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(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|M_ZERO);
180 nd->chlim = IPV6_DEFHLIM;
181 nd->basereachable = REACHABLE_TIME;
182 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
183 nd->retrans = RETRANS_TIMER;
185 nd->flags = ND6_IFF_PERFORMNUD;
187 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
188 if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
189 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
191 /* A loopback interface does not need to accept RTADV. */
192 if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
193 nd->flags |= ND6_IFF_ACCEPT_RTADV;
194 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
195 nd->flags |= ND6_IFF_NO_RADR;
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;
285 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
286 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
288 if (ndopts->nd_opts_search == NULL)
290 if (ndopts->nd_opts_done)
293 nd_opt = ndopts->nd_opts_search;
295 /* make sure nd_opt_len is inside the buffer */
296 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
297 bzero(ndopts, sizeof(*ndopts));
301 olen = nd_opt->nd_opt_len << 3;
304 * Message validation requires that all included
305 * options have a length that is greater than zero.
307 bzero(ndopts, sizeof(*ndopts));
311 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
312 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
313 /* option overruns the end of buffer, invalid */
314 bzero(ndopts, sizeof(*ndopts));
316 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
317 /* reached the end of options chain */
318 ndopts->nd_opts_done = 1;
319 ndopts->nd_opts_search = NULL;
325 * Parse multiple ND options.
326 * This function is much easier to use, for ND routines that do not need
327 * multiple options of the same type.
330 nd6_options(union nd_opts *ndopts)
332 struct nd_opt_hdr *nd_opt;
335 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
336 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
338 if (ndopts->nd_opts_search == NULL)
342 nd_opt = nd6_option(ndopts);
343 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
345 * Message validation requires that all included
346 * options have a length that is greater than zero.
348 ICMP6STAT_INC(icp6s_nd_badopt);
349 bzero(ndopts, sizeof(*ndopts));
356 switch (nd_opt->nd_opt_type) {
357 case ND_OPT_SOURCE_LINKADDR:
358 case ND_OPT_TARGET_LINKADDR:
360 case ND_OPT_REDIRECTED_HEADER:
361 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
363 "duplicated ND6 option found (type=%d)\n",
364 nd_opt->nd_opt_type));
367 ndopts->nd_opt_array[nd_opt->nd_opt_type]
371 case ND_OPT_PREFIX_INFORMATION:
372 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
373 ndopts->nd_opt_array[nd_opt->nd_opt_type]
376 ndopts->nd_opts_pi_end =
377 (struct nd_opt_prefix_info *)nd_opt;
379 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
380 case ND_OPT_RDNSS: /* RFC 6106 */
381 case ND_OPT_DNSSL: /* RFC 6106 */
383 * Silently ignore options we know and do not care about
389 * Unknown options must be silently ignored,
390 * to accomodate future extension to the protocol.
393 "nd6_options: unsupported option %d - "
394 "option ignored\n", nd_opt->nd_opt_type));
399 if (i > V_nd6_maxndopt) {
400 ICMP6STAT_INC(icp6s_nd_toomanyopt);
401 nd6log((LOG_INFO, "too many loop in nd opt\n"));
405 if (ndopts->nd_opts_done)
413 * ND6 timer routine to handle ND6 entries
416 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
420 LLE_WLOCK_ASSERT(ln);
425 canceled = callout_stop(&ln->ln_timer_ch);
427 ln->la_expire = time_second + tick / hz;
429 if (tick > INT_MAX) {
430 ln->ln_ntick = tick - INT_MAX;
431 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
432 nd6_llinfo_timer, ln);
435 canceled = callout_reset(&ln->ln_timer_ch, tick,
436 nd6_llinfo_timer, ln);
444 nd6_llinfo_settimer(struct llentry *ln, long tick)
448 nd6_llinfo_settimer_locked(ln, tick);
453 nd6_llinfo_timer(void *arg)
456 struct in6_addr *dst;
458 struct nd_ifinfo *ndi = NULL;
460 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
461 ln = (struct llentry *)arg;
462 LLE_WLOCK_ASSERT(ln);
463 ifp = ln->lle_tbl->llt_ifp;
465 CURVNET_SET(ifp->if_vnet);
467 if (ln->ln_ntick > 0) {
468 if (ln->ln_ntick > INT_MAX) {
469 ln->ln_ntick -= INT_MAX;
470 nd6_llinfo_settimer_locked(ln, INT_MAX);
473 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
478 ndi = ND_IFINFO(ifp);
479 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
480 if (ln->la_flags & LLE_STATIC) {
484 if (ln->la_flags & LLE_DELETED) {
485 (void)nd6_free(ln, 0);
490 switch (ln->ln_state) {
491 case ND6_LLINFO_INCOMPLETE:
492 if (ln->la_asked < V_nd6_mmaxtries) {
494 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
496 nd6_ns_output(ifp, NULL, dst, ln, 0);
499 struct mbuf *m = ln->la_hold;
504 * assuming every packet in la_hold has the
505 * same IP header. Send error after unlock.
510 clear_llinfo_pqueue(ln);
512 (void)nd6_free(ln, 0);
515 icmp6_error2(m, ICMP6_DST_UNREACH,
516 ICMP6_DST_UNREACH_ADDR, 0, ifp);
519 case ND6_LLINFO_REACHABLE:
520 if (!ND6_LLINFO_PERMANENT(ln)) {
521 ln->ln_state = ND6_LLINFO_STALE;
522 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
526 case ND6_LLINFO_STALE:
527 /* Garbage Collection(RFC 2461 5.3) */
528 if (!ND6_LLINFO_PERMANENT(ln)) {
529 (void)nd6_free(ln, 1);
534 case ND6_LLINFO_DELAY:
535 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
538 ln->ln_state = ND6_LLINFO_PROBE;
539 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
541 nd6_ns_output(ifp, dst, dst, ln, 0);
544 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
545 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
548 case ND6_LLINFO_PROBE:
549 if (ln->la_asked < V_nd6_umaxtries) {
551 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
553 nd6_ns_output(ifp, dst, dst, ln, 0);
556 (void)nd6_free(ln, 0);
561 panic("%s: paths in a dark night can be confusing: %d",
562 __func__, ln->ln_state);
572 * ND6 timer routine to expire default route list and prefix list
577 CURVNET_SET((struct vnet *) arg);
578 struct nd_defrouter *dr, *ndr;
579 struct nd_prefix *pr, *npr;
580 struct in6_ifaddr *ia6, *nia6;
582 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
585 /* expire default router list */
586 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
587 if (dr->expire && dr->expire < time_second)
592 * expire interface addresses.
593 * in the past the loop was inside prefix expiry processing.
594 * However, from a stricter speci-confrmance standpoint, we should
595 * rather separate address lifetimes and prefix lifetimes.
597 * XXXRW: in6_ifaddrhead locking.
600 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
601 /* check address lifetime */
602 if (IFA6_IS_INVALID(ia6)) {
606 * If the expiring address is temporary, try
607 * regenerating a new one. This would be useful when
608 * we suspended a laptop PC, then turned it on after a
609 * period that could invalidate all temporary
610 * addresses. Although we may have to restart the
611 * loop (see below), it must be after purging the
612 * address. Otherwise, we'd see an infinite loop of
615 if (V_ip6_use_tempaddr &&
616 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
617 if (regen_tmpaddr(ia6) == 0)
621 in6_purgeaddr(&ia6->ia_ifa);
624 goto addrloop; /* XXX: see below */
625 } else if (IFA6_IS_DEPRECATED(ia6)) {
626 int oldflags = ia6->ia6_flags;
628 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
631 * If a temporary address has just become deprecated,
632 * regenerate a new one if possible.
634 if (V_ip6_use_tempaddr &&
635 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
636 (oldflags & IN6_IFF_DEPRECATED) == 0) {
638 if (regen_tmpaddr(ia6) == 0) {
640 * A new temporary address is
642 * XXX: this means the address chain
643 * has changed while we are still in
644 * the loop. Although the change
645 * would not cause disaster (because
646 * it's not a deletion, but an
647 * addition,) we'd rather restart the
648 * loop just for safety. Or does this
649 * significantly reduce performance??
656 * A new RA might have made a deprecated address
659 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
663 /* expire prefix list */
664 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
666 * check prefix lifetime.
667 * since pltime is just for autoconf, pltime processing for
668 * prefix is not necessary.
670 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
671 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
674 * address expiration and prefix expiration are
675 * separate. NEVER perform in6_purgeaddr here.
684 * ia6 - deprecated/invalidated temporary address
687 regen_tmpaddr(struct in6_ifaddr *ia6)
691 struct in6_ifaddr *public_ifa6 = NULL;
693 ifp = ia6->ia_ifa.ifa_ifp;
695 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
696 struct in6_ifaddr *it6;
698 if (ifa->ifa_addr->sa_family != AF_INET6)
701 it6 = (struct in6_ifaddr *)ifa;
703 /* ignore no autoconf addresses. */
704 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
707 /* ignore autoconf addresses with different prefixes. */
708 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
712 * Now we are looking at an autoconf address with the same
713 * prefix as ours. If the address is temporary and is still
714 * preferred, do not create another one. It would be rare, but
715 * could happen, for example, when we resume a laptop PC after
718 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
719 !IFA6_IS_DEPRECATED(it6)) {
725 * This is a public autoconf address that has the same prefix
726 * as ours. If it is preferred, keep it. We can't break the
727 * loop here, because there may be a still-preferred temporary
728 * address with the prefix.
730 if (!IFA6_IS_DEPRECATED(it6))
733 if (public_ifa6 != NULL)
734 ifa_ref(&public_ifa6->ia_ifa);
736 IF_ADDR_RUNLOCK(ifp);
738 if (public_ifa6 != NULL) {
741 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
742 ifa_free(&public_ifa6->ia_ifa);
743 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
744 " tmp addr,errno=%d\n", e);
747 ifa_free(&public_ifa6->ia_ifa);
755 * Nuke neighbor cache/prefix/default router management table, right before
759 nd6_purge(struct ifnet *ifp)
761 struct nd_defrouter *dr, *ndr;
762 struct nd_prefix *pr, *npr;
765 * Nuke default router list entries toward ifp.
766 * We defer removal of default router list entries that is installed
767 * in the routing table, in order to keep additional side effects as
770 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
778 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
786 /* Nuke prefix list entries toward ifp */
787 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
788 if (pr->ndpr_ifp == ifp) {
790 * Because if_detach() does *not* release prefixes
791 * while purging addresses the reference count will
792 * still be above zero. We therefore reset it to
793 * make sure that the prefix really gets purged.
798 * Previously, pr->ndpr_addr is removed as well,
799 * but I strongly believe we don't have to do it.
800 * nd6_purge() is only called from in6_ifdetach(),
801 * which removes all the associated interface addresses
803 * (jinmei@kame.net 20010129)
809 /* cancel default outgoing interface setting */
810 if (V_nd6_defifindex == ifp->if_index)
811 nd6_setdefaultiface(0);
813 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
814 /* Refresh default router list. */
819 * We do not nuke the neighbor cache entries here any more
820 * because the neighbor cache is kept in if_afdata[AF_INET6].
821 * nd6_purge() is invoked by in6_ifdetach() which is called
822 * from if_detach() where everything gets purged. So let
823 * in6_domifdetach() do the actual L2 table purging work.
828 * the caller acquires and releases the lock on the lltbls
829 * Returns the llentry locked
832 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
834 struct sockaddr_in6 sin6;
838 bzero(&sin6, sizeof(sin6));
839 sin6.sin6_len = sizeof(struct sockaddr_in6);
840 sin6.sin6_family = AF_INET6;
841 sin6.sin6_addr = *addr6;
843 IF_AFDATA_LOCK_ASSERT(ifp);
846 if (flags & ND6_CREATE)
847 llflags |= LLE_CREATE;
848 if (flags & ND6_EXCLUSIVE)
849 llflags |= LLE_EXCLUSIVE;
851 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
852 if ((ln != NULL) && (llflags & LLE_CREATE))
853 ln->ln_state = ND6_LLINFO_NOSTATE;
859 * Test whether a given IPv6 address is a neighbor or not, ignoring
860 * the actual neighbor cache. The neighbor cache is ignored in order
861 * to not reenter the routing code from within itself.
864 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
866 struct nd_prefix *pr;
867 struct ifaddr *dstaddr;
870 * A link-local address is always a neighbor.
871 * XXX: a link does not necessarily specify a single interface.
873 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
874 struct sockaddr_in6 sin6_copy;
878 * We need sin6_copy since sa6_recoverscope() may modify the
882 if (sa6_recoverscope(&sin6_copy))
883 return (0); /* XXX: should be impossible */
884 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
886 if (sin6_copy.sin6_scope_id == zone)
893 * If the address matches one of our addresses,
894 * it should be a neighbor.
895 * If the address matches one of our on-link prefixes, it should be a
898 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
899 if (pr->ndpr_ifp != ifp)
902 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
905 /* Always use the default FIB here. */
906 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
907 0, 0, RT_DEFAULT_FIB);
911 * This is the case where multiple interfaces
912 * have the same prefix, but only one is installed
913 * into the routing table and that prefix entry
914 * is not the one being examined here. In the case
915 * where RADIX_MPATH is enabled, multiple route
916 * entries (of the same rt_key value) will be
917 * installed because the interface addresses all
920 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
921 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
928 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
929 &addr->sin6_addr, &pr->ndpr_mask))
934 * If the address is assigned on the node of the other side of
935 * a p2p interface, the address should be a neighbor.
937 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
938 if (dstaddr != NULL) {
939 if (dstaddr->ifa_ifp == ifp) {
947 * If the default router list is empty, all addresses are regarded
948 * as on-link, and thus, as a neighbor.
950 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
951 TAILQ_EMPTY(&V_nd_defrouter) &&
952 V_nd6_defifindex == ifp->if_index) {
961 * Detect if a given IPv6 address identifies a neighbor on a given link.
962 * XXX: should take care of the destination of a p2p link?
965 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
970 IF_AFDATA_UNLOCK_ASSERT(ifp);
971 if (nd6_is_new_addr_neighbor(addr, ifp))
975 * Even if the address matches none of our addresses, it might be
976 * in the neighbor cache.
978 IF_AFDATA_RLOCK(ifp);
979 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
983 IF_AFDATA_RUNLOCK(ifp);
988 * Free an nd6 llinfo entry.
989 * Since the function would cause significant changes in the kernel, DO NOT
990 * make it global, unless you have a strong reason for the change, and are sure
991 * that the change is safe.
993 static struct llentry *
994 nd6_free(struct llentry *ln, int gc)
996 struct llentry *next;
997 struct nd_defrouter *dr;
1000 LLE_WLOCK_ASSERT(ln);
1003 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1004 * even though it is not harmful, it was not really necessary.
1008 nd6_llinfo_settimer_locked(ln, -1);
1010 ifp = ln->lle_tbl->llt_ifp;
1012 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1013 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1015 if (dr != NULL && dr->expire &&
1016 ln->ln_state == ND6_LLINFO_STALE && gc) {
1018 * If the reason for the deletion is just garbage
1019 * collection, and the neighbor is an active default
1020 * router, do not delete it. Instead, reset the GC
1021 * timer using the router's lifetime.
1022 * Simply deleting the entry would affect default
1023 * router selection, which is not necessarily a good
1024 * thing, especially when we're using router preference
1026 * XXX: the check for ln_state would be redundant,
1027 * but we intentionally keep it just in case.
1029 if (dr->expire > time_second)
1030 nd6_llinfo_settimer_locked(ln,
1031 (dr->expire - time_second) * hz);
1033 nd6_llinfo_settimer_locked(ln,
1034 (long)V_nd6_gctimer * hz);
1036 next = LIST_NEXT(ln, lle_next);
1044 * Unreachablity of a router might affect the default
1045 * router selection and on-link detection of advertised
1050 * Temporarily fake the state to choose a new default
1051 * router and to perform on-link determination of
1052 * prefixes correctly.
1053 * Below the state will be set correctly,
1054 * or the entry itself will be deleted.
1056 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1059 if (ln->ln_router || dr) {
1062 * We need to unlock to avoid a LOR with rt6_flush() with the
1063 * rnh and for the calls to pfxlist_onlink_check() and
1064 * defrouter_select() in the block further down for calls
1065 * into nd6_lookup(). We still hold a ref.
1070 * rt6_flush must be called whether or not the neighbor
1071 * is in the Default Router List.
1072 * See a corresponding comment in nd6_na_input().
1074 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1079 * Since defrouter_select() does not affect the
1080 * on-link determination and MIP6 needs the check
1081 * before the default router selection, we perform
1084 pfxlist_onlink_check();
1087 * Refresh default router list.
1092 if (ln->ln_router || dr)
1097 * Before deleting the entry, remember the next entry as the
1098 * return value. We need this because pfxlist_onlink_check() above
1099 * might have freed other entries (particularly the old next entry) as
1100 * a side effect (XXX).
1102 next = LIST_NEXT(ln, lle_next);
1105 * Save to unlock. We still hold an extra reference and will not
1106 * free(9) in llentry_free() if someone else holds one as well.
1109 IF_AFDATA_LOCK(ifp);
1112 /* Guard against race with other llentry_free(). */
1113 if (ln->la_flags & LLE_LINKED) {
1117 LLE_FREE_LOCKED(ln);
1119 IF_AFDATA_UNLOCK(ifp);
1125 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1127 * XXX cost-effective methods?
1130 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1135 if ((dst6 == NULL) || (rt == NULL))
1139 IF_AFDATA_LOCK(ifp);
1140 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1141 IF_AFDATA_UNLOCK(ifp);
1145 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1149 * if we get upper-layer reachability confirmation many times,
1150 * it is possible we have false information.
1154 if (ln->ln_byhint > V_nd6_maxnudhint) {
1159 ln->ln_state = ND6_LLINFO_REACHABLE;
1160 if (!ND6_LLINFO_PERMANENT(ln)) {
1161 nd6_llinfo_settimer_locked(ln,
1162 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1170 * Rejuvenate this function for routing operations related
1174 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1176 struct sockaddr_in6 *gateway;
1177 struct nd_defrouter *dr;
1181 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1192 * Only indirect routes are interesting.
1194 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1197 * check for default route
1199 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1200 &SIN6(rt_key(rt))->sin6_addr)) {
1202 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1212 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1214 struct in6_drlist *drl = (struct in6_drlist *)data;
1215 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1216 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1217 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1218 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1219 struct nd_defrouter *dr;
1220 struct nd_prefix *pr;
1221 int i = 0, error = 0;
1224 case SIOCGDRLST_IN6:
1226 * obsolete API, use sysctl under net.inet6.icmp6
1228 bzero(drl, sizeof(*drl));
1229 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1232 drl->defrouter[i].rtaddr = dr->rtaddr;
1233 in6_clearscope(&drl->defrouter[i].rtaddr);
1235 drl->defrouter[i].flags = dr->flags;
1236 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1237 drl->defrouter[i].expire = dr->expire;
1238 drl->defrouter[i].if_index = dr->ifp->if_index;
1242 case SIOCGPRLST_IN6:
1244 * obsolete API, use sysctl under net.inet6.icmp6
1246 * XXX the structure in6_prlist was changed in backward-
1247 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1248 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1251 * XXX meaning of fields, especialy "raflags", is very
1252 * differnet between RA prefix list and RR/static prefix list.
1253 * how about separating ioctls into two?
1255 bzero(oprl, sizeof(*oprl));
1256 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1257 struct nd_pfxrouter *pfr;
1262 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1263 oprl->prefix[i].raflags = pr->ndpr_raf;
1264 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1265 oprl->prefix[i].vltime = pr->ndpr_vltime;
1266 oprl->prefix[i].pltime = pr->ndpr_pltime;
1267 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1268 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1269 oprl->prefix[i].expire = 0;
1273 /* XXX: we assume time_t is signed. */
1276 ((sizeof(maxexpire) * 8) - 1));
1277 if (pr->ndpr_vltime <
1278 maxexpire - pr->ndpr_lastupdate) {
1279 oprl->prefix[i].expire =
1280 pr->ndpr_lastupdate +
1283 oprl->prefix[i].expire = maxexpire;
1287 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1289 #define RTRADDR oprl->prefix[i].advrtr[j]
1290 RTRADDR = pfr->router->rtaddr;
1291 in6_clearscope(&RTRADDR);
1296 oprl->prefix[i].advrtrs = j;
1297 oprl->prefix[i].origin = PR_ORIG_RA;
1303 case OSIOCGIFINFO_IN6:
1305 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1306 bzero(&ND, sizeof(ND));
1307 ND.linkmtu = IN6_LINKMTU(ifp);
1308 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1309 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1310 ND.reachable = ND_IFINFO(ifp)->reachable;
1311 ND.retrans = ND_IFINFO(ifp)->retrans;
1312 ND.flags = ND_IFINFO(ifp)->flags;
1313 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1314 ND.chlim = ND_IFINFO(ifp)->chlim;
1316 case SIOCGIFINFO_IN6:
1317 ND = *ND_IFINFO(ifp);
1319 case SIOCSIFINFO_IN6:
1321 * used to change host variables from userland.
1322 * intented for a use on router to reflect RA configurations.
1324 /* 0 means 'unspecified' */
1325 if (ND.linkmtu != 0) {
1326 if (ND.linkmtu < IPV6_MMTU ||
1327 ND.linkmtu > IN6_LINKMTU(ifp)) {
1331 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1334 if (ND.basereachable != 0) {
1335 int obasereachable = ND_IFINFO(ifp)->basereachable;
1337 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1338 if (ND.basereachable != obasereachable)
1339 ND_IFINFO(ifp)->reachable =
1340 ND_COMPUTE_RTIME(ND.basereachable);
1342 if (ND.retrans != 0)
1343 ND_IFINFO(ifp)->retrans = ND.retrans;
1345 ND_IFINFO(ifp)->chlim = ND.chlim;
1347 case SIOCSIFINFO_FLAGS:
1350 struct in6_ifaddr *ia;
1352 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1353 !(ND.flags & ND6_IFF_IFDISABLED)) {
1354 /* ifdisabled 1->0 transision */
1357 * If the interface is marked as ND6_IFF_IFDISABLED and
1358 * has an link-local address with IN6_IFF_DUPLICATED,
1359 * do not clear ND6_IFF_IFDISABLED.
1360 * See RFC 4862, Section 5.4.5.
1362 int duplicated_linklocal = 0;
1365 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1366 if (ifa->ifa_addr->sa_family != AF_INET6)
1368 ia = (struct in6_ifaddr *)ifa;
1369 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1370 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1371 duplicated_linklocal = 1;
1375 IF_ADDR_RUNLOCK(ifp);
1377 if (duplicated_linklocal) {
1378 ND.flags |= ND6_IFF_IFDISABLED;
1379 log(LOG_ERR, "Cannot enable an interface"
1380 " with a link-local address marked"
1383 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1384 if (ifp->if_flags & IFF_UP)
1387 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1388 (ND.flags & ND6_IFF_IFDISABLED)) {
1389 /* ifdisabled 0->1 transision */
1390 /* Mark all IPv6 address as tentative. */
1392 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1394 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1395 if (ifa->ifa_addr->sa_family != AF_INET6)
1397 ia = (struct in6_ifaddr *)ifa;
1398 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1400 IF_ADDR_RUNLOCK(ifp);
1403 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1404 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1405 /* auto_linklocal 0->1 transision */
1407 /* If no link-local address on ifp, configure */
1408 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1409 in6_ifattach(ifp, NULL);
1410 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1411 ifp->if_flags & IFF_UP) {
1413 * When the IF already has
1414 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1415 * address is assigned, and IFF_UP, try to
1418 int haslinklocal = 0;
1421 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1422 if (ifa->ifa_addr->sa_family != AF_INET6)
1424 ia = (struct in6_ifaddr *)ifa;
1425 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1430 IF_ADDR_RUNLOCK(ifp);
1432 in6_ifattach(ifp, NULL);
1436 ND_IFINFO(ifp)->flags = ND.flags;
1439 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1440 /* sync kernel routing table with the default router list */
1444 case SIOCSPFXFLUSH_IN6:
1446 /* flush all the prefix advertised by routers */
1447 struct nd_prefix *pr, *next;
1449 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1450 struct in6_ifaddr *ia, *ia_next;
1452 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1455 /* do we really have to remove addresses as well? */
1456 /* XXXRW: in6_ifaddrhead locking. */
1457 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1459 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1462 if (ia->ia6_ndpr == pr)
1463 in6_purgeaddr(&ia->ia_ifa);
1469 case SIOCSRTRFLUSH_IN6:
1471 /* flush all the default routers */
1472 struct nd_defrouter *dr, *next;
1475 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1481 case SIOCGNBRINFO_IN6:
1484 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1486 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1489 IF_AFDATA_RLOCK(ifp);
1490 ln = nd6_lookup(&nb_addr, 0, ifp);
1491 IF_AFDATA_RUNLOCK(ifp);
1497 nbi->state = ln->ln_state;
1498 nbi->asked = ln->la_asked;
1499 nbi->isrouter = ln->ln_router;
1500 nbi->expire = ln->la_expire;
1504 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1505 ndif->ifindex = V_nd6_defifindex;
1507 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1508 return (nd6_setdefaultiface(ndif->ifindex));
1514 * Create neighbor cache entry and cache link-layer address,
1515 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1518 * code - type dependent information
1521 * The caller of this function already acquired the ndp
1522 * cache table lock because the cache entry is returned.
1525 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1526 int lladdrlen, int type, int code)
1528 struct llentry *ln = NULL;
1535 uint16_t router = 0;
1536 struct sockaddr_in6 sin6;
1537 struct mbuf *chain = NULL;
1538 int static_route = 0;
1540 IF_AFDATA_UNLOCK_ASSERT(ifp);
1542 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1543 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1545 /* nothing must be updated for unspecified address */
1546 if (IN6_IS_ADDR_UNSPECIFIED(from))
1550 * Validation about ifp->if_addrlen and lladdrlen must be done in
1553 * XXX If the link does not have link-layer adderss, what should
1554 * we do? (ifp->if_addrlen == 0)
1555 * Spec says nothing in sections for RA, RS and NA. There's small
1556 * description on it in NS section (RFC 2461 7.2.3).
1558 flags = lladdr ? ND6_EXCLUSIVE : 0;
1559 IF_AFDATA_LOCK(ifp);
1560 ln = nd6_lookup(from, flags, ifp);
1563 flags |= ND6_EXCLUSIVE;
1564 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1565 IF_AFDATA_UNLOCK(ifp);
1568 IF_AFDATA_UNLOCK(ifp);
1569 /* do nothing if static ndp is set */
1570 if (ln->la_flags & LLE_STATIC) {
1579 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1580 if (olladdr && lladdr) {
1581 llchange = bcmp(lladdr, &ln->ll_addr,
1587 * newentry olladdr lladdr llchange (*=record)
1590 * 0 n y -- (3) * STALE
1592 * 0 y y y (5) * STALE
1593 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1594 * 1 -- y -- (7) * STALE
1597 if (lladdr) { /* (3-5) and (7) */
1599 * Record source link-layer address
1600 * XXX is it dependent to ifp->if_type?
1602 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1603 ln->la_flags |= LLE_VALID;
1607 if ((!olladdr && lladdr != NULL) || /* (3) */
1608 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1610 newstate = ND6_LLINFO_STALE;
1611 } else /* (1-2,4) */
1615 if (lladdr == NULL) /* (6) */
1616 newstate = ND6_LLINFO_NOSTATE;
1618 newstate = ND6_LLINFO_STALE;
1623 * Update the state of the neighbor cache.
1625 ln->ln_state = newstate;
1627 if (ln->ln_state == ND6_LLINFO_STALE) {
1629 * XXX: since nd6_output() below will cause
1630 * state tansition to DELAY and reset the timer,
1631 * we must set the timer now, although it is actually
1634 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1637 struct mbuf *m_hold, *m_hold_next;
1640 * reset the la_hold in advance, to explicitly
1641 * prevent a la_hold lookup in nd6_output()
1642 * (wouldn't happen, though...)
1644 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1645 m_hold; m_hold = m_hold_next) {
1646 m_hold_next = m_hold->m_nextpkt;
1647 m_hold->m_nextpkt = NULL;
1650 * we assume ifp is not a p2p here, so
1651 * just set the 2nd argument as the
1654 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1657 * If we have mbufs in the chain we need to do
1658 * deferred transmit. Copy the address from the
1659 * llentry before dropping the lock down below.
1662 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1664 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1665 /* probe right away */
1666 nd6_llinfo_settimer_locked((void *)ln, 0);
1671 * ICMP6 type dependent behavior.
1673 * NS: clear IsRouter if new entry
1674 * RS: clear IsRouter
1675 * RA: set IsRouter if there's lladdr
1676 * redir: clear IsRouter if new entry
1679 * The spec says that we must set IsRouter in the following cases:
1680 * - If lladdr exist, set IsRouter. This means (1-5).
1681 * - If it is old entry (!newentry), set IsRouter. This means (7).
1682 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1683 * A quetion arises for (1) case. (1) case has no lladdr in the
1684 * neighbor cache, this is similar to (6).
1685 * This case is rare but we figured that we MUST NOT set IsRouter.
1687 * newentry olladdr lladdr llchange NS RS RA redir
1689 * 0 n n -- (1) c ? s
1690 * 0 y n -- (2) c s s
1691 * 0 n y -- (3) c s s
1694 * 1 -- n -- (6) c c c s
1695 * 1 -- y -- (7) c c s c s
1699 switch (type & 0xff) {
1700 case ND_NEIGHBOR_SOLICIT:
1702 * New entry must have is_router flag cleared.
1704 if (is_newentry) /* (6-7) */
1709 * If the icmp is a redirect to a better router, always set the
1710 * is_router flag. Otherwise, if the entry is newly created,
1711 * clear the flag. [RFC 2461, sec 8.3]
1713 if (code == ND_REDIRECT_ROUTER)
1715 else if (is_newentry) /* (6-7) */
1718 case ND_ROUTER_SOLICIT:
1720 * is_router flag must always be cleared.
1724 case ND_ROUTER_ADVERT:
1726 * Mark an entry with lladdr as a router.
1728 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1729 (is_newentry && lladdr)) { /* (7) */
1736 static_route = (ln->la_flags & LLE_STATIC);
1737 router = ln->ln_router;
1739 if (flags & ND6_EXCLUSIVE)
1747 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1750 * When the link-layer address of a router changes, select the
1751 * best router again. In particular, when the neighbor entry is newly
1752 * created, it might affect the selection policy.
1753 * Question: can we restrict the first condition to the "is_newentry"
1755 * XXX: when we hear an RA from a new router with the link-layer
1756 * address option, defrouter_select() is called twice, since
1757 * defrtrlist_update called the function as well. However, I believe
1758 * we can compromise the overhead, since it only happens the first
1760 * XXX: although defrouter_select() should not have a bad effect
1761 * for those are not autoconfigured hosts, we explicitly avoid such
1764 if (do_update && router &&
1765 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1767 * guaranteed recursion
1775 if (flags & ND6_EXCLUSIVE)
1786 nd6_slowtimo(void *arg)
1788 CURVNET_SET((struct vnet *) arg);
1789 struct nd_ifinfo *nd6if;
1792 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1793 nd6_slowtimo, curvnet);
1794 IFNET_RLOCK_NOSLEEP();
1795 TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1796 nd6if = ND_IFINFO(ifp);
1797 if (nd6if->basereachable && /* already initialized */
1798 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1800 * Since reachable time rarely changes by router
1801 * advertisements, we SHOULD insure that a new random
1802 * value gets recomputed at least once every few hours.
1805 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1806 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1809 IFNET_RUNLOCK_NOSLEEP();
1814 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1815 struct sockaddr_in6 *dst, struct rtentry *rt0)
1818 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1823 * Note that I'm not enforcing any global serialization
1824 * lle state or asked changes here as the logic is too
1825 * complicated to avoid having to always acquire an exclusive
1830 #define senderr(e) { error = (e); goto bad;}
1833 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1834 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1835 struct mbuf **chain)
1837 struct mbuf *m = m0;
1839 struct llentry *ln = lle;
1840 struct ip6_hdr *ip6;
1848 LLE_WLOCK_ASSERT(lle);
1850 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1853 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1856 if (nd6_need_cache(ifp) == 0)
1860 * next hop determination. This routine is derived from ether_output.
1864 * Address resolution or Neighbor Unreachability Detection
1866 * At this point, the destination of the packet must be a unicast
1867 * or an anycast address(i.e. not a multicast).
1870 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1873 IF_AFDATA_LOCK(ifp);
1874 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1875 IF_AFDATA_UNLOCK(ifp);
1876 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1878 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1879 * the condition below is not very efficient. But we believe
1880 * it is tolerable, because this should be a rare case.
1882 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1883 IF_AFDATA_LOCK(ifp);
1884 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1885 IF_AFDATA_UNLOCK(ifp);
1889 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1890 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1891 char ip6buf[INET6_ADDRSTRLEN];
1893 "nd6_output: can't allocate llinfo for %s "
1895 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1896 senderr(EIO); /* XXX: good error? */
1898 goto sendpkt; /* send anyway */
1901 /* We don't have to do link-layer address resolution on a p2p link. */
1902 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1903 ln->ln_state < ND6_LLINFO_REACHABLE) {
1904 if ((flags & LLE_EXCLUSIVE) == 0) {
1905 flags |= LLE_EXCLUSIVE;
1908 ln->ln_state = ND6_LLINFO_STALE;
1909 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1913 * The first time we send a packet to a neighbor whose entry is
1914 * STALE, we have to change the state to DELAY and a sets a timer to
1915 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1916 * neighbor unreachability detection on expiration.
1919 if (ln->ln_state == ND6_LLINFO_STALE) {
1920 if ((flags & LLE_EXCLUSIVE) == 0) {
1921 flags |= LLE_EXCLUSIVE;
1926 ln->ln_state = ND6_LLINFO_DELAY;
1927 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1931 * If the neighbor cache entry has a state other than INCOMPLETE
1932 * (i.e. its link-layer address is already resolved), just
1935 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1939 * There is a neighbor cache entry, but no ethernet address
1940 * response yet. Append this latest packet to the end of the
1941 * packet queue in the mbuf, unless the number of the packet
1942 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1943 * the oldest packet in the queue will be removed.
1945 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1946 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1948 if ((flags & LLE_EXCLUSIVE) == 0) {
1949 flags |= LLE_EXCLUSIVE;
1954 LLE_WLOCK_ASSERT(ln);
1957 struct mbuf *m_hold;
1961 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1963 if (m_hold->m_nextpkt == NULL) {
1964 m_hold->m_nextpkt = m;
1968 while (i >= V_nd6_maxqueuelen) {
1969 m_hold = ln->la_hold;
1970 ln->la_hold = ln->la_hold->m_nextpkt;
1979 * If there has been no NS for the neighbor after entering the
1980 * INCOMPLETE state, send the first solicitation.
1982 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1985 nd6_llinfo_settimer_locked(ln,
1986 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1988 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1989 if (lle != NULL && ln == lle)
1992 } else if (lle == NULL || ln != lle) {
1994 * We did the lookup (no lle arg) so we
1995 * need to do the unlock here.
2003 /* discard the packet if IPv6 operation is disabled on the interface */
2004 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2005 error = ENETDOWN; /* better error? */
2009 * ln is valid and the caller did not pass in
2012 if ((ln != NULL) && (lle == NULL)) {
2013 if (flags & LLE_EXCLUSIVE)
2020 mac_netinet6_nd6_send(ifp, m);
2024 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2025 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2026 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2027 * to be diverted to user space. When re-injected into the kernel,
2028 * send_output() will directly dispatch them to the outgoing interface.
2030 if (send_sendso_input_hook != NULL) {
2031 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2033 ip6 = mtod(m, struct ip6_hdr *);
2034 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2035 /* Use the SEND socket */
2036 error = send_sendso_input_hook(m, ifp, SND_OUT,
2038 /* -1 == no app on SEND socket */
2039 if (error == 0 || error != -1)
2045 * We were passed in a pointer to an lle with the lock held
2046 * this means that we can't call if_output as we will
2047 * recurse on the lle lock - so what we do is we create
2048 * a list of mbufs to send and transmit them in the caller
2049 * after the lock is dropped
2058 * append mbuf to end of deferred chain
2061 while (mb->m_nextpkt != NULL)
2067 /* Reset layer specific mbuf flags to avoid confusing lower layers. */
2068 m->m_flags &= ~(M_PROTOFLAGS);
2069 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2070 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2073 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2078 * ln is valid and the caller did not pass in
2081 if ((ln != NULL) && (lle == NULL)) {
2082 if (flags & LLE_EXCLUSIVE)
2095 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2096 struct sockaddr_in6 *dst, struct route *ro)
2098 struct mbuf *m, *m_head;
2099 struct ifnet *outifp;
2103 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2110 m_head = m_head->m_nextpkt;
2111 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2116 * note that intermediate errors are blindly ignored - but this is
2117 * the same convention as used with nd6_output when called by
2125 nd6_need_cache(struct ifnet *ifp)
2128 * XXX: we currently do not make neighbor cache on any interface
2129 * other than ARCnet, Ethernet, FDDI and GIF.
2132 * - unidirectional tunnels needs no ND
2134 switch (ifp->if_type) {
2142 #ifdef IFT_IEEE80211
2145 case IFT_INFINIBAND:
2146 case IFT_GIF: /* XXX need more cases? */
2150 case IFT_PROPVIRTUAL:
2158 * the callers of this function need to be re-worked to drop
2159 * the lle lock, drop here for now
2162 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2163 struct sockaddr *dst, u_char *desten, struct llentry **lle)
2168 IF_AFDATA_UNLOCK_ASSERT(ifp);
2169 if (m->m_flags & M_MCAST) {
2172 switch (ifp->if_type) {
2178 #ifdef IFT_IEEE80211
2183 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2188 * netbsd can use if_broadcastaddr, but we don't do so
2189 * to reduce # of ifdef.
2191 for (i = 0; i < ifp->if_addrlen; i++)
2199 return (EAFNOSUPPORT);
2205 * the entry should have been created in nd6_store_lladdr
2207 IF_AFDATA_RLOCK(ifp);
2208 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2209 IF_AFDATA_RUNLOCK(ifp);
2210 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2213 /* this could happen, if we could not allocate memory */
2218 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2222 * A *small* use after free race exists here
2228 clear_llinfo_pqueue(struct llentry *ln)
2230 struct mbuf *m_hold, *m_hold_next;
2232 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2233 m_hold_next = m_hold->m_nextpkt;
2241 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2242 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2244 SYSCTL_DECL(_net_inet6_icmp6);
2246 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2247 CTLFLAG_RD, nd6_sysctl_drlist, "");
2248 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2249 CTLFLAG_RD, nd6_sysctl_prlist, "");
2250 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2251 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2254 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2256 struct in6_defrouter d;
2257 struct nd_defrouter *dr;
2263 bzero(&d, sizeof(d));
2264 d.rtaddr.sin6_family = AF_INET6;
2265 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2270 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2271 d.rtaddr.sin6_addr = dr->rtaddr;
2272 error = sa6_recoverscope(&d.rtaddr);
2275 d.flags = dr->flags;
2276 d.rtlifetime = dr->rtlifetime;
2277 d.expire = dr->expire;
2278 d.if_index = dr->ifp->if_index;
2279 error = SYSCTL_OUT(req, &d, sizeof(d));
2287 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2289 struct in6_prefix p;
2290 struct sockaddr_in6 s6;
2291 struct nd_prefix *pr;
2292 struct nd_pfxrouter *pfr;
2295 char ip6buf[INET6_ADDRSTRLEN];
2300 bzero(&p, sizeof(p));
2301 p.origin = PR_ORIG_RA;
2302 bzero(&s6, sizeof(s6));
2303 s6.sin6_family = AF_INET6;
2304 s6.sin6_len = sizeof(s6);
2309 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2310 p.prefix = pr->ndpr_prefix;
2311 if (sa6_recoverscope(&p.prefix)) {
2312 log(LOG_ERR, "scope error in prefix list (%s)\n",
2313 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2314 /* XXX: press on... */
2316 p.raflags = pr->ndpr_raf;
2317 p.prefixlen = pr->ndpr_plen;
2318 p.vltime = pr->ndpr_vltime;
2319 p.pltime = pr->ndpr_pltime;
2320 p.if_index = pr->ndpr_ifp->if_index;
2321 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2324 /* XXX: we assume time_t is signed. */
2326 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2327 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2328 p.expire = pr->ndpr_lastupdate +
2331 p.expire = maxexpire;
2333 p.refcnt = pr->ndpr_refcnt;
2334 p.flags = pr->ndpr_stateflags;
2336 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2338 error = SYSCTL_OUT(req, &p, sizeof(p));
2341 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2342 s6.sin6_addr = pfr->router->rtaddr;
2343 if (sa6_recoverscope(&s6))
2345 "scope error in prefix list (%s)\n",
2346 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2347 error = SYSCTL_OUT(req, &s6, sizeof(s6));