xref: /openbsd-src/sys/netinet6/nd6.c (revision d13be5d47e4149db2549a9828e244d59dbc43f15)
1 /*	$OpenBSD: nd6.c,v 1.87 2011/06/17 07:06:47 mk Exp $	*/
2 /*	$KAME: nd6.c,v 1.280 2002/06/08 19:52:07 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 <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/timeout.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/socket.h>
39 #include <sys/sockio.h>
40 #include <sys/time.h>
41 #include <sys/kernel.h>
42 #include <sys/protosw.h>
43 #include <sys/errno.h>
44 #include <sys/ioctl.h>
45 #include <sys/syslog.h>
46 #include <sys/queue.h>
47 #include <dev/rndvar.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/if_types.h>
52 #include <net/if_fddi.h>
53 #include <net/route.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/if_ether.h>
57 #include <netinet/ip_ipsp.h>
58 
59 #include <netinet6/in6_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet6/nd6.h>
63 #include <netinet/icmp6.h>
64 
65 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
66 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
67 
68 #define SIN6(s) ((struct sockaddr_in6 *)s)
69 #define SDL(s) ((struct sockaddr_dl *)s)
70 
71 /* timer values */
72 int	nd6_prune	= 1;	/* walk list every 1 seconds */
73 int	nd6_delay	= 5;	/* delay first probe time 5 second */
74 int	nd6_umaxtries	= 3;	/* maximum unicast query */
75 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
76 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
77 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
78 
79 /* preventing too many loops in ND option parsing */
80 int nd6_maxndopt = 10;	/* max # of ND options allowed */
81 
82 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
83 
84 #ifdef ND6_DEBUG
85 int nd6_debug = 1;
86 #else
87 int nd6_debug = 0;
88 #endif
89 
90 static int nd6_inuse, nd6_allocated;
91 
92 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
93 struct nd_drhead nd_defrouter;
94 struct nd_prhead nd_prefix = { 0 };
95 
96 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
97 static struct sockaddr_in6 all1_sa;
98 
99 void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
100 void nd6_slowtimo(void *);
101 struct llinfo_nd6 *nd6_free(struct rtentry *, int);
102 void nd6_llinfo_timer(void *);
103 
104 struct timeout nd6_slowtimo_ch;
105 struct timeout nd6_timer_ch;
106 extern struct timeout in6_tmpaddrtimer_ch;
107 
108 int fill_drlist(void *, size_t *, size_t);
109 int fill_prlist(void *, size_t *, size_t);
110 
111 #define LN_DEQUEUE(ln) do { \
112 	(ln)->ln_next->ln_prev = (ln)->ln_prev; \
113 	(ln)->ln_prev->ln_next = (ln)->ln_next; \
114 	} while (0)
115 #define LN_INSERTHEAD(ln) do { \
116 	(ln)->ln_next = llinfo_nd6.ln_next; \
117 	llinfo_nd6.ln_next = (ln); \
118 	(ln)->ln_prev = &llinfo_nd6; \
119 	(ln)->ln_next->ln_prev = (ln); \
120 	} while (0)
121 
122 void
123 nd6_init(void)
124 {
125 	static int nd6_init_done = 0;
126 	int i;
127 
128 	if (nd6_init_done) {
129 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
130 		return;
131 	}
132 
133 	all1_sa.sin6_family = AF_INET6;
134 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
135 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
136 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
137 
138 	/* initialization of the default router list */
139 	TAILQ_INIT(&nd_defrouter);
140 
141 	nd6_init_done = 1;
142 
143 	/* start timer */
144 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
145 	timeout_add_sec(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL);
146 }
147 
148 struct nd_ifinfo *
149 nd6_ifattach(struct ifnet *ifp)
150 {
151 	struct nd_ifinfo *nd;
152 
153 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
154 
155 	nd->initialized = 1;
156 
157 	nd->chlim = IPV6_DEFHLIM;
158 	nd->basereachable = REACHABLE_TIME;
159 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
160 	nd->retrans = RETRANS_TIMER;
161 	/*
162 	 * Note that the default value of ip6_accept_rtadv is 0, which means
163 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
164 	 * here.
165 	 */
166 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
167 
168 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
169 	nd6_setmtu0(ifp, nd);
170 
171 	return nd;
172 }
173 
174 void
175 nd6_ifdetach(struct nd_ifinfo *nd)
176 {
177 
178 	free(nd, M_IP6NDP);
179 }
180 
181 void
182 nd6_setmtu(struct ifnet *ifp)
183 {
184 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
185 }
186 
187 void
188 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
189 {
190 	u_int32_t omaxmtu;
191 
192 	omaxmtu = ndi->maxmtu;
193 
194 	if (ifp->if_type == IFT_FDDI)
195 		ndi->maxmtu = MIN(FDDIMTU, ifp->if_mtu);
196 	else
197 		ndi->maxmtu = ifp->if_mtu;
198 
199 	/*
200 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
201 	 * undesirable situation.  We thus notify the operator of the change
202 	 * explicitly.  The check for omaxmtu is necessary to restrict the
203 	 * log to the case of changing the MTU, not initializing it.
204 	 */
205 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
206 		log(LOG_NOTICE, "nd6_setmtu0: "
207 		    "new link MTU on %s (%lu) is too small for IPv6\n",
208 		    ifp->if_xname, (unsigned long)ndi->maxmtu);
209 	}
210 
211 	if (ndi->maxmtu > in6_maxmtu)
212 		in6_setmaxmtu(); /* check all interfaces just in case */
213 }
214 
215 void
216 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
217 {
218 	bzero(ndopts, sizeof(*ndopts));
219 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
220 	ndopts->nd_opts_last
221 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
222 
223 	if (icmp6len == 0) {
224 		ndopts->nd_opts_done = 1;
225 		ndopts->nd_opts_search = NULL;
226 	}
227 }
228 
229 /*
230  * Take one ND option.
231  */
232 struct nd_opt_hdr *
233 nd6_option(union nd_opts *ndopts)
234 {
235 	struct nd_opt_hdr *nd_opt;
236 	int olen;
237 
238 	if (!ndopts)
239 		panic("ndopts == NULL in nd6_option");
240 	if (!ndopts->nd_opts_last)
241 		panic("uninitialized ndopts in nd6_option");
242 	if (!ndopts->nd_opts_search)
243 		return NULL;
244 	if (ndopts->nd_opts_done)
245 		return NULL;
246 
247 	nd_opt = ndopts->nd_opts_search;
248 
249 	/* make sure nd_opt_len is inside the buffer */
250 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
251 		bzero(ndopts, sizeof(*ndopts));
252 		return NULL;
253 	}
254 
255 	olen = nd_opt->nd_opt_len << 3;
256 	if (olen == 0) {
257 		/*
258 		 * Message validation requires that all included
259 		 * options have a length that is greater than zero.
260 		 */
261 		bzero(ndopts, sizeof(*ndopts));
262 		return NULL;
263 	}
264 
265 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
266 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
267 		/* option overruns the end of buffer, invalid */
268 		bzero(ndopts, sizeof(*ndopts));
269 		return NULL;
270 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
271 		/* reached the end of options chain */
272 		ndopts->nd_opts_done = 1;
273 		ndopts->nd_opts_search = NULL;
274 	}
275 	return nd_opt;
276 }
277 
278 /*
279  * Parse multiple ND options.
280  * This function is much easier to use, for ND routines that do not need
281  * multiple options of the same type.
282  */
283 int
284 nd6_options(union nd_opts *ndopts)
285 {
286 	struct nd_opt_hdr *nd_opt;
287 	int i = 0;
288 
289 	if (!ndopts)
290 		panic("ndopts == NULL in nd6_options");
291 	if (!ndopts->nd_opts_last)
292 		panic("uninitialized ndopts in nd6_options");
293 	if (!ndopts->nd_opts_search)
294 		return 0;
295 
296 	while (1) {
297 		nd_opt = nd6_option(ndopts);
298 		if (!nd_opt && !ndopts->nd_opts_last) {
299 			/*
300 			 * Message validation requires that all included
301 			 * options have a length that is greater than zero.
302 			 */
303 			icmp6stat.icp6s_nd_badopt++;
304 			bzero(ndopts, sizeof(*ndopts));
305 			return -1;
306 		}
307 
308 		if (!nd_opt)
309 			goto skip1;
310 
311 		switch (nd_opt->nd_opt_type) {
312 		case ND_OPT_SOURCE_LINKADDR:
313 		case ND_OPT_TARGET_LINKADDR:
314 		case ND_OPT_MTU:
315 		case ND_OPT_REDIRECTED_HEADER:
316 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
317 				nd6log((LOG_INFO,
318 				    "duplicated ND6 option found (type=%d)\n",
319 				    nd_opt->nd_opt_type));
320 				/* XXX bark? */
321 			} else {
322 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
323 					= nd_opt;
324 			}
325 			break;
326 		case ND_OPT_PREFIX_INFORMATION:
327 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
328 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
329 					= nd_opt;
330 			}
331 			ndopts->nd_opts_pi_end =
332 				(struct nd_opt_prefix_info *)nd_opt;
333 			break;
334 		default:
335 			/*
336 			 * Unknown options must be silently ignored,
337 			 * to accommodate future extension to the protocol.
338 			 */
339 			nd6log((LOG_DEBUG,
340 			    "nd6_options: unsupported option %d - "
341 			    "option ignored\n", nd_opt->nd_opt_type));
342 		}
343 
344 skip1:
345 		i++;
346 		if (i > nd6_maxndopt) {
347 			icmp6stat.icp6s_nd_toomanyopt++;
348 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
349 			break;
350 		}
351 
352 		if (ndopts->nd_opts_done)
353 			break;
354 	}
355 
356 	return 0;
357 }
358 
359 /*
360  * ND6 timer routine to handle ND6 entries
361  */
362 void
363 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long tick)
364 {
365 	int s;
366 
367 	s = splsoftnet();
368 
369 	if (tick < 0) {
370 		ln->ln_expire = 0;
371 		ln->ln_ntick = 0;
372 		timeout_del(&ln->ln_timer_ch);
373 	} else {
374 		ln->ln_expire = time_second + tick / hz;
375 		if (tick > INT_MAX) {
376 			ln->ln_ntick = tick - INT_MAX;
377 			timeout_add(&ln->ln_timer_ch, INT_MAX);
378 		} else {
379 			ln->ln_ntick = 0;
380 			timeout_add(&ln->ln_timer_ch, tick);
381 		}
382 	}
383 
384 	splx(s);
385 }
386 
387 void
388 nd6_llinfo_timer(void *arg)
389 {
390 	int s;
391 	struct llinfo_nd6 *ln;
392 	struct rtentry *rt;
393 	struct sockaddr_in6 *dst;
394 	struct ifnet *ifp;
395 	struct nd_ifinfo *ndi = NULL;
396 
397 	s = splsoftnet();
398 
399 	ln = (struct llinfo_nd6 *)arg;
400 
401 	if (ln->ln_ntick > 0) {
402 		if (ln->ln_ntick > INT_MAX) {
403 			ln->ln_ntick -= INT_MAX;
404 			nd6_llinfo_settimer(ln, INT_MAX);
405 		} else {
406 			ln->ln_ntick = 0;
407 			nd6_llinfo_settimer(ln, ln->ln_ntick);
408 		}
409 		splx(s);
410 		return;
411 	}
412 
413 	if ((rt = ln->ln_rt) == NULL)
414 		panic("ln->ln_rt == NULL");
415 	if ((ifp = rt->rt_ifp) == NULL)
416 		panic("ln->ln_rt->rt_ifp == NULL");
417 	ndi = ND_IFINFO(ifp);
418 	dst = (struct sockaddr_in6 *)rt_key(rt);
419 
420 	/* sanity check */
421 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
422 		panic("rt_llinfo(%p) is not equal to ln(%p)",
423 		      rt->rt_llinfo, ln);
424 	if (!dst)
425 		panic("dst=0 in nd6_timer(ln=%p)", ln);
426 
427 	switch (ln->ln_state) {
428 	case ND6_LLINFO_INCOMPLETE:
429 		if (ln->ln_asked < nd6_mmaxtries) {
430 			ln->ln_asked++;
431 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
432 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
433 		} else {
434 			struct mbuf *m = ln->ln_hold;
435 			if (m) {
436 				ln->ln_hold = NULL;
437 				/*
438 				 * Fake rcvif to make the ICMP error
439 				 * more helpful in diagnosing for the
440 				 * receiver.
441 				 * XXX: should we consider
442 				 * older rcvif?
443 				 */
444 				m->m_pkthdr.rcvif = rt->rt_ifp;
445 
446 				icmp6_error(m, ICMP6_DST_UNREACH,
447 				    ICMP6_DST_UNREACH_ADDR, 0);
448 				if (ln->ln_hold == m) {
449 					/* m is back in ln_hold. Discard. */
450 					m_freem(ln->ln_hold);
451 					ln->ln_hold = NULL;
452 				}
453 			}
454 			(void)nd6_free(rt, 0);
455 			ln = NULL;
456 		}
457 		break;
458 	case ND6_LLINFO_REACHABLE:
459 		if (!ND6_LLINFO_PERMANENT(ln)) {
460 			ln->ln_state = ND6_LLINFO_STALE;
461 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
462 		}
463 		break;
464 
465 	case ND6_LLINFO_STALE:
466 	case ND6_LLINFO_PURGE:
467 		/* Garbage Collection(RFC 2461 5.3) */
468 		if (!ND6_LLINFO_PERMANENT(ln)) {
469 			(void)nd6_free(rt, 1);
470 			ln = NULL;
471 		}
472 		break;
473 
474 	case ND6_LLINFO_DELAY:
475 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
476 			/* We need NUD */
477 			ln->ln_asked = 1;
478 			ln->ln_state = ND6_LLINFO_PROBE;
479 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
480 			nd6_ns_output(ifp, &dst->sin6_addr,
481 			    &dst->sin6_addr, ln, 0);
482 		} else {
483 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
484 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
485 		}
486 		break;
487 	case ND6_LLINFO_PROBE:
488 		if (ln->ln_asked < nd6_umaxtries) {
489 			ln->ln_asked++;
490 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
491 			nd6_ns_output(ifp, &dst->sin6_addr,
492 			    &dst->sin6_addr, ln, 0);
493 		} else {
494 			(void)nd6_free(rt, 0);
495 			ln = NULL;
496 		}
497 		break;
498 	}
499 
500 	splx(s);
501 }
502 
503 /*
504  * ND6 timer routine to expire default route list and prefix list
505  */
506 void
507 nd6_timer(void *ignored_arg)
508 {
509 	int s;
510 	struct nd_defrouter *dr;
511 	struct nd_prefix *pr;
512 	struct in6_ifaddr *ia6, *nia6;
513 
514 	s = splsoftnet();
515 	timeout_set(&nd6_timer_ch, nd6_timer, NULL);
516 	timeout_add_sec(&nd6_timer_ch, nd6_prune);
517 
518 	/* expire default router list */
519 	dr = TAILQ_FIRST(&nd_defrouter);
520 	while (dr) {
521 		if (dr->expire && dr->expire < time_second) {
522 			struct nd_defrouter *t;
523 			t = TAILQ_NEXT(dr, dr_entry);
524 			defrtrlist_del(dr);
525 			dr = t;
526 		} else {
527 			dr = TAILQ_NEXT(dr, dr_entry);
528 		}
529 	}
530 
531 	/*
532 	 * expire interface addresses.
533 	 * in the past the loop was inside prefix expiry processing.
534 	 * However, from a stricter spec-conformance standpoint, we should
535 	 * rather separate address lifetimes and prefix lifetimes.
536 	 */
537 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
538 		nia6 = ia6->ia_next;
539 		/* check address lifetime */
540 		if (IFA6_IS_INVALID(ia6)) {
541 			in6_purgeaddr(&ia6->ia_ifa);
542 		} else if (IFA6_IS_DEPRECATED(ia6)) {
543 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
544 		} else {
545 			/*
546 			 * A new RA might have made a deprecated address
547 			 * preferred.
548 			 */
549 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
550 		}
551 	}
552 
553 	/* expire prefix list */
554 	pr = LIST_FIRST(&nd_prefix);
555 	while (pr != NULL) {
556 		/*
557 		 * check prefix lifetime.
558 		 * since pltime is just for autoconf, pltime processing for
559 		 * prefix is not necessary.
560 		 */
561 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
562 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
563 			struct nd_prefix *t;
564 			t = LIST_NEXT(pr, ndpr_entry);
565 
566 			/*
567 			 * address expiration and prefix expiration are
568 			 * separate.  NEVER perform in6_purgeaddr here.
569 			 */
570 
571 			prelist_remove(pr);
572 			pr = t;
573 		} else
574 			pr = LIST_NEXT(pr, ndpr_entry);
575 	}
576 	splx(s);
577 }
578 
579 /*
580  * Nuke neighbor cache/prefix/default router management table, right before
581  * ifp goes away.
582  */
583 void
584 nd6_purge(struct ifnet *ifp)
585 {
586 	struct llinfo_nd6 *ln, *nln;
587 	struct nd_defrouter *dr, *ndr;
588 	struct nd_prefix *pr, *npr;
589 
590 	/*
591 	 * Nuke default router list entries toward ifp.
592 	 * We defer removal of default router list entries that is installed
593 	 * in the routing table, in order to keep additional side effects as
594 	 * small as possible.
595 	 */
596 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
597 		ndr = TAILQ_NEXT(dr, dr_entry);
598 		if (dr->installed)
599 			continue;
600 
601 		if (dr->ifp == ifp)
602 			defrtrlist_del(dr);
603 	}
604 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
605 		ndr = TAILQ_NEXT(dr, dr_entry);
606 		if (!dr->installed)
607 			continue;
608 
609 		if (dr->ifp == ifp)
610 			defrtrlist_del(dr);
611 	}
612 
613 	/* Nuke prefix list entries toward ifp */
614 	for (pr = LIST_FIRST(&nd_prefix); pr != NULL; pr = npr) {
615 		npr = LIST_NEXT(pr, ndpr_entry);
616 		if (pr->ndpr_ifp == ifp) {
617 			/*
618 			 * Because if_detach() does *not* release prefixes
619 			 * while purging addresses the reference count will
620 			 * still be above zero. We therefore reset it to
621 			 * make sure that the prefix really gets purged.
622 			 */
623 			pr->ndpr_refcnt = 0;
624 			/*
625 			 * Previously, pr->ndpr_addr is removed as well,
626 			 * but I strongly believe we don't have to do it.
627 			 * nd6_purge() is only called from in6_ifdetach(),
628 			 * which removes all the associated interface addresses
629 			 * by itself.
630 			 * (jinmei@kame.net 20010129)
631 			 */
632 			prelist_remove(pr);
633 		}
634 	}
635 
636 	/* cancel default outgoing interface setting */
637 	if (nd6_defifindex == ifp->if_index)
638 		nd6_setdefaultiface(0);
639 
640 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
641 		/* refresh default router list */
642 		defrouter_select();
643 	}
644 
645 	/*
646 	 * Nuke neighbor cache entries for the ifp.
647 	 * Note that rt->rt_ifp may not be the same as ifp,
648 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
649 	 * nd6_rtrequest(), and ip6_input().
650 	 */
651 	ln = llinfo_nd6.ln_next;
652 	while (ln && ln != &llinfo_nd6) {
653 		struct rtentry *rt;
654 		struct sockaddr_dl *sdl;
655 
656 		nln = ln->ln_next;
657 		rt = ln->ln_rt;
658 		if (rt && rt->rt_gateway &&
659 		    rt->rt_gateway->sa_family == AF_LINK) {
660 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
661 			if (sdl->sdl_index == ifp->if_index)
662 				nln = nd6_free(rt, 0);
663 		}
664 		ln = nln;
665 	}
666 }
667 
668 struct rtentry *
669 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
670 {
671 	struct rtentry *rt;
672 	struct sockaddr_in6 sin6;
673 
674 	bzero(&sin6, sizeof(sin6));
675 	sin6.sin6_len = sizeof(struct sockaddr_in6);
676 	sin6.sin6_family = AF_INET6;
677 	sin6.sin6_addr = *addr6;
678 
679 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0);
680 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
681 		/*
682 		 * This is the case for the default route.
683 		 * If we want to create a neighbor cache for the address, we
684 		 * should free the route for the destination and allocate an
685 		 * interface route.
686 		 */
687 		if (create) {
688 			RTFREE(rt);
689 			rt = 0;
690 		}
691 	}
692 	if (!rt) {
693 		if (create && ifp) {
694 			struct rt_addrinfo info;
695 			int e;
696 
697 			/*
698 			 * If no route is available and create is set,
699 			 * we allocate a host route for the destination
700 			 * and treat it like an interface route.
701 			 * This hack is necessary for a neighbor which can't
702 			 * be covered by our own prefix.
703 			 */
704 			struct ifaddr *ifa =
705 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
706 			if (ifa == NULL)
707 				return (NULL);
708 
709 			/*
710 			 * Create a new route.  RTF_LLINFO is necessary
711 			 * to create a Neighbor Cache entry for the
712 			 * destination in nd6_rtrequest which will be
713 			 * called in rtrequest1 via ifa->ifa_rtrequest.
714 			 */
715 			bzero(&info, sizeof(info));
716 			info.rti_flags = (ifa->ifa_flags | RTF_HOST |
717 			    RTF_LLINFO) & ~RTF_CLONING;
718 			info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
719 			info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr;
720 			info.rti_info[RTAX_NETMASK] =
721 			    (struct sockaddr *)&all1_sa;
722 			if ((e = rtrequest1(RTM_ADD, &info, RTP_CONNECTED,
723 			    &rt, 0)) != 0) {
724 #if 0
725 				log(LOG_ERR,
726 				    "nd6_lookup: failed to add route for a "
727 				    "neighbor(%s), errno=%d\n",
728 				    ip6_sprintf(addr6), e);
729 #endif
730 				return (NULL);
731 			}
732 			if (rt == NULL)
733 				return (NULL);
734 			if (rt->rt_llinfo) {
735 				struct llinfo_nd6 *ln =
736 				    (struct llinfo_nd6 *)rt->rt_llinfo;
737 				ln->ln_state = ND6_LLINFO_NOSTATE;
738 			}
739 		} else
740 			return (NULL);
741 	}
742 	rt->rt_refcnt--;
743 	/*
744 	 * Validation for the entry.
745 	 * Note that the check for rt_llinfo is necessary because a cloned
746 	 * route from a parent route that has the L flag (e.g. the default
747 	 * route to a p2p interface) may have the flag, too, while the
748 	 * destination is not actually a neighbor.
749 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
750 	 *      it might be the loopback interface if the entry is for our
751 	 *      own address on a non-loopback interface. Instead, we should
752 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
753 	 *	interface.
754 	 */
755 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
756 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
757 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
758 		if (create) {
759 			nd6log((LOG_DEBUG,
760 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
761 			    ip6_sprintf(addr6),
762 			    ifp ? ifp->if_xname : "unspec"));
763 		}
764 		return (NULL);
765 	}
766 	return (rt);
767 }
768 
769 /*
770  * Detect if a given IPv6 address identifies a neighbor on a given link.
771  * XXX: should take care of the destination of a p2p link?
772  */
773 int
774 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
775 {
776 	struct nd_prefix *pr;
777 	struct rtentry *rt;
778 
779 	/*
780 	 * A link-local address is always a neighbor.
781 	 * XXX: we should use the sin6_scope_id field rather than the embedded
782 	 * interface index.
783 	 * XXX: a link does not necessarily specify a single interface.
784 	 */
785 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
786 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
787 		return (1);
788 
789 	/*
790 	 * If the address matches one of our on-link prefixes, it should be a
791 	 * neighbor.
792 	 */
793 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
794 		if (pr->ndpr_ifp != ifp)
795 			continue;
796 
797 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
798 			continue;
799 
800 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
801 		    &addr->sin6_addr, &pr->ndpr_mask))
802 			return (1);
803 	}
804 
805 	/*
806 	 * If the default router list is empty, all addresses are regarded
807 	 * as on-link, and thus, as a neighbor.
808 	 * XXX: we restrict the condition to hosts, because routers usually do
809 	 * not have the "default router list".
810 	 */
811 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
812 	    nd6_defifindex == ifp->if_index) {
813 		return (1);
814 	}
815 
816 	/*
817 	 * Even if the address matches none of our addresses, it might be
818 	 * in the neighbor cache.
819 	 */
820 	if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL)
821 		return (1);
822 
823 	return (0);
824 }
825 
826 /*
827  * Free an nd6 llinfo entry.
828  * Since the function would cause significant changes in the kernel, DO NOT
829  * make it global, unless you have a strong reason for the change, and are sure
830  * that the change is safe.
831  */
832 struct llinfo_nd6 *
833 nd6_free(struct rtentry *rt, int gc)
834 {
835 	struct rt_addrinfo info;
836 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
837 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
838 	struct nd_defrouter *dr;
839 
840 	/*
841 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
842 	 * even though it is not harmful, it was not really necessary.
843 	 */
844 
845 	if (!ip6_forwarding) {
846 		int s;
847 		s = splsoftnet();
848 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
849 		    rt->rt_ifp);
850 
851 		if (dr != NULL && dr->expire &&
852 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
853 			/*
854 			 * If the reason for the deletion is just garbage
855 			 * collection, and the neighbor is an active default
856 			 * router, do not delete it.  Instead, reset the GC
857 			 * timer using the router's lifetime.
858 			 * Simply deleting the entry would affect default
859 			 * router selection, which is not necessarily a good
860 			 * thing, especially when we're using router preference
861 			 * values.
862 			 * XXX: the check for ln_state would be redundant,
863 			 *      but we intentionally keep it just in case.
864 			 */
865 			if (dr->expire > time_second * hz) {
866 				nd6_llinfo_settimer(ln,
867 				    dr->expire - time_second * hz);
868 			} else
869 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
870 			splx(s);
871 			return (ln->ln_next);
872 		}
873 
874 		if (ln->ln_router || dr) {
875 			/*
876 			 * rt6_flush must be called whether or not the neighbor
877 			 * is in the Default Router List.
878 			 * See a corresponding comment in nd6_na_input().
879 			 */
880 			rt6_flush(&in6, rt->rt_ifp);
881 		}
882 
883 		if (dr) {
884 			/*
885 			 * Unreachability of a router might affect the default
886 			 * router selection and on-link detection of advertised
887 			 * prefixes.
888 			 */
889 
890 			/*
891 			 * Temporarily fake the state to choose a new default
892 			 * router and to perform on-link determination of
893 			 * prefixes correctly.
894 			 * Below the state will be set correctly,
895 			 * or the entry itself will be deleted.
896 			 */
897 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
898 
899 			/*
900 			 * Since defrouter_select() does not affect the
901 			 * on-link determination and MIP6 needs the check
902 			 * before the default router selection, we perform
903 			 * the check now.
904 			 */
905 			pfxlist_onlink_check();
906 
907 			/*
908 			 * refresh default router list
909 			 */
910 			defrouter_select();
911 		}
912 		splx(s);
913 	}
914 
915 	/*
916 	 * Before deleting the entry, remember the next entry as the
917 	 * return value.  We need this because pfxlist_onlink_check() above
918 	 * might have freed other entries (particularly the old next entry) as
919 	 * a side effect (XXX).
920 	 */
921 	next = ln->ln_next;
922 
923 	/*
924 	 * Detach the route from the routing tree and the list of neighbor
925 	 * caches, and disable the route entry not to be used in already
926 	 * cached routes.
927 	 */
928 	bzero(&info, sizeof(info));
929 	info.rti_info[RTAX_DST] = rt_key(rt);
930 	info.rti_info[RTAX_NETMASK] = rt_mask(rt);
931 	rtrequest1(RTM_DELETE, &info, rt->rt_priority, NULL, 0);
932 
933 	return (next);
934 }
935 
936 /*
937  * Upper-layer reachability hint for Neighbor Unreachability Detection.
938  *
939  * XXX cost-effective methods?
940  */
941 void
942 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
943 {
944 	struct llinfo_nd6 *ln;
945 
946 	/*
947 	 * If the caller specified "rt", use that.  Otherwise, resolve the
948 	 * routing table by supplied "dst6".
949 	 */
950 	if (!rt) {
951 		if (!dst6)
952 			return;
953 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
954 			return;
955 	}
956 
957 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
958 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
959 	    !rt->rt_llinfo || !rt->rt_gateway ||
960 	    rt->rt_gateway->sa_family != AF_LINK) {
961 		/* This is not a host route. */
962 		return;
963 	}
964 
965 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
966 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
967 		return;
968 
969 	/*
970 	 * if we get upper-layer reachability confirmation many times,
971 	 * it is possible we have false information.
972 	 */
973 	if (!force) {
974 		ln->ln_byhint++;
975 		if (ln->ln_byhint > nd6_maxnudhint)
976 			return;
977 	}
978 
979 	ln->ln_state = ND6_LLINFO_REACHABLE;
980 	if (!ND6_LLINFO_PERMANENT(ln)) {
981 		nd6_llinfo_settimer(ln,
982 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
983 	}
984 }
985 
986 /*
987  * info - XXX: unused
988  */
989 
990 void
991 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
992 {
993 	struct sockaddr *gate = rt->rt_gateway;
994 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
995 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
996 	struct ifnet *ifp = rt->rt_ifp;
997 	struct ifaddr *ifa;
998 	struct nd_defrouter *dr;
999 
1000 	if (req == RTM_DELETE && (rt->rt_flags & RTF_GATEWAY) &&
1001 	    (IN6_ARE_ADDR_EQUAL(&(satosin6(rt_key(rt)))->sin6_addr,
1002 	    &in6addr_any) && rt_mask(rt) && (rt_mask(rt)->sa_len == 0 ||
1003 	    IN6_ARE_ADDR_EQUAL(&(satosin6(rt_mask(rt)))->sin6_addr,
1004 	    &in6addr_any)))) {
1005 		dr = defrouter_lookup(&SIN6(gate)->sin6_addr, ifp);
1006 		if (dr)
1007 			dr->installed = 0;
1008 	}
1009 
1010 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1011 		return;
1012 
1013 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1014 		/*
1015 		 * This is probably an interface direct route for a link
1016 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1017 		 * We do not need special treatment below for such a route.
1018 		 * Moreover, the RTF_LLINFO flag which would be set below
1019 		 * would annoy the ndp(8) command.
1020 		 */
1021 		return;
1022 	}
1023 
1024 	if (req == RTM_RESOLVE && nd6_need_cache(ifp) == 0) {
1025 		/*
1026 		 * For routing daemons like ospf6d we allow neighbor discovery
1027 		 * based on the cloning route only.  This allows us to sent
1028 		 * packets directly into a network without having an address
1029 		 * with matching prefix on the interface.  If the cloning
1030 		 * route is used for an stf interface, we would mistakenly
1031 		 * make a neighbor cache for the host route, and would see
1032 		 * strange neighbor solicitation for the corresponding
1033 		 * destination.  In order to avoid confusion, we check if the
1034 		 * interface is suitable for neighbor discovery, and stop the
1035 		 * process if not.  Additionally, we remove the LLINFO flag
1036 		 * so that ndp(8) will not try to get the neighbor information
1037 		 * of the destination.
1038 		 */
1039 		rt->rt_flags &= ~RTF_LLINFO;
1040 		return;
1041 	}
1042 
1043 	switch (req) {
1044 	case RTM_ADD:
1045 		/*
1046 		 * There is no backward compatibility :)
1047 		 *
1048 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1049 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1050 		 *	   rt->rt_flags |= RTF_CLONING;
1051 		 */
1052 		if ((rt->rt_flags & RTF_CLONING) ||
1053 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1054 			/*
1055 			 * Case 1: This route should come from a route to
1056 			 * interface (RTF_CLONING case) or the route should be
1057 			 * treated as on-link but is currently not
1058 			 * (RTF_LLINFO && !ln case).
1059 			 */
1060 			rt_setgate(rt, rt_key(rt),
1061 				   (struct sockaddr *)&null_sdl, 0);
1062 			gate = rt->rt_gateway;
1063 			SDL(gate)->sdl_type = ifp->if_type;
1064 			SDL(gate)->sdl_index = ifp->if_index;
1065 			if (ln)
1066 				nd6_llinfo_settimer(ln, 0);
1067 			if ((rt->rt_flags & RTF_CLONING) != 0)
1068 				break;
1069 		}
1070 		/*
1071 		 * In IPv4 code, we try to announce new RTF_ANNOUNCE entry here.
1072 		 * We don't do that here since llinfo is not ready yet.
1073 		 *
1074 		 * There are also couple of other things to be discussed:
1075 		 * - unsolicited NA code needs improvement beforehand
1076 		 * - RFC2461 says we MAY send multicast unsolicited NA
1077 		 *   (7.2.6 paragraph 4), however, it also says that we
1078 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1079 		 *   we don't have anything like it right now.
1080 		 *   note that the mechanism needs a mutual agreement
1081 		 *   between proxies, which means that we need to implement
1082 		 *   a new protocol, or a new kludge.
1083 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1084 		 *   we need to check ip6forwarding before sending it.
1085 		 *   (or should we allow proxy ND configuration only for
1086 		 *   routers?  there's no mention about proxy ND from hosts)
1087 		 */
1088 #if 0
1089 		/* XXX it does not work */
1090 		if (rt->rt_flags & RTF_ANNOUNCE)
1091 			nd6_na_output(ifp,
1092 			      &SIN6(rt_key(rt))->sin6_addr,
1093 			      &SIN6(rt_key(rt))->sin6_addr,
1094 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1095 			      1, NULL);
1096 #endif
1097 		/* FALLTHROUGH */
1098 	case RTM_RESOLVE:
1099 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1100 			/*
1101 			 * Address resolution isn't necessary for a point to
1102 			 * point link, so we can skip this test for a p2p link.
1103 			 */
1104 			if (gate->sa_family != AF_LINK ||
1105 			    gate->sa_len < sizeof(null_sdl)) {
1106 				log(LOG_DEBUG,
1107 				    "nd6_rtrequest: bad gateway value: %s\n",
1108 				    ifp->if_xname);
1109 				break;
1110 			}
1111 			SDL(gate)->sdl_type = ifp->if_type;
1112 			SDL(gate)->sdl_index = ifp->if_index;
1113 		}
1114 		if (ln != NULL)
1115 			break;	/* This happens on a route change */
1116 		/*
1117 		 * Case 2: This route may come from cloning, or a manual route
1118 		 * add with a LL address.
1119 		 */
1120 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1121 		rt->rt_llinfo = (caddr_t)ln;
1122 		if (!ln) {
1123 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1124 			break;
1125 		}
1126 		nd6_inuse++;
1127 		nd6_allocated++;
1128 		Bzero(ln, sizeof(*ln));
1129 		ln->ln_rt = rt;
1130 		timeout_set(&ln->ln_timer_ch, nd6_llinfo_timer, ln);
1131 		/* this is required for "ndp" command. - shin */
1132 		if (req == RTM_ADD) {
1133 		        /*
1134 			 * gate should have some valid AF_LINK entry,
1135 			 * and ln->ln_expire should have some lifetime
1136 			 * which is specified by ndp command.
1137 			 */
1138 			ln->ln_state = ND6_LLINFO_REACHABLE;
1139 			ln->ln_byhint = 0;
1140 		} else {
1141 		        /*
1142 			 * When req == RTM_RESOLVE, rt is created and
1143 			 * initialized in rtrequest(), so rt_expire is 0.
1144 			 */
1145 			ln->ln_state = ND6_LLINFO_NOSTATE;
1146 			nd6_llinfo_settimer(ln, 0);
1147 		}
1148 		rt->rt_flags |= RTF_LLINFO;
1149 		ln->ln_next = llinfo_nd6.ln_next;
1150 		llinfo_nd6.ln_next = ln;
1151 		ln->ln_prev = &llinfo_nd6;
1152 		ln->ln_next->ln_prev = ln;
1153 
1154 		/*
1155 		 * If we have too many cache entries, initiate immediate
1156 		 * purging for some "less recently used" entries.  Note that
1157 		 * we cannot directly call nd6_free() here because it would
1158 		 * cause re-entering rtable related routines triggering an LOR
1159 		 * problem for FreeBSD.
1160 		 */
1161 		if (ip6_neighborgcthresh >= 0 &&
1162 		    nd6_inuse >= ip6_neighborgcthresh) {
1163 			int i;
1164 
1165 			for (i = 0; i < 10 && llinfo_nd6.ln_prev != ln; i++) {
1166 				struct llinfo_nd6 *ln_end = llinfo_nd6.ln_prev;
1167 
1168 				/* Move this entry to the head */
1169 				LN_DEQUEUE(ln_end);
1170 				LN_INSERTHEAD(ln_end);
1171 
1172 				if (ND6_LLINFO_PERMANENT(ln_end))
1173 					continue;
1174 
1175 				if (ln_end->ln_state > ND6_LLINFO_INCOMPLETE)
1176 					ln_end->ln_state = ND6_LLINFO_STALE;
1177 				else
1178 					ln_end->ln_state = ND6_LLINFO_PURGE;
1179 				nd6_llinfo_settimer(ln_end, 0);
1180 			}
1181 		}
1182 
1183 		/*
1184 		 * check if rt_key(rt) is one of my address assigned
1185 		 * to the interface.
1186 		 */
1187 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1188 		    &SIN6(rt_key(rt))->sin6_addr);
1189 		if (ifa) {
1190 			caddr_t macp = nd6_ifptomac(ifp);
1191 			nd6_llinfo_settimer(ln, -1);
1192 			ln->ln_state = ND6_LLINFO_REACHABLE;
1193 			ln->ln_byhint = 0;
1194 			if (macp) {
1195 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1196 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1197 			}
1198 			if (nd6_useloopback) {
1199 				rt->rt_ifp = lo0ifp;	/*XXX*/
1200 				/*
1201 				 * Make sure rt_ifa be equal to the ifaddr
1202 				 * corresponding to the address.
1203 				 * We need this because when we refer
1204 				 * rt_ifa->ia6_flags in ip6_input, we assume
1205 				 * that the rt_ifa points to the address instead
1206 				 * of the loopback address.
1207 				 */
1208 				if (ifa != rt->rt_ifa) {
1209 					IFAFREE(rt->rt_ifa);
1210 					ifa->ifa_refcnt++;
1211 					rt->rt_ifa = ifa;
1212 				}
1213 			}
1214 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1215 			nd6_llinfo_settimer(ln, -1);
1216 			ln->ln_state = ND6_LLINFO_REACHABLE;
1217 			ln->ln_byhint = 0;
1218 
1219 			/* join solicited node multicast for proxy ND */
1220 			if (ifp->if_flags & IFF_MULTICAST) {
1221 				struct in6_addr llsol;
1222 				int error;
1223 
1224 				llsol = SIN6(rt_key(rt))->sin6_addr;
1225 				llsol.s6_addr16[0] = htons(0xff02);
1226 				llsol.s6_addr16[1] = htons(ifp->if_index);
1227 				llsol.s6_addr32[1] = 0;
1228 				llsol.s6_addr32[2] = htonl(1);
1229 				llsol.s6_addr8[12] = 0xff;
1230 
1231 				if (in6_addmulti(&llsol, ifp, &error)) {
1232 					nd6log((LOG_ERR, "%s: failed to join "
1233 					    "%s (errno=%d)\n", ifp->if_xname,
1234 					    ip6_sprintf(&llsol), error));
1235 				}
1236 			}
1237 		}
1238 		break;
1239 
1240 	case RTM_DELETE:
1241 		if (!ln)
1242 			break;
1243 		/* leave from solicited node multicast for proxy ND */
1244 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1245 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1246 			struct in6_addr llsol;
1247 			struct in6_multi *in6m;
1248 
1249 			llsol = SIN6(rt_key(rt))->sin6_addr;
1250 			llsol.s6_addr16[0] = htons(0xff02);
1251 			llsol.s6_addr16[1] = htons(ifp->if_index);
1252 			llsol.s6_addr32[1] = 0;
1253 			llsol.s6_addr32[2] = htonl(1);
1254 			llsol.s6_addr8[12] = 0xff;
1255 
1256 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1257 			if (in6m)
1258 				in6_delmulti(in6m);
1259 		}
1260 		nd6_inuse--;
1261 		ln->ln_next->ln_prev = ln->ln_prev;
1262 		ln->ln_prev->ln_next = ln->ln_next;
1263 		ln->ln_prev = NULL;
1264 		nd6_llinfo_settimer(ln, -1);
1265 		rt->rt_llinfo = 0;
1266 		rt->rt_flags &= ~RTF_LLINFO;
1267 		if (ln->ln_hold)
1268 			m_freem(ln->ln_hold);
1269 		Free((caddr_t)ln);
1270 	}
1271 }
1272 
1273 int
1274 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1275 {
1276 	struct in6_drlist *drl = (struct in6_drlist *)data;
1277 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1278 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1279 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1280 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1281 	struct nd_defrouter *dr;
1282 	struct nd_prefix *pr;
1283 	struct rtentry *rt;
1284 	int i = 0, error = 0;
1285 	int s;
1286 
1287 	switch (cmd) {
1288 	case SIOCGDRLST_IN6:
1289 		/*
1290 		 * obsolete API, use sysctl under net.inet6.icmp6
1291 		 */
1292 		bzero(drl, sizeof(*drl));
1293 		s = splsoftnet();
1294 		dr = TAILQ_FIRST(&nd_defrouter);
1295 		while (dr && i < DRLSTSIZ) {
1296 			drl->defrouter[i].rtaddr = dr->rtaddr;
1297 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1298 				/* XXX: need to this hack for KAME stack */
1299 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1300 			} else
1301 				log(LOG_ERR,
1302 				    "default router list contains a "
1303 				    "non-linklocal address(%s)\n",
1304 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1305 
1306 			drl->defrouter[i].flags = dr->flags;
1307 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1308 			drl->defrouter[i].expire = dr->expire;
1309 			drl->defrouter[i].if_index = dr->ifp->if_index;
1310 			i++;
1311 			dr = TAILQ_NEXT(dr, dr_entry);
1312 		}
1313 		splx(s);
1314 		break;
1315 	case SIOCGPRLST_IN6:
1316 		/*
1317 		 * obsolete API, use sysctl under net.inet6.icmp6
1318 		 *
1319 		 * XXX the structure in6_prlist was changed in backward-
1320 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1321 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1322 		 */
1323 		/*
1324 		 * XXX meaning of fields, especially "raflags", is very
1325 		 * different between RA prefix list and RR/static prefix list.
1326 		 * how about separating ioctls into two?
1327 		 */
1328 		bzero(oprl, sizeof(*oprl));
1329 		s = splsoftnet();
1330 		pr = LIST_FIRST(&nd_prefix);
1331 		while (pr && i < PRLSTSIZ) {
1332 			struct nd_pfxrouter *pfr;
1333 			int j;
1334 
1335 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1336 			oprl->prefix[i].raflags = pr->ndpr_raf;
1337 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1338 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1339 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1340 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1341 			oprl->prefix[i].expire = pr->ndpr_expire;
1342 
1343 			pfr = LIST_FIRST(&pr->ndpr_advrtrs);
1344 			j = 0;
1345 			while(pfr) {
1346 				if (j < DRLSTSIZ) {
1347 #define RTRADDR oprl->prefix[i].advrtr[j]
1348 					RTRADDR = pfr->router->rtaddr;
1349 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1350 						/* XXX: hack for KAME */
1351 						RTRADDR.s6_addr16[1] = 0;
1352 					} else
1353 						log(LOG_ERR,
1354 						    "a router(%s) advertises "
1355 						    "a prefix with "
1356 						    "non-link local address\n",
1357 						    ip6_sprintf(&RTRADDR));
1358 #undef RTRADDR
1359 				}
1360 				j++;
1361 				pfr = LIST_NEXT(pfr, pfr_entry);
1362 			}
1363 			oprl->prefix[i].advrtrs = j;
1364 			oprl->prefix[i].origin = PR_ORIG_RA;
1365 
1366 			i++;
1367 			pr = LIST_NEXT(pr, ndpr_entry);
1368 		}
1369 		splx(s);
1370 
1371 		break;
1372 	case OSIOCGIFINFO_IN6:
1373 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1374 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1375 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1376 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1377 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1378 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1379 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1380 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1381 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1382 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1383 		break;
1384 	case SIOCGIFINFO_IN6:
1385 		ndi->ndi = *ND_IFINFO(ifp);
1386 		break;
1387 	case SIOCSIFINFO_FLAGS:
1388 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1389 		break;
1390 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1391 		/* sync kernel routing table with the default router list */
1392 		defrouter_reset();
1393 		defrouter_select();
1394 		break;
1395 	case SIOCSPFXFLUSH_IN6:
1396 	{
1397 		/* flush all the prefix advertised by routers */
1398 		struct nd_prefix *pr, *next;
1399 
1400 		s = splsoftnet();
1401 		for (pr = LIST_FIRST(&nd_prefix); pr; pr = next) {
1402 			struct in6_ifaddr *ia, *ia_next;
1403 
1404 			next = LIST_NEXT(pr, ndpr_entry);
1405 
1406 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1407 				continue; /* XXX */
1408 
1409 			/* do we really have to remove addresses as well? */
1410 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1411 				/* ia might be removed.  keep the next ptr. */
1412 				ia_next = ia->ia_next;
1413 
1414 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1415 					continue;
1416 
1417 				if (ia->ia6_ndpr == pr)
1418 					in6_purgeaddr(&ia->ia_ifa);
1419 			}
1420 			prelist_remove(pr);
1421 		}
1422 		splx(s);
1423 		break;
1424 	}
1425 	case SIOCSRTRFLUSH_IN6:
1426 	{
1427 		/* flush all the default routers */
1428 		struct nd_defrouter *dr, *next;
1429 
1430 		s = splsoftnet();
1431 		defrouter_reset();
1432 		for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) {
1433 			next = TAILQ_NEXT(dr, dr_entry);
1434 			defrtrlist_del(dr);
1435 		}
1436 		defrouter_select();
1437 		splx(s);
1438 		break;
1439 	}
1440 	case SIOCGNBRINFO_IN6:
1441 	{
1442 		struct llinfo_nd6 *ln;
1443 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1444 
1445 		/*
1446 		 * XXX: KAME specific hack for scoped addresses
1447 		 *      XXXX: for other scopes than link-local?
1448 		 */
1449 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1450 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1451 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1452 
1453 			if (*idp == 0)
1454 				*idp = htons(ifp->if_index);
1455 		}
1456 
1457 		s = splsoftnet();
1458 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1459 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1460 			error = EINVAL;
1461 			splx(s);
1462 			break;
1463 		}
1464 		nbi->state = ln->ln_state;
1465 		nbi->asked = ln->ln_asked;
1466 		nbi->isrouter = ln->ln_router;
1467 		nbi->expire = ln->ln_expire;
1468 		splx(s);
1469 
1470 		break;
1471 	}
1472 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1473 		ndif->ifindex = nd6_defifindex;
1474 		break;
1475 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1476 		return (nd6_setdefaultiface(ndif->ifindex));
1477 		break;
1478 	}
1479 	return (error);
1480 }
1481 
1482 /*
1483  * Create neighbor cache entry and cache link-layer address,
1484  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1485  *
1486  * type - ICMP6 type
1487  * code - type dependent information
1488  */
1489 struct rtentry *
1490 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1491     int lladdrlen, int type, int code)
1492 {
1493 	struct rtentry *rt = NULL;
1494 	struct llinfo_nd6 *ln = NULL;
1495 	int is_newentry;
1496 	struct sockaddr_dl *sdl = NULL;
1497 	int do_update;
1498 	int olladdr;
1499 	int llchange;
1500 	int newstate = 0;
1501 
1502 	if (!ifp)
1503 		panic("ifp == NULL in nd6_cache_lladdr");
1504 	if (!from)
1505 		panic("from == NULL in nd6_cache_lladdr");
1506 
1507 	/* nothing must be updated for unspecified address */
1508 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1509 		return NULL;
1510 
1511 	/*
1512 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1513 	 * the caller.
1514 	 *
1515 	 * XXX If the link does not have link-layer address, what should
1516 	 * we do? (ifp->if_addrlen == 0)
1517 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1518 	 * description on it in NS section (RFC 2461 7.2.3).
1519 	 */
1520 
1521 	rt = nd6_lookup(from, 0, ifp);
1522 	if (!rt) {
1523 #if 0
1524 		/* nothing must be done if there's no lladdr */
1525 		if (!lladdr || !lladdrlen)
1526 			return NULL;
1527 #endif
1528 
1529 		rt = nd6_lookup(from, RT_REPORT, ifp);
1530 		is_newentry = 1;
1531 	} else {
1532 		/* do nothing if static ndp is set */
1533 		if (rt->rt_flags & RTF_STATIC)
1534 			return NULL;
1535 		is_newentry = 0;
1536 	}
1537 
1538 	if (!rt)
1539 		return NULL;
1540 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1541 fail:
1542 		(void)nd6_free(rt, 0);
1543 		return NULL;
1544 	}
1545 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1546 	if (!ln)
1547 		goto fail;
1548 	if (!rt->rt_gateway)
1549 		goto fail;
1550 	if (rt->rt_gateway->sa_family != AF_LINK)
1551 		goto fail;
1552 	sdl = SDL(rt->rt_gateway);
1553 
1554 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1555 	if (olladdr && lladdr) {
1556 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1557 			llchange = 1;
1558 		else
1559 			llchange = 0;
1560 	} else
1561 		llchange = 0;
1562 
1563 	/*
1564 	 * newentry olladdr  lladdr  llchange	(*=record)
1565 	 *	0	n	n	--	(1)
1566 	 *	0	y	n	--	(2)
1567 	 *	0	n	y	--	(3) * STALE
1568 	 *	0	y	y	n	(4) *
1569 	 *	0	y	y	y	(5) * STALE
1570 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1571 	 *	1	--	y	--	(7) * STALE
1572 	 */
1573 
1574 	if (llchange) {
1575 		log(LOG_INFO, "ndp info overwritten for %s by %s on %s\n",
1576 		    ip6_sprintf(from), ether_sprintf(lladdr), ifp->if_xname);
1577 	}
1578 	if (lladdr) {		/* (3-5) and (7) */
1579 		/*
1580 		 * Record source link-layer address
1581 		 * XXX is it dependent to ifp->if_type?
1582 		 */
1583 		sdl->sdl_alen = ifp->if_addrlen;
1584 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1585 	}
1586 
1587 	if (!is_newentry) {
1588 		if ((!olladdr && lladdr) ||		/* (3) */
1589 		    (olladdr && lladdr && llchange)) {	/* (5) */
1590 			do_update = 1;
1591 			newstate = ND6_LLINFO_STALE;
1592 		} else					/* (1-2,4) */
1593 			do_update = 0;
1594 	} else {
1595 		do_update = 1;
1596 		if (!lladdr)				/* (6) */
1597 			newstate = ND6_LLINFO_NOSTATE;
1598 		else					/* (7) */
1599 			newstate = ND6_LLINFO_STALE;
1600 	}
1601 
1602 	if (do_update) {
1603 		/*
1604 		 * Update the state of the neighbor cache.
1605 		 */
1606 		ln->ln_state = newstate;
1607 
1608 		if (ln->ln_state == ND6_LLINFO_STALE) {
1609 			/*
1610 			 * XXX: since nd6_output() below will cause
1611 			 * state transition to DELAY and reset the timer,
1612 			 * we must set the timer now, although it is actually
1613 			 * meaningless.
1614 			 */
1615 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1616 
1617 			if (ln->ln_hold) {
1618 				struct mbuf *n = ln->ln_hold;
1619 				ln->ln_hold = NULL;
1620 				/*
1621 				 * we assume ifp is not a p2p here, so just
1622 				 * set the 2nd argument as the 1st one.
1623 				 */
1624 				nd6_output(ifp, ifp, n,
1625 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1626 				if (ln->ln_hold == n) {
1627 					/* n is back in ln_hold. Discard. */
1628 					m_freem(ln->ln_hold);
1629 					ln->ln_hold = NULL;
1630 				}
1631 			}
1632 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1633 			/* probe right away */
1634 			nd6_llinfo_settimer((void *)ln, 0);
1635 		}
1636 	}
1637 
1638 	/*
1639 	 * ICMP6 type dependent behavior.
1640 	 *
1641 	 * NS: clear IsRouter if new entry
1642 	 * RS: clear IsRouter
1643 	 * RA: set IsRouter if there's lladdr
1644 	 * redir: clear IsRouter if new entry
1645 	 *
1646 	 * RA case, (1):
1647 	 * The spec says that we must set IsRouter in the following cases:
1648 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1649 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1650 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1651 	 * A question arises for (1) case.  (1) case has no lladdr in the
1652 	 * neighbor cache, this is similar to (6).
1653 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1654 	 *
1655 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1656 	 *							D R
1657 	 *	0	n	n	--	(1)	c   ?     s
1658 	 *	0	y	n	--	(2)	c   s     s
1659 	 *	0	n	y	--	(3)	c   s     s
1660 	 *	0	y	y	n	(4)	c   s     s
1661 	 *	0	y	y	y	(5)	c   s     s
1662 	 *	1	--	n	--	(6) c	c 	c s
1663 	 *	1	--	y	--	(7) c	c   s	c s
1664 	 *
1665 	 *					(c=clear s=set)
1666 	 */
1667 	switch (type & 0xff) {
1668 	case ND_NEIGHBOR_SOLICIT:
1669 		/*
1670 		 * New entry must have is_router flag cleared.
1671 		 */
1672 		if (is_newentry)	/* (6-7) */
1673 			ln->ln_router = 0;
1674 		break;
1675 	case ND_REDIRECT:
1676 		/*
1677 		 * If the icmp is a redirect to a better router, always set the
1678 		 * is_router flag.  Otherwise, if the entry is newly created,
1679 		 * clear the flag.  [RFC 2461, sec 8.3]
1680 		 */
1681 		if (code == ND_REDIRECT_ROUTER)
1682 			ln->ln_router = 1;
1683 		else if (is_newentry) /* (6-7) */
1684 			ln->ln_router = 0;
1685 		break;
1686 	case ND_ROUTER_SOLICIT:
1687 		/*
1688 		 * is_router flag must always be cleared.
1689 		 */
1690 		ln->ln_router = 0;
1691 		break;
1692 	case ND_ROUTER_ADVERT:
1693 		/*
1694 		 * Mark an entry with lladdr as a router.
1695 		 */
1696 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1697 		    (is_newentry && lladdr)) {			/* (7) */
1698 			ln->ln_router = 1;
1699 		}
1700 		break;
1701 	}
1702 
1703 	/*
1704 	 * When the link-layer address of a router changes, select the
1705 	 * best router again.  In particular, when the neighbor entry is newly
1706 	 * created, it might affect the selection policy.
1707 	 * Question: can we restrict the first condition to the "is_newentry"
1708 	 * case?
1709 	 * XXX: when we hear an RA from a new router with the link-layer
1710 	 * address option, defrouter_select() is called twice, since
1711 	 * defrtrlist_update called the function as well.  However, I believe
1712 	 * we can compromise the overhead, since it only happens the first
1713 	 * time.
1714 	 * XXX: although defrouter_select() should not have a bad effect
1715 	 * for those are not autoconfigured hosts, we explicitly avoid such
1716 	 * cases for safety.
1717 	 */
1718 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1719 		defrouter_select();
1720 
1721 	return rt;
1722 }
1723 
1724 void
1725 nd6_slowtimo(void *ignored_arg)
1726 {
1727 	int s = splsoftnet();
1728 	struct nd_ifinfo *nd6if;
1729 	struct ifnet *ifp;
1730 
1731 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
1732 	timeout_add_sec(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL);
1733 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1734 	{
1735 		nd6if = ND_IFINFO(ifp);
1736 		if (nd6if->basereachable && /* already initialized */
1737 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1738 			/*
1739 			 * Since reachable time rarely changes by router
1740 			 * advertisements, we SHOULD insure that a new random
1741 			 * value gets recomputed at least once every few hours.
1742 			 * (RFC 2461, 6.3.4)
1743 			 */
1744 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1745 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1746 		}
1747 	}
1748 	splx(s);
1749 }
1750 
1751 #define senderr(e) { error = (e); goto bad;}
1752 int
1753 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1754     struct sockaddr_in6 *dst, struct rtentry *rt0)
1755 {
1756 	struct mbuf *m = m0;
1757 	struct rtentry *rt = rt0;
1758 	struct sockaddr_in6 *gw6 = NULL;
1759 	struct llinfo_nd6 *ln = NULL;
1760 	int error = 0;
1761 #ifdef IPSEC
1762 	struct m_tag *mtag;
1763 #endif /* IPSEC */
1764 
1765 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1766 		goto sendpkt;
1767 
1768 	if (nd6_need_cache(ifp) == 0)
1769 		goto sendpkt;
1770 
1771 	/*
1772 	 * next hop determination.  This routine is derived from ether_output.
1773 	 */
1774 	if (rt) {
1775 		if ((rt->rt_flags & RTF_UP) == 0) {
1776 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1777 			    RT_REPORT, 0)) != NULL)
1778 			{
1779 				rt->rt_refcnt--;
1780 				if (rt->rt_ifp != ifp)
1781 					senderr(EHOSTUNREACH);
1782 			} else
1783 				senderr(EHOSTUNREACH);
1784 		}
1785 
1786 		if (rt->rt_flags & RTF_GATEWAY) {
1787 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1788 
1789 			/*
1790 			 * We skip link-layer address resolution and NUD
1791 			 * if the gateway is not a neighbor from ND point
1792 			 * of view, regardless of the value of nd_ifinfo.flags.
1793 			 * The second condition is a bit tricky; we skip
1794 			 * if the gateway is our own address, which is
1795 			 * sometimes used to install a route to a p2p link.
1796 			 */
1797 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1798 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1799 				/*
1800 				 * We allow this kind of tricky route only
1801 				 * when the outgoing interface is p2p.
1802 				 * XXX: we may need a more generic rule here.
1803 				 */
1804 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1805 					senderr(EHOSTUNREACH);
1806 
1807 				goto sendpkt;
1808 			}
1809 
1810 			if (rt->rt_gwroute == 0)
1811 				goto lookup;
1812 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1813 				rtfree(rt); rt = rt0;
1814 			lookup:
1815 				rt->rt_gwroute = rtalloc1(rt->rt_gateway,
1816 				    RT_REPORT, 0);
1817 				if ((rt = rt->rt_gwroute) == 0)
1818 					senderr(EHOSTUNREACH);
1819 			}
1820 		}
1821 	}
1822 
1823 	/*
1824 	 * Address resolution or Neighbor Unreachability Detection
1825 	 * for the next hop.
1826 	 * At this point, the destination of the packet must be a unicast
1827 	 * or an anycast address(i.e. not a multicast).
1828 	 */
1829 
1830 	/* Look up the neighbor cache for the nexthop */
1831 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1832 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1833 	else {
1834 		/*
1835 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1836 		 * the condition below is not very efficient.  But we believe
1837 		 * it is tolerable, because this should be a rare case.
1838 		 */
1839 		if (nd6_is_addr_neighbor(dst, ifp) &&
1840 		    (rt = nd6_lookup(&dst->sin6_addr, RT_REPORT, ifp)) != NULL)
1841 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1842 	}
1843 	if (!ln || !rt) {
1844 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1845 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1846 			log(LOG_DEBUG,
1847 			    "nd6_output: can't allocate llinfo for %s "
1848 			    "(ln=%p, rt=%p)\n",
1849 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1850 			senderr(EIO);	/* XXX: good error? */
1851 		}
1852 
1853 		goto sendpkt;	/* send anyway */
1854 	}
1855 
1856 	/*
1857 	 * Move this entry to the head of the queue so that it is less likely
1858 	 * for this entry to be a target of forced garbage collection (see
1859 	 * nd6_rtrequest()).
1860 	 */
1861 	LN_DEQUEUE(ln);
1862 	LN_INSERTHEAD(ln);
1863 
1864 	/* We don't have to do link-layer address resolution on a p2p link. */
1865 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1866 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1867 		ln->ln_state = ND6_LLINFO_STALE;
1868 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1869 	}
1870 
1871 	/*
1872 	 * The first time we send a packet to a neighbor whose entry is
1873 	 * STALE, we have to change the state to DELAY and a sets a timer to
1874 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1875 	 * neighbor unreachability detection on expiration.
1876 	 * (RFC 2461 7.3.3)
1877 	 */
1878 	if (ln->ln_state == ND6_LLINFO_STALE) {
1879 		ln->ln_asked = 0;
1880 		ln->ln_state = ND6_LLINFO_DELAY;
1881 		nd6_llinfo_settimer(ln, nd6_delay * hz);
1882 	}
1883 
1884 	/*
1885 	 * If the neighbor cache entry has a state other than INCOMPLETE
1886 	 * (i.e. its link-layer address is already resolved), just
1887 	 * send the packet.
1888 	 */
1889 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1890 		goto sendpkt;
1891 
1892 	/*
1893 	 * There is a neighbor cache entry, but no ethernet address
1894 	 * response yet.  Replace the held mbuf (if any) with this
1895 	 * latest one.
1896 	 */
1897 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1898 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1899 	if (ln->ln_hold)
1900 		m_freem(ln->ln_hold);
1901 	ln->ln_hold = m;
1902 	/*
1903 	 * If there has been no NS for the neighbor after entering the
1904 	 * INCOMPLETE state, send the first solicitation.
1905 	 */
1906 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
1907 		ln->ln_asked++;
1908 		nd6_llinfo_settimer(ln,
1909 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1910 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1911 	}
1912 	return (0);
1913 
1914   sendpkt:
1915 #ifdef IPSEC
1916 	/*
1917 	 * If the packet needs outgoing IPsec crypto processing and the
1918 	 * interface doesn't support it, drop it.
1919 	 */
1920 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
1921 #endif /* IPSEC */
1922 
1923 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1924 #ifdef IPSEC
1925 		if (mtag != NULL &&
1926 		    (origifp->if_capabilities & IFCAP_IPSEC) == 0) {
1927 			/* Tell IPsec to do its own crypto. */
1928 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
1929 			error = EACCES;
1930 			goto bad;
1931 		}
1932 #endif /* IPSEC */
1933 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1934 		    rt));
1935 	}
1936 #ifdef IPSEC
1937 	if (mtag != NULL &&
1938 	    (ifp->if_capabilities & IFCAP_IPSEC) == 0) {
1939 		/* Tell IPsec to do its own crypto. */
1940 		ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
1941 		error = EACCES;
1942 		goto bad;
1943 	}
1944 #endif /* IPSEC */
1945 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1946 
1947   bad:
1948 	if (m)
1949 		m_freem(m);
1950 	return (error);
1951 }
1952 #undef senderr
1953 
1954 int
1955 nd6_need_cache(struct ifnet *ifp)
1956 {
1957 	/*
1958 	 * XXX: we currently do not make neighbor cache on any interface
1959 	 * other than Ethernet, FDDI and GIF.
1960 	 *
1961 	 * RFC2893 says:
1962 	 * - unidirectional tunnels needs no ND
1963 	 */
1964 	switch (ifp->if_type) {
1965 	case IFT_ETHER:
1966 	case IFT_FDDI:
1967 	case IFT_IEEE1394:
1968 	case IFT_PROPVIRTUAL:
1969 	case IFT_L2VLAN:
1970 	case IFT_IEEE80211:
1971 	case IFT_CARP:
1972 	case IFT_GIF:		/* XXX need more cases? */
1973 		return (1);
1974 	default:
1975 		return (0);
1976 	}
1977 }
1978 
1979 int
1980 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt, struct mbuf *m,
1981     struct sockaddr *dst, u_char *desten)
1982 {
1983 	struct sockaddr_dl *sdl;
1984 
1985 	if (m->m_flags & M_MCAST) {
1986 		switch (ifp->if_type) {
1987 		case IFT_ETHER:
1988 		case IFT_FDDI:
1989 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1990 						 desten);
1991 			return (1);
1992 			break;
1993 		default:
1994 			m_freem(m);
1995 			return (0);
1996 		}
1997 	}
1998 
1999 	if (rt == NULL) {
2000 		/* this could happen, if we could not allocate memory */
2001 		m_freem(m);
2002 		return (0);
2003 	}
2004 	if (rt->rt_gateway->sa_family != AF_LINK) {
2005 		printf("nd6_storelladdr: something odd happens\n");
2006 		m_freem(m);
2007 		return (0);
2008 	}
2009 	sdl = SDL(rt->rt_gateway);
2010 	if (sdl->sdl_alen == 0) {
2011 		/* this should be impossible, but we bark here for debugging */
2012 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
2013 		    ip6_sprintf(&SIN6(dst)->sin6_addr), ifp->if_xname);
2014 		m_freem(m);
2015 		return (0);
2016 	}
2017 
2018 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2019 	return (1);
2020 }
2021 
2022 /*
2023  * oldp - syscall arg, need copyout
2024  * newp - syscall arg, need copyin
2025  */
2026 
2027 int
2028 nd6_sysctl(int name, void *oldp, size_t *oldlenp, void *newp, size_t newlen)
2029 {
2030 	void *p;
2031 	size_t ol;
2032 	int error;
2033 
2034 	error = 0;
2035 
2036 	if (newp)
2037 		return EPERM;
2038 	if (oldp && !oldlenp)
2039 		return EINVAL;
2040 	ol = oldlenp ? *oldlenp : 0;
2041 
2042 	if (oldp) {
2043 		p = malloc(*oldlenp, M_TEMP, M_WAITOK | M_CANFAIL);
2044 		if (!p)
2045 			return ENOMEM;
2046 	} else
2047 		p = NULL;
2048 	switch (name) {
2049 	case ICMPV6CTL_ND6_DRLIST:
2050 		error = fill_drlist(p, oldlenp, ol);
2051 		if (!error && p && oldp)
2052 			error = copyout(p, oldp, *oldlenp);
2053 		break;
2054 
2055 	case ICMPV6CTL_ND6_PRLIST:
2056 		error = fill_prlist(p, oldlenp, ol);
2057 		if (!error && p && oldp)
2058 			error = copyout(p, oldp, *oldlenp);
2059 		break;
2060 
2061 	default:
2062 		error = ENOPROTOOPT;
2063 		break;
2064 	}
2065 	if (p)
2066 		free(p, M_TEMP);
2067 
2068 	return (error);
2069 }
2070 
2071 int
2072 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2073 {
2074 	int error = 0, s;
2075 	struct in6_defrouter *d = NULL, *de = NULL;
2076 	struct nd_defrouter *dr;
2077 	size_t l;
2078 
2079 	s = splsoftnet();
2080 
2081 	if (oldp) {
2082 		d = (struct in6_defrouter *)oldp;
2083 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2084 	}
2085 	l = 0;
2086 
2087 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2088 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2089 
2090 		if (oldp && d + 1 <= de) {
2091 			bzero(d, sizeof(*d));
2092 			d->rtaddr.sin6_family = AF_INET6;
2093 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2094 			d->rtaddr.sin6_addr = dr->rtaddr;
2095 			in6_recoverscope(&d->rtaddr, &d->rtaddr.sin6_addr,
2096 			    dr->ifp);
2097 			d->flags = dr->flags;
2098 			d->rtlifetime = dr->rtlifetime;
2099 			d->expire = dr->expire;
2100 			d->if_index = dr->ifp->if_index;
2101 		}
2102 
2103 		l += sizeof(*d);
2104 		if (d)
2105 			d++;
2106 	}
2107 
2108 	if (oldp) {
2109 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2110 		if (l > ol)
2111 			error = ENOMEM;
2112 	} else
2113 		*oldlenp = l;
2114 
2115 	splx(s);
2116 
2117 	return (error);
2118 }
2119 
2120 int
2121 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2122 {
2123 	int error = 0, s;
2124 	struct nd_prefix *pr;
2125 	struct in6_prefix *p = NULL;
2126 	struct in6_prefix *pe = NULL;
2127 	size_t l;
2128 
2129 	s = splsoftnet();
2130 
2131 	if (oldp) {
2132 		p = (struct in6_prefix *)oldp;
2133 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2134 	}
2135 	l = 0;
2136 
2137 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2138 		u_short advrtrs;
2139 		size_t advance;
2140 		struct sockaddr_in6 *sin6;
2141 		struct sockaddr_in6 *s6;
2142 		struct nd_pfxrouter *pfr;
2143 
2144 		if (oldp && p + 1 <= pe)
2145 		{
2146 			bzero(p, sizeof(*p));
2147 			sin6 = (struct sockaddr_in6 *)(p + 1);
2148 
2149 			p->prefix = pr->ndpr_prefix;
2150 			if (in6_recoverscope(&p->prefix,
2151 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2152 				log(LOG_ERR,
2153 				    "scope error in prefix list (%s)\n",
2154 				    ip6_sprintf(&p->prefix.sin6_addr));
2155 			p->raflags = pr->ndpr_raf;
2156 			p->prefixlen = pr->ndpr_plen;
2157 			p->vltime = pr->ndpr_vltime;
2158 			p->pltime = pr->ndpr_pltime;
2159 			p->if_index = pr->ndpr_ifp->if_index;
2160 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2161 				p->expire = 0;
2162 			else {
2163 				time_t maxexpire;
2164 
2165 				/* XXX: we assume time_t is signed. */
2166 				maxexpire = (-1) &
2167 					~(1 << ((sizeof(maxexpire) * 8) - 1));
2168 				if (pr->ndpr_vltime <
2169 				    maxexpire - pr->ndpr_lastupdate) {
2170 					p->expire = pr->ndpr_lastupdate +
2171 						pr->ndpr_vltime;
2172 				} else
2173 					p->expire = maxexpire;
2174 			}
2175 			p->refcnt = pr->ndpr_refcnt;
2176 			p->flags = pr->ndpr_stateflags;
2177 			p->origin = PR_ORIG_RA;
2178 			advrtrs = 0;
2179 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2180 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2181 					advrtrs++;
2182 					continue;
2183 				}
2184 				s6 = &sin6[advrtrs];
2185 				s6->sin6_family = AF_INET6;
2186 				s6->sin6_len = sizeof(struct sockaddr_in6);
2187 				s6->sin6_addr = pfr->router->rtaddr;
2188 				in6_recoverscope(s6, &pfr->router->rtaddr,
2189 				    pfr->router->ifp);
2190 				advrtrs++;
2191 			}
2192 			p->advrtrs = advrtrs;
2193 		}
2194 		else {
2195 			advrtrs = 0;
2196 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2197 				advrtrs++;
2198 		}
2199 
2200 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2201 		l += advance;
2202 		if (p)
2203 			p = (struct in6_prefix *)((caddr_t)p + advance);
2204 	}
2205 
2206 	if (oldp) {
2207 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2208 		if (l > ol)
2209 			error = ENOMEM;
2210 	} else
2211 		*oldlenp = l;
2212 
2213 	splx(s);
2214 
2215 	return (error);
2216 }
2217