Index: head/sys/netinet/tcp_subr.c =================================================================== --- head/sys/netinet/tcp_subr.c (revision 356968) +++ head/sys/netinet/tcp_subr.c (revision 356969) @@ -1,3457 +1,3458 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include "opt_tcpdebug.h" #include #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef TCP_HHOOK #include #endif #ifdef KERN_TLS #include #endif #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #include #include #endif #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include #include VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; #ifdef INET6 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; #endif #ifdef NETFLIX_EXP_DETECTION /* Sack attack detection thresholds and such */ SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, CTLFLAG_RW, 0, "Sack Attack detection thresholds"); int32_t tcp_force_detection = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, CTLFLAG_RW, &tcp_force_detection, 0, "Do we force detection even if the INP has it off?"); int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, CTLFLAG_RW, &tcp_sack_to_ack_thresh, 700, "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, CTLFLAG_RW, &tcp_sack_to_move_thresh, 600, "Percentage of sack moves we must see above (10.1 percent is 101)"); int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, CTLFLAG_RW, &tcp_restoral_thresh, 550, "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); int32_t tcp_sad_decay_val = 800; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, CTLFLAG_RW, &tcp_sad_decay_val, 800, "The decay percentage (10.1 percent equals 101 )"); int32_t tcp_map_minimum = 500; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, CTLFLAG_RW, &tcp_map_minimum, 500, "Number of Map enteries before we start detection"); int32_t tcp_attack_on_turns_on_logging = 0; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, CTLFLAG_RW, &tcp_attack_on_turns_on_logging, 0, "When we have a positive hit on attack, do we turn on logging?"); int32_t tcp_sad_pacing_interval = 2000; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, CTLFLAG_RW, &tcp_sad_pacing_interval, 2000, "What is the minimum pacing interval for a classified attacker?"); int32_t tcp_sad_low_pps = 100; SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, CTLFLAG_RW, &tcp_sad_low_pps, 100, "What is the input pps that below which we do not decay?"); #endif struct rwlock tcp_function_lock; static int sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", "Default TCP Maximum Segment Size"); #ifdef INET6 static int sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_v6mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_v6mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", "Default TCP Maximum Segment Size for IPv6"); #endif /* INET6 */ /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds * of packets instead of one. The effect scales with the available * bandwidth and quickly saturates the CPU and network interface * with packet generation and sending. Set to zero to disable MINMSS * checking. This setting prevents us from sending too small packets. */ VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_minmss), 0, "Minimum TCP Maximum Segment Size"); VNET_DEFINE(int, tcp_do_rfc1323) = 1; SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc1323), 0, "Enable rfc1323 (high performance TCP) extensions"); VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ts_offset_per_conn), 0, "Initialize TCP timestamps per connection instead of per host pair"); static int tcp_log_debug = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); static int tcp_tcbhashsize; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); static int do_tcpdrain = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; #define V_icmp_may_rst VNET(icmp_may_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_may_rst), 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_isn_reseed_interval), 0, "Seconds between reseeding of ISN secret"); static int tcp_soreceive_stream; SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); VNET_DEFINE(uma_zone_t, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ static int sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) { int error; uint32_t new; new = V_tcp_map_entries_limit; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { /* only allow "0" and value > minimum */ if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) error = EINVAL; else V_tcp_map_entries_limit = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(tcp_map_entries_limit), 0, &sysctl_net_inet_tcp_map_limit_check, "IU", "Total sendmap entries limit"); VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_map_split_limit), 0, "Total sendmap split entries limit"); #ifdef TCP_HHOOK VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); #endif #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); #define V_ts_offset_secret VNET(ts_offset_secret) static int tcp_default_fb_init(struct tcpcb *tp); static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); static int tcp_default_handoff_ok(struct tcpcb *tp); static struct inpcb *tcp_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); static void tcp_mtudisc(struct inpcb *, int); static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr); static struct tcp_function_block tcp_def_funcblk = { .tfb_tcp_block_name = "freebsd", .tfb_tcp_output = tcp_output, .tfb_tcp_do_segment = tcp_do_segment, .tfb_tcp_ctloutput = tcp_default_ctloutput, .tfb_tcp_handoff_ok = tcp_default_handoff_ok, .tfb_tcp_fb_init = tcp_default_fb_init, .tfb_tcp_fb_fini = tcp_default_fb_fini, }; static int tcp_fb_cnt = 0; struct tcp_funchead t_functions; static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; static struct tcp_function_block * find_tcp_functions_locked(struct tcp_function_set *fs) { struct tcp_function *f; struct tcp_function_block *blk=NULL; TAILQ_FOREACH(f, &t_functions, tf_next) { if (strcmp(f->tf_name, fs->function_set_name) == 0) { blk = f->tf_fb; break; } } return(blk); } static struct tcp_function_block * find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) { struct tcp_function_block *rblk=NULL; struct tcp_function *f; TAILQ_FOREACH(f, &t_functions, tf_next) { if (f->tf_fb == blk) { rblk = blk; if (s) { *s = f; } break; } } return (rblk); } struct tcp_function_block * find_and_ref_tcp_functions(struct tcp_function_set *fs) { struct tcp_function_block *blk; rw_rlock(&tcp_function_lock); blk = find_tcp_functions_locked(fs); if (blk) refcount_acquire(&blk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(blk); } struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *blk) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = find_tcp_fb_locked(blk, NULL); if (rblk) refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(rblk); } static struct tcp_function_block * find_and_ref_tcp_default_fb(void) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = tcp_func_set_ptr; refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return (rblk); } void tcp_switch_back_to_default(struct tcpcb *tp) { struct tcp_function_block *tfb; KASSERT(tp->t_fb != &tcp_def_funcblk, ("%s: called by the built-in default stack", __func__)); /* * Release the old stack. This function will either find a new one * or panic. */ if (tp->t_fb->tfb_tcp_fb_fini != NULL) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); /* * Now, we'll find a new function block to use. * Start by trying the current user-selected * default, unless this stack is the user-selected * default. */ tfb = find_and_ref_tcp_default_fb(); if (tfb == tp->t_fb) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Does the stack accept this connection? */ if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { refcount_release(&tfb->tfb_refcnt); tfb = NULL; } /* Try to use that stack. */ if (tfb != NULL) { /* Initialize the new stack. If it succeeds, we are done. */ tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init == NULL || (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) return; /* * Initialization failed. Release the reference count on * the stack. */ refcount_release(&tfb->tfb_refcnt); } /* * If that wasn't feasible, use the built-in default * stack which is not allowed to reject anyone. */ tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); if (tfb == NULL) { /* there always should be a default */ panic("Can't refer to tcp_def_funcblk"); } if (tfb->tfb_tcp_handoff_ok != NULL) { if ((*tfb->tfb_tcp_handoff_ok) (tp)) { /* The default stack cannot say no */ panic("Default stack rejects a new session?"); } } tp->t_fb = tfb; if (tp->t_fb->tfb_tcp_fb_init != NULL && (*tp->t_fb->tfb_tcp_fb_init)(tp)) { /* The default stack cannot fail */ panic("Default stack initialization failed"); } } static int sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) { int error=ENOENT; struct tcp_function_set fs; struct tcp_function_block *blk; memset(&fs, 0, sizeof(fs)); rw_rlock(&tcp_function_lock); blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); if (blk) { /* Found him */ strcpy(fs.function_set_name, blk->tfb_tcp_block_name); fs.pcbcnt = blk->tfb_refcnt; } rw_runlock(&tcp_function_lock); error = sysctl_handle_string(oidp, fs.function_set_name, sizeof(fs.function_set_name), req); /* Check for error or no change */ if (error != 0 || req->newptr == NULL) return(error); rw_wlock(&tcp_function_lock); blk = find_tcp_functions_locked(&fs); if ((blk == NULL) || (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { error = ENOENT; goto done; } tcp_func_set_ptr = blk; done: rw_wunlock(&tcp_function_lock); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, CTLTYPE_STRING | CTLFLAG_RW, NULL, 0, sysctl_net_inet_default_tcp_functions, "A", "Set/get the default TCP functions"); static int sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) { int error, cnt, linesz; struct tcp_function *f; char *buffer, *cp; size_t bufsz, outsz; bool alias; cnt = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { cnt++; } rw_runlock(&tcp_function_lock); bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; buffer = malloc(bufsz, M_TEMP, M_WAITOK); error = 0; cp = buffer; linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', "Alias", "PCB count"); cp += linesz; bufsz -= linesz; outsz = linesz; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", f->tf_fb->tfb_tcp_block_name, (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', alias ? f->tf_name : "-", f->tf_fb->tfb_refcnt); if (linesz >= bufsz) { error = EOVERFLOW; break; } cp += linesz; bufsz -= linesz; outsz += linesz; } rw_runlock(&tcp_function_lock); if (error == 0) error = sysctl_handle_string(oidp, buffer, outsz + 1, req); free(buffer, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, CTLTYPE_STRING|CTLFLAG_RD, NULL, 0, sysctl_net_inet_list_available, "A", "list available TCP Function sets"); /* * Exports one (struct tcp_function_info) for each alias/name. */ static int sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) { int cnt, error; struct tcp_function *f; struct tcp_function_info tfi; /* * We don't allow writes. */ if (req->newptr != NULL) return (EINVAL); /* * Wire the old buffer so we can directly copy the functions to * user space without dropping the lock. */ if (req->oldptr != NULL) { error = sysctl_wire_old_buffer(req, 0); if (error) return (error); } /* * Walk the list and copy out matching entries. If INVARIANTS * is compiled in, also walk the list to verify the length of * the list matches what we have recorded. */ rw_rlock(&tcp_function_lock); cnt = 0; #ifndef INVARIANTS if (req->oldptr == NULL) { cnt = tcp_fb_cnt; goto skip_loop; } #endif TAILQ_FOREACH(f, &t_functions, tf_next) { #ifdef INVARIANTS cnt++; #endif if (req->oldptr != NULL) { bzero(&tfi, sizeof(tfi)); tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; tfi.tfi_id = f->tf_fb->tfb_id; (void)strlcpy(tfi.tfi_alias, f->tf_name, sizeof(tfi.tfi_alias)); (void)strlcpy(tfi.tfi_name, f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); /* * Don't stop on error, as that is the * mechanism we use to accumulate length * information if the buffer was too short. */ } } KASSERT(cnt == tcp_fb_cnt, ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); #ifndef INVARIANTS skip_loop: #endif rw_runlock(&tcp_function_lock); if (req->oldptr == NULL) error = SYSCTL_OUT(req, NULL, (cnt + 1) * sizeof(struct tcp_function_info)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", "List TCP function block name-to-ID mappings"); /* * tfb_tcp_handoff_ok() function for the default stack. * Note that we'll basically try to take all comers. */ static int tcp_default_handoff_ok(struct tcpcb *tp) { return (0); } /* * tfb_tcp_fb_init() function for the default stack. * * This handles making sure we have appropriate timers set if you are * transitioning a socket that has some amount of setup done. * * The init() fuction from the default can *never* return non-zero i.e. * it is required to always succeed since it is the stack of last resort! */ static int tcp_default_fb_init(struct tcpcb *tp) { struct socket *so; INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, ("%s: connection %p in unexpected state %d", __func__, tp, tp->t_state)); /* * Nothing to do for ESTABLISHED or LISTEN states. And, we don't * know what to do for unexpected states (which includes TIME_WAIT). */ if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) return (0); /* * Make sure some kind of transmission timer is set if there is * outstanding data. */ so = tp->t_inpcb->inp_socket; if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || tcp_timer_active(tp, TT_PERSIST))) { /* * If the session has established and it looks like it should * be in the persist state, set the persist timer. Otherwise, * set the retransmit timer. */ if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && (int32_t)(tp->snd_nxt - tp->snd_una) < (int32_t)sbavail(&so->so_snd)) tcp_setpersist(tp); else tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); } /* All non-embryonic sessions get a keepalive timer. */ if (!tcp_timer_active(tp, TT_KEEP)) tcp_timer_activate(tp, TT_KEEP, TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : TP_KEEPINIT(tp)); return (0); } /* * tfb_tcp_fb_fini() function for the default stack. * * This changes state as necessary (or prudent) to prepare for another stack * to assume responsibility for the connection. */ static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) { INP_WLOCK_ASSERT(tp->t_inpcb); return; } /* * Target size of TCP PCB hash tables. Must be a power of two. * * Note that this can be overridden by the kernel environment * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE #define TCBHASHSIZE 0 #endif /* * XXX * Callouts should be moved into struct tcp directly. They are currently * separate because the tcpcb structure is exported to userland for sysctl * parsing purposes, which do not know about callouts. */ struct tcpcb_mem { struct tcpcb tcb; struct tcp_timer tt; struct cc_var ccv; #ifdef TCP_HHOOK struct osd osd; #endif }; VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); #define V_tcpcb_zone VNET(tcpcb_zone) MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); static struct mtx isn_mtx; #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) #define ISN_LOCK() mtx_lock(&isn_mtx) #define ISN_UNLOCK() mtx_unlock(&isn_mtx) /* * TCP initialization. */ static void tcp_zone_change(void *tag) { uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_tcpcb_zone, maxsockets); tcp_tw_zone_change(); } static int tcp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp = mem; INP_LOCK_INIT(inp, "inp", "tcpinp"); return (0); } /* * Take a value and get the next power of 2 that doesn't overflow. * Used to size the tcp_inpcb hash buckets. */ static int maketcp_hashsize(int size) { int hashsize; /* * auto tune. * get the next power of 2 higher than maxsockets. */ hashsize = 1 << fls(size); /* catch overflow, and just go one power of 2 smaller */ if (hashsize < size) { hashsize = 1 << (fls(size) - 1); } return (hashsize); } static volatile int next_tcp_stack_id = 1; /* * Register a TCP function block with the name provided in the names * array. (Note that this function does NOT automatically register * blk->tfb_tcp_block_name as a stack name. Therefore, you should * explicitly include blk->tfb_tcp_block_name in the list of names if * you wish to register the stack with that name.) * * Either all name registrations will succeed or all will fail. If * a name registration fails, the function will update the num_names * argument to point to the array index of the name that encountered * the failure. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, const char *names[], int *num_names) { struct tcp_function *n; struct tcp_function_set fs; int error, i; KASSERT(names != NULL && *num_names > 0, ("%s: Called with 0-length name list", __func__)); KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); KASSERT(rw_initialized(&tcp_function_lock), ("%s: called too early", __func__)); if ((blk->tfb_tcp_output == NULL) || (blk->tfb_tcp_do_segment == NULL) || (blk->tfb_tcp_ctloutput == NULL) || (strlen(blk->tfb_tcp_block_name) == 0)) { /* * These functions are required and you * need a name. */ *num_names = 0; return (EINVAL); } if (blk->tfb_tcp_timer_stop_all || blk->tfb_tcp_timer_activate || blk->tfb_tcp_timer_active || blk->tfb_tcp_timer_stop) { /* * If you define one timer function you * must have them all. */ if ((blk->tfb_tcp_timer_stop_all == NULL) || (blk->tfb_tcp_timer_activate == NULL) || (blk->tfb_tcp_timer_active == NULL) || (blk->tfb_tcp_timer_stop == NULL)) { *num_names = 0; return (EINVAL); } } if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { *num_names = 0; return (EINVAL); } refcount_init(&blk->tfb_refcnt, 0); blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); for (i = 0; i < *num_names; i++) { n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); if (n == NULL) { error = ENOMEM; goto cleanup; } n->tf_fb = blk; (void)strlcpy(fs.function_set_name, names[i], sizeof(fs.function_set_name)); rw_wlock(&tcp_function_lock); if (find_tcp_functions_locked(&fs) != NULL) { /* Duplicate name space not allowed */ rw_wunlock(&tcp_function_lock); free(n, M_TCPFUNCTIONS); error = EALREADY; goto cleanup; } (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); TAILQ_INSERT_TAIL(&t_functions, n, tf_next); tcp_fb_cnt++; rw_wunlock(&tcp_function_lock); } return(0); cleanup: /* * Deregister the names we just added. Because registration failed * for names[i], we don't need to deregister that name. */ *num_names = i; rw_wlock(&tcp_function_lock); while (--i >= 0) { TAILQ_FOREACH(n, &t_functions, tf_next) { if (!strncmp(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX)) { TAILQ_REMOVE(&t_functions, n, tf_next); tcp_fb_cnt--; n->tf_fb = NULL; free(n, M_TCPFUNCTIONS); break; } } } rw_wunlock(&tcp_function_lock); return (error); } /* * Register a TCP function block using the name provided in the name * argument. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, int wait) { const char *name_list[1]; int num_names, rv; num_names = 1; if (name != NULL) name_list[0] = name; else name_list[0] = blk->tfb_tcp_block_name; rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); return (rv); } /* * Register a TCP function block using the name defined in * blk->tfb_tcp_block_name. * * Returns 0 on success, or an error code on failure. */ int register_tcp_functions(struct tcp_function_block *blk, int wait) { return (register_tcp_functions_as_name(blk, NULL, wait)); } /* * Deregister all names associated with a function block. This * functionally removes the function block from use within the system. * * When called with a true quiesce argument, mark the function block * as being removed so no more stacks will use it and determine * whether the removal would succeed. * * When called with a false quiesce argument, actually attempt the * removal. * * When called with a force argument, attempt to switch all TCBs to * use the default stack instead of returning EBUSY. * * Returns 0 on success (or if the removal would succeed, or an error * code on failure. */ int deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, bool force) { struct tcp_function *f; if (blk == &tcp_def_funcblk) { /* You can't un-register the default */ return (EPERM); } rw_wlock(&tcp_function_lock); if (blk == tcp_func_set_ptr) { /* You can't free the current default */ rw_wunlock(&tcp_function_lock); return (EBUSY); } /* Mark the block so no more stacks can use it. */ blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; /* * If TCBs are still attached to the stack, attempt to switch them * to the default stack. */ if (force && blk->tfb_refcnt) { struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); rw_wunlock(&tcp_function_lock); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inp); if (inp->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inp); continue; } tp = intotcpcb(inp); if (tp == NULL || tp->t_fb != blk) { INP_WUNLOCK(inp); continue; } tcp_switch_back_to_default(tp); INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); rw_wlock(&tcp_function_lock); } if (blk->tfb_refcnt) { /* TCBs still attached. */ rw_wunlock(&tcp_function_lock); return (EBUSY); } if (quiesce) { /* Skip removal. */ rw_wunlock(&tcp_function_lock); return (0); } /* Remove any function names that map to this function block. */ while (find_tcp_fb_locked(blk, &f) != NULL) { TAILQ_REMOVE(&t_functions, f, tf_next); tcp_fb_cnt--; f->tf_fb = NULL; free(f, M_TCPFUNCTIONS); } rw_wunlock(&tcp_function_lock); return (0); } void tcp_init(void) { const char *tcbhash_tuneable; int hashsize; tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; #ifdef TCP_HHOOK if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); #endif #ifdef STATS if (tcp_stats_init()) printf("%s: WARNING: unable to initialise TCP stats\n", __func__); #endif hashsize = TCBHASHSIZE; TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); if (hashsize == 0) { /* * Auto tune the hash size based on maxsockets. * A perfect hash would have a 1:1 mapping * (hashsize = maxsockets) however it's been * suggested that O(2) average is better. */ hashsize = maketcp_hashsize(maxsockets / 4); /* * Our historical default is 512, * do not autotune lower than this. */ if (hashsize < 512) hashsize = 512; if (bootverbose && IS_DEFAULT_VNET(curvnet)) printf("%s: %s auto tuned to %d\n", __func__, tcbhash_tuneable, hashsize); } /* * We require a hashsize to be a power of two. * Previously if it was not a power of two we would just reset it * back to 512, which could be a nasty surprise if you did not notice * the error message. * Instead what we do is clip it to the closest power of two lower * than the specified hash value. */ if (!powerof2(hashsize)) { int oldhashsize = hashsize; hashsize = maketcp_hashsize(hashsize); /* prevent absurdly low value */ if (hashsize < 16) hashsize = 16; printf("%s: WARNING: TCB hash size not a power of 2, " "clipped from %d to %d.\n", __func__, oldhashsize, hashsize); } in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); /* * These have to be type stable for the benefit of the timers. */ V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_zone_set_max(V_tcpcb_zone, maxsockets); uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); tcp_tw_init(); syncache_init(); tcp_hc_init(); TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); tcp_fastopen_init(); /* Skip initialization of globals for non-default instances. */ if (!IS_DEFAULT_VNET(curvnet)) return; tcp_reass_global_init(); /* XXX virtualize those bellow? */ tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; tcp_rexmit_initial = TCPTV_RTOBASE; if (tcp_rexmit_initial < 1) tcp_rexmit_initial = 1; tcp_rexmit_min = TCPTV_MIN; if (tcp_rexmit_min < 1) tcp_rexmit_min = 1; tcp_persmin = TCPTV_PERSMIN; tcp_persmax = TCPTV_PERSMAX; tcp_rexmit_slop = TCPTV_CPU_VAR; tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; tcp_tcbhashsize = hashsize; /* Setup the tcp function block list */ TAILQ_INIT(&t_functions); rw_init(&tcp_function_lock, "tcp_func_lock"); register_tcp_functions(&tcp_def_funcblk, M_WAITOK); #ifdef TCP_BLACKBOX /* Initialize the TCP logging data. */ tcp_log_init(); #endif arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); if (tcp_soreceive_stream) { #ifdef INET tcp_usrreqs.pru_soreceive = soreceive_stream; #endif #ifdef INET6 tcp6_usrreqs.pru_soreceive = soreceive_stream; #endif /* INET6 */ } #ifdef INET6 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ if (max_protohdr < TCP_MINPROTOHDR) max_protohdr = TCP_MINPROTOHDR; if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) panic("tcp_init"); #undef TCP_MINPROTOHDR ISN_LOCK_INIT(); EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, SHUTDOWN_PRI_DEFAULT); EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, EVENTHANDLER_PRI_ANY); tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); #ifdef TCPPCAP tcp_pcap_init(); #endif } #ifdef VIMAGE static void tcp_destroy(void *unused __unused) { int n; #ifdef TCP_HHOOK int error; #endif /* * All our processes are gone, all our sockets should be cleaned * up, which means, we should be past the tcp_discardcb() calls. * Sleep to let all tcpcb timers really disappear and cleanup. */ for (;;) { INP_LIST_RLOCK(&V_tcbinfo); n = V_tcbinfo.ipi_count; INP_LIST_RUNLOCK(&V_tcbinfo); if (n == 0) break; pause("tcpdes", hz / 10); } tcp_hc_destroy(); syncache_destroy(); tcp_tw_destroy(); in_pcbinfo_destroy(&V_tcbinfo); /* tcp_discardcb() clears the sack_holes up. */ uma_zdestroy(V_sack_hole_zone); uma_zdestroy(V_tcpcb_zone); /* * Cannot free the zone until all tcpcbs are released as we attach * the allocations to them. */ tcp_fastopen_destroy(); #ifdef TCP_HHOOK error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); } error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); } #endif } VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); #endif void tcp_fini(void *xtp) { } /* * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. * tcp_template used to store this data in mbufs, but we now recopy it out * of the tcpcb each time to conserve mbufs. */ void tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) { struct tcphdr *th = (struct tcphdr *)tcp_ptr; INP_WLOCK_ASSERT(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { struct ip *ip; ip = (struct ip *)ip_ptr; ip->ip_v = IPVERSION; ip->ip_hl = 5; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = 0; ip->ip_id = 0; ip->ip_off = 0; ip->ip_ttl = inp->inp_ip_ttl; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } #endif /* INET */ th->th_sport = inp->inp_lport; th->th_dport = inp->inp_fport; th->th_seq = 0; th->th_ack = 0; th->th_x2 = 0; th->th_off = 5; th->th_flags = 0; th->th_win = 0; th->th_urp = 0; th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ } /* * Create template to be used to send tcp packets on a connection. * Allocates an mbuf and fills in a skeletal tcp/ip header. The only * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * tcpip_maketemplate(struct inpcb *inp) { struct tcptemp *t; t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); if (t == NULL) return (NULL); tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); return (t); } /* * Send a single message to the TCP at address specified by * the given TCP/IP header. If m == NULL, then we make a copy * of the tcpiphdr at th and send directly to the addressed host. * This is used to force keep alive messages out using the TCP * template for a connection. If flags are given then we send * a message back to the TCP which originated the segment th, * and discard the mbuf containing it and any other attached mbufs. * * In any case the ack and sequence number of the transmitted * segment are as specified by the parameters. * * NOTE: If m != NULL, then th must point to *inside* the mbuf. */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) { struct tcpopt to; struct inpcb *inp; struct ip *ip; struct mbuf *optm; struct tcphdr *nth; u_char *optp; #ifdef INET6 struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ int optlen, tlen, win; bool incl_opts; KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); + NET_EPOCH_ASSERT(); #ifdef INET6 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); ip6 = ipgen; #endif /* INET6 */ ip = ipgen; if (tp != NULL) { inp = tp->t_inpcb; KASSERT(inp != NULL, ("tcp control block w/o inpcb")); INP_WLOCK_ASSERT(inp); } else inp = NULL; incl_opts = false; win = 0; if (tp != NULL) { if (!(flags & TH_RST)) { win = sbspace(&inp->inp_socket->so_rcv); if (win > TCP_MAXWIN << tp->rcv_scale) win = TCP_MAXWIN << tp->rcv_scale; } if ((tp->t_flags & TF_NOOPT) == 0) incl_opts = true; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; m->m_data += max_linkhdr; #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); flags = TH_ACK; } else if (!M_WRITABLE(m)) { struct mbuf *n; /* Can't reuse 'm', allocate a new mbuf. */ n = m_gethdr(M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return; } if (!m_dup_pkthdr(n, m, M_NOWAIT)) { m_freem(m); m_freem(n); return; } n->m_data += max_linkhdr; /* m_len is set later */ #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(n, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(n, struct ip6_hdr *); xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); ip = mtod(n, struct ip *); xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); xchg(nth->th_dport, nth->th_sport, uint16_t); th = nth; m_freem(m); m = n; } else { /* * reuse the mbuf. * XXX MRT We inherit the FIB, which is lucky. */ m_freem(m->m_next); m->m_next = NULL; m->m_data = (caddr_t)ipgen; /* m_len is set later */ #ifdef INET6 if (isipv6) { xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } if (th != nth) { /* * this is usually a case when an extension header * exists between the IPv6 header and the * TCP header. */ nth->th_sport = th->th_sport; nth->th_dport = th->th_dport; } xchg(nth->th_dport, nth->th_sport, uint16_t); #undef xchg } tlen = 0; #ifdef INET6 if (isipv6) tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET tlen = sizeof (struct tcpiphdr); #endif #ifdef INVARIANTS m->m_len = 0; KASSERT(M_TRAILINGSPACE(m) >= tlen, ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", m, tlen, (long)M_TRAILINGSPACE(m))); #endif m->m_len = tlen; to.to_flags = 0; if (incl_opts) { /* Make sure we have room. */ if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { m->m_next = m_get(M_NOWAIT, MT_DATA); if (m->m_next) { optp = mtod(m->m_next, u_char *); optm = m->m_next; } else incl_opts = false; } else { optp = (u_char *) (nth + 1); optm = m; } } if (incl_opts) { /* Timestamps. */ if (tp->t_flags & TF_RCVD_TSTMP) { to.to_tsval = tcp_ts_getticks() + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* Add the options. */ tlen += optlen = tcp_addoptions(&to, optp); /* Update m_len in the correct mbuf. */ optm->m_len += optlen; } else optlen = 0; #ifdef INET6 if (isipv6) { ip6->ip6_flow = 0; ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(tlen - sizeof(*ip6)); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(tlen); ip->ip_ttl = V_ip_defttl; if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); } #endif m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef MAC if (inp != NULL) { /* * Packet is associated with a socket, so allow the * label of the response to reflect the socket label. */ INP_WLOCK_ASSERT(inp); mac_inpcb_create_mbuf(inp, m); } else { /* * Packet is not associated with a socket, so possibly * update the label in place. */ mac_netinet_tcp_reply(m); } #endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; nth->th_flags = flags; if (tp != NULL) nth->th_win = htons((u_short) (win >> tp->rcv_scale)); else nth->th_win = htons((u_short)win); nth->th_urp = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { m_freem(m); return; } } #endif m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #ifdef INET6 if (isipv6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; nth->th_sum = in6_cksum_pseudo(ip6, tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : NULL, NULL); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); } #endif /* INET */ #ifdef TCPDEBUG if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); #endif TCP_PROBE3(debug__output, tp, th, m); if (flags & TH_RST) TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); #ifdef INET6 if (isipv6) { TCP_PROBE5(send, NULL, tp, ip6, tp, nth); (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { TCP_PROBE5(send, NULL, tp, ip, tp, nth); (void)ip_output(m, NULL, NULL, 0, NULL, inp); } #endif } /* * Create a new TCP control block, making an * empty reassembly queue and hooking it to the argument * protocol control block. The `inp' parameter must have * come from the zone allocator set up in tcp_init(). */ struct tcpcb * tcp_newtcpcb(struct inpcb *inp) { struct tcpcb_mem *tm; struct tcpcb *tp; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) return (NULL); tp = &tm->tcb; /* Initialise cc_var struct for this tcpcb. */ tp->ccv = &tm->ccv; tp->ccv->type = IPPROTO_TCP; tp->ccv->ccvc.tcp = tp; rw_rlock(&tcp_function_lock); tp->t_fb = tcp_func_set_ptr; refcount_acquire(&tp->t_fb->tfb_refcnt); rw_runlock(&tcp_function_lock); /* * Use the current system default CC algorithm. */ CC_LIST_RLOCK(); KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); CC_ALGO(tp) = CC_DEFAULT(); CC_LIST_RUNLOCK(); if (CC_ALGO(tp)->cb_init != NULL) if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #ifdef TCP_HHOOK tp->osd = &tm->osd; if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #endif #ifdef VIMAGE tp->t_vnet = inp->inp_vnet; #endif tp->t_timers = &tm->tt; TAILQ_INIT(&tp->t_segq); tp->t_maxseg = #ifdef INET6 isipv6 ? V_tcp_v6mssdflt : #endif /* INET6 */ V_tcp_mssdflt; /* Set up our timeouts. */ callout_init(&tp->t_timers->tt_rexmt, 1); callout_init(&tp->t_timers->tt_persist, 1); callout_init(&tp->t_timers->tt_keep, 1); callout_init(&tp->t_timers->tt_2msl, 1); callout_init(&tp->t_timers->tt_delack, 1); if (V_tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); if (V_tcp_do_sack) tp->t_flags |= TF_SACK_PERMIT; TAILQ_INIT(&tp->snd_holes); /* * The tcpcb will hold a reference on its inpcb until tcp_discardcb() * is called. */ in_pcbref(inp); /* Reference for tcpcb */ tp->t_inpcb = inp; /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_rexmit_min; tp->t_rxtcur = tcp_rexmit_initial; tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = ticks; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ inp->inp_ip_ttl = V_ip_defttl; inp->inp_ppcb = tp; #ifdef TCPPCAP /* * Init the TCP PCAP queues. */ tcp_pcap_tcpcb_init(tp); #endif #ifdef TCP_BLACKBOX /* Initialize the per-TCPCB log data. */ tcp_log_tcpcbinit(tp); #endif if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } #ifdef STATS if (V_tcp_perconn_stats_enable == 1) tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); #endif return (tp); /* XXX */ } /* * Switch the congestion control algorithm back to NewReno for any active * control blocks using an algorithm which is about to go away. * This ensures the CC framework can allow the unload to proceed without leaving * any dangling pointers which would trigger a panic. * Returning non-zero would inform the CC framework that something went wrong * and it would be unsafe to allow the unload to proceed. However, there is no * way for this to occur with this implementation so we always return zero. */ int tcp_ccalgounload(struct cc_algo *unload_algo) { struct cc_algo *tmpalgo; struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); /* * Check all active control blocks across all network stacks and change * any that are using "unload_algo" back to NewReno. If "unload_algo" * requires cleanup code to be run, call it. */ VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); /* * New connections already part way through being initialised * with the CC algo we're removing will not race with this code * because the INP_INFO_WLOCK is held during initialisation. We * therefore don't enter the loop below until the connection * list has stabilised. */ CK_LIST_FOREACH(inp, &V_tcb, inp_list) { INP_WLOCK(inp); /* Important to skip tcptw structs. */ if (!(inp->inp_flags & INP_TIMEWAIT) && (tp = intotcpcb(inp)) != NULL) { /* * By holding INP_WLOCK here, we are assured * that the connection is not currently * executing inside the CC module's functions * i.e. it is safe to make the switch back to * NewReno. */ if (CC_ALGO(tp) == unload_algo) { tmpalgo = CC_ALGO(tp); if (tmpalgo->cb_destroy != NULL) tmpalgo->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; /* * NewReno may allocate memory on * demand for certain stateful * configuration as needed, but is * coded to never fail on memory * allocation failure so it is a safe * fallback. */ CC_ALGO(tp) = &newreno_cc_algo; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } /* * Drop a TCP connection, reporting * the specified error. If connection is synchronized, * then send a RST to peer. */ struct tcpcb * tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); if (TCPS_HAVERCVDSYN(tp->t_state)) { tcp_state_change(tp, TCPS_CLOSED); (void) tp->t_fb->tfb_tcp_output(tp); TCPSTAT_INC(tcps_drops); } else TCPSTAT_INC(tcps_conndrops); if (errno == ETIMEDOUT && tp->t_softerror) errno = tp->t_softerror; so->so_error = errno; return (tcp_close(tp)); } void tcp_discardcb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ int released __unused; INP_WLOCK_ASSERT(inp); /* * Make sure that all of our timers are stopped before we delete the * PCB. * * If stopping a timer fails, we schedule a discard function in same * callout, and the last discard function called will take care of * deleting the tcpcb. */ tp->t_timers->tt_draincnt = 0; tcp_timer_stop(tp, TT_REXMT); tcp_timer_stop(tp, TT_PERSIST); tcp_timer_stop(tp, TT_KEEP); tcp_timer_stop(tp, TT_2MSL); tcp_timer_stop(tp, TT_DELACK); if (tp->t_fb->tfb_tcp_timer_stop_all) { /* * Call the stop-all function of the methods, * this function should call the tcp_timer_stop() * method with each of the function specific timeouts. * That stop will be called via the tfb_tcp_timer_stop() * which should use the async drain function of the * callout system (see tcp_var.h). */ tp->t_fb->tfb_tcp_timer_stop_all(tp); } /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. * 'Enough' is arbitrarily defined as 4 rtt samples. * 4 samples is enough for the srtt filter to converge * to within enough % of the correct value; fewer samples * and we could save a bogus rtt. The danger is not high * as tcp quickly recovers from everything. * XXX: Works very well but needs some more statistics! */ if (tp->t_rttupdated >= 4) { struct hc_metrics_lite metrics; uint32_t ssthresh; bzero(&metrics, sizeof(metrics)); /* * Update the ssthresh always when the conditions below * are satisfied. This gives us better new start value * for the congestion avoidance for new connections. * ssthresh is only set if packet loss occurred on a session. * * XXXRW: 'so' may be NULL here, and/or socket buffer may be * being torn down. Ideally this code would not use 'so'. */ ssthresh = tp->snd_ssthresh; if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { /* * convert the limit from user data bytes to * packets then to packet data bytes. */ ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; if (ssthresh < 2) ssthresh = 2; ssthresh *= (tp->t_maxseg + #ifdef INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : #endif sizeof (struct tcpiphdr) #ifdef INET6 ) #endif ); } else ssthresh = 0; metrics.rmx_ssthresh = ssthresh; metrics.rmx_rtt = tp->t_srtt; metrics.rmx_rttvar = tp->t_rttvar; metrics.rmx_cwnd = tp->snd_cwnd; metrics.rmx_sendpipe = 0; metrics.rmx_recvpipe = 0; tcp_hc_update(&inp->inp_inc, &metrics); } /* free the reassembly queue, if any */ tcp_reass_flush(tp); #ifdef TCP_OFFLOAD /* Disconnect offload device, if any. */ if (tp->t_flags & TF_TOE) tcp_offload_detach(tp); #endif tcp_free_sackholes(tp); #ifdef TCPPCAP /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tp->t_inpkts)); tcp_pcap_drain(&(tp->t_outpkts)); #endif /* Allow the CC algorithm to clean up after itself. */ if (CC_ALGO(tp)->cb_destroy != NULL) CC_ALGO(tp)->cb_destroy(tp->ccv); CC_DATA(tp) = NULL; #ifdef TCP_HHOOK khelp_destroy_osd(tp->osd); #endif #ifdef STATS stats_blob_destroy(tp->t_stats); #endif CC_ALGO(tp) = NULL; inp->inp_ppcb = NULL; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); released = in_pcbrele_wlocked(inp); KASSERT(!released, ("%s: inp %p should not have been released " "here", __func__, inp)); } } void tcp_timer_discard(void *ptp) { struct inpcb *inp; struct tcpcb *tp; struct epoch_tracker et; tp = (struct tcpcb *)ptp; CURVNET_SET(tp->t_vnet); NET_EPOCH_ENTER(et); inp = tp->t_inpcb; KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp)); INP_WLOCK(inp); KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, ("%s: tcpcb has to be stopped here", __func__)); tp->t_timers->tt_draincnt--; if (tp->t_timers->tt_draincnt == 0) { /* We own the last reference on this tcpcb, let's free it. */ #ifdef TCP_BLACKBOX tcp_log_tcpcbfini(tp); #endif TCPSTATES_DEC(tp->t_state); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); if (in_pcbrele_wlocked(inp)) { NET_EPOCH_EXIT(et); CURVNET_RESTORE(); return; } } INP_WUNLOCK(inp); NET_EPOCH_EXIT(et); CURVNET_RESTORE(); } /* * Attempt to close a TCP control block, marking it as dropped, and freeing * the socket if we hold the only reference. */ struct tcpcb * tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_state == TCPS_LISTEN) tcp_offload_listen_stop(tp); #endif /* * This releases the TFO pending counter resource for TFO listen * sockets as well as passively-created TFO sockets that transition * from SYN_RECEIVED to CLOSED. */ if (tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } in_pcbdrop(inp); TCPSTAT_INC(tcps_closed); if (tp->t_state != TCPS_CLOSED) tcp_state_change(tp, TCPS_CLOSED); KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); so = inp->inp_socket; soisdisconnected(so); if (inp->inp_flags & INP_SOCKREF) { KASSERT(so->so_state & SS_PROTOREF, ("tcp_close: !SS_PROTOREF")); inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); SOCK_LOCK(so); so->so_state &= ~SS_PROTOREF; sofree(so); return (NULL); } return (tp); } void tcp_drain(void) { VNET_ITERATOR_DECL(vnet_iter); if (!do_tcpdrain) return; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct inpcb *inpb; struct tcpcb *tcpb; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... * XXX: The "Net/3" implementation doesn't imply that the TCP * reassembly queue should be flushed, but in a situation * where we're really low on mbufs, this is potentially * useful. */ INP_INFO_WLOCK(&V_tcbinfo); CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { INP_WLOCK(inpb); if (inpb->inp_flags & INP_TIMEWAIT) { INP_WUNLOCK(inpb); continue; } if ((tcpb = intotcpcb(inpb)) != NULL) { tcp_reass_flush(tcpb); tcp_clean_sackreport(tcpb); #ifdef TCP_BLACKBOX tcp_log_drain(tcpb); #endif #ifdef TCPPCAP if (tcp_pcap_aggressive_free) { /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tcpb->t_inpkts)); tcp_pcap_drain(&(tcpb->t_outpkts)); } #endif } INP_WUNLOCK(inpb); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Notify a tcp user of an asynchronous error; * store error as soft error, but wake up user * (for now, won't do anything until can select for soft error). * * Do not wake up user since there currently is no mechanism for * reporting soft errors (yet - a kqueue filter may be added). */ static struct inpcb * tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, * and receives a second error, give up now. This is better * than waiting a long time to establish a connection that * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && (error == EHOSTUNREACH || error == ENETUNREACH || error == EHOSTDOWN)) { if (inp->inp_route.ro_rt) { RTFREE(inp->inp_route.ro_rt); inp->inp_route.ro_rt = (struct rtentry *)NULL; } return (inp); } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) { tp = tcp_drop(tp, error); if (tp != NULL) return (inp); else return (NULL); } else { tp->t_softerror = error; return (inp); } #if 0 wakeup( &so->so_timeo); sorwakeup(so); sowwakeup(so); #endif } static int tcp_pcblist(SYSCTL_HANDLER_ARGS) { struct epoch_tracker et; struct inpcb *inp; struct xinpgen xig; int error; if (req->newptr != NULL) return (EPERM); if (req->oldptr == NULL) { int n; n = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); return (0); } if ((error = sysctl_wire_old_buffer(req, 0)) != 0) return (error); bzero(&xig, sizeof(xig)); xig.xig_len = sizeof xig; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); error = syncache_pcblist(req); if (error) return (error); NET_EPOCH_ENTER(et); for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); inp != NULL; inp = CK_LIST_NEXT(inp, inp_list)) { INP_RLOCK(inp); if (inp->inp_gencnt <= xig.xig_gen) { int crerr; /* * XXX: This use of cr_cansee(), introduced with * TCP state changes, is not quite right, but for * now, better than nothing. */ if (inp->inp_flags & INP_TIMEWAIT) { if (intotw(inp) != NULL) crerr = cr_cansee(req->td->td_ucred, intotw(inp)->tw_cred); else crerr = EINVAL; /* Skip this inp. */ } else crerr = cr_canseeinpcb(req->td->td_ucred, inp); if (crerr == 0) { struct xtcpcb xt; tcp_inptoxtp(inp, &xt); INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xt, sizeof xt); if (error) break; else continue; } } INP_RUNLOCK(inp); } NET_EPOCH_EXIT(et); if (!error) { /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_tcbinfo.ipi_count + counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); error = SYSCTL_OUT(req, &xig, sizeof xig); } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); #ifdef INET static int tcp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct epoch_tracker et; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); NET_EPOCH_ENTER(et); inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); #endif /* INET */ #ifdef INET6 static int tcp6_getcred(SYSCTL_HANDLER_ARGS) { struct epoch_tracker et; struct xucred xuc; struct sockaddr_in6 addrs[2]; struct inpcb *inp; int error; #ifdef INET int mapped = 0; #endif error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { return (error); } if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) mapped = 1; else #endif return (EINVAL); } NET_EPOCH_ENTER(et); #ifdef INET if (mapped == 1) inp = in_pcblookup(&V_tcbinfo, *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], addrs[1].sin6_port, *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); else #endif inp = in6_pcblookup(&V_tcbinfo, &addrs[1].sin6_addr, addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); NET_EPOCH_EXIT(et); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); #endif /* INET6 */ #ifdef INET void tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { struct ip *ip = vip; struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct icmp *icp; struct in_conninfo inc; tcp_seq icmp_tcp_seq; int mtu; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip == NULL) { in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); return; } icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } icmp_tcp_seq = th->th_seq; if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohs(icp->icmp_nextmtu); /* * If no alternative MTU was * proposed, try the next smaller * one. */ if (!mtu) mtu = ip_next_mtu( ntohs(ip->ip_len), 1); if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) mtu = V_tcp_minmss + sizeof(struct tcpiphdr); /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof(struct tcpiphdr)) { bzero(&inc, sizeof(inc)); inc.inc_faddr = faddr; inc.inc_fibnum = inp->inp_inc.inc_fibnum; tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); } } else inp = (*notify)(inp, inetctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fport = th->th_dport; inc.inc_lport = th->th_sport; inc.inc_faddr = faddr; inc.inc_laddr = ip->ip_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); } #endif /* INET */ #ifdef INET6 void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { struct in6_addr *dst; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; struct inpcb *inp; struct tcpcb *tp; struct icmp6_hdr *icmp6; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; struct in_conninfo inc; struct tcp_ports { uint16_t th_sport; uint16_t th_dport; } t_ports; tcp_seq icmp_tcp_seq; unsigned int mtu; unsigned int off; if (sa->sa_family != AF_INET6 || sa->sa_len != sizeof(struct sockaddr_in6)) return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; icmp6 = ip6cp->ip6c_icmp6; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; dst = ip6cp->ip6c_finaldst; } else { m = NULL; ip6 = NULL; off = 0; /* fool gcc */ sa6_src = &sa6_any; dst = NULL; } if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) notify = tcp_drop_syn_sent; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip6 = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) return; if (ip6 == NULL) { in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); return; } /* Check if we can safely get the ports from the tcp hdr */ if (m == NULL || (m->m_pkthdr.len < (int32_t) (off + sizeof(struct tcp_ports)))) { return; } bzero(&t_ports, sizeof(struct tcp_ports)); m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL && PRC_IS_REDIRECT(cmd)) { /* signal EHOSTDOWN, as it flushes the cached route */ inp = (*notify)(inp, EHOSTDOWN); goto out; } off += sizeof(struct tcp_ports); if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { goto out; } m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { tp = intotcpcb(inp); if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohl(icmp6->icmp6_mtu); /* * If no alternative MTU was * proposed, or the proposed * MTU was too small, set to * the min. */ if (mtu < IPV6_MMTU) mtu = IPV6_MMTU - 8; bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc6_faddr = *dst; if (in6_setscope(&inc.inc6_faddr, m->m_pkthdr.rcvif, NULL)) goto out; /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxseg + sizeof (struct tcphdr) + sizeof (struct ip6_hdr)) { tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); ICMP6STAT_INC(icp6s_pmtuchg); } } else inp = (*notify)(inp, inet6ctlerrmap[cmd]); } } } else { bzero(&inc, sizeof(inc)); inc.inc_fibnum = M_GETFIB(m); inc.inc_flags |= INC_ISIPV6; inc.inc_fport = t_ports.th_dport; inc.inc_lport = t_ports.th_sport; inc.inc6_faddr = *dst; inc.inc6_laddr = ip6->ip6_src; syncache_unreach(&inc, icmp_tcp_seq); } out: if (inp != NULL) INP_WUNLOCK(inp); } #endif /* INET6 */ static uint32_t tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) { SIPHASH_CTX ctx; uint32_t hash[2]; KASSERT(len >= SIPHASH_KEY_LENGTH, ("%s: keylen %u too short ", __func__, len)); SipHash24_Init(&ctx); SipHash_SetKey(&ctx, (uint8_t *)key); SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); break; #endif } SipHash_Final((uint8_t *)hash, &ctx); return (hash[0] ^ hash[1]); } uint32_t tcp_new_ts_offset(struct in_conninfo *inc) { struct in_conninfo inc_store, *local_inc; if (!V_tcp_ts_offset_per_conn) { memcpy(&inc_store, inc, sizeof(struct in_conninfo)); inc_store.inc_lport = 0; inc_store.inc_fport = 0; local_inc = &inc_store; } else { local_inc = inc; } return (tcp_keyed_hash(local_inc, V_ts_offset_secret, sizeof(V_ts_offset_secret))); } /* * Following is where TCP initial sequence number generation occurs. * * There are two places where we must use initial sequence numbers: * 1. In SYN-ACK packets. * 2. In SYN packets. * * All ISNs for SYN-ACK packets are generated by the syncache. See * tcp_syncache.c for details. * * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling * depends on this property. In addition, these ISNs should be * unguessable so as to prevent connection hijacking. To satisfy * the requirements of this situation, the algorithm outlined in * RFC 1948 is used, with only small modifications. * * Implementation details: * * Time is based off the system timer, and is corrected so that it * increases by one megabyte per second. This allows for proper * recycling on high speed LANs while still leaving over an hour * before rollover. * * As reading the *exact* system time is too expensive to be done * whenever setting up a TCP connection, we increment the time * offset in two ways. First, a small random positive increment * is added to isn_offset for each connection that is set up. * Second, the function tcp_isn_tick fires once per clock tick * and increments isn_offset as necessary so that sequence numbers * are incremented at approximately ISN_BYTES_PER_SECOND. The * random positive increments serve only to ensure that the same * exact sequence number is never sent out twice (as could otherwise * happen when a port is recycled in less than the system tick * interval.) * * net.inet.tcp.isn_reseed_interval controls the number of seconds * between seeding of isn_secret. This is normally set to zero, * as reseeding should not be necessary. * * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, * isn_offset_old, and isn_ctx is performed using the ISN lock. In * general, this means holding an exclusive (write) lock. */ #define ISN_BYTES_PER_SECOND 1048576 #define ISN_STATIC_INCREMENT 4096 #define ISN_RANDOM_INCREMENT (4096 - 1) #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); VNET_DEFINE_STATIC(int, isn_last); VNET_DEFINE_STATIC(int, isn_last_reseed); VNET_DEFINE_STATIC(u_int32_t, isn_offset); VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); #define V_isn_secret VNET(isn_secret) #define V_isn_last VNET(isn_last) #define V_isn_last_reseed VNET(isn_last_reseed) #define V_isn_offset VNET(isn_offset) #define V_isn_offset_old VNET(isn_offset_old) tcp_seq tcp_new_isn(struct in_conninfo *inc) { tcp_seq new_isn; u_int32_t projected_offset; ISN_LOCK(); /* Seed if this is the first use, reseed if requested. */ if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) < (u_int)ticks))) { arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); V_isn_last_reseed = ticks; } /* Compute the hash and return the ISN. */ new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, sizeof(V_isn_secret)); V_isn_offset += ISN_STATIC_INCREMENT + (arc4random() & ISN_RANDOM_INCREMENT); if (ticks != V_isn_last) { projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); if (SEQ_GT(projected_offset, V_isn_offset)) V_isn_offset = projected_offset; V_isn_offset_old = V_isn_offset; V_isn_last = ticks; } new_isn += V_isn_offset; ISN_UNLOCK(); return (new_isn); } /* * When a specific ICMP unreachable message is received and the * connection state is SYN-SENT, drop the connection. This behavior * is controlled by the icmp_may_rst sysctl. */ struct inpcb * tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp; NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); if (tp->t_state != TCPS_SYN_SENT) return (inp); if (IS_FASTOPEN(tp->t_flags)) tcp_fastopen_disable_path(tp); tp = tcp_drop(tp, errno); if (tp != NULL) return (inp); else return (NULL); } /* * When `need fragmentation' ICMP is received, update our idea of the MSS * based on the new value. Also nudge TCP to send something, since we * know the packet we just sent was dropped. * This duplicates some code in the tcp_mss() function in tcp_input.c. */ static struct inpcb * tcp_mtudisc_notify(struct inpcb *inp, int error) { tcp_mtudisc(inp, -1); return (inp); } static void tcp_mtudisc(struct inpcb *inp, int mtuoffer) { struct tcpcb *tp; struct socket *so; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return; tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); /* If the mss is larger than the socket buffer, decrease the mss. */ if (so->so_snd.sb_hiwat < tp->t_maxseg) tp->t_maxseg = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); TCPSTAT_INC(tcps_mturesent); tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; tcp_free_sackholes(tp); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_SACK_PERMIT) EXIT_FASTRECOVERY(tp->t_flags); tp->t_fb->tfb_tcp_output(tp); } #ifdef INET /* * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated, then return 0. This routine * is called by TCP routines that access the rmx structure and by * tcp_mss_update to get the peer/interface MTU. */ uint32_t tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop4_extended nh4; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); if (inc->inc_faddr.s_addr != INADDR_ANY) { if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, NHR_REF, 0, &nh4) != 0) return (0); ifp = nh4.nh_ifp; maxmtu = nh4.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO4 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib4_free_nh_ext(inc->inc_fibnum, &nh4); } return (maxmtu); } #endif /* INET */ #ifdef INET6 uint32_t tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct nhop6_extended nh6; struct in6_addr dst6; uint32_t scopeid; struct ifnet *ifp; uint32_t maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); if (inc->inc_flags & INC_IPV6MINMTU) return (IPV6_MMTU); if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 0, &nh6) != 0) return (0); ifp = nh6.nh_ifp; maxmtu = nh6.nh_mtu; /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO6 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } fib6_free_nh_ext(inc->inc_fibnum, &nh6); } return (maxmtu); } #endif /* INET6 */ /* * Calculate effective SMSS per RFC5681 definition for a given TCP * connection at its current state, taking into account SACK and etc. */ u_int tcp_maxseg(const struct tcpcb *tp) { u_int optlen; if (tp->t_flags & TF_NOOPT) return (tp->t_maxseg); /* * Here we have a simplified code from tcp_addoptions(), * without a proper loop, and having most of paddings hardcoded. * We might make mistakes with padding here in some edge cases, * but this is harmless, since result of tcp_maxseg() is used * only in cwnd and ssthresh estimations. */ #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) if (TCPS_HAVEESTABLISHED(tp->t_state)) { if (tp->t_flags & TF_RCVD_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = 0; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { optlen += TCPOLEN_SACKHDR; optlen += tp->rcv_numsacks * TCPOLEN_SACK; optlen = PAD(optlen); } } else { if (tp->t_flags & TF_REQ_TSTMP) optlen = TCPOLEN_TSTAMP_APPA; else optlen = PAD(TCPOLEN_MAXSEG); if (tp->t_flags & TF_REQ_SCALE) optlen += PAD(TCPOLEN_WINDOW); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) optlen += PAD(TCPOLEN_SIGNATURE); #endif if (tp->t_flags & TF_SACK_PERMIT) optlen += PAD(TCPOLEN_SACK_PERMITTED); } #undef PAD optlen = min(optlen, TCP_MAXOLEN); return (tp->t_maxseg - optlen); } static int sysctl_drop(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct tcpcb *tp; struct tcptw *tw; struct sockaddr_in *fin, *lin; struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } if (inp != NULL) { if (inp->inp_flags & INP_TIMEWAIT) { /* * XXXRW: There currently exists a state where an * inpcb is present, but its timewait state has been * discarded. For now, don't allow dropping of this * type of inpcb. */ tw = intotw(inp); if (tw != NULL) tcp_twclose(tw, 0); else INP_WUNLOCK(inp); } else if (!(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { tp = intotcpcb(inp); tp = tcp_drop(tp, ECONNABORTED); if (tp != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } else error = ESRCH; NET_EPOCH_EXIT(et); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 0, sysctl_drop, "", "Drop TCP connection"); #ifdef KERN_TLS static int sysctl_switch_tls(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct sockaddr_in *fin, *lin; struct epoch_tracker et; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } NET_EPOCH_ENTER(et); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } NET_EPOCH_EXIT(et); if (inp != NULL) { if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || inp->inp_socket == NULL) { error = ECONNRESET; INP_WUNLOCK(inp); } else { struct socket *so; so = inp->inp_socket; soref(so); error = ktls_set_tx_mode(so, arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); INP_WUNLOCK(inp); SOCK_LOCK(so); sorele(so); } } else error = ESRCH; return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS"); SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS"); #endif /* * Generate a standardized TCP log line for use throughout the * tcp subsystem. Memory allocation is done with M_NOWAIT to * allow use in the interrupt context. * * NB: The caller MUST free(s, M_TCPLOG) the returned string. * NB: The function may return NULL if memory allocation failed. * * Due to header inclusion and ordering limitations the struct ip * and ip6_hdr pointers have to be passed as void pointers. */ char * tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (V_tcp_log_in_vain == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } char * tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_debug == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { char *s, *sp; size_t size; struct ip *ip; #ifdef INET6 const struct ip6_hdr *ip6; ip6 = (const struct ip6_hdr *)ip6hdr; #endif /* INET6 */ ip = (struct ip *)ip4hdr; /* * The log line looks like this: * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2" */ size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + sizeof(PRINT_TH_FLAGS) + 1 + #ifdef INET6 2 * INET6_ADDRSTRLEN; #else 2 * INET_ADDRSTRLEN; #endif /* INET6 */ s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); if (s == NULL) return (NULL); strcat(s, "TCP: ["); sp = s + strlen(s); if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { inet_ntoa_r(inc->inc_faddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); inet_ntoa_r(inc->inc_laddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); #ifdef INET6 } else if (inc) { ip6_sprintf(sp, &inc->inc6_faddr); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); ip6_sprintf(sp, &inc->inc6_laddr); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); } else if (ip6 && th) { ip6_sprintf(sp, &ip6->ip6_src); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); ip6_sprintf(sp, &ip6->ip6_dst); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET6 */ #ifdef INET } else if (ip && th) { inet_ntoa_r(ip->ip_src, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); inet_ntoa_r(ip->ip_dst, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET */ } else { free(s, M_TCPLOG); return (NULL); } sp = s + strlen(s); if (th) sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); if (*(s + size - 1) != '\0') panic("%s: string too long", __func__); return (s); } /* * A subroutine which makes it easy to track TCP state changes with DTrace. * This function shouldn't be called for t_state initializations that don't * correspond to actual TCP state transitions. */ void tcp_state_change(struct tcpcb *tp, int newstate) { #if defined(KDTRACE_HOOKS) int pstate = tp->t_state; #endif TCPSTATES_DEC(tp->t_state); TCPSTATES_INC(newstate); tp->t_state = newstate; TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); } /* * Create an external-format (``xtcpcb'') structure using the information in * the kernel-format tcpcb structure pointed to by tp. This is done to * reduce the spew of irrelevant information over this interface, to isolate * user code from changes in the kernel structure, and potentially to provide * information-hiding if we decide that some of this information should be * hidden from users. */ void tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) { struct tcpcb *tp = intotcpcb(inp); sbintime_t now; bzero(xt, sizeof(*xt)); if (inp->inp_flags & INP_TIMEWAIT) { xt->t_state = TCPS_TIME_WAIT; } else { xt->t_state = tp->t_state; xt->t_logstate = tp->t_logstate; xt->t_flags = tp->t_flags; xt->t_sndzerowin = tp->t_sndzerowin; xt->t_sndrexmitpack = tp->t_sndrexmitpack; xt->t_rcvoopack = tp->t_rcvoopack; now = getsbinuptime(); #define COPYTIMER(ttt) do { \ if (callout_active(&tp->t_timers->ttt)) \ xt->ttt = (tp->t_timers->ttt.c_time - now) / \ SBT_1MS; \ else \ xt->ttt = 0; \ } while (0) COPYTIMER(tt_delack); COPYTIMER(tt_rexmt); COPYTIMER(tt_persist); COPYTIMER(tt_keep); COPYTIMER(tt_2msl); #undef COPYTIMER xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, TCP_FUNCTION_NAME_LEN_MAX); #ifdef TCP_BLACKBOX (void)tcp_log_get_id(tp, xt->xt_logid); #endif } xt->xt_len = sizeof(struct xtcpcb); in_pcbtoxinpcb(inp, &xt->xt_inp); if (inp->inp_socket == NULL) xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; } Index: head/sys/netinet/tcp_syncache.c =================================================================== --- head/sys/netinet/tcp_syncache.c (revision 356968) +++ head/sys/netinet/tcp_syncache.c (revision 356969) @@ -1,2466 +1,2469 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2001 McAfee, Inc. * Copyright (c) 2006,2013 Andre Oppermann, Internet Business Solutions AG * All rights reserved. * * This software was developed for the FreeBSD Project by Jonathan Lemon * and McAfee Research, the Security Research Division of McAfee, Inc. under * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the * DARPA CHATS research program. [2001 McAfee, Inc.] * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_pcbgroup.h" #include #include #include #include #include #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #endif #include #include #include #include #include #include #include #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include VNET_DEFINE_STATIC(int, tcp_syncookies) = 1; #define V_tcp_syncookies VNET(tcp_syncookies) SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_syncookies), 0, "Use TCP SYN cookies if the syncache overflows"); VNET_DEFINE_STATIC(int, tcp_syncookiesonly) = 0; #define V_tcp_syncookiesonly VNET(tcp_syncookiesonly) SYSCTL_INT(_net_inet_tcp, OID_AUTO, syncookies_only, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_syncookiesonly), 0, "Use only TCP SYN cookies"); VNET_DEFINE_STATIC(int, functions_inherit_listen_socket_stack) = 1; #define V_functions_inherit_listen_socket_stack \ VNET(functions_inherit_listen_socket_stack) SYSCTL_INT(_net_inet_tcp, OID_AUTO, functions_inherit_listen_socket_stack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(functions_inherit_listen_socket_stack), 0, "Inherit listen socket's stack"); #ifdef TCP_OFFLOAD #define ADDED_BY_TOE(sc) ((sc)->sc_tod != NULL) #endif static void syncache_drop(struct syncache *, struct syncache_head *); static void syncache_free(struct syncache *); static void syncache_insert(struct syncache *, struct syncache_head *); static int syncache_respond(struct syncache *, const struct mbuf *, int); static struct socket *syncache_socket(struct syncache *, struct socket *, struct mbuf *m); static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout); static void syncache_timer(void *); static uint32_t syncookie_mac(struct in_conninfo *, tcp_seq, uint8_t, uint8_t *, uintptr_t); static tcp_seq syncookie_generate(struct syncache_head *, struct syncache *); static struct syncache *syncookie_lookup(struct in_conninfo *, struct syncache_head *, struct syncache *, struct tcphdr *, struct tcpopt *, struct socket *); static void syncache_pause(struct in_conninfo *); static void syncache_unpause(void *); static void syncookie_reseed(void *); #ifdef INVARIANTS static int syncookie_cmp(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso); #endif /* * Transmit the SYN,ACK fewer times than TCP_MAXRXTSHIFT specifies. * 3 retransmits corresponds to a timeout with default values of * tcp_rexmit_initial * ( 1 + * tcp_backoff[1] + * tcp_backoff[2] + * tcp_backoff[3]) + 3 * tcp_rexmit_slop, * 1000 ms * (1 + 2 + 4 + 8) + 3 * 200 ms = 15600 ms, * the odds are that the user has given up attempting to connect by then. */ #define SYNCACHE_MAXREXMTS 3 /* Arbitrary values */ #define TCP_SYNCACHE_HASHSIZE 512 #define TCP_SYNCACHE_BUCKETLIMIT 30 VNET_DEFINE_STATIC(struct tcp_syncache, tcp_syncache); #define V_tcp_syncache VNET(tcp_syncache) static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, syncache, CTLFLAG_RW, 0, "TCP SYN cache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, bucketlimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.bucket_limit), 0, "Per-bucket hash limit for syncache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, cachelimit, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.cache_limit), 0, "Overall entry limit for syncache"); SYSCTL_UMA_CUR(_net_inet_tcp_syncache, OID_AUTO, count, CTLFLAG_VNET, &VNET_NAME(tcp_syncache.zone), "Current number of entries in syncache"); SYSCTL_UINT(_net_inet_tcp_syncache, OID_AUTO, hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN, &VNET_NAME(tcp_syncache.hashsize), 0, "Size of TCP syncache hashtable"); static int sysctl_net_inet_tcp_syncache_rexmtlimit_check(SYSCTL_HANDLER_ARGS) { int error; u_int new; new = V_tcp_syncache.rexmt_limit; error = sysctl_handle_int(oidp, &new, 0, req); if ((error == 0) && (req->newptr != NULL)) { if (new > TCP_MAXRXTSHIFT) error = EINVAL; else V_tcp_syncache.rexmt_limit = new; } return (error); } SYSCTL_PROC(_net_inet_tcp_syncache, OID_AUTO, rexmtlimit, CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(tcp_syncache.rexmt_limit), 0, sysctl_net_inet_tcp_syncache_rexmtlimit_check, "UI", "Limit on SYN/ACK retransmissions"); VNET_DEFINE(int, tcp_sc_rst_sock_fail) = 1; SYSCTL_INT(_net_inet_tcp_syncache, OID_AUTO, rst_on_sock_fail, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sc_rst_sock_fail), 0, "Send reset on socket allocation failure"); static MALLOC_DEFINE(M_SYNCACHE, "syncache", "TCP syncache"); #define SCH_LOCK(sch) mtx_lock(&(sch)->sch_mtx) #define SCH_UNLOCK(sch) mtx_unlock(&(sch)->sch_mtx) #define SCH_LOCK_ASSERT(sch) mtx_assert(&(sch)->sch_mtx, MA_OWNED) /* * Requires the syncache entry to be already removed from the bucket list. */ static void syncache_free(struct syncache *sc) { if (sc->sc_ipopts) (void) m_free(sc->sc_ipopts); if (sc->sc_cred) crfree(sc->sc_cred); #ifdef MAC mac_syncache_destroy(&sc->sc_label); #endif uma_zfree(V_tcp_syncache.zone, sc); } void syncache_init(void) { int i; V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; V_tcp_syncache.bucket_limit = TCP_SYNCACHE_BUCKETLIMIT; V_tcp_syncache.rexmt_limit = SYNCACHE_MAXREXMTS; V_tcp_syncache.hash_secret = arc4random(); TUNABLE_INT_FETCH("net.inet.tcp.syncache.hashsize", &V_tcp_syncache.hashsize); TUNABLE_INT_FETCH("net.inet.tcp.syncache.bucketlimit", &V_tcp_syncache.bucket_limit); if (!powerof2(V_tcp_syncache.hashsize) || V_tcp_syncache.hashsize == 0) { printf("WARNING: syncache hash size is not a power of 2.\n"); V_tcp_syncache.hashsize = TCP_SYNCACHE_HASHSIZE; } V_tcp_syncache.hashmask = V_tcp_syncache.hashsize - 1; /* Set limits. */ V_tcp_syncache.cache_limit = V_tcp_syncache.hashsize * V_tcp_syncache.bucket_limit; TUNABLE_INT_FETCH("net.inet.tcp.syncache.cachelimit", &V_tcp_syncache.cache_limit); /* Allocate the hash table. */ V_tcp_syncache.hashbase = malloc(V_tcp_syncache.hashsize * sizeof(struct syncache_head), M_SYNCACHE, M_WAITOK | M_ZERO); #ifdef VIMAGE V_tcp_syncache.vnet = curvnet; #endif /* Initialize the hash buckets. */ for (i = 0; i < V_tcp_syncache.hashsize; i++) { TAILQ_INIT(&V_tcp_syncache.hashbase[i].sch_bucket); mtx_init(&V_tcp_syncache.hashbase[i].sch_mtx, "tcp_sc_head", NULL, MTX_DEF); callout_init_mtx(&V_tcp_syncache.hashbase[i].sch_timer, &V_tcp_syncache.hashbase[i].sch_mtx, 0); V_tcp_syncache.hashbase[i].sch_length = 0; V_tcp_syncache.hashbase[i].sch_sc = &V_tcp_syncache; V_tcp_syncache.hashbase[i].sch_last_overflow = -(SYNCOOKIE_LIFETIME + 1); } /* Create the syncache entry zone. */ V_tcp_syncache.zone = uma_zcreate("syncache", sizeof(struct syncache), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_tcp_syncache.cache_limit = uma_zone_set_max(V_tcp_syncache.zone, V_tcp_syncache.cache_limit); /* Start the SYN cookie reseeder callout. */ callout_init(&V_tcp_syncache.secret.reseed, 1); arc4rand(V_tcp_syncache.secret.key[0], SYNCOOKIE_SECRET_SIZE, 0); arc4rand(V_tcp_syncache.secret.key[1], SYNCOOKIE_SECRET_SIZE, 0); callout_reset(&V_tcp_syncache.secret.reseed, SYNCOOKIE_LIFETIME * hz, syncookie_reseed, &V_tcp_syncache); /* Initialize the pause machinery. */ mtx_init(&V_tcp_syncache.pause_mtx, "tcp_sc_pause", NULL, MTX_DEF); callout_init_mtx(&V_tcp_syncache.pause_co, &V_tcp_syncache.pause_mtx, 0); V_tcp_syncache.pause_until = time_uptime - TCP_SYNCACHE_PAUSE_TIME; V_tcp_syncache.pause_backoff = 0; V_tcp_syncache.paused = false; } #ifdef VIMAGE void syncache_destroy(void) { struct syncache_head *sch; struct syncache *sc, *nsc; int i; /* * Stop the re-seed timer before freeing resources. No need to * possibly schedule it another time. */ callout_drain(&V_tcp_syncache.secret.reseed); /* Stop the SYN cache pause callout. */ mtx_lock(&V_tcp_syncache.pause_mtx); if (callout_stop(&V_tcp_syncache.pause_co) == 0) { mtx_unlock(&V_tcp_syncache.pause_mtx); callout_drain(&V_tcp_syncache.pause_co); } else mtx_unlock(&V_tcp_syncache.pause_mtx); /* Cleanup hash buckets: stop timers, free entries, destroy locks. */ for (i = 0; i < V_tcp_syncache.hashsize; i++) { sch = &V_tcp_syncache.hashbase[i]; callout_drain(&sch->sch_timer); SCH_LOCK(sch); TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) syncache_drop(sc, sch); SCH_UNLOCK(sch); KASSERT(TAILQ_EMPTY(&sch->sch_bucket), ("%s: sch->sch_bucket not empty", __func__)); KASSERT(sch->sch_length == 0, ("%s: sch->sch_length %d not 0", __func__, sch->sch_length)); mtx_destroy(&sch->sch_mtx); } KASSERT(uma_zone_get_cur(V_tcp_syncache.zone) == 0, ("%s: cache_count not 0", __func__)); /* Free the allocated global resources. */ uma_zdestroy(V_tcp_syncache.zone); free(V_tcp_syncache.hashbase, M_SYNCACHE); mtx_destroy(&V_tcp_syncache.pause_mtx); } #endif /* * Inserts a syncache entry into the specified bucket row. * Locks and unlocks the syncache_head autonomously. */ static void syncache_insert(struct syncache *sc, struct syncache_head *sch) { struct syncache *sc2; SCH_LOCK(sch); /* * Make sure that we don't overflow the per-bucket limit. * If the bucket is full, toss the oldest element. */ if (sch->sch_length >= V_tcp_syncache.bucket_limit) { KASSERT(!TAILQ_EMPTY(&sch->sch_bucket), ("sch->sch_length incorrect")); syncache_pause(&sc->sc_inc); sc2 = TAILQ_LAST(&sch->sch_bucket, sch_head); sch->sch_last_overflow = time_uptime; syncache_drop(sc2, sch); } /* Put it into the bucket. */ TAILQ_INSERT_HEAD(&sch->sch_bucket, sc, sc_hash); sch->sch_length++; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_added(tod, sc->sc_todctx); } #endif /* Reinitialize the bucket row's timer. */ if (sch->sch_length == 1) sch->sch_nextc = ticks + INT_MAX; syncache_timeout(sc, sch, 1); SCH_UNLOCK(sch); TCPSTATES_INC(TCPS_SYN_RECEIVED); TCPSTAT_INC(tcps_sc_added); } /* * Remove and free entry from syncache bucket row. * Expects locked syncache head. */ static void syncache_drop(struct syncache *sc, struct syncache_head *sch) { SCH_LOCK_ASSERT(sch); TCPSTATES_DEC(TCPS_SYN_RECEIVED); TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); sch->sch_length--; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_removed(tod, sc->sc_todctx); } #endif syncache_free(sc); } /* * Engage/reengage time on bucket row. */ static void syncache_timeout(struct syncache *sc, struct syncache_head *sch, int docallout) { int rexmt; if (sc->sc_rxmits == 0) rexmt = tcp_rexmit_initial; else TCPT_RANGESET(rexmt, tcp_rexmit_initial * tcp_backoff[sc->sc_rxmits], tcp_rexmit_min, TCPTV_REXMTMAX); sc->sc_rxttime = ticks + rexmt; sc->sc_rxmits++; if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) { sch->sch_nextc = sc->sc_rxttime; if (docallout) callout_reset(&sch->sch_timer, sch->sch_nextc - ticks, syncache_timer, (void *)sch); } } /* * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. * If we have retransmitted an entry the maximum number of times, expire it. * One separate timer for each bucket row. */ static void syncache_timer(void *xsch) { struct syncache_head *sch = (struct syncache_head *)xsch; struct syncache *sc, *nsc; int tick = ticks; char *s; bool paused; CURVNET_SET(sch->sch_sc->vnet); /* NB: syncache_head has already been locked by the callout. */ SCH_LOCK_ASSERT(sch); /* * In the following cycle we may remove some entries and/or * advance some timeouts, so re-initialize the bucket timer. */ sch->sch_nextc = tick + INT_MAX; /* * If we have paused processing, unconditionally remove * all syncache entries. */ mtx_lock(&V_tcp_syncache.pause_mtx); paused = V_tcp_syncache.paused; mtx_unlock(&V_tcp_syncache.pause_mtx); TAILQ_FOREACH_SAFE(sc, &sch->sch_bucket, sc_hash, nsc) { if (paused) { syncache_drop(sc, sch); continue; } /* * We do not check if the listen socket still exists * and accept the case where the listen socket may be * gone by the time we resend the SYN/ACK. We do * not expect this to happens often. If it does, * then the RST will be sent by the time the remote * host does the SYN/ACK->ACK. */ if (TSTMP_GT(sc->sc_rxttime, tick)) { if (TSTMP_LT(sc->sc_rxttime, sch->sch_nextc)) sch->sch_nextc = sc->sc_rxttime; continue; } if (sc->sc_rxmits > V_tcp_syncache.rexmt_limit) { if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Retransmits exhausted, " "giving up and removing syncache entry\n", s, __func__); free(s, M_TCPLOG); } syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_stale); continue; } if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Response timeout, " "retransmitting (%u) SYN|ACK\n", s, __func__, sc->sc_rxmits); free(s, M_TCPLOG); } syncache_respond(sc, NULL, TH_SYN|TH_ACK); TCPSTAT_INC(tcps_sc_retransmitted); syncache_timeout(sc, sch, 0); } if (!TAILQ_EMPTY(&(sch)->sch_bucket)) callout_reset(&(sch)->sch_timer, (sch)->sch_nextc - tick, syncache_timer, (void *)(sch)); CURVNET_RESTORE(); } /* * Returns true if the system is only using cookies at the moment. * This could be due to a sysadmin decision to only use cookies, or it * could be due to the system detecting an attack. */ static inline bool syncache_cookiesonly(void) { return (V_tcp_syncookies && (V_tcp_syncache.paused || V_tcp_syncookiesonly)); } /* * Find the hash bucket for the given connection. */ static struct syncache_head * syncache_hashbucket(struct in_conninfo *inc) { uint32_t hash; /* * The hash is built on foreign port + local port + foreign address. * We rely on the fact that struct in_conninfo starts with 16 bits * of foreign port, then 16 bits of local port then followed by 128 * bits of foreign address. In case of IPv4 address, the first 3 * 32-bit words of the address always are zeroes. */ hash = jenkins_hash32((uint32_t *)&inc->inc_ie, 5, V_tcp_syncache.hash_secret) & V_tcp_syncache.hashmask; return (&V_tcp_syncache.hashbase[hash]); } /* * Find an entry in the syncache. * Returns always with locked syncache_head plus a matching entry or NULL. */ static struct syncache * syncache_lookup(struct in_conninfo *inc, struct syncache_head **schp) { struct syncache *sc; struct syncache_head *sch; *schp = sch = syncache_hashbucket(inc); SCH_LOCK(sch); /* Circle through bucket row to find matching entry. */ TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) if (bcmp(&inc->inc_ie, &sc->sc_inc.inc_ie, sizeof(struct in_endpoints)) == 0) break; return (sc); /* Always returns with locked sch. */ } /* * This function is called when we get a RST for a * non-existent connection, so that we can see if the * connection is in the syn cache. If it is, zap it. * If required send a challenge ACK. */ void syncache_chkrst(struct in_conninfo *inc, struct tcphdr *th, struct mbuf *m) { struct syncache *sc; struct syncache_head *sch; char *s = NULL; if (syncache_cookiesonly()) return; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); /* * Any RST to our SYN|ACK must not carry ACK, SYN or FIN flags. * See RFC 793 page 65, section SEGMENT ARRIVES. */ if (th->th_flags & (TH_ACK|TH_SYN|TH_FIN)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious RST with ACK, SYN or " "FIN flag set, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badrst); goto done; } /* * No corresponding connection was found in syncache. * If syncookies are enabled and possibly exclusively * used, or we are under memory pressure, a valid RST * may not find a syncache entry. In that case we're * done and no SYN|ACK retransmissions will happen. * Otherwise the RST was misdirected or spoofed. */ if (sc == NULL) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious RST without matching " "syncache entry (possibly syncookie only), " "segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badrst); goto done; } /* * If the RST bit is set, check the sequence number to see * if this is a valid reset segment. * * RFC 793 page 37: * In all states except SYN-SENT, all reset (RST) segments * are validated by checking their SEQ-fields. A reset is * valid if its sequence number is in the window. * * RFC 793 page 69: * There are four cases for the acceptability test for an incoming * segment: * * Segment Receive Test * Length Window * ------- ------- ------------------------------------------- * 0 0 SEG.SEQ = RCV.NXT * 0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND * >0 0 not acceptable * >0 >0 RCV.NXT =< SEG.SEQ < RCV.NXT+RCV.WND * or RCV.NXT =< SEG.SEQ+SEG.LEN-1 < RCV.NXT+RCV.WND * * Note that when receiving a SYN segment in the LISTEN state, * IRS is set to SEG.SEQ and RCV.NXT is set to SEG.SEQ+1, as * described in RFC 793, page 66. */ if ((SEQ_GEQ(th->th_seq, sc->sc_irs + 1) && SEQ_LT(th->th_seq, sc->sc_irs + 1 + sc->sc_wnd)) || (sc->sc_wnd == 0 && th->th_seq == sc->sc_irs + 1)) { if (V_tcp_insecure_rst || th->th_seq == sc->sc_irs + 1) { syncache_drop(sc, sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Our SYN|ACK was rejected, " "connection attempt aborted by remote " "endpoint\n", s, __func__); TCPSTAT_INC(tcps_sc_reset); } else { TCPSTAT_INC(tcps_badrst); /* Send challenge ACK. */ if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: RST with invalid " " SEQ %u != NXT %u (+WND %u), " "sending challenge ACK\n", s, __func__, th->th_seq, sc->sc_irs + 1, sc->sc_wnd); syncache_respond(sc, m, TH_ACK); } } else { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: RST with invalid SEQ %u != " "NXT %u (+WND %u), segment ignored\n", s, __func__, th->th_seq, sc->sc_irs + 1, sc->sc_wnd); TCPSTAT_INC(tcps_badrst); } done: if (s != NULL) free(s, M_TCPLOG); SCH_UNLOCK(sch); } void syncache_badack(struct in_conninfo *inc) { struct syncache *sc; struct syncache_head *sch; if (syncache_cookiesonly()) return; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); if (sc != NULL) { syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_badack); } SCH_UNLOCK(sch); } void syncache_unreach(struct in_conninfo *inc, tcp_seq th_seq) { struct syncache *sc; struct syncache_head *sch; if (syncache_cookiesonly()) return; sc = syncache_lookup(inc, &sch); /* returns locked sch */ SCH_LOCK_ASSERT(sch); if (sc == NULL) goto done; /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ if (ntohl(th_seq) != sc->sc_iss) goto done; /* * If we've rertransmitted 3 times and this is our second error, * we remove the entry. Otherwise, we allow it to continue on. * This prevents us from incorrectly nuking an entry during a * spurious network outage. * * See tcp_notify(). */ if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxmits < 3 + 1) { sc->sc_flags |= SCF_UNREACH; goto done; } syncache_drop(sc, sch); TCPSTAT_INC(tcps_sc_unreach); done: SCH_UNLOCK(sch); } /* * Build a new TCP socket structure from a syncache entry. * * On success return the newly created socket with its underlying inp locked. */ static struct socket * syncache_socket(struct syncache *sc, struct socket *lso, struct mbuf *m) { struct tcp_function_block *blk; struct inpcb *inp = NULL; struct socket *so; struct tcpcb *tp; int error; char *s; NET_EPOCH_ASSERT(); /* * Ok, create the full blown connection, and set things up * as they would have been set up if we had created the * connection when the SYN arrived. If we can't create * the connection, abort it. */ so = sonewconn(lso, 0); if (so == NULL) { /* * Drop the connection; we will either send a RST or * have the peer retransmit its SYN again after its * RTO and try again. */ TCPSTAT_INC(tcps_listendrop); if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Socket create failed " "due to limits or memory shortage\n", s, __func__); free(s, M_TCPLOG); } goto abort2; } #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif inp = sotoinpcb(so); inp->inp_inc.inc_fibnum = so->so_fibnum; INP_WLOCK(inp); /* * Exclusive pcbinfo lock is not required in syncache socket case even * if two inpcb locks can be acquired simultaneously: * - the inpcb in LISTEN state, * - the newly created inp. * * In this case, an inp cannot be at same time in LISTEN state and * just created by an accept() call. */ INP_HASH_WLOCK(&V_tcbinfo); /* Insert new socket into PCB hash list. */ inp->inp_inc.inc_flags = sc->sc_inc.inc_flags; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { inp->inp_vflag &= ~INP_IPV4; inp->inp_vflag |= INP_IPV6; inp->in6p_laddr = sc->sc_inc.inc6_laddr; } else { inp->inp_vflag &= ~INP_IPV6; inp->inp_vflag |= INP_IPV4; #endif inp->inp_laddr = sc->sc_inc.inc_laddr; #ifdef INET6 } #endif /* * If there's an mbuf and it has a flowid, then let's initialise the * inp with that particular flowid. */ if (m != NULL && M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); #ifdef NUMA inp->inp_numa_domain = m->m_pkthdr.numa_domain; #endif } inp->inp_lport = sc->sc_inc.inc_lport; #ifdef INET6 if (inp->inp_vflag & INP_IPV6PROTO) { struct inpcb *oinp = sotoinpcb(lso); /* * Inherit socket options from the listening socket. * Note that in6p_inputopts are not (and should not be) * copied, since it stores previously received options and is * used to detect if each new option is different than the * previous one and hence should be passed to a user. * If we copied in6p_inputopts, a user would not be able to * receive options just after calling the accept system call. */ inp->inp_flags |= oinp->inp_flags & INP_CONTROLOPTS; if (oinp->in6p_outputopts) inp->in6p_outputopts = ip6_copypktopts(oinp->in6p_outputopts, M_NOWAIT); } if (sc->sc_inc.inc_flags & INC_ISIPV6) { struct in6_addr laddr6; struct sockaddr_in6 sin6; sin6.sin6_family = AF_INET6; sin6.sin6_len = sizeof(sin6); sin6.sin6_addr = sc->sc_inc.inc6_faddr; sin6.sin6_port = sc->sc_inc.inc_fport; sin6.sin6_flowinfo = sin6.sin6_scope_id = 0; laddr6 = inp->in6p_laddr; if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) inp->in6p_laddr = sc->sc_inc.inc6_laddr; if ((error = in6_pcbconnect_mbuf(inp, (struct sockaddr *)&sin6, thread0.td_ucred, m, false)) != 0) { inp->in6p_laddr = laddr6; if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: in6_pcbconnect failed " "with error %i\n", s, __func__, error); free(s, M_TCPLOG); } INP_HASH_WUNLOCK(&V_tcbinfo); goto abort; } /* Override flowlabel from in6_pcbconnect. */ inp->inp_flow &= ~IPV6_FLOWLABEL_MASK; inp->inp_flow |= sc->sc_flowlabel; } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { struct in_addr laddr; struct sockaddr_in sin; inp->inp_options = (m) ? ip_srcroute(m) : NULL; if (inp->inp_options == NULL) { inp->inp_options = sc->sc_ipopts; sc->sc_ipopts = NULL; } sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); sin.sin_addr = sc->sc_inc.inc_faddr; sin.sin_port = sc->sc_inc.inc_fport; bzero((caddr_t)sin.sin_zero, sizeof(sin.sin_zero)); laddr = inp->inp_laddr; if (inp->inp_laddr.s_addr == INADDR_ANY) inp->inp_laddr = sc->sc_inc.inc_laddr; if ((error = in_pcbconnect_mbuf(inp, (struct sockaddr *)&sin, thread0.td_ucred, m, false)) != 0) { inp->inp_laddr = laddr; if ((s = tcp_log_addrs(&sc->sc_inc, NULL, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: in_pcbconnect failed " "with error %i\n", s, __func__, error); free(s, M_TCPLOG); } INP_HASH_WUNLOCK(&V_tcbinfo); goto abort; } } #endif /* INET */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* Copy old policy into new socket's. */ if (ipsec_copy_pcbpolicy(sotoinpcb(lso), inp) != 0) printf("syncache_socket: could not copy policy\n"); #endif INP_HASH_WUNLOCK(&V_tcbinfo); tp = intotcpcb(inp); tcp_state_change(tp, TCPS_SYN_RECEIVED); tp->iss = sc->sc_iss; tp->irs = sc->sc_irs; tcp_rcvseqinit(tp); tcp_sendseqinit(tp); blk = sototcpcb(lso)->t_fb; if (V_functions_inherit_listen_socket_stack && blk != tp->t_fb) { /* * Our parents t_fb was not the default, * we need to release our ref on tp->t_fb and * pickup one on the new entry. */ struct tcp_function_block *rblk; rblk = find_and_ref_tcp_fb(blk); KASSERT(rblk != NULL, ("cannot find blk %p out of syncache?", blk)); if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_fb = rblk; /* * XXXrrs this is quite dangerous, it is possible * for the new function to fail to init. We also * are not asking if the handoff_is_ok though at * the very start thats probalbly ok. */ if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } } tp->snd_wl1 = sc->sc_irs; tp->snd_max = tp->iss + 1; tp->snd_nxt = tp->iss + 1; tp->rcv_up = sc->sc_irs + 1; tp->rcv_wnd = sc->sc_wnd; tp->rcv_adv += tp->rcv_wnd; tp->last_ack_sent = tp->rcv_nxt; tp->t_flags = sototcpcb(lso)->t_flags & (TF_NOPUSH|TF_NODELAY); if (sc->sc_flags & SCF_NOOPT) tp->t_flags |= TF_NOOPT; else { if (sc->sc_flags & SCF_WINSCALE) { tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; tp->snd_scale = sc->sc_requested_s_scale; tp->request_r_scale = sc->sc_requested_r_scale; } if (sc->sc_flags & SCF_TIMESTAMP) { tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; tp->ts_recent = sc->sc_tsreflect; tp->ts_recent_age = tcp_ts_getticks(); tp->ts_offset = sc->sc_tsoff; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) tp->t_flags |= TF_SIGNATURE; #endif if (sc->sc_flags & SCF_SACK) tp->t_flags |= TF_SACK_PERMIT; } if (sc->sc_flags & SCF_ECN) tp->t_flags2 |= TF2_ECN_PERMIT; /* * Set up MSS and get cached values from tcp_hostcache. * This might overwrite some of the defaults we just set. */ tcp_mss(tp, sc->sc_peer_mss); /* * If the SYN,ACK was retransmitted, indicate that CWND to be * limited to one segment in cc_conn_init(). * NB: sc_rxmits counts all SYN,ACK transmits, not just retransmits. */ if (sc->sc_rxmits > 1) tp->snd_cwnd = 1; #ifdef TCP_OFFLOAD /* * Allow a TOE driver to install its hooks. Note that we hold the * pcbinfo lock too and that prevents tcp_usr_accept from accepting a * new connection before the TOE driver has done its thing. */ if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_offload_socket(tod, sc->sc_todctx, so); } #endif /* * Copy and activate timers. */ tp->t_keepinit = sototcpcb(lso)->t_keepinit; tp->t_keepidle = sototcpcb(lso)->t_keepidle; tp->t_keepintvl = sototcpcb(lso)->t_keepintvl; tp->t_keepcnt = sototcpcb(lso)->t_keepcnt; tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); TCPSTAT_INC(tcps_accepts); return (so); abort: INP_WUNLOCK(inp); abort2: if (so != NULL) soabort(so); return (NULL); } /* * This function gets called when we receive an ACK for a * socket in the LISTEN state. We look up the connection * in the syncache, and if its there, we pull it out of * the cache and turn it into a full-blown connection in * the SYN-RECEIVED state. * * On syncache_socket() success the newly created socket * has its underlying inp locked. */ int syncache_expand(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, struct socket **lsop, struct mbuf *m) { struct syncache *sc; struct syncache_head *sch; struct syncache scs; char *s; bool locked; NET_EPOCH_ASSERT(); KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK, ("%s: can handle only ACK", __func__)); if (syncache_cookiesonly()) { sc = NULL; sch = syncache_hashbucket(inc); locked = false; } else { sc = syncache_lookup(inc, &sch); /* returns locked sch */ locked = true; SCH_LOCK_ASSERT(sch); } #ifdef INVARIANTS /* * Test code for syncookies comparing the syncache stored * values with the reconstructed values from the cookie. */ if (sc != NULL) syncookie_cmp(inc, sch, sc, th, to, *lsop); #endif if (sc == NULL) { /* * There is no syncache entry, so see if this ACK is * a returning syncookie. To do this, first: * A. Check if syncookies are used in case of syncache * overflows * B. See if this socket has had a syncache entry dropped in * the recent past. We don't want to accept a bogus * syncookie if we've never received a SYN or accept it * twice. * C. check that the syncookie is valid. If it is, then * cobble up a fake syncache entry, and return. */ if (locked && !V_tcp_syncookies) { SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious ACK, " "segment rejected (syncookies disabled)\n", s, __func__); goto failed; } if (locked && !V_tcp_syncookiesonly && sch->sch_last_overflow < time_uptime - SYNCOOKIE_LIFETIME) { SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Spurious ACK, " "segment rejected (no syncache entry)\n", s, __func__); goto failed; } bzero(&scs, sizeof(scs)); sc = syncookie_lookup(inc, sch, &scs, th, to, *lsop); if (locked) SCH_UNLOCK(sch); if (sc == NULL) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Segment failed " "SYNCOOKIE authentication, segment rejected " "(probably spoofed)\n", s, __func__); goto failed; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* If received ACK has MD5 signature, check it. */ if ((to->to_flags & TOF_SIGNATURE) != 0 && (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0)) { /* Drop the ACK. */ if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment rejected, " "MD5 signature doesn't match.\n", s, __func__); free(s, M_TCPLOG); } TCPSTAT_INC(tcps_sig_err_sigopt); return (-1); /* Do not send RST */ } #endif /* TCP_SIGNATURE */ } else { #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, check that * received ACK has signature and it is correct. * If not, drop the ACK and leave sc entry in th cache, * because SYN was received with correct signature. */ if (sc->sc_flags & SCF_SIGNATURE) { if ((to->to_flags & TOF_SIGNATURE) == 0) { /* No signature */ TCPSTAT_INC(tcps_sig_err_nosigopt); SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment " "rejected, MD5 signature wasn't " "provided.\n", s, __func__); free(s, M_TCPLOG); } return (-1); /* Do not send RST */ } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0) { /* Doesn't match or no SA */ SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Segment " "rejected, MD5 signature doesn't " "match.\n", s, __func__); free(s, M_TCPLOG); } return (-1); /* Do not send RST */ } } #endif /* TCP_SIGNATURE */ /* * RFC 7323 PAWS: If we have a timestamp on this segment and * it's less than ts_recent, drop it. * XXXMT: RFC 7323 also requires to send an ACK. * In tcp_input.c this is only done for TCP segments * with user data, so be consistent here and just drop * the segment. */ if (sc->sc_flags & SCF_TIMESTAMP && to->to_flags & TOF_TS && TSTMP_LT(to->to_tsval, sc->sc_tsreflect)) { SCH_UNLOCK(sch); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: SEG.TSval %u < TS.Recent %u, " "segment dropped\n", s, __func__, to->to_tsval, sc->sc_tsreflect); free(s, M_TCPLOG); } return (-1); /* Do not send RST */ } /* * Pull out the entry to unlock the bucket row. * * NOTE: We must decrease TCPS_SYN_RECEIVED count here, not * tcp_state_change(). The tcpcb is not existent at this * moment. A new one will be allocated via syncache_socket-> * sonewconn->tcp_usr_attach in TCPS_CLOSED state, then * syncache_socket() will change it to TCPS_SYN_RECEIVED. */ TCPSTATES_DEC(TCPS_SYN_RECEIVED); TAILQ_REMOVE(&sch->sch_bucket, sc, sc_hash); sch->sch_length--; #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; tod->tod_syncache_removed(tod, sc->sc_todctx); } #endif SCH_UNLOCK(sch); } /* * Segment validation: * ACK must match our initial sequence number + 1 (the SYN|ACK). */ if (th->th_ack != sc->sc_iss + 1) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: ACK %u != ISS+1 %u, segment " "rejected\n", s, __func__, th->th_ack, sc->sc_iss); goto failed; } /* * The SEQ must fall in the window starting at the received * initial receive sequence number + 1 (the SYN). */ if (SEQ_LEQ(th->th_seq, sc->sc_irs) || SEQ_GT(th->th_seq, sc->sc_irs + sc->sc_wnd)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: SEQ %u != IRS+1 %u, segment " "rejected\n", s, __func__, th->th_seq, sc->sc_irs); goto failed; } /* * If timestamps were not negotiated during SYN/ACK they * must not appear on any segment during this session. */ if (!(sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "segment rejected\n", s, __func__); goto failed; } /* * If timestamps were negotiated during SYN/ACK they should * appear on every segment during this session. * XXXAO: This is only informal as there have been unverified * reports of non-compliants stacks. */ if ((sc->sc_flags & SCF_TIMESTAMP) && !(to->to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "no action\n", s, __func__); free(s, M_TCPLOG); s = NULL; } } *lsop = syncache_socket(sc, *lsop, m); if (*lsop == NULL) TCPSTAT_INC(tcps_sc_aborted); else TCPSTAT_INC(tcps_sc_completed); /* how do we find the inp for the new socket? */ if (sc != &scs) syncache_free(sc); return (1); failed: if (sc != NULL && sc != &scs) syncache_free(sc); if (s != NULL) free(s, M_TCPLOG); *lsop = NULL; return (0); } static void syncache_tfo_expand(struct syncache *sc, struct socket **lsop, struct mbuf *m, uint64_t response_cookie) { struct inpcb *inp; struct tcpcb *tp; unsigned int *pending_counter; NET_EPOCH_ASSERT(); pending_counter = intotcpcb(sotoinpcb(*lsop))->t_tfo_pending; *lsop = syncache_socket(sc, *lsop, m); if (*lsop == NULL) { TCPSTAT_INC(tcps_sc_aborted); atomic_subtract_int(pending_counter, 1); } else { soisconnected(*lsop); inp = sotoinpcb(*lsop); tp = intotcpcb(inp); tp->t_flags |= TF_FASTOPEN; tp->t_tfo_cookie.server = response_cookie; tp->snd_max = tp->iss; tp->snd_nxt = tp->iss; tp->t_tfo_pending = pending_counter; TCPSTAT_INC(tcps_sc_completed); } } /* * Given a LISTEN socket and an inbound SYN request, add * this to the syn cache, and send back a segment: * * to the source. * * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. * Doing so would require that we hold onto the data and deliver it * to the application. However, if we are the target of a SYN-flood * DoS attack, an attacker could send data which would eventually * consume all available buffer space if it were ACKed. By not ACKing * the data, we avoid this DoS scenario. * * The exception to the above is when a SYN with a valid TCP Fast Open (TFO) * cookie is processed and a new socket is created. In this case, any data * accompanying the SYN will be queued to the socket by tcp_input() and will * be ACKed either when the application sends response data or the delayed * ACK timer expires, whichever comes first. */ int syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, struct inpcb *inp, struct socket **lsop, struct mbuf *m, void *tod, void *todctx, uint8_t iptos) { struct tcpcb *tp; struct socket *so; struct syncache *sc = NULL; struct syncache_head *sch; struct mbuf *ipopts = NULL; u_int ltflags; int win, ip_ttl, ip_tos; char *s; int rv = 0; #ifdef INET6 int autoflowlabel = 0; #endif #ifdef MAC struct label *maclabel; #endif struct syncache scs; struct ucred *cred; uint64_t tfo_response_cookie; unsigned int *tfo_pending = NULL; int tfo_cookie_valid = 0; int tfo_response_cookie_valid = 0; bool locked; INP_WLOCK_ASSERT(inp); /* listen socket */ KASSERT((th->th_flags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN, ("%s: unexpected tcp flags", __func__)); /* * Combine all so/tp operations very early to drop the INP lock as * soon as possible. */ so = *lsop; KASSERT(SOLISTENING(so), ("%s: %p not listening", __func__, so)); tp = sototcpcb(so); cred = crhold(so->so_cred); #ifdef INET6 if ((inc->inc_flags & INC_ISIPV6) && (inp->inp_flags & IN6P_AUTOFLOWLABEL)) autoflowlabel = 1; #endif ip_ttl = inp->inp_ip_ttl; ip_tos = inp->inp_ip_tos; win = so->sol_sbrcv_hiwat; ltflags = (tp->t_flags & (TF_NOOPT | TF_SIGNATURE)); if (V_tcp_fastopen_server_enable && IS_FASTOPEN(tp->t_flags) && (tp->t_tfo_pending != NULL) && (to->to_flags & TOF_FASTOPEN)) { /* * Limit the number of pending TFO connections to * approximately half of the queue limit. This prevents TFO * SYN floods from starving the service by filling the * listen queue with bogus TFO connections. */ if (atomic_fetchadd_int(tp->t_tfo_pending, 1) <= (so->sol_qlimit / 2)) { int result; result = tcp_fastopen_check_cookie(inc, to->to_tfo_cookie, to->to_tfo_len, &tfo_response_cookie); tfo_cookie_valid = (result > 0); tfo_response_cookie_valid = (result >= 0); } /* * Remember the TFO pending counter as it will have to be * decremented below if we don't make it to syncache_tfo_expand(). */ tfo_pending = tp->t_tfo_pending; } /* By the time we drop the lock these should no longer be used. */ so = NULL; tp = NULL; #ifdef MAC if (mac_syncache_init(&maclabel) != 0) { INP_WUNLOCK(inp); goto done; } else mac_syncache_create(maclabel, inp); #endif if (!tfo_cookie_valid) INP_WUNLOCK(inp); /* * Remember the IP options, if any. */ #ifdef INET6 if (!(inc->inc_flags & INC_ISIPV6)) #endif #ifdef INET ipopts = (m) ? ip_srcroute(m) : NULL; #else ipopts = NULL; #endif #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, check that received * SYN has signature and it is correct. If signature doesn't match * or TCP_SIGNATURE support isn't enabled, drop the packet. */ if (ltflags & TF_SIGNATURE) { if ((to->to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); goto done; } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to->to_signature) != 0) goto done; } #endif /* TCP_SIGNATURE */ /* * See if we already have an entry for this connection. * If we do, resend the SYN,ACK, and reset the retransmit timer. * * XXX: should the syncache be re-initialized with the contents * of the new SYN here (which may have different options?) * * XXX: We do not check the sequence number to see if this is a * real retransmit or a new connection attempt. The question is * how to handle such a case; either ignore it as spoofed, or * drop the current entry and create a new one? */ if (syncache_cookiesonly()) { sc = NULL; sch = syncache_hashbucket(inc); locked = false; } else { sc = syncache_lookup(inc, &sch); /* returns locked sch */ locked = true; SCH_LOCK_ASSERT(sch); } if (sc != NULL) { if (tfo_cookie_valid) INP_WUNLOCK(inp); TCPSTAT_INC(tcps_sc_dupsyn); if (ipopts) { /* * If we were remembering a previous source route, * forget it and use the new one we've been given. */ if (sc->sc_ipopts) (void) m_free(sc->sc_ipopts); sc->sc_ipopts = ipopts; } /* * Update timestamp if present. */ if ((sc->sc_flags & SCF_TIMESTAMP) && (to->to_flags & TOF_TS)) sc->sc_tsreflect = to->to_tsval; else sc->sc_flags &= ~SCF_TIMESTAMP; #ifdef MAC /* * Since we have already unconditionally allocated label * storage, free it up. The syncache entry will already * have an initialized label we can use. */ mac_syncache_destroy(&maclabel); #endif TCP_PROBE5(receive, NULL, NULL, m, NULL, th); /* Retransmit SYN|ACK and reset retransmit count. */ if ((s = tcp_log_addrs(&sc->sc_inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Received duplicate SYN, " "resetting timer and retransmitting SYN|ACK\n", s, __func__); free(s, M_TCPLOG); } if (syncache_respond(sc, m, TH_SYN|TH_ACK) == 0) { sc->sc_rxmits = 0; syncache_timeout(sc, sch, 1); TCPSTAT_INC(tcps_sndacks); TCPSTAT_INC(tcps_sndtotal); } SCH_UNLOCK(sch); goto donenoprobe; } if (tfo_cookie_valid) { bzero(&scs, sizeof(scs)); sc = &scs; goto skip_alloc; } /* * Skip allocating a syncache entry if we are just going to discard * it later. */ if (!locked) { bzero(&scs, sizeof(scs)); sc = &scs; } else sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); if (sc == NULL) { /* * The zone allocator couldn't provide more entries. * Treat this as if the cache was full; drop the oldest * entry and insert the new one. */ TCPSTAT_INC(tcps_sc_zonefail); if ((sc = TAILQ_LAST(&sch->sch_bucket, sch_head)) != NULL) { sch->sch_last_overflow = time_uptime; syncache_drop(sc, sch); syncache_pause(inc); } sc = uma_zalloc(V_tcp_syncache.zone, M_NOWAIT | M_ZERO); if (sc == NULL) { if (V_tcp_syncookies) { bzero(&scs, sizeof(scs)); sc = &scs; } else { KASSERT(locked, ("%s: bucket unexpectedly unlocked", __func__)); SCH_UNLOCK(sch); if (ipopts) (void) m_free(ipopts); goto done; } } } skip_alloc: if (!tfo_cookie_valid && tfo_response_cookie_valid) sc->sc_tfo_cookie = &tfo_response_cookie; /* * Fill in the syncache values. */ #ifdef MAC sc->sc_label = maclabel; #endif sc->sc_cred = cred; cred = NULL; sc->sc_ipopts = ipopts; bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); #ifdef INET6 if (!(inc->inc_flags & INC_ISIPV6)) #endif { sc->sc_ip_tos = ip_tos; sc->sc_ip_ttl = ip_ttl; } #ifdef TCP_OFFLOAD sc->sc_tod = tod; sc->sc_todctx = todctx; #endif sc->sc_irs = th->th_seq; sc->sc_flags = 0; sc->sc_flowlabel = 0; /* * Initial receive window: clip sbspace to [0 .. TCP_MAXWIN]. * win was derived from socket earlier in the function. */ win = imax(win, 0); win = imin(win, TCP_MAXWIN); sc->sc_wnd = win; if (V_tcp_do_rfc1323) { /* * A timestamp received in a SYN makes * it ok to send timestamp requests and replies. */ if (to->to_flags & TOF_TS) { sc->sc_tsreflect = to->to_tsval; sc->sc_flags |= SCF_TIMESTAMP; sc->sc_tsoff = tcp_new_ts_offset(inc); } if (to->to_flags & TOF_SCALE) { int wscale = 0; /* * Pick the smallest possible scaling factor that * will still allow us to scale up to sb_max, aka * kern.ipc.maxsockbuf. * * We do this because there are broken firewalls that * will corrupt the window scale option, leading to * the other endpoint believing that our advertised * window is unscaled. At scale factors larger than * 5 the unscaled window will drop below 1500 bytes, * leading to serious problems when traversing these * broken firewalls. * * With the default maxsockbuf of 256K, a scale factor * of 3 will be chosen by this algorithm. Those who * choose a larger maxsockbuf should watch out * for the compatibility problems mentioned above. * * RFC1323: The Window field in a SYN (i.e., a * or ) segment itself is never scaled. */ while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < sb_max) wscale++; sc->sc_requested_r_scale = wscale; sc->sc_requested_s_scale = to->to_wscale; sc->sc_flags |= SCF_WINSCALE; } } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* * If listening socket requested TCP digests, flag this in the * syncache so that syncache_respond() will do the right thing * with the SYN+ACK. */ if (ltflags & TF_SIGNATURE) sc->sc_flags |= SCF_SIGNATURE; #endif /* TCP_SIGNATURE */ if (to->to_flags & TOF_SACKPERM) sc->sc_flags |= SCF_SACK; if (to->to_flags & TOF_MSS) sc->sc_peer_mss = to->to_mss; /* peer mss may be zero */ if (ltflags & TF_NOOPT) sc->sc_flags |= SCF_NOOPT; if ((th->th_flags & (TH_ECE|TH_CWR)) && V_tcp_do_ecn) sc->sc_flags |= SCF_ECN; if (V_tcp_syncookies) sc->sc_iss = syncookie_generate(sch, sc); else sc->sc_iss = arc4random(); #ifdef INET6 if (autoflowlabel) { if (V_tcp_syncookies) sc->sc_flowlabel = sc->sc_iss; else sc->sc_flowlabel = ip6_randomflowlabel(); sc->sc_flowlabel = htonl(sc->sc_flowlabel) & IPV6_FLOWLABEL_MASK; } #endif if (locked) SCH_UNLOCK(sch); if (tfo_cookie_valid) { syncache_tfo_expand(sc, lsop, m, tfo_response_cookie); /* INP_WUNLOCK(inp) will be performed by the caller */ rv = 1; goto tfo_expanded; } TCP_PROBE5(receive, NULL, NULL, m, NULL, th); /* * Do a standard 3-way handshake. */ if (syncache_respond(sc, m, TH_SYN|TH_ACK) == 0) { if (V_tcp_syncookies && V_tcp_syncookiesonly && sc != &scs) syncache_free(sc); else if (sc != &scs) syncache_insert(sc, sch); /* locks and unlocks sch */ TCPSTAT_INC(tcps_sndacks); TCPSTAT_INC(tcps_sndtotal); } else { if (sc != &scs) syncache_free(sc); TCPSTAT_INC(tcps_sc_dropped); } goto donenoprobe; done: TCP_PROBE5(receive, NULL, NULL, m, NULL, th); donenoprobe: if (m) { *lsop = NULL; m_freem(m); } /* * If tfo_pending is not NULL here, then a TFO SYN that did not * result in a new socket was processed and the associated pending * counter has not yet been decremented. All such TFO processing paths * transit this point. */ if (tfo_pending != NULL) tcp_fastopen_decrement_counter(tfo_pending); tfo_expanded: if (cred != NULL) crfree(cred); #ifdef MAC if (sc == &scs) mac_syncache_destroy(&maclabel); #endif return (rv); } /* * Send SYN|ACK or ACK to the peer. Either in response to a peer's segment, * i.e. m0 != NULL, or upon 3WHS ACK timeout, i.e. m0 == NULL. */ static int syncache_respond(struct syncache *sc, const struct mbuf *m0, int flags) { struct ip *ip = NULL; struct mbuf *m; struct tcphdr *th = NULL; int optlen, error = 0; /* Make compiler happy */ u_int16_t hlen, tlen, mssopt; struct tcpopt to; #ifdef INET6 struct ip6_hdr *ip6 = NULL; #endif + + NET_EPOCH_ASSERT(); + hlen = #ifdef INET6 (sc->sc_inc.inc_flags & INC_ISIPV6) ? sizeof(struct ip6_hdr) : #endif sizeof(struct ip); tlen = hlen + sizeof(struct tcphdr); /* Determine MSS we advertize to other end of connection. */ mssopt = max(tcp_mssopt(&sc->sc_inc), V_tcp_minmss); /* XXX: Assume that the entire packet will fit in a header mbuf. */ KASSERT(max_linkhdr + tlen + TCP_MAXOLEN <= MHLEN, ("syncache: mbuf too small")); /* Create the IP+TCP header from scratch. */ m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); #ifdef MAC mac_syncache_create_mbuf(sc->sc_label, m); #endif m->m_data += max_linkhdr; m->m_len = tlen; m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_src = sc->sc_inc.inc6_laddr; ip6->ip6_dst = sc->sc_inc.inc6_faddr; ip6->ip6_plen = htons(tlen - hlen); /* ip6_hlim is set after checksum */ ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK; ip6->ip6_flow |= sc->sc_flowlabel; th = (struct tcphdr *)(ip6 + 1); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { ip = mtod(m, struct ip *); ip->ip_v = IPVERSION; ip->ip_hl = sizeof(struct ip) >> 2; ip->ip_len = htons(tlen); ip->ip_id = 0; ip->ip_off = 0; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = sc->sc_inc.inc_laddr; ip->ip_dst = sc->sc_inc.inc_faddr; ip->ip_ttl = sc->sc_ip_ttl; ip->ip_tos = sc->sc_ip_tos; /* * See if we should do MTU discovery. Route lookups are * expensive, so we will only unset the DF bit if: * * 1) path_mtu_discovery is disabled * 2) the SCF_UNREACH flag has been set */ if (V_path_mtu_discovery && ((sc->sc_flags & SCF_UNREACH) == 0)) ip->ip_off |= htons(IP_DF); th = (struct tcphdr *)(ip + 1); } #endif /* INET */ th->th_sport = sc->sc_inc.inc_lport; th->th_dport = sc->sc_inc.inc_fport; if (flags & TH_SYN) th->th_seq = htonl(sc->sc_iss); else th->th_seq = htonl(sc->sc_iss + 1); th->th_ack = htonl(sc->sc_irs + 1); th->th_off = sizeof(struct tcphdr) >> 2; th->th_x2 = 0; th->th_flags = flags; th->th_win = htons(sc->sc_wnd); th->th_urp = 0; if ((flags & TH_SYN) && (sc->sc_flags & SCF_ECN)) { th->th_flags |= TH_ECE; TCPSTAT_INC(tcps_ecn_shs); } /* Tack on the TCP options. */ if ((sc->sc_flags & SCF_NOOPT) == 0) { to.to_flags = 0; if (flags & TH_SYN) { to.to_mss = mssopt; to.to_flags = TOF_MSS; if (sc->sc_flags & SCF_WINSCALE) { to.to_wscale = sc->sc_requested_r_scale; to.to_flags |= TOF_SCALE; } if (sc->sc_flags & SCF_SACK) to.to_flags |= TOF_SACKPERM; #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif if (sc->sc_tfo_cookie) { to.to_flags |= TOF_FASTOPEN; to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; to.to_tfo_cookie = sc->sc_tfo_cookie; /* don't send cookie again when retransmitting response */ sc->sc_tfo_cookie = NULL; } } if (sc->sc_flags & SCF_TIMESTAMP) { to.to_tsval = sc->sc_tsoff + tcp_ts_getticks(); to.to_tsecr = sc->sc_tsreflect; to.to_flags |= TOF_TS; } optlen = tcp_addoptions(&to, (u_char *)(th + 1)); /* Adjust headers by option size. */ th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; m->m_len += optlen; m->m_pkthdr.len += optlen; #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) + optlen); else #endif ip->ip_len = htons(ntohs(ip->ip_len) + optlen); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (sc->sc_flags & SCF_SIGNATURE) { KASSERT(to.to_flags & TOF_SIGNATURE, ("tcp_addoptions() didn't set tcp_signature")); /* NOTE: to.to_signature is inside of mbuf */ if (!TCPMD5_ENABLED() || TCPMD5_OUTPUT(m, th, to.to_signature) != 0) { m_freem(m); return (EACCES); } } #endif } else optlen = 0; M_SETFIB(m, sc->sc_inc.inc_fibnum); m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); /* * If we have peer's SYN and it has a flowid, then let's assign it to * our SYN|ACK. ip6_output() and ip_output() will not assign flowid * to SYN|ACK due to lack of inp here. */ if (m0 != NULL && M_HASHTYPE_GET(m0) != M_HASHTYPE_NONE) { m->m_pkthdr.flowid = m0->m_pkthdr.flowid; M_HASHTYPE_SET(m, M_HASHTYPE_GET(m0)); } #ifdef INET6 if (sc->sc_inc.inc_flags & INC_ISIPV6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; th->th_sum = in6_cksum_pseudo(ip6, tlen + optlen - hlen, IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(NULL, NULL); #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; error = tod->tod_syncache_respond(tod, sc->sc_todctx, m); return (error); } #endif TCP_PROBE5(send, NULL, NULL, ip6, NULL, th); error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(tlen + optlen - hlen + IPPROTO_TCP)); #ifdef TCP_OFFLOAD if (ADDED_BY_TOE(sc)) { struct toedev *tod = sc->sc_tod; error = tod->tod_syncache_respond(tod, sc->sc_todctx, m); return (error); } #endif TCP_PROBE5(send, NULL, NULL, ip, NULL, th); error = ip_output(m, sc->sc_ipopts, NULL, 0, NULL, NULL); } #endif return (error); } /* * The purpose of syncookies is to handle spoofed SYN flooding DoS attacks * that exceed the capacity of the syncache by avoiding the storage of any * of the SYNs we receive. Syncookies defend against blind SYN flooding * attacks where the attacker does not have access to our responses. * * Syncookies encode and include all necessary information about the * connection setup within the SYN|ACK that we send back. That way we * can avoid keeping any local state until the ACK to our SYN|ACK returns * (if ever). Normally the syncache and syncookies are running in parallel * with the latter taking over when the former is exhausted. When matching * syncache entry is found the syncookie is ignored. * * The only reliable information persisting the 3WHS is our initial sequence * number ISS of 32 bits. Syncookies embed a cryptographically sufficient * strong hash (MAC) value and a few bits of TCP SYN options in the ISS * of our SYN|ACK. The MAC can be recomputed when the ACK to our SYN|ACK * returns and signifies a legitimate connection if it matches the ACK. * * The available space of 32 bits to store the hash and to encode the SYN * option information is very tight and we should have at least 24 bits for * the MAC to keep the number of guesses by blind spoofing reasonably high. * * SYN option information we have to encode to fully restore a connection: * MSS: is imporant to chose an optimal segment size to avoid IP level * fragmentation along the path. The common MSS values can be encoded * in a 3-bit table. Uncommon values are captured by the next lower value * in the table leading to a slight increase in packetization overhead. * WSCALE: is necessary to allow large windows to be used for high delay- * bandwidth product links. Not scaling the window when it was initially * negotiated is bad for performance as lack of scaling further decreases * the apparent available send window. We only need to encode the WSCALE * we received from the remote end. Our end can be recalculated at any * time. The common WSCALE values can be encoded in a 3-bit table. * Uncommon values are captured by the next lower value in the table * making us under-estimate the available window size halving our * theoretically possible maximum throughput for that connection. * SACK: Greatly assists in packet loss recovery and requires 1 bit. * TIMESTAMP and SIGNATURE is not encoded because they are permanent options * that are included in all segments on a connection. We enable them when * the ACK has them. * * Security of syncookies and attack vectors: * * The MAC is computed over (faddr||laddr||fport||lport||irs||flags||secmod) * together with the gloabl secret to make it unique per connection attempt. * Thus any change of any of those parameters results in a different MAC output * in an unpredictable way unless a collision is encountered. 24 bits of the * MAC are embedded into the ISS. * * To prevent replay attacks two rotating global secrets are updated with a * new random value every 15 seconds. The life-time of a syncookie is thus * 15-30 seconds. * * Vector 1: Attacking the secret. This requires finding a weakness in the * MAC itself or the way it is used here. The attacker can do a chosen plain * text attack by varying and testing the all parameters under his control. * The strength depends on the size and randomness of the secret, and the * cryptographic security of the MAC function. Due to the constant updating * of the secret the attacker has at most 29.999 seconds to find the secret * and launch spoofed connections. After that he has to start all over again. * * Vector 2: Collision attack on the MAC of a single ACK. With a 24 bit MAC * size an average of 4,823 attempts are required for a 50% chance of success * to spoof a single syncookie (birthday collision paradox). However the * attacker is blind and doesn't know if one of his attempts succeeded unless * he has a side channel to interfere success from. A single connection setup * success average of 90% requires 8,790 packets, 99.99% requires 17,578 packets. * This many attempts are required for each one blind spoofed connection. For * every additional spoofed connection he has to launch another N attempts. * Thus for a sustained rate 100 spoofed connections per second approximately * 1,800,000 packets per second would have to be sent. * * NB: The MAC function should be fast so that it doesn't become a CPU * exhaustion attack vector itself. * * References: * RFC4987 TCP SYN Flooding Attacks and Common Mitigations * SYN cookies were first proposed by cryptographer Dan J. Bernstein in 1996 * http://cr.yp.to/syncookies.html (overview) * http://cr.yp.to/syncookies/archive (details) * * * Schematic construction of a syncookie enabled Initial Sequence Number: * 0 1 2 3 * 12345678901234567890123456789012 * |xxxxxxxxxxxxxxxxxxxxxxxxWWWMMMSP| * * x 24 MAC (truncated) * W 3 Send Window Scale index * M 3 MSS index * S 1 SACK permitted * P 1 Odd/even secret */ /* * Distribution and probability of certain MSS values. Those in between are * rounded down to the next lower one. * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011] * .2% .3% 5% 7% 7% 20% 15% 45% */ static int tcp_sc_msstab[] = { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 }; /* * Distribution and probability of certain WSCALE values. We have to map the * (send) window scale (shift) option with a range of 0-14 from 4 bits into 3 * bits based on prevalence of certain values. Where we don't have an exact * match for are rounded down to the next lower one letting us under-estimate * the true available window. At the moment this would happen only for the * very uncommon values 3, 5 and those above 8 (more than 16MB socket buffer * and window size). The absence of the WSCALE option (no scaling in either * direction) is encoded with index zero. * [WSCALE values histograms, Allman, 2012] * X 10 10 35 5 6 14 10% by host * X 11 4 5 5 18 49 3% by connections */ static int tcp_sc_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 }; /* * Compute the MAC for the SYN cookie. SIPHASH-2-4 is chosen for its speed * and good cryptographic properties. */ static uint32_t syncookie_mac(struct in_conninfo *inc, tcp_seq irs, uint8_t flags, uint8_t *secbits, uintptr_t secmod) { SIPHASH_CTX ctx; uint32_t siphash[2]; SipHash24_Init(&ctx); SipHash_SetKey(&ctx, secbits); switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: SipHash_Update(&ctx, &inc->inc_faddr, sizeof(inc->inc_faddr)); SipHash_Update(&ctx, &inc->inc_laddr, sizeof(inc->inc_laddr)); break; #endif #ifdef INET6 case INC_ISIPV6: SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(inc->inc6_faddr)); SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(inc->inc6_laddr)); break; #endif } SipHash_Update(&ctx, &inc->inc_fport, sizeof(inc->inc_fport)); SipHash_Update(&ctx, &inc->inc_lport, sizeof(inc->inc_lport)); SipHash_Update(&ctx, &irs, sizeof(irs)); SipHash_Update(&ctx, &flags, sizeof(flags)); SipHash_Update(&ctx, &secmod, sizeof(secmod)); SipHash_Final((u_int8_t *)&siphash, &ctx); return (siphash[0] ^ siphash[1]); } static tcp_seq syncookie_generate(struct syncache_head *sch, struct syncache *sc) { u_int i, secbit, wscale; uint32_t iss, hash; uint8_t *secbits; union syncookie cookie; cookie.cookie = 0; /* Map our computed MSS into the 3-bit index. */ for (i = nitems(tcp_sc_msstab) - 1; tcp_sc_msstab[i] > sc->sc_peer_mss && i > 0; i--) ; cookie.flags.mss_idx = i; /* * Map the send window scale into the 3-bit index but only if * the wscale option was received. */ if (sc->sc_flags & SCF_WINSCALE) { wscale = sc->sc_requested_s_scale; for (i = nitems(tcp_sc_wstab) - 1; tcp_sc_wstab[i] > wscale && i > 0; i--) ; cookie.flags.wscale_idx = i; } /* Can we do SACK? */ if (sc->sc_flags & SCF_SACK) cookie.flags.sack_ok = 1; /* Which of the two secrets to use. */ secbit = V_tcp_syncache.secret.oddeven & 0x1; cookie.flags.odd_even = secbit; secbits = V_tcp_syncache.secret.key[secbit]; hash = syncookie_mac(&sc->sc_inc, sc->sc_irs, cookie.cookie, secbits, (uintptr_t)sch); /* * Put the flags into the hash and XOR them to get better ISS number * variance. This doesn't enhance the cryptographic strength and is * done to prevent the 8 cookie bits from showing up directly on the * wire. */ iss = hash & ~0xff; iss |= cookie.cookie ^ (hash >> 24); TCPSTAT_INC(tcps_sc_sendcookie); return (iss); } static struct syncache * syncookie_lookup(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso) { uint32_t hash; uint8_t *secbits; tcp_seq ack, seq; int wnd, wscale = 0; union syncookie cookie; /* * Pull information out of SYN-ACK/ACK and revert sequence number * advances. */ ack = th->th_ack - 1; seq = th->th_seq - 1; /* * Unpack the flags containing enough information to restore the * connection. */ cookie.cookie = (ack & 0xff) ^ (ack >> 24); /* Which of the two secrets to use. */ secbits = V_tcp_syncache.secret.key[cookie.flags.odd_even]; hash = syncookie_mac(inc, seq, cookie.cookie, secbits, (uintptr_t)sch); /* The recomputed hash matches the ACK if this was a genuine cookie. */ if ((ack & ~0xff) != (hash & ~0xff)) return (NULL); /* Fill in the syncache values. */ sc->sc_flags = 0; bcopy(inc, &sc->sc_inc, sizeof(struct in_conninfo)); sc->sc_ipopts = NULL; sc->sc_irs = seq; sc->sc_iss = ack; switch (inc->inc_flags & INC_ISIPV6) { #ifdef INET case 0: sc->sc_ip_ttl = sotoinpcb(lso)->inp_ip_ttl; sc->sc_ip_tos = sotoinpcb(lso)->inp_ip_tos; break; #endif #ifdef INET6 case INC_ISIPV6: if (sotoinpcb(lso)->inp_flags & IN6P_AUTOFLOWLABEL) sc->sc_flowlabel = sc->sc_iss & IPV6_FLOWLABEL_MASK; break; #endif } sc->sc_peer_mss = tcp_sc_msstab[cookie.flags.mss_idx]; /* We can simply recompute receive window scale we sent earlier. */ while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < sb_max) wscale++; /* Only use wscale if it was enabled in the orignal SYN. */ if (cookie.flags.wscale_idx > 0) { sc->sc_requested_r_scale = wscale; sc->sc_requested_s_scale = tcp_sc_wstab[cookie.flags.wscale_idx]; sc->sc_flags |= SCF_WINSCALE; } wnd = lso->sol_sbrcv_hiwat; wnd = imax(wnd, 0); wnd = imin(wnd, TCP_MAXWIN); sc->sc_wnd = wnd; if (cookie.flags.sack_ok) sc->sc_flags |= SCF_SACK; if (to->to_flags & TOF_TS) { sc->sc_flags |= SCF_TIMESTAMP; sc->sc_tsreflect = to->to_tsval; sc->sc_tsoff = tcp_new_ts_offset(inc); } if (to->to_flags & TOF_SIGNATURE) sc->sc_flags |= SCF_SIGNATURE; sc->sc_rxmits = 0; TCPSTAT_INC(tcps_sc_recvcookie); return (sc); } #ifdef INVARIANTS static int syncookie_cmp(struct in_conninfo *inc, struct syncache_head *sch, struct syncache *sc, struct tcphdr *th, struct tcpopt *to, struct socket *lso) { struct syncache scs, *scx; char *s; bzero(&scs, sizeof(scs)); scx = syncookie_lookup(inc, sch, &scs, th, to, lso); if ((s = tcp_log_addrs(inc, th, NULL, NULL)) == NULL) return (0); if (scx != NULL) { if (sc->sc_peer_mss != scx->sc_peer_mss) log(LOG_DEBUG, "%s; %s: mss different %i vs %i\n", s, __func__, sc->sc_peer_mss, scx->sc_peer_mss); if (sc->sc_requested_r_scale != scx->sc_requested_r_scale) log(LOG_DEBUG, "%s; %s: rwscale different %i vs %i\n", s, __func__, sc->sc_requested_r_scale, scx->sc_requested_r_scale); if (sc->sc_requested_s_scale != scx->sc_requested_s_scale) log(LOG_DEBUG, "%s; %s: swscale different %i vs %i\n", s, __func__, sc->sc_requested_s_scale, scx->sc_requested_s_scale); if ((sc->sc_flags & SCF_SACK) != (scx->sc_flags & SCF_SACK)) log(LOG_DEBUG, "%s; %s: SACK different\n", s, __func__); } if (s != NULL) free(s, M_TCPLOG); return (0); } #endif /* INVARIANTS */ static void syncookie_reseed(void *arg) { struct tcp_syncache *sc = arg; uint8_t *secbits; int secbit; /* * Reseeding the secret doesn't have to be protected by a lock. * It only must be ensured that the new random values are visible * to all CPUs in a SMP environment. The atomic with release * semantics ensures that. */ secbit = (sc->secret.oddeven & 0x1) ? 0 : 1; secbits = sc->secret.key[secbit]; arc4rand(secbits, SYNCOOKIE_SECRET_SIZE, 0); atomic_add_rel_int(&sc->secret.oddeven, 1); /* Reschedule ourself. */ callout_schedule(&sc->secret.reseed, SYNCOOKIE_LIFETIME * hz); } /* * We have overflowed a bucket. Let's pause dealing with the syncache. * This function will increment the bucketoverflow statistics appropriately * (once per pause when pausing is enabled; otherwise, once per overflow). */ static void syncache_pause(struct in_conninfo *inc) { time_t delta; const char *s; /* XXX: * 2. Add sysctl read here so we don't get the benefit of this * change without the new sysctl. */ /* * Try an unlocked read. If we already know that another thread * has activated the feature, there is no need to proceed. */ if (V_tcp_syncache.paused) return; /* Are cookied enabled? If not, we can't pause. */ if (!V_tcp_syncookies) { TCPSTAT_INC(tcps_sc_bucketoverflow); return; } /* * We may be the first thread to find an overflow. Get the lock * and evaluate if we need to take action. */ mtx_lock(&V_tcp_syncache.pause_mtx); if (V_tcp_syncache.paused) { mtx_unlock(&V_tcp_syncache.pause_mtx); return; } /* Activate protection. */ V_tcp_syncache.paused = true; TCPSTAT_INC(tcps_sc_bucketoverflow); /* * Determine the last backoff time. If we are seeing a re-newed * attack within that same time after last reactivating the syncache, * consider it an extension of the same attack. */ delta = TCP_SYNCACHE_PAUSE_TIME << V_tcp_syncache.pause_backoff; if (V_tcp_syncache.pause_until + delta - time_uptime > 0) { if (V_tcp_syncache.pause_backoff < TCP_SYNCACHE_MAX_BACKOFF) { delta <<= 1; V_tcp_syncache.pause_backoff++; } } else { delta = TCP_SYNCACHE_PAUSE_TIME; V_tcp_syncache.pause_backoff = 0; } /* Log a warning, including IP addresses, if able. */ if (inc != NULL) s = tcp_log_addrs(inc, NULL, NULL, NULL); else s = (const char *)NULL; log(LOG_WARNING, "TCP syncache overflow detected; using syncookies for " "the next %lld seconds%s%s%s\n", (long long)delta, (s != NULL) ? " (last SYN: " : "", (s != NULL) ? s : "", (s != NULL) ? ")" : ""); free(__DECONST(void *, s), M_TCPLOG); /* Use the calculated delta to set a new pause time. */ V_tcp_syncache.pause_until = time_uptime + delta; callout_reset(&V_tcp_syncache.pause_co, delta * hz, syncache_unpause, &V_tcp_syncache); mtx_unlock(&V_tcp_syncache.pause_mtx); } /* Evaluate whether we need to unpause. */ static void syncache_unpause(void *arg) { struct tcp_syncache *sc; time_t delta; sc = arg; mtx_assert(&sc->pause_mtx, MA_OWNED | MA_NOTRECURSED); callout_deactivate(&sc->pause_co); /* * Check to make sure we are not running early. If the pause * time has expired, then deactivate the protection. */ if ((delta = sc->pause_until - time_uptime) > 0) callout_schedule(&sc->pause_co, delta * hz); else sc->paused = false; } /* * Exports the syncache entries to userland so that netstat can display * them alongside the other sockets. This function is intended to be * called only from tcp_pcblist. * * Due to concurrency on an active system, the number of pcbs exported * may have no relation to max_pcbs. max_pcbs merely indicates the * amount of space the caller allocated for this function to use. */ int syncache_pcblist(struct sysctl_req *req) { struct xtcpcb xt; struct syncache *sc; struct syncache_head *sch; int error, i; bzero(&xt, sizeof(xt)); xt.xt_len = sizeof(xt); xt.t_state = TCPS_SYN_RECEIVED; xt.xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; xt.xt_inp.xi_socket.xso_len = sizeof (struct xsocket); xt.xt_inp.xi_socket.so_type = SOCK_STREAM; xt.xt_inp.xi_socket.so_state = SS_ISCONNECTING; for (i = 0; i < V_tcp_syncache.hashsize; i++) { sch = &V_tcp_syncache.hashbase[i]; SCH_LOCK(sch); TAILQ_FOREACH(sc, &sch->sch_bucket, sc_hash) { if (cr_cansee(req->td->td_ucred, sc->sc_cred) != 0) continue; if (sc->sc_inc.inc_flags & INC_ISIPV6) xt.xt_inp.inp_vflag = INP_IPV6; else xt.xt_inp.inp_vflag = INP_IPV4; bcopy(&sc->sc_inc, &xt.xt_inp.inp_inc, sizeof (struct in_conninfo)); error = SYSCTL_OUT(req, &xt, sizeof xt); if (error) { SCH_UNLOCK(sch); return (0); } } SCH_UNLOCK(sch); } return (0); } Index: head/sys/netinet6/mld6.c =================================================================== --- head/sys/netinet6/mld6.c (revision 356968) +++ head/sys/netinet6/mld6.c (revision 356969) @@ -1,3349 +1,3350 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2009 Bruce Simpson. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote * products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ */ /*- * Copyright (c) 1988 Stephen Deering. * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)igmp.c 8.1 (Berkeley) 7/19/93 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef KTR_MLD #define KTR_MLD KTR_INET6 #endif static struct mld_ifsoftc * mli_alloc_locked(struct ifnet *); static void mli_delete_locked(const struct ifnet *); static void mld_dispatch_packet(struct mbuf *); static void mld_dispatch_queue(struct mbufq *, int); static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *); static void mld_fasttimo_vnet(struct in6_multi_head *inmh); static int mld_handle_state_change(struct in6_multi *, struct mld_ifsoftc *); static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *, const int); #ifdef KTR static char * mld_rec_type_to_str(const int); #endif static void mld_set_version(struct mld_ifsoftc *, const int); static void mld_slowtimo_vnet(void); static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, /*const*/ struct mld_hdr *); static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, /*const*/ struct mld_hdr *); static void mld_v1_process_group_timer(struct in6_multi_head *, struct in6_multi *); static void mld_v1_process_querier_timers(struct mld_ifsoftc *); static int mld_v1_transmit_report(struct in6_multi *, const int); static void mld_v1_update_group(struct in6_multi *, const int); static void mld_v2_cancel_link_timers(struct mld_ifsoftc *); static void mld_v2_dispatch_general_query(struct mld_ifsoftc *); static struct mbuf * mld_v2_encap_report(struct ifnet *, struct mbuf *); static int mld_v2_enqueue_filter_change(struct mbufq *, struct in6_multi *); static int mld_v2_enqueue_group_record(struct mbufq *, struct in6_multi *, const int, const int, const int, const int); static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, struct mbuf *, struct mldv2_query *, const int, const int); static int mld_v2_merge_state_changes(struct in6_multi *, struct mbufq *); static void mld_v2_process_group_timers(struct in6_multi_head *, struct mbufq *, struct mbufq *, struct in6_multi *, const int); static int mld_v2_process_group_query(struct in6_multi *, struct mld_ifsoftc *mli, int, struct mbuf *, struct mldv2_query *, const int); static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); /* * Normative references: RFC 2710, RFC 3590, RFC 3810. * * Locking: * * The MLD subsystem lock ends up being system-wide for the moment, * but could be per-VIMAGE later on. * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. * Any may be taken independently; if any are held at the same * time, the above lock order must be followed. * * IN6_MULTI_LOCK covers in_multi. * * MLD_LOCK covers per-link state and any global variables in this file. * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of * per-link state iterators. * * XXX LOR PREVENTION * A special case for IPv6 is the in6_setscope() routine. ip6_output() * will not accept an ifp; it wants an embedded scope ID, unlike * ip_output(), which happily takes the ifp given to it. The embedded * scope ID is only used by MLD to select the outgoing interface. * * During interface attach and detach, MLD will take MLD_LOCK *after* * the IF_AFDATA_LOCK. * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call * it with MLD_LOCK held without triggering an LOR. A netisr with indirect * dispatch could work around this, but we'd rather not do that, as it * can introduce other races. * * As such, we exploit the fact that the scope ID is just the interface * index, and embed it in the IPv6 destination address accordingly. * This is potentially NOT VALID for MLDv1 reports, as they * are always sent to the multicast group itself; as MLDv2 * reports are always sent to ff02::16, this is not an issue * when MLDv2 is in use. * * This does not however eliminate the LOR when ip6_output() itself * calls in6_setscope() internally whilst MLD_LOCK is held. This will * trigger a LOR warning in WITNESS when the ifnet is detached. * * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given * how it's used across the network stack. Here we're simply exploiting * the fact that MLD runs at a similar layer in the stack to scope6.c. * * VIMAGE: * * Each in6_multi corresponds to an ifp, and each ifp corresponds * to a vnet in ifp->if_vnet. */ static struct mtx mld_mtx; static MALLOC_DEFINE(M_MLD, "mld", "mld state"); #define MLD_EMBEDSCOPE(pin6, zoneid) \ if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ /* * VIMAGE-wide globals. */ VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0}; VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head); VNET_DEFINE_STATIC(int, interface_timers_running6); VNET_DEFINE_STATIC(int, state_change_timers_running6); VNET_DEFINE_STATIC(int, current_state_timers_running6); #define V_mld_gsrdelay VNET(mld_gsrdelay) #define V_mli_head VNET(mli_head) #define V_interface_timers_running6 VNET(interface_timers_running6) #define V_state_change_timers_running6 VNET(state_change_timers_running6) #define V_current_state_timers_running6 VNET(current_state_timers_running6) SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, "IPv6 Multicast Listener Discovery"); /* * Virtualized sysctls. */ SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", "Rate limit for MLDv2 Group-and-Source queries in seconds"); /* * Non-virtualized sysctls. */ static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, "Per-interface MLDv2 state"); static int mld_v1enable = 1; SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN, &mld_v1enable, 0, "Enable fallback to MLDv1"); static int mld_v2enable = 1; SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN, &mld_v2enable, 0, "Enable MLDv2"); static int mld_use_allow = 1; SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN, &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); /* * Packed Router Alert option structure declaration. */ struct mld_raopt { struct ip6_hbh hbh; struct ip6_opt pad; struct ip6_opt_router ra; } __packed; /* * Router Alert hop-by-hop option header. */ static struct mld_raopt mld_ra = { .hbh = { 0, 0 }, .pad = { .ip6o_type = IP6OPT_PADN, 0 }, .ra = { .ip6or_type = IP6OPT_ROUTER_ALERT, .ip6or_len = IP6OPT_RTALERT_LEN - 2, .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) } }; static struct ip6_pktopts mld_po; static __inline void mld_save_context(struct mbuf *m, struct ifnet *ifp) { #ifdef VIMAGE m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; #endif /* VIMAGE */ m->m_pkthdr.flowid = ifp->if_index; } static __inline void mld_scrub_context(struct mbuf *m) { m->m_pkthdr.PH_loc.ptr = NULL; m->m_pkthdr.flowid = 0; } /* * Restore context from a queued output chain. * Return saved ifindex. * * VIMAGE: The assertion is there to make sure that we * actually called CURVNET_SET() with what's in the mbuf chain. */ static __inline uint32_t mld_restore_context(struct mbuf *m) { #if defined(VIMAGE) && defined(INVARIANTS) KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, ("%s: called when curvnet was not restored: cuvnet %p m ptr %p", __func__, curvnet, m->m_pkthdr.PH_loc.ptr)); #endif return (m->m_pkthdr.flowid); } /* * Retrieve or set threshold between group-source queries in seconds. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by MLD lock. */ static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) { int error; int i; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); MLD_LOCK(); i = V_mld_gsrdelay.tv_sec; error = sysctl_handle_int(oidp, &i, 0, req); if (error || !req->newptr) goto out_locked; if (i < -1 || i >= 60) { error = EINVAL; goto out_locked; } CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", V_mld_gsrdelay.tv_sec, i); V_mld_gsrdelay.tv_sec = i; out_locked: MLD_UNLOCK(); return (error); } /* * Expose struct mld_ifsoftc to userland, keyed by ifindex. * For use by ifmcstat(8). * * SMPng: NOTE: Does an unlocked ifindex space read. * VIMAGE: Assume curvnet set by caller. The node handler itself * is not directly virtualized. */ static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) { int *name; int error; u_int namelen; struct ifnet *ifp; struct mld_ifsoftc *mli; name = (int *)arg1; namelen = arg2; if (req->newptr != NULL) return (EPERM); if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); if (error) return (error); IN6_MULTI_LOCK(); IN6_MULTI_LIST_LOCK(); MLD_LOCK(); if (name[0] <= 0 || name[0] > V_if_index) { error = ENOENT; goto out_locked; } error = ENOENT; ifp = ifnet_byindex(name[0]); if (ifp == NULL) goto out_locked; LIST_FOREACH(mli, &V_mli_head, mli_link) { if (ifp == mli->mli_ifp) { struct mld_ifinfo info; info.mli_version = mli->mli_version; info.mli_v1_timer = mli->mli_v1_timer; info.mli_v2_timer = mli->mli_v2_timer; info.mli_flags = mli->mli_flags; info.mli_rv = mli->mli_rv; info.mli_qi = mli->mli_qi; info.mli_qri = mli->mli_qri; info.mli_uri = mli->mli_uri; error = SYSCTL_OUT(req, &info, sizeof(info)); break; } } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); IN6_MULTI_UNLOCK(); return (error); } /* * Dispatch an entire queue of pending packet chains. * VIMAGE: Assumes the vnet pointer has been set. */ static void mld_dispatch_queue(struct mbufq *mq, int limit) { struct mbuf *m; while ((m = mbufq_dequeue(mq)) != NULL) { CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m); mld_dispatch_packet(m); if (--limit == 0) break; } } /* * Filter outgoing MLD report state by group. * * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) * and node-local addresses. However, kernel and socket consumers * always embed the KAME scope ID in the address provided, so strip it * when performing comparison. * Note: This is not the same as the *multicast* scope. * * Return zero if the given group is one for which MLD reports * should be suppressed, or non-zero if reports should be issued. */ static __inline int mld_is_addr_reported(const struct in6_addr *addr) { KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) return (0); if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { struct in6_addr tmp = *addr; in6_clearscope(&tmp); if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) return (0); } return (1); } /* * Attach MLD when PF_INET6 is attached to an interface. * * SMPng: Normally called with IF_AFDATA_LOCK held. */ struct mld_ifsoftc * mld_domifattach(struct ifnet *ifp) { struct mld_ifsoftc *mli; CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK(); mli = mli_alloc_locked(ifp); if (!(ifp->if_flags & IFF_MULTICAST)) mli->mli_flags |= MLIF_SILENT; if (mld_use_allow) mli->mli_flags |= MLIF_USEALLOW; MLD_UNLOCK(); return (mli); } /* * VIMAGE: assume curvnet set by caller. */ static struct mld_ifsoftc * mli_alloc_locked(/*const*/ struct ifnet *ifp) { struct mld_ifsoftc *mli; MLD_LOCK_ASSERT(); mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO); if (mli == NULL) goto out; mli->mli_ifp = ifp; mli->mli_version = MLD_VERSION_2; mli->mli_flags = 0; mli->mli_rv = MLD_RV_INIT; mli->mli_qi = MLD_QI_INIT; mli->mli_qri = MLD_QRI_INIT; mli->mli_uri = MLD_URI_INIT; mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)", ifp, if_name(ifp)); out: return (mli); } /* * Hook for ifdetach. * * NOTE: Some finalization tasks need to run before the protocol domain * is detached, but also before the link layer does its cleanup. * Run before link-layer cleanup; cleanup groups, but do not free MLD state. * * SMPng: Caller must hold IN6_MULTI_LOCK(). * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. * XXX This routine is also bitten by unlocked ifma_protospec access. */ void mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh) { struct epoch_tracker et; struct mld_ifsoftc *mli; struct ifmultiaddr *ifma; struct in6_multi *inm; CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK(); mli = MLD_IFINFO(ifp); IF_ADDR_WLOCK(ifp); /* * Extract list of in6_multi associated with the detaching ifp * which the PF_INET6 layer is about to release. */ NET_EPOCH_ENTER(et); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { inm = in6m_ifmultiaddr_get_inm(ifma); if (inm == NULL) continue; in6m_disconnect_locked(inmh, inm); if (mli->mli_version == MLD_VERSION_2) { in6m_clear_recorded(inm); /* * We need to release the final reference held * for issuing the INCLUDE {}. */ if (inm->in6m_state == MLD_LEAVING_MEMBER) { inm->in6m_state = MLD_NOT_MEMBER; in6m_rele_locked(inmh, inm); } } } NET_EPOCH_EXIT(et); IF_ADDR_WUNLOCK(ifp); MLD_UNLOCK(); } /* * Hook for domifdetach. * Runs after link-layer cleanup; free MLD state. * * SMPng: Normally called with IF_AFDATA_LOCK held. */ void mld_domifdetach(struct ifnet *ifp) { CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK(); mli_delete_locked(ifp); MLD_UNLOCK(); } static void mli_delete_locked(const struct ifnet *ifp) { struct mld_ifsoftc *mli, *tmli; CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)", __func__, ifp, if_name(ifp)); MLD_LOCK_ASSERT(); LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { if (mli->mli_ifp == ifp) { /* * Free deferred General Query responses. */ mbufq_drain(&mli->mli_gq); LIST_REMOVE(mli, mli_link); free(mli, M_MLD); return; } } } /* * Process a received MLDv1 general or address-specific query. * Assumes that the query header has been pulled up to sizeof(mld_hdr). * * NOTE: Can't be fully const correct as we temporarily embed scope ID in * mld_addr. This is OK as we own the mbuf chain. */ static int mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld) { struct ifmultiaddr *ifma; struct mld_ifsoftc *mli; struct in6_multi *inm; int is_general_query; uint16_t timer; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif NET_EPOCH_ASSERT(); is_general_query = 0; if (!mld_v1enable) { CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); return (0); } /* * RFC3810 Section 6.2: MLD queries must originate from * a router's link-local address. */ if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (0); } /* * Do address field validation upfront before we accept * the query. */ if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { /* * MLDv1 General Query. * If this was not sent to the all-nodes group, ignore it. */ struct in6_addr dst; dst = ip6->ip6_dst; in6_clearscope(&dst); if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) return (EINVAL); is_general_query = 1; } else { /* * Embed scope ID of receiving interface in MLD query for * lookup whilst we don't hold other locks. */ in6_setscope(&mld->mld_addr, ifp, NULL); } IN6_MULTI_LIST_LOCK(); MLD_LOCK(); /* * Switch to MLDv1 host compatibility mode. */ mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); mld_set_version(mli, MLD_VERSION_1); timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; if (timer == 0) timer = 1; if (is_general_query) { /* * For each reporting group joined on this * interface, kick the report timer. */ CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", ifp, if_name(ifp)); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { inm = in6m_ifmultiaddr_get_inm(ifma); if (inm == NULL) continue; mld_v1_update_group(inm, timer); } } else { /* * MLDv1 Group-Specific Query. * If this is a group-specific MLDv1 query, we need only * look up the single group to process it. */ inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm != NULL) { CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); mld_v1_update_group(inm, timer); } /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); } MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); return (0); } /* * Update the report timer on a group in response to an MLDv1 query. * * If we are becoming the reporting member for this group, start the timer. * If we already are the reporting member for this group, and timer is * below the threshold, reset it. * * We may be updating the group for the first time since we switched * to MLDv2. If we are, then we must clear any recorded source lists, * and transition to REPORTING state; the group timer is overloaded * for group and group-source query responses. * * Unlike MLDv2, the delay per group should be jittered * to avoid bursts of MLDv1 reports. */ static void mld_v1_update_group(struct in6_multi *inm, const int timer) { #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp), timer); IN6_MULTI_LIST_LOCK_ASSERT(); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: break; case MLD_REPORTING_MEMBER: if (inm->in6m_timer != 0 && inm->in6m_timer <= timer) { CTR1(KTR_MLD, "%s: REPORTING and timer running, " "skipping.", __func__); break; } /* FALLTHROUGH */ case MLD_SG_QUERY_PENDING_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: CTR1(KTR_MLD, "%s: ->REPORTING", __func__); inm->in6m_state = MLD_REPORTING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; break; case MLD_SLEEPING_MEMBER: CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); inm->in6m_state = MLD_AWAKENING_MEMBER; break; case MLD_LEAVING_MEMBER: break; } } /* * Process a received MLDv2 general, group-specific or * group-and-source-specific query. * * Assumes that mld points to a struct mldv2_query which is stored in * contiguous memory. * * Return 0 if successful, otherwise an appropriate error code is returned. */ static int mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len) { struct mld_ifsoftc *mli; struct in6_multi *inm; uint32_t maxdelay, nsrc, qqi; int is_general_query; uint16_t timer; uint8_t qrv; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif NET_EPOCH_ASSERT(); if (!mld_v2enable) { CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (0); } /* * RFC3810 Section 6.2: MLD queries must originate from * a router's link-local address. */ if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (0); } is_general_query = 0; CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ if (maxdelay >= 32768) { maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << (MLD_MRC_EXP(maxdelay) + 3); } timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; if (timer == 0) timer = 1; qrv = MLD_QRV(mld->mld_misc); if (qrv < 2) { CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, qrv, MLD_RV_INIT); qrv = MLD_RV_INIT; } qqi = mld->mld_qqi; if (qqi >= 128) { qqi = MLD_QQIC_MANT(mld->mld_qqi) << (MLD_QQIC_EXP(mld->mld_qqi) + 3); } nsrc = ntohs(mld->mld_numsrc); if (nsrc > MLD_MAX_GS_SOURCES) return (EMSGSIZE); if (icmp6len < sizeof(struct mldv2_query) + (nsrc * sizeof(struct in6_addr))) return (EMSGSIZE); /* * Do further input validation upfront to avoid resetting timers * should we need to discard this query. */ if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { /* * A general query with a source list has undefined * behaviour; discard it. */ if (nsrc > 0) return (EINVAL); is_general_query = 1; } else { /* * Embed scope ID of receiving interface in MLD query for * lookup whilst we don't hold other locks (due to KAME * locking lameness). We own this mbuf chain just now. */ in6_setscope(&mld->mld_addr, ifp, NULL); } IN6_MULTI_LIST_LOCK(); MLD_LOCK(); mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); /* * Discard the v2 query if we're in Compatibility Mode. * The RFC is pretty clear that hosts need to stay in MLDv1 mode * until the Old Version Querier Present timer expires. */ if (mli->mli_version != MLD_VERSION_2) goto out_locked; mld_set_version(mli, MLD_VERSION_2); mli->mli_rv = qrv; mli->mli_qi = qqi; mli->mli_qri = maxdelay; CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, maxdelay); if (is_general_query) { /* * MLDv2 General Query. * * Schedule a current-state report on this ifp for * all groups, possibly containing source lists. * * If there is a pending General Query response * scheduled earlier than the selected delay, do * not schedule any other reports. * Otherwise, reset the interface timer. */ CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", ifp, if_name(ifp)); if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); V_interface_timers_running6 = 1; } } else { /* * MLDv2 Group-specific or Group-and-source-specific Query. * * Group-source-specific queries are throttled on * a per-group basis to defeat denial-of-service attempts. * Queries for groups we are not a member of on this * link are simply ignored. */ inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm == NULL) goto out_locked; if (nsrc > 0) { if (!ratecheck(&inm->in6m_lastgsrtv, &V_mld_gsrdelay)) { CTR1(KTR_MLD, "%s: GS query throttled.", __func__); goto out_locked; } } CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", ifp, if_name(ifp)); /* * If there is a pending General Query response * scheduled sooner than the selected delay, no * further report need be scheduled. * Otherwise, prepare to respond to the * group-specific or group-and-source query. */ if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) mld_v2_process_group_query(inm, mli, timer, m, mld, off); /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received MLDv2 group-specific or group-and-source-specific * query. * Return <0 if any error occurred. Currently this is ignored. */ static int mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli, int timer, struct mbuf *m0, struct mldv2_query *mld, const int off) { int retval; uint16_t nsrc; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); retval = 0; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LEAVING_MEMBER: return (retval); break; case MLD_REPORTING_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: break; } nsrc = ntohs(mld->mld_numsrc); /* Length should be checked by calling function. */ KASSERT((m0->m_flags & M_PKTHDR) == 0 || m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) + nsrc * sizeof(struct in6_addr), ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)", m0->m_pkthdr.len, off + sizeof(struct mldv2_query) + nsrc * sizeof(struct in6_addr), m0)); /* * Deal with group-specific queries upfront. * If any group query is already pending, purge any recorded * source-list state if it exists, and schedule a query response * for this group-specific query. */ if (nsrc == 0) { if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { in6m_clear_recorded(inm); timer = min(inm->in6m_timer, timer); } inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; return (retval); } /* * Deal with the case where a group-and-source-specific query has * been received but a group-specific query is already pending. */ if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { timer = min(inm->in6m_timer, timer); inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; return (retval); } /* * Finally, deal with the case where a group-and-source-specific * query has been received, where a response to a previous g-s-r * query exists, or none exists. * In this case, we need to parse the source-list which the Querier * has provided us with and check if we have any source list filter * entries at T1 for these sources. If we do not, there is no need * schedule a report and the query may be dropped. * If we do, we must record them and schedule a current-state * report for those sources. */ if (inm->in6m_nsrc > 0) { struct in6_addr srcaddr; int i, nrecorded; int soff; soff = off + sizeof(struct mldv2_query); nrecorded = 0; for (i = 0; i < nsrc; i++) { m_copydata(m0, soff, sizeof(struct in6_addr), (caddr_t)&srcaddr); retval = in6m_record_source(inm, &srcaddr); if (retval < 0) break; nrecorded += retval; soff += sizeof(struct in6_addr); } if (nrecorded > 0) { CTR1(KTR_MLD, "%s: schedule response to SG query", __func__); inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; inm->in6m_timer = MLD_RANDOM_DELAY(timer); V_current_state_timers_running6 = 1; } } return (retval); } /* * Process a received MLDv1 host membership report. * Assumes mld points to mld_hdr in pulled up mbuf chain. * * NOTE: Can't be fully const correct as we temporarily embed scope ID in * mld_addr. This is OK as we own the mbuf chain. */ static int mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld) { struct in6_addr src, dst; struct in6_ifaddr *ia; struct in6_multi *inm; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif NET_EPOCH_ASSERT(); if (!mld_v1enable) { CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); return (0); } if (ifp->if_flags & IFF_LOOPBACK) return (0); /* * MLDv1 reports must originate from a host's link-local address, * or the unspecified address (when booting). */ src = ip6->ip6_src; in6_clearscope(&src); if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_src), ifp, if_name(ifp)); return (EINVAL); } /* * RFC2710 Section 4: MLDv1 reports must pertain to a multicast * group, and must be directed to the group itself. */ dst = ip6->ip6_dst; in6_clearscope(&dst); if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &ip6->ip6_dst), ifp, if_name(ifp)); return (EINVAL); } /* * Make sure we don't hear our own membership report, as fast * leave requires knowing that we are the only member of a * group. Assume we used the link-local address if available, * otherwise look for ::. * * XXX Note that scope ID comparison is needed for the address * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be * performed for the on-wire address. */ ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (0); } if (ia != NULL) ifa_free(&ia->ia_ifa); CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); /* * Embed scope ID of receiving interface in MLD query for lookup * whilst we don't hold other locks (due to KAME locking lameness). */ if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) in6_setscope(&mld->mld_addr, ifp, NULL); IN6_MULTI_LIST_LOCK(); MLD_LOCK(); /* * MLDv1 report suppression. * If we are a member of this group, and our membership should be * reported, and our group timer is pending or about to be reset, * stop our group timer by transitioning to the 'lazy' state. */ inm = in6m_lookup_locked(ifp, &mld->mld_addr); if (inm != NULL) { struct mld_ifsoftc *mli; mli = inm->in6m_mli; KASSERT(mli != NULL, ("%s: no mli for ifp %p", __func__, ifp)); /* * If we are in MLDv2 host mode, do not allow the * other host's MLDv1 report to suppress our reports. */ if (mli->mli_version == MLD_VERSION_2) goto out_locked; inm->in6m_timer = 0; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_AWAKENING_MEMBER: CTR3(KTR_MLD, "report suppressed for %s on ifp %p(%s)", ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); case MLD_LAZY_MEMBER: inm->in6m_state = MLD_LAZY_MEMBER; break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); /* XXX Clear embedded scope ID as userland won't expect it. */ in6_clearscope(&mld->mld_addr); return (0); } /* * MLD input path. * * Assume query messages which fit in a single ICMPv6 message header * have been pulled up. * Assume that userland will want to see the message, even if it * otherwise fails kernel input validation; do not free it. * Pullup may however free the mbuf chain m if it fails. * * Return IPPROTO_DONE if we freed m. Otherwise, return 0. */ int mld_input(struct mbuf **mp, int off, int icmp6len) { struct ifnet *ifp; struct ip6_hdr *ip6; struct mbuf *m; struct mld_hdr *mld; int mldlen; m = *mp; CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); ifp = m->m_pkthdr.rcvif; /* Pullup to appropriate size. */ if (m->m_len < off + sizeof(*mld)) { m = m_pullup(m, off + sizeof(*mld)); if (m == NULL) { ICMP6STAT_INC(icp6s_badlen); return (IPPROTO_DONE); } } mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); if (mld->mld_type == MLD_LISTENER_QUERY && icmp6len >= sizeof(struct mldv2_query)) { mldlen = sizeof(struct mldv2_query); } else { mldlen = sizeof(struct mld_hdr); } if (m->m_len < off + mldlen) { m = m_pullup(m, off + mldlen); if (m == NULL) { ICMP6STAT_INC(icp6s_badlen); return (IPPROTO_DONE); } } *mp = m; ip6 = mtod(m, struct ip6_hdr *); mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); /* * Userland needs to see all of this traffic for implementing * the endpoint discovery portion of multicast routing. */ switch (mld->mld_type) { case MLD_LISTENER_QUERY: icmp6_ifstat_inc(ifp, ifs6_in_mldquery); if (icmp6len == sizeof(struct mld_hdr)) { if (mld_v1_input_query(ifp, ip6, mld) != 0) return (0); } else if (icmp6len >= sizeof(struct mldv2_query)) { if (mld_v2_input_query(ifp, ip6, m, (struct mldv2_query *)mld, off, icmp6len) != 0) return (0); } break; case MLD_LISTENER_REPORT: icmp6_ifstat_inc(ifp, ifs6_in_mldreport); if (mld_v1_input_report(ifp, ip6, mld) != 0) return (0); break; case MLDV2_LISTENER_REPORT: icmp6_ifstat_inc(ifp, ifs6_in_mldreport); break; case MLD_LISTENER_DONE: icmp6_ifstat_inc(ifp, ifs6_in_mlddone); break; default: break; } return (0); } /* * Fast timeout handler (global). * VIMAGE: Timeout handlers are expected to service all vimages. */ void mld_fasttimo(void) { struct in6_multi_head inmh; VNET_ITERATOR_DECL(vnet_iter); SLIST_INIT(&inmh); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); mld_fasttimo_vnet(&inmh); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); in6m_release_list_deferred(&inmh); } /* * Fast timeout handler (per-vnet). * * VIMAGE: Assume caller has set up our curvnet. */ static void mld_fasttimo_vnet(struct in6_multi_head *inmh) { struct epoch_tracker et; struct mbufq scq; /* State-change packets */ struct mbufq qrq; /* Query response packets */ struct ifnet *ifp; struct mld_ifsoftc *mli; struct ifmultiaddr *ifma; struct in6_multi *inm; int uri_fasthz; uri_fasthz = 0; /* * Quick check to see if any work needs to be done, in order to * minimize the overhead of fasttimo processing. * SMPng: XXX Unlocked reads. */ if (!V_current_state_timers_running6 && !V_interface_timers_running6 && !V_state_change_timers_running6) return; IN6_MULTI_LIST_LOCK(); MLD_LOCK(); /* * MLDv2 General Query response timer processing. */ if (V_interface_timers_running6) { CTR1(KTR_MLD, "%s: interface timers running", __func__); V_interface_timers_running6 = 0; LIST_FOREACH(mli, &V_mli_head, mli_link) { if (mli->mli_v2_timer == 0) { /* Do nothing. */ } else if (--mli->mli_v2_timer == 0) { mld_v2_dispatch_general_query(mli); } else { V_interface_timers_running6 = 1; } } } if (!V_current_state_timers_running6 && !V_state_change_timers_running6) goto out_locked; V_current_state_timers_running6 = 0; V_state_change_timers_running6 = 0; CTR1(KTR_MLD, "%s: state change timers running", __func__); /* * MLD host report and state-change timer processing. * Note: Processing a v2 group timer may remove a node. */ LIST_FOREACH(mli, &V_mli_head, mli_link) { ifp = mli->mli_ifp; if (mli->mli_version == MLD_VERSION_2) { uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * PR_FASTHZ); mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS); mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS); } NET_EPOCH_ENTER(et); IF_ADDR_WLOCK(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { inm = in6m_ifmultiaddr_get_inm(ifma); if (inm == NULL) continue; switch (mli->mli_version) { case MLD_VERSION_1: mld_v1_process_group_timer(inmh, inm); break; case MLD_VERSION_2: mld_v2_process_group_timers(inmh, &qrq, &scq, inm, uri_fasthz); break; } } IF_ADDR_WUNLOCK(ifp); switch (mli->mli_version) { case MLD_VERSION_1: /* * Transmit reports for this lifecycle. This * is done while not holding IF_ADDR_LOCK * since this can call * in6ifa_ifpforlinklocal() which locks * IF_ADDR_LOCK internally as well as * ip6_output() to transmit a packet. */ while ((inm = SLIST_FIRST(inmh)) != NULL) { SLIST_REMOVE_HEAD(inmh, in6m_defer); (void)mld_v1_transmit_report(inm, MLD_LISTENER_REPORT); } break; case MLD_VERSION_2: mld_dispatch_queue(&qrq, 0); mld_dispatch_queue(&scq, 0); break; } NET_EPOCH_EXIT(et); } out_locked: MLD_UNLOCK(); IN6_MULTI_LIST_UNLOCK(); } /* * Update host report group timer. * Will update the global pending timer flags. */ static void mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm) { int report_timer_expired; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); if (inm->in6m_timer == 0) { report_timer_expired = 0; } else if (--inm->in6m_timer == 0) { report_timer_expired = 1; } else { V_current_state_timers_running6 = 1; return; } switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: break; case MLD_REPORTING_MEMBER: if (report_timer_expired) { inm->in6m_state = MLD_IDLE_MEMBER; SLIST_INSERT_HEAD(inmh, inm, in6m_defer); } break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } /* * Update a group's timers for MLDv2. * Will update the global pending timer flags. * Note: Unlocked read from mli. */ static void mld_v2_process_group_timers(struct in6_multi_head *inmh, struct mbufq *qrq, struct mbufq *scq, struct in6_multi *inm, const int uri_fasthz) { int query_response_timer_expired; int state_change_retransmit_timer_expired; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); query_response_timer_expired = 0; state_change_retransmit_timer_expired = 0; /* * During a transition from compatibility mode back to MLDv2, * a group record in REPORTING state may still have its group * timer active. This is a no-op in this function; it is easier * to deal with it here than to complicate the slow-timeout path. */ if (inm->in6m_timer == 0) { query_response_timer_expired = 0; } else if (--inm->in6m_timer == 0) { query_response_timer_expired = 1; } else { V_current_state_timers_running6 = 1; } if (inm->in6m_sctimer == 0) { state_change_retransmit_timer_expired = 0; } else if (--inm->in6m_sctimer == 0) { state_change_retransmit_timer_expired = 1; } else { V_state_change_timers_running6 = 1; } /* We are in fasttimo, so be quick about it. */ if (!state_change_retransmit_timer_expired && !query_response_timer_expired) return; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_LAZY_MEMBER: case MLD_AWAKENING_MEMBER: case MLD_IDLE_MEMBER: break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: /* * Respond to a previously pending Group-Specific * or Group-and-Source-Specific query by enqueueing * the appropriate Current-State report for * immediate transmission. */ if (query_response_timer_expired) { int retval; retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 0); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); inm->in6m_state = MLD_REPORTING_MEMBER; in6m_clear_recorded(inm); } /* FALLTHROUGH */ case MLD_REPORTING_MEMBER: case MLD_LEAVING_MEMBER: if (state_change_retransmit_timer_expired) { /* * State-change retransmission timer fired. * If there are any further pending retransmissions, * set the global pending state-change flag, and * reset the timer. */ if (--inm->in6m_scrv > 0) { inm->in6m_sctimer = uri_fasthz; V_state_change_timers_running6 = 1; } /* * Retransmit the previously computed state-change * report. If there are no further pending * retransmissions, the mbuf queue will be consumed. * Update T0 state to T1 as we have now sent * a state-change. */ (void)mld_v2_merge_state_changes(inm, scq); in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); /* * If we are leaving the group for good, make sure * we release MLD's reference to it. * This release must be deferred using a SLIST, * as we are called from a loop which traverses * the in_ifmultiaddr TAILQ. */ if (inm->in6m_state == MLD_LEAVING_MEMBER && inm->in6m_scrv == 0) { inm->in6m_state = MLD_NOT_MEMBER; in6m_disconnect_locked(inmh, inm); in6m_rele_locked(inmh, inm); } } break; } } /* * Switch to a different version on the given interface, * as per Section 9.12. */ static void mld_set_version(struct mld_ifsoftc *mli, const int version) { int old_version_timer; MLD_LOCK_ASSERT(); CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, version, mli->mli_ifp, if_name(mli->mli_ifp)); if (version == MLD_VERSION_1) { /* * Compute the "Older Version Querier Present" timer as per * Section 9.12. */ old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; old_version_timer *= PR_SLOWHZ; mli->mli_v1_timer = old_version_timer; } if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { mli->mli_version = MLD_VERSION_1; mld_v2_cancel_link_timers(mli); } } /* * Cancel pending MLDv2 timers for the given link and all groups * joined on it; state-change, general-query, and group-query timers. */ static void mld_v2_cancel_link_timers(struct mld_ifsoftc *mli) { struct epoch_tracker et; struct in6_multi_head inmh; struct ifmultiaddr *ifma; struct ifnet *ifp; struct in6_multi *inm; CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, mli->mli_ifp, if_name(mli->mli_ifp)); SLIST_INIT(&inmh); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); /* * Fast-track this potentially expensive operation * by checking all the global 'timer pending' flags. */ if (!V_interface_timers_running6 && !V_state_change_timers_running6 && !V_current_state_timers_running6) return; mli->mli_v2_timer = 0; ifp = mli->mli_ifp; IF_ADDR_WLOCK(ifp); NET_EPOCH_ENTER(et); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { inm = in6m_ifmultiaddr_get_inm(ifma); if (inm == NULL) continue; switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: break; case MLD_LEAVING_MEMBER: /* * If we are leaving the group and switching * version, we need to release the final * reference held for issuing the INCLUDE {}. */ if (inm->in6m_refcount == 1) in6m_disconnect_locked(&inmh, inm); in6m_rele_locked(&inmh, inm); /* FALLTHROUGH */ case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: in6m_clear_recorded(inm); /* FALLTHROUGH */ case MLD_REPORTING_MEMBER: inm->in6m_sctimer = 0; inm->in6m_timer = 0; inm->in6m_state = MLD_REPORTING_MEMBER; /* * Free any pending MLDv2 state-change records. */ mbufq_drain(&inm->in6m_scq); break; } } NET_EPOCH_EXIT(et); IF_ADDR_WUNLOCK(ifp); in6m_release_list_deferred(&inmh); } /* * Global slowtimo handler. * VIMAGE: Timeout handlers are expected to service all vimages. */ void mld_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); mld_slowtimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Per-vnet slowtimo handler. */ static void mld_slowtimo_vnet(void) { struct mld_ifsoftc *mli; MLD_LOCK(); LIST_FOREACH(mli, &V_mli_head, mli_link) { mld_v1_process_querier_timers(mli); } MLD_UNLOCK(); } /* * Update the Older Version Querier Present timers for a link. * See Section 9.12 of RFC 3810. */ static void mld_v1_process_querier_timers(struct mld_ifsoftc *mli) { MLD_LOCK_ASSERT(); if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { /* * MLDv1 Querier Present timer expired; revert to MLDv2. */ CTR5(KTR_MLD, "%s: transition from v%d -> v%d on %p(%s)", __func__, mli->mli_version, MLD_VERSION_2, mli->mli_ifp, if_name(mli->mli_ifp)); mli->mli_version = MLD_VERSION_2; } } /* * Transmit an MLDv1 report immediately. */ static int mld_v1_transmit_report(struct in6_multi *in6m, const int type) { struct ifnet *ifp; struct in6_ifaddr *ia; struct ip6_hdr *ip6; struct mbuf *mh, *md; struct mld_hdr *mld; NET_EPOCH_ASSERT(); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); ifp = in6m->in6m_ifp; /* in process of being freed */ if (ifp == NULL) return (0); ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); /* ia may be NULL if link-local address is tentative. */ mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { if (ia != NULL) ifa_free(&ia->ia_ifa); return (ENOMEM); } md = m_get(M_NOWAIT, MT_DATA); if (md == NULL) { m_free(mh); if (ia != NULL) ifa_free(&ia->ia_ifa); return (ENOMEM); } mh->m_next = md; /* * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so * that ether_output() does not need to allocate another mbuf * for the header in the most common case. */ M_ALIGN(mh, sizeof(struct ip6_hdr)); mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); mh->m_len = sizeof(struct ip6_hdr); ip6 = mtod(mh, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; ip6->ip6_dst = in6m->in6m_addr; md->m_len = sizeof(struct mld_hdr); mld = mtod(md, struct mld_hdr *); mld->mld_type = type; mld->mld_code = 0; mld->mld_cksum = 0; mld->mld_maxdelay = 0; mld->mld_reserved = 0; mld->mld_addr = in6m->in6m_addr; in6_clearscope(&mld->mld_addr); mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); mld_save_context(mh, ifp); mh->m_flags |= M_MLDV1; mld_dispatch_packet(mh); if (ia != NULL) ifa_free(&ia->ia_ifa); return (0); } /* * Process a state change from the upper layer for the given IPv6 group. * * Each socket holds a reference on the in_multi in its own ip_moptions. * The socket layer will have made the necessary updates to.the group * state, it is now up to MLD to issue a state change report if there * has been any change between T0 (when the last state-change was issued) * and T1 (now). * * We use the MLDv2 state machine at group level. The MLd module * however makes the decision as to which MLD protocol version to speak. * A state change *from* INCLUDE {} always means an initial join. * A state change *to* INCLUDE {} always means a final leave. * * If delay is non-zero, and the state change is an initial multicast * join, the state change report will be delayed by 'delay' ticks * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise * the initial MLDv2 state change report will be delayed by whichever * is sooner, a pending state-change timer or delay itself. * * VIMAGE: curvnet should have been set by caller, as this routine * is called from the socket option handlers. */ int mld_change_state(struct in6_multi *inm, const int delay) { struct mld_ifsoftc *mli; struct ifnet *ifp; int error; IN6_MULTI_LIST_LOCK_ASSERT(); error = 0; /* * Check if the in6_multi has already been disconnected. */ if (inm->in6m_ifp == NULL) { CTR1(KTR_MLD, "%s: inm is disconnected", __func__); return (0); } /* * Try to detect if the upper layer just asked us to change state * for an interface which has now gone away. */ KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); ifp = inm->in6m_ifma->ifma_ifp; if (ifp == NULL) return (0); /* * Sanity check that netinet6's notion of ifp is the * same as net's. */ KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); MLD_LOCK(); mli = MLD_IFINFO(ifp); KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); /* * If we detect a state transition to or from MCAST_UNDEFINED * for this group, then we are starting or finishing an MLD * life cycle for this group. */ if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_MLD, "%s: initial join", __func__); error = mld_initial_join(inm, mli, delay); goto out_locked; } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_MLD, "%s: final leave", __func__); mld_final_leave(inm, mli); goto out_locked; } } else { CTR1(KTR_MLD, "%s: filter set change", __func__); } error = mld_handle_state_change(inm, mli); out_locked: MLD_UNLOCK(); return (error); } /* * Perform the initial join for an MLD group. * * When joining a group: * If the group should have its MLD traffic suppressed, do nothing. * MLDv1 starts sending MLDv1 host membership reports. * MLDv2 will schedule an MLDv2 state-change report containing the * initial state of the membership. * * If the delay argument is non-zero, then we must delay sending the * initial state change for delay ticks (in units of PR_FASTHZ). */ static int mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli, const int delay) { struct epoch_tracker et; struct ifnet *ifp; struct mbufq *mq; int error, retval, syncstates; int odelay; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); error = 0; syncstates = 1; ifp = inm->in6m_ifp; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); /* * Groups joined on loopback or marked as 'not reported', * enter the MLD_SILENT_MEMBER state and * are never reported in any protocol exchanges. * All other groups enter the appropriate state machine * for the version in use on this link. * A link marked as MLIF_SILENT causes MLD to be completely * disabled for the link. */ if ((ifp->if_flags & IFF_LOOPBACK) || (mli->mli_flags & MLIF_SILENT) || !mld_is_addr_reported(&inm->in6m_addr)) { CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); inm->in6m_state = MLD_SILENT_MEMBER; inm->in6m_timer = 0; } else { /* * Deal with overlapping in_multi lifecycle. * If this group was LEAVING, then make sure * we drop the reference we picked up to keep the * group around for the final INCLUDE {} enqueue. */ if (mli->mli_version == MLD_VERSION_2 && inm->in6m_state == MLD_LEAVING_MEMBER) { inm->in6m_refcount--; MPASS(inm->in6m_refcount > 0); } inm->in6m_state = MLD_REPORTING_MEMBER; switch (mli->mli_version) { case MLD_VERSION_1: /* * If a delay was provided, only use it if * it is greater than the delay normally * used for an MLDv1 state change report, * and delay sending the initial MLDv1 report * by not transitioning to the IDLE state. */ odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); if (delay) { inm->in6m_timer = max(delay, odelay); V_current_state_timers_running6 = 1; } else { inm->in6m_state = MLD_IDLE_MEMBER; NET_EPOCH_ENTER(et); error = mld_v1_transmit_report(inm, MLD_LISTENER_REPORT); NET_EPOCH_EXIT(et); if (error == 0) { inm->in6m_timer = odelay; V_current_state_timers_running6 = 1; } } break; case MLD_VERSION_2: /* * Defer update of T0 to T1, until the first copy * of the state change has been transmitted. */ syncstates = 0; /* * Immediately enqueue a State-Change Report for * this interface, freeing any previous reports. * Don't kick the timers if there is nothing to do, * or if an error occurred. */ mq = &inm->in6m_scq; mbufq_drain(mq); retval = mld_v2_enqueue_group_record(mq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) { error = retval * -1; break; } /* * Schedule transmission of pending state-change * report up to RV times for this link. The timer * will fire at the next mld_fasttimo (~200ms), * giving us an opportunity to merge the reports. * * If a delay was provided to this function, only * use this delay if sooner than the existing one. */ KASSERT(mli->mli_rv > 1, ("%s: invalid robustness %d", __func__, mli->mli_rv)); inm->in6m_scrv = mli->mli_rv; if (delay) { if (inm->in6m_sctimer > 1) { inm->in6m_sctimer = min(inm->in6m_sctimer, delay); } else inm->in6m_sctimer = delay; } else inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; error = 0; break; } } /* * Only update the T0 state if state change is atomic, * i.e. we don't need to wait for a timer to fire before we * can consider the state change to have been communicated. */ if (syncstates) { in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); } return (error); } /* * Issue an intermediate state change during the life-cycle. */ static int mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli) { struct ifnet *ifp; int retval; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); ifp = inm->in6m_ifp; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); if ((ifp->if_flags & IFF_LOOPBACK) || (mli->mli_flags & MLIF_SILENT) || !mld_is_addr_reported(&inm->in6m_addr) || (mli->mli_version != MLD_VERSION_2)) { if (!mld_is_addr_reported(&inm->in6m_addr)) { CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); } CTR1(KTR_MLD, "%s: nothing to do", __func__); in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); return (0); } mbufq_drain(&inm->in6m_scq); retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) return (-retval); /* * If record(s) were enqueued, start the state-change * report timer for this group. */ inm->in6m_scrv = mli->mli_rv; inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; return (0); } /* * Perform the final leave for a multicast address. * * When leaving a group: * MLDv1 sends a DONE message, if and only if we are the reporter. * MLDv2 enqueues a state-change report containing a transition * to INCLUDE {} for immediate transmission. */ static void mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli) { struct epoch_tracker et; int syncstates; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif syncstates = 1; CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), inm->in6m_ifp, if_name(inm->in6m_ifp)); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: case MLD_LEAVING_MEMBER: /* Already leaving or left; do nothing. */ CTR1(KTR_MLD, "%s: not kicking state machine for silent group", __func__); break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: if (mli->mli_version == MLD_VERSION_1) { #ifdef INVARIANTS if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) panic("%s: MLDv2 state reached, not MLDv2 mode", __func__); #endif NET_EPOCH_ENTER(et); mld_v1_transmit_report(inm, MLD_LISTENER_DONE); NET_EPOCH_EXIT(et); inm->in6m_state = MLD_NOT_MEMBER; V_current_state_timers_running6 = 1; } else if (mli->mli_version == MLD_VERSION_2) { /* * Stop group timer and all pending reports. * Immediately enqueue a state-change report * TO_IN {} to be sent on the next fast timeout, * giving us an opportunity to merge reports. */ mbufq_drain(&inm->in6m_scq); inm->in6m_timer = 0; inm->in6m_scrv = mli->mli_rv; CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " "pending retransmissions.", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp), inm->in6m_scrv); if (inm->in6m_scrv == 0) { inm->in6m_state = MLD_NOT_MEMBER; inm->in6m_sctimer = 0; } else { int retval; in6m_acquire_locked(inm); retval = mld_v2_enqueue_group_record( &inm->in6m_scq, inm, 1, 0, 0, (mli->mli_flags & MLIF_USEALLOW)); KASSERT(retval != 0, ("%s: enqueue record = %d", __func__, retval)); inm->in6m_state = MLD_LEAVING_MEMBER; inm->in6m_sctimer = 1; V_state_change_timers_running6 = 1; syncstates = 0; } break; } break; case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: /* Our reports are suppressed; do nothing. */ break; } if (syncstates) { in6m_commit(inm); CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); } } /* * Enqueue an MLDv2 group record to the given output queue. * * If is_state_change is zero, a current-state record is appended. * If is_state_change is non-zero, a state-change report is appended. * * If is_group_query is non-zero, an mbuf packet chain is allocated. * If is_group_query is zero, and if there is a packet with free space * at the tail of the queue, it will be appended to providing there * is enough free space. * Otherwise a new mbuf packet chain is allocated. * * If is_source_query is non-zero, each source is checked to see if * it was recorded for a Group-Source query, and will be omitted if * it is not both in-mode and recorded. * * If use_block_allow is non-zero, state change reports for initial join * and final leave, on an inclusive mode group with a source list, will be * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. * * The function will attempt to allocate leading space in the packet * for the IPv6+ICMP headers to be prepended without fragmenting the chain. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm, const int is_state_change, const int is_group_query, const int is_source_query, const int use_block_allow) { struct mldv2_record mr; struct mldv2_record *pmr; struct ifnet *ifp; struct ip6_msource *ims, *nims; struct mbuf *m0, *m, *md; int is_filter_list_change; int minrec0len, m0srcs, msrcs, nbytes, off; int record_has_sources; int now; int type; uint8_t mode; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); ifp = inm->in6m_ifp; is_filter_list_change = 0; m = NULL; m0 = NULL; m0srcs = 0; msrcs = 0; nbytes = 0; nims = NULL; record_has_sources = 1; pmr = NULL; type = MLD_DO_NOTHING; mode = inm->in6m_st[1].iss_fmode; /* * If we did not transition out of ASM mode during t0->t1, * and there are no source nodes to process, we can skip * the generation of source records. */ if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && inm->in6m_nsrc == 0) record_has_sources = 0; if (is_state_change) { /* * Queue a state change record. * If the mode did not change, and there are non-ASM * listeners or source filters present, * we potentially need to issue two records for the group. * If there are ASM listeners, and there was no filter * mode transition of any kind, do nothing. * * If we are transitioning to MCAST_UNDEFINED, we need * not send any sources. A transition to/from this state is * considered inclusive with some special treatment. * * If we are rewriting initial joins/leaves to use * ALLOW/BLOCK, and the group's membership is inclusive, * we need to send sources in all cases. */ if (mode != inm->in6m_st[0].iss_fmode) { if (mode == MCAST_EXCLUDE) { CTR1(KTR_MLD, "%s: change to EXCLUDE", __func__); type = MLD_CHANGE_TO_EXCLUDE_MODE; } else { CTR1(KTR_MLD, "%s: change to INCLUDE", __func__); if (use_block_allow) { /* * XXX * Here we're interested in state * edges either direction between * MCAST_UNDEFINED and MCAST_INCLUDE. * Perhaps we should just check * the group state, rather than * the filter mode. */ if (mode == MCAST_UNDEFINED) { type = MLD_BLOCK_OLD_SOURCES; } else { type = MLD_ALLOW_NEW_SOURCES; } } else { type = MLD_CHANGE_TO_INCLUDE_MODE; if (mode == MCAST_UNDEFINED) record_has_sources = 0; } } } else { if (record_has_sources) { is_filter_list_change = 1; } else { type = MLD_DO_NOTHING; } } } else { /* * Queue a current state record. */ if (mode == MCAST_EXCLUDE) { type = MLD_MODE_IS_EXCLUDE; } else if (mode == MCAST_INCLUDE) { type = MLD_MODE_IS_INCLUDE; KASSERT(inm->in6m_st[1].iss_asm == 0, ("%s: inm %p is INCLUDE but ASM count is %d", __func__, inm, inm->in6m_st[1].iss_asm)); } } /* * Generate the filter list changes using a separate function. */ if (is_filter_list_change) return (mld_v2_enqueue_filter_change(mq, inm)); if (type == MLD_DO_NOTHING) { CTR3(KTR_MLD, "%s: nothing to do for %s/%s", __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); return (0); } /* * If any sources are present, we must be able to fit at least * one in the trailing space of the tail packet's mbuf, * ideally more. */ minrec0len = sizeof(struct mldv2_record); if (record_has_sources) minrec0len += sizeof(struct in6_addr); CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, mld_rec_type_to_str(type), ip6_sprintf(ip6tbuf, &inm->in6m_addr), if_name(inm->in6m_ifp)); /* * Check if we have a packet in the tail of the queue for this * group into which the first group record for this group will fit. * Otherwise allocate a new packet. * Always allocate leading space for IP6+RA+ICMPV6+REPORT. * Note: Group records for G/GSR query responses MUST be sent * in their own packet. */ m0 = mbufq_last(mq); if (!is_group_query && m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && (m0->m_pkthdr.len + minrec0len) < (ifp->if_mtu - MLD_MTUSPACE)) { m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); m = m0; CTR1(KTR_MLD, "%s: use existing packet", __func__); } else { if (mbufq_full(mq)) { CTR1(KTR_MLD, "%s: outbound queue full", __func__); return (-ENOMEM); } m = NULL; m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); if (!is_state_change && !is_group_query) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (-ENOMEM); mld_save_context(m, ifp); CTR1(KTR_MLD, "%s: allocated first packet", __func__); } /* * Append group record. * If we have sources, we don't know how many yet. */ mr.mr_type = type; mr.mr_datalen = 0; mr.mr_numsrc = 0; mr.mr_addr = inm->in6m_addr; in6_clearscope(&mr.mr_addr); if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct mldv2_record); /* * Append as many sources as will fit in the first packet. * If we are appending to a new packet, the chain allocation * may potentially use clusters; use m_getptr() in this case. * If we are appending to an existing packet, we need to obtain * a pointer to the group record after m_append(), in case a new * mbuf was allocated. * * Only append sources which are in-mode at t1. If we are * transitioning to MCAST_UNDEFINED state on the group, and * use_block_allow is zero, do not include source entries. * Otherwise, we need to include this source in the report. * * Only report recorded sources in our filter set when responding * to a group-source query. */ if (record_has_sources) { if (m == m0) { md = m_last(m); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + md->m_len - nbytes); } else { md = m_getptr(m, 0, &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); } msrcs = 0; RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); CTR2(KTR_MLD, "%s: node is %d", __func__, now); if ((now != mode) || (now == mode && (!use_block_allow && mode == MCAST_UNDEFINED))) { CTR1(KTR_MLD, "%s: skip node", __func__); continue; } if (is_source_query && ims->im6s_stp == 0) { CTR1(KTR_MLD, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_MLD, "%s: append node", __func__); if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct in6_addr); ++msrcs; if (msrcs == m0srcs) break; } CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, msrcs); pmr->mr_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(struct in6_addr)); } if (is_source_query && msrcs == 0) { CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); if (m != m0) m_freem(m); return (0); } /* * We are good to go with first packet. */ if (m != m0) { CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); m->m_pkthdr.PH_vt.vt_nrecs = 1; mbufq_enqueue(mq, m); } else m->m_pkthdr.PH_vt.vt_nrecs++; /* * No further work needed if no source list in packet(s). */ if (!record_has_sources) return (nbytes); /* * Whilst sources remain to be announced, we need to allocate * a new packet and fill out as many sources as will fit. * Always try for a cluster first. */ while (nims != NULL) { if (mbufq_full(mq)) { CTR1(KTR_MLD, "%s: outbound queue full", __func__); return (-ENOMEM); } m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (-ENOMEM); mld_save_context(m, ifp); md = m_getptr(m, 0, &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); CTR1(KTR_MLD, "%s: allocated next packet", __func__); if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 1; nbytes += sizeof(struct mldv2_record); m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); msrcs = 0; RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); if ((now != mode) || (now == mode && (!use_block_allow && mode == MCAST_UNDEFINED))) { CTR1(KTR_MLD, "%s: skip node", __func__); continue; } if (is_source_query && ims->im6s_stp == 0) { CTR1(KTR_MLD, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_MLD, "%s: append node", __func__); if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed.", __func__); return (-ENOMEM); } ++msrcs; if (msrcs == m0srcs) break; } pmr->mr_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(struct in6_addr)); CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); mbufq_enqueue(mq, m); } return (nbytes); } /* * Type used to mark record pass completion. * We exploit the fact we can cast to this easily from the * current filter modes on each ip_msource node. */ typedef enum { REC_NONE = 0x00, /* MCAST_UNDEFINED */ REC_ALLOW = 0x01, /* MCAST_INCLUDE */ REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ REC_FULL = REC_ALLOW | REC_BLOCK } rectype_t; /* * Enqueue an MLDv2 filter list change to the given output queue. * * Source list filter state is held in an RB-tree. When the filter list * for a group is changed without changing its mode, we need to compute * the deltas between T0 and T1 for each source in the filter set, * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. * * As we may potentially queue two record types, and the entire R-B tree * needs to be walked at once, we break this out into its own function * so we can generate a tightly packed queue of packets. * * XXX This could be written to only use one tree walk, although that makes * serializing into the mbuf chains a bit harder. For now we do two walks * which makes things easier on us, and it may or may not be harder on * the L2 cache. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm) { static const int MINRECLEN = sizeof(struct mldv2_record) + sizeof(struct in6_addr); struct ifnet *ifp; struct mldv2_record mr; struct mldv2_record *pmr; struct ip6_msource *ims, *nims; struct mbuf *m, *m0, *md; int m0srcs, nbytes, npbytes, off, rsrcs, schanged; int nallow, nblock; uint8_t mode, now, then; rectype_t crt, drt, nrt; #ifdef KTR char ip6tbuf[INET6_ADDRSTRLEN]; #endif IN6_MULTI_LIST_LOCK_ASSERT(); if (inm->in6m_nsrc == 0 || (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) return (0); ifp = inm->in6m_ifp; /* interface */ mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ crt = REC_NONE; /* current group record type */ drt = REC_NONE; /* mask of completed group record types */ nrt = REC_NONE; /* record type for current node */ m0srcs = 0; /* # source which will fit in current mbuf chain */ npbytes = 0; /* # of bytes appended this packet */ nbytes = 0; /* # of bytes appended to group's state-change queue */ rsrcs = 0; /* # sources encoded in current record */ schanged = 0; /* # nodes encoded in overall filter change */ nallow = 0; /* # of source entries in ALLOW_NEW */ nblock = 0; /* # of source entries in BLOCK_OLD */ nims = NULL; /* next tree node pointer */ /* * For each possible filter record mode. * The first kind of source we encounter tells us which * is the first kind of record we start appending. * If a node transitioned to UNDEFINED at t1, its mode is treated * as the inverse of the group's filter mode. */ while (drt != REC_FULL) { do { m0 = mbufq_last(mq); if (m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && (m0->m_pkthdr.len + MINRECLEN) < (ifp->if_mtu - MLD_MTUSPACE)) { m = m0; m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); CTR1(KTR_MLD, "%s: use previous packet", __func__); } else { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m == NULL) m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { CTR1(KTR_MLD, "%s: m_get*() failed", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 0; mld_save_context(m, ifp); m0srcs = (ifp->if_mtu - MLD_MTUSPACE - sizeof(struct mldv2_record)) / sizeof(struct in6_addr); npbytes = 0; CTR1(KTR_MLD, "%s: allocated new packet", __func__); } /* * Append the MLD group record header to the * current packet's data area. * Recalculate pointer to free space for next * group record, in case m_append() allocated * a new mbuf or cluster. */ memset(&mr, 0, sizeof(mr)); mr.mr_addr = inm->in6m_addr; in6_clearscope(&mr.mr_addr); if (!m_append(m, sizeof(mr), (void *)&mr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed", __func__); return (-ENOMEM); } npbytes += sizeof(struct mldv2_record); if (m != m0) { /* new packet; offset in chain */ md = m_getptr(m, npbytes - sizeof(struct mldv2_record), &off); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); } else { /* current packet; offset from last append */ md = m_last(m); pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + md->m_len - sizeof(struct mldv2_record)); } /* * Begin walking the tree for this record type * pass, or continue from where we left off * previously if we had to allocate a new packet. * Only report deltas in-mode at t1. * We need not report included sources as allowed * if we are in inclusive mode on the group, * however the converse is not true. */ rsrcs = 0; if (nims == NULL) { nims = RB_MIN(ip6_msource_tree, &inm->in6m_srcs); } RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { CTR2(KTR_MLD, "%s: visit node %s", __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); now = im6s_get_mode(inm, ims, 1); then = im6s_get_mode(inm, ims, 0); CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", __func__, then, now); if (now == then) { CTR1(KTR_MLD, "%s: skip unchanged", __func__); continue; } if (mode == MCAST_EXCLUDE && now == MCAST_INCLUDE) { CTR1(KTR_MLD, "%s: skip IN src on EX group", __func__); continue; } nrt = (rectype_t)now; if (nrt == REC_NONE) nrt = (rectype_t)(~mode & REC_FULL); if (schanged++ == 0) { crt = nrt; } else if (crt != nrt) continue; if (!m_append(m, sizeof(struct in6_addr), (void *)&ims->im6s_addr)) { if (m != m0) m_freem(m); CTR1(KTR_MLD, "%s: m_append() failed", __func__); return (-ENOMEM); } nallow += !!(crt == REC_ALLOW); nblock += !!(crt == REC_BLOCK); if (++rsrcs == m0srcs) break; } /* * If we did not append any tree nodes on this * pass, back out of allocations. */ if (rsrcs == 0) { npbytes -= sizeof(struct mldv2_record); if (m != m0) { CTR1(KTR_MLD, "%s: m_free(m)", __func__); m_freem(m); } else { CTR1(KTR_MLD, "%s: m_adj(m, -mr)", __func__); m_adj(m, -((int)sizeof( struct mldv2_record))); } continue; } npbytes += (rsrcs * sizeof(struct in6_addr)); if (crt == REC_ALLOW) pmr->mr_type = MLD_ALLOW_NEW_SOURCES; else if (crt == REC_BLOCK) pmr->mr_type = MLD_BLOCK_OLD_SOURCES; pmr->mr_numsrc = htons(rsrcs); /* * Count the new group record, and enqueue this * packet if it wasn't already queued. */ m->m_pkthdr.PH_vt.vt_nrecs++; if (m != m0) mbufq_enqueue(mq, m); nbytes += npbytes; } while (nims != NULL); drt |= crt; crt = (~crt & REC_FULL); } CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, nallow, nblock); return (nbytes); } static int mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq) { struct mbufq *gq; struct mbuf *m; /* pending state-change */ struct mbuf *m0; /* copy of pending state-change */ struct mbuf *mt; /* last state-change in packet */ int docopy, domerge; u_int recslen; docopy = 0; domerge = 0; recslen = 0; IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); /* * If there are further pending retransmissions, make a writable * copy of each queued state-change message before merging. */ if (inm->in6m_scrv > 0) docopy = 1; gq = &inm->in6m_scq; #ifdef KTR if (mbufq_first(gq) == NULL) { CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", __func__, inm); } #endif m = mbufq_first(gq); while (m != NULL) { /* * Only merge the report into the current packet if * there is sufficient space to do so; an MLDv2 report * packet may only contain 65,535 group records. * Always use a simple mbuf chain concatentation to do this, * as large state changes for single groups may have * allocated clusters. */ domerge = 0; mt = mbufq_last(scq); if (mt != NULL) { recslen = m_length(m, NULL); if ((mt->m_pkthdr.PH_vt.vt_nrecs + m->m_pkthdr.PH_vt.vt_nrecs <= MLD_V2_REPORT_MAXRECS) && (mt->m_pkthdr.len + recslen <= (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) domerge = 1; } if (!domerge && mbufq_full(gq)) { CTR2(KTR_MLD, "%s: outbound queue full, skipping whole packet %p", __func__, m); mt = m->m_nextpkt; if (!docopy) m_freem(m); m = mt; continue; } if (!docopy) { CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); m0 = mbufq_dequeue(gq); m = m0->m_nextpkt; } else { CTR2(KTR_MLD, "%s: copying %p", __func__, m); m0 = m_dup(m, M_NOWAIT); if (m0 == NULL) return (ENOMEM); m0->m_nextpkt = NULL; m = m->m_nextpkt; } if (!domerge) { CTR3(KTR_MLD, "%s: queueing %p to scq %p)", __func__, m0, scq); mbufq_enqueue(scq, m0); } else { struct mbuf *mtl; /* last mbuf of packet mt */ CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", __func__, m0, mt); mtl = m_last(mt); m0->m_flags &= ~M_PKTHDR; mt->m_pkthdr.len += recslen; mt->m_pkthdr.PH_vt.vt_nrecs += m0->m_pkthdr.PH_vt.vt_nrecs; mtl->m_next = m0; } } return (0); } /* * Respond to a pending MLDv2 General Query. */ static void mld_v2_dispatch_general_query(struct mld_ifsoftc *mli) { struct ifmultiaddr *ifma; struct ifnet *ifp; struct in6_multi *inm; int retval; NET_EPOCH_ASSERT(); IN6_MULTI_LIST_LOCK_ASSERT(); MLD_LOCK_ASSERT(); KASSERT(mli->mli_version == MLD_VERSION_2, ("%s: called when version %d", __func__, mli->mli_version)); /* * Check that there are some packets queued. If so, send them first. * For large number of groups the reply to general query can take * many packets, we should finish sending them before starting of * queuing the new reply. */ if (mbufq_len(&mli->mli_gq) != 0) goto send; ifp = mli->mli_ifp; CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { inm = in6m_ifmultiaddr_get_inm(ifma); if (inm == NULL) continue; KASSERT(ifp == inm->in6m_ifp, ("%s: inconsistent ifp", __func__)); switch (inm->in6m_state) { case MLD_NOT_MEMBER: case MLD_SILENT_MEMBER: break; case MLD_REPORTING_MEMBER: case MLD_IDLE_MEMBER: case MLD_LAZY_MEMBER: case MLD_SLEEPING_MEMBER: case MLD_AWAKENING_MEMBER: inm->in6m_state = MLD_REPORTING_MEMBER; retval = mld_v2_enqueue_group_record(&mli->mli_gq, inm, 0, 0, 0, 0); CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); break; case MLD_G_QUERY_PENDING_MEMBER: case MLD_SG_QUERY_PENDING_MEMBER: case MLD_LEAVING_MEMBER: break; } } send: mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); /* * Slew transmission of bursts over 500ms intervals. */ if (mbufq_first(&mli->mli_gq) != NULL) { mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( MLD_RESPONSE_BURST_INTERVAL); V_interface_timers_running6 = 1; } } /* * Transmit the next pending message in the output queue. * * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. * MRT: Nothing needs to be done, as MLD traffic is always local to * a link and uses a link-scope multicast address. */ static void mld_dispatch_packet(struct mbuf *m) { struct ip6_moptions im6o; struct ifnet *ifp; struct ifnet *oifp; struct mbuf *m0; struct mbuf *md; struct ip6_hdr *ip6; struct mld_hdr *mld; int error; int off; int type; uint32_t ifindex; CTR2(KTR_MLD, "%s: transmit %p", __func__, m); + NET_EPOCH_ASSERT(); /* * Set VNET image pointer from enqueued mbuf chain * before doing anything else. Whilst we use interface * indexes to guard against interface detach, they are * unique to each VIMAGE and must be retrieved. */ ifindex = mld_restore_context(m); /* * Check if the ifnet still exists. This limits the scope of * any race in the absence of a global ifp lock for low cost * (an array lookup). */ ifp = ifnet_byindex(ifindex); if (ifp == NULL) { CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", __func__, m, ifindex); m_freem(m); IP6STAT_INC(ip6s_noroute); goto out; } im6o.im6o_multicast_hlim = 1; im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); im6o.im6o_multicast_ifp = ifp; if (m->m_flags & M_MLDV1) { m0 = m; } else { m0 = mld_v2_encap_report(ifp, m); if (m0 == NULL) { CTR2(KTR_MLD, "%s: dropped %p", __func__, m); IP6STAT_INC(ip6s_odropped); goto out; } } mld_scrub_context(m0); m_clrprotoflags(m); m0->m_pkthdr.rcvif = V_loif; ip6 = mtod(m0, struct ip6_hdr *); #if 0 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ #else /* * XXX XXX Break some KPI rules to prevent an LOR which would * occur if we called in6_setscope() at transmission. * See comments at top of file. */ MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); #endif /* * Retrieve the ICMPv6 type before handoff to ip6_output(), * so we can bump the stats. */ md = m_getptr(m0, sizeof(struct ip6_hdr), &off); mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); type = mld->mld_type; error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, &oifp, NULL); if (error) { CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); goto out; } ICMP6STAT_INC(icp6s_outhist[type]); if (oifp != NULL) { icmp6_ifstat_inc(oifp, ifs6_out_msg); switch (type) { case MLD_LISTENER_REPORT: case MLDV2_LISTENER_REPORT: icmp6_ifstat_inc(oifp, ifs6_out_mldreport); break; case MLD_LISTENER_DONE: icmp6_ifstat_inc(oifp, ifs6_out_mlddone); break; } } out: return; } /* * Encapsulate an MLDv2 report. * * KAME IPv6 requires that hop-by-hop options be passed separately, * and that the IPv6 header be prepended in a separate mbuf. * * Returns a pointer to the new mbuf chain head, or NULL if the * allocation failed. */ static struct mbuf * mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) { struct mbuf *mh; struct mldv2_report *mld; struct ip6_hdr *ip6; struct in6_ifaddr *ia; int mldreclen; KASSERT(ifp != NULL, ("%s: null ifp", __func__)); KASSERT((m->m_flags & M_PKTHDR), ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); /* * RFC3590: OK to send as :: or tentative during DAD. */ NET_EPOCH_ASSERT(); ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); if (ia == NULL) CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); mh = m_gethdr(M_NOWAIT, MT_DATA); if (mh == NULL) { if (ia != NULL) ifa_free(&ia->ia_ifa); m_freem(m); return (NULL); } M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); mldreclen = m_length(m, NULL); CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report) + mldreclen; ip6 = mtod(mh, struct ip6_hdr *); ip6->ip6_flow = 0; ip6->ip6_vfc &= ~IPV6_VERSION_MASK; ip6->ip6_vfc |= IPV6_VERSION; ip6->ip6_nxt = IPPROTO_ICMPV6; ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; if (ia != NULL) ifa_free(&ia->ia_ifa); ip6->ip6_dst = in6addr_linklocal_allv2routers; /* scope ID will be set in netisr */ mld = (struct mldv2_report *)(ip6 + 1); mld->mld_type = MLDV2_LISTENER_REPORT; mld->mld_code = 0; mld->mld_cksum = 0; mld->mld_v2_reserved = 0; mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); m->m_pkthdr.PH_vt.vt_nrecs = 0; mh->m_next = m; mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); return (mh); } #ifdef KTR static char * mld_rec_type_to_str(const int type) { switch (type) { case MLD_CHANGE_TO_EXCLUDE_MODE: return "TO_EX"; break; case MLD_CHANGE_TO_INCLUDE_MODE: return "TO_IN"; break; case MLD_MODE_IS_EXCLUDE: return "MODE_EX"; break; case MLD_MODE_IS_INCLUDE: return "MODE_IN"; break; case MLD_ALLOW_NEW_SOURCES: return "ALLOW_NEW"; break; case MLD_BLOCK_OLD_SOURCES: return "BLOCK_OLD"; break; default: break; } return "unknown"; } #endif static void mld_init(void *unused __unused) { CTR1(KTR_MLD, "%s: initializing", __func__); MLD_LOCK_INIT(); ip6_initpktopts(&mld_po); mld_po.ip6po_hlim = 1; mld_po.ip6po_hbh = &mld_ra.hbh; mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; mld_po.ip6po_flags = IP6PO_DONTFRAG; } SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL); static void mld_uninit(void *unused __unused) { CTR1(KTR_MLD, "%s: tearing down", __func__); MLD_LOCK_DESTROY(); } SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL); static void vnet_mld_init(const void *unused __unused) { CTR1(KTR_MLD, "%s: initializing", __func__); LIST_INIT(&V_mli_head); } VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init, NULL); static void vnet_mld_uninit(const void *unused __unused) { /* This can happen if we shutdown the network stack. */ CTR1(KTR_MLD, "%s: tearing down", __func__); } VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit, NULL); static int mld_modevent(module_t mod, int type, void *unused __unused) { switch (type) { case MOD_LOAD: case MOD_UNLOAD: break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t mld_mod = { "mld", mld_modevent, 0 }; DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);