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 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/send.h>
79 #include <sys/limits.h>
81 #include <security/mac/mac_framework.h>
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
86 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
89 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */
90 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */
91 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */
92 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */
93 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for
95 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
98 /* preventing too many loops in ND option parsing */
99 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
101 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper
103 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
105 #define V_nd6_maxndopt VNET(nd6_maxndopt)
106 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
109 VNET_DEFINE(int, nd6_debug) = 1;
111 VNET_DEFINE(int, nd6_debug) = 0;
114 static eventhandler_tag lle_event_eh;
118 static int nd6_inuse, nd6_allocated;
121 VNET_DEFINE(struct nd_drhead, nd_defrouter);
122 VNET_DEFINE(struct nd_prhead, nd_prefix);
124 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
125 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
127 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
129 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *,
131 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
132 static void nd6_slowtimo(void *);
133 static int regen_tmpaddr(struct in6_ifaddr *);
134 static struct llentry *nd6_free(struct llentry *, int);
135 static void nd6_llinfo_timer(void *);
136 static void clear_llinfo_pqueue(struct llentry *);
137 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
138 static int nd6_output_lle(struct ifnet *, struct ifnet *, struct mbuf *,
139 struct sockaddr_in6 *);
140 static int nd6_output_ifp(struct ifnet *, struct ifnet *, struct mbuf *,
141 struct sockaddr_in6 *);
143 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
144 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
146 VNET_DEFINE(struct callout, nd6_timer_ch);
149 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
151 struct rt_addrinfo rtinfo;
152 struct sockaddr_in6 dst;
153 struct sockaddr_dl gw;
157 LLE_WLOCK_ASSERT(lle);
159 if (lltable_get_af(lle->lle_tbl) != AF_INET6)
163 case LLENTRY_RESOLVED:
165 KASSERT(lle->la_flags & LLE_VALID,
166 ("%s: %p resolved but not valid?", __func__, lle));
168 case LLENTRY_EXPIRED:
175 ifp = lltable_get_ifp(lle->lle_tbl);
177 bzero(&dst, sizeof(dst));
178 bzero(&gw, sizeof(gw));
179 bzero(&rtinfo, sizeof(rtinfo));
180 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
181 dst.sin6_scope_id = in6_getscopezone(ifp,
182 in6_addrscope(&dst.sin6_addr));
183 gw.sdl_len = sizeof(struct sockaddr_dl);
184 gw.sdl_family = AF_LINK;
185 gw.sdl_alen = ifp->if_addrlen;
186 gw.sdl_index = ifp->if_index;
187 gw.sdl_type = ifp->if_type;
188 if (evt == LLENTRY_RESOLVED)
189 bcopy(&lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
190 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
191 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
192 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
193 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
194 type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
201 LIST_INIT(&V_nd_prefix);
203 /* initialization of the default router list */
204 TAILQ_INIT(&V_nd_defrouter);
207 callout_init(&V_nd6_slowtimo_ch, 0);
208 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
209 nd6_slowtimo, curvnet);
212 if (IS_DEFAULT_VNET(curvnet))
213 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
214 NULL, EVENTHANDLER_PRI_ANY);
222 callout_drain(&V_nd6_slowtimo_ch);
223 callout_drain(&V_nd6_timer_ch);
224 if (IS_DEFAULT_VNET(curvnet))
225 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
230 nd6_ifattach(struct ifnet *ifp)
232 struct nd_ifinfo *nd;
234 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
237 nd->chlim = IPV6_DEFHLIM;
238 nd->basereachable = REACHABLE_TIME;
239 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
240 nd->retrans = RETRANS_TIMER;
242 nd->flags = ND6_IFF_PERFORMNUD;
244 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
245 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
246 * default regardless of the V_ip6_auto_linklocal configuration to
247 * give a reasonable default behavior.
249 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
250 (ifp->if_flags & IFF_LOOPBACK))
251 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
253 * A loopback interface does not need to accept RTADV.
254 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
255 * default regardless of the V_ip6_accept_rtadv configuration to
256 * prevent the interface from accepting RA messages arrived
257 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
259 if (V_ip6_accept_rtadv &&
260 !(ifp->if_flags & IFF_LOOPBACK) &&
261 (ifp->if_type != IFT_BRIDGE))
262 nd->flags |= ND6_IFF_ACCEPT_RTADV;
263 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
264 nd->flags |= ND6_IFF_NO_RADR;
266 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
267 nd6_setmtu0(ifp, nd);
273 nd6_ifdetach(struct nd_ifinfo *nd)
280 * Reset ND level link MTU. This function is called when the physical MTU
281 * changes, which means we might have to adjust the ND level MTU.
284 nd6_setmtu(struct ifnet *ifp)
287 nd6_setmtu0(ifp, ND_IFINFO(ifp));
290 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
292 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
296 omaxmtu = ndi->maxmtu;
298 switch (ifp->if_type) {
300 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
303 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
306 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
309 ndi->maxmtu = ifp->if_mtu;
314 * Decreasing the interface MTU under IPV6 minimum MTU may cause
315 * undesirable situation. We thus notify the operator of the change
316 * explicitly. The check for omaxmtu is necessary to restrict the
317 * log to the case of changing the MTU, not initializing it.
319 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
320 log(LOG_NOTICE, "nd6_setmtu0: "
321 "new link MTU on %s (%lu) is too small for IPv6\n",
322 if_name(ifp), (unsigned long)ndi->maxmtu);
325 if (ndi->maxmtu > V_in6_maxmtu)
326 in6_setmaxmtu(); /* check all interfaces just in case */
331 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
334 bzero(ndopts, sizeof(*ndopts));
335 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
337 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
340 ndopts->nd_opts_done = 1;
341 ndopts->nd_opts_search = NULL;
346 * Take one ND option.
349 nd6_option(union nd_opts *ndopts)
351 struct nd_opt_hdr *nd_opt;
354 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
355 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
357 if (ndopts->nd_opts_search == NULL)
359 if (ndopts->nd_opts_done)
362 nd_opt = ndopts->nd_opts_search;
364 /* make sure nd_opt_len is inside the buffer */
365 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
366 bzero(ndopts, sizeof(*ndopts));
370 olen = nd_opt->nd_opt_len << 3;
373 * Message validation requires that all included
374 * options have a length that is greater than zero.
376 bzero(ndopts, sizeof(*ndopts));
380 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
381 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
382 /* option overruns the end of buffer, invalid */
383 bzero(ndopts, sizeof(*ndopts));
385 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
386 /* reached the end of options chain */
387 ndopts->nd_opts_done = 1;
388 ndopts->nd_opts_search = NULL;
394 * Parse multiple ND options.
395 * This function is much easier to use, for ND routines that do not need
396 * multiple options of the same type.
399 nd6_options(union nd_opts *ndopts)
401 struct nd_opt_hdr *nd_opt;
404 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
405 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
407 if (ndopts->nd_opts_search == NULL)
411 nd_opt = nd6_option(ndopts);
412 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
414 * Message validation requires that all included
415 * options have a length that is greater than zero.
417 ICMP6STAT_INC(icp6s_nd_badopt);
418 bzero(ndopts, sizeof(*ndopts));
425 switch (nd_opt->nd_opt_type) {
426 case ND_OPT_SOURCE_LINKADDR:
427 case ND_OPT_TARGET_LINKADDR:
429 case ND_OPT_REDIRECTED_HEADER:
431 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
433 "duplicated ND6 option found (type=%d)\n",
434 nd_opt->nd_opt_type));
437 ndopts->nd_opt_array[nd_opt->nd_opt_type]
441 case ND_OPT_PREFIX_INFORMATION:
442 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
443 ndopts->nd_opt_array[nd_opt->nd_opt_type]
446 ndopts->nd_opts_pi_end =
447 (struct nd_opt_prefix_info *)nd_opt;
449 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
450 case ND_OPT_RDNSS: /* RFC 6106 */
451 case ND_OPT_DNSSL: /* RFC 6106 */
453 * Silently ignore options we know and do not care about
459 * Unknown options must be silently ignored,
460 * to accomodate future extension to the protocol.
463 "nd6_options: unsupported option %d - "
464 "option ignored\n", nd_opt->nd_opt_type));
469 if (i > V_nd6_maxndopt) {
470 ICMP6STAT_INC(icp6s_nd_toomanyopt);
471 nd6log((LOG_INFO, "too many loop in nd opt\n"));
475 if (ndopts->nd_opts_done)
483 * ND6 timer routine to handle ND6 entries
486 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
490 LLE_WLOCK_ASSERT(ln);
495 canceled = callout_stop(&ln->lle_timer);
497 ln->la_expire = time_uptime + tick / hz;
499 if (tick > INT_MAX) {
500 ln->ln_ntick = tick - INT_MAX;
501 canceled = callout_reset(&ln->lle_timer, INT_MAX,
502 nd6_llinfo_timer, ln);
505 canceled = callout_reset(&ln->lle_timer, tick,
506 nd6_llinfo_timer, ln);
514 * Gets source address of the first packet in hold queue
515 * and stores it in @src.
516 * Returns pointer to @src (if hold queue is not empty) or NULL.
519 static struct in6_addr *
520 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
525 if (ln->la_hold == NULL)
529 * assume every packet in la_hold has the same IP header
532 if (sizeof(hdr) < m->m_len)
535 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
542 nd6_llinfo_settimer(struct llentry *ln, long tick)
546 nd6_llinfo_settimer_locked(ln, tick);
551 nd6_llinfo_timer(void *arg)
554 struct in6_addr *dst, *pdst, *psrc, src;
556 struct nd_ifinfo *ndi = NULL;
559 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
560 ln = (struct llentry *)arg;
562 if (callout_pending(&ln->lle_timer)) {
564 * Here we are a bit odd here in the treatment of
565 * active/pending. If the pending bit is set, it got
566 * rescheduled before I ran. The active
567 * bit we ignore, since if it was stopped
568 * in ll_tablefree() and was currently running
569 * it would have return 0 so the code would
570 * not have deleted it since the callout could
571 * not be stopped so we want to go through
572 * with the delete here now. If the callout
573 * was restarted, the pending bit will be back on and
574 * we just want to bail since the callout_reset would
575 * return 1 and our reference would have been removed
576 * by nd6_llinfo_settimer_locked above since canceled
582 ifp = ln->lle_tbl->llt_ifp;
583 CURVNET_SET(ifp->if_vnet);
584 ndi = ND_IFINFO(ifp);
586 dst = &ln->r_l3addr.addr6;
589 if (ln->ln_ntick > 0) {
590 if (ln->ln_ntick > INT_MAX) {
591 ln->ln_ntick -= INT_MAX;
592 nd6_llinfo_settimer_locked(ln, INT_MAX);
595 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
600 if (ln->la_flags & LLE_STATIC) {
604 if (ln->la_flags & LLE_DELETED) {
605 (void)nd6_free(ln, 0);
610 switch (ln->ln_state) {
611 case ND6_LLINFO_INCOMPLETE:
612 if (ln->la_asked < V_nd6_mmaxtries) {
615 /* Send NS to multicast address */
618 struct mbuf *m = ln->la_hold;
623 * assuming every packet in la_hold has the
624 * same IP header. Send error after unlock.
629 clear_llinfo_pqueue(ln);
631 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
632 (void)nd6_free(ln, 0);
635 icmp6_error2(m, ICMP6_DST_UNREACH,
636 ICMP6_DST_UNREACH_ADDR, 0, ifp);
639 case ND6_LLINFO_REACHABLE:
640 if (!ND6_LLINFO_PERMANENT(ln)) {
641 ln->ln_state = ND6_LLINFO_STALE;
642 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
646 case ND6_LLINFO_STALE:
647 /* Garbage Collection(RFC 2461 5.3) */
648 if (!ND6_LLINFO_PERMANENT(ln)) {
649 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
650 (void)nd6_free(ln, 1);
655 case ND6_LLINFO_DELAY:
656 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
659 ln->ln_state = ND6_LLINFO_PROBE;
662 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
663 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
666 case ND6_LLINFO_PROBE:
667 if (ln->la_asked < V_nd6_umaxtries) {
671 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
672 (void)nd6_free(ln, 0);
677 panic("%s: paths in a dark night can be confusing: %d",
678 __func__, ln->ln_state);
682 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
683 psrc = nd6_llinfo_get_holdsrc(ln, &src);
686 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
696 * ND6 timer routine to expire default route list and prefix list
701 CURVNET_SET((struct vnet *) arg);
702 struct nd_defrouter *dr, *ndr;
703 struct nd_prefix *pr, *npr;
704 struct in6_ifaddr *ia6, *nia6;
706 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
709 /* expire default router list */
710 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
711 if (dr->expire && dr->expire < time_uptime)
716 * expire interface addresses.
717 * in the past the loop was inside prefix expiry processing.
718 * However, from a stricter speci-confrmance standpoint, we should
719 * rather separate address lifetimes and prefix lifetimes.
721 * XXXRW: in6_ifaddrhead locking.
724 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
725 /* check address lifetime */
726 if (IFA6_IS_INVALID(ia6)) {
730 * If the expiring address is temporary, try
731 * regenerating a new one. This would be useful when
732 * we suspended a laptop PC, then turned it on after a
733 * period that could invalidate all temporary
734 * addresses. Although we may have to restart the
735 * loop (see below), it must be after purging the
736 * address. Otherwise, we'd see an infinite loop of
739 if (V_ip6_use_tempaddr &&
740 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
741 if (regen_tmpaddr(ia6) == 0)
745 in6_purgeaddr(&ia6->ia_ifa);
748 goto addrloop; /* XXX: see below */
749 } else if (IFA6_IS_DEPRECATED(ia6)) {
750 int oldflags = ia6->ia6_flags;
752 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
755 * If a temporary address has just become deprecated,
756 * regenerate a new one if possible.
758 if (V_ip6_use_tempaddr &&
759 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
760 (oldflags & IN6_IFF_DEPRECATED) == 0) {
762 if (regen_tmpaddr(ia6) == 0) {
764 * A new temporary address is
766 * XXX: this means the address chain
767 * has changed while we are still in
768 * the loop. Although the change
769 * would not cause disaster (because
770 * it's not a deletion, but an
771 * addition,) we'd rather restart the
772 * loop just for safety. Or does this
773 * significantly reduce performance??
780 * A new RA might have made a deprecated address
783 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
787 /* expire prefix list */
788 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
790 * check prefix lifetime.
791 * since pltime is just for autoconf, pltime processing for
792 * prefix is not necessary.
794 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
795 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
798 * address expiration and prefix expiration are
799 * separate. NEVER perform in6_purgeaddr here.
808 * ia6 - deprecated/invalidated temporary address
811 regen_tmpaddr(struct in6_ifaddr *ia6)
815 struct in6_ifaddr *public_ifa6 = NULL;
817 ifp = ia6->ia_ifa.ifa_ifp;
819 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
820 struct in6_ifaddr *it6;
822 if (ifa->ifa_addr->sa_family != AF_INET6)
825 it6 = (struct in6_ifaddr *)ifa;
827 /* ignore no autoconf addresses. */
828 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
831 /* ignore autoconf addresses with different prefixes. */
832 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
836 * Now we are looking at an autoconf address with the same
837 * prefix as ours. If the address is temporary and is still
838 * preferred, do not create another one. It would be rare, but
839 * could happen, for example, when we resume a laptop PC after
842 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
843 !IFA6_IS_DEPRECATED(it6)) {
849 * This is a public autoconf address that has the same prefix
850 * as ours. If it is preferred, keep it. We can't break the
851 * loop here, because there may be a still-preferred temporary
852 * address with the prefix.
854 if (!IFA6_IS_DEPRECATED(it6))
857 if (public_ifa6 != NULL)
858 ifa_ref(&public_ifa6->ia_ifa);
859 IF_ADDR_RUNLOCK(ifp);
861 if (public_ifa6 != NULL) {
864 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
865 ifa_free(&public_ifa6->ia_ifa);
866 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
867 " tmp addr,errno=%d\n", e);
870 ifa_free(&public_ifa6->ia_ifa);
878 * Nuke neighbor cache/prefix/default router management table, right before
882 nd6_purge(struct ifnet *ifp)
884 struct nd_defrouter *dr, *ndr;
885 struct nd_prefix *pr, *npr;
888 * Nuke default router list entries toward ifp.
889 * We defer removal of default router list entries that is installed
890 * in the routing table, in order to keep additional side effects as
893 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
901 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
909 /* Nuke prefix list entries toward ifp */
910 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
911 if (pr->ndpr_ifp == ifp) {
913 * Because if_detach() does *not* release prefixes
914 * while purging addresses the reference count will
915 * still be above zero. We therefore reset it to
916 * make sure that the prefix really gets purged.
921 * Previously, pr->ndpr_addr is removed as well,
922 * but I strongly believe we don't have to do it.
923 * nd6_purge() is only called from in6_ifdetach(),
924 * which removes all the associated interface addresses
926 * (jinmei@kame.net 20010129)
932 /* cancel default outgoing interface setting */
933 if (V_nd6_defifindex == ifp->if_index)
934 nd6_setdefaultiface(0);
936 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
937 /* Refresh default router list. */
942 * We do not nuke the neighbor cache entries here any more
943 * because the neighbor cache is kept in if_afdata[AF_INET6].
944 * nd6_purge() is invoked by in6_ifdetach() which is called
945 * from if_detach() where everything gets purged. So let
946 * in6_domifdetach() do the actual L2 table purging work.
951 * the caller acquires and releases the lock on the lltbls
952 * Returns the llentry locked
955 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
957 struct sockaddr_in6 sin6;
961 bzero(&sin6, sizeof(sin6));
962 sin6.sin6_len = sizeof(struct sockaddr_in6);
963 sin6.sin6_family = AF_INET6;
964 sin6.sin6_addr = *addr6;
966 IF_AFDATA_LOCK_ASSERT(ifp);
968 llflags = (flags & ND6_EXCLUSIVE) ? LLE_EXCLUSIVE : 0;
969 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
975 nd6_alloc(struct in6_addr *addr6, int flags, struct ifnet *ifp)
977 struct sockaddr_in6 sin6;
980 bzero(&sin6, sizeof(sin6));
981 sin6.sin6_len = sizeof(struct sockaddr_in6);
982 sin6.sin6_family = AF_INET6;
983 sin6.sin6_addr = *addr6;
985 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
987 ln->ln_state = ND6_LLINFO_NOSTATE;
993 * Test whether a given IPv6 address is a neighbor or not, ignoring
994 * the actual neighbor cache. The neighbor cache is ignored in order
995 * to not reenter the routing code from within itself.
998 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1000 struct nd_prefix *pr;
1001 struct ifaddr *dstaddr;
1004 * A link-local address is always a neighbor.
1005 * XXX: a link does not necessarily specify a single interface.
1007 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1008 struct sockaddr_in6 sin6_copy;
1012 * We need sin6_copy since sa6_recoverscope() may modify the
1016 if (sa6_recoverscope(&sin6_copy))
1017 return (0); /* XXX: should be impossible */
1018 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1020 if (sin6_copy.sin6_scope_id == zone)
1027 * If the address matches one of our addresses,
1028 * it should be a neighbor.
1029 * If the address matches one of our on-link prefixes, it should be a
1032 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1033 if (pr->ndpr_ifp != ifp)
1036 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1039 /* Always use the default FIB here. */
1040 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
1041 0, 0, RT_DEFAULT_FIB);
1045 * This is the case where multiple interfaces
1046 * have the same prefix, but only one is installed
1047 * into the routing table and that prefix entry
1048 * is not the one being examined here. In the case
1049 * where RADIX_MPATH is enabled, multiple route
1050 * entries (of the same rt_key value) will be
1051 * installed because the interface addresses all
1054 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1055 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
1062 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1063 &addr->sin6_addr, &pr->ndpr_mask))
1068 * If the address is assigned on the node of the other side of
1069 * a p2p interface, the address should be a neighbor.
1071 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr, RT_ALL_FIBS);
1072 if (dstaddr != NULL) {
1073 if (dstaddr->ifa_ifp == ifp) {
1081 * If the default router list is empty, all addresses are regarded
1082 * as on-link, and thus, as a neighbor.
1084 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1085 TAILQ_EMPTY(&V_nd_defrouter) &&
1086 V_nd6_defifindex == ifp->if_index) {
1095 * Detect if a given IPv6 address identifies a neighbor on a given link.
1096 * XXX: should take care of the destination of a p2p link?
1099 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1101 struct llentry *lle;
1104 IF_AFDATA_UNLOCK_ASSERT(ifp);
1105 if (nd6_is_new_addr_neighbor(addr, ifp))
1109 * Even if the address matches none of our addresses, it might be
1110 * in the neighbor cache.
1112 IF_AFDATA_RLOCK(ifp);
1113 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1117 IF_AFDATA_RUNLOCK(ifp);
1122 * Free an nd6 llinfo entry.
1123 * Since the function would cause significant changes in the kernel, DO NOT
1124 * make it global, unless you have a strong reason for the change, and are sure
1125 * that the change is safe.
1127 static struct llentry *
1128 nd6_free(struct llentry *ln, int gc)
1130 struct llentry *next;
1131 struct nd_defrouter *dr;
1134 LLE_WLOCK_ASSERT(ln);
1137 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1138 * even though it is not harmful, it was not really necessary.
1142 nd6_llinfo_settimer_locked(ln, -1);
1144 ifp = ln->lle_tbl->llt_ifp;
1146 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1147 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1149 if (dr != NULL && dr->expire &&
1150 ln->ln_state == ND6_LLINFO_STALE && gc) {
1152 * If the reason for the deletion is just garbage
1153 * collection, and the neighbor is an active default
1154 * router, do not delete it. Instead, reset the GC
1155 * timer using the router's lifetime.
1156 * Simply deleting the entry would affect default
1157 * router selection, which is not necessarily a good
1158 * thing, especially when we're using router preference
1160 * XXX: the check for ln_state would be redundant,
1161 * but we intentionally keep it just in case.
1163 if (dr->expire > time_uptime)
1164 nd6_llinfo_settimer_locked(ln,
1165 (dr->expire - time_uptime) * hz);
1167 nd6_llinfo_settimer_locked(ln,
1168 (long)V_nd6_gctimer * hz);
1170 next = LIST_NEXT(ln, lle_next);
1178 * Unreachablity of a router might affect the default
1179 * router selection and on-link detection of advertised
1184 * Temporarily fake the state to choose a new default
1185 * router and to perform on-link determination of
1186 * prefixes correctly.
1187 * Below the state will be set correctly,
1188 * or the entry itself will be deleted.
1190 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1193 if (ln->ln_router || dr) {
1196 * We need to unlock to avoid a LOR with rt6_flush() with the
1197 * rnh and for the calls to pfxlist_onlink_check() and
1198 * defrouter_select() in the block further down for calls
1199 * into nd6_lookup(). We still hold a ref.
1204 * rt6_flush must be called whether or not the neighbor
1205 * is in the Default Router List.
1206 * See a corresponding comment in nd6_na_input().
1208 rt6_flush(&ln->r_l3addr.addr6, ifp);
1213 * Since defrouter_select() does not affect the
1214 * on-link determination and MIP6 needs the check
1215 * before the default router selection, we perform
1218 pfxlist_onlink_check();
1221 * Refresh default router list.
1226 if (ln->ln_router || dr)
1231 * Before deleting the entry, remember the next entry as the
1232 * return value. We need this because pfxlist_onlink_check() above
1233 * might have freed other entries (particularly the old next entry) as
1234 * a side effect (XXX).
1236 next = LIST_NEXT(ln, lle_next);
1239 * Save to unlock. We still hold an extra reference and will not
1240 * free(9) in llentry_free() if someone else holds one as well.
1243 IF_AFDATA_LOCK(ifp);
1246 /* Guard against race with other llentry_free(). */
1247 if (ln->la_flags & LLE_LINKED) {
1251 LLE_FREE_LOCKED(ln);
1253 IF_AFDATA_UNLOCK(ifp);
1259 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1261 * XXX cost-effective methods?
1264 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1269 if ((dst6 == NULL) || (rt == NULL))
1273 IF_AFDATA_RLOCK(ifp);
1274 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1275 IF_AFDATA_RUNLOCK(ifp);
1279 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1283 * if we get upper-layer reachability confirmation many times,
1284 * it is possible we have false information.
1288 if (ln->ln_byhint > V_nd6_maxnudhint) {
1293 ln->ln_state = ND6_LLINFO_REACHABLE;
1294 if (!ND6_LLINFO_PERMANENT(ln)) {
1295 nd6_llinfo_settimer_locked(ln,
1296 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1304 * Rejuvenate this function for routing operations related
1308 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1310 struct sockaddr_in6 *gateway;
1311 struct nd_defrouter *dr;
1314 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1325 * Only indirect routes are interesting.
1327 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1330 * check for default route
1332 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1333 &SIN6(rt_key(rt))->sin6_addr)) {
1335 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1345 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1347 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1348 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1349 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1352 if (ifp->if_afdata[AF_INET6] == NULL)
1353 return (EPFNOSUPPORT);
1355 case OSIOCGIFINFO_IN6:
1357 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1358 bzero(&ND, sizeof(ND));
1359 ND.linkmtu = IN6_LINKMTU(ifp);
1360 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1361 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1362 ND.reachable = ND_IFINFO(ifp)->reachable;
1363 ND.retrans = ND_IFINFO(ifp)->retrans;
1364 ND.flags = ND_IFINFO(ifp)->flags;
1365 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1366 ND.chlim = ND_IFINFO(ifp)->chlim;
1368 case SIOCGIFINFO_IN6:
1369 ND = *ND_IFINFO(ifp);
1371 case SIOCSIFINFO_IN6:
1373 * used to change host variables from userland.
1374 * intented for a use on router to reflect RA configurations.
1376 /* 0 means 'unspecified' */
1377 if (ND.linkmtu != 0) {
1378 if (ND.linkmtu < IPV6_MMTU ||
1379 ND.linkmtu > IN6_LINKMTU(ifp)) {
1383 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1386 if (ND.basereachable != 0) {
1387 int obasereachable = ND_IFINFO(ifp)->basereachable;
1389 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1390 if (ND.basereachable != obasereachable)
1391 ND_IFINFO(ifp)->reachable =
1392 ND_COMPUTE_RTIME(ND.basereachable);
1394 if (ND.retrans != 0)
1395 ND_IFINFO(ifp)->retrans = ND.retrans;
1397 ND_IFINFO(ifp)->chlim = ND.chlim;
1399 case SIOCSIFINFO_FLAGS:
1402 struct in6_ifaddr *ia;
1404 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1405 !(ND.flags & ND6_IFF_IFDISABLED)) {
1406 /* ifdisabled 1->0 transision */
1409 * If the interface is marked as ND6_IFF_IFDISABLED and
1410 * has an link-local address with IN6_IFF_DUPLICATED,
1411 * do not clear ND6_IFF_IFDISABLED.
1412 * See RFC 4862, Section 5.4.5.
1415 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1416 if (ifa->ifa_addr->sa_family != AF_INET6)
1418 ia = (struct in6_ifaddr *)ifa;
1419 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1420 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1423 IF_ADDR_RUNLOCK(ifp);
1426 /* LLA is duplicated. */
1427 ND.flags |= ND6_IFF_IFDISABLED;
1428 log(LOG_ERR, "Cannot enable an interface"
1429 " with a link-local address marked"
1432 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1433 if (ifp->if_flags & IFF_UP)
1436 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1437 (ND.flags & ND6_IFF_IFDISABLED)) {
1438 /* ifdisabled 0->1 transision */
1439 /* Mark all IPv6 address as tentative. */
1441 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1442 if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1444 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1446 if (ifa->ifa_addr->sa_family !=
1449 ia = (struct in6_ifaddr *)ifa;
1450 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1452 IF_ADDR_RUNLOCK(ifp);
1456 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1457 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1458 /* auto_linklocal 0->1 transision */
1460 /* If no link-local address on ifp, configure */
1461 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1462 in6_ifattach(ifp, NULL);
1463 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1464 ifp->if_flags & IFF_UP) {
1466 * When the IF already has
1467 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1468 * address is assigned, and IFF_UP, try to
1472 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1474 if (ifa->ifa_addr->sa_family !=
1477 ia = (struct in6_ifaddr *)ifa;
1478 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1481 IF_ADDR_RUNLOCK(ifp);
1483 /* No LLA is configured. */
1484 in6_ifattach(ifp, NULL);
1488 ND_IFINFO(ifp)->flags = ND.flags;
1491 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1492 /* sync kernel routing table with the default router list */
1496 case SIOCSPFXFLUSH_IN6:
1498 /* flush all the prefix advertised by routers */
1499 struct nd_prefix *pr, *next;
1501 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1502 struct in6_ifaddr *ia, *ia_next;
1504 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1507 /* do we really have to remove addresses as well? */
1508 /* XXXRW: in6_ifaddrhead locking. */
1509 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1511 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1514 if (ia->ia6_ndpr == pr)
1515 in6_purgeaddr(&ia->ia_ifa);
1521 case SIOCSRTRFLUSH_IN6:
1523 /* flush all the default routers */
1524 struct nd_defrouter *dr, *next;
1527 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1533 case SIOCGNBRINFO_IN6:
1536 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1538 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1541 IF_AFDATA_RLOCK(ifp);
1542 ln = nd6_lookup(&nb_addr, 0, ifp);
1543 IF_AFDATA_RUNLOCK(ifp);
1549 nbi->state = ln->ln_state;
1550 nbi->asked = ln->la_asked;
1551 nbi->isrouter = ln->ln_router;
1552 if (ln->la_expire == 0)
1555 nbi->expire = ln->la_expire +
1556 (time_second - time_uptime);
1560 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1561 ndif->ifindex = V_nd6_defifindex;
1563 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1564 return (nd6_setdefaultiface(ndif->ifindex));
1570 * Create neighbor cache entry and cache link-layer address,
1571 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1574 * code - type dependent information
1577 * The caller of this function already acquired the ndp
1578 * cache table lock because the cache entry is returned.
1581 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1582 int lladdrlen, int type, int code)
1584 struct llentry *ln = NULL, *ln_tmp;
1591 uint16_t router = 0;
1592 struct sockaddr_in6 sin6;
1593 struct mbuf *chain = NULL;
1594 int static_route = 0;
1596 IF_AFDATA_UNLOCK_ASSERT(ifp);
1598 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1599 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1601 /* nothing must be updated for unspecified address */
1602 if (IN6_IS_ADDR_UNSPECIFIED(from))
1606 * Validation about ifp->if_addrlen and lladdrlen must be done in
1609 * XXX If the link does not have link-layer adderss, what should
1610 * we do? (ifp->if_addrlen == 0)
1611 * Spec says nothing in sections for RA, RS and NA. There's small
1612 * description on it in NS section (RFC 2461 7.2.3).
1614 flags = lladdr ? ND6_EXCLUSIVE : 0;
1615 IF_AFDATA_RLOCK(ifp);
1616 ln = nd6_lookup(from, flags, ifp);
1617 IF_AFDATA_RUNLOCK(ifp);
1620 flags |= ND6_EXCLUSIVE;
1621 ln = nd6_alloc(from, 0, ifp);
1624 IF_AFDATA_WLOCK(ifp);
1626 /* Prefer any existing lle over newly-created one */
1627 ln_tmp = nd6_lookup(from, ND6_EXCLUSIVE, ifp);
1629 lltable_link_entry(LLTABLE6(ifp), ln);
1630 IF_AFDATA_WUNLOCK(ifp);
1632 /* No existing lle, mark as new entry */
1635 lltable_free_entry(LLTABLE6(ifp), ln);
1640 /* do nothing if static ndp is set */
1641 if ((ln->la_flags & LLE_STATIC)) {
1646 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1647 if (olladdr && lladdr) {
1648 llchange = bcmp(lladdr, &ln->ll_addr,
1654 * newentry olladdr lladdr llchange (*=record)
1657 * 0 n y -- (3) * STALE
1659 * 0 y y y (5) * STALE
1660 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1661 * 1 -- y -- (7) * STALE
1664 if (lladdr) { /* (3-5) and (7) */
1666 * Record source link-layer address
1667 * XXX is it dependent to ifp->if_type?
1669 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1670 ln->la_flags |= LLE_VALID;
1671 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1675 if ((!olladdr && lladdr != NULL) || /* (3) */
1676 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1678 newstate = ND6_LLINFO_STALE;
1679 } else /* (1-2,4) */
1683 if (lladdr == NULL) /* (6) */
1684 newstate = ND6_LLINFO_NOSTATE;
1686 newstate = ND6_LLINFO_STALE;
1691 * Update the state of the neighbor cache.
1693 ln->ln_state = newstate;
1695 if (ln->ln_state == ND6_LLINFO_STALE) {
1696 if (ln->la_hold != NULL)
1697 nd6_grab_holdchain(ln, &chain, &sin6);
1698 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1699 /* probe right away */
1700 nd6_llinfo_settimer_locked((void *)ln, 0);
1705 * ICMP6 type dependent behavior.
1707 * NS: clear IsRouter if new entry
1708 * RS: clear IsRouter
1709 * RA: set IsRouter if there's lladdr
1710 * redir: clear IsRouter if new entry
1713 * The spec says that we must set IsRouter in the following cases:
1714 * - If lladdr exist, set IsRouter. This means (1-5).
1715 * - If it is old entry (!newentry), set IsRouter. This means (7).
1716 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1717 * A quetion arises for (1) case. (1) case has no lladdr in the
1718 * neighbor cache, this is similar to (6).
1719 * This case is rare but we figured that we MUST NOT set IsRouter.
1721 * newentry olladdr lladdr llchange NS RS RA redir
1723 * 0 n n -- (1) c ? s
1724 * 0 y n -- (2) c s s
1725 * 0 n y -- (3) c s s
1728 * 1 -- n -- (6) c c c s
1729 * 1 -- y -- (7) c c s c s
1733 switch (type & 0xff) {
1734 case ND_NEIGHBOR_SOLICIT:
1736 * New entry must have is_router flag cleared.
1738 if (is_newentry) /* (6-7) */
1743 * If the icmp is a redirect to a better router, always set the
1744 * is_router flag. Otherwise, if the entry is newly created,
1745 * clear the flag. [RFC 2461, sec 8.3]
1747 if (code == ND_REDIRECT_ROUTER)
1749 else if (is_newentry) /* (6-7) */
1752 case ND_ROUTER_SOLICIT:
1754 * is_router flag must always be cleared.
1758 case ND_ROUTER_ADVERT:
1760 * Mark an entry with lladdr as a router.
1762 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1763 (is_newentry && lladdr)) { /* (7) */
1770 static_route = (ln->la_flags & LLE_STATIC);
1771 router = ln->ln_router;
1773 if (flags & ND6_EXCLUSIVE)
1781 nd6_flush_holdchain(ifp, ifp, chain, &sin6);
1784 * When the link-layer address of a router changes, select the
1785 * best router again. In particular, when the neighbor entry is newly
1786 * created, it might affect the selection policy.
1787 * Question: can we restrict the first condition to the "is_newentry"
1789 * XXX: when we hear an RA from a new router with the link-layer
1790 * address option, defrouter_select() is called twice, since
1791 * defrtrlist_update called the function as well. However, I believe
1792 * we can compromise the overhead, since it only happens the first
1794 * XXX: although defrouter_select() should not have a bad effect
1795 * for those are not autoconfigured hosts, we explicitly avoid such
1798 if (do_update && router &&
1799 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1801 * guaranteed recursion
1809 if (flags & ND6_EXCLUSIVE)
1820 nd6_slowtimo(void *arg)
1822 CURVNET_SET((struct vnet *) arg);
1823 struct nd_ifinfo *nd6if;
1826 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1827 nd6_slowtimo, curvnet);
1828 IFNET_RLOCK_NOSLEEP();
1829 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1830 if (ifp->if_afdata[AF_INET6] == NULL)
1832 nd6if = ND_IFINFO(ifp);
1833 if (nd6if->basereachable && /* already initialized */
1834 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1836 * Since reachable time rarely changes by router
1837 * advertisements, we SHOULD insure that a new random
1838 * value gets recomputed at least once every few hours.
1841 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1842 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1845 IFNET_RUNLOCK_NOSLEEP();
1850 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
1851 struct sockaddr_in6 *sin6)
1854 LLE_WLOCK_ASSERT(ln);
1856 *chain = ln->la_hold;
1858 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
1860 if (ln->ln_state == ND6_LLINFO_STALE) {
1863 * The first time we send a packet to a
1864 * neighbor whose entry is STALE, we have
1865 * to change the state to DELAY and a sets
1866 * a timer to expire in DELAY_FIRST_PROBE_TIME
1867 * seconds to ensure do neighbor unreachability
1868 * detection on expiration.
1872 ln->ln_state = ND6_LLINFO_DELAY;
1873 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1878 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1879 struct sockaddr_in6 *dst)
1883 struct ip6_hdr *ip6;
1887 mac_netinet6_nd6_send(ifp, m);
1891 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1892 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1893 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1894 * to be diverted to user space. When re-injected into the kernel,
1895 * send_output() will directly dispatch them to the outgoing interface.
1897 if (send_sendso_input_hook != NULL) {
1898 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1900 ip6 = mtod(m, struct ip6_hdr *);
1901 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1902 /* Use the SEND socket */
1903 error = send_sendso_input_hook(m, ifp, SND_OUT,
1905 /* -1 == no app on SEND socket */
1906 if (error == 0 || error != -1)
1911 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1912 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1913 mtod(m, struct ip6_hdr *));
1915 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1918 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1923 * IPv6 packet output - light version.
1924 * Checks if destination LLE exists and is in proper state
1925 * (e.g no modification required). If not true, fall back to
1929 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1930 struct sockaddr_in6 *dst, struct rtentry *rt0)
1932 struct llentry *ln = NULL;
1934 /* discard the packet if IPv6 operation is disabled on the interface */
1935 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1937 return (ENETDOWN); /* better error? */
1940 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1943 if (nd6_need_cache(ifp) == 0)
1946 IF_AFDATA_RLOCK(ifp);
1947 ln = nd6_lookup(&dst->sin6_addr, 0, ifp);
1948 IF_AFDATA_RUNLOCK(ifp);
1951 * Perform fast path for the following cases:
1952 * 1) lle state is REACHABLE
1953 * 2) lle state is DELAY (NS message sentNS message sent)
1955 * Every other case involves lle modification, so we handle
1958 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
1959 ln->ln_state != ND6_LLINFO_DELAY)) {
1960 /* Fall back to slow processing path */
1963 return (nd6_output_lle(ifp, origifp, m, dst));
1970 return (nd6_output_ifp(ifp, origifp, m, dst));
1975 * Output IPv6 packet - heavy version.
1976 * Function assume that either
1977 * 1) destination LLE does not exist, is invalid or stale, so
1978 * ND6_EXCLUSIVE lock needs to be acquired
1979 * 2) destination lle is provided (with ND6_EXCLUSIVE lock),
1980 * in that case packets are queued in &chain.
1984 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1985 struct sockaddr_in6 *dst)
1987 struct llentry *lle = NULL, *lle_tmp;
1989 KASSERT(m != NULL, ("NULL mbuf, nothing to send"));
1990 /* discard the packet if IPv6 operation is disabled on the interface */
1991 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1993 return (ENETDOWN); /* better error? */
1996 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1999 if (nd6_need_cache(ifp) == 0)
2003 * Address resolution or Neighbor Unreachability Detection
2005 * At this point, the destination of the packet must be a unicast
2006 * or an anycast address(i.e. not a multicast).
2009 IF_AFDATA_RLOCK(ifp);
2010 lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
2011 IF_AFDATA_RUNLOCK(ifp);
2012 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2014 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2015 * the condition below is not very efficient. But we believe
2016 * it is tolerable, because this should be a rare case.
2018 lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2020 char ip6buf[INET6_ADDRSTRLEN];
2022 "nd6_output: can't allocate llinfo for %s "
2024 ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2028 lle->ln_state = ND6_LLINFO_NOSTATE;
2030 IF_AFDATA_WLOCK(ifp);
2032 /* Prefer any existing entry over newly-created one */
2033 lle_tmp = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
2034 if (lle_tmp == NULL)
2035 lltable_link_entry(LLTABLE6(ifp), lle);
2036 IF_AFDATA_WUNLOCK(ifp);
2037 if (lle_tmp != NULL) {
2038 lltable_free_entry(LLTABLE6(ifp), lle);
2045 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2046 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2050 goto sendpkt; /* send anyway */
2053 LLE_WLOCK_ASSERT(lle);
2055 /* We don't have to do link-layer address resolution on a p2p link. */
2056 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2057 lle->ln_state < ND6_LLINFO_REACHABLE) {
2058 lle->ln_state = ND6_LLINFO_STALE;
2059 nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz);
2063 * The first time we send a packet to a neighbor whose entry is
2064 * STALE, we have to change the state to DELAY and a sets a timer to
2065 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2066 * neighbor unreachability detection on expiration.
2069 if (lle->ln_state == ND6_LLINFO_STALE) {
2071 lle->ln_state = ND6_LLINFO_DELAY;
2072 nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
2076 * If the neighbor cache entry has a state other than INCOMPLETE
2077 * (i.e. its link-layer address is already resolved), just
2080 if (lle->ln_state > ND6_LLINFO_INCOMPLETE)
2084 * There is a neighbor cache entry, but no ethernet address
2085 * response yet. Append this latest packet to the end of the
2086 * packet queue in the mbuf, unless the number of the packet
2087 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
2088 * the oldest packet in the queue will be removed.
2090 if (lle->ln_state == ND6_LLINFO_NOSTATE)
2091 lle->ln_state = ND6_LLINFO_INCOMPLETE;
2093 if (lle->la_hold != NULL) {
2094 struct mbuf *m_hold;
2098 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2100 if (m_hold->m_nextpkt == NULL) {
2101 m_hold->m_nextpkt = m;
2105 while (i >= V_nd6_maxqueuelen) {
2106 m_hold = lle->la_hold;
2107 lle->la_hold = lle->la_hold->m_nextpkt;
2116 * If there has been no NS for the neighbor after entering the
2117 * INCOMPLETE state, send the first solicitation.
2119 if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) {
2120 struct in6_addr src, *psrc;
2123 nd6_llinfo_settimer_locked(lle,
2124 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2125 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2127 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2129 /* We did the lookup so we need to do the unlock here. */
2139 return (nd6_output_ifp(ifp, origifp, m, dst));
2144 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2145 struct sockaddr_in6 *dst)
2147 struct mbuf *m, *m_head;
2148 struct ifnet *outifp;
2152 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2159 m_head = m_head->m_nextpkt;
2160 error = nd6_output_ifp(ifp, origifp, m, dst);
2165 * note that intermediate errors are blindly ignored - but this is
2166 * the same convention as used with nd6_output when called by
2174 nd6_need_cache(struct ifnet *ifp)
2177 * XXX: we currently do not make neighbor cache on any interface
2178 * other than ARCnet, Ethernet, FDDI and GIF.
2181 * - unidirectional tunnels needs no ND
2183 switch (ifp->if_type) {
2190 case IFT_INFINIBAND:
2192 case IFT_PROPVIRTUAL:
2200 * Add pernament ND6 link-layer record for given
2201 * interface address.
2203 * Very similar to IPv4 arp_ifinit(), but:
2204 * 1) IPv6 DAD is performed in different place
2205 * 2) It is called by IPv6 protocol stack in contrast to
2206 * arp_ifinit() which is typically called in SIOCSIFADDR
2207 * driver ioctl handler.
2211 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2214 struct llentry *ln, *ln_tmp;
2215 struct sockaddr *dst;
2217 ifp = ia->ia_ifa.ifa_ifp;
2218 if (nd6_need_cache(ifp) == 0)
2221 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2222 dst = (struct sockaddr *)&ia->ia_addr;
2223 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2227 IF_AFDATA_WLOCK(ifp);
2229 /* Unlink any entry if exists */
2230 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2232 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2233 lltable_link_entry(LLTABLE6(ifp), ln);
2234 IF_AFDATA_WUNLOCK(ifp);
2237 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2238 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2242 llentry_free(ln_tmp);
2248 * Removes ALL lle records for interface address prefix.
2249 * XXXME: That's probably not we really want to do, we need
2250 * to remove address record only and keep other records
2251 * until we determine if given prefix is really going
2255 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2257 struct sockaddr_in6 mask, addr;
2260 ifp = ia->ia_ifa.ifa_ifp;
2261 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2262 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2263 lltable_prefix_free(AF_INET6, (struct sockaddr *)&addr,
2264 (struct sockaddr *)&mask, LLE_STATIC);
2268 * the callers of this function need to be re-worked to drop
2269 * the lle lock, drop here for now
2272 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2273 const struct sockaddr *dst, u_char *desten, uint32_t *pflags)
2279 IF_AFDATA_UNLOCK_ASSERT(ifp);
2280 if (m != NULL && m->m_flags & M_MCAST) {
2281 switch (ifp->if_type) {
2288 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2293 return (EAFNOSUPPORT);
2299 * the entry should have been created in nd6_store_lladdr
2301 IF_AFDATA_RLOCK(ifp);
2302 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2303 IF_AFDATA_RUNLOCK(ifp);
2304 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2307 /* this could happen, if we could not allocate memory */
2312 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2314 *pflags = ln->la_flags;
2317 * A *small* use after free race exists here
2323 clear_llinfo_pqueue(struct llentry *ln)
2325 struct mbuf *m_hold, *m_hold_next;
2327 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2328 m_hold_next = m_hold->m_nextpkt;
2336 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2337 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2339 SYSCTL_DECL(_net_inet6_icmp6);
2341 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2342 CTLFLAG_RD, nd6_sysctl_drlist, "");
2343 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2344 CTLFLAG_RD, nd6_sysctl_prlist, "");
2345 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2346 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2347 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2348 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2351 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2353 struct in6_defrouter d;
2354 struct nd_defrouter *dr;
2360 bzero(&d, sizeof(d));
2361 d.rtaddr.sin6_family = AF_INET6;
2362 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2367 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2368 d.rtaddr.sin6_addr = dr->rtaddr;
2369 error = sa6_recoverscope(&d.rtaddr);
2372 d.flags = dr->flags;
2373 d.rtlifetime = dr->rtlifetime;
2374 d.expire = dr->expire + (time_second - time_uptime);
2375 d.if_index = dr->ifp->if_index;
2376 error = SYSCTL_OUT(req, &d, sizeof(d));
2384 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2386 struct in6_prefix p;
2387 struct sockaddr_in6 s6;
2388 struct nd_prefix *pr;
2389 struct nd_pfxrouter *pfr;
2392 char ip6buf[INET6_ADDRSTRLEN];
2397 bzero(&p, sizeof(p));
2398 p.origin = PR_ORIG_RA;
2399 bzero(&s6, sizeof(s6));
2400 s6.sin6_family = AF_INET6;
2401 s6.sin6_len = sizeof(s6);
2406 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2407 p.prefix = pr->ndpr_prefix;
2408 if (sa6_recoverscope(&p.prefix)) {
2409 log(LOG_ERR, "scope error in prefix list (%s)\n",
2410 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2411 /* XXX: press on... */
2413 p.raflags = pr->ndpr_raf;
2414 p.prefixlen = pr->ndpr_plen;
2415 p.vltime = pr->ndpr_vltime;
2416 p.pltime = pr->ndpr_pltime;
2417 p.if_index = pr->ndpr_ifp->if_index;
2418 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2421 /* XXX: we assume time_t is signed. */
2423 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2424 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2425 p.expire = pr->ndpr_lastupdate +
2427 (time_second - time_uptime);
2429 p.expire = maxexpire;
2431 p.refcnt = pr->ndpr_refcnt;
2432 p.flags = pr->ndpr_stateflags;
2434 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2436 error = SYSCTL_OUT(req, &p, sizeof(p));
2439 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2440 s6.sin6_addr = pfr->router->rtaddr;
2441 if (sa6_recoverscope(&s6))
2443 "scope error in prefix list (%s)\n",
2444 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2445 error = SYSCTL_OUT(req, &s6, sizeof(s6));