xref: /netbsd-src/sys/netinet6/nd6.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: nd6.c,v 1.103 2006/06/07 22:34:04 kardel Exp $	*/
2 /*	$KAME: nd6.c,v 1.279 2002/06/08 11:16:51 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/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.103 2006/06/07 22:34:04 kardel Exp $");
35 
36 #include "opt_ipsec.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/ioctl.h>
50 #include <sys/syslog.h>
51 #include <sys/queue.h>
52 
53 #include <net/if.h>
54 #include <net/if_dl.h>
55 #include <net/if_types.h>
56 #include <net/route.h>
57 #include <net/if_ether.h>
58 #include <net/if_fddi.h>
59 #include <net/if_arc.h>
60 
61 #include <netinet/in.h>
62 #include <netinet6/in6_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet6/scope6_var.h>
66 #include <netinet6/nd6.h>
67 #include <netinet/icmp6.h>
68 
69 #ifdef IPSEC
70 #include <netinet6/ipsec.h>
71 #endif
72 
73 #include <net/net_osdep.h>
74 
75 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
76 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
77 
78 #define SIN6(s) ((struct sockaddr_in6 *)s)
79 #define SDL(s) ((struct sockaddr_dl *)s)
80 
81 /* timer values */
82 int	nd6_prune	= 1;	/* walk list every 1 seconds */
83 int	nd6_delay	= 5;	/* delay first probe time 5 second */
84 int	nd6_umaxtries	= 3;	/* maximum unicast query */
85 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
86 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
87 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
88 
89 /* preventing too many loops in ND option parsing */
90 int nd6_maxndopt = 10;	/* max # of ND options allowed */
91 
92 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
93 
94 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
95 
96 #ifdef ND6_DEBUG
97 int nd6_debug = 1;
98 #else
99 int nd6_debug = 0;
100 #endif
101 
102 /* for debugging? */
103 static int nd6_inuse, nd6_allocated;
104 
105 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
106 struct nd_drhead nd_defrouter;
107 struct nd_prhead nd_prefix = { 0 };
108 
109 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
110 static struct sockaddr_in6 all1_sa;
111 
112 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
113 static void nd6_slowtimo __P((void *));
114 static int regen_tmpaddr __P((struct in6_ifaddr *));
115 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
116 static void nd6_llinfo_timer __P((void *));
117 static void clear_llinfo_pqueue __P((struct llinfo_nd6 *));
118 
119 struct callout nd6_slowtimo_ch = CALLOUT_INITIALIZER;
120 struct callout nd6_timer_ch = CALLOUT_INITIALIZER;
121 extern struct callout in6_tmpaddrtimer_ch;
122 
123 static int fill_drlist __P((void *, size_t *, size_t));
124 static int fill_prlist __P((void *, size_t *, size_t));
125 
126 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
127 
128 void
129 nd6_init()
130 {
131 	static int nd6_init_done = 0;
132 	int i;
133 
134 	if (nd6_init_done) {
135 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
136 		return;
137 	}
138 
139 	all1_sa.sin6_family = AF_INET6;
140 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
141 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
142 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
143 
144 	/* initialization of the default router list */
145 	TAILQ_INIT(&nd_defrouter);
146 
147 	nd6_init_done = 1;
148 
149 	/* start timer */
150 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
151 	    nd6_slowtimo, NULL);
152 }
153 
154 struct nd_ifinfo *
155 nd6_ifattach(ifp)
156 	struct ifnet *ifp;
157 {
158 	struct nd_ifinfo *nd;
159 
160 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
161 	bzero(nd, sizeof(*nd));
162 
163 	nd->initialized = 1;
164 
165 	nd->chlim = IPV6_DEFHLIM;
166 	nd->basereachable = REACHABLE_TIME;
167 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
168 	nd->retrans = RETRANS_TIMER;
169 	/*
170 	 * Note that the default value of ip6_accept_rtadv is 0, which means
171 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
172 	 * here.
173 	 */
174 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
175 
176 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
177 	nd6_setmtu0(ifp, nd);
178 
179 	return nd;
180 }
181 
182 void
183 nd6_ifdetach(nd)
184 	struct nd_ifinfo *nd;
185 {
186 
187 	free(nd, M_IP6NDP);
188 }
189 
190 void
191 nd6_setmtu(ifp)
192 	struct ifnet *ifp;
193 {
194 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
195 }
196 
197 void
198 nd6_setmtu0(ifp, ndi)
199 	struct ifnet *ifp;
200 	struct nd_ifinfo *ndi;
201 {
202 	u_int32_t omaxmtu;
203 
204 	omaxmtu = ndi->maxmtu;
205 
206 	switch (ifp->if_type) {
207 	case IFT_ARCNET:
208 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
209 		break;
210 	case IFT_FDDI:
211 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
212 		break;
213 	default:
214 		ndi->maxmtu = ifp->if_mtu;
215 		break;
216 	}
217 
218 	/*
219 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
220 	 * undesirable situation.  We thus notify the operator of the change
221 	 * explicitly.  The check for omaxmtu is necessary to restrict the
222 	 * log to the case of changing the MTU, not initializing it.
223 	 */
224 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
225 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
226 		    " small for IPv6 which needs %lu\n",
227 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
228 		    IPV6_MMTU);
229 	}
230 
231 	if (ndi->maxmtu > in6_maxmtu)
232 		in6_setmaxmtu(); /* check all interfaces just in case */
233 }
234 
235 void
236 nd6_option_init(opt, icmp6len, ndopts)
237 	void *opt;
238 	int icmp6len;
239 	union nd_opts *ndopts;
240 {
241 
242 	bzero(ndopts, sizeof(*ndopts));
243 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
244 	ndopts->nd_opts_last
245 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
246 
247 	if (icmp6len == 0) {
248 		ndopts->nd_opts_done = 1;
249 		ndopts->nd_opts_search = NULL;
250 	}
251 }
252 
253 /*
254  * Take one ND option.
255  */
256 struct nd_opt_hdr *
257 nd6_option(ndopts)
258 	union nd_opts *ndopts;
259 {
260 	struct nd_opt_hdr *nd_opt;
261 	int olen;
262 
263 	if (ndopts == NULL)
264 		panic("ndopts == NULL in nd6_option");
265 	if (ndopts->nd_opts_last == NULL)
266 		panic("uninitialized ndopts in nd6_option");
267 	if (ndopts->nd_opts_search == NULL)
268 		return NULL;
269 	if (ndopts->nd_opts_done)
270 		return NULL;
271 
272 	nd_opt = ndopts->nd_opts_search;
273 
274 	/* make sure nd_opt_len is inside the buffer */
275 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
276 		bzero(ndopts, sizeof(*ndopts));
277 		return NULL;
278 	}
279 
280 	olen = nd_opt->nd_opt_len << 3;
281 	if (olen == 0) {
282 		/*
283 		 * Message validation requires that all included
284 		 * options have a length that is greater than zero.
285 		 */
286 		bzero(ndopts, sizeof(*ndopts));
287 		return NULL;
288 	}
289 
290 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
291 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
292 		/* option overruns the end of buffer, invalid */
293 		bzero(ndopts, sizeof(*ndopts));
294 		return NULL;
295 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
296 		/* reached the end of options chain */
297 		ndopts->nd_opts_done = 1;
298 		ndopts->nd_opts_search = NULL;
299 	}
300 	return nd_opt;
301 }
302 
303 /*
304  * Parse multiple ND options.
305  * This function is much easier to use, for ND routines that do not need
306  * multiple options of the same type.
307  */
308 int
309 nd6_options(ndopts)
310 	union nd_opts *ndopts;
311 {
312 	struct nd_opt_hdr *nd_opt;
313 	int i = 0;
314 
315 	if (ndopts == NULL)
316 		panic("ndopts == NULL in nd6_options");
317 	if (ndopts->nd_opts_last == NULL)
318 		panic("uninitialized ndopts in nd6_options");
319 	if (ndopts->nd_opts_search == NULL)
320 		return 0;
321 
322 	while (1) {
323 		nd_opt = nd6_option(ndopts);
324 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
325 			/*
326 			 * Message validation requires that all included
327 			 * options have a length that is greater than zero.
328 			 */
329 			icmp6stat.icp6s_nd_badopt++;
330 			bzero(ndopts, sizeof(*ndopts));
331 			return -1;
332 		}
333 
334 		if (nd_opt == NULL)
335 			goto skip1;
336 
337 		switch (nd_opt->nd_opt_type) {
338 		case ND_OPT_SOURCE_LINKADDR:
339 		case ND_OPT_TARGET_LINKADDR:
340 		case ND_OPT_MTU:
341 		case ND_OPT_REDIRECTED_HEADER:
342 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
343 				nd6log((LOG_INFO,
344 				    "duplicated ND6 option found (type=%d)\n",
345 				    nd_opt->nd_opt_type));
346 				/* XXX bark? */
347 			} else {
348 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
349 					= nd_opt;
350 			}
351 			break;
352 		case ND_OPT_PREFIX_INFORMATION:
353 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
354 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
355 					= nd_opt;
356 			}
357 			ndopts->nd_opts_pi_end =
358 				(struct nd_opt_prefix_info *)nd_opt;
359 			break;
360 		default:
361 			/*
362 			 * Unknown options must be silently ignored,
363 			 * to accomodate future extension to the protocol.
364 			 */
365 			nd6log((LOG_DEBUG,
366 			    "nd6_options: unsupported option %d - "
367 			    "option ignored\n", nd_opt->nd_opt_type));
368 		}
369 
370 skip1:
371 		i++;
372 		if (i > nd6_maxndopt) {
373 			icmp6stat.icp6s_nd_toomanyopt++;
374 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
375 			break;
376 		}
377 
378 		if (ndopts->nd_opts_done)
379 			break;
380 	}
381 
382 	return 0;
383 }
384 
385 /*
386  * ND6 timer routine to handle ND6 entries
387  */
388 void
389 nd6_llinfo_settimer(ln, xtick)
390 	struct llinfo_nd6 *ln;
391 	long xtick;
392 {
393 	int s;
394 
395 	s = splsoftnet();
396 
397 	if (xtick < 0) {
398 		ln->ln_expire = 0;
399 		ln->ln_ntick = 0;
400 		callout_stop(&ln->ln_timer_ch);
401 	} else {
402 		ln->ln_expire = time_second + xtick / hz;
403 		if (xtick > INT_MAX) {
404 			ln->ln_ntick = xtick - INT_MAX;
405 			callout_reset(&ln->ln_timer_ch, INT_MAX,
406 			    nd6_llinfo_timer, ln);
407 		} else {
408 			ln->ln_ntick = 0;
409 			callout_reset(&ln->ln_timer_ch, xtick,
410 			    nd6_llinfo_timer, ln);
411 		}
412 	}
413 
414 	splx(s);
415 }
416 
417 static void
418 nd6_llinfo_timer(arg)
419 	void *arg;
420 {
421 	int s;
422 	struct llinfo_nd6 *ln;
423 	struct rtentry *rt;
424 	const struct sockaddr_in6 *dst;
425 	struct ifnet *ifp;
426 	struct nd_ifinfo *ndi = NULL;
427 
428 	s = splsoftnet();
429 
430 	ln = (struct llinfo_nd6 *)arg;
431 
432 	if (ln->ln_ntick > 0) {
433 		nd6_llinfo_settimer(ln, ln->ln_ntick);
434 		splx(s);
435 		return;
436 	}
437 
438 	if ((rt = ln->ln_rt) == NULL)
439 		panic("ln->ln_rt == NULL");
440 	if ((ifp = rt->rt_ifp) == NULL)
441 		panic("ln->ln_rt->rt_ifp == NULL");
442 	ndi = ND_IFINFO(ifp);
443 	dst = (struct sockaddr_in6 *)rt_key(rt);
444 
445 	/* sanity check */
446 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
447 		panic("rt_llinfo(%p) is not equal to ln(%p)",
448 		      rt->rt_llinfo, ln);
449 	if (!dst)
450 		panic("dst=0 in nd6_timer(ln=%p)", ln);
451 
452 	switch (ln->ln_state) {
453 	case ND6_LLINFO_INCOMPLETE:
454 		if (ln->ln_asked < nd6_mmaxtries) {
455 			ln->ln_asked++;
456 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
457 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
458 		} else {
459 			struct mbuf *m = ln->ln_hold;
460 			if (m) {
461 				struct mbuf *m0;
462 
463 				/*
464 				 * assuming every packet in ln_hold has
465 				 * the same IP header
466 				 */
467 				m0 = m->m_nextpkt;
468 				m->m_nextpkt = NULL;
469 				icmp6_error2(m, ICMP6_DST_UNREACH,
470 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
471 
472 				ln->ln_hold = m0;
473 				clear_llinfo_pqueue(ln);
474  			}
475 			(void)nd6_free(rt, 0);
476 			ln = NULL;
477 		}
478 		break;
479 	case ND6_LLINFO_REACHABLE:
480 		if (!ND6_LLINFO_PERMANENT(ln)) {
481 			ln->ln_state = ND6_LLINFO_STALE;
482 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
483 		}
484 		break;
485 
486 	case ND6_LLINFO_STALE:
487 		/* Garbage Collection(RFC 2461 5.3) */
488 		if (!ND6_LLINFO_PERMANENT(ln)) {
489 			(void)nd6_free(rt, 1);
490 			ln = NULL;
491 		}
492 		break;
493 
494 	case ND6_LLINFO_DELAY:
495 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
496 			/* We need NUD */
497 			ln->ln_asked = 1;
498 			ln->ln_state = ND6_LLINFO_PROBE;
499 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
500 			nd6_ns_output(ifp, &dst->sin6_addr,
501 			    &dst->sin6_addr, ln, 0);
502 		} else {
503 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
504 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
505 		}
506 		break;
507 	case ND6_LLINFO_PROBE:
508 		if (ln->ln_asked < nd6_umaxtries) {
509 			ln->ln_asked++;
510 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
511 			nd6_ns_output(ifp, &dst->sin6_addr,
512 			    &dst->sin6_addr, ln, 0);
513 		} else {
514 			(void)nd6_free(rt, 0);
515 			ln = NULL;
516 		}
517 		break;
518 	}
519 
520 	splx(s);
521 }
522 
523 /*
524  * ND6 timer routine to expire default route list and prefix list
525  */
526 void
527 nd6_timer(ignored_arg)
528 	void	*ignored_arg;
529 {
530 	int s;
531 	struct nd_defrouter *dr;
532 	struct nd_prefix *pr;
533 	struct in6_ifaddr *ia6, *nia6;
534 	struct in6_addrlifetime *lt6;
535 
536 	s = splsoftnet();
537 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
538 	    nd6_timer, NULL);
539 
540 	/* expire default router list */
541 	dr = TAILQ_FIRST(&nd_defrouter);
542 	while (dr) {
543 		if (dr->expire && dr->expire < time_second) {
544 			struct nd_defrouter *t;
545 			t = TAILQ_NEXT(dr, dr_entry);
546 			defrtrlist_del(dr);
547 			dr = t;
548 		} else {
549 			dr = TAILQ_NEXT(dr, dr_entry);
550 		}
551 	}
552 
553 	/*
554 	 * expire interface addresses.
555 	 * in the past the loop was inside prefix expiry processing.
556 	 * However, from a stricter speci-confrmance standpoint, we should
557 	 * rather separate address lifetimes and prefix lifetimes.
558 	 */
559   addrloop:
560 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
561 		nia6 = ia6->ia_next;
562 		/* check address lifetime */
563 		lt6 = &ia6->ia6_lifetime;
564 		if (IFA6_IS_INVALID(ia6)) {
565 			int regen = 0;
566 
567 			/*
568 			 * If the expiring address is temporary, try
569 			 * regenerating a new one.  This would be useful when
570 			 * we suspended a laptop PC, then turned it on after a
571 			 * period that could invalidate all temporary
572 			 * addresses.  Although we may have to restart the
573 			 * loop (see below), it must be after purging the
574 			 * address.  Otherwise, we'd see an infinite loop of
575 			 * regeneration.
576 			 */
577 			if (ip6_use_tempaddr &&
578 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
579 				if (regen_tmpaddr(ia6) == 0)
580 					regen = 1;
581 			}
582 
583  			in6_purgeaddr(&ia6->ia_ifa);
584 
585 			if (regen)
586 				goto addrloop; /* XXX: see below */
587 		} else if (IFA6_IS_DEPRECATED(ia6)) {
588 			int oldflags = ia6->ia6_flags;
589 
590  			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
591 
592 			/*
593 			 * If a temporary address has just become deprecated,
594 			 * regenerate a new one if possible.
595 			 */
596 			if (ip6_use_tempaddr &&
597 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
598 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
599 
600 				if (regen_tmpaddr(ia6) == 0) {
601 					/*
602 					 * A new temporary address is
603 					 * generated.
604 					 * XXX: this means the address chain
605 					 * has changed while we are still in
606 					 * the loop.  Although the change
607 					 * would not cause disaster (because
608 					 * it's not a deletion, but an
609 					 * addition,) we'd rather restart the
610 					 * loop just for safety.  Or does this
611 					 * significantly reduce performance??
612 					 */
613 					goto addrloop;
614 				}
615 			}
616 		} else {
617 			/*
618 			 * A new RA might have made a deprecated address
619 			 * preferred.
620 			 */
621 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
622 		}
623 	}
624 
625 	/* expire prefix list */
626 	pr = nd_prefix.lh_first;
627 	while (pr) {
628 		/*
629 		 * check prefix lifetime.
630 		 * since pltime is just for autoconf, pltime processing for
631 		 * prefix is not necessary.
632 		 */
633 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
634 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
635 			struct nd_prefix *t;
636 			t = pr->ndpr_next;
637 
638 			/*
639 			 * address expiration and prefix expiration are
640 			 * separate.  NEVER perform in6_purgeaddr here.
641 			 */
642 
643 			prelist_remove(pr);
644 			pr = t;
645 		} else
646 			pr = pr->ndpr_next;
647 	}
648 	splx(s);
649 }
650 
651 static int
652 regen_tmpaddr(ia6)
653 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
654 {
655 	struct ifaddr *ifa;
656 	struct ifnet *ifp;
657 	struct in6_ifaddr *public_ifa6 = NULL;
658 
659 	ifp = ia6->ia_ifa.ifa_ifp;
660 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
661 	     ifa = ifa->ifa_list.tqe_next) {
662 		struct in6_ifaddr *it6;
663 
664 		if (ifa->ifa_addr->sa_family != AF_INET6)
665 			continue;
666 
667 		it6 = (struct in6_ifaddr *)ifa;
668 
669 		/* ignore no autoconf addresses. */
670 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
671 			continue;
672 
673 		/* ignore autoconf addresses with different prefixes. */
674 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
675 			continue;
676 
677 		/*
678 		 * Now we are looking at an autoconf address with the same
679 		 * prefix as ours.  If the address is temporary and is still
680 		 * preferred, do not create another one.  It would be rare, but
681 		 * could happen, for example, when we resume a laptop PC after
682 		 * a long period.
683 		 */
684 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
685 		    !IFA6_IS_DEPRECATED(it6)) {
686 			public_ifa6 = NULL;
687 			break;
688 		}
689 
690 		/*
691 		 * This is a public autoconf address that has the same prefix
692 		 * as ours.  If it is preferred, keep it.  We can't break the
693 		 * loop here, because there may be a still-preferred temporary
694 		 * address with the prefix.
695 		 */
696 		if (!IFA6_IS_DEPRECATED(it6))
697 		    public_ifa6 = it6;
698 	}
699 
700 	if (public_ifa6 != NULL) {
701 		int e;
702 
703 		/*
704 		 * Random factor is introduced in the preferred lifetime, so
705 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
706 		 */
707 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
708 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
709 			    " tmp addr, errno=%d\n", e);
710 			return (-1);
711 		}
712 		return (0);
713 	}
714 
715 	return (-1);
716 }
717 
718 /*
719  * Nuke neighbor cache/prefix/default router management table, right before
720  * ifp goes away.
721  */
722 void
723 nd6_purge(ifp)
724 	struct ifnet *ifp;
725 {
726 	struct llinfo_nd6 *ln, *nln;
727 	struct nd_defrouter *dr, *ndr;
728 	struct nd_prefix *pr, *npr;
729 
730 	/*
731 	 * Nuke default router list entries toward ifp.
732 	 * We defer removal of default router list entries that is installed
733 	 * in the routing table, in order to keep additional side effects as
734 	 * small as possible.
735 	 */
736 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
737 		ndr = TAILQ_NEXT(dr, dr_entry);
738 		if (dr->installed)
739 			continue;
740 
741 		if (dr->ifp == ifp)
742 			defrtrlist_del(dr);
743 	}
744 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
745 		ndr = TAILQ_NEXT(dr, dr_entry);
746 		if (!dr->installed)
747 			continue;
748 
749 		if (dr->ifp == ifp)
750 			defrtrlist_del(dr);
751 	}
752 
753 	/* Nuke prefix list entries toward ifp */
754 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
755 		npr = pr->ndpr_next;
756 		if (pr->ndpr_ifp == ifp) {
757 			/*
758 			 * Because if_detach() does *not* release prefixes
759 			 * while purging addresses the reference count will
760 			 * still be above zero. We therefore reset it to
761 			 * make sure that the prefix really gets purged.
762 			 */
763 			pr->ndpr_refcnt = 0;
764 			/*
765 			 * Previously, pr->ndpr_addr is removed as well,
766 			 * but I strongly believe we don't have to do it.
767 			 * nd6_purge() is only called from in6_ifdetach(),
768 			 * which removes all the associated interface addresses
769 			 * by itself.
770 			 * (jinmei@kame.net 20010129)
771 			 */
772 			prelist_remove(pr);
773 		}
774 	}
775 
776 	/* cancel default outgoing interface setting */
777 	if (nd6_defifindex == ifp->if_index)
778 		nd6_setdefaultiface(0);
779 
780 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
781 		/* refresh default router list */
782 		defrouter_select();
783 	}
784 
785 	/*
786 	 * Nuke neighbor cache entries for the ifp.
787 	 * Note that rt->rt_ifp may not be the same as ifp,
788 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
789 	 * nd6_rtrequest(), and ip6_input().
790 	 */
791 	ln = llinfo_nd6.ln_next;
792 	while (ln && ln != &llinfo_nd6) {
793 		struct rtentry *rt;
794 		struct sockaddr_dl *sdl;
795 
796 		nln = ln->ln_next;
797 		rt = ln->ln_rt;
798 		if (rt && rt->rt_gateway &&
799 		    rt->rt_gateway->sa_family == AF_LINK) {
800 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
801 			if (sdl->sdl_index == ifp->if_index)
802 				nln = nd6_free(rt, 0);
803 		}
804 		ln = nln;
805 	}
806 }
807 
808 struct rtentry *
809 nd6_lookup(addr6, create, ifp)
810 	struct in6_addr *addr6;
811 	int create;
812 	struct ifnet *ifp;
813 {
814 	struct rtentry *rt;
815 	struct sockaddr_in6 sin6;
816 
817 	bzero(&sin6, sizeof(sin6));
818 	sin6.sin6_len = sizeof(struct sockaddr_in6);
819 	sin6.sin6_family = AF_INET6;
820 	sin6.sin6_addr = *addr6;
821 	rt = rtalloc1((struct sockaddr *)&sin6, create);
822 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
823 		/*
824 		 * This is the case for the default route.
825 		 * If we want to create a neighbor cache for the address, we
826 		 * should free the route for the destination and allocate an
827 		 * interface route.
828 		 */
829 		if (create) {
830 			RTFREE(rt);
831 			rt = NULL;
832 		}
833 	}
834 	if (rt == NULL) {
835 		if (create && ifp) {
836 			int e;
837 
838 			/*
839 			 * If no route is available and create is set,
840 			 * we allocate a host route for the destination
841 			 * and treat it like an interface route.
842 			 * This hack is necessary for a neighbor which can't
843 			 * be covered by our own prefix.
844 			 */
845 			struct ifaddr *ifa =
846 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
847 			if (ifa == NULL)
848 				return (NULL);
849 
850 			/*
851 			 * Create a new route.  RTF_LLINFO is necessary
852 			 * to create a Neighbor Cache entry for the
853 			 * destination in nd6_rtrequest which will be
854 			 * called in rtrequest via ifa->ifa_rtrequest.
855 			 */
856 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
857 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
858 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
859 			    ~RTF_CLONING, &rt)) != 0) {
860 #if 0
861 				log(LOG_ERR,
862 				    "nd6_lookup: failed to add route for a "
863 				    "neighbor(%s), errno=%d\n",
864 				    ip6_sprintf(addr6), e);
865 #endif
866 				return (NULL);
867 			}
868 			if (rt == NULL)
869 				return (NULL);
870 			if (rt->rt_llinfo) {
871 				struct llinfo_nd6 *ln =
872 				    (struct llinfo_nd6 *)rt->rt_llinfo;
873 				ln->ln_state = ND6_LLINFO_NOSTATE;
874 			}
875 		} else
876 			return (NULL);
877 	}
878 	rt->rt_refcnt--;
879 	/*
880 	 * Validation for the entry.
881 	 * Note that the check for rt_llinfo is necessary because a cloned
882 	 * route from a parent route that has the L flag (e.g. the default
883 	 * route to a p2p interface) may have the flag, too, while the
884 	 * destination is not actually a neighbor.
885 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
886 	 *      it might be the loopback interface if the entry is for our
887 	 *      own address on a non-loopback interface. Instead, we should
888 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
889 	 *	interface.
890 	 * Note also that ifa_ifp and ifp may differ when we connect two
891 	 * interfaces to a same link, install a link prefix to an interface,
892 	 * and try to install a neighbor cache on an interface that does not
893 	 * have a route to the prefix.
894 	 */
895 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
896 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
897 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
898 		if (create) {
899 			nd6log((LOG_DEBUG,
900 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
901 			    ip6_sprintf(addr6),
902 			    ifp ? if_name(ifp) : "unspec"));
903 		}
904 		return (NULL);
905 	}
906 	return (rt);
907 }
908 
909 /*
910  * Detect if a given IPv6 address identifies a neighbor on a given link.
911  * XXX: should take care of the destination of a p2p link?
912  */
913 int
914 nd6_is_addr_neighbor(addr, ifp)
915 	struct sockaddr_in6 *addr;
916 	struct ifnet *ifp;
917 {
918 	struct nd_prefix *pr;
919 
920 	/*
921 	 * A link-local address is always a neighbor.
922 	 * XXX: a link does not necessarily specify a single interface.
923 	 */
924 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
925 		struct sockaddr_in6 sin6_copy;
926 		u_int32_t zone;
927 
928 		/*
929 		 * We need sin6_copy since sa6_recoverscope() may modify the
930 		 * content (XXX).
931 		 */
932 		sin6_copy = *addr;
933 		if (sa6_recoverscope(&sin6_copy))
934 			return (0); /* XXX: should be impossible */
935 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
936 			return (0);
937 		if (sin6_copy.sin6_scope_id == zone)
938 			return (1);
939 		else
940 			return (0);
941 	}
942 
943 	/*
944 	 * If the address matches one of our on-link prefixes, it should be a
945 	 * neighbor.
946 	 */
947 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
948 		if (pr->ndpr_ifp != ifp)
949 			continue;
950 
951 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
952 			continue;
953 
954 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
955 		    &addr->sin6_addr, &pr->ndpr_mask))
956 			return (1);
957 	}
958 
959 	/*
960 	 * If the default router list is empty, all addresses are regarded
961 	 * as on-link, and thus, as a neighbor.
962 	 * XXX: we restrict the condition to hosts, because routers usually do
963 	 * not have the "default router list".
964 	 */
965 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
966 	    nd6_defifindex == ifp->if_index) {
967 		return (1);
968 	}
969 
970 	/*
971 	 * Even if the address matches none of our addresses, it might be
972 	 * in the neighbor cache.
973 	 */
974 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
975 		return (1);
976 
977 	return (0);
978 }
979 
980 /*
981  * Free an nd6 llinfo entry.
982  * Since the function would cause significant changes in the kernel, DO NOT
983  * make it global, unless you have a strong reason for the change, and are sure
984  * that the change is safe.
985  */
986 static struct llinfo_nd6 *
987 nd6_free(rt, gc)
988 	struct rtentry *rt;
989 	int gc;
990 {
991 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
992 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
993 	struct nd_defrouter *dr;
994 
995 	/*
996 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
997 	 * even though it is not harmful, it was not really necessary.
998 	 */
999 
1000 	/* cancel timer */
1001 	nd6_llinfo_settimer(ln, -1);
1002 
1003 	if (!ip6_forwarding) {
1004 		int s;
1005 		s = splsoftnet();
1006 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1007 		    rt->rt_ifp);
1008 
1009 		if (dr != NULL && dr->expire &&
1010 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1011 			/*
1012 			 * If the reason for the deletion is just garbage
1013 			 * collection, and the neighbor is an active default
1014 			 * router, do not delete it.  Instead, reset the GC
1015 			 * timer using the router's lifetime.
1016 			 * Simply deleting the entry would affect default
1017 			 * router selection, which is not necessarily a good
1018 			 * thing, especially when we're using router preference
1019 			 * values.
1020 			 * XXX: the check for ln_state would be redundant,
1021 			 *      but we intentionally keep it just in case.
1022 			 */
1023 			if (dr->expire > time_second)
1024 				nd6_llinfo_settimer(ln,
1025 				    (dr->expire - time_second) * hz);
1026 			else
1027 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1028 			splx(s);
1029 			return (ln->ln_next);
1030 		}
1031 
1032 		if (ln->ln_router || dr) {
1033 			/*
1034 			 * rt6_flush must be called whether or not the neighbor
1035 			 * is in the Default Router List.
1036 			 * See a corresponding comment in nd6_na_input().
1037 			 */
1038 			rt6_flush(&in6, rt->rt_ifp);
1039 		}
1040 
1041 		if (dr) {
1042 			/*
1043 			 * Unreachablity of a router might affect the default
1044 			 * router selection and on-link detection of advertised
1045 			 * prefixes.
1046 			 */
1047 
1048 			/*
1049 			 * Temporarily fake the state to choose a new default
1050 			 * router and to perform on-link determination of
1051 			 * prefixes correctly.
1052 			 * Below the state will be set correctly,
1053 			 * or the entry itself will be deleted.
1054 			 */
1055 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1056 
1057 			/*
1058 			 * Since defrouter_select() does not affect the
1059 			 * on-link determination and MIP6 needs the check
1060 			 * before the default router selection, we perform
1061 			 * the check now.
1062 			 */
1063 			pfxlist_onlink_check();
1064 
1065 			/*
1066 			 * refresh default router list
1067 			 */
1068 			defrouter_select();
1069 		}
1070 		splx(s);
1071 	}
1072 
1073 	/*
1074 	 * Before deleting the entry, remember the next entry as the
1075 	 * return value.  We need this because pfxlist_onlink_check() above
1076 	 * might have freed other entries (particularly the old next entry) as
1077 	 * a side effect (XXX).
1078 	 */
1079 	next = ln->ln_next;
1080 
1081 	/*
1082 	 * Detach the route from the routing tree and the list of neighbor
1083 	 * caches, and disable the route entry not to be used in already
1084 	 * cached routes.
1085 	 */
1086 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1087 	    rt_mask(rt), 0, (struct rtentry **)0);
1088 
1089 	return (next);
1090 }
1091 
1092 /*
1093  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1094  *
1095  * XXX cost-effective methods?
1096  */
1097 void
1098 nd6_nud_hint(rt, dst6, force)
1099 	struct rtentry *rt;
1100 	struct in6_addr *dst6;
1101 	int force;
1102 {
1103 	struct llinfo_nd6 *ln;
1104 
1105 	/*
1106 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1107 	 * routing table by supplied "dst6".
1108 	 */
1109 	if (rt == NULL) {
1110 		if (dst6 == NULL)
1111 			return;
1112 		if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1113 			return;
1114 	}
1115 
1116 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1117 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1118 	    !rt->rt_llinfo || !rt->rt_gateway ||
1119 	    rt->rt_gateway->sa_family != AF_LINK) {
1120 		/* This is not a host route. */
1121 		return;
1122 	}
1123 
1124 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1125 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1126 		return;
1127 
1128 	/*
1129 	 * if we get upper-layer reachability confirmation many times,
1130 	 * it is possible we have false information.
1131 	 */
1132 	if (!force) {
1133 		ln->ln_byhint++;
1134 		if (ln->ln_byhint > nd6_maxnudhint)
1135 			return;
1136 	}
1137 
1138 	ln->ln_state = ND6_LLINFO_REACHABLE;
1139 	if (!ND6_LLINFO_PERMANENT(ln)) {
1140 		nd6_llinfo_settimer(ln,
1141 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1142 	}
1143 }
1144 
1145 void
1146 nd6_rtrequest(req, rt, info)
1147 	int	req;
1148 	struct rtentry *rt;
1149 	struct rt_addrinfo *info; /* xxx unused */
1150 {
1151 	struct sockaddr *gate = rt->rt_gateway;
1152 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1153 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1154 	struct ifnet *ifp = rt->rt_ifp;
1155 	struct ifaddr *ifa;
1156 
1157 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1158 		return;
1159 
1160 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1161 		/*
1162 		 * This is probably an interface direct route for a link
1163 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1164 		 * We do not need special treatment below for such a route.
1165 		 * Moreover, the RTF_LLINFO flag which would be set below
1166 		 * would annoy the ndp(8) command.
1167 		 */
1168 		return;
1169 	}
1170 
1171 	if (req == RTM_RESOLVE &&
1172 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1173 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1174 		/*
1175 		 * FreeBSD and BSD/OS often make a cloned host route based
1176 		 * on a less-specific route (e.g. the default route).
1177 		 * If the less specific route does not have a "gateway"
1178 		 * (this is the case when the route just goes to a p2p or an
1179 		 * stf interface), we'll mistakenly make a neighbor cache for
1180 		 * the host route, and will see strange neighbor solicitation
1181 		 * for the corresponding destination.  In order to avoid the
1182 		 * confusion, we check if the destination of the route is
1183 		 * a neighbor in terms of neighbor discovery, and stop the
1184 		 * process if not.  Additionally, we remove the LLINFO flag
1185 		 * so that ndp(8) will not try to get the neighbor information
1186 		 * of the destination.
1187 		 */
1188 		rt->rt_flags &= ~RTF_LLINFO;
1189 		return;
1190 	}
1191 
1192 	switch (req) {
1193 	case RTM_ADD:
1194 		/*
1195 		 * There is no backward compatibility :)
1196 		 *
1197 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1198 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1199 		 *	   rt->rt_flags |= RTF_CLONING;
1200 		 */
1201 		if ((rt->rt_flags & RTF_CLONING) ||
1202 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1203 			/*
1204 			 * Case 1: This route should come from a route to
1205 			 * interface (RTF_CLONING case) or the route should be
1206 			 * treated as on-link but is currently not
1207 			 * (RTF_LLINFO && !ln case).
1208 			 */
1209 			rt_setgate(rt, rt_key(rt),
1210 				   (struct sockaddr *)&null_sdl);
1211 			gate = rt->rt_gateway;
1212 			SDL(gate)->sdl_type = ifp->if_type;
1213 			SDL(gate)->sdl_index = ifp->if_index;
1214 			if (ln)
1215 				nd6_llinfo_settimer(ln, 0);
1216 			if ((rt->rt_flags & RTF_CLONING) != 0)
1217 				break;
1218 		}
1219 		/*
1220 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1221 		 * We don't do that here since llinfo is not ready yet.
1222 		 *
1223 		 * There are also couple of other things to be discussed:
1224 		 * - unsolicited NA code needs improvement beforehand
1225 		 * - RFC2461 says we MAY send multicast unsolicited NA
1226 		 *   (7.2.6 paragraph 4), however, it also says that we
1227 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1228 		 *   we don't have anything like it right now.
1229 		 *   note that the mechanism needs a mutual agreement
1230 		 *   between proxies, which means that we need to implement
1231 		 *   a new protocol, or a new kludge.
1232 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1233 		 *   we need to check ip6forwarding before sending it.
1234 		 *   (or should we allow proxy ND configuration only for
1235 		 *   routers?  there's no mention about proxy ND from hosts)
1236 		 */
1237 #if 0
1238 		/* XXX it does not work */
1239 		if (rt->rt_flags & RTF_ANNOUNCE)
1240 			nd6_na_output(ifp,
1241 			      &SIN6(rt_key(rt))->sin6_addr,
1242 			      &SIN6(rt_key(rt))->sin6_addr,
1243 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1244 			      1, NULL);
1245 #endif
1246 		/* FALLTHROUGH */
1247 	case RTM_RESOLVE:
1248 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1249 			/*
1250 			 * Address resolution isn't necessary for a point to
1251 			 * point link, so we can skip this test for a p2p link.
1252 			 */
1253 			if (gate->sa_family != AF_LINK ||
1254 			    gate->sa_len < sizeof(null_sdl)) {
1255 				log(LOG_DEBUG,
1256 				    "nd6_rtrequest: bad gateway value: %s\n",
1257 				    if_name(ifp));
1258 				break;
1259 			}
1260 			SDL(gate)->sdl_type = ifp->if_type;
1261 			SDL(gate)->sdl_index = ifp->if_index;
1262 		}
1263 		if (ln != NULL)
1264 			break;	/* This happens on a route change */
1265 		/*
1266 		 * Case 2: This route may come from cloning, or a manual route
1267 		 * add with a LL address.
1268 		 */
1269 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1270 		rt->rt_llinfo = (caddr_t)ln;
1271 		if (ln == NULL) {
1272 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1273 			break;
1274 		}
1275 		nd6_inuse++;
1276 		nd6_allocated++;
1277 		bzero(ln, sizeof(*ln));
1278 		ln->ln_rt = rt;
1279 		callout_init(&ln->ln_timer_ch);
1280 		/* this is required for "ndp" command. - shin */
1281 		if (req == RTM_ADD) {
1282 		        /*
1283 			 * gate should have some valid AF_LINK entry,
1284 			 * and ln->ln_expire should have some lifetime
1285 			 * which is specified by ndp command.
1286 			 */
1287 			ln->ln_state = ND6_LLINFO_REACHABLE;
1288 			ln->ln_byhint = 0;
1289 		} else {
1290 		        /*
1291 			 * When req == RTM_RESOLVE, rt is created and
1292 			 * initialized in rtrequest(), so rt_expire is 0.
1293 			 */
1294 			ln->ln_state = ND6_LLINFO_NOSTATE;
1295 			nd6_llinfo_settimer(ln, 0);
1296 		}
1297 		rt->rt_flags |= RTF_LLINFO;
1298 		ln->ln_next = llinfo_nd6.ln_next;
1299 		llinfo_nd6.ln_next = ln;
1300 		ln->ln_prev = &llinfo_nd6;
1301 		ln->ln_next->ln_prev = ln;
1302 
1303 		/*
1304 		 * check if rt_key(rt) is one of my address assigned
1305 		 * to the interface.
1306 		 */
1307 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1308 		    &SIN6(rt_key(rt))->sin6_addr);
1309 		if (ifa) {
1310 			caddr_t macp = nd6_ifptomac(ifp);
1311 			nd6_llinfo_settimer(ln, -1);
1312 			ln->ln_state = ND6_LLINFO_REACHABLE;
1313 			ln->ln_byhint = 0;
1314 			if (macp) {
1315 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1316 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1317 			}
1318 			if (nd6_useloopback) {
1319 				rt->rt_ifp = lo0ifp;	/* XXX */
1320 				/*
1321 				 * Make sure rt_ifa be equal to the ifaddr
1322 				 * corresponding to the address.
1323 				 * We need this because when we refer
1324 				 * rt_ifa->ia6_flags in ip6_input, we assume
1325 				 * that the rt_ifa points to the address instead
1326 				 * of the loopback address.
1327 				 */
1328 				if (ifa != rt->rt_ifa) {
1329 					IFAFREE(rt->rt_ifa);
1330 					IFAREF(ifa);
1331 					rt->rt_ifa = ifa;
1332 				}
1333 			}
1334 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1335 			nd6_llinfo_settimer(ln, -1);
1336 			ln->ln_state = ND6_LLINFO_REACHABLE;
1337 			ln->ln_byhint = 0;
1338 
1339 			/* join solicited node multicast for proxy ND */
1340 			if (ifp->if_flags & IFF_MULTICAST) {
1341 				struct in6_addr llsol;
1342 				int error;
1343 
1344 				llsol = SIN6(rt_key(rt))->sin6_addr;
1345 				llsol.s6_addr32[0] = htonl(0xff020000);
1346 				llsol.s6_addr32[1] = 0;
1347 				llsol.s6_addr32[2] = htonl(1);
1348 				llsol.s6_addr8[12] = 0xff;
1349 				if (in6_setscope(&llsol, ifp, NULL))
1350 					break;
1351 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1352 					nd6log((LOG_ERR, "%s: failed to join "
1353 					    "%s (errno=%d)\n", if_name(ifp),
1354 					    ip6_sprintf(&llsol), error));
1355 				}
1356 			}
1357 		}
1358 		break;
1359 
1360 	case RTM_DELETE:
1361 		if (ln == NULL)
1362 			break;
1363 		/* leave from solicited node multicast for proxy ND */
1364 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1365 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1366 			struct in6_addr llsol;
1367 			struct in6_multi *in6m;
1368 
1369 			llsol = SIN6(rt_key(rt))->sin6_addr;
1370 			llsol.s6_addr32[0] = htonl(0xff020000);
1371 			llsol.s6_addr32[1] = 0;
1372 			llsol.s6_addr32[2] = htonl(1);
1373 			llsol.s6_addr8[12] = 0xff;
1374 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1375 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1376 				if (in6m)
1377 					in6_delmulti(in6m);
1378 			} else
1379 				; /* XXX: should not happen. bark here? */
1380 		}
1381 		nd6_inuse--;
1382 		ln->ln_next->ln_prev = ln->ln_prev;
1383 		ln->ln_prev->ln_next = ln->ln_next;
1384 		ln->ln_prev = NULL;
1385 		nd6_llinfo_settimer(ln, -1);
1386 		rt->rt_llinfo = 0;
1387 		rt->rt_flags &= ~RTF_LLINFO;
1388 		clear_llinfo_pqueue(ln);
1389 		Free((caddr_t)ln);
1390 	}
1391 }
1392 
1393 int
1394 nd6_ioctl(cmd, data, ifp)
1395 	u_long cmd;
1396 	caddr_t	data;
1397 	struct ifnet *ifp;
1398 {
1399 	struct in6_drlist *drl = (struct in6_drlist *)data;
1400 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1401 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1402 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1403 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1404 	struct nd_defrouter *dr;
1405 	struct nd_prefix *pr;
1406 	struct rtentry *rt;
1407 	int i = 0, error = 0;
1408 	int s;
1409 
1410 	switch (cmd) {
1411 	case SIOCGDRLST_IN6:
1412 		/*
1413 		 * obsolete API, use sysctl under net.inet6.icmp6
1414 		 */
1415 		bzero(drl, sizeof(*drl));
1416 		s = splsoftnet();
1417 		dr = TAILQ_FIRST(&nd_defrouter);
1418 		while (dr && i < DRLSTSIZ) {
1419 			drl->defrouter[i].rtaddr = dr->rtaddr;
1420 			in6_clearscope(&drl->defrouter[i].rtaddr);
1421 
1422 			drl->defrouter[i].flags = dr->flags;
1423 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1424 			drl->defrouter[i].expire = dr->expire;
1425 			drl->defrouter[i].if_index = dr->ifp->if_index;
1426 			i++;
1427 			dr = TAILQ_NEXT(dr, dr_entry);
1428 		}
1429 		splx(s);
1430 		break;
1431 	case SIOCGPRLST_IN6:
1432 		/*
1433 		 * obsolete API, use sysctl under net.inet6.icmp6
1434 		 *
1435 		 * XXX the structure in6_prlist was changed in backward-
1436 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1437 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1438 		 */
1439 		/*
1440 		 * XXX meaning of fields, especialy "raflags", is very
1441 		 * differnet between RA prefix list and RR/static prefix list.
1442 		 * how about separating ioctls into two?
1443 		 */
1444 		bzero(oprl, sizeof(*oprl));
1445 		s = splsoftnet();
1446 		pr = nd_prefix.lh_first;
1447 		while (pr && i < PRLSTSIZ) {
1448 			struct nd_pfxrouter *pfr;
1449 			int j;
1450 
1451 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1452 			oprl->prefix[i].raflags = pr->ndpr_raf;
1453 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1454 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1455 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1456 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1457 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1458 				oprl->prefix[i].expire = 0;
1459 			else {
1460 				time_t maxexpire;
1461 
1462 				/* XXX: we assume time_t is signed. */
1463 				maxexpire = (-1) &
1464 				    ~((time_t)1 <<
1465 				    ((sizeof(maxexpire) * 8) - 1));
1466 				if (pr->ndpr_vltime <
1467 				    maxexpire - pr->ndpr_lastupdate) {
1468 					oprl->prefix[i].expire =
1469 						 pr->ndpr_lastupdate +
1470 						pr->ndpr_vltime;
1471 				} else
1472 					oprl->prefix[i].expire = maxexpire;
1473 			}
1474 
1475 			pfr = pr->ndpr_advrtrs.lh_first;
1476 			j = 0;
1477 			while (pfr) {
1478 				if (j < DRLSTSIZ) {
1479 #define RTRADDR oprl->prefix[i].advrtr[j]
1480 					RTRADDR = pfr->router->rtaddr;
1481 					in6_clearscope(&RTRADDR);
1482 #undef RTRADDR
1483 				}
1484 				j++;
1485 				pfr = pfr->pfr_next;
1486 			}
1487 			oprl->prefix[i].advrtrs = j;
1488 			oprl->prefix[i].origin = PR_ORIG_RA;
1489 
1490 			i++;
1491 			pr = pr->ndpr_next;
1492 		}
1493 		splx(s);
1494 
1495 		break;
1496 	case OSIOCGIFINFO_IN6:
1497 #define ND	ndi->ndi
1498 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1499 		memset(&ND, 0, sizeof(ND));
1500 		ND.linkmtu = IN6_LINKMTU(ifp);
1501 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1502 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1503 		ND.reachable = ND_IFINFO(ifp)->reachable;
1504 		ND.retrans = ND_IFINFO(ifp)->retrans;
1505 		ND.flags = ND_IFINFO(ifp)->flags;
1506 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1507 		ND.chlim = ND_IFINFO(ifp)->chlim;
1508 		break;
1509 	case SIOCGIFINFO_IN6:
1510 		ND = *ND_IFINFO(ifp);
1511 		break;
1512 	case SIOCSIFINFO_IN6:
1513 		/*
1514 		 * used to change host variables from userland.
1515 		 * intented for a use on router to reflect RA configurations.
1516 		 */
1517 		/* 0 means 'unspecified' */
1518 		if (ND.linkmtu != 0) {
1519 			if (ND.linkmtu < IPV6_MMTU ||
1520 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1521 				error = EINVAL;
1522 				break;
1523 			}
1524 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1525 		}
1526 
1527 		if (ND.basereachable != 0) {
1528 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1529 
1530 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1531 			if (ND.basereachable != obasereachable)
1532 				ND_IFINFO(ifp)->reachable =
1533 				    ND_COMPUTE_RTIME(ND.basereachable);
1534 		}
1535 		if (ND.retrans != 0)
1536 			ND_IFINFO(ifp)->retrans = ND.retrans;
1537 		if (ND.chlim != 0)
1538 			ND_IFINFO(ifp)->chlim = ND.chlim;
1539 		/* FALLTHROUGH */
1540 	case SIOCSIFINFO_FLAGS:
1541 		ND_IFINFO(ifp)->flags = ND.flags;
1542 		break;
1543 #undef ND
1544 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1545 		/* sync kernel routing table with the default router list */
1546 		defrouter_reset();
1547 		defrouter_select();
1548 		break;
1549 	case SIOCSPFXFLUSH_IN6:
1550 	{
1551 		/* flush all the prefix advertised by routers */
1552 		struct nd_prefix *pfx, *next;
1553 
1554 		s = splsoftnet();
1555 		for (pfx = nd_prefix.lh_first; pfx; pfx = next) {
1556 			struct in6_ifaddr *ia, *ia_next;
1557 
1558 			next = pfx->ndpr_next;
1559 
1560 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1561 				continue; /* XXX */
1562 
1563 			/* do we really have to remove addresses as well? */
1564 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1565 				/* ia might be removed.  keep the next ptr. */
1566 				ia_next = ia->ia_next;
1567 
1568 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1569 					continue;
1570 
1571 				if (ia->ia6_ndpr == pfx)
1572 					in6_purgeaddr(&ia->ia_ifa);
1573 			}
1574 			prelist_remove(pfx);
1575 		}
1576 		splx(s);
1577 		break;
1578 	}
1579 	case SIOCSRTRFLUSH_IN6:
1580 	{
1581 		/* flush all the default routers */
1582 		struct nd_defrouter *drtr, *next;
1583 
1584 		s = splsoftnet();
1585 		defrouter_reset();
1586 		for (drtr = TAILQ_FIRST(&nd_defrouter); drtr; drtr = next) {
1587 			next = TAILQ_NEXT(drtr, dr_entry);
1588 			defrtrlist_del(drtr);
1589 		}
1590 		defrouter_select();
1591 		splx(s);
1592 		break;
1593 	}
1594 	case SIOCGNBRINFO_IN6:
1595 	{
1596 		struct llinfo_nd6 *ln;
1597 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1598 
1599 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1600 			return (error);
1601 
1602 		s = splsoftnet();
1603 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1604 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1605 			error = EINVAL;
1606 			splx(s);
1607 			break;
1608 		}
1609 		nbi->state = ln->ln_state;
1610 		nbi->asked = ln->ln_asked;
1611 		nbi->isrouter = ln->ln_router;
1612 		nbi->expire = ln->ln_expire;
1613 		splx(s);
1614 
1615 		break;
1616 	}
1617 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1618 		ndif->ifindex = nd6_defifindex;
1619 		break;
1620 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1621 		return (nd6_setdefaultiface(ndif->ifindex));
1622 	}
1623 	return (error);
1624 }
1625 
1626 /*
1627  * Create neighbor cache entry and cache link-layer address,
1628  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1629  */
1630 struct rtentry *
1631 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1632 	struct ifnet *ifp;
1633 	struct in6_addr *from;
1634 	char *lladdr;
1635 	int lladdrlen;
1636 	int type;	/* ICMP6 type */
1637 	int code;	/* type dependent information */
1638 {
1639 	struct rtentry *rt = NULL;
1640 	struct llinfo_nd6 *ln = NULL;
1641 	int is_newentry;
1642 	struct sockaddr_dl *sdl = NULL;
1643 	int do_update;
1644 	int olladdr;
1645 	int llchange;
1646 	int newstate = 0;
1647 
1648 	if (ifp == NULL)
1649 		panic("ifp == NULL in nd6_cache_lladdr");
1650 	if (from == NULL)
1651 		panic("from == NULL in nd6_cache_lladdr");
1652 
1653 	/* nothing must be updated for unspecified address */
1654 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1655 		return NULL;
1656 
1657 	/*
1658 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1659 	 * the caller.
1660 	 *
1661 	 * XXX If the link does not have link-layer adderss, what should
1662 	 * we do? (ifp->if_addrlen == 0)
1663 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1664 	 * description on it in NS section (RFC 2461 7.2.3).
1665 	 */
1666 
1667 	rt = nd6_lookup(from, 0, ifp);
1668 	if (rt == NULL) {
1669 #if 0
1670 		/* nothing must be done if there's no lladdr */
1671 		if (!lladdr || !lladdrlen)
1672 			return NULL;
1673 #endif
1674 
1675 		rt = nd6_lookup(from, 1, ifp);
1676 		is_newentry = 1;
1677 	} else {
1678 		/* do nothing if static ndp is set */
1679 		if (rt->rt_flags & RTF_STATIC)
1680 			return NULL;
1681 		is_newentry = 0;
1682 	}
1683 
1684 	if (rt == NULL)
1685 		return NULL;
1686 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1687 fail:
1688 		(void)nd6_free(rt, 0);
1689 		return NULL;
1690 	}
1691 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1692 	if (ln == NULL)
1693 		goto fail;
1694 	if (rt->rt_gateway == NULL)
1695 		goto fail;
1696 	if (rt->rt_gateway->sa_family != AF_LINK)
1697 		goto fail;
1698 	sdl = SDL(rt->rt_gateway);
1699 
1700 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1701 	if (olladdr && lladdr) {
1702 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1703 			llchange = 1;
1704 		else
1705 			llchange = 0;
1706 	} else
1707 		llchange = 0;
1708 
1709 	/*
1710 	 * newentry olladdr  lladdr  llchange	(*=record)
1711 	 *	0	n	n	--	(1)
1712 	 *	0	y	n	--	(2)
1713 	 *	0	n	y	--	(3) * STALE
1714 	 *	0	y	y	n	(4) *
1715 	 *	0	y	y	y	(5) * STALE
1716 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1717 	 *	1	--	y	--	(7) * STALE
1718 	 */
1719 
1720 	if (lladdr) {		/* (3-5) and (7) */
1721 		/*
1722 		 * Record source link-layer address
1723 		 * XXX is it dependent to ifp->if_type?
1724 		 */
1725 		sdl->sdl_alen = ifp->if_addrlen;
1726 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1727 	}
1728 
1729 	if (!is_newentry) {
1730 		if ((!olladdr && lladdr) ||		/* (3) */
1731 		    (olladdr && lladdr && llchange)) {	/* (5) */
1732 			do_update = 1;
1733 			newstate = ND6_LLINFO_STALE;
1734 		} else					/* (1-2,4) */
1735 			do_update = 0;
1736 	} else {
1737 		do_update = 1;
1738 		if (lladdr == NULL)			/* (6) */
1739 			newstate = ND6_LLINFO_NOSTATE;
1740 		else					/* (7) */
1741 			newstate = ND6_LLINFO_STALE;
1742 	}
1743 
1744 	if (do_update) {
1745 		/*
1746 		 * Update the state of the neighbor cache.
1747 		 */
1748 		ln->ln_state = newstate;
1749 
1750 		if (ln->ln_state == ND6_LLINFO_STALE) {
1751 			/*
1752 			 * XXX: since nd6_output() below will cause
1753 			 * state tansition to DELAY and reset the timer,
1754 			 * we must set the timer now, although it is actually
1755 			 * meaningless.
1756 			 */
1757 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1758 
1759  			if (ln->ln_hold) {
1760 				struct mbuf *m_hold, *m_hold_next;
1761 				for (m_hold = ln->ln_hold; m_hold;
1762 				     m_hold = m_hold_next) {
1763 					struct mbuf *mpkt = NULL;
1764 
1765 					m_hold_next = m_hold->m_nextpkt;
1766 					mpkt = m_copym(m_hold, 0, M_COPYALL, M_DONTWAIT);
1767 					if (mpkt == NULL) {
1768 						m_freem(m_hold);
1769 						break;
1770 					}
1771 					mpkt->m_nextpkt = NULL;
1772 
1773 					/*
1774 					 * we assume ifp is not a p2p here, so
1775 					 * just set the 2nd argument as the
1776 					 * 1st one.
1777 					 */
1778 					nd6_output(ifp, ifp, mpkt,
1779 					     (struct sockaddr_in6 *)rt_key(rt),
1780 					     rt);
1781 				}
1782  				ln->ln_hold = NULL;
1783  			}
1784 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1785 			/* probe right away */
1786 			nd6_llinfo_settimer((void *)ln, 0);
1787 		}
1788 	}
1789 
1790 	/*
1791 	 * ICMP6 type dependent behavior.
1792 	 *
1793 	 * NS: clear IsRouter if new entry
1794 	 * RS: clear IsRouter
1795 	 * RA: set IsRouter if there's lladdr
1796 	 * redir: clear IsRouter if new entry
1797 	 *
1798 	 * RA case, (1):
1799 	 * The spec says that we must set IsRouter in the following cases:
1800 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1801 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1802 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1803 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1804 	 * neighbor cache, this is similar to (6).
1805 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1806 	 *
1807 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1808 	 *							D R
1809 	 *	0	n	n	--	(1)	c   ?     s
1810 	 *	0	y	n	--	(2)	c   s     s
1811 	 *	0	n	y	--	(3)	c   s     s
1812 	 *	0	y	y	n	(4)	c   s     s
1813 	 *	0	y	y	y	(5)	c   s     s
1814 	 *	1	--	n	--	(6) c	c 	c s
1815 	 *	1	--	y	--	(7) c	c   s	c s
1816 	 *
1817 	 *					(c=clear s=set)
1818 	 */
1819 	switch (type & 0xff) {
1820 	case ND_NEIGHBOR_SOLICIT:
1821 		/*
1822 		 * New entry must have is_router flag cleared.
1823 		 */
1824 		if (is_newentry)	/* (6-7) */
1825 			ln->ln_router = 0;
1826 		break;
1827 	case ND_REDIRECT:
1828 		/*
1829 		 * If the icmp is a redirect to a better router, always set the
1830 		 * is_router flag.  Otherwise, if the entry is newly created,
1831 		 * clear the flag.  [RFC 2461, sec 8.3]
1832 		 */
1833 		if (code == ND_REDIRECT_ROUTER)
1834 			ln->ln_router = 1;
1835 		else if (is_newentry) /* (6-7) */
1836 			ln->ln_router = 0;
1837 		break;
1838 	case ND_ROUTER_SOLICIT:
1839 		/*
1840 		 * is_router flag must always be cleared.
1841 		 */
1842 		ln->ln_router = 0;
1843 		break;
1844 	case ND_ROUTER_ADVERT:
1845 		/*
1846 		 * Mark an entry with lladdr as a router.
1847 		 */
1848 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1849 		    (is_newentry && lladdr)) {			/* (7) */
1850 			ln->ln_router = 1;
1851 		}
1852 		break;
1853 	}
1854 
1855 	/*
1856 	 * When the link-layer address of a router changes, select the
1857 	 * best router again.  In particular, when the neighbor entry is newly
1858 	 * created, it might affect the selection policy.
1859 	 * Question: can we restrict the first condition to the "is_newentry"
1860 	 * case?
1861 	 * XXX: when we hear an RA from a new router with the link-layer
1862 	 * address option, defrouter_select() is called twice, since
1863 	 * defrtrlist_update called the function as well.  However, I believe
1864 	 * we can compromise the overhead, since it only happens the first
1865 	 * time.
1866 	 * XXX: although defrouter_select() should not have a bad effect
1867 	 * for those are not autoconfigured hosts, we explicitly avoid such
1868 	 * cases for safety.
1869 	 */
1870 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1871 		defrouter_select();
1872 
1873 	return rt;
1874 }
1875 
1876 static void
1877 nd6_slowtimo(ignored_arg)
1878     void *ignored_arg;
1879 {
1880 	int s = splsoftnet();
1881 	struct nd_ifinfo *nd6if;
1882 	struct ifnet *ifp;
1883 
1884 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1885 	    nd6_slowtimo, NULL);
1886 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1887 	{
1888 		nd6if = ND_IFINFO(ifp);
1889 		if (nd6if->basereachable && /* already initialized */
1890 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1891 			/*
1892 			 * Since reachable time rarely changes by router
1893 			 * advertisements, we SHOULD insure that a new random
1894 			 * value gets recomputed at least once every few hours.
1895 			 * (RFC 2461, 6.3.4)
1896 			 */
1897 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1898 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1899 		}
1900 	}
1901 	splx(s);
1902 }
1903 
1904 #define senderr(e) { error = (e); goto bad;}
1905 int
1906 nd6_output(ifp, origifp, m0, dst, rt0)
1907 	struct ifnet *ifp;
1908 	struct ifnet *origifp;
1909 	struct mbuf *m0;
1910 	struct sockaddr_in6 *dst;
1911 	struct rtentry *rt0;
1912 {
1913 	struct mbuf *m = m0;
1914 	struct rtentry *rt = rt0;
1915 	struct sockaddr_in6 *gw6 = NULL;
1916 	struct llinfo_nd6 *ln = NULL;
1917 	int error = 0;
1918 
1919 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1920 		goto sendpkt;
1921 
1922 	if (nd6_need_cache(ifp) == 0)
1923 		goto sendpkt;
1924 
1925 	/*
1926 	 * next hop determination.  This routine is derived from ether_output.
1927 	 */
1928 	if (rt) {
1929 		if ((rt->rt_flags & RTF_UP) == 0) {
1930 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1931 			    1)) != NULL)
1932 			{
1933 				rt->rt_refcnt--;
1934 				if (rt->rt_ifp != ifp)
1935 					senderr(EHOSTUNREACH);
1936 			} else
1937 				senderr(EHOSTUNREACH);
1938 		}
1939 
1940 		if (rt->rt_flags & RTF_GATEWAY) {
1941 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1942 
1943 			/*
1944 			 * We skip link-layer address resolution and NUD
1945 			 * if the gateway is not a neighbor from ND point
1946 			 * of view, regardless of the value of nd_ifinfo.flags.
1947 			 * The second condition is a bit tricky; we skip
1948 			 * if the gateway is our own address, which is
1949 			 * sometimes used to install a route to a p2p link.
1950 			 */
1951 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1952 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1953 				/*
1954 				 * We allow this kind of tricky route only
1955 				 * when the outgoing interface is p2p.
1956 				 * XXX: we may need a more generic rule here.
1957 				 */
1958 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1959 					senderr(EHOSTUNREACH);
1960 
1961 				goto sendpkt;
1962 			}
1963 
1964 			if (rt->rt_gwroute == 0)
1965 				goto lookup;
1966 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1967 				rtfree(rt); rt = rt0;
1968 			lookup:
1969 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1970 				if ((rt = rt->rt_gwroute) == 0)
1971 					senderr(EHOSTUNREACH);
1972 				/* the "G" test below also prevents rt == rt0 */
1973 				if ((rt->rt_flags & RTF_GATEWAY) ||
1974 				    (rt->rt_ifp != ifp)) {
1975 					rt->rt_refcnt--;
1976 					rt0->rt_gwroute = 0;
1977 					senderr(EHOSTUNREACH);
1978 				}
1979 			}
1980 		}
1981 	}
1982 
1983 	/*
1984 	 * Address resolution or Neighbor Unreachability Detection
1985 	 * for the next hop.
1986 	 * At this point, the destination of the packet must be a unicast
1987 	 * or an anycast address(i.e. not a multicast).
1988 	 */
1989 
1990 	/* Look up the neighbor cache for the nexthop */
1991 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1992 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1993 	else {
1994 		/*
1995 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1996 		 * the condition below is not very efficient.  But we believe
1997 		 * it is tolerable, because this should be a rare case.
1998 		 */
1999 		if (nd6_is_addr_neighbor(dst, ifp) &&
2000 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
2001 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2002 	}
2003 	if (ln == NULL || rt == NULL) {
2004 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2005 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2006 			log(LOG_DEBUG,
2007 			    "nd6_output: can't allocate llinfo for %s "
2008 			    "(ln=%p, rt=%p)\n",
2009 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2010 			senderr(EIO);	/* XXX: good error? */
2011 		}
2012 
2013 		goto sendpkt;	/* send anyway */
2014 	}
2015 
2016 	/* We don't have to do link-layer address resolution on a p2p link. */
2017 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2018 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2019 		ln->ln_state = ND6_LLINFO_STALE;
2020 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2021 	}
2022 
2023 	/*
2024 	 * The first time we send a packet to a neighbor whose entry is
2025 	 * STALE, we have to change the state to DELAY and a sets a timer to
2026 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2027 	 * neighbor unreachability detection on expiration.
2028 	 * (RFC 2461 7.3.3)
2029 	 */
2030 	if (ln->ln_state == ND6_LLINFO_STALE) {
2031 		ln->ln_asked = 0;
2032 		ln->ln_state = ND6_LLINFO_DELAY;
2033 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2034 	}
2035 
2036 	/*
2037 	 * If the neighbor cache entry has a state other than INCOMPLETE
2038 	 * (i.e. its link-layer address is already resolved), just
2039 	 * send the packet.
2040 	 */
2041 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2042 		goto sendpkt;
2043 
2044 	/*
2045 	 * There is a neighbor cache entry, but no ethernet address
2046 	 * response yet.  Append this latest packet to the end of the
2047 	 * packet queue in the mbuf, unless the number of the packet
2048 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2049 	 * the oldest packet in the queue will be removed.
2050 	 */
2051 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2052 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2053 	if (ln->ln_hold) {
2054 		struct mbuf *m_hold;
2055 		int i;
2056 
2057 		i = 0;
2058 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2059 			i++;
2060 			if (m_hold->m_nextpkt == NULL) {
2061 				m_hold->m_nextpkt = m;
2062 				break;
2063 			}
2064 		}
2065 		while (i >= nd6_maxqueuelen) {
2066 			m_hold = ln->ln_hold;
2067 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2068 			m_freem(m_hold);
2069 			i--;
2070 		}
2071 	} else {
2072 		ln->ln_hold = m;
2073 	}
2074 
2075 	/*
2076 	 * If there has been no NS for the neighbor after entering the
2077 	 * INCOMPLETE state, send the first solicitation.
2078 	 */
2079 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2080 		ln->ln_asked++;
2081 		nd6_llinfo_settimer(ln,
2082 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2083 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2084 	}
2085 	return (0);
2086 
2087   sendpkt:
2088 	/* discard the packet if IPv6 operation is disabled on the interface */
2089 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2090 		error = ENETDOWN; /* better error? */
2091 		goto bad;
2092 	}
2093 
2094 #ifdef IPSEC
2095 	/* clean ipsec history once it goes out of the node */
2096 	ipsec_delaux(m);
2097 #endif
2098 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2099 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2100 					 rt));
2101 	}
2102 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2103 
2104   bad:
2105 	if (m)
2106 		m_freem(m);
2107 	return (error);
2108 }
2109 #undef senderr
2110 
2111 int
2112 nd6_need_cache(ifp)
2113 	struct ifnet *ifp;
2114 {
2115 	/*
2116 	 * XXX: we currently do not make neighbor cache on any interface
2117 	 * other than ARCnet, Ethernet, FDDI and GIF.
2118 	 *
2119 	 * RFC2893 says:
2120 	 * - unidirectional tunnels needs no ND
2121 	 */
2122 	switch (ifp->if_type) {
2123 	case IFT_ARCNET:
2124 	case IFT_ETHER:
2125 	case IFT_FDDI:
2126 	case IFT_IEEE1394:
2127 	case IFT_CARP:
2128 	case IFT_GIF:		/* XXX need more cases? */
2129 	case IFT_PPP:
2130 	case IFT_TUNNEL:
2131 		return (1);
2132 	default:
2133 		return (0);
2134 	}
2135 }
2136 
2137 int
2138 nd6_storelladdr(ifp, rt, m, dst, desten)
2139 	struct ifnet *ifp;
2140 	struct rtentry *rt;
2141 	struct mbuf *m;
2142 	struct sockaddr *dst;
2143 	u_char *desten;
2144 {
2145 	struct sockaddr_dl *sdl;
2146 
2147 	if (m->m_flags & M_MCAST) {
2148 		switch (ifp->if_type) {
2149 		case IFT_ETHER:
2150 		case IFT_FDDI:
2151 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2152 						 desten);
2153 			return (1);
2154 		case IFT_IEEE1394:
2155 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2156 			return (1);
2157 		case IFT_ARCNET:
2158 			*desten = 0;
2159 			return (1);
2160 		default:
2161 			m_freem(m);
2162 			return (0);
2163 		}
2164 	}
2165 
2166 	if (rt == NULL) {
2167 		/* this could happen, if we could not allocate memory */
2168 		m_freem(m);
2169 		return (0);
2170 	}
2171 	if (rt->rt_gateway->sa_family != AF_LINK) {
2172 		printf("nd6_storelladdr: something odd happens\n");
2173 		m_freem(m);
2174 		return (0);
2175 	}
2176 	sdl = SDL(rt->rt_gateway);
2177 	if (sdl->sdl_alen == 0) {
2178 		/* this should be impossible, but we bark here for debugging */
2179 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
2180 		    ip6_sprintf(&SIN6(dst)->sin6_addr), if_name(ifp));
2181 		m_freem(m);
2182 		return (0);
2183 	}
2184 
2185 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2186 	return (1);
2187 }
2188 
2189 static void
2190 clear_llinfo_pqueue(ln)
2191 	struct llinfo_nd6 *ln;
2192 {
2193 	struct mbuf *m_hold, *m_hold_next;
2194 
2195 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2196 		m_hold_next = m_hold->m_nextpkt;
2197 		m_hold->m_nextpkt = NULL;
2198 		m_freem(m_hold);
2199 	}
2200 
2201 	ln->ln_hold = NULL;
2202 	return;
2203 }
2204 
2205 int
2206 nd6_sysctl(name, oldp, oldlenp, newp, newlen)
2207 	int name;
2208 	void *oldp;	/* syscall arg, need copyout */
2209 	size_t *oldlenp;
2210 	void *newp;	/* syscall arg, need copyin */
2211 	size_t newlen;
2212 {
2213 	void *p;
2214 	size_t ol;
2215 	int error;
2216 
2217 	error = 0;
2218 
2219 	if (newp)
2220 		return EPERM;
2221 	if (oldp && !oldlenp)
2222 		return EINVAL;
2223 	ol = oldlenp ? *oldlenp : 0;
2224 
2225 	if (oldp) {
2226 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2227 		if (p == NULL)
2228 			return ENOMEM;
2229 	} else
2230 		p = NULL;
2231 	switch (name) {
2232 	case ICMPV6CTL_ND6_DRLIST:
2233 		error = fill_drlist(p, oldlenp, ol);
2234 		if (!error && p != NULL && oldp != NULL)
2235 			error = copyout(p, oldp, *oldlenp);
2236 		break;
2237 
2238 	case ICMPV6CTL_ND6_PRLIST:
2239 		error = fill_prlist(p, oldlenp, ol);
2240 		if (!error && p != NULL && oldp != NULL)
2241 			error = copyout(p, oldp, *oldlenp);
2242 		break;
2243 
2244 	case ICMPV6CTL_ND6_MAXQLEN:
2245 		break;
2246 
2247 	default:
2248 		error = ENOPROTOOPT;
2249 		break;
2250 	}
2251 	if (p)
2252 		free(p, M_TEMP);
2253 
2254 	return (error);
2255 }
2256 
2257 static int
2258 fill_drlist(oldp, oldlenp, ol)
2259 	void *oldp;
2260 	size_t *oldlenp, ol;
2261 {
2262 	int error = 0, s;
2263 	struct in6_defrouter *d = NULL, *de = NULL;
2264 	struct nd_defrouter *dr;
2265 	size_t l;
2266 
2267 	s = splsoftnet();
2268 
2269 	if (oldp) {
2270 		d = (struct in6_defrouter *)oldp;
2271 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2272 	}
2273 	l = 0;
2274 
2275 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2276 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2277 
2278 		if (oldp && d + 1 <= de) {
2279 			bzero(d, sizeof(*d));
2280 			d->rtaddr.sin6_family = AF_INET6;
2281 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2282 			d->rtaddr.sin6_addr = dr->rtaddr;
2283 			if (sa6_recoverscope(&d->rtaddr)) {
2284 				log(LOG_ERR,
2285 				    "scope error in router list (%s)\n",
2286 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2287 				/* XXX: press on... */
2288 			}
2289 			d->flags = dr->flags;
2290 			d->rtlifetime = dr->rtlifetime;
2291 			d->expire = dr->expire;
2292 			d->if_index = dr->ifp->if_index;
2293 		}
2294 
2295 		l += sizeof(*d);
2296 		if (d)
2297 			d++;
2298 	}
2299 
2300 	if (oldp) {
2301 		if (l > ol)
2302 			error = ENOMEM;
2303 	}
2304 	if (oldlenp)
2305 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2306 
2307 	splx(s);
2308 
2309 	return (error);
2310 }
2311 
2312 static int
2313 fill_prlist(oldp, oldlenp, ol)
2314 	void *oldp;
2315 	size_t *oldlenp, ol;
2316 {
2317 	int error = 0, s;
2318 	struct nd_prefix *pr;
2319 	struct in6_prefix *p = NULL;
2320 	struct in6_prefix *pe = NULL;
2321 	size_t l;
2322 
2323 	s = splsoftnet();
2324 
2325 	if (oldp) {
2326 		p = (struct in6_prefix *)oldp;
2327 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2328 	}
2329 	l = 0;
2330 
2331 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2332 		u_short advrtrs;
2333 		size_t advance;
2334 		struct sockaddr_in6 *sin6;
2335 		struct sockaddr_in6 *s6;
2336 		struct nd_pfxrouter *pfr;
2337 
2338 		if (oldp && p + 1 <= pe)
2339 		{
2340 			bzero(p, sizeof(*p));
2341 			sin6 = (struct sockaddr_in6 *)(p + 1);
2342 
2343 			p->prefix = pr->ndpr_prefix;
2344 			if (sa6_recoverscope(&p->prefix)) {
2345 				log(LOG_ERR,
2346 				    "scope error in prefix list (%s)\n",
2347 				    ip6_sprintf(&p->prefix.sin6_addr));
2348 				/* XXX: press on... */
2349 			}
2350 			p->raflags = pr->ndpr_raf;
2351 			p->prefixlen = pr->ndpr_plen;
2352 			p->vltime = pr->ndpr_vltime;
2353 			p->pltime = pr->ndpr_pltime;
2354 			p->if_index = pr->ndpr_ifp->if_index;
2355 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2356 				p->expire = 0;
2357 			else {
2358 				time_t maxexpire;
2359 
2360 				/* XXX: we assume time_t is signed. */
2361 				maxexpire = (-1) &
2362 				    ~((time_t)1 <<
2363 				    ((sizeof(maxexpire) * 8) - 1));
2364 				if (pr->ndpr_vltime <
2365 				    maxexpire - pr->ndpr_lastupdate) {
2366 					p->expire = pr->ndpr_lastupdate +
2367 						pr->ndpr_vltime;
2368 				} else
2369 					p->expire = maxexpire;
2370 			}
2371 			p->refcnt = pr->ndpr_refcnt;
2372 			p->flags = pr->ndpr_stateflags;
2373 			p->origin = PR_ORIG_RA;
2374 			advrtrs = 0;
2375 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2376 			     pfr = pfr->pfr_next) {
2377 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2378 					advrtrs++;
2379 					continue;
2380 				}
2381 				s6 = &sin6[advrtrs];
2382 				s6->sin6_family = AF_INET6;
2383 				s6->sin6_len = sizeof(struct sockaddr_in6);
2384 				s6->sin6_addr = pfr->router->rtaddr;
2385 				s6->sin6_scope_id = 0;
2386 				if (sa6_recoverscope(s6)) {
2387 					log(LOG_ERR,
2388 					    "scope error in "
2389 					    "prefix list (%s)\n",
2390 					    ip6_sprintf(&pfr->router->rtaddr));
2391 				}
2392 				advrtrs++;
2393 			}
2394 			p->advrtrs = advrtrs;
2395 		}
2396 		else {
2397 			advrtrs = 0;
2398 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2399 			     pfr = pfr->pfr_next)
2400 				advrtrs++;
2401 		}
2402 
2403 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2404 		l += advance;
2405 		if (p)
2406 			p = (struct in6_prefix *)((caddr_t)p + advance);
2407 	}
2408 
2409 	if (oldp) {
2410 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2411 		if (l > ol)
2412 			error = ENOMEM;
2413 	} else
2414 		*oldlenp = l;
2415 
2416 	splx(s);
2417 
2418 	return (error);
2419 }
2420