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"
37 #include "opt_kdtrace.h"
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/callout.h>
42 #include <sys/malloc.h>
44 #include <sys/socket.h>
45 #include <sys/sockio.h>
47 #include <sys/kernel.h>
48 #include <sys/protosw.h>
49 #include <sys/errno.h>
50 #include <sys/syslog.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
55 #include <sys/sysctl.h>
58 #include <net/if_arc.h>
59 #include <net/if_dl.h>
60 #include <net/if_types.h>
61 #include <net/iso88025.h>
63 #include <net/route.h>
66 #include <netinet/in.h>
67 #include <netinet/in_kdtrace.h>
68 #include <net/if_llatbl.h>
69 #define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le))
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/send.h>
80 #include <sys/limits.h>
82 #include <security/mac/mac_framework.h>
84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
87 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
90 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */
91 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */
92 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */
93 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */
94 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for
96 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
99 /* preventing too many loops in ND option parsing */
100 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
102 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper
104 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
106 #define V_nd6_maxndopt VNET(nd6_maxndopt)
107 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
110 VNET_DEFINE(int, nd6_debug) = 1;
112 VNET_DEFINE(int, nd6_debug) = 0;
117 static int nd6_inuse, nd6_allocated;
120 VNET_DEFINE(struct nd_drhead, nd_defrouter);
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
123 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
124 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
126 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
128 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *,
130 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
131 static void nd6_slowtimo(void *);
132 static int regen_tmpaddr(struct in6_ifaddr *);
133 static struct llentry *nd6_free(struct llentry *, int);
134 static void nd6_llinfo_timer(void *);
135 static void clear_llinfo_pqueue(struct llentry *);
137 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
138 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
140 VNET_DEFINE(struct callout, nd6_timer_ch);
146 LIST_INIT(&V_nd_prefix);
148 /* initialization of the default router list */
149 TAILQ_INIT(&V_nd_defrouter);
152 callout_init(&V_nd6_slowtimo_ch, 0);
153 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
154 nd6_slowtimo, curvnet);
162 callout_drain(&V_nd6_slowtimo_ch);
163 callout_drain(&V_nd6_timer_ch);
168 nd6_ifattach(struct ifnet *ifp)
170 struct nd_ifinfo *nd;
172 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
175 nd->chlim = IPV6_DEFHLIM;
176 nd->basereachable = REACHABLE_TIME;
177 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
178 nd->retrans = RETRANS_TIMER;
180 nd->flags = ND6_IFF_PERFORMNUD;
182 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
183 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
184 * default regardless of the V_ip6_auto_linklocal configuration to
185 * give a reasonable default behavior.
187 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
188 (ifp->if_flags & IFF_LOOPBACK))
189 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
191 * A loopback interface does not need to accept RTADV.
192 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
193 * default regardless of the V_ip6_accept_rtadv configuration to
194 * prevent the interface from accepting RA messages arrived
195 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
197 if (V_ip6_accept_rtadv &&
198 !(ifp->if_flags & IFF_LOOPBACK) &&
199 (ifp->if_type != IFT_BRIDGE))
200 nd->flags |= ND6_IFF_ACCEPT_RTADV;
201 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
202 nd->flags |= ND6_IFF_NO_RADR;
204 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
205 nd6_setmtu0(ifp, nd);
211 nd6_ifdetach(struct nd_ifinfo *nd)
218 * Reset ND level link MTU. This function is called when the physical MTU
219 * changes, which means we might have to adjust the ND level MTU.
222 nd6_setmtu(struct ifnet *ifp)
225 nd6_setmtu0(ifp, ND_IFINFO(ifp));
228 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
230 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
234 omaxmtu = ndi->maxmtu;
236 switch (ifp->if_type) {
238 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
241 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
244 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
247 ndi->maxmtu = ifp->if_mtu;
252 * Decreasing the interface MTU under IPV6 minimum MTU may cause
253 * undesirable situation. We thus notify the operator of the change
254 * explicitly. The check for omaxmtu is necessary to restrict the
255 * log to the case of changing the MTU, not initializing it.
257 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
258 log(LOG_NOTICE, "nd6_setmtu0: "
259 "new link MTU on %s (%lu) is too small for IPv6\n",
260 if_name(ifp), (unsigned long)ndi->maxmtu);
263 if (ndi->maxmtu > V_in6_maxmtu)
264 in6_setmaxmtu(); /* check all interfaces just in case */
269 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
272 bzero(ndopts, sizeof(*ndopts));
273 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
275 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
278 ndopts->nd_opts_done = 1;
279 ndopts->nd_opts_search = NULL;
284 * Take one ND option.
287 nd6_option(union nd_opts *ndopts)
289 struct nd_opt_hdr *nd_opt;
292 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
293 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
295 if (ndopts->nd_opts_search == NULL)
297 if (ndopts->nd_opts_done)
300 nd_opt = ndopts->nd_opts_search;
302 /* make sure nd_opt_len is inside the buffer */
303 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
304 bzero(ndopts, sizeof(*ndopts));
308 olen = nd_opt->nd_opt_len << 3;
311 * Message validation requires that all included
312 * options have a length that is greater than zero.
314 bzero(ndopts, sizeof(*ndopts));
318 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
319 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
320 /* option overruns the end of buffer, invalid */
321 bzero(ndopts, sizeof(*ndopts));
323 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
324 /* reached the end of options chain */
325 ndopts->nd_opts_done = 1;
326 ndopts->nd_opts_search = NULL;
332 * Parse multiple ND options.
333 * This function is much easier to use, for ND routines that do not need
334 * multiple options of the same type.
337 nd6_options(union nd_opts *ndopts)
339 struct nd_opt_hdr *nd_opt;
342 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
343 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
345 if (ndopts->nd_opts_search == NULL)
349 nd_opt = nd6_option(ndopts);
350 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
352 * Message validation requires that all included
353 * options have a length that is greater than zero.
355 ICMP6STAT_INC(icp6s_nd_badopt);
356 bzero(ndopts, sizeof(*ndopts));
363 switch (nd_opt->nd_opt_type) {
364 case ND_OPT_SOURCE_LINKADDR:
365 case ND_OPT_TARGET_LINKADDR:
367 case ND_OPT_REDIRECTED_HEADER:
368 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
370 "duplicated ND6 option found (type=%d)\n",
371 nd_opt->nd_opt_type));
374 ndopts->nd_opt_array[nd_opt->nd_opt_type]
378 case ND_OPT_PREFIX_INFORMATION:
379 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
380 ndopts->nd_opt_array[nd_opt->nd_opt_type]
383 ndopts->nd_opts_pi_end =
384 (struct nd_opt_prefix_info *)nd_opt;
386 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
387 case ND_OPT_RDNSS: /* RFC 6106 */
388 case ND_OPT_DNSSL: /* RFC 6106 */
390 * Silently ignore options we know and do not care about
396 * Unknown options must be silently ignored,
397 * to accomodate future extension to the protocol.
400 "nd6_options: unsupported option %d - "
401 "option ignored\n", nd_opt->nd_opt_type));
406 if (i > V_nd6_maxndopt) {
407 ICMP6STAT_INC(icp6s_nd_toomanyopt);
408 nd6log((LOG_INFO, "too many loop in nd opt\n"));
412 if (ndopts->nd_opts_done)
420 * ND6 timer routine to handle ND6 entries
423 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
427 LLE_WLOCK_ASSERT(ln);
432 canceled = callout_stop(&ln->ln_timer_ch);
434 ln->la_expire = time_uptime + tick / hz;
436 if (tick > INT_MAX) {
437 ln->ln_ntick = tick - INT_MAX;
438 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
439 nd6_llinfo_timer, ln);
442 canceled = callout_reset(&ln->ln_timer_ch, tick,
443 nd6_llinfo_timer, ln);
451 nd6_llinfo_settimer(struct llentry *ln, long tick)
455 nd6_llinfo_settimer_locked(ln, tick);
460 nd6_llinfo_timer(void *arg)
463 struct in6_addr *dst;
465 struct nd_ifinfo *ndi = NULL;
467 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
468 ln = (struct llentry *)arg;
470 if (callout_pending(&ln->la_timer)) {
472 * Here we are a bit odd here in the treatment of
473 * active/pending. If the pending bit is set, it got
474 * rescheduled before I ran. The active
475 * bit we ignore, since if it was stopped
476 * in ll_tablefree() and was currently running
477 * it would have return 0 so the code would
478 * not have deleted it since the callout could
479 * not be stopped so we want to go through
480 * with the delete here now. If the callout
481 * was restarted, the pending bit will be back on and
482 * we just want to bail since the callout_reset would
483 * return 1 and our reference would have been removed
484 * by nd6_llinfo_settimer_locked above since canceled
490 ifp = ln->lle_tbl->llt_ifp;
491 CURVNET_SET(ifp->if_vnet);
493 if (ln->ln_ntick > 0) {
494 if (ln->ln_ntick > INT_MAX) {
495 ln->ln_ntick -= INT_MAX;
496 nd6_llinfo_settimer_locked(ln, INT_MAX);
499 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
504 ndi = ND_IFINFO(ifp);
505 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
506 if (ln->la_flags & LLE_STATIC) {
510 if (ln->la_flags & LLE_DELETED) {
511 (void)nd6_free(ln, 0);
516 switch (ln->ln_state) {
517 case ND6_LLINFO_INCOMPLETE:
518 if (ln->la_asked < V_nd6_mmaxtries) {
520 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
522 nd6_ns_output(ifp, NULL, dst, ln, 0);
525 struct mbuf *m = ln->la_hold;
530 * assuming every packet in la_hold has the
531 * same IP header. Send error after unlock.
536 clear_llinfo_pqueue(ln);
538 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
539 (void)nd6_free(ln, 0);
542 icmp6_error2(m, ICMP6_DST_UNREACH,
543 ICMP6_DST_UNREACH_ADDR, 0, ifp);
546 case ND6_LLINFO_REACHABLE:
547 if (!ND6_LLINFO_PERMANENT(ln)) {
548 ln->ln_state = ND6_LLINFO_STALE;
549 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
553 case ND6_LLINFO_STALE:
554 /* Garbage Collection(RFC 2461 5.3) */
555 if (!ND6_LLINFO_PERMANENT(ln)) {
556 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
557 (void)nd6_free(ln, 1);
562 case ND6_LLINFO_DELAY:
563 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
566 ln->ln_state = ND6_LLINFO_PROBE;
567 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
569 nd6_ns_output(ifp, dst, dst, ln, 0);
572 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
573 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
576 case ND6_LLINFO_PROBE:
577 if (ln->la_asked < V_nd6_umaxtries) {
579 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
581 nd6_ns_output(ifp, dst, dst, ln, 0);
584 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
585 (void)nd6_free(ln, 0);
590 panic("%s: paths in a dark night can be confusing: %d",
591 __func__, ln->ln_state);
601 * ND6 timer routine to expire default route list and prefix list
606 CURVNET_SET((struct vnet *) arg);
607 struct nd_defrouter *dr, *ndr;
608 struct nd_prefix *pr, *npr;
609 struct in6_ifaddr *ia6, *nia6;
611 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
614 /* expire default router list */
615 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
616 if (dr->expire && dr->expire < time_uptime)
621 * expire interface addresses.
622 * in the past the loop was inside prefix expiry processing.
623 * However, from a stricter speci-confrmance standpoint, we should
624 * rather separate address lifetimes and prefix lifetimes.
626 * XXXRW: in6_ifaddrhead locking.
629 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
630 /* check address lifetime */
631 if (IFA6_IS_INVALID(ia6)) {
635 * If the expiring address is temporary, try
636 * regenerating a new one. This would be useful when
637 * we suspended a laptop PC, then turned it on after a
638 * period that could invalidate all temporary
639 * addresses. Although we may have to restart the
640 * loop (see below), it must be after purging the
641 * address. Otherwise, we'd see an infinite loop of
644 if (V_ip6_use_tempaddr &&
645 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
646 if (regen_tmpaddr(ia6) == 0)
650 in6_purgeaddr(&ia6->ia_ifa);
653 goto addrloop; /* XXX: see below */
654 } else if (IFA6_IS_DEPRECATED(ia6)) {
655 int oldflags = ia6->ia6_flags;
657 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
660 * If a temporary address has just become deprecated,
661 * regenerate a new one if possible.
663 if (V_ip6_use_tempaddr &&
664 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
665 (oldflags & IN6_IFF_DEPRECATED) == 0) {
667 if (regen_tmpaddr(ia6) == 0) {
669 * A new temporary address is
671 * XXX: this means the address chain
672 * has changed while we are still in
673 * the loop. Although the change
674 * would not cause disaster (because
675 * it's not a deletion, but an
676 * addition,) we'd rather restart the
677 * loop just for safety. Or does this
678 * significantly reduce performance??
685 * A new RA might have made a deprecated address
688 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
692 /* expire prefix list */
693 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
695 * check prefix lifetime.
696 * since pltime is just for autoconf, pltime processing for
697 * prefix is not necessary.
699 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
700 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
703 * address expiration and prefix expiration are
704 * separate. NEVER perform in6_purgeaddr here.
713 * ia6 - deprecated/invalidated temporary address
716 regen_tmpaddr(struct in6_ifaddr *ia6)
720 struct in6_ifaddr *public_ifa6 = NULL;
722 ifp = ia6->ia_ifa.ifa_ifp;
724 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
725 struct in6_ifaddr *it6;
727 if (ifa->ifa_addr->sa_family != AF_INET6)
730 it6 = (struct in6_ifaddr *)ifa;
732 /* ignore no autoconf addresses. */
733 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
736 /* ignore autoconf addresses with different prefixes. */
737 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
741 * Now we are looking at an autoconf address with the same
742 * prefix as ours. If the address is temporary and is still
743 * preferred, do not create another one. It would be rare, but
744 * could happen, for example, when we resume a laptop PC after
747 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
748 !IFA6_IS_DEPRECATED(it6)) {
754 * This is a public autoconf address that has the same prefix
755 * as ours. If it is preferred, keep it. We can't break the
756 * loop here, because there may be a still-preferred temporary
757 * address with the prefix.
759 if (!IFA6_IS_DEPRECATED(it6))
762 if (public_ifa6 != NULL)
763 ifa_ref(&public_ifa6->ia_ifa);
765 IF_ADDR_RUNLOCK(ifp);
767 if (public_ifa6 != NULL) {
770 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
771 ifa_free(&public_ifa6->ia_ifa);
772 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
773 " tmp addr,errno=%d\n", e);
776 ifa_free(&public_ifa6->ia_ifa);
784 * Nuke neighbor cache/prefix/default router management table, right before
788 nd6_purge(struct ifnet *ifp)
790 struct nd_defrouter *dr, *ndr;
791 struct nd_prefix *pr, *npr;
794 * Nuke default router list entries toward ifp.
795 * We defer removal of default router list entries that is installed
796 * in the routing table, in order to keep additional side effects as
799 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
807 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
815 /* Nuke prefix list entries toward ifp */
816 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
817 if (pr->ndpr_ifp == ifp) {
819 * Because if_detach() does *not* release prefixes
820 * while purging addresses the reference count will
821 * still be above zero. We therefore reset it to
822 * make sure that the prefix really gets purged.
827 * Previously, pr->ndpr_addr is removed as well,
828 * but I strongly believe we don't have to do it.
829 * nd6_purge() is only called from in6_ifdetach(),
830 * which removes all the associated interface addresses
832 * (jinmei@kame.net 20010129)
838 /* cancel default outgoing interface setting */
839 if (V_nd6_defifindex == ifp->if_index)
840 nd6_setdefaultiface(0);
842 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
843 /* Refresh default router list. */
848 * We do not nuke the neighbor cache entries here any more
849 * because the neighbor cache is kept in if_afdata[AF_INET6].
850 * nd6_purge() is invoked by in6_ifdetach() which is called
851 * from if_detach() where everything gets purged. So let
852 * in6_domifdetach() do the actual L2 table purging work.
857 * the caller acquires and releases the lock on the lltbls
858 * Returns the llentry locked
861 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
863 struct sockaddr_in6 sin6;
867 bzero(&sin6, sizeof(sin6));
868 sin6.sin6_len = sizeof(struct sockaddr_in6);
869 sin6.sin6_family = AF_INET6;
870 sin6.sin6_addr = *addr6;
872 IF_AFDATA_LOCK_ASSERT(ifp);
875 if (flags & ND6_CREATE)
876 llflags |= LLE_CREATE;
877 if (flags & ND6_EXCLUSIVE)
878 llflags |= LLE_EXCLUSIVE;
880 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
881 if ((ln != NULL) && (llflags & LLE_CREATE))
882 ln->ln_state = ND6_LLINFO_NOSTATE;
888 * Test whether a given IPv6 address is a neighbor or not, ignoring
889 * the actual neighbor cache. The neighbor cache is ignored in order
890 * to not reenter the routing code from within itself.
893 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
895 struct nd_prefix *pr;
896 struct ifaddr *dstaddr;
899 * A link-local address is always a neighbor.
900 * XXX: a link does not necessarily specify a single interface.
902 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
903 struct sockaddr_in6 sin6_copy;
907 * We need sin6_copy since sa6_recoverscope() may modify the
911 if (sa6_recoverscope(&sin6_copy))
912 return (0); /* XXX: should be impossible */
913 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
915 if (sin6_copy.sin6_scope_id == zone)
922 * If the address matches one of our addresses,
923 * it should be a neighbor.
924 * If the address matches one of our on-link prefixes, it should be a
927 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
928 if (pr->ndpr_ifp != ifp)
931 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
934 /* Always use the default FIB here. */
935 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
936 0, 0, RT_DEFAULT_FIB);
940 * This is the case where multiple interfaces
941 * have the same prefix, but only one is installed
942 * into the routing table and that prefix entry
943 * is not the one being examined here. In the case
944 * where RADIX_MPATH is enabled, multiple route
945 * entries (of the same rt_key value) will be
946 * installed because the interface addresses all
949 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
950 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
957 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
958 &addr->sin6_addr, &pr->ndpr_mask))
963 * If the address is assigned on the node of the other side of
964 * a p2p interface, the address should be a neighbor.
966 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
967 if (dstaddr != NULL) {
968 if (dstaddr->ifa_ifp == ifp) {
976 * If the default router list is empty, all addresses are regarded
977 * as on-link, and thus, as a neighbor.
979 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
980 TAILQ_EMPTY(&V_nd_defrouter) &&
981 V_nd6_defifindex == ifp->if_index) {
990 * Detect if a given IPv6 address identifies a neighbor on a given link.
991 * XXX: should take care of the destination of a p2p link?
994 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
999 IF_AFDATA_UNLOCK_ASSERT(ifp);
1000 if (nd6_is_new_addr_neighbor(addr, ifp))
1004 * Even if the address matches none of our addresses, it might be
1005 * in the neighbor cache.
1007 IF_AFDATA_RLOCK(ifp);
1008 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1012 IF_AFDATA_RUNLOCK(ifp);
1017 * Free an nd6 llinfo entry.
1018 * Since the function would cause significant changes in the kernel, DO NOT
1019 * make it global, unless you have a strong reason for the change, and are sure
1020 * that the change is safe.
1022 static struct llentry *
1023 nd6_free(struct llentry *ln, int gc)
1025 struct llentry *next;
1026 struct nd_defrouter *dr;
1029 LLE_WLOCK_ASSERT(ln);
1032 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1033 * even though it is not harmful, it was not really necessary.
1037 nd6_llinfo_settimer_locked(ln, -1);
1039 ifp = ln->lle_tbl->llt_ifp;
1041 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1042 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1044 if (dr != NULL && dr->expire &&
1045 ln->ln_state == ND6_LLINFO_STALE && gc) {
1047 * If the reason for the deletion is just garbage
1048 * collection, and the neighbor is an active default
1049 * router, do not delete it. Instead, reset the GC
1050 * timer using the router's lifetime.
1051 * Simply deleting the entry would affect default
1052 * router selection, which is not necessarily a good
1053 * thing, especially when we're using router preference
1055 * XXX: the check for ln_state would be redundant,
1056 * but we intentionally keep it just in case.
1058 if (dr->expire > time_uptime)
1059 nd6_llinfo_settimer_locked(ln,
1060 (dr->expire - time_uptime) * hz);
1062 nd6_llinfo_settimer_locked(ln,
1063 (long)V_nd6_gctimer * hz);
1065 next = LIST_NEXT(ln, lle_next);
1073 * Unreachablity of a router might affect the default
1074 * router selection and on-link detection of advertised
1079 * Temporarily fake the state to choose a new default
1080 * router and to perform on-link determination of
1081 * prefixes correctly.
1082 * Below the state will be set correctly,
1083 * or the entry itself will be deleted.
1085 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1088 if (ln->ln_router || dr) {
1091 * We need to unlock to avoid a LOR with rt6_flush() with the
1092 * rnh and for the calls to pfxlist_onlink_check() and
1093 * defrouter_select() in the block further down for calls
1094 * into nd6_lookup(). We still hold a ref.
1099 * rt6_flush must be called whether or not the neighbor
1100 * is in the Default Router List.
1101 * See a corresponding comment in nd6_na_input().
1103 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1108 * Since defrouter_select() does not affect the
1109 * on-link determination and MIP6 needs the check
1110 * before the default router selection, we perform
1113 pfxlist_onlink_check();
1116 * Refresh default router list.
1121 if (ln->ln_router || dr)
1126 * Before deleting the entry, remember the next entry as the
1127 * return value. We need this because pfxlist_onlink_check() above
1128 * might have freed other entries (particularly the old next entry) as
1129 * a side effect (XXX).
1131 next = LIST_NEXT(ln, lle_next);
1134 * Save to unlock. We still hold an extra reference and will not
1135 * free(9) in llentry_free() if someone else holds one as well.
1138 IF_AFDATA_LOCK(ifp);
1141 /* Guard against race with other llentry_free(). */
1142 if (ln->la_flags & LLE_LINKED) {
1146 LLE_FREE_LOCKED(ln);
1148 IF_AFDATA_UNLOCK(ifp);
1154 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1156 * XXX cost-effective methods?
1159 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1164 if ((dst6 == NULL) || (rt == NULL))
1168 IF_AFDATA_RLOCK(ifp);
1169 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1170 IF_AFDATA_RUNLOCK(ifp);
1174 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1178 * if we get upper-layer reachability confirmation many times,
1179 * it is possible we have false information.
1183 if (ln->ln_byhint > V_nd6_maxnudhint) {
1188 ln->ln_state = ND6_LLINFO_REACHABLE;
1189 if (!ND6_LLINFO_PERMANENT(ln)) {
1190 nd6_llinfo_settimer_locked(ln,
1191 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1199 * Rejuvenate this function for routing operations related
1203 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1205 struct sockaddr_in6 *gateway;
1206 struct nd_defrouter *dr;
1210 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1221 * Only indirect routes are interesting.
1223 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1226 * check for default route
1228 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1229 &SIN6(rt_key(rt))->sin6_addr)) {
1231 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1241 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1243 struct in6_drlist *drl = (struct in6_drlist *)data;
1244 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1245 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1246 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1247 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1248 struct nd_defrouter *dr;
1249 struct nd_prefix *pr;
1250 int i = 0, error = 0;
1252 if (ifp->if_afdata[AF_INET6] == NULL)
1253 return (EPFNOSUPPORT);
1255 case SIOCGDRLST_IN6:
1257 * obsolete API, use sysctl under net.inet6.icmp6
1259 bzero(drl, sizeof(*drl));
1260 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1263 drl->defrouter[i].rtaddr = dr->rtaddr;
1264 in6_clearscope(&drl->defrouter[i].rtaddr);
1266 drl->defrouter[i].flags = dr->flags;
1267 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1268 drl->defrouter[i].expire = dr->expire +
1269 (time_second - time_uptime);
1270 drl->defrouter[i].if_index = dr->ifp->if_index;
1274 case SIOCGPRLST_IN6:
1276 * obsolete API, use sysctl under net.inet6.icmp6
1278 * XXX the structure in6_prlist was changed in backward-
1279 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1280 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1283 * XXX meaning of fields, especialy "raflags", is very
1284 * differnet between RA prefix list and RR/static prefix list.
1285 * how about separating ioctls into two?
1287 bzero(oprl, sizeof(*oprl));
1288 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1289 struct nd_pfxrouter *pfr;
1294 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1295 oprl->prefix[i].raflags = pr->ndpr_raf;
1296 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1297 oprl->prefix[i].vltime = pr->ndpr_vltime;
1298 oprl->prefix[i].pltime = pr->ndpr_pltime;
1299 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1300 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1301 oprl->prefix[i].expire = 0;
1305 /* XXX: we assume time_t is signed. */
1308 ((sizeof(maxexpire) * 8) - 1));
1309 if (pr->ndpr_vltime <
1310 maxexpire - pr->ndpr_lastupdate) {
1311 oprl->prefix[i].expire =
1312 pr->ndpr_lastupdate +
1314 (time_second - time_uptime);
1316 oprl->prefix[i].expire = maxexpire;
1320 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1322 #define RTRADDR oprl->prefix[i].advrtr[j]
1323 RTRADDR = pfr->router->rtaddr;
1324 in6_clearscope(&RTRADDR);
1329 oprl->prefix[i].advrtrs = j;
1330 oprl->prefix[i].origin = PR_ORIG_RA;
1336 case OSIOCGIFINFO_IN6:
1338 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1339 bzero(&ND, sizeof(ND));
1340 ND.linkmtu = IN6_LINKMTU(ifp);
1341 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1342 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1343 ND.reachable = ND_IFINFO(ifp)->reachable;
1344 ND.retrans = ND_IFINFO(ifp)->retrans;
1345 ND.flags = ND_IFINFO(ifp)->flags;
1346 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1347 ND.chlim = ND_IFINFO(ifp)->chlim;
1349 case SIOCGIFINFO_IN6:
1350 ND = *ND_IFINFO(ifp);
1352 case SIOCSIFINFO_IN6:
1354 * used to change host variables from userland.
1355 * intented for a use on router to reflect RA configurations.
1357 /* 0 means 'unspecified' */
1358 if (ND.linkmtu != 0) {
1359 if (ND.linkmtu < IPV6_MMTU ||
1360 ND.linkmtu > IN6_LINKMTU(ifp)) {
1364 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1367 if (ND.basereachable != 0) {
1368 int obasereachable = ND_IFINFO(ifp)->basereachable;
1370 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1371 if (ND.basereachable != obasereachable)
1372 ND_IFINFO(ifp)->reachable =
1373 ND_COMPUTE_RTIME(ND.basereachable);
1375 if (ND.retrans != 0)
1376 ND_IFINFO(ifp)->retrans = ND.retrans;
1378 ND_IFINFO(ifp)->chlim = ND.chlim;
1380 case SIOCSIFINFO_FLAGS:
1383 struct in6_ifaddr *ia;
1385 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1386 !(ND.flags & ND6_IFF_IFDISABLED)) {
1387 /* ifdisabled 1->0 transision */
1390 * If the interface is marked as ND6_IFF_IFDISABLED and
1391 * has an link-local address with IN6_IFF_DUPLICATED,
1392 * do not clear ND6_IFF_IFDISABLED.
1393 * See RFC 4862, Section 5.4.5.
1395 int duplicated_linklocal = 0;
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 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1403 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1404 duplicated_linklocal = 1;
1408 IF_ADDR_RUNLOCK(ifp);
1410 if (duplicated_linklocal) {
1411 ND.flags |= ND6_IFF_IFDISABLED;
1412 log(LOG_ERR, "Cannot enable an interface"
1413 " with a link-local address marked"
1416 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1417 if (ifp->if_flags & IFF_UP)
1420 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1421 (ND.flags & ND6_IFF_IFDISABLED)) {
1422 /* ifdisabled 0->1 transision */
1423 /* Mark all IPv6 address as tentative. */
1425 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1427 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1428 if (ifa->ifa_addr->sa_family != AF_INET6)
1430 ia = (struct in6_ifaddr *)ifa;
1431 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1433 IF_ADDR_RUNLOCK(ifp);
1436 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1437 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1438 /* auto_linklocal 0->1 transision */
1440 /* If no link-local address on ifp, configure */
1441 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1442 in6_ifattach(ifp, NULL);
1443 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1444 ifp->if_flags & IFF_UP) {
1446 * When the IF already has
1447 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1448 * address is assigned, and IFF_UP, try to
1451 int haslinklocal = 0;
1454 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1455 if (ifa->ifa_addr->sa_family != AF_INET6)
1457 ia = (struct in6_ifaddr *)ifa;
1458 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1463 IF_ADDR_RUNLOCK(ifp);
1465 in6_ifattach(ifp, NULL);
1469 ND_IFINFO(ifp)->flags = ND.flags;
1472 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1473 /* sync kernel routing table with the default router list */
1477 case SIOCSPFXFLUSH_IN6:
1479 /* flush all the prefix advertised by routers */
1480 struct nd_prefix *pr, *next;
1482 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1483 struct in6_ifaddr *ia, *ia_next;
1485 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1488 /* do we really have to remove addresses as well? */
1489 /* XXXRW: in6_ifaddrhead locking. */
1490 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1492 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1495 if (ia->ia6_ndpr == pr)
1496 in6_purgeaddr(&ia->ia_ifa);
1502 case SIOCSRTRFLUSH_IN6:
1504 /* flush all the default routers */
1505 struct nd_defrouter *dr, *next;
1508 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1514 case SIOCGNBRINFO_IN6:
1517 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1519 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1522 IF_AFDATA_RLOCK(ifp);
1523 ln = nd6_lookup(&nb_addr, 0, ifp);
1524 IF_AFDATA_RUNLOCK(ifp);
1530 nbi->state = ln->ln_state;
1531 nbi->asked = ln->la_asked;
1532 nbi->isrouter = ln->ln_router;
1533 if (ln->la_expire == 0)
1536 nbi->expire = ln->la_expire +
1537 (time_second - time_uptime);
1541 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1542 ndif->ifindex = V_nd6_defifindex;
1544 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1545 return (nd6_setdefaultiface(ndif->ifindex));
1551 * Create neighbor cache entry and cache link-layer address,
1552 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1555 * code - type dependent information
1558 * The caller of this function already acquired the ndp
1559 * cache table lock because the cache entry is returned.
1562 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1563 int lladdrlen, int type, int code)
1565 struct llentry *ln = NULL;
1572 uint16_t router = 0;
1573 struct sockaddr_in6 sin6;
1574 struct mbuf *chain = NULL;
1575 int static_route = 0;
1577 IF_AFDATA_UNLOCK_ASSERT(ifp);
1579 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1580 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1582 /* nothing must be updated for unspecified address */
1583 if (IN6_IS_ADDR_UNSPECIFIED(from))
1587 * Validation about ifp->if_addrlen and lladdrlen must be done in
1590 * XXX If the link does not have link-layer adderss, what should
1591 * we do? (ifp->if_addrlen == 0)
1592 * Spec says nothing in sections for RA, RS and NA. There's small
1593 * description on it in NS section (RFC 2461 7.2.3).
1595 flags = lladdr ? ND6_EXCLUSIVE : 0;
1596 IF_AFDATA_RLOCK(ifp);
1597 ln = nd6_lookup(from, flags, ifp);
1598 IF_AFDATA_RUNLOCK(ifp);
1600 flags |= ND6_EXCLUSIVE;
1601 IF_AFDATA_LOCK(ifp);
1602 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1603 IF_AFDATA_UNLOCK(ifp);
1606 /* do nothing if static ndp is set */
1607 if (ln->la_flags & LLE_STATIC) {
1616 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1617 if (olladdr && lladdr) {
1618 llchange = bcmp(lladdr, &ln->ll_addr,
1624 * newentry olladdr lladdr llchange (*=record)
1627 * 0 n y -- (3) * STALE
1629 * 0 y y y (5) * STALE
1630 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1631 * 1 -- y -- (7) * STALE
1634 if (lladdr) { /* (3-5) and (7) */
1636 * Record source link-layer address
1637 * XXX is it dependent to ifp->if_type?
1639 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1640 ln->la_flags |= LLE_VALID;
1641 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1645 if ((!olladdr && lladdr != NULL) || /* (3) */
1646 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1648 newstate = ND6_LLINFO_STALE;
1649 } else /* (1-2,4) */
1653 if (lladdr == NULL) /* (6) */
1654 newstate = ND6_LLINFO_NOSTATE;
1656 newstate = ND6_LLINFO_STALE;
1661 * Update the state of the neighbor cache.
1663 ln->ln_state = newstate;
1665 if (ln->ln_state == ND6_LLINFO_STALE) {
1667 * XXX: since nd6_output() below will cause
1668 * state tansition to DELAY and reset the timer,
1669 * we must set the timer now, although it is actually
1672 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1675 struct mbuf *m_hold, *m_hold_next;
1678 * reset the la_hold in advance, to explicitly
1679 * prevent a la_hold lookup in nd6_output()
1680 * (wouldn't happen, though...)
1682 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1683 m_hold; m_hold = m_hold_next) {
1684 m_hold_next = m_hold->m_nextpkt;
1685 m_hold->m_nextpkt = NULL;
1688 * we assume ifp is not a p2p here, so
1689 * just set the 2nd argument as the
1692 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1695 * If we have mbufs in the chain we need to do
1696 * deferred transmit. Copy the address from the
1697 * llentry before dropping the lock down below.
1700 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1702 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1703 /* probe right away */
1704 nd6_llinfo_settimer_locked((void *)ln, 0);
1709 * ICMP6 type dependent behavior.
1711 * NS: clear IsRouter if new entry
1712 * RS: clear IsRouter
1713 * RA: set IsRouter if there's lladdr
1714 * redir: clear IsRouter if new entry
1717 * The spec says that we must set IsRouter in the following cases:
1718 * - If lladdr exist, set IsRouter. This means (1-5).
1719 * - If it is old entry (!newentry), set IsRouter. This means (7).
1720 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1721 * A quetion arises for (1) case. (1) case has no lladdr in the
1722 * neighbor cache, this is similar to (6).
1723 * This case is rare but we figured that we MUST NOT set IsRouter.
1725 * newentry olladdr lladdr llchange NS RS RA redir
1727 * 0 n n -- (1) c ? s
1728 * 0 y n -- (2) c s s
1729 * 0 n y -- (3) c s s
1732 * 1 -- n -- (6) c c c s
1733 * 1 -- y -- (7) c c s c s
1737 switch (type & 0xff) {
1738 case ND_NEIGHBOR_SOLICIT:
1740 * New entry must have is_router flag cleared.
1742 if (is_newentry) /* (6-7) */
1747 * If the icmp is a redirect to a better router, always set the
1748 * is_router flag. Otherwise, if the entry is newly created,
1749 * clear the flag. [RFC 2461, sec 8.3]
1751 if (code == ND_REDIRECT_ROUTER)
1753 else if (is_newentry) /* (6-7) */
1756 case ND_ROUTER_SOLICIT:
1758 * is_router flag must always be cleared.
1762 case ND_ROUTER_ADVERT:
1764 * Mark an entry with lladdr as a router.
1766 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1767 (is_newentry && lladdr)) { /* (7) */
1774 static_route = (ln->la_flags & LLE_STATIC);
1775 router = ln->ln_router;
1777 if (flags & ND6_EXCLUSIVE)
1785 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1788 * When the link-layer address of a router changes, select the
1789 * best router again. In particular, when the neighbor entry is newly
1790 * created, it might affect the selection policy.
1791 * Question: can we restrict the first condition to the "is_newentry"
1793 * XXX: when we hear an RA from a new router with the link-layer
1794 * address option, defrouter_select() is called twice, since
1795 * defrtrlist_update called the function as well. However, I believe
1796 * we can compromise the overhead, since it only happens the first
1798 * XXX: although defrouter_select() should not have a bad effect
1799 * for those are not autoconfigured hosts, we explicitly avoid such
1802 if (do_update && router &&
1803 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1805 * guaranteed recursion
1813 if (flags & ND6_EXCLUSIVE)
1824 nd6_slowtimo(void *arg)
1826 CURVNET_SET((struct vnet *) arg);
1827 struct nd_ifinfo *nd6if;
1830 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1831 nd6_slowtimo, curvnet);
1832 IFNET_RLOCK_NOSLEEP();
1833 TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1834 if (ifp->if_afdata[AF_INET6] == NULL)
1836 nd6if = ND_IFINFO(ifp);
1837 if (nd6if->basereachable && /* already initialized */
1838 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1840 * Since reachable time rarely changes by router
1841 * advertisements, we SHOULD insure that a new random
1842 * value gets recomputed at least once every few hours.
1845 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1846 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1849 IFNET_RUNLOCK_NOSLEEP();
1854 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1855 struct sockaddr_in6 *dst, struct rtentry *rt0)
1858 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1863 * Note that I'm not enforcing any global serialization
1864 * lle state or asked changes here as the logic is too
1865 * complicated to avoid having to always acquire an exclusive
1870 #define senderr(e) { error = (e); goto bad;}
1873 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1874 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1875 struct mbuf **chain)
1877 struct mbuf *m = m0;
1879 struct llentry *ln = lle;
1880 struct ip6_hdr *ip6;
1888 LLE_WLOCK_ASSERT(lle);
1890 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1893 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1896 if (nd6_need_cache(ifp) == 0)
1900 * next hop determination. This routine is derived from ether_output.
1904 * Address resolution or Neighbor Unreachability Detection
1906 * At this point, the destination of the packet must be a unicast
1907 * or an anycast address(i.e. not a multicast).
1910 flags = (lle != NULL) ? LLE_EXCLUSIVE : 0;
1913 IF_AFDATA_RLOCK(ifp);
1914 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1915 IF_AFDATA_RUNLOCK(ifp);
1916 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1918 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1919 * the condition below is not very efficient. But we believe
1920 * it is tolerable, because this should be a rare case.
1922 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1923 IF_AFDATA_LOCK(ifp);
1924 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1925 IF_AFDATA_UNLOCK(ifp);
1929 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1930 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1931 char ip6buf[INET6_ADDRSTRLEN];
1933 "nd6_output: can't allocate llinfo for %s "
1935 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1936 senderr(EIO); /* XXX: good error? */
1938 goto sendpkt; /* send anyway */
1941 /* We don't have to do link-layer address resolution on a p2p link. */
1942 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1943 ln->ln_state < ND6_LLINFO_REACHABLE) {
1944 if ((flags & LLE_EXCLUSIVE) == 0) {
1945 flags |= LLE_EXCLUSIVE;
1949 ln->ln_state = ND6_LLINFO_STALE;
1950 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1954 * The first time we send a packet to a neighbor whose entry is
1955 * STALE, we have to change the state to DELAY and a sets a timer to
1956 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1957 * neighbor unreachability detection on expiration.
1960 if (ln->ln_state == ND6_LLINFO_STALE) {
1961 if ((flags & LLE_EXCLUSIVE) == 0) {
1962 flags |= LLE_EXCLUSIVE;
1967 ln->ln_state = ND6_LLINFO_DELAY;
1968 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1972 * If the neighbor cache entry has a state other than INCOMPLETE
1973 * (i.e. its link-layer address is already resolved), just
1976 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1980 * There is a neighbor cache entry, but no ethernet address
1981 * response yet. Append this latest packet to the end of the
1982 * packet queue in the mbuf, unless the number of the packet
1983 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1984 * the oldest packet in the queue will be removed.
1986 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1987 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1989 if ((flags & LLE_EXCLUSIVE) == 0) {
1990 flags |= LLE_EXCLUSIVE;
1995 LLE_WLOCK_ASSERT(ln);
1998 struct mbuf *m_hold;
2002 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2004 if (m_hold->m_nextpkt == NULL) {
2005 m_hold->m_nextpkt = m;
2009 while (i >= V_nd6_maxqueuelen) {
2010 m_hold = ln->la_hold;
2011 ln->la_hold = ln->la_hold->m_nextpkt;
2020 * If there has been no NS for the neighbor after entering the
2021 * INCOMPLETE state, send the first solicitation.
2023 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
2026 nd6_llinfo_settimer_locked(ln,
2027 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2029 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2030 if (lle != NULL && ln == lle)
2033 } else if (lle == NULL || ln != lle) {
2035 * We did the lookup (no lle arg) so we
2036 * need to do the unlock here.
2044 /* discard the packet if IPv6 operation is disabled on the interface */
2045 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2046 error = ENETDOWN; /* better error? */
2050 * ln is valid and the caller did not pass in
2053 if ((ln != NULL) && (lle == NULL)) {
2054 if (flags & LLE_EXCLUSIVE)
2061 mac_netinet6_nd6_send(ifp, m);
2065 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2066 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2067 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2068 * to be diverted to user space. When re-injected into the kernel,
2069 * send_output() will directly dispatch them to the outgoing interface.
2071 if (send_sendso_input_hook != NULL) {
2072 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2074 ip6 = mtod(m, struct ip6_hdr *);
2075 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2076 /* Use the SEND socket */
2077 error = send_sendso_input_hook(m, ifp, SND_OUT,
2079 /* -1 == no app on SEND socket */
2080 if (error == 0 || error != -1)
2086 * We were passed in a pointer to an lle with the lock held
2087 * this means that we can't call if_output as we will
2088 * recurse on the lle lock - so what we do is we create
2089 * a list of mbufs to send and transmit them in the caller
2090 * after the lock is dropped
2099 * append mbuf to end of deferred chain
2102 while (mb->m_nextpkt != NULL)
2108 m_clrprotoflags(m); /* Avoid confusing lower layers. */
2109 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2110 mtod(m, struct ip6_hdr *));
2111 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2112 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2115 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2120 * ln is valid and the caller did not pass in
2123 if ((ln != NULL) && (lle == NULL)) {
2124 if (flags & LLE_EXCLUSIVE)
2137 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2138 struct sockaddr_in6 *dst, struct route *ro)
2140 struct mbuf *m, *m_head;
2141 struct ifnet *outifp;
2145 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2152 m_head = m_head->m_nextpkt;
2153 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2158 * note that intermediate errors are blindly ignored - but this is
2159 * the same convention as used with nd6_output when called by
2167 nd6_need_cache(struct ifnet *ifp)
2170 * XXX: we currently do not make neighbor cache on any interface
2171 * other than ARCnet, Ethernet, FDDI and GIF.
2174 * - unidirectional tunnels needs no ND
2176 switch (ifp->if_type) {
2184 #ifdef IFT_IEEE80211
2187 case IFT_INFINIBAND:
2189 case IFT_PROPVIRTUAL:
2197 * the callers of this function need to be re-worked to drop
2198 * the lle lock, drop here for now
2201 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2202 const struct sockaddr *dst, u_char *desten, struct llentry **lle)
2207 IF_AFDATA_UNLOCK_ASSERT(ifp);
2208 if (m != NULL && m->m_flags & M_MCAST) {
2211 switch (ifp->if_type) {
2217 #ifdef IFT_IEEE80211
2222 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2227 * netbsd can use if_broadcastaddr, but we don't do so
2228 * to reduce # of ifdef.
2230 for (i = 0; i < ifp->if_addrlen; i++)
2238 return (EAFNOSUPPORT);
2244 * the entry should have been created in nd6_store_lladdr
2246 IF_AFDATA_RLOCK(ifp);
2247 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2248 IF_AFDATA_RUNLOCK(ifp);
2249 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2252 /* this could happen, if we could not allocate memory */
2257 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2261 * A *small* use after free race exists here
2267 clear_llinfo_pqueue(struct llentry *ln)
2269 struct mbuf *m_hold, *m_hold_next;
2271 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2272 m_hold_next = m_hold->m_nextpkt;
2280 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2281 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2283 SYSCTL_DECL(_net_inet6_icmp6);
2285 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2286 CTLFLAG_RD, nd6_sysctl_drlist, "");
2287 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2288 CTLFLAG_RD, nd6_sysctl_prlist, "");
2289 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2290 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2291 SYSCTL_VNET_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2292 CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2295 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2297 struct in6_defrouter d;
2298 struct nd_defrouter *dr;
2304 bzero(&d, sizeof(d));
2305 d.rtaddr.sin6_family = AF_INET6;
2306 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2311 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2312 d.rtaddr.sin6_addr = dr->rtaddr;
2313 error = sa6_recoverscope(&d.rtaddr);
2316 d.flags = dr->flags;
2317 d.rtlifetime = dr->rtlifetime;
2318 d.expire = dr->expire + (time_second - time_uptime);
2319 d.if_index = dr->ifp->if_index;
2320 error = SYSCTL_OUT(req, &d, sizeof(d));
2328 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2330 struct in6_prefix p;
2331 struct sockaddr_in6 s6;
2332 struct nd_prefix *pr;
2333 struct nd_pfxrouter *pfr;
2336 char ip6buf[INET6_ADDRSTRLEN];
2341 bzero(&p, sizeof(p));
2342 p.origin = PR_ORIG_RA;
2343 bzero(&s6, sizeof(s6));
2344 s6.sin6_family = AF_INET6;
2345 s6.sin6_len = sizeof(s6);
2350 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2351 p.prefix = pr->ndpr_prefix;
2352 if (sa6_recoverscope(&p.prefix)) {
2353 log(LOG_ERR, "scope error in prefix list (%s)\n",
2354 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2355 /* XXX: press on... */
2357 p.raflags = pr->ndpr_raf;
2358 p.prefixlen = pr->ndpr_plen;
2359 p.vltime = pr->ndpr_vltime;
2360 p.pltime = pr->ndpr_pltime;
2361 p.if_index = pr->ndpr_ifp->if_index;
2362 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2365 /* XXX: we assume time_t is signed. */
2367 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2368 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2369 p.expire = pr->ndpr_lastupdate +
2371 (time_second - time_uptime);
2373 p.expire = maxexpire;
2375 p.refcnt = pr->ndpr_refcnt;
2376 p.flags = pr->ndpr_stateflags;
2378 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2380 error = SYSCTL_OUT(req, &p, sizeof(p));
2383 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2384 s6.sin6_addr = pfr->router->rtaddr;
2385 if (sa6_recoverscope(&s6))
2387 "scope error in prefix list (%s)\n",
2388 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2389 error = SYSCTL_OUT(req, &s6, sizeof(s6));