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 <netinet6/in6_ifattach.h>
74 #include <netinet/icmp6.h>
75 #include <netinet6/send.h>
77 #include <sys/limits.h>
79 #include <security/mac/mac_framework.h>
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
84 #define SIN6(s) ((struct sockaddr_in6 *)s)
87 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */
88 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */
89 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */
90 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */
91 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for
93 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
96 /* preventing too many loops in ND option parsing */
97 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
99 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper
101 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
103 #define V_nd6_maxndopt VNET(nd6_maxndopt)
104 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
107 VNET_DEFINE(int, nd6_debug) = 1;
109 VNET_DEFINE(int, nd6_debug) = 0;
114 static int nd6_inuse, nd6_allocated;
117 VNET_DEFINE(struct nd_drhead, nd_defrouter);
118 VNET_DEFINE(struct nd_prhead, nd_prefix);
120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
121 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
123 static struct sockaddr_in6 all1_sa;
125 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
127 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
130 static void nd6_slowtimo(void *);
131 static int regen_tmpaddr(struct in6_ifaddr *);
132 static struct llentry *nd6_free(struct llentry *, int);
133 static void nd6_llinfo_timer(void *);
134 static void clear_llinfo_pqueue(struct llentry *);
136 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
137 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
139 VNET_DEFINE(struct callout, nd6_timer_ch);
146 LIST_INIT(&V_nd_prefix);
148 all1_sa.sin6_family = AF_INET6;
149 all1_sa.sin6_len = sizeof(struct sockaddr_in6);
150 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
151 all1_sa.sin6_addr.s6_addr[i] = 0xff;
153 /* initialization of the default router list */
154 TAILQ_INIT(&V_nd_defrouter);
157 callout_init(&V_nd6_slowtimo_ch, 0);
158 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
159 nd6_slowtimo, curvnet);
167 callout_drain(&V_nd6_slowtimo_ch);
168 callout_drain(&V_nd6_timer_ch);
173 nd6_ifattach(struct ifnet *ifp)
175 struct nd_ifinfo *nd;
177 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
178 bzero(nd, sizeof(*nd));
182 nd->chlim = IPV6_DEFHLIM;
183 nd->basereachable = REACHABLE_TIME;
184 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
185 nd->retrans = RETRANS_TIMER;
187 nd->flags = ND6_IFF_PERFORMNUD;
189 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
190 if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
191 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
193 /* A loopback interface does not need to accept RTADV. */
194 if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
195 nd->flags |= ND6_IFF_ACCEPT_RTADV;
197 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
198 nd6_setmtu0(ifp, nd);
204 nd6_ifdetach(struct nd_ifinfo *nd)
211 * Reset ND level link MTU. This function is called when the physical MTU
212 * changes, which means we might have to adjust the ND level MTU.
215 nd6_setmtu(struct ifnet *ifp)
218 nd6_setmtu0(ifp, ND_IFINFO(ifp));
221 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
223 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
227 omaxmtu = ndi->maxmtu;
229 switch (ifp->if_type) {
231 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
234 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
237 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
240 ndi->maxmtu = ifp->if_mtu;
245 * Decreasing the interface MTU under IPV6 minimum MTU may cause
246 * undesirable situation. We thus notify the operator of the change
247 * explicitly. The check for omaxmtu is necessary to restrict the
248 * log to the case of changing the MTU, not initializing it.
250 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
251 log(LOG_NOTICE, "nd6_setmtu0: "
252 "new link MTU on %s (%lu) is too small for IPv6\n",
253 if_name(ifp), (unsigned long)ndi->maxmtu);
256 if (ndi->maxmtu > V_in6_maxmtu)
257 in6_setmaxmtu(); /* check all interfaces just in case */
262 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
265 bzero(ndopts, sizeof(*ndopts));
266 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
268 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
271 ndopts->nd_opts_done = 1;
272 ndopts->nd_opts_search = NULL;
277 * Take one ND option.
280 nd6_option(union nd_opts *ndopts)
282 struct nd_opt_hdr *nd_opt;
286 panic("ndopts == NULL in nd6_option");
287 if (ndopts->nd_opts_last == NULL)
288 panic("uninitialized ndopts in nd6_option");
289 if (ndopts->nd_opts_search == NULL)
291 if (ndopts->nd_opts_done)
294 nd_opt = ndopts->nd_opts_search;
296 /* make sure nd_opt_len is inside the buffer */
297 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
298 bzero(ndopts, sizeof(*ndopts));
302 olen = nd_opt->nd_opt_len << 3;
305 * Message validation requires that all included
306 * options have a length that is greater than zero.
308 bzero(ndopts, sizeof(*ndopts));
312 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
313 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
314 /* option overruns the end of buffer, invalid */
315 bzero(ndopts, sizeof(*ndopts));
317 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
318 /* reached the end of options chain */
319 ndopts->nd_opts_done = 1;
320 ndopts->nd_opts_search = NULL;
326 * Parse multiple ND options.
327 * This function is much easier to use, for ND routines that do not need
328 * multiple options of the same type.
331 nd6_options(union nd_opts *ndopts)
333 struct nd_opt_hdr *nd_opt;
337 panic("ndopts == NULL in nd6_options");
338 if (ndopts->nd_opts_last == NULL)
339 panic("uninitialized ndopts in nd6_options");
340 if (ndopts->nd_opts_search == NULL)
344 nd_opt = nd6_option(ndopts);
345 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
347 * Message validation requires that all included
348 * options have a length that is greater than zero.
350 ICMP6STAT_INC(icp6s_nd_badopt);
351 bzero(ndopts, sizeof(*ndopts));
358 switch (nd_opt->nd_opt_type) {
359 case ND_OPT_SOURCE_LINKADDR:
360 case ND_OPT_TARGET_LINKADDR:
362 case ND_OPT_REDIRECTED_HEADER:
363 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
365 "duplicated ND6 option found (type=%d)\n",
366 nd_opt->nd_opt_type));
369 ndopts->nd_opt_array[nd_opt->nd_opt_type]
373 case ND_OPT_PREFIX_INFORMATION:
374 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
375 ndopts->nd_opt_array[nd_opt->nd_opt_type]
378 ndopts->nd_opts_pi_end =
379 (struct nd_opt_prefix_info *)nd_opt;
383 * Unknown options must be silently ignored,
384 * to accomodate future extension to the protocol.
387 "nd6_options: unsupported option %d - "
388 "option ignored\n", nd_opt->nd_opt_type));
393 if (i > V_nd6_maxndopt) {
394 ICMP6STAT_INC(icp6s_nd_toomanyopt);
395 nd6log((LOG_INFO, "too many loop in nd opt\n"));
399 if (ndopts->nd_opts_done)
407 * ND6 timer routine to handle ND6 entries
410 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
414 LLE_WLOCK_ASSERT(ln);
419 canceled = callout_stop(&ln->ln_timer_ch);
421 ln->la_expire = time_second + tick / hz;
423 if (tick > INT_MAX) {
424 ln->ln_ntick = tick - INT_MAX;
425 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
426 nd6_llinfo_timer, ln);
429 canceled = callout_reset(&ln->ln_timer_ch, tick,
430 nd6_llinfo_timer, ln);
438 nd6_llinfo_settimer(struct llentry *ln, long tick)
442 nd6_llinfo_settimer_locked(ln, tick);
447 nd6_llinfo_timer(void *arg)
450 struct in6_addr *dst;
452 struct nd_ifinfo *ndi = NULL;
454 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
455 ln = (struct llentry *)arg;
456 LLE_WLOCK_ASSERT(ln);
457 ifp = ln->lle_tbl->llt_ifp;
459 CURVNET_SET(ifp->if_vnet);
461 if (ln->ln_ntick > 0) {
462 if (ln->ln_ntick > INT_MAX) {
463 ln->ln_ntick -= INT_MAX;
464 nd6_llinfo_settimer_locked(ln, INT_MAX);
467 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
472 ndi = ND_IFINFO(ifp);
473 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
474 if (ln->la_flags & LLE_STATIC) {
478 if (ln->la_flags & LLE_DELETED) {
479 (void)nd6_free(ln, 0);
484 switch (ln->ln_state) {
485 case ND6_LLINFO_INCOMPLETE:
486 if (ln->la_asked < V_nd6_mmaxtries) {
488 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
490 nd6_ns_output(ifp, NULL, dst, ln, 0);
493 struct mbuf *m = ln->la_hold;
498 * assuming every packet in la_hold has the
499 * same IP header. Send error after unlock.
504 clear_llinfo_pqueue(ln);
506 (void)nd6_free(ln, 0);
509 icmp6_error2(m, ICMP6_DST_UNREACH,
510 ICMP6_DST_UNREACH_ADDR, 0, ifp);
513 case ND6_LLINFO_REACHABLE:
514 if (!ND6_LLINFO_PERMANENT(ln)) {
515 ln->ln_state = ND6_LLINFO_STALE;
516 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
520 case ND6_LLINFO_STALE:
521 /* Garbage Collection(RFC 2461 5.3) */
522 if (!ND6_LLINFO_PERMANENT(ln)) {
523 (void)nd6_free(ln, 1);
528 case ND6_LLINFO_DELAY:
529 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
532 ln->ln_state = ND6_LLINFO_PROBE;
533 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
535 nd6_ns_output(ifp, dst, dst, ln, 0);
538 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
539 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
542 case ND6_LLINFO_PROBE:
543 if (ln->la_asked < V_nd6_umaxtries) {
545 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
547 nd6_ns_output(ifp, dst, dst, ln, 0);
550 (void)nd6_free(ln, 0);
555 panic("%s: paths in a dark night can be confusing: %d",
556 __func__, ln->ln_state);
566 * ND6 timer routine to expire default route list and prefix list
571 CURVNET_SET((struct vnet *) arg);
573 struct nd_defrouter *dr;
574 struct nd_prefix *pr;
575 struct in6_ifaddr *ia6, *nia6;
576 struct in6_addrlifetime *lt6;
578 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
581 /* expire default router list */
583 dr = TAILQ_FIRST(&V_nd_defrouter);
585 if (dr->expire && dr->expire < time_second) {
586 struct nd_defrouter *t;
587 t = TAILQ_NEXT(dr, dr_entry);
591 dr = TAILQ_NEXT(dr, dr_entry);
596 * expire interface addresses.
597 * in the past the loop was inside prefix expiry processing.
598 * However, from a stricter speci-confrmance standpoint, we should
599 * rather separate address lifetimes and prefix lifetimes.
601 * XXXRW: in6_ifaddrhead locking.
604 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
605 /* check address lifetime */
606 lt6 = &ia6->ia6_lifetime;
607 if (IFA6_IS_INVALID(ia6)) {
611 * If the expiring address is temporary, try
612 * regenerating a new one. This would be useful when
613 * we suspended a laptop PC, then turned it on after a
614 * period that could invalidate all temporary
615 * addresses. Although we may have to restart the
616 * loop (see below), it must be after purging the
617 * address. Otherwise, we'd see an infinite loop of
620 if (V_ip6_use_tempaddr &&
621 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
622 if (regen_tmpaddr(ia6) == 0)
626 in6_purgeaddr(&ia6->ia_ifa);
629 goto addrloop; /* XXX: see below */
630 } else if (IFA6_IS_DEPRECATED(ia6)) {
631 int oldflags = ia6->ia6_flags;
633 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
636 * If a temporary address has just become deprecated,
637 * regenerate a new one if possible.
639 if (V_ip6_use_tempaddr &&
640 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
641 (oldflags & IN6_IFF_DEPRECATED) == 0) {
643 if (regen_tmpaddr(ia6) == 0) {
645 * A new temporary address is
647 * XXX: this means the address chain
648 * has changed while we are still in
649 * the loop. Although the change
650 * would not cause disaster (because
651 * it's not a deletion, but an
652 * addition,) we'd rather restart the
653 * loop just for safety. Or does this
654 * significantly reduce performance??
661 * A new RA might have made a deprecated address
664 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
668 /* expire prefix list */
669 pr = V_nd_prefix.lh_first;
672 * check prefix lifetime.
673 * since pltime is just for autoconf, pltime processing for
674 * prefix is not necessary.
676 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
677 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
682 * address expiration and prefix expiration are
683 * separate. NEVER perform in6_purgeaddr here.
696 * ia6 - deprecated/invalidated temporary address
699 regen_tmpaddr(struct in6_ifaddr *ia6)
703 struct in6_ifaddr *public_ifa6 = NULL;
705 ifp = ia6->ia_ifa.ifa_ifp;
707 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
708 struct in6_ifaddr *it6;
710 if (ifa->ifa_addr->sa_family != AF_INET6)
713 it6 = (struct in6_ifaddr *)ifa;
715 /* ignore no autoconf addresses. */
716 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
719 /* ignore autoconf addresses with different prefixes. */
720 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
724 * Now we are looking at an autoconf address with the same
725 * prefix as ours. If the address is temporary and is still
726 * preferred, do not create another one. It would be rare, but
727 * could happen, for example, when we resume a laptop PC after
730 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
731 !IFA6_IS_DEPRECATED(it6)) {
737 * This is a public autoconf address that has the same prefix
738 * as ours. If it is preferred, keep it. We can't break the
739 * loop here, because there may be a still-preferred temporary
740 * address with the prefix.
742 if (!IFA6_IS_DEPRECATED(it6))
745 if (public_ifa6 != NULL)
746 ifa_ref(&public_ifa6->ia_ifa);
750 if (public_ifa6 != NULL) {
753 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
754 ifa_free(&public_ifa6->ia_ifa);
755 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
756 " tmp addr,errno=%d\n", e);
759 ifa_free(&public_ifa6->ia_ifa);
767 * Nuke neighbor cache/prefix/default router management table, right before
771 nd6_purge(struct ifnet *ifp)
773 struct nd_defrouter *dr, *ndr;
774 struct nd_prefix *pr, *npr;
777 * Nuke default router list entries toward ifp.
778 * We defer removal of default router list entries that is installed
779 * in the routing table, in order to keep additional side effects as
782 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
783 ndr = TAILQ_NEXT(dr, dr_entry);
791 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
792 ndr = TAILQ_NEXT(dr, dr_entry);
800 /* Nuke prefix list entries toward ifp */
801 for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
803 if (pr->ndpr_ifp == ifp) {
805 * Because if_detach() does *not* release prefixes
806 * while purging addresses the reference count will
807 * still be above zero. We therefore reset it to
808 * make sure that the prefix really gets purged.
813 * Previously, pr->ndpr_addr is removed as well,
814 * but I strongly believe we don't have to do it.
815 * nd6_purge() is only called from in6_ifdetach(),
816 * which removes all the associated interface addresses
818 * (jinmei@kame.net 20010129)
824 /* cancel default outgoing interface setting */
825 if (V_nd6_defifindex == ifp->if_index)
826 nd6_setdefaultiface(0);
828 if (!V_ip6_forwarding && ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
829 /* Refresh default router list. */
834 * We do not nuke the neighbor cache entries here any more
835 * because the neighbor cache is kept in if_afdata[AF_INET6].
836 * nd6_purge() is invoked by in6_ifdetach() which is called
837 * from if_detach() where everything gets purged. So let
838 * in6_domifdetach() do the actual L2 table purging work.
843 * the caller acquires and releases the lock on the lltbls
844 * Returns the llentry locked
847 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
849 struct sockaddr_in6 sin6;
853 bzero(&sin6, sizeof(sin6));
854 sin6.sin6_len = sizeof(struct sockaddr_in6);
855 sin6.sin6_family = AF_INET6;
856 sin6.sin6_addr = *addr6;
858 IF_AFDATA_LOCK_ASSERT(ifp);
861 if (flags & ND6_CREATE)
862 llflags |= LLE_CREATE;
863 if (flags & ND6_EXCLUSIVE)
864 llflags |= LLE_EXCLUSIVE;
866 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
867 if ((ln != NULL) && (llflags & LLE_CREATE))
868 ln->ln_state = ND6_LLINFO_NOSTATE;
874 * Test whether a given IPv6 address is a neighbor or not, ignoring
875 * the actual neighbor cache. The neighbor cache is ignored in order
876 * to not reenter the routing code from within itself.
879 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
881 struct nd_prefix *pr;
882 struct ifaddr *dstaddr;
885 * A link-local address is always a neighbor.
886 * XXX: a link does not necessarily specify a single interface.
888 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
889 struct sockaddr_in6 sin6_copy;
893 * We need sin6_copy since sa6_recoverscope() may modify the
897 if (sa6_recoverscope(&sin6_copy))
898 return (0); /* XXX: should be impossible */
899 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
901 if (sin6_copy.sin6_scope_id == zone)
908 * If the address matches one of our addresses,
909 * it should be a neighbor.
910 * If the address matches one of our on-link prefixes, it should be a
913 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
914 if (pr->ndpr_ifp != ifp)
917 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
919 rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
923 * This is the case where multiple interfaces
924 * have the same prefix, but only one is installed
925 * into the routing table and that prefix entry
926 * is not the one being examined here. In the case
927 * where RADIX_MPATH is enabled, multiple route
928 * entries (of the same rt_key value) will be
929 * installed because the interface addresses all
932 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
933 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
940 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
941 &addr->sin6_addr, &pr->ndpr_mask))
946 * If the address is assigned on the node of the other side of
947 * a p2p interface, the address should be a neighbor.
949 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
950 if (dstaddr != NULL) {
951 if (dstaddr->ifa_ifp == ifp) {
959 * If the default router list is empty, all addresses are regarded
960 * as on-link, and thus, as a neighbor.
961 * XXX: we restrict the condition to hosts, because routers usually do
962 * not have the "default router list".
964 if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
965 V_nd6_defifindex == ifp->if_index) {
974 * Detect if a given IPv6 address identifies a neighbor on a given link.
975 * XXX: should take care of the destination of a p2p link?
978 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
983 IF_AFDATA_UNLOCK_ASSERT(ifp);
984 if (nd6_is_new_addr_neighbor(addr, ifp))
988 * Even if the address matches none of our addresses, it might be
989 * in the neighbor cache.
992 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
996 IF_AFDATA_UNLOCK(ifp);
1001 * Free an nd6 llinfo entry.
1002 * Since the function would cause significant changes in the kernel, DO NOT
1003 * make it global, unless you have a strong reason for the change, and are sure
1004 * that the change is safe.
1006 static struct llentry *
1007 nd6_free(struct llentry *ln, int gc)
1009 struct llentry *next;
1010 struct nd_defrouter *dr;
1013 LLE_WLOCK_ASSERT(ln);
1016 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1017 * even though it is not harmful, it was not really necessary.
1021 nd6_llinfo_settimer_locked(ln, -1);
1023 ifp = ln->lle_tbl->llt_ifp;
1025 if (!V_ip6_forwarding) {
1027 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1029 if (dr != NULL && dr->expire &&
1030 ln->ln_state == ND6_LLINFO_STALE && gc) {
1032 * If the reason for the deletion is just garbage
1033 * collection, and the neighbor is an active default
1034 * router, do not delete it. Instead, reset the GC
1035 * timer using the router's lifetime.
1036 * Simply deleting the entry would affect default
1037 * router selection, which is not necessarily a good
1038 * thing, especially when we're using router preference
1040 * XXX: the check for ln_state would be redundant,
1041 * but we intentionally keep it just in case.
1043 if (dr->expire > time_second)
1044 nd6_llinfo_settimer_locked(ln,
1045 (dr->expire - time_second) * hz);
1047 nd6_llinfo_settimer_locked(ln,
1048 (long)V_nd6_gctimer * hz);
1050 next = LIST_NEXT(ln, lle_next);
1056 if (ln->ln_router || dr) {
1058 * rt6_flush must be called whether or not the neighbor
1059 * is in the Default Router List.
1060 * See a corresponding comment in nd6_na_input().
1062 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1067 * Unreachablity of a router might affect the default
1068 * router selection and on-link detection of advertised
1073 * Temporarily fake the state to choose a new default
1074 * router and to perform on-link determination of
1075 * prefixes correctly.
1076 * Below the state will be set correctly,
1077 * or the entry itself will be deleted.
1079 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1082 * Since defrouter_select() does not affect the
1083 * on-link determination and MIP6 needs the check
1084 * before the default router selection, we perform
1087 pfxlist_onlink_check();
1090 * Refresh default router list. Have to unlock as
1091 * it calls into nd6_lookup(), still holding a ref.
1100 * Before deleting the entry, remember the next entry as the
1101 * return value. We need this because pfxlist_onlink_check() above
1102 * might have freed other entries (particularly the old next entry) as
1103 * a side effect (XXX).
1105 next = LIST_NEXT(ln, lle_next);
1108 * Save to unlock. We still hold an extra reference and will not
1109 * free(9) in llentry_free() if someone else holds one as well.
1112 IF_AFDATA_LOCK(ifp);
1116 IF_AFDATA_UNLOCK(ifp);
1122 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1124 * XXX cost-effective methods?
1127 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1132 if ((dst6 == NULL) || (rt == NULL))
1136 IF_AFDATA_LOCK(ifp);
1137 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1138 IF_AFDATA_UNLOCK(ifp);
1142 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1146 * if we get upper-layer reachability confirmation many times,
1147 * it is possible we have false information.
1151 if (ln->ln_byhint > V_nd6_maxnudhint) {
1156 ln->ln_state = ND6_LLINFO_REACHABLE;
1157 if (!ND6_LLINFO_PERMANENT(ln)) {
1158 nd6_llinfo_settimer_locked(ln,
1159 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1167 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1169 struct in6_drlist *drl = (struct in6_drlist *)data;
1170 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1171 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1172 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1173 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1174 struct nd_defrouter *dr;
1175 struct nd_prefix *pr;
1176 int i = 0, error = 0;
1180 case SIOCGDRLST_IN6:
1182 * obsolete API, use sysctl under net.inet6.icmp6
1184 bzero(drl, sizeof(*drl));
1186 dr = TAILQ_FIRST(&V_nd_defrouter);
1187 while (dr && i < DRLSTSIZ) {
1188 drl->defrouter[i].rtaddr = dr->rtaddr;
1189 in6_clearscope(&drl->defrouter[i].rtaddr);
1191 drl->defrouter[i].flags = dr->flags;
1192 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1193 drl->defrouter[i].expire = dr->expire;
1194 drl->defrouter[i].if_index = dr->ifp->if_index;
1196 dr = TAILQ_NEXT(dr, dr_entry);
1200 case SIOCGPRLST_IN6:
1202 * obsolete API, use sysctl under net.inet6.icmp6
1204 * XXX the structure in6_prlist was changed in backward-
1205 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1206 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1209 * XXX meaning of fields, especialy "raflags", is very
1210 * differnet between RA prefix list and RR/static prefix list.
1211 * how about separating ioctls into two?
1213 bzero(oprl, sizeof(*oprl));
1215 pr = V_nd_prefix.lh_first;
1216 while (pr && i < PRLSTSIZ) {
1217 struct nd_pfxrouter *pfr;
1220 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1221 oprl->prefix[i].raflags = pr->ndpr_raf;
1222 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1223 oprl->prefix[i].vltime = pr->ndpr_vltime;
1224 oprl->prefix[i].pltime = pr->ndpr_pltime;
1225 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1226 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1227 oprl->prefix[i].expire = 0;
1231 /* XXX: we assume time_t is signed. */
1234 ((sizeof(maxexpire) * 8) - 1));
1235 if (pr->ndpr_vltime <
1236 maxexpire - pr->ndpr_lastupdate) {
1237 oprl->prefix[i].expire =
1238 pr->ndpr_lastupdate +
1241 oprl->prefix[i].expire = maxexpire;
1244 pfr = pr->ndpr_advrtrs.lh_first;
1248 #define RTRADDR oprl->prefix[i].advrtr[j]
1249 RTRADDR = pfr->router->rtaddr;
1250 in6_clearscope(&RTRADDR);
1254 pfr = pfr->pfr_next;
1256 oprl->prefix[i].advrtrs = j;
1257 oprl->prefix[i].origin = PR_ORIG_RA;
1265 case OSIOCGIFINFO_IN6:
1267 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1268 bzero(&ND, sizeof(ND));
1269 ND.linkmtu = IN6_LINKMTU(ifp);
1270 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1271 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1272 ND.reachable = ND_IFINFO(ifp)->reachable;
1273 ND.retrans = ND_IFINFO(ifp)->retrans;
1274 ND.flags = ND_IFINFO(ifp)->flags;
1275 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1276 ND.chlim = ND_IFINFO(ifp)->chlim;
1278 case SIOCGIFINFO_IN6:
1279 ND = *ND_IFINFO(ifp);
1281 case SIOCSIFINFO_IN6:
1283 * used to change host variables from userland.
1284 * intented for a use on router to reflect RA configurations.
1286 /* 0 means 'unspecified' */
1287 if (ND.linkmtu != 0) {
1288 if (ND.linkmtu < IPV6_MMTU ||
1289 ND.linkmtu > IN6_LINKMTU(ifp)) {
1293 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1296 if (ND.basereachable != 0) {
1297 int obasereachable = ND_IFINFO(ifp)->basereachable;
1299 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1300 if (ND.basereachable != obasereachable)
1301 ND_IFINFO(ifp)->reachable =
1302 ND_COMPUTE_RTIME(ND.basereachable);
1304 if (ND.retrans != 0)
1305 ND_IFINFO(ifp)->retrans = ND.retrans;
1307 ND_IFINFO(ifp)->chlim = ND.chlim;
1309 case SIOCSIFINFO_FLAGS:
1312 struct in6_ifaddr *ia;
1314 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1315 !(ND.flags & ND6_IFF_IFDISABLED)) {
1316 /* ifdisabled 1->0 transision */
1319 * If the interface is marked as ND6_IFF_IFDISABLED and
1320 * has an link-local address with IN6_IFF_DUPLICATED,
1321 * do not clear ND6_IFF_IFDISABLED.
1322 * See RFC 4862, Section 5.4.5.
1324 int duplicated_linklocal = 0;
1327 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1328 if (ifa->ifa_addr->sa_family != AF_INET6)
1330 ia = (struct in6_ifaddr *)ifa;
1331 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1332 IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1333 duplicated_linklocal = 1;
1337 IF_ADDR_UNLOCK(ifp);
1339 if (duplicated_linklocal) {
1340 ND.flags |= ND6_IFF_IFDISABLED;
1341 log(LOG_ERR, "Cannot enable an interface"
1342 " with a link-local address marked"
1345 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1348 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1349 (ND.flags & ND6_IFF_IFDISABLED)) {
1350 /* ifdisabled 0->1 transision */
1351 /* Mark all IPv6 address as tentative. */
1353 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1355 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1356 if (ifa->ifa_addr->sa_family != AF_INET6)
1358 ia = (struct in6_ifaddr *)ifa;
1359 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1361 IF_ADDR_UNLOCK(ifp);
1364 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) &&
1365 (ND.flags & ND6_IFF_AUTO_LINKLOCAL)) {
1366 /* auto_linklocal 0->1 transision */
1368 /* If no link-local address on ifp, configure */
1369 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1370 in6_ifattach(ifp, NULL);
1373 ND_IFINFO(ifp)->flags = ND.flags;
1376 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1377 /* sync kernel routing table with the default router list */
1381 case SIOCSPFXFLUSH_IN6:
1383 /* flush all the prefix advertised by routers */
1384 struct nd_prefix *pr, *next;
1387 for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1388 struct in6_ifaddr *ia, *ia_next;
1390 next = pr->ndpr_next;
1392 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1395 /* do we really have to remove addresses as well? */
1396 /* XXXRW: in6_ifaddrhead locking. */
1397 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1399 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1402 if (ia->ia6_ndpr == pr)
1403 in6_purgeaddr(&ia->ia_ifa);
1410 case SIOCSRTRFLUSH_IN6:
1412 /* flush all the default routers */
1413 struct nd_defrouter *dr, *next;
1417 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1418 next = TAILQ_NEXT(dr, dr_entry);
1425 case SIOCGNBRINFO_IN6:
1428 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1430 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1433 IF_AFDATA_LOCK(ifp);
1434 ln = nd6_lookup(&nb_addr, 0, ifp);
1435 IF_AFDATA_UNLOCK(ifp);
1441 nbi->state = ln->ln_state;
1442 nbi->asked = ln->la_asked;
1443 nbi->isrouter = ln->ln_router;
1444 nbi->expire = ln->la_expire;
1448 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1449 ndif->ifindex = V_nd6_defifindex;
1451 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1452 return (nd6_setdefaultiface(ndif->ifindex));
1458 * Create neighbor cache entry and cache link-layer address,
1459 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1462 * code - type dependent information
1465 * The caller of this function already acquired the ndp
1466 * cache table lock because the cache entry is returned.
1469 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1470 int lladdrlen, int type, int code)
1472 struct llentry *ln = NULL;
1479 uint16_t router = 0;
1480 struct sockaddr_in6 sin6;
1481 struct mbuf *chain = NULL;
1482 int static_route = 0;
1484 IF_AFDATA_UNLOCK_ASSERT(ifp);
1487 panic("ifp == NULL in nd6_cache_lladdr");
1489 panic("from == NULL in nd6_cache_lladdr");
1491 /* nothing must be updated for unspecified address */
1492 if (IN6_IS_ADDR_UNSPECIFIED(from))
1496 * Validation about ifp->if_addrlen and lladdrlen must be done in
1499 * XXX If the link does not have link-layer adderss, what should
1500 * we do? (ifp->if_addrlen == 0)
1501 * Spec says nothing in sections for RA, RS and NA. There's small
1502 * description on it in NS section (RFC 2461 7.2.3).
1504 flags = lladdr ? ND6_EXCLUSIVE : 0;
1505 IF_AFDATA_LOCK(ifp);
1506 ln = nd6_lookup(from, flags, ifp);
1509 flags |= ND6_EXCLUSIVE;
1510 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1511 IF_AFDATA_UNLOCK(ifp);
1514 IF_AFDATA_UNLOCK(ifp);
1515 /* do nothing if static ndp is set */
1516 if (ln->la_flags & LLE_STATIC) {
1525 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1526 if (olladdr && lladdr) {
1527 llchange = bcmp(lladdr, &ln->ll_addr,
1533 * newentry olladdr lladdr llchange (*=record)
1536 * 0 n y -- (3) * STALE
1538 * 0 y y y (5) * STALE
1539 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1540 * 1 -- y -- (7) * STALE
1543 if (lladdr) { /* (3-5) and (7) */
1545 * Record source link-layer address
1546 * XXX is it dependent to ifp->if_type?
1548 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1549 ln->la_flags |= LLE_VALID;
1553 if ((!olladdr && lladdr != NULL) || /* (3) */
1554 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1556 newstate = ND6_LLINFO_STALE;
1557 } else /* (1-2,4) */
1561 if (lladdr == NULL) /* (6) */
1562 newstate = ND6_LLINFO_NOSTATE;
1564 newstate = ND6_LLINFO_STALE;
1569 * Update the state of the neighbor cache.
1571 ln->ln_state = newstate;
1573 if (ln->ln_state == ND6_LLINFO_STALE) {
1575 * XXX: since nd6_output() below will cause
1576 * state tansition to DELAY and reset the timer,
1577 * we must set the timer now, although it is actually
1580 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1583 struct mbuf *m_hold, *m_hold_next;
1586 * reset the la_hold in advance, to explicitly
1587 * prevent a la_hold lookup in nd6_output()
1588 * (wouldn't happen, though...)
1590 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1591 m_hold; m_hold = m_hold_next) {
1592 m_hold_next = m_hold->m_nextpkt;
1593 m_hold->m_nextpkt = NULL;
1596 * we assume ifp is not a p2p here, so
1597 * just set the 2nd argument as the
1600 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1603 * If we have mbufs in the chain we need to do
1604 * deferred transmit. Copy the address from the
1605 * llentry before dropping the lock down below.
1608 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1610 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1611 /* probe right away */
1612 nd6_llinfo_settimer_locked((void *)ln, 0);
1617 * ICMP6 type dependent behavior.
1619 * NS: clear IsRouter if new entry
1620 * RS: clear IsRouter
1621 * RA: set IsRouter if there's lladdr
1622 * redir: clear IsRouter if new entry
1625 * The spec says that we must set IsRouter in the following cases:
1626 * - If lladdr exist, set IsRouter. This means (1-5).
1627 * - If it is old entry (!newentry), set IsRouter. This means (7).
1628 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1629 * A quetion arises for (1) case. (1) case has no lladdr in the
1630 * neighbor cache, this is similar to (6).
1631 * This case is rare but we figured that we MUST NOT set IsRouter.
1633 * newentry olladdr lladdr llchange NS RS RA redir
1635 * 0 n n -- (1) c ? s
1636 * 0 y n -- (2) c s s
1637 * 0 n y -- (3) c s s
1640 * 1 -- n -- (6) c c c s
1641 * 1 -- y -- (7) c c s c s
1645 switch (type & 0xff) {
1646 case ND_NEIGHBOR_SOLICIT:
1648 * New entry must have is_router flag cleared.
1650 if (is_newentry) /* (6-7) */
1655 * If the icmp is a redirect to a better router, always set the
1656 * is_router flag. Otherwise, if the entry is newly created,
1657 * clear the flag. [RFC 2461, sec 8.3]
1659 if (code == ND_REDIRECT_ROUTER)
1661 else if (is_newentry) /* (6-7) */
1664 case ND_ROUTER_SOLICIT:
1666 * is_router flag must always be cleared.
1670 case ND_ROUTER_ADVERT:
1672 * Mark an entry with lladdr as a router.
1674 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1675 (is_newentry && lladdr)) { /* (7) */
1682 static_route = (ln->la_flags & LLE_STATIC);
1683 router = ln->ln_router;
1685 if (flags & ND6_EXCLUSIVE)
1693 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1696 * When the link-layer address of a router changes, select the
1697 * best router again. In particular, when the neighbor entry is newly
1698 * created, it might affect the selection policy.
1699 * Question: can we restrict the first condition to the "is_newentry"
1701 * XXX: when we hear an RA from a new router with the link-layer
1702 * address option, defrouter_select() is called twice, since
1703 * defrtrlist_update called the function as well. However, I believe
1704 * we can compromise the overhead, since it only happens the first
1706 * XXX: although defrouter_select() should not have a bad effect
1707 * for those are not autoconfigured hosts, we explicitly avoid such
1710 if (do_update && router && !V_ip6_forwarding &&
1711 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1713 * guaranteed recursion
1721 if (flags & ND6_EXCLUSIVE)
1732 nd6_slowtimo(void *arg)
1734 CURVNET_SET((struct vnet *) arg);
1735 struct nd_ifinfo *nd6if;
1738 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1739 nd6_slowtimo, curvnet);
1740 IFNET_RLOCK_NOSLEEP();
1741 for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1742 ifp = TAILQ_NEXT(ifp, if_list)) {
1743 nd6if = ND_IFINFO(ifp);
1744 if (nd6if->basereachable && /* already initialized */
1745 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1747 * Since reachable time rarely changes by router
1748 * advertisements, we SHOULD insure that a new random
1749 * value gets recomputed at least once every few hours.
1752 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1753 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1756 IFNET_RUNLOCK_NOSLEEP();
1761 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1762 struct sockaddr_in6 *dst, struct rtentry *rt0)
1765 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1770 * Note that I'm not enforcing any global serialization
1771 * lle state or asked changes here as the logic is too
1772 * complicated to avoid having to always acquire an exclusive
1777 #define senderr(e) { error = (e); goto bad;}
1780 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1781 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1782 struct mbuf **chain)
1784 struct mbuf *m = m0;
1786 struct llentry *ln = lle;
1787 struct ip6_hdr *ip6;
1795 LLE_WLOCK_ASSERT(lle);
1797 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1800 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1803 if (nd6_need_cache(ifp) == 0)
1807 * next hop determination. This routine is derived from ether_output.
1811 * Address resolution or Neighbor Unreachability Detection
1813 * At this point, the destination of the packet must be a unicast
1814 * or an anycast address(i.e. not a multicast).
1817 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1820 IF_AFDATA_LOCK(ifp);
1821 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1822 IF_AFDATA_UNLOCK(ifp);
1823 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1825 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1826 * the condition below is not very efficient. But we believe
1827 * it is tolerable, because this should be a rare case.
1829 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1830 IF_AFDATA_LOCK(ifp);
1831 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1832 IF_AFDATA_UNLOCK(ifp);
1836 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1837 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1838 char ip6buf[INET6_ADDRSTRLEN];
1840 "nd6_output: can't allocate llinfo for %s "
1842 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1843 senderr(EIO); /* XXX: good error? */
1845 goto sendpkt; /* send anyway */
1848 /* We don't have to do link-layer address resolution on a p2p link. */
1849 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1850 ln->ln_state < ND6_LLINFO_REACHABLE) {
1851 if ((flags & LLE_EXCLUSIVE) == 0) {
1852 flags |= LLE_EXCLUSIVE;
1855 ln->ln_state = ND6_LLINFO_STALE;
1856 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1860 * The first time we send a packet to a neighbor whose entry is
1861 * STALE, we have to change the state to DELAY and a sets a timer to
1862 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1863 * neighbor unreachability detection on expiration.
1866 if (ln->ln_state == ND6_LLINFO_STALE) {
1867 if ((flags & LLE_EXCLUSIVE) == 0) {
1868 flags |= LLE_EXCLUSIVE;
1873 ln->ln_state = ND6_LLINFO_DELAY;
1874 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1878 * If the neighbor cache entry has a state other than INCOMPLETE
1879 * (i.e. its link-layer address is already resolved), just
1882 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1886 * There is a neighbor cache entry, but no ethernet address
1887 * response yet. Append this latest packet to the end of the
1888 * packet queue in the mbuf, unless the number of the packet
1889 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1890 * the oldest packet in the queue will be removed.
1892 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1893 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1895 if ((flags & LLE_EXCLUSIVE) == 0) {
1896 flags |= LLE_EXCLUSIVE;
1901 LLE_WLOCK_ASSERT(ln);
1904 struct mbuf *m_hold;
1908 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1910 if (m_hold->m_nextpkt == NULL) {
1911 m_hold->m_nextpkt = m;
1915 while (i >= V_nd6_maxqueuelen) {
1916 m_hold = ln->la_hold;
1917 ln->la_hold = ln->la_hold->m_nextpkt;
1926 * If there has been no NS for the neighbor after entering the
1927 * INCOMPLETE state, send the first solicitation.
1929 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1932 nd6_llinfo_settimer_locked(ln,
1933 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1935 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1936 if (lle != NULL && ln == lle)
1939 } else if (lle == NULL || ln != lle) {
1941 * We did the lookup (no lle arg) so we
1942 * need to do the unlock here.
1950 /* discard the packet if IPv6 operation is disabled on the interface */
1951 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1952 error = ENETDOWN; /* better error? */
1956 * ln is valid and the caller did not pass in
1959 if ((ln != NULL) && (lle == NULL)) {
1960 if (flags & LLE_EXCLUSIVE)
1967 mac_netinet6_nd6_send(ifp, m);
1971 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1972 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1973 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1974 * to be diverted to user space. When re-injected into the kernel,
1975 * send_output() will directly dispatch them to the outgoing interface.
1977 if (send_sendso_input_hook != NULL) {
1978 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1980 ip6 = mtod(m, struct ip6_hdr *);
1981 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1982 /* Use the SEND socket */
1983 error = send_sendso_input_hook(m, ifp, SND_OUT,
1985 /* -1 == no app on SEND socket */
1986 if (error == 0 || error != -1)
1992 * We were passed in a pointer to an lle with the lock held
1993 * this means that we can't call if_output as we will
1994 * recurse on the lle lock - so what we do is we create
1995 * a list of mbufs to send and transmit them in the caller
1996 * after the lock is dropped
2002 struct mbuf *m = *chain;
2005 * append mbuf to end of deferred chain
2007 while (m->m_nextpkt != NULL)
2013 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2014 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2017 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2022 * ln is valid and the caller did not pass in
2025 if ((ln != NULL) && (lle == NULL)) {
2026 if (flags & LLE_EXCLUSIVE)
2039 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2040 struct sockaddr_in6 *dst, struct route *ro)
2042 struct mbuf *m, *m_head;
2043 struct ifnet *outifp;
2047 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2054 m_head = m_head->m_nextpkt;
2055 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2060 * note that intermediate errors are blindly ignored - but this is
2061 * the same convention as used with nd6_output when called by
2069 nd6_need_cache(struct ifnet *ifp)
2072 * XXX: we currently do not make neighbor cache on any interface
2073 * other than ARCnet, Ethernet, FDDI and GIF.
2076 * - unidirectional tunnels needs no ND
2078 switch (ifp->if_type) {
2086 #ifdef IFT_IEEE80211
2092 case IFT_GIF: /* XXX need more cases? */
2096 case IFT_PROPVIRTUAL:
2104 * the callers of this function need to be re-worked to drop
2105 * the lle lock, drop here for now
2108 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2109 struct sockaddr *dst, u_char *desten, struct llentry **lle)
2114 IF_AFDATA_UNLOCK_ASSERT(ifp);
2115 if (m->m_flags & M_MCAST) {
2118 switch (ifp->if_type) {
2124 #ifdef IFT_IEEE80211
2129 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2134 * netbsd can use if_broadcastaddr, but we don't do so
2135 * to reduce # of ifdef.
2137 for (i = 0; i < ifp->if_addrlen; i++)
2145 return (EAFNOSUPPORT);
2151 * the entry should have been created in nd6_store_lladdr
2153 IF_AFDATA_LOCK(ifp);
2154 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2155 IF_AFDATA_UNLOCK(ifp);
2156 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2159 /* this could happen, if we could not allocate memory */
2164 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2168 * A *small* use after free race exists here
2174 clear_llinfo_pqueue(struct llentry *ln)
2176 struct mbuf *m_hold, *m_hold_next;
2178 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2179 m_hold_next = m_hold->m_nextpkt;
2180 m_hold->m_nextpkt = NULL;
2188 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2189 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2191 SYSCTL_DECL(_net_inet6_icmp6);
2193 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2194 CTLFLAG_RD, nd6_sysctl_drlist, "");
2195 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2196 CTLFLAG_RD, nd6_sysctl_prlist, "");
2197 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2198 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2201 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2204 char buf[1024] __aligned(4);
2205 struct in6_defrouter *d, *de;
2206 struct nd_defrouter *dr;
2212 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2213 dr = TAILQ_NEXT(dr, dr_entry)) {
2214 d = (struct in6_defrouter *)buf;
2215 de = (struct in6_defrouter *)(buf + sizeof(buf));
2218 bzero(d, sizeof(*d));
2219 d->rtaddr.sin6_family = AF_INET6;
2220 d->rtaddr.sin6_len = sizeof(d->rtaddr);
2221 d->rtaddr.sin6_addr = dr->rtaddr;
2222 error = sa6_recoverscope(&d->rtaddr);
2225 d->flags = dr->flags;
2226 d->rtlifetime = dr->rtlifetime;
2227 d->expire = dr->expire;
2228 d->if_index = dr->ifp->if_index;
2230 panic("buffer too short");
2232 error = SYSCTL_OUT(req, buf, sizeof(*d));
2241 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2244 char buf[1024] __aligned(4);
2245 struct in6_prefix *p, *pe;
2246 struct nd_prefix *pr;
2247 char ip6buf[INET6_ADDRSTRLEN];
2253 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2256 struct sockaddr_in6 *sin6, *s6;
2257 struct nd_pfxrouter *pfr;
2259 p = (struct in6_prefix *)buf;
2260 pe = (struct in6_prefix *)(buf + sizeof(buf));
2263 bzero(p, sizeof(*p));
2264 sin6 = (struct sockaddr_in6 *)(p + 1);
2266 p->prefix = pr->ndpr_prefix;
2267 if (sa6_recoverscope(&p->prefix)) {
2269 "scope error in prefix list (%s)\n",
2270 ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2271 /* XXX: press on... */
2273 p->raflags = pr->ndpr_raf;
2274 p->prefixlen = pr->ndpr_plen;
2275 p->vltime = pr->ndpr_vltime;
2276 p->pltime = pr->ndpr_pltime;
2277 p->if_index = pr->ndpr_ifp->if_index;
2278 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2283 /* XXX: we assume time_t is signed. */
2286 ((sizeof(maxexpire) * 8) - 1));
2287 if (pr->ndpr_vltime <
2288 maxexpire - pr->ndpr_lastupdate) {
2289 p->expire = pr->ndpr_lastupdate +
2292 p->expire = maxexpire;
2294 p->refcnt = pr->ndpr_refcnt;
2295 p->flags = pr->ndpr_stateflags;
2296 p->origin = PR_ORIG_RA;
2298 for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2299 pfr = pfr->pfr_next) {
2300 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2304 s6 = &sin6[advrtrs];
2305 bzero(s6, sizeof(*s6));
2306 s6->sin6_family = AF_INET6;
2307 s6->sin6_len = sizeof(*sin6);
2308 s6->sin6_addr = pfr->router->rtaddr;
2309 if (sa6_recoverscope(s6)) {
2312 "prefix list (%s)\n",
2314 &pfr->router->rtaddr));
2318 p->advrtrs = advrtrs;
2320 panic("buffer too short");
2322 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2323 error = SYSCTL_OUT(req, buf, advance);