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(const 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_resolve_slow(struct ifnet *, struct mbuf *,
140 const struct sockaddr_in6 *, u_char *, uint32_t *);
141 static int nd6_need_cache(struct ifnet *);
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 * Switch @lle state to new state optionally arming timers.
546 nd6_llinfo_setstate(struct llentry *lle, int newstate)
554 case ND6_LLINFO_INCOMPLETE:
555 ifp = lle->lle_tbl->llt_ifp;
556 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
558 case ND6_LLINFO_REACHABLE:
559 if (!ND6_LLINFO_PERMANENT(lle)) {
560 ifp = lle->lle_tbl->llt_ifp;
561 delay = (long)ND_IFINFO(ifp)->reachable * hz;
564 case ND6_LLINFO_STALE:
565 delay = (long)V_nd6_gctimer * hz;
567 case ND6_LLINFO_DELAY:
569 delay = (long)V_nd6_delay * hz;
574 nd6_llinfo_settimer_locked(lle, delay);
576 lle->ln_state = newstate;
581 nd6_llinfo_settimer(struct llentry *ln, long tick)
585 nd6_llinfo_settimer_locked(ln, tick);
590 nd6_llinfo_timer(void *arg)
593 struct in6_addr *dst, *pdst, *psrc, src;
595 struct nd_ifinfo *ndi = NULL;
598 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
599 ln = (struct llentry *)arg;
601 if (callout_pending(&ln->lle_timer)) {
603 * Here we are a bit odd here in the treatment of
604 * active/pending. If the pending bit is set, it got
605 * rescheduled before I ran. The active
606 * bit we ignore, since if it was stopped
607 * in ll_tablefree() and was currently running
608 * it would have return 0 so the code would
609 * not have deleted it since the callout could
610 * not be stopped so we want to go through
611 * with the delete here now. If the callout
612 * was restarted, the pending bit will be back on and
613 * we just want to bail since the callout_reset would
614 * return 1 and our reference would have been removed
615 * by nd6_llinfo_settimer_locked above since canceled
621 ifp = ln->lle_tbl->llt_ifp;
622 CURVNET_SET(ifp->if_vnet);
623 ndi = ND_IFINFO(ifp);
625 dst = &ln->r_l3addr.addr6;
628 if (ln->ln_ntick > 0) {
629 if (ln->ln_ntick > INT_MAX) {
630 ln->ln_ntick -= INT_MAX;
631 nd6_llinfo_settimer_locked(ln, INT_MAX);
634 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
639 if (ln->la_flags & LLE_STATIC) {
643 if (ln->la_flags & LLE_DELETED) {
649 switch (ln->ln_state) {
650 case ND6_LLINFO_INCOMPLETE:
651 if (ln->la_asked < V_nd6_mmaxtries) {
654 /* Send NS to multicast address */
657 struct mbuf *m = ln->la_hold;
662 * assuming every packet in la_hold has the
663 * same IP header. Send error after unlock.
668 clear_llinfo_pqueue(ln);
670 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
674 icmp6_error2(m, ICMP6_DST_UNREACH,
675 ICMP6_DST_UNREACH_ADDR, 0, ifp);
678 case ND6_LLINFO_REACHABLE:
679 if (!ND6_LLINFO_PERMANENT(ln))
680 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
683 case ND6_LLINFO_STALE:
684 /* Garbage Collection(RFC 2461 5.3) */
685 if (!ND6_LLINFO_PERMANENT(ln)) {
686 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
692 case ND6_LLINFO_DELAY:
693 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
696 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
699 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
701 case ND6_LLINFO_PROBE:
702 if (ln->la_asked < V_nd6_umaxtries) {
706 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
712 panic("%s: paths in a dark night can be confusing: %d",
713 __func__, ln->ln_state);
717 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
718 psrc = nd6_llinfo_get_holdsrc(ln, &src);
721 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
731 * ND6 timer routine to expire default route list and prefix list
736 CURVNET_SET((struct vnet *) arg);
737 struct nd_defrouter *dr, *ndr;
738 struct nd_prefix *pr, *npr;
739 struct in6_ifaddr *ia6, *nia6;
741 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
744 /* expire default router list */
745 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
746 if (dr->expire && dr->expire < time_uptime)
751 * expire interface addresses.
752 * in the past the loop was inside prefix expiry processing.
753 * However, from a stricter speci-confrmance standpoint, we should
754 * rather separate address lifetimes and prefix lifetimes.
756 * XXXRW: in6_ifaddrhead locking.
759 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
760 /* check address lifetime */
761 if (IFA6_IS_INVALID(ia6)) {
765 * If the expiring address is temporary, try
766 * regenerating a new one. This would be useful when
767 * we suspended a laptop PC, then turned it on after a
768 * period that could invalidate all temporary
769 * addresses. Although we may have to restart the
770 * loop (see below), it must be after purging the
771 * address. Otherwise, we'd see an infinite loop of
774 if (V_ip6_use_tempaddr &&
775 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
776 if (regen_tmpaddr(ia6) == 0)
780 in6_purgeaddr(&ia6->ia_ifa);
783 goto addrloop; /* XXX: see below */
784 } else if (IFA6_IS_DEPRECATED(ia6)) {
785 int oldflags = ia6->ia6_flags;
787 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
790 * If a temporary address has just become deprecated,
791 * regenerate a new one if possible.
793 if (V_ip6_use_tempaddr &&
794 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
795 (oldflags & IN6_IFF_DEPRECATED) == 0) {
797 if (regen_tmpaddr(ia6) == 0) {
799 * A new temporary address is
801 * XXX: this means the address chain
802 * has changed while we are still in
803 * the loop. Although the change
804 * would not cause disaster (because
805 * it's not a deletion, but an
806 * addition,) we'd rather restart the
807 * loop just for safety. Or does this
808 * significantly reduce performance??
815 * A new RA might have made a deprecated address
818 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
822 /* expire prefix list */
823 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
825 * check prefix lifetime.
826 * since pltime is just for autoconf, pltime processing for
827 * prefix is not necessary.
829 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
830 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
833 * address expiration and prefix expiration are
834 * separate. NEVER perform in6_purgeaddr here.
843 * ia6 - deprecated/invalidated temporary address
846 regen_tmpaddr(struct in6_ifaddr *ia6)
850 struct in6_ifaddr *public_ifa6 = NULL;
852 ifp = ia6->ia_ifa.ifa_ifp;
854 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
855 struct in6_ifaddr *it6;
857 if (ifa->ifa_addr->sa_family != AF_INET6)
860 it6 = (struct in6_ifaddr *)ifa;
862 /* ignore no autoconf addresses. */
863 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
866 /* ignore autoconf addresses with different prefixes. */
867 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
871 * Now we are looking at an autoconf address with the same
872 * prefix as ours. If the address is temporary and is still
873 * preferred, do not create another one. It would be rare, but
874 * could happen, for example, when we resume a laptop PC after
877 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
878 !IFA6_IS_DEPRECATED(it6)) {
884 * This is a public autoconf address that has the same prefix
885 * as ours. If it is preferred, keep it. We can't break the
886 * loop here, because there may be a still-preferred temporary
887 * address with the prefix.
889 if (!IFA6_IS_DEPRECATED(it6))
892 if (public_ifa6 != NULL)
893 ifa_ref(&public_ifa6->ia_ifa);
894 IF_ADDR_RUNLOCK(ifp);
896 if (public_ifa6 != NULL) {
899 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
900 ifa_free(&public_ifa6->ia_ifa);
901 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
902 " tmp addr,errno=%d\n", e);
905 ifa_free(&public_ifa6->ia_ifa);
913 * Nuke neighbor cache/prefix/default router management table, right before
917 nd6_purge(struct ifnet *ifp)
919 struct nd_defrouter *dr, *ndr;
920 struct nd_prefix *pr, *npr;
923 * Nuke default router list entries toward ifp.
924 * We defer removal of default router list entries that is installed
925 * in the routing table, in order to keep additional side effects as
928 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
936 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
944 /* Nuke prefix list entries toward ifp */
945 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
946 if (pr->ndpr_ifp == ifp) {
948 * Because if_detach() does *not* release prefixes
949 * while purging addresses the reference count will
950 * still be above zero. We therefore reset it to
951 * make sure that the prefix really gets purged.
956 * Previously, pr->ndpr_addr is removed as well,
957 * but I strongly believe we don't have to do it.
958 * nd6_purge() is only called from in6_ifdetach(),
959 * which removes all the associated interface addresses
961 * (jinmei@kame.net 20010129)
967 /* cancel default outgoing interface setting */
968 if (V_nd6_defifindex == ifp->if_index)
969 nd6_setdefaultiface(0);
971 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
972 /* Refresh default router list. */
977 * We do not nuke the neighbor cache entries here any more
978 * because the neighbor cache is kept in if_afdata[AF_INET6].
979 * nd6_purge() is invoked by in6_ifdetach() which is called
980 * from if_detach() where everything gets purged. So let
981 * in6_domifdetach() do the actual L2 table purging work.
986 * the caller acquires and releases the lock on the lltbls
987 * Returns the llentry locked
990 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
992 struct sockaddr_in6 sin6;
995 bzero(&sin6, sizeof(sin6));
996 sin6.sin6_len = sizeof(struct sockaddr_in6);
997 sin6.sin6_family = AF_INET6;
998 sin6.sin6_addr = *addr6;
1000 IF_AFDATA_LOCK_ASSERT(ifp);
1002 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1008 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1010 struct sockaddr_in6 sin6;
1013 bzero(&sin6, sizeof(sin6));
1014 sin6.sin6_len = sizeof(struct sockaddr_in6);
1015 sin6.sin6_family = AF_INET6;
1016 sin6.sin6_addr = *addr6;
1018 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1020 ln->ln_state = ND6_LLINFO_NOSTATE;
1026 * Test whether a given IPv6 address is a neighbor or not, ignoring
1027 * the actual neighbor cache. The neighbor cache is ignored in order
1028 * to not reenter the routing code from within itself.
1031 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1033 struct nd_prefix *pr;
1034 struct ifaddr *dstaddr;
1037 * A link-local address is always a neighbor.
1038 * XXX: a link does not necessarily specify a single interface.
1040 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1041 struct sockaddr_in6 sin6_copy;
1045 * We need sin6_copy since sa6_recoverscope() may modify the
1049 if (sa6_recoverscope(&sin6_copy))
1050 return (0); /* XXX: should be impossible */
1051 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1053 if (sin6_copy.sin6_scope_id == zone)
1060 * If the address matches one of our addresses,
1061 * it should be a neighbor.
1062 * If the address matches one of our on-link prefixes, it should be a
1065 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1066 if (pr->ndpr_ifp != ifp)
1069 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1072 /* Always use the default FIB here. */
1073 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
1074 0, 0, RT_DEFAULT_FIB);
1078 * This is the case where multiple interfaces
1079 * have the same prefix, but only one is installed
1080 * into the routing table and that prefix entry
1081 * is not the one being examined here. In the case
1082 * where RADIX_MPATH is enabled, multiple route
1083 * entries (of the same rt_key value) will be
1084 * installed because the interface addresses all
1087 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1088 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
1095 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1096 &addr->sin6_addr, &pr->ndpr_mask))
1101 * If the address is assigned on the node of the other side of
1102 * a p2p interface, the address should be a neighbor.
1104 dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1105 if (dstaddr != NULL) {
1106 if (dstaddr->ifa_ifp == ifp) {
1114 * If the default router list is empty, all addresses are regarded
1115 * as on-link, and thus, as a neighbor.
1117 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1118 TAILQ_EMPTY(&V_nd_defrouter) &&
1119 V_nd6_defifindex == ifp->if_index) {
1128 * Detect if a given IPv6 address identifies a neighbor on a given link.
1129 * XXX: should take care of the destination of a p2p link?
1132 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1134 struct llentry *lle;
1137 IF_AFDATA_UNLOCK_ASSERT(ifp);
1138 if (nd6_is_new_addr_neighbor(addr, ifp))
1142 * Even if the address matches none of our addresses, it might be
1143 * in the neighbor cache.
1145 IF_AFDATA_RLOCK(ifp);
1146 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1150 IF_AFDATA_RUNLOCK(ifp);
1155 * Free an nd6 llinfo entry.
1156 * Since the function would cause significant changes in the kernel, DO NOT
1157 * make it global, unless you have a strong reason for the change, and are sure
1158 * that the change is safe.
1161 nd6_free(struct llentry *ln, int gc)
1163 struct nd_defrouter *dr;
1166 LLE_WLOCK_ASSERT(ln);
1169 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1170 * even though it is not harmful, it was not really necessary.
1174 nd6_llinfo_settimer_locked(ln, -1);
1176 ifp = ln->lle_tbl->llt_ifp;
1178 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1179 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1181 if (dr != NULL && dr->expire &&
1182 ln->ln_state == ND6_LLINFO_STALE && gc) {
1184 * If the reason for the deletion is just garbage
1185 * collection, and the neighbor is an active default
1186 * router, do not delete it. Instead, reset the GC
1187 * timer using the router's lifetime.
1188 * Simply deleting the entry would affect default
1189 * router selection, which is not necessarily a good
1190 * thing, especially when we're using router preference
1192 * XXX: the check for ln_state would be redundant,
1193 * but we intentionally keep it just in case.
1195 if (dr->expire > time_uptime)
1196 nd6_llinfo_settimer_locked(ln,
1197 (dr->expire - time_uptime) * hz);
1199 nd6_llinfo_settimer_locked(ln,
1200 (long)V_nd6_gctimer * hz);
1209 * Unreachablity of a router might affect the default
1210 * router selection and on-link detection of advertised
1215 * Temporarily fake the state to choose a new default
1216 * router and to perform on-link determination of
1217 * prefixes correctly.
1218 * Below the state will be set correctly,
1219 * or the entry itself will be deleted.
1221 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1224 if (ln->ln_router || dr) {
1227 * We need to unlock to avoid a LOR with rt6_flush() with the
1228 * rnh and for the calls to pfxlist_onlink_check() and
1229 * defrouter_select() in the block further down for calls
1230 * into nd6_lookup(). We still hold a ref.
1235 * rt6_flush must be called whether or not the neighbor
1236 * is in the Default Router List.
1237 * See a corresponding comment in nd6_na_input().
1239 rt6_flush(&ln->r_l3addr.addr6, ifp);
1244 * Since defrouter_select() does not affect the
1245 * on-link determination and MIP6 needs the check
1246 * before the default router selection, we perform
1249 pfxlist_onlink_check();
1252 * Refresh default router list.
1258 * If this entry was added by an on-link redirect, remove the
1259 * corresponding host route.
1261 if (ln->la_flags & LLE_REDIRECT)
1262 nd6_free_redirect(ln);
1264 if (ln->ln_router || dr)
1269 * Save to unlock. We still hold an extra reference and will not
1270 * free(9) in llentry_free() if someone else holds one as well.
1273 IF_AFDATA_LOCK(ifp);
1275 /* Guard against race with other llentry_free(). */
1276 if (ln->la_flags & LLE_LINKED) {
1277 /* Remove callout reference */
1279 lltable_unlink_entry(ln->lle_tbl, ln);
1281 IF_AFDATA_UNLOCK(ifp);
1287 * Remove the rtentry for the given llentry,
1288 * both of which were installed by a redirect.
1291 nd6_free_redirect(const struct llentry *ln)
1295 struct radix_node_head *rnh;
1296 struct sockaddr_in6 sin6;
1298 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1299 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1300 rnh = rt_tables_get_rnh(fibnum, AF_INET6);
1304 RADIX_NODE_HEAD_LOCK(rnh);
1305 rt = in6_rtalloc1((struct sockaddr *)&sin6, 0,
1306 RTF_RNH_LOCKED, fibnum);
1308 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1309 rt_expunge(rnh, rt);
1312 RADIX_NODE_HEAD_UNLOCK(rnh);
1317 * Upper-layer reachability hint for Neighbor Unreachability Detection.
1319 * XXX cost-effective methods?
1322 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1327 if ((dst6 == NULL) || (rt == NULL))
1331 IF_AFDATA_RLOCK(ifp);
1332 ln = nd6_lookup(dst6, LLE_EXCLUSIVE, NULL);
1333 IF_AFDATA_RUNLOCK(ifp);
1337 if (ln->ln_state < ND6_LLINFO_REACHABLE)
1341 * if we get upper-layer reachability confirmation many times,
1342 * it is possible we have false information.
1346 if (ln->ln_byhint > V_nd6_maxnudhint) {
1351 nd6_llinfo_setstate(ln, ND6_LLINFO_REACHABLE);
1358 * Rejuvenate this function for routing operations related
1362 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1364 struct sockaddr_in6 *gateway;
1365 struct nd_defrouter *dr;
1368 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1379 * Only indirect routes are interesting.
1381 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1384 * check for default route
1386 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1387 &SIN6(rt_key(rt))->sin6_addr)) {
1389 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1399 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1401 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1402 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1403 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1406 if (ifp->if_afdata[AF_INET6] == NULL)
1407 return (EPFNOSUPPORT);
1409 case OSIOCGIFINFO_IN6:
1411 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1412 bzero(&ND, sizeof(ND));
1413 ND.linkmtu = IN6_LINKMTU(ifp);
1414 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1415 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1416 ND.reachable = ND_IFINFO(ifp)->reachable;
1417 ND.retrans = ND_IFINFO(ifp)->retrans;
1418 ND.flags = ND_IFINFO(ifp)->flags;
1419 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1420 ND.chlim = ND_IFINFO(ifp)->chlim;
1422 case SIOCGIFINFO_IN6:
1423 ND = *ND_IFINFO(ifp);
1425 case SIOCSIFINFO_IN6:
1427 * used to change host variables from userland.
1428 * intented for a use on router to reflect RA configurations.
1430 /* 0 means 'unspecified' */
1431 if (ND.linkmtu != 0) {
1432 if (ND.linkmtu < IPV6_MMTU ||
1433 ND.linkmtu > IN6_LINKMTU(ifp)) {
1437 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1440 if (ND.basereachable != 0) {
1441 int obasereachable = ND_IFINFO(ifp)->basereachable;
1443 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1444 if (ND.basereachable != obasereachable)
1445 ND_IFINFO(ifp)->reachable =
1446 ND_COMPUTE_RTIME(ND.basereachable);
1448 if (ND.retrans != 0)
1449 ND_IFINFO(ifp)->retrans = ND.retrans;
1451 ND_IFINFO(ifp)->chlim = ND.chlim;
1453 case SIOCSIFINFO_FLAGS:
1456 struct in6_ifaddr *ia;
1458 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1459 !(ND.flags & ND6_IFF_IFDISABLED)) {
1460 /* ifdisabled 1->0 transision */
1463 * If the interface is marked as ND6_IFF_IFDISABLED and
1464 * has an link-local address with IN6_IFF_DUPLICATED,
1465 * do not clear ND6_IFF_IFDISABLED.
1466 * See RFC 4862, Section 5.4.5.
1469 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1470 if (ifa->ifa_addr->sa_family != AF_INET6)
1472 ia = (struct in6_ifaddr *)ifa;
1473 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1474 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1477 IF_ADDR_RUNLOCK(ifp);
1480 /* LLA is duplicated. */
1481 ND.flags |= ND6_IFF_IFDISABLED;
1482 log(LOG_ERR, "Cannot enable an interface"
1483 " with a link-local address marked"
1486 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1487 if (ifp->if_flags & IFF_UP)
1490 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1491 (ND.flags & ND6_IFF_IFDISABLED)) {
1492 /* ifdisabled 0->1 transision */
1493 /* Mark all IPv6 address as tentative. */
1495 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1496 if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1498 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1500 if (ifa->ifa_addr->sa_family !=
1503 ia = (struct in6_ifaddr *)ifa;
1504 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1506 IF_ADDR_RUNLOCK(ifp);
1510 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1511 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1512 /* auto_linklocal 0->1 transision */
1514 /* If no link-local address on ifp, configure */
1515 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1516 in6_ifattach(ifp, NULL);
1517 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1518 ifp->if_flags & IFF_UP) {
1520 * When the IF already has
1521 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1522 * address is assigned, and IFF_UP, try to
1526 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1528 if (ifa->ifa_addr->sa_family !=
1531 ia = (struct in6_ifaddr *)ifa;
1532 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1535 IF_ADDR_RUNLOCK(ifp);
1537 /* No LLA is configured. */
1538 in6_ifattach(ifp, NULL);
1542 ND_IFINFO(ifp)->flags = ND.flags;
1545 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1546 /* sync kernel routing table with the default router list */
1550 case SIOCSPFXFLUSH_IN6:
1552 /* flush all the prefix advertised by routers */
1553 struct nd_prefix *pr, *next;
1555 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1556 struct in6_ifaddr *ia, *ia_next;
1558 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1561 /* do we really have to remove addresses as well? */
1562 /* XXXRW: in6_ifaddrhead locking. */
1563 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1565 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1568 if (ia->ia6_ndpr == pr)
1569 in6_purgeaddr(&ia->ia_ifa);
1575 case SIOCSRTRFLUSH_IN6:
1577 /* flush all the default routers */
1578 struct nd_defrouter *dr, *next;
1581 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1587 case SIOCGNBRINFO_IN6:
1590 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1592 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1595 IF_AFDATA_RLOCK(ifp);
1596 ln = nd6_lookup(&nb_addr, 0, ifp);
1597 IF_AFDATA_RUNLOCK(ifp);
1603 nbi->state = ln->ln_state;
1604 nbi->asked = ln->la_asked;
1605 nbi->isrouter = ln->ln_router;
1606 if (ln->la_expire == 0)
1609 nbi->expire = ln->la_expire +
1610 (time_second - time_uptime);
1614 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1615 ndif->ifindex = V_nd6_defifindex;
1617 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1618 return (nd6_setdefaultiface(ndif->ifindex));
1624 * Calculates new isRouter value based on provided parameters and
1628 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1633 * ICMP6 type dependent behavior.
1635 * NS: clear IsRouter if new entry
1636 * RS: clear IsRouter
1637 * RA: set IsRouter if there's lladdr
1638 * redir: clear IsRouter if new entry
1641 * The spec says that we must set IsRouter in the following cases:
1642 * - If lladdr exist, set IsRouter. This means (1-5).
1643 * - If it is old entry (!newentry), set IsRouter. This means (7).
1644 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1645 * A quetion arises for (1) case. (1) case has no lladdr in the
1646 * neighbor cache, this is similar to (6).
1647 * This case is rare but we figured that we MUST NOT set IsRouter.
1649 * is_new old_addr new_addr NS RS RA redir
1656 * 1 -- n (6) c c c s
1657 * 1 -- y (7) c c s c s
1661 switch (type & 0xff) {
1662 case ND_NEIGHBOR_SOLICIT:
1664 * New entry must have is_router flag cleared.
1666 if (is_new) /* (6-7) */
1671 * If the icmp is a redirect to a better router, always set the
1672 * is_router flag. Otherwise, if the entry is newly created,
1673 * clear the flag. [RFC 2461, sec 8.3]
1675 if (code == ND_REDIRECT_ROUTER)
1678 if (is_new) /* (6-7) */
1682 case ND_ROUTER_SOLICIT:
1684 * is_router flag must always be cleared.
1688 case ND_ROUTER_ADVERT:
1690 * Mark an entry with lladdr as a router.
1692 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */
1693 (is_new && new_addr)) { /* (7) */
1703 * Create neighbor cache entry and cache link-layer address,
1704 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1707 * code - type dependent information
1711 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1712 int lladdrlen, int type, int code)
1714 struct llentry *ln = NULL, *ln_tmp;
1720 uint16_t router = 0;
1721 struct sockaddr_in6 sin6;
1722 struct mbuf *chain = NULL;
1724 IF_AFDATA_UNLOCK_ASSERT(ifp);
1726 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1727 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1729 /* nothing must be updated for unspecified address */
1730 if (IN6_IS_ADDR_UNSPECIFIED(from))
1734 * Validation about ifp->if_addrlen and lladdrlen must be done in
1737 * XXX If the link does not have link-layer adderss, what should
1738 * we do? (ifp->if_addrlen == 0)
1739 * Spec says nothing in sections for RA, RS and NA. There's small
1740 * description on it in NS section (RFC 2461 7.2.3).
1742 flags = lladdr ? LLE_EXCLUSIVE : 0;
1743 IF_AFDATA_RLOCK(ifp);
1744 ln = nd6_lookup(from, flags, ifp);
1745 IF_AFDATA_RUNLOCK(ifp);
1748 flags |= LLE_EXCLUSIVE;
1749 ln = nd6_alloc(from, 0, ifp);
1754 * Since we already know all the data for the new entry,
1755 * fill it before insertion.
1757 if (lladdr != NULL) {
1758 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1759 ln->la_flags |= LLE_VALID;
1761 IF_AFDATA_WLOCK(ifp);
1763 /* Prefer any existing lle over newly-created one */
1764 ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1766 lltable_link_entry(LLTABLE6(ifp), ln);
1767 IF_AFDATA_WUNLOCK(ifp);
1768 if (ln_tmp == NULL) {
1769 /* No existing lle, mark as new entry */
1771 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1773 lltable_free_entry(LLTABLE6(ifp), ln);
1778 /* do nothing if static ndp is set */
1779 if ((ln->la_flags & LLE_STATIC)) {
1780 if (flags & LLE_EXCLUSIVE)
1787 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1788 if (olladdr && lladdr) {
1789 llchange = bcmp(lladdr, &ln->ll_addr,
1791 } else if (!olladdr && lladdr)
1797 * newentry olladdr lladdr llchange (*=record)
1800 * 0 n y y (3) * STALE
1802 * 0 y y y (5) * STALE
1803 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1804 * 1 -- y -- (7) * STALE
1808 if (!is_newentry && llchange != 0)
1809 do_update = 1; /* (3,5) */
1811 if (lladdr) { /* (3-5) and (7) */
1813 * Record source link-layer address
1814 * XXX is it dependent to ifp->if_type?
1816 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1817 ln->la_flags |= LLE_VALID;
1818 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1820 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1823 if (ln->la_hold != NULL)
1824 nd6_grab_holdchain(ln, &chain, &sin6);
1828 /* Calculates new router status */
1829 router = nd6_is_router(type, code, is_newentry, olladdr,
1830 lladdr != NULL ? 1 : 0, ln->ln_router);
1832 ln->ln_router = router;
1833 /* Mark non-router redirects with special flag */
1834 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
1835 ln->la_flags |= LLE_REDIRECT;
1837 if (flags & LLE_EXCLUSIVE)
1843 nd6_flush_holdchain(ifp, ifp, chain, &sin6);
1846 * When the link-layer address of a router changes, select the
1847 * best router again. In particular, when the neighbor entry is newly
1848 * created, it might affect the selection policy.
1849 * Question: can we restrict the first condition to the "is_newentry"
1851 * XXX: when we hear an RA from a new router with the link-layer
1852 * address option, defrouter_select() is called twice, since
1853 * defrtrlist_update called the function as well. However, I believe
1854 * we can compromise the overhead, since it only happens the first
1856 * XXX: although defrouter_select() should not have a bad effect
1857 * for those are not autoconfigured hosts, we explicitly avoid such
1860 if ((do_update || is_newentry) && router &&
1861 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1863 * guaranteed recursion
1870 nd6_slowtimo(void *arg)
1872 CURVNET_SET((struct vnet *) arg);
1873 struct nd_ifinfo *nd6if;
1876 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1877 nd6_slowtimo, curvnet);
1878 IFNET_RLOCK_NOSLEEP();
1879 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1880 if (ifp->if_afdata[AF_INET6] == NULL)
1882 nd6if = ND_IFINFO(ifp);
1883 if (nd6if->basereachable && /* already initialized */
1884 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1886 * Since reachable time rarely changes by router
1887 * advertisements, we SHOULD insure that a new random
1888 * value gets recomputed at least once every few hours.
1891 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1892 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1895 IFNET_RUNLOCK_NOSLEEP();
1900 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
1901 struct sockaddr_in6 *sin6)
1904 LLE_WLOCK_ASSERT(ln);
1906 *chain = ln->la_hold;
1908 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
1910 if (ln->ln_state == ND6_LLINFO_STALE) {
1913 * The first time we send a packet to a
1914 * neighbor whose entry is STALE, we have
1915 * to change the state to DELAY and a sets
1916 * a timer to expire in DELAY_FIRST_PROBE_TIME
1917 * seconds to ensure do neighbor unreachability
1918 * detection on expiration.
1921 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
1926 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1927 struct sockaddr_in6 *dst)
1931 struct ip6_hdr *ip6;
1935 mac_netinet6_nd6_send(ifp, m);
1939 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1940 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1941 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1942 * to be diverted to user space. When re-injected into the kernel,
1943 * send_output() will directly dispatch them to the outgoing interface.
1945 if (send_sendso_input_hook != NULL) {
1946 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1948 ip6 = mtod(m, struct ip6_hdr *);
1949 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1950 /* Use the SEND socket */
1951 error = send_sendso_input_hook(m, ifp, SND_OUT,
1953 /* -1 == no app on SEND socket */
1954 if (error == 0 || error != -1)
1959 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1960 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1961 mtod(m, struct ip6_hdr *));
1963 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1966 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1971 * Do L2 address resolution for @sa_dst address. Stores found
1972 * address in @desten buffer. Copy of lle ln_flags can be also
1973 * saved in @pflags if @pflags is non-NULL.
1975 * If destination LLE does not exists or lle state modification
1976 * is required, call "slow" version.
1979 * - 0 on success (address copied to buffer).
1980 * - EWOULDBLOCK (no local error, but address is still unresolved)
1981 * - other errors (alloc failure, etc)
1984 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
1985 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
1987 struct llentry *ln = NULL;
1988 const struct sockaddr_in6 *dst6;
1993 dst6 = (const struct sockaddr_in6 *)sa_dst;
1995 /* discard the packet if IPv6 operation is disabled on the interface */
1996 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1998 return (ENETDOWN); /* better error? */
2001 if (m != NULL && m->m_flags & M_MCAST) {
2002 switch (ifp->if_type) {
2009 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2014 return (EAFNOSUPPORT);
2018 IF_AFDATA_RLOCK(ifp);
2019 ln = nd6_lookup(&dst6->sin6_addr, 0, ifp);
2020 IF_AFDATA_RUNLOCK(ifp);
2023 * Perform fast path for the following cases:
2024 * 1) lle state is REACHABLE
2025 * 2) lle state is DELAY (NS message sent)
2027 * Every other case involves lle modification, so we handle
2030 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
2031 ln->ln_state != ND6_LLINFO_DELAY)) {
2032 /* Fall back to slow processing path */
2035 return (nd6_resolve_slow(ifp, m, dst6, desten, pflags));
2039 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2041 *pflags = ln->la_flags;
2048 * Do L2 address resolution for @sa_dst address. Stores found
2049 * address in @desten buffer. Copy of lle ln_flags can be also
2050 * saved in @pflags if @pflags is non-NULL.
2053 * Function assume that destination LLE does not exist,
2054 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2057 nd6_resolve_slow(struct ifnet *ifp, struct mbuf *m,
2058 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2060 struct llentry *lle = NULL, *lle_tmp;
2061 struct in6_addr *psrc, src;
2065 * Address resolution or Neighbor Unreachability Detection
2067 * At this point, the destination of the packet must be a unicast
2068 * or an anycast address(i.e. not a multicast).
2071 IF_AFDATA_RLOCK(ifp);
2072 lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2073 IF_AFDATA_RUNLOCK(ifp);
2074 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2076 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2077 * the condition below is not very efficient. But we believe
2078 * it is tolerable, because this should be a rare case.
2080 lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2082 char ip6buf[INET6_ADDRSTRLEN];
2084 "nd6_output: can't allocate llinfo for %s "
2086 ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2091 IF_AFDATA_WLOCK(ifp);
2093 /* Prefer any existing entry over newly-created one */
2094 lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2095 if (lle_tmp == NULL)
2096 lltable_link_entry(LLTABLE6(ifp), lle);
2097 IF_AFDATA_WUNLOCK(ifp);
2098 if (lle_tmp != NULL) {
2099 lltable_free_entry(LLTABLE6(ifp), lle);
2106 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2116 LLE_WLOCK_ASSERT(lle);
2119 * The first time we send a packet to a neighbor whose entry is
2120 * STALE, we have to change the state to DELAY and a sets a timer to
2121 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2122 * neighbor unreachability detection on expiration.
2125 if (lle->ln_state == ND6_LLINFO_STALE)
2126 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2129 * If the neighbor cache entry has a state other than INCOMPLETE
2130 * (i.e. its link-layer address is already resolved), just
2133 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2134 bcopy(&lle->ll_addr, desten, ifp->if_addrlen);
2136 *pflags = lle->la_flags;
2142 * There is a neighbor cache entry, but no ethernet address
2143 * response yet. Append this latest packet to the end of the
2144 * packet queue in the mbuf, unless the number of the packet
2145 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
2146 * the oldest packet in the queue will be removed.
2149 if (lle->la_hold != NULL) {
2150 struct mbuf *m_hold;
2154 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2156 if (m_hold->m_nextpkt == NULL) {
2157 m_hold->m_nextpkt = m;
2161 while (i >= V_nd6_maxqueuelen) {
2162 m_hold = lle->la_hold;
2163 lle->la_hold = lle->la_hold->m_nextpkt;
2172 * If there has been no NS for the neighbor after entering the
2173 * INCOMPLETE state, send the first solicitation.
2174 * Note that for newly-created lle la_asked will be 0,
2175 * so we will transition from ND6_LLINFO_NOSTATE to
2176 * ND6_LLINFO_INCOMPLETE state here.
2180 if (lle->la_asked == 0) {
2183 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2185 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2189 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2191 return (EWOULDBLOCK);
2196 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2197 struct sockaddr_in6 *dst)
2199 struct mbuf *m, *m_head;
2200 struct ifnet *outifp;
2204 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2211 m_head = m_head->m_nextpkt;
2212 error = nd6_output_ifp(ifp, origifp, m, dst);
2217 * note that intermediate errors are blindly ignored
2223 nd6_need_cache(struct ifnet *ifp)
2226 * XXX: we currently do not make neighbor cache on any interface
2227 * other than ARCnet, Ethernet, FDDI and GIF.
2230 * - unidirectional tunnels needs no ND
2232 switch (ifp->if_type) {
2239 case IFT_INFINIBAND:
2241 case IFT_PROPVIRTUAL:
2249 * Add pernament ND6 link-layer record for given
2250 * interface address.
2252 * Very similar to IPv4 arp_ifinit(), but:
2253 * 1) IPv6 DAD is performed in different place
2254 * 2) It is called by IPv6 protocol stack in contrast to
2255 * arp_ifinit() which is typically called in SIOCSIFADDR
2256 * driver ioctl handler.
2260 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2263 struct llentry *ln, *ln_tmp;
2264 struct sockaddr *dst;
2266 ifp = ia->ia_ifa.ifa_ifp;
2267 if (nd6_need_cache(ifp) == 0)
2270 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2271 dst = (struct sockaddr *)&ia->ia_addr;
2272 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2276 IF_AFDATA_WLOCK(ifp);
2278 /* Unlink any entry if exists */
2279 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2281 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2282 lltable_link_entry(LLTABLE6(ifp), ln);
2283 IF_AFDATA_WUNLOCK(ifp);
2286 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2287 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2291 llentry_free(ln_tmp);
2297 * Removes either all lle entries for given @ia, or lle
2298 * corresponding to @ia address.
2301 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2303 struct sockaddr_in6 mask, addr;
2304 struct sockaddr *saddr, *smask;
2307 ifp = ia->ia_ifa.ifa_ifp;
2308 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2309 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2310 saddr = (struct sockaddr *)&addr;
2311 smask = (struct sockaddr *)&mask;
2314 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2316 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2320 clear_llinfo_pqueue(struct llentry *ln)
2322 struct mbuf *m_hold, *m_hold_next;
2324 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2325 m_hold_next = m_hold->m_nextpkt;
2333 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2334 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2336 SYSCTL_DECL(_net_inet6_icmp6);
2338 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2339 CTLFLAG_RD, nd6_sysctl_drlist, "");
2340 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2341 CTLFLAG_RD, nd6_sysctl_prlist, "");
2342 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2343 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2344 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2345 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2348 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2350 struct in6_defrouter d;
2351 struct nd_defrouter *dr;
2357 bzero(&d, sizeof(d));
2358 d.rtaddr.sin6_family = AF_INET6;
2359 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2364 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2365 d.rtaddr.sin6_addr = dr->rtaddr;
2366 error = sa6_recoverscope(&d.rtaddr);
2369 d.flags = dr->flags;
2370 d.rtlifetime = dr->rtlifetime;
2371 d.expire = dr->expire + (time_second - time_uptime);
2372 d.if_index = dr->ifp->if_index;
2373 error = SYSCTL_OUT(req, &d, sizeof(d));
2381 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2383 struct in6_prefix p;
2384 struct sockaddr_in6 s6;
2385 struct nd_prefix *pr;
2386 struct nd_pfxrouter *pfr;
2389 char ip6buf[INET6_ADDRSTRLEN];
2394 bzero(&p, sizeof(p));
2395 p.origin = PR_ORIG_RA;
2396 bzero(&s6, sizeof(s6));
2397 s6.sin6_family = AF_INET6;
2398 s6.sin6_len = sizeof(s6);
2403 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2404 p.prefix = pr->ndpr_prefix;
2405 if (sa6_recoverscope(&p.prefix)) {
2406 log(LOG_ERR, "scope error in prefix list (%s)\n",
2407 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2408 /* XXX: press on... */
2410 p.raflags = pr->ndpr_raf;
2411 p.prefixlen = pr->ndpr_plen;
2412 p.vltime = pr->ndpr_vltime;
2413 p.pltime = pr->ndpr_pltime;
2414 p.if_index = pr->ndpr_ifp->if_index;
2415 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2418 /* XXX: we assume time_t is signed. */
2420 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2421 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2422 p.expire = pr->ndpr_lastupdate +
2424 (time_second - time_uptime);
2426 p.expire = maxexpire;
2428 p.refcnt = pr->ndpr_refcnt;
2429 p.flags = pr->ndpr_stateflags;
2431 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2433 error = SYSCTL_OUT(req, &p, sizeof(p));
2436 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2437 s6.sin6_addr = pfr->router->rtaddr;
2438 if (sa6_recoverscope(&s6))
2440 "scope error in prefix list (%s)\n",
2441 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2442 error = SYSCTL_OUT(req, &s6, sizeof(s6));