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>
54 #include <sys/sysctl.h>
57 #include <net/if_var.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 *);
136 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
138 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
139 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
141 VNET_DEFINE(struct callout, nd6_timer_ch);
147 LIST_INIT(&V_nd_prefix);
149 /* initialization of the default router list */
150 TAILQ_INIT(&V_nd_defrouter);
153 callout_init(&V_nd6_slowtimo_ch, 0);
154 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
155 nd6_slowtimo, curvnet);
163 callout_drain(&V_nd6_slowtimo_ch);
164 callout_drain(&V_nd6_timer_ch);
169 nd6_ifattach(struct ifnet *ifp)
171 struct nd_ifinfo *nd;
173 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
176 nd->chlim = IPV6_DEFHLIM;
177 nd->basereachable = REACHABLE_TIME;
178 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
179 nd->retrans = RETRANS_TIMER;
181 nd->flags = ND6_IFF_PERFORMNUD;
183 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
184 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
185 * default regardless of the V_ip6_auto_linklocal configuration to
186 * give a reasonable default behavior.
188 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
189 (ifp->if_flags & IFF_LOOPBACK))
190 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
192 * A loopback interface does not need to accept RTADV.
193 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
194 * default regardless of the V_ip6_accept_rtadv configuration to
195 * prevent the interface from accepting RA messages arrived
196 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
198 if (V_ip6_accept_rtadv &&
199 !(ifp->if_flags & IFF_LOOPBACK) &&
200 (ifp->if_type != IFT_BRIDGE))
201 nd->flags |= ND6_IFF_ACCEPT_RTADV;
202 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
203 nd->flags |= ND6_IFF_NO_RADR;
205 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
206 nd6_setmtu0(ifp, nd);
212 nd6_ifdetach(struct nd_ifinfo *nd)
219 * Reset ND level link MTU. This function is called when the physical MTU
220 * changes, which means we might have to adjust the ND level MTU.
223 nd6_setmtu(struct ifnet *ifp)
226 nd6_setmtu0(ifp, ND_IFINFO(ifp));
229 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
231 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
235 omaxmtu = ndi->maxmtu;
237 switch (ifp->if_type) {
239 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
242 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
245 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
248 ndi->maxmtu = ifp->if_mtu;
253 * Decreasing the interface MTU under IPV6 minimum MTU may cause
254 * undesirable situation. We thus notify the operator of the change
255 * explicitly. The check for omaxmtu is necessary to restrict the
256 * log to the case of changing the MTU, not initializing it.
258 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
259 log(LOG_NOTICE, "nd6_setmtu0: "
260 "new link MTU on %s (%lu) is too small for IPv6\n",
261 if_name(ifp), (unsigned long)ndi->maxmtu);
264 if (ndi->maxmtu > V_in6_maxmtu)
265 in6_setmaxmtu(); /* check all interfaces just in case */
270 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
273 bzero(ndopts, sizeof(*ndopts));
274 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
276 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
279 ndopts->nd_opts_done = 1;
280 ndopts->nd_opts_search = NULL;
285 * Take one ND option.
288 nd6_option(union nd_opts *ndopts)
290 struct nd_opt_hdr *nd_opt;
293 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
294 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
296 if (ndopts->nd_opts_search == NULL)
298 if (ndopts->nd_opts_done)
301 nd_opt = ndopts->nd_opts_search;
303 /* make sure nd_opt_len is inside the buffer */
304 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
305 bzero(ndopts, sizeof(*ndopts));
309 olen = nd_opt->nd_opt_len << 3;
312 * Message validation requires that all included
313 * options have a length that is greater than zero.
315 bzero(ndopts, sizeof(*ndopts));
319 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
320 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
321 /* option overruns the end of buffer, invalid */
322 bzero(ndopts, sizeof(*ndopts));
324 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
325 /* reached the end of options chain */
326 ndopts->nd_opts_done = 1;
327 ndopts->nd_opts_search = NULL;
333 * Parse multiple ND options.
334 * This function is much easier to use, for ND routines that do not need
335 * multiple options of the same type.
338 nd6_options(union nd_opts *ndopts)
340 struct nd_opt_hdr *nd_opt;
343 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
344 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
346 if (ndopts->nd_opts_search == NULL)
350 nd_opt = nd6_option(ndopts);
351 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
353 * Message validation requires that all included
354 * options have a length that is greater than zero.
356 ICMP6STAT_INC(icp6s_nd_badopt);
357 bzero(ndopts, sizeof(*ndopts));
364 switch (nd_opt->nd_opt_type) {
365 case ND_OPT_SOURCE_LINKADDR:
366 case ND_OPT_TARGET_LINKADDR:
368 case ND_OPT_REDIRECTED_HEADER:
369 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
371 "duplicated ND6 option found (type=%d)\n",
372 nd_opt->nd_opt_type));
375 ndopts->nd_opt_array[nd_opt->nd_opt_type]
379 case ND_OPT_PREFIX_INFORMATION:
380 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
381 ndopts->nd_opt_array[nd_opt->nd_opt_type]
384 ndopts->nd_opts_pi_end =
385 (struct nd_opt_prefix_info *)nd_opt;
387 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
388 case ND_OPT_RDNSS: /* RFC 6106 */
389 case ND_OPT_DNSSL: /* RFC 6106 */
391 * Silently ignore options we know and do not care about
397 * Unknown options must be silently ignored,
398 * to accomodate future extension to the protocol.
401 "nd6_options: unsupported option %d - "
402 "option ignored\n", nd_opt->nd_opt_type));
407 if (i > V_nd6_maxndopt) {
408 ICMP6STAT_INC(icp6s_nd_toomanyopt);
409 nd6log((LOG_INFO, "too many loop in nd opt\n"));
413 if (ndopts->nd_opts_done)
421 * ND6 timer routine to handle ND6 entries
424 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
428 LLE_WLOCK_ASSERT(ln);
433 canceled = callout_stop(&ln->ln_timer_ch);
435 ln->la_expire = time_uptime + tick / hz;
437 if (tick > INT_MAX) {
438 ln->ln_ntick = tick - INT_MAX;
439 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
440 nd6_llinfo_timer, ln);
443 canceled = callout_reset(&ln->ln_timer_ch, tick,
444 nd6_llinfo_timer, ln);
452 nd6_llinfo_settimer(struct llentry *ln, long tick)
456 nd6_llinfo_settimer_locked(ln, tick);
461 nd6_llinfo_timer(void *arg)
464 struct in6_addr *dst;
466 struct nd_ifinfo *ndi = NULL;
468 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
469 ln = (struct llentry *)arg;
470 LLE_WLOCK_ASSERT(ln);
471 ifp = ln->lle_tbl->llt_ifp;
473 CURVNET_SET(ifp->if_vnet);
475 if (ln->ln_ntick > 0) {
476 if (ln->ln_ntick > INT_MAX) {
477 ln->ln_ntick -= INT_MAX;
478 nd6_llinfo_settimer_locked(ln, INT_MAX);
481 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
486 ndi = ND_IFINFO(ifp);
487 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
488 if (ln->la_flags & LLE_STATIC) {
492 if (ln->la_flags & LLE_DELETED) {
493 (void)nd6_free(ln, 0);
498 switch (ln->ln_state) {
499 case ND6_LLINFO_INCOMPLETE:
500 if (ln->la_asked < V_nd6_mmaxtries) {
502 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
504 nd6_ns_output(ifp, NULL, dst, ln, 0);
507 struct mbuf *m = ln->la_hold;
512 * assuming every packet in la_hold has the
513 * same IP header. Send error after unlock.
518 clear_llinfo_pqueue(ln);
520 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
521 (void)nd6_free(ln, 0);
524 icmp6_error2(m, ICMP6_DST_UNREACH,
525 ICMP6_DST_UNREACH_ADDR, 0, ifp);
528 case ND6_LLINFO_REACHABLE:
529 if (!ND6_LLINFO_PERMANENT(ln)) {
530 ln->ln_state = ND6_LLINFO_STALE;
531 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
535 case ND6_LLINFO_STALE:
536 /* Garbage Collection(RFC 2461 5.3) */
537 if (!ND6_LLINFO_PERMANENT(ln)) {
538 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
539 (void)nd6_free(ln, 1);
544 case ND6_LLINFO_DELAY:
545 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
548 ln->ln_state = ND6_LLINFO_PROBE;
549 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
551 nd6_ns_output(ifp, dst, dst, ln, 0);
554 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
555 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
558 case ND6_LLINFO_PROBE:
559 if (ln->la_asked < V_nd6_umaxtries) {
561 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
563 nd6_ns_output(ifp, dst, dst, ln, 0);
566 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
567 (void)nd6_free(ln, 0);
572 panic("%s: paths in a dark night can be confusing: %d",
573 __func__, ln->ln_state);
583 * ND6 timer routine to expire default route list and prefix list
588 CURVNET_SET((struct vnet *) arg);
589 struct nd_defrouter *dr, *ndr;
590 struct nd_prefix *pr, *npr;
591 struct in6_ifaddr *ia6, *nia6;
593 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
596 /* expire default router list */
597 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
598 if (dr->expire && dr->expire < time_uptime)
603 * expire interface addresses.
604 * in the past the loop was inside prefix expiry processing.
605 * However, from a stricter speci-confrmance standpoint, we should
606 * rather separate address lifetimes and prefix lifetimes.
608 * XXXRW: in6_ifaddrhead locking.
611 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
612 /* check address lifetime */
613 if (IFA6_IS_INVALID(ia6)) {
617 * If the expiring address is temporary, try
618 * regenerating a new one. This would be useful when
619 * we suspended a laptop PC, then turned it on after a
620 * period that could invalidate all temporary
621 * addresses. Although we may have to restart the
622 * loop (see below), it must be after purging the
623 * address. Otherwise, we'd see an infinite loop of
626 if (V_ip6_use_tempaddr &&
627 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
628 if (regen_tmpaddr(ia6) == 0)
632 in6_purgeaddr(&ia6->ia_ifa);
635 goto addrloop; /* XXX: see below */
636 } else if (IFA6_IS_DEPRECATED(ia6)) {
637 int oldflags = ia6->ia6_flags;
639 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
642 * If a temporary address has just become deprecated,
643 * regenerate a new one if possible.
645 if (V_ip6_use_tempaddr &&
646 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
647 (oldflags & IN6_IFF_DEPRECATED) == 0) {
649 if (regen_tmpaddr(ia6) == 0) {
651 * A new temporary address is
653 * XXX: this means the address chain
654 * has changed while we are still in
655 * the loop. Although the change
656 * would not cause disaster (because
657 * it's not a deletion, but an
658 * addition,) we'd rather restart the
659 * loop just for safety. Or does this
660 * significantly reduce performance??
667 * A new RA might have made a deprecated address
670 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
674 /* expire prefix list */
675 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
677 * check prefix lifetime.
678 * since pltime is just for autoconf, pltime processing for
679 * prefix is not necessary.
681 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
682 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
685 * address expiration and prefix expiration are
686 * separate. NEVER perform in6_purgeaddr here.
695 * ia6 - deprecated/invalidated temporary address
698 regen_tmpaddr(struct in6_ifaddr *ia6)
702 struct in6_ifaddr *public_ifa6 = NULL;
704 ifp = ia6->ia_ifa.ifa_ifp;
706 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
707 struct in6_ifaddr *it6;
709 if (ifa->ifa_addr->sa_family != AF_INET6)
712 it6 = (struct in6_ifaddr *)ifa;
714 /* ignore no autoconf addresses. */
715 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
718 /* ignore autoconf addresses with different prefixes. */
719 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
723 * Now we are looking at an autoconf address with the same
724 * prefix as ours. If the address is temporary and is still
725 * preferred, do not create another one. It would be rare, but
726 * could happen, for example, when we resume a laptop PC after
729 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
730 !IFA6_IS_DEPRECATED(it6)) {
736 * This is a public autoconf address that has the same prefix
737 * as ours. If it is preferred, keep it. We can't break the
738 * loop here, because there may be a still-preferred temporary
739 * address with the prefix.
741 if (!IFA6_IS_DEPRECATED(it6))
744 if (public_ifa6 != NULL)
745 ifa_ref(&public_ifa6->ia_ifa);
747 IF_ADDR_RUNLOCK(ifp);
749 if (public_ifa6 != NULL) {
752 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
753 ifa_free(&public_ifa6->ia_ifa);
754 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
755 " tmp addr,errno=%d\n", e);
758 ifa_free(&public_ifa6->ia_ifa);
766 * Nuke neighbor cache/prefix/default router management table, right before
770 nd6_purge(struct ifnet *ifp)
772 struct nd_defrouter *dr, *ndr;
773 struct nd_prefix *pr, *npr;
776 * Nuke default router list entries toward ifp.
777 * We defer removal of default router list entries that is installed
778 * in the routing table, in order to keep additional side effects as
781 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
789 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
797 /* Nuke prefix list entries toward ifp */
798 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
799 if (pr->ndpr_ifp == ifp) {
801 * Because if_detach() does *not* release prefixes
802 * while purging addresses the reference count will
803 * still be above zero. We therefore reset it to
804 * make sure that the prefix really gets purged.
809 * Previously, pr->ndpr_addr is removed as well,
810 * but I strongly believe we don't have to do it.
811 * nd6_purge() is only called from in6_ifdetach(),
812 * which removes all the associated interface addresses
814 * (jinmei@kame.net 20010129)
820 /* cancel default outgoing interface setting */
821 if (V_nd6_defifindex == ifp->if_index)
822 nd6_setdefaultiface(0);
824 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
825 /* Refresh default router list. */
830 * We do not nuke the neighbor cache entries here any more
831 * because the neighbor cache is kept in if_afdata[AF_INET6].
832 * nd6_purge() is invoked by in6_ifdetach() which is called
833 * from if_detach() where everything gets purged. So let
834 * in6_domifdetach() do the actual L2 table purging work.
839 * the caller acquires and releases the lock on the lltbls
840 * Returns the llentry locked
843 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
845 struct sockaddr_in6 sin6;
849 bzero(&sin6, sizeof(sin6));
850 sin6.sin6_len = sizeof(struct sockaddr_in6);
851 sin6.sin6_family = AF_INET6;
852 sin6.sin6_addr = *addr6;
854 IF_AFDATA_LOCK_ASSERT(ifp);
857 if (flags & ND6_CREATE)
858 llflags |= LLE_CREATE;
859 if (flags & ND6_EXCLUSIVE)
860 llflags |= LLE_EXCLUSIVE;
862 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
863 if ((ln != NULL) && (llflags & LLE_CREATE))
864 ln->ln_state = ND6_LLINFO_NOSTATE;
870 * Test whether a given IPv6 address is a neighbor or not, ignoring
871 * the actual neighbor cache. The neighbor cache is ignored in order
872 * to not reenter the routing code from within itself.
875 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
877 struct nd_prefix *pr;
878 struct ifaddr *dstaddr;
881 * A link-local address is always a neighbor.
882 * XXX: a link does not necessarily specify a single interface.
884 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
885 struct sockaddr_in6 sin6_copy;
889 * We need sin6_copy since sa6_recoverscope() may modify the
893 if (sa6_recoverscope(&sin6_copy))
894 return (0); /* XXX: should be impossible */
895 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
897 if (sin6_copy.sin6_scope_id == zone)
904 * If the address matches one of our addresses,
905 * it should be a neighbor.
906 * If the address matches one of our on-link prefixes, it should be a
909 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
910 if (pr->ndpr_ifp != ifp)
913 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
916 /* Always use the default FIB here. */
917 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
918 0, 0, RT_DEFAULT_FIB);
922 * This is the case where multiple interfaces
923 * have the same prefix, but only one is installed
924 * into the routing table and that prefix entry
925 * is not the one being examined here. In the case
926 * where RADIX_MPATH is enabled, multiple route
927 * entries (of the same rt_key value) will be
928 * installed because the interface addresses all
931 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
932 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
939 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
940 &addr->sin6_addr, &pr->ndpr_mask))
945 * If the address is assigned on the node of the other side of
946 * a p2p interface, the address should be a neighbor.
948 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
949 if (dstaddr != NULL) {
950 if (dstaddr->ifa_ifp == ifp) {
958 * If the default router list is empty, all addresses are regarded
959 * as on-link, and thus, as a neighbor.
961 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
962 TAILQ_EMPTY(&V_nd_defrouter) &&
963 V_nd6_defifindex == ifp->if_index) {
972 * Detect if a given IPv6 address identifies a neighbor on a given link.
973 * XXX: should take care of the destination of a p2p link?
976 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
981 IF_AFDATA_UNLOCK_ASSERT(ifp);
982 if (nd6_is_new_addr_neighbor(addr, ifp))
986 * Even if the address matches none of our addresses, it might be
987 * in the neighbor cache.
989 IF_AFDATA_RLOCK(ifp);
990 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
994 IF_AFDATA_RUNLOCK(ifp);
999 * Free an nd6 llinfo entry.
1000 * Since the function would cause significant changes in the kernel, DO NOT
1001 * make it global, unless you have a strong reason for the change, and are sure
1002 * that the change is safe.
1004 static struct llentry *
1005 nd6_free(struct llentry *ln, int gc)
1007 struct llentry *next;
1008 struct nd_defrouter *dr;
1011 LLE_WLOCK_ASSERT(ln);
1014 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1015 * even though it is not harmful, it was not really necessary.
1019 nd6_llinfo_settimer_locked(ln, -1);
1021 ifp = ln->lle_tbl->llt_ifp;
1023 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1024 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1026 if (dr != NULL && dr->expire &&
1027 ln->ln_state == ND6_LLINFO_STALE && gc) {
1029 * If the reason for the deletion is just garbage
1030 * collection, and the neighbor is an active default
1031 * router, do not delete it. Instead, reset the GC
1032 * timer using the router's lifetime.
1033 * Simply deleting the entry would affect default
1034 * router selection, which is not necessarily a good
1035 * thing, especially when we're using router preference
1037 * XXX: the check for ln_state would be redundant,
1038 * but we intentionally keep it just in case.
1040 if (dr->expire > time_uptime)
1041 nd6_llinfo_settimer_locked(ln,
1042 (dr->expire - time_uptime) * hz);
1044 nd6_llinfo_settimer_locked(ln,
1045 (long)V_nd6_gctimer * hz);
1047 next = LIST_NEXT(ln, lle_next);
1055 * Unreachablity of a router might affect the default
1056 * router selection and on-link detection of advertised
1061 * Temporarily fake the state to choose a new default
1062 * router and to perform on-link determination of
1063 * prefixes correctly.
1064 * Below the state will be set correctly,
1065 * or the entry itself will be deleted.
1067 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1070 if (ln->ln_router || dr) {
1073 * We need to unlock to avoid a LOR with rt6_flush() with the
1074 * rnh and for the calls to pfxlist_onlink_check() and
1075 * defrouter_select() in the block further down for calls
1076 * into nd6_lookup(). We still hold a ref.
1081 * rt6_flush must be called whether or not the neighbor
1082 * is in the Default Router List.
1083 * See a corresponding comment in nd6_na_input().
1085 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1090 * Since defrouter_select() does not affect the
1091 * on-link determination and MIP6 needs the check
1092 * before the default router selection, we perform
1095 pfxlist_onlink_check();
1098 * Refresh default router list.
1103 if (ln->ln_router || dr)
1108 * Before deleting the entry, remember the next entry as the
1109 * return value. We need this because pfxlist_onlink_check() above
1110 * might have freed other entries (particularly the old next entry) as
1111 * a side effect (XXX).
1113 next = LIST_NEXT(ln, lle_next);
1116 * Save to unlock. We still hold an extra reference and will not
1117 * free(9) in llentry_free() if someone else holds one as well.
1120 IF_AFDATA_LOCK(ifp);
1123 /* Guard against race with other llentry_free(). */
1124 if (ln->la_flags & LLE_LINKED) {
1128 LLE_FREE_LOCKED(ln);
1130 IF_AFDATA_UNLOCK(ifp);
1136 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1138 * XXX cost-effective methods?
1141 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1146 if ((dst6 == NULL) || (rt == NULL))
1150 IF_AFDATA_RLOCK(ifp);
1151 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1152 IF_AFDATA_RUNLOCK(ifp);
1156 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1160 * if we get upper-layer reachability confirmation many times,
1161 * it is possible we have false information.
1165 if (ln->ln_byhint > V_nd6_maxnudhint) {
1170 ln->ln_state = ND6_LLINFO_REACHABLE;
1171 if (!ND6_LLINFO_PERMANENT(ln)) {
1172 nd6_llinfo_settimer_locked(ln,
1173 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1181 * Rejuvenate this function for routing operations related
1185 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1187 struct sockaddr_in6 *gateway;
1188 struct nd_defrouter *dr;
1192 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1203 * Only indirect routes are interesting.
1205 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1208 * check for default route
1210 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1211 &SIN6(rt_key(rt))->sin6_addr)) {
1213 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1223 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1225 struct in6_drlist *drl = (struct in6_drlist *)data;
1226 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1227 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1228 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1229 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1230 struct nd_defrouter *dr;
1231 struct nd_prefix *pr;
1232 int i = 0, error = 0;
1234 if (ifp->if_afdata[AF_INET6] == NULL)
1235 return (EPFNOSUPPORT);
1237 case SIOCGDRLST_IN6:
1239 * obsolete API, use sysctl under net.inet6.icmp6
1241 bzero(drl, sizeof(*drl));
1242 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1245 drl->defrouter[i].rtaddr = dr->rtaddr;
1246 in6_clearscope(&drl->defrouter[i].rtaddr);
1248 drl->defrouter[i].flags = dr->flags;
1249 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1250 drl->defrouter[i].expire = dr->expire +
1251 (time_second - time_uptime);
1252 drl->defrouter[i].if_index = dr->ifp->if_index;
1256 case SIOCGPRLST_IN6:
1258 * obsolete API, use sysctl under net.inet6.icmp6
1260 * XXX the structure in6_prlist was changed in backward-
1261 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1262 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1265 * XXX meaning of fields, especialy "raflags", is very
1266 * differnet between RA prefix list and RR/static prefix list.
1267 * how about separating ioctls into two?
1269 bzero(oprl, sizeof(*oprl));
1270 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1271 struct nd_pfxrouter *pfr;
1276 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1277 oprl->prefix[i].raflags = pr->ndpr_raf;
1278 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1279 oprl->prefix[i].vltime = pr->ndpr_vltime;
1280 oprl->prefix[i].pltime = pr->ndpr_pltime;
1281 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1282 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1283 oprl->prefix[i].expire = 0;
1287 /* XXX: we assume time_t is signed. */
1290 ((sizeof(maxexpire) * 8) - 1));
1291 if (pr->ndpr_vltime <
1292 maxexpire - pr->ndpr_lastupdate) {
1293 oprl->prefix[i].expire =
1294 pr->ndpr_lastupdate +
1296 (time_second - time_uptime);
1298 oprl->prefix[i].expire = maxexpire;
1302 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1304 #define RTRADDR oprl->prefix[i].advrtr[j]
1305 RTRADDR = pfr->router->rtaddr;
1306 in6_clearscope(&RTRADDR);
1311 oprl->prefix[i].advrtrs = j;
1312 oprl->prefix[i].origin = PR_ORIG_RA;
1318 case OSIOCGIFINFO_IN6:
1320 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1321 bzero(&ND, sizeof(ND));
1322 ND.linkmtu = IN6_LINKMTU(ifp);
1323 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1324 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1325 ND.reachable = ND_IFINFO(ifp)->reachable;
1326 ND.retrans = ND_IFINFO(ifp)->retrans;
1327 ND.flags = ND_IFINFO(ifp)->flags;
1328 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1329 ND.chlim = ND_IFINFO(ifp)->chlim;
1331 case SIOCGIFINFO_IN6:
1332 ND = *ND_IFINFO(ifp);
1334 case SIOCSIFINFO_IN6:
1336 * used to change host variables from userland.
1337 * intented for a use on router to reflect RA configurations.
1339 /* 0 means 'unspecified' */
1340 if (ND.linkmtu != 0) {
1341 if (ND.linkmtu < IPV6_MMTU ||
1342 ND.linkmtu > IN6_LINKMTU(ifp)) {
1346 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1349 if (ND.basereachable != 0) {
1350 int obasereachable = ND_IFINFO(ifp)->basereachable;
1352 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1353 if (ND.basereachable != obasereachable)
1354 ND_IFINFO(ifp)->reachable =
1355 ND_COMPUTE_RTIME(ND.basereachable);
1357 if (ND.retrans != 0)
1358 ND_IFINFO(ifp)->retrans = ND.retrans;
1360 ND_IFINFO(ifp)->chlim = ND.chlim;
1362 case SIOCSIFINFO_FLAGS:
1365 struct in6_ifaddr *ia;
1367 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1368 !(ND.flags & ND6_IFF_IFDISABLED)) {
1369 /* ifdisabled 1->0 transision */
1372 * If the interface is marked as ND6_IFF_IFDISABLED and
1373 * has an link-local address with IN6_IFF_DUPLICATED,
1374 * do not clear ND6_IFF_IFDISABLED.
1375 * See RFC 4862, Section 5.4.5.
1377 int duplicated_linklocal = 0;
1380 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1381 if (ifa->ifa_addr->sa_family != AF_INET6)
1383 ia = (struct in6_ifaddr *)ifa;
1384 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1385 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1386 duplicated_linklocal = 1;
1390 IF_ADDR_RUNLOCK(ifp);
1392 if (duplicated_linklocal) {
1393 ND.flags |= ND6_IFF_IFDISABLED;
1394 log(LOG_ERR, "Cannot enable an interface"
1395 " with a link-local address marked"
1398 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1399 if (ifp->if_flags & IFF_UP)
1402 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1403 (ND.flags & ND6_IFF_IFDISABLED)) {
1404 /* ifdisabled 0->1 transision */
1405 /* Mark all IPv6 address as tentative. */
1407 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1409 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1410 if (ifa->ifa_addr->sa_family != AF_INET6)
1412 ia = (struct in6_ifaddr *)ifa;
1413 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1415 IF_ADDR_RUNLOCK(ifp);
1418 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1419 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1420 /* auto_linklocal 0->1 transision */
1422 /* If no link-local address on ifp, configure */
1423 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1424 in6_ifattach(ifp, NULL);
1425 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1426 ifp->if_flags & IFF_UP) {
1428 * When the IF already has
1429 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1430 * address is assigned, and IFF_UP, try to
1433 int haslinklocal = 0;
1436 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1437 if (ifa->ifa_addr->sa_family != AF_INET6)
1439 ia = (struct in6_ifaddr *)ifa;
1440 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1445 IF_ADDR_RUNLOCK(ifp);
1447 in6_ifattach(ifp, NULL);
1451 ND_IFINFO(ifp)->flags = ND.flags;
1454 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1455 /* sync kernel routing table with the default router list */
1459 case SIOCSPFXFLUSH_IN6:
1461 /* flush all the prefix advertised by routers */
1462 struct nd_prefix *pr, *next;
1464 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1465 struct in6_ifaddr *ia, *ia_next;
1467 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1470 /* do we really have to remove addresses as well? */
1471 /* XXXRW: in6_ifaddrhead locking. */
1472 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1474 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1477 if (ia->ia6_ndpr == pr)
1478 in6_purgeaddr(&ia->ia_ifa);
1484 case SIOCSRTRFLUSH_IN6:
1486 /* flush all the default routers */
1487 struct nd_defrouter *dr, *next;
1490 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1496 case SIOCGNBRINFO_IN6:
1499 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1501 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1504 IF_AFDATA_RLOCK(ifp);
1505 ln = nd6_lookup(&nb_addr, 0, ifp);
1506 IF_AFDATA_RUNLOCK(ifp);
1512 nbi->state = ln->ln_state;
1513 nbi->asked = ln->la_asked;
1514 nbi->isrouter = ln->ln_router;
1515 if (ln->la_expire == 0)
1518 nbi->expire = ln->la_expire +
1519 (time_second - time_uptime);
1523 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1524 ndif->ifindex = V_nd6_defifindex;
1526 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1527 return (nd6_setdefaultiface(ndif->ifindex));
1533 * Create neighbor cache entry and cache link-layer address,
1534 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1537 * code - type dependent information
1540 * The caller of this function already acquired the ndp
1541 * cache table lock because the cache entry is returned.
1544 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1545 int lladdrlen, int type, int code)
1547 struct llentry *ln = NULL;
1554 uint16_t router = 0;
1555 struct sockaddr_in6 sin6;
1556 struct mbuf *chain = NULL;
1557 int static_route = 0;
1559 IF_AFDATA_UNLOCK_ASSERT(ifp);
1561 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1562 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1564 /* nothing must be updated for unspecified address */
1565 if (IN6_IS_ADDR_UNSPECIFIED(from))
1569 * Validation about ifp->if_addrlen and lladdrlen must be done in
1572 * XXX If the link does not have link-layer adderss, what should
1573 * we do? (ifp->if_addrlen == 0)
1574 * Spec says nothing in sections for RA, RS and NA. There's small
1575 * description on it in NS section (RFC 2461 7.2.3).
1577 flags = lladdr ? ND6_EXCLUSIVE : 0;
1578 IF_AFDATA_RLOCK(ifp);
1579 ln = nd6_lookup(from, flags, ifp);
1580 IF_AFDATA_RUNLOCK(ifp);
1582 flags |= ND6_EXCLUSIVE;
1583 IF_AFDATA_LOCK(ifp);
1584 ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1585 IF_AFDATA_UNLOCK(ifp);
1588 /* do nothing if static ndp is set */
1589 if (ln->la_flags & LLE_STATIC) {
1598 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1599 if (olladdr && lladdr) {
1600 llchange = bcmp(lladdr, &ln->ll_addr,
1606 * newentry olladdr lladdr llchange (*=record)
1609 * 0 n y -- (3) * STALE
1611 * 0 y y y (5) * STALE
1612 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1613 * 1 -- y -- (7) * STALE
1616 if (lladdr) { /* (3-5) and (7) */
1618 * Record source link-layer address
1619 * XXX is it dependent to ifp->if_type?
1621 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1622 ln->la_flags |= LLE_VALID;
1623 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1627 if ((!olladdr && lladdr != NULL) || /* (3) */
1628 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1630 newstate = ND6_LLINFO_STALE;
1631 } else /* (1-2,4) */
1635 if (lladdr == NULL) /* (6) */
1636 newstate = ND6_LLINFO_NOSTATE;
1638 newstate = ND6_LLINFO_STALE;
1643 * Update the state of the neighbor cache.
1645 ln->ln_state = newstate;
1647 if (ln->ln_state == ND6_LLINFO_STALE) {
1649 * XXX: since nd6_output() below will cause
1650 * state tansition to DELAY and reset the timer,
1651 * we must set the timer now, although it is actually
1654 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1657 struct mbuf *m_hold, *m_hold_next;
1660 * reset the la_hold in advance, to explicitly
1661 * prevent a la_hold lookup in nd6_output()
1662 * (wouldn't happen, though...)
1664 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1665 m_hold; m_hold = m_hold_next) {
1666 m_hold_next = m_hold->m_nextpkt;
1667 m_hold->m_nextpkt = NULL;
1670 * we assume ifp is not a p2p here, so
1671 * just set the 2nd argument as the
1674 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1677 * If we have mbufs in the chain we need to do
1678 * deferred transmit. Copy the address from the
1679 * llentry before dropping the lock down below.
1682 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1684 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1685 /* probe right away */
1686 nd6_llinfo_settimer_locked((void *)ln, 0);
1691 * ICMP6 type dependent behavior.
1693 * NS: clear IsRouter if new entry
1694 * RS: clear IsRouter
1695 * RA: set IsRouter if there's lladdr
1696 * redir: clear IsRouter if new entry
1699 * The spec says that we must set IsRouter in the following cases:
1700 * - If lladdr exist, set IsRouter. This means (1-5).
1701 * - If it is old entry (!newentry), set IsRouter. This means (7).
1702 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1703 * A quetion arises for (1) case. (1) case has no lladdr in the
1704 * neighbor cache, this is similar to (6).
1705 * This case is rare but we figured that we MUST NOT set IsRouter.
1707 * newentry olladdr lladdr llchange NS RS RA redir
1709 * 0 n n -- (1) c ? s
1710 * 0 y n -- (2) c s s
1711 * 0 n y -- (3) c s s
1714 * 1 -- n -- (6) c c c s
1715 * 1 -- y -- (7) c c s c s
1719 switch (type & 0xff) {
1720 case ND_NEIGHBOR_SOLICIT:
1722 * New entry must have is_router flag cleared.
1724 if (is_newentry) /* (6-7) */
1729 * If the icmp is a redirect to a better router, always set the
1730 * is_router flag. Otherwise, if the entry is newly created,
1731 * clear the flag. [RFC 2461, sec 8.3]
1733 if (code == ND_REDIRECT_ROUTER)
1735 else if (is_newentry) /* (6-7) */
1738 case ND_ROUTER_SOLICIT:
1740 * is_router flag must always be cleared.
1744 case ND_ROUTER_ADVERT:
1746 * Mark an entry with lladdr as a router.
1748 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1749 (is_newentry && lladdr)) { /* (7) */
1756 static_route = (ln->la_flags & LLE_STATIC);
1757 router = ln->ln_router;
1759 if (flags & ND6_EXCLUSIVE)
1767 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1770 * When the link-layer address of a router changes, select the
1771 * best router again. In particular, when the neighbor entry is newly
1772 * created, it might affect the selection policy.
1773 * Question: can we restrict the first condition to the "is_newentry"
1775 * XXX: when we hear an RA from a new router with the link-layer
1776 * address option, defrouter_select() is called twice, since
1777 * defrtrlist_update called the function as well. However, I believe
1778 * we can compromise the overhead, since it only happens the first
1780 * XXX: although defrouter_select() should not have a bad effect
1781 * for those are not autoconfigured hosts, we explicitly avoid such
1784 if (do_update && router &&
1785 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1787 * guaranteed recursion
1795 if (flags & ND6_EXCLUSIVE)
1806 nd6_slowtimo(void *arg)
1808 CURVNET_SET((struct vnet *) arg);
1809 struct nd_ifinfo *nd6if;
1812 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1813 nd6_slowtimo, curvnet);
1814 IFNET_RLOCK_NOSLEEP();
1815 TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1816 if (ifp->if_afdata[AF_INET6] == NULL)
1818 nd6if = ND_IFINFO(ifp);
1819 if (nd6if->basereachable && /* already initialized */
1820 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1822 * Since reachable time rarely changes by router
1823 * advertisements, we SHOULD insure that a new random
1824 * value gets recomputed at least once every few hours.
1827 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1828 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1831 IFNET_RUNLOCK_NOSLEEP();
1836 * IPv6 packet output - light version.
1837 * Checks if destination LLE exists and is in proper state
1838 * (e.g no modification required). If not true, fall back to
1842 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1843 struct sockaddr_in6 *dst, struct rtentry *rt0)
1845 struct llentry *ln = NULL;
1848 /* discard the packet if IPv6 operation is disabled on the interface */
1849 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1851 return (ENETDOWN); /* better error? */
1854 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1857 if (nd6_need_cache(ifp) == 0)
1860 IF_AFDATA_RLOCK(ifp);
1861 ln = nd6_lookup(&dst->sin6_addr, 0, ifp);
1862 IF_AFDATA_RUNLOCK(ifp);
1865 * Perform fast path for the following cases:
1866 * 1) lle state is REACHABLE
1867 * 2) lle state is DELAY (NS message sentNS message sent)
1869 * Every other case involves lle modification, so we handle
1872 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
1873 ln->ln_state != ND6_LLINFO_DELAY)) {
1874 /* Fall back to slow processing path */
1877 return (nd6_output_lle(ifp, origifp, m, dst, rt0, NULL, NULL));
1885 mac_netinet6_nd6_send(ifp, m);
1889 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1890 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1891 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1892 * to be diverted to user space. When re-injected into the kernel,
1893 * send_output() will directly dispatch them to the outgoing interface.
1895 if (send_sendso_input_hook != NULL) {
1897 struct ip6_hdr *ip6;
1899 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1901 ip6 = mtod(m, struct ip6_hdr *);
1902 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1903 /* Use the SEND socket */
1904 error = send_sendso_input_hook(m, ifp, SND_OUT,
1906 /* -1 == no app on SEND socket */
1907 if (error == 0 || error != -1)
1912 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1913 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1914 mtod(m, struct ip6_hdr *));
1916 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1919 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1925 * Output IPv6 packet - heavy version.
1926 * Function assume that either
1927 * 1) destination LLE does not exist, is invalid or stale, so
1928 * ND6_EXCLUSIVE lock needs to be acquired
1929 * 2) destination lle is provided (with ND6_EXCLUSIVE lock),
1930 * in that case packets are queued in &chain.
1934 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1935 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1936 struct mbuf **chain)
1939 struct ip6_hdr *ip6;
1948 LLE_WLOCK_ASSERT(lle);
1950 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1953 KASSERT(m != NULL, ("NULL mbuf, nothing to send"));
1954 /* discard the packet if IPv6 operation is disabled on the interface */
1955 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1957 return (ENETDOWN); /* better error? */
1963 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1966 if (nd6_need_cache(ifp) == 0)
1970 * Address resolution or Neighbor Unreachability Detection
1972 * At this point, the destination of the packet must be a unicast
1973 * or an anycast address(i.e. not a multicast).
1976 IF_AFDATA_RLOCK(ifp);
1977 lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
1978 IF_AFDATA_RUNLOCK(ifp);
1979 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1981 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1982 * the condition below is not very efficient. But we believe
1983 * it is tolerable, because this should be a rare case.
1985 flags = ND6_CREATE | ND6_EXCLUSIVE;
1986 IF_AFDATA_LOCK(ifp);
1987 lle = nd6_lookup(&dst->sin6_addr, flags, ifp);
1988 IF_AFDATA_UNLOCK(ifp);
1992 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1993 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1994 char ip6buf[INET6_ADDRSTRLEN];
1996 "nd6_output: can't allocate llinfo for %s "
1998 ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2002 goto sendpkt; /* send anyway */
2005 LLE_WLOCK_ASSERT(lle);
2007 /* We don't have to do link-layer address resolution on a p2p link. */
2008 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2009 lle->ln_state < ND6_LLINFO_REACHABLE) {
2010 lle->ln_state = ND6_LLINFO_STALE;
2011 nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz);
2015 * The first time we send a packet to a neighbor whose entry is
2016 * STALE, we have to change the state to DELAY and a sets a timer to
2017 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2018 * neighbor unreachability detection on expiration.
2021 if (lle->ln_state == ND6_LLINFO_STALE) {
2023 lle->ln_state = ND6_LLINFO_DELAY;
2024 nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
2028 * If the neighbor cache entry has a state other than INCOMPLETE
2029 * (i.e. its link-layer address is already resolved), just
2032 if (lle->ln_state > ND6_LLINFO_INCOMPLETE)
2036 * There is a neighbor cache entry, but no ethernet address
2037 * response yet. Append this latest packet to the end of the
2038 * packet queue in the mbuf, unless the number of the packet
2039 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
2040 * the oldest packet in the queue will be removed.
2042 if (lle->ln_state == ND6_LLINFO_NOSTATE)
2043 lle->ln_state = ND6_LLINFO_INCOMPLETE;
2045 if (lle->la_hold != NULL) {
2046 struct mbuf *m_hold;
2050 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2052 if (m_hold->m_nextpkt == NULL) {
2053 m_hold->m_nextpkt = m;
2057 while (i >= V_nd6_maxqueuelen) {
2058 m_hold = lle->la_hold;
2059 lle->la_hold = lle->la_hold->m_nextpkt;
2068 * If there has been no NS for the neighbor after entering the
2069 * INCOMPLETE state, send the first solicitation.
2071 if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) {
2074 nd6_llinfo_settimer_locked(lle,
2075 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2077 nd6_ns_output(ifp, NULL, &dst->sin6_addr, lle, 0);
2080 } else if (has_lle == 0) {
2082 * We did the lookup (no lle arg) so we
2083 * need to do the unlock here.
2092 * ln is valid and the caller did not pass in
2095 if (lle != NULL && has_lle == 0)
2099 mac_netinet6_nd6_send(ifp, m);
2103 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2104 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2105 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2106 * to be diverted to user space. When re-injected into the kernel,
2107 * send_output() will directly dispatch them to the outgoing interface.
2109 if (send_sendso_input_hook != NULL) {
2110 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2112 ip6 = mtod(m, struct ip6_hdr *);
2113 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2114 /* Use the SEND socket */
2115 error = send_sendso_input_hook(m, ifp, SND_OUT,
2117 /* -1 == no app on SEND socket */
2118 if (error == 0 || error != -1)
2124 * We were passed in a pointer to an lle with the lock held
2125 * this means that we can't call if_output as we will
2126 * recurse on the lle lock - so what we do is we create
2127 * a list of mbufs to send and transmit them in the caller
2128 * after the lock is dropped
2137 * append mbuf to end of deferred chain
2140 while (mb->m_nextpkt != NULL)
2146 m_clrprotoflags(m); /* Avoid confusing lower layers. */
2147 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2148 mtod(m, struct ip6_hdr *));
2150 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2153 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
2159 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2160 struct sockaddr_in6 *dst, struct route *ro)
2162 struct mbuf *m, *m_head;
2163 struct ifnet *outifp;
2167 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2174 m_head = m_head->m_nextpkt;
2175 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2180 * note that intermediate errors are blindly ignored - but this is
2181 * the same convention as used with nd6_output when called by
2189 nd6_need_cache(struct ifnet *ifp)
2192 * XXX: we currently do not make neighbor cache on any interface
2193 * other than ARCnet, Ethernet, FDDI and GIF.
2196 * - unidirectional tunnels needs no ND
2198 switch (ifp->if_type) {
2206 #ifdef IFT_IEEE80211
2209 case IFT_INFINIBAND:
2210 case IFT_GIF: /* XXX need more cases? */
2214 case IFT_PROPVIRTUAL:
2222 * Add pernament ND6 link-layer record for given
2223 * interface address.
2225 * Very similar to IPv4 arp_ifinit(), but:
2226 * 1) IPv6 DAD is performed in different place
2227 * 2) It is called by IPv6 protocol stack in contrast to
2228 * arp_ifinit() which is typically called in SIOCSIFADDR
2229 * driver ioctl handler.
2233 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2238 ifp = ia->ia_ifa.ifa_ifp;
2239 IF_AFDATA_LOCK(ifp);
2240 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2241 ln = lla_lookup(LLTABLE6(ifp), (LLE_CREATE | LLE_IFADDR |
2242 LLE_EXCLUSIVE), (struct sockaddr *)&ia->ia_addr);
2243 IF_AFDATA_UNLOCK(ifp);
2245 ln->la_expire = 0; /* for IPv6 this means permanent */
2246 ln->ln_state = ND6_LLINFO_REACHABLE;
2248 in6_newaddrmsg(ia, RTM_ADD);
2256 * Removes ALL lle records for interface address prefix.
2257 * XXXME: That's probably not we really want to do, we need
2258 * to remove address record only and keep other records
2259 * until we determine if given prefix is really going
2263 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2265 struct sockaddr_in6 mask, addr;
2268 in6_newaddrmsg(ia, RTM_DELETE);
2270 ifp = ia->ia_ifa.ifa_ifp;
2271 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2272 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2273 lltable_prefix_free(AF_INET6, (struct sockaddr *)&addr,
2274 (struct sockaddr *)&mask, LLE_STATIC);
2278 * the callers of this function need to be re-worked to drop
2279 * the lle lock, drop here for now
2282 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2283 const struct sockaddr *dst, u_char *desten, struct llentry **lle)
2288 IF_AFDATA_UNLOCK_ASSERT(ifp);
2289 if (m != NULL && m->m_flags & M_MCAST) {
2292 switch (ifp->if_type) {
2298 #ifdef IFT_IEEE80211
2303 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2308 * netbsd can use if_broadcastaddr, but we don't do so
2309 * to reduce # of ifdef.
2311 for (i = 0; i < ifp->if_addrlen; i++)
2319 return (EAFNOSUPPORT);
2325 * the entry should have been created in nd6_store_lladdr
2327 IF_AFDATA_RLOCK(ifp);
2328 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2329 IF_AFDATA_RUNLOCK(ifp);
2330 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2333 /* this could happen, if we could not allocate memory */
2338 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2342 * A *small* use after free race exists here
2348 clear_llinfo_pqueue(struct llentry *ln)
2350 struct mbuf *m_hold, *m_hold_next;
2352 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2353 m_hold_next = m_hold->m_nextpkt;
2361 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2362 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2364 SYSCTL_DECL(_net_inet6_icmp6);
2366 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2367 CTLFLAG_RD, nd6_sysctl_drlist, "");
2368 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2369 CTLFLAG_RD, nd6_sysctl_prlist, "");
2370 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2371 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2374 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2376 struct in6_defrouter d;
2377 struct nd_defrouter *dr;
2383 bzero(&d, sizeof(d));
2384 d.rtaddr.sin6_family = AF_INET6;
2385 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2390 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2391 d.rtaddr.sin6_addr = dr->rtaddr;
2392 error = sa6_recoverscope(&d.rtaddr);
2395 d.flags = dr->flags;
2396 d.rtlifetime = dr->rtlifetime;
2397 d.expire = dr->expire + (time_second - time_uptime);
2398 d.if_index = dr->ifp->if_index;
2399 error = SYSCTL_OUT(req, &d, sizeof(d));
2407 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2409 struct in6_prefix p;
2410 struct sockaddr_in6 s6;
2411 struct nd_prefix *pr;
2412 struct nd_pfxrouter *pfr;
2415 char ip6buf[INET6_ADDRSTRLEN];
2420 bzero(&p, sizeof(p));
2421 p.origin = PR_ORIG_RA;
2422 bzero(&s6, sizeof(s6));
2423 s6.sin6_family = AF_INET6;
2424 s6.sin6_len = sizeof(s6);
2429 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2430 p.prefix = pr->ndpr_prefix;
2431 if (sa6_recoverscope(&p.prefix)) {
2432 log(LOG_ERR, "scope error in prefix list (%s)\n",
2433 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2434 /* XXX: press on... */
2436 p.raflags = pr->ndpr_raf;
2437 p.prefixlen = pr->ndpr_plen;
2438 p.vltime = pr->ndpr_vltime;
2439 p.pltime = pr->ndpr_pltime;
2440 p.if_index = pr->ndpr_ifp->if_index;
2441 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2444 /* XXX: we assume time_t is signed. */
2446 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2447 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2448 p.expire = pr->ndpr_lastupdate +
2450 (time_second - time_uptime);
2452 p.expire = maxexpire;
2454 p.refcnt = pr->ndpr_refcnt;
2455 p.flags = pr->ndpr_stateflags;
2457 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2459 error = SYSCTL_OUT(req, &p, sizeof(p));
2462 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2463 s6.sin6_addr = pfr->router->rtaddr;
2464 if (sa6_recoverscope(&s6))
2466 "scope error in prefix list (%s)\n",
2467 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2468 error = SYSCTL_OUT(req, &s6, sizeof(s6));