2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the project nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sysctl.h>
56 #include <net/if_arc.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/iso88025.h>
61 #include <net/route.h>
64 #include <netinet/in.h>
65 #include <net/if_llatbl.h>
66 #define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le))
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet/icmp6.h>
75 #include <sys/limits.h>
77 #include <security/mac/mac_framework.h>
79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 #define SIN6(s) ((struct sockaddr_in6 *)s)
85 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */
86 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */
87 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */
88 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */
89 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for
91 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
94 /* preventing too many loops in ND option parsing */
95 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
97 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper
99 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
101 #define V_nd6_maxndopt VNET(nd6_maxndopt)
102 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
105 VNET_DEFINE(int, nd6_debug) = 1;
107 VNET_DEFINE(int, nd6_debug) = 0;
112 static int nd6_inuse, nd6_allocated;
115 VNET_DEFINE(struct nd_drhead, nd_defrouter);
116 VNET_DEFINE(struct nd_prhead, nd_prefix);
118 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
119 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
121 static struct sockaddr_in6 all1_sa;
123 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
125 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
126 static void nd6_slowtimo(void *);
127 static int regen_tmpaddr(struct in6_ifaddr *);
128 static struct llentry *nd6_free(struct llentry *, int);
129 static void nd6_llinfo_timer(void *);
130 static void clear_llinfo_pqueue(struct llentry *);
132 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
133 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
135 VNET_DEFINE(struct callout, nd6_timer_ch);
142 LIST_INIT(&V_nd_prefix);
144 all1_sa.sin6_family = AF_INET6;
145 all1_sa.sin6_len = sizeof(struct sockaddr_in6);
146 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
147 all1_sa.sin6_addr.s6_addr[i] = 0xff;
149 /* initialization of the default router list */
150 TAILQ_INIT(&V_nd_defrouter);
153 callout_init(&V_nd6_slowtimo_ch, 0);
154 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
155 nd6_slowtimo, curvnet);
163 callout_drain(&V_nd6_slowtimo_ch);
164 callout_drain(&V_nd6_timer_ch);
169 nd6_ifattach(struct ifnet *ifp)
171 struct nd_ifinfo *nd;
173 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
174 bzero(nd, sizeof(*nd));
178 nd->chlim = IPV6_DEFHLIM;
179 nd->basereachable = REACHABLE_TIME;
180 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
181 nd->retrans = RETRANS_TIMER;
183 * Note that the default value of ip6_accept_rtadv is 0, which means
184 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
187 nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
189 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
190 nd6_setmtu0(ifp, nd);
196 nd6_ifdetach(struct nd_ifinfo *nd)
203 * Reset ND level link MTU. This function is called when the physical MTU
204 * changes, which means we might have to adjust the ND level MTU.
207 nd6_setmtu(struct ifnet *ifp)
210 nd6_setmtu0(ifp, ND_IFINFO(ifp));
213 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
215 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
219 omaxmtu = ndi->maxmtu;
221 switch (ifp->if_type) {
223 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
226 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
229 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
232 ndi->maxmtu = ifp->if_mtu;
237 * Decreasing the interface MTU under IPV6 minimum MTU may cause
238 * undesirable situation. We thus notify the operator of the change
239 * explicitly. The check for omaxmtu is necessary to restrict the
240 * log to the case of changing the MTU, not initializing it.
242 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
243 log(LOG_NOTICE, "nd6_setmtu0: "
244 "new link MTU on %s (%lu) is too small for IPv6\n",
245 if_name(ifp), (unsigned long)ndi->maxmtu);
248 if (ndi->maxmtu > V_in6_maxmtu)
249 in6_setmaxmtu(); /* check all interfaces just in case */
254 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
257 bzero(ndopts, sizeof(*ndopts));
258 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
260 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
263 ndopts->nd_opts_done = 1;
264 ndopts->nd_opts_search = NULL;
269 * Take one ND option.
272 nd6_option(union nd_opts *ndopts)
274 struct nd_opt_hdr *nd_opt;
278 panic("ndopts == NULL in nd6_option");
279 if (ndopts->nd_opts_last == NULL)
280 panic("uninitialized ndopts in nd6_option");
281 if (ndopts->nd_opts_search == NULL)
283 if (ndopts->nd_opts_done)
286 nd_opt = ndopts->nd_opts_search;
288 /* make sure nd_opt_len is inside the buffer */
289 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
290 bzero(ndopts, sizeof(*ndopts));
294 olen = nd_opt->nd_opt_len << 3;
297 * Message validation requires that all included
298 * options have a length that is greater than zero.
300 bzero(ndopts, sizeof(*ndopts));
304 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
305 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
306 /* option overruns the end of buffer, invalid */
307 bzero(ndopts, sizeof(*ndopts));
309 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
310 /* reached the end of options chain */
311 ndopts->nd_opts_done = 1;
312 ndopts->nd_opts_search = NULL;
318 * Parse multiple ND options.
319 * This function is much easier to use, for ND routines that do not need
320 * multiple options of the same type.
323 nd6_options(union nd_opts *ndopts)
325 struct nd_opt_hdr *nd_opt;
329 panic("ndopts == NULL in nd6_options");
330 if (ndopts->nd_opts_last == NULL)
331 panic("uninitialized ndopts in nd6_options");
332 if (ndopts->nd_opts_search == NULL)
336 nd_opt = nd6_option(ndopts);
337 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
339 * Message validation requires that all included
340 * options have a length that is greater than zero.
342 ICMP6STAT_INC(icp6s_nd_badopt);
343 bzero(ndopts, sizeof(*ndopts));
350 switch (nd_opt->nd_opt_type) {
351 case ND_OPT_SOURCE_LINKADDR:
352 case ND_OPT_TARGET_LINKADDR:
354 case ND_OPT_REDIRECTED_HEADER:
355 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
357 "duplicated ND6 option found (type=%d)\n",
358 nd_opt->nd_opt_type));
361 ndopts->nd_opt_array[nd_opt->nd_opt_type]
365 case ND_OPT_PREFIX_INFORMATION:
366 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
367 ndopts->nd_opt_array[nd_opt->nd_opt_type]
370 ndopts->nd_opts_pi_end =
371 (struct nd_opt_prefix_info *)nd_opt;
375 * Unknown options must be silently ignored,
376 * to accomodate future extension to the protocol.
379 "nd6_options: unsupported option %d - "
380 "option ignored\n", nd_opt->nd_opt_type));
385 if (i > V_nd6_maxndopt) {
386 ICMP6STAT_INC(icp6s_nd_toomanyopt);
387 nd6log((LOG_INFO, "too many loop in nd opt\n"));
391 if (ndopts->nd_opts_done)
399 * ND6 timer routine to handle ND6 entries
402 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
406 LLE_WLOCK_ASSERT(ln);
411 canceled = callout_stop(&ln->ln_timer_ch);
413 ln->la_expire = time_second + tick / hz;
415 if (tick > INT_MAX) {
416 ln->ln_ntick = tick - INT_MAX;
417 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
418 nd6_llinfo_timer, ln);
421 canceled = callout_reset(&ln->ln_timer_ch, tick,
422 nd6_llinfo_timer, ln);
430 nd6_llinfo_settimer(struct llentry *ln, long tick)
434 nd6_llinfo_settimer_locked(ln, tick);
439 nd6_llinfo_timer(void *arg)
442 struct in6_addr *dst;
444 struct nd_ifinfo *ndi = NULL;
446 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
447 ln = (struct llentry *)arg;
448 LLE_WLOCK_ASSERT(ln);
449 ifp = ln->lle_tbl->llt_ifp;
451 CURVNET_SET(ifp->if_vnet);
453 if (ln->ln_ntick > 0) {
454 if (ln->ln_ntick > INT_MAX) {
455 ln->ln_ntick -= INT_MAX;
456 nd6_llinfo_settimer_locked(ln, INT_MAX);
459 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
464 ndi = ND_IFINFO(ifp);
465 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
466 if (ln->la_flags & LLE_STATIC) {
470 if (ln->la_flags & LLE_DELETED) {
471 (void)nd6_free(ln, 0);
476 switch (ln->ln_state) {
477 case ND6_LLINFO_INCOMPLETE:
478 if (ln->la_asked < V_nd6_mmaxtries) {
480 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
482 nd6_ns_output(ifp, NULL, dst, ln, 0);
485 struct mbuf *m = ln->la_hold;
490 * assuming every packet in la_hold has the
491 * same IP header. Send error after unlock.
496 clear_llinfo_pqueue(ln);
498 (void)nd6_free(ln, 0);
501 icmp6_error2(m, ICMP6_DST_UNREACH,
502 ICMP6_DST_UNREACH_ADDR, 0, ifp);
505 case ND6_LLINFO_REACHABLE:
506 if (!ND6_LLINFO_PERMANENT(ln)) {
507 ln->ln_state = ND6_LLINFO_STALE;
508 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
512 case ND6_LLINFO_STALE:
513 /* Garbage Collection(RFC 2461 5.3) */
514 if (!ND6_LLINFO_PERMANENT(ln)) {
515 (void)nd6_free(ln, 1);
520 case ND6_LLINFO_DELAY:
521 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
524 ln->ln_state = ND6_LLINFO_PROBE;
525 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
527 nd6_ns_output(ifp, dst, dst, ln, 0);
530 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
531 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
534 case ND6_LLINFO_PROBE:
535 if (ln->la_asked < V_nd6_umaxtries) {
537 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
539 nd6_ns_output(ifp, dst, dst, ln, 0);
542 (void)nd6_free(ln, 0);
547 panic("%s: paths in a dark night can be confusing: %d",
548 __func__, ln->ln_state);
558 * ND6 timer routine to expire default route list and prefix list
563 CURVNET_SET((struct vnet *) arg);
565 struct nd_defrouter *dr, *ndr;
566 struct nd_prefix *pr, *npr;
567 struct in6_ifaddr *ia6, *nia6;
569 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
572 /* expire default router list */
574 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
575 if (dr->expire && dr->expire < time_second)
580 * expire interface addresses.
581 * in the past the loop was inside prefix expiry processing.
582 * However, from a stricter speci-confrmance standpoint, we should
583 * rather separate address lifetimes and prefix lifetimes.
585 * XXXRW: in6_ifaddrhead locking.
588 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
589 /* check address lifetime */
590 if (IFA6_IS_INVALID(ia6)) {
594 * If the expiring address is temporary, try
595 * regenerating a new one. This would be useful when
596 * we suspended a laptop PC, then turned it on after a
597 * period that could invalidate all temporary
598 * addresses. Although we may have to restart the
599 * loop (see below), it must be after purging the
600 * address. Otherwise, we'd see an infinite loop of
603 if (V_ip6_use_tempaddr &&
604 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
605 if (regen_tmpaddr(ia6) == 0)
609 in6_purgeaddr(&ia6->ia_ifa);
612 goto addrloop; /* XXX: see below */
613 } else if (IFA6_IS_DEPRECATED(ia6)) {
614 int oldflags = ia6->ia6_flags;
616 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
619 * If a temporary address has just become deprecated,
620 * regenerate a new one if possible.
622 if (V_ip6_use_tempaddr &&
623 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
624 (oldflags & IN6_IFF_DEPRECATED) == 0) {
626 if (regen_tmpaddr(ia6) == 0) {
628 * A new temporary address is
630 * XXX: this means the address chain
631 * has changed while we are still in
632 * the loop. Although the change
633 * would not cause disaster (because
634 * it's not a deletion, but an
635 * addition,) we'd rather restart the
636 * loop just for safety. Or does this
637 * significantly reduce performance??
644 * A new RA might have made a deprecated address
647 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
651 /* expire prefix list */
652 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
654 * check prefix lifetime.
655 * since pltime is just for autoconf, pltime processing for
656 * prefix is not necessary.
658 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
659 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
662 * address expiration and prefix expiration are
663 * separate. NEVER perform in6_purgeaddr here.
673 * ia6 - deprecated/invalidated temporary address
676 regen_tmpaddr(struct in6_ifaddr *ia6)
680 struct in6_ifaddr *public_ifa6 = NULL;
682 ifp = ia6->ia_ifa.ifa_ifp;
684 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
685 struct in6_ifaddr *it6;
687 if (ifa->ifa_addr->sa_family != AF_INET6)
690 it6 = (struct in6_ifaddr *)ifa;
692 /* ignore no autoconf addresses. */
693 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
696 /* ignore autoconf addresses with different prefixes. */
697 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
701 * Now we are looking at an autoconf address with the same
702 * prefix as ours. If the address is temporary and is still
703 * preferred, do not create another one. It would be rare, but
704 * could happen, for example, when we resume a laptop PC after
707 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
708 !IFA6_IS_DEPRECATED(it6)) {
714 * This is a public autoconf address that has the same prefix
715 * as ours. If it is preferred, keep it. We can't break the
716 * loop here, because there may be a still-preferred temporary
717 * address with the prefix.
719 if (!IFA6_IS_DEPRECATED(it6))
722 if (public_ifa6 != NULL)
723 ifa_ref(&public_ifa6->ia_ifa);
725 IF_ADDR_RUNLOCK(ifp);
727 if (public_ifa6 != NULL) {
730 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
731 ifa_free(&public_ifa6->ia_ifa);
732 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
733 " tmp addr,errno=%d\n", e);
736 ifa_free(&public_ifa6->ia_ifa);
744 * Nuke neighbor cache/prefix/default router management table, right before
748 nd6_purge(struct ifnet *ifp)
750 struct nd_defrouter *dr, *ndr;
751 struct nd_prefix *pr, *npr;
754 * Nuke default router list entries toward ifp.
755 * We defer removal of default router list entries that is installed
756 * in the routing table, in order to keep additional side effects as
759 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
767 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
775 /* Nuke prefix list entries toward ifp */
776 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
777 if (pr->ndpr_ifp == ifp) {
779 * Because if_detach() does *not* release prefixes
780 * while purging addresses the reference count will
781 * still be above zero. We therefore reset it to
782 * make sure that the prefix really gets purged.
787 * Previously, pr->ndpr_addr is removed as well,
788 * but I strongly believe we don't have to do it.
789 * nd6_purge() is only called from in6_ifdetach(),
790 * which removes all the associated interface addresses
792 * (jinmei@kame.net 20010129)
798 /* cancel default outgoing interface setting */
799 if (V_nd6_defifindex == ifp->if_index)
800 nd6_setdefaultiface(0);
802 if (!V_ip6_forwarding && V_ip6_accept_rtadv) { /* XXX: too restrictive? */
803 /* refresh default router list
812 * We do not nuke the neighbor cache entries here any more
813 * because the neighbor cache is kept in if_afdata[AF_INET6].
814 * nd6_purge() is invoked by in6_ifdetach() which is called
815 * from if_detach() where everything gets purged. So let
816 * in6_domifdetach() do the actual L2 table purging work.
821 * the caller acquires and releases the lock on the lltbls
822 * Returns the llentry locked
825 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
827 struct sockaddr_in6 sin6;
831 bzero(&sin6, sizeof(sin6));
832 sin6.sin6_len = sizeof(struct sockaddr_in6);
833 sin6.sin6_family = AF_INET6;
834 sin6.sin6_addr = *addr6;
836 IF_AFDATA_LOCK_ASSERT(ifp);
839 if (flags & ND6_CREATE)
840 llflags |= LLE_CREATE;
841 if (flags & ND6_EXCLUSIVE)
842 llflags |= LLE_EXCLUSIVE;
844 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
845 if ((ln != NULL) && (llflags & LLE_CREATE))
846 ln->ln_state = ND6_LLINFO_NOSTATE;
852 * Test whether a given IPv6 address is a neighbor or not, ignoring
853 * the actual neighbor cache. The neighbor cache is ignored in order
854 * to not reenter the routing code from within itself.
857 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
859 struct nd_prefix *pr;
860 struct ifaddr *dstaddr;
863 * A link-local address is always a neighbor.
864 * XXX: a link does not necessarily specify a single interface.
866 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
867 struct sockaddr_in6 sin6_copy;
871 * We need sin6_copy since sa6_recoverscope() may modify the
875 if (sa6_recoverscope(&sin6_copy))
876 return (0); /* XXX: should be impossible */
877 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
879 if (sin6_copy.sin6_scope_id == zone)
886 * If the address matches one of our addresses,
887 * it should be a neighbor.
888 * If the address matches one of our on-link prefixes, it should be a
891 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
892 if (pr->ndpr_ifp != ifp)
895 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
898 /* Always use the default FIB here. */
899 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
900 0, 0, RT_DEFAULT_FIB);
904 * This is the case where multiple interfaces
905 * have the same prefix, but only one is installed
906 * into the routing table and that prefix entry
907 * is not the one being examined here. In the case
908 * where RADIX_MPATH is enabled, multiple route
909 * entries (of the same rt_key value) will be
910 * installed because the interface addresses all
913 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
914 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
921 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
922 &addr->sin6_addr, &pr->ndpr_mask))
927 * If the address is assigned on the node of the other side of
928 * a p2p interface, the address should be a neighbor.
930 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
931 if (dstaddr != NULL) {
932 if (dstaddr->ifa_ifp == ifp) {
940 * If the default router list is empty, all addresses are regarded
941 * as on-link, and thus, as a neighbor.
942 * XXX: we restrict the condition to hosts, because routers usually do
943 * not have the "default router list".
945 if (!V_ip6_forwarding && TAILQ_EMPTY(&V_nd_defrouter) &&
946 V_nd6_defifindex == ifp->if_index) {
955 * Detect if a given IPv6 address identifies a neighbor on a given link.
956 * XXX: should take care of the destination of a p2p link?
959 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
964 IF_AFDATA_UNLOCK_ASSERT(ifp);
965 if (nd6_is_new_addr_neighbor(addr, ifp))
969 * Even if the address matches none of our addresses, it might be
970 * in the neighbor cache.
973 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
977 IF_AFDATA_UNLOCK(ifp);
982 * Free an nd6 llinfo entry.
983 * Since the function would cause significant changes in the kernel, DO NOT
984 * make it global, unless you have a strong reason for the change, and are sure
985 * that the change is safe.
987 static struct llentry *
988 nd6_free(struct llentry *ln, int gc)
990 struct llentry *next;
991 struct nd_defrouter *dr;
994 LLE_WLOCK_ASSERT(ln);
997 * we used to have pfctlinput(PRC_HOSTDEAD) here.
998 * even though it is not harmful, it was not really necessary.
1002 nd6_llinfo_settimer_locked(ln, -1);
1004 ifp = ln->lle_tbl->llt_ifp;
1006 if (!V_ip6_forwarding) {
1008 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1010 if (dr != NULL && dr->expire &&
1011 ln->ln_state == ND6_LLINFO_STALE && gc) {
1013 * If the reason for the deletion is just garbage
1014 * collection, and the neighbor is an active default
1015 * router, do not delete it. Instead, reset the GC
1016 * timer using the router's lifetime.
1017 * Simply deleting the entry would affect default
1018 * router selection, which is not necessarily a good
1019 * thing, especially when we're using router preference
1021 * XXX: the check for ln_state would be redundant,
1022 * but we intentionally keep it just in case.
1024 if (dr->expire > time_second)
1025 nd6_llinfo_settimer_locked(ln,
1026 (dr->expire - time_second) * hz);
1028 nd6_llinfo_settimer_locked(ln,
1029 (long)V_nd6_gctimer * hz);
1031 next = LIST_NEXT(ln, lle_next);
1039 * Unreachablity of a router might affect the default
1040 * router selection and on-link detection of advertised
1045 * Temporarily fake the state to choose a new default
1046 * router and to perform on-link determination of
1047 * prefixes correctly.
1048 * Below the state will be set correctly,
1049 * or the entry itself will be deleted.
1051 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1054 if (ln->ln_router || dr) {
1057 * We need to unlock to avoid a LOR with rt6_flush() with the
1058 * rnh and for the calls to pfxlist_onlink_check() and
1059 * defrouter_select() in the block further down for calls
1060 * into nd6_lookup(). We still hold a ref.
1065 * rt6_flush must be called whether or not the neighbor
1066 * is in the Default Router List.
1067 * See a corresponding comment in nd6_na_input().
1069 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1074 * Since defrouter_select() does not affect the
1075 * on-link determination and MIP6 needs the check
1076 * before the default router selection, we perform
1079 pfxlist_onlink_check();
1082 * Refresh default router list.
1087 if (ln->ln_router || dr)
1092 * Before deleting the entry, remember the next entry as the
1093 * return value. We need this because pfxlist_onlink_check() above
1094 * might have freed other entries (particularly the old next entry) as
1095 * a side effect (XXX).
1097 next = LIST_NEXT(ln, lle_next);
1100 * Save to unlock. We still hold an extra reference and will not
1101 * free(9) in llentry_free() if someone else holds one as well.
1104 IF_AFDATA_LOCK(ifp);
1108 IF_AFDATA_UNLOCK(ifp);
1114 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1116 * XXX cost-effective methods?
1119 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1124 if ((dst6 == NULL) || (rt == NULL))
1128 IF_AFDATA_LOCK(ifp);
1129 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1130 IF_AFDATA_UNLOCK(ifp);
1134 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1138 * if we get upper-layer reachability confirmation many times,
1139 * it is possible we have false information.
1143 if (ln->ln_byhint > V_nd6_maxnudhint) {
1148 ln->ln_state = ND6_LLINFO_REACHABLE;
1149 if (!ND6_LLINFO_PERMANENT(ln)) {
1150 nd6_llinfo_settimer_locked(ln,
1151 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1159 * Rejuvenate this function for routing operations related
1163 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1165 struct sockaddr_in6 *gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1166 struct nd_defrouter *dr;
1167 struct ifnet *ifp = rt->rt_ifp;
1179 * Only indirect routes are interesting.
1181 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1184 * check for default route
1186 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1187 &SIN6(rt_key(rt))->sin6_addr)) {
1189 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1199 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1201 struct in6_drlist *drl = (struct in6_drlist *)data;
1202 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1203 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1204 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1205 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1206 struct nd_defrouter *dr;
1207 struct nd_prefix *pr;
1208 int i = 0, error = 0;
1212 case SIOCGDRLST_IN6:
1214 * obsolete API, use sysctl under net.inet6.icmp6
1216 bzero(drl, sizeof(*drl));
1218 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1221 drl->defrouter[i].rtaddr = dr->rtaddr;
1222 in6_clearscope(&drl->defrouter[i].rtaddr);
1224 drl->defrouter[i].flags = dr->flags;
1225 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1226 drl->defrouter[i].expire = dr->expire;
1227 drl->defrouter[i].if_index = dr->ifp->if_index;
1232 case SIOCGPRLST_IN6:
1234 * obsolete API, use sysctl under net.inet6.icmp6
1236 * XXX the structure in6_prlist was changed in backward-
1237 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1238 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1241 * XXX meaning of fields, especialy "raflags", is very
1242 * differnet between RA prefix list and RR/static prefix list.
1243 * how about separating ioctls into two?
1245 bzero(oprl, sizeof(*oprl));
1247 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1248 struct nd_pfxrouter *pfr;
1253 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1254 oprl->prefix[i].raflags = pr->ndpr_raf;
1255 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1256 oprl->prefix[i].vltime = pr->ndpr_vltime;
1257 oprl->prefix[i].pltime = pr->ndpr_pltime;
1258 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1259 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1260 oprl->prefix[i].expire = 0;
1264 /* XXX: we assume time_t is signed. */
1267 ((sizeof(maxexpire) * 8) - 1));
1268 if (pr->ndpr_vltime <
1269 maxexpire - pr->ndpr_lastupdate) {
1270 oprl->prefix[i].expire =
1271 pr->ndpr_lastupdate +
1274 oprl->prefix[i].expire = maxexpire;
1278 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1280 #define RTRADDR oprl->prefix[i].advrtr[j]
1281 RTRADDR = pfr->router->rtaddr;
1282 in6_clearscope(&RTRADDR);
1287 oprl->prefix[i].advrtrs = j;
1288 oprl->prefix[i].origin = PR_ORIG_RA;
1295 case OSIOCGIFINFO_IN6:
1297 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1298 bzero(&ND, sizeof(ND));
1299 ND.linkmtu = IN6_LINKMTU(ifp);
1300 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1301 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1302 ND.reachable = ND_IFINFO(ifp)->reachable;
1303 ND.retrans = ND_IFINFO(ifp)->retrans;
1304 ND.flags = ND_IFINFO(ifp)->flags;
1305 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1306 ND.chlim = ND_IFINFO(ifp)->chlim;
1308 case SIOCGIFINFO_IN6:
1309 ND = *ND_IFINFO(ifp);
1311 case SIOCSIFINFO_IN6:
1313 * used to change host variables from userland.
1314 * intented for a use on router to reflect RA configurations.
1316 /* 0 means 'unspecified' */
1317 if (ND.linkmtu != 0) {
1318 if (ND.linkmtu < IPV6_MMTU ||
1319 ND.linkmtu > IN6_LINKMTU(ifp)) {
1323 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1326 if (ND.basereachable != 0) {
1327 int obasereachable = ND_IFINFO(ifp)->basereachable;
1329 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1330 if (ND.basereachable != obasereachable)
1331 ND_IFINFO(ifp)->reachable =
1332 ND_COMPUTE_RTIME(ND.basereachable);
1334 if (ND.retrans != 0)
1335 ND_IFINFO(ifp)->retrans = ND.retrans;
1337 ND_IFINFO(ifp)->chlim = ND.chlim;
1339 case SIOCSIFINFO_FLAGS:
1340 ND_IFINFO(ifp)->flags = ND.flags;
1343 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1344 /* sync kernel routing table with the default router list */
1348 case SIOCSPFXFLUSH_IN6:
1350 /* flush all the prefix advertised by routers */
1351 struct nd_prefix *pr, *next;
1354 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1355 struct in6_ifaddr *ia, *ia_next;
1357 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1360 /* do we really have to remove addresses as well? */
1361 /* XXXRW: in6_ifaddrhead locking. */
1362 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1364 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1367 if (ia->ia6_ndpr == pr)
1368 in6_purgeaddr(&ia->ia_ifa);
1375 case SIOCSRTRFLUSH_IN6:
1377 /* flush all the default routers */
1378 struct nd_defrouter *dr, *next;
1382 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1389 case SIOCGNBRINFO_IN6:
1392 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1394 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1397 IF_AFDATA_LOCK(ifp);
1398 ln = nd6_lookup(&nb_addr, 0, ifp);
1399 IF_AFDATA_UNLOCK(ifp);
1405 nbi->state = ln->ln_state;
1406 nbi->asked = ln->la_asked;
1407 nbi->isrouter = ln->ln_router;
1408 nbi->expire = ln->la_expire;
1412 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1413 ndif->ifindex = V_nd6_defifindex;
1415 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1416 return (nd6_setdefaultiface(ndif->ifindex));
1422 * Create neighbor cache entry and cache link-layer address,
1423 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1426 * code - type dependent information
1429 * The caller of this function already acquired the ndp
1430 * cache table lock because the cache entry is returned.
1433 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1434 int lladdrlen, int type, int code)
1436 struct llentry *ln = NULL;
1443 uint16_t router = 0;
1444 struct sockaddr_in6 sin6;
1445 struct mbuf *chain = NULL;
1446 int static_route = 0;
1448 IF_AFDATA_UNLOCK_ASSERT(ifp);
1451 panic("ifp == NULL in nd6_cache_lladdr");
1453 panic("from == NULL in nd6_cache_lladdr");
1455 /* nothing must be updated for unspecified address */
1456 if (IN6_IS_ADDR_UNSPECIFIED(from))
1460 * Validation about ifp->if_addrlen and lladdrlen must be done in
1463 * XXX If the link does not have link-layer adderss, what should
1464 * we do? (ifp->if_addrlen == 0)
1465 * Spec says nothing in sections for RA, RS and NA. There's small
1466 * description on it in NS section (RFC 2461 7.2.3).
1468 flags = lladdr ? ND6_EXCLUSIVE : 0;
1469 IF_AFDATA_LOCK(ifp);
1470 ln = nd6_lookup(from, flags, ifp);
1473 flags |= ND6_EXCLUSIVE;
1474 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1475 IF_AFDATA_UNLOCK(ifp);
1478 IF_AFDATA_UNLOCK(ifp);
1479 /* do nothing if static ndp is set */
1480 if (ln->la_flags & LLE_STATIC) {
1489 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1490 if (olladdr && lladdr) {
1491 llchange = bcmp(lladdr, &ln->ll_addr,
1497 * newentry olladdr lladdr llchange (*=record)
1500 * 0 n y -- (3) * STALE
1502 * 0 y y y (5) * STALE
1503 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1504 * 1 -- y -- (7) * STALE
1507 if (lladdr) { /* (3-5) and (7) */
1509 * Record source link-layer address
1510 * XXX is it dependent to ifp->if_type?
1512 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1513 ln->la_flags |= LLE_VALID;
1517 if ((!olladdr && lladdr != NULL) || /* (3) */
1518 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1520 newstate = ND6_LLINFO_STALE;
1521 } else /* (1-2,4) */
1525 if (lladdr == NULL) /* (6) */
1526 newstate = ND6_LLINFO_NOSTATE;
1528 newstate = ND6_LLINFO_STALE;
1533 * Update the state of the neighbor cache.
1535 ln->ln_state = newstate;
1537 if (ln->ln_state == ND6_LLINFO_STALE) {
1539 * XXX: since nd6_output() below will cause
1540 * state tansition to DELAY and reset the timer,
1541 * we must set the timer now, although it is actually
1544 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1547 struct mbuf *m_hold, *m_hold_next;
1550 * reset the la_hold in advance, to explicitly
1551 * prevent a la_hold lookup in nd6_output()
1552 * (wouldn't happen, though...)
1554 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1555 m_hold; m_hold = m_hold_next) {
1556 m_hold_next = m_hold->m_nextpkt;
1557 m_hold->m_nextpkt = NULL;
1560 * we assume ifp is not a p2p here, so
1561 * just set the 2nd argument as the
1564 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1567 * If we have mbufs in the chain we need to do
1568 * deferred transmit. Copy the address from the
1569 * llentry before dropping the lock down below.
1572 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1574 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1575 /* probe right away */
1576 nd6_llinfo_settimer_locked((void *)ln, 0);
1581 * ICMP6 type dependent behavior.
1583 * NS: clear IsRouter if new entry
1584 * RS: clear IsRouter
1585 * RA: set IsRouter if there's lladdr
1586 * redir: clear IsRouter if new entry
1589 * The spec says that we must set IsRouter in the following cases:
1590 * - If lladdr exist, set IsRouter. This means (1-5).
1591 * - If it is old entry (!newentry), set IsRouter. This means (7).
1592 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1593 * A quetion arises for (1) case. (1) case has no lladdr in the
1594 * neighbor cache, this is similar to (6).
1595 * This case is rare but we figured that we MUST NOT set IsRouter.
1597 * newentry olladdr lladdr llchange NS RS RA redir
1599 * 0 n n -- (1) c ? s
1600 * 0 y n -- (2) c s s
1601 * 0 n y -- (3) c s s
1604 * 1 -- n -- (6) c c c s
1605 * 1 -- y -- (7) c c s c s
1609 switch (type & 0xff) {
1610 case ND_NEIGHBOR_SOLICIT:
1612 * New entry must have is_router flag cleared.
1614 if (is_newentry) /* (6-7) */
1619 * If the icmp is a redirect to a better router, always set the
1620 * is_router flag. Otherwise, if the entry is newly created,
1621 * clear the flag. [RFC 2461, sec 8.3]
1623 if (code == ND_REDIRECT_ROUTER)
1625 else if (is_newentry) /* (6-7) */
1628 case ND_ROUTER_SOLICIT:
1630 * is_router flag must always be cleared.
1634 case ND_ROUTER_ADVERT:
1636 * Mark an entry with lladdr as a router.
1638 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1639 (is_newentry && lladdr)) { /* (7) */
1646 static_route = (ln->la_flags & LLE_STATIC);
1647 router = ln->ln_router;
1649 if (flags & ND6_EXCLUSIVE)
1657 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1660 * When the link-layer address of a router changes, select the
1661 * best router again. In particular, when the neighbor entry is newly
1662 * created, it might affect the selection policy.
1663 * Question: can we restrict the first condition to the "is_newentry"
1665 * XXX: when we hear an RA from a new router with the link-layer
1666 * address option, defrouter_select() is called twice, since
1667 * defrtrlist_update called the function as well. However, I believe
1668 * we can compromise the overhead, since it only happens the first
1670 * XXX: although defrouter_select() should not have a bad effect
1671 * for those are not autoconfigured hosts, we explicitly avoid such
1674 if (do_update && router && !V_ip6_forwarding && V_ip6_accept_rtadv) {
1676 * guaranteed recursion
1684 if (flags & ND6_EXCLUSIVE)
1695 nd6_slowtimo(void *arg)
1697 CURVNET_SET((struct vnet *) arg);
1698 struct nd_ifinfo *nd6if;
1701 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1702 nd6_slowtimo, curvnet);
1703 IFNET_RLOCK_NOSLEEP();
1704 TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1705 nd6if = ND_IFINFO(ifp);
1706 if (nd6if->basereachable && /* already initialized */
1707 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1709 * Since reachable time rarely changes by router
1710 * advertisements, we SHOULD insure that a new random
1711 * value gets recomputed at least once every few hours.
1714 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1715 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1718 IFNET_RUNLOCK_NOSLEEP();
1723 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1724 struct sockaddr_in6 *dst, struct rtentry *rt0)
1727 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1732 * Note that I'm not enforcing any global serialization
1733 * lle state or asked changes here as the logic is too
1734 * complicated to avoid having to always acquire an exclusive
1739 #define senderr(e) { error = (e); goto bad;}
1742 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1743 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1744 struct mbuf **chain)
1746 struct mbuf *m = m0;
1747 struct llentry *ln = lle;
1754 LLE_WLOCK_ASSERT(lle);
1756 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1759 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1762 if (nd6_need_cache(ifp) == 0)
1766 * next hop determination. This routine is derived from ether_output.
1770 * Address resolution or Neighbor Unreachability Detection
1772 * At this point, the destination of the packet must be a unicast
1773 * or an anycast address(i.e. not a multicast).
1776 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1779 IF_AFDATA_LOCK(ifp);
1780 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1781 IF_AFDATA_UNLOCK(ifp);
1782 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1784 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1785 * the condition below is not very efficient. But we believe
1786 * it is tolerable, because this should be a rare case.
1788 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1789 IF_AFDATA_LOCK(ifp);
1790 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1791 IF_AFDATA_UNLOCK(ifp);
1795 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1796 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1797 char ip6buf[INET6_ADDRSTRLEN];
1799 "nd6_output: can't allocate llinfo for %s "
1801 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1802 senderr(EIO); /* XXX: good error? */
1804 goto sendpkt; /* send anyway */
1807 /* We don't have to do link-layer address resolution on a p2p link. */
1808 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1809 ln->ln_state < ND6_LLINFO_REACHABLE) {
1810 if ((flags & LLE_EXCLUSIVE) == 0) {
1811 flags |= LLE_EXCLUSIVE;
1814 ln->ln_state = ND6_LLINFO_STALE;
1815 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1819 * The first time we send a packet to a neighbor whose entry is
1820 * STALE, we have to change the state to DELAY and a sets a timer to
1821 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1822 * neighbor unreachability detection on expiration.
1825 if (ln->ln_state == ND6_LLINFO_STALE) {
1826 if ((flags & LLE_EXCLUSIVE) == 0) {
1827 flags |= LLE_EXCLUSIVE;
1832 ln->ln_state = ND6_LLINFO_DELAY;
1833 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1837 * If the neighbor cache entry has a state other than INCOMPLETE
1838 * (i.e. its link-layer address is already resolved), just
1841 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1845 * There is a neighbor cache entry, but no ethernet address
1846 * response yet. Append this latest packet to the end of the
1847 * packet queue in the mbuf, unless the number of the packet
1848 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1849 * the oldest packet in the queue will be removed.
1851 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1852 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1854 if ((flags & LLE_EXCLUSIVE) == 0) {
1855 flags |= LLE_EXCLUSIVE;
1860 LLE_WLOCK_ASSERT(ln);
1863 struct mbuf *m_hold;
1867 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1869 if (m_hold->m_nextpkt == NULL) {
1870 m_hold->m_nextpkt = m;
1874 while (i >= V_nd6_maxqueuelen) {
1875 m_hold = ln->la_hold;
1876 ln->la_hold = ln->la_hold->m_nextpkt;
1885 * If there has been no NS for the neighbor after entering the
1886 * INCOMPLETE state, send the first solicitation.
1888 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1891 nd6_llinfo_settimer_locked(ln,
1892 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1894 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1895 if (lle != NULL && ln == lle)
1898 } else if (lle == NULL || ln != lle) {
1900 * We did the lookup (no lle arg) so we
1901 * need to do the unlock here.
1909 /* discard the packet if IPv6 operation is disabled on the interface */
1910 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1911 error = ENETDOWN; /* better error? */
1915 * ln is valid and the caller did not pass in
1918 if ((ln != NULL) && (lle == NULL)) {
1919 if (flags & LLE_EXCLUSIVE)
1926 mac_netinet6_nd6_send(ifp, m);
1929 * We were passed in a pointer to an lle with the lock held
1930 * this means that we can't call if_output as we will
1931 * recurse on the lle lock - so what we do is we create
1932 * a list of mbufs to send and transmit them in the caller
1933 * after the lock is dropped
1942 * append mbuf to end of deferred chain
1945 while (mb->m_nextpkt != NULL)
1951 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1952 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1955 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1960 * ln is valid and the caller did not pass in
1963 if ((ln != NULL) && (lle == NULL)) {
1964 if (flags & LLE_EXCLUSIVE)
1977 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
1978 struct sockaddr_in6 *dst, struct route *ro)
1980 struct mbuf *m, *m_head;
1981 struct ifnet *outifp;
1985 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
1992 m_head = m_head->m_nextpkt;
1993 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
1998 * note that intermediate errors are blindly ignored - but this is
1999 * the same convention as used with nd6_output when called by
2007 nd6_need_cache(struct ifnet *ifp)
2010 * XXX: we currently do not make neighbor cache on any interface
2011 * other than ARCnet, Ethernet, FDDI and GIF.
2014 * - unidirectional tunnels needs no ND
2016 switch (ifp->if_type) {
2024 #ifdef IFT_IEEE80211
2030 case IFT_GIF: /* XXX need more cases? */
2034 case IFT_PROPVIRTUAL:
2042 * the callers of this function need to be re-worked to drop
2043 * the lle lock, drop here for now
2046 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2047 struct sockaddr *dst, u_char *desten, struct llentry **lle)
2052 IF_AFDATA_UNLOCK_ASSERT(ifp);
2053 if (m->m_flags & M_MCAST) {
2056 switch (ifp->if_type) {
2062 #ifdef IFT_IEEE80211
2067 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2072 * netbsd can use if_broadcastaddr, but we don't do so
2073 * to reduce # of ifdef.
2075 for (i = 0; i < ifp->if_addrlen; i++)
2083 return (EAFNOSUPPORT);
2089 * the entry should have been created in nd6_store_lladdr
2091 IF_AFDATA_LOCK(ifp);
2092 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2093 IF_AFDATA_UNLOCK(ifp);
2094 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2097 /* this could happen, if we could not allocate memory */
2102 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2106 * A *small* use after free race exists here
2112 clear_llinfo_pqueue(struct llentry *ln)
2114 struct mbuf *m_hold, *m_hold_next;
2116 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2117 m_hold_next = m_hold->m_nextpkt;
2118 m_hold->m_nextpkt = NULL;
2126 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2127 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2129 SYSCTL_DECL(_net_inet6_icmp6);
2131 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2132 CTLFLAG_RD, nd6_sysctl_drlist, "");
2133 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2134 CTLFLAG_RD, nd6_sysctl_prlist, "");
2135 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2136 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2139 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2142 char buf[1024] __aligned(4);
2143 struct in6_defrouter *d, *de;
2144 struct nd_defrouter *dr;
2150 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2151 d = (struct in6_defrouter *)buf;
2152 de = (struct in6_defrouter *)(buf + sizeof(buf));
2155 bzero(d, sizeof(*d));
2156 d->rtaddr.sin6_family = AF_INET6;
2157 d->rtaddr.sin6_len = sizeof(d->rtaddr);
2158 d->rtaddr.sin6_addr = dr->rtaddr;
2159 error = sa6_recoverscope(&d->rtaddr);
2162 d->flags = dr->flags;
2163 d->rtlifetime = dr->rtlifetime;
2164 d->expire = dr->expire;
2165 d->if_index = dr->ifp->if_index;
2167 panic("buffer too short");
2169 error = SYSCTL_OUT(req, buf, sizeof(*d));
2178 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2181 char buf[1024] __aligned(4);
2182 struct in6_prefix *p, *pe;
2183 struct nd_prefix *pr;
2184 char ip6buf[INET6_ADDRSTRLEN];
2190 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2193 struct sockaddr_in6 *sin6, *s6;
2194 struct nd_pfxrouter *pfr;
2196 p = (struct in6_prefix *)buf;
2197 pe = (struct in6_prefix *)(buf + sizeof(buf));
2200 bzero(p, sizeof(*p));
2201 sin6 = (struct sockaddr_in6 *)(p + 1);
2203 p->prefix = pr->ndpr_prefix;
2204 if (sa6_recoverscope(&p->prefix)) {
2206 "scope error in prefix list (%s)\n",
2207 ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2208 /* XXX: press on... */
2210 p->raflags = pr->ndpr_raf;
2211 p->prefixlen = pr->ndpr_plen;
2212 p->vltime = pr->ndpr_vltime;
2213 p->pltime = pr->ndpr_pltime;
2214 p->if_index = pr->ndpr_ifp->if_index;
2215 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2220 /* XXX: we assume time_t is signed. */
2223 ((sizeof(maxexpire) * 8) - 1));
2224 if (pr->ndpr_vltime <
2225 maxexpire - pr->ndpr_lastupdate) {
2226 p->expire = pr->ndpr_lastupdate +
2229 p->expire = maxexpire;
2231 p->refcnt = pr->ndpr_refcnt;
2232 p->flags = pr->ndpr_stateflags;
2233 p->origin = PR_ORIG_RA;
2235 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2236 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2240 s6 = &sin6[advrtrs];
2241 bzero(s6, sizeof(*s6));
2242 s6->sin6_family = AF_INET6;
2243 s6->sin6_len = sizeof(*sin6);
2244 s6->sin6_addr = pfr->router->rtaddr;
2245 if (sa6_recoverscope(s6)) {
2248 "prefix list (%s)\n",
2250 &pfr->router->rtaddr));
2254 p->advrtrs = advrtrs;
2256 panic("buffer too short");
2258 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2259 error = SYSCTL_OUT(req, buf, advance);