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 nd6_llinfo_settimer_locked(struct llentry *, long);
138 static void clear_llinfo_pqueue(struct llentry *);
139 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
140 static int nd6_resolve_slow(struct ifnet *, struct mbuf *,
141 const struct sockaddr_in6 *, u_char *, uint32_t *);
142 static int nd6_need_cache(struct ifnet *);
145 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
146 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
148 VNET_DEFINE(struct callout, nd6_timer_ch);
151 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
153 struct rt_addrinfo rtinfo;
154 struct sockaddr_in6 dst;
155 struct sockaddr_dl gw;
159 LLE_WLOCK_ASSERT(lle);
161 if (lltable_get_af(lle->lle_tbl) != AF_INET6)
165 case LLENTRY_RESOLVED:
167 KASSERT(lle->la_flags & LLE_VALID,
168 ("%s: %p resolved but not valid?", __func__, lle));
170 case LLENTRY_EXPIRED:
177 ifp = lltable_get_ifp(lle->lle_tbl);
179 bzero(&dst, sizeof(dst));
180 bzero(&gw, sizeof(gw));
181 bzero(&rtinfo, sizeof(rtinfo));
182 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
183 dst.sin6_scope_id = in6_getscopezone(ifp,
184 in6_addrscope(&dst.sin6_addr));
185 gw.sdl_len = sizeof(struct sockaddr_dl);
186 gw.sdl_family = AF_LINK;
187 gw.sdl_alen = ifp->if_addrlen;
188 gw.sdl_index = ifp->if_index;
189 gw.sdl_type = ifp->if_type;
190 if (evt == LLENTRY_RESOLVED)
191 bcopy(&lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
192 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
193 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
194 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
195 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
196 type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
203 LIST_INIT(&V_nd_prefix);
205 /* initialization of the default router list */
206 TAILQ_INIT(&V_nd_defrouter);
209 callout_init(&V_nd6_slowtimo_ch, 0);
210 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
211 nd6_slowtimo, curvnet);
214 if (IS_DEFAULT_VNET(curvnet))
215 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
216 NULL, EVENTHANDLER_PRI_ANY);
224 callout_drain(&V_nd6_slowtimo_ch);
225 callout_drain(&V_nd6_timer_ch);
226 if (IS_DEFAULT_VNET(curvnet))
227 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
232 nd6_ifattach(struct ifnet *ifp)
234 struct nd_ifinfo *nd;
236 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
239 nd->chlim = IPV6_DEFHLIM;
240 nd->basereachable = REACHABLE_TIME;
241 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
242 nd->retrans = RETRANS_TIMER;
244 nd->flags = ND6_IFF_PERFORMNUD;
246 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
247 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
248 * default regardless of the V_ip6_auto_linklocal configuration to
249 * give a reasonable default behavior.
251 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
252 (ifp->if_flags & IFF_LOOPBACK))
253 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
255 * A loopback interface does not need to accept RTADV.
256 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
257 * default regardless of the V_ip6_accept_rtadv configuration to
258 * prevent the interface from accepting RA messages arrived
259 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
261 if (V_ip6_accept_rtadv &&
262 !(ifp->if_flags & IFF_LOOPBACK) &&
263 (ifp->if_type != IFT_BRIDGE))
264 nd->flags |= ND6_IFF_ACCEPT_RTADV;
265 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
266 nd->flags |= ND6_IFF_NO_RADR;
268 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
269 nd6_setmtu0(ifp, nd);
275 nd6_ifdetach(struct nd_ifinfo *nd)
282 * Reset ND level link MTU. This function is called when the physical MTU
283 * changes, which means we might have to adjust the ND level MTU.
286 nd6_setmtu(struct ifnet *ifp)
289 nd6_setmtu0(ifp, ND_IFINFO(ifp));
292 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
294 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
298 omaxmtu = ndi->maxmtu;
300 switch (ifp->if_type) {
302 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
305 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
308 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
311 ndi->maxmtu = ifp->if_mtu;
316 * Decreasing the interface MTU under IPV6 minimum MTU may cause
317 * undesirable situation. We thus notify the operator of the change
318 * explicitly. The check for omaxmtu is necessary to restrict the
319 * log to the case of changing the MTU, not initializing it.
321 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
322 log(LOG_NOTICE, "nd6_setmtu0: "
323 "new link MTU on %s (%lu) is too small for IPv6\n",
324 if_name(ifp), (unsigned long)ndi->maxmtu);
327 if (ndi->maxmtu > V_in6_maxmtu)
328 in6_setmaxmtu(); /* check all interfaces just in case */
333 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
336 bzero(ndopts, sizeof(*ndopts));
337 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
339 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
342 ndopts->nd_opts_done = 1;
343 ndopts->nd_opts_search = NULL;
348 * Take one ND option.
351 nd6_option(union nd_opts *ndopts)
353 struct nd_opt_hdr *nd_opt;
356 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
357 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
359 if (ndopts->nd_opts_search == NULL)
361 if (ndopts->nd_opts_done)
364 nd_opt = ndopts->nd_opts_search;
366 /* make sure nd_opt_len is inside the buffer */
367 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
368 bzero(ndopts, sizeof(*ndopts));
372 olen = nd_opt->nd_opt_len << 3;
375 * Message validation requires that all included
376 * options have a length that is greater than zero.
378 bzero(ndopts, sizeof(*ndopts));
382 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
383 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
384 /* option overruns the end of buffer, invalid */
385 bzero(ndopts, sizeof(*ndopts));
387 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
388 /* reached the end of options chain */
389 ndopts->nd_opts_done = 1;
390 ndopts->nd_opts_search = NULL;
396 * Parse multiple ND options.
397 * This function is much easier to use, for ND routines that do not need
398 * multiple options of the same type.
401 nd6_options(union nd_opts *ndopts)
403 struct nd_opt_hdr *nd_opt;
406 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
407 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
409 if (ndopts->nd_opts_search == NULL)
413 nd_opt = nd6_option(ndopts);
414 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
416 * Message validation requires that all included
417 * options have a length that is greater than zero.
419 ICMP6STAT_INC(icp6s_nd_badopt);
420 bzero(ndopts, sizeof(*ndopts));
427 switch (nd_opt->nd_opt_type) {
428 case ND_OPT_SOURCE_LINKADDR:
429 case ND_OPT_TARGET_LINKADDR:
431 case ND_OPT_REDIRECTED_HEADER:
433 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
435 "duplicated ND6 option found (type=%d)\n",
436 nd_opt->nd_opt_type));
439 ndopts->nd_opt_array[nd_opt->nd_opt_type]
443 case ND_OPT_PREFIX_INFORMATION:
444 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
445 ndopts->nd_opt_array[nd_opt->nd_opt_type]
448 ndopts->nd_opts_pi_end =
449 (struct nd_opt_prefix_info *)nd_opt;
451 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
452 case ND_OPT_RDNSS: /* RFC 6106 */
453 case ND_OPT_DNSSL: /* RFC 6106 */
455 * Silently ignore options we know and do not care about
461 * Unknown options must be silently ignored,
462 * to accomodate future extension to the protocol.
465 "nd6_options: unsupported option %d - "
466 "option ignored\n", nd_opt->nd_opt_type));
471 if (i > V_nd6_maxndopt) {
472 ICMP6STAT_INC(icp6s_nd_toomanyopt);
473 nd6log((LOG_INFO, "too many loop in nd opt\n"));
477 if (ndopts->nd_opts_done)
485 * ND6 timer routine to handle ND6 entries
488 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
492 LLE_WLOCK_ASSERT(ln);
497 canceled = callout_stop(&ln->lle_timer);
499 ln->la_expire = time_uptime + tick / hz;
501 if (tick > INT_MAX) {
502 ln->ln_ntick = tick - INT_MAX;
503 canceled = callout_reset(&ln->lle_timer, INT_MAX,
504 nd6_llinfo_timer, ln);
507 canceled = callout_reset(&ln->lle_timer, tick,
508 nd6_llinfo_timer, ln);
516 * Gets source address of the first packet in hold queue
517 * and stores it in @src.
518 * Returns pointer to @src (if hold queue is not empty) or NULL.
520 * Set noinline to be dtrace-friendly
522 static __noinline struct in6_addr *
523 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
528 if (ln->la_hold == NULL)
532 * assume every packet in la_hold has the same IP header
535 if (sizeof(hdr) > m->m_len)
538 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
545 * Switch @lle state to new state optionally arming timers.
547 * Set noinline to be dtrace-friendly
550 nd6_llinfo_setstate(struct llentry *lle, int newstate)
558 case ND6_LLINFO_INCOMPLETE:
559 ifp = lle->lle_tbl->llt_ifp;
560 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
562 case ND6_LLINFO_REACHABLE:
563 if (!ND6_LLINFO_PERMANENT(lle)) {
564 ifp = lle->lle_tbl->llt_ifp;
565 delay = (long)ND_IFINFO(ifp)->reachable * hz;
568 case ND6_LLINFO_STALE:
569 delay = (long)V_nd6_gctimer * hz;
571 case ND6_LLINFO_DELAY:
573 delay = (long)V_nd6_delay * hz;
578 nd6_llinfo_settimer_locked(lle, delay);
580 lle->ln_state = newstate;
584 * Timer-dependent part of nd state machine.
586 * Set noinline to be dtrace-friendly
588 static __noinline void
589 nd6_llinfo_timer(void *arg)
592 struct in6_addr *dst, *pdst, *psrc, src;
594 struct nd_ifinfo *ndi = NULL;
597 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
598 ln = (struct llentry *)arg;
600 if (callout_pending(&ln->lle_timer)) {
602 * Here we are a bit odd here in the treatment of
603 * active/pending. If the pending bit is set, it got
604 * rescheduled before I ran. The active
605 * bit we ignore, since if it was stopped
606 * in ll_tablefree() and was currently running
607 * it would have return 0 so the code would
608 * not have deleted it since the callout could
609 * not be stopped so we want to go through
610 * with the delete here now. If the callout
611 * was restarted, the pending bit will be back on and
612 * we just want to bail since the callout_reset would
613 * return 1 and our reference would have been removed
614 * by nd6_llinfo_settimer_locked above since canceled
620 ifp = ln->lle_tbl->llt_ifp;
621 CURVNET_SET(ifp->if_vnet);
622 ndi = ND_IFINFO(ifp);
624 dst = &ln->r_l3addr.addr6;
627 if (ln->ln_ntick > 0) {
628 if (ln->ln_ntick > INT_MAX) {
629 ln->ln_ntick -= INT_MAX;
630 nd6_llinfo_settimer_locked(ln, INT_MAX);
633 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
638 if (ln->la_flags & LLE_STATIC) {
642 if (ln->la_flags & LLE_DELETED) {
648 switch (ln->ln_state) {
649 case ND6_LLINFO_INCOMPLETE:
650 if (ln->la_asked < V_nd6_mmaxtries) {
653 /* Send NS to multicast address */
656 struct mbuf *m = ln->la_hold;
661 * assuming every packet in la_hold has the
662 * same IP header. Send error after unlock.
667 clear_llinfo_pqueue(ln);
669 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
673 icmp6_error2(m, ICMP6_DST_UNREACH,
674 ICMP6_DST_UNREACH_ADDR, 0, ifp);
677 case ND6_LLINFO_REACHABLE:
678 if (!ND6_LLINFO_PERMANENT(ln))
679 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
682 case ND6_LLINFO_STALE:
683 /* Garbage Collection(RFC 2461 5.3) */
684 if (!ND6_LLINFO_PERMANENT(ln)) {
685 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
691 case ND6_LLINFO_DELAY:
692 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
695 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
698 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
700 case ND6_LLINFO_PROBE:
701 if (ln->la_asked < V_nd6_umaxtries) {
705 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
711 panic("%s: paths in a dark night can be confusing: %d",
712 __func__, ln->ln_state);
716 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
717 psrc = nd6_llinfo_get_holdsrc(ln, &src);
720 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
730 * ND6 timer routine to expire default route list and prefix list
735 CURVNET_SET((struct vnet *) arg);
736 struct nd_defrouter *dr, *ndr;
737 struct nd_prefix *pr, *npr;
738 struct in6_ifaddr *ia6, *nia6;
740 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
743 /* expire default router list */
744 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
745 if (dr->expire && dr->expire < time_uptime)
750 * expire interface addresses.
751 * in the past the loop was inside prefix expiry processing.
752 * However, from a stricter speci-confrmance standpoint, we should
753 * rather separate address lifetimes and prefix lifetimes.
755 * XXXRW: in6_ifaddrhead locking.
758 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
759 /* check address lifetime */
760 if (IFA6_IS_INVALID(ia6)) {
764 * If the expiring address is temporary, try
765 * regenerating a new one. This would be useful when
766 * we suspended a laptop PC, then turned it on after a
767 * period that could invalidate all temporary
768 * addresses. Although we may have to restart the
769 * loop (see below), it must be after purging the
770 * address. Otherwise, we'd see an infinite loop of
773 if (V_ip6_use_tempaddr &&
774 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
775 if (regen_tmpaddr(ia6) == 0)
779 in6_purgeaddr(&ia6->ia_ifa);
782 goto addrloop; /* XXX: see below */
783 } else if (IFA6_IS_DEPRECATED(ia6)) {
784 int oldflags = ia6->ia6_flags;
786 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
789 * If a temporary address has just become deprecated,
790 * regenerate a new one if possible.
792 if (V_ip6_use_tempaddr &&
793 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
794 (oldflags & IN6_IFF_DEPRECATED) == 0) {
796 if (regen_tmpaddr(ia6) == 0) {
798 * A new temporary address is
800 * XXX: this means the address chain
801 * has changed while we are still in
802 * the loop. Although the change
803 * would not cause disaster (because
804 * it's not a deletion, but an
805 * addition,) we'd rather restart the
806 * loop just for safety. Or does this
807 * significantly reduce performance??
812 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
814 * Schedule DAD for a tentative address. This happens
815 * if the interface was down or not running
816 * when the address was configured.
820 delay = arc4random() %
821 (MAX_RTR_SOLICITATION_DELAY * hz);
822 nd6_dad_start((struct ifaddr *)ia6, delay);
825 * Check status of the interface. If it is down,
826 * mark the address as tentative for future DAD.
828 if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
829 (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
831 (ND_IFINFO(ia6->ia_ifp)->flags &
832 ND6_IFF_IFDISABLED) != 0) {
833 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
834 ia6->ia6_flags |= IN6_IFF_TENTATIVE;
837 * A new RA might have made a deprecated address
840 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
844 /* expire prefix list */
845 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
847 * check prefix lifetime.
848 * since pltime is just for autoconf, pltime processing for
849 * prefix is not necessary.
851 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
852 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
855 * address expiration and prefix expiration are
856 * separate. NEVER perform in6_purgeaddr here.
865 * ia6 - deprecated/invalidated temporary address
868 regen_tmpaddr(struct in6_ifaddr *ia6)
872 struct in6_ifaddr *public_ifa6 = NULL;
874 ifp = ia6->ia_ifa.ifa_ifp;
876 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
877 struct in6_ifaddr *it6;
879 if (ifa->ifa_addr->sa_family != AF_INET6)
882 it6 = (struct in6_ifaddr *)ifa;
884 /* ignore no autoconf addresses. */
885 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
888 /* ignore autoconf addresses with different prefixes. */
889 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
893 * Now we are looking at an autoconf address with the same
894 * prefix as ours. If the address is temporary and is still
895 * preferred, do not create another one. It would be rare, but
896 * could happen, for example, when we resume a laptop PC after
899 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
900 !IFA6_IS_DEPRECATED(it6)) {
906 * This is a public autoconf address that has the same prefix
907 * as ours. If it is preferred, keep it. We can't break the
908 * loop here, because there may be a still-preferred temporary
909 * address with the prefix.
911 if (!IFA6_IS_DEPRECATED(it6))
914 if (public_ifa6 != NULL)
915 ifa_ref(&public_ifa6->ia_ifa);
916 IF_ADDR_RUNLOCK(ifp);
918 if (public_ifa6 != NULL) {
921 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
922 ifa_free(&public_ifa6->ia_ifa);
923 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
924 " tmp addr,errno=%d\n", e);
927 ifa_free(&public_ifa6->ia_ifa);
935 * Nuke neighbor cache/prefix/default router management table, right before
939 nd6_purge(struct ifnet *ifp)
941 struct nd_defrouter *dr, *ndr;
942 struct nd_prefix *pr, *npr;
945 * Nuke default router list entries toward ifp.
946 * We defer removal of default router list entries that is installed
947 * in the routing table, in order to keep additional side effects as
950 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
958 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
966 /* Nuke prefix list entries toward ifp */
967 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
968 if (pr->ndpr_ifp == ifp) {
970 * Because if_detach() does *not* release prefixes
971 * while purging addresses the reference count will
972 * still be above zero. We therefore reset it to
973 * make sure that the prefix really gets purged.
978 * Previously, pr->ndpr_addr is removed as well,
979 * but I strongly believe we don't have to do it.
980 * nd6_purge() is only called from in6_ifdetach(),
981 * which removes all the associated interface addresses
983 * (jinmei@kame.net 20010129)
989 /* cancel default outgoing interface setting */
990 if (V_nd6_defifindex == ifp->if_index)
991 nd6_setdefaultiface(0);
993 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
994 /* Refresh default router list. */
999 * We do not nuke the neighbor cache entries here any more
1000 * because the neighbor cache is kept in if_afdata[AF_INET6].
1001 * nd6_purge() is invoked by in6_ifdetach() which is called
1002 * from if_detach() where everything gets purged. So let
1003 * in6_domifdetach() do the actual L2 table purging work.
1008 * the caller acquires and releases the lock on the lltbls
1009 * Returns the llentry locked
1012 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1014 struct sockaddr_in6 sin6;
1017 bzero(&sin6, sizeof(sin6));
1018 sin6.sin6_len = sizeof(struct sockaddr_in6);
1019 sin6.sin6_family = AF_INET6;
1020 sin6.sin6_addr = *addr6;
1022 IF_AFDATA_LOCK_ASSERT(ifp);
1024 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1030 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1032 struct sockaddr_in6 sin6;
1035 bzero(&sin6, sizeof(sin6));
1036 sin6.sin6_len = sizeof(struct sockaddr_in6);
1037 sin6.sin6_family = AF_INET6;
1038 sin6.sin6_addr = *addr6;
1040 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1042 ln->ln_state = ND6_LLINFO_NOSTATE;
1048 * Test whether a given IPv6 address is a neighbor or not, ignoring
1049 * the actual neighbor cache. The neighbor cache is ignored in order
1050 * to not reenter the routing code from within itself.
1053 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1055 struct nd_prefix *pr;
1056 struct ifaddr *dstaddr;
1059 * A link-local address is always a neighbor.
1060 * XXX: a link does not necessarily specify a single interface.
1062 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1063 struct sockaddr_in6 sin6_copy;
1067 * We need sin6_copy since sa6_recoverscope() may modify the
1071 if (sa6_recoverscope(&sin6_copy))
1072 return (0); /* XXX: should be impossible */
1073 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1075 if (sin6_copy.sin6_scope_id == zone)
1082 * If the address matches one of our addresses,
1083 * it should be a neighbor.
1084 * If the address matches one of our on-link prefixes, it should be a
1087 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1088 if (pr->ndpr_ifp != ifp)
1091 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1094 /* Always use the default FIB here. */
1095 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
1096 0, 0, RT_DEFAULT_FIB);
1100 * This is the case where multiple interfaces
1101 * have the same prefix, but only one is installed
1102 * into the routing table and that prefix entry
1103 * is not the one being examined here. In the case
1104 * where RADIX_MPATH is enabled, multiple route
1105 * entries (of the same rt_key value) will be
1106 * installed because the interface addresses all
1109 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1110 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
1117 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1118 &addr->sin6_addr, &pr->ndpr_mask))
1123 * If the address is assigned on the node of the other side of
1124 * a p2p interface, the address should be a neighbor.
1126 dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1127 if (dstaddr != NULL) {
1128 if (dstaddr->ifa_ifp == ifp) {
1136 * If the default router list is empty, all addresses are regarded
1137 * as on-link, and thus, as a neighbor.
1139 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1140 TAILQ_EMPTY(&V_nd_defrouter) &&
1141 V_nd6_defifindex == ifp->if_index) {
1150 * Detect if a given IPv6 address identifies a neighbor on a given link.
1151 * XXX: should take care of the destination of a p2p link?
1154 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1156 struct llentry *lle;
1159 IF_AFDATA_UNLOCK_ASSERT(ifp);
1160 if (nd6_is_new_addr_neighbor(addr, ifp))
1164 * Even if the address matches none of our addresses, it might be
1165 * in the neighbor cache.
1167 IF_AFDATA_RLOCK(ifp);
1168 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1172 IF_AFDATA_RUNLOCK(ifp);
1177 * Free an nd6 llinfo entry.
1178 * Since the function would cause significant changes in the kernel, DO NOT
1179 * make it global, unless you have a strong reason for the change, and are sure
1180 * that the change is safe.
1182 * Set noinline to be dtrace-friendly
1184 static __noinline void
1185 nd6_free(struct llentry *ln, int gc)
1187 struct nd_defrouter *dr;
1190 LLE_WLOCK_ASSERT(ln);
1193 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1194 * even though it is not harmful, it was not really necessary.
1198 nd6_llinfo_settimer_locked(ln, -1);
1200 ifp = ln->lle_tbl->llt_ifp;
1202 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1203 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1205 if (dr != NULL && dr->expire &&
1206 ln->ln_state == ND6_LLINFO_STALE && gc) {
1208 * If the reason for the deletion is just garbage
1209 * collection, and the neighbor is an active default
1210 * router, do not delete it. Instead, reset the GC
1211 * timer using the router's lifetime.
1212 * Simply deleting the entry would affect default
1213 * router selection, which is not necessarily a good
1214 * thing, especially when we're using router preference
1216 * XXX: the check for ln_state would be redundant,
1217 * but we intentionally keep it just in case.
1219 if (dr->expire > time_uptime)
1220 nd6_llinfo_settimer_locked(ln,
1221 (dr->expire - time_uptime) * hz);
1223 nd6_llinfo_settimer_locked(ln,
1224 (long)V_nd6_gctimer * hz);
1233 * Unreachablity of a router might affect the default
1234 * router selection and on-link detection of advertised
1239 * Temporarily fake the state to choose a new default
1240 * router and to perform on-link determination of
1241 * prefixes correctly.
1242 * Below the state will be set correctly,
1243 * or the entry itself will be deleted.
1245 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1248 if (ln->ln_router || dr) {
1251 * We need to unlock to avoid a LOR with rt6_flush() with the
1252 * rnh and for the calls to pfxlist_onlink_check() and
1253 * defrouter_select() in the block further down for calls
1254 * into nd6_lookup(). We still hold a ref.
1259 * rt6_flush must be called whether or not the neighbor
1260 * is in the Default Router List.
1261 * See a corresponding comment in nd6_na_input().
1263 rt6_flush(&ln->r_l3addr.addr6, ifp);
1268 * Since defrouter_select() does not affect the
1269 * on-link determination and MIP6 needs the check
1270 * before the default router selection, we perform
1273 pfxlist_onlink_check();
1276 * Refresh default router list.
1282 * If this entry was added by an on-link redirect, remove the
1283 * corresponding host route.
1285 if (ln->la_flags & LLE_REDIRECT)
1286 nd6_free_redirect(ln);
1288 if (ln->ln_router || dr)
1293 * Save to unlock. We still hold an extra reference and will not
1294 * free(9) in llentry_free() if someone else holds one as well.
1297 IF_AFDATA_LOCK(ifp);
1299 /* Guard against race with other llentry_free(). */
1300 if (ln->la_flags & LLE_LINKED) {
1301 /* Remove callout reference */
1303 lltable_unlink_entry(ln->lle_tbl, ln);
1305 IF_AFDATA_UNLOCK(ifp);
1311 * Remove the rtentry for the given llentry,
1312 * both of which were installed by a redirect.
1315 nd6_free_redirect(const struct llentry *ln)
1319 struct radix_node_head *rnh;
1320 struct sockaddr_in6 sin6;
1322 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1323 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) {
1324 rnh = rt_tables_get_rnh(fibnum, AF_INET6);
1328 RADIX_NODE_HEAD_LOCK(rnh);
1329 rt = in6_rtalloc1((struct sockaddr *)&sin6, 0,
1330 RTF_RNH_LOCKED, fibnum);
1332 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1333 rt_expunge(rnh, rt);
1336 RADIX_NODE_HEAD_UNLOCK(rnh);
1341 * Rejuvenate this function for routing operations related
1345 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1347 struct sockaddr_in6 *gateway;
1348 struct nd_defrouter *dr;
1351 gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1362 * Only indirect routes are interesting.
1364 if ((rt->rt_flags & RTF_GATEWAY) == 0)
1367 * check for default route
1369 if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1370 &SIN6(rt_key(rt))->sin6_addr)) {
1372 dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1382 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1384 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1385 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1386 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1389 if (ifp->if_afdata[AF_INET6] == NULL)
1390 return (EPFNOSUPPORT);
1392 case OSIOCGIFINFO_IN6:
1394 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1395 bzero(&ND, sizeof(ND));
1396 ND.linkmtu = IN6_LINKMTU(ifp);
1397 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1398 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1399 ND.reachable = ND_IFINFO(ifp)->reachable;
1400 ND.retrans = ND_IFINFO(ifp)->retrans;
1401 ND.flags = ND_IFINFO(ifp)->flags;
1402 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1403 ND.chlim = ND_IFINFO(ifp)->chlim;
1405 case SIOCGIFINFO_IN6:
1406 ND = *ND_IFINFO(ifp);
1408 case SIOCSIFINFO_IN6:
1410 * used to change host variables from userland.
1411 * intented for a use on router to reflect RA configurations.
1413 /* 0 means 'unspecified' */
1414 if (ND.linkmtu != 0) {
1415 if (ND.linkmtu < IPV6_MMTU ||
1416 ND.linkmtu > IN6_LINKMTU(ifp)) {
1420 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1423 if (ND.basereachable != 0) {
1424 int obasereachable = ND_IFINFO(ifp)->basereachable;
1426 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1427 if (ND.basereachable != obasereachable)
1428 ND_IFINFO(ifp)->reachable =
1429 ND_COMPUTE_RTIME(ND.basereachable);
1431 if (ND.retrans != 0)
1432 ND_IFINFO(ifp)->retrans = ND.retrans;
1434 ND_IFINFO(ifp)->chlim = ND.chlim;
1436 case SIOCSIFINFO_FLAGS:
1439 struct in6_ifaddr *ia;
1441 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1442 !(ND.flags & ND6_IFF_IFDISABLED)) {
1443 /* ifdisabled 1->0 transision */
1446 * If the interface is marked as ND6_IFF_IFDISABLED and
1447 * has an link-local address with IN6_IFF_DUPLICATED,
1448 * do not clear ND6_IFF_IFDISABLED.
1449 * See RFC 4862, Section 5.4.5.
1452 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1453 if (ifa->ifa_addr->sa_family != AF_INET6)
1455 ia = (struct in6_ifaddr *)ifa;
1456 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1457 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1460 IF_ADDR_RUNLOCK(ifp);
1463 /* LLA is duplicated. */
1464 ND.flags |= ND6_IFF_IFDISABLED;
1465 log(LOG_ERR, "Cannot enable an interface"
1466 " with a link-local address marked"
1469 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1470 if (ifp->if_flags & IFF_UP)
1473 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1474 (ND.flags & ND6_IFF_IFDISABLED)) {
1475 /* ifdisabled 0->1 transision */
1476 /* Mark all IPv6 address as tentative. */
1478 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1479 if (V_ip6_dad_count > 0 &&
1480 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1482 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1484 if (ifa->ifa_addr->sa_family !=
1487 ia = (struct in6_ifaddr *)ifa;
1488 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1490 IF_ADDR_RUNLOCK(ifp);
1494 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1495 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1496 /* auto_linklocal 0->1 transision */
1498 /* If no link-local address on ifp, configure */
1499 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1500 in6_ifattach(ifp, NULL);
1501 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1502 ifp->if_flags & IFF_UP) {
1504 * When the IF already has
1505 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1506 * address is assigned, and IFF_UP, try to
1510 TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1512 if (ifa->ifa_addr->sa_family !=
1515 ia = (struct in6_ifaddr *)ifa;
1516 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1519 IF_ADDR_RUNLOCK(ifp);
1521 /* No LLA is configured. */
1522 in6_ifattach(ifp, NULL);
1526 ND_IFINFO(ifp)->flags = ND.flags;
1529 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1530 /* sync kernel routing table with the default router list */
1534 case SIOCSPFXFLUSH_IN6:
1536 /* flush all the prefix advertised by routers */
1537 struct nd_prefix *pr, *next;
1539 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1540 struct in6_ifaddr *ia, *ia_next;
1542 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1545 /* do we really have to remove addresses as well? */
1546 /* XXXRW: in6_ifaddrhead locking. */
1547 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1549 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1552 if (ia->ia6_ndpr == pr)
1553 in6_purgeaddr(&ia->ia_ifa);
1559 case SIOCSRTRFLUSH_IN6:
1561 /* flush all the default routers */
1562 struct nd_defrouter *dr, *next;
1565 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1571 case SIOCGNBRINFO_IN6:
1574 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1576 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1579 IF_AFDATA_RLOCK(ifp);
1580 ln = nd6_lookup(&nb_addr, 0, ifp);
1581 IF_AFDATA_RUNLOCK(ifp);
1587 nbi->state = ln->ln_state;
1588 nbi->asked = ln->la_asked;
1589 nbi->isrouter = ln->ln_router;
1590 if (ln->la_expire == 0)
1593 nbi->expire = ln->la_expire +
1594 (time_second - time_uptime);
1598 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1599 ndif->ifindex = V_nd6_defifindex;
1601 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1602 return (nd6_setdefaultiface(ndif->ifindex));
1608 * Calculates new isRouter value based on provided parameters and
1612 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1617 * ICMP6 type dependent behavior.
1619 * NS: clear IsRouter if new entry
1620 * RS: clear IsRouter
1621 * RA: set IsRouter if there's lladdr
1622 * redir: clear IsRouter if new entry
1625 * The spec says that we must set IsRouter in the following cases:
1626 * - If lladdr exist, set IsRouter. This means (1-5).
1627 * - If it is old entry (!newentry), set IsRouter. This means (7).
1628 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1629 * A quetion arises for (1) case. (1) case has no lladdr in the
1630 * neighbor cache, this is similar to (6).
1631 * This case is rare but we figured that we MUST NOT set IsRouter.
1633 * is_new old_addr new_addr NS RS RA redir
1640 * 1 -- n (6) c c c s
1641 * 1 -- y (7) c c s c s
1645 switch (type & 0xff) {
1646 case ND_NEIGHBOR_SOLICIT:
1648 * New entry must have is_router flag cleared.
1650 if (is_new) /* (6-7) */
1655 * If the icmp is a redirect to a better router, always set the
1656 * is_router flag. Otherwise, if the entry is newly created,
1657 * clear the flag. [RFC 2461, sec 8.3]
1659 if (code == ND_REDIRECT_ROUTER)
1662 if (is_new) /* (6-7) */
1666 case ND_ROUTER_SOLICIT:
1668 * is_router flag must always be cleared.
1672 case ND_ROUTER_ADVERT:
1674 * Mark an entry with lladdr as a router.
1676 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */
1677 (is_new && new_addr)) { /* (7) */
1687 * Create neighbor cache entry and cache link-layer address,
1688 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1691 * code - type dependent information
1695 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1696 int lladdrlen, int type, int code)
1698 struct llentry *ln = NULL, *ln_tmp;
1704 uint16_t router = 0;
1705 struct sockaddr_in6 sin6;
1706 struct mbuf *chain = NULL;
1708 IF_AFDATA_UNLOCK_ASSERT(ifp);
1710 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1711 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1713 /* nothing must be updated for unspecified address */
1714 if (IN6_IS_ADDR_UNSPECIFIED(from))
1718 * Validation about ifp->if_addrlen and lladdrlen must be done in
1721 * XXX If the link does not have link-layer adderss, what should
1722 * we do? (ifp->if_addrlen == 0)
1723 * Spec says nothing in sections for RA, RS and NA. There's small
1724 * description on it in NS section (RFC 2461 7.2.3).
1726 flags = lladdr ? LLE_EXCLUSIVE : 0;
1727 IF_AFDATA_RLOCK(ifp);
1728 ln = nd6_lookup(from, flags, ifp);
1729 IF_AFDATA_RUNLOCK(ifp);
1732 flags |= LLE_EXCLUSIVE;
1733 ln = nd6_alloc(from, 0, ifp);
1738 * Since we already know all the data for the new entry,
1739 * fill it before insertion.
1741 if (lladdr != NULL) {
1742 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1743 ln->la_flags |= LLE_VALID;
1745 IF_AFDATA_WLOCK(ifp);
1747 /* Prefer any existing lle over newly-created one */
1748 ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1750 lltable_link_entry(LLTABLE6(ifp), ln);
1751 IF_AFDATA_WUNLOCK(ifp);
1752 if (ln_tmp == NULL) {
1753 /* No existing lle, mark as new entry */
1755 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1757 lltable_free_entry(LLTABLE6(ifp), ln);
1762 /* do nothing if static ndp is set */
1763 if ((ln->la_flags & LLE_STATIC)) {
1764 if (flags & LLE_EXCLUSIVE)
1771 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1772 if (olladdr && lladdr) {
1773 llchange = bcmp(lladdr, &ln->ll_addr,
1775 } else if (!olladdr && lladdr)
1781 * newentry olladdr lladdr llchange (*=record)
1784 * 0 n y y (3) * STALE
1786 * 0 y y y (5) * STALE
1787 * 1 -- n -- (6) NOSTATE(= PASSIVE)
1788 * 1 -- y -- (7) * STALE
1792 if (!is_newentry && llchange != 0)
1793 do_update = 1; /* (3,5) */
1795 if (lladdr) { /* (3-5) and (7) */
1797 * Record source link-layer address
1798 * XXX is it dependent to ifp->if_type?
1800 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1801 ln->la_flags |= LLE_VALID;
1802 if (do_update != 0) /* 3,5,7 */
1803 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1805 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1808 if (ln->la_hold != NULL)
1809 nd6_grab_holdchain(ln, &chain, &sin6);
1813 /* Calculates new router status */
1814 router = nd6_is_router(type, code, is_newentry, olladdr,
1815 lladdr != NULL ? 1 : 0, ln->ln_router);
1817 ln->ln_router = router;
1818 /* Mark non-router redirects with special flag */
1819 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
1820 ln->la_flags |= LLE_REDIRECT;
1822 if (flags & LLE_EXCLUSIVE)
1828 nd6_flush_holdchain(ifp, ifp, chain, &sin6);
1831 * When the link-layer address of a router changes, select the
1832 * best router again. In particular, when the neighbor entry is newly
1833 * created, it might affect the selection policy.
1834 * Question: can we restrict the first condition to the "is_newentry"
1836 * XXX: when we hear an RA from a new router with the link-layer
1837 * address option, defrouter_select() is called twice, since
1838 * defrtrlist_update called the function as well. However, I believe
1839 * we can compromise the overhead, since it only happens the first
1841 * XXX: although defrouter_select() should not have a bad effect
1842 * for those are not autoconfigured hosts, we explicitly avoid such
1845 if ((do_update || is_newentry) && router &&
1846 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1848 * guaranteed recursion
1855 nd6_slowtimo(void *arg)
1857 CURVNET_SET((struct vnet *) arg);
1858 struct nd_ifinfo *nd6if;
1861 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1862 nd6_slowtimo, curvnet);
1863 IFNET_RLOCK_NOSLEEP();
1864 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1865 if (ifp->if_afdata[AF_INET6] == NULL)
1867 nd6if = ND_IFINFO(ifp);
1868 if (nd6if->basereachable && /* already initialized */
1869 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1871 * Since reachable time rarely changes by router
1872 * advertisements, we SHOULD insure that a new random
1873 * value gets recomputed at least once every few hours.
1876 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1877 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1880 IFNET_RUNLOCK_NOSLEEP();
1885 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
1886 struct sockaddr_in6 *sin6)
1889 LLE_WLOCK_ASSERT(ln);
1891 *chain = ln->la_hold;
1893 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
1895 if (ln->ln_state == ND6_LLINFO_STALE) {
1898 * The first time we send a packet to a
1899 * neighbor whose entry is STALE, we have
1900 * to change the state to DELAY and a sets
1901 * a timer to expire in DELAY_FIRST_PROBE_TIME
1902 * seconds to ensure do neighbor unreachability
1903 * detection on expiration.
1906 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
1911 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1912 struct sockaddr_in6 *dst)
1916 struct ip6_hdr *ip6;
1920 mac_netinet6_nd6_send(ifp, m);
1924 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1925 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1926 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1927 * to be diverted to user space. When re-injected into the kernel,
1928 * send_output() will directly dispatch them to the outgoing interface.
1930 if (send_sendso_input_hook != NULL) {
1931 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1933 ip6 = mtod(m, struct ip6_hdr *);
1934 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1935 /* Use the SEND socket */
1936 error = send_sendso_input_hook(m, ifp, SND_OUT,
1938 /* -1 == no app on SEND socket */
1939 if (error == 0 || error != -1)
1944 m_clrprotoflags(m); /* Avoid confusing lower layers. */
1945 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
1946 mtod(m, struct ip6_hdr *));
1948 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
1951 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL);
1956 * Do L2 address resolution for @sa_dst address. Stores found
1957 * address in @desten buffer. Copy of lle ln_flags can be also
1958 * saved in @pflags if @pflags is non-NULL.
1960 * If destination LLE does not exists or lle state modification
1961 * is required, call "slow" version.
1964 * - 0 on success (address copied to buffer).
1965 * - EWOULDBLOCK (no local error, but address is still unresolved)
1966 * - other errors (alloc failure, etc)
1969 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
1970 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
1972 struct llentry *ln = NULL;
1973 const struct sockaddr_in6 *dst6;
1978 dst6 = (const struct sockaddr_in6 *)sa_dst;
1980 /* discard the packet if IPv6 operation is disabled on the interface */
1981 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1983 return (ENETDOWN); /* better error? */
1986 if (m != NULL && m->m_flags & M_MCAST) {
1987 switch (ifp->if_type) {
1994 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
1999 return (EAFNOSUPPORT);
2003 IF_AFDATA_RLOCK(ifp);
2004 ln = nd6_lookup(&dst6->sin6_addr, 0, ifp);
2005 IF_AFDATA_RUNLOCK(ifp);
2008 * Perform fast path for the following cases:
2009 * 1) lle state is REACHABLE
2010 * 2) lle state is DELAY (NS message sent)
2012 * Every other case involves lle modification, so we handle
2015 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE &&
2016 ln->ln_state != ND6_LLINFO_DELAY)) {
2017 /* Fall back to slow processing path */
2020 return (nd6_resolve_slow(ifp, m, dst6, desten, pflags));
2024 bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2026 *pflags = ln->la_flags;
2033 * Do L2 address resolution for @sa_dst address. Stores found
2034 * address in @desten buffer. Copy of lle ln_flags can be also
2035 * saved in @pflags if @pflags is non-NULL.
2038 * Function assume that destination LLE does not exist,
2039 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2041 * Set noinline to be dtrace-friendly
2043 static __noinline int
2044 nd6_resolve_slow(struct ifnet *ifp, struct mbuf *m,
2045 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2047 struct llentry *lle = NULL, *lle_tmp;
2048 struct in6_addr *psrc, src;
2052 * Address resolution or Neighbor Unreachability Detection
2054 * At this point, the destination of the packet must be a unicast
2055 * or an anycast address(i.e. not a multicast).
2058 IF_AFDATA_RLOCK(ifp);
2059 lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2060 IF_AFDATA_RUNLOCK(ifp);
2061 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2063 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2064 * the condition below is not very efficient. But we believe
2065 * it is tolerable, because this should be a rare case.
2067 lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2069 char ip6buf[INET6_ADDRSTRLEN];
2071 "nd6_output: can't allocate llinfo for %s "
2073 ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2078 IF_AFDATA_WLOCK(ifp);
2080 /* Prefer any existing entry over newly-created one */
2081 lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2082 if (lle_tmp == NULL)
2083 lltable_link_entry(LLTABLE6(ifp), lle);
2084 IF_AFDATA_WUNLOCK(ifp);
2085 if (lle_tmp != NULL) {
2086 lltable_free_entry(LLTABLE6(ifp), lle);
2093 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2103 LLE_WLOCK_ASSERT(lle);
2106 * The first time we send a packet to a neighbor whose entry is
2107 * STALE, we have to change the state to DELAY and a sets a timer to
2108 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2109 * neighbor unreachability detection on expiration.
2112 if (lle->ln_state == ND6_LLINFO_STALE)
2113 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2116 * If the neighbor cache entry has a state other than INCOMPLETE
2117 * (i.e. its link-layer address is already resolved), just
2120 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2121 bcopy(&lle->ll_addr, desten, ifp->if_addrlen);
2123 *pflags = lle->la_flags;
2129 * There is a neighbor cache entry, but no ethernet address
2130 * response yet. Append this latest packet to the end of the
2131 * packet queue in the mbuf, unless the number of the packet
2132 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen,
2133 * the oldest packet in the queue will be removed.
2136 if (lle->la_hold != NULL) {
2137 struct mbuf *m_hold;
2141 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2143 if (m_hold->m_nextpkt == NULL) {
2144 m_hold->m_nextpkt = m;
2148 while (i >= V_nd6_maxqueuelen) {
2149 m_hold = lle->la_hold;
2150 lle->la_hold = lle->la_hold->m_nextpkt;
2159 * If there has been no NS for the neighbor after entering the
2160 * INCOMPLETE state, send the first solicitation.
2161 * Note that for newly-created lle la_asked will be 0,
2162 * so we will transition from ND6_LLINFO_NOSTATE to
2163 * ND6_LLINFO_INCOMPLETE state here.
2167 if (lle->la_asked == 0) {
2170 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2172 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2176 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2178 return (EWOULDBLOCK);
2183 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2184 struct sockaddr_in6 *dst)
2186 struct mbuf *m, *m_head;
2187 struct ifnet *outifp;
2191 if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2198 m_head = m_head->m_nextpkt;
2199 error = nd6_output_ifp(ifp, origifp, m, dst);
2204 * note that intermediate errors are blindly ignored
2210 nd6_need_cache(struct ifnet *ifp)
2213 * XXX: we currently do not make neighbor cache on any interface
2214 * other than ARCnet, Ethernet, FDDI and GIF.
2217 * - unidirectional tunnels needs no ND
2219 switch (ifp->if_type) {
2226 case IFT_INFINIBAND:
2228 case IFT_PROPVIRTUAL:
2236 * Add pernament ND6 link-layer record for given
2237 * interface address.
2239 * Very similar to IPv4 arp_ifinit(), but:
2240 * 1) IPv6 DAD is performed in different place
2241 * 2) It is called by IPv6 protocol stack in contrast to
2242 * arp_ifinit() which is typically called in SIOCSIFADDR
2243 * driver ioctl handler.
2247 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2250 struct llentry *ln, *ln_tmp;
2251 struct sockaddr *dst;
2253 ifp = ia->ia_ifa.ifa_ifp;
2254 if (nd6_need_cache(ifp) == 0)
2257 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2258 dst = (struct sockaddr *)&ia->ia_addr;
2259 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2263 IF_AFDATA_WLOCK(ifp);
2265 /* Unlink any entry if exists */
2266 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2268 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2269 lltable_link_entry(LLTABLE6(ifp), ln);
2270 IF_AFDATA_WUNLOCK(ifp);
2273 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2274 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2278 llentry_free(ln_tmp);
2284 * Removes either all lle entries for given @ia, or lle
2285 * corresponding to @ia address.
2288 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2290 struct sockaddr_in6 mask, addr;
2291 struct sockaddr *saddr, *smask;
2294 ifp = ia->ia_ifa.ifa_ifp;
2295 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2296 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2297 saddr = (struct sockaddr *)&addr;
2298 smask = (struct sockaddr *)&mask;
2301 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2303 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2307 clear_llinfo_pqueue(struct llentry *ln)
2309 struct mbuf *m_hold, *m_hold_next;
2311 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2312 m_hold_next = m_hold->m_nextpkt;
2320 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2321 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2323 SYSCTL_DECL(_net_inet6_icmp6);
2325 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2326 CTLFLAG_RD, nd6_sysctl_drlist, "");
2327 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2328 CTLFLAG_RD, nd6_sysctl_prlist, "");
2329 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2330 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2331 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2332 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2335 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2337 struct in6_defrouter d;
2338 struct nd_defrouter *dr;
2344 bzero(&d, sizeof(d));
2345 d.rtaddr.sin6_family = AF_INET6;
2346 d.rtaddr.sin6_len = sizeof(d.rtaddr);
2351 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2352 d.rtaddr.sin6_addr = dr->rtaddr;
2353 error = sa6_recoverscope(&d.rtaddr);
2356 d.flags = dr->flags;
2357 d.rtlifetime = dr->rtlifetime;
2358 d.expire = dr->expire + (time_second - time_uptime);
2359 d.if_index = dr->ifp->if_index;
2360 error = SYSCTL_OUT(req, &d, sizeof(d));
2368 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2370 struct in6_prefix p;
2371 struct sockaddr_in6 s6;
2372 struct nd_prefix *pr;
2373 struct nd_pfxrouter *pfr;
2376 char ip6buf[INET6_ADDRSTRLEN];
2381 bzero(&p, sizeof(p));
2382 p.origin = PR_ORIG_RA;
2383 bzero(&s6, sizeof(s6));
2384 s6.sin6_family = AF_INET6;
2385 s6.sin6_len = sizeof(s6);
2390 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2391 p.prefix = pr->ndpr_prefix;
2392 if (sa6_recoverscope(&p.prefix)) {
2393 log(LOG_ERR, "scope error in prefix list (%s)\n",
2394 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2395 /* XXX: press on... */
2397 p.raflags = pr->ndpr_raf;
2398 p.prefixlen = pr->ndpr_plen;
2399 p.vltime = pr->ndpr_vltime;
2400 p.pltime = pr->ndpr_pltime;
2401 p.if_index = pr->ndpr_ifp->if_index;
2402 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2405 /* XXX: we assume time_t is signed. */
2407 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2408 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2409 p.expire = pr->ndpr_lastupdate +
2411 (time_second - time_uptime);
2413 p.expire = maxexpire;
2415 p.refcnt = pr->ndpr_refcnt;
2416 p.flags = pr->ndpr_stateflags;
2418 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2420 error = SYSCTL_OUT(req, &p, sizeof(p));
2423 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2424 s6.sin6_addr = pfr->router->rtaddr;
2425 if (sa6_recoverscope(&s6))
2427 "scope error in prefix list (%s)\n",
2428 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2429 error = SYSCTL_OUT(req, &s6, sizeof(s6));