xref: /netbsd-src/sys/netinet6/nd6.c (revision a6f3f22f245acb8ee3bbf6871d7dce989204fa97)
1 /*	$NetBSD: nd6.c,v 1.177 2015/09/11 10:33:32 roy 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 1.177 2015/09/11 10:33:32 roy Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_net_mpsafe.h"
38 #endif
39 
40 #include "bridge.h"
41 #include "carp.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/ioctl.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/cprng.h>
59 
60 #include <net/if.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/route.h>
64 #include <net/if_ether.h>
65 #include <net/if_fddi.h>
66 #include <net/if_arc.h>
67 
68 #include <netinet/in.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/scope6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/in6_ifattach.h>
75 #include <netinet/icmp6.h>
76 #include <netinet6/icmp6_private.h>
77 
78 #include <net/net_osdep.h>
79 
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 
83 /* timer values */
84 int	nd6_prune	= 1;	/* walk list every 1 seconds */
85 int	nd6_delay	= 5;	/* delay first probe time 5 second */
86 int	nd6_umaxtries	= 3;	/* maximum unicast query */
87 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
88 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
89 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
90 
91 /* preventing too many loops in ND option parsing */
92 int nd6_maxndopt = 10;	/* max # of ND options allowed */
93 
94 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
95 
96 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
97 
98 #ifdef ND6_DEBUG
99 int nd6_debug = 1;
100 #else
101 int nd6_debug = 0;
102 #endif
103 
104 /* for debugging? */
105 static int nd6_inuse, nd6_allocated;
106 
107 struct llinfo_nd6 llinfo_nd6 = {
108 	.ln_prev = &llinfo_nd6,
109 	.ln_next = &llinfo_nd6,
110 };
111 struct nd_drhead nd_defrouter;
112 struct nd_prhead nd_prefix = { 0 };
113 
114 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
115 static const struct sockaddr_in6 all1_sa = {
116 	  .sin6_family = AF_INET6
117 	, .sin6_len = sizeof(struct sockaddr_in6)
118 	, .sin6_addr = {.s6_addr = {0xff, 0xff, 0xff, 0xff,
119 				    0xff, 0xff, 0xff, 0xff,
120 				    0xff, 0xff, 0xff, 0xff,
121 				    0xff, 0xff, 0xff, 0xff}}
122 };
123 
124 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
125 static void nd6_slowtimo(void *);
126 static int regen_tmpaddr(struct in6_ifaddr *);
127 static struct llinfo_nd6 *nd6_free(struct rtentry *, int);
128 static void nd6_llinfo_timer(void *);
129 static void clear_llinfo_pqueue(struct llinfo_nd6 *);
130 
131 callout_t nd6_slowtimo_ch;
132 callout_t nd6_timer_ch;
133 extern callout_t in6_tmpaddrtimer_ch;
134 
135 static int fill_drlist(void *, size_t *, size_t);
136 static int fill_prlist(void *, size_t *, size_t);
137 
138 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
139 
140 #define LN_DEQUEUE(ln) do { \
141 	(ln)->ln_next->ln_prev = (ln)->ln_prev; \
142 	(ln)->ln_prev->ln_next = (ln)->ln_next; \
143 	} while (/*CONSTCOND*/0)
144 #define LN_INSERTHEAD(ln) do { \
145 	(ln)->ln_next = llinfo_nd6.ln_next; \
146 	llinfo_nd6.ln_next = (ln); \
147 	(ln)->ln_prev = &llinfo_nd6; \
148 	(ln)->ln_next->ln_prev = (ln); \
149 	} while (/*CONSTCOND*/0)
150 void
151 nd6_init(void)
152 {
153 	static int nd6_init_done = 0;
154 
155 	if (nd6_init_done) {
156 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
157 		return;
158 	}
159 
160 	/* initialization of the default router list */
161 	TAILQ_INIT(&nd_defrouter);
162 
163 	nd6_init_done = 1;
164 
165 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
166 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
167 
168 	/* start timer */
169 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
170 	    nd6_slowtimo, NULL);
171 }
172 
173 struct nd_ifinfo *
174 nd6_ifattach(struct ifnet *ifp)
175 {
176 	struct nd_ifinfo *nd;
177 
178 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
179 
180 	nd->initialized = 1;
181 
182 	nd->chlim = IPV6_DEFHLIM;
183 	nd->basereachable = REACHABLE_TIME;
184 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
185 	nd->retrans = RETRANS_TIMER;
186 
187 	nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV;
188 
189 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
190 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
191 	 * because one of its members should. */
192 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
193 	    (ifp->if_flags & IFF_LOOPBACK))
194 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
195 
196 	/* A loopback interface does not need to accept RTADV.
197 	 * A bridge interface should not accept RTADV
198 	 * because one of its members should. */
199 	if (ip6_accept_rtadv &&
200 	    !(ifp->if_flags & IFF_LOOPBACK) &&
201 	    !(ifp->if_type != IFT_BRIDGE))
202 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
203 
204 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
205 	nd6_setmtu0(ifp, nd);
206 
207 	return nd;
208 }
209 
210 void
211 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
212 {
213 
214 	nd6_purge(ifp, ext);
215 	free(ext->nd_ifinfo, M_IP6NDP);
216 }
217 
218 void
219 nd6_setmtu(struct ifnet *ifp)
220 {
221 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
222 }
223 
224 void
225 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
226 {
227 	u_int32_t omaxmtu;
228 
229 	omaxmtu = ndi->maxmtu;
230 
231 	switch (ifp->if_type) {
232 	case IFT_ARCNET:
233 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
234 		break;
235 	case IFT_FDDI:
236 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
237 		break;
238 	default:
239 		ndi->maxmtu = ifp->if_mtu;
240 		break;
241 	}
242 
243 	/*
244 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
245 	 * undesirable situation.  We thus notify the operator of the change
246 	 * explicitly.  The check for omaxmtu is necessary to restrict the
247 	 * log to the case of changing the MTU, not initializing it.
248 	 */
249 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
250 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
251 		    " small for IPv6 which needs %lu\n",
252 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
253 		    IPV6_MMTU);
254 	}
255 
256 	if (ndi->maxmtu > in6_maxmtu)
257 		in6_setmaxmtu(); /* check all interfaces just in case */
258 }
259 
260 void
261 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
262 {
263 
264 	memset(ndopts, 0, sizeof(*ndopts));
265 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
266 	ndopts->nd_opts_last
267 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
268 
269 	if (icmp6len == 0) {
270 		ndopts->nd_opts_done = 1;
271 		ndopts->nd_opts_search = NULL;
272 	}
273 }
274 
275 /*
276  * Take one ND option.
277  */
278 struct nd_opt_hdr *
279 nd6_option(union nd_opts *ndopts)
280 {
281 	struct nd_opt_hdr *nd_opt;
282 	int olen;
283 
284 	KASSERT(ndopts != NULL);
285 	KASSERT(ndopts->nd_opts_last != NULL);
286 
287 	if (ndopts->nd_opts_search == NULL)
288 		return NULL;
289 	if (ndopts->nd_opts_done)
290 		return NULL;
291 
292 	nd_opt = ndopts->nd_opts_search;
293 
294 	/* make sure nd_opt_len is inside the buffer */
295 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
296 		memset(ndopts, 0, sizeof(*ndopts));
297 		return NULL;
298 	}
299 
300 	olen = nd_opt->nd_opt_len << 3;
301 	if (olen == 0) {
302 		/*
303 		 * Message validation requires that all included
304 		 * options have a length that is greater than zero.
305 		 */
306 		memset(ndopts, 0, sizeof(*ndopts));
307 		return NULL;
308 	}
309 
310 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
311 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
312 		/* option overruns the end of buffer, invalid */
313 		memset(ndopts, 0, sizeof(*ndopts));
314 		return NULL;
315 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
316 		/* reached the end of options chain */
317 		ndopts->nd_opts_done = 1;
318 		ndopts->nd_opts_search = NULL;
319 	}
320 	return nd_opt;
321 }
322 
323 /*
324  * Parse multiple ND options.
325  * This function is much easier to use, for ND routines that do not need
326  * multiple options of the same type.
327  */
328 int
329 nd6_options(union nd_opts *ndopts)
330 {
331 	struct nd_opt_hdr *nd_opt;
332 	int i = 0;
333 
334 	KASSERT(ndopts != NULL);
335 	KASSERT(ndopts->nd_opts_last != NULL);
336 
337 	if (ndopts->nd_opts_search == NULL)
338 		return 0;
339 
340 	while (1) {
341 		nd_opt = nd6_option(ndopts);
342 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
343 			/*
344 			 * Message validation requires that all included
345 			 * options have a length that is greater than zero.
346 			 */
347 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
348 			memset(ndopts, 0, sizeof(*ndopts));
349 			return -1;
350 		}
351 
352 		if (nd_opt == NULL)
353 			goto skip1;
354 
355 		switch (nd_opt->nd_opt_type) {
356 		case ND_OPT_SOURCE_LINKADDR:
357 		case ND_OPT_TARGET_LINKADDR:
358 		case ND_OPT_MTU:
359 		case ND_OPT_REDIRECTED_HEADER:
360 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
361 				nd6log((LOG_INFO,
362 				    "duplicated ND6 option found (type=%d)\n",
363 				    nd_opt->nd_opt_type));
364 				/* XXX bark? */
365 			} else {
366 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
367 					= nd_opt;
368 			}
369 			break;
370 		case ND_OPT_PREFIX_INFORMATION:
371 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
372 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
373 					= nd_opt;
374 			}
375 			ndopts->nd_opts_pi_end =
376 				(struct nd_opt_prefix_info *)nd_opt;
377 			break;
378 		default:
379 			/*
380 			 * Unknown options must be silently ignored,
381 			 * to accommodate future extension to the protocol.
382 			 */
383 			nd6log((LOG_DEBUG,
384 			    "nd6_options: unsupported option %d - "
385 			    "option ignored\n", nd_opt->nd_opt_type));
386 		}
387 
388 skip1:
389 		i++;
390 		if (i > nd6_maxndopt) {
391 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
392 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
393 			break;
394 		}
395 
396 		if (ndopts->nd_opts_done)
397 			break;
398 	}
399 
400 	return 0;
401 }
402 
403 /*
404  * ND6 timer routine to handle ND6 entries
405  */
406 void
407 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long xtick)
408 {
409 	int s;
410 
411 	s = splsoftnet();
412 
413 	if (xtick < 0) {
414 		ln->ln_expire = 0;
415 		ln->ln_ntick = 0;
416 		callout_stop(&ln->ln_timer_ch);
417 	} else {
418 		ln->ln_expire = time_uptime + xtick / hz;
419 		if (xtick > INT_MAX) {
420 			ln->ln_ntick = xtick - INT_MAX;
421 			callout_reset(&ln->ln_timer_ch, INT_MAX,
422 			    nd6_llinfo_timer, ln);
423 		} else {
424 			ln->ln_ntick = 0;
425 			callout_reset(&ln->ln_timer_ch, xtick,
426 			    nd6_llinfo_timer, ln);
427 		}
428 	}
429 
430 	splx(s);
431 }
432 
433 static void
434 nd6_llinfo_timer(void *arg)
435 {
436 	struct llinfo_nd6 *ln;
437 	struct rtentry *rt;
438 	const struct sockaddr_in6 *dst;
439 	struct ifnet *ifp;
440 	struct nd_ifinfo *ndi = NULL;
441 
442 	mutex_enter(softnet_lock);
443 	KERNEL_LOCK(1, NULL);
444 
445 	ln = (struct llinfo_nd6 *)arg;
446 
447 	if (ln->ln_ntick > 0) {
448 		nd6_llinfo_settimer(ln, ln->ln_ntick);
449 		KERNEL_UNLOCK_ONE(NULL);
450 		mutex_exit(softnet_lock);
451 		return;
452 	}
453 
454 	rt = ln->ln_rt;
455 	KASSERT(rt != NULL);
456 	ifp = rt->rt_ifp;
457 	KASSERT(ifp != NULL);
458 
459 	ndi = ND_IFINFO(ifp);
460 	dst = satocsin6(rt_getkey(rt));
461 
462 	/* sanity check */
463 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
464 		panic("rt_llinfo(%p) is not equal to ln(%p)",
465 		      rt->rt_llinfo, ln);
466 	if (!dst)
467 		panic("dst=0 in nd6_timer(ln=%p)", ln);
468 
469 	switch (ln->ln_state) {
470 	case ND6_LLINFO_INCOMPLETE:
471 		if (ln->ln_asked < nd6_mmaxtries) {
472 			ln->ln_asked++;
473 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
474 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
475 		} else {
476 			struct mbuf *m = ln->ln_hold;
477 			if (m) {
478 				struct mbuf *m0;
479 
480 				/*
481 				 * assuming every packet in ln_hold has
482 				 * the same IP header
483 				 */
484 				m0 = m->m_nextpkt;
485 				m->m_nextpkt = NULL;
486 				icmp6_error2(m, ICMP6_DST_UNREACH,
487 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
488 
489 				ln->ln_hold = m0;
490 				clear_llinfo_pqueue(ln);
491  			}
492 			(void)nd6_free(rt, 0);
493 			ln = NULL;
494 		}
495 		break;
496 	case ND6_LLINFO_REACHABLE:
497 		if (!ND6_LLINFO_PERMANENT(ln)) {
498 			ln->ln_state = ND6_LLINFO_STALE;
499 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
500 		}
501 		break;
502 
503 	case ND6_LLINFO_PURGE:
504 	case ND6_LLINFO_STALE:
505 		/* Garbage Collection(RFC 2461 5.3) */
506 		if (!ND6_LLINFO_PERMANENT(ln)) {
507 			(void)nd6_free(rt, 1);
508 			ln = NULL;
509 		}
510 		break;
511 
512 	case ND6_LLINFO_DELAY:
513 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
514 			/* We need NUD */
515 			ln->ln_asked = 1;
516 			ln->ln_state = ND6_LLINFO_PROBE;
517 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
518 			nd6_ns_output(ifp, &dst->sin6_addr,
519 			    &dst->sin6_addr, ln, 0);
520 		} else {
521 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
522 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
523 		}
524 		break;
525 	case ND6_LLINFO_PROBE:
526 		if (ln->ln_asked < nd6_umaxtries) {
527 			ln->ln_asked++;
528 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
529 			nd6_ns_output(ifp, &dst->sin6_addr,
530 			    &dst->sin6_addr, ln, 0);
531 		} else {
532 			(void)nd6_free(rt, 0);
533 			ln = NULL;
534 		}
535 		break;
536 	}
537 
538 	KERNEL_UNLOCK_ONE(NULL);
539 	mutex_exit(softnet_lock);
540 }
541 
542 /*
543  * ND6 timer routine to expire default route list and prefix list
544  */
545 void
546 nd6_timer(void *ignored_arg)
547 {
548 	struct nd_defrouter *next_dr, *dr;
549 	struct nd_prefix *next_pr, *pr;
550 	struct in6_ifaddr *ia6, *nia6;
551 
552 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
553 	    nd6_timer, NULL);
554 
555 	mutex_enter(softnet_lock);
556 	KERNEL_LOCK(1, NULL);
557 
558 	/* expire default router list */
559 
560 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, next_dr) {
561 		if (dr->expire && dr->expire < time_uptime) {
562 			defrtrlist_del(dr, NULL);
563 		}
564 	}
565 
566 	/*
567 	 * expire interface addresses.
568 	 * in the past the loop was inside prefix expiry processing.
569 	 * However, from a stricter speci-confrmance standpoint, we should
570 	 * rather separate address lifetimes and prefix lifetimes.
571 	 */
572   addrloop:
573 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
574 		nia6 = ia6->ia_next;
575 		/* check address lifetime */
576 		if (IFA6_IS_INVALID(ia6)) {
577 			int regen = 0;
578 
579 			/*
580 			 * If the expiring address is temporary, try
581 			 * regenerating a new one.  This would be useful when
582 			 * we suspended a laptop PC, then turned it on after a
583 			 * period that could invalidate all temporary
584 			 * addresses.  Although we may have to restart the
585 			 * loop (see below), it must be after purging the
586 			 * address.  Otherwise, we'd see an infinite loop of
587 			 * regeneration.
588 			 */
589 			if (ip6_use_tempaddr &&
590 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
591 				if (regen_tmpaddr(ia6) == 0)
592 					regen = 1;
593 			}
594 
595  			in6_purgeaddr(&ia6->ia_ifa);
596 
597 			if (regen)
598 				goto addrloop; /* XXX: see below */
599 		} else if (IFA6_IS_DEPRECATED(ia6)) {
600 			int oldflags = ia6->ia6_flags;
601 
602 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
603 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
604 				rt_newaddrmsg(RTM_NEWADDR,
605 				    (struct ifaddr *)ia6, 0, NULL);
606 			}
607 
608 			/*
609 			 * If a temporary address has just become deprecated,
610 			 * regenerate a new one if possible.
611 			 */
612 			if (ip6_use_tempaddr &&
613 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
614 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
615 
616 				if (regen_tmpaddr(ia6) == 0) {
617 					/*
618 					 * A new temporary address is
619 					 * generated.
620 					 * XXX: this means the address chain
621 					 * has changed while we are still in
622 					 * the loop.  Although the change
623 					 * would not cause disaster (because
624 					 * it's not a deletion, but an
625 					 * addition,) we'd rather restart the
626 					 * loop just for safety.  Or does this
627 					 * significantly reduce performance??
628 					 */
629 					goto addrloop;
630 				}
631 			}
632 		} else {
633 			/*
634 			 * A new RA might have made a deprecated address
635 			 * preferred.
636 			 */
637 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
638 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
639 				rt_newaddrmsg(RTM_NEWADDR,
640 				    (struct ifaddr *)ia6, 0, NULL);
641 			}
642 		}
643 	}
644 
645 	/* expire prefix list */
646 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, next_pr) {
647 		/*
648 		 * check prefix lifetime.
649 		 * since pltime is just for autoconf, pltime processing for
650 		 * prefix is not necessary.
651 		 */
652 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
653 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
654 
655 			/*
656 			 * address expiration and prefix expiration are
657 			 * separate.  NEVER perform in6_purgeaddr here.
658 			 */
659 
660 			prelist_remove(pr);
661 		}
662 	}
663 
664 	KERNEL_UNLOCK_ONE(NULL);
665 	mutex_exit(softnet_lock);
666 }
667 
668 /* ia6: deprecated/invalidated temporary address */
669 static int
670 regen_tmpaddr(struct in6_ifaddr *ia6)
671 {
672 	struct ifaddr *ifa;
673 	struct ifnet *ifp;
674 	struct in6_ifaddr *public_ifa6 = NULL;
675 
676 	ifp = ia6->ia_ifa.ifa_ifp;
677 	IFADDR_FOREACH(ifa, ifp) {
678 		struct in6_ifaddr *it6;
679 
680 		if (ifa->ifa_addr->sa_family != AF_INET6)
681 			continue;
682 
683 		it6 = (struct in6_ifaddr *)ifa;
684 
685 		/* ignore no autoconf addresses. */
686 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
687 			continue;
688 
689 		/* ignore autoconf addresses with different prefixes. */
690 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
691 			continue;
692 
693 		/*
694 		 * Now we are looking at an autoconf address with the same
695 		 * prefix as ours.  If the address is temporary and is still
696 		 * preferred, do not create another one.  It would be rare, but
697 		 * could happen, for example, when we resume a laptop PC after
698 		 * a long period.
699 		 */
700 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
701 		    !IFA6_IS_DEPRECATED(it6)) {
702 			public_ifa6 = NULL;
703 			break;
704 		}
705 
706 		/*
707 		 * This is a public autoconf address that has the same prefix
708 		 * as ours.  If it is preferred, keep it.  We can't break the
709 		 * loop here, because there may be a still-preferred temporary
710 		 * address with the prefix.
711 		 */
712 		if (!IFA6_IS_DEPRECATED(it6))
713 		    public_ifa6 = it6;
714 	}
715 
716 	if (public_ifa6 != NULL) {
717 		int e;
718 
719 		/*
720 		 * Random factor is introduced in the preferred lifetime, so
721 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
722 		 */
723 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
724 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
725 			    " tmp addr, errno=%d\n", e);
726 			return -1;
727 		}
728 		return 0;
729 	}
730 
731 	return -1;
732 }
733 
734 bool
735 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
736 {
737 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
738 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
739 		return true;
740 	case ND6_IFF_ACCEPT_RTADV:
741 		return ip6_accept_rtadv != 0;
742 	case ND6_IFF_OVERRIDE_RTADV:
743 	case 0:
744 	default:
745 		return false;
746 	}
747 }
748 
749 /*
750  * Nuke neighbor cache/prefix/default router management table, right before
751  * ifp goes away.
752  */
753 void
754 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
755 {
756 	struct llinfo_nd6 *ln, *nln;
757 	struct nd_defrouter *dr, *ndr;
758 	struct nd_prefix *pr, *npr;
759 
760 	/*
761 	 * During detach, the ND info might be already removed, but
762 	 * then is explitly passed as argument.
763 	 * Otherwise get it from ifp->if_afdata.
764 	 */
765 	if (ext == NULL)
766 		ext = ifp->if_afdata[AF_INET6];
767 	if (ext == NULL)
768 		return;
769 
770 	/*
771 	 * Nuke default router list entries toward ifp.
772 	 * We defer removal of default router list entries that is installed
773 	 * in the routing table, in order to keep additional side effects as
774 	 * small as possible.
775 	 */
776 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
777 		if (dr->installed)
778 			continue;
779 
780 		if (dr->ifp == ifp) {
781 			KASSERT(ext != NULL);
782 			defrtrlist_del(dr, ext);
783 		}
784 	}
785 
786 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
787 		if (!dr->installed)
788 			continue;
789 
790 		if (dr->ifp == ifp) {
791 			KASSERT(ext != NULL);
792 			defrtrlist_del(dr, ext);
793 		}
794 	}
795 
796 	/* Nuke prefix list entries toward ifp */
797 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, npr) {
798 		if (pr->ndpr_ifp == ifp) {
799 			/*
800 			 * Because if_detach() does *not* release prefixes
801 			 * while purging addresses the reference count will
802 			 * still be above zero. We therefore reset it to
803 			 * make sure that the prefix really gets purged.
804 			 */
805 			pr->ndpr_refcnt = 0;
806 			/*
807 			 * Previously, pr->ndpr_addr is removed as well,
808 			 * but I strongly believe we don't have to do it.
809 			 * nd6_purge() is only called from in6_ifdetach(),
810 			 * which removes all the associated interface addresses
811 			 * by itself.
812 			 * (jinmei@kame.net 20010129)
813 			 */
814 			prelist_remove(pr);
815 		}
816 	}
817 
818 	/* cancel default outgoing interface setting */
819 	if (nd6_defifindex == ifp->if_index)
820 		nd6_setdefaultiface(0);
821 
822 	/* XXX: too restrictive? */
823 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
824 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
825 		if (ndi && nd6_accepts_rtadv(ndi)) {
826 			/* refresh default router list */
827 			defrouter_select();
828 		}
829 	}
830 
831 	/*
832 	 * Nuke neighbor cache entries for the ifp.
833 	 * Note that rt->rt_ifp may not be the same as ifp,
834 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
835 	 * nd6_rtrequest(), and ip6_input().
836 	 */
837 	ln = llinfo_nd6.ln_next;
838 	while (ln != NULL && ln != &llinfo_nd6) {
839 		struct rtentry *rt;
840 		const struct sockaddr_dl *sdl;
841 
842 		nln = ln->ln_next;
843 		rt = ln->ln_rt;
844 		if (rt && rt->rt_gateway &&
845 		    rt->rt_gateway->sa_family == AF_LINK) {
846 			sdl = satocsdl(rt->rt_gateway);
847 			if (sdl->sdl_index == ifp->if_index)
848 				nln = nd6_free(rt, 0);
849 		}
850 		ln = nln;
851 	}
852 }
853 
854 static struct rtentry *
855 nd6_lookup1(const struct in6_addr *addr6, int create, struct ifnet *ifp,
856     int cloning)
857 {
858 	struct rtentry *rt;
859 	struct sockaddr_in6 sin6;
860 
861 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
862 	rt = rtalloc1((struct sockaddr *)&sin6, create);
863 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) == 0) {
864 		/*
865 		 * This is the case for the default route.
866 		 * If we want to create a neighbor cache for the address, we
867 		 * should free the route for the destination and allocate an
868 		 * interface route.
869 		 */
870 		if (create) {
871 			rtfree(rt);
872 			rt = NULL;
873 		}
874 	}
875 	if (rt != NULL)
876 		;
877 	else if (create && ifp) {
878 		int e;
879 
880 		/*
881 		 * If no route is available and create is set,
882 		 * we allocate a host route for the destination
883 		 * and treat it like an interface route.
884 		 * This hack is necessary for a neighbor which can't
885 		 * be covered by our own prefix.
886 		 */
887 		struct ifaddr *ifa =
888 		    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
889 		if (ifa == NULL)
890 			return NULL;
891 
892 		/*
893 		 * Create a new route.  RTF_LLINFO is necessary
894 		 * to create a Neighbor Cache entry for the
895 		 * destination in nd6_rtrequest which will be
896 		 * called in rtrequest via ifa->ifa_rtrequest.
897 		 */
898 		if ((e = rtrequest(RTM_ADD, (const struct sockaddr *)&sin6,
899 		    ifa->ifa_addr, (const struct sockaddr *)&all1_sa,
900 		    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
901 		    ~RTF_CLONING, &rt)) != 0) {
902 #if 0
903 			log(LOG_ERR,
904 			    "nd6_lookup: failed to add route for a "
905 			    "neighbor(%s), errno=%d\n",
906 			    ip6_sprintf(addr6), e);
907 #endif
908 			return NULL;
909 		}
910 		if (rt == NULL)
911 			return NULL;
912 		if (rt->rt_llinfo) {
913 			struct llinfo_nd6 *ln =
914 			    (struct llinfo_nd6 *)rt->rt_llinfo;
915 			ln->ln_state = ND6_LLINFO_NOSTATE;
916 		}
917 	} else
918 		return NULL;
919 
920 	/*
921 	 * Check for a cloning route to match the address.
922 	 * This should only be set from in6_is_addr_neighbor so we avoid
923 	 * a potentially expensive second call to rtalloc1.
924 	 */
925 	if (cloning &&
926 	    rt->rt_flags & (RTF_CLONING | RTF_CLONED) &&
927 	    (rt->rt_ifp == ifp
928 #if NBRIDGE > 0
929 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
930 #endif
931 #if NCARP > 0
932 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
933 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
934 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
935 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
936 #endif
937 	    ))
938 		return rt;
939 
940 	/*
941 	 * Validation for the entry.
942 	 * Note that the check for rt_llinfo is necessary because a cloned
943 	 * route from a parent route that has the L flag (e.g. the default
944 	 * route to a p2p interface) may have the flag, too, while the
945 	 * destination is not actually a neighbor.
946 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
947 	 *      it might be the loopback interface if the entry is for our
948 	 *      own address on a non-loopback interface. Instead, we should
949 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
950 	 *	interface.
951 	 * Note also that ifa_ifp and ifp may differ when we connect two
952 	 * interfaces to a same link, install a link prefix to an interface,
953 	 * and try to install a neighbor cache on an interface that does not
954 	 * have a route to the prefix.
955 	 */
956 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
957 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
958 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
959 		if (create) {
960 			nd6log((LOG_DEBUG,
961 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
962 			    ip6_sprintf(addr6),
963 			    ifp ? if_name(ifp) : "unspec"));
964 		}
965 		rtfree(rt);
966 		return NULL;
967 	}
968 	return rt;
969 }
970 
971 struct rtentry *
972 nd6_lookup(const struct in6_addr *addr6, int create, struct ifnet *ifp)
973 {
974 
975 	return nd6_lookup1(addr6, create, ifp, 0);
976 }
977 
978 /*
979  * Detect if a given IPv6 address identifies a neighbor on a given link.
980  * XXX: should take care of the destination of a p2p link?
981  */
982 int
983 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
984 {
985 	struct nd_prefix *pr;
986 	struct rtentry *rt;
987 
988 	/*
989 	 * A link-local address is always a neighbor.
990 	 * XXX: a link does not necessarily specify a single interface.
991 	 */
992 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
993 		struct sockaddr_in6 sin6_copy;
994 		u_int32_t zone;
995 
996 		/*
997 		 * We need sin6_copy since sa6_recoverscope() may modify the
998 		 * content (XXX).
999 		 */
1000 		sin6_copy = *addr;
1001 		if (sa6_recoverscope(&sin6_copy))
1002 			return 0; /* XXX: should be impossible */
1003 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1004 			return 0;
1005 		if (sin6_copy.sin6_scope_id == zone)
1006 			return 1;
1007 		else
1008 			return 0;
1009 	}
1010 
1011 	/*
1012 	 * If the address matches one of our on-link prefixes, it should be a
1013 	 * neighbor.
1014 	 */
1015 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1016 		if (pr->ndpr_ifp != ifp)
1017 			continue;
1018 
1019 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1020 			continue;
1021 
1022 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1023 		    &addr->sin6_addr, &pr->ndpr_mask))
1024 			return 1;
1025 	}
1026 
1027 	/*
1028 	 * If the default router list is empty, all addresses are regarded
1029 	 * as on-link, and thus, as a neighbor.
1030 	 * XXX: we restrict the condition to hosts, because routers usually do
1031 	 * not have the "default router list".
1032 	 */
1033 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
1034 	    nd6_defifindex == ifp->if_index) {
1035 		return 1;
1036 	}
1037 
1038 	/*
1039 	 * Even if the address matches none of our addresses, it might match
1040 	 * a cloning route or be in the neighbor cache.
1041 	 */
1042 	rt = nd6_lookup1(&addr->sin6_addr, 0, ifp, 1);
1043 	if (rt != NULL) {
1044 		rtfree(rt);
1045 		return 1;
1046 	}
1047 
1048 	return 0;
1049 }
1050 
1051 /*
1052  * Free an nd6 llinfo entry.
1053  * Since the function would cause significant changes in the kernel, DO NOT
1054  * make it global, unless you have a strong reason for the change, and are sure
1055  * that the change is safe.
1056  */
1057 static struct llinfo_nd6 *
1058 nd6_free(struct rtentry *rt, int gc)
1059 {
1060 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1061 	struct in6_addr in6 = satocsin6(rt_getkey(rt))->sin6_addr;
1062 	struct nd_defrouter *dr;
1063 	int error;
1064 
1065 	/*
1066 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1067 	 * even though it is not harmful, it was not really necessary.
1068 	 */
1069 
1070 	/* cancel timer */
1071 	nd6_llinfo_settimer(ln, -1);
1072 
1073 	if (!ip6_forwarding) {
1074 		int s;
1075 		s = splsoftnet();
1076 		dr = defrouter_lookup(&satocsin6(rt_getkey(rt))->sin6_addr,
1077 		    rt->rt_ifp);
1078 
1079 		if (dr != NULL && dr->expire &&
1080 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1081 			/*
1082 			 * If the reason for the deletion is just garbage
1083 			 * collection, and the neighbor is an active default
1084 			 * router, do not delete it.  Instead, reset the GC
1085 			 * timer using the router's lifetime.
1086 			 * Simply deleting the entry would affect default
1087 			 * router selection, which is not necessarily a good
1088 			 * thing, especially when we're using router preference
1089 			 * values.
1090 			 * XXX: the check for ln_state would be redundant,
1091 			 *      but we intentionally keep it just in case.
1092 			 */
1093 			if (dr->expire > time_uptime)
1094 				nd6_llinfo_settimer(ln,
1095 				    (dr->expire - time_uptime) * hz);
1096 			else
1097 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1098 			splx(s);
1099 			return ln->ln_next;
1100 		}
1101 
1102 		if (ln->ln_router || dr) {
1103 			/*
1104 			 * rt6_flush must be called whether or not the neighbor
1105 			 * is in the Default Router List.
1106 			 * See a corresponding comment in nd6_na_input().
1107 			 */
1108 			rt6_flush(&in6, rt->rt_ifp);
1109 		}
1110 
1111 		if (dr) {
1112 			/*
1113 			 * Unreachablity of a router might affect the default
1114 			 * router selection and on-link detection of advertised
1115 			 * prefixes.
1116 			 */
1117 
1118 			/*
1119 			 * Temporarily fake the state to choose a new default
1120 			 * router and to perform on-link determination of
1121 			 * prefixes correctly.
1122 			 * Below the state will be set correctly,
1123 			 * or the entry itself will be deleted.
1124 			 */
1125 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1126 
1127 			/*
1128 			 * Since defrouter_select() does not affect the
1129 			 * on-link determination and MIP6 needs the check
1130 			 * before the default router selection, we perform
1131 			 * the check now.
1132 			 */
1133 			pfxlist_onlink_check();
1134 
1135 			/*
1136 			 * refresh default router list
1137 			 */
1138 			defrouter_select();
1139 		}
1140 		splx(s);
1141 	}
1142 
1143 	/*
1144 	 * Before deleting the entry, remember the next entry as the
1145 	 * return value.  We need this because pfxlist_onlink_check() above
1146 	 * might have freed other entries (particularly the old next entry) as
1147 	 * a side effect (XXX).
1148 	 */
1149 	next = ln->ln_next;
1150 
1151 	/*
1152 	 * Detach the route from the routing tree and the list of neighbor
1153 	 * caches, and disable the route entry not to be used in already
1154 	 * cached routes.
1155 	 */
1156 	error = rtrequest_newmsg(RTM_DELETE, rt_getkey(rt), NULL,
1157 	    rt_mask(rt), 0);
1158 	if (error != 0) {
1159 		/* XXX need error message? */;
1160 	}
1161 
1162 	return next;
1163 }
1164 
1165 /*
1166  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1167  *
1168  * XXX cost-effective methods?
1169  */
1170 void
1171 nd6_nud_hint(struct rtentry *rt)
1172 {
1173 	struct llinfo_nd6 *ln;
1174 
1175 	if (rt == NULL)
1176 		return;
1177 
1178 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1179 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1180 	    !rt->rt_llinfo || !rt->rt_gateway ||
1181 	    rt->rt_gateway->sa_family != AF_LINK) {
1182 		/* This is not a host route. */
1183 		return;
1184 	}
1185 
1186 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1187 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1188 		return;
1189 
1190 	/*
1191 	 * if we get upper-layer reachability confirmation many times,
1192 	 * it is possible we have false information.
1193 	 */
1194 	ln->ln_byhint++;
1195 	if (ln->ln_byhint > nd6_maxnudhint)
1196 		return;
1197 
1198 	ln->ln_state = ND6_LLINFO_REACHABLE;
1199 	if (!ND6_LLINFO_PERMANENT(ln)) {
1200 		nd6_llinfo_settimer(ln,
1201 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1202 	}
1203 
1204 	return;
1205 }
1206 
1207 void
1208 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1209 {
1210 	struct sockaddr *gate = rt->rt_gateway;
1211 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1212 	struct ifnet *ifp = rt->rt_ifp;
1213 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1214 	struct ifaddr *ifa;
1215 
1216 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1217 
1218 	if (req == RTM_LLINFO_UPD) {
1219 		int rc;
1220 		struct in6_addr *in6;
1221 		struct in6_addr in6_all;
1222 		int anycast;
1223 
1224 		if ((ifa = info->rti_ifa) == NULL)
1225 			return;
1226 
1227 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1228 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1229 
1230 		in6_all = in6addr_linklocal_allnodes;
1231 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1232 			log(LOG_ERR, "%s: failed to set scope %s "
1233 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1234 			return;
1235 		}
1236 
1237 		/* XXX don't set Override for proxy addresses */
1238 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1239 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1240 #if 0
1241 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1242 #endif
1243 		    , 1, NULL);
1244 		return;
1245 	}
1246 
1247 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1248 		return;
1249 
1250 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1251 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1252 		/*
1253 		 * This is probably an interface direct route for a link
1254 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1255 		 * We do not need special treatment below for such a route.
1256 		 * Moreover, the RTF_LLINFO flag which would be set below
1257 		 * would annoy the ndp(8) command.
1258 		 */
1259 		return;
1260 	}
1261 
1262 	if (req == RTM_RESOLVE &&
1263 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1264 	     !nd6_is_addr_neighbor(satocsin6(rt_getkey(rt)), ifp))) {
1265 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1266 		/*
1267 		 * FreeBSD and BSD/OS often make a cloned host route based
1268 		 * on a less-specific route (e.g. the default route).
1269 		 * If the less specific route does not have a "gateway"
1270 		 * (this is the case when the route just goes to a p2p or an
1271 		 * stf interface), we'll mistakenly make a neighbor cache for
1272 		 * the host route, and will see strange neighbor solicitation
1273 		 * for the corresponding destination.  In order to avoid the
1274 		 * confusion, we check if the destination of the route is
1275 		 * a neighbor in terms of neighbor discovery, and stop the
1276 		 * process if not.  Additionally, we remove the LLINFO flag
1277 		 * so that ndp(8) will not try to get the neighbor information
1278 		 * of the destination.
1279 		 */
1280 		rt->rt_flags &= ~RTF_LLINFO;
1281 		return;
1282 	}
1283 
1284 	switch (req) {
1285 	case RTM_ADD:
1286 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1287 		/*
1288 		 * There is no backward compatibility :)
1289 		 *
1290 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1291 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1292 		 *	   rt->rt_flags |= RTF_CLONING;
1293 		 */
1294 		if ((rt->rt_flags & RTF_CLONING) ||
1295 		    ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) {
1296 			union {
1297 				struct sockaddr sa;
1298 				struct sockaddr_dl sdl;
1299 				struct sockaddr_storage ss;
1300 			} u;
1301 			/*
1302 			 * Case 1: This route should come from a route to
1303 			 * interface (RTF_CLONING case) or the route should be
1304 			 * treated as on-link but is currently not
1305 			 * (RTF_LLINFO && ln == NULL case).
1306 			 */
1307 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1308 			    ifp->if_index, ifp->if_type,
1309 			    NULL, namelen, NULL, addrlen) == NULL) {
1310 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1311 				    "failed on %s\n", __func__, __LINE__,
1312 				    sizeof(u.ss), if_name(ifp));
1313 			}
1314 			rt_setgate(rt, &u.sa);
1315 			gate = rt->rt_gateway;
1316 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1317 			if (ln != NULL)
1318 				nd6_llinfo_settimer(ln, 0);
1319 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1320 			if ((rt->rt_flags & RTF_CLONING) != 0)
1321 				break;
1322 		}
1323 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1324 		/*
1325 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1326 		 * We don't do that here since llinfo is not ready yet.
1327 		 *
1328 		 * There are also couple of other things to be discussed:
1329 		 * - unsolicited NA code needs improvement beforehand
1330 		 * - RFC2461 says we MAY send multicast unsolicited NA
1331 		 *   (7.2.6 paragraph 4), however, it also says that we
1332 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1333 		 *   we don't have anything like it right now.
1334 		 *   note that the mechanism needs a mutual agreement
1335 		 *   between proxies, which means that we need to implement
1336 		 *   a new protocol, or a new kludge.
1337 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1338 		 *   we need to check ip6forwarding before sending it.
1339 		 *   (or should we allow proxy ND configuration only for
1340 		 *   routers?  there's no mention about proxy ND from hosts)
1341 		 */
1342 #if 0
1343 		/* XXX it does not work */
1344 		if (rt->rt_flags & RTF_ANNOUNCE)
1345 			nd6_na_output(ifp,
1346 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1347 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1348 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1349 			      1, NULL);
1350 #endif
1351 		/* FALLTHROUGH */
1352 	case RTM_RESOLVE:
1353 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1354 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1355 			/*
1356 			 * Address resolution isn't necessary for a point to
1357 			 * point link, so we can skip this test for a p2p link.
1358 			 */
1359 			if (gate->sa_family != AF_LINK ||
1360 			    gate->sa_len <
1361 			    sockaddr_dl_measure(namelen, addrlen)) {
1362 				log(LOG_DEBUG,
1363 				    "nd6_rtrequest: bad gateway value: %s\n",
1364 				    if_name(ifp));
1365 				break;
1366 			}
1367 			satosdl(gate)->sdl_type = ifp->if_type;
1368 			satosdl(gate)->sdl_index = ifp->if_index;
1369 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1370 		}
1371 		if (ln != NULL)
1372 			break;	/* This happens on a route change */
1373 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1374 		/*
1375 		 * Case 2: This route may come from cloning, or a manual route
1376 		 * add with a LL address.
1377 		 */
1378 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1379 		rt->rt_llinfo = ln;
1380 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1381 		if (ln == NULL) {
1382 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1383 			break;
1384 		}
1385 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1386 		nd6_inuse++;
1387 		nd6_allocated++;
1388 		memset(ln, 0, sizeof(*ln));
1389 		ln->ln_rt = rt;
1390 		callout_init(&ln->ln_timer_ch, CALLOUT_MPSAFE);
1391 		/* this is required for "ndp" command. - shin */
1392 		if (req == RTM_ADD) {
1393 		        /*
1394 			 * gate should have some valid AF_LINK entry,
1395 			 * and ln->ln_expire should have some lifetime
1396 			 * which is specified by ndp command.
1397 			 */
1398 			ln->ln_state = ND6_LLINFO_REACHABLE;
1399 			ln->ln_byhint = 0;
1400 		} else {
1401 		        /*
1402 			 * When req == RTM_RESOLVE, rt is created and
1403 			 * initialized in rtrequest(), so rt_expire is 0.
1404 			 */
1405 			ln->ln_state = ND6_LLINFO_NOSTATE;
1406 			nd6_llinfo_settimer(ln, 0);
1407 		}
1408 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1409 		rt->rt_flags |= RTF_LLINFO;
1410 		ln->ln_next = llinfo_nd6.ln_next;
1411 		llinfo_nd6.ln_next = ln;
1412 		ln->ln_prev = &llinfo_nd6;
1413 		ln->ln_next->ln_prev = ln;
1414 
1415 		/*
1416 		 * If we have too many cache entries, initiate immediate
1417 		 * purging for some "less recently used" entries.  Note that
1418 		 * we cannot directly call nd6_free() here because it would
1419 		 * cause re-entering rtable related routines triggering an LOR
1420 		 * problem for FreeBSD.
1421 		 */
1422 		if (ip6_neighborgcthresh >= 0 &&
1423 		    nd6_inuse >= ip6_neighborgcthresh) {
1424 			int i;
1425 
1426 			for (i = 0; i < 10 && llinfo_nd6.ln_prev != ln; i++) {
1427 				struct llinfo_nd6 *ln_end = llinfo_nd6.ln_prev;
1428 
1429 				/* Move this entry to the head */
1430 				LN_DEQUEUE(ln_end);
1431 				LN_INSERTHEAD(ln_end);
1432 
1433 				if (ND6_LLINFO_PERMANENT(ln_end))
1434 					continue;
1435 
1436 				if (ln_end->ln_state > ND6_LLINFO_INCOMPLETE)
1437 					ln_end->ln_state = ND6_LLINFO_STALE;
1438 				else
1439 					ln_end->ln_state = ND6_LLINFO_PURGE;
1440 				nd6_llinfo_settimer(ln_end, 0);
1441 			}
1442 		}
1443 
1444 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1445 		/*
1446 		 * check if rt_getkey(rt) is an address assigned
1447 		 * to the interface.
1448 		 */
1449 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1450 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1451 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1452 		if (ifa != NULL) {
1453 			const void *mac;
1454 			nd6_llinfo_settimer(ln, -1);
1455 			ln->ln_state = ND6_LLINFO_REACHABLE;
1456 			ln->ln_byhint = 0;
1457 			if ((mac = nd6_ifptomac(ifp)) != NULL) {
1458 				/* XXX check for error */
1459 				if (sockaddr_dl_setaddr(satosdl(gate),
1460 				    gate->sa_len, mac,
1461 				    ifp->if_addrlen) == NULL) {
1462 					printf("%s.%d: "
1463 					    "sockaddr_dl_setaddr(, %d, ) "
1464 					    "failed on %s\n", __func__,
1465 					    __LINE__, gate->sa_len,
1466 					    if_name(ifp));
1467 				}
1468 			}
1469 			if (nd6_useloopback) {
1470 				ifp = rt->rt_ifp = lo0ifp;	/* XXX */
1471 				/*
1472 				 * Make sure rt_ifa be equal to the ifaddr
1473 				 * corresponding to the address.
1474 				 * We need this because when we refer
1475 				 * rt_ifa->ia6_flags in ip6_input, we assume
1476 				 * that the rt_ifa points to the address instead
1477 				 * of the loopback address.
1478 				 */
1479 				if (ifa != rt->rt_ifa)
1480 					rt_replace_ifa(rt, ifa);
1481 				rt->rt_rmx.rmx_mtu = 0;
1482 				rt->rt_flags &= ~RTF_CLONED;
1483 			}
1484 			rt->rt_flags |= RTF_LOCAL;
1485 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1486 			nd6_llinfo_settimer(ln, -1);
1487 			ln->ln_state = ND6_LLINFO_REACHABLE;
1488 			ln->ln_byhint = 0;
1489 
1490 			/* join solicited node multicast for proxy ND */
1491 			if (ifp->if_flags & IFF_MULTICAST) {
1492 				struct in6_addr llsol;
1493 				int error;
1494 
1495 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1496 				llsol.s6_addr32[0] = htonl(0xff020000);
1497 				llsol.s6_addr32[1] = 0;
1498 				llsol.s6_addr32[2] = htonl(1);
1499 				llsol.s6_addr8[12] = 0xff;
1500 				if (in6_setscope(&llsol, ifp, NULL))
1501 					break;
1502 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1503 					nd6log((LOG_ERR, "%s: failed to join "
1504 					    "%s (errno=%d)\n", if_name(ifp),
1505 					    ip6_sprintf(&llsol), error));
1506 				}
1507 			}
1508 		}
1509 		break;
1510 
1511 	case RTM_DELETE:
1512 		if (ln == NULL)
1513 			break;
1514 		/* leave from solicited node multicast for proxy ND */
1515 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1516 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1517 			struct in6_addr llsol;
1518 			struct in6_multi *in6m;
1519 
1520 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1521 			llsol.s6_addr32[0] = htonl(0xff020000);
1522 			llsol.s6_addr32[1] = 0;
1523 			llsol.s6_addr32[2] = htonl(1);
1524 			llsol.s6_addr8[12] = 0xff;
1525 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1526 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1527 				if (in6m)
1528 					in6_delmulti(in6m);
1529 			}
1530 		}
1531 		nd6_inuse--;
1532 		ln->ln_next->ln_prev = ln->ln_prev;
1533 		ln->ln_prev->ln_next = ln->ln_next;
1534 		ln->ln_prev = NULL;
1535 		nd6_llinfo_settimer(ln, -1);
1536 		rt->rt_llinfo = 0;
1537 		rt->rt_flags &= ~RTF_LLINFO;
1538 		clear_llinfo_pqueue(ln);
1539 		Free(ln);
1540 	}
1541 }
1542 
1543 int
1544 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1545 {
1546 	struct in6_drlist *drl = (struct in6_drlist *)data;
1547 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1548 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1549 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1550 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1551 	struct nd_defrouter *dr;
1552 	struct nd_prefix *pr;
1553 	int i = 0, error = 0;
1554 	int s;
1555 
1556 	switch (cmd) {
1557 	case SIOCGDRLST_IN6:
1558 		/*
1559 		 * obsolete API, use sysctl under net.inet6.icmp6
1560 		 */
1561 		memset(drl, 0, sizeof(*drl));
1562 		s = splsoftnet();
1563 		TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
1564 			if (i >= DRLSTSIZ)
1565 				break;
1566 			drl->defrouter[i].rtaddr = dr->rtaddr;
1567 			in6_clearscope(&drl->defrouter[i].rtaddr);
1568 
1569 			drl->defrouter[i].flags = dr->flags;
1570 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1571 			drl->defrouter[i].expire = dr->expire ?
1572 			    time_mono_to_wall(dr->expire) : 0;
1573 			drl->defrouter[i].if_index = dr->ifp->if_index;
1574 			i++;
1575 		}
1576 		splx(s);
1577 		break;
1578 	case SIOCGPRLST_IN6:
1579 		/*
1580 		 * obsolete API, use sysctl under net.inet6.icmp6
1581 		 *
1582 		 * XXX the structure in6_prlist was changed in backward-
1583 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1584 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1585 		 */
1586 		/*
1587 		 * XXX meaning of fields, especialy "raflags", is very
1588 		 * differnet between RA prefix list and RR/static prefix list.
1589 		 * how about separating ioctls into two?
1590 		 */
1591 		memset(oprl, 0, sizeof(*oprl));
1592 		s = splsoftnet();
1593 		LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1594 			struct nd_pfxrouter *pfr;
1595 			int j;
1596 
1597 			if (i >= PRLSTSIZ)
1598 				break;
1599 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1600 			oprl->prefix[i].raflags = pr->ndpr_raf;
1601 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1602 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1603 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1604 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1605 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1606 				oprl->prefix[i].expire = 0;
1607 			else {
1608 				time_t maxexpire;
1609 
1610 				/* XXX: we assume time_t is signed. */
1611 				maxexpire = (-1) &
1612 				    ~((time_t)1 <<
1613 				    ((sizeof(maxexpire) * 8) - 1));
1614 				if (pr->ndpr_vltime <
1615 				    maxexpire - pr->ndpr_lastupdate) {
1616 					time_t expire;
1617 					expire = pr->ndpr_lastupdate +
1618 					    pr->ndpr_vltime;
1619 					oprl->prefix[i].expire = expire ?
1620 					    time_mono_to_wall(expire) : 0;
1621 				} else
1622 					oprl->prefix[i].expire = maxexpire;
1623 			}
1624 
1625 			j = 0;
1626 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1627 				if (j < DRLSTSIZ) {
1628 #define RTRADDR oprl->prefix[i].advrtr[j]
1629 					RTRADDR = pfr->router->rtaddr;
1630 					in6_clearscope(&RTRADDR);
1631 #undef RTRADDR
1632 				}
1633 				j++;
1634 			}
1635 			oprl->prefix[i].advrtrs = j;
1636 			oprl->prefix[i].origin = PR_ORIG_RA;
1637 
1638 			i++;
1639 		}
1640 		splx(s);
1641 
1642 		break;
1643 	case OSIOCGIFINFO_IN6:
1644 #define ND	ndi->ndi
1645 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1646 		memset(&ND, 0, sizeof(ND));
1647 		ND.linkmtu = IN6_LINKMTU(ifp);
1648 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1649 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1650 		ND.reachable = ND_IFINFO(ifp)->reachable;
1651 		ND.retrans = ND_IFINFO(ifp)->retrans;
1652 		ND.flags = ND_IFINFO(ifp)->flags;
1653 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1654 		ND.chlim = ND_IFINFO(ifp)->chlim;
1655 		break;
1656 	case SIOCGIFINFO_IN6:
1657 		ND = *ND_IFINFO(ifp);
1658 		break;
1659 	case SIOCSIFINFO_IN6:
1660 		/*
1661 		 * used to change host variables from userland.
1662 		 * intented for a use on router to reflect RA configurations.
1663 		 */
1664 		/* 0 means 'unspecified' */
1665 		if (ND.linkmtu != 0) {
1666 			if (ND.linkmtu < IPV6_MMTU ||
1667 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1668 				error = EINVAL;
1669 				break;
1670 			}
1671 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1672 		}
1673 
1674 		if (ND.basereachable != 0) {
1675 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1676 
1677 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1678 			if (ND.basereachable != obasereachable)
1679 				ND_IFINFO(ifp)->reachable =
1680 				    ND_COMPUTE_RTIME(ND.basereachable);
1681 		}
1682 		if (ND.retrans != 0)
1683 			ND_IFINFO(ifp)->retrans = ND.retrans;
1684 		if (ND.chlim != 0)
1685 			ND_IFINFO(ifp)->chlim = ND.chlim;
1686 		/* FALLTHROUGH */
1687 	case SIOCSIFINFO_FLAGS:
1688 	{
1689 		struct ifaddr *ifa;
1690 		struct in6_ifaddr *ia;
1691 
1692 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1693 		    !(ND.flags & ND6_IFF_IFDISABLED))
1694 		{
1695 			/*
1696 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1697 			 * has a link-local address with IN6_IFF_DUPLICATED,
1698 			 * do not clear ND6_IFF_IFDISABLED.
1699 			 * See RFC 4862, section 5.4.5.
1700 			 */
1701 			int duplicated_linklocal = 0;
1702 
1703 			IFADDR_FOREACH(ifa, ifp) {
1704 				if (ifa->ifa_addr->sa_family != AF_INET6)
1705 					continue;
1706 				ia = (struct in6_ifaddr *)ifa;
1707 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1708 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1709 				{
1710 					duplicated_linklocal = 1;
1711 					break;
1712 				}
1713 			}
1714 
1715 			if (duplicated_linklocal) {
1716 				ND.flags |= ND6_IFF_IFDISABLED;
1717 				log(LOG_ERR, "Cannot enable an interface"
1718 				    " with a link-local address marked"
1719 				    " duplicate.\n");
1720 			} else {
1721 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1722 				if (ifp->if_flags & IFF_UP)
1723 					in6_if_up(ifp);
1724 			}
1725 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1726 		    (ND.flags & ND6_IFF_IFDISABLED))
1727 		{
1728 			/* Mark all IPv6 addresses as tentative. */
1729 
1730 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1731 			IFADDR_FOREACH(ifa, ifp) {
1732 				if (ifa->ifa_addr->sa_family != AF_INET6)
1733 					continue;
1734 				nd6_dad_stop(ifa);
1735 				ia = (struct in6_ifaddr *)ifa;
1736 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1737 			}
1738 		}
1739 
1740 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1741 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1742 				/* auto_linklocal 0->1 transition */
1743 
1744 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1745 				in6_ifattach(ifp, NULL);
1746 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1747 			    ifp->if_flags & IFF_UP)
1748 			{
1749 				/*
1750 				 * When the IF already has
1751 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1752 				 * address is assigned, and IFF_UP, try to
1753 				 * assign one.
1754 				 */
1755 				 int haslinklocal = 0;
1756 
1757 				 IFADDR_FOREACH(ifa, ifp) {
1758 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1759 						continue;
1760 					ia = (struct in6_ifaddr *)ifa;
1761 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1762 						haslinklocal = 1;
1763 						break;
1764 					}
1765 				 }
1766 				 if (!haslinklocal)
1767 					in6_ifattach(ifp, NULL);
1768 			}
1769 		}
1770 	}
1771 		ND_IFINFO(ifp)->flags = ND.flags;
1772 		break;
1773 #undef ND
1774 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1775 		/* sync kernel routing table with the default router list */
1776 		defrouter_reset();
1777 		defrouter_select();
1778 		break;
1779 	case SIOCSPFXFLUSH_IN6:
1780 	{
1781 		/* flush all the prefix advertised by routers */
1782 		struct nd_prefix *pfx, *next;
1783 
1784 		s = splsoftnet();
1785 		LIST_FOREACH_SAFE(pfx, &nd_prefix, ndpr_entry, next) {
1786 			struct in6_ifaddr *ia, *ia_next;
1787 
1788 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1789 				continue; /* XXX */
1790 
1791 			/* do we really have to remove addresses as well? */
1792 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1793 				/* ia might be removed.  keep the next ptr. */
1794 				ia_next = ia->ia_next;
1795 
1796 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1797 					continue;
1798 
1799 				if (ia->ia6_ndpr == pfx)
1800 					in6_purgeaddr(&ia->ia_ifa);
1801 			}
1802 			prelist_remove(pfx);
1803 		}
1804 		splx(s);
1805 		break;
1806 	}
1807 	case SIOCSRTRFLUSH_IN6:
1808 	{
1809 		/* flush all the default routers */
1810 		struct nd_defrouter *drtr, *next;
1811 
1812 		s = splsoftnet();
1813 		defrouter_reset();
1814 		TAILQ_FOREACH_SAFE(drtr, &nd_defrouter, dr_entry, next) {
1815 			defrtrlist_del(drtr, NULL);
1816 		}
1817 		defrouter_select();
1818 		splx(s);
1819 		break;
1820 	}
1821 	case SIOCGNBRINFO_IN6:
1822 	{
1823 		struct llinfo_nd6 *ln;
1824 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1825 		struct rtentry *rt;
1826 
1827 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1828 			return error;
1829 
1830 		s = splsoftnet();
1831 		rt = nd6_lookup(&nb_addr, 0, ifp);
1832 		if (rt == NULL) {
1833 			error = EINVAL;
1834 			splx(s);
1835 			break;
1836 		}
1837 
1838 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1839 		rtfree(rt);
1840 		if (ln == NULL) {
1841 			error = EINVAL;
1842 			splx(s);
1843 			break;
1844 		}
1845 		nbi->state = ln->ln_state;
1846 		nbi->asked = ln->ln_asked;
1847 		nbi->isrouter = ln->ln_router;
1848 		nbi->expire = ln->ln_expire ?
1849 		    time_mono_to_wall(ln->ln_expire) : 0;
1850 		splx(s);
1851 
1852 		break;
1853 	}
1854 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1855 		ndif->ifindex = nd6_defifindex;
1856 		break;
1857 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1858 		return nd6_setdefaultiface(ndif->ifindex);
1859 	}
1860 	return error;
1861 }
1862 
1863 void
1864 nd6_llinfo_release_pkts(struct llinfo_nd6 *ln, struct ifnet *ifp,
1865     struct rtentry *rt)
1866 {
1867 	struct mbuf *m_hold, *m_hold_next;
1868 
1869 	for (m_hold = ln->ln_hold, ln->ln_hold = NULL;
1870 	     m_hold != NULL;
1871 	     m_hold = m_hold_next) {
1872 		m_hold_next = m_hold->m_nextpkt;
1873 		m_hold->m_nextpkt = NULL;
1874 
1875 		/*
1876 		 * we assume ifp is not a p2p here, so
1877 		 * just set the 2nd argument as the
1878 		 * 1st one.
1879 		 */
1880 		nd6_output(ifp, ifp, m_hold, satocsin6(rt_getkey(rt)), rt);
1881 	}
1882 }
1883 
1884 /*
1885  * Create neighbor cache entry and cache link-layer address,
1886  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1887  */
1888 void
1889 nd6_cache_lladdr(
1890     struct ifnet *ifp,
1891     struct in6_addr *from,
1892     char *lladdr,
1893     int lladdrlen,
1894     int type,	/* ICMP6 type */
1895     int code	/* type dependent information */
1896 )
1897 {
1898 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
1899 	struct rtentry *rt = NULL;
1900 	struct llinfo_nd6 *ln = NULL;
1901 	int is_newentry;
1902 	struct sockaddr_dl *sdl = NULL;
1903 	int do_update;
1904 	int olladdr;
1905 	int llchange;
1906 	int newstate = 0;
1907 
1908 	KASSERT(ifp != NULL);
1909 	KASSERT(from != NULL);
1910 
1911 	/* nothing must be updated for unspecified address */
1912 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1913 		return;
1914 
1915 	/*
1916 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1917 	 * the caller.
1918 	 *
1919 	 * XXX If the link does not have link-layer adderss, what should
1920 	 * we do? (ifp->if_addrlen == 0)
1921 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1922 	 * description on it in NS section (RFC 2461 7.2.3).
1923 	 */
1924 
1925 	rt = nd6_lookup(from, 0, ifp);
1926 	if (rt == NULL) {
1927 #if 0
1928 		/* nothing must be done if there's no lladdr */
1929 		if (!lladdr || !lladdrlen)
1930 			return NULL;
1931 #endif
1932 
1933 		rt = nd6_lookup(from, 1, ifp);
1934 		is_newentry = 1;
1935 	} else {
1936 		/* do nothing if static ndp is set */
1937 		if (rt->rt_flags & RTF_STATIC) {
1938 			rtfree(rt);
1939 			return;
1940 		}
1941 		is_newentry = 0;
1942 	}
1943 
1944 	if (rt == NULL)
1945 		return;
1946 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1947 fail:
1948 		(void)nd6_free(rt, 0);
1949 		rtfree(rt);
1950 		return;
1951 	}
1952 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1953 	if (ln == NULL)
1954 		goto fail;
1955 	if (rt->rt_gateway == NULL)
1956 		goto fail;
1957 	if (rt->rt_gateway->sa_family != AF_LINK)
1958 		goto fail;
1959 	sdl = satosdl(rt->rt_gateway);
1960 
1961 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1962 	if (olladdr && lladdr) {
1963 		if (memcmp(lladdr, CLLADDR(sdl), ifp->if_addrlen))
1964 			llchange = 1;
1965 		else
1966 			llchange = 0;
1967 	} else
1968 		llchange = 0;
1969 
1970 	/*
1971 	 * newentry olladdr  lladdr  llchange	(*=record)
1972 	 *	0	n	n	--	(1)
1973 	 *	0	y	n	--	(2)
1974 	 *	0	n	y	--	(3) * STALE
1975 	 *	0	y	y	n	(4) *
1976 	 *	0	y	y	y	(5) * STALE
1977 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1978 	 *	1	--	y	--	(7) * STALE
1979 	 */
1980 
1981 	if (lladdr) {		/* (3-5) and (7) */
1982 		/*
1983 		 * Record source link-layer address
1984 		 * XXX is it dependent to ifp->if_type?
1985 		 */
1986 		/* XXX check for error */
1987 		if (sockaddr_dl_setaddr(sdl, sdl->sdl_len, lladdr,
1988 		    ifp->if_addrlen) == NULL) {
1989 			printf("%s.%d: sockaddr_dl_setaddr(, %d, ) "
1990 			    "failed on %s\n", __func__, __LINE__,
1991 			    sdl->sdl_len, if_name(ifp));
1992 		}
1993 	}
1994 
1995 	if (!is_newentry) {
1996 		if ((!olladdr && lladdr) ||		/* (3) */
1997 		    (olladdr && lladdr && llchange)) {	/* (5) */
1998 			do_update = 1;
1999 			newstate = ND6_LLINFO_STALE;
2000 		} else					/* (1-2,4) */
2001 			do_update = 0;
2002 	} else {
2003 		do_update = 1;
2004 		if (lladdr == NULL)			/* (6) */
2005 			newstate = ND6_LLINFO_NOSTATE;
2006 		else					/* (7) */
2007 			newstate = ND6_LLINFO_STALE;
2008 	}
2009 
2010 	if (do_update) {
2011 		/*
2012 		 * Update the state of the neighbor cache.
2013 		 */
2014 		ln->ln_state = newstate;
2015 
2016 		if (ln->ln_state == ND6_LLINFO_STALE) {
2017 			/*
2018 			 * XXX: since nd6_output() below will cause
2019 			 * state tansition to DELAY and reset the timer,
2020 			 * we must set the timer now, although it is actually
2021 			 * meaningless.
2022 			 */
2023 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2024 
2025 			nd6_llinfo_release_pkts(ln, ifp, rt);
2026 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2027 			/* probe right away */
2028 			nd6_llinfo_settimer((void *)ln, 0);
2029 		}
2030 	}
2031 
2032 	/*
2033 	 * ICMP6 type dependent behavior.
2034 	 *
2035 	 * NS: clear IsRouter if new entry
2036 	 * RS: clear IsRouter
2037 	 * RA: set IsRouter if there's lladdr
2038 	 * redir: clear IsRouter if new entry
2039 	 *
2040 	 * RA case, (1):
2041 	 * The spec says that we must set IsRouter in the following cases:
2042 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2043 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2044 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2045 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2046 	 * neighbor cache, this is similar to (6).
2047 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2048 	 *
2049 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2050 	 *							D R
2051 	 *	0	n	n	--	(1)	c   ?     s
2052 	 *	0	y	n	--	(2)	c   s     s
2053 	 *	0	n	y	--	(3)	c   s     s
2054 	 *	0	y	y	n	(4)	c   s     s
2055 	 *	0	y	y	y	(5)	c   s     s
2056 	 *	1	--	n	--	(6) c	c 	c s
2057 	 *	1	--	y	--	(7) c	c   s	c s
2058 	 *
2059 	 *					(c=clear s=set)
2060 	 */
2061 	switch (type & 0xff) {
2062 	case ND_NEIGHBOR_SOLICIT:
2063 		/*
2064 		 * New entry must have is_router flag cleared.
2065 		 */
2066 		if (is_newentry)	/* (6-7) */
2067 			ln->ln_router = 0;
2068 		break;
2069 	case ND_REDIRECT:
2070 		/*
2071 		 * If the icmp is a redirect to a better router, always set the
2072 		 * is_router flag.  Otherwise, if the entry is newly created,
2073 		 * clear the flag.  [RFC 2461, sec 8.3]
2074 		 */
2075 		if (code == ND_REDIRECT_ROUTER)
2076 			ln->ln_router = 1;
2077 		else if (is_newentry) /* (6-7) */
2078 			ln->ln_router = 0;
2079 		break;
2080 	case ND_ROUTER_SOLICIT:
2081 		/*
2082 		 * is_router flag must always be cleared.
2083 		 */
2084 		ln->ln_router = 0;
2085 		break;
2086 	case ND_ROUTER_ADVERT:
2087 		/*
2088 		 * Mark an entry with lladdr as a router.
2089 		 */
2090 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2091 		    (is_newentry && lladdr)) {			/* (7) */
2092 			ln->ln_router = 1;
2093 		}
2094 		break;
2095 	}
2096 
2097 	if (do_update)
2098 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2099 
2100 	/*
2101 	 * When the link-layer address of a router changes, select the
2102 	 * best router again.  In particular, when the neighbor entry is newly
2103 	 * created, it might affect the selection policy.
2104 	 * Question: can we restrict the first condition to the "is_newentry"
2105 	 * case?
2106 	 * XXX: when we hear an RA from a new router with the link-layer
2107 	 * address option, defrouter_select() is called twice, since
2108 	 * defrtrlist_update called the function as well.  However, I believe
2109 	 * we can compromise the overhead, since it only happens the first
2110 	 * time.
2111 	 * XXX: although defrouter_select() should not have a bad effect
2112 	 * for those are not autoconfigured hosts, we explicitly avoid such
2113 	 * cases for safety.
2114 	 */
2115 	if (do_update && ln->ln_router && !ip6_forwarding &&
2116 	    nd6_accepts_rtadv(ndi))
2117 		defrouter_select();
2118 
2119 	rtfree(rt);
2120 }
2121 
2122 static void
2123 nd6_slowtimo(void *ignored_arg)
2124 {
2125 	struct nd_ifinfo *nd6if;
2126 	struct ifnet *ifp;
2127 
2128 	mutex_enter(softnet_lock);
2129 	KERNEL_LOCK(1, NULL);
2130 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2131 	    nd6_slowtimo, NULL);
2132 	IFNET_FOREACH(ifp) {
2133 		nd6if = ND_IFINFO(ifp);
2134 		if (nd6if->basereachable && /* already initialized */
2135 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2136 			/*
2137 			 * Since reachable time rarely changes by router
2138 			 * advertisements, we SHOULD insure that a new random
2139 			 * value gets recomputed at least once every few hours.
2140 			 * (RFC 2461, 6.3.4)
2141 			 */
2142 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2143 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2144 		}
2145 	}
2146 	KERNEL_UNLOCK_ONE(NULL);
2147 	mutex_exit(softnet_lock);
2148 }
2149 
2150 /*
2151  * Next hop determination.  This routine was derived from ether_output.
2152  */
2153 static int
2154 nd6_determine_nexthop(struct ifnet *ifp, const struct sockaddr_in6 *dst,
2155     struct rtentry *rt00, struct rtentry **ret_rt, bool *sendpkt)
2156 {
2157 	struct rtentry *rt, *rt0;
2158 	struct rtentry *gwrt;
2159 	struct sockaddr_in6 *gw6;
2160 
2161 #define RTFREE_IF_NEEDED(_rt) \
2162 	if ((_rt) != NULL && (_rt) != rt00) \
2163 		rtfree((_rt));
2164 
2165 	KASSERT(rt00 != NULL);
2166 
2167 	rt = rt0 = rt00;
2168 
2169 	if ((rt->rt_flags & RTF_UP) == 0) {
2170 		rt0 = rt = rtalloc1(sin6tocsa(dst), 1);
2171 		if (rt == NULL)
2172 			goto hostunreach;
2173 		if (rt->rt_ifp != ifp)
2174 			goto hostunreach;
2175 	}
2176 
2177 	if ((rt->rt_flags & RTF_GATEWAY) == 0)
2178 		goto out;
2179 
2180 	gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
2181 
2182 	/*
2183 	 * We skip link-layer address resolution and NUD
2184 	 * if the gateway is not a neighbor from ND point
2185 	 * of view, regardless of the value of nd_ifinfo.flags.
2186 	 * The second condition is a bit tricky; we skip
2187 	 * if the gateway is our own address, which is
2188 	 * sometimes used to install a route to a p2p link.
2189 	 */
2190 	if (!nd6_is_addr_neighbor(gw6, ifp) ||
2191 	    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
2192 		/*
2193 		 * We allow this kind of tricky route only
2194 		 * when the outgoing interface is p2p.
2195 		 * XXX: we may need a more generic rule here.
2196 		 */
2197 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
2198 			goto hostunreach;
2199 
2200 		*sendpkt = true;
2201 		goto out;
2202 	}
2203 
2204 	/* Try to use a cached nexthop route (gwroute) if exists */
2205 	gwrt = rt_get_gwroute(rt);
2206 	if (gwrt == NULL)
2207 		goto lookup;
2208 
2209 	RTFREE_IF_NEEDED(rt);
2210 	rt = gwrt;
2211 	if ((rt->rt_flags & RTF_UP) == 0) {
2212 		RTFREE_IF_NEEDED(rt);
2213 		rt = rt0;
2214 	lookup:
2215 		/* Look up a nexthop route */
2216 		gwrt = rtalloc1(rt->rt_gateway, 1);
2217 		rt_set_gwroute(rt, gwrt);
2218 		RTFREE_IF_NEEDED(rt);
2219 		rt = gwrt;
2220 		if (rt == NULL)
2221 			goto hostunreach;
2222 		/* the "G" test below also prevents rt == rt0 */
2223 		if ((rt->rt_flags & RTF_GATEWAY) ||
2224 		    (rt->rt_ifp != ifp)) {
2225 			if (rt0->rt_gwroute != NULL)
2226 				rtfree(rt0->rt_gwroute);
2227 			rt0->rt_gwroute = NULL;
2228 			goto hostunreach;
2229 		}
2230 	}
2231 
2232 out:
2233 	*ret_rt = rt;
2234 	return 0;
2235 
2236 hostunreach:
2237 	RTFREE_IF_NEEDED(rt);
2238 
2239 	return EHOSTUNREACH;
2240 #undef RTFREE_IF_NEEDED
2241 }
2242 
2243 #define senderr(e) { error = (e); goto bad;}
2244 int
2245 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
2246     const struct sockaddr_in6 *dst, struct rtentry *rt0)
2247 {
2248 	struct mbuf *m = m0;
2249 	struct rtentry *rt = rt0;
2250 	struct llinfo_nd6 *ln = NULL;
2251 	int error = 0;
2252 
2253 #define RTFREE_IF_NEEDED(_rt) \
2254 	if ((_rt) != NULL && (_rt) != rt0) \
2255 		rtfree((_rt));
2256 
2257 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2258 		goto sendpkt;
2259 
2260 	if (nd6_need_cache(ifp) == 0)
2261 		goto sendpkt;
2262 
2263 	if (rt) {
2264 		struct rtentry *nexthop = NULL;
2265 		bool sendpkt = false;
2266 
2267 		error = nd6_determine_nexthop(ifp, dst, rt, &nexthop, &sendpkt);
2268 		if (error != 0)
2269 			senderr(error);
2270 		rt = nexthop;
2271 		if (sendpkt)
2272 			goto sendpkt;
2273 	}
2274 
2275 	/*
2276 	 * Address resolution or Neighbor Unreachability Detection
2277 	 * for the next hop.
2278 	 * At this point, the destination of the packet must be a unicast
2279 	 * or an anycast address(i.e. not a multicast).
2280 	 */
2281 
2282 	/* Look up the neighbor cache for the nexthop */
2283 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) != 0)
2284 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2285 	else {
2286 		/*
2287 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2288 		 * the condition below is not very efficient.  But we believe
2289 		 * it is tolerable, because this should be a rare case.
2290 		 */
2291 		if (nd6_is_addr_neighbor(dst, ifp)) {
2292 			RTFREE_IF_NEEDED(rt);
2293 			rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
2294 			if (rt != NULL)
2295 				ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2296 		}
2297 	}
2298 	if (ln == NULL || rt == NULL) {
2299 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2300 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2301 			log(LOG_DEBUG,
2302 			    "nd6_output: can't allocate llinfo for %s "
2303 			    "(ln=%p, rt=%p)\n",
2304 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2305 			senderr(EIO);	/* XXX: good error? */
2306 		}
2307 
2308 		goto sendpkt;	/* send anyway */
2309 	}
2310 
2311 	/*
2312 	 * Move this entry to the head of the queue so that it is less likely
2313 	 * for this entry to be a target of forced garbage collection (see
2314 	 * nd6_rtrequest()).
2315 	 */
2316 	LN_DEQUEUE(ln);
2317 	LN_INSERTHEAD(ln);
2318 
2319 	/* We don't have to do link-layer address resolution on a p2p link. */
2320 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2321 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2322 		ln->ln_state = ND6_LLINFO_STALE;
2323 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2324 	}
2325 
2326 	/*
2327 	 * The first time we send a packet to a neighbor whose entry is
2328 	 * STALE, we have to change the state to DELAY and a sets a timer to
2329 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2330 	 * neighbor unreachability detection on expiration.
2331 	 * (RFC 2461 7.3.3)
2332 	 */
2333 	if (ln->ln_state == ND6_LLINFO_STALE) {
2334 		ln->ln_asked = 0;
2335 		ln->ln_state = ND6_LLINFO_DELAY;
2336 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2337 	}
2338 
2339 	/*
2340 	 * If the neighbor cache entry has a state other than INCOMPLETE
2341 	 * (i.e. its link-layer address is already resolved), just
2342 	 * send the packet.
2343 	 */
2344 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2345 		goto sendpkt;
2346 
2347 	/*
2348 	 * There is a neighbor cache entry, but no ethernet address
2349 	 * response yet.  Append this latest packet to the end of the
2350 	 * packet queue in the mbuf, unless the number of the packet
2351 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2352 	 * the oldest packet in the queue will be removed.
2353 	 */
2354 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2355 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2356 	if (ln->ln_hold) {
2357 		struct mbuf *m_hold;
2358 		int i;
2359 
2360 		i = 0;
2361 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2362 			i++;
2363 			if (m_hold->m_nextpkt == NULL) {
2364 				m_hold->m_nextpkt = m;
2365 				break;
2366 			}
2367 		}
2368 		while (i >= nd6_maxqueuelen) {
2369 			m_hold = ln->ln_hold;
2370 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2371 			m_freem(m_hold);
2372 			i--;
2373 		}
2374 	} else {
2375 		ln->ln_hold = m;
2376 	}
2377 
2378 	/*
2379 	 * If there has been no NS for the neighbor after entering the
2380 	 * INCOMPLETE state, send the first solicitation.
2381 	 */
2382 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2383 		ln->ln_asked++;
2384 		nd6_llinfo_settimer(ln,
2385 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2386 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2387 	}
2388 	error = 0;
2389 	goto exit;
2390 
2391   sendpkt:
2392 	/* discard the packet if IPv6 operation is disabled on the interface */
2393 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2394 		error = ENETDOWN; /* better error? */
2395 		goto bad;
2396 	}
2397 
2398 #ifndef NET_MPSAFE
2399 	KERNEL_LOCK(1, NULL);
2400 #endif
2401 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2402 		error = (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt);
2403 	else
2404 		error = (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt);
2405 #ifndef NET_MPSAFE
2406 	KERNEL_UNLOCK_ONE(NULL);
2407 #endif
2408 	goto exit;
2409 
2410   bad:
2411 	if (m != NULL)
2412 		m_freem(m);
2413   exit:
2414 	RTFREE_IF_NEEDED(rt);
2415 
2416 	return error;
2417 #undef RTFREE_IF_NEEDED
2418 }
2419 #undef senderr
2420 
2421 int
2422 nd6_need_cache(struct ifnet *ifp)
2423 {
2424 	/*
2425 	 * XXX: we currently do not make neighbor cache on any interface
2426 	 * other than ARCnet, Ethernet, FDDI and GIF.
2427 	 *
2428 	 * RFC2893 says:
2429 	 * - unidirectional tunnels needs no ND
2430 	 */
2431 	switch (ifp->if_type) {
2432 	case IFT_ARCNET:
2433 	case IFT_ETHER:
2434 	case IFT_FDDI:
2435 	case IFT_IEEE1394:
2436 	case IFT_CARP:
2437 	case IFT_GIF:		/* XXX need more cases? */
2438 	case IFT_PPP:
2439 	case IFT_TUNNEL:
2440 		return 1;
2441 	default:
2442 		return 0;
2443 	}
2444 }
2445 
2446 int
2447 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2448     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2449     size_t dstsize)
2450 {
2451 	const struct sockaddr_dl *sdl;
2452 
2453 	if (m->m_flags & M_MCAST) {
2454 		switch (ifp->if_type) {
2455 		case IFT_ETHER:
2456 		case IFT_FDDI:
2457 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2458 			    lldst);
2459 			return 1;
2460 		case IFT_IEEE1394:
2461 			memcpy(lldst, ifp->if_broadcastaddr,
2462 			    MIN(dstsize, ifp->if_addrlen));
2463 			return 1;
2464 		case IFT_ARCNET:
2465 			*lldst = 0;
2466 			return 1;
2467 		default:
2468 			m_freem(m);
2469 			return 0;
2470 		}
2471 	}
2472 
2473 	if (rt == NULL) {
2474 		/* this could happen, if we could not allocate memory */
2475 		m_freem(m);
2476 		return 0;
2477 	}
2478 	if (rt->rt_gateway->sa_family != AF_LINK) {
2479 		char gbuf[256];
2480 		char dbuf[LINK_ADDRSTRLEN];
2481 		sockaddr_format(rt->rt_gateway, gbuf, sizeof(gbuf));
2482 		printf("%s: bad gateway address type %s for dst %s"
2483 		    " through interface %s\n", __func__, gbuf,
2484 		    IN6_PRINT(dbuf, &satocsin6(dst)->sin6_addr),
2485 		    if_name(ifp));
2486 		m_freem(m);
2487 		return 0;
2488 	}
2489 	sdl = satocsdl(rt->rt_gateway);
2490 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2491 		char sbuf[INET6_ADDRSTRLEN];
2492 		char dbuf[LINK_ADDRSTRLEN];
2493 		/* this should be impossible, but we bark here for debugging */
2494 		printf("%s: sdl_alen == %" PRIu8 ", if=%s, dst=%s, sdl=%s\n",
2495 		    __func__, sdl->sdl_alen, if_name(ifp),
2496 		    IN6_PRINT(sbuf, &satocsin6(dst)->sin6_addr),
2497 		    DL_PRINT(dbuf, &sdl->sdl_addr));
2498 		m_freem(m);
2499 		return 0;
2500 	}
2501 
2502 	memcpy(lldst, CLLADDR(sdl), MIN(dstsize, sdl->sdl_alen));
2503 	return 1;
2504 }
2505 
2506 static void
2507 clear_llinfo_pqueue(struct llinfo_nd6 *ln)
2508 {
2509 	struct mbuf *m_hold, *m_hold_next;
2510 
2511 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2512 		m_hold_next = m_hold->m_nextpkt;
2513 		m_hold->m_nextpkt = NULL;
2514 		m_freem(m_hold);
2515 	}
2516 
2517 	ln->ln_hold = NULL;
2518 	return;
2519 }
2520 
2521 int
2522 nd6_sysctl(
2523     int name,
2524     void *oldp,	/* syscall arg, need copyout */
2525     size_t *oldlenp,
2526     void *newp,	/* syscall arg, need copyin */
2527     size_t newlen
2528 )
2529 {
2530 	void *p;
2531 	size_t ol;
2532 	int error;
2533 
2534 	error = 0;
2535 
2536 	if (newp)
2537 		return EPERM;
2538 	if (oldp && !oldlenp)
2539 		return EINVAL;
2540 	ol = oldlenp ? *oldlenp : 0;
2541 
2542 	if (oldp) {
2543 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2544 		if (p == NULL)
2545 			return ENOMEM;
2546 	} else
2547 		p = NULL;
2548 	switch (name) {
2549 	case ICMPV6CTL_ND6_DRLIST:
2550 		error = fill_drlist(p, oldlenp, ol);
2551 		if (!error && p != NULL && oldp != NULL)
2552 			error = copyout(p, oldp, *oldlenp);
2553 		break;
2554 
2555 	case ICMPV6CTL_ND6_PRLIST:
2556 		error = fill_prlist(p, oldlenp, ol);
2557 		if (!error && p != NULL && oldp != NULL)
2558 			error = copyout(p, oldp, *oldlenp);
2559 		break;
2560 
2561 	case ICMPV6CTL_ND6_MAXQLEN:
2562 		break;
2563 
2564 	default:
2565 		error = ENOPROTOOPT;
2566 		break;
2567 	}
2568 	if (p)
2569 		free(p, M_TEMP);
2570 
2571 	return error;
2572 }
2573 
2574 static int
2575 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2576 {
2577 	int error = 0, s;
2578 	struct in6_defrouter *d = NULL, *de = NULL;
2579 	struct nd_defrouter *dr;
2580 	size_t l;
2581 
2582 	s = splsoftnet();
2583 
2584 	if (oldp) {
2585 		d = (struct in6_defrouter *)oldp;
2586 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2587 	}
2588 	l = 0;
2589 
2590 	TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
2591 
2592 		if (oldp && d + 1 <= de) {
2593 			memset(d, 0, sizeof(*d));
2594 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2595 			if (sa6_recoverscope(&d->rtaddr)) {
2596 				log(LOG_ERR,
2597 				    "scope error in router list (%s)\n",
2598 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2599 				/* XXX: press on... */
2600 			}
2601 			d->flags = dr->flags;
2602 			d->rtlifetime = dr->rtlifetime;
2603 			d->expire = dr->expire ?
2604 			    time_mono_to_wall(dr->expire) : 0;
2605 			d->if_index = dr->ifp->if_index;
2606 		}
2607 
2608 		l += sizeof(*d);
2609 		if (d)
2610 			d++;
2611 	}
2612 
2613 	if (oldp) {
2614 		if (l > ol)
2615 			error = ENOMEM;
2616 	}
2617 	if (oldlenp)
2618 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2619 
2620 	splx(s);
2621 
2622 	return error;
2623 }
2624 
2625 static int
2626 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2627 {
2628 	int error = 0, s;
2629 	struct nd_prefix *pr;
2630 	uint8_t *p = NULL, *ps = NULL;
2631 	uint8_t *pe = NULL;
2632 	size_t l;
2633 
2634 	s = splsoftnet();
2635 
2636 	if (oldp) {
2637 		ps = p = (uint8_t*)oldp;
2638 		pe = (uint8_t*)oldp + *oldlenp;
2639 	}
2640 	l = 0;
2641 
2642 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2643 		u_short advrtrs;
2644 		struct sockaddr_in6 sin6;
2645 		struct nd_pfxrouter *pfr;
2646 		struct in6_prefix pfx;
2647 
2648 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2649 		{
2650 			memset(&pfx, 0, sizeof(pfx));
2651 			ps = p;
2652 			pfx.prefix = pr->ndpr_prefix;
2653 
2654 			if (sa6_recoverscope(&pfx.prefix)) {
2655 				log(LOG_ERR,
2656 				    "scope error in prefix list (%s)\n",
2657 				    ip6_sprintf(&pfx.prefix.sin6_addr));
2658 				/* XXX: press on... */
2659 			}
2660 			pfx.raflags = pr->ndpr_raf;
2661 			pfx.prefixlen = pr->ndpr_plen;
2662 			pfx.vltime = pr->ndpr_vltime;
2663 			pfx.pltime = pr->ndpr_pltime;
2664 			pfx.if_index = pr->ndpr_ifp->if_index;
2665 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2666 				pfx.expire = 0;
2667 			else {
2668 				time_t maxexpire;
2669 
2670 				/* XXX: we assume time_t is signed. */
2671 				maxexpire = (-1) &
2672 				    ~((time_t)1 <<
2673 				    ((sizeof(maxexpire) * 8) - 1));
2674 				if (pr->ndpr_vltime <
2675 				    maxexpire - pr->ndpr_lastupdate) {
2676 					pfx.expire = pr->ndpr_lastupdate +
2677 						pr->ndpr_vltime;
2678 				} else
2679 					pfx.expire = maxexpire;
2680 			}
2681 			pfx.refcnt = pr->ndpr_refcnt;
2682 			pfx.flags = pr->ndpr_stateflags;
2683 			pfx.origin = PR_ORIG_RA;
2684 
2685 			p += sizeof(pfx); l += sizeof(pfx);
2686 
2687 			advrtrs = 0;
2688 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2689 				if (p + sizeof(sin6) > pe) {
2690 					advrtrs++;
2691 					continue;
2692 				}
2693 
2694 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2695 				    0, 0, 0);
2696 				if (sa6_recoverscope(&sin6)) {
2697 					log(LOG_ERR,
2698 					    "scope error in "
2699 					    "prefix list (%s)\n",
2700 					    ip6_sprintf(&pfr->router->rtaddr));
2701 				}
2702 				advrtrs++;
2703 				memcpy(p, &sin6, sizeof(sin6));
2704 				p += sizeof(sin6);
2705 				l += sizeof(sin6);
2706 			}
2707 			pfx.advrtrs = advrtrs;
2708 			memcpy(ps, &pfx, sizeof(pfx));
2709 		}
2710 		else {
2711 			l += sizeof(pfx);
2712 			advrtrs = 0;
2713 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2714 				advrtrs++;
2715 				l += sizeof(sin6);
2716 			}
2717 		}
2718 	}
2719 
2720 	if (oldp) {
2721 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2722 		if (l > ol)
2723 			error = ENOMEM;
2724 	} else
2725 		*oldlenp = l;
2726 
2727 	splx(s);
2728 
2729 	return error;
2730 }
2731