3c9525847ebffe0a85cd5a1a22c8f9a1c8ff9454
[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 /*      $DragonFly: src/sys/netinet6/nd6.c,v 1.25 2007/08/27 16:15:42 hasso Exp $       */
3 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
4
5 /*
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * XXX
36  * KAME 970409 note:
37  * BSD/OS version heavily modifies this code, related to llinfo.
38  * Since we don't have BSD/OS version of net/route.c in our hand,
39  * I left the code mostly as it was in 970310.  -- itojun
40  */
41
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/callout.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/time.h>
53 #include <sys/kernel.h>
54 #include <sys/protosw.h>
55 #include <sys/errno.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/sysctl.h>
59 #include <sys/thread2.h>
60
61 #include <net/if.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/if_atm.h>
65 #include <net/route.h>
66
67 #include <netinet/in.h>
68 #include <netinet/if_ether.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_prefix.h>
74 #include <netinet/icmp6.h>
75
76 #include "use_loop.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
112 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
113 static struct sockaddr_in6 all1_sa;
114
115 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
116 static void nd6_slowtimo (void *);
117 static int regen_tmpaddr (struct in6_ifaddr *);
118
119 struct callout nd6_slowtimo_ch;
120 struct callout nd6_timer_ch;
121 extern struct callout in6_tmpaddrtimer_ch;
122
123 void
124 nd6_init(void)
125 {
126         static int nd6_init_done = 0;
127         int i;
128
129         if (nd6_init_done) {
130                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
131                 return;
132         }
133
134         all1_sa.sin6_family = AF_INET6;
135         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
136         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
137                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
138
139         /* initialization of the default router list */
140         TAILQ_INIT(&nd_defrouter);
141
142         nd6_init_done = 1;
143
144         /* start timer */
145         callout_init(&nd6_slowtimo_ch);
146         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
147             nd6_slowtimo, NULL);
148 }
149
150 struct nd_ifinfo *
151 nd6_ifattach(struct ifnet *ifp)
152 {
153         struct nd_ifinfo *nd;
154
155         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
156         bzero(nd, sizeof(*nd));
157
158         nd->initialized = 1;
159
160         nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
161         nd->chlim = IPV6_DEFHLIM;
162         nd->basereachable = REACHABLE_TIME;
163         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
164         nd->retrans = RETRANS_TIMER;
165         nd->receivedra = 0;
166
167         /*
168          * Note that the default value of ip6_accept_rtadv is 0, which means
169          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
170          * here.
171          */
172         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
173
174         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
175         nd6_setmtu0(ifp, nd);
176         return nd;
177 }
178
179 void
180 nd6_ifdetach(struct nd_ifinfo *nd)
181 {
182         kfree(nd, M_IP6NDP);
183 }
184
185 /*
186  * Reset ND level link MTU. This function is called when the physical MTU
187  * changes, which means we might have to adjust the ND level MTU.
188  */
189 void
190 nd6_setmtu(struct ifnet *ifp)
191 {
192         nd6_setmtu0(ifp, ND_IFINFO(ifp));
193 }
194
195 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
196 void
197 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
198 {
199         u_long oldmaxmtu;
200         u_long oldlinkmtu;
201
202         oldmaxmtu = ndi->maxmtu;
203         oldlinkmtu = ndi->linkmtu;
204
205         switch (ifp->if_type) {
206         case IFT_ETHER:
207                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
208                 break;
209         case IFT_ATM:
210                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
211                 break;
212         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
213                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
214                 break;
215 #ifdef IFT_IEEE80211
216         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
217                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
218                 break;
219 #endif
220         default:
221                 ndi->maxmtu = ifp->if_mtu;
222                 break;
223         }
224
225         if (oldmaxmtu != ndi->maxmtu) {
226                 /*
227                  * If the ND level MTU is not set yet, or if the maxmtu
228                  * is reset to a smaller value than the ND level MTU,
229                  * also reset the ND level MTU.
230                  */
231                 if (ndi->linkmtu == 0 ||
232                     ndi->maxmtu < ndi->linkmtu) {
233                         ndi->linkmtu = ndi->maxmtu;
234                         /* also adjust in6_maxmtu if necessary. */
235                         if (oldlinkmtu == 0) {
236                                 /*
237                                  * XXX: the case analysis is grotty, but
238                                  * it is not efficient to call in6_setmaxmtu()
239                                  * here when we are during the initialization
240                                  * procedure.
241                                  */
242                                 if (in6_maxmtu < ndi->linkmtu)
243                                         in6_maxmtu = ndi->linkmtu;
244                         } else
245                                 in6_setmaxmtu();
246                 }
247         }
248 #undef MIN
249 }
250
251 void
252 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
253 {
254         bzero(ndopts, sizeof(*ndopts));
255         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
256         ndopts->nd_opts_last
257                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
258
259         if (icmp6len == 0) {
260                 ndopts->nd_opts_done = 1;
261                 ndopts->nd_opts_search = NULL;
262         }
263 }
264
265 /*
266  * Take one ND option.
267  */
268 struct nd_opt_hdr *
269 nd6_option(union nd_opts *ndopts)
270 {
271         struct nd_opt_hdr *nd_opt;
272         int olen;
273
274         if (!ndopts)
275                 panic("ndopts == NULL in nd6_option");
276         if (!ndopts->nd_opts_last)
277                 panic("uninitialized ndopts in nd6_option");
278         if (!ndopts->nd_opts_search)
279                 return NULL;
280         if (ndopts->nd_opts_done)
281                 return NULL;
282
283         nd_opt = ndopts->nd_opts_search;
284
285         /* make sure nd_opt_len is inside the buffer */
286         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
287                 bzero(ndopts, sizeof(*ndopts));
288                 return NULL;
289         }
290
291         olen = nd_opt->nd_opt_len << 3;
292         if (olen == 0) {
293                 /*
294                  * Message validation requires that all included
295                  * options have a length that is greater than zero.
296                  */
297                 bzero(ndopts, sizeof(*ndopts));
298                 return NULL;
299         }
300
301         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
302         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
303                 /* option overruns the end of buffer, invalid */
304                 bzero(ndopts, sizeof(*ndopts));
305                 return NULL;
306         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
307                 /* reached the end of options chain */
308                 ndopts->nd_opts_done = 1;
309                 ndopts->nd_opts_search = NULL;
310         }
311         return nd_opt;
312 }
313
314 /*
315  * Parse multiple ND options.
316  * This function is much easier to use, for ND routines that do not need
317  * multiple options of the same type.
318  */
319 int
320 nd6_options(union nd_opts *ndopts)
321 {
322         struct nd_opt_hdr *nd_opt;
323         int i = 0;
324
325         if (!ndopts)
326                 panic("ndopts == NULL in nd6_options");
327         if (!ndopts->nd_opts_last)
328                 panic("uninitialized ndopts in nd6_options");
329         if (!ndopts->nd_opts_search)
330                 return 0;
331
332         while (1) {
333                 nd_opt = nd6_option(ndopts);
334                 if (!nd_opt && !ndopts->nd_opts_last) {
335                         /*
336                          * Message validation requires that all included
337                          * options have a length that is greater than zero.
338                          */
339                         icmp6stat.icp6s_nd_badopt++;
340                         bzero(ndopts, sizeof(*ndopts));
341                         return -1;
342                 }
343
344                 if (!nd_opt)
345                         goto skip1;
346
347                 switch (nd_opt->nd_opt_type) {
348                 case ND_OPT_SOURCE_LINKADDR:
349                 case ND_OPT_TARGET_LINKADDR:
350                 case ND_OPT_MTU:
351                 case ND_OPT_REDIRECTED_HEADER:
352                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
353                                 nd6log((LOG_INFO,
354                                     "duplicated ND6 option found (type=%d)\n",
355                                     nd_opt->nd_opt_type));
356                                 /* XXX bark? */
357                         } else {
358                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
359                                         = nd_opt;
360                         }
361                         break;
362                 case ND_OPT_PREFIX_INFORMATION:
363                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
364                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
365                                         = nd_opt;
366                         }
367                         ndopts->nd_opts_pi_end =
368                                 (struct nd_opt_prefix_info *)nd_opt;
369                         break;
370                 default:
371                         /*
372                          * Unknown options must be silently ignored,
373                          * to accomodate future extension to the protocol.
374                          */
375                         nd6log((LOG_DEBUG,
376                             "nd6_options: unsupported option %d - "
377                             "option ignored\n", nd_opt->nd_opt_type));
378                 }
379
380 skip1:
381                 i++;
382                 if (i > nd6_maxndopt) {
383                         icmp6stat.icp6s_nd_toomanyopt++;
384                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
385                         break;
386                 }
387
388                 if (ndopts->nd_opts_done)
389                         break;
390         }
391
392         return 0;
393 }
394
395 /*
396  * ND6 timer routine to expire default route list and prefix list
397  */
398 void
399 nd6_timer(void *ignored_arg)
400 {
401         struct llinfo_nd6 *ln;
402         struct nd_defrouter *dr;
403         struct nd_prefix *pr;
404         struct ifnet *ifp;
405         struct in6_ifaddr *ia6, *nia6;
406         struct in6_addrlifetime *lt6;
407
408         crit_enter();
409         callout_reset(&nd6_timer_ch, nd6_prune * hz,
410                       nd6_timer, NULL);
411
412         ln = llinfo_nd6.ln_next;
413         while (ln && ln != &llinfo_nd6) {
414                 struct rtentry *rt;
415                 struct sockaddr_in6 *dst;
416                 struct llinfo_nd6 *next = ln->ln_next;
417                 /* XXX: used for the DELAY case only: */
418                 struct nd_ifinfo *ndi = NULL;
419
420                 if ((rt = ln->ln_rt) == NULL) {
421                         ln = next;
422                         continue;
423                 }
424                 if ((ifp = rt->rt_ifp) == NULL) {
425                         ln = next;
426                         continue;
427                 }
428                 ndi = ND_IFINFO(ifp);
429                 dst = (struct sockaddr_in6 *)rt_key(rt);
430
431                 if (ln->ln_expire > time_second) {
432                         ln = next;
433                         continue;
434                 }
435
436                 /* sanity check */
437                 if (!rt)
438                         panic("rt=0 in nd6_timer(ln=%p)", ln);
439                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
440                         panic("rt_llinfo(%p) is not equal to ln(%p)",
441                               rt->rt_llinfo, ln);
442                 if (!dst)
443                         panic("dst=0 in nd6_timer(ln=%p)", ln);
444
445                 switch (ln->ln_state) {
446                 case ND6_LLINFO_INCOMPLETE:
447                         if (ln->ln_asked < nd6_mmaxtries) {
448                                 ln->ln_asked++;
449                                 ln->ln_expire = time_second +
450                                         ND_IFINFO(ifp)->retrans / 1000;
451                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
452                                         ln, 0);
453                         } else {
454                                 struct mbuf *m = ln->ln_hold;
455                                 if (m) {
456                                         if (rt->rt_ifp) {
457                                                 /*
458                                                  * Fake rcvif to make ICMP error
459                                                  * more helpful in diagnosing
460                                                  * for the receiver.
461                                                  * XXX: should we consider
462                                                  * older rcvif?
463                                                  */
464                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
465                                         }
466                                         icmp6_error(m, ICMP6_DST_UNREACH,
467                                                     ICMP6_DST_UNREACH_ADDR, 0);
468                                         ln->ln_hold = NULL;
469                                 }
470                                 next = nd6_free(rt);
471                         }
472                         break;
473                 case ND6_LLINFO_REACHABLE:
474                         if (ln->ln_expire) {
475                                 ln->ln_state = ND6_LLINFO_STALE;
476                                 ln->ln_expire = time_second + nd6_gctimer;
477                         }
478                         break;
479
480                 case ND6_LLINFO_STALE:
481                         /* Garbage Collection(RFC 2461 5.3) */
482                         if (ln->ln_expire)
483                                 next = nd6_free(rt);
484                         break;
485
486                 case ND6_LLINFO_DELAY:
487                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
488                                 /* We need NUD */
489                                 ln->ln_asked = 1;
490                                 ln->ln_state = ND6_LLINFO_PROBE;
491                                 ln->ln_expire = time_second +
492                                         ndi->retrans / 1000;
493                                 nd6_ns_output(ifp, &dst->sin6_addr,
494                                               &dst->sin6_addr,
495                                               ln, 0);
496                         } else {
497                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
498                                 ln->ln_expire = time_second + nd6_gctimer;
499                         }
500                         break;
501                 case ND6_LLINFO_PROBE:
502                         if (ln->ln_asked < nd6_umaxtries) {
503                                 ln->ln_asked++;
504                                 ln->ln_expire = time_second +
505                                         ND_IFINFO(ifp)->retrans / 1000;
506                                 nd6_ns_output(ifp, &dst->sin6_addr,
507                                                &dst->sin6_addr, ln, 0);
508                         } else {
509                                 next = nd6_free(rt);
510                         }
511                         break;
512                 }
513                 ln = next;
514         }
515
516         /* expire default router list */
517         dr = TAILQ_FIRST(&nd_defrouter);
518         while (dr) {
519                 if (dr->expire && dr->expire < time_second) {
520                         struct nd_defrouter *t;
521                         t = TAILQ_NEXT(dr, dr_entry);
522                         defrtrlist_del(dr);
523                         dr = t;
524                 } else {
525                         dr = TAILQ_NEXT(dr, dr_entry);
526                 }
527         }
528
529         /*
530          * expire interface addresses.
531          * in the past the loop was inside prefix expiry processing.
532          * However, from a stricter speci-confrmance standpoint, we should
533          * rather separate address lifetimes and prefix lifetimes.
534          */
535 addrloop:
536         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
537                 nia6 = ia6->ia_next;
538                 /* check address lifetime */
539                 lt6 = &ia6->ia6_lifetime;
540                 if (IFA6_IS_INVALID(ia6)) {
541                         int regen = 0;
542
543                         /*
544                          * If the expiring address is temporary, try
545                          * regenerating a new one.  This would be useful when
546                          * we suspended a laptop PC, then turned it on after a
547                          * period that could invalidate all temporary
548                          * addresses.  Although we may have to restart the
549                          * loop (see below), it must be after purging the
550                          * address.  Otherwise, we'd see an infinite loop of
551                          * regeneration.
552                          */
553                         if (ip6_use_tempaddr &&
554                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
555                                 if (regen_tmpaddr(ia6) == 0)
556                                         regen = 1;
557                         }
558
559                         in6_purgeaddr(&ia6->ia_ifa);
560
561                         if (regen)
562                                 goto addrloop; /* XXX: see below */
563                 }
564                 if (IFA6_IS_DEPRECATED(ia6)) {
565                         int oldflags = ia6->ia6_flags;
566
567                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
568
569                         /*
570                          * If a temporary address has just become deprecated,
571                          * regenerate a new one if possible.
572                          */
573                         if (ip6_use_tempaddr &&
574                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
575                             !(oldflags & IN6_IFF_DEPRECATED)) {
576
577                                 if (regen_tmpaddr(ia6) == 0) {
578                                         /*
579                                          * A new temporary address is
580                                          * generated.
581                                          * XXX: this means the address chain
582                                          * has changed while we are still in
583                                          * the loop.  Although the change
584                                          * would not cause disaster (because
585                                          * it's not a deletion, but an
586                                          * addition,) we'd rather restart the
587                                          * loop just for safety.  Or does this
588                                          * significantly reduce performance??
589                                          */
590                                         goto addrloop;
591                                 }
592                         }
593                 } else {
594                         /*
595                          * A new RA might have made a deprecated address
596                          * preferred.
597                          */
598                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
599                 }
600         }
601
602         /* expire prefix list */
603         pr = nd_prefix.lh_first;
604         while (pr) {
605                 /*
606                  * check prefix lifetime.
607                  * since pltime is just for autoconf, pltime processing for
608                  * prefix is not necessary.
609                  */
610                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
611                         struct nd_prefix *t;
612                         t = pr->ndpr_next;
613
614                         /*
615                          * address expiration and prefix expiration are
616                          * separate.  NEVER perform in6_purgeaddr here.
617                          */
618
619                         prelist_remove(pr);
620                         pr = t;
621                 } else
622                         pr = pr->ndpr_next;
623         }
624         crit_exit();
625 }
626
627 static int
628 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
629                                          address */
630 {
631         struct ifaddr *ifa;
632         struct ifnet *ifp;
633         struct in6_ifaddr *public_ifa6 = NULL;
634
635         ifp = ia6->ia_ifa.ifa_ifp;
636         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
637                 struct in6_ifaddr *it6;
638
639                 if (ifa->ifa_addr->sa_family != AF_INET6)
640                         continue;
641
642                 it6 = (struct in6_ifaddr *)ifa;
643
644                 /* ignore no autoconf addresses. */
645                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
646                         continue;
647
648                 /* ignore autoconf addresses with different prefixes. */
649                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
650                         continue;
651
652                 /*
653                  * Now we are looking at an autoconf address with the same
654                  * prefix as ours.  If the address is temporary and is still
655                  * preferred, do not create another one.  It would be rare, but
656                  * could happen, for example, when we resume a laptop PC after
657                  * a long period.
658                  */
659                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
660                     !IFA6_IS_DEPRECATED(it6)) {
661                         public_ifa6 = NULL;
662                         break;
663                 }
664
665                 /*
666                  * This is a public autoconf address that has the same prefix
667                  * as ours.  If it is preferred, keep it.  We can't break the
668                  * loop here, because there may be a still-preferred temporary
669                  * address with the prefix.
670                  */
671                 if (!IFA6_IS_DEPRECATED(it6))
672                     public_ifa6 = it6;
673         }
674
675         if (public_ifa6 != NULL) {
676                 int e;
677
678                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
679                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
680                             " tmp addr,errno=%d\n", e);
681                         return (-1);
682                 }
683                 return (0);
684         }
685
686         return (-1);
687 }
688
689 /*
690  * Nuke neighbor cache/prefix/default router management table, right before
691  * ifp goes away.
692  */
693 void
694 nd6_purge(struct ifnet *ifp)
695 {
696         struct llinfo_nd6 *ln, *nln;
697         struct nd_defrouter *dr, *ndr, drany;
698         struct nd_prefix *pr, *npr;
699
700         /* Nuke default router list entries toward ifp */
701         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
702                 /*
703                  * The first entry of the list may be stored in
704                  * the routing table, so we'll delete it later.
705                  */
706                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
707                         ndr = TAILQ_NEXT(dr, dr_entry);
708                         if (dr->ifp == ifp)
709                                 defrtrlist_del(dr);
710                 }
711                 dr = TAILQ_FIRST(&nd_defrouter);
712                 if (dr->ifp == ifp)
713                         defrtrlist_del(dr);
714         }
715
716         /* Nuke prefix list entries toward ifp */
717         for (pr = nd_prefix.lh_first; pr; pr = npr) {
718                 npr = pr->ndpr_next;
719                 if (pr->ndpr_ifp == ifp) {
720                         /*
721                          * Previously, pr->ndpr_addr is removed as well,
722                          * but I strongly believe we don't have to do it.
723                          * nd6_purge() is only called from in6_ifdetach(),
724                          * which removes all the associated interface addresses
725                          * by itself.
726                          * (jinmei@kame.net 20010129)
727                          */
728                         prelist_remove(pr);
729                 }
730         }
731
732         /* cancel default outgoing interface setting */
733         if (nd6_defifindex == ifp->if_index)
734                 nd6_setdefaultiface(0);
735
736         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
737                 /* refresh default router list */
738                 bzero(&drany, sizeof(drany));
739                 defrouter_delreq(&drany, 0);
740                 defrouter_select();
741         }
742
743         /*
744          * Nuke neighbor cache entries for the ifp.
745          * Note that rt->rt_ifp may not be the same as ifp,
746          * due to KAME goto ours hack.  See RTM_RESOLVE case in
747          * nd6_rtrequest(), and ip6_input().
748          */
749         ln = llinfo_nd6.ln_next;
750         while (ln && ln != &llinfo_nd6) {
751                 struct rtentry *rt;
752                 struct sockaddr_dl *sdl;
753
754                 nln = ln->ln_next;
755                 rt = ln->ln_rt;
756                 if (rt && rt->rt_gateway &&
757                     rt->rt_gateway->sa_family == AF_LINK) {
758                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
759                         if (sdl->sdl_index == ifp->if_index)
760                                 nln = nd6_free(rt);
761                 }
762                 ln = nln;
763         }
764 }
765
766 struct rtentry *
767 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
768 {
769         struct rtentry *rt;
770         struct sockaddr_in6 sin6;
771
772         bzero(&sin6, sizeof(sin6));
773         sin6.sin6_len = sizeof(struct sockaddr_in6);
774         sin6.sin6_family = AF_INET6;
775         sin6.sin6_addr = *addr6;
776
777         if (create)
778                 rt = rtlookup((struct sockaddr *)&sin6);
779         else
780                 rt = rtpurelookup((struct sockaddr *)&sin6);
781         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
782                 /*
783                  * This is the case for the default route.
784                  * If we want to create a neighbor cache for the address, we
785                  * should free the route for the destination and allocate an
786                  * interface route.
787                  */
788                 if (create) {
789                         --rt->rt_refcnt;
790                         rt = NULL;
791                 }
792         }
793         if (!rt) {
794                 if (create && ifp) {
795                         int e;
796
797                         /*
798                          * If no route is available and create is set,
799                          * we allocate a host route for the destination
800                          * and treat it like an interface route.
801                          * This hack is necessary for a neighbor which can't
802                          * be covered by our own prefix.
803                          */
804                         struct ifaddr *ifa =
805                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
806                         if (ifa == NULL)
807                                 return (NULL);
808
809                         /*
810                          * Create a new route.  RTF_LLINFO is necessary
811                          * to create a Neighbor Cache entry for the
812                          * destination in nd6_rtrequest which will be
813                          * called in rtrequest via ifa->ifa_rtrequest.
814                          */
815                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
816                                            ifa->ifa_addr,
817                                            (struct sockaddr *)&all1_sa,
818                                            (ifa->ifa_flags |
819                                             RTF_HOST | RTF_LLINFO) &
820                                            ~RTF_CLONING,
821                                            &rt)) != 0)
822                                 log(LOG_ERR,
823                                     "nd6_lookup: failed to add route for a "
824                                     "neighbor(%s), errno=%d\n",
825                                     ip6_sprintf(addr6), e);
826                         if (rt == NULL)
827                                 return (NULL);
828                         if (rt->rt_llinfo) {
829                                 struct llinfo_nd6 *ln =
830                                         (struct llinfo_nd6 *)rt->rt_llinfo;
831                                 ln->ln_state = ND6_LLINFO_NOSTATE;
832                         }
833                 } else
834                         return (NULL);
835         }
836         rt->rt_refcnt--;
837         /*
838          * Validation for the entry.
839          * Note that the check for rt_llinfo is necessary because a cloned
840          * route from a parent route that has the L flag (e.g. the default
841          * route to a p2p interface) may have the flag, too, while the
842          * destination is not actually a neighbor.
843          * XXX: we can't use rt->rt_ifp to check for the interface, since
844          *      it might be the loopback interface if the entry is for our
845          *      own address on a non-loopback interface. Instead, we should
846          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
847          *      interface.
848          */
849         if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
850             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
851             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
852                 if (create) {
853                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
854                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
855                         /* xxx more logs... kazu */
856                 }
857                 return (NULL);
858         }
859         return (rt);
860 }
861
862 /*
863  * Detect if a given IPv6 address identifies a neighbor on a given link.
864  * XXX: should take care of the destination of a p2p link?
865  */
866 int
867 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
868 {
869         struct ifaddr *ifa;
870         int i;
871
872 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
873 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
874
875         /*
876          * A link-local address is always a neighbor.
877          * XXX: we should use the sin6_scope_id field rather than the embedded
878          * interface index.
879          */
880         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
881             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
882                 return (1);
883
884         /*
885          * If the address matches one of our addresses,
886          * it should be a neighbor.
887          */
888         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
889                 if (ifa->ifa_addr->sa_family != AF_INET6)
890                         next: continue;
891
892                 for (i = 0; i < 4; i++) {
893                         if ((IFADDR6(ifa).s6_addr32[i] ^
894                              addr->sin6_addr.s6_addr32[i]) &
895                             IFMASK6(ifa).s6_addr32[i])
896                                 goto next;
897                 }
898                 return (1);
899         }
900
901         /*
902          * Even if the address matches none of our addresses, it might be
903          * in the neighbor cache.
904          */
905         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
906                 return (1);
907
908         return (0);
909 #undef IFADDR6
910 #undef IFMASK6
911 }
912
913 /*
914  * Free an nd6 llinfo entry.
915  */
916 struct llinfo_nd6 *
917 nd6_free(struct rtentry *rt)
918 {
919         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
920         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
921         struct nd_defrouter *dr;
922
923         /*
924          * we used to have pfctlinput(PRC_HOSTDEAD) here.
925          * even though it is not harmful, it was not really necessary.
926          */
927
928         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
929                 crit_enter();
930                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
931                                       rt->rt_ifp);
932
933                 if (ln->ln_router || dr) {
934                         /*
935                          * rt6_flush must be called whether or not the neighbor
936                          * is in the Default Router List.
937                          * See a corresponding comment in nd6_na_input().
938                          */
939                         rt6_flush(&in6, rt->rt_ifp);
940                 }
941
942                 if (dr) {
943                         /*
944                          * Unreachablity of a router might affect the default
945                          * router selection and on-link detection of advertised
946                          * prefixes.
947                          */
948
949                         /*
950                          * Temporarily fake the state to choose a new default
951                          * router and to perform on-link determination of
952                          * prefixes correctly.
953                          * Below the state will be set correctly,
954                          * or the entry itself will be deleted.
955                          */
956                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
957
958                         /*
959                          * Since defrouter_select() does not affect the
960                          * on-link determination and MIP6 needs the check
961                          * before the default router selection, we perform
962                          * the check now.
963                          */
964                         pfxlist_onlink_check();
965
966                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
967                                 /*
968                                  * It is used as the current default router,
969                                  * so we have to move it to the end of the
970                                  * list and choose a new one.
971                                  * XXX: it is not very efficient if this is
972                                  *      the only router.
973                                  */
974                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
975                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
976
977                                 defrouter_select();
978                         }
979                 }
980                 crit_exit();
981         }
982
983         /*
984          * Before deleting the entry, remember the next entry as the
985          * return value.  We need this because pfxlist_onlink_check() above
986          * might have freed other entries (particularly the old next entry) as
987          * a side effect (XXX).
988          */
989         next = ln->ln_next;
990
991         /*
992          * Detach the route from the routing tree and the list of neighbor
993          * caches, and disable the route entry not to be used in already
994          * cached routes.
995          */
996         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
997                   rt_mask(rt), 0, (struct rtentry **)0);
998
999         return (next);
1000 }
1001
1002 /*
1003  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1004  *
1005  * XXX cost-effective metods?
1006  */
1007 void
1008 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1009 {
1010         struct llinfo_nd6 *ln;
1011
1012         /*
1013          * If the caller specified "rt", use that.  Otherwise, resolve the
1014          * routing table by supplied "dst6".
1015          */
1016         if (!rt) {
1017                 if (!dst6)
1018                         return;
1019                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1020                         return;
1021         }
1022
1023         if ((rt->rt_flags & RTF_GATEWAY) ||
1024             !(rt->rt_flags & RTF_LLINFO) ||
1025             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1026             rt->rt_gateway->sa_family != AF_LINK) {
1027                 /* This is not a host route. */
1028                 return;
1029         }
1030
1031         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1032         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1033                 return;
1034
1035         /*
1036          * if we get upper-layer reachability confirmation many times,
1037          * it is possible we have false information.
1038          */
1039         if (!force) {
1040                 ln->ln_byhint++;
1041                 if (ln->ln_byhint > nd6_maxnudhint)
1042                         return;
1043         }
1044
1045         ln->ln_state = ND6_LLINFO_REACHABLE;
1046         if (ln->ln_expire)
1047                 ln->ln_expire = time_second +
1048                         ND_IFINFO(rt->rt_ifp)->reachable;
1049 }
1050
1051 void
1052 nd6_rtrequest(int req, struct rtentry *rt,
1053               struct rt_addrinfo *info) /* xxx unused */
1054 {
1055         struct sockaddr *gate = rt->rt_gateway;
1056         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1057         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1058         struct ifnet *ifp = rt->rt_ifp;
1059         struct ifaddr *ifa;
1060
1061         if ((rt->rt_flags & RTF_GATEWAY))
1062                 return;
1063
1064         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1065                 /*
1066                  * This is probably an interface direct route for a link
1067                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1068                  * We do not need special treatment below for such a route.
1069                  * Moreover, the RTF_LLINFO flag which would be set below
1070                  * would annoy the ndp(8) command.
1071                  */
1072                 return;
1073         }
1074
1075         if (req == RTM_RESOLVE &&
1076             (nd6_need_cache(ifp) == 0 || /* stf case */
1077              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1078                 /*
1079                  * FreeBSD and BSD/OS often make a cloned host route based
1080                  * on a less-specific route (e.g. the default route).
1081                  * If the less specific route does not have a "gateway"
1082                  * (this is the case when the route just goes to a p2p or an
1083                  * stf interface), we'll mistakenly make a neighbor cache for
1084                  * the host route, and will see strange neighbor solicitation
1085                  * for the corresponding destination.  In order to avoid the
1086                  * confusion, we check if the destination of the route is
1087                  * a neighbor in terms of neighbor discovery, and stop the
1088                  * process if not.  Additionally, we remove the LLINFO flag
1089                  * so that ndp(8) will not try to get the neighbor information
1090                  * of the destination.
1091                  */
1092                 rt->rt_flags &= ~RTF_LLINFO;
1093                 return;
1094         }
1095
1096         switch (req) {
1097         case RTM_ADD:
1098                 /*
1099                  * There is no backward compatibility :)
1100                  *
1101                  * if (!(rt->rt_flags & RTF_HOST) &&
1102                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1103                  *         rt->rt_flags |= RTF_CLONING;
1104                  */
1105                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1106                         /*
1107                          * Case 1: This route should come from
1108                          * a route to interface.  RTF_LLINFO flag is set
1109                          * for a host route whose destination should be
1110                          * treated as on-link.
1111                          */
1112                         rt_setgate(rt, rt_key(rt),
1113                                    (struct sockaddr *)&null_sdl);
1114                         gate = rt->rt_gateway;
1115                         SDL(gate)->sdl_type = ifp->if_type;
1116                         SDL(gate)->sdl_index = ifp->if_index;
1117                         if (ln)
1118                                 ln->ln_expire = time_second;
1119 #if 1
1120                         if (ln && ln->ln_expire == 0) {
1121                                 /* kludge for desktops */
1122 #if 0
1123                                 kprintf("nd6_rtequest: time.tv_sec is zero; "
1124                                        "treat it as 1\n");
1125 #endif
1126                                 ln->ln_expire = 1;
1127                         }
1128 #endif
1129                         if ((rt->rt_flags & RTF_CLONING))
1130                                 break;
1131                 }
1132                 /*
1133                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1134                  * We don't do that here since llinfo is not ready yet.
1135                  *
1136                  * There are also couple of other things to be discussed:
1137                  * - unsolicited NA code needs improvement beforehand
1138                  * - RFC2461 says we MAY send multicast unsolicited NA
1139                  *   (7.2.6 paragraph 4), however, it also says that we
1140                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1141                  *   we don't have anything like it right now.
1142                  *   note that the mechanism needs a mutual agreement
1143                  *   between proxies, which means that we need to implement
1144                  *   a new protocol, or a new kludge.
1145                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1146                  *   we need to check ip6forwarding before sending it.
1147                  *   (or should we allow proxy ND configuration only for
1148                  *   routers?  there's no mention about proxy ND from hosts)
1149                  */
1150 #if 0
1151                 /* XXX it does not work */
1152                 if (rt->rt_flags & RTF_ANNOUNCE)
1153                         nd6_na_output(ifp,
1154                               &SIN6(rt_key(rt))->sin6_addr,
1155                               &SIN6(rt_key(rt))->sin6_addr,
1156                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1157                               1, NULL);
1158 #endif
1159                 /* FALLTHROUGH */
1160         case RTM_RESOLVE:
1161                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1162                         /*
1163                          * Address resolution isn't necessary for a point to
1164                          * point link, so we can skip this test for a p2p link.
1165                          */
1166                         if (gate->sa_family != AF_LINK ||
1167                             gate->sa_len < sizeof(null_sdl)) {
1168                                 log(LOG_DEBUG,
1169                                     "nd6_rtrequest: bad gateway value: %s\n",
1170                                     if_name(ifp));
1171                                 break;
1172                         }
1173                         SDL(gate)->sdl_type = ifp->if_type;
1174                         SDL(gate)->sdl_index = ifp->if_index;
1175                 }
1176                 if (ln != NULL)
1177                         break;  /* This happens on a route change */
1178                 /*
1179                  * Case 2: This route may come from cloning, or a manual route
1180                  * add with a LL address.
1181                  */
1182                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1183                 rt->rt_llinfo = (caddr_t)ln;
1184                 if (!ln) {
1185                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1186                         break;
1187                 }
1188                 nd6_inuse++;
1189                 nd6_allocated++;
1190                 bzero(ln, sizeof(*ln));
1191                 ln->ln_rt = rt;
1192                 /* this is required for "ndp" command. - shin */
1193                 if (req == RTM_ADD) {
1194                         /*
1195                          * gate should have some valid AF_LINK entry,
1196                          * and ln->ln_expire should have some lifetime
1197                          * which is specified by ndp command.
1198                          */
1199                         ln->ln_state = ND6_LLINFO_REACHABLE;
1200                         ln->ln_byhint = 0;
1201                 } else {
1202                         /*
1203                          * When req == RTM_RESOLVE, rt is created and
1204                          * initialized in rtrequest(), so rt_expire is 0.
1205                          */
1206                         ln->ln_state = ND6_LLINFO_NOSTATE;
1207                         ln->ln_expire = time_second;
1208                 }
1209                 rt->rt_flags |= RTF_LLINFO;
1210                 ln->ln_next = llinfo_nd6.ln_next;
1211                 llinfo_nd6.ln_next = ln;
1212                 ln->ln_prev = &llinfo_nd6;
1213                 ln->ln_next->ln_prev = ln;
1214
1215                 /*
1216                  * check if rt_key(rt) is one of my address assigned
1217                  * to the interface.
1218                  */
1219                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1220                                           &SIN6(rt_key(rt))->sin6_addr);
1221                 if (ifa) {
1222                         caddr_t macp = nd6_ifptomac(ifp);
1223                         ln->ln_expire = 0;
1224                         ln->ln_state = ND6_LLINFO_REACHABLE;
1225                         ln->ln_byhint = 0;
1226                         if (macp) {
1227                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1228                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1229                         }
1230                         if (nd6_useloopback) {
1231                                 rt->rt_ifp = &loif[0];  /* XXX */
1232                                 /*
1233                                  * Make sure rt_ifa be equal to the ifaddr
1234                                  * corresponding to the address.
1235                                  * We need this because when we refer
1236                                  * rt_ifa->ia6_flags in ip6_input, we assume
1237                                  * that the rt_ifa points to the address instead
1238                                  * of the loopback address.
1239                                  */
1240                                 if (ifa != rt->rt_ifa) {
1241                                         IFAFREE(rt->rt_ifa);
1242                                         IFAREF(ifa);
1243                                         rt->rt_ifa = ifa;
1244                                 }
1245                         }
1246                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1247                         ln->ln_expire = 0;
1248                         ln->ln_state = ND6_LLINFO_REACHABLE;
1249                         ln->ln_byhint = 0;
1250
1251                         /* join solicited node multicast for proxy ND */
1252                         if (ifp->if_flags & IFF_MULTICAST) {
1253                                 struct in6_addr llsol;
1254                                 int error;
1255
1256                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1257                                 llsol.s6_addr16[0] = htons(0xff02);
1258                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1259                                 llsol.s6_addr32[1] = 0;
1260                                 llsol.s6_addr32[2] = htonl(1);
1261                                 llsol.s6_addr8[12] = 0xff;
1262
1263                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1264                                         nd6log((LOG_ERR, "%s: failed to join "
1265                                             "%s (errno=%d)\n", if_name(ifp),
1266                                             ip6_sprintf(&llsol), error));
1267                                 }
1268                         }
1269                 }
1270                 break;
1271
1272         case RTM_DELETE:
1273                 if (!ln)
1274                         break;
1275                 /* leave from solicited node multicast for proxy ND */
1276                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1277                     (ifp->if_flags & IFF_MULTICAST)) {
1278                         struct in6_addr llsol;
1279                         struct in6_multi *in6m;
1280
1281                         llsol = SIN6(rt_key(rt))->sin6_addr;
1282                         llsol.s6_addr16[0] = htons(0xff02);
1283                         llsol.s6_addr16[1] = htons(ifp->if_index);
1284                         llsol.s6_addr32[1] = 0;
1285                         llsol.s6_addr32[2] = htonl(1);
1286                         llsol.s6_addr8[12] = 0xff;
1287
1288                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1289                         if (in6m)
1290                                 in6_delmulti(in6m);
1291                 }
1292                 nd6_inuse--;
1293                 ln->ln_next->ln_prev = ln->ln_prev;
1294                 ln->ln_prev->ln_next = ln->ln_next;
1295                 ln->ln_prev = NULL;
1296                 rt->rt_llinfo = 0;
1297                 rt->rt_flags &= ~RTF_LLINFO;
1298                 if (ln->ln_hold)
1299                         m_freem(ln->ln_hold);
1300                 Free((caddr_t)ln);
1301         }
1302 }
1303
1304 int
1305 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1306 {
1307         struct in6_drlist *drl = (struct in6_drlist *)data;
1308         struct in6_prlist *prl = (struct in6_prlist *)data;
1309         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1310         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1311         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1312         struct nd_defrouter *dr, any;
1313         struct nd_prefix *pr;
1314         struct rtentry *rt;
1315         int i = 0, error = 0;
1316
1317         switch (cmd) {
1318         case SIOCGDRLST_IN6:
1319                 /*
1320                  * obsolete API, use sysctl under net.inet6.icmp6
1321                  */
1322                 bzero(drl, sizeof(*drl));
1323                 crit_enter();
1324                 dr = TAILQ_FIRST(&nd_defrouter);
1325                 while (dr && i < DRLSTSIZ) {
1326                         drl->defrouter[i].rtaddr = dr->rtaddr;
1327                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1328                                 /* XXX: need to this hack for KAME stack */
1329                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1330                         } else
1331                                 log(LOG_ERR,
1332                                     "default router list contains a "
1333                                     "non-linklocal address(%s)\n",
1334                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1335
1336                         drl->defrouter[i].flags = dr->flags;
1337                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1338                         drl->defrouter[i].expire = dr->expire;
1339                         drl->defrouter[i].if_index = dr->ifp->if_index;
1340                         i++;
1341                         dr = TAILQ_NEXT(dr, dr_entry);
1342                 }
1343                 crit_exit();
1344                 break;
1345         case SIOCGPRLST_IN6:
1346                 /*
1347                  * obsolete API, use sysctl under net.inet6.icmp6
1348                  */
1349                 /*
1350                  * XXX meaning of fields, especialy "raflags", is very
1351                  * differnet between RA prefix list and RR/static prefix list.
1352                  * how about separating ioctls into two?
1353                  */
1354                 bzero(prl, sizeof(*prl));
1355                 crit_enter();
1356                 pr = nd_prefix.lh_first;
1357                 while (pr && i < PRLSTSIZ) {
1358                         struct nd_pfxrouter *pfr;
1359                         int j;
1360
1361                         in6_embedscope(&prl->prefix[i].prefix,
1362                             &pr->ndpr_prefix, NULL, NULL);
1363                         prl->prefix[i].raflags = pr->ndpr_raf;
1364                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1365                         prl->prefix[i].vltime = pr->ndpr_vltime;
1366                         prl->prefix[i].pltime = pr->ndpr_pltime;
1367                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1368                         prl->prefix[i].expire = pr->ndpr_expire;
1369
1370                         pfr = pr->ndpr_advrtrs.lh_first;
1371                         j = 0;
1372                         while (pfr) {
1373                                 if (j < DRLSTSIZ) {
1374 #define RTRADDR prl->prefix[i].advrtr[j]
1375                                         RTRADDR = pfr->router->rtaddr;
1376                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1377                                                 /* XXX: hack for KAME */
1378                                                 RTRADDR.s6_addr16[1] = 0;
1379                                         } else
1380                                                 log(LOG_ERR,
1381                                                     "a router(%s) advertises "
1382                                                     "a prefix with "
1383                                                     "non-link local address\n",
1384                                                     ip6_sprintf(&RTRADDR));
1385 #undef RTRADDR
1386                                 }
1387                                 j++;
1388                                 pfr = pfr->pfr_next;
1389                         }
1390                         prl->prefix[i].advrtrs = j;
1391                         prl->prefix[i].origin = PR_ORIG_RA;
1392
1393                         i++;
1394                         pr = pr->ndpr_next;
1395                 }
1396               {
1397                 struct rr_prefix *rpp;
1398
1399                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1400                      rpp = LIST_NEXT(rpp, rp_entry)) {
1401                         if (i >= PRLSTSIZ)
1402                                 break;
1403                         in6_embedscope(&prl->prefix[i].prefix,
1404                             &pr->ndpr_prefix, NULL, NULL);
1405                         prl->prefix[i].raflags = rpp->rp_raf;
1406                         prl->prefix[i].prefixlen = rpp->rp_plen;
1407                         prl->prefix[i].vltime = rpp->rp_vltime;
1408                         prl->prefix[i].pltime = rpp->rp_pltime;
1409                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1410                         prl->prefix[i].expire = rpp->rp_expire;
1411                         prl->prefix[i].advrtrs = 0;
1412                         prl->prefix[i].origin = rpp->rp_origin;
1413                         i++;
1414                 }
1415               }
1416                 crit_exit();
1417
1418                 break;
1419         case OSIOCGIFINFO_IN6:
1420                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1421                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1422                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1423                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1424                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1425                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1426                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1427                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1428                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1429                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1430                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1431                 break;
1432         case SIOCGIFINFO_IN6:
1433                 ndi->ndi = *ND_IFINFO(ifp);
1434                 break;
1435         case SIOCSIFINFO_FLAGS:
1436                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1437                 break;
1438         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1439                 /* flush default router list */
1440                 /*
1441                  * xxx sumikawa: should not delete route if default
1442                  * route equals to the top of default router list
1443                  */
1444                 bzero(&any, sizeof(any));
1445                 defrouter_delreq(&any, 0);
1446                 defrouter_select();
1447                 /* xxx sumikawa: flush prefix list */
1448                 break;
1449         case SIOCSPFXFLUSH_IN6:
1450             {
1451                 /* flush all the prefix advertised by routers */
1452                 struct nd_prefix *pr, *next;
1453
1454                 crit_enter();
1455                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1456                         struct in6_ifaddr *ia, *ia_next;
1457
1458                         next = pr->ndpr_next;
1459
1460                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1461                                 continue; /* XXX */
1462
1463                         /* do we really have to remove addresses as well? */
1464                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1465                                 /* ia might be removed.  keep the next ptr. */
1466                                 ia_next = ia->ia_next;
1467
1468                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1469                                         continue;
1470
1471                                 if (ia->ia6_ndpr == pr)
1472                                         in6_purgeaddr(&ia->ia_ifa);
1473                         }
1474                         prelist_remove(pr);
1475                 }
1476                 crit_exit();
1477                 break;
1478             }
1479         case SIOCSRTRFLUSH_IN6:
1480             {
1481                 /* flush all the default routers */
1482                 struct nd_defrouter *dr, *next;
1483
1484                 crit_enter();
1485                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1486                         /*
1487                          * The first entry of the list may be stored in
1488                          * the routing table, so we'll delete it later.
1489                          */
1490                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1491                                 next = TAILQ_NEXT(dr, dr_entry);
1492                                 defrtrlist_del(dr);
1493                         }
1494                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1495                 }
1496                 crit_exit();
1497                 break;
1498             }
1499         case SIOCGNBRINFO_IN6:
1500             {
1501                 struct llinfo_nd6 *ln;
1502                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1503
1504                 /*
1505                  * XXX: KAME specific hack for scoped addresses
1506                  *      XXXX: for other scopes than link-local?
1507                  */
1508                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1509                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1510                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1511
1512                         if (*idp == 0)
1513                                 *idp = htons(ifp->if_index);
1514                 }
1515
1516                 crit_enter();
1517                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1518                         error = EINVAL;
1519                         crit_exit();
1520                         break;
1521                 }
1522                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1523                 nbi->state = ln->ln_state;
1524                 nbi->asked = ln->ln_asked;
1525                 nbi->isrouter = ln->ln_router;
1526                 nbi->expire = ln->ln_expire;
1527                 crit_exit();
1528
1529                 break;
1530             }
1531         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1532                 ndif->ifindex = nd6_defifindex;
1533                 break;
1534         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1535                 return (nd6_setdefaultiface(ndif->ifindex));
1536                 break;
1537         }
1538         return (error);
1539 }
1540
1541 /*
1542  * Create neighbor cache entry and cache link-layer address,
1543  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1544  */
1545 struct rtentry *
1546 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1547                  int lladdrlen,
1548                  int type,      /* ICMP6 type */
1549                  int code       /* type dependent information */)
1550 {
1551         struct rtentry *rt = NULL;
1552         struct llinfo_nd6 *ln = NULL;
1553         int is_newentry;
1554         struct sockaddr_dl *sdl = NULL;
1555         int do_update;
1556         int olladdr;
1557         int llchange;
1558         int newstate = 0;
1559
1560         if (!ifp)
1561                 panic("ifp == NULL in nd6_cache_lladdr");
1562         if (!from)
1563                 panic("from == NULL in nd6_cache_lladdr");
1564
1565         /* nothing must be updated for unspecified address */
1566         if (IN6_IS_ADDR_UNSPECIFIED(from))
1567                 return NULL;
1568
1569         /*
1570          * Validation about ifp->if_addrlen and lladdrlen must be done in
1571          * the caller.
1572          *
1573          * XXX If the link does not have link-layer adderss, what should
1574          * we do? (ifp->if_addrlen == 0)
1575          * Spec says nothing in sections for RA, RS and NA.  There's small
1576          * description on it in NS section (RFC 2461 7.2.3).
1577          */
1578
1579         rt = nd6_lookup(from, 0, ifp);
1580         if (!rt) {
1581 #if 0
1582                 /* nothing must be done if there's no lladdr */
1583                 if (!lladdr || !lladdrlen)
1584                         return NULL;
1585 #endif
1586
1587                 rt = nd6_lookup(from, 1, ifp);
1588                 is_newentry = 1;
1589         } else {
1590                 /* do nothing if static ndp is set */
1591                 if (rt->rt_flags & RTF_STATIC)
1592                         return NULL;
1593                 is_newentry = 0;
1594         }
1595
1596         if (!rt)
1597                 return NULL;
1598         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1599 fail:
1600                 nd6_free(rt);
1601                 return NULL;
1602         }
1603         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1604         if (!ln)
1605                 goto fail;
1606         if (!rt->rt_gateway)
1607                 goto fail;
1608         if (rt->rt_gateway->sa_family != AF_LINK)
1609                 goto fail;
1610         sdl = SDL(rt->rt_gateway);
1611
1612         olladdr = (sdl->sdl_alen) ? 1 : 0;
1613         if (olladdr && lladdr) {
1614                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1615                         llchange = 1;
1616                 else
1617                         llchange = 0;
1618         } else
1619                 llchange = 0;
1620
1621         /*
1622          * newentry olladdr  lladdr  llchange   (*=record)
1623          *      0       n       n       --      (1)
1624          *      0       y       n       --      (2)
1625          *      0       n       y       --      (3) * STALE
1626          *      0       y       y       n       (4) *
1627          *      0       y       y       y       (5) * STALE
1628          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1629          *      1       --      y       --      (7) * STALE
1630          */
1631
1632         if (lladdr) {           /* (3-5) and (7) */
1633                 /*
1634                  * Record source link-layer address
1635                  * XXX is it dependent to ifp->if_type?
1636                  */
1637                 sdl->sdl_alen = ifp->if_addrlen;
1638                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1639         }
1640
1641         if (!is_newentry) {
1642                 if ((!olladdr && lladdr)                /* (3) */
1643                  || (olladdr && lladdr && llchange)) {  /* (5) */
1644                         do_update = 1;
1645                         newstate = ND6_LLINFO_STALE;
1646                 } else                                  /* (1-2,4) */
1647                         do_update = 0;
1648         } else {
1649                 do_update = 1;
1650                 if (!lladdr)                            /* (6) */
1651                         newstate = ND6_LLINFO_NOSTATE;
1652                 else                                    /* (7) */
1653                         newstate = ND6_LLINFO_STALE;
1654         }
1655
1656         if (do_update) {
1657                 /*
1658                  * Update the state of the neighbor cache.
1659                  */
1660                 ln->ln_state = newstate;
1661
1662                 if (ln->ln_state == ND6_LLINFO_STALE) {
1663                         /*
1664                          * XXX: since nd6_output() below will cause
1665                          * state tansition to DELAY and reset the timer,
1666                          * we must set the timer now, although it is actually
1667                          * meaningless.
1668                          */
1669                         ln->ln_expire = time_second + nd6_gctimer;
1670
1671                         if (ln->ln_hold) {
1672                                 /*
1673                                  * we assume ifp is not a p2p here, so just
1674                                  * set the 2nd argument as the 1st one.
1675                                  */
1676                                 nd6_output(ifp, ifp, ln->ln_hold,
1677                                            (struct sockaddr_in6 *)rt_key(rt),
1678                                            rt);
1679                                 ln->ln_hold = NULL;
1680                         }
1681                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1682                         /* probe right away */
1683                         ln->ln_expire = time_second;
1684                 }
1685         }
1686
1687         /*
1688          * ICMP6 type dependent behavior.
1689          *
1690          * NS: clear IsRouter if new entry
1691          * RS: clear IsRouter
1692          * RA: set IsRouter if there's lladdr
1693          * redir: clear IsRouter if new entry
1694          *
1695          * RA case, (1):
1696          * The spec says that we must set IsRouter in the following cases:
1697          * - If lladdr exist, set IsRouter.  This means (1-5).
1698          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1699          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1700          * A quetion arises for (1) case.  (1) case has no lladdr in the
1701          * neighbor cache, this is similar to (6).
1702          * This case is rare but we figured that we MUST NOT set IsRouter.
1703          *
1704          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1705          *                                                      D R
1706          *      0       n       n       --      (1)     c   ?     s
1707          *      0       y       n       --      (2)     c   s     s
1708          *      0       n       y       --      (3)     c   s     s
1709          *      0       y       y       n       (4)     c   s     s
1710          *      0       y       y       y       (5)     c   s     s
1711          *      1       --      n       --      (6) c   c       c s
1712          *      1       --      y       --      (7) c   c   s   c s
1713          *
1714          *                                      (c=clear s=set)
1715          */
1716         switch (type & 0xff) {
1717         case ND_NEIGHBOR_SOLICIT:
1718                 /*
1719                  * New entry must have is_router flag cleared.
1720                  */
1721                 if (is_newentry)        /* (6-7) */
1722                         ln->ln_router = 0;
1723                 break;
1724         case ND_REDIRECT:
1725                 /*
1726                  * If the icmp is a redirect to a better router, always set the
1727                  * is_router flag. Otherwise, if the entry is newly created,
1728                  * clear the flag. [RFC 2461, sec 8.3]
1729                  */
1730                 if (code == ND_REDIRECT_ROUTER)
1731                         ln->ln_router = 1;
1732                 else if (is_newentry) /* (6-7) */
1733                         ln->ln_router = 0;
1734                 break;
1735         case ND_ROUTER_SOLICIT:
1736                 /*
1737                  * is_router flag must always be cleared.
1738                  */
1739                 ln->ln_router = 0;
1740                 break;
1741         case ND_ROUTER_ADVERT:
1742                 /*
1743                  * Mark an entry with lladdr as a router.
1744                  */
1745                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1746                  || (is_newentry && lladdr)) {                  /* (7) */
1747                         ln->ln_router = 1;
1748                 }
1749                 break;
1750         }
1751
1752         /*
1753          * When the link-layer address of a router changes, select the
1754          * best router again.  In particular, when the neighbor entry is newly
1755          * created, it might affect the selection policy.
1756          * Question: can we restrict the first condition to the "is_newentry"
1757          * case?
1758          * XXX: when we hear an RA from a new router with the link-layer
1759          * address option, defrouter_select() is called twice, since
1760          * defrtrlist_update called the function as well.  However, I believe
1761          * we can compromise the overhead, since it only happens the first
1762          * time.
1763          * XXX: although defrouter_select() should not have a bad effect
1764          * for those are not autoconfigured hosts, we explicitly avoid such
1765          * cases for safety.
1766          */
1767         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1768                 defrouter_select();
1769
1770         return rt;
1771 }
1772
1773 static void
1774 nd6_slowtimo(void *ignored_arg)
1775 {
1776         struct nd_ifinfo *nd6if;
1777         struct ifnet *ifp;
1778
1779         crit_enter();
1780         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1781             nd6_slowtimo, NULL);
1782         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1783                 nd6if = ND_IFINFO(ifp);
1784                 if (nd6if->basereachable && /* already initialized */
1785                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1786                         /*
1787                          * Since reachable time rarely changes by router
1788                          * advertisements, we SHOULD insure that a new random
1789                          * value gets recomputed at least once every few hours.
1790                          * (RFC 2461, 6.3.4)
1791                          */
1792                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1793                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1794                 }
1795         }
1796         crit_exit();
1797 }
1798
1799 #define gotoerr(e) { error = (e); goto bad;}
1800
1801 int
1802 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1803            struct sockaddr_in6 *dst, struct rtentry *rt)
1804 {
1805         struct llinfo_nd6 *ln = NULL;
1806         int error = 0;
1807
1808         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1809                 goto sendpkt;
1810
1811         if (nd6_need_cache(ifp) == 0)
1812                 goto sendpkt;
1813
1814         /*
1815          * next hop determination.  This routine is derived from ether_outpout.
1816          */
1817         if (rt != NULL) {
1818                 if (!(rt->rt_flags & RTF_UP)) {
1819                         rt = rtlookup((struct sockaddr *)dst);
1820                         if (rt == NULL)
1821                                 gotoerr(EHOSTUNREACH);
1822                         rt->rt_refcnt--;
1823                         if (rt->rt_ifp != ifp) {
1824                                 /* XXX: loop care? */
1825                                 return nd6_output(ifp, origifp, m, dst, rt);
1826                         }
1827                 }
1828                 if (rt->rt_flags & RTF_GATEWAY) {
1829                         struct sockaddr_in6 *gw6;
1830
1831                         /*
1832                          * We skip link-layer address resolution and NUD
1833                          * if the gateway is not a neighbor from ND point
1834                          * of view, regardless of the value of nd_ifinfo.flags.
1835                          * The second condition is a bit tricky; we skip
1836                          * if the gateway is our own address, which is
1837                          * sometimes used to install a route to a p2p link.
1838                          */
1839                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1840                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1841                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1842                                 /*
1843                                  * We allow this kind of tricky route only
1844                                  * when the outgoing interface is p2p.
1845                                  * XXX: we may need a more generic rule here.
1846                                  */
1847                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1848                                         gotoerr(EHOSTUNREACH);
1849
1850                                 goto sendpkt;
1851                         }
1852
1853                         if (rt->rt_gwroute == NULL) {
1854                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1855                                 if (rt->rt_gwroute == NULL)
1856                                         gotoerr(EHOSTUNREACH);
1857                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1858                                 rtfree(rt->rt_gwroute);
1859                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1860                                 if (rt->rt_gwroute == NULL)
1861                                         gotoerr(EHOSTUNREACH);
1862                         }
1863                 }
1864         }
1865
1866         /*
1867          * Address resolution or Neighbor Unreachability Detection
1868          * for the next hop.
1869          * At this point, the destination of the packet must be a unicast
1870          * or an anycast address(i.e. not a multicast).
1871          */
1872
1873         /* Look up the neighbor cache for the nexthop */
1874         if (rt && (rt->rt_flags & RTF_LLINFO))
1875                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1876         else {
1877                 /*
1878                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1879                  * the condition below is not very efficient.  But we believe
1880                  * it is tolerable, because this should be a rare case.
1881                  */
1882                 if (nd6_is_addr_neighbor(dst, ifp) &&
1883                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1884                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1885         }
1886         if (!ln || !rt) {
1887                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1888                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1889                         log(LOG_DEBUG,
1890                             "nd6_output: can't allocate llinfo for %s "
1891                             "(ln=%p, rt=%p)\n",
1892                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1893                         gotoerr(EIO);   /* XXX: good error? */
1894                 }
1895
1896                 goto sendpkt;   /* send anyway */
1897         }
1898
1899         /* We don't have to do link-layer address resolution on a p2p link. */
1900         if ((ifp->if_flags & IFF_POINTOPOINT) &&
1901             ln->ln_state < ND6_LLINFO_REACHABLE) {
1902                 ln->ln_state = ND6_LLINFO_STALE;
1903                 ln->ln_expire = time_second + nd6_gctimer;
1904         }
1905
1906         /*
1907          * The first time we send a packet to a neighbor whose entry is
1908          * STALE, we have to change the state to DELAY and a sets a timer to
1909          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1910          * neighbor unreachability detection on expiration.
1911          * (RFC 2461 7.3.3)
1912          */
1913         if (ln->ln_state == ND6_LLINFO_STALE) {
1914                 ln->ln_asked = 0;
1915                 ln->ln_state = ND6_LLINFO_DELAY;
1916                 ln->ln_expire = time_second + nd6_delay;
1917         }
1918
1919         /*
1920          * If the neighbor cache entry has a state other than INCOMPLETE
1921          * (i.e. its link-layer address is already resolved), just
1922          * send the packet.
1923          */
1924         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1925                 goto sendpkt;
1926
1927         /*
1928          * There is a neighbor cache entry, but no ethernet address
1929          * response yet.  Replace the held mbuf (if any) with this
1930          * latest one.
1931          *
1932          * This code conforms to the rate-limiting rule described in Section
1933          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1934          * an NS below.
1935          */
1936         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1937                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1938         if (ln->ln_hold)
1939                 m_freem(ln->ln_hold);
1940         ln->ln_hold = m;
1941         if (ln->ln_expire) {
1942                 if (ln->ln_asked < nd6_mmaxtries &&
1943                     ln->ln_expire < time_second) {
1944                         ln->ln_asked++;
1945                         ln->ln_expire = time_second +
1946                                 ND_IFINFO(ifp)->retrans / 1000;
1947                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1948                 }
1949         }
1950         return (0);
1951
1952 sendpkt:
1953         lwkt_serialize_enter(ifp->if_serializer);
1954         if (ifp->if_flags & IFF_LOOPBACK) {
1955                 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1956                                           rt);
1957         } else {
1958                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt);
1959         }
1960         lwkt_serialize_exit(ifp->if_serializer);
1961         return (error);
1962
1963 bad:
1964         m_freem(m);
1965         return (error);
1966 }
1967 #undef gotoerr
1968
1969 int
1970 nd6_need_cache(struct ifnet *ifp)
1971 {
1972         /*
1973          * XXX: we currently do not make neighbor cache on any interface
1974          * other than Ethernet and GIF.
1975          *
1976          * RFC2893 says:
1977          * - unidirectional tunnels needs no ND
1978          */
1979         switch (ifp->if_type) {
1980         case IFT_ETHER:
1981         case IFT_IEEE1394:
1982 #ifdef IFT_L2VLAN
1983         case IFT_L2VLAN:
1984 #endif
1985 #ifdef IFT_IEEE80211
1986         case IFT_IEEE80211:
1987 #endif
1988 #ifdef IFT_CARP
1989         case IFT_CARP:
1990 #endif
1991         case IFT_GIF:           /* XXX need more cases? */
1992                 return (1);
1993         default:
1994                 return (0);
1995         }
1996 }
1997
1998 int
1999 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2000                 struct sockaddr *dst, u_char *desten)
2001 {
2002         struct sockaddr_dl *sdl;
2003         struct rtentry *rt;
2004
2005
2006         if (m->m_flags & M_MCAST) {
2007                 switch (ifp->if_type) {
2008                 case IFT_ETHER:
2009 #ifdef IFT_L2VLAN
2010         case IFT_L2VLAN:
2011 #endif
2012 #ifdef IFT_IEEE80211
2013                 case IFT_IEEE80211:
2014 #endif
2015                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2016                                                  desten);
2017                         return (1);
2018                 case IFT_IEEE1394:
2019                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2020                         return (1);
2021                 default:
2022                         m_freem(m);
2023                         return (0);
2024                 }
2025         }
2026         if (rt0 == NULL) {
2027                 /* this could happen, if we could not allocate memory */
2028                 m_freem(m);
2029                 return (0);
2030         }
2031         if (rt_llroute(dst, rt0, &rt) != 0) {
2032                 m_freem(m);
2033                 return (0);
2034         }
2035         if (rt->rt_gateway->sa_family != AF_LINK) {
2036                 kprintf("nd6_storelladdr: something odd happens\n");
2037                 m_freem(m);
2038                 return (0);
2039         }
2040         sdl = SDL(rt->rt_gateway);
2041         if (sdl->sdl_alen == 0) {
2042                 /* this should be impossible, but we bark here for debugging */
2043                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2044                 m_freem(m);
2045                 return (0);
2046         }
2047
2048         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2049         return (1);
2050 }
2051
2052 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2053 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2054 #ifdef SYSCTL_DECL
2055 SYSCTL_DECL(_net_inet6_icmp6);
2056 #endif
2057 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2058         CTLFLAG_RD, nd6_sysctl_drlist, "");
2059 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2060         CTLFLAG_RD, nd6_sysctl_prlist, "");
2061
2062 static int
2063 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2064 {
2065         int error;
2066         char buf[1024];
2067         struct in6_defrouter *d, *de;
2068         struct nd_defrouter *dr;
2069
2070         if (req->newptr)
2071                 return EPERM;
2072         error = 0;
2073
2074         for (dr = TAILQ_FIRST(&nd_defrouter);
2075              dr;
2076              dr = TAILQ_NEXT(dr, dr_entry)) {
2077                 d = (struct in6_defrouter *)buf;
2078                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2079
2080                 if (d + 1 <= de) {
2081                         bzero(d, sizeof(*d));
2082                         d->rtaddr.sin6_family = AF_INET6;
2083                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2084                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2085                             dr->ifp) != 0)
2086                                 log(LOG_ERR,
2087                                     "scope error in "
2088                                     "default router list (%s)\n",
2089                                     ip6_sprintf(&dr->rtaddr));
2090                         d->flags = dr->flags;
2091                         d->rtlifetime = dr->rtlifetime;
2092                         d->expire = dr->expire;
2093                         d->if_index = dr->ifp->if_index;
2094                 } else
2095                         panic("buffer too short");
2096
2097                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2098                 if (error)
2099                         break;
2100         }
2101         return error;
2102 }
2103
2104 static int
2105 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2106 {
2107         int error;
2108         char buf[1024];
2109         struct in6_prefix *p, *pe;
2110         struct nd_prefix *pr;
2111
2112         if (req->newptr)
2113                 return EPERM;
2114         error = 0;
2115
2116         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2117                 u_short advrtrs;
2118                 size_t advance;
2119                 struct sockaddr_in6 *sin6, *s6;
2120                 struct nd_pfxrouter *pfr;
2121
2122                 p = (struct in6_prefix *)buf;
2123                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2124
2125                 if (p + 1 <= pe) {
2126                         bzero(p, sizeof(*p));
2127                         sin6 = (struct sockaddr_in6 *)(p + 1);
2128
2129                         p->prefix = pr->ndpr_prefix;
2130                         if (in6_recoverscope(&p->prefix,
2131                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2132                                 log(LOG_ERR,
2133                                     "scope error in prefix list (%s)\n",
2134                                     ip6_sprintf(&p->prefix.sin6_addr));
2135                         p->raflags = pr->ndpr_raf;
2136                         p->prefixlen = pr->ndpr_plen;
2137                         p->vltime = pr->ndpr_vltime;
2138                         p->pltime = pr->ndpr_pltime;
2139                         p->if_index = pr->ndpr_ifp->if_index;
2140                         p->expire = pr->ndpr_expire;
2141                         p->refcnt = pr->ndpr_refcnt;
2142                         p->flags = pr->ndpr_stateflags;
2143                         p->origin = PR_ORIG_RA;
2144                         advrtrs = 0;
2145                         for (pfr = pr->ndpr_advrtrs.lh_first;
2146                              pfr;
2147                              pfr = pfr->pfr_next) {
2148                                 if ((void *)&sin6[advrtrs + 1] >
2149                                     (void *)pe) {
2150                                         advrtrs++;
2151                                         continue;
2152                                 }
2153                                 s6 = &sin6[advrtrs];
2154                                 bzero(s6, sizeof(*s6));
2155                                 s6->sin6_family = AF_INET6;
2156                                 s6->sin6_len = sizeof(*sin6);
2157                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2158                                                      pfr->router->ifp) != 0)
2159                                         log(LOG_ERR,
2160                                             "scope error in "
2161                                             "prefix list (%s)\n",
2162                                             ip6_sprintf(&pfr->router->rtaddr));
2163                                 advrtrs++;
2164                         }
2165                         p->advrtrs = advrtrs;
2166                 } else
2167                         panic("buffer too short");
2168
2169                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2170                 error = SYSCTL_OUT(req, buf, advance);
2171                 if (error)
2172                         break;
2173         }
2174         return error;
2175 }