Make all network interrupt service routines MPSAFE part 1/3.
[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 1.17 2005/11/28 17:13:46 dillon 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 <netinet/if_fddi.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/in6_prefix.h>
75 #include <netinet/icmp6.h>
76
77 #include "use_loop.h"
78
79 #include <net/net_osdep.h>
80
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
83
84 #define SIN6(s) ((struct sockaddr_in6 *)s)
85 #define SDL(s) ((struct sockaddr_dl *)s)
86
87 /* timer values */
88 int     nd6_prune       = 1;    /* walk list every 1 seconds */
89 int     nd6_delay       = 5;    /* delay first probe time 5 second */
90 int     nd6_umaxtries   = 3;    /* maximum unicast query */
91 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
92 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
93 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
94
95 /* preventing too many loops in ND option parsing */
96 int nd6_maxndopt = 10;  /* max # of ND options allowed */
97
98 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
99
100 #ifdef ND6_DEBUG
101 int nd6_debug = 1;
102 #else
103 int nd6_debug = 0;
104 #endif
105
106 /* for debugging? */
107 static int nd6_inuse, nd6_allocated;
108
109 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
110 struct nd_drhead nd_defrouter;
111 struct nd_prhead nd_prefix = { 0 };
112
113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
114 static struct sockaddr_in6 all1_sa;
115
116 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
117 static void nd6_slowtimo (void *);
118 static int regen_tmpaddr (struct in6_ifaddr *);
119
120 struct callout nd6_slowtimo_ch;
121 struct callout nd6_timer_ch;
122 extern struct callout in6_tmpaddrtimer_ch;
123
124 void
125 nd6_init(void)
126 {
127         static int nd6_init_done = 0;
128         int i;
129
130         if (nd6_init_done) {
131                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
132                 return;
133         }
134
135         all1_sa.sin6_family = AF_INET6;
136         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
137         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
138                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
139
140         /* initialization of the default router list */
141         TAILQ_INIT(&nd_defrouter);
142
143         nd6_init_done = 1;
144
145         /* start timer */
146         callout_init(&nd6_slowtimo_ch);
147         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
148             nd6_slowtimo, NULL);
149 }
150
151 struct nd_ifinfo *
152 nd6_ifattach(struct ifnet *ifp)
153 {
154         struct nd_ifinfo *nd;
155
156         nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
157         bzero(nd, sizeof(*nd));
158
159         nd->initialized = 1;
160
161         nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
162         nd->chlim = IPV6_DEFHLIM;
163         nd->basereachable = REACHABLE_TIME;
164         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
165         nd->retrans = RETRANS_TIMER;
166         nd->receivedra = 0;
167
168         /*
169          * Note that the default value of ip6_accept_rtadv is 0, which means
170          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
171          * here.
172          */
173         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
174
175         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
176         nd6_setmtu0(ifp, nd);
177         return nd;
178 }
179
180 void
181 nd6_ifdetach(struct nd_ifinfo *nd)
182 {
183         free(nd, M_IP6NDP);
184 }
185
186 /*
187  * Reset ND level link MTU. This function is called when the physical MTU
188  * changes, which means we might have to adjust the ND level MTU.
189  */
190 void
191 nd6_setmtu(struct ifnet *ifp)
192 {
193         nd6_setmtu0(ifp, ND_IFINFO(ifp));
194 }
195
196 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
197 void
198 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
199 {
200         u_long oldmaxmtu;
201         u_long oldlinkmtu;
202
203         oldmaxmtu = ndi->maxmtu;
204         oldlinkmtu = ndi->linkmtu;
205
206         switch (ifp->if_type) {
207         case IFT_ARCNET:        /* XXX MTU handling needs more work */
208                 ndi->maxmtu = MIN(60480, ifp->if_mtu);
209                 break;
210         case IFT_ETHER:
211                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
212                 break;
213         case IFT_FDDI:
214                 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
215                 break;
216         case IFT_ATM:
217                 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
218                 break;
219         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
220                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
221                 break;
222 #ifdef IFT_IEEE80211
223         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
224                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
225                 break;
226 #endif
227         default:
228                 ndi->maxmtu = ifp->if_mtu;
229                 break;
230         }
231
232         if (oldmaxmtu != ndi->maxmtu) {
233                 /*
234                  * If the ND level MTU is not set yet, or if the maxmtu
235                  * is reset to a smaller value than the ND level MTU,
236                  * also reset the ND level MTU.
237                  */
238                 if (ndi->linkmtu == 0 ||
239                     ndi->maxmtu < ndi->linkmtu) {
240                         ndi->linkmtu = ndi->maxmtu;
241                         /* also adjust in6_maxmtu if necessary. */
242                         if (oldlinkmtu == 0) {
243                                 /*
244                                  * XXX: the case analysis is grotty, but
245                                  * it is not efficient to call in6_setmaxmtu()
246                                  * here when we are during the initialization
247                                  * procedure.
248                                  */
249                                 if (in6_maxmtu < ndi->linkmtu)
250                                         in6_maxmtu = ndi->linkmtu;
251                         } else
252                                 in6_setmaxmtu();
253                 }
254         }
255 #undef MIN
256 }
257
258 void
259 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
260 {
261         bzero(ndopts, sizeof(*ndopts));
262         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
263         ndopts->nd_opts_last
264                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
265
266         if (icmp6len == 0) {
267                 ndopts->nd_opts_done = 1;
268                 ndopts->nd_opts_search = NULL;
269         }
270 }
271
272 /*
273  * Take one ND option.
274  */
275 struct nd_opt_hdr *
276 nd6_option(union nd_opts *ndopts)
277 {
278         struct nd_opt_hdr *nd_opt;
279         int olen;
280
281         if (!ndopts)
282                 panic("ndopts == NULL in nd6_option");
283         if (!ndopts->nd_opts_last)
284                 panic("uninitialized ndopts in nd6_option");
285         if (!ndopts->nd_opts_search)
286                 return NULL;
287         if (ndopts->nd_opts_done)
288                 return NULL;
289
290         nd_opt = ndopts->nd_opts_search;
291
292         /* make sure nd_opt_len is inside the buffer */
293         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
294                 bzero(ndopts, sizeof(*ndopts));
295                 return NULL;
296         }
297
298         olen = nd_opt->nd_opt_len << 3;
299         if (olen == 0) {
300                 /*
301                  * Message validation requires that all included
302                  * options have a length that is greater than zero.
303                  */
304                 bzero(ndopts, sizeof(*ndopts));
305                 return NULL;
306         }
307
308         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
309         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
310                 /* option overruns the end of buffer, invalid */
311                 bzero(ndopts, sizeof(*ndopts));
312                 return NULL;
313         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
314                 /* reached the end of options chain */
315                 ndopts->nd_opts_done = 1;
316                 ndopts->nd_opts_search = NULL;
317         }
318         return nd_opt;
319 }
320
321 /*
322  * Parse multiple ND options.
323  * This function is much easier to use, for ND routines that do not need
324  * multiple options of the same type.
325  */
326 int
327 nd6_options(union nd_opts *ndopts)
328 {
329         struct nd_opt_hdr *nd_opt;
330         int i = 0;
331
332         if (!ndopts)
333                 panic("ndopts == NULL in nd6_options");
334         if (!ndopts->nd_opts_last)
335                 panic("uninitialized ndopts in nd6_options");
336         if (!ndopts->nd_opts_search)
337                 return 0;
338
339         while (1) {
340                 nd_opt = nd6_option(ndopts);
341                 if (!nd_opt && !ndopts->nd_opts_last) {
342                         /*
343                          * Message validation requires that all included
344                          * options have a length that is greater than zero.
345                          */
346                         icmp6stat.icp6s_nd_badopt++;
347                         bzero(ndopts, sizeof(*ndopts));
348                         return -1;
349                 }
350
351                 if (!nd_opt)
352                         goto skip1;
353
354                 switch (nd_opt->nd_opt_type) {
355                 case ND_OPT_SOURCE_LINKADDR:
356                 case ND_OPT_TARGET_LINKADDR:
357                 case ND_OPT_MTU:
358                 case ND_OPT_REDIRECTED_HEADER:
359                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
360                                 nd6log((LOG_INFO,
361                                     "duplicated ND6 option found (type=%d)\n",
362                                     nd_opt->nd_opt_type));
363                                 /* XXX bark? */
364                         } else {
365                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
366                                         = nd_opt;
367                         }
368                         break;
369                 case ND_OPT_PREFIX_INFORMATION:
370                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
371                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
372                                         = nd_opt;
373                         }
374                         ndopts->nd_opts_pi_end =
375                                 (struct nd_opt_prefix_info *)nd_opt;
376                         break;
377                 default:
378                         /*
379                          * Unknown options must be silently ignored,
380                          * to accomodate future extension to the protocol.
381                          */
382                         nd6log((LOG_DEBUG,
383                             "nd6_options: unsupported option %d - "
384                             "option ignored\n", nd_opt->nd_opt_type));
385                 }
386
387 skip1:
388                 i++;
389                 if (i > nd6_maxndopt) {
390                         icmp6stat.icp6s_nd_toomanyopt++;
391                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
392                         break;
393                 }
394
395                 if (ndopts->nd_opts_done)
396                         break;
397         }
398
399         return 0;
400 }
401
402 /*
403  * ND6 timer routine to expire default route list and prefix list
404  */
405 void
406 nd6_timer(void *ignored_arg)
407 {
408         struct llinfo_nd6 *ln;
409         struct nd_defrouter *dr;
410         struct nd_prefix *pr;
411         struct ifnet *ifp;
412         struct in6_ifaddr *ia6, *nia6;
413         struct in6_addrlifetime *lt6;
414         
415         crit_enter();
416         callout_reset(&nd6_timer_ch, nd6_prune * hz,
417                       nd6_timer, NULL);
418
419         ln = llinfo_nd6.ln_next;
420         while (ln && ln != &llinfo_nd6) {
421                 struct rtentry *rt;
422                 struct sockaddr_in6 *dst;
423                 struct llinfo_nd6 *next = ln->ln_next;
424                 /* XXX: used for the DELAY case only: */
425                 struct nd_ifinfo *ndi = NULL;
426
427                 if ((rt = ln->ln_rt) == NULL) {
428                         ln = next;
429                         continue;
430                 }
431                 if ((ifp = rt->rt_ifp) == NULL) {
432                         ln = next;
433                         continue;
434                 }
435                 ndi = ND_IFINFO(ifp);
436                 dst = (struct sockaddr_in6 *)rt_key(rt);
437
438                 if (ln->ln_expire > time_second) {
439                         ln = next;
440                         continue;
441                 }
442
443                 /* sanity check */
444                 if (!rt)
445                         panic("rt=0 in nd6_timer(ln=%p)", ln);
446                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
447                         panic("rt_llinfo(%p) is not equal to ln(%p)",
448                               rt->rt_llinfo, ln);
449                 if (!dst)
450                         panic("dst=0 in nd6_timer(ln=%p)", ln);
451
452                 switch (ln->ln_state) {
453                 case ND6_LLINFO_INCOMPLETE:
454                         if (ln->ln_asked < nd6_mmaxtries) {
455                                 ln->ln_asked++;
456                                 ln->ln_expire = time_second +
457                                         ND_IFINFO(ifp)->retrans / 1000;
458                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
459                                         ln, 0);
460                         } else {
461                                 struct mbuf *m = ln->ln_hold;
462                                 if (m) {
463                                         if (rt->rt_ifp) {
464                                                 /*
465                                                  * Fake rcvif to make ICMP error
466                                                  * more helpful in diagnosing
467                                                  * for the receiver.
468                                                  * XXX: should we consider
469                                                  * older rcvif?
470                                                  */
471                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
472                                         }
473                                         icmp6_error(m, ICMP6_DST_UNREACH,
474                                                     ICMP6_DST_UNREACH_ADDR, 0);
475                                         ln->ln_hold = NULL;
476                                 }
477                                 next = nd6_free(rt);
478                         }
479                         break;
480                 case ND6_LLINFO_REACHABLE:
481                         if (ln->ln_expire) {
482                                 ln->ln_state = ND6_LLINFO_STALE;
483                                 ln->ln_expire = time_second + nd6_gctimer;
484                         }
485                         break;
486
487                 case ND6_LLINFO_STALE:
488                         /* Garbage Collection(RFC 2461 5.3) */
489                         if (ln->ln_expire)
490                                 next = nd6_free(rt);
491                         break;
492
493                 case ND6_LLINFO_DELAY:
494                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
495                                 /* We need NUD */
496                                 ln->ln_asked = 1;
497                                 ln->ln_state = ND6_LLINFO_PROBE;
498                                 ln->ln_expire = time_second +
499                                         ndi->retrans / 1000;
500                                 nd6_ns_output(ifp, &dst->sin6_addr,
501                                               &dst->sin6_addr,
502                                               ln, 0);
503                         } else {
504                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
505                                 ln->ln_expire = time_second + nd6_gctimer;
506                         }
507                         break;
508                 case ND6_LLINFO_PROBE:
509                         if (ln->ln_asked < nd6_umaxtries) {
510                                 ln->ln_asked++;
511                                 ln->ln_expire = time_second +
512                                         ND_IFINFO(ifp)->retrans / 1000;
513                                 nd6_ns_output(ifp, &dst->sin6_addr,
514                                                &dst->sin6_addr, ln, 0);
515                         } else {
516                                 next = nd6_free(rt);
517                         }
518                         break;
519                 }
520                 ln = next;
521         }
522         
523         /* expire default router list */
524         dr = TAILQ_FIRST(&nd_defrouter);
525         while (dr) {
526                 if (dr->expire && dr->expire < time_second) {
527                         struct nd_defrouter *t;
528                         t = TAILQ_NEXT(dr, dr_entry);
529                         defrtrlist_del(dr);
530                         dr = t;
531                 } else {
532                         dr = TAILQ_NEXT(dr, dr_entry);
533                 }
534         }
535
536         /*
537          * expire interface addresses.
538          * in the past the loop was inside prefix expiry processing.
539          * However, from a stricter speci-confrmance standpoint, we should
540          * rather separate address lifetimes and prefix lifetimes.
541          */
542   addrloop:
543         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
544                 nia6 = ia6->ia_next;
545                 /* check address lifetime */
546                 lt6 = &ia6->ia6_lifetime;
547                 if (IFA6_IS_INVALID(ia6)) {
548                         int regen = 0;
549
550                         /*
551                          * If the expiring address is temporary, try
552                          * regenerating a new one.  This would be useful when
553                          * we suspended a laptop PC, then turned it on after a
554                          * period that could invalidate all temporary
555                          * addresses.  Although we may have to restart the
556                          * loop (see below), it must be after purging the
557                          * address.  Otherwise, we'd see an infinite loop of
558                          * regeneration. 
559                          */
560                         if (ip6_use_tempaddr &&
561                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
562                                 if (regen_tmpaddr(ia6) == 0)
563                                         regen = 1;
564                         }
565
566                         in6_purgeaddr(&ia6->ia_ifa);
567
568                         if (regen)
569                                 goto addrloop; /* XXX: see below */
570                 }
571                 if (IFA6_IS_DEPRECATED(ia6)) {
572                         int oldflags = ia6->ia6_flags;
573
574                         ia6->ia6_flags |= IN6_IFF_DEPRECATED;
575
576                         /*
577                          * If a temporary address has just become deprecated,
578                          * regenerate a new one if possible.
579                          */
580                         if (ip6_use_tempaddr &&
581                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
582                             (oldflags & IN6_IFF_DEPRECATED) == 0) {
583
584                                 if (regen_tmpaddr(ia6) == 0) {
585                                         /*
586                                          * A new temporary address is
587                                          * generated.
588                                          * XXX: this means the address chain
589                                          * has changed while we are still in
590                                          * the loop.  Although the change
591                                          * would not cause disaster (because
592                                          * it's not a deletion, but an
593                                          * addition,) we'd rather restart the
594                                          * loop just for safety.  Or does this 
595                                          * significantly reduce performance??
596                                          */
597                                         goto addrloop;
598                                 }
599                         }
600                 } else {
601                         /*
602                          * A new RA might have made a deprecated address
603                          * preferred.
604                          */
605                         ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
606                 }
607         }
608
609         /* expire prefix list */
610         pr = nd_prefix.lh_first;
611         while (pr) {
612                 /*
613                  * check prefix lifetime.
614                  * since pltime is just for autoconf, pltime processing for
615                  * prefix is not necessary.
616                  */
617                 if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
618                         struct nd_prefix *t;
619                         t = pr->ndpr_next;
620
621                         /*
622                          * address expiration and prefix expiration are
623                          * separate.  NEVER perform in6_purgeaddr here.
624                          */
625
626                         prelist_remove(pr);
627                         pr = t;
628                 } else
629                         pr = pr->ndpr_next;
630         }
631         crit_exit();
632 }
633
634 static int
635 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
636                                          address */
637 {
638         struct ifaddr *ifa;
639         struct ifnet *ifp;
640         struct in6_ifaddr *public_ifa6 = NULL;
641
642         ifp = ia6->ia_ifa.ifa_ifp;
643         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
644                 struct in6_ifaddr *it6;
645
646                 if (ifa->ifa_addr->sa_family != AF_INET6)
647                         continue;
648
649                 it6 = (struct in6_ifaddr *)ifa;
650
651                 /* ignore no autoconf addresses. */
652                 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
653                         continue;
654
655                 /* ignore autoconf addresses with different prefixes. */
656                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
657                         continue;
658
659                 /*
660                  * Now we are looking at an autoconf address with the same
661                  * prefix as ours.  If the address is temporary and is still
662                  * preferred, do not create another one.  It would be rare, but
663                  * could happen, for example, when we resume a laptop PC after
664                  * a long period.
665                  */
666                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
667                     !IFA6_IS_DEPRECATED(it6)) {
668                         public_ifa6 = NULL;
669                         break;
670                 }
671
672                 /*
673                  * This is a public autoconf address that has the same prefix
674                  * as ours.  If it is preferred, keep it.  We can't break the
675                  * loop here, because there may be a still-preferred temporary
676                  * address with the prefix.
677                  */
678                 if (!IFA6_IS_DEPRECATED(it6))
679                     public_ifa6 = it6;
680         }
681
682         if (public_ifa6 != NULL) {
683                 int e;
684
685                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
686                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
687                             " tmp addr,errno=%d\n", e);
688                         return(-1);
689                 }
690                 return(0);
691         }
692
693         return(-1);
694 }
695
696 /*
697  * Nuke neighbor cache/prefix/default router management table, right before
698  * ifp goes away.
699  */
700 void
701 nd6_purge(struct ifnet *ifp)
702 {
703         struct llinfo_nd6 *ln, *nln;
704         struct nd_defrouter *dr, *ndr, drany;
705         struct nd_prefix *pr, *npr;
706
707         /* Nuke default router list entries toward ifp */
708         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
709                 /*
710                  * The first entry of the list may be stored in
711                  * the routing table, so we'll delete it later.
712                  */
713                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
714                         ndr = TAILQ_NEXT(dr, dr_entry);
715                         if (dr->ifp == ifp)
716                                 defrtrlist_del(dr);
717                 }
718                 dr = TAILQ_FIRST(&nd_defrouter);
719                 if (dr->ifp == ifp)
720                         defrtrlist_del(dr);
721         }
722
723         /* Nuke prefix list entries toward ifp */
724         for (pr = nd_prefix.lh_first; pr; pr = npr) {
725                 npr = pr->ndpr_next;
726                 if (pr->ndpr_ifp == ifp) {
727                         /*
728                          * Previously, pr->ndpr_addr is removed as well,
729                          * but I strongly believe we don't have to do it.
730                          * nd6_purge() is only called from in6_ifdetach(),
731                          * which removes all the associated interface addresses
732                          * by itself.
733                          * (jinmei@kame.net 20010129)
734                          */
735                         prelist_remove(pr);
736                 }
737         }
738
739         /* cancel default outgoing interface setting */
740         if (nd6_defifindex == ifp->if_index)
741                 nd6_setdefaultiface(0);
742
743         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
744                 /* refresh default router list */
745                 bzero(&drany, sizeof(drany));
746                 defrouter_delreq(&drany, 0);
747                 defrouter_select();
748         }
749
750         /*
751          * Nuke neighbor cache entries for the ifp.
752          * Note that rt->rt_ifp may not be the same as ifp,
753          * due to KAME goto ours hack.  See RTM_RESOLVE case in
754          * nd6_rtrequest(), and ip6_input().
755          */
756         ln = llinfo_nd6.ln_next;
757         while (ln && ln != &llinfo_nd6) {
758                 struct rtentry *rt;
759                 struct sockaddr_dl *sdl;
760
761                 nln = ln->ln_next;
762                 rt = ln->ln_rt;
763                 if (rt && rt->rt_gateway &&
764                     rt->rt_gateway->sa_family == AF_LINK) {
765                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
766                         if (sdl->sdl_index == ifp->if_index)
767                                 nln = nd6_free(rt);
768                 }
769                 ln = nln;
770         }
771 }
772
773 struct rtentry *
774 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
775 {
776         struct rtentry *rt;
777         struct sockaddr_in6 sin6;
778
779         bzero(&sin6, sizeof(sin6));
780         sin6.sin6_len = sizeof(struct sockaddr_in6);
781         sin6.sin6_family = AF_INET6;
782         sin6.sin6_addr = *addr6;
783
784         if (create)
785                 rt = rtlookup((struct sockaddr *)&sin6);
786         else
787                 rt = rtpurelookup((struct sockaddr *)&sin6);
788         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
789                 /*
790                  * This is the case for the default route.
791                  * If we want to create a neighbor cache for the address, we
792                  * should free the route for the destination and allocate an
793                  * interface route.
794                  */
795                 if (create) {
796                         --rt->rt_refcnt;
797                         rt = NULL;
798                 }
799         }
800         if (!rt) {
801                 if (create && ifp) {
802                         int e;
803
804                         /*
805                          * If no route is available and create is set,
806                          * we allocate a host route for the destination
807                          * and treat it like an interface route.
808                          * This hack is necessary for a neighbor which can't
809                          * be covered by our own prefix.
810                          */
811                         struct ifaddr *ifa =
812                                 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
813                         if (ifa == NULL)
814                                 return(NULL);
815
816                         /*
817                          * Create a new route.  RTF_LLINFO is necessary
818                          * to create a Neighbor Cache entry for the
819                          * destination in nd6_rtrequest which will be
820                          * called in rtrequest via ifa->ifa_rtrequest.
821                          */
822                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
823                                            ifa->ifa_addr,
824                                            (struct sockaddr *)&all1_sa,
825                                            (ifa->ifa_flags |
826                                             RTF_HOST | RTF_LLINFO) &
827                                            ~RTF_CLONING,
828                                            &rt)) != 0)
829                                 log(LOG_ERR,
830                                     "nd6_lookup: failed to add route for a "
831                                     "neighbor(%s), errno=%d\n",
832                                     ip6_sprintf(addr6), e);
833                         if (rt == NULL)
834                                 return(NULL);
835                         if (rt->rt_llinfo) {
836                                 struct llinfo_nd6 *ln =
837                                         (struct llinfo_nd6 *)rt->rt_llinfo;
838                                 ln->ln_state = ND6_LLINFO_NOSTATE;
839                         }
840                 } else
841                         return(NULL);
842         }
843         rt->rt_refcnt--;
844         /*
845          * Validation for the entry.
846          * Note that the check for rt_llinfo is necessary because a cloned
847          * route from a parent route that has the L flag (e.g. the default
848          * route to a p2p interface) may have the flag, too, while the
849          * destination is not actually a neighbor.
850          * XXX: we can't use rt->rt_ifp to check for the interface, since
851          *      it might be the loopback interface if the entry is for our
852          *      own address on a non-loopback interface. Instead, we should
853          *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
854          *      interface.
855          */
856         if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
857             rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
858             (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
859                 if (create) {
860                         log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
861                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
862                         /* xxx more logs... kazu */
863                 }
864                 return(NULL);
865         }
866         return(rt);
867 }
868
869 /*
870  * Detect if a given IPv6 address identifies a neighbor on a given link.
871  * XXX: should take care of the destination of a p2p link?
872  */
873 int
874 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
875 {
876         struct ifaddr *ifa;
877         int i;
878
879 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
880 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
881
882         /*
883          * A link-local address is always a neighbor.
884          * XXX: we should use the sin6_scope_id field rather than the embedded
885          * interface index.
886          */
887         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
888             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
889                 return(1);
890
891         /*
892          * If the address matches one of our addresses,
893          * it should be a neighbor.
894          */
895         TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
896                 if (ifa->ifa_addr->sa_family != AF_INET6)
897                         next: continue;
898
899                 for (i = 0; i < 4; i++) {
900                         if ((IFADDR6(ifa).s6_addr32[i] ^
901                              addr->sin6_addr.s6_addr32[i]) &
902                             IFMASK6(ifa).s6_addr32[i])
903                                 goto next;
904                 }
905                 return(1);
906         }
907
908         /*
909          * Even if the address matches none of our addresses, it might be
910          * in the neighbor cache.
911          */
912         if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
913                 return(1);
914
915         return(0);
916 #undef IFADDR6
917 #undef IFMASK6
918 }
919
920 /*
921  * Free an nd6 llinfo entry.
922  */
923 struct llinfo_nd6 *
924 nd6_free(struct rtentry *rt)
925 {
926         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
927         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
928         struct nd_defrouter *dr;
929
930         /*
931          * we used to have pfctlinput(PRC_HOSTDEAD) here. 
932          * even though it is not harmful, it was not really necessary.
933          */
934
935         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
936                 crit_enter();
937                 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
938                                       rt->rt_ifp);
939
940                 if (ln->ln_router || dr) {
941                         /*
942                          * rt6_flush must be called whether or not the neighbor
943                          * is in the Default Router List.
944                          * See a corresponding comment in nd6_na_input().
945                          */
946                         rt6_flush(&in6, rt->rt_ifp);
947                 }
948
949                 if (dr) {
950                         /*
951                          * Unreachablity of a router might affect the default
952                          * router selection and on-link detection of advertised
953                          * prefixes.
954                          */
955
956                         /*
957                          * Temporarily fake the state to choose a new default
958                          * router and to perform on-link determination of
959                          * prefixes correctly.
960                          * Below the state will be set correctly,
961                          * or the entry itself will be deleted.
962                          */
963                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
964
965                         /*
966                          * Since defrouter_select() does not affect the
967                          * on-link determination and MIP6 needs the check
968                          * before the default router selection, we perform
969                          * the check now.
970                          */
971                         pfxlist_onlink_check();
972
973                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
974                                 /*
975                                  * It is used as the current default router,
976                                  * so we have to move it to the end of the
977                                  * list and choose a new one.
978                                  * XXX: it is not very efficient if this is
979                                  *      the only router.
980                                  */
981                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
982                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
983
984                                 defrouter_select();
985                         }
986                 }
987                 crit_exit();
988         }
989
990         /*
991          * Before deleting the entry, remember the next entry as the
992          * return value.  We need this because pfxlist_onlink_check() above
993          * might have freed other entries (particularly the old next entry) as
994          * a side effect (XXX).
995          */
996         next = ln->ln_next;
997
998         /*
999          * Detach the route from the routing tree and the list of neighbor
1000          * caches, and disable the route entry not to be used in already
1001          * cached routes.
1002          */
1003         rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1004                   rt_mask(rt), 0, (struct rtentry **)0);
1005
1006         return(next);
1007 }
1008
1009 /*
1010  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1011  *
1012  * XXX cost-effective metods?
1013  */
1014 void
1015 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1016 {
1017         struct llinfo_nd6 *ln;
1018
1019         /*
1020          * If the caller specified "rt", use that.  Otherwise, resolve the
1021          * routing table by supplied "dst6".
1022          */
1023         if (!rt) {
1024                 if (!dst6)
1025                         return;
1026                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1027                         return;
1028         }
1029
1030         if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1031             (rt->rt_flags & RTF_LLINFO) == 0 ||
1032             !rt->rt_llinfo || !rt->rt_gateway ||
1033             rt->rt_gateway->sa_family != AF_LINK) {
1034                 /* This is not a host route. */
1035                 return;
1036         }
1037
1038         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1039         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1040                 return;
1041
1042         /*
1043          * if we get upper-layer reachability confirmation many times,
1044          * it is possible we have false information.
1045          */
1046         if (!force) {
1047                 ln->ln_byhint++;
1048                 if (ln->ln_byhint > nd6_maxnudhint)
1049                         return;
1050         }
1051
1052         ln->ln_state = ND6_LLINFO_REACHABLE;
1053         if (ln->ln_expire)
1054                 ln->ln_expire = time_second +
1055                         ND_IFINFO(rt->rt_ifp)->reachable;
1056 }
1057
1058 void
1059 nd6_rtrequest(int req, struct rtentry *rt,
1060               struct rt_addrinfo *info) /* xxx unused */
1061 {
1062         struct sockaddr *gate = rt->rt_gateway;
1063         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1064         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1065         struct ifnet *ifp = rt->rt_ifp;
1066         struct ifaddr *ifa;
1067
1068         if ((rt->rt_flags & RTF_GATEWAY))
1069                 return;
1070
1071         if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1072                 /*
1073                  * This is probably an interface direct route for a link
1074                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1075                  * We do not need special treatment below for such a route.
1076                  * Moreover, the RTF_LLINFO flag which would be set below
1077                  * would annoy the ndp(8) command.
1078                  */
1079                 return;
1080         }
1081
1082         if (req == RTM_RESOLVE &&
1083             (nd6_need_cache(ifp) == 0 || /* stf case */
1084              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1085                 /*
1086                  * FreeBSD and BSD/OS often make a cloned host route based
1087                  * on a less-specific route (e.g. the default route).
1088                  * If the less specific route does not have a "gateway"
1089                  * (this is the case when the route just goes to a p2p or an
1090                  * stf interface), we'll mistakenly make a neighbor cache for
1091                  * the host route, and will see strange neighbor solicitation
1092                  * for the corresponding destination.  In order to avoid the
1093                  * confusion, we check if the destination of the route is
1094                  * a neighbor in terms of neighbor discovery, and stop the
1095                  * process if not.  Additionally, we remove the LLINFO flag
1096                  * so that ndp(8) will not try to get the neighbor information
1097                  * of the destination.
1098                  */
1099                 rt->rt_flags &= ~RTF_LLINFO;
1100                 return;
1101         }
1102
1103         switch (req) {
1104         case RTM_ADD:
1105                 /*
1106                  * There is no backward compatibility :)
1107                  *
1108                  * if ((rt->rt_flags & RTF_HOST) == 0 &&
1109                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1110                  *         rt->rt_flags |= RTF_CLONING;
1111                  */
1112                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1113                         /*
1114                          * Case 1: This route should come from
1115                          * a route to interface.  RTF_LLINFO flag is set
1116                          * for a host route whose destination should be
1117                          * treated as on-link.
1118                          */
1119                         rt_setgate(rt, rt_key(rt),
1120                                    (struct sockaddr *)&null_sdl);
1121                         gate = rt->rt_gateway;
1122                         SDL(gate)->sdl_type = ifp->if_type;
1123                         SDL(gate)->sdl_index = ifp->if_index;
1124                         if (ln)
1125                                 ln->ln_expire = time_second;
1126 #if 1
1127                         if (ln && ln->ln_expire == 0) {
1128                                 /* kludge for desktops */
1129 #if 0
1130                                 printf("nd6_rtequest: time.tv_sec is zero; "
1131                                        "treat it as 1\n");
1132 #endif
1133                                 ln->ln_expire = 1;
1134                         }
1135 #endif
1136                         if ((rt->rt_flags & RTF_CLONING))
1137                                 break;
1138                 }
1139                 /*
1140                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1141                  * We don't do that here since llinfo is not ready yet.
1142                  *
1143                  * There are also couple of other things to be discussed:
1144                  * - unsolicited NA code needs improvement beforehand
1145                  * - RFC2461 says we MAY send multicast unsolicited NA
1146                  *   (7.2.6 paragraph 4), however, it also says that we
1147                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1148                  *   we don't have anything like it right now.
1149                  *   note that the mechanism needs a mutual agreement
1150                  *   between proxies, which means that we need to implement
1151                  *   a new protocol, or a new kludge.
1152                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1153                  *   we need to check ip6forwarding before sending it.
1154                  *   (or should we allow proxy ND configuration only for
1155                  *   routers?  there's no mention about proxy ND from hosts)
1156                  */
1157 #if 0
1158                 /* XXX it does not work */
1159                 if (rt->rt_flags & RTF_ANNOUNCE)
1160                         nd6_na_output(ifp,
1161                               &SIN6(rt_key(rt))->sin6_addr,
1162                               &SIN6(rt_key(rt))->sin6_addr,
1163                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1164                               1, NULL);
1165 #endif
1166                 /* FALLTHROUGH */
1167         case RTM_RESOLVE:
1168                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1169                         /*
1170                          * Address resolution isn't necessary for a point to
1171                          * point link, so we can skip this test for a p2p link.
1172                          */
1173                         if (gate->sa_family != AF_LINK ||
1174                             gate->sa_len < sizeof(null_sdl)) {
1175                                 log(LOG_DEBUG,
1176                                     "nd6_rtrequest: bad gateway value: %s\n",
1177                                     if_name(ifp));
1178                                 break;
1179                         }
1180                         SDL(gate)->sdl_type = ifp->if_type;
1181                         SDL(gate)->sdl_index = ifp->if_index;
1182                 }
1183                 if (ln != NULL)
1184                         break;  /* This happens on a route change */
1185                 /*
1186                  * Case 2: This route may come from cloning, or a manual route
1187                  * add with a LL address.
1188                  */
1189                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1190                 rt->rt_llinfo = (caddr_t)ln;
1191                 if (!ln) {
1192                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1193                         break;
1194                 }
1195                 nd6_inuse++;
1196                 nd6_allocated++;
1197                 bzero(ln, sizeof(*ln));
1198                 ln->ln_rt = rt;
1199                 /* this is required for "ndp" command. - shin */
1200                 if (req == RTM_ADD) {
1201                         /*
1202                          * gate should have some valid AF_LINK entry,
1203                          * and ln->ln_expire should have some lifetime
1204                          * which is specified by ndp command.
1205                          */
1206                         ln->ln_state = ND6_LLINFO_REACHABLE;
1207                         ln->ln_byhint = 0;
1208                 } else {
1209                         /*
1210                          * When req == RTM_RESOLVE, rt is created and
1211                          * initialized in rtrequest(), so rt_expire is 0.
1212                          */
1213                         ln->ln_state = ND6_LLINFO_NOSTATE;
1214                         ln->ln_expire = time_second;
1215                 }
1216                 rt->rt_flags |= RTF_LLINFO;
1217                 ln->ln_next = llinfo_nd6.ln_next;
1218                 llinfo_nd6.ln_next = ln;
1219                 ln->ln_prev = &llinfo_nd6;
1220                 ln->ln_next->ln_prev = ln;
1221
1222                 /*
1223                  * check if rt_key(rt) is one of my address assigned
1224                  * to the interface.
1225                  */
1226                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1227                                           &SIN6(rt_key(rt))->sin6_addr);
1228                 if (ifa) {
1229                         caddr_t macp = nd6_ifptomac(ifp);
1230                         ln->ln_expire = 0;
1231                         ln->ln_state = ND6_LLINFO_REACHABLE;
1232                         ln->ln_byhint = 0;
1233                         if (macp) {
1234                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1235                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1236                         }
1237                         if (nd6_useloopback) {
1238                                 rt->rt_ifp = &loif[0];  /* XXX */
1239                                 /*
1240                                  * Make sure rt_ifa be equal to the ifaddr
1241                                  * corresponding to the address.
1242                                  * We need this because when we refer
1243                                  * rt_ifa->ia6_flags in ip6_input, we assume
1244                                  * that the rt_ifa points to the address instead
1245                                  * of the loopback address.
1246                                  */
1247                                 if (ifa != rt->rt_ifa) {
1248                                         IFAFREE(rt->rt_ifa);
1249                                         IFAREF(ifa);
1250                                         rt->rt_ifa = ifa;
1251                                 }
1252                         }
1253                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1254                         ln->ln_expire = 0;
1255                         ln->ln_state = ND6_LLINFO_REACHABLE;
1256                         ln->ln_byhint = 0;
1257
1258                         /* join solicited node multicast for proxy ND */
1259                         if (ifp->if_flags & IFF_MULTICAST) {
1260                                 struct in6_addr llsol;
1261                                 int error;
1262
1263                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1264                                 llsol.s6_addr16[0] = htons(0xff02);
1265                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1266                                 llsol.s6_addr32[1] = 0;
1267                                 llsol.s6_addr32[2] = htonl(1);
1268                                 llsol.s6_addr8[12] = 0xff;
1269
1270                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1271                                         nd6log((LOG_ERR, "%s: failed to join "
1272                                             "%s (errno=%d)\n", if_name(ifp),
1273                                             ip6_sprintf(&llsol), error));
1274                                 }
1275                         }
1276                 }
1277                 break;
1278
1279         case RTM_DELETE:
1280                 if (!ln)
1281                         break;
1282                 /* leave from solicited node multicast for proxy ND */
1283                 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1284                     (ifp->if_flags & IFF_MULTICAST) != 0) {
1285                         struct in6_addr llsol;
1286                         struct in6_multi *in6m;
1287
1288                         llsol = SIN6(rt_key(rt))->sin6_addr;
1289                         llsol.s6_addr16[0] = htons(0xff02);
1290                         llsol.s6_addr16[1] = htons(ifp->if_index);
1291                         llsol.s6_addr32[1] = 0;
1292                         llsol.s6_addr32[2] = htonl(1);
1293                         llsol.s6_addr8[12] = 0xff;
1294
1295                         IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1296                         if (in6m)
1297                                 in6_delmulti(in6m);
1298                 }
1299                 nd6_inuse--;
1300                 ln->ln_next->ln_prev = ln->ln_prev;
1301                 ln->ln_prev->ln_next = ln->ln_next;
1302                 ln->ln_prev = NULL;
1303                 rt->rt_llinfo = 0;
1304                 rt->rt_flags &= ~RTF_LLINFO;
1305                 if (ln->ln_hold)
1306                         m_freem(ln->ln_hold);
1307                 Free((caddr_t)ln);
1308         }
1309 }
1310
1311 int
1312 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1313 {
1314         struct in6_drlist *drl = (struct in6_drlist *)data;
1315         struct in6_prlist *prl = (struct in6_prlist *)data;
1316         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1317         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1318         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1319         struct nd_defrouter *dr, any;
1320         struct nd_prefix *pr;
1321         struct rtentry *rt;
1322         int i = 0, error = 0;
1323
1324         switch (cmd) {
1325         case SIOCGDRLST_IN6:
1326                 /*
1327                  * obsolete API, use sysctl under net.inet6.icmp6
1328                  */
1329                 bzero(drl, sizeof(*drl));
1330                 crit_enter();
1331                 dr = TAILQ_FIRST(&nd_defrouter);
1332                 while (dr && i < DRLSTSIZ) {
1333                         drl->defrouter[i].rtaddr = dr->rtaddr;
1334                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1335                                 /* XXX: need to this hack for KAME stack */
1336                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1337                         } else
1338                                 log(LOG_ERR,
1339                                     "default router list contains a "
1340                                     "non-linklocal address(%s)\n",
1341                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1342
1343                         drl->defrouter[i].flags = dr->flags;
1344                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1345                         drl->defrouter[i].expire = dr->expire;
1346                         drl->defrouter[i].if_index = dr->ifp->if_index;
1347                         i++;
1348                         dr = TAILQ_NEXT(dr, dr_entry);
1349                 }
1350                 crit_exit();
1351                 break;
1352         case SIOCGPRLST_IN6:
1353                 /*
1354                  * obsolete API, use sysctl under net.inet6.icmp6
1355                  */
1356                 /*
1357                  * XXX meaning of fields, especialy "raflags", is very
1358                  * differnet between RA prefix list and RR/static prefix list.
1359                  * how about separating ioctls into two?
1360                  */
1361                 bzero(prl, sizeof(*prl));
1362                 crit_enter();
1363                 pr = nd_prefix.lh_first;
1364                 while (pr && i < PRLSTSIZ) {
1365                         struct nd_pfxrouter *pfr;
1366                         int j;
1367
1368                         (void)in6_embedscope(&prl->prefix[i].prefix,
1369                             &pr->ndpr_prefix, NULL, NULL);
1370                         prl->prefix[i].raflags = pr->ndpr_raf;
1371                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1372                         prl->prefix[i].vltime = pr->ndpr_vltime;
1373                         prl->prefix[i].pltime = pr->ndpr_pltime;
1374                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1375                         prl->prefix[i].expire = pr->ndpr_expire;
1376
1377                         pfr = pr->ndpr_advrtrs.lh_first;
1378                         j = 0;
1379                         while (pfr) {
1380                                 if (j < DRLSTSIZ) {
1381 #define RTRADDR prl->prefix[i].advrtr[j]
1382                                         RTRADDR = pfr->router->rtaddr;
1383                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1384                                                 /* XXX: hack for KAME */
1385                                                 RTRADDR.s6_addr16[1] = 0;
1386                                         } else
1387                                                 log(LOG_ERR,
1388                                                     "a router(%s) advertises "
1389                                                     "a prefix with "
1390                                                     "non-link local address\n",
1391                                                     ip6_sprintf(&RTRADDR));
1392 #undef RTRADDR
1393                                 }
1394                                 j++;
1395                                 pfr = pfr->pfr_next;
1396                         }
1397                         prl->prefix[i].advrtrs = j;
1398                         prl->prefix[i].origin = PR_ORIG_RA;
1399
1400                         i++;
1401                         pr = pr->ndpr_next;
1402                 }
1403               {
1404                 struct rr_prefix *rpp;
1405
1406                 for (rpp = LIST_FIRST(&rr_prefix); rpp;
1407                      rpp = LIST_NEXT(rpp, rp_entry)) {
1408                         if (i >= PRLSTSIZ)
1409                                 break;
1410                         (void)in6_embedscope(&prl->prefix[i].prefix,
1411                             &pr->ndpr_prefix, NULL, NULL);
1412                         prl->prefix[i].raflags = rpp->rp_raf;
1413                         prl->prefix[i].prefixlen = rpp->rp_plen;
1414                         prl->prefix[i].vltime = rpp->rp_vltime;
1415                         prl->prefix[i].pltime = rpp->rp_pltime;
1416                         prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1417                         prl->prefix[i].expire = rpp->rp_expire;
1418                         prl->prefix[i].advrtrs = 0;
1419                         prl->prefix[i].origin = rpp->rp_origin;
1420                         i++;
1421                 }
1422               }
1423                 crit_exit();
1424
1425                 break;
1426         case OSIOCGIFINFO_IN6:
1427                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1428                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1429                 ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1430                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1431                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1432                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1433                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1434                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1435                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1436                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1437                 ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1438                 break;
1439         case SIOCGIFINFO_IN6:
1440                 ndi->ndi = *ND_IFINFO(ifp);
1441                 break;
1442         case SIOCSIFINFO_FLAGS:
1443                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1444                 break;
1445         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1446                 /* flush default router list */
1447                 /*
1448                  * xxx sumikawa: should not delete route if default
1449                  * route equals to the top of default router list
1450                  */
1451                 bzero(&any, sizeof(any));
1452                 defrouter_delreq(&any, 0);
1453                 defrouter_select();
1454                 /* xxx sumikawa: flush prefix list */
1455                 break;
1456         case SIOCSPFXFLUSH_IN6:
1457             {
1458                 /* flush all the prefix advertised by routers */
1459                 struct nd_prefix *pr, *next;
1460
1461                 crit_enter();
1462                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1463                         struct in6_ifaddr *ia, *ia_next;
1464
1465                         next = pr->ndpr_next;
1466
1467                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1468                                 continue; /* XXX */
1469
1470                         /* do we really have to remove addresses as well? */
1471                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1472                                 /* ia might be removed.  keep the next ptr. */
1473                                 ia_next = ia->ia_next;
1474
1475                                 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1476                                         continue;
1477
1478                                 if (ia->ia6_ndpr == pr)
1479                                         in6_purgeaddr(&ia->ia_ifa);
1480                         }
1481                         prelist_remove(pr);
1482                 }
1483                 crit_exit();
1484                 break;
1485             }
1486         case SIOCSRTRFLUSH_IN6:
1487             {
1488                 /* flush all the default routers */
1489                 struct nd_defrouter *dr, *next;
1490
1491                 crit_enter();
1492                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1493                         /*
1494                          * The first entry of the list may be stored in
1495                          * the routing table, so we'll delete it later.
1496                          */
1497                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1498                                 next = TAILQ_NEXT(dr, dr_entry);
1499                                 defrtrlist_del(dr);
1500                         }
1501                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1502                 }
1503                 crit_exit();
1504                 break;
1505             }
1506         case SIOCGNBRINFO_IN6:
1507             {
1508                 struct llinfo_nd6 *ln;
1509                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1510
1511                 /*
1512                  * XXX: KAME specific hack for scoped addresses
1513                  *      XXXX: for other scopes than link-local?
1514                  */
1515                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1516                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1517                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1518
1519                         if (*idp == 0)
1520                                 *idp = htons(ifp->if_index);
1521                 }
1522
1523                 crit_enter();
1524                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1525                         error = EINVAL;
1526                         crit_exit();
1527                         break;
1528                 }
1529                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1530                 nbi->state = ln->ln_state;
1531                 nbi->asked = ln->ln_asked;
1532                 nbi->isrouter = ln->ln_router;
1533                 nbi->expire = ln->ln_expire;
1534                 crit_exit();
1535                 
1536                 break;
1537             }
1538         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1539                 ndif->ifindex = nd6_defifindex;
1540                 break;
1541         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1542                 return(nd6_setdefaultiface(ndif->ifindex));
1543                 break;
1544         }
1545         return(error);
1546 }
1547
1548 /*
1549  * Create neighbor cache entry and cache link-layer address,
1550  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1551  */
1552 struct rtentry *
1553 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1554                  int lladdrlen,
1555                  int type,      /* ICMP6 type */
1556                  int code       /* type dependent information */)
1557 {
1558         struct rtentry *rt = NULL;
1559         struct llinfo_nd6 *ln = NULL;
1560         int is_newentry;
1561         struct sockaddr_dl *sdl = NULL;
1562         int do_update;
1563         int olladdr;
1564         int llchange;
1565         int newstate = 0;
1566
1567         if (!ifp)
1568                 panic("ifp == NULL in nd6_cache_lladdr");
1569         if (!from)
1570                 panic("from == NULL in nd6_cache_lladdr");
1571
1572         /* nothing must be updated for unspecified address */
1573         if (IN6_IS_ADDR_UNSPECIFIED(from))
1574                 return NULL;
1575
1576         /*
1577          * Validation about ifp->if_addrlen and lladdrlen must be done in
1578          * the caller.
1579          *
1580          * XXX If the link does not have link-layer adderss, what should
1581          * we do? (ifp->if_addrlen == 0)
1582          * Spec says nothing in sections for RA, RS and NA.  There's small
1583          * description on it in NS section (RFC 2461 7.2.3).
1584          */
1585
1586         rt = nd6_lookup(from, 0, ifp);
1587         if (!rt) {
1588 #if 0
1589                 /* nothing must be done if there's no lladdr */
1590                 if (!lladdr || !lladdrlen)
1591                         return NULL;
1592 #endif
1593
1594                 rt = nd6_lookup(from, 1, ifp);
1595                 is_newentry = 1;
1596         } else {
1597                 /* do nothing if static ndp is set */
1598                 if (rt->rt_flags & RTF_STATIC)
1599                         return NULL;
1600                 is_newentry = 0;
1601         }
1602
1603         if (!rt)
1604                 return NULL;
1605         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1606 fail:
1607                 (void)nd6_free(rt);
1608                 return NULL;
1609         }
1610         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1611         if (!ln)
1612                 goto fail;
1613         if (!rt->rt_gateway)
1614                 goto fail;
1615         if (rt->rt_gateway->sa_family != AF_LINK)
1616                 goto fail;
1617         sdl = SDL(rt->rt_gateway);
1618
1619         olladdr = (sdl->sdl_alen) ? 1 : 0;
1620         if (olladdr && lladdr) {
1621                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1622                         llchange = 1;
1623                 else
1624                         llchange = 0;
1625         } else
1626                 llchange = 0;
1627
1628         /*
1629          * newentry olladdr  lladdr  llchange   (*=record)
1630          *      0       n       n       --      (1)
1631          *      0       y       n       --      (2)
1632          *      0       n       y       --      (3) * STALE
1633          *      0       y       y       n       (4) *
1634          *      0       y       y       y       (5) * STALE
1635          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1636          *      1       --      y       --      (7) * STALE
1637          */
1638
1639         if (lladdr) {           /* (3-5) and (7) */
1640                 /*
1641                  * Record source link-layer address
1642                  * XXX is it dependent to ifp->if_type?
1643                  */
1644                 sdl->sdl_alen = ifp->if_addrlen;
1645                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1646         }
1647
1648         if (!is_newentry) {
1649                 if ((!olladdr && lladdr)                /* (3) */
1650                  || (olladdr && lladdr && llchange)) {  /* (5) */
1651                         do_update = 1;
1652                         newstate = ND6_LLINFO_STALE;
1653                 } else                                  /* (1-2,4) */
1654                         do_update = 0;
1655         } else {
1656                 do_update = 1;
1657                 if (!lladdr)                            /* (6) */
1658                         newstate = ND6_LLINFO_NOSTATE;
1659                 else                                    /* (7) */
1660                         newstate = ND6_LLINFO_STALE;
1661         }
1662
1663         if (do_update) {
1664                 /*
1665                  * Update the state of the neighbor cache.
1666                  */
1667                 ln->ln_state = newstate;
1668
1669                 if (ln->ln_state == ND6_LLINFO_STALE) {
1670                         /*
1671                          * XXX: since nd6_output() below will cause
1672                          * state tansition to DELAY and reset the timer,
1673                          * we must set the timer now, although it is actually
1674                          * meaningless.
1675                          */
1676                         ln->ln_expire = time_second + nd6_gctimer;
1677
1678                         if (ln->ln_hold) {
1679                                 /*
1680                                  * we assume ifp is not a p2p here, so just
1681                                  * set the 2nd argument as the 1st one.
1682                                  */
1683                                 nd6_output(ifp, ifp, ln->ln_hold,
1684                                            (struct sockaddr_in6 *)rt_key(rt),
1685                                            rt);
1686                                 ln->ln_hold = NULL;
1687                         }
1688                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1689                         /* probe right away */
1690                         ln->ln_expire = time_second;
1691                 }
1692         }
1693
1694         /*
1695          * ICMP6 type dependent behavior.
1696          *
1697          * NS: clear IsRouter if new entry
1698          * RS: clear IsRouter
1699          * RA: set IsRouter if there's lladdr
1700          * redir: clear IsRouter if new entry
1701          *
1702          * RA case, (1):
1703          * The spec says that we must set IsRouter in the following cases:
1704          * - If lladdr exist, set IsRouter.  This means (1-5).
1705          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1706          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1707          * A quetion arises for (1) case.  (1) case has no lladdr in the
1708          * neighbor cache, this is similar to (6).
1709          * This case is rare but we figured that we MUST NOT set IsRouter.
1710          *
1711          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1712          *                                                      D R
1713          *      0       n       n       --      (1)     c   ?     s
1714          *      0       y       n       --      (2)     c   s     s
1715          *      0       n       y       --      (3)     c   s     s
1716          *      0       y       y       n       (4)     c   s     s
1717          *      0       y       y       y       (5)     c   s     s
1718          *      1       --      n       --      (6) c   c       c s
1719          *      1       --      y       --      (7) c   c   s   c s
1720          *
1721          *                                      (c=clear s=set)
1722          */
1723         switch (type & 0xff) {
1724         case ND_NEIGHBOR_SOLICIT:
1725                 /*
1726                  * New entry must have is_router flag cleared.
1727                  */
1728                 if (is_newentry)        /* (6-7) */
1729                         ln->ln_router = 0;
1730                 break;
1731         case ND_REDIRECT:
1732                 /*
1733                  * If the icmp is a redirect to a better router, always set the
1734                  * is_router flag. Otherwise, if the entry is newly created,
1735                  * clear the flag. [RFC 2461, sec 8.3]
1736                  */
1737                 if (code == ND_REDIRECT_ROUTER)
1738                         ln->ln_router = 1;
1739                 else if (is_newentry) /* (6-7) */
1740                         ln->ln_router = 0;
1741                 break;
1742         case ND_ROUTER_SOLICIT:
1743                 /*
1744                  * is_router flag must always be cleared.
1745                  */
1746                 ln->ln_router = 0;
1747                 break;
1748         case ND_ROUTER_ADVERT:
1749                 /*
1750                  * Mark an entry with lladdr as a router.
1751                  */
1752                 if ((!is_newentry && (olladdr || lladdr))       /* (2-5) */
1753                  || (is_newentry && lladdr)) {                  /* (7) */
1754                         ln->ln_router = 1;
1755                 }
1756                 break;
1757         }
1758
1759         /*
1760          * When the link-layer address of a router changes, select the
1761          * best router again.  In particular, when the neighbor entry is newly
1762          * created, it might affect the selection policy.
1763          * Question: can we restrict the first condition to the "is_newentry"
1764          * case?
1765          * XXX: when we hear an RA from a new router with the link-layer
1766          * address option, defrouter_select() is called twice, since
1767          * defrtrlist_update called the function as well.  However, I believe
1768          * we can compromise the overhead, since it only happens the first
1769          * time.
1770          * XXX: although defrouter_select() should not have a bad effect
1771          * for those are not autoconfigured hosts, we explicitly avoid such
1772          * cases for safety.
1773          */
1774         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1775                 defrouter_select();
1776
1777         return rt;
1778 }
1779
1780 static void
1781 nd6_slowtimo(void *ignored_arg)
1782 {
1783         struct nd_ifinfo *nd6if;
1784         struct ifnet *ifp;
1785
1786         crit_enter();
1787         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1788             nd6_slowtimo, NULL);
1789         for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1790                 nd6if = ND_IFINFO(ifp);
1791                 if (nd6if->basereachable && /* already initialized */
1792                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1793                         /*
1794                          * Since reachable time rarely changes by router
1795                          * advertisements, we SHOULD insure that a new random
1796                          * value gets recomputed at least once every few hours.
1797                          * (RFC 2461, 6.3.4)
1798                          */
1799                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1800                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1801                 }
1802         }
1803         crit_exit();
1804 }
1805
1806 #define senderr(e) { error = (e); goto bad;}
1807 int
1808 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1809            struct sockaddr_in6 *dst, struct rtentry *rt0)
1810 {
1811         struct mbuf *m = m0;
1812         struct rtentry *rt = rt0;
1813         struct sockaddr_in6 *gw6 = NULL;
1814         struct llinfo_nd6 *ln = NULL;
1815         int error = 0;
1816
1817         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1818                 goto sendpkt;
1819
1820         if (nd6_need_cache(ifp) == 0)
1821                 goto sendpkt;
1822
1823         /*
1824          * next hop determination.  This routine is derived from ether_outpout.
1825          */
1826         if (rt) {
1827                 if (!(rt->rt_flags & RTF_UP)) {
1828                         if ((rt0 = rt = rtlookup((struct sockaddr *)dst))) {
1829                                 rt->rt_refcnt--;
1830                                 if (rt->rt_ifp != ifp) {
1831                                         /* XXX: loop care? */
1832                                         return nd6_output(ifp, origifp, m0,
1833                                                           dst, rt);
1834                                 }
1835                         } else
1836                                 senderr(EHOSTUNREACH);
1837                 }
1838
1839                 if (rt->rt_flags & RTF_GATEWAY) {
1840                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1841
1842                         /*
1843                          * We skip link-layer address resolution and NUD
1844                          * if the gateway is not a neighbor from ND point
1845                          * of view, regardless of the value of nd_ifinfo.flags.
1846                          * The second condition is a bit tricky; we skip
1847                          * if the gateway is our own address, which is
1848                          * sometimes used to install a route to a p2p link.
1849                          */
1850                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1851                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1852                                 /*
1853                                  * We allow this kind of tricky route only
1854                                  * when the outgoing interface is p2p.
1855                                  * XXX: we may need a more generic rule here.
1856                                  */
1857                                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1858                                         senderr(EHOSTUNREACH);
1859
1860                                 goto sendpkt;
1861                         }
1862
1863                         if (rt->rt_gwroute == NULL)
1864                                 goto lookup;
1865                         if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1866                                 rtfree(rt->rt_gwroute);
1867 lookup:                         rt->rt_gwroute = rtlookup(rt->rt_gateway);
1868                                 if (rt->rt_gwroute == NULL)
1869                                         senderr(EHOSTUNREACH);
1870                         }
1871                 }
1872         }
1873
1874         /*
1875          * Address resolution or Neighbor Unreachability Detection
1876          * for the next hop.
1877          * At this point, the destination of the packet must be a unicast
1878          * or an anycast address(i.e. not a multicast).
1879          */
1880
1881         /* Look up the neighbor cache for the nexthop */
1882         if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1883                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1884         else {
1885                 /*
1886                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1887                  * the condition below is not very efficient.  But we believe
1888                  * it is tolerable, because this should be a rare case.
1889                  */
1890                 if (nd6_is_addr_neighbor(dst, ifp) &&
1891                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1892                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1893         }
1894         if (!ln || !rt) {
1895                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1896                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1897                         log(LOG_DEBUG,
1898                             "nd6_output: can't allocate llinfo for %s "
1899                             "(ln=%p, rt=%p)\n",
1900                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1901                         senderr(EIO);   /* XXX: good error? */
1902                 }
1903
1904                 goto sendpkt;   /* send anyway */
1905         }
1906
1907         /* We don't have to do link-layer address resolution on a p2p link. */
1908         if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1909             ln->ln_state < ND6_LLINFO_REACHABLE) {
1910                 ln->ln_state = ND6_LLINFO_STALE;
1911                 ln->ln_expire = time_second + nd6_gctimer;
1912         }
1913
1914         /*
1915          * The first time we send a packet to a neighbor whose entry is
1916          * STALE, we have to change the state to DELAY and a sets a timer to
1917          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1918          * neighbor unreachability detection on expiration.
1919          * (RFC 2461 7.3.3)
1920          */
1921         if (ln->ln_state == ND6_LLINFO_STALE) {
1922                 ln->ln_asked = 0;
1923                 ln->ln_state = ND6_LLINFO_DELAY;
1924                 ln->ln_expire = time_second + nd6_delay;
1925         }
1926
1927         /*
1928          * If the neighbor cache entry has a state other than INCOMPLETE
1929          * (i.e. its link-layer address is already resolved), just
1930          * send the packet.
1931          */
1932         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1933                 goto sendpkt;
1934
1935         /*
1936          * There is a neighbor cache entry, but no ethernet address
1937          * response yet.  Replace the held mbuf (if any) with this
1938          * latest one.
1939          *
1940          * This code conforms to the rate-limiting rule described in Section
1941          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1942          * an NS below.
1943          */
1944         if (ln->ln_state == ND6_LLINFO_NOSTATE)
1945                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1946         if (ln->ln_hold)
1947                 m_freem(ln->ln_hold);
1948         ln->ln_hold = m;
1949         if (ln->ln_expire) {
1950                 if (ln->ln_asked < nd6_mmaxtries &&
1951                     ln->ln_expire < time_second) {
1952                         ln->ln_asked++;
1953                         ln->ln_expire = time_second +
1954                                 ND_IFINFO(ifp)->retrans / 1000;
1955                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1956                 }
1957         }
1958         return(0);
1959         
1960   sendpkt:
1961
1962         lwkt_serialize_enter(ifp->if_serializer);
1963         if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1964                 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1965                                           rt);
1966         } else {
1967                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt);
1968         }
1969         lwkt_serialize_exit(ifp->if_serializer);
1970         return (error);
1971
1972   bad:
1973         if (m)
1974                 m_freem(m);
1975         return (error);
1976 }
1977 #undef senderr
1978
1979 int
1980 nd6_need_cache(struct ifnet *ifp)
1981 {
1982         /*
1983          * XXX: we currently do not make neighbor cache on any interface
1984          * other than ARCnet, Ethernet, FDDI and GIF.
1985          *
1986          * RFC2893 says:
1987          * - unidirectional tunnels needs no ND
1988          */
1989         switch (ifp->if_type) {
1990         case IFT_ARCNET:
1991         case IFT_ETHER:
1992         case IFT_FDDI:
1993         case IFT_IEEE1394:
1994 #ifdef IFT_L2VLAN
1995         case IFT_L2VLAN:
1996 #endif
1997 #ifdef IFT_IEEE80211
1998         case IFT_IEEE80211:
1999 #endif
2000         case IFT_GIF:           /* XXX need more cases? */
2001                 return(1);
2002         default:
2003                 return(0);
2004         }
2005 }
2006
2007 int
2008 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2009                 struct sockaddr *dst, u_char *desten)
2010 {
2011         struct sockaddr_dl *sdl;
2012         struct rtentry *rt;
2013
2014
2015         if (m->m_flags & M_MCAST) {
2016                 switch (ifp->if_type) {
2017                 case IFT_ETHER:
2018                 case IFT_FDDI:
2019 #ifdef IFT_L2VLAN
2020         case IFT_L2VLAN:
2021 #endif
2022 #ifdef IFT_IEEE80211
2023                 case IFT_IEEE80211:
2024 #endif
2025                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2026                                                  desten);
2027                         return(1);
2028                 case IFT_IEEE1394:
2029                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen); 
2030                         return(1);
2031                 case IFT_ARCNET:
2032                         *desten = 0;
2033                         return(1);
2034                 default:
2035                         m_freem(m);
2036                         return(0);
2037                 }
2038         }
2039         if (rt0 == NULL) {
2040                 /* this could happen, if we could not allocate memory */
2041                 m_freem(m);
2042                 return(0);
2043         }
2044         if (rt_llroute(dst, rt0, &rt) != 0) {
2045                 m_freem(m);
2046                 return (0);
2047         }
2048         if (rt->rt_gateway->sa_family != AF_LINK) {
2049                 printf("nd6_storelladdr: something odd happens\n");
2050                 m_freem(m);
2051                 return(0);
2052         }
2053         sdl = SDL(rt->rt_gateway);
2054         if (sdl->sdl_alen == 0) {
2055                 /* this should be impossible, but we bark here for debugging */
2056                 printf("nd6_storelladdr: sdl_alen == 0\n");
2057                 m_freem(m);
2058                 return(0);
2059         }
2060
2061         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2062         return(1);
2063 }
2064
2065 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2066 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2067 #ifdef SYSCTL_DECL
2068 SYSCTL_DECL(_net_inet6_icmp6);
2069 #endif
2070 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2071         CTLFLAG_RD, nd6_sysctl_drlist, "");
2072 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2073         CTLFLAG_RD, nd6_sysctl_prlist, "");
2074
2075 static int
2076 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2077 {
2078         int error;
2079         char buf[1024];
2080         struct in6_defrouter *d, *de;
2081         struct nd_defrouter *dr;
2082
2083         if (req->newptr)
2084                 return EPERM;
2085         error = 0;
2086
2087         for (dr = TAILQ_FIRST(&nd_defrouter);
2088              dr;
2089              dr = TAILQ_NEXT(dr, dr_entry)) {
2090                 d = (struct in6_defrouter *)buf;
2091                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2092
2093                 if (d + 1 <= de) {
2094                         bzero(d, sizeof(*d));
2095                         d->rtaddr.sin6_family = AF_INET6;
2096                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2097                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2098                             dr->ifp) != 0)
2099                                 log(LOG_ERR,
2100                                     "scope error in "
2101                                     "default router list (%s)\n",
2102                                     ip6_sprintf(&dr->rtaddr));
2103                         d->flags = dr->flags;
2104                         d->rtlifetime = dr->rtlifetime;
2105                         d->expire = dr->expire;
2106                         d->if_index = dr->ifp->if_index;
2107                 } else
2108                         panic("buffer too short");
2109
2110                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2111                 if (error)
2112                         break;
2113         }
2114         return error;
2115 }
2116
2117 static int
2118 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2119 {
2120         int error;
2121         char buf[1024];
2122         struct in6_prefix *p, *pe;
2123         struct nd_prefix *pr;
2124
2125         if (req->newptr)
2126                 return EPERM;
2127         error = 0;
2128
2129         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2130                 u_short advrtrs;
2131                 size_t advance;
2132                 struct sockaddr_in6 *sin6, *s6;
2133                 struct nd_pfxrouter *pfr;
2134
2135                 p = (struct in6_prefix *)buf;
2136                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2137
2138                 if (p + 1 <= pe) {
2139                         bzero(p, sizeof(*p));
2140                         sin6 = (struct sockaddr_in6 *)(p + 1);
2141
2142                         p->prefix = pr->ndpr_prefix;
2143                         if (in6_recoverscope(&p->prefix,
2144                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2145                                 log(LOG_ERR,
2146                                     "scope error in prefix list (%s)\n",
2147                                     ip6_sprintf(&p->prefix.sin6_addr));
2148                         p->raflags = pr->ndpr_raf;
2149                         p->prefixlen = pr->ndpr_plen;
2150                         p->vltime = pr->ndpr_vltime;
2151                         p->pltime = pr->ndpr_pltime;
2152                         p->if_index = pr->ndpr_ifp->if_index;
2153                         p->expire = pr->ndpr_expire;
2154                         p->refcnt = pr->ndpr_refcnt;
2155                         p->flags = pr->ndpr_stateflags;
2156                         p->origin = PR_ORIG_RA;
2157                         advrtrs = 0;
2158                         for (pfr = pr->ndpr_advrtrs.lh_first;
2159                              pfr;
2160                              pfr = pfr->pfr_next) {
2161                                 if ((void *)&sin6[advrtrs + 1] >
2162                                     (void *)pe) {
2163                                         advrtrs++;
2164                                         continue;
2165                                 }
2166                                 s6 = &sin6[advrtrs];
2167                                 bzero(s6, sizeof(*s6));
2168                                 s6->sin6_family = AF_INET6;
2169                                 s6->sin6_len = sizeof(*sin6);
2170                                 if (in6_recoverscope(s6,
2171                                     &pfr->router->rtaddr,
2172                                     pfr->router->ifp) != 0)
2173                                         log(LOG_ERR,
2174                                             "scope error in "
2175                                             "prefix list (%s)\n",
2176                                             ip6_sprintf(&pfr->router->rtaddr));
2177                                 advrtrs++;
2178                         }
2179                         p->advrtrs = advrtrs;
2180                 } else 
2181                         panic("buffer too short");
2182
2183                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2184                 error = SYSCTL_OUT(req, buf, advance);
2185                 if (error)
2186                         break;
2187         }
2188         return error;
2189 }