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 void nd6_free(struct llentry *, int);
135 static void nd6_free_redirect(const struct llentry *);
136 static void nd6_llinfo_timer(void *);
137 static void clear_llinfo_pqueue(struct llentry *);
138 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
139 static int nd6_output_lle(struct ifnet *, struct ifnet *, struct mbuf *,
140 struct sockaddr_in6 *);
141 static int nd6_output_ifp(struct ifnet *, struct ifnet *, struct mbuf *,
142 struct sockaddr_in6 *);
144 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
145 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
147 VNET_DEFINE(struct callout, nd6_timer_ch);
150 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
152 struct rt_addrinfo rtinfo;
153 struct sockaddr_in6 dst;
154 struct sockaddr_dl gw;
158 LLE_WLOCK_ASSERT(lle);
160 if (lltable_get_af(lle->lle_tbl) != AF_INET6)
164 case LLENTRY_RESOLVED:
166 KASSERT(lle->la_flags & LLE_VALID,
167 ("%s: %p resolved but not valid?", __func__, lle));
169 case LLENTRY_EXPIRED:
176 ifp = lltable_get_ifp(lle->lle_tbl);
178 bzero(&dst, sizeof(dst));
179 bzero(&gw, sizeof(gw));
180 bzero(&rtinfo, sizeof(rtinfo));
181 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
182 dst.sin6_scope_id = in6_getscopezone(ifp,
183 in6_addrscope(&dst.sin6_addr));
184 gw.sdl_len = sizeof(struct sockaddr_dl);
185 gw.sdl_family = AF_LINK;
186 gw.sdl_alen = ifp->if_addrlen;
187 gw.sdl_index = ifp->if_index;
188 gw.sdl_type = ifp->if_type;
189 if (evt == LLENTRY_RESOLVED)
190 bcopy(&lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
191 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
192 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
193 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
194 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
195 type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
202 LIST_INIT(&V_nd_prefix);
204 /* initialization of the default router list */
205 TAILQ_INIT(&V_nd_defrouter);
208 callout_init(&V_nd6_slowtimo_ch, 0);
209 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
210 nd6_slowtimo, curvnet);
213 if (IS_DEFAULT_VNET(curvnet))
214 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
215 NULL, EVENTHANDLER_PRI_ANY);
223 callout_drain(&V_nd6_slowtimo_ch);
224 callout_drain(&V_nd6_timer_ch);
225 if (IS_DEFAULT_VNET(curvnet))
226 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
231 nd6_ifattach(struct ifnet *ifp)
233 struct nd_ifinfo *nd;
235 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
238 nd->chlim = IPV6_DEFHLIM;
239 nd->basereachable = REACHABLE_TIME;
240 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
241 nd->retrans = RETRANS_TIMER;
243 nd->flags = ND6_IFF_PERFORMNUD;
245 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
246 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
247 * default regardless of the V_ip6_auto_linklocal configuration to
248 * give a reasonable default behavior.
250 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
251 (ifp->if_flags & IFF_LOOPBACK))
252 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
254 * A loopback interface does not need to accept RTADV.
255 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
256 * default regardless of the V_ip6_accept_rtadv configuration to
257 * prevent the interface from accepting RA messages arrived
258 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
260 if (V_ip6_accept_rtadv &&
261 !(ifp->if_flags & IFF_LOOPBACK) &&
262 (ifp->if_type != IFT_BRIDGE))
263 nd->flags |= ND6_IFF_ACCEPT_RTADV;
264 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
265 nd->flags |= ND6_IFF_NO_RADR;
267 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
268 nd6_setmtu0(ifp, nd);
274 nd6_ifdetach(struct nd_ifinfo *nd)
281 * Reset ND level link MTU. This function is called when the physical MTU
282 * changes, which means we might have to adjust the ND level MTU.
285 nd6_setmtu(struct ifnet *ifp)
288 nd6_setmtu0(ifp, ND_IFINFO(ifp));
291 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
293 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
297 omaxmtu = ndi->maxmtu;
299 switch (ifp->if_type) {
301 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
304 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
307 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
310 ndi->maxmtu = ifp->if_mtu;
315 * Decreasing the interface MTU under IPV6 minimum MTU may cause
316 * undesirable situation. We thus notify the operator of the change
317 * explicitly. The check for omaxmtu is necessary to restrict the
318 * log to the case of changing the MTU, not initializing it.
320 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
321 log(LOG_NOTICE, "nd6_setmtu0: "
322 "new link MTU on %s (%lu) is too small for IPv6\n",
323 if_name(ifp), (unsigned long)ndi->maxmtu);
326 if (ndi->maxmtu > V_in6_maxmtu)
327 in6_setmaxmtu(); /* check all interfaces just in case */
332 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
335 bzero(ndopts, sizeof(*ndopts));
336 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
338 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
341 ndopts->nd_opts_done = 1;
342 ndopts->nd_opts_search = NULL;
347 * Take one ND option.
350 nd6_option(union nd_opts *ndopts)
352 struct nd_opt_hdr *nd_opt;
355 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
356 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
358 if (ndopts->nd_opts_search == NULL)
360 if (ndopts->nd_opts_done)
363 nd_opt = ndopts->nd_opts_search;
365 /* make sure nd_opt_len is inside the buffer */
366 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
367 bzero(ndopts, sizeof(*ndopts));
371 olen = nd_opt->nd_opt_len << 3;
374 * Message validation requires that all included
375 * options have a length that is greater than zero.
377 bzero(ndopts, sizeof(*ndopts));
381 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
382 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
383 /* option overruns the end of buffer, invalid */
384 bzero(ndopts, sizeof(*ndopts));
386 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
387 /* reached the end of options chain */
388 ndopts->nd_opts_done = 1;
389 ndopts->nd_opts_search = NULL;
395 * Parse multiple ND options.
396 * This function is much easier to use, for ND routines that do not need
397 * multiple options of the same type.
400 nd6_options(union nd_opts *ndopts)
402 struct nd_opt_hdr *nd_opt;
405 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
406 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
408 if (ndopts->nd_opts_search == NULL)
412 nd_opt = nd6_option(ndopts);
413 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
415 * Message validation requires that all included
416 * options have a length that is greater than zero.
418 ICMP6STAT_INC(icp6s_nd_badopt);
419 bzero(ndopts, sizeof(*ndopts));
426 switch (nd_opt->nd_opt_type) {
427 case ND_OPT_SOURCE_LINKADDR:
428 case ND_OPT_TARGET_LINKADDR:
430 case ND_OPT_REDIRECTED_HEADER:
432 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
434 "duplicated ND6 option found (type=%d)\n",
435 nd_opt->nd_opt_type));
438 ndopts->nd_opt_array[nd_opt->nd_opt_type]
442 case ND_OPT_PREFIX_INFORMATION:
443 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
444 ndopts->nd_opt_array[nd_opt->nd_opt_type]
447 ndopts->nd_opts_pi_end =
448 (struct nd_opt_prefix_info *)nd_opt;
450 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
451 case ND_OPT_RDNSS: /* RFC 6106 */
452 case ND_OPT_DNSSL: /* RFC 6106 */
454 * Silently ignore options we know and do not care about
460 * Unknown options must be silently ignored,
461 * to accomodate future extension to the protocol.
464 "nd6_options: unsupported option %d - "
465 "option ignored\n", nd_opt->nd_opt_type));
470 if (i > V_nd6_maxndopt) {
471 ICMP6STAT_INC(icp6s_nd_toomanyopt);
472 nd6log((LOG_INFO, "too many loop in nd opt\n"));
476 if (ndopts->nd_opts_done)
484 * ND6 timer routine to handle ND6 entries
487 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
491 LLE_WLOCK_ASSERT(ln);
496 canceled = callout_stop(&ln->lle_timer);
498 ln->la_expire = time_uptime + tick / hz;
500 if (tick > INT_MAX) {
501 ln->ln_ntick = tick - INT_MAX;
502 canceled = callout_reset(&ln->lle_timer, INT_MAX,
503 nd6_llinfo_timer, ln);
506 canceled = callout_reset(&ln->lle_timer, tick,
507 nd6_llinfo_timer, ln);
515 * Gets source address of the first packet in hold queue
516 * and stores it in @src.
517 * Returns pointer to @src (if hold queue is not empty) or NULL.
520 static struct in6_addr *
521 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
526 if (ln->la_hold == NULL)
530 * assume every packet in la_hold has the same IP header
533 if (sizeof(hdr) < m->m_len)
536 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
543 nd6_llinfo_settimer(struct llentry *ln, long tick)
547 nd6_llinfo_settimer_locked(ln, tick);
552 nd6_llinfo_timer(void *arg)
555 struct in6_addr *dst, *pdst, *psrc, src;
557 struct nd_ifinfo *ndi = NULL;
560 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
561 ln = (struct llentry *)arg;
563 if (callout_pending(&ln->lle_timer)) {
565 * Here we are a bit odd here in the treatment of
566 * active/pending. If the pending bit is set, it got
567 * rescheduled before I ran. The active
568 * bit we ignore, since if it was stopped
569 * in ll_tablefree() and was currently running
570 * it would have return 0 so the code would
571 * not have deleted it since the callout could
572 * not be stopped so we want to go through
573 * with the delete here now. If the callout
574 * was restarted, the pending bit will be back on and
575 * we just want to bail since the callout_reset would
576 * return 1 and our reference would have been removed
577 * by nd6_llinfo_settimer_locked above since canceled
583 ifp = ln->lle_tbl->llt_ifp;
584 CURVNET_SET(ifp->if_vnet);
585 ndi = ND_IFINFO(ifp);
587 dst = &ln->r_l3addr.addr6;
590 if (ln->ln_ntick > 0) {
591 if (ln->ln_ntick > INT_MAX) {
592 ln->ln_ntick -= INT_MAX;
593 nd6_llinfo_settimer_locked(ln, INT_MAX);
596 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
601 if (ln->la_flags & LLE_STATIC) {
605 if (ln->la_flags & LLE_DELETED) {
611 switch (ln->ln_state) {
612 case ND6_LLINFO_INCOMPLETE:
613 if (ln->la_asked < V_nd6_mmaxtries) {
616 /* Send NS to multicast address */
619 struct mbuf *m = ln->la_hold;
624 * assuming every packet in la_hold has the
625 * same IP header. Send error after unlock.
630 clear_llinfo_pqueue(ln);
632 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
636 icmp6_error2(m, ICMP6_DST_UNREACH,
637 ICMP6_DST_UNREACH_ADDR, 0, ifp);
640 case ND6_LLINFO_REACHABLE:
641 if (!ND6_LLINFO_PERMANENT(ln)) {
642 ln->ln_state = ND6_LLINFO_STALE;
643 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
647 case ND6_LLINFO_STALE:
648 /* Garbage Collection(RFC 2461 5.3) */
649 if (!ND6_LLINFO_PERMANENT(ln)) {
650 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
656 case ND6_LLINFO_DELAY:
657 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
660 ln->ln_state = ND6_LLINFO_PROBE;
663 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
664 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
667 case ND6_LLINFO_PROBE:
668 if (ln->la_asked < V_nd6_umaxtries) {
672 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
678 panic("%s: paths in a dark night can be confusing: %d",
679 __func__, ln->ln_state);
683 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
684 psrc = nd6_llinfo_get_holdsrc(ln, &src);
687 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
697 * ND6 timer routine to expire default route list and prefix list
702 CURVNET_SET((struct vnet *) arg);
703 struct nd_defrouter *dr, *ndr;
704 struct nd_prefix *pr, *npr;
705 struct in6_ifaddr *ia6, *nia6;
707 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
710 /* expire default router list */
711 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
712 if (dr->expire && dr->expire < time_uptime)
717 * expire interface addresses.
718 * in the past the loop was inside prefix expiry processing.
719 * However, from a stricter speci-confrmance standpoint, we should
720 * rather separate address lifetimes and prefix lifetimes.
722 * XXXRW: in6_ifaddrhead locking.
725 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
726 /* check address lifetime */
727 if (IFA6_IS_INVALID(ia6)) {
731 * If the expiring address is temporary, try
732 * regenerating a new one. This would be useful when
733 * we suspended a laptop PC, then turned it on after a
734 * period that could invalidate all temporary
735 * addresses. Although we may have to restart the
736 * loop (see below), it must be after purging the
737 * address. Otherwise, we'd see an infinite loop of
740 if (V_ip6_use_tempaddr &&
741 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
742 if (regen_tmpaddr(ia6) == 0)
746 in6_purgeaddr(&ia6->ia_ifa);
749 goto addrloop; /* XXX: see below */
750 } else if (IFA6_IS_DEPRECATED(ia6)) {
751 int oldflags = ia6->ia6_flags;
753 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
756 * If a temporary address has just become deprecated,
757 * regenerate a new one if possible.
759 if (V_ip6_use_tempaddr &&
760 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
761 (oldflags & IN6_IFF_DEPRECATED) == 0) {
763 if (regen_tmpaddr(ia6) == 0) {
765 * A new temporary address is
767 * XXX: this means the address chain
768 * has changed while we are still in
769 * the loop. Although the change
770 * would not cause disaster (because
771 * it's not a deletion, but an
772 * addition,) we'd rather restart the
773 * loop just for safety. Or does this
774 * significantly reduce performance??
781 * A new RA might have made a deprecated address
784 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
788 /* expire prefix list */
789 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
791 * check prefix lifetime.
792 * since pltime is just for autoconf, pltime processing for
793 * prefix is not necessary.
795 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
796 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
799 * address expiration and prefix expiration are
800 * separate. NEVER perform in6_purgeaddr here.
809 * ia6 - deprecated/invalidated temporary address
812 regen_tmpaddr(struct in6_ifaddr *ia6)
816 struct in6_ifaddr *public_ifa6 = NULL;
818 ifp = ia6->ia_ifa.ifa_ifp;
820 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
821 struct in6_ifaddr *it6;
823 if (ifa->ifa_addr->sa_family != AF_INET6)
826 it6 = (struct in6_ifaddr *)ifa;
828 /* ignore no autoconf addresses. */
829 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
832 /* ignore autoconf addresses with different prefixes. */
833 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
837 * Now we are looking at an autoconf address with the same
838 * prefix as ours. If the address is temporary and is still
839 * preferred, do not create another one. It would be rare, but
840 * could happen, for example, when we resume a laptop PC after
843 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
844 !IFA6_IS_DEPRECATED(it6)) {
850 * This is a public autoconf address that has the same prefix
851 * as ours. If it is preferred, keep it. We can't break the
852 * loop here, because there may be a still-preferred temporary
853 * address with the prefix.
855 if (!IFA6_IS_DEPRECATED(it6))
858 if (public_ifa6 != NULL)
859 ifa_ref(&public_ifa6->ia_ifa);
860 IF_ADDR_RUNLOCK(ifp);
862 if (public_ifa6 != NULL) {
865 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
866 ifa_free(&public_ifa6->ia_ifa);
867 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
868 " tmp addr,errno=%d\n", e);
871 ifa_free(&public_ifa6->ia_ifa);
879 * Nuke neighbor cache/prefix/default router management table, right before
883 nd6_purge(struct ifnet *ifp)
885 struct nd_defrouter *dr, *ndr;
886 struct nd_prefix *pr, *npr;
889 * Nuke default router list entries toward ifp.
890 * We defer removal of default router list entries that is installed
891 * in the routing table, in order to keep additional side effects as
894 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
902 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
910 /* Nuke prefix list entries toward ifp */
911 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
912 if (pr->ndpr_ifp == ifp) {
914 * Because if_detach() does *not* release prefixes
915 * while purging addresses the reference count will
916 * still be above zero. We therefore reset it to
917 * make sure that the prefix really gets purged.
922 * Previously, pr->ndpr_addr is removed as well,
923 * but I strongly believe we don't have to do it.
924 * nd6_purge() is only called from in6_ifdetach(),
925 * which removes all the associated interface addresses
927 * (jinmei@kame.net 20010129)
933 /* cancel default outgoing interface setting */
934 if (V_nd6_defifindex == ifp->if_index)
935 nd6_setdefaultiface(0);
937 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
938 /* Refresh default router list. */
943 * We do not nuke the neighbor cache entries here any more
944 * because the neighbor cache is kept in if_afdata[AF_INET6].
945 * nd6_purge() is invoked by in6_ifdetach() which is called
946 * from if_detach() where everything gets purged. So let
947 * in6_domifdetach() do the actual L2 table purging work.
952 * the caller acquires and releases the lock on the lltbls
953 * Returns the llentry locked
956 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
958 struct sockaddr_in6 sin6;
962 bzero(&sin6, sizeof(sin6));
963 sin6.sin6_len = sizeof(struct sockaddr_in6);
964 sin6.sin6_family = AF_INET6;
965 sin6.sin6_addr = *addr6;
967 IF_AFDATA_LOCK_ASSERT(ifp);
969 llflags = (flags & ND6_EXCLUSIVE) ? LLE_EXCLUSIVE : 0;
970 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
976 nd6_alloc(struct in6_addr *addr6, int flags, struct ifnet *ifp)
978 struct sockaddr_in6 sin6;
981 bzero(&sin6, sizeof(sin6));
982 sin6.sin6_len = sizeof(struct sockaddr_in6);
983 sin6.sin6_family = AF_INET6;
984 sin6.sin6_addr = *addr6;
986 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
988 ln->ln_state = ND6_LLINFO_NOSTATE;
994 * Test whether a given IPv6 address is a neighbor or not, ignoring
995 * the actual neighbor cache. The neighbor cache is ignored in order
996 * to not reenter the routing code from within itself.
999 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1001 struct nd_prefix *pr;
1002 struct ifaddr *dstaddr;
1005 * A link-local address is always a neighbor.
1006 * XXX: a link does not necessarily specify a single interface.
1008 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1009 struct sockaddr_in6 sin6_copy;
1013 * We need sin6_copy since sa6_recoverscope() may modify the
1017 if (sa6_recoverscope(&sin6_copy))
1018 return (0); /* XXX: should be impossible */
1019 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1021 if (sin6_copy.sin6_scope_id == zone)
1028 * If the address matches one of our addresses,
1029 * it should be a neighbor.
1030 * If the address matches one of our on-link prefixes, it should be a
1033 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1034 if (pr->ndpr_ifp != ifp)
1037 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1040 /* Always use the default FIB here. */
1041 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
1042 0, 0, RT_DEFAULT_FIB);
1046 * This is the case where multiple interfaces
1047 * have the same prefix, but only one is installed
1048 * into the routing table and that prefix entry
1049 * is not the one being examined here. In the case
1050 * where RADIX_MPATH is enabled, multiple route
1051 * entries (of the same rt_key value) will be
1052 * installed because the interface addresses all
1055 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1056 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
1063 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1064 &addr->sin6_addr, &pr->ndpr_mask))
1069 * If the address is assigned on the node of the other side of
1070 * a p2p interface, the address should be a neighbor.
1072 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr, RT_ALL_FIBS);
1073 if (dstaddr != NULL) {
1074 if (dstaddr->ifa_ifp == ifp) {
1082 * If the default router list is empty, all addresses are regarded
1083 * as on-link, and thus, as a neighbor.
1085 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1086 TAILQ_EMPTY(&V_nd_defrouter) &&
1087 V_nd6_defifindex == ifp->if_index) {
1096 * Detect if a given IPv6 address identifies a neighbor on a given link.
1097 * XXX: should take care of the destination of a p2p link?
1100 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1102 struct llentry *lle;
1105 IF_AFDATA_UNLOCK_ASSERT(ifp);
1106 if (nd6_is_new_addr_neighbor(addr, ifp))
1110 * Even if the address matches none of our addresses, it might be
1111 * in the neighbor cache.
1113 IF_AFDATA_RLOCK(ifp);
1114 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1118 IF_AFDATA_RUNLOCK(ifp);
1123 * Free an nd6 llinfo entry.
1124 * Since the function would cause significant changes in the kernel, DO NOT
1125 * make it global, unless you have a strong reason for the change, and are sure
1126 * that the change is safe.
1129 nd6_free(struct llentry *ln, int gc)
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);
1177 * Unreachablity of a router might affect the default
1178 * router selection and on-link detection of advertised
1183 * Temporarily fake the state to choose a new default
1184 * router and to perform on-link determination of
1185 * prefixes correctly.
1186 * Below the state will be set correctly,
1187 * or the entry itself will be deleted.
1189 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1192 if (ln->ln_router || dr) {
1195 * We need to unlock to avoid a LOR with rt6_flush() with the
1196 * rnh and for the calls to pfxlist_onlink_check() and
1197 * defrouter_select() in the block further down for calls
1198 * into nd6_lookup(). We still hold a ref.
1203 * rt6_flush must be called whether or not the neighbor
1204 * is in the Default Router List.
1205 * See a corresponding comment in nd6_na_input().
1207 rt6_flush(&ln->r_l3addr.addr6, ifp);
1212 * Since defrouter_select() does not affect the
1213 * on-link determination and MIP6 needs the check
1214 * before the default router selection, we perform
1217 pfxlist_onlink_check();
1220 * Refresh default router list.
1226 * If this entry was added by an on-link redirect, remove the
1227 * corresponding host route.
1229 if (ln->la_flags & LLE_REDIRECT)
1230 nd6_free_redirect(ln);
1232 if (ln->ln_router || dr)
1237 * Save to unlock. We still hold an extra reference and will not
1238 * free(9) in llentry_free() if someone else holds one as well.
1241 IF_AFDATA_LOCK(ifp);
1243 /* Guard against race with other llentry_free(). */
1244 if (ln->la_flags & LLE_LINKED) {
1245 /* Remove callout reference */
1247 lltable_unlink_entry(ln->lle_tbl, ln);
1249 IF_AFDATA_UNLOCK(ifp);
1255 * Remove the rtentry for the given llentry,
1256 * both of which were installed by a redirect.
1259 nd6_free_redirect(const struct llentry *ln)
1263 struct radix_node_head *rnh;
1264 struct sockaddr_in6 sin6;
1266 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1267 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1268 rnh = rt_tables_get_rnh(fibnum, AF_INET6);
1272 RADIX_NODE_HEAD_LOCK(rnh);
1273 rt = in6_rtalloc1((struct sockaddr *)&sin6, 0,
1274 RTF_RNH_LOCKED, fibnum);
1276 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1277 rt_expunge(rnh, rt);
1280 RADIX_NODE_HEAD_UNLOCK(rnh);
1285 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1287 * XXX cost-effective methods?
1290 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1295 if ((dst6 == NULL) || (rt == NULL))
1299 IF_AFDATA_RLOCK(ifp);
1300 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1301 IF_AFDATA_RUNLOCK(ifp);
1305 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1309 * if we get upper-layer reachability confirmation many times,
1310 * it is possible we have false information.
1314 if (ln->ln_byhint > V_nd6_maxnudhint) {
1319 ln->ln_state = ND6_LLINFO_REACHABLE;
1320 if (!ND6_LLINFO_PERMANENT(ln)) {
1321 nd6_llinfo_settimer_locked(ln,
1322 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1330 * Rejuvenate this function for routing operations related
1334 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1336 struct sockaddr_in6 *gateway;
1337 struct nd_defrouter *dr;
1340 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1351 * Only indirect routes are interesting.
1353 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1356 * check for default route
1358 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1359 &SIN6(rt_key(rt))->sin6_addr)) {
1361 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1371 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1373 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1374 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1375 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1378 if (ifp->if_afdata[AF_INET6] == NULL)
1379 return (EPFNOSUPPORT);
1381 case OSIOCGIFINFO_IN6:
1383 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1384 bzero(&ND, sizeof(ND));
1385 ND.linkmtu = IN6_LINKMTU(ifp);
1386 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1387 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1388 ND.reachable = ND_IFINFO(ifp)->reachable;
1389 ND.retrans = ND_IFINFO(ifp)->retrans;
1390 ND.flags = ND_IFINFO(ifp)->flags;
1391 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1392 ND.chlim = ND_IFINFO(ifp)->chlim;
1394 case SIOCGIFINFO_IN6:
1395 ND = *ND_IFINFO(ifp);
1397 case SIOCSIFINFO_IN6:
1399 * used to change host variables from userland.
1400 * intented for a use on router to reflect RA configurations.
1402 /* 0 means 'unspecified' */
1403 if (ND.linkmtu != 0) {
1404 if (ND.linkmtu < IPV6_MMTU ||
1405 ND.linkmtu > IN6_LINKMTU(ifp)) {
1409 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1412 if (ND.basereachable != 0) {
1413 int obasereachable = ND_IFINFO(ifp)->basereachable;
1415 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1416 if (ND.basereachable != obasereachable)
1417 ND_IFINFO(ifp)->reachable =
1418 ND_COMPUTE_RTIME(ND.basereachable);
1420 if (ND.retrans != 0)
1421 ND_IFINFO(ifp)->retrans = ND.retrans;
1423 ND_IFINFO(ifp)->chlim = ND.chlim;
1425 case SIOCSIFINFO_FLAGS:
1428 struct in6_ifaddr *ia;
1430 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1431 !(ND.flags & ND6_IFF_IFDISABLED)) {
1432 /* ifdisabled 1->0 transision */
1435 * If the interface is marked as ND6_IFF_IFDISABLED and
1436 * has an link-local address with IN6_IFF_DUPLICATED,
1437 * do not clear ND6_IFF_IFDISABLED.
1438 * See RFC 4862, Section 5.4.5.
1441 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1442 if (ifa->ifa_addr->sa_family != AF_INET6)
1444 ia = (struct in6_ifaddr *)ifa;
1445 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1446 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1449 IF_ADDR_RUNLOCK(ifp);
1452 /* LLA is duplicated. */
1453 ND.flags |= ND6_IFF_IFDISABLED;
1454 log(LOG_ERR, "Cannot enable an interface"
1455 " with a link-local address marked"
1458 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1459 if (ifp->if_flags & IFF_UP)
1462 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1463 (ND.flags & ND6_IFF_IFDISABLED)) {
1464 /* ifdisabled 0->1 transision */
1465 /* Mark all IPv6 address as tentative. */
1467 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1468 if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1470 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1472 if (ifa->ifa_addr->sa_family !=
1475 ia = (struct in6_ifaddr *)ifa;
1476 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1478 IF_ADDR_RUNLOCK(ifp);
1482 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1483 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1484 /* auto_linklocal 0->1 transision */
1486 /* If no link-local address on ifp, configure */
1487 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1488 in6_ifattach(ifp, NULL);
1489 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1490 ifp->if_flags & IFF_UP) {
1492 * When the IF already has
1493 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1494 * address is assigned, and IFF_UP, try to
1498 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1500 if (ifa->ifa_addr->sa_family !=
1503 ia = (struct in6_ifaddr *)ifa;
1504 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1507 IF_ADDR_RUNLOCK(ifp);
1509 /* No LLA is configured. */
1510 in6_ifattach(ifp, NULL);
1514 ND_IFINFO(ifp)->flags = ND.flags;
1517 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1518 /* sync kernel routing table with the default router list */
1522 case SIOCSPFXFLUSH_IN6:
1524 /* flush all the prefix advertised by routers */
1525 struct nd_prefix *pr, *next;
1527 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1528 struct in6_ifaddr *ia, *ia_next;
1530 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1533 /* do we really have to remove addresses as well? */
1534 /* XXXRW: in6_ifaddrhead locking. */
1535 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1537 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1540 if (ia->ia6_ndpr == pr)
1541 in6_purgeaddr(&ia->ia_ifa);
1547 case SIOCSRTRFLUSH_IN6:
1549 /* flush all the default routers */
1550 struct nd_defrouter *dr, *next;
1553 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1559 case SIOCGNBRINFO_IN6:
1562 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1564 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1567 IF_AFDATA_RLOCK(ifp);
1568 ln = nd6_lookup(&nb_addr, 0, ifp);
1569 IF_AFDATA_RUNLOCK(ifp);
1575 nbi->state = ln->ln_state;
1576 nbi->asked = ln->la_asked;
1577 nbi->isrouter = ln->ln_router;
1578 if (ln->la_expire == 0)
1581 nbi->expire = ln->la_expire +
1582 (time_second - time_uptime);
1586 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1587 ndif->ifindex = V_nd6_defifindex;
1589 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1590 return (nd6_setdefaultiface(ndif->ifindex));
1596 * Create neighbor cache entry and cache link-layer address,
1597 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1600 * code - type dependent information
1603 * The caller of this function already acquired the ndp
1604 * cache table lock because the cache entry is returned.
1607 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1608 int lladdrlen, int type, int code)
1610 struct llentry *ln = NULL, *ln_tmp;
1617 uint16_t router = 0;
1618 struct sockaddr_in6 sin6;
1619 struct mbuf *chain = NULL;
1620 int static_route = 0;
1622 IF_AFDATA_UNLOCK_ASSERT(ifp);
1624 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1625 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1627 /* nothing must be updated for unspecified address */
1628 if (IN6_IS_ADDR_UNSPECIFIED(from))
1632 * Validation about ifp->if_addrlen and lladdrlen must be done in
1635 * XXX If the link does not have link-layer adderss, what should
1636 * we do? (ifp->if_addrlen == 0)
1637 * Spec says nothing in sections for RA, RS and NA. There's small
1638 * description on it in NS section (RFC 2461 7.2.3).
1640 flags = lladdr ? ND6_EXCLUSIVE : 0;
1641 IF_AFDATA_RLOCK(ifp);
1642 ln = nd6_lookup(from, flags, ifp);
1643 IF_AFDATA_RUNLOCK(ifp);
1646 flags |= ND6_EXCLUSIVE;
1647 ln = nd6_alloc(from, 0, ifp);
1650 IF_AFDATA_WLOCK(ifp);
1652 /* Prefer any existing lle over newly-created one */
1653 ln_tmp = nd6_lookup(from, ND6_EXCLUSIVE, ifp);
1655 lltable_link_entry(LLTABLE6(ifp), ln);
1656 IF_AFDATA_WUNLOCK(ifp);
1658 /* No existing lle, mark as new entry */
1661 lltable_free_entry(LLTABLE6(ifp), ln);
1666 /* do nothing if static ndp is set */
1667 if ((ln->la_flags & LLE_STATIC)) {
1672 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1673 if (olladdr && lladdr) {
1674 llchange = bcmp(lladdr, &ln->ll_addr,
1680 * newentry olladdr lladdr llchange (*=record)
1683 * 0 n y -- (3) * STALE
1685 * 0 y y y (5) * STALE
1686 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1687 * 1 -- y -- (7) * STALE
1690 if (lladdr) { /* (3-5) and (7) */
1692 * Record source link-layer address
1693 * XXX is it dependent to ifp->if_type?
1695 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1696 ln->la_flags |= LLE_VALID;
1697 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1701 if ((!olladdr && lladdr != NULL) || /* (3) */
1702 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1704 newstate = ND6_LLINFO_STALE;
1705 } else /* (1-2,4) */
1709 if (lladdr == NULL) /* (6) */
1710 newstate = ND6_LLINFO_NOSTATE;
1712 newstate = ND6_LLINFO_STALE;
1717 * Update the state of the neighbor cache.
1719 ln->ln_state = newstate;
1721 if (ln->ln_state == ND6_LLINFO_STALE) {
1722 if (ln->la_hold != NULL)
1723 nd6_grab_holdchain(ln, &chain, &sin6);
1724 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1725 /* probe right away */
1726 nd6_llinfo_settimer_locked((void *)ln, 0);
1731 * ICMP6 type dependent behavior.
1733 * NS: clear IsRouter if new entry
1734 * RS: clear IsRouter
1735 * RA: set IsRouter if there's lladdr
1736 * redir: clear IsRouter if new entry
1739 * The spec says that we must set IsRouter in the following cases:
1740 * - If lladdr exist, set IsRouter. This means (1-5).
1741 * - If it is old entry (!newentry), set IsRouter. This means (7).
1742 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1743 * A quetion arises for (1) case. (1) case has no lladdr in the
1744 * neighbor cache, this is similar to (6).
1745 * This case is rare but we figured that we MUST NOT set IsRouter.
1747 * newentry olladdr lladdr llchange NS RS RA redir
1749 * 0 n n -- (1) c ? s
1750 * 0 y n -- (2) c s s
1751 * 0 n y -- (3) c s s
1754 * 1 -- n -- (6) c c c s
1755 * 1 -- y -- (7) c c s c s
1759 switch (type & 0xff) {
1760 case ND_NEIGHBOR_SOLICIT:
1762 * New entry must have is_router flag cleared.
1764 if (is_newentry) /* (6-7) */
1769 * If the icmp is a redirect to a better router, always set the
1770 * is_router flag. Otherwise, if the entry is newly created,
1771 * clear the flag. [RFC 2461, sec 8.3]
1773 if (code == ND_REDIRECT_ROUTER)
1776 if (is_newentry) /* (6-7) */
1778 ln->la_flags |= LLE_REDIRECT;
1781 case ND_ROUTER_SOLICIT:
1783 * is_router flag must always be cleared.
1787 case ND_ROUTER_ADVERT:
1789 * Mark an entry with lladdr as a router.
1791 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1792 (is_newentry && lladdr)) { /* (7) */
1799 static_route = (ln->la_flags & LLE_STATIC);
1800 router = ln->ln_router;
1802 if (flags & ND6_EXCLUSIVE)
1810 nd6_flush_holdchain(ifp, ifp, chain, &sin6);
1813 * When the link-layer address of a router changes, select the
1814 * best router again. In particular, when the neighbor entry is newly
1815 * created, it might affect the selection policy.
1816 * Question: can we restrict the first condition to the "is_newentry"
1818 * XXX: when we hear an RA from a new router with the link-layer
1819 * address option, defrouter_select() is called twice, since
1820 * defrtrlist_update called the function as well. However, I believe
1821 * we can compromise the overhead, since it only happens the first
1823 * XXX: although defrouter_select() should not have a bad effect
1824 * for those are not autoconfigured hosts, we explicitly avoid such
1827 if (do_update && router &&
1828 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1830 * guaranteed recursion
1838 if (flags & ND6_EXCLUSIVE)
1849 nd6_slowtimo(void *arg)
1851 CURVNET_SET((struct vnet *) arg);
1852 struct nd_ifinfo *nd6if;
1855 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1856 nd6_slowtimo, curvnet);
1857 IFNET_RLOCK_NOSLEEP();
1858 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1859 if (ifp->if_afdata[AF_INET6] == NULL)
1861 nd6if = ND_IFINFO(ifp);
1862 if (nd6if->basereachable && /* already initialized */
1863 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1865 * Since reachable time rarely changes by router
1866 * advertisements, we SHOULD insure that a new random
1867 * value gets recomputed at least once every few hours.
1870 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1871 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1874 IFNET_RUNLOCK_NOSLEEP();
1879 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
1880 struct sockaddr_in6 *sin6)
1883 LLE_WLOCK_ASSERT(ln);
1885 *chain = ln->la_hold;
1887 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
1889 if (ln->ln_state == ND6_LLINFO_STALE) {
1892 * The first time we send a packet to a
1893 * neighbor whose entry is STALE, we have
1894 * to change the state to DELAY and a sets
1895 * a timer to expire in DELAY_FIRST_PROBE_TIME
1896 * seconds to ensure do neighbor unreachability
1897 * detection on expiration.
1901 ln->ln_state = ND6_LLINFO_DELAY;
1902 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1907 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1908 struct sockaddr_in6 *dst)
1912 struct ip6_hdr *ip6;
1916 mac_netinet6_nd6_send(ifp, m);
1920 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1921 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1922 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1923 * to be diverted to user space. When re-injected into the kernel,
1924 * send_output() will directly dispatch them to the outgoing interface.
1926 if (send_sendso_input_hook != NULL) {
1927 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1929 ip6 = mtod(m, struct ip6_hdr *);
1930 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1931 /* Use the SEND socket */
1932 error = send_sendso_input_hook(m, ifp, SND_OUT,
1934 /* -1 == no app on SEND socket */
1935 if (error == 0 || error != -1)
1940 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1941 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1942 mtod(m, struct ip6_hdr *));
1944 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1947 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1952 * IPv6 packet output - light version.
1953 * Checks if destination LLE exists and is in proper state
1954 * (e.g no modification required). If not true, fall back to
1958 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1959 struct sockaddr_in6 *dst, struct rtentry *rt0)
1961 struct llentry *ln = NULL;
1963 /* discard the packet if IPv6 operation is disabled on the interface */
1964 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1966 return (ENETDOWN); /* better error? */
1969 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1972 if (nd6_need_cache(ifp) == 0)
1975 IF_AFDATA_RLOCK(ifp);
1976 ln = nd6_lookup(&dst->sin6_addr, 0, ifp);
1977 IF_AFDATA_RUNLOCK(ifp);
1980 * Perform fast path for the following cases:
1981 * 1) lle state is REACHABLE
1982 * 2) lle state is DELAY (NS message sentNS message sent)
1984 * Every other case involves lle modification, so we handle
1987 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
1988 ln->ln_state != ND6_LLINFO_DELAY)) {
1989 /* Fall back to slow processing path */
1992 return (nd6_output_lle(ifp, origifp, m, dst));
1999 return (nd6_output_ifp(ifp, origifp, m, dst));
2004 * Output IPv6 packet - heavy version.
2005 * Function assume that either
2006 * 1) destination LLE does not exist, is invalid or stale, so
2007 * ND6_EXCLUSIVE lock needs to be acquired
2008 * 2) destination lle is provided (with ND6_EXCLUSIVE lock),
2009 * in that case packets are queued in &chain.
2013 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2014 struct sockaddr_in6 *dst)
2016 struct llentry *lle = NULL, *lle_tmp;
2018 KASSERT(m != NULL, ("NULL mbuf, nothing to send"));
2019 /* discard the packet if IPv6 operation is disabled on the interface */
2020 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2022 return (ENETDOWN); /* better error? */
2025 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2028 if (nd6_need_cache(ifp) == 0)
2032 * Address resolution or Neighbor Unreachability Detection
2034 * At this point, the destination of the packet must be a unicast
2035 * or an anycast address(i.e. not a multicast).
2038 IF_AFDATA_RLOCK(ifp);
2039 lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
2040 IF_AFDATA_RUNLOCK(ifp);
2041 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2043 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2044 * the condition below is not very efficient. But we believe
2045 * it is tolerable, because this should be a rare case.
2047 lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2049 char ip6buf[INET6_ADDRSTRLEN];
2051 "nd6_output: can't allocate llinfo for %s "
2053 ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2057 lle->ln_state = ND6_LLINFO_NOSTATE;
2059 IF_AFDATA_WLOCK(ifp);
2061 /* Prefer any existing entry over newly-created one */
2062 lle_tmp = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp);
2063 if (lle_tmp == NULL)
2064 lltable_link_entry(LLTABLE6(ifp), lle);
2065 IF_AFDATA_WUNLOCK(ifp);
2066 if (lle_tmp != NULL) {
2067 lltable_free_entry(LLTABLE6(ifp), lle);
2074 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2075 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2079 goto sendpkt; /* send anyway */
2082 LLE_WLOCK_ASSERT(lle);
2084 /* We don't have to do link-layer address resolution on a p2p link. */
2085 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2086 lle->ln_state < ND6_LLINFO_REACHABLE) {
2087 lle->ln_state = ND6_LLINFO_STALE;
2088 nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz);
2092 * The first time we send a packet to a neighbor whose entry is
2093 * STALE, we have to change the state to DELAY and a sets a timer to
2094 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2095 * neighbor unreachability detection on expiration.
2098 if (lle->ln_state == ND6_LLINFO_STALE) {
2100 lle->ln_state = ND6_LLINFO_DELAY;
2101 nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz);
2105 * If the neighbor cache entry has a state other than INCOMPLETE
2106 * (i.e. its link-layer address is already resolved), just
2109 if (lle->ln_state > ND6_LLINFO_INCOMPLETE)
2113 * There is a neighbor cache entry, but no ethernet address
2114 * response yet. Append this latest packet to the end of the
2115 * packet queue in the mbuf, unless the number of the packet
2116 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
2117 * the oldest packet in the queue will be removed.
2119 if (lle->ln_state == ND6_LLINFO_NOSTATE)
2120 lle->ln_state = ND6_LLINFO_INCOMPLETE;
2122 if (lle->la_hold != NULL) {
2123 struct mbuf *m_hold;
2127 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2129 if (m_hold->m_nextpkt == NULL) {
2130 m_hold->m_nextpkt = m;
2134 while (i >= V_nd6_maxqueuelen) {
2135 m_hold = lle->la_hold;
2136 lle->la_hold = lle->la_hold->m_nextpkt;
2145 * If there has been no NS for the neighbor after entering the
2146 * INCOMPLETE state, send the first solicitation.
2148 if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) {
2149 struct in6_addr src, *psrc;
2152 nd6_llinfo_settimer_locked(lle,
2153 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2154 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2156 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2158 /* We did the lookup so we need to do the unlock here. */
2168 return (nd6_output_ifp(ifp, origifp, m, dst));
2173 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2174 struct sockaddr_in6 *dst)
2176 struct mbuf *m, *m_head;
2177 struct ifnet *outifp;
2181 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2188 m_head = m_head->m_nextpkt;
2189 error = nd6_output_ifp(ifp, origifp, m, dst);
2194 * note that intermediate errors are blindly ignored - but this is
2195 * the same convention as used with nd6_output when called by
2203 nd6_need_cache(struct ifnet *ifp)
2206 * XXX: we currently do not make neighbor cache on any interface
2207 * other than ARCnet, Ethernet, FDDI and GIF.
2210 * - unidirectional tunnels needs no ND
2212 switch (ifp->if_type) {
2219 case IFT_INFINIBAND:
2221 case IFT_PROPVIRTUAL:
2229 * Add pernament ND6 link-layer record for given
2230 * interface address.
2232 * Very similar to IPv4 arp_ifinit(), but:
2233 * 1) IPv6 DAD is performed in different place
2234 * 2) It is called by IPv6 protocol stack in contrast to
2235 * arp_ifinit() which is typically called in SIOCSIFADDR
2236 * driver ioctl handler.
2240 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2243 struct llentry *ln, *ln_tmp;
2244 struct sockaddr *dst;
2246 ifp = ia->ia_ifa.ifa_ifp;
2247 if (nd6_need_cache(ifp) == 0)
2250 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2251 dst = (struct sockaddr *)&ia->ia_addr;
2252 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2256 IF_AFDATA_WLOCK(ifp);
2258 /* Unlink any entry if exists */
2259 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2261 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2262 lltable_link_entry(LLTABLE6(ifp), ln);
2263 IF_AFDATA_WUNLOCK(ifp);
2266 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2267 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2271 llentry_free(ln_tmp);
2277 * Removes either all lle entries for given @ia, or lle
2278 * corresponding to @ia address.
2281 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2283 struct sockaddr_in6 mask, addr;
2284 struct sockaddr *saddr, *smask;
2287 ifp = ia->ia_ifa.ifa_ifp;
2288 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2289 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2290 saddr = (struct sockaddr *)&addr;
2291 smask = (struct sockaddr *)&mask;
2294 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2296 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2300 * the callers of this function need to be re-worked to drop
2301 * the lle lock, drop here for now
2304 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2305 const struct sockaddr *dst, u_char *desten, uint32_t *pflags)
2311 IF_AFDATA_UNLOCK_ASSERT(ifp);
2312 if (m != NULL && m->m_flags & M_MCAST) {
2313 switch (ifp->if_type) {
2320 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2325 return (EAFNOSUPPORT);
2331 * the entry should have been created in nd6_store_lladdr
2333 IF_AFDATA_RLOCK(ifp);
2334 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2335 IF_AFDATA_RUNLOCK(ifp);
2336 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2339 /* this could happen, if we could not allocate memory */
2344 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2346 *pflags = ln->la_flags;
2349 * A *small* use after free race exists here
2355 clear_llinfo_pqueue(struct llentry *ln)
2357 struct mbuf *m_hold, *m_hold_next;
2359 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2360 m_hold_next = m_hold->m_nextpkt;
2368 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2369 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2371 SYSCTL_DECL(_net_inet6_icmp6);
2373 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2374 CTLFLAG_RD, nd6_sysctl_drlist, "");
2375 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2376 CTLFLAG_RD, nd6_sysctl_prlist, "");
2377 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2378 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2379 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2380 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2383 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2385 struct in6_defrouter d;
2386 struct nd_defrouter *dr;
2392 bzero(&d, sizeof(d));
2393 d.rtaddr.sin6_family = AF_INET6;
2394 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2399 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2400 d.rtaddr.sin6_addr = dr->rtaddr;
2401 error = sa6_recoverscope(&d.rtaddr);
2404 d.flags = dr->flags;
2405 d.rtlifetime = dr->rtlifetime;
2406 d.expire = dr->expire + (time_second - time_uptime);
2407 d.if_index = dr->ifp->if_index;
2408 error = SYSCTL_OUT(req, &d, sizeof(d));
2416 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2418 struct in6_prefix p;
2419 struct sockaddr_in6 s6;
2420 struct nd_prefix *pr;
2421 struct nd_pfxrouter *pfr;
2424 char ip6buf[INET6_ADDRSTRLEN];
2429 bzero(&p, sizeof(p));
2430 p.origin = PR_ORIG_RA;
2431 bzero(&s6, sizeof(s6));
2432 s6.sin6_family = AF_INET6;
2433 s6.sin6_len = sizeof(s6);
2438 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2439 p.prefix = pr->ndpr_prefix;
2440 if (sa6_recoverscope(&p.prefix)) {
2441 log(LOG_ERR, "scope error in prefix list (%s)\n",
2442 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2443 /* XXX: press on... */
2445 p.raflags = pr->ndpr_raf;
2446 p.prefixlen = pr->ndpr_plen;
2447 p.vltime = pr->ndpr_vltime;
2448 p.pltime = pr->ndpr_pltime;
2449 p.if_index = pr->ndpr_ifp->if_index;
2450 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2453 /* XXX: we assume time_t is signed. */
2455 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2456 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2457 p.expire = pr->ndpr_lastupdate +
2459 (time_second - time_uptime);
2461 p.expire = maxexpire;
2463 p.refcnt = pr->ndpr_refcnt;
2464 p.flags = pr->ndpr_stateflags;
2466 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2468 error = SYSCTL_OUT(req, &p, sizeof(p));
2471 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2472 s6.sin6_addr = pfr->router->rtaddr;
2473 if (sa6_recoverscope(&s6))
2475 "scope error in prefix list (%s)\n",
2476 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2477 error = SYSCTL_OUT(req, &s6, sizeof(s6));