route: Seperate route messages from creation/lookup.
[dragonfly.git] / sys / netinet6 / nd6.c
1 /*      $FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $       */
2 /*      $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $   */
3
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/callout.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/syslog.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/mutex.h>
51
52 #include <sys/thread2.h>
53 #include <sys/mutex2.h>
54
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/route.h>
59 #include <net/netisr2.h>
60 #include <net/netmsg2.h>
61
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69
70 #include <net/net_osdep.h>
71
72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
74
75 #define SIN6(s) ((struct sockaddr_in6 *)s)
76 #define SDL(s) ((struct sockaddr_dl *)s)
77
78 /*
79  * Note that the check for rt_llinfo is necessary because a cloned
80  * route from a parent route that has the L flag (e.g. the default
81  * route to a p2p interface) may have the flag, too, while the
82  * destination is not actually a neighbor.
83  * XXX: we can't use rt->rt_ifp to check for the interface, since
84  *      it might be the loopback interface if the entry is for our
85  *      own address on a non-loopback interface. Instead, we should
86  *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
87  *      interface.
88  */
89 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)                        \
90         !(((rt)->rt_flags & RTF_GATEWAY) ||                     \
91           ((rt)->rt_flags & RTF_LLINFO) == 0 ||                 \
92           (rt)->rt_gateway->sa_family != AF_LINK ||             \
93           (rt)->rt_llinfo == NULL ||                            \
94           ((ifp) != NULL && (rt)->rt_ifa->ifa_ifp != (ifp)))
95
96 #define ND6_RTENTRY_IS_LLCLONING(rt)                            \
97         (((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==     \
98          (RTF_PRCLONING | RTF_LLINFO) ||                        \
99          ((rt)->rt_flags & RTF_CLONING))
100
101 /* timer values */
102 int     nd6_prune       = 1;    /* walk list every 1 seconds */
103 int     nd6_delay       = 5;    /* delay first probe time 5 second */
104 int     nd6_umaxtries   = 3;    /* maximum unicast query */
105 int     nd6_mmaxtries   = 3;    /* maximum multicast query */
106 int     nd6_useloopback = 1;    /* use loopback interface for local traffic */
107 int     nd6_gctimer     = (60 * 60 * 24); /* 1 day: garbage collection timer */
108
109 /* preventing too many loops in ND option parsing */
110 int nd6_maxndopt = 10;  /* max # of ND options allowed */
111
112 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */
113
114 #ifdef ND6_DEBUG
115 int nd6_debug = 1;
116 #else
117 int nd6_debug = 0;
118 #endif
119
120 /* for debugging? */
121 static int nd6_inuse, nd6_allocated;
122
123 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
124 struct nd_drhead nd_defrouter;
125 struct nd_prhead nd_prefix = { 0 };
126 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
127
128 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
129 static struct sockaddr_in6 all1_sa;
130
131 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
132 static int regen_tmpaddr (struct in6_ifaddr *);
133 static void nd6_slowtimo(void *);
134 static void nd6_slowtimo_dispatch(netmsg_t);
135 static void nd6_timer(void *);
136 static void nd6_timer_dispatch(netmsg_t);
137
138 static struct callout nd6_slowtimo_ch;
139 static struct netmsg_base nd6_slowtimo_netmsg;
140
141 static struct callout nd6_timer_ch;
142 static struct netmsg_base nd6_timer_netmsg;
143
144 void
145 nd6_init(void)
146 {
147         static int nd6_init_done = 0;
148         int i;
149
150         if (nd6_init_done) {
151                 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
152                 return;
153         }
154
155         all1_sa.sin6_family = AF_INET6;
156         all1_sa.sin6_len = sizeof(struct sockaddr_in6);
157         for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
158                 all1_sa.sin6_addr.s6_addr[i] = 0xff;
159
160         /* initialization of the default router list */
161         TAILQ_INIT(&nd_defrouter);
162
163         nd6_init_done = 1;
164
165         /* start timer */
166         callout_init_mp(&nd6_slowtimo_ch);
167         netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
168             MSGF_PRIORITY, nd6_slowtimo_dispatch);
169         callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
170             nd6_slowtimo, NULL, 0);
171 }
172
173 struct nd_ifinfo *
174 nd6_ifattach(struct ifnet *ifp)
175 {
176         struct nd_ifinfo *nd;
177
178         nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
179             M_WAITOK | M_ZERO);
180
181         nd->initialized = 1;
182
183         nd->chlim = IPV6_DEFHLIM;
184         nd->basereachable = REACHABLE_TIME;
185         nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
186         nd->retrans = RETRANS_TIMER;
187
188         /*
189          * Note that the default value of ip6_accept_rtadv is 0, which means
190          * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
191          * here.
192          */
193         nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
194
195         /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
196         nd6_setmtu0(ifp, nd);
197         return nd;
198 }
199
200 void
201 nd6_ifdetach(struct nd_ifinfo *nd)
202 {
203         kfree(nd, M_IP6NDP);
204 }
205
206 /*
207  * Reset ND level link MTU. This function is called when the physical MTU
208  * changes, which means we might have to adjust the ND level MTU.
209  */
210 void
211 nd6_setmtu(struct ifnet *ifp)
212 {
213         nd6_setmtu0(ifp, ND_IFINFO(ifp));
214 }
215
216 struct netmsg_nd6setmtu {
217         struct netmsg_base      nmsg;
218         struct ifnet            *ifp;
219         struct nd_ifinfo        *ndi;
220 };
221
222 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
223 static void
224 nd6_setmtu0_dispatch(netmsg_t msg)
225 {
226         struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
227         struct ifnet *ifp = nmsg->ifp;
228         struct nd_ifinfo *ndi = nmsg->ndi;
229         uint32_t omaxmtu;
230
231         omaxmtu = ndi->maxmtu;
232
233         switch (ifp->if_type) {
234         case IFT_ETHER:
235                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
236                 break;
237         case IFT_IEEE1394:      /* XXX should be IEEE1394MTU(1500) */
238                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
239                 break;
240 #ifdef IFT_IEEE80211
241         case IFT_IEEE80211:     /* XXX should be IEEE80211MTU(1500) */
242                 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
243                 break;
244 #endif
245         default:
246                 ndi->maxmtu = ifp->if_mtu;
247                 break;
248         }
249
250         /*
251          * Decreasing the interface MTU under IPV6 minimum MTU may cause
252          * undesirable situation.  We thus notify the operator of the change
253          * explicitly.  The check for omaxmtu is necessary to restrict the
254          * log to the case of changing the MTU, not initializing it.
255          */
256         if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
257                 log(LOG_NOTICE, "nd6_setmtu0: "
258                     "new link MTU on %s (%lu) is too small for IPv6\n",
259                     if_name(ifp), (unsigned long)ndi->maxmtu);
260         }
261
262         if (ndi->maxmtu > in6_maxmtu)
263                 in6_setmaxmtu(); /* check all interfaces just in case */
264
265         lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
266 }
267
268 void
269 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
270 {
271         struct netmsg_nd6setmtu nmsg;
272
273         netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
274             nd6_setmtu0_dispatch);
275         nmsg.ifp = ifp;
276         nmsg.ndi = ndi;
277         lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
278 }
279
280 void
281 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
282 {
283         bzero(ndopts, sizeof(*ndopts));
284         ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
285         ndopts->nd_opts_last
286                 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
287
288         if (icmp6len == 0) {
289                 ndopts->nd_opts_done = 1;
290                 ndopts->nd_opts_search = NULL;
291         }
292 }
293
294 /*
295  * Take one ND option.
296  */
297 struct nd_opt_hdr *
298 nd6_option(union nd_opts *ndopts)
299 {
300         struct nd_opt_hdr *nd_opt;
301         int olen;
302
303         if (!ndopts)
304                 panic("ndopts == NULL in nd6_option");
305         if (!ndopts->nd_opts_last)
306                 panic("uninitialized ndopts in nd6_option");
307         if (!ndopts->nd_opts_search)
308                 return NULL;
309         if (ndopts->nd_opts_done)
310                 return NULL;
311
312         nd_opt = ndopts->nd_opts_search;
313
314         /* make sure nd_opt_len is inside the buffer */
315         if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
316                 bzero(ndopts, sizeof(*ndopts));
317                 return NULL;
318         }
319
320         olen = nd_opt->nd_opt_len << 3;
321         if (olen == 0) {
322                 /*
323                  * Message validation requires that all included
324                  * options have a length that is greater than zero.
325                  */
326                 bzero(ndopts, sizeof(*ndopts));
327                 return NULL;
328         }
329
330         ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
331         if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
332                 /* option overruns the end of buffer, invalid */
333                 bzero(ndopts, sizeof(*ndopts));
334                 return NULL;
335         } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
336                 /* reached the end of options chain */
337                 ndopts->nd_opts_done = 1;
338                 ndopts->nd_opts_search = NULL;
339         }
340         return nd_opt;
341 }
342
343 /*
344  * Parse multiple ND options.
345  * This function is much easier to use, for ND routines that do not need
346  * multiple options of the same type.
347  */
348 int
349 nd6_options(union nd_opts *ndopts)
350 {
351         struct nd_opt_hdr *nd_opt;
352         int i = 0;
353
354         if (!ndopts)
355                 panic("ndopts == NULL in nd6_options");
356         if (!ndopts->nd_opts_last)
357                 panic("uninitialized ndopts in nd6_options");
358         if (!ndopts->nd_opts_search)
359                 return 0;
360
361         while (1) {
362                 nd_opt = nd6_option(ndopts);
363                 if (!nd_opt && !ndopts->nd_opts_last) {
364                         /*
365                          * Message validation requires that all included
366                          * options have a length that is greater than zero.
367                          */
368                         icmp6stat.icp6s_nd_badopt++;
369                         bzero(ndopts, sizeof(*ndopts));
370                         return -1;
371                 }
372
373                 if (!nd_opt)
374                         goto skip1;
375
376                 switch (nd_opt->nd_opt_type) {
377                 case ND_OPT_SOURCE_LINKADDR:
378                 case ND_OPT_TARGET_LINKADDR:
379                 case ND_OPT_MTU:
380                 case ND_OPT_REDIRECTED_HEADER:
381                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
382                                 nd6log((LOG_INFO,
383                                     "duplicated ND6 option found (type=%d)\n",
384                                     nd_opt->nd_opt_type));
385                                 /* XXX bark? */
386                         } else {
387                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
388                                         = nd_opt;
389                         }
390                         break;
391                 case ND_OPT_PREFIX_INFORMATION:
392                         if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
393                                 ndopts->nd_opt_array[nd_opt->nd_opt_type]
394                                         = nd_opt;
395                         }
396                         ndopts->nd_opts_pi_end =
397                                 (struct nd_opt_prefix_info *)nd_opt;
398                         break;
399                 default:
400                         /*
401                          * Unknown options must be silently ignored,
402                          * to accomodate future extension to the protocol.
403                          */
404                         nd6log((LOG_DEBUG,
405                             "nd6_options: unsupported option %d - "
406                             "option ignored\n", nd_opt->nd_opt_type));
407                 }
408
409 skip1:
410                 i++;
411                 if (i > nd6_maxndopt) {
412                         icmp6stat.icp6s_nd_toomanyopt++;
413                         nd6log((LOG_INFO, "too many loop in nd opt\n"));
414                         break;
415                 }
416
417                 if (ndopts->nd_opts_done)
418                         break;
419         }
420
421         return 0;
422 }
423
424 /*
425  * ND6 timer routine to expire default route list and prefix list
426  */
427 static void
428 nd6_timer_dispatch(netmsg_t nmsg)
429 {
430         struct llinfo_nd6 *ln;
431         struct nd_defrouter *dr;
432         struct nd_prefix *pr;
433         struct ifnet *ifp;
434         struct in6_ifaddr *ia6, *nia6;
435
436         ASSERT_NETISR0;
437
438         crit_enter();
439         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
440         crit_exit();
441
442         mtx_lock(&nd6_mtx);
443
444         ln = llinfo_nd6.ln_next;
445         while (ln && ln != &llinfo_nd6) {
446                 struct rtentry *rt;
447                 struct sockaddr_in6 *dst;
448                 struct llinfo_nd6 *next = ln->ln_next;
449                 /* XXX: used for the DELAY case only: */
450                 struct nd_ifinfo *ndi = NULL;
451
452                 if ((rt = ln->ln_rt) == NULL) {
453                         ln = next;
454                         continue;
455                 }
456                 if ((ifp = rt->rt_ifp) == NULL) {
457                         ln = next;
458                         continue;
459                 }
460                 ndi = ND_IFINFO(ifp);
461                 dst = (struct sockaddr_in6 *)rt_key(rt);
462
463                 if (ln->ln_expire > time_uptime) {
464                         ln = next;
465                         continue;
466                 }
467
468                 /* sanity check */
469                 if (!rt)
470                         panic("rt=0 in nd6_timer(ln=%p)", ln);
471                 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
472                         panic("rt_llinfo(%p) is not equal to ln(%p)",
473                               rt->rt_llinfo, ln);
474                 if (!dst)
475                         panic("dst=0 in nd6_timer(ln=%p)", ln);
476
477                 switch (ln->ln_state) {
478                 case ND6_LLINFO_INCOMPLETE:
479                         if (ln->ln_asked < nd6_mmaxtries) {
480                                 ln->ln_asked++;
481                                 ln->ln_expire = time_uptime +
482                                         ND_IFINFO(ifp)->retrans / 1000;
483                                 nd6_ns_output(ifp, NULL, &dst->sin6_addr,
484                                         ln, 0);
485                         } else {
486                                 struct mbuf *m = ln->ln_hold;
487                                 if (m) {
488                                         if (rt->rt_ifp) {
489                                                 /*
490                                                  * Fake rcvif to make ICMP error
491                                                  * more helpful in diagnosing
492                                                  * for the receiver.
493                                                  * XXX: should we consider
494                                                  * older rcvif?
495                                                  */
496                                                 m->m_pkthdr.rcvif = rt->rt_ifp;
497                                         }
498                                         icmp6_error(m, ICMP6_DST_UNREACH,
499                                                     ICMP6_DST_UNREACH_ADDR, 0);
500                                         ln->ln_hold = NULL;
501                                 }
502                                 rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
503                                 next = nd6_free(rt);
504                         }
505                         break;
506                 case ND6_LLINFO_REACHABLE:
507                         if (ln->ln_expire) {
508                                 ln->ln_state = ND6_LLINFO_STALE;
509                                 ln->ln_expire = time_uptime + nd6_gctimer;
510                         }
511                         break;
512
513                 case ND6_LLINFO_STALE:
514                         /* Garbage Collection(RFC 2461 5.3) */
515                         if (ln->ln_expire)
516                                 next = nd6_free(rt);
517                         break;
518
519                 case ND6_LLINFO_DELAY:
520                         if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
521                                 /* We need NUD */
522                                 ln->ln_asked = 1;
523                                 ln->ln_state = ND6_LLINFO_PROBE;
524                                 ln->ln_expire = time_uptime +
525                                         ndi->retrans / 1000;
526                                 nd6_ns_output(ifp, &dst->sin6_addr,
527                                               &dst->sin6_addr,
528                                               ln, 0);
529                         } else {
530                                 ln->ln_state = ND6_LLINFO_STALE; /* XXX */
531                                 ln->ln_expire = time_uptime + nd6_gctimer;
532                         }
533                         break;
534                 case ND6_LLINFO_PROBE:
535                         if (ln->ln_asked < nd6_umaxtries) {
536                                 ln->ln_asked++;
537                                 ln->ln_expire = time_uptime +
538                                         ND_IFINFO(ifp)->retrans / 1000;
539                                 nd6_ns_output(ifp, &dst->sin6_addr,
540                                                &dst->sin6_addr, ln, 0);
541                         } else {
542                                 next = nd6_free(rt);
543                         }
544                         break;
545                 }
546                 ln = next;
547         }
548
549         /* expire default router list */
550         dr = TAILQ_FIRST(&nd_defrouter);
551         while (dr) {
552                 if (dr->expire && dr->expire < time_uptime) {
553                         struct nd_defrouter *t;
554                         t = TAILQ_NEXT(dr, dr_entry);
555                         defrtrlist_del(dr);
556                         dr = t;
557                 } else {
558                         dr = TAILQ_NEXT(dr, dr_entry);
559                 }
560         }
561
562         /*
563          * expire interface addresses.
564          * in the past the loop was inside prefix expiry processing.
565          * However, from a stricter speci-confrmance standpoint, we should
566          * rather separate address lifetimes and prefix lifetimes.
567          */
568 addrloop:
569         for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
570                 nia6 = ia6->ia_next;
571                 /* check address lifetime */
572                 if (IFA6_IS_INVALID(ia6)) {
573                         int regen = 0;
574
575                         /*
576                          * If the expiring address is temporary, try
577                          * regenerating a new one.  This would be useful when
578                          * we suspended a laptop PC, then turned it on after a
579                          * period that could invalidate all temporary
580                          * addresses.  Although we may have to restart the
581                          * loop (see below), it must be after purging the
582                          * address.  Otherwise, we'd see an infinite loop of
583                          * regeneration.
584                          */
585                         if (ip6_use_tempaddr &&
586                             (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
587                                 if (regen_tmpaddr(ia6) == 0)
588                                         regen = 1;
589                         }
590
591                         in6_purgeaddr(&ia6->ia_ifa);
592
593                         if (regen)
594                                 goto addrloop; /* XXX: see below */
595                 }
596                 if (IFA6_IS_DEPRECATED(ia6)) {
597                         int oldflags = ia6->ia6_flags;
598
599                         if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
600                                 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
601                                 in6_newaddrmsg((struct ifaddr *)ia6);
602                         }
603
604                         /*
605                          * If a temporary address has just become deprecated,
606                          * regenerate a new one if possible.
607                          */
608                         if (ip6_use_tempaddr &&
609                             (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
610                             !(oldflags & IN6_IFF_DEPRECATED)) {
611
612                                 if (regen_tmpaddr(ia6) == 0) {
613                                         /*
614                                          * A new temporary address is
615                                          * generated.
616                                          * XXX: this means the address chain
617                                          * has changed while we are still in
618                                          * the loop.  Although the change
619                                          * would not cause disaster (because
620                                          * it's not a deletion, but an
621                                          * addition,) we'd rather restart the
622                                          * loop just for safety.  Or does this
623                                          * significantly reduce performance??
624                                          */
625                                         goto addrloop;
626                                 }
627                         }
628                 } else {
629                         /*
630                          * A new RA might have made a deprecated address
631                          * preferred.
632                          */
633                         if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
634                                 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
635                                 in6_newaddrmsg((struct ifaddr *)ia6);
636                         }
637                 }
638         }
639
640         /* expire prefix list */
641         pr = nd_prefix.lh_first;
642         while (pr) {
643                 /*
644                  * check prefix lifetime.
645                  * since pltime is just for autoconf, pltime processing for
646                  * prefix is not necessary.
647                  */
648                 if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
649                         struct nd_prefix *t;
650                         t = pr->ndpr_next;
651
652                         /*
653                          * address expiration and prefix expiration are
654                          * separate.  NEVER perform in6_purgeaddr here.
655                          */
656
657                         prelist_remove(pr);
658                         pr = t;
659                 } else
660                         pr = pr->ndpr_next;
661         }
662
663         mtx_unlock(&nd6_mtx);
664
665         callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
666 }
667
668 static void
669 nd6_timer(void *arg __unused)
670 {
671         struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
672
673         KASSERT(mycpuid == 0, ("not on cpu0"));
674         crit_enter();
675         if (lmsg->ms_flags & MSGF_DONE)
676                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
677         crit_exit();
678 }
679
680 void
681 nd6_timer_init(void)
682 {
683         callout_init_mp(&nd6_timer_ch);
684         netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
685             MSGF_PRIORITY, nd6_timer_dispatch);
686         callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
687 }
688
689 static int
690 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
691                                          address */
692 {
693         struct ifaddr_container *ifac;
694         struct ifnet *ifp;
695         struct in6_ifaddr *public_ifa6 = NULL;
696
697         ifp = ia6->ia_ifa.ifa_ifp;
698         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
699                 struct ifaddr *ifa = ifac->ifa;
700                 struct in6_ifaddr *it6;
701
702                 if (ifa->ifa_addr->sa_family != AF_INET6)
703                         continue;
704
705                 it6 = (struct in6_ifaddr *)ifa;
706
707                 /* ignore no autoconf addresses. */
708                 if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
709                         continue;
710
711                 /* ignore autoconf addresses with different prefixes. */
712                 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
713                         continue;
714
715                 /*
716                  * Now we are looking at an autoconf address with the same
717                  * prefix as ours.  If the address is temporary and is still
718                  * preferred, do not create another one.  It would be rare, but
719                  * could happen, for example, when we resume a laptop PC after
720                  * a long period.
721                  */
722                 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
723                     !IFA6_IS_DEPRECATED(it6)) {
724                         public_ifa6 = NULL;
725                         break;
726                 }
727
728                 /*
729                  * This is a public autoconf address that has the same prefix
730                  * as ours.  If it is preferred, keep it.  We can't break the
731                  * loop here, because there may be a still-preferred temporary
732                  * address with the prefix.
733                  */
734                 if (!IFA6_IS_DEPRECATED(it6))
735                     public_ifa6 = it6;
736         }
737
738         if (public_ifa6 != NULL) {
739                 int e;
740
741                 if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
742                         log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
743                             " tmp addr,errno=%d\n", e);
744                         return (-1);
745                 }
746                 return (0);
747         }
748
749         return (-1);
750 }
751
752 /*
753  * Nuke neighbor cache/prefix/default router management table, right before
754  * ifp goes away.
755  */
756 void
757 nd6_purge(struct ifnet *ifp)
758 {
759         struct llinfo_nd6 *ln, *nln;
760         struct nd_defrouter *dr, *ndr, drany;
761         struct nd_prefix *pr, *npr;
762
763         /* Nuke default router list entries toward ifp */
764         if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
765                 /*
766                  * The first entry of the list may be stored in
767                  * the routing table, so we'll delete it later.
768                  */
769                 for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
770                         ndr = TAILQ_NEXT(dr, dr_entry);
771                         if (dr->ifp == ifp)
772                                 defrtrlist_del(dr);
773                 }
774                 dr = TAILQ_FIRST(&nd_defrouter);
775                 if (dr->ifp == ifp)
776                         defrtrlist_del(dr);
777         }
778
779         /* Nuke prefix list entries toward ifp */
780         for (pr = nd_prefix.lh_first; pr; pr = npr) {
781                 npr = pr->ndpr_next;
782                 if (pr->ndpr_ifp == ifp) {
783                         /*
784                          * Previously, pr->ndpr_addr is removed as well,
785                          * but I strongly believe we don't have to do it.
786                          * nd6_purge() is only called from in6_ifdetach(),
787                          * which removes all the associated interface addresses
788                          * by itself.
789                          * (jinmei@kame.net 20010129)
790                          */
791                         prelist_remove(pr);
792                 }
793         }
794
795         /* cancel default outgoing interface setting */
796         if (nd6_defifindex == ifp->if_index)
797                 nd6_setdefaultiface(0);
798
799         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
800                 /* refresh default router list */
801                 bzero(&drany, sizeof(drany));
802                 defrouter_delreq(&drany, 0);
803                 defrouter_select();
804         }
805
806         /*
807          * Nuke neighbor cache entries for the ifp.
808          * Note that rt->rt_ifp may not be the same as ifp,
809          * due to KAME goto ours hack.  See RTM_RESOLVE case in
810          * nd6_rtrequest(), and ip6_input().
811          */
812         ln = llinfo_nd6.ln_next;
813         while (ln && ln != &llinfo_nd6) {
814                 struct rtentry *rt;
815                 struct sockaddr_dl *sdl;
816
817                 nln = ln->ln_next;
818                 rt = ln->ln_rt;
819                 if (rt && rt->rt_gateway &&
820                     rt->rt_gateway->sa_family == AF_LINK) {
821                         sdl = (struct sockaddr_dl *)rt->rt_gateway;
822                         if (sdl->sdl_index == ifp->if_index)
823                                 nln = nd6_free(rt);
824                 }
825                 ln = nln;
826         }
827 }
828
829 struct rtentry *
830 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
831 {
832         struct rtentry *rt;
833         struct sockaddr_in6 sin6;
834
835         bzero(&sin6, sizeof(sin6));
836         sin6.sin6_len = sizeof(struct sockaddr_in6);
837         sin6.sin6_family = AF_INET6;
838         sin6.sin6_addr = *addr6;
839
840         if (create)
841                 rt = rtlookup((struct sockaddr *)&sin6);
842         else
843                 rt = rtpurelookup((struct sockaddr *)&sin6);
844         if (rt && !(rt->rt_flags & RTF_LLINFO)) {
845                 /*
846                  * This is the case for the default route.
847                  * If we want to create a neighbor cache for the address, we
848                  * should free the route for the destination and allocate an
849                  * interface route.
850                  */
851                 if (create) {
852                         --rt->rt_refcnt;
853                         rt = NULL;
854                 }
855         }
856         if (!rt) {
857                 if (create && ifp) {
858                         int e;
859
860                         /*
861                          * If no route is available and create is set,
862                          * we allocate a host route for the destination
863                          * and treat it like an interface route.
864                          * This hack is necessary for a neighbor which can't
865                          * be covered by our own prefix.
866                          */
867                         struct ifaddr *ifa;
868
869                         ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
870                         if (ifa == NULL)
871                                 return (NULL);
872
873                         /*
874                          * Create a new route.  RTF_LLINFO is necessary
875                          * to create a Neighbor Cache entry for the
876                          * destination in nd6_rtrequest which will be
877                          * called in rtrequest via ifa->ifa_rtrequest.
878                          */
879                         if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
880                              ifa->ifa_addr, (struct sockaddr *)&all1_sa,
881                              (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
882                              ~RTF_CLONING, &rt)) != 0) {
883                                 log(LOG_ERR,
884                                     "nd6_lookup: failed to add route for a "
885                                     "neighbor(%s), errno=%d\n",
886                                     ip6_sprintf(addr6), e);
887                         }
888                         if (rt == NULL)
889                                 return (NULL);
890                         if (rt->rt_llinfo) {
891                                 struct llinfo_nd6 *ln =
892                                     (struct llinfo_nd6 *)rt->rt_llinfo;
893
894                                 ln->ln_state = ND6_LLINFO_NOSTATE;
895                         }
896                 } else
897                         return (NULL);
898         }
899         rt->rt_refcnt--;
900
901         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
902                 if (create) {
903                         log(LOG_DEBUG,
904                             "nd6_lookup: failed to lookup %s (if = %s)\n",
905                             ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
906                         /* xxx more logs... kazu */
907                 }
908                 return (NULL);
909         }
910         return (rt);
911 }
912
913 static struct rtentry *
914 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
915 {
916         struct rtentry *rt;
917         struct sockaddr_in6 sin6;
918
919         bzero(&sin6, sizeof(sin6));
920         sin6.sin6_len = sizeof(struct sockaddr_in6);
921         sin6.sin6_family = AF_INET6;
922         sin6.sin6_addr = *addr6;
923
924         rt = rtpurelookup((struct sockaddr *)&sin6);
925         if (rt == NULL)
926                 return (NULL);
927         rt->rt_refcnt--;
928
929         if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
930                 if (nd6_onlink_ns_rfc4861 &&
931                     (ND6_RTENTRY_IS_LLCLONING(rt) ||    /* not cloned yet */
932                      (rt->rt_parent != NULL &&          /* cloning */
933                       ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
934                         /*
935                          * If cloning ever happened or is happening,
936                          * rtentry for addr6 would or will become a
937                          * neighbor cache.
938                          */
939                 } else {
940                         rt = NULL;
941                 }
942         }
943         return (rt);
944 }
945
946 /*
947  * Detect if a given IPv6 address identifies a neighbor on a given link.
948  * XXX: should take care of the destination of a p2p link?
949  */
950 int
951 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
952 {
953         struct ifaddr_container *ifac;
954         int i;
955
956 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
957 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
958
959         /*
960          * A link-local address is always a neighbor.
961          * XXX: we should use the sin6_scope_id field rather than the embedded
962          * interface index.
963          */
964         if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
965             ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
966                 return (1);
967
968         /*
969          * If the address matches one of our addresses,
970          * it should be a neighbor.
971          */
972         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
973                 struct ifaddr *ifa = ifac->ifa;
974
975                 if (ifa->ifa_addr->sa_family != AF_INET6)
976                         next: continue;
977
978                 for (i = 0; i < 4; i++) {
979                         if ((IFADDR6(ifa).s6_addr32[i] ^
980                              addr->sin6_addr.s6_addr32[i]) &
981                             IFMASK6(ifa).s6_addr32[i])
982                                 goto next;
983                 }
984                 return (1);
985         }
986
987         /*
988          * Even if the address matches none of our addresses, it might be
989          * in the neighbor cache.
990          */
991         if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
992                 return (1);
993
994         return (0);
995 #undef IFADDR6
996 #undef IFMASK6
997 }
998
999 /*
1000  * Free an nd6 llinfo entry.
1001  */
1002 struct llinfo_nd6 *
1003 nd6_free(struct rtentry *rt)
1004 {
1005         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1006         struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1007         struct nd_defrouter *dr;
1008         int error;
1009         struct rtentry *nrt = NULL;
1010
1011         /*
1012          * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1013          * even though it is not harmful, it was not really necessary.
1014          */
1015
1016         if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
1017                 mtx_lock(&nd6_mtx);
1018                 dr = defrouter_lookup(
1019                     &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1020                     rt->rt_ifp);
1021
1022                 if (ln->ln_router || dr) {
1023                         /*
1024                          * rt6_flush must be called whether or not the neighbor
1025                          * is in the Default Router List.
1026                          * See a corresponding comment in nd6_na_input().
1027                          */
1028                         rt6_flush(&in6, rt->rt_ifp);
1029                 }
1030
1031                 if (dr) {
1032                         /*
1033                          * Unreachablity of a router might affect the default
1034                          * router selection and on-link detection of advertised
1035                          * prefixes.
1036                          */
1037
1038                         /*
1039                          * Temporarily fake the state to choose a new default
1040                          * router and to perform on-link determination of
1041                          * prefixes correctly.
1042                          * Below the state will be set correctly,
1043                          * or the entry itself will be deleted.
1044                          */
1045                         ln->ln_state = ND6_LLINFO_INCOMPLETE;
1046
1047                         /*
1048                          * Since defrouter_select() does not affect the
1049                          * on-link determination and MIP6 needs the check
1050                          * before the default router selection, we perform
1051                          * the check now.
1052                          */
1053                         pfxlist_onlink_check();
1054
1055                         if (dr == TAILQ_FIRST(&nd_defrouter)) {
1056                                 /*
1057                                  * It is used as the current default router,
1058                                  * so we have to move it to the end of the
1059                                  * list and choose a new one.
1060                                  * XXX: it is not very efficient if this is
1061                                  *      the only router.
1062                                  */
1063                                 TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1064                                 TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1065
1066                                 defrouter_select();
1067                         }
1068                 }
1069                 mtx_unlock(&nd6_mtx);
1070         }
1071
1072         /*
1073          * Before deleting the entry, remember the next entry as the
1074          * return value.  We need this because pfxlist_onlink_check() above
1075          * might have freed other entries (particularly the old next entry) as
1076          * a side effect (XXX).
1077          */
1078         next = ln->ln_next;
1079
1080         /*
1081          * Detach the route from the routing tree and the list of neighbor
1082          * caches, and disable the route entry not to be used in already
1083          * cached routes.
1084          */
1085         error = rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, &nrt);
1086         if (error == 0 && nrt != NULL) {
1087                 struct sockaddr_dl *sdl;
1088
1089                 sdl = (struct sockaddr_dl *)nrt->rt_gateway;
1090                 if (sdl->sdl_alen != 0)
1091                         rt_rtmsg(RTM_DELETE, nrt, nrt->rt_ifp, 0);
1092                 rtfree(nrt);
1093         }
1094
1095         return (next);
1096 }
1097
1098 /*
1099  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1100  *
1101  * XXX cost-effective metods?
1102  */
1103 void
1104 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1105 {
1106         struct llinfo_nd6 *ln;
1107
1108         /*
1109          * If the caller specified "rt", use that.  Otherwise, resolve the
1110          * routing table by supplied "dst6".
1111          */
1112         if (!rt) {
1113                 if (!dst6)
1114                         return;
1115                 if (!(rt = nd6_lookup(dst6, 0, NULL)))
1116                         return;
1117         }
1118
1119         if ((rt->rt_flags & RTF_GATEWAY) ||
1120             !(rt->rt_flags & RTF_LLINFO) ||
1121             rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1122             rt->rt_gateway->sa_family != AF_LINK) {
1123                 /* This is not a host route. */
1124                 return;
1125         }
1126
1127         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1128         if (ln->ln_state < ND6_LLINFO_REACHABLE)
1129                 return;
1130
1131         /*
1132          * if we get upper-layer reachability confirmation many times,
1133          * it is possible we have false information.
1134          */
1135         if (!force) {
1136                 ln->ln_byhint++;
1137                 if (ln->ln_byhint > nd6_maxnudhint)
1138                         return;
1139         }
1140
1141         ln->ln_state = ND6_LLINFO_REACHABLE;
1142         if (ln->ln_expire)
1143                 ln->ln_expire = time_uptime +
1144                         ND_IFINFO(rt->rt_ifp)->reachable;
1145 }
1146
1147 void
1148 nd6_rtrequest(int req, struct rtentry *rt)
1149 {
1150         struct sockaddr *gate = rt->rt_gateway;
1151         struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1152         static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1153         struct ifnet *ifp = rt->rt_ifp;
1154         struct ifaddr *ifa;
1155
1156         if ((rt->rt_flags & RTF_GATEWAY))
1157                 return;
1158
1159         if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1160                 /*
1161                  * This is probably an interface direct route for a link
1162                  * which does not need neighbor caches (e.g. fe80::%lo0/64).
1163                  * We do not need special treatment below for such a route.
1164                  * Moreover, the RTF_LLINFO flag which would be set below
1165                  * would annoy the ndp(8) command.
1166                  */
1167                 return;
1168         }
1169
1170         if (req == RTM_RESOLVE &&
1171             (nd6_need_cache(ifp) == 0 || /* stf case */
1172              !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1173                 /*
1174                  * FreeBSD and BSD/OS often make a cloned host route based
1175                  * on a less-specific route (e.g. the default route).
1176                  * If the less specific route does not have a "gateway"
1177                  * (this is the case when the route just goes to a p2p or an
1178                  * stf interface), we'll mistakenly make a neighbor cache for
1179                  * the host route, and will see strange neighbor solicitation
1180                  * for the corresponding destination.  In order to avoid the
1181                  * confusion, we check if the destination of the route is
1182                  * a neighbor in terms of neighbor discovery, and stop the
1183                  * process if not.  Additionally, we remove the LLINFO flag
1184                  * so that ndp(8) will not try to get the neighbor information
1185                  * of the destination.
1186                  */
1187                 rt->rt_flags &= ~RTF_LLINFO;
1188                 return;
1189         }
1190
1191         switch (req) {
1192         case RTM_ADD:
1193                 /*
1194                  * There is no backward compatibility :)
1195                  *
1196                  * if (!(rt->rt_flags & RTF_HOST) &&
1197                  *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1198                  *         rt->rt_flags |= RTF_CLONING;
1199                  */
1200                 if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1201                         /*
1202                          * Case 1: This route should come from
1203                          * a route to interface.  RTF_LLINFO flag is set
1204                          * for a host route whose destination should be
1205                          * treated as on-link.
1206                          */
1207                         rt_setgate(rt, rt_key(rt),
1208                                    (struct sockaddr *)&null_sdl);
1209                         gate = rt->rt_gateway;
1210                         SDL(gate)->sdl_type = ifp->if_type;
1211                         SDL(gate)->sdl_index = ifp->if_index;
1212                         if (ln)
1213                                 ln->ln_expire = time_uptime;
1214                         if (ln && ln->ln_expire == 0) {
1215                                 /* kludge for desktops */
1216                                 ln->ln_expire = 1;
1217                         }
1218                         if ((rt->rt_flags & RTF_CLONING))
1219                                 break;
1220                 }
1221                 /*
1222                  * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1223                  * We don't do that here since llinfo is not ready yet.
1224                  *
1225                  * There are also couple of other things to be discussed:
1226                  * - unsolicited NA code needs improvement beforehand
1227                  * - RFC2461 says we MAY send multicast unsolicited NA
1228                  *   (7.2.6 paragraph 4), however, it also says that we
1229                  *   SHOULD provide a mechanism to prevent multicast NA storm.
1230                  *   we don't have anything like it right now.
1231                  *   note that the mechanism needs a mutual agreement
1232                  *   between proxies, which means that we need to implement
1233                  *   a new protocol, or a new kludge.
1234                  * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1235                  *   we need to check ip6forwarding before sending it.
1236                  *   (or should we allow proxy ND configuration only for
1237                  *   routers?  there's no mention about proxy ND from hosts)
1238                  */
1239 #if 0
1240                 /* XXX it does not work */
1241                 if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1242                         nd6_na_output(ifp,
1243                               &SIN6(rt_key(rt))->sin6_addr,
1244                               &SIN6(rt_key(rt))->sin6_addr,
1245                               ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1246                               1, NULL);
1247                 }
1248 #endif
1249                 /* FALLTHROUGH */
1250         case RTM_RESOLVE:
1251                 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1252                         /*
1253                          * Address resolution isn't necessary for a point to
1254                          * point link, so we can skip this test for a p2p link.
1255                          */
1256                         if (gate->sa_family != AF_LINK ||
1257                             gate->sa_len < sizeof(null_sdl)) {
1258                                 log(LOG_DEBUG,
1259                                     "nd6_rtrequest: bad gateway value: %s\n",
1260                                     if_name(ifp));
1261                                 break;
1262                         }
1263                         SDL(gate)->sdl_type = ifp->if_type;
1264                         SDL(gate)->sdl_index = ifp->if_index;
1265                 }
1266                 if (ln != NULL)
1267                         break;  /* This happens on a route change */
1268                 /*
1269                  * Case 2: This route may come from cloning, or a manual route
1270                  * add with a LL address.
1271                  */
1272                 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1273                 rt->rt_llinfo = (caddr_t)ln;
1274                 if (!ln) {
1275                         log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1276                         break;
1277                 }
1278                 nd6_inuse++;
1279                 nd6_allocated++;
1280                 bzero(ln, sizeof(*ln));
1281                 ln->ln_rt = rt;
1282                 /* this is required for "ndp" command. - shin */
1283                 if (req == RTM_ADD) {
1284                         /*
1285                          * gate should have some valid AF_LINK entry,
1286                          * and ln->ln_expire should have some lifetime
1287                          * which is specified by ndp command.
1288                          */
1289                         ln->ln_state = ND6_LLINFO_REACHABLE;
1290                         ln->ln_byhint = 0;
1291                 } else {
1292                         /*
1293                          * When req == RTM_RESOLVE, rt is created and
1294                          * initialized in rtrequest(), so rt_expire is 0.
1295                          */
1296                         ln->ln_state = ND6_LLINFO_NOSTATE;
1297                         ln->ln_expire = time_uptime;
1298                 }
1299                 rt->rt_flags |= RTF_LLINFO;
1300                 ln->ln_next = llinfo_nd6.ln_next;
1301                 llinfo_nd6.ln_next = ln;
1302                 ln->ln_prev = &llinfo_nd6;
1303                 ln->ln_next->ln_prev = ln;
1304
1305                 /*
1306                  * check if rt_key(rt) is one of my address assigned
1307                  * to the interface.
1308                  */
1309                 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1310                     &SIN6(rt_key(rt))->sin6_addr);
1311                 if (ifa) {
1312                         caddr_t macp = nd6_ifptomac(ifp);
1313                         ln->ln_expire = 0;
1314                         ln->ln_state = ND6_LLINFO_REACHABLE;
1315                         ln->ln_byhint = 0;
1316                         if (macp) {
1317                                 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1318                                 SDL(gate)->sdl_alen = ifp->if_addrlen;
1319                         }
1320                         if (nd6_useloopback) {
1321                                 rt->rt_ifp = loif;      /* XXX */
1322                                 /*
1323                                  * Make sure rt_ifa be equal to the ifaddr
1324                                  * corresponding to the address.
1325                                  * We need this because when we refer
1326                                  * rt_ifa->ia6_flags in ip6_input, we assume
1327                                  * that the rt_ifa points to the address instead
1328                                  * of the loopback address.
1329                                  */
1330                                 if (ifa != rt->rt_ifa) {
1331                                         IFAFREE(rt->rt_ifa);
1332                                         IFAREF(ifa);
1333                                         rt->rt_ifa = ifa;
1334                                 }
1335                         }
1336                 } else if (rt->rt_flags & RTF_ANNOUNCE) {
1337                         ln->ln_expire = 0;
1338                         ln->ln_state = ND6_LLINFO_REACHABLE;
1339                         ln->ln_byhint = 0;
1340
1341                         /*
1342                          * Join solicited node multicast for proxy ND, and only
1343                          * join it once on cpu0.
1344                          */
1345                         if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1346                                 struct in6_addr llsol;
1347                                 int error;
1348
1349                                 llsol = SIN6(rt_key(rt))->sin6_addr;
1350                                 llsol.s6_addr16[0] = htons(0xff02);
1351                                 llsol.s6_addr16[1] = htons(ifp->if_index);
1352                                 llsol.s6_addr32[1] = 0;
1353                                 llsol.s6_addr32[2] = htonl(1);
1354                                 llsol.s6_addr8[12] = 0xff;
1355
1356                                 if (!in6_addmulti(&llsol, ifp, &error)) {
1357                                         nd6log((LOG_ERR, "%s: failed to join "
1358                                             "%s (errno=%d)\n", if_name(ifp),
1359                                             ip6_sprintf(&llsol), error));
1360                                 }
1361                         }
1362                 }
1363                 break;
1364
1365         case RTM_DELETE:
1366                 if (!ln)
1367                         break;
1368                 /*
1369                  * Leave from solicited node multicast for proxy ND, and only
1370                  * leave it once on cpu0 (since we joined it once on cpu0).
1371                  */
1372                 if ((rt->rt_flags & RTF_ANNOUNCE) &&
1373                     (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1374                         struct in6_addr llsol;
1375                         struct in6_multi *in6m;
1376
1377                         llsol = SIN6(rt_key(rt))->sin6_addr;
1378                         llsol.s6_addr16[0] = htons(0xff02);
1379                         llsol.s6_addr16[1] = htons(ifp->if_index);
1380                         llsol.s6_addr32[1] = 0;
1381                         llsol.s6_addr32[2] = htonl(1);
1382                         llsol.s6_addr8[12] = 0xff;
1383
1384                         in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1385                         if (in6m)
1386                                 in6_delmulti(in6m);
1387                 }
1388                 nd6_inuse--;
1389                 ln->ln_next->ln_prev = ln->ln_prev;
1390                 ln->ln_prev->ln_next = ln->ln_next;
1391                 ln->ln_prev = NULL;
1392                 rt->rt_llinfo = 0;
1393                 rt->rt_flags &= ~RTF_LLINFO;
1394                 if (ln->ln_hold)
1395                         m_freem(ln->ln_hold);
1396                 Free((caddr_t)ln);
1397         }
1398 }
1399
1400 int
1401 nd6_ioctl(u_long cmd, caddr_t   data, struct ifnet *ifp)
1402 {
1403         struct in6_drlist *drl = (struct in6_drlist *)data;
1404         struct in6_prlist *prl = (struct in6_prlist *)data;
1405         struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1406         struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1407         struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1408         struct nd_defrouter *dr, any;
1409         struct nd_prefix *pr;
1410         struct rtentry *rt;
1411         int i = 0, error = 0;
1412
1413         switch (cmd) {
1414         case SIOCGDRLST_IN6:
1415                 /*
1416                  * obsolete API, use sysctl under net.inet6.icmp6
1417                  */
1418                 bzero(drl, sizeof(*drl));
1419                 mtx_lock(&nd6_mtx);
1420                 dr = TAILQ_FIRST(&nd_defrouter);
1421                 while (dr && i < DRLSTSIZ) {
1422                         drl->defrouter[i].rtaddr = dr->rtaddr;
1423                         if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1424                                 /* XXX: need to this hack for KAME stack */
1425                                 drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1426                         } else
1427                                 log(LOG_ERR,
1428                                     "default router list contains a "
1429                                     "non-linklocal address(%s)\n",
1430                                     ip6_sprintf(&drl->defrouter[i].rtaddr));
1431
1432                         drl->defrouter[i].flags = dr->flags;
1433                         drl->defrouter[i].rtlifetime = dr->rtlifetime;
1434                         drl->defrouter[i].expire = dr->expire;
1435                         drl->defrouter[i].if_index = dr->ifp->if_index;
1436                         i++;
1437                         dr = TAILQ_NEXT(dr, dr_entry);
1438                 }
1439                 mtx_unlock(&nd6_mtx);
1440                 break;
1441         case SIOCGPRLST_IN6:
1442                 /*
1443                  * obsolete API, use sysctl under net.inet6.icmp6
1444                  */
1445                 /*
1446                  * XXX meaning of fields, especialy "raflags", is very
1447                  * differnet between RA prefix list and RR/static prefix list.
1448                  * how about separating ioctls into two?
1449                  */
1450                 bzero(prl, sizeof(*prl));
1451                 mtx_lock(&nd6_mtx);
1452                 pr = nd_prefix.lh_first;
1453                 while (pr && i < PRLSTSIZ) {
1454                         struct nd_pfxrouter *pfr;
1455                         int j;
1456
1457                         in6_embedscope(&prl->prefix[i].prefix,
1458                             &pr->ndpr_prefix, NULL, NULL);
1459                         prl->prefix[i].raflags = pr->ndpr_raf;
1460                         prl->prefix[i].prefixlen = pr->ndpr_plen;
1461                         prl->prefix[i].vltime = pr->ndpr_vltime;
1462                         prl->prefix[i].pltime = pr->ndpr_pltime;
1463                         prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1464                         prl->prefix[i].expire = pr->ndpr_expire;
1465
1466                         pfr = pr->ndpr_advrtrs.lh_first;
1467                         j = 0;
1468                         while (pfr) {
1469                                 if (j < DRLSTSIZ) {
1470 #define RTRADDR prl->prefix[i].advrtr[j]
1471                                         RTRADDR = pfr->router->rtaddr;
1472                                         if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1473                                                 /* XXX: hack for KAME */
1474                                                 RTRADDR.s6_addr16[1] = 0;
1475                                         } else
1476                                                 log(LOG_ERR,
1477                                                     "a router(%s) advertises "
1478                                                     "a prefix with "
1479                                                     "non-link local address\n",
1480                                                     ip6_sprintf(&RTRADDR));
1481 #undef RTRADDR
1482                                 }
1483                                 j++;
1484                                 pfr = pfr->pfr_next;
1485                         }
1486                         prl->prefix[i].advrtrs = j;
1487                         prl->prefix[i].origin = PR_ORIG_RA;
1488
1489                         i++;
1490                         pr = pr->ndpr_next;
1491                 }
1492                 mtx_unlock(&nd6_mtx);
1493
1494                 break;
1495         case OSIOCGIFINFO_IN6:
1496                 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1497                 bzero(&ndi->ndi, sizeof(ndi->ndi));
1498                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1499                 ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1500                 ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1501                 ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1502                 ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1503                 ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1504                 ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1505                 ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1506                 break;
1507         case SIOCGIFINFO_IN6:
1508                 ndi->ndi = *ND_IFINFO(ifp);
1509                 ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1510                 break;
1511         case SIOCSIFINFO_FLAGS:
1512                 ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1513                 break;
1514         case SIOCSNDFLUSH_IN6:  /* XXX: the ioctl name is confusing... */
1515                 /* flush default router list */
1516                 /*
1517                  * xxx sumikawa: should not delete route if default
1518                  * route equals to the top of default router list
1519                  */
1520                 bzero(&any, sizeof(any));
1521                 defrouter_delreq(&any, 0);
1522                 defrouter_select();
1523                 /* xxx sumikawa: flush prefix list */
1524                 break;
1525         case SIOCSPFXFLUSH_IN6:
1526         {
1527                 /* flush all the prefix advertised by routers */
1528                 struct nd_prefix *pr, *next;
1529
1530                 mtx_lock(&nd6_mtx);
1531                 for (pr = nd_prefix.lh_first; pr; pr = next) {
1532                         struct in6_ifaddr *ia, *ia_next;
1533
1534                         next = pr->ndpr_next;
1535
1536                         if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1537                                 continue; /* XXX */
1538
1539                         /* do we really have to remove addresses as well? */
1540                         for (ia = in6_ifaddr; ia; ia = ia_next) {
1541                                 /* ia might be removed.  keep the next ptr. */
1542                                 ia_next = ia->ia_next;
1543
1544                                 if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1545                                         continue;
1546
1547                                 if (ia->ia6_ndpr == pr)
1548                                         in6_purgeaddr(&ia->ia_ifa);
1549                         }
1550                         prelist_remove(pr);
1551                 }
1552                 mtx_unlock(&nd6_mtx);
1553                 break;
1554         }
1555         case SIOCSRTRFLUSH_IN6:
1556         {
1557                 /* flush all the default routers */
1558                 struct nd_defrouter *dr, *next;
1559
1560                 mtx_lock(&nd6_mtx);
1561                 if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1562                         /*
1563                          * The first entry of the list may be stored in
1564                          * the routing table, so we'll delete it later.
1565                          */
1566                         for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1567                                 next = TAILQ_NEXT(dr, dr_entry);
1568                                 defrtrlist_del(dr);
1569                         }
1570                         defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1571                 }
1572                 mtx_unlock(&nd6_mtx);
1573                 break;
1574         }
1575         case SIOCGNBRINFO_IN6:
1576         {
1577                 struct llinfo_nd6 *ln;
1578                 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1579
1580                 /*
1581                  * XXX: KAME specific hack for scoped addresses
1582                  *      XXXX: for other scopes than link-local?
1583                  */
1584                 if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1585                     IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1586                         u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1587
1588                         if (*idp == 0)
1589                                 *idp = htons(ifp->if_index);
1590                 }
1591
1592                 mtx_lock(&nd6_mtx);
1593                 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1594                         error = EINVAL;
1595                         mtx_unlock(&nd6_mtx);
1596                         break;
1597                 }
1598                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1599                 nbi->state = ln->ln_state;
1600                 nbi->asked = ln->ln_asked;
1601                 nbi->isrouter = ln->ln_router;
1602                 nbi->expire = ln->ln_expire;
1603                 mtx_unlock(&nd6_mtx);
1604
1605                 break;
1606         }
1607         case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1608                 ndif->ifindex = nd6_defifindex;
1609                 break;
1610         case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1611                 return (nd6_setdefaultiface(ndif->ifindex));
1612         }
1613         return (error);
1614 }
1615
1616 /*
1617  * Create neighbor cache entry and cache link-layer address,
1618  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1619  */
1620 struct rtentry *
1621 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1622                  int lladdrlen,
1623                  int type,      /* ICMP6 type */
1624                  int code       /* type dependent information */)
1625 {
1626         struct rtentry *rt = NULL;
1627         struct llinfo_nd6 *ln = NULL;
1628         int is_newentry;
1629         struct sockaddr_dl *sdl = NULL;
1630         int do_update;
1631         int olladdr;
1632         int llchange;
1633         int newstate = 0;
1634
1635         if (!ifp)
1636                 panic("ifp == NULL in nd6_cache_lladdr");
1637         if (!from)
1638                 panic("from == NULL in nd6_cache_lladdr");
1639
1640         /* nothing must be updated for unspecified address */
1641         if (IN6_IS_ADDR_UNSPECIFIED(from))
1642                 return NULL;
1643
1644         /*
1645          * Validation about ifp->if_addrlen and lladdrlen must be done in
1646          * the caller.
1647          *
1648          * XXX If the link does not have link-layer adderss, what should
1649          * we do? (ifp->if_addrlen == 0)
1650          * Spec says nothing in sections for RA, RS and NA.  There's small
1651          * description on it in NS section (RFC 2461 7.2.3).
1652          */
1653
1654         rt = nd6_lookup(from, 0, ifp);
1655         if (!rt) {
1656 #if 0
1657                 /* nothing must be done if there's no lladdr */
1658                 if (!lladdr || !lladdrlen)
1659                         return NULL;
1660 #endif
1661
1662                 rt = nd6_lookup(from, 1, ifp);
1663                 is_newentry = 1;
1664         } else {
1665                 /* do nothing if static ndp is set */
1666                 if (rt->rt_flags & RTF_STATIC)
1667                         return NULL;
1668                 is_newentry = 0;
1669         }
1670
1671         if (!rt)
1672                 return NULL;
1673         if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1674 fail:
1675                 nd6_free(rt);
1676                 return NULL;
1677         }
1678         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1679         if (!ln)
1680                 goto fail;
1681         if (!rt->rt_gateway)
1682                 goto fail;
1683         if (rt->rt_gateway->sa_family != AF_LINK)
1684                 goto fail;
1685         sdl = SDL(rt->rt_gateway);
1686
1687         olladdr = (sdl->sdl_alen) ? 1 : 0;
1688         if (olladdr && lladdr) {
1689                 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1690                         llchange = 1;
1691                 else
1692                         llchange = 0;
1693         } else
1694                 llchange = 0;
1695
1696         /*
1697          * newentry olladdr  lladdr  llchange   (*=record)
1698          *      0       n       n       --      (1)
1699          *      0       y       n       --      (2)
1700          *      0       n       y       --      (3) * STALE
1701          *      0       y       y       n       (4) *
1702          *      0       y       y       y       (5) * STALE
1703          *      1       --      n       --      (6)   NOSTATE(= PASSIVE)
1704          *      1       --      y       --      (7) * STALE
1705          */
1706
1707         if (lladdr) {           /* (3-5) and (7) */
1708                 /*
1709                  * Record source link-layer address
1710                  * XXX is it dependent to ifp->if_type?
1711                  */
1712                 sdl->sdl_alen = ifp->if_addrlen;
1713                 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1714         }
1715
1716         if (!is_newentry) {
1717                 if ((!olladdr && lladdr) ||             /* (3) */
1718                     (olladdr && lladdr && llchange)) {  /* (5) */
1719                         do_update = 1;
1720                         newstate = ND6_LLINFO_STALE;
1721                 } else {                                /* (1-2,4) */
1722                         do_update = 0;
1723                 }
1724         } else {
1725                 do_update = 1;
1726                 if (!lladdr)                            /* (6) */
1727                         newstate = ND6_LLINFO_NOSTATE;
1728                 else                                    /* (7) */
1729                         newstate = ND6_LLINFO_STALE;
1730         }
1731
1732         if (do_update) {
1733                 /*
1734                  * Update the state of the neighbor cache.
1735                  */
1736                 ln->ln_state = newstate;
1737
1738                 if (ln->ln_state == ND6_LLINFO_STALE) {
1739                         /*
1740                          * XXX: since nd6_output() below will cause
1741                          * state tansition to DELAY and reset the timer,
1742                          * we must set the timer now, although it is actually
1743                          * meaningless.
1744                          */
1745                         ln->ln_expire = time_uptime + nd6_gctimer;
1746
1747                         if (ln->ln_hold) {
1748                                 /*
1749                                  * we assume ifp is not a p2p here, so just
1750                                  * set the 2nd argument as the 1st one.
1751                                  */
1752                                 nd6_output(ifp, ifp, ln->ln_hold,
1753                                     (struct sockaddr_in6 *)rt_key(rt), rt);
1754                                 ln->ln_hold = NULL;
1755                         }
1756                 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1757                         /* probe right away */
1758                         ln->ln_expire = time_uptime;
1759                 }
1760         }
1761
1762         /*
1763          * ICMP6 type dependent behavior.
1764          *
1765          * NS: clear IsRouter if new entry
1766          * RS: clear IsRouter
1767          * RA: set IsRouter if there's lladdr
1768          * redir: clear IsRouter if new entry
1769          *
1770          * RA case, (1):
1771          * The spec says that we must set IsRouter in the following cases:
1772          * - If lladdr exist, set IsRouter.  This means (1-5).
1773          * - If it is old entry (!newentry), set IsRouter.  This means (7).
1774          * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1775          * A quetion arises for (1) case.  (1) case has no lladdr in the
1776          * neighbor cache, this is similar to (6).
1777          * This case is rare but we figured that we MUST NOT set IsRouter.
1778          *
1779          * newentry olladdr  lladdr  llchange       NS  RS  RA  redir
1780          *                                                      D R
1781          *      0       n       n       --      (1)     c   ?     s
1782          *      0       y       n       --      (2)     c   s     s
1783          *      0       n       y       --      (3)     c   s     s
1784          *      0       y       y       n       (4)     c   s     s
1785          *      0       y       y       y       (5)     c   s     s
1786          *      1       --      n       --      (6) c   c       c s
1787          *      1       --      y       --      (7) c   c   s   c s
1788          *
1789          *                                      (c=clear s=set)
1790          */
1791         switch (type & 0xff) {
1792         case ND_NEIGHBOR_SOLICIT:
1793                 /*
1794                  * New entry must have is_router flag cleared.
1795                  */
1796                 if (is_newentry)        /* (6-7) */
1797                         ln->ln_router = 0;
1798                 break;
1799         case ND_REDIRECT:
1800                 /*
1801                  * If the icmp is a redirect to a better router, always set the
1802                  * is_router flag.  Otherwise, if the entry is newly created,
1803                  * clear the flag.  [RFC 2461, sec 8.3]
1804                  */
1805                 if (code == ND_REDIRECT_ROUTER)
1806                         ln->ln_router = 1;
1807                 else if (is_newentry) /* (6-7) */
1808                         ln->ln_router = 0;
1809                 break;
1810         case ND_ROUTER_SOLICIT:
1811                 /*
1812                  * is_router flag must always be cleared.
1813                  */
1814                 ln->ln_router = 0;
1815                 break;
1816         case ND_ROUTER_ADVERT:
1817                 /*
1818                  * Mark an entry with lladdr as a router.
1819                  */
1820                 if ((!is_newentry && (olladdr || lladdr)) ||    /* (2-5) */
1821                     (is_newentry && lladdr)) {                  /* (7) */
1822                         ln->ln_router = 1;
1823                 }
1824                 break;
1825         }
1826
1827         if (llchange || lladdr)
1828                 rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1829
1830         /*
1831          * When the link-layer address of a router changes, select the
1832          * best router again.  In particular, when the neighbor entry is newly
1833          * created, it might affect the selection policy.
1834          * Question: can we restrict the first condition to the "is_newentry"
1835          * case?
1836          * XXX: when we hear an RA from a new router with the link-layer
1837          * address option, defrouter_select() is called twice, since
1838          * defrtrlist_update called the function as well.  However, I believe
1839          * we can compromise the overhead, since it only happens the first
1840          * time.
1841          * XXX: although defrouter_select() should not have a bad effect
1842          * for those are not autoconfigured hosts, we explicitly avoid such
1843          * cases for safety.
1844          */
1845         if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1846                 defrouter_select();
1847
1848         return rt;
1849 }
1850
1851 static void
1852 nd6_slowtimo(void *arg __unused)
1853 {
1854         struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1855
1856         KASSERT(mycpuid == 0, ("not on cpu0"));
1857         crit_enter();
1858         if (lmsg->ms_flags & MSGF_DONE)
1859                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1860         crit_exit();
1861 }
1862
1863 static void
1864 nd6_slowtimo_dispatch(netmsg_t nmsg)
1865 {
1866         const struct ifnet_array *arr;
1867         struct nd_ifinfo *nd6if;
1868         int i;
1869
1870         ASSERT_NETISR0;
1871
1872         crit_enter();
1873         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1874         crit_exit();
1875
1876         arr = ifnet_array_get();
1877
1878         mtx_lock(&nd6_mtx);
1879         for (i = 0; i < arr->ifnet_count; ++i) {
1880                 struct ifnet *ifp = arr->ifnet_arr[i];
1881
1882                 if (ifp->if_afdata[AF_INET6] == NULL)
1883                         continue;
1884                 nd6if = ND_IFINFO(ifp);
1885                 if (nd6if->basereachable && /* already initialized */
1886                     (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1887                         /*
1888                          * Since reachable time rarely changes by router
1889                          * advertisements, we SHOULD insure that a new random
1890                          * value gets recomputed at least once every few hours.
1891                          * (RFC 2461, 6.3.4)
1892                          */
1893                         nd6if->recalctm = nd6_recalc_reachtm_interval;
1894                         nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1895                 }
1896         }
1897         mtx_unlock(&nd6_mtx);
1898
1899         callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1900             nd6_slowtimo, NULL);
1901 }
1902
1903 #define gotoerr(e) { error = (e); goto bad;}
1904
1905 int
1906 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1907            struct sockaddr_in6 *dst, struct rtentry *rt)
1908 {
1909         struct llinfo_nd6 *ln = NULL;
1910         int error = 0;
1911
1912         if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1913                 goto sendpkt;
1914
1915         if (nd6_need_cache(ifp) == 0)
1916                 goto sendpkt;
1917
1918         /*
1919          * Next hop determination.  This routine is derived from rt_llroute.
1920          */
1921         if (rt != NULL) {
1922                 if (!(rt->rt_flags & RTF_UP)) {
1923                         rt = rtlookup((struct sockaddr *)dst);
1924                         if (rt == NULL)
1925                                 gotoerr(EHOSTUNREACH);
1926                         rt->rt_refcnt--;
1927                         if (rt->rt_ifp != ifp) {
1928                                 /* XXX: loop care? */
1929                                 return nd6_output(ifp, origifp, m, dst, rt);
1930                         }
1931                 }
1932                 if (rt->rt_flags & RTF_GATEWAY) {
1933                         struct sockaddr_in6 *gw6;
1934
1935                         /*
1936                          * We skip link-layer address resolution and NUD
1937                          * if the gateway is not a neighbor from ND point
1938                          * of view, regardless of the value of nd_ifinfo.flags.
1939                          * The second condition is a bit tricky; we skip
1940                          * if the gateway is our own address, which is
1941                          * sometimes used to install a route to a p2p link.
1942                          */
1943                         gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1944                         if (!nd6_is_addr_neighbor(gw6, ifp) ||
1945                             in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1946                                 /*
1947                                  * We allow this kind of tricky route only
1948                                  * when the outgoing interface is p2p.
1949                                  * XXX: we may need a more generic rule here.
1950                                  */
1951                                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1952                                         gotoerr(EHOSTUNREACH);
1953
1954                                 goto sendpkt;
1955                         }
1956
1957                         if (rt->rt_gwroute == NULL) {
1958                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1959                                 if (rt->rt_gwroute == NULL)
1960                                         gotoerr(EHOSTUNREACH);
1961                         } else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1962                                 rtfree(rt->rt_gwroute);
1963                                 rt->rt_gwroute = rtlookup(rt->rt_gateway);
1964                                 if (rt->rt_gwroute == NULL)
1965                                         gotoerr(EHOSTUNREACH);
1966                         }
1967                         rt = rt->rt_gwroute;
1968                 }
1969         }
1970
1971         /*
1972          * Address resolution or Neighbor Unreachability Detection
1973          * for the next hop.
1974          * At this point, the destination of the packet must be a unicast
1975          * or an anycast address(i.e. not a multicast).
1976          */
1977
1978         /* Look up the neighbor cache for the nexthop */
1979         if (rt && (rt->rt_flags & RTF_LLINFO))
1980                 ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1981         else {
1982                 /*
1983                  * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1984                  * the condition below is not very efficient.  But we believe
1985                  * it is tolerable, because this should be a rare case.
1986                  */
1987                 if (nd6_is_addr_neighbor(dst, ifp) &&
1988                     (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1989                         ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1990         }
1991         if (!ln || !rt) {
1992                 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1993                     !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1994                         log(LOG_DEBUG,
1995                             "nd6_output: can't allocate llinfo for %s "
1996                             "(ln=%p, rt=%p)\n",
1997                             ip6_sprintf(&dst->sin6_addr), ln, rt);
1998                         gotoerr(EIO);   /* XXX: good error? */
1999                 }
2000
2001                 goto sendpkt;   /* send anyway */
2002         }
2003
2004         /* We don't have to do link-layer address resolution on a p2p link. */
2005         if ((ifp->if_flags & IFF_POINTOPOINT) &&
2006             ln->ln_state < ND6_LLINFO_REACHABLE) {
2007                 ln->ln_state = ND6_LLINFO_STALE;
2008                 ln->ln_expire = time_uptime + nd6_gctimer;
2009         }
2010
2011         /*
2012          * The first time we send a packet to a neighbor whose entry is
2013          * STALE, we have to change the state to DELAY and a sets a timer to
2014          * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2015          * neighbor unreachability detection on expiration.
2016          * (RFC 2461 7.3.3)
2017          */
2018         if (ln->ln_state == ND6_LLINFO_STALE) {
2019                 ln->ln_asked = 0;
2020                 ln->ln_state = ND6_LLINFO_DELAY;
2021                 ln->ln_expire = time_uptime + nd6_delay;
2022         }
2023
2024         /*
2025          * If the neighbor cache entry has a state other than INCOMPLETE
2026          * (i.e. its link-layer address is already resolved), just
2027          * send the packet.
2028          */
2029         if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2030                 goto sendpkt;
2031
2032         /*
2033          * There is a neighbor cache entry, but no ethernet address
2034          * response yet.  Replace the held mbuf (if any) with this
2035          * latest one.
2036          *
2037          * This code conforms to the rate-limiting rule described in Section
2038          * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2039          * an NS below.
2040          */
2041         if (ln->ln_state == ND6_LLINFO_NOSTATE) {
2042                 /*
2043                  * This neighbor cache entry was just created; change its
2044                  * state to INCOMPLETE and start its life cycle.
2045                  *
2046                  * We force an NS output below by setting ln_expire to 1
2047                  * (nd6_rtrequest() could set it to the current time_uptime)
2048                  * and zeroing out ln_asked (XXX this may not be necessary).
2049                  */
2050                 ln->ln_state = ND6_LLINFO_INCOMPLETE;
2051                 ln->ln_expire = 1;
2052                 ln->ln_asked = 0;
2053         }
2054         if (ln->ln_hold)
2055                 m_freem(ln->ln_hold);
2056         ln->ln_hold = m;
2057         if (ln->ln_expire) {
2058                 if (ln->ln_asked < nd6_mmaxtries &&
2059                     ln->ln_expire < time_uptime) {
2060                         ln->ln_asked++;
2061                         ln->ln_expire = time_uptime +
2062                                 ND_IFINFO(ifp)->retrans / 1000;
2063                         nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2064                 }
2065         }
2066         return (0);
2067
2068 sendpkt:
2069         if (ifp->if_flags & IFF_LOOPBACK)
2070                 error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
2071         else
2072                 error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
2073         return (error);
2074
2075 bad:
2076         m_freem(m);
2077         return (error);
2078 }
2079 #undef gotoerr
2080
2081 int
2082 nd6_need_cache(struct ifnet *ifp)
2083 {
2084         /*
2085          * XXX: we currently do not make neighbor cache on any interface
2086          * other than Ethernet and GIF.
2087          *
2088          * RFC2893 says:
2089          * - unidirectional tunnels needs no ND
2090          */
2091         switch (ifp->if_type) {
2092         case IFT_ETHER:
2093         case IFT_IEEE1394:
2094 #ifdef IFT_L2VLAN
2095         case IFT_L2VLAN:
2096 #endif
2097 #ifdef IFT_IEEE80211
2098         case IFT_IEEE80211:
2099 #endif
2100 #ifdef IFT_CARP
2101         case IFT_CARP:
2102 #endif
2103         case IFT_GIF:           /* XXX need more cases? */
2104                 return (1);
2105         default:
2106                 return (0);
2107         }
2108 }
2109
2110 int
2111 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2112                 struct sockaddr *dst, u_char *desten)
2113 {
2114         struct sockaddr_dl *sdl;
2115         struct rtentry *rt;
2116
2117
2118         if (m->m_flags & M_MCAST) {
2119                 switch (ifp->if_type) {
2120                 case IFT_ETHER:
2121 #ifdef IFT_L2VLAN
2122         case IFT_L2VLAN:
2123 #endif
2124 #ifdef IFT_IEEE80211
2125                 case IFT_IEEE80211:
2126 #endif
2127                         ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2128                                                  desten);
2129                         return (1);
2130                 case IFT_IEEE1394:
2131                         bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2132                         return (1);
2133                 default:
2134                         m_freem(m);
2135                         return (0);
2136                 }
2137         }
2138         if (rt0 == NULL) {
2139                 /* this could happen, if we could not allocate memory */
2140                 m_freem(m);
2141                 return (0);
2142         }
2143         if (rt_llroute(dst, rt0, &rt) != 0) {
2144                 m_freem(m);
2145                 return (0);
2146         }
2147         if (rt->rt_gateway->sa_family != AF_LINK) {
2148                 kprintf("nd6_storelladdr: something odd happens\n");
2149                 m_freem(m);
2150                 return (0);
2151         }
2152         sdl = SDL(rt->rt_gateway);
2153         if (sdl->sdl_alen == 0) {
2154                 /* this should be impossible, but we bark here for debugging */
2155                 kprintf("nd6_storelladdr: sdl_alen == 0\n");
2156                 m_freem(m);
2157                 return (0);
2158         }
2159
2160         bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2161         return (1);
2162 }
2163
2164 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2165 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2166 #ifdef SYSCTL_DECL
2167 SYSCTL_DECL(_net_inet6_icmp6);
2168 #endif
2169 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2170         CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2171 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2172         CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2173
2174 static int
2175 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2176 {
2177         int error;
2178         char buf[1024];
2179         struct in6_defrouter *d, *de;
2180         struct nd_defrouter *dr;
2181
2182         if (req->newptr)
2183                 return EPERM;
2184         error = 0;
2185
2186         for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2187              dr = TAILQ_NEXT(dr, dr_entry)) {
2188                 d = (struct in6_defrouter *)buf;
2189                 de = (struct in6_defrouter *)(buf + sizeof(buf));
2190
2191                 if (d + 1 <= de) {
2192                         bzero(d, sizeof(*d));
2193                         d->rtaddr.sin6_family = AF_INET6;
2194                         d->rtaddr.sin6_len = sizeof(d->rtaddr);
2195                         if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2196                             dr->ifp) != 0)
2197                                 log(LOG_ERR,
2198                                     "scope error in "
2199                                     "default router list (%s)\n",
2200                                     ip6_sprintf(&dr->rtaddr));
2201                         d->flags = dr->flags;
2202                         d->rtlifetime = dr->rtlifetime;
2203                         d->expire = dr->expire;
2204                         d->if_index = dr->ifp->if_index;
2205                 } else
2206                         panic("buffer too short");
2207
2208                 error = SYSCTL_OUT(req, buf, sizeof(*d));
2209                 if (error)
2210                         break;
2211         }
2212         return error;
2213 }
2214
2215 static int
2216 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2217 {
2218         int error;
2219         char buf[1024];
2220         struct in6_prefix *p, *pe;
2221         struct nd_prefix *pr;
2222
2223         if (req->newptr)
2224                 return EPERM;
2225         error = 0;
2226
2227         for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2228                 u_short advrtrs;
2229                 size_t advance;
2230                 struct sockaddr_in6 *sin6, *s6;
2231                 struct nd_pfxrouter *pfr;
2232
2233                 p = (struct in6_prefix *)buf;
2234                 pe = (struct in6_prefix *)(buf + sizeof(buf));
2235
2236                 if (p + 1 <= pe) {
2237                         bzero(p, sizeof(*p));
2238                         sin6 = (struct sockaddr_in6 *)(p + 1);
2239
2240                         p->prefix = pr->ndpr_prefix;
2241                         if (in6_recoverscope(&p->prefix,
2242                             &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2243                                 log(LOG_ERR,
2244                                     "scope error in prefix list (%s)\n",
2245                                     ip6_sprintf(&p->prefix.sin6_addr));
2246                         p->raflags = pr->ndpr_raf;
2247                         p->prefixlen = pr->ndpr_plen;
2248                         p->vltime = pr->ndpr_vltime;
2249                         p->pltime = pr->ndpr_pltime;
2250                         p->if_index = pr->ndpr_ifp->if_index;
2251                         p->expire = pr->ndpr_expire;
2252                         p->refcnt = pr->ndpr_refcnt;
2253                         p->flags = pr->ndpr_stateflags;
2254                         p->origin = PR_ORIG_RA;
2255                         advrtrs = 0;
2256                         for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2257                              pfr = pfr->pfr_next) {
2258                                 if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2259                                         advrtrs++;
2260                                         continue;
2261                                 }
2262                                 s6 = &sin6[advrtrs];
2263                                 bzero(s6, sizeof(*s6));
2264                                 s6->sin6_family = AF_INET6;
2265                                 s6->sin6_len = sizeof(*sin6);
2266                                 if (in6_recoverscope(s6, &pfr->router->rtaddr,
2267                                     pfr->router->ifp) != 0)
2268                                         log(LOG_ERR,
2269                                             "scope error in "
2270                                             "prefix list (%s)\n",
2271                                             ip6_sprintf(&pfr->router->rtaddr));
2272                                 advrtrs++;
2273                         }
2274                         p->advrtrs = advrtrs;
2275                 } else {
2276                         panic("buffer too short");
2277                 }
2278
2279                 advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2280                 error = SYSCTL_OUT(req, buf, advance);
2281                 if (error)
2282                         break;
2283         }
2284         return error;
2285 }