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;
196 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
197 nd->flags |= ND6_IFF_NO_RADR;
199 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
200 nd6_setmtu0(ifp, nd);
206 nd6_ifdetach(struct nd_ifinfo *nd)
213 * Reset ND level link MTU. This function is called when the physical MTU
214 * changes, which means we might have to adjust the ND level MTU.
217 nd6_setmtu(struct ifnet *ifp)
220 nd6_setmtu0(ifp, ND_IFINFO(ifp));
223 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
225 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
229 omaxmtu = ndi->maxmtu;
231 switch (ifp->if_type) {
233 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
236 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
239 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
242 ndi->maxmtu = ifp->if_mtu;
247 * Decreasing the interface MTU under IPV6 minimum MTU may cause
248 * undesirable situation. We thus notify the operator of the change
249 * explicitly. The check for omaxmtu is necessary to restrict the
250 * log to the case of changing the MTU, not initializing it.
252 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
253 log(LOG_NOTICE, "nd6_setmtu0: "
254 "new link MTU on %s (%lu) is too small for IPv6\n",
255 if_name(ifp), (unsigned long)ndi->maxmtu);
258 if (ndi->maxmtu > V_in6_maxmtu)
259 in6_setmaxmtu(); /* check all interfaces just in case */
264 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
267 bzero(ndopts, sizeof(*ndopts));
268 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
270 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
273 ndopts->nd_opts_done = 1;
274 ndopts->nd_opts_search = NULL;
279 * Take one ND option.
282 nd6_option(union nd_opts *ndopts)
284 struct nd_opt_hdr *nd_opt;
288 panic("ndopts == NULL in nd6_option");
289 if (ndopts->nd_opts_last == NULL)
290 panic("uninitialized ndopts in nd6_option");
291 if (ndopts->nd_opts_search == NULL)
293 if (ndopts->nd_opts_done)
296 nd_opt = ndopts->nd_opts_search;
298 /* make sure nd_opt_len is inside the buffer */
299 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
300 bzero(ndopts, sizeof(*ndopts));
304 olen = nd_opt->nd_opt_len << 3;
307 * Message validation requires that all included
308 * options have a length that is greater than zero.
310 bzero(ndopts, sizeof(*ndopts));
314 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
315 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
316 /* option overruns the end of buffer, invalid */
317 bzero(ndopts, sizeof(*ndopts));
319 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
320 /* reached the end of options chain */
321 ndopts->nd_opts_done = 1;
322 ndopts->nd_opts_search = NULL;
328 * Parse multiple ND options.
329 * This function is much easier to use, for ND routines that do not need
330 * multiple options of the same type.
333 nd6_options(union nd_opts *ndopts)
335 struct nd_opt_hdr *nd_opt;
339 panic("ndopts == NULL in nd6_options");
340 if (ndopts->nd_opts_last == NULL)
341 panic("uninitialized ndopts in nd6_options");
342 if (ndopts->nd_opts_search == NULL)
346 nd_opt = nd6_option(ndopts);
347 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
349 * Message validation requires that all included
350 * options have a length that is greater than zero.
352 ICMP6STAT_INC(icp6s_nd_badopt);
353 bzero(ndopts, sizeof(*ndopts));
360 switch (nd_opt->nd_opt_type) {
361 case ND_OPT_SOURCE_LINKADDR:
362 case ND_OPT_TARGET_LINKADDR:
364 case ND_OPT_REDIRECTED_HEADER:
365 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
367 "duplicated ND6 option found (type=%d)\n",
368 nd_opt->nd_opt_type));
371 ndopts->nd_opt_array[nd_opt->nd_opt_type]
375 case ND_OPT_PREFIX_INFORMATION:
376 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
377 ndopts->nd_opt_array[nd_opt->nd_opt_type]
380 ndopts->nd_opts_pi_end =
381 (struct nd_opt_prefix_info *)nd_opt;
383 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
384 case ND_OPT_RDNSS: /* RFC 6106 */
385 case ND_OPT_DNSSL: /* RFC 6106 */
387 * Silently ignore options we know and do not care about
393 * Unknown options must be silently ignored,
394 * to accomodate future extension to the protocol.
397 "nd6_options: unsupported option %d - "
398 "option ignored\n", nd_opt->nd_opt_type));
403 if (i > V_nd6_maxndopt) {
404 ICMP6STAT_INC(icp6s_nd_toomanyopt);
405 nd6log((LOG_INFO, "too many loop in nd opt\n"));
409 if (ndopts->nd_opts_done)
417 * ND6 timer routine to handle ND6 entries
420 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
424 LLE_WLOCK_ASSERT(ln);
429 canceled = callout_stop(&ln->ln_timer_ch);
431 ln->la_expire = time_second + tick / hz;
433 if (tick > INT_MAX) {
434 ln->ln_ntick = tick - INT_MAX;
435 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
436 nd6_llinfo_timer, ln);
439 canceled = callout_reset(&ln->ln_timer_ch, tick,
440 nd6_llinfo_timer, ln);
448 nd6_llinfo_settimer(struct llentry *ln, long tick)
452 nd6_llinfo_settimer_locked(ln, tick);
457 nd6_llinfo_timer(void *arg)
460 struct in6_addr *dst;
462 struct nd_ifinfo *ndi = NULL;
464 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
465 ln = (struct llentry *)arg;
466 LLE_WLOCK_ASSERT(ln);
467 ifp = ln->lle_tbl->llt_ifp;
469 CURVNET_SET(ifp->if_vnet);
471 if (ln->ln_ntick > 0) {
472 if (ln->ln_ntick > INT_MAX) {
473 ln->ln_ntick -= INT_MAX;
474 nd6_llinfo_settimer_locked(ln, INT_MAX);
477 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
482 ndi = ND_IFINFO(ifp);
483 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
484 if (ln->la_flags & LLE_STATIC) {
488 if (ln->la_flags & LLE_DELETED) {
489 (void)nd6_free(ln, 0);
494 switch (ln->ln_state) {
495 case ND6_LLINFO_INCOMPLETE:
496 if (ln->la_asked < V_nd6_mmaxtries) {
498 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
500 nd6_ns_output(ifp, NULL, dst, ln, 0);
503 struct mbuf *m = ln->la_hold;
508 * assuming every packet in la_hold has the
509 * same IP header. Send error after unlock.
514 clear_llinfo_pqueue(ln);
516 (void)nd6_free(ln, 0);
519 icmp6_error2(m, ICMP6_DST_UNREACH,
520 ICMP6_DST_UNREACH_ADDR, 0, ifp);
523 case ND6_LLINFO_REACHABLE:
524 if (!ND6_LLINFO_PERMANENT(ln)) {
525 ln->ln_state = ND6_LLINFO_STALE;
526 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
530 case ND6_LLINFO_STALE:
531 /* Garbage Collection(RFC 2461 5.3) */
532 if (!ND6_LLINFO_PERMANENT(ln)) {
533 (void)nd6_free(ln, 1);
538 case ND6_LLINFO_DELAY:
539 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
542 ln->ln_state = ND6_LLINFO_PROBE;
543 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
545 nd6_ns_output(ifp, dst, dst, ln, 0);
548 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
549 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
552 case ND6_LLINFO_PROBE:
553 if (ln->la_asked < V_nd6_umaxtries) {
555 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
557 nd6_ns_output(ifp, dst, dst, ln, 0);
560 (void)nd6_free(ln, 0);
565 panic("%s: paths in a dark night can be confusing: %d",
566 __func__, ln->ln_state);
576 * ND6 timer routine to expire default route list and prefix list
581 CURVNET_SET((struct vnet *) arg);
583 struct nd_defrouter *dr, *ndr;
584 struct nd_prefix *pr, *npr;
585 struct in6_ifaddr *ia6, *nia6;
587 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
590 /* expire default router list */
592 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
593 if (dr->expire && dr->expire < time_second)
598 * expire interface addresses.
599 * in the past the loop was inside prefix expiry processing.
600 * However, from a stricter speci-confrmance standpoint, we should
601 * rather separate address lifetimes and prefix lifetimes.
603 * XXXRW: in6_ifaddrhead locking.
606 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
607 /* check address lifetime */
608 if (IFA6_IS_INVALID(ia6)) {
612 * If the expiring address is temporary, try
613 * regenerating a new one. This would be useful when
614 * we suspended a laptop PC, then turned it on after a
615 * period that could invalidate all temporary
616 * addresses. Although we may have to restart the
617 * loop (see below), it must be after purging the
618 * address. Otherwise, we'd see an infinite loop of
621 if (V_ip6_use_tempaddr &&
622 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
623 if (regen_tmpaddr(ia6) == 0)
627 in6_purgeaddr(&ia6->ia_ifa);
630 goto addrloop; /* XXX: see below */
631 } else if (IFA6_IS_DEPRECATED(ia6)) {
632 int oldflags = ia6->ia6_flags;
634 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
637 * If a temporary address has just become deprecated,
638 * regenerate a new one if possible.
640 if (V_ip6_use_tempaddr &&
641 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
642 (oldflags & IN6_IFF_DEPRECATED) == 0) {
644 if (regen_tmpaddr(ia6) == 0) {
646 * A new temporary address is
648 * XXX: this means the address chain
649 * has changed while we are still in
650 * the loop. Although the change
651 * would not cause disaster (because
652 * it's not a deletion, but an
653 * addition,) we'd rather restart the
654 * loop just for safety. Or does this
655 * significantly reduce performance??
662 * A new RA might have made a deprecated address
665 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
669 /* expire prefix list */
670 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
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) {
680 * address expiration and prefix expiration are
681 * separate. NEVER perform in6_purgeaddr here.
691 * ia6 - deprecated/invalidated temporary address
694 regen_tmpaddr(struct in6_ifaddr *ia6)
698 struct in6_ifaddr *public_ifa6 = NULL;
700 ifp = ia6->ia_ifa.ifa_ifp;
702 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
703 struct in6_ifaddr *it6;
705 if (ifa->ifa_addr->sa_family != AF_INET6)
708 it6 = (struct in6_ifaddr *)ifa;
710 /* ignore no autoconf addresses. */
711 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
714 /* ignore autoconf addresses with different prefixes. */
715 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
719 * Now we are looking at an autoconf address with the same
720 * prefix as ours. If the address is temporary and is still
721 * preferred, do not create another one. It would be rare, but
722 * could happen, for example, when we resume a laptop PC after
725 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
726 !IFA6_IS_DEPRECATED(it6)) {
732 * This is a public autoconf address that has the same prefix
733 * as ours. If it is preferred, keep it. We can't break the
734 * loop here, because there may be a still-preferred temporary
735 * address with the prefix.
737 if (!IFA6_IS_DEPRECATED(it6))
740 if (public_ifa6 != NULL)
741 ifa_ref(&public_ifa6->ia_ifa);
743 IF_ADDR_RUNLOCK(ifp);
745 if (public_ifa6 != NULL) {
748 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
749 ifa_free(&public_ifa6->ia_ifa);
750 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
751 " tmp addr,errno=%d\n", e);
754 ifa_free(&public_ifa6->ia_ifa);
762 * Nuke neighbor cache/prefix/default router management table, right before
766 nd6_purge(struct ifnet *ifp)
768 struct nd_defrouter *dr, *ndr;
769 struct nd_prefix *pr, *npr;
772 * Nuke default router list entries toward ifp.
773 * We defer removal of default router list entries that is installed
774 * in the routing table, in order to keep additional side effects as
777 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
785 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
793 /* Nuke prefix list entries toward ifp */
794 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
795 if (pr->ndpr_ifp == ifp) {
797 * Because if_detach() does *not* release prefixes
798 * while purging addresses the reference count will
799 * still be above zero. We therefore reset it to
800 * make sure that the prefix really gets purged.
805 * Previously, pr->ndpr_addr is removed as well,
806 * but I strongly believe we don't have to do it.
807 * nd6_purge() is only called from in6_ifdetach(),
808 * which removes all the associated interface addresses
810 * (jinmei@kame.net 20010129)
816 /* cancel default outgoing interface setting */
817 if (V_nd6_defifindex == ifp->if_index)
818 nd6_setdefaultiface(0);
820 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
821 /* Refresh default router list. */
826 * We do not nuke the neighbor cache entries here any more
827 * because the neighbor cache is kept in if_afdata[AF_INET6].
828 * nd6_purge() is invoked by in6_ifdetach() which is called
829 * from if_detach() where everything gets purged. So let
830 * in6_domifdetach() do the actual L2 table purging work.
835 * the caller acquires and releases the lock on the lltbls
836 * Returns the llentry locked
839 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
841 struct sockaddr_in6 sin6;
845 bzero(&sin6, sizeof(sin6));
846 sin6.sin6_len = sizeof(struct sockaddr_in6);
847 sin6.sin6_family = AF_INET6;
848 sin6.sin6_addr = *addr6;
850 IF_AFDATA_LOCK_ASSERT(ifp);
853 if (flags & ND6_CREATE)
854 llflags |= LLE_CREATE;
855 if (flags & ND6_EXCLUSIVE)
856 llflags |= LLE_EXCLUSIVE;
858 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
859 if ((ln != NULL) && (llflags & LLE_CREATE))
860 ln->ln_state = ND6_LLINFO_NOSTATE;
866 * Test whether a given IPv6 address is a neighbor or not, ignoring
867 * the actual neighbor cache. The neighbor cache is ignored in order
868 * to not reenter the routing code from within itself.
871 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
873 struct nd_prefix *pr;
874 struct ifaddr *dstaddr;
877 * A link-local address is always a neighbor.
878 * XXX: a link does not necessarily specify a single interface.
880 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
881 struct sockaddr_in6 sin6_copy;
885 * We need sin6_copy since sa6_recoverscope() may modify the
889 if (sa6_recoverscope(&sin6_copy))
890 return (0); /* XXX: should be impossible */
891 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
893 if (sin6_copy.sin6_scope_id == zone)
900 * If the address matches one of our addresses,
901 * it should be a neighbor.
902 * If the address matches one of our on-link prefixes, it should be a
905 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
906 if (pr->ndpr_ifp != ifp)
909 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
912 /* Always use the default FIB here. */
913 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
914 0, 0, RT_DEFAULT_FIB);
918 * This is the case where multiple interfaces
919 * have the same prefix, but only one is installed
920 * into the routing table and that prefix entry
921 * is not the one being examined here. In the case
922 * where RADIX_MPATH is enabled, multiple route
923 * entries (of the same rt_key value) will be
924 * installed because the interface addresses all
927 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
928 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
935 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
936 &addr->sin6_addr, &pr->ndpr_mask))
941 * If the address is assigned on the node of the other side of
942 * a p2p interface, the address should be a neighbor.
944 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
945 if (dstaddr != NULL) {
946 if (dstaddr->ifa_ifp == ifp) {
954 * If the default router list is empty, all addresses are regarded
955 * as on-link, and thus, as a neighbor.
957 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
958 TAILQ_EMPTY(&V_nd_defrouter) &&
959 V_nd6_defifindex == ifp->if_index) {
968 * Detect if a given IPv6 address identifies a neighbor on a given link.
969 * XXX: should take care of the destination of a p2p link?
972 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
977 IF_AFDATA_UNLOCK_ASSERT(ifp);
978 if (nd6_is_new_addr_neighbor(addr, ifp))
982 * Even if the address matches none of our addresses, it might be
983 * in the neighbor cache.
986 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
990 IF_AFDATA_UNLOCK(ifp);
995 * Free an nd6 llinfo entry.
996 * Since the function would cause significant changes in the kernel, DO NOT
997 * make it global, unless you have a strong reason for the change, and are sure
998 * that the change is safe.
1000 static struct llentry *
1001 nd6_free(struct llentry *ln, int gc)
1003 struct llentry *next;
1004 struct nd_defrouter *dr;
1007 LLE_WLOCK_ASSERT(ln);
1010 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1011 * even though it is not harmful, it was not really necessary.
1015 nd6_llinfo_settimer_locked(ln, -1);
1017 ifp = ln->lle_tbl->llt_ifp;
1019 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1020 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1022 if (dr != NULL && dr->expire &&
1023 ln->ln_state == ND6_LLINFO_STALE && gc) {
1025 * If the reason for the deletion is just garbage
1026 * collection, and the neighbor is an active default
1027 * router, do not delete it. Instead, reset the GC
1028 * timer using the router's lifetime.
1029 * Simply deleting the entry would affect default
1030 * router selection, which is not necessarily a good
1031 * thing, especially when we're using router preference
1033 * XXX: the check for ln_state would be redundant,
1034 * but we intentionally keep it just in case.
1036 if (dr->expire > time_second)
1037 nd6_llinfo_settimer_locked(ln,
1038 (dr->expire - time_second) * hz);
1040 nd6_llinfo_settimer_locked(ln,
1041 (long)V_nd6_gctimer * hz);
1043 next = LIST_NEXT(ln, lle_next);
1051 * Unreachablity of a router might affect the default
1052 * router selection and on-link detection of advertised
1057 * Temporarily fake the state to choose a new default
1058 * router and to perform on-link determination of
1059 * prefixes correctly.
1060 * Below the state will be set correctly,
1061 * or the entry itself will be deleted.
1063 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1066 if (ln->ln_router || dr) {
1069 * We need to unlock to avoid a LOR with rt6_flush() with the
1070 * rnh and for the calls to pfxlist_onlink_check() and
1071 * defrouter_select() in the block further down for calls
1072 * into nd6_lookup(). We still hold a ref.
1077 * rt6_flush must be called whether or not the neighbor
1078 * is in the Default Router List.
1079 * See a corresponding comment in nd6_na_input().
1081 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1086 * Since defrouter_select() does not affect the
1087 * on-link determination and MIP6 needs the check
1088 * before the default router selection, we perform
1091 pfxlist_onlink_check();
1094 * Refresh default router list.
1099 if (ln->ln_router || dr)
1104 * Before deleting the entry, remember the next entry as the
1105 * return value. We need this because pfxlist_onlink_check() above
1106 * might have freed other entries (particularly the old next entry) as
1107 * a side effect (XXX).
1109 next = LIST_NEXT(ln, lle_next);
1112 * Save to unlock. We still hold an extra reference and will not
1113 * free(9) in llentry_free() if someone else holds one as well.
1116 IF_AFDATA_LOCK(ifp);
1120 IF_AFDATA_UNLOCK(ifp);
1126 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1128 * XXX cost-effective methods?
1131 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1136 if ((dst6 == NULL) || (rt == NULL))
1140 IF_AFDATA_LOCK(ifp);
1141 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1142 IF_AFDATA_UNLOCK(ifp);
1146 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1150 * if we get upper-layer reachability confirmation many times,
1151 * it is possible we have false information.
1155 if (ln->ln_byhint > V_nd6_maxnudhint) {
1160 ln->ln_state = ND6_LLINFO_REACHABLE;
1161 if (!ND6_LLINFO_PERMANENT(ln)) {
1162 nd6_llinfo_settimer_locked(ln,
1163 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1171 * Rejuvenate this function for routing operations related
1175 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1177 struct sockaddr_in6 *gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1178 struct nd_defrouter *dr;
1179 struct ifnet *ifp = rt->rt_ifp;
1191 * Only indirect routes are interesting.
1193 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1196 * check for default route
1198 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1199 &SIN6(rt_key(rt))->sin6_addr)) {
1201 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1211 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1213 struct in6_drlist *drl = (struct in6_drlist *)data;
1214 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1215 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1216 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1217 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1218 struct nd_defrouter *dr;
1219 struct nd_prefix *pr;
1220 int i = 0, error = 0;
1224 case SIOCGDRLST_IN6:
1226 * obsolete API, use sysctl under net.inet6.icmp6
1228 bzero(drl, sizeof(*drl));
1230 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1233 drl->defrouter[i].rtaddr = dr->rtaddr;
1234 in6_clearscope(&drl->defrouter[i].rtaddr);
1236 drl->defrouter[i].flags = dr->flags;
1237 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1238 drl->defrouter[i].expire = dr->expire;
1239 drl->defrouter[i].if_index = dr->ifp->if_index;
1244 case SIOCGPRLST_IN6:
1246 * obsolete API, use sysctl under net.inet6.icmp6
1248 * XXX the structure in6_prlist was changed in backward-
1249 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1250 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1253 * XXX meaning of fields, especialy "raflags", is very
1254 * differnet between RA prefix list and RR/static prefix list.
1255 * how about separating ioctls into two?
1257 bzero(oprl, sizeof(*oprl));
1259 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1260 struct nd_pfxrouter *pfr;
1265 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1266 oprl->prefix[i].raflags = pr->ndpr_raf;
1267 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1268 oprl->prefix[i].vltime = pr->ndpr_vltime;
1269 oprl->prefix[i].pltime = pr->ndpr_pltime;
1270 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1271 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1272 oprl->prefix[i].expire = 0;
1276 /* XXX: we assume time_t is signed. */
1279 ((sizeof(maxexpire) * 8) - 1));
1280 if (pr->ndpr_vltime <
1281 maxexpire - pr->ndpr_lastupdate) {
1282 oprl->prefix[i].expire =
1283 pr->ndpr_lastupdate +
1286 oprl->prefix[i].expire = maxexpire;
1290 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1292 #define RTRADDR oprl->prefix[i].advrtr[j]
1293 RTRADDR = pfr->router->rtaddr;
1294 in6_clearscope(&RTRADDR);
1299 oprl->prefix[i].advrtrs = j;
1300 oprl->prefix[i].origin = PR_ORIG_RA;
1307 case OSIOCGIFINFO_IN6:
1309 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1310 bzero(&ND, sizeof(ND));
1311 ND.linkmtu = IN6_LINKMTU(ifp);
1312 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1313 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1314 ND.reachable = ND_IFINFO(ifp)->reachable;
1315 ND.retrans = ND_IFINFO(ifp)->retrans;
1316 ND.flags = ND_IFINFO(ifp)->flags;
1317 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1318 ND.chlim = ND_IFINFO(ifp)->chlim;
1320 case SIOCGIFINFO_IN6:
1321 ND = *ND_IFINFO(ifp);
1323 case SIOCSIFINFO_IN6:
1325 * used to change host variables from userland.
1326 * intented for a use on router to reflect RA configurations.
1328 /* 0 means 'unspecified' */
1329 if (ND.linkmtu != 0) {
1330 if (ND.linkmtu < IPV6_MMTU ||
1331 ND.linkmtu > IN6_LINKMTU(ifp)) {
1335 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1338 if (ND.basereachable != 0) {
1339 int obasereachable = ND_IFINFO(ifp)->basereachable;
1341 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1342 if (ND.basereachable != obasereachable)
1343 ND_IFINFO(ifp)->reachable =
1344 ND_COMPUTE_RTIME(ND.basereachable);
1346 if (ND.retrans != 0)
1347 ND_IFINFO(ifp)->retrans = ND.retrans;
1349 ND_IFINFO(ifp)->chlim = ND.chlim;
1351 case SIOCSIFINFO_FLAGS:
1354 struct in6_ifaddr *ia;
1356 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1357 !(ND.flags & ND6_IFF_IFDISABLED)) {
1358 /* ifdisabled 1->0 transision */
1361 * If the interface is marked as ND6_IFF_IFDISABLED and
1362 * has an link-local address with IN6_IFF_DUPLICATED,
1363 * do not clear ND6_IFF_IFDISABLED.
1364 * See RFC 4862, Section 5.4.5.
1366 int duplicated_linklocal = 0;
1369 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1370 if (ifa->ifa_addr->sa_family != AF_INET6)
1372 ia = (struct in6_ifaddr *)ifa;
1373 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1374 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1375 duplicated_linklocal = 1;
1379 IF_ADDR_RUNLOCK(ifp);
1381 if (duplicated_linklocal) {
1382 ND.flags |= ND6_IFF_IFDISABLED;
1383 log(LOG_ERR, "Cannot enable an interface"
1384 " with a link-local address marked"
1387 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1388 if (ifp->if_flags & IFF_UP)
1391 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1392 (ND.flags & ND6_IFF_IFDISABLED)) {
1393 /* ifdisabled 0->1 transision */
1394 /* Mark all IPv6 address as tentative. */
1396 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1398 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1399 if (ifa->ifa_addr->sa_family != AF_INET6)
1401 ia = (struct in6_ifaddr *)ifa;
1402 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1404 IF_ADDR_RUNLOCK(ifp);
1407 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1408 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1409 /* auto_linklocal 0->1 transision */
1411 /* If no link-local address on ifp, configure */
1412 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1413 in6_ifattach(ifp, NULL);
1414 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1415 ifp->if_flags & IFF_UP) {
1417 * When the IF already has
1418 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1419 * address is assigned, and IFF_UP, try to
1422 int haslinklocal = 0;
1425 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1426 if (ifa->ifa_addr->sa_family != AF_INET6)
1428 ia = (struct in6_ifaddr *)ifa;
1429 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1434 IF_ADDR_RUNLOCK(ifp);
1436 in6_ifattach(ifp, NULL);
1440 ND_IFINFO(ifp)->flags = ND.flags;
1443 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1444 /* sync kernel routing table with the default router list */
1448 case SIOCSPFXFLUSH_IN6:
1450 /* flush all the prefix advertised by routers */
1451 struct nd_prefix *pr, *next;
1454 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1455 struct in6_ifaddr *ia, *ia_next;
1457 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1460 /* do we really have to remove addresses as well? */
1461 /* XXXRW: in6_ifaddrhead locking. */
1462 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1464 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1467 if (ia->ia6_ndpr == pr)
1468 in6_purgeaddr(&ia->ia_ifa);
1475 case SIOCSRTRFLUSH_IN6:
1477 /* flush all the default routers */
1478 struct nd_defrouter *dr, *next;
1482 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1489 case SIOCGNBRINFO_IN6:
1492 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1494 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1497 IF_AFDATA_LOCK(ifp);
1498 ln = nd6_lookup(&nb_addr, 0, ifp);
1499 IF_AFDATA_UNLOCK(ifp);
1505 nbi->state = ln->ln_state;
1506 nbi->asked = ln->la_asked;
1507 nbi->isrouter = ln->ln_router;
1508 nbi->expire = ln->la_expire;
1512 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1513 ndif->ifindex = V_nd6_defifindex;
1515 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1516 return (nd6_setdefaultiface(ndif->ifindex));
1522 * Create neighbor cache entry and cache link-layer address,
1523 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1526 * code - type dependent information
1529 * The caller of this function already acquired the ndp
1530 * cache table lock because the cache entry is returned.
1533 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1534 int lladdrlen, int type, int code)
1536 struct llentry *ln = NULL;
1543 uint16_t router = 0;
1544 struct sockaddr_in6 sin6;
1545 struct mbuf *chain = NULL;
1546 int static_route = 0;
1548 IF_AFDATA_UNLOCK_ASSERT(ifp);
1551 panic("ifp == NULL in nd6_cache_lladdr");
1553 panic("from == NULL in nd6_cache_lladdr");
1555 /* nothing must be updated for unspecified address */
1556 if (IN6_IS_ADDR_UNSPECIFIED(from))
1560 * Validation about ifp->if_addrlen and lladdrlen must be done in
1563 * XXX If the link does not have link-layer adderss, what should
1564 * we do? (ifp->if_addrlen == 0)
1565 * Spec says nothing in sections for RA, RS and NA. There's small
1566 * description on it in NS section (RFC 2461 7.2.3).
1568 flags = lladdr ? ND6_EXCLUSIVE : 0;
1569 IF_AFDATA_LOCK(ifp);
1570 ln = nd6_lookup(from, flags, ifp);
1573 flags |= ND6_EXCLUSIVE;
1574 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1575 IF_AFDATA_UNLOCK(ifp);
1578 IF_AFDATA_UNLOCK(ifp);
1579 /* do nothing if static ndp is set */
1580 if (ln->la_flags & LLE_STATIC) {
1589 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1590 if (olladdr && lladdr) {
1591 llchange = bcmp(lladdr, &ln->ll_addr,
1597 * newentry olladdr lladdr llchange (*=record)
1600 * 0 n y -- (3) * STALE
1602 * 0 y y y (5) * STALE
1603 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1604 * 1 -- y -- (7) * STALE
1607 if (lladdr) { /* (3-5) and (7) */
1609 * Record source link-layer address
1610 * XXX is it dependent to ifp->if_type?
1612 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1613 ln->la_flags |= LLE_VALID;
1617 if ((!olladdr && lladdr != NULL) || /* (3) */
1618 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1620 newstate = ND6_LLINFO_STALE;
1621 } else /* (1-2,4) */
1625 if (lladdr == NULL) /* (6) */
1626 newstate = ND6_LLINFO_NOSTATE;
1628 newstate = ND6_LLINFO_STALE;
1633 * Update the state of the neighbor cache.
1635 ln->ln_state = newstate;
1637 if (ln->ln_state == ND6_LLINFO_STALE) {
1639 * XXX: since nd6_output() below will cause
1640 * state tansition to DELAY and reset the timer,
1641 * we must set the timer now, although it is actually
1644 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1647 struct mbuf *m_hold, *m_hold_next;
1650 * reset the la_hold in advance, to explicitly
1651 * prevent a la_hold lookup in nd6_output()
1652 * (wouldn't happen, though...)
1654 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1655 m_hold; m_hold = m_hold_next) {
1656 m_hold_next = m_hold->m_nextpkt;
1657 m_hold->m_nextpkt = NULL;
1660 * we assume ifp is not a p2p here, so
1661 * just set the 2nd argument as the
1664 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1667 * If we have mbufs in the chain we need to do
1668 * deferred transmit. Copy the address from the
1669 * llentry before dropping the lock down below.
1672 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1674 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1675 /* probe right away */
1676 nd6_llinfo_settimer_locked((void *)ln, 0);
1681 * ICMP6 type dependent behavior.
1683 * NS: clear IsRouter if new entry
1684 * RS: clear IsRouter
1685 * RA: set IsRouter if there's lladdr
1686 * redir: clear IsRouter if new entry
1689 * The spec says that we must set IsRouter in the following cases:
1690 * - If lladdr exist, set IsRouter. This means (1-5).
1691 * - If it is old entry (!newentry), set IsRouter. This means (7).
1692 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1693 * A quetion arises for (1) case. (1) case has no lladdr in the
1694 * neighbor cache, this is similar to (6).
1695 * This case is rare but we figured that we MUST NOT set IsRouter.
1697 * newentry olladdr lladdr llchange NS RS RA redir
1699 * 0 n n -- (1) c ? s
1700 * 0 y n -- (2) c s s
1701 * 0 n y -- (3) c s s
1704 * 1 -- n -- (6) c c c s
1705 * 1 -- y -- (7) c c s c s
1709 switch (type & 0xff) {
1710 case ND_NEIGHBOR_SOLICIT:
1712 * New entry must have is_router flag cleared.
1714 if (is_newentry) /* (6-7) */
1719 * If the icmp is a redirect to a better router, always set the
1720 * is_router flag. Otherwise, if the entry is newly created,
1721 * clear the flag. [RFC 2461, sec 8.3]
1723 if (code == ND_REDIRECT_ROUTER)
1725 else if (is_newentry) /* (6-7) */
1728 case ND_ROUTER_SOLICIT:
1730 * is_router flag must always be cleared.
1734 case ND_ROUTER_ADVERT:
1736 * Mark an entry with lladdr as a router.
1738 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1739 (is_newentry && lladdr)) { /* (7) */
1746 static_route = (ln->la_flags & LLE_STATIC);
1747 router = ln->ln_router;
1749 if (flags & ND6_EXCLUSIVE)
1757 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1760 * When the link-layer address of a router changes, select the
1761 * best router again. In particular, when the neighbor entry is newly
1762 * created, it might affect the selection policy.
1763 * Question: can we restrict the first condition to the "is_newentry"
1765 * XXX: when we hear an RA from a new router with the link-layer
1766 * address option, defrouter_select() is called twice, since
1767 * defrtrlist_update called the function as well. However, I believe
1768 * we can compromise the overhead, since it only happens the first
1770 * XXX: although defrouter_select() should not have a bad effect
1771 * for those are not autoconfigured hosts, we explicitly avoid such
1774 if (do_update && router &&
1775 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1777 * guaranteed recursion
1785 if (flags & ND6_EXCLUSIVE)
1796 nd6_slowtimo(void *arg)
1798 CURVNET_SET((struct vnet *) arg);
1799 struct nd_ifinfo *nd6if;
1802 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1803 nd6_slowtimo, curvnet);
1804 IFNET_RLOCK_NOSLEEP();
1805 TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1806 nd6if = ND_IFINFO(ifp);
1807 if (nd6if->basereachable && /* already initialized */
1808 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1810 * Since reachable time rarely changes by router
1811 * advertisements, we SHOULD insure that a new random
1812 * value gets recomputed at least once every few hours.
1815 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1816 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1819 IFNET_RUNLOCK_NOSLEEP();
1824 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1825 struct sockaddr_in6 *dst, struct rtentry *rt0)
1828 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1833 * Note that I'm not enforcing any global serialization
1834 * lle state or asked changes here as the logic is too
1835 * complicated to avoid having to always acquire an exclusive
1840 #define senderr(e) { error = (e); goto bad;}
1843 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1844 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1845 struct mbuf **chain)
1847 struct mbuf *m = m0;
1849 struct llentry *ln = lle;
1850 struct ip6_hdr *ip6;
1858 LLE_WLOCK_ASSERT(lle);
1860 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1863 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1866 if (nd6_need_cache(ifp) == 0)
1870 * next hop determination. This routine is derived from ether_output.
1874 * Address resolution or Neighbor Unreachability Detection
1876 * At this point, the destination of the packet must be a unicast
1877 * or an anycast address(i.e. not a multicast).
1880 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1883 IF_AFDATA_LOCK(ifp);
1884 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1885 IF_AFDATA_UNLOCK(ifp);
1886 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1888 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1889 * the condition below is not very efficient. But we believe
1890 * it is tolerable, because this should be a rare case.
1892 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1893 IF_AFDATA_LOCK(ifp);
1894 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1895 IF_AFDATA_UNLOCK(ifp);
1899 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1900 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1901 char ip6buf[INET6_ADDRSTRLEN];
1903 "nd6_output: can't allocate llinfo for %s "
1905 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1906 senderr(EIO); /* XXX: good error? */
1908 goto sendpkt; /* send anyway */
1911 /* We don't have to do link-layer address resolution on a p2p link. */
1912 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1913 ln->ln_state < ND6_LLINFO_REACHABLE) {
1914 if ((flags & LLE_EXCLUSIVE) == 0) {
1915 flags |= LLE_EXCLUSIVE;
1918 ln->ln_state = ND6_LLINFO_STALE;
1919 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1923 * The first time we send a packet to a neighbor whose entry is
1924 * STALE, we have to change the state to DELAY and a sets a timer to
1925 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1926 * neighbor unreachability detection on expiration.
1929 if (ln->ln_state == ND6_LLINFO_STALE) {
1930 if ((flags & LLE_EXCLUSIVE) == 0) {
1931 flags |= LLE_EXCLUSIVE;
1936 ln->ln_state = ND6_LLINFO_DELAY;
1937 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1941 * If the neighbor cache entry has a state other than INCOMPLETE
1942 * (i.e. its link-layer address is already resolved), just
1945 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1949 * There is a neighbor cache entry, but no ethernet address
1950 * response yet. Append this latest packet to the end of the
1951 * packet queue in the mbuf, unless the number of the packet
1952 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1953 * the oldest packet in the queue will be removed.
1955 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1956 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1958 if ((flags & LLE_EXCLUSIVE) == 0) {
1959 flags |= LLE_EXCLUSIVE;
1964 LLE_WLOCK_ASSERT(ln);
1967 struct mbuf *m_hold;
1971 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1973 if (m_hold->m_nextpkt == NULL) {
1974 m_hold->m_nextpkt = m;
1978 while (i >= V_nd6_maxqueuelen) {
1979 m_hold = ln->la_hold;
1980 ln->la_hold = ln->la_hold->m_nextpkt;
1989 * If there has been no NS for the neighbor after entering the
1990 * INCOMPLETE state, send the first solicitation.
1992 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1995 nd6_llinfo_settimer_locked(ln,
1996 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1998 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1999 if (lle != NULL && ln == lle)
2002 } else if (lle == NULL || ln != lle) {
2004 * We did the lookup (no lle arg) so we
2005 * need to do the unlock here.
2013 /* discard the packet if IPv6 operation is disabled on the interface */
2014 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2015 error = ENETDOWN; /* better error? */
2019 * ln is valid and the caller did not pass in
2022 if ((ln != NULL) && (lle == NULL)) {
2023 if (flags & LLE_EXCLUSIVE)
2030 mac_netinet6_nd6_send(ifp, m);
2034 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2035 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2036 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2037 * to be diverted to user space. When re-injected into the kernel,
2038 * send_output() will directly dispatch them to the outgoing interface.
2040 if (send_sendso_input_hook != NULL) {
2041 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2043 ip6 = mtod(m, struct ip6_hdr *);
2044 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2045 /* Use the SEND socket */
2046 error = send_sendso_input_hook(m, ifp, SND_OUT,
2048 /* -1 == no app on SEND socket */
2049 if (error == 0 || error != -1)
2055 * We were passed in a pointer to an lle with the lock held
2056 * this means that we can't call if_output as we will
2057 * recurse on the lle lock - so what we do is we create
2058 * a list of mbufs to send and transmit them in the caller
2059 * after the lock is dropped
2068 * append mbuf to end of deferred chain
2071 while (mb->m_nextpkt != NULL)
2077 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2078 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2081 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2086 * ln is valid and the caller did not pass in
2089 if ((ln != NULL) && (lle == NULL)) {
2090 if (flags & LLE_EXCLUSIVE)
2103 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2104 struct sockaddr_in6 *dst, struct route *ro)
2106 struct mbuf *m, *m_head;
2107 struct ifnet *outifp;
2111 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2118 m_head = m_head->m_nextpkt;
2119 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2124 * note that intermediate errors are blindly ignored - but this is
2125 * the same convention as used with nd6_output when called by
2133 nd6_need_cache(struct ifnet *ifp)
2136 * XXX: we currently do not make neighbor cache on any interface
2137 * other than ARCnet, Ethernet, FDDI and GIF.
2140 * - unidirectional tunnels needs no ND
2142 switch (ifp->if_type) {
2150 #ifdef IFT_IEEE80211
2153 case IFT_INFINIBAND:
2154 case IFT_GIF: /* XXX need more cases? */
2158 case IFT_PROPVIRTUAL:
2166 * the callers of this function need to be re-worked to drop
2167 * the lle lock, drop here for now
2170 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2171 struct sockaddr *dst, u_char *desten, struct llentry **lle)
2176 IF_AFDATA_UNLOCK_ASSERT(ifp);
2177 if (m->m_flags & M_MCAST) {
2180 switch (ifp->if_type) {
2186 #ifdef IFT_IEEE80211
2191 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2196 * netbsd can use if_broadcastaddr, but we don't do so
2197 * to reduce # of ifdef.
2199 for (i = 0; i < ifp->if_addrlen; i++)
2207 return (EAFNOSUPPORT);
2213 * the entry should have been created in nd6_store_lladdr
2215 IF_AFDATA_LOCK(ifp);
2216 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2217 IF_AFDATA_UNLOCK(ifp);
2218 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2221 /* this could happen, if we could not allocate memory */
2226 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2230 * A *small* use after free race exists here
2236 clear_llinfo_pqueue(struct llentry *ln)
2238 struct mbuf *m_hold, *m_hold_next;
2240 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2241 m_hold_next = m_hold->m_nextpkt;
2242 m_hold->m_nextpkt = NULL;
2250 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2251 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2253 SYSCTL_DECL(_net_inet6_icmp6);
2255 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2256 CTLFLAG_RD, nd6_sysctl_drlist, "");
2257 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2258 CTLFLAG_RD, nd6_sysctl_prlist, "");
2259 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2260 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2263 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2266 char buf[1024] __aligned(4);
2267 struct in6_defrouter *d, *de;
2268 struct nd_defrouter *dr;
2274 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2275 d = (struct in6_defrouter *)buf;
2276 de = (struct in6_defrouter *)(buf + sizeof(buf));
2279 bzero(d, sizeof(*d));
2280 d->rtaddr.sin6_family = AF_INET6;
2281 d->rtaddr.sin6_len = sizeof(d->rtaddr);
2282 d->rtaddr.sin6_addr = dr->rtaddr;
2283 error = sa6_recoverscope(&d->rtaddr);
2286 d->flags = dr->flags;
2287 d->rtlifetime = dr->rtlifetime;
2288 d->expire = dr->expire;
2289 d->if_index = dr->ifp->if_index;
2291 panic("buffer too short");
2293 error = SYSCTL_OUT(req, buf, sizeof(*d));
2302 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2305 char buf[1024] __aligned(4);
2306 struct in6_prefix *p, *pe;
2307 struct nd_prefix *pr;
2308 char ip6buf[INET6_ADDRSTRLEN];
2314 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2317 struct sockaddr_in6 *sin6, *s6;
2318 struct nd_pfxrouter *pfr;
2320 p = (struct in6_prefix *)buf;
2321 pe = (struct in6_prefix *)(buf + sizeof(buf));
2324 bzero(p, sizeof(*p));
2325 sin6 = (struct sockaddr_in6 *)(p + 1);
2327 p->prefix = pr->ndpr_prefix;
2328 if (sa6_recoverscope(&p->prefix)) {
2330 "scope error in prefix list (%s)\n",
2331 ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2332 /* XXX: press on... */
2334 p->raflags = pr->ndpr_raf;
2335 p->prefixlen = pr->ndpr_plen;
2336 p->vltime = pr->ndpr_vltime;
2337 p->pltime = pr->ndpr_pltime;
2338 p->if_index = pr->ndpr_ifp->if_index;
2339 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2344 /* XXX: we assume time_t is signed. */
2347 ((sizeof(maxexpire) * 8) - 1));
2348 if (pr->ndpr_vltime <
2349 maxexpire - pr->ndpr_lastupdate) {
2350 p->expire = pr->ndpr_lastupdate +
2353 p->expire = maxexpire;
2355 p->refcnt = pr->ndpr_refcnt;
2356 p->flags = pr->ndpr_stateflags;
2357 p->origin = PR_ORIG_RA;
2359 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2360 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2364 s6 = &sin6[advrtrs];
2365 bzero(s6, sizeof(*s6));
2366 s6->sin6_family = AF_INET6;
2367 s6->sin6_len = sizeof(*sin6);
2368 s6->sin6_addr = pfr->router->rtaddr;
2369 if (sa6_recoverscope(s6)) {
2372 "prefix list (%s)\n",
2374 &pfr->router->rtaddr));
2378 p->advrtrs = advrtrs;
2380 panic("buffer too short");
2382 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2383 error = SYSCTL_OUT(req, buf, advance);