Local changes for the flex upgrade.
[dragonfly.git] / sys / netinet6 / nd6.c
1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
3
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/callout.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/syslog.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/mutex.h>
51
52 #include <sys/thread2.h>
53 #include <sys/mutex2.h>
54
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/route.h>
59 #include <net/netisr2.h>
60 #include <net/netmsg2.h>
61
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69
70 #include <net/net_osdep.h>
71
72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
74
75 #define SIN6(s) ((struct sockaddr_in6 *)s)
76 #define SDL(s) ((struct sockaddr_dl *)s)
77
78 /*
79  * Note that the check for rt_llinfo is necessary because a cloned
80  * route from a parent route that has the L flag (e.g. the default
81  * route to a p2p interface) may have the flag, too, while the
82  * destination is not actually a neighbor.
83  * XXX: we can't use rt->rt_ifp to check for the interface, since
84  *      it might be the loopback interface if the entry is for our
85  *      own address on a non-loopback interface. Instead, we should
86  *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
87  *      interface.
88  */
89 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)                        \
90         !(((rt)->rt_flags & RTF_GATEWAY) ||                     \
91           ((rt)->rt_flags & RTF_LLINFO) == 0 ||                 \
92           (rt)->rt_gateway->sa_family != AF_LINK ||             \
93           (rt)->rt_llinfo == NULL ||                            \
94           ((ifp) != NULL && (rt)->rt_ifa->ifa_ifp != (ifp)))
95
96 #define ND6_RTENTRY_IS_LLCLONING(rt)                            \
97         (((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==     \
98          (RTF_PRCLONING | RTF_LLINFO) ||                        \
99          ((rt)->rt_flags & RTF_CLONING))
100
101 /* timer values */
102 int     nd6_prune       = 1;    /* walk list every 1 seconds */
103 int     nd6_delay       = 5;    /* delay first probe time 5 second */
104 int     nd6_umaxtries   = 3;    /* maximum unicast query */
105 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
106 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
107 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
108
109 /* preventing too many loops in ND option parsing */
110 int nd6_maxndopt = 10;  /* max # of ND options allowed */
111
112 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
113
114 #ifdef ND6_DEBUG
115 int nd6_debug = 1;
116 #else
117 int nd6_debug = 0;
118 #endif
119
120 /* for debugging? */
121 static int nd6_inuse, nd6_allocated;
122
123 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
124 struct nd_drhead nd_defrouter;
125 struct nd_prhead nd_prefix = { 0 };
126 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
127
128 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
129 static struct sockaddr_in6 all1_sa;
130
131 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
132 static int regen_tmpaddr (struct in6_ifaddr *);
133 static void nd6_slowtimo(void *);
134 static void nd6_slowtimo_dispatch(netmsg_t);
135 static void nd6_timer(void *);
136 static void nd6_timer_dispatch(netmsg_t);
137
138 static struct callout nd6_slowtimo_ch;
139 static struct netmsg_base nd6_slowtimo_netmsg;
140
141 static struct callout nd6_timer_ch;
142 static struct netmsg_base nd6_timer_netmsg;
143
144 void
145 nd6_init(void)
146 {
147         static int nd6_init_done = 0;
148         int i;
149
150         if (nd6_init_done) {
151                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
152                 return;
153         }
154
155         all1_sa.sin6_family = AF_INET6;
156         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
157         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
158                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
159
160         /* initialization of the default router list */
161         TAILQ_INIT(&nd_defrouter);
162
163         nd6_init_done = 1;
164
165         /* start timer */
166         callout_init_mp(&nd6_slowtimo_ch);
167         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
168             MSGF_PRIORITY, nd6_slowtimo_dispatch);
169         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
170             nd6_slowtimo, NULL, 0);
171 }
172
173 struct nd_ifinfo *
174 nd6_ifattach(struct ifnet *ifp)
175 {
176         struct nd_ifinfo *nd;
177
178         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
179             M_WAITOK | M_ZERO);
180
181         nd->initialized = 1;
182
183         nd->chlim = IPV6_DEFHLIM;
184         nd->basereachable = REACHABLE_TIME;
185         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
186         nd->retrans = RETRANS_TIMER;
187
188         /*
189          * Note that the default value of ip6_accept_rtadv is 0, which means
190          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
191          * here.
192          */
193         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
194
195         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
196         nd6_setmtu0(ifp, nd);
197         return nd;
198 }
199
200 void
201 nd6_ifdetach(struct nd_ifinfo *nd)
202 {
203         kfree(nd, M_IP6NDP);
204 }
205
206 /*
207  * Reset ND level link MTU. This function is called when the physical MTU
208  * changes, which means we might have to adjust the ND level MTU.
209  */
210 void
211 nd6_setmtu(struct ifnet *ifp)
212 {
213         nd6_setmtu0(ifp, ND_IFINFO(ifp));
214 }
215
216 struct netmsg_nd6setmtu {
217         struct netmsg_base      nmsg;
218         struct ifnet            *ifp;
219         struct nd_ifinfo        *ndi;
220 };
221
222 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
223 static void
224 nd6_setmtu0_dispatch(netmsg_t msg)
225 {
226         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
227         struct ifnet *ifp = nmsg->ifp;
228         struct nd_ifinfo *ndi = nmsg->ndi;
229         uint32_t omaxmtu;
230
231         omaxmtu = ndi->maxmtu;
232
233         switch (ifp->if_type) {
234         case IFT_ETHER:
235                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
236                 break;
237         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
238                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
239                 break;
240 #ifdef IFT_IEEE80211
241         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
242                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
243                 break;
244 #endif
245         default:
246                 ndi->maxmtu = ifp->if_mtu;
247                 break;
248         }
249
250         /*
251          * Decreasing the interface MTU under IPV6 minimum MTU may cause
252          * undesirable situation.  We thus notify the operator of the change
253          * explicitly.  The check for omaxmtu is necessary to restrict the
254          * log to the case of changing the MTU, not initializing it.
255          */
256         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
257                 log(LOG_NOTICE, "nd6_setmtu0: "
258                     "new link MTU on %s (%lu) is too small for IPv6\n",
259                     if_name(ifp), (unsigned long)ndi->maxmtu);
260         }
261
262         if (ndi->maxmtu > in6_maxmtu)
263                 in6_setmaxmtu(); /* check all interfaces just in case */
264
265         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
266 }
267
268 void
269 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
270 {
271         struct netmsg_nd6setmtu nmsg;
272
273         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
274             nd6_setmtu0_dispatch);
275         nmsg.ifp = ifp;
276         nmsg.ndi = ndi;
277         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
278 }
279
280 void
281 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
282 {
283         bzero(ndopts, sizeof(*ndopts));
284         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
285         ndopts->nd_opts_last
286                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
287
288         if (icmp6len == 0) {
289                 ndopts->nd_opts_done = 1;
290                 ndopts->nd_opts_search = NULL;
291         }
292 }
293
294 /*
295  * Take one ND option.
296  */
297 struct nd_opt_hdr *
298 nd6_option(union nd_opts *ndopts)
299 {
300         struct nd_opt_hdr *nd_opt;
301         int olen;
302
303         if (!ndopts)
304                 panic("ndopts == NULL in nd6_option");
305         if (!ndopts->nd_opts_last)
306                 panic("uninitialized ndopts in nd6_option");
307         if (!ndopts->nd_opts_search)
308                 return NULL;
309         if (ndopts->nd_opts_done)
310                 return NULL;
311
312         nd_opt = ndopts->nd_opts_search;
313
314         /* make sure nd_opt_len is inside the buffer */
315         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
316                 bzero(ndopts, sizeof(*ndopts));
317                 return NULL;
318         }
319
320         olen = nd_opt->nd_opt_len << 3;
321         if (olen == 0) {
322                 /*
323                  * Message validation requires that all included
324                  * options have a length that is greater than zero.
325                  */
326                 bzero(ndopts, sizeof(*ndopts));
327                 return NULL;
328         }
329
330         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
331         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
332                 /* option overruns the end of buffer, invalid */
333                 bzero(ndopts, sizeof(*ndopts));
334                 return NULL;
335         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
336                 /* reached the end of options chain */
337                 ndopts->nd_opts_done = 1;
338                 ndopts->nd_opts_search = NULL;
339         }
340         return nd_opt;
341 }
342
343 /*
344  * Parse multiple ND options.
345  * This function is much easier to use, for ND routines that do not need
346  * multiple options of the same type.
347  */
348 int
349 nd6_options(union nd_opts *ndopts)
350 {
351         struct nd_opt_hdr *nd_opt;
352         int i = 0;
353
354         if (!ndopts)
355                 panic("ndopts == NULL in nd6_options");
356         if (!ndopts->nd_opts_last)
357                 panic("uninitialized ndopts in nd6_options");
358         if (!ndopts->nd_opts_search)
359                 return 0;
360
361         while (1) {
362                 nd_opt = nd6_option(ndopts);
363                 if (!nd_opt && !ndopts->nd_opts_last) {
364                         /*
365                          * Message validation requires that all included
366                          * options have a length that is greater than zero.
367                          */
368                         icmp6stat.icp6s_nd_badopt++;
369                         bzero(ndopts, sizeof(*ndopts));
370                         return -1;
371                 }
372
373                 if (!nd_opt)
374                         goto skip1;
375
376                 switch (nd_opt->nd_opt_type) {
377                 case ND_OPT_SOURCE_LINKADDR:
378                 case ND_OPT_TARGET_LINKADDR:
379                 case ND_OPT_MTU:
380                 case ND_OPT_REDIRECTED_HEADER:
381                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
382                                 nd6log((LOG_INFO,
383                                     "duplicated ND6 option found (type=%d)\n",
384                                     nd_opt->nd_opt_type));
385                                 /* XXX bark? */
386                         } else {
387                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
388                                         = nd_opt;
389                         }
390                         break;
391                 case ND_OPT_PREFIX_INFORMATION:
392                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
393                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
394                                         = nd_opt;
395                         }
396                         ndopts->nd_opts_pi_end =
397                                 (struct nd_opt_prefix_info *)nd_opt;
398                         break;
399                 default:
400                         /*
401                          * Unknown options must be silently ignored,
402                          * to accomodate future extension to the protocol.
403                          */
404                         nd6log((LOG_DEBUG,
405                             "nd6_options: unsupported option %d - "
406                             "option ignored\n", nd_opt->nd_opt_type));
407                 }
408
409 skip1:
410                 i++;
411                 if (i > nd6_maxndopt) {
412                         icmp6stat.icp6s_nd_toomanyopt++;
413                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
414                         break;
415                 }
416
417                 if (ndopts->nd_opts_done)
418                         break;
419         }
420
421         return 0;
422 }
423
424 /*
425  * ND6 timer routine to expire default route list and prefix list
426  */
427 static void
428 nd6_timer_dispatch(netmsg_t nmsg)
429 {
430         struct llinfo_nd6 *ln;
431         struct nd_defrouter *dr;
432         struct nd_prefix *pr;
433         struct ifnet *ifp;
434         struct in6_ifaddr *ia6, *nia6;
435
436         ASSERT_NETISR0;
437
438         crit_enter();
439         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
440         crit_exit();
441
442         mtx_lock(&nd6_mtx);
443
444         ln = llinfo_nd6.ln_next;
445         while (ln && ln != &llinfo_nd6) {
446                 struct rtentry *rt;
447                 struct sockaddr_in6 *dst;
448                 struct llinfo_nd6 *next = ln->ln_next;
449                 /* XXX: used for the DELAY case only: */
450                 struct nd_ifinfo *ndi = NULL;
451
452                 if ((rt = ln->ln_rt) == NULL) {
453                         ln = next;
454                         continue;
455                 }
456                 if ((ifp = rt->rt_ifp) == NULL) {
457                         ln = next;
458                         continue;
459                 }
460                 ndi = ND_IFINFO(ifp);
461                 dst = (struct sockaddr_in6 *)rt_key(rt);
462
463                 if (ln->ln_expire > time_uptime) {
464                         ln = next;
465                         continue;
466                 }
467
468                 /* sanity check */
469                 if (!rt)
470                         panic("rt=0 in nd6_timer(ln=%p)", ln);
471                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
472                         panic("rt_llinfo(%p) is not equal to ln(%p)",
473                               rt->rt_llinfo, ln);
474                 if (!dst)
475                         panic("dst=0 in nd6_timer(ln=%p)", ln);
476
477                 switch (ln->ln_state) {
478                 case ND6_LLINFO_WAITDELETE:
479                         next = nd6_free(rt);
480                         break;
481                 case ND6_LLINFO_INCOMPLETE:
482                         if (ln->ln_asked++ >= nd6_mmaxtries) {
483                                 struct mbuf *m = ln->ln_hold;
484                                 if (m) {
485                                         if (rt->rt_ifp) {
486                                                 /*
487                                                  * Fake rcvif to make ICMP error
488                                                  * more helpful in diagnosing
489                                                  * for the receiver.
490                                                  * XXX: should we consider
491                                                  * older rcvif?
492                                                  */
493                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
494                                         }
495                                         /* XXX This will log a diagnostic
496                                          * that it's not an IPv6 header. */
497                                         icmp6_error(m, ICMP6_DST_UNREACH,
498                                                     ICMP6_DST_UNREACH_ADDR, 0);
499                                         ln->ln_hold = NULL;
500                                 }
501                                 ln->ln_state = ND6_LLINFO_WAITDELETE;
502                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
503                         }
504                         ln->ln_expire = time_uptime +
505                                 ND_IFINFO(ifp)->retrans / 1000;
506                         nd6_ns_output(ifp, NULL, &dst->sin6_addr,
507                                 ln, 0);
508                         break;
509                 case ND6_LLINFO_REACHABLE:
510                         if (ln->ln_expire) {
511                                 ln->ln_state = ND6_LLINFO_STALE;
512                                 ln->ln_expire = time_uptime + nd6_gctimer;
513                         }
514                         break;
515
516                 case ND6_LLINFO_STALE:
517                         /* Garbage Collection(RFC 2461 5.3) */
518                         if (ln->ln_expire)
519                                 next = nd6_free(rt);
520                         break;
521
522                 case ND6_LLINFO_DELAY:
523                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
524                                 /* We need NUD */
525                                 ln->ln_asked = 1;
526                                 ln->ln_state = ND6_LLINFO_PROBE;
527                                 ln->ln_expire = time_uptime +
528                                         ndi->retrans / 1000;
529                                 nd6_ns_output(ifp, &dst->sin6_addr,
530                                               &dst->sin6_addr,
531                                               ln, 0);
532                         } else {
533                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
534                                 ln->ln_expire = time_uptime + nd6_gctimer;
535                         }
536                         break;
537                 case ND6_LLINFO_PROBE:
538                         if (ln->ln_asked < nd6_umaxtries) {
539                                 ln->ln_asked++;
540                                 ln->ln_expire = time_uptime +
541                                         ND_IFINFO(ifp)->retrans / 1000;
542                                 nd6_ns_output(ifp, &dst->sin6_addr,
543                                                &dst->sin6_addr, ln, 0);
544                         } else {
545                                 next = nd6_free(rt);
546                         }
547                         break;
548                 }
549                 ln = next;
550         }
551
552         /* expire default router list */
553         dr = TAILQ_FIRST(&nd_defrouter);
554         while (dr) {
555                 if (dr->expire && dr->expire < time_uptime) {
556                         struct nd_defrouter *t;
557                         t = TAILQ_NEXT(dr, dr_entry);
558                         defrtrlist_del(dr);
559                         dr = t;
560                 } else {
561                         dr = TAILQ_NEXT(dr, dr_entry);
562                 }
563         }
564
565         /*
566          * expire interface addresses.
567          * in the past the loop was inside prefix expiry processing.
568          * However, from a stricter speci-confrmance standpoint, we should
569          * rather separate address lifetimes and prefix lifetimes.
570          */
571 addrloop:
572         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
573                 nia6 = ia6->ia_next;
574                 /* check address lifetime */
575                 if (IFA6_IS_INVALID(ia6)) {
576                         int regen = 0;
577
578                         /*
579                          * If the expiring address is temporary, try
580                          * regenerating a new one.  This would be useful when
581                          * we suspended a laptop PC, then turned it on after a
582                          * period that could invalidate all temporary
583                          * addresses.  Although we may have to restart the
584                          * loop (see below), it must be after purging the
585                          * address.  Otherwise, we'd see an infinite loop of
586                          * regeneration.
587                          */
588                         if (ip6_use_tempaddr &&
589                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
590                                 if (regen_tmpaddr(ia6) == 0)
591                                         regen = 1;
592                         }
593
594                         in6_purgeaddr(&ia6->ia_ifa);
595
596                         if (regen)
597                                 goto addrloop; /* XXX: see below */
598                 }
599                 if (IFA6_IS_DEPRECATED(ia6)) {
600                         int oldflags = ia6->ia6_flags;
601
602                         if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
603                                 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
604                                 in6_newaddrmsg((struct ifaddr *)ia6);
605                         }
606
607                         /*
608                          * If a temporary address has just become deprecated,
609                          * regenerate a new one if possible.
610                          */
611                         if (ip6_use_tempaddr &&
612                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
613                             !(oldflags & IN6_IFF_DEPRECATED)) {
614
615                                 if (regen_tmpaddr(ia6) == 0) {
616                                         /*
617                                          * A new temporary address is
618                                          * generated.
619                                          * XXX: this means the address chain
620                                          * has changed while we are still in
621                                          * the loop.  Although the change
622                                          * would not cause disaster (because
623                                          * it's not a deletion, but an
624                                          * addition,) we'd rather restart the
625                                          * loop just for safety.  Or does this
626                                          * significantly reduce performance??
627                                          */
628                                         goto addrloop;
629                                 }
630                         }
631                 } else {
632                         /*
633                          * A new RA might have made a deprecated address
634                          * preferred.
635                          */
636                         if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
637                                 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
638                                 in6_newaddrmsg((struct ifaddr *)ia6);
639                         }
640                 }
641         }
642
643         /* expire prefix list */
644         pr = nd_prefix.lh_first;
645         while (pr) {
646                 /*
647                  * check prefix lifetime.
648                  * since pltime is just for autoconf, pltime processing for
649                  * prefix is not necessary.
650                  */
651                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
652                         struct nd_prefix *t;
653                         t = pr->ndpr_next;
654
655                         /*
656                          * address expiration and prefix expiration are
657                          * separate.  NEVER perform in6_purgeaddr here.
658                          */
659
660                         prelist_remove(pr);
661                         pr = t;
662                 } else
663                         pr = pr->ndpr_next;
664         }
665
666         mtx_unlock(&nd6_mtx);
667
668         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
669 }
670
671 static void
672 nd6_timer(void *arg __unused)
673 {
674         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
675
676         KASSERT(mycpuid == 0, ("not on cpu0"));
677         crit_enter();
678         if (lmsg->ms_flags & MSGF_DONE)
679                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
680         crit_exit();
681 }
682
683 void
684 nd6_timer_init(void)
685 {
686         callout_init_mp(&nd6_timer_ch);
687         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
688             MSGF_PRIORITY, nd6_timer_dispatch);
689         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
690 }
691
692 static int
693 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
694                                          address */
695 {
696         struct ifaddr_container *ifac;
697         struct ifnet *ifp;
698         struct in6_ifaddr *public_ifa6 = NULL;
699
700         ifp = ia6->ia_ifa.ifa_ifp;
701         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
702                 struct ifaddr *ifa = ifac->ifa;
703                 struct in6_ifaddr *it6;
704
705                 if (ifa->ifa_addr->sa_family != AF_INET6)
706                         continue;
707
708                 it6 = (struct in6_ifaddr *)ifa;
709
710                 /* ignore no autoconf addresses. */
711                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
712                         continue;
713
714                 /* ignore autoconf addresses with different prefixes. */
715                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
716                         continue;
717
718                 /*
719                  * Now we are looking at an autoconf address with the same
720                  * prefix as ours.  If the address is temporary and is still
721                  * preferred, do not create another one.  It would be rare, but
722                  * could happen, for example, when we resume a laptop PC after
723                  * a long period.
724                  */
725                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
726                     !IFA6_IS_DEPRECATED(it6)) {
727                         public_ifa6 = NULL;
728                         break;
729                 }
730
731                 /*
732                  * This is a public autoconf address that has the same prefix
733                  * as ours.  If it is preferred, keep it.  We can't break the
734                  * loop here, because there may be a still-preferred temporary
735                  * address with the prefix.
736                  */
737                 if (!IFA6_IS_DEPRECATED(it6))
738                     public_ifa6 = it6;
739         }
740
741         if (public_ifa6 != NULL) {
742                 int e;
743
744                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
745                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
746                             " tmp addr,errno=%d\n", e);
747                         return (-1);
748                 }
749                 return (0);
750         }
751
752         return (-1);
753 }
754
755 /*
756  * Nuke neighbor cache/prefix/default router management table, right before
757  * ifp goes away.
758  */
759 void
760 nd6_purge(struct ifnet *ifp)
761 {
762         struct llinfo_nd6 *ln, *nln;
763         struct nd_defrouter *dr, *ndr, drany;
764         struct nd_prefix *pr, *npr;
765
766         /* Nuke default router list entries toward ifp */
767         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
768                 /*
769                  * The first entry of the list may be stored in
770                  * the routing table, so we'll delete it later.
771                  */
772                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
773                         ndr = TAILQ_NEXT(dr, dr_entry);
774                         if (dr->ifp == ifp)
775                                 defrtrlist_del(dr);
776                 }
777                 dr = TAILQ_FIRST(&nd_defrouter);
778                 if (dr->ifp == ifp)
779                         defrtrlist_del(dr);
780         }
781
782         /* Nuke prefix list entries toward ifp */
783         for (pr = nd_prefix.lh_first; pr; pr = npr) {
784                 npr = pr->ndpr_next;
785                 if (pr->ndpr_ifp == ifp) {
786                         /*
787                          * Previously, pr->ndpr_addr is removed as well,
788                          * but I strongly believe we don't have to do it.
789                          * nd6_purge() is only called from in6_ifdetach(),
790                          * which removes all the associated interface addresses
791                          * by itself.
792                          * (jinmei@kame.net 20010129)
793                          */
794                         prelist_remove(pr);
795                 }
796         }
797
798         /* cancel default outgoing interface setting */
799         if (nd6_defifindex == ifp->if_index)
800                 nd6_setdefaultiface(0);
801
802         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
803                 /* refresh default router list */
804                 bzero(&drany, sizeof(drany));
805                 defrouter_delreq(&drany, 0);
806                 defrouter_select();
807         }
808
809         /*
810          * Nuke neighbor cache entries for the ifp.
811          * Note that rt->rt_ifp may not be the same as ifp,
812          * due to KAME goto ours hack.  See RTM_RESOLVE case in
813          * nd6_rtrequest(), and ip6_input().
814          */
815         ln = llinfo_nd6.ln_next;
816         while (ln && ln != &llinfo_nd6) {
817                 struct rtentry *rt;
818                 struct sockaddr_dl *sdl;
819
820                 nln = ln->ln_next;
821                 rt = ln->ln_rt;
822                 if (rt && rt->rt_gateway &&
823                     rt->rt_gateway->sa_family == AF_LINK) {
824                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
825                         if (sdl->sdl_index == ifp->if_index)
826                                 nln = nd6_free(rt);
827                 }
828                 ln = nln;
829         }
830 }
831
832 struct rtentry *
833 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
834 {
835         struct rtentry *rt;
836         struct sockaddr_in6 sin6;
837
838         bzero(&sin6, sizeof(sin6));
839         sin6.sin6_len = sizeof(struct sockaddr_in6);
840         sin6.sin6_family = AF_INET6;
841         sin6.sin6_addr = *addr6;
842
843         if (create)
844                 rt = rtlookup((struct sockaddr *)&sin6);
845         else
846                 rt = rtpurelookup((struct sockaddr *)&sin6);
847         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
848                 /*
849                  * This is the case for the default route.
850                  * If we want to create a neighbor cache for the address, we
851                  * should free the route for the destination and allocate an
852                  * interface route.
853                  */
854                 if (create) {
855                         --rt->rt_refcnt;
856                         rt = NULL;
857                 }
858         }
859         if (!rt) {
860                 if (create && ifp) {
861                         int e;
862
863                         /*
864                          * If no route is available and create is set,
865                          * we allocate a host route for the destination
866                          * and treat it like an interface route.
867                          * This hack is necessary for a neighbor which can't
868                          * be covered by our own prefix.
869                          */
870                         struct ifaddr *ifa;
871
872                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
873                         if (ifa == NULL)
874                                 return (NULL);
875
876                         /*
877                          * Create a new route.  RTF_LLINFO is necessary
878                          * to create a Neighbor Cache entry for the
879                          * destination in nd6_rtrequest which will be
880                          * called in rtrequest via ifa->ifa_rtrequest.
881                          */
882                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
883                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
884                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
885                              ~RTF_CLONING, &rt)) != 0) {
886                                 log(LOG_ERR,
887                                     "nd6_lookup: failed to add route for a "
888                                     "neighbor(%s), errno=%d\n",
889                                     ip6_sprintf(addr6), e);
890                         }
891                         if (rt == NULL)
892                                 return (NULL);
893                         if (rt->rt_llinfo) {
894                                 struct llinfo_nd6 *ln =
895                                     (struct llinfo_nd6 *)rt->rt_llinfo;
896
897                                 ln->ln_state = ND6_LLINFO_NOSTATE;
898                         }
899                 } else
900                         return (NULL);
901         }
902         rt->rt_refcnt--;
903
904         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
905                 if (create) {
906                         log(LOG_DEBUG,
907                             "nd6_lookup: failed to lookup %s (if = %s)\n",
908                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
909                         /* xxx more logs... kazu */
910                 }
911                 return (NULL);
912         }
913         return (rt);
914 }
915
916 static struct rtentry *
917 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
918 {
919         struct rtentry *rt;
920         struct sockaddr_in6 sin6;
921
922         bzero(&sin6, sizeof(sin6));
923         sin6.sin6_len = sizeof(struct sockaddr_in6);
924         sin6.sin6_family = AF_INET6;
925         sin6.sin6_addr = *addr6;
926
927         rt = rtpurelookup((struct sockaddr *)&sin6);
928         if (rt == NULL)
929                 return (NULL);
930         rt->rt_refcnt--;
931
932         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
933                 if (nd6_onlink_ns_rfc4861 &&
934                     (ND6_RTENTRY_IS_LLCLONING(rt) ||    /* not cloned yet */
935                      (rt->rt_parent != NULL &&          /* cloning */
936                       ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
937                         /*
938                          * If cloning ever happened or is happening,
939                          * rtentry for addr6 would or will become a
940                          * neighbor cache.
941                          */
942                 } else {
943                         rt = NULL;
944                 }
945         }
946         return (rt);
947 }
948
949 /*
950  * Detect if a given IPv6 address identifies a neighbor on a given link.
951  * XXX: should take care of the destination of a p2p link?
952  */
953 int
954 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
955 {
956         struct ifaddr_container *ifac;
957         int i;
958
959 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
960 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
961
962         /*
963          * A link-local address is always a neighbor.
964          * XXX: we should use the sin6_scope_id field rather than the embedded
965          * interface index.
966          */
967         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
968             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
969                 return (1);
970
971         /*
972          * If the address matches one of our addresses,
973          * it should be a neighbor.
974          */
975         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
976                 struct ifaddr *ifa = ifac->ifa;
977
978                 if (ifa->ifa_addr->sa_family != AF_INET6)
979                         next: continue;
980
981                 for (i = 0; i < 4; i++) {
982                         if ((IFADDR6(ifa).s6_addr32[i] ^
983                              addr->sin6_addr.s6_addr32[i]) &
984                             IFMASK6(ifa).s6_addr32[i])
985                                 goto next;
986                 }
987                 return (1);
988         }
989
990         /*
991          * Even if the address matches none of our addresses, it might be
992          * in the neighbor cache.
993          */
994         if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
995                 return (1);
996
997         return (0);
998 #undef IFADDR6
999 #undef IFMASK6
1000 }
1001
1002 /*
1003  * Free an nd6 llinfo entry.
1004  */
1005 struct llinfo_nd6 *
1006 nd6_free(struct rtentry *rt)
1007 {
1008         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1009         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1010         struct nd_defrouter *dr;
1011
1012         /*
1013          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1014          * even though it is not harmful, it was not really necessary.
1015          */
1016
1017         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
1018                 mtx_lock(&nd6_mtx);
1019                 dr = defrouter_lookup(
1020                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1021                     rt->rt_ifp);
1022
1023                 if (ln->ln_router || dr) {
1024                         /*
1025                          * rt6_flush must be called whether or not the neighbor
1026                          * is in the Default Router List.
1027                          * See a corresponding comment in nd6_na_input().
1028                          */
1029                         rt6_flush(&in6, rt->rt_ifp);
1030                 }
1031
1032                 if (dr) {
1033                         /*
1034                          * Unreachablity of a router might affect the default
1035                          * router selection and on-link detection of advertised
1036                          * prefixes.
1037                          */
1038
1039                         /*
1040                          * Temporarily fake the state to choose a new default
1041                          * router and to perform on-link determination of
1042                          * prefixes correctly.
1043                          * Below the state will be set correctly,
1044                          * or the entry itself will be deleted.
1045                          */
1046                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1047
1048                         /*
1049                          * Since defrouter_select() does not affect the
1050                          * on-link determination and MIP6 needs the check
1051                          * before the default router selection, we perform
1052                          * the check now.
1053                          */
1054                         pfxlist_onlink_check();
1055
1056                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1057                                 /*
1058                                  * It is used as the current default router,
1059                                  * so we have to move it to the end of the
1060                                  * list and choose a new one.
1061                                  * XXX: it is not very efficient if this is
1062                                  *      the only router.
1063                                  */
1064                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1065                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1066
1067                                 defrouter_select();
1068                         }
1069                 }
1070                 mtx_unlock(&nd6_mtx);
1071         }
1072
1073         /*
1074          * Before deleting the entry, remember the next entry as the
1075          * return value.  We need this because pfxlist_onlink_check() above
1076          * might have freed other entries (particularly the old next entry) as
1077          * a side effect (XXX).
1078          */
1079         next = ln->ln_next;
1080
1081         /*
1082          * Detach the route from the routing tree and the list of neighbor
1083          * caches, and disable the route entry not to be used in already
1084          * cached routes.
1085          *
1086          * ND expiry happens under one big timer.
1087          * To avoid overflowing the route socket, don't report this.
1088          * Now that RTM_MISS is reported when an address is unresolvable
1089          * the benefit of reporting this deletion is questionable.
1090          */
1091         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1092
1093         return (next);
1094 }
1095
1096 /*
1097  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1098  *
1099  * XXX cost-effective metods?
1100  */
1101 void
1102 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1103 {
1104         struct llinfo_nd6 *ln;
1105
1106         /*
1107          * If the caller specified "rt", use that.  Otherwise, resolve the
1108          * routing table by supplied "dst6".
1109          */
1110         if (!rt) {
1111                 if (!dst6)
1112                         return;
1113                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1114                         return;
1115         }
1116
1117         if ((rt->rt_flags & RTF_GATEWAY) ||
1118             !(rt->rt_flags & RTF_LLINFO) ||
1119             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1120             rt->rt_gateway->sa_family != AF_LINK) {
1121                 /* This is not a host route. */
1122                 return;
1123         }
1124
1125         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1126         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1127                 return;
1128
1129         /*
1130          * if we get upper-layer reachability confirmation many times,
1131          * it is possible we have false information.
1132          */
1133         if (!force) {
1134                 ln->ln_byhint++;
1135                 if (ln->ln_byhint > nd6_maxnudhint)
1136                         return;
1137         }
1138
1139         ln->ln_state = ND6_LLINFO_REACHABLE;
1140         if (ln->ln_expire)
1141                 ln->ln_expire = time_uptime +
1142                         ND_IFINFO(rt->rt_ifp)->reachable;
1143 }
1144
1145 void
1146 nd6_rtrequest(int req, struct rtentry *rt)
1147 {
1148         struct sockaddr *gate = rt->rt_gateway;
1149         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1150         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1151         struct ifnet *ifp = rt->rt_ifp;
1152         struct ifaddr *ifa;
1153
1154         if ((rt->rt_flags & RTF_GATEWAY))
1155                 return;
1156
1157         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1158                 /*
1159                  * This is probably an interface direct route for a link
1160                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1161                  * We do not need special treatment below for such a route.
1162                  * Moreover, the RTF_LLINFO flag which would be set below
1163                  * would annoy the ndp(8) command.
1164                  */
1165                 return;
1166         }
1167
1168         if (req == RTM_RESOLVE &&
1169             (nd6_need_cache(ifp) == 0 || /* stf case */
1170              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1171                 /*
1172                  * FreeBSD and BSD/OS often make a cloned host route based
1173                  * on a less-specific route (e.g. the default route).
1174                  * If the less specific route does not have a "gateway"
1175                  * (this is the case when the route just goes to a p2p or an
1176                  * stf interface), we'll mistakenly make a neighbor cache for
1177                  * the host route, and will see strange neighbor solicitation
1178                  * for the corresponding destination.  In order to avoid the
1179                  * confusion, we check if the destination of the route is
1180                  * a neighbor in terms of neighbor discovery, and stop the
1181                  * process if not.  Additionally, we remove the LLINFO flag
1182                  * so that ndp(8) will not try to get the neighbor information
1183                  * of the destination.
1184                  */
1185                 rt->rt_flags &= ~RTF_LLINFO;
1186                 return;
1187         }
1188
1189         switch (req) {
1190         case RTM_ADD:
1191                 /*
1192                  * There is no backward compatibility :)
1193                  *
1194                  * if (!(rt->rt_flags & RTF_HOST) &&
1195                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1196                  *         rt->rt_flags |= RTF_CLONING;
1197                  */
1198                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1199                         /*
1200                          * Case 1: This route should come from
1201                          * a route to interface.  RTF_LLINFO flag is set
1202                          * for a host route whose destination should be
1203                          * treated as on-link.
1204                          */
1205                         rt_setgate(rt, rt_key(rt),
1206                                    (struct sockaddr *)&null_sdl);
1207                         gate = rt->rt_gateway;
1208                         SDL(gate)->sdl_type = ifp->if_type;
1209                         SDL(gate)->sdl_index = ifp->if_index;
1210                         if (ln)
1211                                 ln->ln_expire = time_uptime;
1212                         if (ln && ln->ln_expire == 0) {
1213                                 /* kludge for desktops */
1214                                 ln->ln_expire = 1;
1215                         }
1216                         if ((rt->rt_flags & RTF_CLONING))
1217                                 break;
1218                 }
1219                 /*
1220                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1221                  * We don't do that here since llinfo is not ready yet.
1222                  *
1223                  * There are also couple of other things to be discussed:
1224                  * - unsolicited NA code needs improvement beforehand
1225                  * - RFC2461 says we MAY send multicast unsolicited NA
1226                  *   (7.2.6 paragraph 4), however, it also says that we
1227                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1228                  *   we don't have anything like it right now.
1229                  *   note that the mechanism needs a mutual agreement
1230                  *   between proxies, which means that we need to implement
1231                  *   a new protocol, or a new kludge.
1232                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1233                  *   we need to check ip6forwarding before sending it.
1234                  *   (or should we allow proxy ND configuration only for
1235                  *   routers?  there's no mention about proxy ND from hosts)
1236                  */
1237 #if 0
1238                 /* XXX it does not work */
1239                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1240                         nd6_na_output(ifp,
1241                               &SIN6(rt_key(rt))->sin6_addr,
1242                               &SIN6(rt_key(rt))->sin6_addr,
1243                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1244                               1, NULL);
1245                 }
1246 #endif
1247                 /* FALLTHROUGH */
1248         case RTM_RESOLVE:
1249                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1250                         /*
1251                          * Address resolution isn't necessary for a point to
1252                          * point link, so we can skip this test for a p2p link.
1253                          */
1254                         if (gate->sa_family != AF_LINK ||
1255                             gate->sa_len < sizeof(null_sdl)) {
1256                                 log(LOG_DEBUG,
1257                                     "nd6_rtrequest: bad gateway value: %s\n",
1258                                     if_name(ifp));
1259                                 break;
1260                         }
1261                         SDL(gate)->sdl_type = ifp->if_type;
1262                         SDL(gate)->sdl_index = ifp->if_index;
1263                 }
1264                 if (ln != NULL)
1265                         break;  /* This happens on a route change */
1266                 /*
1267                  * Case 2: This route may come from cloning, or a manual route
1268                  * add with a LL address.
1269                  */
1270                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1271                 rt->rt_llinfo = (caddr_t)ln;
1272                 if (!ln) {
1273                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1274                         break;
1275                 }
1276                 nd6_inuse++;
1277                 nd6_allocated++;
1278                 bzero(ln, sizeof(*ln));
1279                 ln->ln_rt = rt;
1280                 /* this is required for "ndp" command. - shin */
1281                 if (req == RTM_ADD) {
1282                         /*
1283                          * gate should have some valid AF_LINK entry,
1284                          * and ln->ln_expire should have some lifetime
1285                          * which is specified by ndp command.
1286                          */
1287                         ln->ln_state = ND6_LLINFO_REACHABLE;
1288                         ln->ln_byhint = 0;
1289                 } else {
1290                         /*
1291                          * When req == RTM_RESOLVE, rt is created and
1292                          * initialized in rtrequest(), so rt_expire is 0.
1293                          */
1294                         ln->ln_state = ND6_LLINFO_NOSTATE;
1295                         ln->ln_expire = time_uptime;
1296                 }
1297                 rt->rt_flags |= RTF_LLINFO;
1298                 ln->ln_next = llinfo_nd6.ln_next;
1299                 llinfo_nd6.ln_next = ln;
1300                 ln->ln_prev = &llinfo_nd6;
1301                 ln->ln_next->ln_prev = ln;
1302
1303                 /*
1304                  * check if rt_key(rt) is one of my address assigned
1305                  * to the interface.
1306                  */
1307                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1308                     &SIN6(rt_key(rt))->sin6_addr);
1309                 if (ifa) {
1310                         caddr_t macp = nd6_ifptomac(ifp);
1311                         ln->ln_expire = 0;
1312                         ln->ln_state = ND6_LLINFO_REACHABLE;
1313                         ln->ln_byhint = 0;
1314                         if (macp) {
1315                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1316                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1317                         }
1318                         if (nd6_useloopback) {
1319                                 rt->rt_ifp = loif;      /* XXX */
1320                                 /*
1321                                  * Make sure rt_ifa be equal to the ifaddr
1322                                  * corresponding to the address.
1323                                  * We need this because when we refer
1324                                  * rt_ifa->ia6_flags in ip6_input, we assume
1325                                  * that the rt_ifa points to the address instead
1326                                  * of the loopback address.
1327                                  */
1328                                 if (ifa != rt->rt_ifa) {
1329                                         IFAFREE(rt->rt_ifa);
1330                                         IFAREF(ifa);
1331                                         rt->rt_ifa = ifa;
1332                                 }
1333                         }
1334                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1335                         ln->ln_expire = 0;
1336                         ln->ln_state = ND6_LLINFO_REACHABLE;
1337                         ln->ln_byhint = 0;
1338
1339                         /*
1340                          * Join solicited node multicast for proxy ND, and only
1341                          * join it once on cpu0.
1342                          */
1343                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1344                                 struct in6_addr llsol;
1345                                 int error;
1346
1347                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1348                                 llsol.s6_addr16[0] = htons(0xff02);
1349                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1350                                 llsol.s6_addr32[1] = 0;
1351                                 llsol.s6_addr32[2] = htonl(1);
1352                                 llsol.s6_addr8[12] = 0xff;
1353
1354                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1355                                         nd6log((LOG_ERR, "%s: failed to join "
1356                                             "%s (errno=%d)\n", if_name(ifp),
1357                                             ip6_sprintf(&llsol), error));
1358                                 }
1359                         }
1360                 }
1361                 break;
1362
1363         case RTM_DELETE:
1364                 if (!ln)
1365                         break;
1366                 /*
1367                  * Leave from solicited node multicast for proxy ND, and only
1368                  * leave it once on cpu0 (since we joined it once on cpu0).
1369                  */
1370                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1371                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1372                         struct in6_addr llsol;
1373                         struct in6_multi *in6m;
1374
1375                         llsol = SIN6(rt_key(rt))->sin6_addr;
1376                         llsol.s6_addr16[0] = htons(0xff02);
1377                         llsol.s6_addr16[1] = htons(ifp->if_index);
1378                         llsol.s6_addr32[1] = 0;
1379                         llsol.s6_addr32[2] = htonl(1);
1380                         llsol.s6_addr8[12] = 0xff;
1381
1382                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1383                         if (in6m)
1384                                 in6_delmulti(in6m);
1385                 }
1386                 nd6_inuse--;
1387                 ln->ln_next->ln_prev = ln->ln_prev;
1388                 ln->ln_prev->ln_next = ln->ln_next;
1389                 ln->ln_prev = NULL;
1390                 rt->rt_llinfo = 0;
1391                 rt->rt_flags &= ~RTF_LLINFO;
1392                 if (ln->ln_hold)
1393                         m_freem(ln->ln_hold);
1394                 Free((caddr_t)ln);
1395         }
1396 }
1397
1398 int
1399 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1400 {
1401         struct in6_drlist *drl = (struct in6_drlist *)data;
1402         struct in6_prlist *prl = (struct in6_prlist *)data;
1403         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1404         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1405         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1406         struct nd_defrouter *dr, any;
1407         struct nd_prefix *pr;
1408         struct rtentry *rt;
1409         int i = 0, error = 0;
1410
1411         switch (cmd) {
1412         case SIOCGDRLST_IN6:
1413                 /*
1414                  * obsolete API, use sysctl under net.inet6.icmp6
1415                  */
1416                 bzero(drl, sizeof(*drl));
1417                 mtx_lock(&nd6_mtx);
1418                 dr = TAILQ_FIRST(&nd_defrouter);
1419                 while (dr && i < DRLSTSIZ) {
1420                         drl->defrouter[i].rtaddr = dr->rtaddr;
1421                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1422                                 /* XXX: need to this hack for KAME stack */
1423                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1424                         } else
1425                                 log(LOG_ERR,
1426                                     "default router list contains a "
1427                                     "non-linklocal address(%s)\n",
1428                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1429
1430                         drl->defrouter[i].flags = dr->flags;
1431                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1432                         drl->defrouter[i].expire = dr->expire;
1433                         drl->defrouter[i].if_index = dr->ifp->if_index;
1434                         i++;
1435                         dr = TAILQ_NEXT(dr, dr_entry);
1436                 }
1437                 mtx_unlock(&nd6_mtx);
1438                 break;
1439         case SIOCGPRLST_IN6:
1440                 /*
1441                  * obsolete API, use sysctl under net.inet6.icmp6
1442                  */
1443                 /*
1444                  * XXX meaning of fields, especialy "raflags", is very
1445                  * differnet between RA prefix list and RR/static prefix list.
1446                  * how about separating ioctls into two?
1447                  */
1448                 bzero(prl, sizeof(*prl));
1449                 mtx_lock(&nd6_mtx);
1450                 pr = nd_prefix.lh_first;
1451                 while (pr && i < PRLSTSIZ) {
1452                         struct nd_pfxrouter *pfr;
1453                         int j;
1454
1455                         in6_embedscope(&prl->prefix[i].prefix,
1456                             &pr->ndpr_prefix, NULL, NULL);
1457                         prl->prefix[i].raflags = pr->ndpr_raf;
1458                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1459                         prl->prefix[i].vltime = pr->ndpr_vltime;
1460                         prl->prefix[i].pltime = pr->ndpr_pltime;
1461                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1462                         prl->prefix[i].expire = pr->ndpr_expire;
1463
1464                         pfr = pr->ndpr_advrtrs.lh_first;
1465                         j = 0;
1466                         while (pfr) {
1467                                 if (j < DRLSTSIZ) {
1468 #define RTRADDR prl->prefix[i].advrtr[j]
1469                                         RTRADDR = pfr->router->rtaddr;
1470                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1471                                                 /* XXX: hack for KAME */
1472                                                 RTRADDR.s6_addr16[1] = 0;
1473                                         } else
1474                                                 log(LOG_ERR,
1475                                                     "a router(%s) advertises "
1476                                                     "a prefix with "
1477                                                     "non-link local address\n",
1478                                                     ip6_sprintf(&RTRADDR));
1479 #undef RTRADDR
1480                                 }
1481                                 j++;
1482                                 pfr = pfr->pfr_next;
1483                         }
1484                         prl->prefix[i].advrtrs = j;
1485                         prl->prefix[i].origin = PR_ORIG_RA;
1486
1487                         i++;
1488                         pr = pr->ndpr_next;
1489                 }
1490                 mtx_unlock(&nd6_mtx);
1491
1492                 break;
1493         case OSIOCGIFINFO_IN6:
1494                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1495                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1496                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1497                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1498                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1499                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1500                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1501                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1502                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1503                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1504                 break;
1505         case SIOCGIFINFO_IN6:
1506                 ndi->ndi = *ND_IFINFO(ifp);
1507                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1508                 break;
1509         case SIOCSIFINFO_IN6:
1510                 /*
1511                  * used to change host variables from userland.
1512                  * intented for a use on router to reflect RA configurations.
1513                  */
1514                 /* 0 means 'unspecified' */
1515                 if (ndi->ndi.linkmtu != 0) {
1516                         if (ndi->ndi.linkmtu < IPV6_MMTU ||
1517                             ndi->ndi.linkmtu > IN6_LINKMTU(ifp)) {
1518                                 error = EINVAL;
1519                                 break;
1520                         }
1521                         ND_IFINFO(ifp)->linkmtu = ndi->ndi.linkmtu;
1522                 }
1523
1524                 if (ndi->ndi.basereachable != 0) {
1525                         int obasereachable = ND_IFINFO(ifp)->basereachable;
1526
1527                         ND_IFINFO(ifp)->basereachable = ndi->ndi.basereachable;
1528                         if (ndi->ndi.basereachable != obasereachable)
1529                                 ND_IFINFO(ifp)->reachable =
1530                                     ND_COMPUTE_RTIME(ndi->ndi.basereachable);
1531                 }
1532                 if (ndi->ndi.retrans != 0)
1533                         ND_IFINFO(ifp)->retrans = ndi->ndi.retrans;
1534                 if (ndi->ndi.chlim != 0)
1535                         ND_IFINFO(ifp)->chlim = ndi->ndi.chlim;
1536                 /* FALLTHROUGH */
1537         case SIOCSIFINFO_FLAGS:
1538                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1539                 break;
1540         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1541                 /* flush default router list */
1542                 /*
1543                  * xxx sumikawa: should not delete route if default
1544                  * route equals to the top of default router list
1545                  */
1546                 bzero(&any, sizeof(any));
1547                 defrouter_delreq(&any, 0);
1548                 defrouter_select();
1549                 /* xxx sumikawa: flush prefix list */
1550                 break;
1551         case SIOCSPFXFLUSH_IN6:
1552         {
1553                 /* flush all the prefix advertised by routers */
1554                 struct nd_prefix *pr, *next;
1555
1556                 mtx_lock(&nd6_mtx);
1557                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1558                         struct in6_ifaddr *ia, *ia_next;
1559
1560                         next = pr->ndpr_next;
1561
1562                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1563                                 continue; /* XXX */
1564
1565                         /* do we really have to remove addresses as well? */
1566                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1567                                 /* ia might be removed.  keep the next ptr. */
1568                                 ia_next = ia->ia_next;
1569
1570                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1571                                         continue;
1572
1573                                 if (ia->ia6_ndpr == pr)
1574                                         in6_purgeaddr(&ia->ia_ifa);
1575                         }
1576                         prelist_remove(pr);
1577                 }
1578                 mtx_unlock(&nd6_mtx);
1579                 break;
1580         }
1581         case SIOCSRTRFLUSH_IN6:
1582         {
1583                 /* flush all the default routers */
1584                 struct nd_defrouter *dr, *next;
1585
1586                 mtx_lock(&nd6_mtx);
1587                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1588                         /*
1589                          * The first entry of the list may be stored in
1590                          * the routing table, so we'll delete it later.
1591                          */
1592                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1593                                 next = TAILQ_NEXT(dr, dr_entry);
1594                                 defrtrlist_del(dr);
1595                         }
1596                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1597                 }
1598                 mtx_unlock(&nd6_mtx);
1599                 break;
1600         }
1601         case SIOCGNBRINFO_IN6:
1602         {
1603                 struct llinfo_nd6 *ln;
1604                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1605
1606                 /*
1607                  * XXX: KAME specific hack for scoped addresses
1608                  *      XXXX: for other scopes than link-local?
1609                  */
1610                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1611                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1612                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1613
1614                         if (*idp == 0)
1615                                 *idp = htons(ifp->if_index);
1616                 }
1617
1618                 mtx_lock(&nd6_mtx);
1619                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1620                         error = EINVAL;
1621                         mtx_unlock(&nd6_mtx);
1622                         break;
1623                 }
1624                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1625                 nbi->state = ln->ln_state;
1626                 nbi->asked = ln->ln_asked;
1627                 nbi->isrouter = ln->ln_router;
1628                 nbi->expire = ln->ln_expire;
1629                 mtx_unlock(&nd6_mtx);
1630
1631                 break;
1632         }
1633         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1634                 ndif->ifindex = nd6_defifindex;
1635                 break;
1636         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1637                 return (nd6_setdefaultiface(ndif->ifindex));
1638         }
1639         return (error);
1640 }
1641
1642 /*
1643  * Create neighbor cache entry and cache link-layer address,
1644  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1645  */
1646 struct rtentry *
1647 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1648                  int lladdrlen,
1649                  int type,      /* ICMP6 type */
1650                  int code       /* type dependent information */)
1651 {
1652         struct rtentry *rt = NULL;
1653         struct llinfo_nd6 *ln = NULL;
1654         int is_newentry;
1655         struct sockaddr_dl *sdl = NULL;
1656         int do_update;
1657         int olladdr;
1658         int llchange;
1659         int newstate = 0;
1660
1661         if (!ifp)
1662                 panic("ifp == NULL in nd6_cache_lladdr");
1663         if (!from)
1664                 panic("from == NULL in nd6_cache_lladdr");
1665
1666         /* nothing must be updated for unspecified address */
1667         if (IN6_IS_ADDR_UNSPECIFIED(from))
1668                 return NULL;
1669
1670         /*
1671          * Validation about ifp->if_addrlen and lladdrlen must be done in
1672          * the caller.
1673          *
1674          * XXX If the link does not have link-layer adderss, what should
1675          * we do? (ifp->if_addrlen == 0)
1676          * Spec says nothing in sections for RA, RS and NA.  There's small
1677          * description on it in NS section (RFC 2461 7.2.3).
1678          */
1679
1680         rt = nd6_lookup(from, 0, ifp);
1681         if (!rt) {
1682 #if 0
1683                 /* nothing must be done if there's no lladdr */
1684                 if (!lladdr || !lladdrlen)
1685                         return NULL;
1686 #endif
1687
1688                 rt = nd6_lookup(from, 1, ifp);
1689                 is_newentry = 1;
1690         } else {
1691                 /* do nothing if static ndp is set */
1692                 if (rt->rt_flags & RTF_STATIC)
1693                         return NULL;
1694                 is_newentry = 0;
1695         }
1696
1697         if (!rt)
1698                 return NULL;
1699         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1700 fail:
1701                 nd6_free(rt);
1702                 return NULL;
1703         }
1704         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1705         if (!ln)
1706                 goto fail;
1707         if (!rt->rt_gateway)
1708                 goto fail;
1709         if (rt->rt_gateway->sa_family != AF_LINK)
1710                 goto fail;
1711         sdl = SDL(rt->rt_gateway);
1712
1713         olladdr = (sdl->sdl_alen) ? 1 : 0;
1714         if (olladdr && lladdr) {
1715                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1716                         llchange = 1;
1717                 else
1718                         llchange = 0;
1719         } else
1720                 llchange = 0;
1721
1722         /*
1723          * newentry olladdr  lladdr  llchange   (*=record)
1724          *      0       n       n       --      (1)
1725          *      0       y       n       --      (2)
1726          *      0       n       y       --      (3) * STALE
1727          *      0       y       y       n       (4) *
1728          *      0       y       y       y       (5) * STALE
1729          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1730          *      1       --      y       --      (7) * STALE
1731          */
1732
1733         if (lladdr) {           /* (3-5) and (7) */
1734                 /*
1735                  * Record source link-layer address
1736                  * XXX is it dependent to ifp->if_type?
1737                  */
1738                 sdl->sdl_alen = ifp->if_addrlen;
1739                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1740         }
1741
1742         if (!is_newentry) {
1743                 if ((!olladdr && lladdr) ||             /* (3) */
1744                     (olladdr && lladdr && llchange)) {  /* (5) */
1745                         do_update = 1;
1746                         newstate = ND6_LLINFO_STALE;
1747                 } else {                                /* (1-2,4) */
1748                         do_update = 0;
1749                 }
1750         } else {
1751                 do_update = 1;
1752                 if (!lladdr)                            /* (6) */
1753                         newstate = ND6_LLINFO_NOSTATE;
1754                 else                                    /* (7) */
1755                         newstate = ND6_LLINFO_STALE;
1756         }
1757
1758         if (do_update) {
1759                 /*
1760                  * Update the state of the neighbor cache.
1761                  */
1762                 ln->ln_state = newstate;
1763
1764                 if (ln->ln_state == ND6_LLINFO_STALE) {
1765                         /*
1766                          * XXX: since nd6_output() below will cause
1767                          * state tansition to DELAY and reset the timer,
1768                          * we must set the timer now, although it is actually
1769                          * meaningless.
1770                          */
1771                         ln->ln_expire = time_uptime + nd6_gctimer;
1772
1773                         if (ln->ln_hold) {
1774                                 /*
1775                                  * we assume ifp is not a p2p here, so just
1776                                  * set the 2nd argument as the 1st one.
1777                                  */
1778                                 nd6_output(ifp, ifp, ln->ln_hold,
1779                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1780                                 ln->ln_hold = NULL;
1781                         }
1782                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1783                         /* probe right away */
1784                         ln->ln_expire = time_uptime;
1785                 }
1786         }
1787
1788         /*
1789          * ICMP6 type dependent behavior.
1790          *
1791          * NS: clear IsRouter if new entry
1792          * RS: clear IsRouter
1793          * RA: set IsRouter if there's lladdr
1794          * redir: clear IsRouter if new entry
1795          *
1796          * RA case, (1):
1797          * The spec says that we must set IsRouter in the following cases:
1798          * - If lladdr exist, set IsRouter.  This means (1-5).
1799          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1800          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1801          * A quetion arises for (1) case.  (1) case has no lladdr in the
1802          * neighbor cache, this is similar to (6).
1803          * This case is rare but we figured that we MUST NOT set IsRouter.
1804          *
1805          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1806          *                                                      D R
1807          *      0       n       n       --      (1)     c   ?     s
1808          *      0       y       n       --      (2)     c   s     s
1809          *      0       n       y       --      (3)     c   s     s
1810          *      0       y       y       n       (4)     c   s     s
1811          *      0       y       y       y       (5)     c   s     s
1812          *      1       --      n       --      (6) c   c       c s
1813          *      1       --      y       --      (7) c   c   s   c s
1814          *
1815          *                                      (c=clear s=set)
1816          */
1817         switch (type & 0xff) {
1818         case ND_NEIGHBOR_SOLICIT:
1819                 /*
1820                  * New entry must have is_router flag cleared.
1821                  */
1822                 if (is_newentry)        /* (6-7) */
1823                         ln->ln_router = 0;
1824                 break;
1825         case ND_REDIRECT:
1826                 /*
1827                  * If the icmp is a redirect to a better router, always set the
1828                  * is_router flag.  Otherwise, if the entry is newly created,
1829                  * clear the flag.  [RFC 2461, sec 8.3]
1830                  */
1831                 if (code == ND_REDIRECT_ROUTER)
1832                         ln->ln_router = 1;
1833                 else if (is_newentry) /* (6-7) */
1834                         ln->ln_router = 0;
1835                 break;
1836         case ND_ROUTER_SOLICIT:
1837                 /*
1838                  * is_router flag must always be cleared.
1839                  */
1840                 ln->ln_router = 0;
1841                 break;
1842         case ND_ROUTER_ADVERT:
1843                 /*
1844                  * Mark an entry with lladdr as a router.
1845                  */
1846                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1847                     (is_newentry && lladdr)) {                  /* (7) */
1848                         ln->ln_router = 1;
1849                 }
1850                 break;
1851         }
1852
1853         if (llchange || lladdr)
1854                 rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1855
1856         /*
1857          * When the link-layer address of a router changes, select the
1858          * best router again.  In particular, when the neighbor entry is newly
1859          * created, it might affect the selection policy.
1860          * Question: can we restrict the first condition to the "is_newentry"
1861          * case?
1862          * XXX: when we hear an RA from a new router with the link-layer
1863          * address option, defrouter_select() is called twice, since
1864          * defrtrlist_update called the function as well.  However, I believe
1865          * we can compromise the overhead, since it only happens the first
1866          * time.
1867          * XXX: although defrouter_select() should not have a bad effect
1868          * for those are not autoconfigured hosts, we explicitly avoid such
1869          * cases for safety.
1870          */
1871         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1872                 defrouter_select();
1873
1874         return rt;
1875 }
1876
1877 static void
1878 nd6_slowtimo(void *arg __unused)
1879 {
1880         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1881
1882         KASSERT(mycpuid == 0, ("not on cpu0"));
1883         crit_enter();
1884         if (lmsg->ms_flags & MSGF_DONE)
1885                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1886         crit_exit();
1887 }
1888
1889 static void
1890 nd6_slowtimo_dispatch(netmsg_t nmsg)
1891 {
1892         const struct ifnet_array *arr;
1893         struct nd_ifinfo *nd6if;
1894         int i;
1895
1896         ASSERT_NETISR0;
1897
1898         crit_enter();
1899         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1900         crit_exit();
1901
1902         arr = ifnet_array_get();
1903
1904         mtx_lock(&nd6_mtx);
1905         for (i = 0; i < arr->ifnet_count; ++i) {
1906                 struct ifnet *ifp = arr->ifnet_arr[i];
1907
1908                 if (ifp->if_afdata[AF_INET6] == NULL)
1909                         continue;
1910                 nd6if = ND_IFINFO(ifp);
1911                 if (nd6if->basereachable && /* already initialized */
1912                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1913                         /*
1914                          * Since reachable time rarely changes by router
1915                          * advertisements, we SHOULD insure that a new random
1916                          * value gets recomputed at least once every few hours.
1917                          * (RFC 2461, 6.3.4)
1918                          */
1919                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1920                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1921                 }
1922         }
1923         mtx_unlock(&nd6_mtx);
1924
1925         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1926             nd6_slowtimo, NULL);
1927 }
1928
1929 int
1930 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1931            struct sockaddr_in6 *dst, struct rtentry *rt)
1932 {
1933         int error;
1934
1935         if (ifp->if_flags & IFF_LOOPBACK)
1936                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1937         else
1938                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1939         return error;
1940 }
1941
1942 int
1943 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1944     struct sockaddr *dst0, u_char *desten)
1945 {
1946         struct sockaddr_in6 *dst = SIN6(dst0);
1947         struct rtentry *rt = NULL;
1948         struct llinfo_nd6 *ln = NULL;
1949         int error;
1950
1951         if (m->m_flags & M_MCAST) {
1952                 switch (ifp->if_type) {
1953                 case IFT_ETHER:
1954 #ifdef IFT_L2VLAN
1955                 case IFT_L2VLAN:
1956 #endif
1957 #ifdef IFT_IEEE80211
1958                 case IFT_IEEE80211:
1959 #endif
1960                         ETHER_MAP_IPV6_MULTICAST(&dst->sin6_addr,
1961                                                  desten);
1962                         return 0;
1963                 case IFT_IEEE1394:
1964                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1965                         return 0;
1966                 default:
1967                         error = EAFNOSUPPORT;
1968                         goto bad;
1969                 }
1970         }
1971
1972         if (rt0 != NULL) {
1973                 error = rt_llroute(dst0, rt0, &rt);
1974                 if (error != 0)
1975                         goto bad;
1976                 ln = rt->rt_llinfo;
1977         }
1978
1979         /*
1980          * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1981          * the condition below is not very efficient.  But we believe
1982          * it is tolerable, because this should be a rare case.
1983          */
1984         if (ln == NULL && nd6_is_addr_neighbor(dst, ifp)) {
1985                 rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
1986                 if (rt != NULL)
1987                         ln = rt->rt_llinfo;
1988         }
1989
1990         if (ln == NULL || rt == NULL) {
1991                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1992                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1993                         log(LOG_DEBUG,
1994                             "nd6_output: can't allocate llinfo for %s "
1995                             "(ln=%p, rt=%p)\n",
1996                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1997                         error = ENOBUFS;
1998                         goto bad;
1999                 }
2000                 return 0;
2001         }
2002
2003         /* We don't have to do link-layer address resolution on a p2p link. */
2004         if ((ifp->if_flags & IFF_POINTOPOINT) &&
2005             ln->ln_state < ND6_LLINFO_REACHABLE) {
2006                 ln->ln_state = ND6_LLINFO_STALE;
2007                 ln->ln_expire = time_uptime + nd6_gctimer;
2008         }
2009
2010         /*
2011          * The first time we send a packet to a neighbor whose entry is
2012          * STALE, we have to change the state to DELAY and a sets a timer to
2013          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2014          * neighbor unreachability detection on expiration.
2015          * (RFC 2461 7.3.3)
2016          */
2017         if (ln->ln_state == ND6_LLINFO_STALE) {
2018                 ln->ln_asked = 0;
2019                 ln->ln_state = ND6_LLINFO_DELAY;
2020                 ln->ln_expire = time_uptime + nd6_delay;
2021         }
2022
2023         /*
2024          * If the neighbor cache entry has a state other than INCOMPLETE
2025          * (i.e. its link-layer address is already resolved), return it.
2026          */
2027         if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2028                 struct sockaddr_dl *sdl = SDL(rt->rt_gateway);
2029
2030                 KKASSERT(sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0);
2031                 bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2032                 return 0;
2033         }
2034
2035         /*
2036          * There is a neighbor cache entry, but no ethernet address
2037          * response yet.  Replace the held mbuf (if any) with this
2038          * latest one.
2039          */
2040         if (ln->ln_hold)
2041                 m_freem(ln->ln_hold);
2042         ln->ln_hold = m;
2043
2044         /*
2045          * This code conforms to the rate-limiting rule described in Section
2046          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2047          * an NS below.
2048          */
2049         if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2050             ln->ln_state == ND6_LLINFO_WAITDELETE) {
2051                 /*
2052                  * This neighbor cache entry was just created; change its
2053                  * state to INCOMPLETE and start its life cycle.
2054                  *
2055                  * We force an NS output below by setting ln_expire to 1
2056                  * (nd6_rtrequest() could set it to the current time_uptime)
2057                  * and zeroing out ln_asked (XXX this may not be necessary).
2058                  */
2059                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2060                 ln->ln_expire = 1;
2061                 ln->ln_asked = 0;
2062         }
2063         if (ln->ln_expire && ln->ln_expire < time_uptime && ln->ln_asked == 0) {
2064                 ln->ln_asked++;
2065                 ln->ln_expire = time_uptime + ND_IFINFO(ifp)->retrans / 1000;
2066                 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2067         }
2068
2069         if (ln->ln_asked >= nd6_mmaxtries)
2070                 return (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2071                     EHOSTUNREACH : EHOSTDOWN;
2072         return EWOULDBLOCK;
2073
2074 bad:
2075         m_freem(m);
2076         return error;
2077 }
2078
2079 int
2080 nd6_need_cache(struct ifnet *ifp)
2081 {
2082         /*
2083          * XXX: we currently do not make neighbor cache on any interface
2084          * other than Ethernet and GIF.
2085          *
2086          * RFC2893 says:
2087          * - unidirectional tunnels needs no ND
2088          */
2089         switch (ifp->if_type) {
2090         case IFT_ETHER:
2091         case IFT_IEEE1394:
2092 #ifdef IFT_L2VLAN
2093         case IFT_L2VLAN:
2094 #endif
2095 #ifdef IFT_IEEE80211
2096         case IFT_IEEE80211:
2097 #endif
2098 #ifdef IFT_CARP
2099         case IFT_CARP:
2100 #endif
2101         case IFT_GIF:           /* XXX need more cases? */
2102                 return (1);
2103         default:
2104                 return (0);
2105         }
2106 }
2107
2108 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2109 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2110 #ifdef SYSCTL_DECL
2111 SYSCTL_DECL(_net_inet6_icmp6);
2112 #endif
2113 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2114         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2115 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2116         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2117
2118 static int
2119 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2120 {
2121         int error;
2122         char buf[1024];
2123         struct in6_defrouter *d, *de;
2124         struct nd_defrouter *dr;
2125
2126         if (req->newptr)
2127                 return EPERM;
2128         error = 0;
2129
2130         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2131              dr = TAILQ_NEXT(dr, dr_entry)) {
2132                 d = (struct in6_defrouter *)buf;
2133                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2134
2135                 if (d + 1 <= de) {
2136                         bzero(d, sizeof(*d));
2137                         d->rtaddr.sin6_family = AF_INET6;
2138                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2139                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2140                             dr->ifp) != 0)
2141                                 log(LOG_ERR,
2142                                     "scope error in "
2143                                     "default router list (%s)\n",
2144                                     ip6_sprintf(&dr->rtaddr));
2145                         d->flags = dr->flags;
2146                         d->rtlifetime = dr->rtlifetime;
2147                         d->expire = dr->expire;
2148                         d->if_index = dr->ifp->if_index;
2149                 } else
2150                         panic("buffer too short");
2151
2152                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2153                 if (error)
2154                         break;
2155         }
2156         return error;
2157 }
2158
2159 static int
2160 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2161 {
2162         int error;
2163         char buf[1024];
2164         struct in6_prefix *p, *pe;
2165         struct nd_prefix *pr;
2166
2167         if (req->newptr)
2168                 return EPERM;
2169         error = 0;
2170
2171         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2172                 u_short advrtrs;
2173                 size_t advance;
2174                 struct sockaddr_in6 *sin6, *s6;
2175                 struct nd_pfxrouter *pfr;
2176
2177                 p = (struct in6_prefix *)buf;
2178                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2179
2180                 if (p + 1 <= pe) {
2181                         bzero(p, sizeof(*p));
2182                         sin6 = (struct sockaddr_in6 *)(p + 1);
2183
2184                         p->prefix = pr->ndpr_prefix;
2185                         if (in6_recoverscope(&p->prefix,
2186                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2187                                 log(LOG_ERR,
2188                                     "scope error in prefix list (%s)\n",
2189                                     ip6_sprintf(&p->prefix.sin6_addr));
2190                         p->raflags = pr->ndpr_raf;
2191                         p->prefixlen = pr->ndpr_plen;
2192                         p->vltime = pr->ndpr_vltime;
2193                         p->pltime = pr->ndpr_pltime;
2194                         p->if_index = pr->ndpr_ifp->if_index;
2195                         p->expire = pr->ndpr_expire;
2196                         p->refcnt = pr->ndpr_refcnt;
2197                         p->flags = pr->ndpr_stateflags;
2198                         p->origin = PR_ORIG_RA;
2199                         advrtrs = 0;
2200                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2201                              pfr = pfr->pfr_next) {
2202                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2203                                         advrtrs++;
2204                                         continue;
2205                                 }
2206                                 s6 = &sin6[advrtrs];
2207                                 bzero(s6, sizeof(*s6));
2208                                 s6->sin6_family = AF_INET6;
2209                                 s6->sin6_len = sizeof(*sin6);
2210                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2211                                     pfr->router->ifp) != 0)
2212                                         log(LOG_ERR,
2213                                             "scope error in "
2214                                             "prefix list (%s)\n",
2215                                             ip6_sprintf(&pfr->router->rtaddr));
2216                                 advrtrs++;
2217                         }
2218                         p->advrtrs = advrtrs;
2219                 } else {
2220                         panic("buffer too short");
2221                 }
2222
2223                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2224                 error = SYSCTL_OUT(req, buf, advance);
2225                 if (error)
2226                         break;
2227         }
2228         return error;
2229 }