addfd46f02aa50c57315627d718d109b8cf4bf9d
[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 /*
34  * XXX
35  * KAME 970409 note:
36  * BSD/OS version heavily modifies this code, related to llinfo.
37  * Since we don't have BSD/OS version of net/route.c in our hand,
38  * I left the code mostly as it was in 970310.  -- itojun
39  */
40
41 #include "opt_inet.h"
42 #include "opt_inet6.h"
43
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/sysctl.h>
58 #include <sys/mutex.h>
59
60 #include <sys/thread2.h>
61 #include <sys/mutex2.h>
62
63 #include <net/if.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/route.h>
67 #include <net/netisr2.h>
68 #include <net/netmsg2.h>
69
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/icmp6.h>
77
78 #include <net/net_osdep.h>
79
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 #define SDL(s) ((struct sockaddr_dl *)s)
85
86 /* timer values */
87 int     nd6_prune       = 1;    /* walk list every 1 seconds */
88 int     nd6_delay       = 5;    /* delay first probe time 5 second */
89 int     nd6_umaxtries   = 3;    /* maximum unicast query */
90 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
91 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
92 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
93
94 /* preventing too many loops in ND option parsing */
95 int nd6_maxndopt = 10;  /* max # of ND options allowed */
96
97 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
98
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104
105 /* for debugging? */
106 static int nd6_inuse, nd6_allocated;
107
108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
109 struct nd_drhead nd_defrouter;
110 struct nd_prhead nd_prefix = { 0 };
111 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
112
113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
114 static struct sockaddr_in6 all1_sa;
115
116 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
117 static int regen_tmpaddr (struct in6_ifaddr *);
118 static void nd6_slowtimo(void *);
119 static void nd6_slowtimo_dispatch(netmsg_t);
120 static void nd6_timer(void *);
121 static void nd6_timer_dispatch(netmsg_t);
122
123 static struct callout nd6_slowtimo_ch;
124 static struct netmsg_base nd6_slowtimo_netmsg;
125
126 static struct callout nd6_timer_ch;
127 static struct netmsg_base nd6_timer_netmsg;
128
129 void
130 nd6_init(void)
131 {
132         static int nd6_init_done = 0;
133         int i;
134
135         if (nd6_init_done) {
136                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
137                 return;
138         }
139
140         all1_sa.sin6_family = AF_INET6;
141         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
142         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
143                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
144
145         /* initialization of the default router list */
146         TAILQ_INIT(&nd_defrouter);
147
148         nd6_init_done = 1;
149
150         /* start timer */
151         callout_init_mp(&nd6_slowtimo_ch);
152         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
153             MSGF_PRIORITY, nd6_slowtimo_dispatch);
154         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
155             nd6_slowtimo, NULL, 0);
156 }
157
158 struct nd_ifinfo *
159 nd6_ifattach(struct ifnet *ifp)
160 {
161         struct nd_ifinfo *nd;
162
163         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
164             M_WAITOK | M_ZERO);
165
166         nd->initialized = 1;
167
168         nd->linkmtu = ifp->if_mtu;
169         nd->chlim = IPV6_DEFHLIM;
170         nd->basereachable = REACHABLE_TIME;
171         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
172         nd->retrans = RETRANS_TIMER;
173         nd->receivedra = 0;
174
175         /*
176          * Note that the default value of ip6_accept_rtadv is 0, which means
177          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
178          * here.
179          */
180         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
181
182         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
183         nd6_setmtu0(ifp, nd);
184         return nd;
185 }
186
187 void
188 nd6_ifdetach(struct nd_ifinfo *nd)
189 {
190         kfree(nd, M_IP6NDP);
191 }
192
193 /*
194  * Reset ND level link MTU. This function is called when the physical MTU
195  * changes, which means we might have to adjust the ND level MTU.
196  */
197 void
198 nd6_setmtu(struct ifnet *ifp)
199 {
200         nd6_setmtu0(ifp, ND_IFINFO(ifp));
201 }
202
203 struct netmsg_nd6setmtu {
204         struct netmsg_base      nmsg;
205         struct ifnet            *ifp;
206         struct nd_ifinfo        *ndi;
207 };
208
209 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
210 static void
211 nd6_setmtu0_dispatch(netmsg_t msg)
212 {
213         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
214         struct ifnet *ifp = nmsg->ifp;
215         struct nd_ifinfo *ndi = nmsg->ndi;
216         u_long oldmaxmtu;
217         u_long oldlinkmtu;
218
219         oldmaxmtu = ndi->maxmtu;
220         oldlinkmtu = ndi->linkmtu;
221
222         switch (ifp->if_type) {
223         case IFT_ETHER:
224                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
225                 break;
226         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
227                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
228                 break;
229 #ifdef IFT_IEEE80211
230         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
231                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
232                 break;
233 #endif
234         default:
235                 ndi->maxmtu = ifp->if_mtu;
236                 break;
237         }
238
239         if (oldmaxmtu != ndi->maxmtu) {
240                 /*
241                  * If the ND level MTU is not set yet, or if the maxmtu
242                  * is reset to a smaller value than the ND level MTU,
243                  * also reset the ND level MTU.
244                  */
245                 if (ndi->linkmtu == 0 ||
246                     ndi->maxmtu < ndi->linkmtu) {
247                         ndi->linkmtu = ndi->maxmtu;
248                         /* also adjust in6_maxmtu if necessary. */
249                         if (oldlinkmtu == 0) {
250                                 /*
251                                  * XXX: the case analysis is grotty, but
252                                  * it is not efficient to call in6_setmaxmtu()
253                                  * here when we are during the initialization
254                                  * procedure.
255                                  */
256                                 if (in6_maxmtu < ndi->linkmtu)
257                                         in6_maxmtu = ndi->linkmtu;
258                         } else
259                                 in6_setmaxmtu();
260                 }
261         }
262 #undef MIN
263
264         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
265 }
266
267 void
268 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
269 {
270         struct netmsg_nd6setmtu nmsg;
271
272         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
273             nd6_setmtu0_dispatch);
274         nmsg.ifp = ifp;
275         nmsg.ndi = ndi;
276         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
277 }
278
279 void
280 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
281 {
282         bzero(ndopts, sizeof(*ndopts));
283         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
284         ndopts->nd_opts_last
285                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
286
287         if (icmp6len == 0) {
288                 ndopts->nd_opts_done = 1;
289                 ndopts->nd_opts_search = NULL;
290         }
291 }
292
293 /*
294  * Take one ND option.
295  */
296 struct nd_opt_hdr *
297 nd6_option(union nd_opts *ndopts)
298 {
299         struct nd_opt_hdr *nd_opt;
300         int olen;
301
302         if (!ndopts)
303                 panic("ndopts == NULL in nd6_option");
304         if (!ndopts->nd_opts_last)
305                 panic("uninitialized ndopts in nd6_option");
306         if (!ndopts->nd_opts_search)
307                 return NULL;
308         if (ndopts->nd_opts_done)
309                 return NULL;
310
311         nd_opt = ndopts->nd_opts_search;
312
313         /* make sure nd_opt_len is inside the buffer */
314         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
315                 bzero(ndopts, sizeof(*ndopts));
316                 return NULL;
317         }
318
319         olen = nd_opt->nd_opt_len << 3;
320         if (olen == 0) {
321                 /*
322                  * Message validation requires that all included
323                  * options have a length that is greater than zero.
324                  */
325                 bzero(ndopts, sizeof(*ndopts));
326                 return NULL;
327         }
328
329         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
330         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
331                 /* option overruns the end of buffer, invalid */
332                 bzero(ndopts, sizeof(*ndopts));
333                 return NULL;
334         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
335                 /* reached the end of options chain */
336                 ndopts->nd_opts_done = 1;
337                 ndopts->nd_opts_search = NULL;
338         }
339         return nd_opt;
340 }
341
342 /*
343  * Parse multiple ND options.
344  * This function is much easier to use, for ND routines that do not need
345  * multiple options of the same type.
346  */
347 int
348 nd6_options(union nd_opts *ndopts)
349 {
350         struct nd_opt_hdr *nd_opt;
351         int i = 0;
352
353         if (!ndopts)
354                 panic("ndopts == NULL in nd6_options");
355         if (!ndopts->nd_opts_last)
356                 panic("uninitialized ndopts in nd6_options");
357         if (!ndopts->nd_opts_search)
358                 return 0;
359
360         while (1) {
361                 nd_opt = nd6_option(ndopts);
362                 if (!nd_opt && !ndopts->nd_opts_last) {
363                         /*
364                          * Message validation requires that all included
365                          * options have a length that is greater than zero.
366                          */
367                         icmp6stat.icp6s_nd_badopt++;
368                         bzero(ndopts, sizeof(*ndopts));
369                         return -1;
370                 }
371
372                 if (!nd_opt)
373                         goto skip1;
374
375                 switch (nd_opt->nd_opt_type) {
376                 case ND_OPT_SOURCE_LINKADDR:
377                 case ND_OPT_TARGET_LINKADDR:
378                 case ND_OPT_MTU:
379                 case ND_OPT_REDIRECTED_HEADER:
380                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
381                                 nd6log((LOG_INFO,
382                                     "duplicated ND6 option found (type=%d)\n",
383                                     nd_opt->nd_opt_type));
384                                 /* XXX bark? */
385                         } else {
386                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
387                                         = nd_opt;
388                         }
389                         break;
390                 case ND_OPT_PREFIX_INFORMATION:
391                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
392                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
393                                         = nd_opt;
394                         }
395                         ndopts->nd_opts_pi_end =
396                                 (struct nd_opt_prefix_info *)nd_opt;
397                         break;
398                 default:
399                         /*
400                          * Unknown options must be silently ignored,
401                          * to accomodate future extension to the protocol.
402                          */
403                         nd6log((LOG_DEBUG,
404                             "nd6_options: unsupported option %d - "
405                             "option ignored\n", nd_opt->nd_opt_type));
406                 }
407
408 skip1:
409                 i++;
410                 if (i > nd6_maxndopt) {
411                         icmp6stat.icp6s_nd_toomanyopt++;
412                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
413                         break;
414                 }
415
416                 if (ndopts->nd_opts_done)
417                         break;
418         }
419
420         return 0;
421 }
422
423 /*
424  * ND6 timer routine to expire default route list and prefix list
425  */
426 static void
427 nd6_timer_dispatch(netmsg_t nmsg)
428 {
429         struct llinfo_nd6 *ln;
430         struct nd_defrouter *dr;
431         struct nd_prefix *pr;
432         struct ifnet *ifp;
433         struct in6_ifaddr *ia6, *nia6;
434
435         ASSERT_IN_NETISR(0);
436
437         crit_enter();
438         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
439         crit_exit();
440
441         mtx_lock(&nd6_mtx);
442
443         ln = llinfo_nd6.ln_next;
444         while (ln && ln != &llinfo_nd6) {
445                 struct rtentry *rt;
446                 struct sockaddr_in6 *dst;
447                 struct llinfo_nd6 *next = ln->ln_next;
448                 /* XXX: used for the DELAY case only: */
449                 struct nd_ifinfo *ndi = NULL;
450
451                 if ((rt = ln->ln_rt) == NULL) {
452                         ln = next;
453                         continue;
454                 }
455                 if ((ifp = rt->rt_ifp) == NULL) {
456                         ln = next;
457                         continue;
458                 }
459                 ndi = ND_IFINFO(ifp);
460                 dst = (struct sockaddr_in6 *)rt_key(rt);
461
462                 if (ln->ln_expire > time_uptime) {
463                         ln = next;
464                         continue;
465                 }
466
467                 /* sanity check */
468                 if (!rt)
469                         panic("rt=0 in nd6_timer(ln=%p)", ln);
470                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
471                         panic("rt_llinfo(%p) is not equal to ln(%p)",
472                               rt->rt_llinfo, ln);
473                 if (!dst)
474                         panic("dst=0 in nd6_timer(ln=%p)", ln);
475
476                 switch (ln->ln_state) {
477                 case ND6_LLINFO_INCOMPLETE:
478                         if (ln->ln_asked < nd6_mmaxtries) {
479                                 ln->ln_asked++;
480                                 ln->ln_expire = time_uptime +
481                                         ND_IFINFO(ifp)->retrans / 1000;
482                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
483                                         ln, 0);
484                         } else {
485                                 struct mbuf *m = ln->ln_hold;
486                                 if (m) {
487                                         if (rt->rt_ifp) {
488                                                 /*
489                                                  * Fake rcvif to make ICMP error
490                                                  * more helpful in diagnosing
491                                                  * for the receiver.
492                                                  * XXX: should we consider
493                                                  * older rcvif?
494                                                  */
495                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
496                                         }
497                                         icmp6_error(m, ICMP6_DST_UNREACH,
498                                                     ICMP6_DST_UNREACH_ADDR, 0);
499                                         ln->ln_hold = NULL;
500                                 }
501                                 next = nd6_free(rt);
502                         }
503                         break;
504                 case ND6_LLINFO_REACHABLE:
505                         if (ln->ln_expire) {
506                                 ln->ln_state = ND6_LLINFO_STALE;
507                                 ln->ln_expire = time_uptime + nd6_gctimer;
508                         }
509                         break;
510
511                 case ND6_LLINFO_STALE:
512                         /* Garbage Collection(RFC 2461 5.3) */
513                         if (ln->ln_expire)
514                                 next = nd6_free(rt);
515                         break;
516
517                 case ND6_LLINFO_DELAY:
518                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
519                                 /* We need NUD */
520                                 ln->ln_asked = 1;
521                                 ln->ln_state = ND6_LLINFO_PROBE;
522                                 ln->ln_expire = time_uptime +
523                                         ndi->retrans / 1000;
524                                 nd6_ns_output(ifp, &dst->sin6_addr,
525                                               &dst->sin6_addr,
526                                               ln, 0);
527                         } else {
528                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
529                                 ln->ln_expire = time_uptime + nd6_gctimer;
530                         }
531                         break;
532                 case ND6_LLINFO_PROBE:
533                         if (ln->ln_asked < nd6_umaxtries) {
534                                 ln->ln_asked++;
535                                 ln->ln_expire = time_uptime +
536                                         ND_IFINFO(ifp)->retrans / 1000;
537                                 nd6_ns_output(ifp, &dst->sin6_addr,
538                                                &dst->sin6_addr, ln, 0);
539                         } else {
540                                 next = nd6_free(rt);
541                         }
542                         break;
543                 }
544                 ln = next;
545         }
546
547         /* expire default router list */
548         dr = TAILQ_FIRST(&nd_defrouter);
549         while (dr) {
550                 if (dr->expire && dr->expire < time_uptime) {
551                         struct nd_defrouter *t;
552                         t = TAILQ_NEXT(dr, dr_entry);
553                         defrtrlist_del(dr);
554                         dr = t;
555                 } else {
556                         dr = TAILQ_NEXT(dr, dr_entry);
557                 }
558         }
559
560         /*
561          * expire interface addresses.
562          * in the past the loop was inside prefix expiry processing.
563          * However, from a stricter speci-confrmance standpoint, we should
564          * rather separate address lifetimes and prefix lifetimes.
565          */
566 addrloop:
567         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
568                 nia6 = ia6->ia_next;
569                 /* check address lifetime */
570                 if (IFA6_IS_INVALID(ia6)) {
571                         int regen = 0;
572
573                         /*
574                          * If the expiring address is temporary, try
575                          * regenerating a new one.  This would be useful when
576                          * we suspended a laptop PC, then turned it on after a
577                          * period that could invalidate all temporary
578                          * addresses.  Although we may have to restart the
579                          * loop (see below), it must be after purging the
580                          * address.  Otherwise, we'd see an infinite loop of
581                          * regeneration.
582                          */
583                         if (ip6_use_tempaddr &&
584                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
585                                 if (regen_tmpaddr(ia6) == 0)
586                                         regen = 1;
587                         }
588
589                         in6_purgeaddr(&ia6->ia_ifa);
590
591                         if (regen)
592                                 goto addrloop; /* XXX: see below */
593                 }
594                 if (IFA6_IS_DEPRECATED(ia6)) {
595                         int oldflags = ia6->ia6_flags;
596
597                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
598
599                         /*
600                          * If a temporary address has just become deprecated,
601                          * regenerate a new one if possible.
602                          */
603                         if (ip6_use_tempaddr &&
604                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
605                             !(oldflags & IN6_IFF_DEPRECATED)) {
606
607                                 if (regen_tmpaddr(ia6) == 0) {
608                                         /*
609                                          * A new temporary address is
610                                          * generated.
611                                          * XXX: this means the address chain
612                                          * has changed while we are still in
613                                          * the loop.  Although the change
614                                          * would not cause disaster (because
615                                          * it's not a deletion, but an
616                                          * addition,) we'd rather restart the
617                                          * loop just for safety.  Or does this
618                                          * significantly reduce performance??
619                                          */
620                                         goto addrloop;
621                                 }
622                         }
623                 } else {
624                         /*
625                          * A new RA might have made a deprecated address
626                          * preferred.
627                          */
628                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
629                 }
630         }
631
632         /* expire prefix list */
633         pr = nd_prefix.lh_first;
634         while (pr) {
635                 /*
636                  * check prefix lifetime.
637                  * since pltime is just for autoconf, pltime processing for
638                  * prefix is not necessary.
639                  */
640                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
641                         struct nd_prefix *t;
642                         t = pr->ndpr_next;
643
644                         /*
645                          * address expiration and prefix expiration are
646                          * separate.  NEVER perform in6_purgeaddr here.
647                          */
648
649                         prelist_remove(pr);
650                         pr = t;
651                 } else
652                         pr = pr->ndpr_next;
653         }
654
655         mtx_unlock(&nd6_mtx);
656
657         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
658 }
659
660 static void
661 nd6_timer(void *arg __unused)
662 {
663         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
664
665         KASSERT(mycpuid == 0, ("not on cpu0"));
666         crit_enter();
667         if (lmsg->ms_flags & MSGF_DONE)
668                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
669         crit_exit();
670 }
671
672 void
673 nd6_timer_init(void)
674 {
675         callout_init_mp(&nd6_timer_ch);
676         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
677             MSGF_PRIORITY, nd6_timer_dispatch);
678         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
679 }
680
681 static int
682 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
683                                          address */
684 {
685         struct ifaddr_container *ifac;
686         struct ifnet *ifp;
687         struct in6_ifaddr *public_ifa6 = NULL;
688
689         ifp = ia6->ia_ifa.ifa_ifp;
690         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
691                 struct ifaddr *ifa = ifac->ifa;
692                 struct in6_ifaddr *it6;
693
694                 if (ifa->ifa_addr->sa_family != AF_INET6)
695                         continue;
696
697                 it6 = (struct in6_ifaddr *)ifa;
698
699                 /* ignore no autoconf addresses. */
700                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
701                         continue;
702
703                 /* ignore autoconf addresses with different prefixes. */
704                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
705                         continue;
706
707                 /*
708                  * Now we are looking at an autoconf address with the same
709                  * prefix as ours.  If the address is temporary and is still
710                  * preferred, do not create another one.  It would be rare, but
711                  * could happen, for example, when we resume a laptop PC after
712                  * a long period.
713                  */
714                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
715                     !IFA6_IS_DEPRECATED(it6)) {
716                         public_ifa6 = NULL;
717                         break;
718                 }
719
720                 /*
721                  * This is a public autoconf address that has the same prefix
722                  * as ours.  If it is preferred, keep it.  We can't break the
723                  * loop here, because there may be a still-preferred temporary
724                  * address with the prefix.
725                  */
726                 if (!IFA6_IS_DEPRECATED(it6))
727                     public_ifa6 = it6;
728         }
729
730         if (public_ifa6 != NULL) {
731                 int e;
732
733                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
734                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
735                             " tmp addr,errno=%d\n", e);
736                         return (-1);
737                 }
738                 return (0);
739         }
740
741         return (-1);
742 }
743
744 /*
745  * Nuke neighbor cache/prefix/default router management table, right before
746  * ifp goes away.
747  */
748 void
749 nd6_purge(struct ifnet *ifp)
750 {
751         struct llinfo_nd6 *ln, *nln;
752         struct nd_defrouter *dr, *ndr, drany;
753         struct nd_prefix *pr, *npr;
754
755         /* Nuke default router list entries toward ifp */
756         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
757                 /*
758                  * The first entry of the list may be stored in
759                  * the routing table, so we'll delete it later.
760                  */
761                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
762                         ndr = TAILQ_NEXT(dr, dr_entry);
763                         if (dr->ifp == ifp)
764                                 defrtrlist_del(dr);
765                 }
766                 dr = TAILQ_FIRST(&nd_defrouter);
767                 if (dr->ifp == ifp)
768                         defrtrlist_del(dr);
769         }
770
771         /* Nuke prefix list entries toward ifp */
772         for (pr = nd_prefix.lh_first; pr; pr = npr) {
773                 npr = pr->ndpr_next;
774                 if (pr->ndpr_ifp == ifp) {
775                         /*
776                          * Previously, pr->ndpr_addr is removed as well,
777                          * but I strongly believe we don't have to do it.
778                          * nd6_purge() is only called from in6_ifdetach(),
779                          * which removes all the associated interface addresses
780                          * by itself.
781                          * (jinmei@kame.net 20010129)
782                          */
783                         prelist_remove(pr);
784                 }
785         }
786
787         /* cancel default outgoing interface setting */
788         if (nd6_defifindex == ifp->if_index)
789                 nd6_setdefaultiface(0);
790
791         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
792                 /* refresh default router list */
793                 bzero(&drany, sizeof(drany));
794                 defrouter_delreq(&drany, 0);
795                 defrouter_select();
796         }
797
798         /*
799          * Nuke neighbor cache entries for the ifp.
800          * Note that rt->rt_ifp may not be the same as ifp,
801          * due to KAME goto ours hack.  See RTM_RESOLVE case in
802          * nd6_rtrequest(), and ip6_input().
803          */
804         ln = llinfo_nd6.ln_next;
805         while (ln && ln != &llinfo_nd6) {
806                 struct rtentry *rt;
807                 struct sockaddr_dl *sdl;
808
809                 nln = ln->ln_next;
810                 rt = ln->ln_rt;
811                 if (rt && rt->rt_gateway &&
812                     rt->rt_gateway->sa_family == AF_LINK) {
813                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
814                         if (sdl->sdl_index == ifp->if_index)
815                                 nln = nd6_free(rt);
816                 }
817                 ln = nln;
818         }
819 }
820
821 struct rtentry *
822 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
823 {
824         struct rtentry *rt;
825         struct sockaddr_in6 sin6;
826
827         bzero(&sin6, sizeof(sin6));
828         sin6.sin6_len = sizeof(struct sockaddr_in6);
829         sin6.sin6_family = AF_INET6;
830         sin6.sin6_addr = *addr6;
831
832         if (create)
833                 rt = rtlookup((struct sockaddr *)&sin6);
834         else
835                 rt = rtpurelookup((struct sockaddr *)&sin6);
836         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
837                 /*
838                  * This is the case for the default route.
839                  * If we want to create a neighbor cache for the address, we
840                  * should free the route for the destination and allocate an
841                  * interface route.
842                  */
843                 if (create) {
844                         --rt->rt_refcnt;
845                         rt = NULL;
846                 }
847         }
848         if (!rt) {
849                 if (create && ifp) {
850                         int e;
851
852                         /*
853                          * If no route is available and create is set,
854                          * we allocate a host route for the destination
855                          * and treat it like an interface route.
856                          * This hack is necessary for a neighbor which can't
857                          * be covered by our own prefix.
858                          */
859                         struct ifaddr *ifa =
860                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
861                         if (ifa == NULL)
862                                 return (NULL);
863
864                         /*
865                          * Create a new route.  RTF_LLINFO is necessary
866                          * to create a Neighbor Cache entry for the
867                          * destination in nd6_rtrequest which will be
868                          * called in rtrequest via ifa->ifa_rtrequest.
869                          */
870                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
871                                            ifa->ifa_addr,
872                                            (struct sockaddr *)&all1_sa,
873                                            (ifa->ifa_flags |
874                                             RTF_HOST | RTF_LLINFO) &
875                                            ~RTF_CLONING,
876                                            &rt)) != 0)
877                                 log(LOG_ERR,
878                                     "nd6_lookup: failed to add route for a "
879                                     "neighbor(%s), errno=%d\n",
880                                     ip6_sprintf(addr6), e);
881                         if (rt == NULL)
882                                 return (NULL);
883                         if (rt->rt_llinfo) {
884                                 struct llinfo_nd6 *ln =
885                                         (struct llinfo_nd6 *)rt->rt_llinfo;
886                                 ln->ln_state = ND6_LLINFO_NOSTATE;
887                         }
888                 } else
889                         return (NULL);
890         }
891         rt->rt_refcnt--;
892         /*
893          * Validation for the entry.
894          * Note that the check for rt_llinfo is necessary because a cloned
895          * route from a parent route that has the L flag (e.g. the default
896          * route to a p2p interface) may have the flag, too, while the
897          * destination is not actually a neighbor.
898          * XXX: we can't use rt->rt_ifp to check for the interface, since
899          *      it might be the loopback interface if the entry is for our
900          *      own address on a non-loopback interface. Instead, we should
901          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
902          *      interface.
903          */
904         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
905             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
906             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
907                 if (create) {
908                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
909                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
910                         /* xxx more logs... kazu */
911                 }
912                 return (NULL);
913         }
914         return (rt);
915 }
916
917 /*
918  * Detect if a given IPv6 address identifies a neighbor on a given link.
919  * XXX: should take care of the destination of a p2p link?
920  */
921 int
922 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
923 {
924         struct ifaddr_container *ifac;
925         int i;
926
927 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
928 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
929
930         /*
931          * A link-local address is always a neighbor.
932          * XXX: we should use the sin6_scope_id field rather than the embedded
933          * interface index.
934          */
935         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
936             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
937                 return (1);
938
939         /*
940          * If the address matches one of our addresses,
941          * it should be a neighbor.
942          */
943         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
944                 struct ifaddr *ifa = ifac->ifa;
945
946                 if (ifa->ifa_addr->sa_family != AF_INET6)
947                         next: continue;
948
949                 for (i = 0; i < 4; i++) {
950                         if ((IFADDR6(ifa).s6_addr32[i] ^
951                              addr->sin6_addr.s6_addr32[i]) &
952                             IFMASK6(ifa).s6_addr32[i])
953                                 goto next;
954                 }
955                 return (1);
956         }
957
958         /*
959          * Even if the address matches none of our addresses, it might be
960          * in the neighbor cache.
961          */
962         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
963                 return (1);
964
965         return (0);
966 #undef IFADDR6
967 #undef IFMASK6
968 }
969
970 /*
971  * Free an nd6 llinfo entry.
972  */
973 struct llinfo_nd6 *
974 nd6_free(struct rtentry *rt)
975 {
976         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
977         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
978         struct nd_defrouter *dr;
979
980         /*
981          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
982          * even though it is not harmful, it was not really necessary.
983          */
984
985         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
986                 mtx_lock(&nd6_mtx);
987                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
988                                       rt->rt_ifp);
989
990                 if (ln->ln_router || dr) {
991                         /*
992                          * rt6_flush must be called whether or not the neighbor
993                          * is in the Default Router List.
994                          * See a corresponding comment in nd6_na_input().
995                          */
996                         rt6_flush(&in6, rt->rt_ifp);
997                 }
998
999                 if (dr) {
1000                         /*
1001                          * Unreachablity of a router might affect the default
1002                          * router selection and on-link detection of advertised
1003                          * prefixes.
1004                          */
1005
1006                         /*
1007                          * Temporarily fake the state to choose a new default
1008                          * router and to perform on-link determination of
1009                          * prefixes correctly.
1010                          * Below the state will be set correctly,
1011                          * or the entry itself will be deleted.
1012                          */
1013                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1014
1015                         /*
1016                          * Since defrouter_select() does not affect the
1017                          * on-link determination and MIP6 needs the check
1018                          * before the default router selection, we perform
1019                          * the check now.
1020                          */
1021                         pfxlist_onlink_check();
1022
1023                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1024                                 /*
1025                                  * It is used as the current default router,
1026                                  * so we have to move it to the end of the
1027                                  * list and choose a new one.
1028                                  * XXX: it is not very efficient if this is
1029                                  *      the only router.
1030                                  */
1031                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1032                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1033
1034                                 defrouter_select();
1035                         }
1036                 }
1037                 mtx_unlock(&nd6_mtx);
1038         }
1039
1040         /*
1041          * Before deleting the entry, remember the next entry as the
1042          * return value.  We need this because pfxlist_onlink_check() above
1043          * might have freed other entries (particularly the old next entry) as
1044          * a side effect (XXX).
1045          */
1046         next = ln->ln_next;
1047
1048         /*
1049          * Detach the route from the routing tree and the list of neighbor
1050          * caches, and disable the route entry not to be used in already
1051          * cached routes.
1052          */
1053         rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1054
1055         return (next);
1056 }
1057
1058 /*
1059  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1060  *
1061  * XXX cost-effective metods?
1062  */
1063 void
1064 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1065 {
1066         struct llinfo_nd6 *ln;
1067
1068         /*
1069          * If the caller specified "rt", use that.  Otherwise, resolve the
1070          * routing table by supplied "dst6".
1071          */
1072         if (!rt) {
1073                 if (!dst6)
1074                         return;
1075                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1076                         return;
1077         }
1078
1079         if ((rt->rt_flags & RTF_GATEWAY) ||
1080             !(rt->rt_flags & RTF_LLINFO) ||
1081             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1082             rt->rt_gateway->sa_family != AF_LINK) {
1083                 /* This is not a host route. */
1084                 return;
1085         }
1086
1087         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1088         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1089                 return;
1090
1091         /*
1092          * if we get upper-layer reachability confirmation many times,
1093          * it is possible we have false information.
1094          */
1095         if (!force) {
1096                 ln->ln_byhint++;
1097                 if (ln->ln_byhint > nd6_maxnudhint)
1098                         return;
1099         }
1100
1101         ln->ln_state = ND6_LLINFO_REACHABLE;
1102         if (ln->ln_expire)
1103                 ln->ln_expire = time_uptime +
1104                         ND_IFINFO(rt->rt_ifp)->reachable;
1105 }
1106
1107 void
1108 nd6_rtrequest(int req, struct rtentry *rt)
1109 {
1110         struct sockaddr *gate = rt->rt_gateway;
1111         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1112         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1113         struct ifnet *ifp = rt->rt_ifp;
1114         struct ifaddr *ifa;
1115
1116         if ((rt->rt_flags & RTF_GATEWAY))
1117                 return;
1118
1119         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1120                 /*
1121                  * This is probably an interface direct route for a link
1122                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1123                  * We do not need special treatment below for such a route.
1124                  * Moreover, the RTF_LLINFO flag which would be set below
1125                  * would annoy the ndp(8) command.
1126                  */
1127                 return;
1128         }
1129
1130         if (req == RTM_RESOLVE &&
1131             (nd6_need_cache(ifp) == 0 || /* stf case */
1132              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1133                 /*
1134                  * FreeBSD and BSD/OS often make a cloned host route based
1135                  * on a less-specific route (e.g. the default route).
1136                  * If the less specific route does not have a "gateway"
1137                  * (this is the case when the route just goes to a p2p or an
1138                  * stf interface), we'll mistakenly make a neighbor cache for
1139                  * the host route, and will see strange neighbor solicitation
1140                  * for the corresponding destination.  In order to avoid the
1141                  * confusion, we check if the destination of the route is
1142                  * a neighbor in terms of neighbor discovery, and stop the
1143                  * process if not.  Additionally, we remove the LLINFO flag
1144                  * so that ndp(8) will not try to get the neighbor information
1145                  * of the destination.
1146                  */
1147                 rt->rt_flags &= ~RTF_LLINFO;
1148                 return;
1149         }
1150
1151         switch (req) {
1152         case RTM_ADD:
1153                 /*
1154                  * There is no backward compatibility :)
1155                  *
1156                  * if (!(rt->rt_flags & RTF_HOST) &&
1157                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1158                  *         rt->rt_flags |= RTF_CLONING;
1159                  */
1160                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1161                         /*
1162                          * Case 1: This route should come from
1163                          * a route to interface.  RTF_LLINFO flag is set
1164                          * for a host route whose destination should be
1165                          * treated as on-link.
1166                          */
1167                         rt_setgate(rt, rt_key(rt),
1168                                    (struct sockaddr *)&null_sdl,
1169                                    RTL_DONTREPORT);
1170                         gate = rt->rt_gateway;
1171                         SDL(gate)->sdl_type = ifp->if_type;
1172                         SDL(gate)->sdl_index = ifp->if_index;
1173                         if (ln)
1174                                 ln->ln_expire = time_uptime;
1175 #if 1
1176                         if (ln && ln->ln_expire == 0) {
1177                                 /* kludge for desktops */
1178 #if 0
1179                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1180                                        "treat it as 1\n");
1181 #endif
1182                                 ln->ln_expire = 1;
1183                         }
1184 #endif
1185                         if ((rt->rt_flags & RTF_CLONING))
1186                                 break;
1187                 }
1188                 /*
1189                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1190                  * We don't do that here since llinfo is not ready yet.
1191                  *
1192                  * There are also couple of other things to be discussed:
1193                  * - unsolicited NA code needs improvement beforehand
1194                  * - RFC2461 says we MAY send multicast unsolicited NA
1195                  *   (7.2.6 paragraph 4), however, it also says that we
1196                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1197                  *   we don't have anything like it right now.
1198                  *   note that the mechanism needs a mutual agreement
1199                  *   between proxies, which means that we need to implement
1200                  *   a new protocol, or a new kludge.
1201                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1202                  *   we need to check ip6forwarding before sending it.
1203                  *   (or should we allow proxy ND configuration only for
1204                  *   routers?  there's no mention about proxy ND from hosts)
1205                  */
1206 #if 0
1207                 /* XXX it does not work */
1208                 if (rt->rt_flags & RTF_ANNOUNCE)
1209                         nd6_na_output(ifp,
1210                               &SIN6(rt_key(rt))->sin6_addr,
1211                               &SIN6(rt_key(rt))->sin6_addr,
1212                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1213                               1, NULL);
1214 #endif
1215                 /* FALLTHROUGH */
1216         case RTM_RESOLVE:
1217                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1218                         /*
1219                          * Address resolution isn't necessary for a point to
1220                          * point link, so we can skip this test for a p2p link.
1221                          */
1222                         if (gate->sa_family != AF_LINK ||
1223                             gate->sa_len < sizeof(null_sdl)) {
1224                                 log(LOG_DEBUG,
1225                                     "nd6_rtrequest: bad gateway value: %s\n",
1226                                     if_name(ifp));
1227                                 break;
1228                         }
1229                         SDL(gate)->sdl_type = ifp->if_type;
1230                         SDL(gate)->sdl_index = ifp->if_index;
1231                 }
1232                 if (ln != NULL)
1233                         break;  /* This happens on a route change */
1234                 /*
1235                  * Case 2: This route may come from cloning, or a manual route
1236                  * add with a LL address.
1237                  */
1238                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1239                 rt->rt_llinfo = (caddr_t)ln;
1240                 if (!ln) {
1241                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1242                         break;
1243                 }
1244                 nd6_inuse++;
1245                 nd6_allocated++;
1246                 bzero(ln, sizeof(*ln));
1247                 ln->ln_rt = rt;
1248                 /* this is required for "ndp" command. - shin */
1249                 if (req == RTM_ADD) {
1250                         /*
1251                          * gate should have some valid AF_LINK entry,
1252                          * and ln->ln_expire should have some lifetime
1253                          * which is specified by ndp command.
1254                          */
1255                         ln->ln_state = ND6_LLINFO_REACHABLE;
1256                         ln->ln_byhint = 0;
1257                 } else {
1258                         /*
1259                          * When req == RTM_RESOLVE, rt is created and
1260                          * initialized in rtrequest(), so rt_expire is 0.
1261                          */
1262                         ln->ln_state = ND6_LLINFO_NOSTATE;
1263                         ln->ln_expire = time_uptime;
1264                 }
1265                 rt->rt_flags |= RTF_LLINFO;
1266                 ln->ln_next = llinfo_nd6.ln_next;
1267                 llinfo_nd6.ln_next = ln;
1268                 ln->ln_prev = &llinfo_nd6;
1269                 ln->ln_next->ln_prev = ln;
1270
1271                 /*
1272                  * check if rt_key(rt) is one of my address assigned
1273                  * to the interface.
1274                  */
1275                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1276                                           &SIN6(rt_key(rt))->sin6_addr);
1277                 if (ifa) {
1278                         caddr_t macp = nd6_ifptomac(ifp);
1279                         ln->ln_expire = 0;
1280                         ln->ln_state = ND6_LLINFO_REACHABLE;
1281                         ln->ln_byhint = 0;
1282                         if (macp) {
1283                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1284                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1285                         }
1286                         if (nd6_useloopback) {
1287                                 rt->rt_ifp = &loif[0];  /* XXX */
1288                                 /*
1289                                  * Make sure rt_ifa be equal to the ifaddr
1290                                  * corresponding to the address.
1291                                  * We need this because when we refer
1292                                  * rt_ifa->ia6_flags in ip6_input, we assume
1293                                  * that the rt_ifa points to the address instead
1294                                  * of the loopback address.
1295                                  */
1296                                 if (ifa != rt->rt_ifa) {
1297                                         IFAFREE(rt->rt_ifa);
1298                                         IFAREF(ifa);
1299                                         rt->rt_ifa = ifa;
1300                                 }
1301                         }
1302                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1303                         ln->ln_expire = 0;
1304                         ln->ln_state = ND6_LLINFO_REACHABLE;
1305                         ln->ln_byhint = 0;
1306
1307                         /* join solicited node multicast for proxy ND */
1308                         if (ifp->if_flags & IFF_MULTICAST) {
1309                                 struct in6_addr llsol;
1310                                 int error;
1311
1312                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1313                                 llsol.s6_addr16[0] = htons(0xff02);
1314                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1315                                 llsol.s6_addr32[1] = 0;
1316                                 llsol.s6_addr32[2] = htonl(1);
1317                                 llsol.s6_addr8[12] = 0xff;
1318
1319                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1320                                         nd6log((LOG_ERR, "%s: failed to join "
1321                                             "%s (errno=%d)\n", if_name(ifp),
1322                                             ip6_sprintf(&llsol), error));
1323                                 }
1324                         }
1325                 }
1326                 break;
1327
1328         case RTM_DELETE:
1329                 if (!ln)
1330                         break;
1331                 /* leave from solicited node multicast for proxy ND */
1332                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1333                     (ifp->if_flags & IFF_MULTICAST)) {
1334                         struct in6_addr llsol;
1335                         struct in6_multi *in6m;
1336
1337                         llsol = SIN6(rt_key(rt))->sin6_addr;
1338                         llsol.s6_addr16[0] = htons(0xff02);
1339                         llsol.s6_addr16[1] = htons(ifp->if_index);
1340                         llsol.s6_addr32[1] = 0;
1341                         llsol.s6_addr32[2] = htonl(1);
1342                         llsol.s6_addr8[12] = 0xff;
1343
1344                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1345                         if (in6m)
1346                                 in6_delmulti(in6m);
1347                 }
1348                 nd6_inuse--;
1349                 ln->ln_next->ln_prev = ln->ln_prev;
1350                 ln->ln_prev->ln_next = ln->ln_next;
1351                 ln->ln_prev = NULL;
1352                 rt->rt_llinfo = 0;
1353                 rt->rt_flags &= ~RTF_LLINFO;
1354                 if (ln->ln_hold)
1355                         m_freem(ln->ln_hold);
1356                 Free((caddr_t)ln);
1357         }
1358 }
1359
1360 int
1361 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1362 {
1363         struct in6_drlist *drl = (struct in6_drlist *)data;
1364         struct in6_prlist *prl = (struct in6_prlist *)data;
1365         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1366         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1367         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1368         struct nd_defrouter *dr, any;
1369         struct nd_prefix *pr;
1370         struct rtentry *rt;
1371         int i = 0, error = 0;
1372
1373         switch (cmd) {
1374         case SIOCGDRLST_IN6:
1375                 /*
1376                  * obsolete API, use sysctl under net.inet6.icmp6
1377                  */
1378                 bzero(drl, sizeof(*drl));
1379                 mtx_lock(&nd6_mtx);
1380                 dr = TAILQ_FIRST(&nd_defrouter);
1381                 while (dr && i < DRLSTSIZ) {
1382                         drl->defrouter[i].rtaddr = dr->rtaddr;
1383                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1384                                 /* XXX: need to this hack for KAME stack */
1385                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1386                         } else
1387                                 log(LOG_ERR,
1388                                     "default router list contains a "
1389                                     "non-linklocal address(%s)\n",
1390                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1391
1392                         drl->defrouter[i].flags = dr->flags;
1393                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1394                         drl->defrouter[i].expire = dr->expire;
1395                         drl->defrouter[i].if_index = dr->ifp->if_index;
1396                         i++;
1397                         dr = TAILQ_NEXT(dr, dr_entry);
1398                 }
1399                 mtx_unlock(&nd6_mtx);
1400                 break;
1401         case SIOCGPRLST_IN6:
1402                 /*
1403                  * obsolete API, use sysctl under net.inet6.icmp6
1404                  */
1405                 /*
1406                  * XXX meaning of fields, especialy "raflags", is very
1407                  * differnet between RA prefix list and RR/static prefix list.
1408                  * how about separating ioctls into two?
1409                  */
1410                 bzero(prl, sizeof(*prl));
1411                 mtx_lock(&nd6_mtx);
1412                 pr = nd_prefix.lh_first;
1413                 while (pr && i < PRLSTSIZ) {
1414                         struct nd_pfxrouter *pfr;
1415                         int j;
1416
1417                         in6_embedscope(&prl->prefix[i].prefix,
1418                             &pr->ndpr_prefix, NULL, NULL);
1419                         prl->prefix[i].raflags = pr->ndpr_raf;
1420                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1421                         prl->prefix[i].vltime = pr->ndpr_vltime;
1422                         prl->prefix[i].pltime = pr->ndpr_pltime;
1423                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1424                         prl->prefix[i].expire = pr->ndpr_expire;
1425
1426                         pfr = pr->ndpr_advrtrs.lh_first;
1427                         j = 0;
1428                         while (pfr) {
1429                                 if (j < DRLSTSIZ) {
1430 #define RTRADDR prl->prefix[i].advrtr[j]
1431                                         RTRADDR = pfr->router->rtaddr;
1432                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1433                                                 /* XXX: hack for KAME */
1434                                                 RTRADDR.s6_addr16[1] = 0;
1435                                         } else
1436                                                 log(LOG_ERR,
1437                                                     "a router(%s) advertises "
1438                                                     "a prefix with "
1439                                                     "non-link local address\n",
1440                                                     ip6_sprintf(&RTRADDR));
1441 #undef RTRADDR
1442                                 }
1443                                 j++;
1444                                 pfr = pfr->pfr_next;
1445                         }
1446                         prl->prefix[i].advrtrs = j;
1447                         prl->prefix[i].origin = PR_ORIG_RA;
1448
1449                         i++;
1450                         pr = pr->ndpr_next;
1451                 }
1452                 mtx_unlock(&nd6_mtx);
1453
1454                 break;
1455         case OSIOCGIFINFO_IN6:
1456                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1457                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1458                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1459                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1460                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1461                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1462                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1463                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1464                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1465                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1466                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1467                 break;
1468         case SIOCGIFINFO_IN6:
1469                 ndi->ndi = *ND_IFINFO(ifp);
1470                 break;
1471         case SIOCSIFINFO_FLAGS:
1472                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1473                 break;
1474         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1475                 /* flush default router list */
1476                 /*
1477                  * xxx sumikawa: should not delete route if default
1478                  * route equals to the top of default router list
1479                  */
1480                 bzero(&any, sizeof(any));
1481                 defrouter_delreq(&any, 0);
1482                 defrouter_select();
1483                 /* xxx sumikawa: flush prefix list */
1484                 break;
1485         case SIOCSPFXFLUSH_IN6:
1486             {
1487                 /* flush all the prefix advertised by routers */
1488                 struct nd_prefix *pr, *next;
1489
1490                 mtx_lock(&nd6_mtx);
1491                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1492                         struct in6_ifaddr *ia, *ia_next;
1493
1494                         next = pr->ndpr_next;
1495
1496                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1497                                 continue; /* XXX */
1498
1499                         /* do we really have to remove addresses as well? */
1500                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1501                                 /* ia might be removed.  keep the next ptr. */
1502                                 ia_next = ia->ia_next;
1503
1504                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1505                                         continue;
1506
1507                                 if (ia->ia6_ndpr == pr)
1508                                         in6_purgeaddr(&ia->ia_ifa);
1509                         }
1510                         prelist_remove(pr);
1511                 }
1512                 mtx_unlock(&nd6_mtx);
1513                 break;
1514             }
1515         case SIOCSRTRFLUSH_IN6:
1516             {
1517                 /* flush all the default routers */
1518                 struct nd_defrouter *dr, *next;
1519
1520                 mtx_lock(&nd6_mtx);
1521                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1522                         /*
1523                          * The first entry of the list may be stored in
1524                          * the routing table, so we'll delete it later.
1525                          */
1526                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1527                                 next = TAILQ_NEXT(dr, dr_entry);
1528                                 defrtrlist_del(dr);
1529                         }
1530                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1531                 }
1532                 mtx_unlock(&nd6_mtx);
1533                 break;
1534             }
1535         case SIOCGNBRINFO_IN6:
1536             {
1537                 struct llinfo_nd6 *ln;
1538                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1539
1540                 /*
1541                  * XXX: KAME specific hack for scoped addresses
1542                  *      XXXX: for other scopes than link-local?
1543                  */
1544                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1545                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1546                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1547
1548                         if (*idp == 0)
1549                                 *idp = htons(ifp->if_index);
1550                 }
1551
1552                 mtx_lock(&nd6_mtx);
1553                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1554                         error = EINVAL;
1555                         mtx_unlock(&nd6_mtx);
1556                         break;
1557                 }
1558                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1559                 nbi->state = ln->ln_state;
1560                 nbi->asked = ln->ln_asked;
1561                 nbi->isrouter = ln->ln_router;
1562                 nbi->expire = ln->ln_expire;
1563                 mtx_unlock(&nd6_mtx);
1564
1565                 break;
1566             }
1567         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1568                 ndif->ifindex = nd6_defifindex;
1569                 break;
1570         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1571                 return (nd6_setdefaultiface(ndif->ifindex));
1572                 break;
1573         }
1574         return (error);
1575 }
1576
1577 /*
1578  * Create neighbor cache entry and cache link-layer address,
1579  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1580  */
1581 struct rtentry *
1582 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1583                  int lladdrlen,
1584                  int type,      /* ICMP6 type */
1585                  int code       /* type dependent information */)
1586 {
1587         struct rtentry *rt = NULL;
1588         struct llinfo_nd6 *ln = NULL;
1589         int is_newentry;
1590         struct sockaddr_dl *sdl = NULL;
1591         int do_update;
1592         int olladdr;
1593         int llchange;
1594         int newstate = 0;
1595
1596         if (!ifp)
1597                 panic("ifp == NULL in nd6_cache_lladdr");
1598         if (!from)
1599                 panic("from == NULL in nd6_cache_lladdr");
1600
1601         /* nothing must be updated for unspecified address */
1602         if (IN6_IS_ADDR_UNSPECIFIED(from))
1603                 return NULL;
1604
1605         /*
1606          * Validation about ifp->if_addrlen and lladdrlen must be done in
1607          * the caller.
1608          *
1609          * XXX If the link does not have link-layer adderss, what should
1610          * we do? (ifp->if_addrlen == 0)
1611          * Spec says nothing in sections for RA, RS and NA.  There's small
1612          * description on it in NS section (RFC 2461 7.2.3).
1613          */
1614
1615         rt = nd6_lookup(from, 0, ifp);
1616         if (!rt) {
1617 #if 0
1618                 /* nothing must be done if there's no lladdr */
1619                 if (!lladdr || !lladdrlen)
1620                         return NULL;
1621 #endif
1622
1623                 rt = nd6_lookup(from, 1, ifp);
1624                 is_newentry = 1;
1625         } else {
1626                 /* do nothing if static ndp is set */
1627                 if (rt->rt_flags & RTF_STATIC)
1628                         return NULL;
1629                 is_newentry = 0;
1630         }
1631
1632         if (!rt)
1633                 return NULL;
1634         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1635 fail:
1636                 nd6_free(rt);
1637                 return NULL;
1638         }
1639         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1640         if (!ln)
1641                 goto fail;
1642         if (!rt->rt_gateway)
1643                 goto fail;
1644         if (rt->rt_gateway->sa_family != AF_LINK)
1645                 goto fail;
1646         sdl = SDL(rt->rt_gateway);
1647
1648         olladdr = (sdl->sdl_alen) ? 1 : 0;
1649         if (olladdr && lladdr) {
1650                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1651                         llchange = 1;
1652                 else
1653                         llchange = 0;
1654         } else
1655                 llchange = 0;
1656
1657         /*
1658          * newentry olladdr  lladdr  llchange   (*=record)
1659          *      0       n       n       --      (1)
1660          *      0       y       n       --      (2)
1661          *      0       n       y       --      (3) * STALE
1662          *      0       y       y       n       (4) *
1663          *      0       y       y       y       (5) * STALE
1664          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1665          *      1       --      y       --      (7) * STALE
1666          */
1667
1668         if (lladdr) {           /* (3-5) and (7) */
1669                 /*
1670                  * Record source link-layer address
1671                  * XXX is it dependent to ifp->if_type?
1672                  */
1673                 sdl->sdl_alen = ifp->if_addrlen;
1674                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1675         }
1676
1677         if (!is_newentry) {
1678                 if ((!olladdr && lladdr)                /* (3) */
1679                  || (olladdr && lladdr && llchange)) {  /* (5) */
1680                         do_update = 1;
1681                         newstate = ND6_LLINFO_STALE;
1682                 } else                                  /* (1-2,4) */
1683                         do_update = 0;
1684         } else {
1685                 do_update = 1;
1686                 if (!lladdr)                            /* (6) */
1687                         newstate = ND6_LLINFO_NOSTATE;
1688                 else                                    /* (7) */
1689                         newstate = ND6_LLINFO_STALE;
1690         }
1691
1692         if (do_update) {
1693                 /*
1694                  * Update the state of the neighbor cache.
1695                  */
1696                 ln->ln_state = newstate;
1697
1698                 if (ln->ln_state == ND6_LLINFO_STALE) {
1699                         /*
1700                          * XXX: since nd6_output() below will cause
1701                          * state tansition to DELAY and reset the timer,
1702                          * we must set the timer now, although it is actually
1703                          * meaningless.
1704                          */
1705                         ln->ln_expire = time_uptime + nd6_gctimer;
1706
1707                         if (ln->ln_hold) {
1708                                 /*
1709                                  * we assume ifp is not a p2p here, so just
1710                                  * set the 2nd argument as the 1st one.
1711                                  */
1712                                 nd6_output(ifp, ifp, ln->ln_hold,
1713                                            (struct sockaddr_in6 *)rt_key(rt),
1714                                            rt);
1715                                 ln->ln_hold = NULL;
1716                         }
1717                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1718                         /* probe right away */
1719                         ln->ln_expire = time_uptime;
1720                 }
1721         }
1722
1723         /*
1724          * ICMP6 type dependent behavior.
1725          *
1726          * NS: clear IsRouter if new entry
1727          * RS: clear IsRouter
1728          * RA: set IsRouter if there's lladdr
1729          * redir: clear IsRouter if new entry
1730          *
1731          * RA case, (1):
1732          * The spec says that we must set IsRouter in the following cases:
1733          * - If lladdr exist, set IsRouter.  This means (1-5).
1734          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1735          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1736          * A quetion arises for (1) case.  (1) case has no lladdr in the
1737          * neighbor cache, this is similar to (6).
1738          * This case is rare but we figured that we MUST NOT set IsRouter.
1739          *
1740          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1741          *                                                      D R
1742          *      0       n       n       --      (1)     c   ?     s
1743          *      0       y       n       --      (2)     c   s     s
1744          *      0       n       y       --      (3)     c   s     s
1745          *      0       y       y       n       (4)     c   s     s
1746          *      0       y       y       y       (5)     c   s     s
1747          *      1       --      n       --      (6) c   c       c s
1748          *      1       --      y       --      (7) c   c   s   c s
1749          *
1750          *                                      (c=clear s=set)
1751          */
1752         switch (type & 0xff) {
1753         case ND_NEIGHBOR_SOLICIT:
1754                 /*
1755                  * New entry must have is_router flag cleared.
1756                  */
1757                 if (is_newentry)        /* (6-7) */
1758                         ln->ln_router = 0;
1759                 break;
1760         case ND_REDIRECT:
1761                 /*
1762                  * If the icmp is a redirect to a better router, always set the
1763                  * is_router flag. Otherwise, if the entry is newly created,
1764                  * clear the flag. [RFC 2461, sec 8.3]
1765                  */
1766                 if (code == ND_REDIRECT_ROUTER)
1767                         ln->ln_router = 1;
1768                 else if (is_newentry) /* (6-7) */
1769                         ln->ln_router = 0;
1770                 break;
1771         case ND_ROUTER_SOLICIT:
1772                 /*
1773                  * is_router flag must always be cleared.
1774                  */
1775                 ln->ln_router = 0;
1776                 break;
1777         case ND_ROUTER_ADVERT:
1778                 /*
1779                  * Mark an entry with lladdr as a router.
1780                  */
1781                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1782                  || (is_newentry && lladdr)) {                  /* (7) */
1783                         ln->ln_router = 1;
1784                 }
1785                 break;
1786         }
1787
1788         /*
1789          * When the link-layer address of a router changes, select the
1790          * best router again.  In particular, when the neighbor entry is newly
1791          * created, it might affect the selection policy.
1792          * Question: can we restrict the first condition to the "is_newentry"
1793          * case?
1794          * XXX: when we hear an RA from a new router with the link-layer
1795          * address option, defrouter_select() is called twice, since
1796          * defrtrlist_update called the function as well.  However, I believe
1797          * we can compromise the overhead, since it only happens the first
1798          * time.
1799          * XXX: although defrouter_select() should not have a bad effect
1800          * for those are not autoconfigured hosts, we explicitly avoid such
1801          * cases for safety.
1802          */
1803         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1804                 defrouter_select();
1805
1806         return rt;
1807 }
1808
1809 static void
1810 nd6_slowtimo(void *arg __unused)
1811 {
1812         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1813
1814         KASSERT(mycpuid == 0, ("not on cpu0"));
1815         crit_enter();
1816         if (lmsg->ms_flags & MSGF_DONE)
1817                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1818         crit_exit();
1819 }
1820
1821 static void
1822 nd6_slowtimo_dispatch(netmsg_t nmsg)
1823 {
1824         const struct ifnet_array *arr;
1825         struct nd_ifinfo *nd6if;
1826         int i;
1827
1828         ASSERT_IN_NETISR(0);
1829
1830         crit_enter();
1831         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1832         crit_exit();
1833
1834         arr = ifnet_array_get();
1835
1836         mtx_lock(&nd6_mtx);
1837         for (i = 0; i < arr->ifnet_count; ++i) {
1838                 struct ifnet *ifp = arr->ifnet_arr[i];
1839
1840                 if (ifp->if_afdata[AF_INET6] == NULL)
1841                         continue;
1842                 nd6if = ND_IFINFO(ifp);
1843                 if (nd6if->basereachable && /* already initialized */
1844                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1845                         /*
1846                          * Since reachable time rarely changes by router
1847                          * advertisements, we SHOULD insure that a new random
1848                          * value gets recomputed at least once every few hours.
1849                          * (RFC 2461, 6.3.4)
1850                          */
1851                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1852                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1853                 }
1854         }
1855         mtx_unlock(&nd6_mtx);
1856
1857         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1858             nd6_slowtimo, NULL);
1859 }
1860
1861 #define gotoerr(e) { error = (e); goto bad;}
1862
1863 int
1864 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1865            struct sockaddr_in6 *dst, struct rtentry *rt)
1866 {
1867         struct llinfo_nd6 *ln = NULL;
1868         int error = 0;
1869
1870         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1871                 goto sendpkt;
1872
1873         if (nd6_need_cache(ifp) == 0)
1874                 goto sendpkt;
1875
1876         /*
1877          * next hop determination.  This routine is derived from ether_outpout.
1878          */
1879         if (rt != NULL) {
1880                 if (!(rt->rt_flags & RTF_UP)) {
1881                         rt = rtlookup((struct sockaddr *)dst);
1882                         if (rt == NULL)
1883                                 gotoerr(EHOSTUNREACH);
1884                         rt->rt_refcnt--;
1885                         if (rt->rt_ifp != ifp) {
1886                                 /* XXX: loop care? */
1887                                 return nd6_output(ifp, origifp, m, dst, rt);
1888                         }
1889                 }
1890                 if (rt->rt_flags & RTF_GATEWAY) {
1891                         struct sockaddr_in6 *gw6;
1892
1893                         /*
1894                          * We skip link-layer address resolution and NUD
1895                          * if the gateway is not a neighbor from ND point
1896                          * of view, regardless of the value of nd_ifinfo.flags.
1897                          * The second condition is a bit tricky; we skip
1898                          * if the gateway is our own address, which is
1899                          * sometimes used to install a route to a p2p link.
1900                          */
1901                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1902                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1903                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1904                                 /*
1905                                  * We allow this kind of tricky route only
1906                                  * when the outgoing interface is p2p.
1907                                  * XXX: we may need a more generic rule here.
1908                                  */
1909                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1910                                         gotoerr(EHOSTUNREACH);
1911
1912                                 goto sendpkt;
1913                         }
1914
1915                         if (rt->rt_gwroute == NULL) {
1916                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1917                                 if (rt->rt_gwroute == NULL)
1918                                         gotoerr(EHOSTUNREACH);
1919                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1920                                 rtfree(rt->rt_gwroute);
1921                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1922                                 if (rt->rt_gwroute == NULL)
1923                                         gotoerr(EHOSTUNREACH);
1924                         }
1925                 }
1926         }
1927
1928         /*
1929          * Address resolution or Neighbor Unreachability Detection
1930          * for the next hop.
1931          * At this point, the destination of the packet must be a unicast
1932          * or an anycast address(i.e. not a multicast).
1933          */
1934
1935         /* Look up the neighbor cache for the nexthop */
1936         if (rt && (rt->rt_flags & RTF_LLINFO))
1937                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1938         else {
1939                 /*
1940                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1941                  * the condition below is not very efficient.  But we believe
1942                  * it is tolerable, because this should be a rare case.
1943                  */
1944                 if (nd6_is_addr_neighbor(dst, ifp) &&
1945                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1946                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1947         }
1948         if (!ln || !rt) {
1949                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1950                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1951                         log(LOG_DEBUG,
1952                             "nd6_output: can't allocate llinfo for %s "
1953                             "(ln=%p, rt=%p)\n",
1954                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1955                         gotoerr(EIO);   /* XXX: good error? */
1956                 }
1957
1958                 goto sendpkt;   /* send anyway */
1959         }
1960
1961         /* We don't have to do link-layer address resolution on a p2p link. */
1962         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1963             ln->ln_state < ND6_LLINFO_REACHABLE) {
1964                 ln->ln_state = ND6_LLINFO_STALE;
1965                 ln->ln_expire = time_uptime + nd6_gctimer;
1966         }
1967
1968         /*
1969          * The first time we send a packet to a neighbor whose entry is
1970          * STALE, we have to change the state to DELAY and a sets a timer to
1971          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1972          * neighbor unreachability detection on expiration.
1973          * (RFC 2461 7.3.3)
1974          */
1975         if (ln->ln_state == ND6_LLINFO_STALE) {
1976                 ln->ln_asked = 0;
1977                 ln->ln_state = ND6_LLINFO_DELAY;
1978                 ln->ln_expire = time_uptime + nd6_delay;
1979         }
1980
1981         /*
1982          * If the neighbor cache entry has a state other than INCOMPLETE
1983          * (i.e. its link-layer address is already resolved), just
1984          * send the packet.
1985          */
1986         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1987                 goto sendpkt;
1988
1989         /*
1990          * There is a neighbor cache entry, but no ethernet address
1991          * response yet.  Replace the held mbuf (if any) with this
1992          * latest one.
1993          *
1994          * This code conforms to the rate-limiting rule described in Section
1995          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1996          * an NS below.
1997          */
1998         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1999                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2000         if (ln->ln_hold)
2001                 m_freem(ln->ln_hold);
2002         ln->ln_hold = m;
2003         if (ln->ln_expire) {
2004                 if (ln->ln_asked < nd6_mmaxtries &&
2005                     ln->ln_expire < time_uptime) {
2006                         ln->ln_asked++;
2007                         ln->ln_expire = time_uptime +
2008                                 ND_IFINFO(ifp)->retrans / 1000;
2009                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2010                 }
2011         }
2012         return (0);
2013
2014 sendpkt:
2015         if (ifp->if_flags & IFF_LOOPBACK)
2016                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
2017         else
2018                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
2019         return (error);
2020
2021 bad:
2022         m_freem(m);
2023         return (error);
2024 }
2025 #undef gotoerr
2026
2027 int
2028 nd6_need_cache(struct ifnet *ifp)
2029 {
2030         /*
2031          * XXX: we currently do not make neighbor cache on any interface
2032          * other than Ethernet and GIF.
2033          *
2034          * RFC2893 says:
2035          * - unidirectional tunnels needs no ND
2036          */
2037         switch (ifp->if_type) {
2038         case IFT_ETHER:
2039         case IFT_IEEE1394:
2040 #ifdef IFT_L2VLAN
2041         case IFT_L2VLAN:
2042 #endif
2043 #ifdef IFT_IEEE80211
2044         case IFT_IEEE80211:
2045 #endif
2046 #ifdef IFT_CARP
2047         case IFT_CARP:
2048 #endif
2049         case IFT_GIF:           /* XXX need more cases? */
2050                 return (1);
2051         default:
2052                 return (0);
2053         }
2054 }
2055
2056 int
2057 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2058                 struct sockaddr *dst, u_char *desten)
2059 {
2060         struct sockaddr_dl *sdl;
2061         struct rtentry *rt;
2062
2063
2064         if (m->m_flags & M_MCAST) {
2065                 switch (ifp->if_type) {
2066                 case IFT_ETHER:
2067 #ifdef IFT_L2VLAN
2068         case IFT_L2VLAN:
2069 #endif
2070 #ifdef IFT_IEEE80211
2071                 case IFT_IEEE80211:
2072 #endif
2073                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2074                                                  desten);
2075                         return (1);
2076                 case IFT_IEEE1394:
2077                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2078                         return (1);
2079                 default:
2080                         m_freem(m);
2081                         return (0);
2082                 }
2083         }
2084         if (rt0 == NULL) {
2085                 /* this could happen, if we could not allocate memory */
2086                 m_freem(m);
2087                 return (0);
2088         }
2089         if (rt_llroute(dst, rt0, &rt) != 0) {
2090                 m_freem(m);
2091                 return (0);
2092         }
2093         if (rt->rt_gateway->sa_family != AF_LINK) {
2094                 kprintf("nd6_storelladdr: something odd happens\n");
2095                 m_freem(m);
2096                 return (0);
2097         }
2098         sdl = SDL(rt->rt_gateway);
2099         if (sdl->sdl_alen == 0) {
2100                 /* this should be impossible, but we bark here for debugging */
2101                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2102                 m_freem(m);
2103                 return (0);
2104         }
2105
2106         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2107         return (1);
2108 }
2109
2110 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2111 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2112 #ifdef SYSCTL_DECL
2113 SYSCTL_DECL(_net_inet6_icmp6);
2114 #endif
2115 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2116         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2117 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2118         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2119
2120 static int
2121 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2122 {
2123         int error;
2124         char buf[1024];
2125         struct in6_defrouter *d, *de;
2126         struct nd_defrouter *dr;
2127
2128         if (req->newptr)
2129                 return EPERM;
2130         error = 0;
2131
2132         for (dr = TAILQ_FIRST(&nd_defrouter);
2133              dr;
2134              dr = TAILQ_NEXT(dr, dr_entry)) {
2135                 d = (struct in6_defrouter *)buf;
2136                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2137
2138                 if (d + 1 <= de) {
2139                         bzero(d, sizeof(*d));
2140                         d->rtaddr.sin6_family = AF_INET6;
2141                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2142                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2143                             dr->ifp) != 0)
2144                                 log(LOG_ERR,
2145                                     "scope error in "
2146                                     "default router list (%s)\n",
2147                                     ip6_sprintf(&dr->rtaddr));
2148                         d->flags = dr->flags;
2149                         d->rtlifetime = dr->rtlifetime;
2150                         d->expire = dr->expire;
2151                         d->if_index = dr->ifp->if_index;
2152                 } else
2153                         panic("buffer too short");
2154
2155                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2156                 if (error)
2157                         break;
2158         }
2159         return error;
2160 }
2161
2162 static int
2163 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2164 {
2165         int error;
2166         char buf[1024];
2167         struct in6_prefix *p, *pe;
2168         struct nd_prefix *pr;
2169
2170         if (req->newptr)
2171                 return EPERM;
2172         error = 0;
2173
2174         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2175                 u_short advrtrs;
2176                 size_t advance;
2177                 struct sockaddr_in6 *sin6, *s6;
2178                 struct nd_pfxrouter *pfr;
2179
2180                 p = (struct in6_prefix *)buf;
2181                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2182
2183                 if (p + 1 <= pe) {
2184                         bzero(p, sizeof(*p));
2185                         sin6 = (struct sockaddr_in6 *)(p + 1);
2186
2187                         p->prefix = pr->ndpr_prefix;
2188                         if (in6_recoverscope(&p->prefix,
2189                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2190                                 log(LOG_ERR,
2191                                     "scope error in prefix list (%s)\n",
2192                                     ip6_sprintf(&p->prefix.sin6_addr));
2193                         p->raflags = pr->ndpr_raf;
2194                         p->prefixlen = pr->ndpr_plen;
2195                         p->vltime = pr->ndpr_vltime;
2196                         p->pltime = pr->ndpr_pltime;
2197                         p->if_index = pr->ndpr_ifp->if_index;
2198                         p->expire = pr->ndpr_expire;
2199                         p->refcnt = pr->ndpr_refcnt;
2200                         p->flags = pr->ndpr_stateflags;
2201                         p->origin = PR_ORIG_RA;
2202                         advrtrs = 0;
2203                         for (pfr = pr->ndpr_advrtrs.lh_first;
2204                              pfr;
2205                              pfr = pfr->pfr_next) {
2206                                 if ((void *)&sin6[advrtrs + 1] >
2207                                     (void *)pe) {
2208                                         advrtrs++;
2209                                         continue;
2210                                 }
2211                                 s6 = &sin6[advrtrs];
2212                                 bzero(s6, sizeof(*s6));
2213                                 s6->sin6_family = AF_INET6;
2214                                 s6->sin6_len = sizeof(*sin6);
2215                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2216                                                      pfr->router->ifp) != 0)
2217                                         log(LOG_ERR,
2218                                             "scope error in "
2219                                             "prefix list (%s)\n",
2220                                             ip6_sprintf(&pfr->router->rtaddr));
2221                                 advrtrs++;
2222                         }
2223                         p->advrtrs = advrtrs;
2224                 } else
2225                         panic("buffer too short");
2226
2227                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2228                 error = SYSCTL_OUT(req, buf, advance);
2229                 if (error)
2230                         break;
2231         }
2232         return error;
2233 }