lookup_addr(kt, ad, 1); if (p != NULL) p->pfrke_mark = 0; } } SLIST_INIT(&workq); for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) senderr(EINVAL); p = pfr_lookup_addr(kt, ad, 1); if (flags & PFR_FLAG_FEEDBACK) { if (p == NULL) ad->pfra_fback = PFR_FB_NONE; else if (p->pfrke_not != ad->pfra_not) ad->pfra_fback = PFR_FB_CONFLICT; else if (p->pfrke_mark) ad->pfra_fback = PFR_FB_DUPLICATE; else ad->pfra_fback = PFR_FB_DELETED; } if (p != NULL && p->pfrke_not == ad->pfra_not && !p->pfrke_mark) { p->pfrke_mark = 1; SLIST_INSERT_HEAD(&workq, p, pfrke_workq); xdel++; } } if (!(flags & PFR_FLAG_DUMMY)) pfr_remove_kentries(kt, &workq); if (ndel != NULL) *ndel = xdel; return (0); _bad: if (flags & PFR_FLAG_FEEDBACK) pfr_reset_feedback(addr, size); return (rv); } int pfr_set_addrs(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *size2, int *nadd, int *ndel, int *nchange, int flags, u_int32_t ignore_pfrt_flags) { struct pfr_ktable *kt, *tmpkt; struct pfr_kentryworkq addq, delq, changeq; struct pfr_kentry *p, *q; struct pfr_addr ad; int i, rv, xadd = 0, xdel = 0, xchange = 0; long tzero = time_second; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_FEEDBACK); if (pfr_validate_table(tbl, ignore_pfrt_flags, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_flags & PFR_TFLAG_CONST) return (EPERM); tmpkt = pfr_create_ktable(&V_pfr_nulltable, 0, 0); if (tmpkt == NULL) return (ENOMEM); pfr_mark_addrs(kt); SLIST_INIT(&addq); SLIST_INIT(&delq); SLIST_INIT(&changeq); for (i = 0; i < size; i++) { /* * XXXGL: undertand pf_if usage of this function * and make ad a moving pointer */ bcopy(addr + i, &ad, sizeof(ad)); if (pfr_validate_addr(&ad)) senderr(EINVAL); ad.pfra_fback = PFR_FB_NONE; p = pfr_lookup_addr(kt, &ad, 1); if (p != NULL) { if (p->pfrke_mark) { ad.pfra_fback = PFR_FB_DUPLICATE; goto _skip; } p->pfrke_mark = 1; if (p->pfrke_not != ad.pfra_not) { SLIST_INSERT_HEAD(&changeq, p, pfrke_workq); ad.pfra_fback = PFR_FB_CHANGED; xchange++; } } else { q = pfr_lookup_addr(tmpkt, &ad, 1); if (q != NULL) { ad.pfra_fback = PFR_FB_DUPLICATE; goto _skip; } p = pfr_create_kentry(&ad); if (p == NULL) senderr(ENOMEM); if (pfr_route_kentry(tmpkt, p)) { pfr_destroy_kentry(p); ad.pfra_fback = PFR_FB_NONE; } else { SLIST_INSERT_HEAD(&addq, p, pfrke_workq); ad.pfra_fback = PFR_FB_ADDED; xadd++; } } _skip: if (flags & PFR_FLAG_FEEDBACK) bcopy(&ad, addr + i, sizeof(ad)); } pfr_enqueue_addrs(kt, &delq, &xdel, ENQUEUE_UNMARKED_ONLY); if ((flags & PFR_FLAG_FEEDBACK) && *size2) { if (*size2 < size+xdel) { *size2 = size+xdel; senderr(0); } i = 0; SLIST_FOREACH(p, &delq, pfrke_workq) { pfr_copyout_addr(&ad, p); ad.pfra_fback = PFR_FB_DELETED; bcopy(&ad, addr + size + i, sizeof(ad)); i++; } } pfr_clean_node_mask(tmpkt, &addq); if (!(flags & PFR_FLAG_DUMMY)) { pfr_insert_kentries(kt, &addq, tzero); pfr_remove_kentries(kt, &delq); pfr_clstats_kentries(&changeq, tzero, INVERT_NEG_FLAG); } else pfr_destroy_kentries(&addq); if (nadd != NULL) *nadd = xadd; if (ndel != NULL) *ndel = xdel; if (nchange != NULL) *nchange = xchange; if ((flags & PFR_FLAG_FEEDBACK) && size2) *size2 = size+xdel; pfr_destroy_ktable(tmpkt, 0); return (0); _bad: pfr_clean_node_mask(tmpkt, &addq); pfr_destroy_kentries(&addq); if (flags & PFR_FLAG_FEEDBACK) pfr_reset_feedback(addr, size); pfr_destroy_ktable(tmpkt, 0); return (rv); } int pfr_tst_addrs(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *nmatch, int flags) { struct pfr_ktable *kt; struct pfr_kentry *p; struct pfr_addr *ad; int i, xmatch = 0; PF_RULES_RASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_REPLACE); if (pfr_validate_table(tbl, 0, 0)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) return (EINVAL); if (ADDR_NETWORK(ad)) return (EINVAL); p = pfr_lookup_addr(kt, ad, 0); if (flags & PFR_FLAG_REPLACE) pfr_copyout_addr(ad, p); ad->pfra_fback = (p == NULL) ? PFR_FB_NONE : (p->pfrke_not ? PFR_FB_NOTMATCH : PFR_FB_MATCH); if (p != NULL && !p->pfrke_not) xmatch++; } if (nmatch != NULL) *nmatch = xmatch; return (0); } int pfr_get_addrs(struct pfr_table *tbl, struct pfr_addr *addr, int *size, int flags) { struct pfr_ktable *kt; struct pfr_walktree w; int rv; PF_RULES_RASSERT(); ACCEPT_FLAGS(flags, 0); if (pfr_validate_table(tbl, 0, 0)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_cnt > *size) { *size = kt->pfrkt_cnt; return (0); } bzero(&w, sizeof(w)); w.pfrw_op = PFRW_GET_ADDRS; w.pfrw_addr = addr; w.pfrw_free = kt->pfrkt_cnt; rv = kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w); if (!rv) rv = kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w); if (rv) return (rv); KASSERT(w.pfrw_free == 0, ("%s: corruption detected (%d)", __func__, w.pfrw_free)); *size = kt->pfrkt_cnt; return (0); } int pfr_get_astats(struct pfr_table *tbl, struct pfr_astats *addr, int *size, int flags) { struct pfr_ktable *kt; struct pfr_walktree w; struct pfr_kentryworkq workq; int rv; long tzero = time_second; PF_RULES_RASSERT(); /* XXX PFR_FLAG_CLSTATS disabled */ ACCEPT_FLAGS(flags, 0); if (pfr_validate_table(tbl, 0, 0)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); if (kt->pfrkt_cnt > *size) { *size = kt->pfrkt_cnt; return (0); } bzero(&w, sizeof(w)); w.pfrw_op = PFRW_GET_ASTATS; w.pfrw_astats = addr; w.pfrw_free = kt->pfrkt_cnt; rv = kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w); if (!rv) rv = kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w); if (!rv && (flags & PFR_FLAG_CLSTATS)) { pfr_enqueue_addrs(kt, &workq, NULL, 0); pfr_clstats_kentries(&workq, tzero, 0); } if (rv) return (rv); if (w.pfrw_free) { printf("pfr_get_astats: corruption detected (%d).\n", w.pfrw_free); return (ENOTTY); } *size = kt->pfrkt_cnt; return (0); } int pfr_clr_astats(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *nzero, int flags) { struct pfr_ktable *kt; struct pfr_kentryworkq workq; struct pfr_kentry *p; struct pfr_addr *ad; int i, rv, xzero = 0; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_FEEDBACK); if (pfr_validate_table(tbl, 0, 0)) return (EINVAL); kt = pfr_lookup_table(tbl); if (kt == NULL || !(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (ESRCH); SLIST_INIT(&workq); for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) senderr(EINVAL); p = pfr_lookup_addr(kt, ad, 1); if (flags & PFR_FLAG_FEEDBACK) { ad->pfra_fback = (p != NULL) ? PFR_FB_CLEARED : PFR_FB_NONE; } if (p != NULL) { SLIST_INSERT_HEAD(&workq, p, pfrke_workq); xzero++; } } if (!(flags & PFR_FLAG_DUMMY)) pfr_clstats_kentries(&workq, 0, 0); if (nzero != NULL) *nzero = xzero; return (0); _bad: if (flags & PFR_FLAG_FEEDBACK) pfr_reset_feedback(addr, size); return (rv); } static int pfr_validate_addr(struct pfr_addr *ad) { int i; switch (ad->pfra_af) { #ifdef INET case AF_INET: if (ad->pfra_net > 32) return (-1); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (ad->pfra_net > 128) return (-1); break; #endif /* INET6 */ default: return (-1); } if (ad->pfra_net < 128 && (((caddr_t)ad)[ad->pfra_net/8] & (0xFF >> (ad->pfra_net%8)))) return (-1); for (i = (ad->pfra_net+7)/8; i < sizeof(ad->pfra_u); i++) if (((caddr_t)ad)[i]) return (-1); if (ad->pfra_not && ad->pfra_not != 1) return (-1); if (ad->pfra_fback) return (-1); return (0); } static void pfr_enqueue_addrs(struct pfr_ktable *kt, struct pfr_kentryworkq *workq, int *naddr, int sweep) { struct pfr_walktree w; SLIST_INIT(workq); bzero(&w, sizeof(w)); w.pfrw_op = sweep ? PFRW_SWEEP : PFRW_ENQUEUE; w.pfrw_workq = workq; if (kt->pfrkt_ip4 != NULL) if (kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w)) printf("pfr_enqueue_addrs: IPv4 walktree failed.\n"); if (kt->pfrkt_ip6 != NULL) if (kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w)) printf("pfr_enqueue_addrs: IPv6 walktree failed.\n"); if (naddr != NULL) *naddr = w.pfrw_cnt; } static void pfr_mark_addrs(struct pfr_ktable *kt) { struct pfr_walktree w; bzero(&w, sizeof(w)); w.pfrw_op = PFRW_MARK; if (kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w)) printf("pfr_mark_addrs: IPv4 walktree failed.\n"); if (kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w)) printf("pfr_mark_addrs: IPv6 walktree failed.\n"); } static struct pfr_kentry * pfr_lookup_addr(struct pfr_ktable *kt, struct pfr_addr *ad, int exact) { union sockaddr_union sa, mask; struct radix_head *head = NULL; struct pfr_kentry *ke; PF_RULES_ASSERT(); bzero(&sa, sizeof(sa)); if (ad->pfra_af == AF_INET) { FILLIN_SIN(sa.sin, ad->pfra_ip4addr); head = &kt->pfrkt_ip4->rh; } else if ( ad->pfra_af == AF_INET6 ) { FILLIN_SIN6(sa.sin6, ad->pfra_ip6addr); head = &kt->pfrkt_ip6->rh; } if (ADDR_NETWORK(ad)) { pfr_prepare_network(&mask, ad->pfra_af, ad->pfra_net); ke = (struct pfr_kentry *)rn_lookup(&sa, &mask, head); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; } else { ke = (struct pfr_kentry *)rn_match(&sa, head); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; if (exact && ke && KENTRY_NETWORK(ke)) ke = NULL; } return (ke); } static struct pfr_kentry * pfr_create_kentry(struct pfr_addr *ad) { struct pfr_kentry *ke; ke = uma_zalloc(V_pfr_kentry_z, M_NOWAIT | M_ZERO); if (ke == NULL) return (NULL); if (ad->pfra_af == AF_INET) FILLIN_SIN(ke->pfrke_sa.sin, ad->pfra_ip4addr); else if (ad->pfra_af == AF_INET6) FILLIN_SIN6(ke->pfrke_sa.sin6, ad->pfra_ip6addr); ke->pfrke_af = ad->pfra_af; ke->pfrke_net = ad->pfra_net; ke->pfrke_not = ad->pfra_not; return (ke); } static void pfr_destroy_kentries(struct pfr_kentryworkq *workq) { struct pfr_kentry *p, *q; for (p = SLIST_FIRST(workq); p != NULL; p = q) { q = SLIST_NEXT(p, pfrke_workq); pfr_destroy_kentry(p); } } static void pfr_destroy_kentry(struct pfr_kentry *ke) { if (ke->pfrke_counters) uma_zfree(V_pfr_kcounters_z, ke->pfrke_counters); uma_zfree(V_pfr_kentry_z, ke); } static void pfr_insert_kentries(struct pfr_ktable *kt, struct pfr_kentryworkq *workq, long tzero) { struct pfr_kentry *p; int rv, n = 0; SLIST_FOREACH(p, workq, pfrke_workq) { rv = pfr_route_kentry(kt, p); if (rv) { printf("pfr_insert_kentries: cannot route entry " "(code=%d).\n", rv); break; } p->pfrke_tzero = tzero; n++; } kt->pfrkt_cnt += n; } int pfr_insert_kentry(struct pfr_ktable *kt, struct pfr_addr *ad, long tzero) { struct pfr_kentry *p; int rv; p = pfr_lookup_addr(kt, ad, 1); if (p != NULL) return (0); p = pfr_create_kentry(ad); if (p == NULL) return (ENOMEM); rv = pfr_route_kentry(kt, p); if (rv) return (rv); p->pfrke_tzero = tzero; kt->pfrkt_cnt++; return (0); } static void pfr_remove_kentries(struct pfr_ktable *kt, struct pfr_kentryworkq *workq) { struct pfr_kentry *p; int n = 0; SLIST_FOREACH(p, workq, pfrke_workq) { pfr_unroute_kentry(kt, p); n++; } kt->pfrkt_cnt -= n; pfr_destroy_kentries(workq); } static void pfr_clean_node_mask(struct pfr_ktable *kt, struct pfr_kentryworkq *workq) { struct pfr_kentry *p; SLIST_FOREACH(p, workq, pfrke_workq) pfr_unroute_kentry(kt, p); } static void pfr_clstats_kentries(struct pfr_kentryworkq *workq, long tzero, int negchange) { struct pfr_kentry *p; SLIST_FOREACH(p, workq, pfrke_workq) { if (negchange) p->pfrke_not = !p->pfrke_not; if (p->pfrke_counters) { uma_zfree(V_pfr_kcounters_z, p->pfrke_counters); p->pfrke_counters = NULL; } p->pfrke_tzero = tzero; } } static void pfr_reset_feedback(struct pfr_addr *addr, int size) { struct pfr_addr *ad; int i; for (i = 0, ad = addr; i < size; i++, ad++) ad->pfra_fback = PFR_FB_NONE; } static void pfr_prepare_network(union sockaddr_union *sa, int af, int net) { int i; bzero(sa, sizeof(*sa)); if (af == AF_INET) { sa->sin.sin_len = sizeof(sa->sin); sa->sin.sin_family = AF_INET; sa->sin.sin_addr.s_addr = net ? htonl(-1 << (32-net)) : 0; } else if (af == AF_INET6) { sa->sin6.sin6_len = sizeof(sa->sin6); sa->sin6.sin6_family = AF_INET6; for (i = 0; i < 4; i++) { if (net <= 32) { sa->sin6.sin6_addr.s6_addr32[i] = net ? htonl(-1 << (32-net)) : 0; break; } sa->sin6.sin6_addr.s6_addr32[i] = 0xFFFFFFFF; net -= 32; } } } static int pfr_route_kentry(struct pfr_ktable *kt, struct pfr_kentry *ke) { union sockaddr_union mask; struct radix_node *rn; struct radix_head *head = NULL; PF_RULES_WASSERT(); bzero(ke->pfrke_node, sizeof(ke->pfrke_node)); if (ke->pfrke_af == AF_INET) head = &kt->pfrkt_ip4->rh; else if (ke->pfrke_af == AF_INET6) head = &kt->pfrkt_ip6->rh; if (KENTRY_NETWORK(ke)) { pfr_prepare_network(&mask, ke->pfrke_af, ke->pfrke_net); rn = rn_addroute(&ke->pfrke_sa, &mask, head, ke->pfrke_node); } else rn = rn_addroute(&ke->pfrke_sa, NULL, head, ke->pfrke_node); return (rn == NULL ? -1 : 0); } static int pfr_unroute_kentry(struct pfr_ktable *kt, struct pfr_kentry *ke) { union sockaddr_union mask; struct radix_node *rn; struct radix_head *head = NULL; if (ke->pfrke_af == AF_INET) head = &kt->pfrkt_ip4->rh; else if (ke->pfrke_af == AF_INET6) head = &kt->pfrkt_ip6->rh; if (KENTRY_NETWORK(ke)) { pfr_prepare_network(&mask, ke->pfrke_af, ke->pfrke_net); rn = rn_delete(&ke->pfrke_sa, &mask, head); } else rn = rn_delete(&ke->pfrke_sa, NULL, head); if (rn == NULL) { printf("pfr_unroute_kentry: delete failed.\n"); return (-1); } return (0); } static void pfr_copyout_addr(struct pfr_addr *ad, struct pfr_kentry *ke) { bzero(ad, sizeof(*ad)); if (ke == NULL) return; ad->pfra_af = ke->pfrke_af; ad->pfra_net = ke->pfrke_net; ad->pfra_not = ke->pfrke_not; if (ad->pfra_af == AF_INET) ad->pfra_ip4addr = ke->pfrke_sa.sin.sin_addr; else if (ad->pfra_af == AF_INET6) ad->pfra_ip6addr = ke->pfrke_sa.sin6.sin6_addr; } static int pfr_walktree(struct radix_node *rn, void *arg) { struct pfr_kentry *ke = (struct pfr_kentry *)rn; struct pfr_walktree *w = arg; switch (w->pfrw_op) { case PFRW_MARK: ke->pfrke_mark = 0; break; case PFRW_SWEEP: if (ke->pfrke_mark) break; /* FALLTHROUGH */ case PFRW_ENQUEUE: SLIST_INSERT_HEAD(w->pfrw_workq, ke, pfrke_workq); w->pfrw_cnt++; break; case PFRW_GET_ADDRS: if (w->pfrw_free-- > 0) { pfr_copyout_addr(w->pfrw_addr, ke); w->pfrw_addr++; } break; case PFRW_GET_ASTATS: if (w->pfrw_free-- > 0) { struct pfr_astats as; pfr_copyout_addr(&as.pfras_a, ke); if (ke->pfrke_counters) { bcopy(ke->pfrke_counters->pfrkc_packets, as.pfras_packets, sizeof(as.pfras_packets)); bcopy(ke->pfrke_counters->pfrkc_bytes, as.pfras_bytes, sizeof(as.pfras_bytes)); } else { bzero(as.pfras_packets, sizeof(as.pfras_packets)); bzero(as.pfras_bytes, sizeof(as.pfras_bytes)); as.pfras_a.pfra_fback = PFR_FB_NOCOUNT; } as.pfras_tzero = ke->pfrke_tzero; bcopy(&as, w->pfrw_astats, sizeof(as)); w->pfrw_astats++; } break; case PFRW_POOL_GET: if (ke->pfrke_not) break; /* negative entries are ignored */ if (!w->pfrw_cnt--) { w->pfrw_kentry = ke; return (1); /* finish search */ } break; case PFRW_DYNADDR_UPDATE: { union sockaddr_union pfr_mask; if (ke->pfrke_af == AF_INET) { if (w->pfrw_dyn->pfid_acnt4++ > 0) break; pfr_prepare_network(&pfr_mask, AF_INET, ke->pfrke_net); w->pfrw_dyn->pfid_addr4 = *SUNION2PF(&ke->pfrke_sa, AF_INET); w->pfrw_dyn->pfid_mask4 = *SUNION2PF(&pfr_mask, AF_INET); } else if (ke->pfrke_af == AF_INET6){ if (w->pfrw_dyn->pfid_acnt6++ > 0) break; pfr_prepare_network(&pfr_mask, AF_INET6, ke->pfrke_net); w->pfrw_dyn->pfid_addr6 = *SUNION2PF(&ke->pfrke_sa, AF_INET6); w->pfrw_dyn->pfid_mask6 = *SUNION2PF(&pfr_mask, AF_INET6); } break; } } return (0); } int pfr_clr_tables(struct pfr_table *filter, int *ndel, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p; int xdel = 0; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_ALLRSETS); if (pfr_fix_anchor(filter->pfrt_anchor)) return (EINVAL); if (pfr_table_count(filter, flags) < 0) return (ENOENT); SLIST_INIT(&workq); RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (pfr_skip_table(filter, p, flags)) continue; if (!strcmp(p->pfrkt_anchor, PF_RESERVED_ANCHOR)) continue; if (!(p->pfrkt_flags & PFR_TFLAG_ACTIVE)) continue; p->pfrkt_nflags = p->pfrkt_flags & ~PFR_TFLAG_ACTIVE; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); xdel++; } if (!(flags & PFR_FLAG_DUMMY)) pfr_setflags_ktables(&workq); if (ndel != NULL) *ndel = xdel; return (0); } int pfr_add_tables(struct pfr_table *tbl, int size, int *nadd, int flags) { struct pfr_ktableworkq addq, changeq; struct pfr_ktable *p, *q, *r, key; int i, rv, xadd = 0; long tzero = time_second; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); SLIST_INIT(&addq); SLIST_INIT(&changeq); for (i = 0; i < size; i++) { bcopy(tbl+i, &key.pfrkt_t, sizeof(key.pfrkt_t)); if (pfr_validate_table(&key.pfrkt_t, PFR_TFLAG_USRMASK, flags & PFR_FLAG_USERIOCTL)) senderr(EINVAL); key.pfrkt_flags |= PFR_TFLAG_ACTIVE; p = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (p == NULL) { p = pfr_create_ktable(&key.pfrkt_t, tzero, 1); if (p == NULL) senderr(ENOMEM); SLIST_FOREACH(q, &addq, pfrkt_workq) { if (!pfr_ktable_compare(p, q)) { pfr_destroy_ktable(p, 0); goto _skip; } } SLIST_INSERT_HEAD(&addq, p, pfrkt_workq); xadd++; if (!key.pfrkt_anchor[0]) goto _skip; /* find or create root table */ bzero(key.pfrkt_anchor, sizeof(key.pfrkt_anchor)); r = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (r != NULL) { p->pfrkt_root = r; goto _skip; } SLIST_FOREACH(q, &addq, pfrkt_workq) { if (!pfr_ktable_compare(&key, q)) { p->pfrkt_root = q; goto _skip; } } key.pfrkt_flags = 0; r = pfr_create_ktable(&key.pfrkt_t, 0, 1); if (r == NULL) senderr(ENOMEM); SLIST_INSERT_HEAD(&addq, r, pfrkt_workq); p->pfrkt_root = r; } else if (!(p->pfrkt_flags & PFR_TFLAG_ACTIVE)) { SLIST_FOREACH(q, &changeq, pfrkt_workq) if (!pfr_ktable_compare(&key, q)) goto _skip; p->pfrkt_nflags = (p->pfrkt_flags & ~PFR_TFLAG_USRMASK) | key.pfrkt_flags; SLIST_INSERT_HEAD(&changeq, p, pfrkt_workq); xadd++; } _skip: ; } if (!(flags & PFR_FLAG_DUMMY)) { pfr_insert_ktables(&addq); pfr_setflags_ktables(&changeq); } else pfr_destroy_ktables(&addq, 0); if (nadd != NULL) *nadd = xadd; return (0); _bad: pfr_destroy_ktables(&addq, 0); return (rv); } int pfr_del_tables(struct pfr_table *tbl, int size, int *ndel, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p, *q, key; int i, xdel = 0; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); SLIST_INIT(&workq); for (i = 0; i < size; i++) { bcopy(tbl+i, &key.pfrkt_t, sizeof(key.pfrkt_t)); if (pfr_validate_table(&key.pfrkt_t, 0, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); p = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (p != NULL && (p->pfrkt_flags & PFR_TFLAG_ACTIVE)) { SLIST_FOREACH(q, &workq, pfrkt_workq) if (!pfr_ktable_compare(p, q)) goto _skip; p->pfrkt_nflags = p->pfrkt_flags & ~PFR_TFLAG_ACTIVE; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); xdel++; } _skip: ; } if (!(flags & PFR_FLAG_DUMMY)) pfr_setflags_ktables(&workq); if (ndel != NULL) *ndel = xdel; return (0); } int pfr_get_tables(struct pfr_table *filter, struct pfr_table *tbl, int *size, int flags) { struct pfr_ktable *p; int n, nn; PF_RULES_RASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_ALLRSETS); if (pfr_fix_anchor(filter->pfrt_anchor)) return (EINVAL); n = nn = pfr_table_count(filter, flags); if (n < 0) return (ENOENT); if (n > *size) { *size = n; return (0); } RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (pfr_skip_table(filter, p, flags)) continue; if (n-- <= 0) continue; bcopy(&p->pfrkt_t, tbl++, sizeof(*tbl)); } KASSERT(n == 0, ("%s: corruption detected (%d)", __func__, n)); *size = nn; return (0); } int pfr_get_tstats(struct pfr_table *filter, struct pfr_tstats *tbl, int *size, int flags) { struct pfr_ktable *p; struct pfr_ktableworkq workq; int n, nn; long tzero = time_second; /* XXX PFR_FLAG_CLSTATS disabled */ ACCEPT_FLAGS(flags, PFR_FLAG_ALLRSETS); if (pfr_fix_anchor(filter->pfrt_anchor)) return (EINVAL); n = nn = pfr_table_count(filter, flags); if (n < 0) return (ENOENT); if (n > *size) { *size = n; return (0); } SLIST_INIT(&workq); RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (pfr_skip_table(filter, p, flags)) continue; if (n-- <= 0) continue; bcopy(&p->pfrkt_ts, tbl++, sizeof(*tbl)); SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); } if (flags & PFR_FLAG_CLSTATS) pfr_clstats_ktables(&workq, tzero, flags & PFR_FLAG_ADDRSTOO); KASSERT(n == 0, ("%s: corruption detected (%d)", __func__, n)); *size = nn; return (0); } int pfr_clr_tstats(struct pfr_table *tbl, int size, int *nzero, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p, key; int i, xzero = 0; long tzero = time_second; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_ADDRSTOO); SLIST_INIT(&workq); for (i = 0; i < size; i++) { bcopy(tbl + i, &key.pfrkt_t, sizeof(key.pfrkt_t)); if (pfr_validate_table(&key.pfrkt_t, 0, 0)) return (EINVAL); p = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (p != NULL) { SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); xzero++; } } if (!(flags & PFR_FLAG_DUMMY)) pfr_clstats_ktables(&workq, tzero, flags & PFR_FLAG_ADDRSTOO); if (nzero != NULL) *nzero = xzero; return (0); } int pfr_set_tflags(struct pfr_table *tbl, int size, int setflag, int clrflag, int *nchange, int *ndel, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p, *q, key; int i, xchange = 0, xdel = 0; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); if ((setflag & ~PFR_TFLAG_USRMASK) || (clrflag & ~PFR_TFLAG_USRMASK) || (setflag & clrflag)) return (EINVAL); SLIST_INIT(&workq); for (i = 0; i < size; i++) { bcopy(tbl + i, &key.pfrkt_t, sizeof(key.pfrkt_t)); if (pfr_validate_table(&key.pfrkt_t, 0, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); p = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (p != NULL && (p->pfrkt_flags & PFR_TFLAG_ACTIVE)) { p->pfrkt_nflags = (p->pfrkt_flags | setflag) & ~clrflag; if (p->pfrkt_nflags == p->pfrkt_flags) goto _skip; SLIST_FOREACH(q, &workq, pfrkt_workq) if (!pfr_ktable_compare(p, q)) goto _skip; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); if ((p->pfrkt_flags & PFR_TFLAG_PERSIST) && (clrflag & PFR_TFLAG_PERSIST) && !(p->pfrkt_flags & PFR_TFLAG_REFERENCED)) xdel++; else xchange++; } _skip: ; } if (!(flags & PFR_FLAG_DUMMY)) pfr_setflags_ktables(&workq); if (nchange != NULL) *nchange = xchange; if (ndel != NULL) *ndel = xdel; return (0); } int pfr_ina_begin(struct pfr_table *trs, u_int32_t *ticket, int *ndel, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p; struct pf_ruleset *rs; int xdel = 0; ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); rs = pf_find_or_create_ruleset(trs->pfrt_anchor); if (rs == NULL) return (ENOMEM); SLIST_INIT(&workq); RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (!(p->pfrkt_flags & PFR_TFLAG_INACTIVE) || pfr_skip_table(trs, p, 0)) continue; p->pfrkt_nflags = p->pfrkt_flags & ~PFR_TFLAG_INACTIVE; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); xdel++; } if (!(flags & PFR_FLAG_DUMMY)) { pfr_setflags_ktables(&workq); if (ticket != NULL) *ticket = ++rs->tticket; rs->topen = 1; } else pf_remove_if_empty_ruleset(rs); if (ndel != NULL) *ndel = xdel; return (0); } int pfr_ina_define(struct pfr_table *tbl, struct pfr_addr *addr, int size, int *nadd, int *naddr, u_int32_t ticket, int flags) { struct pfr_ktableworkq tableq; struct pfr_kentryworkq addrq; struct pfr_ktable *kt, *rt, *shadow, key; struct pfr_kentry *p; struct pfr_addr *ad; struct pf_ruleset *rs; int i, rv, xadd = 0, xaddr = 0; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY | PFR_FLAG_ADDRSTOO); if (size && !(flags & PFR_FLAG_ADDRSTOO)) return (EINVAL); if (pfr_validate_table(tbl, PFR_TFLAG_USRMASK, flags & PFR_FLAG_USERIOCTL)) return (EINVAL); rs = pf_find_ruleset(tbl->pfrt_anchor); if (rs == NULL || !rs->topen || ticket != rs->tticket) return (EBUSY); tbl->pfrt_flags |= PFR_TFLAG_INACTIVE; SLIST_INIT(&tableq); kt = RB_FIND(pfr_ktablehead, &V_pfr_ktables, (struct pfr_ktable *)tbl); if (kt == NULL) { kt = pfr_create_ktable(tbl, 0, 1); if (kt == NULL) return (ENOMEM); SLIST_INSERT_HEAD(&tableq, kt, pfrkt_workq); xadd++; if (!tbl->pfrt_anchor[0]) goto _skip; /* find or create root table */ bzero(&key, sizeof(key)); strlcpy(key.pfrkt_name, tbl->pfrt_name, sizeof(key.pfrkt_name)); rt = RB_FIND(pfr_ktablehead, &V_pfr_ktables, &key); if (rt != NULL) { kt->pfrkt_root = rt; goto _skip; } rt = pfr_create_ktable(&key.pfrkt_t, 0, 1); if (rt == NULL) { pfr_destroy_ktables(&tableq, 0); return (ENOMEM); } SLIST_INSERT_HEAD(&tableq, rt, pfrkt_workq); kt->pfrkt_root = rt; } else if (!(kt->pfrkt_flags & PFR_TFLAG_INACTIVE)) xadd++; _skip: shadow = pfr_create_ktable(tbl, 0, 0); if (shadow == NULL) { pfr_destroy_ktables(&tableq, 0); return (ENOMEM); } SLIST_INIT(&addrq); for (i = 0, ad = addr; i < size; i++, ad++) { if (pfr_validate_addr(ad)) senderr(EINVAL); if (pfr_lookup_addr(shadow, ad, 1) != NULL) continue; p = pfr_create_kentry(ad); if (p == NULL) senderr(ENOMEM); if (pfr_route_kentry(shadow, p)) { pfr_destroy_kentry(p); continue; } SLIST_INSERT_HEAD(&addrq, p, pfrke_workq); xaddr++; } if (!(flags & PFR_FLAG_DUMMY)) { if (kt->pfrkt_shadow != NULL) pfr_destroy_ktable(kt->pfrkt_shadow, 1); kt->pfrkt_flags |= PFR_TFLAG_INACTIVE; pfr_insert_ktables(&tableq); shadow->pfrkt_cnt = (flags & PFR_FLAG_ADDRSTOO) ? xaddr : NO_ADDRESSES; kt->pfrkt_shadow = shadow; } else { pfr_clean_node_mask(shadow, &addrq); pfr_destroy_ktable(shadow, 0); pfr_destroy_ktables(&tableq, 0); pfr_destroy_kentries(&addrq); } if (nadd != NULL) *nadd = xadd; if (naddr != NULL) *naddr = xaddr; return (0); _bad: pfr_destroy_ktable(shadow, 0); pfr_destroy_ktables(&tableq, 0); pfr_destroy_kentries(&addrq); return (rv); } int pfr_ina_rollback(struct pfr_table *trs, u_int32_t ticket, int *ndel, int flags) { struct pfr_ktableworkq workq; struct pfr_ktable *p; struct pf_ruleset *rs; int xdel = 0; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); rs = pf_find_ruleset(trs->pfrt_anchor); if (rs == NULL || !rs->topen || ticket != rs->tticket) return (0); SLIST_INIT(&workq); RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (!(p->pfrkt_flags & PFR_TFLAG_INACTIVE) || pfr_skip_table(trs, p, 0)) continue; p->pfrkt_nflags = p->pfrkt_flags & ~PFR_TFLAG_INACTIVE; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); xdel++; } if (!(flags & PFR_FLAG_DUMMY)) { pfr_setflags_ktables(&workq); rs->topen = 0; pf_remove_if_empty_ruleset(rs); } if (ndel != NULL) *ndel = xdel; return (0); } int pfr_ina_commit(struct pfr_table *trs, u_int32_t ticket, int *nadd, int *nchange, int flags) { struct pfr_ktable *p, *q; struct pfr_ktableworkq workq; struct pf_ruleset *rs; int xadd = 0, xchange = 0; long tzero = time_second; PF_RULES_WASSERT(); ACCEPT_FLAGS(flags, PFR_FLAG_DUMMY); rs = pf_find_ruleset(trs->pfrt_anchor); if (rs == NULL || !rs->topen || ticket != rs->tticket) return (EBUSY); SLIST_INIT(&workq); RB_FOREACH(p, pfr_ktablehead, &V_pfr_ktables) { if (!(p->pfrkt_flags & PFR_TFLAG_INACTIVE) || pfr_skip_table(trs, p, 0)) continue; SLIST_INSERT_HEAD(&workq, p, pfrkt_workq); if (p->pfrkt_flags & PFR_TFLAG_ACTIVE) xchange++; else xadd++; } if (!(flags & PFR_FLAG_DUMMY)) { for (p = SLIST_FIRST(&workq); p != NULL; p = q) { q = SLIST_NEXT(p, pfrkt_workq); pfr_commit_ktable(p, tzero); } rs->topen = 0; pf_remove_if_empty_ruleset(rs); } if (nadd != NULL) *nadd = xadd; if (nchange != NULL) *nchange = xchange; return (0); } static void pfr_commit_ktable(struct pfr_ktable *kt, long tzero) { struct pfr_ktable *shadow = kt->pfrkt_shadow; int nflags; PF_RULES_WASSERT(); if (shadow->pfrkt_cnt == NO_ADDRESSES) { if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) pfr_clstats_ktable(kt, tzero, 1); } else if (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) { /* kt might contain addresses */ struct pfr_kentryworkq addrq, addq, changeq, delq, garbageq; struct pfr_kentry *p, *q, *next; struct pfr_addr ad; pfr_enqueue_addrs(shadow, &addrq, NULL, 0); pfr_mark_addrs(kt); SLIST_INIT(&addq); SLIST_INIT(&changeq); SLIST_INIT(&delq); SLIST_INIT(&garbageq); pfr_clean_node_mask(shadow, &addrq); for (p = SLIST_FIRST(&addrq); p != NULL; p = next) { next = SLIST_NEXT(p, pfrke_workq); /* XXX */ pfr_copyout_addr(&ad, p); q = pfr_lookup_addr(kt, &ad, 1); if (q != NULL) { if (q->pfrke_not != p->pfrke_not) SLIST_INSERT_HEAD(&changeq, q, pfrke_workq); q->pfrke_mark = 1; SLIST_INSERT_HEAD(&garbageq, p, pfrke_workq); } else { p->pfrke_tzero = tzero; SLIST_INSERT_HEAD(&addq, p, pfrke_workq); } } pfr_enqueue_addrs(kt, &delq, NULL, ENQUEUE_UNMARKED_ONLY); pfr_insert_kentries(kt, &addq, tzero); pfr_remove_kentries(kt, &delq); pfr_clstats_kentries(&changeq, tzero, INVERT_NEG_FLAG); pfr_destroy_kentries(&garbageq); } else { /* kt cannot contain addresses */ SWAP(struct radix_node_head *, kt->pfrkt_ip4, shadow->pfrkt_ip4); SWAP(struct radix_node_head *, kt->pfrkt_ip6, shadow->pfrkt_ip6); SWAP(int, kt->pfrkt_cnt, shadow->pfrkt_cnt); pfr_clstats_ktable(kt, tzero, 1); } nflags = ((shadow->pfrkt_flags & PFR_TFLAG_USRMASK) | (kt->pfrkt_flags & PFR_TFLAG_SETMASK) | PFR_TFLAG_ACTIVE) & ~PFR_TFLAG_INACTIVE; pfr_destroy_ktable(shadow, 0); kt->pfrkt_shadow = NULL; pfr_setflags_ktable(kt, nflags); } static int pfr_validate_table(struct pfr_table *tbl, int allowedflags, int no_reserved) { int i; if (!tbl->pfrt_name[0]) return (-1); if (no_reserved && !strcmp(tbl->pfrt_anchor, PF_RESERVED_ANCHOR)) return (-1); if (tbl->pfrt_name[PF_TABLE_NAME_SIZE-1]) return (-1); for (i = strlen(tbl->pfrt_name); i < PF_TABLE_NAME_SIZE; i++) if (tbl->pfrt_name[i]) return (-1); if (pfr_fix_anchor(tbl->pfrt_anchor)) return (-1); if (tbl->pfrt_flags & ~allowedflags) return (-1); return (0); } /* * Rewrite anchors referenced by tables to remove slashes * and check for validity. */ static int pfr_fix_anchor(char *anchor) { size_t siz = MAXPATHLEN; int i; if (anchor[0] == '/') { char *path; int off; path = anchor; off = 1; while (*++path == '/') off++; bcopy(path, anchor, siz - off); memset(anchor + siz - off, 0, off); } if (anchor[siz - 1]) return (-1); for (i = strlen(anchor); i < siz; i++) if (anchor[i]) return (-1); return (0); } int pfr_table_count(struct pfr_table *filter, int flags) { struct pf_ruleset *rs; PF_RULES_ASSERT(); if (flags & PFR_FLAG_ALLRSETS) return (V_pfr_ktable_cnt); if (filter->pfrt_anchor[0]) { rs = pf_find_ruleset(filter->pfrt_anchor); return ((rs != NULL) ? rs->tables : -1); } return (pf_main_ruleset.tables); } static int pfr_skip_table(struct pfr_table *filter, struct pfr_ktable *kt, int flags) { if (flags & PFR_FLAG_ALLRSETS) return (0); if (strcmp(filter->pfrt_anchor, kt->pfrkt_anchor)) return (1); return (0); } static void pfr_insert_ktables(struct pfr_ktableworkq *workq) { struct pfr_ktable *p; SLIST_FOREACH(p, workq, pfrkt_workq) pfr_insert_ktable(p); } static void pfr_insert_ktable(struct pfr_ktable *kt) { PF_RULES_WASSERT(); RB_INSERT(pfr_ktablehead, &V_pfr_ktables, kt); V_pfr_ktable_cnt++; if (kt->pfrkt_root != NULL) if (!kt->pfrkt_root->pfrkt_refcnt[PFR_REFCNT_ANCHOR]++) pfr_setflags_ktable(kt->pfrkt_root, kt->pfrkt_root->pfrkt_flags|PFR_TFLAG_REFDANCHOR); } static void pfr_setflags_ktables(struct pfr_ktableworkq *workq) { struct pfr_ktable *p, *q; for (p = SLIST_FIRST(workq); p; p = q) { q = SLIST_NEXT(p, pfrkt_workq); pfr_setflags_ktable(p, p->pfrkt_nflags); } } static void pfr_setflags_ktable(struct pfr_ktable *kt, int newf) { struct pfr_kentryworkq addrq; PF_RULES_WASSERT(); if (!(newf & PFR_TFLAG_REFERENCED) && !(newf & PFR_TFLAG_PERSIST)) newf &= ~PFR_TFLAG_ACTIVE; if (!(newf & PFR_TFLAG_ACTIVE)) newf &= ~PFR_TFLAG_USRMASK; if (!(newf & PFR_TFLAG_SETMASK)) { RB_REMOVE(pfr_ktablehead, &V_pfr_ktables, kt); if (kt->pfrkt_root != NULL) if (!--kt->pfrkt_root->pfrkt_refcnt[PFR_REFCNT_ANCHOR]) pfr_setflags_ktable(kt->pfrkt_root, kt->pfrkt_root->pfrkt_flags & ~PFR_TFLAG_REFDANCHOR); pfr_destroy_ktable(kt, 1); V_pfr_ktable_cnt--; return; } if (!(newf & PFR_TFLAG_ACTIVE) && kt->pfrkt_cnt) { pfr_enqueue_addrs(kt, &addrq, NULL, 0); pfr_remove_kentries(kt, &addrq); } if (!(newf & PFR_TFLAG_INACTIVE) && kt->pfrkt_shadow != NULL) { pfr_destroy_ktable(kt->pfrkt_shadow, 1); kt->pfrkt_shadow = NULL; } kt->pfrkt_flags = newf; } static void pfr_clstats_ktables(struct pfr_ktableworkq *workq, long tzero, int recurse) { struct pfr_ktable *p; SLIST_FOREACH(p, workq, pfrkt_workq) pfr_clstats_ktable(p, tzero, recurse); } static void pfr_clstats_ktable(struct pfr_ktable *kt, long tzero, int recurse) { struct pfr_kentryworkq addrq; if (recurse) { pfr_enqueue_addrs(kt, &addrq, NULL, 0); pfr_clstats_kentries(&addrq, tzero, 0); } bzero(kt->pfrkt_packets, sizeof(kt->pfrkt_packets)); bzero(kt->pfrkt_bytes, sizeof(kt->pfrkt_bytes)); kt->pfrkt_match = kt->pfrkt_nomatch = 0; kt->pfrkt_tzero = tzero; } static struct pfr_ktable * pfr_create_ktable(struct pfr_table *tbl, long tzero, int attachruleset) { struct pfr_ktable *kt; struct pf_ruleset *rs; PF_RULES_WASSERT(); kt = malloc(sizeof(*kt), M_PFTABLE, M_NOWAIT|M_ZERO); if (kt == NULL) return (NULL); kt->pfrkt_t = *tbl; if (attachruleset) { rs = pf_find_or_create_ruleset(tbl->pfrt_anchor); if (!rs) { pfr_destroy_ktable(kt, 0); return (NULL); } kt->pfrkt_rs = rs; rs->tables++; } if (!rn_inithead((void **)&kt->pfrkt_ip4, offsetof(struct sockaddr_in, sin_addr) * 8) || !rn_inithead((void **)&kt->pfrkt_ip6, offsetof(struct sockaddr_in6, sin6_addr) * 8)) { pfr_destroy_ktable(kt, 0); return (NULL); } kt->pfrkt_tzero = tzero; return (kt); } static void pfr_destroy_ktables(struct pfr_ktableworkq *workq, int flushaddr) { struct pfr_ktable *p, *q; for (p = SLIST_FIRST(workq); p; p = q) { q = SLIST_NEXT(p, pfrkt_workq); pfr_destroy_ktable(p, flushaddr); } } static void pfr_destroy_ktable(struct pfr_ktable *kt, int flushaddr) { struct pfr_kentryworkq addrq; if (flushaddr) { pfr_enqueue_addrs(kt, &addrq, NULL, 0); pfr_clean_node_mask(kt, &addrq); pfr_destroy_kentries(&addrq); } if (kt->pfrkt_ip4 != NULL) rn_detachhead((void **)&kt->pfrkt_ip4); if (kt->pfrkt_ip6 != NULL) rn_detachhead((void **)&kt->pfrkt_ip6); if (kt->pfrkt_shadow != NULL) pfr_destroy_ktable(kt->pfrkt_shadow, flushaddr); if (kt->pfrkt_rs != NULL) { kt->pfrkt_rs->tables--; pf_remove_if_empty_ruleset(kt->pfrkt_rs); } free(kt, M_PFTABLE); } static int pfr_ktable_compare(struct pfr_ktable *p, struct pfr_ktable *q) { int d; if ((d = strncmp(p->pfrkt_name, q->pfrkt_name, PF_TABLE_NAME_SIZE))) return (d); return (strcmp(p->pfrkt_anchor, q->pfrkt_anchor)); } static struct pfr_ktable * pfr_lookup_table(struct pfr_table *tbl) { /* struct pfr_ktable start like a struct pfr_table */ return (RB_FIND(pfr_ktablehead, &V_pfr_ktables, (struct pfr_ktable *)tbl)); } int pfr_match_addr(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af) { struct pfr_kentry *ke = NULL; int match; PF_RULES_RASSERT(); if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) kt = kt->pfrkt_root; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (0); switch (af) { #ifdef INET case AF_INET: { struct sockaddr_in sin; bzero(&sin, sizeof(sin)); sin.sin_len = sizeof(sin); sin.sin_family = AF_INET; sin.sin_addr.s_addr = a->addr32[0]; ke = (struct pfr_kentry *)rn_match(&sin, &kt->pfrkt_ip4->rh); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { struct sockaddr_in6 sin6; bzero(&sin6, sizeof(sin6)); sin6.sin6_len = sizeof(sin6); sin6.sin6_family = AF_INET6; bcopy(a, &sin6.sin6_addr, sizeof(sin6.sin6_addr)); ke = (struct pfr_kentry *)rn_match(&sin6, &kt->pfrkt_ip6->rh); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; break; } #endif /* INET6 */ } match = (ke && !ke->pfrke_not); if (match) kt->pfrkt_match++; else kt->pfrkt_nomatch++; return (match); } void pfr_update_stats(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af, u_int64_t len, int dir_out, int op_pass, int notrule) { struct pfr_kentry *ke = NULL; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) kt = kt->pfrkt_root; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return; switch (af) { #ifdef INET case AF_INET: { struct sockaddr_in sin; bzero(&sin, sizeof(sin)); sin.sin_len = sizeof(sin); sin.sin_family = AF_INET; sin.sin_addr.s_addr = a->addr32[0]; ke = (struct pfr_kentry *)rn_match(&sin, &kt->pfrkt_ip4->rh); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { struct sockaddr_in6 sin6; bzero(&sin6, sizeof(sin6)); sin6.sin6_len = sizeof(sin6); sin6.sin6_family = AF_INET6; bcopy(a, &sin6.sin6_addr, sizeof(sin6.sin6_addr)); ke = (struct pfr_kentry *)rn_match(&sin6, &kt->pfrkt_ip6->rh); if (ke && KENTRY_RNF_ROOT(ke)) ke = NULL; break; } #endif /* INET6 */ default: panic("%s: unknown address family %u", __func__, af); } if ((ke == NULL || ke->pfrke_not) != notrule) { if (op_pass != PFR_OP_PASS) printf("pfr_update_stats: assertion failed.\n"); op_pass = PFR_OP_XPASS; } kt->pfrkt_packets[dir_out][op_pass]++; kt->pfrkt_bytes[dir_out][op_pass] += len; if (ke != NULL && op_pass != PFR_OP_XPASS && (kt->pfrkt_flags & PFR_TFLAG_COUNTERS)) { if (ke->pfrke_counters == NULL) ke->pfrke_counters = uma_zalloc(V_pfr_kcounters_z, M_NOWAIT | M_ZERO); if (ke->pfrke_counters != NULL) { ke->pfrke_counters->pfrkc_packets[dir_out][op_pass]++; ke->pfrke_counters->pfrkc_bytes[dir_out][op_pass] += len; } } } struct pfr_ktable * pfr_attach_table(struct pf_ruleset *rs, char *name) { struct pfr_ktable *kt, *rt; struct pfr_table tbl; struct pf_anchor *ac = rs->anchor; PF_RULES_WASSERT(); bzero(&tbl, sizeof(tbl)); strlcpy(tbl.pfrt_name, name, sizeof(tbl.pfrt_name)); if (ac != NULL) strlcpy(tbl.pfrt_anchor, ac->path, sizeof(tbl.pfrt_anchor)); kt = pfr_lookup_table(&tbl); if (kt == NULL) { kt = pfr_create_ktable(&tbl, time_second, 1); if (kt == NULL) return (NULL); if (ac != NULL) { bzero(tbl.pfrt_anchor, sizeof(tbl.pfrt_anchor)); rt = pfr_lookup_table(&tbl); if (rt == NULL) { rt = pfr_create_ktable(&tbl, 0, 1); if (rt == NULL) { pfr_destroy_ktable(kt, 0); return (NULL); } pfr_insert_ktable(rt); } kt->pfrkt_root = rt; } pfr_insert_ktable(kt); } if (!kt->pfrkt_refcnt[PFR_REFCNT_RULE]++) pfr_setflags_ktable(kt, kt->pfrkt_flags|PFR_TFLAG_REFERENCED); return (kt); } void pfr_detach_table(struct pfr_ktable *kt) { PF_RULES_WASSERT(); KASSERT(kt->pfrkt_refcnt[PFR_REFCNT_RULE] > 0, ("%s: refcount %d\n", __func__, kt->pfrkt_refcnt[PFR_REFCNT_RULE])); if (!--kt->pfrkt_refcnt[PFR_REFCNT_RULE]) pfr_setflags_ktable(kt, kt->pfrkt_flags&~PFR_TFLAG_REFERENCED); } int pfr_pool_get(struct pfr_ktable *kt, int *pidx, struct pf_addr *counter, sa_family_t af) { struct pf_addr *addr, *cur, *mask; union sockaddr_union uaddr, umask; struct pfr_kentry *ke, *ke2 = NULL; int idx = -1, use_counter = 0; switch (af) { case AF_INET: uaddr.sin.sin_len = sizeof(struct sockaddr_in); uaddr.sin.sin_family = AF_INET; break; case AF_INET6: uaddr.sin6.sin6_len = sizeof(struct sockaddr_in6); uaddr.sin6.sin6_family = AF_INET6; break; } addr = SUNION2PF(&uaddr, af); if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) kt = kt->pfrkt_root; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) return (-1); if (pidx != NULL) idx = *pidx; if (counter != NULL && idx >= 0) use_counter = 1; if (idx < 0) idx = 0; _next_block: ke = pfr_kentry_byidx(kt, idx, af); if (ke == NULL) { kt->pfrkt_nomatch++; return (1); } pfr_prepare_network(&umask, af, ke->pfrke_net); cur = SUNION2PF(&ke->pfrke_sa, af); mask = SUNION2PF(&umask, af); if (use_counter) { /* is supplied address within block? */ if (!PF_MATCHA(0, cur, mask, counter, af)) { /* no, go to next block in table */ idx++; use_counter = 0; goto _next_block; } PF_ACPY(addr, counter, af); } else { /* use first address of block */ PF_ACPY(addr, cur, af); } if (!KENTRY_NETWORK(ke)) { /* this is a single IP address - no possible nested block */ PF_ACPY(counter, addr, af); *pidx = idx; kt->pfrkt_match++; return (0); } for (;;) { /* we don't want to use a nested block */ switch (af) { case AF_INET: ke2 = (struct pfr_kentry *)rn_match(&uaddr, &kt->pfrkt_ip4->rh); break; case AF_INET6: ke2 = (struct pfr_kentry *)rn_match(&uaddr, &kt->pfrkt_ip6->rh); break; } /* no need to check KENTRY_RNF_ROOT() here */ if (ke2 == ke) { /* lookup return the same block - perfect */ PF_ACPY(counter, addr, af); *pidx = idx; kt->pfrkt_match++; return (0); } /* we need to increase the counter past the nested block */ pfr_prepare_network(&umask, AF_INET, ke2->pfrke_net); PF_POOLMASK(addr, addr, SUNION2PF(&umask, af), &pfr_ffaddr, af); PF_AINC(addr, af); if (!PF_MATCHA(0, cur, mask, addr, af)) { /* ok, we reached the end of our main block */ /* go to next block in table */ idx++; use_counter = 0; goto _next_block; } } } static struct pfr_kentry * pfr_kentry_byidx(struct pfr_ktable *kt, int idx, int af) { struct pfr_walktree w; bzero(&w, sizeof(w)); w.pfrw_op = PFRW_POOL_GET; w.pfrw_cnt = idx; switch (af) { #ifdef INET case AF_INET: kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w); return (w.pfrw_kentry); #endif /* INET */ #ifdef INET6 case AF_INET6: kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w); return (w.pfrw_kentry); #endif /* INET6 */ default: return (NULL); } } void pfr_dynaddr_update(struct pfr_ktable *kt, struct pfi_dynaddr *dyn) { struct pfr_walktree w; bzero(&w, sizeof(w)); w.pfrw_op = PFRW_DYNADDR_UPDATE; w.pfrw_dyn = dyn; dyn->pfid_acnt4 = 0; dyn->pfid_acnt6 = 0; if (!dyn->pfid_af || dyn->pfid_af == AF_INET) kt->pfrkt_ip4->rnh_walktree(&kt->pfrkt_ip4->rh, pfr_walktree, &w); if (!dyn->pfid_af || dyn->pfid_af == AF_INET6) kt->pfrkt_ip6->rnh_walktree(&kt->pfrkt_ip6->rh, pfr_walktree, &w); }