xref: /netbsd-src/sys/netinet6/nd6.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: nd6.c,v 1.123 2007/12/04 10:27:34 dyoung 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 1.123 2007/12/04 10:27:34 dyoung 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 /* timer values */
79 int	nd6_prune	= 1;	/* walk list every 1 seconds */
80 int	nd6_delay	= 5;	/* delay first probe time 5 second */
81 int	nd6_umaxtries	= 3;	/* maximum unicast query */
82 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
83 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
84 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
85 
86 /* preventing too many loops in ND option parsing */
87 int nd6_maxndopt = 10;	/* max # of ND options allowed */
88 
89 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
90 
91 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
92 
93 #ifdef ND6_DEBUG
94 int nd6_debug = 1;
95 #else
96 int nd6_debug = 0;
97 #endif
98 
99 /* for debugging? */
100 static int nd6_inuse, nd6_allocated;
101 
102 struct llinfo_nd6 llinfo_nd6 = {
103 	.ln_prev = &llinfo_nd6,
104 	.ln_next = &llinfo_nd6,
105 };
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 const struct sockaddr_in6 all1_sa = {
111 	  .sin6_family = AF_INET6
112 	, .sin6_len = sizeof(struct sockaddr_in6)
113 	, .sin6_addr = {.s6_addr = {0xff, 0xff, 0xff, 0xff,
114 				    0xff, 0xff, 0xff, 0xff,
115 				    0xff, 0xff, 0xff, 0xff,
116 				    0xff, 0xff, 0xff, 0xff}}
117 };
118 
119 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
120 static void nd6_slowtimo(void *);
121 static int regen_tmpaddr(struct in6_ifaddr *);
122 static struct llinfo_nd6 *nd6_free(struct rtentry *, int);
123 static void nd6_llinfo_timer(void *);
124 static void clear_llinfo_pqueue(struct llinfo_nd6 *);
125 
126 callout_t nd6_slowtimo_ch;
127 callout_t nd6_timer_ch;
128 extern callout_t in6_tmpaddrtimer_ch;
129 
130 static int fill_drlist(void *, size_t *, size_t);
131 static int fill_prlist(void *, size_t *, size_t);
132 
133 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
134 
135 void
136 nd6_init(void)
137 {
138 	static int nd6_init_done = 0;
139 
140 	if (nd6_init_done) {
141 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
142 		return;
143 	}
144 
145 	/* initialization of the default router list */
146 	TAILQ_INIT(&nd_defrouter);
147 
148 	nd6_init_done = 1;
149 
150 	callout_init(&nd6_slowtimo_ch, 0);
151 	callout_init(&nd6_timer_ch, 0);
152 
153 	/* start timer */
154 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
155 	    nd6_slowtimo, NULL);
156 }
157 
158 struct nd_ifinfo *
159 nd6_ifattach(struct ifnet *ifp)
160 {
161 	struct nd_ifinfo *nd;
162 
163 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
164 	bzero(nd, sizeof(*nd));
165 
166 	nd->initialized = 1;
167 
168 	nd->chlim = IPV6_DEFHLIM;
169 	nd->basereachable = REACHABLE_TIME;
170 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
171 	nd->retrans = RETRANS_TIMER;
172 	/*
173 	 * Note that the default value of ip6_accept_rtadv is 0, which means
174 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
175 	 * here.
176 	 */
177 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
178 
179 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
180 	nd6_setmtu0(ifp, nd);
181 
182 	return nd;
183 }
184 
185 void
186 nd6_ifdetach(struct nd_ifinfo *nd)
187 {
188 
189 	free(nd, M_IP6NDP);
190 }
191 
192 void
193 nd6_setmtu(struct ifnet *ifp)
194 {
195 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
196 }
197 
198 void
199 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
200 {
201 	u_int32_t omaxmtu;
202 
203 	omaxmtu = ndi->maxmtu;
204 
205 	switch (ifp->if_type) {
206 	case IFT_ARCNET:
207 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
208 		break;
209 	case IFT_FDDI:
210 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
211 		break;
212 	default:
213 		ndi->maxmtu = ifp->if_mtu;
214 		break;
215 	}
216 
217 	/*
218 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
219 	 * undesirable situation.  We thus notify the operator of the change
220 	 * explicitly.  The check for omaxmtu is necessary to restrict the
221 	 * log to the case of changing the MTU, not initializing it.
222 	 */
223 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
224 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
225 		    " small for IPv6 which needs %lu\n",
226 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
227 		    IPV6_MMTU);
228 	}
229 
230 	if (ndi->maxmtu > in6_maxmtu)
231 		in6_setmaxmtu(); /* check all interfaces just in case */
232 }
233 
234 void
235 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
236 {
237 
238 	bzero(ndopts, sizeof(*ndopts));
239 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
240 	ndopts->nd_opts_last
241 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
242 
243 	if (icmp6len == 0) {
244 		ndopts->nd_opts_done = 1;
245 		ndopts->nd_opts_search = NULL;
246 	}
247 }
248 
249 /*
250  * Take one ND option.
251  */
252 struct nd_opt_hdr *
253 nd6_option(union nd_opts *ndopts)
254 {
255 	struct nd_opt_hdr *nd_opt;
256 	int olen;
257 
258 	if (ndopts == NULL)
259 		panic("ndopts == NULL in nd6_option");
260 	if (ndopts->nd_opts_last == NULL)
261 		panic("uninitialized ndopts in nd6_option");
262 	if (ndopts->nd_opts_search == NULL)
263 		return NULL;
264 	if (ndopts->nd_opts_done)
265 		return NULL;
266 
267 	nd_opt = ndopts->nd_opts_search;
268 
269 	/* make sure nd_opt_len is inside the buffer */
270 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
271 		bzero(ndopts, sizeof(*ndopts));
272 		return NULL;
273 	}
274 
275 	olen = nd_opt->nd_opt_len << 3;
276 	if (olen == 0) {
277 		/*
278 		 * Message validation requires that all included
279 		 * options have a length that is greater than zero.
280 		 */
281 		bzero(ndopts, sizeof(*ndopts));
282 		return NULL;
283 	}
284 
285 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
286 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
287 		/* option overruns the end of buffer, invalid */
288 		bzero(ndopts, sizeof(*ndopts));
289 		return NULL;
290 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
291 		/* reached the end of options chain */
292 		ndopts->nd_opts_done = 1;
293 		ndopts->nd_opts_search = NULL;
294 	}
295 	return nd_opt;
296 }
297 
298 /*
299  * Parse multiple ND options.
300  * This function is much easier to use, for ND routines that do not need
301  * multiple options of the same type.
302  */
303 int
304 nd6_options(union nd_opts *ndopts)
305 {
306 	struct nd_opt_hdr *nd_opt;
307 	int i = 0;
308 
309 	if (ndopts == NULL)
310 		panic("ndopts == NULL in nd6_options");
311 	if (ndopts->nd_opts_last == NULL)
312 		panic("uninitialized ndopts in nd6_options");
313 	if (ndopts->nd_opts_search == NULL)
314 		return 0;
315 
316 	while (1) {
317 		nd_opt = nd6_option(ndopts);
318 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
319 			/*
320 			 * Message validation requires that all included
321 			 * options have a length that is greater than zero.
322 			 */
323 			icmp6stat.icp6s_nd_badopt++;
324 			bzero(ndopts, sizeof(*ndopts));
325 			return -1;
326 		}
327 
328 		if (nd_opt == NULL)
329 			goto skip1;
330 
331 		switch (nd_opt->nd_opt_type) {
332 		case ND_OPT_SOURCE_LINKADDR:
333 		case ND_OPT_TARGET_LINKADDR:
334 		case ND_OPT_MTU:
335 		case ND_OPT_REDIRECTED_HEADER:
336 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
337 				nd6log((LOG_INFO,
338 				    "duplicated ND6 option found (type=%d)\n",
339 				    nd_opt->nd_opt_type));
340 				/* XXX bark? */
341 			} else {
342 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
343 					= nd_opt;
344 			}
345 			break;
346 		case ND_OPT_PREFIX_INFORMATION:
347 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
348 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
349 					= nd_opt;
350 			}
351 			ndopts->nd_opts_pi_end =
352 				(struct nd_opt_prefix_info *)nd_opt;
353 			break;
354 		default:
355 			/*
356 			 * Unknown options must be silently ignored,
357 			 * to accommodate future extension to the protocol.
358 			 */
359 			nd6log((LOG_DEBUG,
360 			    "nd6_options: unsupported option %d - "
361 			    "option ignored\n", nd_opt->nd_opt_type));
362 		}
363 
364 skip1:
365 		i++;
366 		if (i > nd6_maxndopt) {
367 			icmp6stat.icp6s_nd_toomanyopt++;
368 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
369 			break;
370 		}
371 
372 		if (ndopts->nd_opts_done)
373 			break;
374 	}
375 
376 	return 0;
377 }
378 
379 /*
380  * ND6 timer routine to handle ND6 entries
381  */
382 void
383 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long xtick)
384 {
385 	int s;
386 
387 	s = splsoftnet();
388 
389 	if (xtick < 0) {
390 		ln->ln_expire = 0;
391 		ln->ln_ntick = 0;
392 		callout_stop(&ln->ln_timer_ch);
393 	} else {
394 		ln->ln_expire = time_second + xtick / hz;
395 		if (xtick > INT_MAX) {
396 			ln->ln_ntick = xtick - INT_MAX;
397 			callout_reset(&ln->ln_timer_ch, INT_MAX,
398 			    nd6_llinfo_timer, ln);
399 		} else {
400 			ln->ln_ntick = 0;
401 			callout_reset(&ln->ln_timer_ch, xtick,
402 			    nd6_llinfo_timer, ln);
403 		}
404 	}
405 
406 	splx(s);
407 }
408 
409 static void
410 nd6_llinfo_timer(void *arg)
411 {
412 	int s;
413 	struct llinfo_nd6 *ln;
414 	struct rtentry *rt;
415 	const struct sockaddr_in6 *dst;
416 	struct ifnet *ifp;
417 	struct nd_ifinfo *ndi = NULL;
418 
419 	s = splsoftnet();
420 
421 	ln = (struct llinfo_nd6 *)arg;
422 
423 	if (ln->ln_ntick > 0) {
424 		nd6_llinfo_settimer(ln, ln->ln_ntick);
425 		splx(s);
426 		return;
427 	}
428 
429 	if ((rt = ln->ln_rt) == NULL)
430 		panic("ln->ln_rt == NULL");
431 	if ((ifp = rt->rt_ifp) == NULL)
432 		panic("ln->ln_rt->rt_ifp == NULL");
433 	ndi = ND_IFINFO(ifp);
434 	dst = satocsin6(rt_getkey(rt));
435 
436 	/* sanity check */
437 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
438 		panic("rt_llinfo(%p) is not equal to ln(%p)",
439 		      rt->rt_llinfo, ln);
440 	if (!dst)
441 		panic("dst=0 in nd6_timer(ln=%p)", ln);
442 
443 	switch (ln->ln_state) {
444 	case ND6_LLINFO_INCOMPLETE:
445 		if (ln->ln_asked < nd6_mmaxtries) {
446 			ln->ln_asked++;
447 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
448 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
449 		} else {
450 			struct mbuf *m = ln->ln_hold;
451 			if (m) {
452 				struct mbuf *m0;
453 
454 				/*
455 				 * assuming every packet in ln_hold has
456 				 * the same IP header
457 				 */
458 				m0 = m->m_nextpkt;
459 				m->m_nextpkt = NULL;
460 				icmp6_error2(m, ICMP6_DST_UNREACH,
461 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
462 
463 				ln->ln_hold = m0;
464 				clear_llinfo_pqueue(ln);
465  			}
466 			(void)nd6_free(rt, 0);
467 			ln = NULL;
468 		}
469 		break;
470 	case ND6_LLINFO_REACHABLE:
471 		if (!ND6_LLINFO_PERMANENT(ln)) {
472 			ln->ln_state = ND6_LLINFO_STALE;
473 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
474 		}
475 		break;
476 
477 	case ND6_LLINFO_STALE:
478 		/* Garbage Collection(RFC 2461 5.3) */
479 		if (!ND6_LLINFO_PERMANENT(ln)) {
480 			(void)nd6_free(rt, 1);
481 			ln = NULL;
482 		}
483 		break;
484 
485 	case ND6_LLINFO_DELAY:
486 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
487 			/* We need NUD */
488 			ln->ln_asked = 1;
489 			ln->ln_state = ND6_LLINFO_PROBE;
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 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
495 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
496 		}
497 		break;
498 	case ND6_LLINFO_PROBE:
499 		if (ln->ln_asked < nd6_umaxtries) {
500 			ln->ln_asked++;
501 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
502 			nd6_ns_output(ifp, &dst->sin6_addr,
503 			    &dst->sin6_addr, ln, 0);
504 		} else {
505 			(void)nd6_free(rt, 0);
506 			ln = NULL;
507 		}
508 		break;
509 	}
510 
511 	splx(s);
512 }
513 
514 /*
515  * ND6 timer routine to expire default route list and prefix list
516  */
517 void
518 nd6_timer(void *ignored_arg)
519 {
520 	int s;
521 	struct nd_defrouter *next_dr, *dr;
522 	struct nd_prefix *next_pr, *pr;
523 	struct in6_ifaddr *ia6, *nia6;
524 	struct in6_addrlifetime *lt6;
525 
526 	s = splsoftnet();
527 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
528 	    nd6_timer, NULL);
529 
530 	/* expire default router list */
531 
532 	for (dr = TAILQ_FIRST(&nd_defrouter); dr != NULL; dr = next_dr) {
533 		next_dr = TAILQ_NEXT(dr, dr_entry);
534 		if (dr->expire && dr->expire < time_second) {
535 			defrtrlist_del(dr);
536 		}
537 	}
538 
539 	/*
540 	 * expire interface addresses.
541 	 * in the past the loop was inside prefix expiry processing.
542 	 * However, from a stricter speci-confrmance standpoint, we should
543 	 * rather separate address lifetimes and prefix lifetimes.
544 	 */
545   addrloop:
546 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
547 		nia6 = ia6->ia_next;
548 		/* check address lifetime */
549 		lt6 = &ia6->ia6_lifetime;
550 		if (IFA6_IS_INVALID(ia6)) {
551 			int regen = 0;
552 
553 			/*
554 			 * If the expiring address is temporary, try
555 			 * regenerating a new one.  This would be useful when
556 			 * we suspended a laptop PC, then turned it on after a
557 			 * period that could invalidate all temporary
558 			 * addresses.  Although we may have to restart the
559 			 * loop (see below), it must be after purging the
560 			 * address.  Otherwise, we'd see an infinite loop of
561 			 * regeneration.
562 			 */
563 			if (ip6_use_tempaddr &&
564 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
565 				if (regen_tmpaddr(ia6) == 0)
566 					regen = 1;
567 			}
568 
569  			in6_purgeaddr(&ia6->ia_ifa);
570 
571 			if (regen)
572 				goto addrloop; /* XXX: see below */
573 		} else if (IFA6_IS_DEPRECATED(ia6)) {
574 			int oldflags = ia6->ia6_flags;
575 
576  			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
577 
578 			/*
579 			 * If a temporary address has just become deprecated,
580 			 * regenerate a new one if possible.
581 			 */
582 			if (ip6_use_tempaddr &&
583 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
584 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
585 
586 				if (regen_tmpaddr(ia6) == 0) {
587 					/*
588 					 * A new temporary address is
589 					 * generated.
590 					 * XXX: this means the address chain
591 					 * has changed while we are still in
592 					 * the loop.  Although the change
593 					 * would not cause disaster (because
594 					 * it's not a deletion, but an
595 					 * addition,) we'd rather restart the
596 					 * loop just for safety.  Or does this
597 					 * significantly reduce performance??
598 					 */
599 					goto addrloop;
600 				}
601 			}
602 		} else {
603 			/*
604 			 * A new RA might have made a deprecated address
605 			 * preferred.
606 			 */
607 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
608 		}
609 	}
610 
611 	/* expire prefix list */
612 	for (pr = LIST_FIRST(&nd_prefix); pr != NULL; pr = next_pr) {
613 		next_pr = LIST_NEXT(pr, ndpr_entry);
614 		/*
615 		 * check prefix lifetime.
616 		 * since pltime is just for autoconf, pltime processing for
617 		 * prefix is not necessary.
618 		 */
619 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
620 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
621 
622 			/*
623 			 * address expiration and prefix expiration are
624 			 * separate.  NEVER perform in6_purgeaddr here.
625 			 */
626 
627 			prelist_remove(pr);
628 		}
629 	}
630 	splx(s);
631 }
632 
633 /* ia6: deprecated/invalidated temporary address */
634 static int
635 regen_tmpaddr(struct in6_ifaddr *ia6)
636 {
637 	struct ifaddr *ifa;
638 	struct ifnet *ifp;
639 	struct in6_ifaddr *public_ifa6 = NULL;
640 
641 	ifp = ia6->ia_ifa.ifa_ifp;
642 	IFADDR_FOREACH(ifa, ifp) {
643 		struct in6_ifaddr *it6;
644 
645 		if (ifa->ifa_addr->sa_family != AF_INET6)
646 			continue;
647 
648 		it6 = (struct in6_ifaddr *)ifa;
649 
650 		/* ignore no autoconf addresses. */
651 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
652 			continue;
653 
654 		/* ignore autoconf addresses with different prefixes. */
655 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
656 			continue;
657 
658 		/*
659 		 * Now we are looking at an autoconf address with the same
660 		 * prefix as ours.  If the address is temporary and is still
661 		 * preferred, do not create another one.  It would be rare, but
662 		 * could happen, for example, when we resume a laptop PC after
663 		 * a long period.
664 		 */
665 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
666 		    !IFA6_IS_DEPRECATED(it6)) {
667 			public_ifa6 = NULL;
668 			break;
669 		}
670 
671 		/*
672 		 * This is a public autoconf address that has the same prefix
673 		 * as ours.  If it is preferred, keep it.  We can't break the
674 		 * loop here, because there may be a still-preferred temporary
675 		 * address with the prefix.
676 		 */
677 		if (!IFA6_IS_DEPRECATED(it6))
678 		    public_ifa6 = it6;
679 	}
680 
681 	if (public_ifa6 != NULL) {
682 		int e;
683 
684 		/*
685 		 * Random factor is introduced in the preferred lifetime, so
686 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
687 		 */
688 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
689 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
690 			    " tmp addr, errno=%d\n", e);
691 			return -1;
692 		}
693 		return 0;
694 	}
695 
696 	return -1;
697 }
698 
699 /*
700  * Nuke neighbor cache/prefix/default router management table, right before
701  * ifp goes away.
702  */
703 void
704 nd6_purge(struct ifnet *ifp)
705 {
706 	struct llinfo_nd6 *ln, *nln;
707 	struct nd_defrouter *dr, *ndr;
708 	struct nd_prefix *pr, *npr;
709 
710 	/*
711 	 * Nuke default router list entries toward ifp.
712 	 * We defer removal of default router list entries that is installed
713 	 * in the routing table, in order to keep additional side effects as
714 	 * small as possible.
715 	 */
716 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
717 		ndr = TAILQ_NEXT(dr, dr_entry);
718 		if (dr->installed)
719 			continue;
720 
721 		if (dr->ifp == ifp)
722 			defrtrlist_del(dr);
723 	}
724 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
725 		ndr = TAILQ_NEXT(dr, dr_entry);
726 		if (!dr->installed)
727 			continue;
728 
729 		if (dr->ifp == ifp)
730 			defrtrlist_del(dr);
731 	}
732 
733 	/* Nuke prefix list entries toward ifp */
734 	for (pr = LIST_FIRST(&nd_prefix); pr != NULL; pr = npr) {
735 		npr = LIST_NEXT(pr, ndpr_entry);
736 		if (pr->ndpr_ifp == ifp) {
737 			/*
738 			 * Because if_detach() does *not* release prefixes
739 			 * while purging addresses the reference count will
740 			 * still be above zero. We therefore reset it to
741 			 * make sure that the prefix really gets purged.
742 			 */
743 			pr->ndpr_refcnt = 0;
744 			/*
745 			 * Previously, pr->ndpr_addr is removed as well,
746 			 * but I strongly believe we don't have to do it.
747 			 * nd6_purge() is only called from in6_ifdetach(),
748 			 * which removes all the associated interface addresses
749 			 * by itself.
750 			 * (jinmei@kame.net 20010129)
751 			 */
752 			prelist_remove(pr);
753 		}
754 	}
755 
756 	/* cancel default outgoing interface setting */
757 	if (nd6_defifindex == ifp->if_index)
758 		nd6_setdefaultiface(0);
759 
760 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
761 		/* refresh default router list */
762 		defrouter_select();
763 	}
764 
765 	/*
766 	 * Nuke neighbor cache entries for the ifp.
767 	 * Note that rt->rt_ifp may not be the same as ifp,
768 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
769 	 * nd6_rtrequest(), and ip6_input().
770 	 */
771 	ln = llinfo_nd6.ln_next;
772 	while (ln && ln != &llinfo_nd6) {
773 		struct rtentry *rt;
774 		const struct sockaddr_dl *sdl;
775 
776 		nln = ln->ln_next;
777 		rt = ln->ln_rt;
778 		if (rt && rt->rt_gateway &&
779 		    rt->rt_gateway->sa_family == AF_LINK) {
780 			sdl = satocsdl(rt->rt_gateway);
781 			if (sdl->sdl_index == ifp->if_index)
782 				nln = nd6_free(rt, 0);
783 		}
784 		ln = nln;
785 	}
786 }
787 
788 struct rtentry *
789 nd6_lookup(const struct in6_addr *addr6, int create, struct ifnet *ifp)
790 {
791 	struct rtentry *rt;
792 	struct sockaddr_in6 sin6;
793 
794 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
795 	rt = rtalloc1((struct sockaddr *)&sin6, create);
796 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
797 		/*
798 		 * This is the case for the default route.
799 		 * If we want to create a neighbor cache for the address, we
800 		 * should free the route for the destination and allocate an
801 		 * interface route.
802 		 */
803 		if (create) {
804 			RTFREE(rt);
805 			rt = NULL;
806 		}
807 	}
808 	if (rt != NULL)
809 		;
810 	else if (create && ifp) {
811 		int e;
812 
813 		/*
814 		 * If no route is available and create is set,
815 		 * we allocate a host route for the destination
816 		 * and treat it like an interface route.
817 		 * This hack is necessary for a neighbor which can't
818 		 * be covered by our own prefix.
819 		 */
820 		struct ifaddr *ifa =
821 		    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
822 		if (ifa == NULL)
823 			return NULL;
824 
825 		/*
826 		 * Create a new route.  RTF_LLINFO is necessary
827 		 * to create a Neighbor Cache entry for the
828 		 * destination in nd6_rtrequest which will be
829 		 * called in rtrequest via ifa->ifa_rtrequest.
830 		 */
831 		if ((e = rtrequest(RTM_ADD, (const struct sockaddr *)&sin6,
832 		    ifa->ifa_addr, (const struct sockaddr *)&all1_sa,
833 		    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
834 		    ~RTF_CLONING, &rt)) != 0) {
835 #if 0
836 			log(LOG_ERR,
837 			    "nd6_lookup: failed to add route for a "
838 			    "neighbor(%s), errno=%d\n",
839 			    ip6_sprintf(addr6), e);
840 #endif
841 			return NULL;
842 		}
843 		if (rt == NULL)
844 			return NULL;
845 		if (rt->rt_llinfo) {
846 			struct llinfo_nd6 *ln =
847 			    (struct llinfo_nd6 *)rt->rt_llinfo;
848 			ln->ln_state = ND6_LLINFO_NOSTATE;
849 		}
850 	} else
851 		return NULL;
852 	rt->rt_refcnt--;
853 	/*
854 	 * Validation for the entry.
855 	 * Note that the check for rt_llinfo is necessary because a cloned
856 	 * route from a parent route that has the L flag (e.g. the default
857 	 * route to a p2p interface) may have the flag, too, while the
858 	 * destination is not actually a neighbor.
859 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
860 	 *      it might be the loopback interface if the entry is for our
861 	 *      own address on a non-loopback interface. Instead, we should
862 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
863 	 *	interface.
864 	 * Note also that ifa_ifp and ifp may differ when we connect two
865 	 * interfaces to a same link, install a link prefix to an interface,
866 	 * and try to install a neighbor cache on an interface that does not
867 	 * have a route to the prefix.
868 	 */
869 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
870 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
871 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
872 		if (create) {
873 			nd6log((LOG_DEBUG,
874 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
875 			    ip6_sprintf(addr6),
876 			    ifp ? if_name(ifp) : "unspec"));
877 		}
878 		return NULL;
879 	}
880 	return rt;
881 }
882 
883 /*
884  * Detect if a given IPv6 address identifies a neighbor on a given link.
885  * XXX: should take care of the destination of a p2p link?
886  */
887 int
888 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
889 {
890 	struct nd_prefix *pr;
891 
892 	/*
893 	 * A link-local address is always a neighbor.
894 	 * XXX: a link does not necessarily specify a single interface.
895 	 */
896 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
897 		struct sockaddr_in6 sin6_copy;
898 		u_int32_t zone;
899 
900 		/*
901 		 * We need sin6_copy since sa6_recoverscope() may modify the
902 		 * content (XXX).
903 		 */
904 		sin6_copy = *addr;
905 		if (sa6_recoverscope(&sin6_copy))
906 			return 0; /* XXX: should be impossible */
907 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
908 			return 0;
909 		if (sin6_copy.sin6_scope_id == zone)
910 			return 1;
911 		else
912 			return 0;
913 	}
914 
915 	/*
916 	 * If the address matches one of our on-link prefixes, it should be a
917 	 * neighbor.
918 	 */
919 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
920 		if (pr->ndpr_ifp != ifp)
921 			continue;
922 
923 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
924 			continue;
925 
926 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
927 		    &addr->sin6_addr, &pr->ndpr_mask))
928 			return 1;
929 	}
930 
931 	/*
932 	 * If the default router list is empty, all addresses are regarded
933 	 * as on-link, and thus, as a neighbor.
934 	 * XXX: we restrict the condition to hosts, because routers usually do
935 	 * not have the "default router list".
936 	 */
937 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
938 	    nd6_defifindex == ifp->if_index) {
939 		return 1;
940 	}
941 
942 	/*
943 	 * Even if the address matches none of our addresses, it might be
944 	 * in the neighbor cache.
945 	 */
946 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
947 		return 1;
948 
949 	return 0;
950 }
951 
952 /*
953  * Free an nd6 llinfo entry.
954  * Since the function would cause significant changes in the kernel, DO NOT
955  * make it global, unless you have a strong reason for the change, and are sure
956  * that the change is safe.
957  */
958 static struct llinfo_nd6 *
959 nd6_free(struct rtentry *rt, int gc)
960 {
961 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
962 	struct in6_addr in6 = satocsin6(rt_getkey(rt))->sin6_addr;
963 	struct nd_defrouter *dr;
964 
965 	/*
966 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
967 	 * even though it is not harmful, it was not really necessary.
968 	 */
969 
970 	/* cancel timer */
971 	nd6_llinfo_settimer(ln, -1);
972 
973 	if (!ip6_forwarding) {
974 		int s;
975 		s = splsoftnet();
976 		dr = defrouter_lookup(&satocsin6(rt_getkey(rt))->sin6_addr,
977 		    rt->rt_ifp);
978 
979 		if (dr != NULL && dr->expire &&
980 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
981 			/*
982 			 * If the reason for the deletion is just garbage
983 			 * collection, and the neighbor is an active default
984 			 * router, do not delete it.  Instead, reset the GC
985 			 * timer using the router's lifetime.
986 			 * Simply deleting the entry would affect default
987 			 * router selection, which is not necessarily a good
988 			 * thing, especially when we're using router preference
989 			 * values.
990 			 * XXX: the check for ln_state would be redundant,
991 			 *      but we intentionally keep it just in case.
992 			 */
993 			if (dr->expire > time_second)
994 				nd6_llinfo_settimer(ln,
995 				    (dr->expire - time_second) * hz);
996 			else
997 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
998 			splx(s);
999 			return ln->ln_next;
1000 		}
1001 
1002 		if (ln->ln_router || dr) {
1003 			/*
1004 			 * rt6_flush must be called whether or not the neighbor
1005 			 * is in the Default Router List.
1006 			 * See a corresponding comment in nd6_na_input().
1007 			 */
1008 			rt6_flush(&in6, rt->rt_ifp);
1009 		}
1010 
1011 		if (dr) {
1012 			/*
1013 			 * Unreachablity of a router might affect the default
1014 			 * router selection and on-link detection of advertised
1015 			 * prefixes.
1016 			 */
1017 
1018 			/*
1019 			 * Temporarily fake the state to choose a new default
1020 			 * router and to perform on-link determination of
1021 			 * prefixes correctly.
1022 			 * Below the state will be set correctly,
1023 			 * or the entry itself will be deleted.
1024 			 */
1025 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1026 
1027 			/*
1028 			 * Since defrouter_select() does not affect the
1029 			 * on-link determination and MIP6 needs the check
1030 			 * before the default router selection, we perform
1031 			 * the check now.
1032 			 */
1033 			pfxlist_onlink_check();
1034 
1035 			/*
1036 			 * refresh default router list
1037 			 */
1038 			defrouter_select();
1039 		}
1040 		splx(s);
1041 	}
1042 
1043 	/*
1044 	 * Before deleting the entry, remember the next entry as the
1045 	 * return value.  We need this because pfxlist_onlink_check() above
1046 	 * might have freed other entries (particularly the old next entry) as
1047 	 * a side effect (XXX).
1048 	 */
1049 	next = ln->ln_next;
1050 
1051 	/*
1052 	 * Detach the route from the routing tree and the list of neighbor
1053 	 * caches, and disable the route entry not to be used in already
1054 	 * cached routes.
1055 	 */
1056 	rtrequest(RTM_DELETE, rt_getkey(rt), (struct sockaddr *)0,
1057 	    rt_mask(rt), 0, (struct rtentry **)0);
1058 
1059 	return next;
1060 }
1061 
1062 /*
1063  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1064  *
1065  * XXX cost-effective methods?
1066  */
1067 void
1068 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1069 {
1070 	struct llinfo_nd6 *ln;
1071 
1072 	/*
1073 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1074 	 * routing table by supplied "dst6".
1075 	 */
1076 	if (rt == NULL) {
1077 		if (dst6 == NULL)
1078 			return;
1079 		if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1080 			return;
1081 	}
1082 
1083 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1084 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1085 	    !rt->rt_llinfo || !rt->rt_gateway ||
1086 	    rt->rt_gateway->sa_family != AF_LINK) {
1087 		/* This is not a host route. */
1088 		return;
1089 	}
1090 
1091 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1092 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1093 		return;
1094 
1095 	/*
1096 	 * if we get upper-layer reachability confirmation many times,
1097 	 * it is possible we have false information.
1098 	 */
1099 	if (!force) {
1100 		ln->ln_byhint++;
1101 		if (ln->ln_byhint > nd6_maxnudhint)
1102 			return;
1103 	}
1104 
1105 	ln->ln_state = ND6_LLINFO_REACHABLE;
1106 	if (!ND6_LLINFO_PERMANENT(ln)) {
1107 		nd6_llinfo_settimer(ln,
1108 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1109 	}
1110 }
1111 
1112 void
1113 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1114 {
1115 	struct sockaddr *gate = rt->rt_gateway;
1116 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1117 	struct ifnet *ifp = rt->rt_ifp;
1118 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1119 	struct ifaddr *ifa;
1120 
1121 	RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1122 	    __LINE__, (void *)rt->_rt_key);
1123 
1124 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1125 		return;
1126 
1127 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1128 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1129 		    __LINE__, (void *)rt->_rt_key);
1130 		/*
1131 		 * This is probably an interface direct route for a link
1132 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1133 		 * We do not need special treatment below for such a route.
1134 		 * Moreover, the RTF_LLINFO flag which would be set below
1135 		 * would annoy the ndp(8) command.
1136 		 */
1137 		return;
1138 	}
1139 
1140 	if (req == RTM_RESOLVE &&
1141 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1142 	     !nd6_is_addr_neighbor(satocsin6(rt_getkey(rt)), ifp))) {
1143 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1144 		    __LINE__, (void *)rt->_rt_key);
1145 		/*
1146 		 * FreeBSD and BSD/OS often make a cloned host route based
1147 		 * on a less-specific route (e.g. the default route).
1148 		 * If the less specific route does not have a "gateway"
1149 		 * (this is the case when the route just goes to a p2p or an
1150 		 * stf interface), we'll mistakenly make a neighbor cache for
1151 		 * the host route, and will see strange neighbor solicitation
1152 		 * for the corresponding destination.  In order to avoid the
1153 		 * confusion, we check if the destination of the route is
1154 		 * a neighbor in terms of neighbor discovery, and stop the
1155 		 * process if not.  Additionally, we remove the LLINFO flag
1156 		 * so that ndp(8) will not try to get the neighbor information
1157 		 * of the destination.
1158 		 */
1159 		rt->rt_flags &= ~RTF_LLINFO;
1160 		return;
1161 	}
1162 
1163 	switch (req) {
1164 	case RTM_ADD:
1165 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1166 		    __LINE__, (void *)rt->_rt_key);
1167 		/*
1168 		 * There is no backward compatibility :)
1169 		 *
1170 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1171 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1172 		 *	   rt->rt_flags |= RTF_CLONING;
1173 		 */
1174 		if ((rt->rt_flags & RTF_CLONING) ||
1175 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1176 			union {
1177 				struct sockaddr sa;
1178 				struct sockaddr_dl sdl;
1179 				struct sockaddr_storage ss;
1180 			} u;
1181 			/*
1182 			 * Case 1: This route should come from a route to
1183 			 * interface (RTF_CLONING case) or the route should be
1184 			 * treated as on-link but is currently not
1185 			 * (RTF_LLINFO && !ln case).
1186 			 */
1187 			sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1188 			    ifp->if_index, ifp->if_type,
1189 			    NULL, namelen, NULL, addrlen);
1190 			rt_setgate(rt, &u.sa);
1191 			gate = rt->rt_gateway;
1192 			RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1193 			    __LINE__, (void *)rt->_rt_key);
1194 			if (ln)
1195 				nd6_llinfo_settimer(ln, 0);
1196 			RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1197 			    __LINE__, (void *)rt->_rt_key);
1198 			if ((rt->rt_flags & RTF_CLONING) != 0)
1199 				break;
1200 		}
1201 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1202 		    __LINE__, (void *)rt->_rt_key);
1203 		/*
1204 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1205 		 * We don't do that here since llinfo is not ready yet.
1206 		 *
1207 		 * There are also couple of other things to be discussed:
1208 		 * - unsolicited NA code needs improvement beforehand
1209 		 * - RFC2461 says we MAY send multicast unsolicited NA
1210 		 *   (7.2.6 paragraph 4), however, it also says that we
1211 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1212 		 *   we don't have anything like it right now.
1213 		 *   note that the mechanism needs a mutual agreement
1214 		 *   between proxies, which means that we need to implement
1215 		 *   a new protocol, or a new kludge.
1216 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1217 		 *   we need to check ip6forwarding before sending it.
1218 		 *   (or should we allow proxy ND configuration only for
1219 		 *   routers?  there's no mention about proxy ND from hosts)
1220 		 */
1221 #if 0
1222 		/* XXX it does not work */
1223 		if (rt->rt_flags & RTF_ANNOUNCE)
1224 			nd6_na_output(ifp,
1225 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1226 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1227 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1228 			      1, NULL);
1229 #endif
1230 		/* FALLTHROUGH */
1231 	case RTM_RESOLVE:
1232 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1233 			RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1234 			    __LINE__, (void *)rt->_rt_key);
1235 			/*
1236 			 * Address resolution isn't necessary for a point to
1237 			 * point link, so we can skip this test for a p2p link.
1238 			 */
1239 			if (gate->sa_family != AF_LINK ||
1240 			    gate->sa_len <
1241 			    sockaddr_dl_measure(namelen, addrlen)) {
1242 				log(LOG_DEBUG,
1243 				    "nd6_rtrequest: bad gateway value: %s\n",
1244 				    if_name(ifp));
1245 				break;
1246 			}
1247 			satosdl(gate)->sdl_type = ifp->if_type;
1248 			satosdl(gate)->sdl_index = ifp->if_index;
1249 			RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1250 			    __LINE__, (void *)rt->_rt_key);
1251 		}
1252 		if (ln != NULL)
1253 			break;	/* This happens on a route change */
1254 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1255 		    __LINE__, (void *)rt->_rt_key);
1256 		/*
1257 		 * Case 2: This route may come from cloning, or a manual route
1258 		 * add with a LL address.
1259 		 */
1260 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1261 		rt->rt_llinfo = (void *)ln;
1262 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1263 		    __LINE__, (void *)rt->_rt_key);
1264 		if (ln == NULL) {
1265 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1266 			break;
1267 		}
1268 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1269 		    __LINE__, (void *)rt->_rt_key);
1270 		nd6_inuse++;
1271 		nd6_allocated++;
1272 		bzero(ln, sizeof(*ln));
1273 		ln->ln_rt = rt;
1274 		callout_init(&ln->ln_timer_ch, 0);
1275 		/* this is required for "ndp" command. - shin */
1276 		if (req == RTM_ADD) {
1277 		        /*
1278 			 * gate should have some valid AF_LINK entry,
1279 			 * and ln->ln_expire should have some lifetime
1280 			 * which is specified by ndp command.
1281 			 */
1282 			ln->ln_state = ND6_LLINFO_REACHABLE;
1283 			ln->ln_byhint = 0;
1284 		} else {
1285 		        /*
1286 			 * When req == RTM_RESOLVE, rt is created and
1287 			 * initialized in rtrequest(), so rt_expire is 0.
1288 			 */
1289 			ln->ln_state = ND6_LLINFO_NOSTATE;
1290 			nd6_llinfo_settimer(ln, 0);
1291 		}
1292 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1293 		    __LINE__, (void *)rt->_rt_key);
1294 		rt->rt_flags |= RTF_LLINFO;
1295 		ln->ln_next = llinfo_nd6.ln_next;
1296 		llinfo_nd6.ln_next = ln;
1297 		ln->ln_prev = &llinfo_nd6;
1298 		ln->ln_next->ln_prev = ln;
1299 
1300 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1301 		    __LINE__, (void *)rt->_rt_key);
1302 		/*
1303 		 * check if rt_getkey(rt) is an address assigned
1304 		 * to the interface.
1305 		 */
1306 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1307 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1308 		RT_DPRINTF("%s l.%d: rt->_rt_key = %p\n", __func__,
1309 		    __LINE__, (void *)rt->_rt_key);
1310 		if (ifa) {
1311 			const void *mac;
1312 			nd6_llinfo_settimer(ln, -1);
1313 			ln->ln_state = ND6_LLINFO_REACHABLE;
1314 			ln->ln_byhint = 0;
1315 			if ((mac = nd6_ifptomac(ifp)) != NULL) {
1316 				/* XXX check for error */
1317 				(void)sockaddr_dl_setaddr(satosdl(gate),
1318 				    gate->sa_len, mac, ifp->if_addrlen);
1319 			}
1320 			if (nd6_useloopback) {
1321 				ifp = rt->rt_ifp = lo0ifp;	/* XXX */
1322 				/*
1323 				 * Make sure rt_ifa be equal to the ifaddr
1324 				 * corresponding to the address.
1325 				 * We need this because when we refer
1326 				 * rt_ifa->ia6_flags in ip6_input, we assume
1327 				 * that the rt_ifa points to the address instead
1328 				 * of the loopback address.
1329 				 */
1330 				if (ifa != rt->rt_ifa)
1331 					rt_replace_ifa(rt, ifa);
1332 				rt->rt_flags &= ~RTF_CLONED;
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 = satocsin6(rt_getkey(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 = satocsin6(rt_getkey(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 			}
1379 		}
1380 		nd6_inuse--;
1381 		ln->ln_next->ln_prev = ln->ln_prev;
1382 		ln->ln_prev->ln_next = ln->ln_next;
1383 		ln->ln_prev = NULL;
1384 		nd6_llinfo_settimer(ln, -1);
1385 		rt->rt_llinfo = 0;
1386 		rt->rt_flags &= ~RTF_LLINFO;
1387 		clear_llinfo_pqueue(ln);
1388 		Free((void *)ln);
1389 	}
1390 }
1391 
1392 int
1393 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1394 {
1395 	struct in6_drlist *drl = (struct in6_drlist *)data;
1396 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1397 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1398 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1399 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1400 	struct nd_defrouter *dr;
1401 	struct nd_prefix *pr;
1402 	struct rtentry *rt;
1403 	int i = 0, error = 0;
1404 	int s;
1405 
1406 	switch (cmd) {
1407 	case SIOCGDRLST_IN6:
1408 		/*
1409 		 * obsolete API, use sysctl under net.inet6.icmp6
1410 		 */
1411 		bzero(drl, sizeof(*drl));
1412 		s = splsoftnet();
1413 		TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
1414 			if (i >= DRLSTSIZ)
1415 				break;
1416 			drl->defrouter[i].rtaddr = dr->rtaddr;
1417 			in6_clearscope(&drl->defrouter[i].rtaddr);
1418 
1419 			drl->defrouter[i].flags = dr->flags;
1420 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1421 			drl->defrouter[i].expire = dr->expire;
1422 			drl->defrouter[i].if_index = dr->ifp->if_index;
1423 			i++;
1424 		}
1425 		splx(s);
1426 		break;
1427 	case SIOCGPRLST_IN6:
1428 		/*
1429 		 * obsolete API, use sysctl under net.inet6.icmp6
1430 		 *
1431 		 * XXX the structure in6_prlist was changed in backward-
1432 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1433 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1434 		 */
1435 		/*
1436 		 * XXX meaning of fields, especialy "raflags", is very
1437 		 * differnet between RA prefix list and RR/static prefix list.
1438 		 * how about separating ioctls into two?
1439 		 */
1440 		bzero(oprl, sizeof(*oprl));
1441 		s = splsoftnet();
1442 		LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1443 			struct nd_pfxrouter *pfr;
1444 			int j;
1445 
1446 			if (i >= PRLSTSIZ)
1447 				break;
1448 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1449 			oprl->prefix[i].raflags = pr->ndpr_raf;
1450 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1451 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1452 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1453 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1454 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1455 				oprl->prefix[i].expire = 0;
1456 			else {
1457 				time_t maxexpire;
1458 
1459 				/* XXX: we assume time_t is signed. */
1460 				maxexpire = (-1) &
1461 				    ~((time_t)1 <<
1462 				    ((sizeof(maxexpire) * 8) - 1));
1463 				if (pr->ndpr_vltime <
1464 				    maxexpire - pr->ndpr_lastupdate) {
1465 					oprl->prefix[i].expire =
1466 						 pr->ndpr_lastupdate +
1467 						pr->ndpr_vltime;
1468 				} else
1469 					oprl->prefix[i].expire = maxexpire;
1470 			}
1471 
1472 			j = 0;
1473 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1474 				if (j < DRLSTSIZ) {
1475 #define RTRADDR oprl->prefix[i].advrtr[j]
1476 					RTRADDR = pfr->router->rtaddr;
1477 					in6_clearscope(&RTRADDR);
1478 #undef RTRADDR
1479 				}
1480 				j++;
1481 			}
1482 			oprl->prefix[i].advrtrs = j;
1483 			oprl->prefix[i].origin = PR_ORIG_RA;
1484 
1485 			i++;
1486 		}
1487 		splx(s);
1488 
1489 		break;
1490 	case OSIOCGIFINFO_IN6:
1491 #define ND	ndi->ndi
1492 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1493 		memset(&ND, 0, sizeof(ND));
1494 		ND.linkmtu = IN6_LINKMTU(ifp);
1495 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1496 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1497 		ND.reachable = ND_IFINFO(ifp)->reachable;
1498 		ND.retrans = ND_IFINFO(ifp)->retrans;
1499 		ND.flags = ND_IFINFO(ifp)->flags;
1500 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1501 		ND.chlim = ND_IFINFO(ifp)->chlim;
1502 		break;
1503 	case SIOCGIFINFO_IN6:
1504 		ND = *ND_IFINFO(ifp);
1505 		break;
1506 	case SIOCSIFINFO_IN6:
1507 		/*
1508 		 * used to change host variables from userland.
1509 		 * intented for a use on router to reflect RA configurations.
1510 		 */
1511 		/* 0 means 'unspecified' */
1512 		if (ND.linkmtu != 0) {
1513 			if (ND.linkmtu < IPV6_MMTU ||
1514 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1515 				error = EINVAL;
1516 				break;
1517 			}
1518 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1519 		}
1520 
1521 		if (ND.basereachable != 0) {
1522 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1523 
1524 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1525 			if (ND.basereachable != obasereachable)
1526 				ND_IFINFO(ifp)->reachable =
1527 				    ND_COMPUTE_RTIME(ND.basereachable);
1528 		}
1529 		if (ND.retrans != 0)
1530 			ND_IFINFO(ifp)->retrans = ND.retrans;
1531 		if (ND.chlim != 0)
1532 			ND_IFINFO(ifp)->chlim = ND.chlim;
1533 		/* FALLTHROUGH */
1534 	case SIOCSIFINFO_FLAGS:
1535 		ND_IFINFO(ifp)->flags = ND.flags;
1536 		break;
1537 #undef ND
1538 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1539 		/* sync kernel routing table with the default router list */
1540 		defrouter_reset();
1541 		defrouter_select();
1542 		break;
1543 	case SIOCSPFXFLUSH_IN6:
1544 	{
1545 		/* flush all the prefix advertised by routers */
1546 		struct nd_prefix *pfx, *next;
1547 
1548 		s = splsoftnet();
1549 		for (pfx = LIST_FIRST(&nd_prefix); pfx; pfx = next) {
1550 			struct in6_ifaddr *ia, *ia_next;
1551 
1552 			next = LIST_NEXT(pfx, ndpr_entry);
1553 
1554 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1555 				continue; /* XXX */
1556 
1557 			/* do we really have to remove addresses as well? */
1558 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1559 				/* ia might be removed.  keep the next ptr. */
1560 				ia_next = ia->ia_next;
1561 
1562 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1563 					continue;
1564 
1565 				if (ia->ia6_ndpr == pfx)
1566 					in6_purgeaddr(&ia->ia_ifa);
1567 			}
1568 			prelist_remove(pfx);
1569 		}
1570 		splx(s);
1571 		break;
1572 	}
1573 	case SIOCSRTRFLUSH_IN6:
1574 	{
1575 		/* flush all the default routers */
1576 		struct nd_defrouter *drtr, *next;
1577 
1578 		s = splsoftnet();
1579 		defrouter_reset();
1580 		for (drtr = TAILQ_FIRST(&nd_defrouter); drtr; drtr = next) {
1581 			next = TAILQ_NEXT(drtr, dr_entry);
1582 			defrtrlist_del(drtr);
1583 		}
1584 		defrouter_select();
1585 		splx(s);
1586 		break;
1587 	}
1588 	case SIOCGNBRINFO_IN6:
1589 	{
1590 		struct llinfo_nd6 *ln;
1591 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1592 
1593 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1594 			return error;
1595 
1596 		s = splsoftnet();
1597 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1598 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1599 			error = EINVAL;
1600 			splx(s);
1601 			break;
1602 		}
1603 		nbi->state = ln->ln_state;
1604 		nbi->asked = ln->ln_asked;
1605 		nbi->isrouter = ln->ln_router;
1606 		nbi->expire = ln->ln_expire;
1607 		splx(s);
1608 
1609 		break;
1610 	}
1611 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1612 		ndif->ifindex = nd6_defifindex;
1613 		break;
1614 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1615 		return nd6_setdefaultiface(ndif->ifindex);
1616 	}
1617 	return error;
1618 }
1619 
1620 void
1621 nd6_llinfo_release_pkts(struct llinfo_nd6 *ln, struct ifnet *ifp,
1622     struct rtentry *rt)
1623 {
1624 	struct mbuf *m_hold, *m_hold_next;
1625 
1626 	for (m_hold = ln->ln_hold, ln->ln_hold = NULL;
1627 	     m_hold != NULL;
1628 	     m_hold = m_hold_next) {
1629 		m_hold_next = m_hold->m_nextpkt;
1630 		m_hold->m_nextpkt = NULL;
1631 
1632 		/*
1633 		 * we assume ifp is not a p2p here, so
1634 		 * just set the 2nd argument as the
1635 		 * 1st one.
1636 		 */
1637 		nd6_output(ifp, ifp, m_hold, satocsin6(rt_getkey(rt)), rt);
1638 	}
1639 }
1640 
1641 /*
1642  * Create neighbor cache entry and cache link-layer address,
1643  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1644  */
1645 struct rtentry *
1646 nd6_cache_lladdr(
1647     struct ifnet *ifp,
1648     struct in6_addr *from,
1649     char *lladdr,
1650     int lladdrlen,
1651     int type,	/* ICMP6 type */
1652     int code	/* type dependent information */
1653 )
1654 {
1655 	struct rtentry *rt = NULL;
1656 	struct llinfo_nd6 *ln = NULL;
1657 	int is_newentry;
1658 	struct sockaddr_dl *sdl = NULL;
1659 	int do_update;
1660 	int olladdr;
1661 	int llchange;
1662 	int newstate = 0;
1663 
1664 	if (ifp == NULL)
1665 		panic("ifp == NULL in nd6_cache_lladdr");
1666 	if (from == NULL)
1667 		panic("from == NULL in nd6_cache_lladdr");
1668 
1669 	/* nothing must be updated for unspecified address */
1670 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1671 		return NULL;
1672 
1673 	/*
1674 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1675 	 * the caller.
1676 	 *
1677 	 * XXX If the link does not have link-layer adderss, what should
1678 	 * we do? (ifp->if_addrlen == 0)
1679 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1680 	 * description on it in NS section (RFC 2461 7.2.3).
1681 	 */
1682 
1683 	rt = nd6_lookup(from, 0, ifp);
1684 	if (rt == NULL) {
1685 #if 0
1686 		/* nothing must be done if there's no lladdr */
1687 		if (!lladdr || !lladdrlen)
1688 			return NULL;
1689 #endif
1690 
1691 		rt = nd6_lookup(from, 1, ifp);
1692 		is_newentry = 1;
1693 	} else {
1694 		/* do nothing if static ndp is set */
1695 		if (rt->rt_flags & RTF_STATIC)
1696 			return NULL;
1697 		is_newentry = 0;
1698 	}
1699 
1700 	if (rt == NULL)
1701 		return NULL;
1702 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1703 fail:
1704 		(void)nd6_free(rt, 0);
1705 		return NULL;
1706 	}
1707 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1708 	if (ln == NULL)
1709 		goto fail;
1710 	if (rt->rt_gateway == NULL)
1711 		goto fail;
1712 	if (rt->rt_gateway->sa_family != AF_LINK)
1713 		goto fail;
1714 	sdl = satosdl(rt->rt_gateway);
1715 
1716 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1717 	if (olladdr && lladdr) {
1718 		if (memcmp(lladdr, CLLADDR(sdl), ifp->if_addrlen))
1719 			llchange = 1;
1720 		else
1721 			llchange = 0;
1722 	} else
1723 		llchange = 0;
1724 
1725 	/*
1726 	 * newentry olladdr  lladdr  llchange	(*=record)
1727 	 *	0	n	n	--	(1)
1728 	 *	0	y	n	--	(2)
1729 	 *	0	n	y	--	(3) * STALE
1730 	 *	0	y	y	n	(4) *
1731 	 *	0	y	y	y	(5) * STALE
1732 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1733 	 *	1	--	y	--	(7) * STALE
1734 	 */
1735 
1736 	if (lladdr) {		/* (3-5) and (7) */
1737 		/*
1738 		 * Record source link-layer address
1739 		 * XXX is it dependent to ifp->if_type?
1740 		 */
1741 		/* XXX check for error */
1742 		(void)sockaddr_dl_setaddr(sdl, sdl->sdl_len, lladdr,
1743 		    ifp->if_addrlen);
1744 	}
1745 
1746 	if (!is_newentry) {
1747 		if ((!olladdr && lladdr) ||		/* (3) */
1748 		    (olladdr && lladdr && llchange)) {	/* (5) */
1749 			do_update = 1;
1750 			newstate = ND6_LLINFO_STALE;
1751 		} else					/* (1-2,4) */
1752 			do_update = 0;
1753 	} else {
1754 		do_update = 1;
1755 		if (lladdr == NULL)			/* (6) */
1756 			newstate = ND6_LLINFO_NOSTATE;
1757 		else					/* (7) */
1758 			newstate = ND6_LLINFO_STALE;
1759 	}
1760 
1761 	if (do_update) {
1762 		/*
1763 		 * Update the state of the neighbor cache.
1764 		 */
1765 		ln->ln_state = newstate;
1766 
1767 		if (ln->ln_state == ND6_LLINFO_STALE) {
1768 			/*
1769 			 * XXX: since nd6_output() below will cause
1770 			 * state tansition to DELAY and reset the timer,
1771 			 * we must set the timer now, although it is actually
1772 			 * meaningless.
1773 			 */
1774 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1775 
1776 			nd6_llinfo_release_pkts(ln, ifp, rt);
1777 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1778 			/* probe right away */
1779 			nd6_llinfo_settimer((void *)ln, 0);
1780 		}
1781 	}
1782 
1783 	/*
1784 	 * ICMP6 type dependent behavior.
1785 	 *
1786 	 * NS: clear IsRouter if new entry
1787 	 * RS: clear IsRouter
1788 	 * RA: set IsRouter if there's lladdr
1789 	 * redir: clear IsRouter if new entry
1790 	 *
1791 	 * RA case, (1):
1792 	 * The spec says that we must set IsRouter in the following cases:
1793 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1794 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1795 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1796 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1797 	 * neighbor cache, this is similar to (6).
1798 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1799 	 *
1800 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1801 	 *							D R
1802 	 *	0	n	n	--	(1)	c   ?     s
1803 	 *	0	y	n	--	(2)	c   s     s
1804 	 *	0	n	y	--	(3)	c   s     s
1805 	 *	0	y	y	n	(4)	c   s     s
1806 	 *	0	y	y	y	(5)	c   s     s
1807 	 *	1	--	n	--	(6) c	c 	c s
1808 	 *	1	--	y	--	(7) c	c   s	c s
1809 	 *
1810 	 *					(c=clear s=set)
1811 	 */
1812 	switch (type & 0xff) {
1813 	case ND_NEIGHBOR_SOLICIT:
1814 		/*
1815 		 * New entry must have is_router flag cleared.
1816 		 */
1817 		if (is_newentry)	/* (6-7) */
1818 			ln->ln_router = 0;
1819 		break;
1820 	case ND_REDIRECT:
1821 		/*
1822 		 * If the icmp is a redirect to a better router, always set the
1823 		 * is_router flag.  Otherwise, if the entry is newly created,
1824 		 * clear the flag.  [RFC 2461, sec 8.3]
1825 		 */
1826 		if (code == ND_REDIRECT_ROUTER)
1827 			ln->ln_router = 1;
1828 		else if (is_newentry) /* (6-7) */
1829 			ln->ln_router = 0;
1830 		break;
1831 	case ND_ROUTER_SOLICIT:
1832 		/*
1833 		 * is_router flag must always be cleared.
1834 		 */
1835 		ln->ln_router = 0;
1836 		break;
1837 	case ND_ROUTER_ADVERT:
1838 		/*
1839 		 * Mark an entry with lladdr as a router.
1840 		 */
1841 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1842 		    (is_newentry && lladdr)) {			/* (7) */
1843 			ln->ln_router = 1;
1844 		}
1845 		break;
1846 	}
1847 
1848 	/*
1849 	 * When the link-layer address of a router changes, select the
1850 	 * best router again.  In particular, when the neighbor entry is newly
1851 	 * created, it might affect the selection policy.
1852 	 * Question: can we restrict the first condition to the "is_newentry"
1853 	 * case?
1854 	 * XXX: when we hear an RA from a new router with the link-layer
1855 	 * address option, defrouter_select() is called twice, since
1856 	 * defrtrlist_update called the function as well.  However, I believe
1857 	 * we can compromise the overhead, since it only happens the first
1858 	 * time.
1859 	 * XXX: although defrouter_select() should not have a bad effect
1860 	 * for those are not autoconfigured hosts, we explicitly avoid such
1861 	 * cases for safety.
1862 	 */
1863 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1864 		defrouter_select();
1865 
1866 	return rt;
1867 }
1868 
1869 static void
1870 nd6_slowtimo(void *ignored_arg)
1871 {
1872 	int s = splsoftnet();
1873 	struct nd_ifinfo *nd6if;
1874 	struct ifnet *ifp;
1875 
1876 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1877 	    nd6_slowtimo, NULL);
1878 	TAILQ_FOREACH(ifp, &ifnet, if_list) {
1879 		nd6if = ND_IFINFO(ifp);
1880 		if (nd6if->basereachable && /* already initialized */
1881 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1882 			/*
1883 			 * Since reachable time rarely changes by router
1884 			 * advertisements, we SHOULD insure that a new random
1885 			 * value gets recomputed at least once every few hours.
1886 			 * (RFC 2461, 6.3.4)
1887 			 */
1888 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1889 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1890 		}
1891 	}
1892 	splx(s);
1893 }
1894 
1895 #define senderr(e) { error = (e); goto bad;}
1896 int
1897 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1898     const struct sockaddr_in6 *dst, struct rtentry *rt0)
1899 {
1900 	struct mbuf *m = m0;
1901 	struct rtentry *rt = rt0;
1902 	struct sockaddr_in6 *gw6 = NULL;
1903 	struct llinfo_nd6 *ln = NULL;
1904 	int error = 0;
1905 
1906 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1907 		goto sendpkt;
1908 
1909 	if (nd6_need_cache(ifp) == 0)
1910 		goto sendpkt;
1911 
1912 	/*
1913 	 * next hop determination.  This routine is derived from ether_output.
1914 	 */
1915 	if (rt) {
1916 		if ((rt->rt_flags & RTF_UP) == 0) {
1917 			if ((rt0 = rt = rtalloc1(sin6tocsa(dst), 1)) != NULL) {
1918 				rt->rt_refcnt--;
1919 				if (rt->rt_ifp != ifp)
1920 					senderr(EHOSTUNREACH);
1921 			} else
1922 				senderr(EHOSTUNREACH);
1923 		}
1924 
1925 		if (rt->rt_flags & RTF_GATEWAY) {
1926 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1927 
1928 			/*
1929 			 * We skip link-layer address resolution and NUD
1930 			 * if the gateway is not a neighbor from ND point
1931 			 * of view, regardless of the value of nd_ifinfo.flags.
1932 			 * The second condition is a bit tricky; we skip
1933 			 * if the gateway is our own address, which is
1934 			 * sometimes used to install a route to a p2p link.
1935 			 */
1936 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1937 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1938 				/*
1939 				 * We allow this kind of tricky route only
1940 				 * when the outgoing interface is p2p.
1941 				 * XXX: we may need a more generic rule here.
1942 				 */
1943 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1944 					senderr(EHOSTUNREACH);
1945 
1946 				goto sendpkt;
1947 			}
1948 
1949 			if (rt->rt_gwroute == 0)
1950 				goto lookup;
1951 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1952 				rtfree(rt); rt = rt0;
1953 			lookup:
1954 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1955 				if ((rt = rt->rt_gwroute) == 0)
1956 					senderr(EHOSTUNREACH);
1957 				/* the "G" test below also prevents rt == rt0 */
1958 				if ((rt->rt_flags & RTF_GATEWAY) ||
1959 				    (rt->rt_ifp != ifp)) {
1960 					rt->rt_refcnt--;
1961 					rt0->rt_gwroute = 0;
1962 					senderr(EHOSTUNREACH);
1963 				}
1964 			}
1965 		}
1966 	}
1967 
1968 	/*
1969 	 * Address resolution or Neighbor Unreachability Detection
1970 	 * for the next hop.
1971 	 * At this point, the destination of the packet must be a unicast
1972 	 * or an anycast address(i.e. not a multicast).
1973 	 */
1974 
1975 	/* Look up the neighbor cache for the nexthop */
1976 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1977 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1978 	else {
1979 		/*
1980 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1981 		 * the condition below is not very efficient.  But we believe
1982 		 * it is tolerable, because this should be a rare case.
1983 		 */
1984 		if (nd6_is_addr_neighbor(dst, ifp) &&
1985 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1986 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1987 	}
1988 	if (ln == NULL || rt == NULL) {
1989 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1990 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1991 			log(LOG_DEBUG,
1992 			    "nd6_output: can't allocate llinfo for %s "
1993 			    "(ln=%p, rt=%p)\n",
1994 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1995 			senderr(EIO);	/* XXX: good error? */
1996 		}
1997 
1998 		goto sendpkt;	/* send anyway */
1999 	}
2000 
2001 	/* We don't have to do link-layer address resolution on a p2p link. */
2002 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2003 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2004 		ln->ln_state = ND6_LLINFO_STALE;
2005 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2006 	}
2007 
2008 	/*
2009 	 * The first time we send a packet to a neighbor whose entry is
2010 	 * STALE, we have to change the state to DELAY and a sets a timer to
2011 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2012 	 * neighbor unreachability detection on expiration.
2013 	 * (RFC 2461 7.3.3)
2014 	 */
2015 	if (ln->ln_state == ND6_LLINFO_STALE) {
2016 		ln->ln_asked = 0;
2017 		ln->ln_state = ND6_LLINFO_DELAY;
2018 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2019 	}
2020 
2021 	/*
2022 	 * If the neighbor cache entry has a state other than INCOMPLETE
2023 	 * (i.e. its link-layer address is already resolved), just
2024 	 * send the packet.
2025 	 */
2026 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2027 		goto sendpkt;
2028 
2029 	/*
2030 	 * There is a neighbor cache entry, but no ethernet address
2031 	 * response yet.  Append this latest packet to the end of the
2032 	 * packet queue in the mbuf, unless the number of the packet
2033 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2034 	 * the oldest packet in the queue will be removed.
2035 	 */
2036 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2037 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2038 	if (ln->ln_hold) {
2039 		struct mbuf *m_hold;
2040 		int i;
2041 
2042 		i = 0;
2043 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2044 			i++;
2045 			if (m_hold->m_nextpkt == NULL) {
2046 				m_hold->m_nextpkt = m;
2047 				break;
2048 			}
2049 		}
2050 		while (i >= nd6_maxqueuelen) {
2051 			m_hold = ln->ln_hold;
2052 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2053 			m_freem(m_hold);
2054 			i--;
2055 		}
2056 	} else {
2057 		ln->ln_hold = m;
2058 	}
2059 
2060 	/*
2061 	 * If there has been no NS for the neighbor after entering the
2062 	 * INCOMPLETE state, send the first solicitation.
2063 	 */
2064 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2065 		ln->ln_asked++;
2066 		nd6_llinfo_settimer(ln,
2067 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2068 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2069 	}
2070 	return 0;
2071 
2072   sendpkt:
2073 	/* discard the packet if IPv6 operation is disabled on the interface */
2074 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2075 		error = ENETDOWN; /* better error? */
2076 		goto bad;
2077 	}
2078 
2079 #ifdef IPSEC
2080 	/* clean ipsec history once it goes out of the node */
2081 	ipsec_delaux(m);
2082 #endif
2083 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2084 		return (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt);
2085 	}
2086 	return (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt);
2087 
2088   bad:
2089 	if (m != NULL)
2090 		m_freem(m);
2091 	return error;
2092 }
2093 #undef senderr
2094 
2095 int
2096 nd6_need_cache(struct ifnet *ifp)
2097 {
2098 	/*
2099 	 * XXX: we currently do not make neighbor cache on any interface
2100 	 * other than ARCnet, Ethernet, FDDI and GIF.
2101 	 *
2102 	 * RFC2893 says:
2103 	 * - unidirectional tunnels needs no ND
2104 	 */
2105 	switch (ifp->if_type) {
2106 	case IFT_ARCNET:
2107 	case IFT_ETHER:
2108 	case IFT_FDDI:
2109 	case IFT_IEEE1394:
2110 	case IFT_CARP:
2111 	case IFT_GIF:		/* XXX need more cases? */
2112 	case IFT_PPP:
2113 	case IFT_TUNNEL:
2114 		return 1;
2115 	default:
2116 		return 0;
2117 	}
2118 }
2119 
2120 int
2121 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2122     struct mbuf *m, const struct sockaddr *dst, u_char *lldst,
2123     size_t dstsize)
2124 {
2125 	const struct sockaddr_dl *sdl;
2126 
2127 	if (m->m_flags & M_MCAST) {
2128 		switch (ifp->if_type) {
2129 		case IFT_ETHER:
2130 		case IFT_FDDI:
2131 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2132 			    lldst);
2133 			return 1;
2134 		case IFT_IEEE1394:
2135 			memcpy(lldst, ifp->if_broadcastaddr,
2136 			    MIN(dstsize, ifp->if_addrlen));
2137 			return 1;
2138 		case IFT_ARCNET:
2139 			*lldst = 0;
2140 			return 1;
2141 		default:
2142 			m_freem(m);
2143 			return 0;
2144 		}
2145 	}
2146 
2147 	if (rt == NULL) {
2148 		/* this could happen, if we could not allocate memory */
2149 		m_freem(m);
2150 		return 0;
2151 	}
2152 	if (rt->rt_gateway->sa_family != AF_LINK) {
2153 		printf("%s: something odd happens\n", __func__);
2154 		m_freem(m);
2155 		return 0;
2156 	}
2157 	sdl = satocsdl(rt->rt_gateway);
2158 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2159 		/* this should be impossible, but we bark here for debugging */
2160 		printf("%s: sdl_alen == 0, dst=%s, if=%s\n", __func__,
2161 		    ip6_sprintf(&satocsin6(dst)->sin6_addr), if_name(ifp));
2162 		m_freem(m);
2163 		return 0;
2164 	}
2165 
2166 	memcpy(lldst, CLLADDR(sdl), MIN(dstsize, sdl->sdl_alen));
2167 	return 1;
2168 }
2169 
2170 static void
2171 clear_llinfo_pqueue(struct llinfo_nd6 *ln)
2172 {
2173 	struct mbuf *m_hold, *m_hold_next;
2174 
2175 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2176 		m_hold_next = m_hold->m_nextpkt;
2177 		m_hold->m_nextpkt = NULL;
2178 		m_freem(m_hold);
2179 	}
2180 
2181 	ln->ln_hold = NULL;
2182 	return;
2183 }
2184 
2185 int
2186 nd6_sysctl(
2187     int name,
2188     void *oldp,	/* syscall arg, need copyout */
2189     size_t *oldlenp,
2190     void *newp,	/* syscall arg, need copyin */
2191     size_t newlen
2192 )
2193 {
2194 	void *p;
2195 	size_t ol;
2196 	int error;
2197 
2198 	error = 0;
2199 
2200 	if (newp)
2201 		return EPERM;
2202 	if (oldp && !oldlenp)
2203 		return EINVAL;
2204 	ol = oldlenp ? *oldlenp : 0;
2205 
2206 	if (oldp) {
2207 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2208 		if (p == NULL)
2209 			return ENOMEM;
2210 	} else
2211 		p = NULL;
2212 	switch (name) {
2213 	case ICMPV6CTL_ND6_DRLIST:
2214 		error = fill_drlist(p, oldlenp, ol);
2215 		if (!error && p != NULL && oldp != NULL)
2216 			error = copyout(p, oldp, *oldlenp);
2217 		break;
2218 
2219 	case ICMPV6CTL_ND6_PRLIST:
2220 		error = fill_prlist(p, oldlenp, ol);
2221 		if (!error && p != NULL && oldp != NULL)
2222 			error = copyout(p, oldp, *oldlenp);
2223 		break;
2224 
2225 	case ICMPV6CTL_ND6_MAXQLEN:
2226 		break;
2227 
2228 	default:
2229 		error = ENOPROTOOPT;
2230 		break;
2231 	}
2232 	if (p)
2233 		free(p, M_TEMP);
2234 
2235 	return error;
2236 }
2237 
2238 static int
2239 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2240 {
2241 	int error = 0, s;
2242 	struct in6_defrouter *d = NULL, *de = NULL;
2243 	struct nd_defrouter *dr;
2244 	size_t l;
2245 
2246 	s = splsoftnet();
2247 
2248 	if (oldp) {
2249 		d = (struct in6_defrouter *)oldp;
2250 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2251 	}
2252 	l = 0;
2253 
2254 	TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
2255 
2256 		if (oldp && d + 1 <= de) {
2257 			memset(d, 0, sizeof(*d));
2258 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2259 			if (sa6_recoverscope(&d->rtaddr)) {
2260 				log(LOG_ERR,
2261 				    "scope error in router list (%s)\n",
2262 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2263 				/* XXX: press on... */
2264 			}
2265 			d->flags = dr->flags;
2266 			d->rtlifetime = dr->rtlifetime;
2267 			d->expire = dr->expire;
2268 			d->if_index = dr->ifp->if_index;
2269 		}
2270 
2271 		l += sizeof(*d);
2272 		if (d)
2273 			d++;
2274 	}
2275 
2276 	if (oldp) {
2277 		if (l > ol)
2278 			error = ENOMEM;
2279 	}
2280 	if (oldlenp)
2281 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2282 
2283 	splx(s);
2284 
2285 	return error;
2286 }
2287 
2288 static int
2289 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2290 {
2291 	int error = 0, s;
2292 	struct nd_prefix *pr;
2293 	struct in6_prefix *p = NULL;
2294 	struct in6_prefix *pe = NULL;
2295 	size_t l;
2296 
2297 	s = splsoftnet();
2298 
2299 	if (oldp) {
2300 		p = (struct in6_prefix *)oldp;
2301 		pe = (struct in6_prefix *)((char *)oldp + *oldlenp);
2302 	}
2303 	l = 0;
2304 
2305 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2306 		u_short advrtrs;
2307 		size_t advance;
2308 		struct sockaddr_in6 *sin6;
2309 		struct sockaddr_in6 *s6;
2310 		struct nd_pfxrouter *pfr;
2311 
2312 		if (oldp && p + 1 <= pe)
2313 		{
2314 			bzero(p, sizeof(*p));
2315 			sin6 = (struct sockaddr_in6 *)(p + 1);
2316 
2317 			p->prefix = pr->ndpr_prefix;
2318 			if (sa6_recoverscope(&p->prefix)) {
2319 				log(LOG_ERR,
2320 				    "scope error in prefix list (%s)\n",
2321 				    ip6_sprintf(&p->prefix.sin6_addr));
2322 				/* XXX: press on... */
2323 			}
2324 			p->raflags = pr->ndpr_raf;
2325 			p->prefixlen = pr->ndpr_plen;
2326 			p->vltime = pr->ndpr_vltime;
2327 			p->pltime = pr->ndpr_pltime;
2328 			p->if_index = pr->ndpr_ifp->if_index;
2329 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2330 				p->expire = 0;
2331 			else {
2332 				time_t maxexpire;
2333 
2334 				/* XXX: we assume time_t is signed. */
2335 				maxexpire = (-1) &
2336 				    ~((time_t)1 <<
2337 				    ((sizeof(maxexpire) * 8) - 1));
2338 				if (pr->ndpr_vltime <
2339 				    maxexpire - pr->ndpr_lastupdate) {
2340 					p->expire = pr->ndpr_lastupdate +
2341 						pr->ndpr_vltime;
2342 				} else
2343 					p->expire = maxexpire;
2344 			}
2345 			p->refcnt = pr->ndpr_refcnt;
2346 			p->flags = pr->ndpr_stateflags;
2347 			p->origin = PR_ORIG_RA;
2348 			advrtrs = 0;
2349 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2350 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2351 					advrtrs++;
2352 					continue;
2353 				}
2354 				s6 = &sin6[advrtrs];
2355 				sockaddr_in6_init(s6, &pfr->router->rtaddr,
2356 				    0, 0, 0);
2357 				if (sa6_recoverscope(s6)) {
2358 					log(LOG_ERR,
2359 					    "scope error in "
2360 					    "prefix list (%s)\n",
2361 					    ip6_sprintf(&pfr->router->rtaddr));
2362 				}
2363 				advrtrs++;
2364 			}
2365 			p->advrtrs = advrtrs;
2366 		}
2367 		else {
2368 			advrtrs = 0;
2369 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2370 				advrtrs++;
2371 		}
2372 
2373 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2374 		l += advance;
2375 		if (p)
2376 			p = (struct in6_prefix *)((char *)p + advance);
2377 	}
2378 
2379 	if (oldp) {
2380 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2381 		if (l > ol)
2382 			error = ENOMEM;
2383 	} else
2384 		*oldlenp = l;
2385 
2386 	splx(s);
2387 
2388 	return error;
2389 }
2390