Index: head/sys/netinet/in_pcb.c =================================================================== --- head/sys/netinet/in_pcb.c (revision 367147) +++ head/sys/netinet/in_pcb.c (revision 367148) @@ -1,3536 +1,3526 @@ /*- * 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_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 #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #if defined(INET) || defined(INET6) #include #include #ifdef INET #include #include #endif #include #include #ifdef TCPHPTS #include #endif #include #include #ifdef INET6 #include #include #include #include #endif /* INET6 */ #include #endif #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); 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 | CTLFLAG_MPSAFE, 0, "IP Ports"); SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &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 | CTLFLAG_NEEDGIANT, &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 | CTLFLAG_NEEDGIANT, &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 | CTLFLAG_NEEDGIANT, &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 | CTLFLAG_NEEDGIANT, &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 | CTLFLAG_NEEDGIANT, &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"); #ifdef RATELIMIT counter_u64_t rate_limit_active; counter_u64_t rate_limit_alloc_fail; counter_u64_t rate_limit_set_ok; static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "IP Rate Limiting"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD, &rate_limit_active, "Active rate limited connections"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD, &rate_limit_alloc_fail, "Rate limited connection failures"); SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, &rate_limit_set_ok, "Rate limited setting succeeded"); #endif /* RATELIMIT */ #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; CK_LIST_INSERT_HEAD(hdr, grp, il_list); return (grp); } static void in_pcblbgroup_free_deferred(epoch_context_t ctx) { struct inpcblbgroup *grp; grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx); free(grp, M_PCB); } static void in_pcblbgroup_free(struct inpcblbgroup *grp) { CK_LIST_REMOVE(grp, il_list); NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx); } 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 == NULL) 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, *new_grp; 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. */ new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2); if (new_grp != NULL) *grpp = new_grp; } } /* * Add PCB to load balance group for SO_REUSEPORT_LB option. */ static int in_pcbinslbgrouphash(struct inpcb *inp) { const static struct timeval interval = { 60, 0 }; static struct timeval lastprint; struct inpcbinfo *pcbinfo; struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; uint32_t idx; pcbinfo = inp->inp_pcbinfo; INP_WLOCK_ASSERT(inp); INP_HASH_WLOCK_ASSERT(pcbinfo); /* * Don't allow jailed socket to join local group. */ if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_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 idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask); hdr = &pcbinfo->ipi_lbgrouphashbase[idx]; CK_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 == NULL) return (ENOBUFS); } else if (grp->il_inpcnt == grp->il_inpsiz) { if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) { if (ratecheck(&lastprint, &interval)) 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 == NULL) 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); hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)]; CK_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) { porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1); 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(porthash_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; error = 0; inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT); if (inp == NULL) return (ENOBUFS); bzero(&inp->inp_start_zero, inp_zero_size); #ifdef NUMA inp->inp_numa_domain = M_NODOM; #endif 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 #if defined(INET) || defined(INET6) /* * Assign a local port like in_pcb_lport(), but also used with connect() * and a foreign address and port. If fsa is non-NULL, choose a local port * that is unused with those, otherwise one that is completely unused. * lsa can be NULL for IPv6. */ int in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, struct sockaddr *fsa, u_short fport, 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, faddr; #endif #ifdef INET6 struct in6_addr *laddr6, *faddr6; #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); 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 laddr.s_addr = INADDR_ANY; if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) { if (lsa != NULL) laddr = ((struct sockaddr_in *)lsa)->sin_addr; if (fsa != NULL) faddr = ((struct sockaddr_in *)fsa)->sin_addr; } #endif #ifdef INET6 laddr6 = NULL; if ((inp->inp_vflag & INP_IPV6) != 0) { if (lsa != NULL) laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr; if (fsa != NULL) faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr; } #endif tmpinp = NULL; 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); if (fsa != NULL) { #ifdef INET if (lsa->sa_family == AF_INET) { tmpinp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport, lookupflags, NULL); } #endif #ifdef INET6 if (lsa->sa_family == AF_INET6) { tmpinp = in6_pcblookup_hash_locked(pcbinfo, faddr6, fport, laddr6, lport, lookupflags, NULL); } #endif } else { #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); *lportp = lport; return (0); } /* * Select a local port (number) to use. */ int in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp, struct ucred *cred, int lookupflags) { struct sockaddr_in laddr; if (laddrp) { bzero(&laddr, sizeof(laddr)); laddr.sin_family = AF_INET; laddr.sin_addr = *laddrp; } return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr : NULL, lportp, NULL, 0, cred, lookupflags)); } /* * 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)) return (EACCES); if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) && priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 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, bool rehash) { 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) { KASSERT(rehash == true, ("Rehashing required for unbound inps")); 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; if (rehash) { in_pcbrehash_mbuf(inp, m); } else { in_pcbinshash_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, true)); } /* * 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, dst; struct nhop_object *nh; int error; NET_EPOCH_ASSERT(); 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; nh = NULL; bzero(&dst, sizeof(dst)); sin = &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) nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr, 0, NHR_NONE, 0); /* * 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. */ if (nh == NULL || nh->nh_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 ((nh->nh_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 *)nh->nh_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 *)nh->nh_ifa->ifa_addr; if (prison_check_ip4(cred, &sin->sin_addr) == 0) { ia = (struct in_ifaddr *)nh->nh_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 = nh->nh_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 ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) { struct in_ifaddr *ia; ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst), inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0, inp->inp_socket->so_fibnum)); if (ia == NULL) ia = ifatoia(ifa_ifwithaddr(sintosa(&dst))); 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: 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. */ NET_EPOCH_ASSERT(); 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; #ifdef ROUTE_MPATH if (CALC_FLOWID_OUTBOUND) { uint32_t hash_val, hash_type; hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport, inp->inp_socket->so_proto->pr_protocol, &hash_type); inp->inp_flowid = hash_val; inp->inp_flowtype = hash_type; } #endif 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); } if (lport != 0) { oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport, laddr, lport, 0, NULL); if (oinp != NULL) { if (oinpp != NULL) *oinpp = oinp; return (EADDRINUSE); } } else { struct sockaddr_in lsin, fsin; bzero(&lsin, sizeof(lsin)); bzero(&fsin, sizeof(fsin)); lsin.sin_family = AF_INET; lsin.sin_addr = laddr; fsin.sin_family = AF_INET; fsin.sin_addr = faddr; error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin, &lport, (struct sockaddr *)& fsin, fport, cred, INPLOOKUP_WILDCARD); 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); CURVNET_SET(inp->inp_vnet); #ifdef INET struct ip_moptions *imo = 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 struct ip6_moptions *im6o = NULL; if (inp->inp_vflag & INP_IPV6PROTO) { ip6_freepcbopts(inp->in6p_outputopts); im6o = 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); #ifdef INET6 ip6_freemoptions(im6o); #endif #ifdef INET inp_freemoptions(imo); #endif CURVNET_RESTORE(); } /* * 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; } 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); NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx); } /* * 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); NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); } 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 in_multi *inm; struct in_mfilter *imf; struct ip_moptions *imo; 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 */ restart: IP_MFILTER_FOREACH(imf, &imo->imo_head) { if ((inm = imf->imf_inm) == NULL) continue; if (inm->inm_ifp != ifp) continue; ip_mfilter_remove(&imo->imo_head, imf); IN_MULTI_LOCK_ASSERT(); in_leavegroup_locked(inm, NULL); ip_mfilter_free(imf); goto restart; } } 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; const struct inpcblbgrouphead *hdr; struct inpcblbgroup *grp; uint32_t idx; INP_HASH_LOCK_ASSERT(pcbinfo); hdr = &pcbinfo->ipi_lbgrouphashbase[ INP_PCBPORTHASH(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 */ local_wild = NULL; CK_LIST_FOREACH(grp, hdr, il_list) { #ifdef INET6 if (!(grp->il_vflag & INP_IPV4)) continue; #endif if (grp->il_lport != lport) continue; idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) % grp->il_inpcnt; if (grp->il_laddr.s_addr == laddr->s_addr) return (grp->il_inp[idx]); if (grp->il_laddr.s_addr == INADDR_ANY && (lookupflags & INPLOOKUP_WILDCARD) != 0) local_wild = grp->il_inp[idx]; } return (local_wild); } #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; KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0, ("%s: invalid lookup flags %d", __func__, lookupflags)); INP_HASH_LOCK_ASSERT(pcbinfo); /* * 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 ((lookupflags & INPLOOKUP_WILDCARD) != 0) { 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 = 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 } 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, struct mbuf *m) { 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 (m != NULL) { in_pcbgroup_update_mbuf(inp, m); } else { in_pcbgroup_update(inp); } #endif return (0); } int in_pcbinshash(struct inpcb *inp) { return (in_pcbinshash_internal(inp, NULL)); } int in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m) { return (in_pcbinshash_internal(inp, m)); } /* * 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; 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); NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx); } 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) { bzero(xi, sizeof(*xi)); xi->xi_len = sizeof(struct xinpcb); if (inp->inp_socket) sotoxsocket(inp->inp_socket, &xi->xi_socket); 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, .rate_limit.flags = M_NOWAIT, }; 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, struct m_snd_tag **st) { 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.hdr.numa_domain = inp->inp_numa_domain, .rate_limit.max_rate = max_pacing_rate, .rate_limit.flags = M_NOWAIT, }; int error; INP_WLOCK_ASSERT(inp); if (*st != NULL) return (EINVAL); if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_alloc(ifp, ¶ms, st); #ifdef INET if (error == 0) { counter_u64_add(rate_limit_set_ok, 1); counter_u64_add(rate_limit_active, 1); } else counter_u64_add(rate_limit_alloc_fail, 1); #endif } return (error); } void -in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst) +in_pcbdetach_tag(struct m_snd_tag *mst) { - 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); + m_snd_tag_rele(mst); #ifdef INET counter_u64_add(rate_limit_active, -1); #endif } /* * 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; INP_WLOCK_ASSERT(inp); mst = inp->inp_snd_tag; inp->inp_snd_tag = NULL; if (mst == NULL) return; m_snd_tag_rele(mst); } int in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate) { int error; /* * If the existing send tag is for the wrong interface due to * a route change, first drop the existing tag. Set the * CHANGED flag so that we will keep trying to allocate a new * tag if we fail to allocate one this time. */ if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) { in_pcbdetach_txrtlmt(inp); inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED; } /* * 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, &inp->inp_snd_tag); } } else { error = in_pcbmodify_txrtlmt(inp, max_pacing_rate); } if (error == 0 || error == EOPNOTSUPP) inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED; return (error); } /* * 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; error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate); if (did_upgrade) INP_DOWNGRADE(inp); } /* * Track route changes for TX rate limiting. */ void in_pcboutput_eagain(struct inpcb *inp) { bool did_upgrade; if (inp == 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); } #ifdef INET static void rl_init(void *st) { rate_limit_active = counter_u64_alloc(M_WAITOK); rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK); rate_limit_set_ok = counter_u64_alloc(M_WAITOK); } SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL); #endif #endif /* RATELIMIT */ Index: head/sys/netinet/in_pcb.h =================================================================== --- head/sys/netinet/in_pcb.h (revision 367147) +++ head/sys/netinet/in_pcb.h (revision 367148) @@ -1,896 +1,896 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1990, 1993 * The Regents of the University of California. * 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.h 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #ifndef _NETINET_IN_PCB_H_ #define _NETINET_IN_PCB_H_ #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #include #endif #include /* * struct inpcb is the common protocol control block structure used in most * IP transport protocols. * * Pointers to local and foreign host table entries, local and foreign socket * numbers, and pointers up (to a socket structure) and down (to a * protocol-specific control block) are stored here. */ CK_LIST_HEAD(inpcbhead, inpcb); CK_LIST_HEAD(inpcbporthead, inpcbport); CK_LIST_HEAD(inpcblbgrouphead, inpcblbgroup); typedef uint64_t inp_gen_t; /* * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet. * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing * the following structure. */ struct in_addr_4in6 { u_int32_t ia46_pad32[3]; struct in_addr ia46_addr4; }; union in_dependaddr { struct in_addr_4in6 id46_addr; struct in6_addr id6_addr; }; /* * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has * some extra padding to accomplish this. * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport, * lport, faddr to generate hash, so these fields shouldn't be moved. */ struct in_endpoints { u_int16_t ie_fport; /* foreign port */ u_int16_t ie_lport; /* local port */ /* protocol dependent part, local and foreign addr */ union in_dependaddr ie_dependfaddr; /* foreign host table entry */ union in_dependaddr ie_dependladdr; /* local host table entry */ #define ie_faddr ie_dependfaddr.id46_addr.ia46_addr4 #define ie_laddr ie_dependladdr.id46_addr.ia46_addr4 #define ie6_faddr ie_dependfaddr.id6_addr #define ie6_laddr ie_dependladdr.id6_addr u_int32_t ie6_zoneid; /* scope zone id */ }; /* * XXX The defines for inc_* are hacks and should be changed to direct * references. */ struct in_conninfo { u_int8_t inc_flags; u_int8_t inc_len; u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */ /* protocol dependent part */ struct in_endpoints inc_ie; }; /* * Flags for inc_flags. */ #define INC_ISIPV6 0x01 #define INC_IPV6MINMTU 0x02 #define inc_fport inc_ie.ie_fport #define inc_lport inc_ie.ie_lport #define inc_faddr inc_ie.ie_faddr #define inc_laddr inc_ie.ie_laddr #define inc6_faddr inc_ie.ie6_faddr #define inc6_laddr inc_ie.ie6_laddr #define inc6_zoneid inc_ie.ie6_zoneid #if defined(_KERNEL) || defined(_WANT_INPCB) /* * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and * IPv6 sockets. In the case of TCP and UDP, further per-connection state is * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb * are static after creation or protected by a per-inpcb rwlock, inp_lock. A * few fields are protected by multiple locks as indicated in the locking notes * below. For these fields, all of the listed locks must be write-locked for * any modifications. However, these fields can be safely read while any one of * the listed locks are read-locked. This model can permit greater concurrency * for read operations. For example, connections can be looked up while only * holding a read lock on the global pcblist lock. This is important for * performance when attempting to find the connection for a packet given its IP * and port tuple. * * One noteworthy exception is that the global pcbinfo lock follows a different * set of rules in relation to the inp_list field. Rather than being * write-locked for modifications and read-locked for list iterations, it must * be read-locked during modifications and write-locked during list iterations. * This ensures that the relatively rare global list iterations safely walk a * stable snapshot of connections while allowing more common list modifications * to safely grab the pcblist lock just while adding or removing a connection * from the global list. * * Key: * (b) - Protected by the hpts lock. * (c) - Constant after initialization * (e) - Protected by the net_epoch_prempt epoch * (g) - Protected by the pcbgroup lock * (i) - Protected by the inpcb lock * (p) - Protected by the pcbinfo lock for the inpcb * (l) - Protected by the pcblist lock for the inpcb * (h) - Protected by the pcbhash lock for the inpcb * (s) - Protected by another subsystem's locks * (x) - Undefined locking * * Notes on the tcp_hpts: * * First Hpts lock order is * 1) INP_WLOCK() * 2) HPTS_LOCK() i.e. hpts->pmtx * * To insert a TCB on the hpts you *must* be holding the INP_WLOCK(). * You may check the inp->inp_in_hpts flag without the hpts lock. * The hpts is the only one that will clear this flag holding * only the hpts lock. This means that in your tcp_output() * routine when you test for the inp_in_hpts flag to be 1 * it may be transitioning to 0 (by the hpts). * That's ok since that will just mean an extra call to tcp_output * that most likely will find the call you executed * (when the mis-match occured) will have put the TCB back * on the hpts and it will return. If your * call did not add the inp back to the hpts then you will either * over-send or the cwnd will block you from sending more. * * Note you should also be holding the INP_WLOCK() when you * call the remove from the hpts as well. Though usually * you are either doing this from a timer, where you need and have * the INP_WLOCK() or from destroying your TCB where again * you should already have the INP_WLOCK(). * * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and * inp_input_cpu_set fields are controlled completely by * the hpts. Do not ever set these. The inp_hpts_cpu_set * and inp_input_cpu_set fields indicate if the hpts has * setup the respective cpu field. It is advised if this * field is 0, to enqueue the packet with the appropriate * hpts_immediate() call. If the _set field is 1, then * you may compare the inp_*_cpu field to the curcpu and * may want to again insert onto the hpts if these fields * are not equal (i.e. you are not on the expected CPU). * * A note on inp_hpts_calls and inp_input_calls, these * flags are set when the hpts calls either the output * or do_segment routines respectively. If the routine * being called wants to use this, then it needs to * clear the flag before returning. The hpts will not * clear the flag. The flags can be used to tell if * the hpts is the function calling the respective * routine. * * A few other notes: * * When a read lock is held, stability of the field is guaranteed; to write * to a field, a write lock must generally be held. * * netinet/netinet6-layer code should not assume that the inp_socket pointer * is safe to dereference without inp_lock being held, even for protocols * other than TCP (where the inpcb persists during TIMEWAIT even after the * socket has been freed), or there may be close(2)-related races. * * The inp_vflag field is overloaded, and would otherwise ideally be (c). * * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock * read-lock usage during modification, this model can be applied to other * protocols (especially SCTP). */ struct icmp6_filter; struct inpcbpolicy; struct m_snd_tag; struct inpcb { /* Cache line #1 (amd64) */ CK_LIST_ENTRY(inpcb) inp_hash; /* [w](h/i) [r](e/i) hash list */ CK_LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */ struct rwlock inp_lock; /* Cache line #2 (amd64) */ #define inp_start_zero inp_hpts #define inp_zero_size (sizeof(struct inpcb) - \ offsetof(struct inpcb, inp_start_zero)) TAILQ_ENTRY(inpcb) inp_hpts; /* pacing out queue next lock(b) */ uint32_t inp_hpts_request; /* Current hpts request, zero if * fits in the pacing window (i&b). */ /* * Note the next fields are protected by a * different lock (hpts-lock). This means that * they must correspond in size to the smallest * protectable bit field (uint8_t on x86, and * other platfomrs potentially uint32_t?). Also * since CPU switches can occur at different times the two * fields can *not* be collapsed into a signal bit field. */ #if defined(__amd64__) || defined(__i386__) volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */ volatile uint8_t inp_in_input; /* on input hpts (lock b) */ #else volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */ volatile uint32_t inp_in_input; /* on input hpts (lock b) */ #endif volatile uint16_t inp_hpts_cpu; /* Lock (i) */ u_int inp_refcount; /* (i) refcount */ int inp_flags; /* (i) generic IP/datagram flags */ int inp_flags2; /* (i) generic IP/datagram flags #2*/ volatile uint16_t inp_input_cpu; /* Lock (i) */ volatile uint8_t inp_hpts_cpu_set :1, /* on output hpts (i) */ inp_input_cpu_set : 1, /* on input hpts (i) */ inp_hpts_calls :1, /* (i) from output hpts */ inp_input_calls :1, /* (i) from input hpts */ inp_spare_bits2 : 4; uint8_t inp_numa_domain; /* numa domain */ void *inp_ppcb; /* (i) pointer to per-protocol pcb */ struct socket *inp_socket; /* (i) back pointer to socket */ uint32_t inp_hptsslot; /* Hpts wheel slot this tcb is Lock(i&b) */ uint32_t inp_hpts_drop_reas; /* reason we are dropping the PCB (lock i&b) */ TAILQ_ENTRY(inpcb) inp_input; /* pacing in queue next lock(b) */ struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */ struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */ CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */ struct ucred *inp_cred; /* (c) cache of socket cred */ u_int32_t inp_flow; /* (i) IPv6 flow information */ u_char inp_vflag; /* (i) IP version flag (v4/v6) */ u_char inp_ip_ttl; /* (i) time to live proto */ u_char inp_ip_p; /* (c) protocol proto */ u_char inp_ip_minttl; /* (i) minimum TTL or drop */ uint32_t inp_flowid; /* (x) flow id / queue id */ struct m_snd_tag *inp_snd_tag; /* (i) send tag for outgoing mbufs */ uint32_t inp_flowtype; /* (x) M_HASHTYPE value */ uint32_t inp_rss_listen_bucket; /* (x) overridden RSS listen bucket */ /* Local and foreign ports, local and foreign addr. */ struct in_conninfo inp_inc; /* (i) list for PCB's local port */ /* MAC and IPSEC policy information. */ struct label *inp_label; /* (i) MAC label */ struct inpcbpolicy *inp_sp; /* (s) for IPSEC */ /* Protocol-dependent part; options. */ struct { u_char inp_ip_tos; /* (i) type of service proto */ struct mbuf *inp_options; /* (i) IP options */ struct ip_moptions *inp_moptions; /* (i) mcast options */ }; struct { /* (i) IP options */ struct mbuf *in6p_options; /* (i) IP6 options for outgoing packets */ struct ip6_pktopts *in6p_outputopts; /* (i) IP multicast options */ struct ip6_moptions *in6p_moptions; /* (i) ICMPv6 code type filter */ struct icmp6_filter *in6p_icmp6filt; /* (i) IPV6_CHECKSUM setsockopt */ int in6p_cksum; short in6p_hops; }; CK_LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */ struct inpcbport *inp_phd; /* (i/h) head of this list */ inp_gen_t inp_gencnt; /* (c) generation count */ void *spare_ptr; /* Spare pointer. */ rt_gen_t inp_rt_cookie; /* generation for route entry */ union { /* cached L3 information */ struct route inp_route; struct route_in6 inp_route6; }; CK_LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */ /* (e[r]) for list iteration */ /* (p[w]/l) for addition/removal */ struct epoch_context inp_epoch_ctx; }; #endif /* _KERNEL */ #define inp_fport inp_inc.inc_fport #define inp_lport inp_inc.inc_lport #define inp_faddr inp_inc.inc_faddr #define inp_laddr inp_inc.inc_laddr #define in6p_faddr inp_inc.inc6_faddr #define in6p_laddr inp_inc.inc6_laddr #define in6p_zoneid inp_inc.inc6_zoneid #define inp_vnet inp_pcbinfo->ipi_vnet /* * The range of the generation count, as used in this implementation, is 9e19. * We would have to create 300 billion connections per second for this number * to roll over in a year. This seems sufficiently unlikely that we simply * don't concern ourselves with that possibility. */ /* * Interface exported to userland by various protocols which use inpcbs. Hack * alert -- only define if struct xsocket is in scope. * Fields prefixed with "xi_" are unique to this structure, and the rest * match fields in the struct inpcb, to ease coding and porting. * * Legend: * (s) - used by userland utilities in src * (p) - used by utilities in ports * (3) - is known to be used by third party software not in ports * (n) - no known usage */ #ifdef _SYS_SOCKETVAR_H_ struct xinpcb { ksize_t xi_len; /* length of this structure */ struct xsocket xi_socket; /* (s,p) */ struct in_conninfo inp_inc; /* (s,p) */ uint64_t inp_gencnt; /* (s,p) */ kvaddr_t inp_ppcb; /* (s) netstat(1) */ int64_t inp_spare64[4]; uint32_t inp_flow; /* (s) */ uint32_t inp_flowid; /* (s) */ uint32_t inp_flowtype; /* (s) */ int32_t inp_flags; /* (s,p) */ int32_t inp_flags2; /* (s) */ int32_t inp_rss_listen_bucket; /* (n) */ int32_t in6p_cksum; /* (n) */ int32_t inp_spare32[4]; uint16_t in6p_hops; /* (n) */ uint8_t inp_ip_tos; /* (n) */ int8_t pad8; uint8_t inp_vflag; /* (s,p) */ uint8_t inp_ip_ttl; /* (n) */ uint8_t inp_ip_p; /* (n) */ uint8_t inp_ip_minttl; /* (n) */ int8_t inp_spare8[4]; } __aligned(8); struct xinpgen { ksize_t xig_len; /* length of this structure */ u_int xig_count; /* number of PCBs at this time */ uint32_t _xig_spare32; inp_gen_t xig_gen; /* generation count at this time */ so_gen_t xig_sogen; /* socket generation count this time */ uint64_t _xig_spare64[4]; } __aligned(8); #ifdef _KERNEL void in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *); #endif #endif /* _SYS_SOCKETVAR_H_ */ struct inpcbport { struct epoch_context phd_epoch_ctx; CK_LIST_ENTRY(inpcbport) phd_hash; struct inpcbhead phd_pcblist; u_short phd_port; }; struct in_pcblist { int il_count; struct epoch_context il_epoch_ctx; struct inpcbinfo *il_pcbinfo; struct inpcb *il_inp_list[0]; }; /*- * Global data structure for each high-level protocol (UDP, TCP, ...) in both * IPv4 and IPv6. Holds inpcb lists and information for managing them. * * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and * ipi_list_lock: * - ipi_lock covering the global pcb list stability during loop iteration, * - ipi_hash_lock covering the hashed lookup tables, * - ipi_list_lock covering mutable global fields (such as the global * pcb list) * * The lock order is: * * ipi_lock (before) * inpcb locks (before) * ipi_list locks (before) * {ipi_hash_lock, pcbgroup locks} * * Locking key: * * (c) Constant or nearly constant after initialisation * (e) - Protected by the net_epoch_prempt epoch * (g) Locked by ipi_lock * (l) Locked by ipi_list_lock * (h) Read using either net_epoch_preempt or inpcb lock; write requires both ipi_hash_lock and inpcb lock * (p) Protected by one or more pcbgroup locks * (x) Synchronisation properties poorly defined */ struct inpcbinfo { /* * Global lock protecting inpcb list modification */ struct mtx ipi_lock; /* * Global list of inpcbs on the protocol. */ struct inpcbhead *ipi_listhead; /* [r](e) [w](g/l) */ u_int ipi_count; /* (l) */ /* * Generation count -- incremented each time a connection is allocated * or freed. */ u_quad_t ipi_gencnt; /* (l) */ /* * Fields associated with port lookup and allocation. */ u_short ipi_lastport; /* (x) */ u_short ipi_lastlow; /* (x) */ u_short ipi_lasthi; /* (x) */ /* * UMA zone from which inpcbs are allocated for this protocol. */ struct uma_zone *ipi_zone; /* (c) */ /* * Connection groups associated with this protocol. These fields are * constant, but pcbgroup structures themselves are protected by * per-pcbgroup locks. */ struct inpcbgroup *ipi_pcbgroups; /* (c) */ u_int ipi_npcbgroups; /* (c) */ u_int ipi_hashfields; /* (c) */ /* * Global lock protecting modification non-pcbgroup hash lookup tables. */ struct mtx ipi_hash_lock; /* * Global hash of inpcbs, hashed by local and foreign addresses and * port numbers. */ struct inpcbhead *ipi_hashbase; /* (h) */ u_long ipi_hashmask; /* (h) */ /* * Global hash of inpcbs, hashed by only local port number. */ struct inpcbporthead *ipi_porthashbase; /* (h) */ u_long ipi_porthashmask; /* (h) */ /* * List of wildcard inpcbs for use with pcbgroups. In the past, was * per-pcbgroup but is now global. All pcbgroup locks must be held * to modify the list, so any is sufficient to read it. */ struct inpcbhead *ipi_wildbase; /* (p) */ u_long ipi_wildmask; /* (p) */ /* * Load balance groups used for the SO_REUSEPORT_LB option, * hashed by local port. */ struct inpcblbgrouphead *ipi_lbgrouphashbase; /* (h) */ u_long ipi_lbgrouphashmask; /* (h) */ /* * Pointer to network stack instance */ struct vnet *ipi_vnet; /* (c) */ /* * general use 2 */ void *ipi_pspare[2]; /* * Global lock protecting global inpcb list, inpcb count, etc. */ struct rwlock ipi_list_lock; }; #ifdef _KERNEL /* * Connection groups hold sets of connections that have similar CPU/thread * affinity. Each connection belongs to exactly one connection group. */ struct inpcbgroup { /* * Per-connection group hash of inpcbs, hashed by local and foreign * addresses and port numbers. */ struct inpcbhead *ipg_hashbase; /* (c) */ u_long ipg_hashmask; /* (c) */ /* * Notional affinity of this pcbgroup. */ u_int ipg_cpu; /* (p) */ /* * Per-connection group lock, not to be confused with ipi_lock. * Protects the hash table hung off the group, but also the global * wildcard list in inpcbinfo. */ struct mtx ipg_lock; } __aligned(CACHE_LINE_SIZE); /* * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group * (or unique address:port combination) can be re-used at most * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which * is dynamically resized as processes bind/unbind to that specific group. */ struct inpcblbgroup { CK_LIST_ENTRY(inpcblbgroup) il_list; struct epoch_context il_epoch_ctx; uint16_t il_lport; /* (c) */ u_char il_vflag; /* (c) */ u_char il_pad; uint32_t il_pad2; union in_dependaddr il_dependladdr; /* (c) */ #define il_laddr il_dependladdr.id46_addr.ia46_addr4 #define il6_laddr il_dependladdr.id6_addr uint32_t il_inpsiz; /* max count in il_inp[] (h) */ uint32_t il_inpcnt; /* cur count in il_inp[] (h) */ struct inpcb *il_inp[]; /* (h) */ }; #define INP_LOCK_INIT(inp, d, t) \ rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK) #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock) #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock) #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock) #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock) #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock) #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock) #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock) #define INP_UNLOCK(inp) rw_unlock(&(inp)->inp_lock) #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock) #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock) #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock) #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED) #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED) #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED) #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED) /* * These locking functions are for inpcb consumers outside of sys/netinet, * more specifically, they were added for the benefit of TOE drivers. The * macros are reserved for use by the stack. */ void inp_wlock(struct inpcb *); void inp_wunlock(struct inpcb *); void inp_rlock(struct inpcb *); void inp_runlock(struct inpcb *); #ifdef INVARIANT_SUPPORT void inp_lock_assert(struct inpcb *); void inp_unlock_assert(struct inpcb *); #else #define inp_lock_assert(inp) do {} while (0) #define inp_unlock_assert(inp) do {} while (0) #endif void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg); int inp_ip_tos_get(const struct inpcb *inp); void inp_ip_tos_set(struct inpcb *inp, int val); struct socket * inp_inpcbtosocket(struct inpcb *inp); struct tcpcb * inp_inpcbtotcpcb(struct inpcb *inp); void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, uint32_t *faddr, uint16_t *fp); int inp_so_options(const struct inpcb *inp); #endif /* _KERNEL */ #define INP_INFO_LOCK_INIT(ipi, d) \ mtx_init(&(ipi)->ipi_lock, (d), NULL, MTX_DEF| MTX_RECURSE) #define INP_INFO_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_lock) #define INP_INFO_WLOCK(ipi) mtx_lock(&(ipi)->ipi_lock) #define INP_INFO_TRY_WLOCK(ipi) mtx_trylock(&(ipi)->ipi_lock) #define INP_INFO_WLOCKED(ipi) mtx_owned(&(ipi)->ipi_lock) #define INP_INFO_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_lock) #define INP_INFO_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_lock)) #define INP_INFO_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_lock, MA_OWNED) #define INP_INFO_WUNLOCK_ASSERT(ipi) \ mtx_assert(&(ipi)->ipi_lock, MA_NOTOWNED) #define INP_LIST_LOCK_INIT(ipi, d) \ rw_init_flags(&(ipi)->ipi_list_lock, (d), 0) #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock) #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock) #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock) #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock) #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock) #define INP_LIST_LOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED) #define INP_LIST_RLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED) #define INP_LIST_WLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED) #define INP_LIST_UNLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED) #define INP_HASH_LOCK_INIT(ipi, d) mtx_init(&(ipi)->ipi_hash_lock, (d), NULL, MTX_DEF) #define INP_HASH_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_hash_lock) #define INP_HASH_WLOCK(ipi) mtx_lock(&(ipi)->ipi_hash_lock) #define INP_HASH_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_hash_lock) #define INP_HASH_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_hash_lock)) #define INP_HASH_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_hash_lock, MA_OWNED); #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \ MTX_DEF | MTX_DUPOK) #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock) #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock) #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED) #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock) #define INP_PCBHASH(faddr, lport, fport, mask) \ (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask)) #define INP_PCBPORTHASH(lport, mask) \ (ntohs((lport)) & (mask)) #define INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \ ((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) #define INP6_PCBHASHKEY(faddr) ((faddr)->s6_addr32[3]) /* * Flags for inp_vflags -- historically version flags only */ #define INP_IPV4 0x1 #define INP_IPV6 0x2 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */ /* * Flags for inp_flags. */ #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */ #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */ #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */ #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */ #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */ #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */ #define INP_ANONPORT 0x00000040 /* port chosen for user */ #define INP_RECVIF 0x00000080 /* receive incoming interface */ #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */ /* 0x000200 unused: was INP_FAITH */ #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */ #define INP_DONTFRAG 0x00000800 /* don't fragment packet */ #define INP_BINDANY 0x00001000 /* allow bind to any address */ #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */ #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */ #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */ #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */ #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */ #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */ #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */ #define IN6P_RTHDR 0x00100000 /* receive routing header */ #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */ #define IN6P_TCLASS 0x00400000 /* receive traffic class value */ #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */ #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */ #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */ #define INP_DROPPED 0x04000000 /* protocol drop flag */ #define INP_SOCKREF 0x08000000 /* strong socket reference */ #define INP_RESERVED_0 0x10000000 /* reserved field */ #define INP_RESERVED_1 0x20000000 /* reserved field */ #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */ #define IN6P_MTU 0x80000000 /* receive path MTU */ #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\ INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\ IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\ IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\ IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\ IN6P_MTU) /* * Flags for inp_flags2. */ #define INP_2UNUSED1 0x00000001 #define INP_2UNUSED2 0x00000002 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */ #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */ #define INP_FREED 0x00000010 /* inp itself is not valid */ #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */ #define INP_BINDMULTI 0x00000040 /* IP_BINDMULTI option is set */ #define INP_RSS_BUCKET_SET 0x00000080 /* IP_RSS_LISTEN_BUCKET is set */ #define INP_RECVFLOWID 0x00000100 /* populate recv datagram with flow info */ #define INP_RECVRSSBUCKETID 0x00000200 /* populate recv datagram with bucket id */ #define INP_RATE_LIMIT_CHANGED 0x00000400 /* rate limit needs attention */ #define INP_ORIGDSTADDR 0x00000800 /* receive IP dst address/port */ #define INP_CANNOT_DO_ECN 0x00001000 /* The stack does not do ECN */ #define INP_REUSEPORT_LB 0x00002000 /* SO_REUSEPORT_LB option is set */ #define INP_SUPPORTS_MBUFQ 0x00004000 /* Supports the mbuf queue method of LRO */ #define INP_MBUF_QUEUE_READY 0x00008000 /* The transport is pacing, inputs can be queued */ #define INP_DONT_SACK_QUEUE 0x00010000 /* If a sack arrives do not wake me */ #define INP_2PCP_SET 0x00020000 /* If the Eth PCP should be set explicitly */ #define INP_2PCP_BIT0 0x00040000 /* Eth PCP Bit 0 */ #define INP_2PCP_BIT1 0x00080000 /* Eth PCP Bit 1 */ #define INP_2PCP_BIT2 0x00100000 /* Eth PCP Bit 2 */ #define INP_2PCP_BASE INP_2PCP_BIT0 #define INP_2PCP_MASK (INP_2PCP_BIT0 | INP_2PCP_BIT1 | INP_2PCP_BIT2) #define INP_2PCP_SHIFT 18 /* shift PCP field in/out of inp_flags2 */ /* * Flags passed to in_pcblookup*() functions. */ #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */ #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */ #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */ #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \ INPLOOKUP_WLOCKPCB) #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb) #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af) /* * Constants for pcbinfo.ipi_hashfields. */ #define IPI_HASHFIELDS_NONE 0 #define IPI_HASHFIELDS_2TUPLE 1 #define IPI_HASHFIELDS_4TUPLE 2 #ifdef _KERNEL VNET_DECLARE(int, ipport_reservedhigh); VNET_DECLARE(int, ipport_reservedlow); VNET_DECLARE(int, ipport_lowfirstauto); VNET_DECLARE(int, ipport_lowlastauto); VNET_DECLARE(int, ipport_firstauto); VNET_DECLARE(int, ipport_lastauto); VNET_DECLARE(int, ipport_hifirstauto); VNET_DECLARE(int, ipport_hilastauto); VNET_DECLARE(int, ipport_randomized); VNET_DECLARE(int, ipport_randomcps); VNET_DECLARE(int, ipport_randomtime); VNET_DECLARE(int, ipport_stoprandom); VNET_DECLARE(int, ipport_tcpallocs); #define V_ipport_reservedhigh VNET(ipport_reservedhigh) #define V_ipport_reservedlow VNET(ipport_reservedlow) #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto) #define V_ipport_lowlastauto VNET(ipport_lowlastauto) #define V_ipport_firstauto VNET(ipport_firstauto) #define V_ipport_lastauto VNET(ipport_lastauto) #define V_ipport_hifirstauto VNET(ipport_hifirstauto) #define V_ipport_hilastauto VNET(ipport_hilastauto) #define V_ipport_randomized VNET(ipport_randomized) #define V_ipport_randomcps VNET(ipport_randomcps) #define V_ipport_randomtime VNET(ipport_randomtime) #define V_ipport_stoprandom VNET(ipport_stoprandom) #define V_ipport_tcpallocs VNET(ipport_tcpallocs) void in_pcbinfo_destroy(struct inpcbinfo *); void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *, int, int, char *, uma_init, u_int); int in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi); struct inpcbgroup * in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t); struct inpcbgroup * in_pcbgroup_byinpcb(struct inpcb *); struct inpcbgroup * in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short, struct in_addr, u_short); void in_pcbgroup_destroy(struct inpcbinfo *); int in_pcbgroup_enabled(struct inpcbinfo *); void in_pcbgroup_init(struct inpcbinfo *, u_int, int); void in_pcbgroup_remove(struct inpcb *); void in_pcbgroup_update(struct inpcb *); void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *); void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *); int in_pcballoc(struct socket *, struct inpcbinfo *); int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *); int in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp, struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags); int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *, struct ucred *, int); int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *, u_short *, struct ucred *); int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *); int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *, struct mbuf *, bool); int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *, u_short *, in_addr_t *, u_short *, struct inpcb **, struct ucred *); void in_pcbdetach(struct inpcb *); void in_pcbdisconnect(struct inpcb *); void in_pcbdrop(struct inpcb *); void in_pcbfree(struct inpcb *); int in_pcbinshash(struct inpcb *); int in_pcbinshash_mbuf(struct inpcb *, struct mbuf *); int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *, struct ucred *); struct inpcb * in_pcblookup_local(struct inpcbinfo *, struct in_addr, u_short, int, struct ucred *); struct inpcb * in_pcblookup(struct inpcbinfo *, struct in_addr, u_int, struct in_addr, u_int, int, struct ifnet *); struct inpcb * in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int, struct in_addr, u_int, int, struct ifnet *, struct mbuf *); void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr, int, struct inpcb *(*)(struct inpcb *, int)); void in_pcbref(struct inpcb *); void in_pcbrehash(struct inpcb *); void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *); int in_pcbrele(struct inpcb *); int in_pcbrele_rlocked(struct inpcb *); int in_pcbrele_wlocked(struct inpcb *); void in_pcblist_rele_rlocked(epoch_context_t ctx); void in_losing(struct inpcb *); void in_pcbsetsolabel(struct socket *so); int in_getpeeraddr(struct socket *so, struct sockaddr **nam); int in_getsockaddr(struct socket *so, struct sockaddr **nam); struct sockaddr * in_sockaddr(in_port_t port, struct in_addr *addr); void in_pcbsosetlabel(struct socket *so); #ifdef RATELIMIT int in_pcboutput_txrtlmt_locked(struct inpcb *, struct ifnet *, struct mbuf *, uint32_t); int in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t, uint32_t, struct m_snd_tag **); void in_pcbdetach_txrtlmt(struct inpcb *); -void in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst); +void in_pcbdetach_tag(struct m_snd_tag *); int in_pcbmodify_txrtlmt(struct inpcb *, uint32_t); int in_pcbquery_txrtlmt(struct inpcb *, uint32_t *); int in_pcbquery_txrlevel(struct inpcb *, uint32_t *); void in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *); void in_pcboutput_eagain(struct inpcb *); #endif #endif /* _KERNEL */ #endif /* !_NETINET_IN_PCB_H_ */ Index: head/sys/netinet/tcp_ratelimit.c =================================================================== --- head/sys/netinet/tcp_ratelimit.c (revision 367147) +++ head/sys/netinet/tcp_ratelimit.c (revision 367148) @@ -1,1561 +1,1557 @@ /*- * * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2018-2020 * Netflix 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 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. * */ /** * Author: Randall Stewart */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #ifdef INET6 #include #endif #include #ifndef USECS_IN_SECOND #define USECS_IN_SECOND 1000000 #endif /* * For the purposes of each send, what is the size * of an ethernet frame. */ MALLOC_DEFINE(M_TCPPACE, "tcp_hwpace", "TCP Hardware pacing memory"); #ifdef RATELIMIT /* * The following preferred table will seem weird to * the casual viewer. Why do we not have any rates below * 1Mbps? Why do we have a rate at 1.44Mbps called common? * Why do the rates cluster in the 1-100Mbps range more * than others? Why does the table jump around at the beginnign * and then be more consistently raising? * * Let me try to answer those questions. A lot of * this is dependant on the hardware. We have three basic * supporters of rate limiting * * Chelsio - Supporting 16 configurable rates. * Mlx - c4 supporting 13 fixed rates. * Mlx - c5 & c6 supporting 127 configurable rates. * * The c4 is why we have a common rate that is available * in all rate tables. This is a selected rate from the * c4 table and we assure its available in all ratelimit * tables. This way the tcp_ratelimit code has an assured * rate it should always be able to get. This answers a * couple of the questions above. * * So what about the rest, well the table is built to * try to get the most out of a joint hardware/software * pacing system. The software pacer will always pick * a rate higher than the b/w that it is estimating * * on the path. This is done for two reasons. * a) So we can discover more b/w * and * b) So we can send a block of MSS's down and then * have the software timer go off after the previous * send is completely out of the hardware. * * But when we do we don't want to have the delay * between the last packet sent by the hardware be * excessively long (to reach our desired rate). * * So let me give an example for clarity. * * Lets assume that the tcp stack sees that 29,110,000 bps is * what the bw of the path is. The stack would select the * rate 31Mbps. 31Mbps means that each send that is done * by the hardware will cause a 390 micro-second gap between * the packets sent at that rate. For 29,110,000 bps we * would need 416 micro-seconds gap between each send. * * Note that are calculating a complete time for pacing * which includes the ethernet, IP and TCP overhead. So * a full 1514 bytes is used for the above calculations. * My testing has shown that both cards are also using this * as their basis i.e. full payload size of the ethernet frame. * The TCP stack caller needs to be aware of this and make the * appropriate overhead calculations be included in its choices. * * Now, continuing our example, we pick a MSS size based on the * delta between the two rates (416 - 390) divided into the rate * we really wish to send at rounded up. That results in a MSS * send of 17 mss's at once. The hardware then will * run out of data in a single 17MSS send in 6,630 micro-seconds. * * On the other hand the software pacer will send more data * in 7,072 micro-seconds. This means that we will refill * the hardware 52 microseconds after it would have sent * next if it had not ran out of data. This is a win since we are * only sending every 7ms or so and yet all the packets are spaced on * the wire with 94% of what they should be and only * the last packet is delayed extra to make up for the * difference. * * Note that the above formula has two important caveat. * If we are above (b/w wise) over 100Mbps we double the result * of the MSS calculation. The second caveat is if we are 500Mbps * or more we just send the maximum MSS at once i.e. 45MSS. At * the higher b/w's even the cards have limits to what times (timer granularity) * they can insert between packets and start to send more than one * packet at a time on the wire. * */ #define COMMON_RATE 180500 const uint64_t desired_rates[] = { 122500, /* 1Mbps - rate 1 */ 180500, /* 1.44Mpbs - rate 2 common rate */ 375000, /* 3Mbps - rate 3 */ 625000, /* 5Mbps - rate 4 */ 875000, /* 7Mbps - rate 5 */ 1125000, /* 9Mbps - rate 6 */ 1375000, /* 11Mbps - rate 7 */ 1625000, /* 13Mbps - rate 8 */ 2625000, /* 21Mbps - rate 9 */ 3875000, /* 31Mbps - rate 10 */ 5125000, /* 41Meg - rate 11 */ 12500000, /* 100Mbps - rate 12 */ 25000000, /* 200Mbps - rate 13 */ 50000000, /* 400Mbps - rate 14 */ 63750000, /* 51Mbps - rate 15 */ 100000000, /* 800Mbps - rate 16 */ 1875000, /* 15Mbps - rate 17 */ 2125000, /* 17Mbps - rate 18 */ 2375000, /* 19Mbps - rate 19 */ 2875000, /* 23Mbps - rate 20 */ 3125000, /* 25Mbps - rate 21 */ 3375000, /* 27Mbps - rate 22 */ 3625000, /* 29Mbps - rate 23 */ 4125000, /* 33Mbps - rate 24 */ 4375000, /* 35Mbps - rate 25 */ 4625000, /* 37Mbps - rate 26 */ 4875000, /* 39Mbps - rate 27 */ 5375000, /* 43Mbps - rate 28 */ 5625000, /* 45Mbps - rate 29 */ 5875000, /* 47Mbps - rate 30 */ 6125000, /* 49Mbps - rate 31 */ 6625000, /* 53Mbps - rate 32 */ 6875000, /* 55Mbps - rate 33 */ 7125000, /* 57Mbps - rate 34 */ 7375000, /* 59Mbps - rate 35 */ 7625000, /* 61Mbps - rate 36 */ 7875000, /* 63Mbps - rate 37 */ 8125000, /* 65Mbps - rate 38 */ 8375000, /* 67Mbps - rate 39 */ 8625000, /* 69Mbps - rate 40 */ 8875000, /* 71Mbps - rate 41 */ 9125000, /* 73Mbps - rate 42 */ 9375000, /* 75Mbps - rate 43 */ 9625000, /* 77Mbps - rate 44 */ 9875000, /* 79Mbps - rate 45 */ 10125000, /* 81Mbps - rate 46 */ 10375000, /* 83Mbps - rate 47 */ 10625000, /* 85Mbps - rate 48 */ 10875000, /* 87Mbps - rate 49 */ 11125000, /* 89Mbps - rate 50 */ 11375000, /* 91Mbps - rate 51 */ 11625000, /* 93Mbps - rate 52 */ 11875000, /* 95Mbps - rate 53 */ 13125000, /* 105Mbps - rate 54 */ 13750000, /* 110Mbps - rate 55 */ 14375000, /* 115Mbps - rate 56 */ 15000000, /* 120Mbps - rate 57 */ 15625000, /* 125Mbps - rate 58 */ 16250000, /* 130Mbps - rate 59 */ 16875000, /* 135Mbps - rate 60 */ 17500000, /* 140Mbps - rate 61 */ 18125000, /* 145Mbps - rate 62 */ 18750000, /* 150Mbps - rate 64 */ 20000000, /* 160Mbps - rate 65 */ 21250000, /* 170Mbps - rate 66 */ 22500000, /* 180Mbps - rate 67 */ 23750000, /* 190Mbps - rate 68 */ 26250000, /* 210Mbps - rate 69 */ 27500000, /* 220Mbps - rate 70 */ 28750000, /* 230Mbps - rate 71 */ 30000000, /* 240Mbps - rate 72 */ 31250000, /* 250Mbps - rate 73 */ 34375000, /* 275Mbps - rate 74 */ 37500000, /* 300Mbps - rate 75 */ 40625000, /* 325Mbps - rate 76 */ 43750000, /* 350Mbps - rate 77 */ 46875000, /* 375Mbps - rate 78 */ 53125000, /* 425Mbps - rate 79 */ 56250000, /* 450Mbps - rate 80 */ 59375000, /* 475Mbps - rate 81 */ 62500000, /* 500Mbps - rate 82 */ 68750000, /* 550Mbps - rate 83 */ 75000000, /* 600Mbps - rate 84 */ 81250000, /* 650Mbps - rate 85 */ 87500000, /* 700Mbps - rate 86 */ 93750000, /* 750Mbps - rate 87 */ 106250000, /* 850Mbps - rate 88 */ 112500000, /* 900Mbps - rate 89 */ 125000000, /* 1Gbps - rate 90 */ 156250000, /* 1.25Gps - rate 91 */ 187500000, /* 1.5Gps - rate 92 */ 218750000, /* 1.75Gps - rate 93 */ 250000000, /* 2Gbps - rate 94 */ 281250000, /* 2.25Gps - rate 95 */ 312500000, /* 2.5Gbps - rate 96 */ 343750000, /* 2.75Gbps - rate 97 */ 375000000, /* 3Gbps - rate 98 */ 500000000, /* 4Gbps - rate 99 */ 625000000, /* 5Gbps - rate 100 */ 750000000, /* 6Gbps - rate 101 */ 875000000, /* 7Gbps - rate 102 */ 1000000000, /* 8Gbps - rate 103 */ 1125000000, /* 9Gbps - rate 104 */ 1250000000, /* 10Gbps - rate 105 */ 1875000000, /* 15Gbps - rate 106 */ 2500000000 /* 20Gbps - rate 107 */ }; #define MAX_HDWR_RATES (sizeof(desired_rates)/sizeof(uint64_t)) #define RS_ORDERED_COUNT 16 /* * Number that are in order * at the beginning of the table, * over this a sort is required. */ #define RS_NEXT_ORDER_GROUP 16 /* * The point in our table where * we come fill in a second ordered * group (index wise means -1). */ #define ALL_HARDWARE_RATES 1004 /* * 1Meg - 1Gig in 1 Meg steps * plus 100, 200k and 500k and * 10Gig */ #define RS_ONE_MEGABIT_PERSEC 1000000 #define RS_ONE_GIGABIT_PERSEC 1000000000 #define RS_TEN_GIGABIT_PERSEC 10000000000 static struct head_tcp_rate_set int_rs; static struct mtx rs_mtx; uint32_t rs_number_alive; uint32_t rs_number_dead; SYSCTL_NODE(_net_inet_tcp, OID_AUTO, rl, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP Ratelimit stats"); SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, alive, CTLFLAG_RW, &rs_number_alive, 0, "Number of interfaces initialized for ratelimiting"); SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, dead, CTLFLAG_RW, &rs_number_dead, 0, "Number of interfaces departing from ratelimiting"); static void rl_add_syctl_entries(struct sysctl_oid *rl_sysctl_root, struct tcp_rate_set *rs) { /* * Add sysctl entries for thus interface. */ if (rs->rs_flags & RS_INTF_NO_SUP) { SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "disable", CTLFLAG_RD, &rs->rs_disable, 0, "Disable this interface from new hdwr limiting?"); } else { SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "disable", CTLFLAG_RW, &rs->rs_disable, 0, "Disable this interface from new hdwr limiting?"); } SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "minseg", CTLFLAG_RW, &rs->rs_min_seg, 0, "What is the minimum we need to send on this interface?"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flow_limit", CTLFLAG_RW, &rs->rs_flow_limit, 0, "What is the limit for number of flows (0=unlimited)?"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "highest", CTLFLAG_RD, &rs->rs_highest_valid, 0, "Highest valid rate"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "lowest", CTLFLAG_RD, &rs->rs_lowest_valid, 0, "Lowest valid rate"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flags", CTLFLAG_RD, &rs->rs_flags, 0, "What lags are on the entry?"); SYSCTL_ADD_S32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "numrates", CTLFLAG_RD, &rs->rs_rate_cnt, 0, "How many rates re there?"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "flows_using", CTLFLAG_RD, &rs->rs_flows_using, 0, "How many flows are using this interface now?"); #ifdef DETAILED_RATELIMIT_SYSCTL if (rs->rs_rlt && rs->rs_rate_cnt > 0) { /* Lets display the rates */ int i; struct sysctl_oid *rl_rates; struct sysctl_oid *rl_rate_num; char rate_num[16]; rl_rates = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_sysctl_root), OID_AUTO, "rate", CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Ratelist"); for( i = 0; i < rs->rs_rate_cnt; i++) { sprintf(rate_num, "%d", i); rl_rate_num = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rates), OID_AUTO, rate_num, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Individual Rate"); SYSCTL_ADD_U32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "flags", CTLFLAG_RD, &rs->rs_rlt[i].flags, 0, "Flags on this rate"); SYSCTL_ADD_U32(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "pacetime", CTLFLAG_RD, &rs->rs_rlt[i].time_between, 0, "Time hardware inserts between 1500 byte sends"); SYSCTL_ADD_U64(&rs->sysctl_ctx, SYSCTL_CHILDREN(rl_rate_num), OID_AUTO, "rate", CTLFLAG_RD, &rs->rs_rlt[i].rate, 0, "Rate in bytes per second"); } } #endif } static void rs_destroy(epoch_context_t ctx) { struct tcp_rate_set *rs; bool do_free_rs; rs = __containerof(ctx, struct tcp_rate_set, rs_epoch_ctx); mtx_lock(&rs_mtx); rs->rs_flags &= ~RS_FUNERAL_SCHD; /* * In theory its possible (but unlikely) * that while the delete was occuring * and we were applying the DEAD flag * someone slipped in and found the * interface in a lookup. While we * decided rs_flows_using were 0 and * scheduling the epoch_call, the other * thread incremented rs_flow_using. This * is because users have a pointer and * we only use the rs_flows_using in an * atomic fashion, i.e. the other entities * are not protected. To assure this did * not occur, we check rs_flows_using here * before deleting. */ do_free_rs = (rs->rs_flows_using == 0); rs_number_dead--; mtx_unlock(&rs_mtx); if (do_free_rs) { sysctl_ctx_free(&rs->sysctl_ctx); free(rs->rs_rlt, M_TCPPACE); free(rs, M_TCPPACE); } } static void rs_defer_destroy(struct tcp_rate_set *rs) { mtx_assert(&rs_mtx, MA_OWNED); /* Check if already pending. */ if (rs->rs_flags & RS_FUNERAL_SCHD) return; rs_number_dead++; /* Set flag to only defer once. */ rs->rs_flags |= RS_FUNERAL_SCHD; NET_EPOCH_CALL(rs_destroy, &rs->rs_epoch_ctx); } #ifdef INET extern counter_u64_t rate_limit_set_ok; extern counter_u64_t rate_limit_active; extern counter_u64_t rate_limit_alloc_fail; #endif static int rl_attach_txrtlmt(struct ifnet *ifp, uint32_t flowtype, int flowid, uint64_t cfg_rate, struct m_snd_tag **tag) { int error; union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = flowid, .rate_limit.hdr.flowtype = flowtype, .rate_limit.max_rate = cfg_rate, .rate_limit.flags = M_NOWAIT, }; if (ifp->if_snd_tag_alloc == NULL) { error = EOPNOTSUPP; } else { error = ifp->if_snd_tag_alloc(ifp, ¶ms, tag); #ifdef INET if (error == 0) { counter_u64_add(rate_limit_set_ok, 1); counter_u64_add(rate_limit_active, 1); } else counter_u64_add(rate_limit_alloc_fail, 1); #endif } return (error); } static void populate_canned_table(struct tcp_rate_set *rs, const uint64_t *rate_table_act) { /* * The internal table is "special", it * is two seperate ordered tables that * must be merged. We get here when the * adapter specifies a number of rates that * covers both ranges in the table in some * form. */ int i, at_low, at_high; uint8_t low_disabled = 0, high_disabled = 0; for(i = 0, at_low = 0, at_high = RS_NEXT_ORDER_GROUP; i < rs->rs_rate_cnt; i++) { rs->rs_rlt[i].flags = 0; rs->rs_rlt[i].time_between = 0; if ((low_disabled == 0) && (high_disabled || (rate_table_act[at_low] < rate_table_act[at_high]))) { rs->rs_rlt[i].rate = rate_table_act[at_low]; at_low++; if (at_low == RS_NEXT_ORDER_GROUP) low_disabled = 1; } else if (high_disabled == 0) { rs->rs_rlt[i].rate = rate_table_act[at_high]; at_high++; if (at_high == MAX_HDWR_RATES) high_disabled = 1; } } } static struct tcp_rate_set * rt_setup_new_rs(struct ifnet *ifp, int *error) { struct tcp_rate_set *rs; const uint64_t *rate_table_act; uint64_t lentim, res; size_t sz; uint32_t hash_type; int i; struct if_ratelimit_query_results rl; struct sysctl_oid *rl_sysctl_root; /* * We expect to enter with the * mutex locked. */ if (ifp->if_ratelimit_query == NULL) { /* * We can do nothing if we cannot * get a query back from the driver. */ printf("Warning:No query functions for %s:%d-- failed\n", ifp->if_dname, ifp->if_dunit); return (NULL); } rs = malloc(sizeof(struct tcp_rate_set), M_TCPPACE, M_NOWAIT | M_ZERO); if (rs == NULL) { if (error) *error = ENOMEM; printf("Warning:No memory for malloc of tcp_rate_set\n"); return (NULL); } memset(&rl, 0, sizeof(rl)); rl.flags = RT_NOSUPPORT; ifp->if_ratelimit_query(ifp, &rl); if (rl.flags & RT_IS_UNUSABLE) { /* * The interface does not really support * the rate-limiting. */ memset(rs, 0, sizeof(struct tcp_rate_set)); rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_flags = RS_INTF_NO_SUP; rs->rs_disable = 1; rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } else if ((rl.flags & RT_IS_INDIRECT) == RT_IS_INDIRECT) { memset(rs, 0, sizeof(struct tcp_rate_set)); rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_flags = RS_IS_DEFF; rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } else if ((rl.flags & RT_IS_FIXED_TABLE) == RT_IS_FIXED_TABLE) { /* Mellanox C4 likely */ rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_rate_cnt = rl.number_of_rates; rs->rs_min_seg = rl.min_segment_burst; rs->rs_highest_valid = 0; rs->rs_flow_limit = rl.max_flows; rs->rs_flags = RS_IS_INTF | RS_NO_PRE; rs->rs_disable = 0; rate_table_act = rl.rate_table; } else if ((rl.flags & RT_IS_SELECTABLE) == RT_IS_SELECTABLE) { /* Chelsio, C5 and C6 of Mellanox? */ rs->rs_ifp = ifp; rs->rs_if_dunit = ifp->if_dunit; rs->rs_rate_cnt = rl.number_of_rates; rs->rs_min_seg = rl.min_segment_burst; rs->rs_disable = 0; rs->rs_flow_limit = rl.max_flows; rate_table_act = desired_rates; if ((rs->rs_rate_cnt > MAX_HDWR_RATES) && (rs->rs_rate_cnt < ALL_HARDWARE_RATES)) { /* * Our desired table is not big * enough, do what we can. */ rs->rs_rate_cnt = MAX_HDWR_RATES; } if (rs->rs_rate_cnt <= RS_ORDERED_COUNT) rs->rs_flags = RS_IS_INTF; else rs->rs_flags = RS_IS_INTF | RS_INT_TBL; if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) rs->rs_rate_cnt = ALL_HARDWARE_RATES; } else { free(rs, M_TCPPACE); return (NULL); } sz = sizeof(struct tcp_hwrate_limit_table) * rs->rs_rate_cnt; rs->rs_rlt = malloc(sz, M_TCPPACE, M_NOWAIT); if (rs->rs_rlt == NULL) { if (error) *error = ENOMEM; bail: free(rs, M_TCPPACE); return (NULL); } if (rs->rs_rate_cnt >= ALL_HARDWARE_RATES) { /* * The interface supports all * the rates we could possibly want. */ uint64_t rat; rs->rs_rlt[0].rate = 12500; /* 100k */ rs->rs_rlt[1].rate = 25000; /* 200k */ rs->rs_rlt[2].rate = 62500; /* 500k */ /* Note 125000 == 1Megabit * populate 1Meg - 1000meg. */ for(i = 3, rat = 125000; i< (ALL_HARDWARE_RATES-1); i++) { rs->rs_rlt[i].rate = rat; rat += 125000; } rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate = 1250000000; } else if (rs->rs_flags & RS_INT_TBL) { /* We populate this in a special way */ populate_canned_table(rs, rate_table_act); } else { /* * Just copy in the rates from * the table, it is in order. */ for (i=0; irs_rate_cnt; i++) { rs->rs_rlt[i].rate = rate_table_act[i]; rs->rs_rlt[i].time_between = 0; rs->rs_rlt[i].flags = 0; } } for (i = (rs->rs_rate_cnt - 1); i >= 0; i--) { /* * We go backwards through the list so that if we can't get * a rate and fail to init one, we have at least a chance of * getting the highest one. */ rs->rs_rlt[i].ptbl = rs; rs->rs_rlt[i].tag = NULL; /* * Calculate the time between. */ lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND; res = lentim / rs->rs_rlt[i].rate; if (res > 0) rs->rs_rlt[i].time_between = res; else rs->rs_rlt[i].time_between = 1; if (rs->rs_flags & RS_NO_PRE) { rs->rs_rlt[i].flags = HDWRPACE_INITED; rs->rs_lowest_valid = i; } else { int err; if ((rl.flags & RT_IS_SETUP_REQ) && (ifp->if_ratelimit_query)) { err = ifp->if_ratelimit_setup(ifp, rs->rs_rlt[i].rate, i); if (err) goto handle_err; } #ifdef RSS hash_type = M_HASHTYPE_RSS_TCP_IPV4; #else hash_type = M_HASHTYPE_OPAQUE_HASH; #endif err = rl_attach_txrtlmt(ifp, hash_type, (i + 1), rs->rs_rlt[i].rate, &rs->rs_rlt[i].tag); if (err) { handle_err: if (i == (rs->rs_rate_cnt - 1)) { /* * Huh - first rate and we can't get * it? */ free(rs->rs_rlt, M_TCPPACE); if (error) *error = err; goto bail; } else { if (error) *error = err; } break; } else { rs->rs_rlt[i].flags = HDWRPACE_INITED | HDWRPACE_TAGPRESENT; rs->rs_lowest_valid = i; } } } /* Did we get at least 1 rate? */ if (rs->rs_rlt[(rs->rs_rate_cnt - 1)].flags & HDWRPACE_INITED) rs->rs_highest_valid = rs->rs_rate_cnt - 1; else { free(rs->rs_rlt, M_TCPPACE); goto bail; } rs_number_alive++; sysctl_ctx_init(&rs->sysctl_ctx); rl_sysctl_root = SYSCTL_ADD_NODE(&rs->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_rl), OID_AUTO, rs->rs_ifp->if_xname, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, ""); rl_add_syctl_entries(rl_sysctl_root, rs); mtx_lock(&rs_mtx); CK_LIST_INSERT_HEAD(&int_rs, rs, next); mtx_unlock(&rs_mtx); return (rs); } static const struct tcp_hwrate_limit_table * tcp_int_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags) { struct tcp_hwrate_limit_table *arte = NULL, *rte = NULL; uint64_t mbits_per_sec, ind_calc; int i; mbits_per_sec = (bytes_per_sec * 8); if (flags & RS_PACING_LT) { if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ /* * Smaller than 1Meg, only * 3 entries can match it. */ for(i = rs->rs_lowest_valid; i < 3; i++) { if (bytes_per_sec <= rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } else if (rs->rs_rlt[i].flags & HDWRPACE_INITED) { arte = &rs->rs_rlt[i]; } } goto done; } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ /* * Larger than 1G (the majority of * our table. */ if (mbits_per_sec < RS_TEN_GIGABIT_PERSEC) rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; else arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; goto done; } /* * If we reach here its in our table (between 1Meg - 1000Meg), * just take the rounded down mbits per second, and add * 1Megabit to it, from this we can calculate * the index in the table. */ ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) != mbits_per_sec) ind_calc++; /* our table is offset by 3, we add 2 */ ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if ((ind_calc >= rs->rs_lowest_valid) && (ind_calc <= rs->rs_highest_valid)) rte = &rs->rs_rlt[ind_calc]; } else if (flags & RS_PACING_EXACT_MATCH) { if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ for(i = rs->rs_lowest_valid; i < 3; i++) { if (bytes_per_sec == rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } } } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { /* > 1Gbps only one rate */ if (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) { /* Its 10G wow */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } } else { /* Ok it must be a exact meg (its between 1G and 1Meg) */ ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { /* its an exact Mbps */ ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) rte = &rs->rs_rlt[ind_calc]; } } } else { /* we want greater than the requested rate */ if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && (rs->rs_lowest_valid <= 2)){ arte = &rs->rs_rlt[3]; /* set alternate to 1Meg */ for (i=2; i>=rs->rs_lowest_valid; i--) { if (bytes_per_sec < rs->rs_rlt[i].rate) { rte = &rs->rs_rlt[i]; break; } else if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; break; } else { arte = &rs->rs_rlt[i]; /* new alternate */ } } } else if (mbits_per_sec > RS_ONE_GIGABIT_PERSEC) { if ((bytes_per_sec < rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)){ /* Our top rate is larger than the request */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } else if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[(ALL_HARDWARE_RATES-1)].rate) && (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED)) { /* It matches our top rate */ rte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } else if (rs->rs_rlt[(ALL_HARDWARE_RATES-1)].flags & HDWRPACE_INITED) { /* The top rate is an alternative */ arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; } } else { /* Its in our range 1Meg - 1Gig */ if (flags & RS_PACING_GEQ) { ind_calc = mbits_per_sec/RS_ONE_MEGABIT_PERSEC; if ((ind_calc * RS_ONE_MEGABIT_PERSEC) == mbits_per_sec) { if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = (ALL_HARDWARE_RATES-1); } rte = &rs->rs_rlt[ind_calc]; } goto done; } ind_calc = (mbits_per_sec + (RS_ONE_MEGABIT_PERSEC-1))/RS_ONE_MEGABIT_PERSEC; ind_calc += 2; if (ind_calc > (ALL_HARDWARE_RATES-1)) { /* This should not happen */ ind_calc = ALL_HARDWARE_RATES-1; } if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) rte = &rs->rs_rlt[ind_calc]; } } done: if ((rte == NULL) && (arte != NULL) && (flags & RS_PACING_SUB_OK)) { /* We can use the substitute */ rte = arte; } return (rte); } static const struct tcp_hwrate_limit_table * tcp_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t bytes_per_sec, uint32_t flags) { /** * Hunt the rate table with the restrictions in flags and find a * suitable rate if possible. * RS_PACING_EXACT_MATCH - look for an exact match to rate. * RS_PACING_GT - must be greater than. * RS_PACING_GEQ - must be greater than or equal. * RS_PACING_LT - must be less than. * RS_PACING_SUB_OK - If we don't meet criteria a * substitute is ok. */ int i, matched; struct tcp_hwrate_limit_table *rte = NULL; if ((rs->rs_flags & RS_INT_TBL) && (rs->rs_rate_cnt >= ALL_HARDWARE_RATES)) { /* * Here we don't want to paw thru * a big table, we have everything * from 1Meg - 1000Meg in 1Meg increments. * Use an alternate method to "lookup". */ return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags)); } if ((flags & RS_PACING_LT) || (flags & RS_PACING_EXACT_MATCH)) { /* * For exact and less than we go forward through the table. * This way when we find one larger we stop (exact was a * toss up). */ for (i = rs->rs_lowest_valid, matched = 0; i <= rs->rs_highest_valid; i++) { if ((flags & RS_PACING_EXACT_MATCH) && (bytes_per_sec == rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; matched = 1; break; } else if ((flags & RS_PACING_LT) && (bytes_per_sec <= rs->rs_rlt[i].rate)) { rte = &rs->rs_rlt[i]; matched = 1; break; } if (bytes_per_sec > rs->rs_rlt[i].rate) break; } if ((matched == 0) && (flags & RS_PACING_LT) && (flags & RS_PACING_SUB_OK)) { /* Kick in a substitute (the lowest) */ rte = &rs->rs_rlt[rs->rs_lowest_valid]; } } else { /* * Here we go backward through the table so that we can find * the one greater in theory faster (but its probably a * wash). */ for (i = rs->rs_highest_valid, matched = 0; i >= rs->rs_lowest_valid; i--) { if (rs->rs_rlt[i].rate > bytes_per_sec) { /* A possible candidate */ rte = &rs->rs_rlt[i]; } if ((flags & RS_PACING_GEQ) && (bytes_per_sec == rs->rs_rlt[i].rate)) { /* An exact match and we want equal */ matched = 1; rte = &rs->rs_rlt[i]; break; } else if (rte) { /* * Found one that is larger than but don't * stop, there may be a more closer match. */ matched = 1; } if (rs->rs_rlt[i].rate < bytes_per_sec) { /* * We found a table entry that is smaller, * stop there will be none greater or equal. */ break; } } if ((matched == 0) && (flags & RS_PACING_SUB_OK)) { /* Kick in a substitute (the highest) */ rte = &rs->rs_rlt[rs->rs_highest_valid]; } } return (rte); } static struct ifnet * rt_find_real_interface(struct ifnet *ifp, struct inpcb *inp, int *error) { struct ifnet *tifp; struct m_snd_tag *tag; union if_snd_tag_alloc_params params = { .rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT, .rate_limit.hdr.flowid = 1, .rate_limit.hdr.numa_domain = inp->inp_numa_domain, .rate_limit.max_rate = COMMON_RATE, .rate_limit.flags = M_NOWAIT, }; int err; #ifdef RSS params.rate_limit.hdr.flowtype = ((inp->inp_vflag & INP_IPV6) ? M_HASHTYPE_RSS_TCP_IPV6 : M_HASHTYPE_RSS_TCP_IPV4); #else params.rate_limit.hdr.flowtype = M_HASHTYPE_OPAQUE_HASH; #endif tag = NULL; if (ifp->if_snd_tag_alloc) { if (error) *error = ENODEV; return (NULL); } err = ifp->if_snd_tag_alloc(ifp, ¶ms, &tag); if (err) { /* Failed to setup a tag? */ if (error) *error = err; return (NULL); } tifp = tag->ifp; - tifp->if_snd_tag_free(tag); + m_snd_tag_rele(tag); return (tifp); } static const struct tcp_hwrate_limit_table * rt_setup_rate(struct inpcb *inp, struct ifnet *ifp, uint64_t bytes_per_sec, uint32_t flags, int *error) { /* First lets find the interface if it exists */ const struct tcp_hwrate_limit_table *rte; struct tcp_rate_set *rs; struct epoch_tracker et; int err; NET_EPOCH_ENTER(et); use_real_interface: CK_LIST_FOREACH(rs, &int_rs, next) { /* * Note we don't look with the lock since we either see a * new entry or will get one when we try to add it. */ if (rs->rs_flags & RS_IS_DEAD) { /* The dead are not looked at */ continue; } if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { /* Ok we found it */ break; } } if ((rs == NULL) || (rs->rs_flags & RS_INTF_NO_SUP) || (rs->rs_flags & RS_IS_DEAD)) { /* * This means we got a packet *before* * the IF-UP was processed below, * while or after we already received an interface * departed event. In either case we really don't * want to do anything with pacing, in * the departing case the packet is not * going to go very far. The new case * might be arguable, but its impossible * to tell from the departing case. */ if (rs->rs_disable && error) *error = ENODEV; NET_EPOCH_EXIT(et); return (NULL); } if ((rs == NULL) || (rs->rs_disable != 0)) { if (rs->rs_disable && error) *error = ENOSPC; NET_EPOCH_EXIT(et); return (NULL); } if (rs->rs_flags & RS_IS_DEFF) { /* We need to find the real interface */ struct ifnet *tifp; tifp = rt_find_real_interface(ifp, inp, error); if (tifp == NULL) { if (rs->rs_disable && error) *error = ENOTSUP; NET_EPOCH_EXIT(et); return (NULL); } goto use_real_interface; } if (rs->rs_flow_limit && ((rs->rs_flows_using + 1) > rs->rs_flow_limit)) { if (error) *error = ENOSPC; NET_EPOCH_EXIT(et); return (NULL); } rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); if (rte) { err = in_pcbattach_txrtlmt(inp, rs->rs_ifp, inp->inp_flowtype, inp->inp_flowid, rte->rate, &inp->inp_snd_tag); if (err) { /* Failed to attach */ if (error) *error = err; rte = NULL; } } if (rte) { /* * We use an atomic here for accounting so we don't have to * use locks when freeing. */ atomic_add_64(&rs->rs_flows_using, 1); } NET_EPOCH_EXIT(et); return (rte); } static void tcp_rl_ifnet_link(void *arg __unused, struct ifnet *ifp, int link_state) { int error; struct tcp_rate_set *rs; if (((ifp->if_capenable & IFCAP_TXRTLMT) == 0) || (link_state != LINK_STATE_UP)) { /* * We only care on an interface going up that is rate-limit * capable. */ return; } mtx_lock(&rs_mtx); CK_LIST_FOREACH(rs, &int_rs, next) { if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { /* We already have initialized this guy */ mtx_unlock(&rs_mtx); return; } } mtx_unlock(&rs_mtx); rt_setup_new_rs(ifp, &error); } static void tcp_rl_ifnet_departure(void *arg __unused, struct ifnet *ifp) { struct tcp_rate_set *rs, *nrs; - struct ifnet *tifp; int i; mtx_lock(&rs_mtx); CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { if ((rs->rs_ifp == ifp) && (rs->rs_if_dunit == ifp->if_dunit)) { CK_LIST_REMOVE(rs, next); rs_number_alive--; rs->rs_flags |= RS_IS_DEAD; for (i = 0; i < rs->rs_rate_cnt; i++) { if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { - tifp = rs->rs_rlt[i].tag->ifp; - in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); + in_pcbdetach_tag(rs->rs_rlt[i].tag); rs->rs_rlt[i].tag = NULL; } rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; } if (rs->rs_flows_using == 0) rs_defer_destroy(rs); break; } } mtx_unlock(&rs_mtx); } static void tcp_rl_shutdown(void *arg __unused, int howto __unused) { struct tcp_rate_set *rs, *nrs; - struct ifnet *tifp; int i; mtx_lock(&rs_mtx); CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { CK_LIST_REMOVE(rs, next); rs_number_alive--; rs->rs_flags |= RS_IS_DEAD; for (i = 0; i < rs->rs_rate_cnt; i++) { if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { - tifp = rs->rs_rlt[i].tag->ifp; - in_pcbdetach_tag(tifp, rs->rs_rlt[i].tag); + in_pcbdetach_tag(rs->rs_rlt[i].tag); rs->rs_rlt[i].tag = NULL; } rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED; } if (rs->rs_flows_using == 0) rs_defer_destroy(rs); } mtx_unlock(&rs_mtx); } const struct tcp_hwrate_limit_table * tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, uint64_t bytes_per_sec, int flags, int *error) { const struct tcp_hwrate_limit_table *rte; #ifdef KERN_TLS struct ktls_session *tls; #endif INP_WLOCK_ASSERT(tp->t_inpcb); if (tp->t_inpcb->inp_snd_tag == NULL) { /* * We are setting up a rate for the first time. */ if ((ifp->if_capenable & IFCAP_TXRTLMT) == 0) { /* Not supported by the egress */ if (error) *error = ENODEV; return (NULL); } #ifdef KERN_TLS tls = NULL; if (tp->t_inpcb->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) { tls = tp->t_inpcb->inp_socket->so_snd.sb_tls_info; if ((ifp->if_capenable & IFCAP_TXTLS_RTLMT) == 0 || tls->mode != TCP_TLS_MODE_IFNET) { if (error) *error = ENODEV; return (NULL); } } #endif rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error); #ifdef KERN_TLS if (rte != NULL && tls != NULL && tls->snd_tag != NULL) { /* * Fake a route change error to reset the TLS * send tag. This will convert the existing * tag to a TLS ratelimit tag. */ MPASS(tls->snd_tag->type == IF_SND_TAG_TYPE_TLS); ktls_output_eagain(tp->t_inpcb, tls); } #endif } else { /* * We are modifying a rate, wrong interface? */ if (error) *error = EINVAL; rte = NULL; } tp->t_pacing_rate = rte->rate; *error = 0; return (rte); } const struct tcp_hwrate_limit_table * tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, struct tcpcb *tp, struct ifnet *ifp, uint64_t bytes_per_sec, int flags, int *error) { const struct tcp_hwrate_limit_table *nrte; const struct tcp_rate_set *rs; #ifdef KERN_TLS struct ktls_session *tls = NULL; #endif int is_indirect = 0; int err; INP_WLOCK_ASSERT(tp->t_inpcb); if (crte == NULL) { /* Wrong interface */ if (error) *error = EINVAL; return (NULL); } #ifdef KERN_TLS if (tp->t_inpcb->inp_socket->so_snd.sb_flags & SB_TLS_IFNET) { tls = tp->t_inpcb->inp_socket->so_snd.sb_tls_info; MPASS(tls->mode == TCP_TLS_MODE_IFNET); if (tls->snd_tag != NULL && tls->snd_tag->type != IF_SND_TAG_TYPE_TLS_RATE_LIMIT) { /* * NIC probably doesn't support ratelimit TLS * tags if it didn't allocate one when an * existing rate was present, so ignore. */ if (error) *error = EOPNOTSUPP; return (NULL); } } #endif if (tp->t_inpcb->inp_snd_tag == NULL) { /* Wrong interface */ if (error) *error = EINVAL; return (NULL); } rs = crte->ptbl; if ((rs->rs_flags & RS_IS_DEAD) || (crte->flags & HDWRPACE_IFPDEPARTED)) { /* Release the rate, and try anew */ re_rate: tcp_rel_pacing_rate(crte, tp); nrte = tcp_set_pacing_rate(tp, ifp, bytes_per_sec, flags, error); return (nrte); } if ((rs->rs_flags & RT_IS_INDIRECT ) == RT_IS_INDIRECT) is_indirect = 1; else is_indirect = 0; if ((is_indirect == 0) && ((ifp != rs->rs_ifp) || (ifp->if_dunit != rs->rs_if_dunit))) { /* * Something changed, the user is not pointing to the same * ifp? Maybe a route updated on this guy? */ goto re_rate; } else if (is_indirect) { /* * For indirect we have to dig in and find the real interface. */ struct ifnet *rifp; rifp = rt_find_real_interface(ifp, tp->t_inpcb, error); if (rifp == NULL) { /* Can't find it? */ goto re_rate; } if ((rifp != rs->rs_ifp) || (ifp->if_dunit != rs->rs_if_dunit)) { goto re_rate; } } nrte = tcp_find_suitable_rate(rs, bytes_per_sec, flags); if (nrte == crte) { /* No change */ if (error) *error = 0; return (crte); } if (nrte == NULL) { /* Release the old rate */ tcp_rel_pacing_rate(crte, tp); return (NULL); } /* Change rates to our new entry */ #ifdef KERN_TLS if (tls != NULL) err = ktls_modify_txrtlmt(tls, nrte->rate); else #endif err = in_pcbmodify_txrtlmt(tp->t_inpcb, nrte->rate); if (err) { if (error) *error = err; return (NULL); } if (error) *error = 0; tp->t_pacing_rate = nrte->rate; return (nrte); } void tcp_rel_pacing_rate(const struct tcp_hwrate_limit_table *crte, struct tcpcb *tp) { const struct tcp_rate_set *crs; struct tcp_rate_set *rs; uint64_t pre; INP_WLOCK_ASSERT(tp->t_inpcb); tp->t_pacing_rate = -1; crs = crte->ptbl; /* * Now we must break the const * in order to release our refcount. */ rs = __DECONST(struct tcp_rate_set *, crs); pre = atomic_fetchadd_64(&rs->rs_flows_using, -1); if (pre == 1) { mtx_lock(&rs_mtx); /* * Is it dead? */ if (rs->rs_flags & RS_IS_DEAD) rs_defer_destroy(rs); mtx_unlock(&rs_mtx); } /* * XXX: If this connection is using ifnet TLS, should we * switch it to using an unlimited rate, or perhaps use * ktls_output_eagain() to reset the send tag to a plain * TLS tag? */ in_pcbdetach_txrtlmt(tp->t_inpcb); } #define ONE_POINT_TWO_MEG 150000 /* 1.2 megabits in bytes */ #define ONE_HUNDRED_MBPS 12500000 /* 100Mbps in bytes per second */ #define FIVE_HUNDRED_MBPS 62500000 /* 500Mbps in bytes per second */ #define MAX_MSS_SENT 43 /* 43 mss = 43 x 1500 = 64,500 bytes */ uint32_t tcp_get_pacing_burst_size (uint64_t bw, uint32_t segsiz, int can_use_1mss, const struct tcp_hwrate_limit_table *te, int *err) { /* * We use the google formula to calculate the * TSO size. I.E. * bw < 24Meg * tso = 2mss * else * tso = min(bw/1000, 64k) * * Note for these calculations we ignore the * packet overhead (enet hdr, ip hdr and tcp hdr). */ uint64_t lentim, res, bytes; uint32_t new_tso, min_tso_segs; bytes = bw / 1000; if (bytes > (64 * 1000)) bytes = 64 * 1000; /* Round up */ new_tso = (bytes + segsiz - 1) / segsiz; if (can_use_1mss && (bw < ONE_POINT_TWO_MEG)) min_tso_segs = 1; else min_tso_segs = 2; if (new_tso < min_tso_segs) new_tso = min_tso_segs; if (new_tso > MAX_MSS_SENT) new_tso = MAX_MSS_SENT; new_tso *= segsiz; /* * If we are not doing hardware pacing * then we are done. */ if (te == NULL) { if (err) *err = 0; return(new_tso); } /* * For hardware pacing we look at the * rate you are sending at and compare * that to the rate you have in hardware. * * If the hardware rate is slower than your * software rate then you are in error and * we will build a queue in our hardware whic * is probably not desired, in such a case * just return the non-hardware TSO size. * * If the rate in hardware is faster (which * it should be) then look at how long it * takes to send one ethernet segment size at * your b/w and compare that to the time it * takes to send at the rate you had selected. * * If your time is greater (which we hope it is) * we get the delta between the two, and then * divide that into your pacing time. This tells * us how many MSS you can send down at once (rounded up). * * Note we also double this value if the b/w is over * 100Mbps. If its over 500meg we just set you to the * max (43 segments). */ if (te->rate > FIVE_HUNDRED_MBPS) return (segsiz * MAX_MSS_SENT); if (te->rate == bw) { /* We are pacing at exactly the hdwr rate */ return (segsiz * MAX_MSS_SENT); } lentim = ETHERNET_SEGMENT_SIZE * USECS_IN_SECOND; res = lentim / bw; if (res > te->time_between) { uint32_t delta, segs; delta = res - te->time_between; segs = (res + delta - 1)/delta; if (te->rate > ONE_HUNDRED_MBPS) segs *= 2; if (segs < min_tso_segs) segs = min_tso_segs; if (segs > MAX_MSS_SENT) segs = MAX_MSS_SENT; segs *= segsiz; if (err) *err = 0; if (segs < new_tso) { /* unexpected ? */ return(new_tso); } else { return (segs); } } else { /* * Your time is smaller which means * we will grow a queue on our * hardware. Send back the non-hardware * rate. */ if (err) *err = -1; return (new_tso); } } static eventhandler_tag rl_ifnet_departs; static eventhandler_tag rl_ifnet_arrives; static eventhandler_tag rl_shutdown_start; static void tcp_rs_init(void *st __unused) { CK_LIST_INIT(&int_rs); rs_number_alive = 0; rs_number_dead = 0; mtx_init(&rs_mtx, "tcp_rs_mtx", "rsmtx", MTX_DEF); rl_ifnet_departs = EVENTHANDLER_REGISTER(ifnet_departure_event, tcp_rl_ifnet_departure, NULL, EVENTHANDLER_PRI_ANY); rl_ifnet_arrives = EVENTHANDLER_REGISTER(ifnet_link_event, tcp_rl_ifnet_link, NULL, EVENTHANDLER_PRI_ANY); rl_shutdown_start = EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_rl_shutdown, NULL, SHUTDOWN_PRI_FIRST); printf("TCP_ratelimit: Is now initialized\n"); } SYSINIT(tcp_rl_init, SI_SUB_SMP + 1, SI_ORDER_ANY, tcp_rs_init, NULL); #endif