2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the project nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sysctl.h>
56 #include <net/if_arc.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/iso88025.h>
61 #include <net/route.h>
64 #include <netinet/in.h>
65 #include <net/if_llatbl.h>
66 #define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le))
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet/icmp6.h>
75 #include <sys/limits.h>
77 #include <security/mac/mac_framework.h>
79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 VNET_DEFINE(int, nd6_prune);
85 VNET_DEFINE(int, nd6_delay);
86 VNET_DEFINE(int, nd6_umaxtries);
87 VNET_DEFINE(int, nd6_mmaxtries);
88 VNET_DEFINE(int, nd6_useloopback);
89 VNET_DEFINE(int, nd6_gctimer);
91 /* preventing too many loops in ND option parsing */
92 static VNET_DEFINE(int, nd6_maxndopt);
93 VNET_DEFINE(int, nd6_maxnudhint);
94 static VNET_DEFINE(int, nd6_maxqueuelen);
95 #define V_nd6_maxndopt VNET(nd6_maxndopt)
96 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
98 VNET_DEFINE(int, nd6_debug);
102 static int nd6_inuse, nd6_allocated;
105 VNET_DEFINE(struct nd_drhead, nd_defrouter);
106 VNET_DEFINE(struct nd_prhead, nd_prefix);
108 VNET_DEFINE(int, nd6_recalc_reachtm_interval);
109 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
111 static struct sockaddr_in6 all1_sa;
113 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
115 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
116 static void nd6_slowtimo(void *);
117 static int regen_tmpaddr(struct in6_ifaddr *);
118 static struct llentry *nd6_free(struct llentry *, int);
119 static void nd6_llinfo_timer(void *);
120 static void clear_llinfo_pqueue(struct llentry *);
122 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
123 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
125 VNET_DEFINE(struct callout, nd6_timer_ch);
127 VNET_DECLARE(int, dad_ignore_ns);
128 VNET_DECLARE(int, dad_maxtry);
129 #define V_dad_ignore_ns VNET(dad_ignore_ns)
130 #define V_dad_maxtry VNET(dad_maxtry)
137 V_nd6_prune = 1; /* walk list every 1 seconds */
138 V_nd6_delay = 5; /* delay first probe time 5 second */
139 V_nd6_umaxtries = 3; /* maximum unicast query */
140 V_nd6_mmaxtries = 3; /* maximum multicast query */
141 V_nd6_useloopback = 1; /* use loopback interface for local traffic */
142 V_nd6_gctimer = (60 * 60 * 24); /* 1 day: garbage collection timer */
144 /* preventing too many loops in ND option parsing */
145 V_nd6_maxndopt = 10; /* max # of ND options allowed */
147 V_nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
148 V_nd6_maxqueuelen = 1; /* max pkts cached in unresolved ND entries */
156 V_nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
158 V_dad_ignore_ns = 0; /* ignore NS in DAD - specwise incorrect*/
159 V_dad_maxtry = 15; /* max # of *tries* to transmit DAD packet */
162 * XXX just to get this to compile KMM
165 V_llinfo_nd6.ln_next = &V_llinfo_nd6;
166 V_llinfo_nd6.ln_prev = &V_llinfo_nd6;
168 LIST_INIT(&V_nd_prefix);
170 V_ip6_use_tempaddr = 0;
171 V_ip6_temp_preferred_lifetime = DEF_TEMP_PREFERRED_LIFETIME;
172 V_ip6_temp_valid_lifetime = DEF_TEMP_VALID_LIFETIME;
173 V_ip6_temp_regen_advance = TEMPADDR_REGEN_ADVANCE;
175 V_ip6_desync_factor = 0;
177 all1_sa.sin6_family = AF_INET6;
178 all1_sa.sin6_len = sizeof(struct sockaddr_in6);
179 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
180 all1_sa.sin6_addr.s6_addr[i] = 0xff;
182 /* initialization of the default router list */
183 TAILQ_INIT(&V_nd_defrouter);
185 callout_init(&V_nd6_slowtimo_ch, 0);
186 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
187 nd6_slowtimo, curvnet);
196 callout_drain(&V_nd6_slowtimo_ch);
197 callout_drain(&V_nd6_timer_ch);
202 nd6_ifattach(struct ifnet *ifp)
204 struct nd_ifinfo *nd;
206 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
207 bzero(nd, sizeof(*nd));
211 nd->chlim = IPV6_DEFHLIM;
212 nd->basereachable = REACHABLE_TIME;
213 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
214 nd->retrans = RETRANS_TIMER;
216 * Note that the default value of ip6_accept_rtadv is 0, which means
217 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
220 nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
222 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
223 nd6_setmtu0(ifp, nd);
229 nd6_ifdetach(struct nd_ifinfo *nd)
236 * Reset ND level link MTU. This function is called when the physical MTU
237 * changes, which means we might have to adjust the ND level MTU.
240 nd6_setmtu(struct ifnet *ifp)
243 nd6_setmtu0(ifp, ND_IFINFO(ifp));
246 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
248 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
252 omaxmtu = ndi->maxmtu;
254 switch (ifp->if_type) {
256 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
259 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
262 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
265 ndi->maxmtu = ifp->if_mtu;
270 * Decreasing the interface MTU under IPV6 minimum MTU may cause
271 * undesirable situation. We thus notify the operator of the change
272 * explicitly. The check for omaxmtu is necessary to restrict the
273 * log to the case of changing the MTU, not initializing it.
275 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
276 log(LOG_NOTICE, "nd6_setmtu0: "
277 "new link MTU on %s (%lu) is too small for IPv6\n",
278 if_name(ifp), (unsigned long)ndi->maxmtu);
281 if (ndi->maxmtu > V_in6_maxmtu)
282 in6_setmaxmtu(); /* check all interfaces just in case */
287 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
290 bzero(ndopts, sizeof(*ndopts));
291 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
293 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
296 ndopts->nd_opts_done = 1;
297 ndopts->nd_opts_search = NULL;
302 * Take one ND option.
305 nd6_option(union nd_opts *ndopts)
307 struct nd_opt_hdr *nd_opt;
311 panic("ndopts == NULL in nd6_option");
312 if (ndopts->nd_opts_last == NULL)
313 panic("uninitialized ndopts in nd6_option");
314 if (ndopts->nd_opts_search == NULL)
316 if (ndopts->nd_opts_done)
319 nd_opt = ndopts->nd_opts_search;
321 /* make sure nd_opt_len is inside the buffer */
322 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
323 bzero(ndopts, sizeof(*ndopts));
327 olen = nd_opt->nd_opt_len << 3;
330 * Message validation requires that all included
331 * options have a length that is greater than zero.
333 bzero(ndopts, sizeof(*ndopts));
337 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
338 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
339 /* option overruns the end of buffer, invalid */
340 bzero(ndopts, sizeof(*ndopts));
342 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
343 /* reached the end of options chain */
344 ndopts->nd_opts_done = 1;
345 ndopts->nd_opts_search = NULL;
351 * Parse multiple ND options.
352 * This function is much easier to use, for ND routines that do not need
353 * multiple options of the same type.
356 nd6_options(union nd_opts *ndopts)
358 struct nd_opt_hdr *nd_opt;
362 panic("ndopts == NULL in nd6_options");
363 if (ndopts->nd_opts_last == NULL)
364 panic("uninitialized ndopts in nd6_options");
365 if (ndopts->nd_opts_search == NULL)
369 nd_opt = nd6_option(ndopts);
370 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
372 * Message validation requires that all included
373 * options have a length that is greater than zero.
375 ICMP6STAT_INC(icp6s_nd_badopt);
376 bzero(ndopts, sizeof(*ndopts));
383 switch (nd_opt->nd_opt_type) {
384 case ND_OPT_SOURCE_LINKADDR:
385 case ND_OPT_TARGET_LINKADDR:
387 case ND_OPT_REDIRECTED_HEADER:
388 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
390 "duplicated ND6 option found (type=%d)\n",
391 nd_opt->nd_opt_type));
394 ndopts->nd_opt_array[nd_opt->nd_opt_type]
398 case ND_OPT_PREFIX_INFORMATION:
399 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
400 ndopts->nd_opt_array[nd_opt->nd_opt_type]
403 ndopts->nd_opts_pi_end =
404 (struct nd_opt_prefix_info *)nd_opt;
408 * Unknown options must be silently ignored,
409 * to accomodate future extension to the protocol.
412 "nd6_options: unsupported option %d - "
413 "option ignored\n", nd_opt->nd_opt_type));
418 if (i > V_nd6_maxndopt) {
419 ICMP6STAT_INC(icp6s_nd_toomanyopt);
420 nd6log((LOG_INFO, "too many loop in nd opt\n"));
424 if (ndopts->nd_opts_done)
432 * ND6 timer routine to handle ND6 entries
435 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
442 canceled = callout_stop(&ln->ln_timer_ch);
444 ln->la_expire = time_second + tick / hz;
446 if (tick > INT_MAX) {
447 ln->ln_ntick = tick - INT_MAX;
448 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
449 nd6_llinfo_timer, ln);
452 canceled = callout_reset(&ln->ln_timer_ch, tick,
453 nd6_llinfo_timer, ln);
461 nd6_llinfo_settimer(struct llentry *ln, long tick)
465 nd6_llinfo_settimer_locked(ln, tick);
470 nd6_llinfo_timer(void *arg)
473 struct in6_addr *dst;
475 struct nd_ifinfo *ndi = NULL;
477 ln = (struct llentry *)arg;
479 panic("%s: NULL entry!\n", __func__);
483 if ((ifp = ((ln->lle_tbl != NULL) ? ln->lle_tbl->llt_ifp : NULL)) == NULL)
484 panic("ln ifp == NULL");
486 CURVNET_SET(ifp->if_vnet);
488 if (ln->ln_ntick > 0) {
489 if (ln->ln_ntick > INT_MAX) {
490 ln->ln_ntick -= INT_MAX;
491 nd6_llinfo_settimer(ln, INT_MAX);
494 nd6_llinfo_settimer(ln, ln->ln_ntick);
499 ndi = ND_IFINFO(ifp);
500 dst = &L3_ADDR_SIN6(ln)->sin6_addr;
501 if (ln->la_flags & LLE_STATIC) {
505 if (ln->la_flags & LLE_DELETED) {
506 (void)nd6_free(ln, 0);
511 switch (ln->ln_state) {
512 case ND6_LLINFO_INCOMPLETE:
513 if (ln->la_asked < V_nd6_mmaxtries) {
515 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
516 nd6_ns_output(ifp, NULL, dst, ln, 0);
518 struct mbuf *m = ln->la_hold;
523 * assuming every packet in la_hold has the
528 icmp6_error2(m, ICMP6_DST_UNREACH,
529 ICMP6_DST_UNREACH_ADDR, 0, ifp);
532 clear_llinfo_pqueue(ln);
534 (void)nd6_free(ln, 0);
538 case ND6_LLINFO_REACHABLE:
539 if (!ND6_LLINFO_PERMANENT(ln)) {
540 ln->ln_state = ND6_LLINFO_STALE;
541 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
545 case ND6_LLINFO_STALE:
546 /* Garbage Collection(RFC 2461 5.3) */
547 if (!ND6_LLINFO_PERMANENT(ln)) {
548 (void)nd6_free(ln, 1);
553 case ND6_LLINFO_DELAY:
554 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
557 ln->ln_state = ND6_LLINFO_PROBE;
558 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
559 nd6_ns_output(ifp, dst, dst, ln, 0);
561 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
562 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
565 case ND6_LLINFO_PROBE:
566 if (ln->la_asked < V_nd6_umaxtries) {
568 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
569 nd6_ns_output(ifp, dst, dst, ln, 0);
571 (void)nd6_free(ln, 0);
584 * ND6 timer routine to expire default route list and prefix list
589 CURVNET_SET((struct vnet *) arg);
591 struct nd_defrouter *dr;
592 struct nd_prefix *pr;
593 struct in6_ifaddr *ia6, *nia6;
594 struct in6_addrlifetime *lt6;
596 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
599 /* expire default router list */
601 dr = TAILQ_FIRST(&V_nd_defrouter);
603 if (dr->expire && dr->expire < time_second) {
604 struct nd_defrouter *t;
605 t = TAILQ_NEXT(dr, dr_entry);
609 dr = TAILQ_NEXT(dr, dr_entry);
614 * expire interface addresses.
615 * in the past the loop was inside prefix expiry processing.
616 * However, from a stricter speci-confrmance standpoint, we should
617 * rather separate address lifetimes and prefix lifetimes.
619 * XXXRW: in6_ifaddrhead locking.
622 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
623 /* check address lifetime */
624 lt6 = &ia6->ia6_lifetime;
625 if (IFA6_IS_INVALID(ia6)) {
629 * If the expiring address is temporary, try
630 * regenerating a new one. This would be useful when
631 * we suspended a laptop PC, then turned it on after a
632 * period that could invalidate all temporary
633 * addresses. Although we may have to restart the
634 * loop (see below), it must be after purging the
635 * address. Otherwise, we'd see an infinite loop of
638 if (V_ip6_use_tempaddr &&
639 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
640 if (regen_tmpaddr(ia6) == 0)
644 in6_purgeaddr(&ia6->ia_ifa);
647 goto addrloop; /* XXX: see below */
648 } else if (IFA6_IS_DEPRECATED(ia6)) {
649 int oldflags = ia6->ia6_flags;
651 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
654 * If a temporary address has just become deprecated,
655 * regenerate a new one if possible.
657 if (V_ip6_use_tempaddr &&
658 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
659 (oldflags & IN6_IFF_DEPRECATED) == 0) {
661 if (regen_tmpaddr(ia6) == 0) {
663 * A new temporary address is
665 * XXX: this means the address chain
666 * has changed while we are still in
667 * the loop. Although the change
668 * would not cause disaster (because
669 * it's not a deletion, but an
670 * addition,) we'd rather restart the
671 * loop just for safety. Or does this
672 * significantly reduce performance??
679 * A new RA might have made a deprecated address
682 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
686 /* expire prefix list */
687 pr = V_nd_prefix.lh_first;
690 * check prefix lifetime.
691 * since pltime is just for autoconf, pltime processing for
692 * prefix is not necessary.
694 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
695 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
700 * address expiration and prefix expiration are
701 * separate. NEVER perform in6_purgeaddr here.
714 * ia6 - deprecated/invalidated temporary address
717 regen_tmpaddr(struct in6_ifaddr *ia6)
721 struct in6_ifaddr *public_ifa6 = NULL;
723 ifp = ia6->ia_ifa.ifa_ifp;
725 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
726 struct in6_ifaddr *it6;
728 if (ifa->ifa_addr->sa_family != AF_INET6)
731 it6 = (struct in6_ifaddr *)ifa;
733 /* ignore no autoconf addresses. */
734 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
737 /* ignore autoconf addresses with different prefixes. */
738 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
742 * Now we are looking at an autoconf address with the same
743 * prefix as ours. If the address is temporary and is still
744 * preferred, do not create another one. It would be rare, but
745 * could happen, for example, when we resume a laptop PC after
748 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
749 !IFA6_IS_DEPRECATED(it6)) {
755 * This is a public autoconf address that has the same prefix
756 * as ours. If it is preferred, keep it. We can't break the
757 * loop here, because there may be a still-preferred temporary
758 * address with the prefix.
760 if (!IFA6_IS_DEPRECATED(it6))
763 if (public_ifa6 != NULL)
764 ifa_ref(&public_ifa6->ia_ifa);
768 if (public_ifa6 != NULL) {
771 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
772 ifa_free(&public_ifa6->ia_ifa);
773 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
774 " tmp addr,errno=%d\n", e);
777 ifa_free(&public_ifa6->ia_ifa);
785 * Nuke neighbor cache/prefix/default router management table, right before
789 nd6_purge(struct ifnet *ifp)
791 struct nd_defrouter *dr, *ndr;
792 struct nd_prefix *pr, *npr;
795 * Nuke default router list entries toward ifp.
796 * We defer removal of default router list entries that is installed
797 * in the routing table, in order to keep additional side effects as
800 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
801 ndr = TAILQ_NEXT(dr, dr_entry);
809 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
810 ndr = TAILQ_NEXT(dr, dr_entry);
818 /* Nuke prefix list entries toward ifp */
819 for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
821 if (pr->ndpr_ifp == ifp) {
823 * Because if_detach() does *not* release prefixes
824 * while purging addresses the reference count will
825 * still be above zero. We therefore reset it to
826 * make sure that the prefix really gets purged.
831 * Previously, pr->ndpr_addr is removed as well,
832 * but I strongly believe we don't have to do it.
833 * nd6_purge() is only called from in6_ifdetach(),
834 * which removes all the associated interface addresses
836 * (jinmei@kame.net 20010129)
842 /* cancel default outgoing interface setting */
843 if (V_nd6_defifindex == ifp->if_index)
844 nd6_setdefaultiface(0);
846 if (!V_ip6_forwarding && V_ip6_accept_rtadv) { /* XXX: too restrictive? */
847 /* refresh default router list
856 * We do not nuke the neighbor cache entries here any more
857 * because the neighbor cache is kept in if_afdata[AF_INET6].
858 * nd6_purge() is invoked by in6_ifdetach() which is called
859 * from if_detach() where everything gets purged. So let
860 * in6_domifdetach() do the actual L2 table purging work.
865 * the caller acquires and releases the lock on the lltbls
866 * Returns the llentry locked
869 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
871 struct sockaddr_in6 sin6;
875 bzero(&sin6, sizeof(sin6));
876 sin6.sin6_len = sizeof(struct sockaddr_in6);
877 sin6.sin6_family = AF_INET6;
878 sin6.sin6_addr = *addr6;
880 IF_AFDATA_LOCK_ASSERT(ifp);
882 if (flags & ND6_CREATE)
883 llflags |= LLE_CREATE;
884 if (flags & ND6_EXCLUSIVE)
885 llflags |= LLE_EXCLUSIVE;
887 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
888 if ((ln != NULL) && (flags & LLE_CREATE)) {
889 ln->ln_state = ND6_LLINFO_NOSTATE;
890 callout_init(&ln->ln_timer_ch, 0);
897 * Test whether a given IPv6 address is a neighbor or not, ignoring
898 * the actual neighbor cache. The neighbor cache is ignored in order
899 * to not reenter the routing code from within itself.
902 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
904 struct nd_prefix *pr;
905 struct ifaddr *dstaddr;
908 * A link-local address is always a neighbor.
909 * XXX: a link does not necessarily specify a single interface.
911 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
912 struct sockaddr_in6 sin6_copy;
916 * We need sin6_copy since sa6_recoverscope() may modify the
920 if (sa6_recoverscope(&sin6_copy))
921 return (0); /* XXX: should be impossible */
922 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
924 if (sin6_copy.sin6_scope_id == zone)
931 * If the address matches one of our addresses,
932 * it should be a neighbor.
933 * If the address matches one of our on-link prefixes, it should be a
936 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
937 if (pr->ndpr_ifp != ifp)
940 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
942 rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
946 * This is the case where multiple interfaces
947 * have the same prefix, but only one is installed
948 * into the routing table and that prefix entry
949 * is not the one being examined here. In the case
950 * where RADIX_MPATH is enabled, multiple route
951 * entries (of the same rt_key value) will be
952 * installed because the interface addresses all
955 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
956 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
963 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
964 &addr->sin6_addr, &pr->ndpr_mask))
969 * If the address is assigned on the node of the other side of
970 * a p2p interface, the address should be a neighbor.
972 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
973 if (dstaddr != NULL) {
974 if (dstaddr->ifa_ifp == ifp) {
982 * If the default router list is empty, all addresses are regarded
983 * as on-link, and thus, as a neighbor.
984 * XXX: we restrict the condition to hosts, because routers usually do
985 * not have the "default router list".
987 if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
988 V_nd6_defifindex == ifp->if_index) {
997 * Detect if a given IPv6 address identifies a neighbor on a given link.
998 * XXX: should take care of the destination of a p2p link?
1001 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
1003 struct llentry *lle;
1006 IF_AFDATA_UNLOCK_ASSERT(ifp);
1007 if (nd6_is_new_addr_neighbor(addr, ifp))
1011 * Even if the address matches none of our addresses, it might be
1012 * in the neighbor cache.
1014 IF_AFDATA_LOCK(ifp);
1015 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1019 IF_AFDATA_UNLOCK(ifp);
1024 * Free an nd6 llinfo entry.
1025 * Since the function would cause significant changes in the kernel, DO NOT
1026 * make it global, unless you have a strong reason for the change, and are sure
1027 * that the change is safe.
1029 static struct llentry *
1030 nd6_free(struct llentry *ln, int gc)
1032 struct llentry *next;
1033 struct nd_defrouter *dr;
1034 struct ifnet *ifp=NULL;
1037 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1038 * even though it is not harmful, it was not really necessary.
1042 nd6_llinfo_settimer(ln, -1);
1044 if (!V_ip6_forwarding) {
1047 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1049 if (dr != NULL && dr->expire &&
1050 ln->ln_state == ND6_LLINFO_STALE && gc) {
1052 * If the reason for the deletion is just garbage
1053 * collection, and the neighbor is an active default
1054 * router, do not delete it. Instead, reset the GC
1055 * timer using the router's lifetime.
1056 * Simply deleting the entry would affect default
1057 * router selection, which is not necessarily a good
1058 * thing, especially when we're using router preference
1060 * XXX: the check for ln_state would be redundant,
1061 * but we intentionally keep it just in case.
1063 if (dr->expire > time_second)
1064 nd6_llinfo_settimer(ln,
1065 (dr->expire - time_second) * hz);
1067 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
1072 return (LIST_NEXT(ln, lle_next));
1075 if (ln->ln_router || dr) {
1077 * rt6_flush must be called whether or not the neighbor
1078 * is in the Default Router List.
1079 * See a corresponding comment in nd6_na_input().
1081 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1086 * Unreachablity of a router might affect the default
1087 * router selection and on-link detection of advertised
1092 * Temporarily fake the state to choose a new default
1093 * router and to perform on-link determination of
1094 * prefixes correctly.
1095 * Below the state will be set correctly,
1096 * or the entry itself will be deleted.
1098 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1101 * Since defrouter_select() does not affect the
1102 * on-link determination and MIP6 needs the check
1103 * before the default router selection, we perform
1106 pfxlist_onlink_check();
1109 * refresh default router list
1117 * Before deleting the entry, remember the next entry as the
1118 * return value. We need this because pfxlist_onlink_check() above
1119 * might have freed other entries (particularly the old next entry) as
1120 * a side effect (XXX).
1122 next = LIST_NEXT(ln, lle_next);
1124 ifp = ln->lle_tbl->llt_ifp;
1125 IF_AFDATA_LOCK(ifp);
1128 IF_AFDATA_UNLOCK(ifp);
1134 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1136 * XXX cost-effective methods?
1139 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1144 if ((dst6 == NULL) || (rt == NULL))
1148 IF_AFDATA_LOCK(ifp);
1149 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1150 IF_AFDATA_UNLOCK(ifp);
1154 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1158 * if we get upper-layer reachability confirmation many times,
1159 * it is possible we have false information.
1163 if (ln->ln_byhint > V_nd6_maxnudhint) {
1168 ln->ln_state = ND6_LLINFO_REACHABLE;
1169 if (!ND6_LLINFO_PERMANENT(ln)) {
1170 nd6_llinfo_settimer_locked(ln,
1171 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1179 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1181 struct in6_drlist *drl = (struct in6_drlist *)data;
1182 struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1183 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1184 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1185 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1186 struct nd_defrouter *dr;
1187 struct nd_prefix *pr;
1188 int i = 0, error = 0;
1192 case SIOCGDRLST_IN6:
1194 * obsolete API, use sysctl under net.inet6.icmp6
1196 bzero(drl, sizeof(*drl));
1198 dr = TAILQ_FIRST(&V_nd_defrouter);
1199 while (dr && i < DRLSTSIZ) {
1200 drl->defrouter[i].rtaddr = dr->rtaddr;
1201 in6_clearscope(&drl->defrouter[i].rtaddr);
1203 drl->defrouter[i].flags = dr->flags;
1204 drl->defrouter[i].rtlifetime = dr->rtlifetime;
1205 drl->defrouter[i].expire = dr->expire;
1206 drl->defrouter[i].if_index = dr->ifp->if_index;
1208 dr = TAILQ_NEXT(dr, dr_entry);
1212 case SIOCGPRLST_IN6:
1214 * obsolete API, use sysctl under net.inet6.icmp6
1216 * XXX the structure in6_prlist was changed in backward-
1217 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6,
1218 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1221 * XXX meaning of fields, especialy "raflags", is very
1222 * differnet between RA prefix list and RR/static prefix list.
1223 * how about separating ioctls into two?
1225 bzero(oprl, sizeof(*oprl));
1227 pr = V_nd_prefix.lh_first;
1228 while (pr && i < PRLSTSIZ) {
1229 struct nd_pfxrouter *pfr;
1232 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1233 oprl->prefix[i].raflags = pr->ndpr_raf;
1234 oprl->prefix[i].prefixlen = pr->ndpr_plen;
1235 oprl->prefix[i].vltime = pr->ndpr_vltime;
1236 oprl->prefix[i].pltime = pr->ndpr_pltime;
1237 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1238 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1239 oprl->prefix[i].expire = 0;
1243 /* XXX: we assume time_t is signed. */
1246 ((sizeof(maxexpire) * 8) - 1));
1247 if (pr->ndpr_vltime <
1248 maxexpire - pr->ndpr_lastupdate) {
1249 oprl->prefix[i].expire =
1250 pr->ndpr_lastupdate +
1253 oprl->prefix[i].expire = maxexpire;
1256 pfr = pr->ndpr_advrtrs.lh_first;
1260 #define RTRADDR oprl->prefix[i].advrtr[j]
1261 RTRADDR = pfr->router->rtaddr;
1262 in6_clearscope(&RTRADDR);
1266 pfr = pfr->pfr_next;
1268 oprl->prefix[i].advrtrs = j;
1269 oprl->prefix[i].origin = PR_ORIG_RA;
1277 case OSIOCGIFINFO_IN6:
1279 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1280 bzero(&ND, sizeof(ND));
1281 ND.linkmtu = IN6_LINKMTU(ifp);
1282 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1283 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1284 ND.reachable = ND_IFINFO(ifp)->reachable;
1285 ND.retrans = ND_IFINFO(ifp)->retrans;
1286 ND.flags = ND_IFINFO(ifp)->flags;
1287 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1288 ND.chlim = ND_IFINFO(ifp)->chlim;
1290 case SIOCGIFINFO_IN6:
1291 ND = *ND_IFINFO(ifp);
1293 case SIOCSIFINFO_IN6:
1295 * used to change host variables from userland.
1296 * intented for a use on router to reflect RA configurations.
1298 /* 0 means 'unspecified' */
1299 if (ND.linkmtu != 0) {
1300 if (ND.linkmtu < IPV6_MMTU ||
1301 ND.linkmtu > IN6_LINKMTU(ifp)) {
1305 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1308 if (ND.basereachable != 0) {
1309 int obasereachable = ND_IFINFO(ifp)->basereachable;
1311 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1312 if (ND.basereachable != obasereachable)
1313 ND_IFINFO(ifp)->reachable =
1314 ND_COMPUTE_RTIME(ND.basereachable);
1316 if (ND.retrans != 0)
1317 ND_IFINFO(ifp)->retrans = ND.retrans;
1319 ND_IFINFO(ifp)->chlim = ND.chlim;
1321 case SIOCSIFINFO_FLAGS:
1322 ND_IFINFO(ifp)->flags = ND.flags;
1325 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1326 /* sync kernel routing table with the default router list */
1330 case SIOCSPFXFLUSH_IN6:
1332 /* flush all the prefix advertised by routers */
1333 struct nd_prefix *pr, *next;
1336 for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1337 struct in6_ifaddr *ia, *ia_next;
1339 next = pr->ndpr_next;
1341 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1344 /* do we really have to remove addresses as well? */
1345 /* XXXRW: in6_ifaddrhead locking. */
1346 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1348 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1351 if (ia->ia6_ndpr == pr)
1352 in6_purgeaddr(&ia->ia_ifa);
1359 case SIOCSRTRFLUSH_IN6:
1361 /* flush all the default routers */
1362 struct nd_defrouter *dr, *next;
1366 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1367 next = TAILQ_NEXT(dr, dr_entry);
1374 case SIOCGNBRINFO_IN6:
1377 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1379 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1382 IF_AFDATA_LOCK(ifp);
1383 ln = nd6_lookup(&nb_addr, 0, ifp);
1384 IF_AFDATA_UNLOCK(ifp);
1390 nbi->state = ln->ln_state;
1391 nbi->asked = ln->la_asked;
1392 nbi->isrouter = ln->ln_router;
1393 nbi->expire = ln->la_expire;
1397 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1398 ndif->ifindex = V_nd6_defifindex;
1400 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1401 return (nd6_setdefaultiface(ndif->ifindex));
1407 * Create neighbor cache entry and cache link-layer address,
1408 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1411 * code - type dependent information
1414 * The caller of this function already acquired the ndp
1415 * cache table lock because the cache entry is returned.
1418 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1419 int lladdrlen, int type, int code)
1421 struct llentry *ln = NULL;
1428 uint16_t router = 0;
1429 struct sockaddr_in6 sin6;
1430 struct mbuf *chain = NULL;
1431 int static_route = 0;
1433 IF_AFDATA_UNLOCK_ASSERT(ifp);
1436 panic("ifp == NULL in nd6_cache_lladdr");
1438 panic("from == NULL in nd6_cache_lladdr");
1440 /* nothing must be updated for unspecified address */
1441 if (IN6_IS_ADDR_UNSPECIFIED(from))
1445 * Validation about ifp->if_addrlen and lladdrlen must be done in
1448 * XXX If the link does not have link-layer adderss, what should
1449 * we do? (ifp->if_addrlen == 0)
1450 * Spec says nothing in sections for RA, RS and NA. There's small
1451 * description on it in NS section (RFC 2461 7.2.3).
1453 flags |= lladdr ? ND6_EXCLUSIVE : 0;
1454 IF_AFDATA_LOCK(ifp);
1455 ln = nd6_lookup(from, flags, ifp);
1458 flags |= LLE_EXCLUSIVE;
1459 ln = nd6_lookup(from, flags |ND6_CREATE, ifp);
1460 IF_AFDATA_UNLOCK(ifp);
1463 IF_AFDATA_UNLOCK(ifp);
1464 /* do nothing if static ndp is set */
1465 if (ln->la_flags & LLE_STATIC) {
1474 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1475 if (olladdr && lladdr) {
1476 llchange = bcmp(lladdr, &ln->ll_addr,
1482 * newentry olladdr lladdr llchange (*=record)
1485 * 0 n y -- (3) * STALE
1487 * 0 y y y (5) * STALE
1488 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1489 * 1 -- y -- (7) * STALE
1492 if (lladdr) { /* (3-5) and (7) */
1494 * Record source link-layer address
1495 * XXX is it dependent to ifp->if_type?
1497 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1498 ln->la_flags |= LLE_VALID;
1502 if ((!olladdr && lladdr != NULL) || /* (3) */
1503 (olladdr && lladdr != NULL && llchange)) { /* (5) */
1505 newstate = ND6_LLINFO_STALE;
1506 } else /* (1-2,4) */
1510 if (lladdr == NULL) /* (6) */
1511 newstate = ND6_LLINFO_NOSTATE;
1513 newstate = ND6_LLINFO_STALE;
1518 * Update the state of the neighbor cache.
1520 ln->ln_state = newstate;
1522 if (ln->ln_state == ND6_LLINFO_STALE) {
1524 * XXX: since nd6_output() below will cause
1525 * state tansition to DELAY and reset the timer,
1526 * we must set the timer now, although it is actually
1529 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1532 struct mbuf *m_hold, *m_hold_next;
1535 * reset the la_hold in advance, to explicitly
1536 * prevent a la_hold lookup in nd6_output()
1537 * (wouldn't happen, though...)
1539 for (m_hold = ln->la_hold, ln->la_hold = NULL;
1540 m_hold; m_hold = m_hold_next) {
1541 m_hold_next = m_hold->m_nextpkt;
1542 m_hold->m_nextpkt = NULL;
1545 * we assume ifp is not a p2p here, so
1546 * just set the 2nd argument as the
1549 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1552 * If we have mbufs in the chain we need to do
1553 * deferred transmit. Copy the address from the
1554 * llentry before dropping the lock down below.
1557 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1559 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1560 /* probe right away */
1561 nd6_llinfo_settimer_locked((void *)ln, 0);
1566 * ICMP6 type dependent behavior.
1568 * NS: clear IsRouter if new entry
1569 * RS: clear IsRouter
1570 * RA: set IsRouter if there's lladdr
1571 * redir: clear IsRouter if new entry
1574 * The spec says that we must set IsRouter in the following cases:
1575 * - If lladdr exist, set IsRouter. This means (1-5).
1576 * - If it is old entry (!newentry), set IsRouter. This means (7).
1577 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1578 * A quetion arises for (1) case. (1) case has no lladdr in the
1579 * neighbor cache, this is similar to (6).
1580 * This case is rare but we figured that we MUST NOT set IsRouter.
1582 * newentry olladdr lladdr llchange NS RS RA redir
1584 * 0 n n -- (1) c ? s
1585 * 0 y n -- (2) c s s
1586 * 0 n y -- (3) c s s
1589 * 1 -- n -- (6) c c c s
1590 * 1 -- y -- (7) c c s c s
1594 switch (type & 0xff) {
1595 case ND_NEIGHBOR_SOLICIT:
1597 * New entry must have is_router flag cleared.
1599 if (is_newentry) /* (6-7) */
1604 * If the icmp is a redirect to a better router, always set the
1605 * is_router flag. Otherwise, if the entry is newly created,
1606 * clear the flag. [RFC 2461, sec 8.3]
1608 if (code == ND_REDIRECT_ROUTER)
1610 else if (is_newentry) /* (6-7) */
1613 case ND_ROUTER_SOLICIT:
1615 * is_router flag must always be cleared.
1619 case ND_ROUTER_ADVERT:
1621 * Mark an entry with lladdr as a router.
1623 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */
1624 (is_newentry && lladdr)) { /* (7) */
1631 static_route = (ln->la_flags & LLE_STATIC);
1632 router = ln->ln_router;
1634 if (flags & ND6_EXCLUSIVE)
1642 nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1645 * When the link-layer address of a router changes, select the
1646 * best router again. In particular, when the neighbor entry is newly
1647 * created, it might affect the selection policy.
1648 * Question: can we restrict the first condition to the "is_newentry"
1650 * XXX: when we hear an RA from a new router with the link-layer
1651 * address option, defrouter_select() is called twice, since
1652 * defrtrlist_update called the function as well. However, I believe
1653 * we can compromise the overhead, since it only happens the first
1655 * XXX: although defrouter_select() should not have a bad effect
1656 * for those are not autoconfigured hosts, we explicitly avoid such
1659 if (do_update && router && !V_ip6_forwarding && V_ip6_accept_rtadv) {
1661 * guaranteed recursion
1669 if (flags & ND6_EXCLUSIVE)
1680 nd6_slowtimo(void *arg)
1682 CURVNET_SET((struct vnet *) arg);
1683 struct nd_ifinfo *nd6if;
1686 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1687 nd6_slowtimo, curvnet);
1688 IFNET_RLOCK_NOSLEEP();
1689 for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1690 ifp = TAILQ_NEXT(ifp, if_list)) {
1691 nd6if = ND_IFINFO(ifp);
1692 if (nd6if->basereachable && /* already initialized */
1693 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1695 * Since reachable time rarely changes by router
1696 * advertisements, we SHOULD insure that a new random
1697 * value gets recomputed at least once every few hours.
1700 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1701 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1704 IFNET_RUNLOCK_NOSLEEP();
1709 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1710 struct sockaddr_in6 *dst, struct rtentry *rt0)
1713 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1718 * Note that I'm not enforcing any global serialization
1719 * lle state or asked changes here as the logic is too
1720 * complicated to avoid having to always acquire an exclusive
1725 #define senderr(e) { error = (e); goto bad;}
1728 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1729 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1730 struct mbuf **chain)
1732 struct mbuf *m = m0;
1733 struct llentry *ln = lle;
1740 LLE_WLOCK_ASSERT(lle);
1742 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1745 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1748 if (nd6_need_cache(ifp) == 0)
1752 * next hop determination. This routine is derived from ether_output.
1756 * Address resolution or Neighbor Unreachability Detection
1758 * At this point, the destination of the packet must be a unicast
1759 * or an anycast address(i.e. not a multicast).
1762 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1765 IF_AFDATA_LOCK(ifp);
1766 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1767 IF_AFDATA_UNLOCK(ifp);
1768 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
1770 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1771 * the condition below is not very efficient. But we believe
1772 * it is tolerable, because this should be a rare case.
1774 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1775 IF_AFDATA_LOCK(ifp);
1776 ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1777 IF_AFDATA_UNLOCK(ifp);
1781 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1782 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1783 char ip6buf[INET6_ADDRSTRLEN];
1785 "nd6_output: can't allocate llinfo for %s "
1787 ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1788 senderr(EIO); /* XXX: good error? */
1790 goto sendpkt; /* send anyway */
1793 /* We don't have to do link-layer address resolution on a p2p link. */
1794 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1795 ln->ln_state < ND6_LLINFO_REACHABLE) {
1796 if ((flags & LLE_EXCLUSIVE) == 0) {
1797 flags |= LLE_EXCLUSIVE;
1800 ln->ln_state = ND6_LLINFO_STALE;
1801 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1805 * The first time we send a packet to a neighbor whose entry is
1806 * STALE, we have to change the state to DELAY and a sets a timer to
1807 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1808 * neighbor unreachability detection on expiration.
1811 if (ln->ln_state == ND6_LLINFO_STALE) {
1812 if ((flags & LLE_EXCLUSIVE) == 0) {
1813 flags |= LLE_EXCLUSIVE;
1818 ln->ln_state = ND6_LLINFO_DELAY;
1819 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1823 * If the neighbor cache entry has a state other than INCOMPLETE
1824 * (i.e. its link-layer address is already resolved), just
1827 if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1831 * There is a neighbor cache entry, but no ethernet address
1832 * response yet. Append this latest packet to the end of the
1833 * packet queue in the mbuf, unless the number of the packet
1834 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
1835 * the oldest packet in the queue will be removed.
1837 if (ln->ln_state == ND6_LLINFO_NOSTATE)
1838 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1840 if ((flags & LLE_EXCLUSIVE) == 0) {
1841 flags |= LLE_EXCLUSIVE;
1846 struct mbuf *m_hold;
1850 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1852 if (m_hold->m_nextpkt == NULL) {
1853 m_hold->m_nextpkt = m;
1857 while (i >= V_nd6_maxqueuelen) {
1858 m_hold = ln->la_hold;
1859 ln->la_hold = ln->la_hold->m_nextpkt;
1867 * We did the lookup (no lle arg) so we
1868 * need to do the unlock here
1871 if (flags & LLE_EXCLUSIVE)
1878 * If there has been no NS for the neighbor after entering the
1879 * INCOMPLETE state, send the first solicitation.
1881 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1884 nd6_llinfo_settimer(ln,
1885 (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1886 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1891 /* discard the packet if IPv6 operation is disabled on the interface */
1892 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1893 error = ENETDOWN; /* better error? */
1897 * ln is valid and the caller did not pass in
1900 if ((ln != NULL) && (lle == NULL)) {
1901 if (flags & LLE_EXCLUSIVE)
1908 mac_netinet6_nd6_send(ifp, m);
1911 * We were passed in a pointer to an lle with the lock held
1912 * this means that we can't call if_output as we will
1913 * recurse on the lle lock - so what we do is we create
1914 * a list of mbufs to send and transmit them in the caller
1915 * after the lock is dropped
1921 struct mbuf *m = *chain;
1924 * append mbuf to end of deferred chain
1926 while (m->m_nextpkt != NULL)
1932 if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1933 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1936 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1941 * ln is valid and the caller did not pass in
1944 if ((ln != NULL) && (lle == NULL)) {
1945 if (flags & LLE_EXCLUSIVE)
1958 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
1959 struct sockaddr_in6 *dst, struct route *ro)
1961 struct mbuf *m, *m_head;
1962 struct ifnet *outifp;
1966 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
1973 m_head = m_head->m_nextpkt;
1974 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
1979 * note that intermediate errors are blindly ignored - but this is
1980 * the same convention as used with nd6_output when called by
1988 nd6_need_cache(struct ifnet *ifp)
1991 * XXX: we currently do not make neighbor cache on any interface
1992 * other than ARCnet, Ethernet, FDDI and GIF.
1995 * - unidirectional tunnels needs no ND
1997 switch (ifp->if_type) {
2005 #ifdef IFT_IEEE80211
2011 case IFT_GIF: /* XXX need more cases? */
2015 case IFT_PROPVIRTUAL:
2023 * the callers of this function need to be re-worked to drop
2024 * the lle lock, drop here for now
2027 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2028 struct sockaddr *dst, u_char *desten, struct llentry **lle)
2033 IF_AFDATA_UNLOCK_ASSERT(ifp);
2034 if (m->m_flags & M_MCAST) {
2037 switch (ifp->if_type) {
2043 #ifdef IFT_IEEE80211
2048 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2053 * netbsd can use if_broadcastaddr, but we don't do so
2054 * to reduce # of ifdef.
2056 for (i = 0; i < ifp->if_addrlen; i++)
2064 return (EAFNOSUPPORT);
2070 * the entry should have been created in nd6_store_lladdr
2072 IF_AFDATA_LOCK(ifp);
2073 ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2074 IF_AFDATA_UNLOCK(ifp);
2075 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2078 /* this could happen, if we could not allocate memory */
2083 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2087 * A *small* use after free race exists here
2093 clear_llinfo_pqueue(struct llentry *ln)
2095 struct mbuf *m_hold, *m_hold_next;
2097 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2098 m_hold_next = m_hold->m_nextpkt;
2099 m_hold->m_nextpkt = NULL;
2107 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2108 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2110 SYSCTL_DECL(_net_inet6_icmp6);
2112 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2113 CTLFLAG_RD, nd6_sysctl_drlist, "");
2114 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2115 CTLFLAG_RD, nd6_sysctl_prlist, "");
2116 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2117 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2120 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2123 char buf[1024] __aligned(4);
2124 struct in6_defrouter *d, *de;
2125 struct nd_defrouter *dr;
2131 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2132 dr = TAILQ_NEXT(dr, dr_entry)) {
2133 d = (struct in6_defrouter *)buf;
2134 de = (struct in6_defrouter *)(buf + sizeof(buf));
2137 bzero(d, sizeof(*d));
2138 d->rtaddr.sin6_family = AF_INET6;
2139 d->rtaddr.sin6_len = sizeof(d->rtaddr);
2140 d->rtaddr.sin6_addr = dr->rtaddr;
2141 error = sa6_recoverscope(&d->rtaddr);
2144 d->flags = dr->flags;
2145 d->rtlifetime = dr->rtlifetime;
2146 d->expire = dr->expire;
2147 d->if_index = dr->ifp->if_index;
2149 panic("buffer too short");
2151 error = SYSCTL_OUT(req, buf, sizeof(*d));
2160 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2163 char buf[1024] __aligned(4);
2164 struct in6_prefix *p, *pe;
2165 struct nd_prefix *pr;
2166 char ip6buf[INET6_ADDRSTRLEN];
2172 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2175 struct sockaddr_in6 *sin6, *s6;
2176 struct nd_pfxrouter *pfr;
2178 p = (struct in6_prefix *)buf;
2179 pe = (struct in6_prefix *)(buf + sizeof(buf));
2182 bzero(p, sizeof(*p));
2183 sin6 = (struct sockaddr_in6 *)(p + 1);
2185 p->prefix = pr->ndpr_prefix;
2186 if (sa6_recoverscope(&p->prefix)) {
2188 "scope error in prefix list (%s)\n",
2189 ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2190 /* XXX: press on... */
2192 p->raflags = pr->ndpr_raf;
2193 p->prefixlen = pr->ndpr_plen;
2194 p->vltime = pr->ndpr_vltime;
2195 p->pltime = pr->ndpr_pltime;
2196 p->if_index = pr->ndpr_ifp->if_index;
2197 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2202 /* XXX: we assume time_t is signed. */
2205 ((sizeof(maxexpire) * 8) - 1));
2206 if (pr->ndpr_vltime <
2207 maxexpire - pr->ndpr_lastupdate) {
2208 p->expire = pr->ndpr_lastupdate +
2211 p->expire = maxexpire;
2213 p->refcnt = pr->ndpr_refcnt;
2214 p->flags = pr->ndpr_stateflags;
2215 p->origin = PR_ORIG_RA;
2217 for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2218 pfr = pfr->pfr_next) {
2219 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2223 s6 = &sin6[advrtrs];
2224 bzero(s6, sizeof(*s6));
2225 s6->sin6_family = AF_INET6;
2226 s6->sin6_len = sizeof(*sin6);
2227 s6->sin6_addr = pfr->router->rtaddr;
2228 if (sa6_recoverscope(s6)) {
2231 "prefix list (%s)\n",
2233 &pfr->router->rtaddr));
2237 p->advrtrs = advrtrs;
2239 panic("buffer too short");
2241 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2242 error = SYSCTL_OUT(req, buf, advance);