xref: /netbsd-src/sys/netinet6/nd6.c (revision 16dce51364ebe8aeafbae46bc5aa167b8115bc45)
1 /*	$NetBSD: nd6.c,v 1.247 2018/03/06 10:57:00 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 1.247 2018/03/06 10:57:00 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/kmem.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/errno.h>
54 #include <sys/ioctl.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/cprng.h>
58 #include <sys/workqueue.h>
59 
60 #include <net/if.h>
61 #include <net/if_dl.h>
62 #include <net/if_llatbl.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/if_ether.h>
66 #include <net/if_fddi.h>
67 #include <net/if_arc.h>
68 
69 #include <netinet/in.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/icmp6_private.h>
78 
79 #include <net/net_osdep.h>
80 
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
83 
84 /* timer values */
85 int	nd6_prune	= 1;	/* walk list every 1 seconds */
86 int	nd6_delay	= 5;	/* delay first probe time 5 second */
87 int	nd6_umaxtries	= 3;	/* maximum unicast query */
88 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
89 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
90 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
91 
92 /* preventing too many loops in ND option parsing */
93 int nd6_maxndopt = 10;	/* max # of ND options allowed */
94 
95 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
96 
97 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
98 
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104 
105 krwlock_t nd6_lock __cacheline_aligned;
106 
107 struct nd_drhead nd_defrouter;
108 struct nd_prhead nd_prefix = { 0 };
109 
110 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
111 
112 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
113 static void nd6_slowtimo(void *);
114 static int regen_tmpaddr(const struct in6_ifaddr *);
115 static void nd6_free(struct llentry *, int);
116 static void nd6_llinfo_timer(void *);
117 static void nd6_timer(void *);
118 static void nd6_timer_work(struct work *, void *);
119 static void clear_llinfo_pqueue(struct llentry *);
120 static struct nd_opt_hdr *nd6_option(union nd_opts *);
121 
122 static callout_t nd6_slowtimo_ch;
123 static callout_t nd6_timer_ch;
124 static struct workqueue	*nd6_timer_wq;
125 static struct work	nd6_timer_wk;
126 
127 static int fill_drlist(void *, size_t *);
128 static int fill_prlist(void *, size_t *);
129 
130 static struct ifnet *nd6_defifp;
131 static int nd6_defifindex;
132 
133 static int nd6_setdefaultiface(int);
134 
135 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
136 
137 void
138 nd6_init(void)
139 {
140 	int error;
141 
142 	rw_init(&nd6_lock);
143 
144 	/* initialization of the default router list */
145 	ND_DEFROUTER_LIST_INIT();
146 
147 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
148 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
149 
150 	error = workqueue_create(&nd6_timer_wq, "nd6_timer",
151 	    nd6_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
152 	if (error)
153 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
154 
155 	/* start timer */
156 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
157 	    nd6_slowtimo, NULL);
158 	callout_reset(&nd6_timer_ch, hz, nd6_timer, NULL);
159 }
160 
161 struct nd_ifinfo *
162 nd6_ifattach(struct ifnet *ifp)
163 {
164 	struct nd_ifinfo *nd;
165 
166 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
167 
168 	nd->initialized = 1;
169 
170 	nd->chlim = IPV6_DEFHLIM;
171 	nd->basereachable = REACHABLE_TIME;
172 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
173 	nd->retrans = RETRANS_TIMER;
174 
175 	nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV;
176 
177 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
178 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
179 	 * because one of its members should. */
180 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
181 	    (ifp->if_flags & IFF_LOOPBACK))
182 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
183 
184 	/* A loopback interface does not need to accept RTADV.
185 	 * A bridge interface should not accept RTADV
186 	 * because one of its members should. */
187 	if (ip6_accept_rtadv &&
188 	    !(ifp->if_flags & IFF_LOOPBACK) &&
189 	    !(ifp->if_type != IFT_BRIDGE))
190 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
191 
192 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
193 	nd6_setmtu0(ifp, nd);
194 
195 	return nd;
196 }
197 
198 void
199 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
200 {
201 
202 	/* Ensure all IPv6 addresses are purged before calling nd6_purge */
203 	if_purgeaddrs(ifp, AF_INET6, in6_purgeaddr);
204 	nd6_purge(ifp, ext);
205 	kmem_free(ext->nd_ifinfo, sizeof(struct nd_ifinfo));
206 }
207 
208 void
209 nd6_setmtu(struct ifnet *ifp)
210 {
211 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
212 }
213 
214 void
215 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
216 {
217 	u_int32_t omaxmtu;
218 
219 	omaxmtu = ndi->maxmtu;
220 
221 	switch (ifp->if_type) {
222 	case IFT_ARCNET:
223 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
224 		break;
225 	case IFT_FDDI:
226 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
227 		break;
228 	default:
229 		ndi->maxmtu = ifp->if_mtu;
230 		break;
231 	}
232 
233 	/*
234 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
235 	 * undesirable situation.  We thus notify the operator of the change
236 	 * explicitly.  The check for omaxmtu is necessary to restrict the
237 	 * log to the case of changing the MTU, not initializing it.
238 	 */
239 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
240 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
241 		    " small for IPv6 which needs %lu\n",
242 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
243 		    IPV6_MMTU);
244 	}
245 
246 	if (ndi->maxmtu > in6_maxmtu)
247 		in6_setmaxmtu(); /* check all interfaces just in case */
248 }
249 
250 void
251 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
252 {
253 
254 	memset(ndopts, 0, sizeof(*ndopts));
255 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
256 	ndopts->nd_opts_last
257 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
258 
259 	if (icmp6len == 0) {
260 		ndopts->nd_opts_done = 1;
261 		ndopts->nd_opts_search = NULL;
262 	}
263 }
264 
265 /*
266  * Take one ND option.
267  */
268 static struct nd_opt_hdr *
269 nd6_option(union nd_opts *ndopts)
270 {
271 	struct nd_opt_hdr *nd_opt;
272 	int olen;
273 
274 	KASSERT(ndopts != NULL);
275 	KASSERT(ndopts->nd_opts_last != NULL);
276 
277 	if (ndopts->nd_opts_search == NULL)
278 		return NULL;
279 	if (ndopts->nd_opts_done)
280 		return NULL;
281 
282 	nd_opt = ndopts->nd_opts_search;
283 
284 	/* make sure nd_opt_len is inside the buffer */
285 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
286 		memset(ndopts, 0, sizeof(*ndopts));
287 		return NULL;
288 	}
289 
290 	olen = nd_opt->nd_opt_len << 3;
291 	if (olen == 0) {
292 		/*
293 		 * Message validation requires that all included
294 		 * options have a length that is greater than zero.
295 		 */
296 		memset(ndopts, 0, sizeof(*ndopts));
297 		return NULL;
298 	}
299 
300 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
301 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
302 		/* option overruns the end of buffer, invalid */
303 		memset(ndopts, 0, sizeof(*ndopts));
304 		return NULL;
305 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
306 		/* reached the end of options chain */
307 		ndopts->nd_opts_done = 1;
308 		ndopts->nd_opts_search = NULL;
309 	}
310 	return nd_opt;
311 }
312 
313 /*
314  * Parse multiple ND options.
315  * This function is much easier to use, for ND routines that do not need
316  * multiple options of the same type.
317  */
318 int
319 nd6_options(union nd_opts *ndopts)
320 {
321 	struct nd_opt_hdr *nd_opt;
322 	int i = 0;
323 
324 	KASSERT(ndopts != NULL);
325 	KASSERT(ndopts->nd_opts_last != NULL);
326 
327 	if (ndopts->nd_opts_search == NULL)
328 		return 0;
329 
330 	while (1) {
331 		nd_opt = nd6_option(ndopts);
332 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
333 			/*
334 			 * Message validation requires that all included
335 			 * options have a length that is greater than zero.
336 			 */
337 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
338 			memset(ndopts, 0, sizeof(*ndopts));
339 			return -1;
340 		}
341 
342 		if (nd_opt == NULL)
343 			goto skip1;
344 
345 		switch (nd_opt->nd_opt_type) {
346 		case ND_OPT_SOURCE_LINKADDR:
347 		case ND_OPT_TARGET_LINKADDR:
348 		case ND_OPT_MTU:
349 		case ND_OPT_REDIRECTED_HEADER:
350 		case ND_OPT_NONCE:
351 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
352 				nd6log(LOG_INFO,
353 				    "duplicated ND6 option found (type=%d)\n",
354 				    nd_opt->nd_opt_type);
355 				/* XXX bark? */
356 			} else {
357 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
358 					= nd_opt;
359 			}
360 			break;
361 		case ND_OPT_PREFIX_INFORMATION:
362 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
363 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
364 					= nd_opt;
365 			}
366 			ndopts->nd_opts_pi_end =
367 				(struct nd_opt_prefix_info *)nd_opt;
368 			break;
369 		default:
370 			/*
371 			 * Unknown options must be silently ignored,
372 			 * to accommodate future extension to the protocol.
373 			 */
374 			nd6log(LOG_DEBUG,
375 			    "nd6_options: unsupported option %d - "
376 			    "option ignored\n", nd_opt->nd_opt_type);
377 		}
378 
379 skip1:
380 		i++;
381 		if (i > nd6_maxndopt) {
382 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
383 			nd6log(LOG_INFO, "too many loop in nd opt\n");
384 			break;
385 		}
386 
387 		if (ndopts->nd_opts_done)
388 			break;
389 	}
390 
391 	return 0;
392 }
393 
394 /*
395  * ND6 timer routine to handle ND6 entries
396  */
397 void
398 nd6_llinfo_settimer(struct llentry *ln, time_t xtick)
399 {
400 
401 	CTASSERT(sizeof(time_t) > sizeof(int));
402 	LLE_WLOCK_ASSERT(ln);
403 
404 	KASSERT(xtick >= 0);
405 
406 	/*
407 	 * We have to take care of a reference leak which occurs if
408 	 * callout_reset overwrites a pending callout schedule.  Unfortunately
409 	 * we don't have a mean to know the overwrite, so we need to know it
410 	 * using callout_stop.  We need to call callout_pending first to exclude
411 	 * the case that the callout has never been scheduled.
412 	 */
413 	if (callout_pending(&ln->la_timer)) {
414 		bool expired = callout_stop(&ln->la_timer);
415 		if (!expired)
416 			LLE_REMREF(ln);
417 	}
418 
419 	ln->ln_expire = time_uptime + xtick / hz;
420 	LLE_ADDREF(ln);
421 	if (xtick > INT_MAX) {
422 		ln->ln_ntick = xtick - INT_MAX;
423 		callout_reset(&ln->ln_timer_ch, INT_MAX,
424 		    nd6_llinfo_timer, ln);
425 	} else {
426 		ln->ln_ntick = 0;
427 		callout_reset(&ln->ln_timer_ch, xtick,
428 		    nd6_llinfo_timer, ln);
429 	}
430 }
431 
432 /*
433  * Gets source address of the first packet in hold queue
434  * and stores it in @src.
435  * Returns pointer to @src (if hold queue is not empty) or NULL.
436  */
437 static struct in6_addr *
438 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
439 {
440 	struct ip6_hdr *hip6;
441 
442 	if (ln == NULL || ln->ln_hold == NULL)
443 		return NULL;
444 
445 	/*
446 	 * assuming every packet in ln_hold has the same IP header
447 	 */
448 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
449 	/* XXX pullup? */
450 	if (sizeof(*hip6) < ln->ln_hold->m_len)
451 		*src = hip6->ip6_src;
452 	else
453 		src = NULL;
454 
455 	return src;
456 }
457 
458 static void
459 nd6_llinfo_timer(void *arg)
460 {
461 	struct llentry *ln = arg;
462 	struct ifnet *ifp;
463 	struct nd_ifinfo *ndi = NULL;
464 	bool send_ns = false;
465 	const struct in6_addr *daddr6 = NULL;
466 
467 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
468 
469 	LLE_WLOCK(ln);
470 	if ((ln->la_flags & LLE_LINKED) == 0)
471 		goto out;
472 	if (ln->ln_ntick > 0) {
473 		nd6_llinfo_settimer(ln, ln->ln_ntick);
474 		goto out;
475 	}
476 
477 
478 	ifp = ln->lle_tbl->llt_ifp;
479 	KASSERT(ifp != NULL);
480 
481 	ndi = ND_IFINFO(ifp);
482 
483 	switch (ln->ln_state) {
484 	case ND6_LLINFO_INCOMPLETE:
485 		if (ln->ln_asked < nd6_mmaxtries) {
486 			ln->ln_asked++;
487 			send_ns = true;
488 		} else {
489 			struct mbuf *m = ln->ln_hold;
490 			if (m) {
491 				struct mbuf *m0;
492 
493 				/*
494 				 * assuming every packet in ln_hold has
495 				 * the same IP header
496 				 */
497 				m0 = m->m_nextpkt;
498 				m->m_nextpkt = NULL;
499 				ln->ln_hold = m0;
500 				clear_llinfo_pqueue(ln);
501  			}
502 			nd6_free(ln, 0);
503 			ln = NULL;
504 			if (m != NULL) {
505 				icmp6_error2(m, ICMP6_DST_UNREACH,
506 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
507 			}
508 		}
509 		break;
510 	case ND6_LLINFO_REACHABLE:
511 		if (!ND6_LLINFO_PERMANENT(ln)) {
512 			ln->ln_state = ND6_LLINFO_STALE;
513 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
514 		}
515 		break;
516 
517 	case ND6_LLINFO_PURGE:
518 	case ND6_LLINFO_STALE:
519 		/* Garbage Collection(RFC 2461 5.3) */
520 		if (!ND6_LLINFO_PERMANENT(ln)) {
521 			nd6_free(ln, 1);
522 			ln = NULL;
523 		}
524 		break;
525 
526 	case ND6_LLINFO_DELAY:
527 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
528 			/* We need NUD */
529 			ln->ln_asked = 1;
530 			ln->ln_state = ND6_LLINFO_PROBE;
531 			daddr6 = &ln->r_l3addr.addr6;
532 			send_ns = true;
533 		} else {
534 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
535 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
536 		}
537 		break;
538 	case ND6_LLINFO_PROBE:
539 		if (ln->ln_asked < nd6_umaxtries) {
540 			ln->ln_asked++;
541 			daddr6 = &ln->r_l3addr.addr6;
542 			send_ns = true;
543 		} else {
544 			nd6_free(ln, 0);
545 			ln = NULL;
546 		}
547 		break;
548 	}
549 
550 	if (send_ns) {
551 		struct in6_addr src, *psrc;
552 		const struct in6_addr *taddr6 = &ln->r_l3addr.addr6;
553 
554 		nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
555 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
556 		LLE_FREE_LOCKED(ln);
557 		ln = NULL;
558 		nd6_ns_output(ifp, daddr6, taddr6, psrc, NULL);
559 	}
560 
561 out:
562 	if (ln != NULL)
563 		LLE_FREE_LOCKED(ln);
564 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
565 }
566 
567 /*
568  * ND6 timer routine to expire default route list and prefix list
569  */
570 static void
571 nd6_timer_work(struct work *wk, void *arg)
572 {
573 	struct nd_defrouter *next_dr, *dr;
574 	struct nd_prefix *next_pr, *pr;
575 	struct in6_ifaddr *ia6, *nia6;
576 	int s, bound;
577 	struct psref psref;
578 
579 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
580 	    nd6_timer, NULL);
581 
582 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
583 
584 	/* expire default router list */
585 
586 	ND6_WLOCK();
587 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, next_dr) {
588 		if (dr->expire && dr->expire < time_uptime) {
589 			nd6_defrtrlist_del(dr, NULL);
590 		}
591 	}
592 	ND6_UNLOCK();
593 
594 	/*
595 	 * expire interface addresses.
596 	 * in the past the loop was inside prefix expiry processing.
597 	 * However, from a stricter speci-confrmance standpoint, we should
598 	 * rather separate address lifetimes and prefix lifetimes.
599 	 */
600 	bound = curlwp_bind();
601   addrloop:
602 	s = pserialize_read_enter();
603 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
604 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
605 
606 		ia6_acquire(ia6, &psref);
607 		pserialize_read_exit(s);
608 
609 		/* check address lifetime */
610 		if (IFA6_IS_INVALID(ia6)) {
611 			int regen = 0;
612 			struct ifnet *ifp;
613 
614 			/*
615 			 * If the expiring address is temporary, try
616 			 * regenerating a new one.  This would be useful when
617 			 * we suspended a laptop PC, then turned it on after a
618 			 * period that could invalidate all temporary
619 			 * addresses.  Although we may have to restart the
620 			 * loop (see below), it must be after purging the
621 			 * address.  Otherwise, we'd see an infinite loop of
622 			 * regeneration.
623 			 */
624 			if (ip6_use_tempaddr &&
625 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
626 				IFNET_LOCK(ia6->ia_ifa.ifa_ifp);
627 				if (regen_tmpaddr(ia6) == 0)
628 					regen = 1;
629 				IFNET_UNLOCK(ia6->ia_ifa.ifa_ifp);
630 			}
631 
632 			ifp = ia6->ia_ifa.ifa_ifp;
633 			IFNET_LOCK(ifp);
634 			/*
635 			 * Need to take the lock first to prevent if_detach
636 			 * from running in6_purgeaddr concurrently.
637 			 */
638 			if (!if_is_deactivated(ifp)) {
639 				ia6_release(ia6, &psref);
640 				in6_purgeaddr(&ia6->ia_ifa);
641 			} else {
642 				/*
643 				 * ifp is being destroyed, ia6 will be destroyed
644 				 * by if_detach.
645 				 */
646 				ia6_release(ia6, &psref);
647 			}
648 			ia6 = NULL;
649 			IFNET_UNLOCK(ifp);
650 
651 			if (regen)
652 				goto addrloop; /* XXX: see below */
653 		} else if (IFA6_IS_DEPRECATED(ia6)) {
654 			int oldflags = ia6->ia6_flags;
655 
656 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
657 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
658 				rt_newaddrmsg(RTM_NEWADDR,
659 				    (struct ifaddr *)ia6, 0, NULL);
660 			}
661 
662 			/*
663 			 * If a temporary address has just become deprecated,
664 			 * regenerate a new one if possible.
665 			 */
666 			if (ip6_use_tempaddr &&
667 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
668 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
669 
670 				if (regen_tmpaddr(ia6) == 0) {
671 					/*
672 					 * A new temporary address is
673 					 * generated.
674 					 * XXX: this means the address chain
675 					 * has changed while we are still in
676 					 * the loop.  Although the change
677 					 * would not cause disaster (because
678 					 * it's not a deletion, but an
679 					 * addition,) we'd rather restart the
680 					 * loop just for safety.  Or does this
681 					 * significantly reduce performance??
682 					 */
683 					ia6_release(ia6, &psref);
684 					goto addrloop;
685 				}
686 			}
687 		} else {
688 			/*
689 			 * A new RA might have made a deprecated address
690 			 * preferred.
691 			 */
692 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
693 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
694 				rt_newaddrmsg(RTM_NEWADDR,
695 				    (struct ifaddr *)ia6, 0, NULL);
696 			}
697 		}
698 		s = pserialize_read_enter();
699 		ia6_release(ia6, &psref);
700 	}
701 	pserialize_read_exit(s);
702 	curlwp_bindx(bound);
703 
704 	/* expire prefix list */
705 	ND6_WLOCK();
706 	ND_PREFIX_LIST_FOREACH_SAFE(pr, next_pr) {
707 		/*
708 		 * check prefix lifetime.
709 		 * since pltime is just for autoconf, pltime processing for
710 		 * prefix is not necessary.
711 		 */
712 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
713 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
714 			/*
715 			 * Just invalidate the prefix here. Removing it
716 			 * will be done when purging an associated address.
717 			 */
718 			KASSERT(pr->ndpr_refcnt > 0);
719 			nd6_invalidate_prefix(pr);
720 		}
721 	}
722 	ND6_UNLOCK();
723 
724 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
725 }
726 
727 static void
728 nd6_timer(void *ignored_arg)
729 {
730 
731 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
732 }
733 
734 /* ia6: deprecated/invalidated temporary address */
735 static int
736 regen_tmpaddr(const struct in6_ifaddr *ia6)
737 {
738 	struct ifaddr *ifa;
739 	struct ifnet *ifp;
740 	struct in6_ifaddr *public_ifa6 = NULL;
741 	int s;
742 
743 	ifp = ia6->ia_ifa.ifa_ifp;
744 	s = pserialize_read_enter();
745 	IFADDR_READER_FOREACH(ifa, ifp) {
746 		struct in6_ifaddr *it6;
747 
748 		if (ifa->ifa_addr->sa_family != AF_INET6)
749 			continue;
750 
751 		it6 = (struct in6_ifaddr *)ifa;
752 
753 		/* ignore no autoconf addresses. */
754 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
755 			continue;
756 
757 		/* ignore autoconf addresses with different prefixes. */
758 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
759 			continue;
760 
761 		/*
762 		 * Now we are looking at an autoconf address with the same
763 		 * prefix as ours.  If the address is temporary and is still
764 		 * preferred, do not create another one.  It would be rare, but
765 		 * could happen, for example, when we resume a laptop PC after
766 		 * a long period.
767 		 */
768 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
769 		    !IFA6_IS_DEPRECATED(it6)) {
770 			public_ifa6 = NULL;
771 			break;
772 		}
773 
774 		/*
775 		 * This is a public autoconf address that has the same prefix
776 		 * as ours.  If it is preferred, keep it.  We can't break the
777 		 * loop here, because there may be a still-preferred temporary
778 		 * address with the prefix.
779 		 */
780 		if (!IFA6_IS_DEPRECATED(it6))
781 			public_ifa6 = it6;
782 	}
783 
784 	if (public_ifa6 != NULL) {
785 		int e;
786 		struct psref psref;
787 
788 		ia6_acquire(public_ifa6, &psref);
789 		pserialize_read_exit(s);
790 		/*
791 		 * Random factor is introduced in the preferred lifetime, so
792 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
793 		 */
794 		ND6_WLOCK();
795 		e = in6_tmpifadd(public_ifa6, 0, 0);
796 		ND6_UNLOCK();
797 		if (e != 0) {
798 			ia6_release(public_ifa6, &psref);
799 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
800 			    " tmp addr, errno=%d\n", e);
801 			return -1;
802 		}
803 		ia6_release(public_ifa6, &psref);
804 		return 0;
805 	}
806 	pserialize_read_exit(s);
807 
808 	return -1;
809 }
810 
811 bool
812 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
813 {
814 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
815 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
816 		return true;
817 	case ND6_IFF_ACCEPT_RTADV:
818 		return ip6_accept_rtadv != 0;
819 	case ND6_IFF_OVERRIDE_RTADV:
820 	case 0:
821 	default:
822 		return false;
823 	}
824 }
825 
826 /*
827  * Nuke neighbor cache/prefix/default router management table, right before
828  * ifp goes away.
829  */
830 void
831 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
832 {
833 	struct nd_defrouter *dr, *ndr;
834 	struct nd_prefix *pr, *npr;
835 
836 	/*
837 	 * During detach, the ND info might be already removed, but
838 	 * then is explitly passed as argument.
839 	 * Otherwise get it from ifp->if_afdata.
840 	 */
841 	if (ext == NULL)
842 		ext = ifp->if_afdata[AF_INET6];
843 	if (ext == NULL)
844 		return;
845 
846 	ND6_WLOCK();
847 	/*
848 	 * Nuke default router list entries toward ifp.
849 	 * We defer removal of default router list entries that is installed
850 	 * in the routing table, in order to keep additional side effects as
851 	 * small as possible.
852 	 */
853 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
854 		if (dr->installed)
855 			continue;
856 
857 		if (dr->ifp == ifp) {
858 			KASSERT(ext != NULL);
859 			nd6_defrtrlist_del(dr, ext);
860 		}
861 	}
862 
863 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
864 		if (!dr->installed)
865 			continue;
866 
867 		if (dr->ifp == ifp) {
868 			KASSERT(ext != NULL);
869 			nd6_defrtrlist_del(dr, ext);
870 		}
871 	}
872 
873 	/* Nuke prefix list entries toward ifp */
874 	ND_PREFIX_LIST_FOREACH_SAFE(pr, npr) {
875 		if (pr->ndpr_ifp == ifp) {
876 			/*
877 			 * All addresses referencing pr should be already freed.
878 			 */
879 			KASSERT(pr->ndpr_refcnt == 0);
880 			nd6_prelist_remove(pr);
881 		}
882 	}
883 
884 	/* cancel default outgoing interface setting */
885 	if (nd6_defifindex == ifp->if_index)
886 		nd6_setdefaultiface(0);
887 
888 	/* XXX: too restrictive? */
889 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
890 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
891 		if (ndi && nd6_accepts_rtadv(ndi)) {
892 			/* refresh default router list */
893 			nd6_defrouter_select();
894 		}
895 	}
896 	ND6_UNLOCK();
897 
898 	/*
899 	 * We may not need to nuke the neighbor cache entries here
900 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
901 	 * nd6_purge() is invoked by in6_ifdetach() which is called
902 	 * from if_detach() where everything gets purged. However
903 	 * in6_ifdetach is directly called from vlan(4), so we still
904 	 * need to purge entries here.
905 	 */
906 	if (ext->lltable != NULL)
907 		lltable_purge_entries(ext->lltable);
908 }
909 
910 void
911 nd6_assert_purged(struct ifnet *ifp)
912 {
913 	struct nd_defrouter *dr;
914 	struct nd_prefix *pr;
915 	char ip6buf[INET6_ADDRSTRLEN] __diagused;
916 
917 	ND6_RLOCK();
918 	ND_DEFROUTER_LIST_FOREACH(dr) {
919 		KASSERTMSG(dr->ifp != ifp,
920 		    "defrouter %s remains on %s",
921 		    IN6_PRINT(ip6buf, &dr->rtaddr), ifp->if_xname);
922 	}
923 
924 	ND_PREFIX_LIST_FOREACH(pr) {
925 		KASSERTMSG(pr->ndpr_ifp != ifp,
926 		    "prefix %s/%d remains on %s",
927 		    IN6_PRINT(ip6buf, &pr->ndpr_prefix.sin6_addr),
928 		    pr->ndpr_plen, ifp->if_xname);
929 	}
930 	ND6_UNLOCK();
931 }
932 
933 struct llentry *
934 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
935 {
936 	struct sockaddr_in6 sin6;
937 	struct llentry *ln;
938 
939 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
940 
941 	IF_AFDATA_RLOCK(ifp);
942 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
943 	    sin6tosa(&sin6));
944 	IF_AFDATA_RUNLOCK(ifp);
945 
946 	return ln;
947 }
948 
949 struct llentry *
950 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
951 {
952 	struct sockaddr_in6 sin6;
953 	struct llentry *ln;
954 	struct rtentry *rt;
955 
956 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
957 	rt = rtalloc1(sin6tosa(&sin6), 0);
958 
959 	IF_AFDATA_WLOCK(ifp);
960 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE, sin6tosa(&sin6), rt);
961 	IF_AFDATA_WUNLOCK(ifp);
962 
963 	if (rt != NULL)
964 		rt_unref(rt);
965 	if (ln != NULL)
966 		ln->ln_state = ND6_LLINFO_NOSTATE;
967 
968 	return ln;
969 }
970 
971 /*
972  * Test whether a given IPv6 address is a neighbor or not, ignoring
973  * the actual neighbor cache.  The neighbor cache is ignored in order
974  * to not reenter the routing code from within itself.
975  */
976 static int
977 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
978 {
979 	struct nd_prefix *pr;
980 	struct ifaddr *dstaddr;
981 	int s;
982 
983 	/*
984 	 * A link-local address is always a neighbor.
985 	 * XXX: a link does not necessarily specify a single interface.
986 	 */
987 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
988 		struct sockaddr_in6 sin6_copy;
989 		u_int32_t zone;
990 
991 		/*
992 		 * We need sin6_copy since sa6_recoverscope() may modify the
993 		 * content (XXX).
994 		 */
995 		sin6_copy = *addr;
996 		if (sa6_recoverscope(&sin6_copy))
997 			return 0; /* XXX: should be impossible */
998 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
999 			return 0;
1000 		if (sin6_copy.sin6_scope_id == zone)
1001 			return 1;
1002 		else
1003 			return 0;
1004 	}
1005 
1006 	/*
1007 	 * If the address matches one of our addresses,
1008 	 * it should be a neighbor.
1009 	 * If the address matches one of our on-link prefixes, it should be a
1010 	 * neighbor.
1011 	 */
1012 	ND6_RLOCK();
1013 	ND_PREFIX_LIST_FOREACH(pr) {
1014 		if (pr->ndpr_ifp != ifp)
1015 			continue;
1016 
1017 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1018 			struct rtentry *rt;
1019 
1020 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
1021 			if (rt == NULL)
1022 				continue;
1023 			/*
1024 			 * This is the case where multiple interfaces
1025 			 * have the same prefix, but only one is installed
1026 			 * into the routing table and that prefix entry
1027 			 * is not the one being examined here. In the case
1028 			 * where RADIX_MPATH is enabled, multiple route
1029 			 * entries (of the same rt_key value) will be
1030 			 * installed because the interface addresses all
1031 			 * differ.
1032 			 */
1033 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1034 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
1035 				rt_unref(rt);
1036 				continue;
1037 			}
1038 			rt_unref(rt);
1039 		}
1040 
1041 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1042 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1043 			ND6_UNLOCK();
1044 			return 1;
1045 		}
1046 	}
1047 	ND6_UNLOCK();
1048 
1049 	/*
1050 	 * If the address is assigned on the node of the other side of
1051 	 * a p2p interface, the address should be a neighbor.
1052 	 */
1053 	s = pserialize_read_enter();
1054 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1055 	if (dstaddr != NULL) {
1056 		if (dstaddr->ifa_ifp == ifp) {
1057 			pserialize_read_exit(s);
1058 			return 1;
1059 		}
1060 	}
1061 	pserialize_read_exit(s);
1062 
1063 	/*
1064 	 * If the default router list is empty, all addresses are regarded
1065 	 * as on-link, and thus, as a neighbor.
1066 	 */
1067 	ND6_RLOCK();
1068 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1069 	    ND_DEFROUTER_LIST_EMPTY() && nd6_defifindex == ifp->if_index) {
1070 		ND6_UNLOCK();
1071 		return 1;
1072 	}
1073 	ND6_UNLOCK();
1074 
1075 	return 0;
1076 }
1077 
1078 /*
1079  * Detect if a given IPv6 address identifies a neighbor on a given link.
1080  * XXX: should take care of the destination of a p2p link?
1081  */
1082 int
1083 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1084 {
1085 	struct nd_prefix *pr;
1086 	struct llentry *ln;
1087 	struct rtentry *rt;
1088 
1089 	/*
1090 	 * A link-local address is always a neighbor.
1091 	 * XXX: a link does not necessarily specify a single interface.
1092 	 */
1093 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1094 		struct sockaddr_in6 sin6_copy;
1095 		u_int32_t zone;
1096 
1097 		/*
1098 		 * We need sin6_copy since sa6_recoverscope() may modify the
1099 		 * content (XXX).
1100 		 */
1101 		sin6_copy = *addr;
1102 		if (sa6_recoverscope(&sin6_copy))
1103 			return 0; /* XXX: should be impossible */
1104 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1105 			return 0;
1106 		if (sin6_copy.sin6_scope_id == zone)
1107 			return 1;
1108 		else
1109 			return 0;
1110 	}
1111 
1112 	/*
1113 	 * If the address matches one of our on-link prefixes, it should be a
1114 	 * neighbor.
1115 	 */
1116 	ND6_RLOCK();
1117 	ND_PREFIX_LIST_FOREACH(pr) {
1118 		if (pr->ndpr_ifp != ifp)
1119 			continue;
1120 
1121 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1122 			continue;
1123 
1124 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1125 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1126 			ND6_UNLOCK();
1127 			return 1;
1128 		}
1129 	}
1130 
1131 	/*
1132 	 * If the default router list is empty, all addresses are regarded
1133 	 * as on-link, and thus, as a neighbor.
1134 	 * XXX: we restrict the condition to hosts, because routers usually do
1135 	 * not have the "default router list".
1136 	 */
1137 	if (!ip6_forwarding && ND_DEFROUTER_LIST_EMPTY() &&
1138 	    nd6_defifindex == ifp->if_index) {
1139 		ND6_UNLOCK();
1140 		return 1;
1141 	}
1142 	ND6_UNLOCK();
1143 
1144 	if (nd6_is_new_addr_neighbor(addr, ifp))
1145 		return 1;
1146 
1147 	/*
1148 	 * Even if the address matches none of our addresses, it might be
1149 	 * in the neighbor cache or a connected route.
1150 	 */
1151 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1152 	if (ln != NULL) {
1153 		LLE_RUNLOCK(ln);
1154 		return 1;
1155 	}
1156 
1157 	rt = rtalloc1(sin6tocsa(addr), 0);
1158 	if (rt == NULL)
1159 		return 0;
1160 
1161 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1162 #if NBRIDGE > 0
1163 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1164 #endif
1165 #if NCARP > 0
1166 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1167 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1168 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1169 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1170 #endif
1171 	    )) {
1172 		rt_unref(rt);
1173 		return 1;
1174 	}
1175 	rt_unref(rt);
1176 
1177 	return 0;
1178 }
1179 
1180 /*
1181  * Free an nd6 llinfo entry.
1182  * Since the function would cause significant changes in the kernel, DO NOT
1183  * make it global, unless you have a strong reason for the change, and are sure
1184  * that the change is safe.
1185  */
1186 static void
1187 nd6_free(struct llentry *ln, int gc)
1188 {
1189 	struct nd_defrouter *dr;
1190 	struct ifnet *ifp;
1191 	struct in6_addr *in6;
1192 
1193 	KASSERT(ln != NULL);
1194 	LLE_WLOCK_ASSERT(ln);
1195 
1196 	ifp = ln->lle_tbl->llt_ifp;
1197 	in6 = &ln->r_l3addr.addr6;
1198 	/*
1199 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1200 	 * even though it is not harmful, it was not really necessary.
1201 	 */
1202 
1203 	if (!ip6_forwarding) {
1204 		ND6_WLOCK();
1205 		dr = nd6_defrouter_lookup(in6, ifp);
1206 
1207 		if (dr != NULL && dr->expire &&
1208 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1209 			/*
1210 			 * If the reason for the deletion is just garbage
1211 			 * collection, and the neighbor is an active default
1212 			 * router, do not delete it.  Instead, reset the GC
1213 			 * timer using the router's lifetime.
1214 			 * Simply deleting the entry would affect default
1215 			 * router selection, which is not necessarily a good
1216 			 * thing, especially when we're using router preference
1217 			 * values.
1218 			 * XXX: the check for ln_state would be redundant,
1219 			 *      but we intentionally keep it just in case.
1220 			 */
1221 			if (dr->expire > time_uptime)
1222 				nd6_llinfo_settimer(ln,
1223 				    (dr->expire - time_uptime) * hz);
1224 			else
1225 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1226 			ND6_UNLOCK();
1227 			LLE_WUNLOCK(ln);
1228 			return;
1229 		}
1230 
1231 		if (ln->ln_router || dr) {
1232 			/*
1233 			 * We need to unlock to avoid a LOR with nd6_rt_flush()
1234 			 * with the rnh and for the calls to
1235 			 * nd6_pfxlist_onlink_check() and nd6_defrouter_select() in the
1236 			 * block further down for calls into nd6_lookup().
1237 			 * We still hold a ref.
1238 			 */
1239 			LLE_WUNLOCK(ln);
1240 
1241 			/*
1242 			 * nd6_rt_flush must be called whether or not the neighbor
1243 			 * is in the Default Router List.
1244 			 * See a corresponding comment in nd6_na_input().
1245 			 */
1246 			nd6_rt_flush(in6, ifp);
1247 		}
1248 
1249 		if (dr) {
1250 			/*
1251 			 * Unreachablity of a router might affect the default
1252 			 * router selection and on-link detection of advertised
1253 			 * prefixes.
1254 			 */
1255 
1256 			/*
1257 			 * Temporarily fake the state to choose a new default
1258 			 * router and to perform on-link determination of
1259 			 * prefixes correctly.
1260 			 * Below the state will be set correctly,
1261 			 * or the entry itself will be deleted.
1262 			 */
1263 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1264 
1265 			/*
1266 			 * Since nd6_defrouter_select() does not affect the
1267 			 * on-link determination and MIP6 needs the check
1268 			 * before the default router selection, we perform
1269 			 * the check now.
1270 			 */
1271 			nd6_pfxlist_onlink_check();
1272 
1273 			/*
1274 			 * refresh default router list
1275 			 */
1276 			nd6_defrouter_select();
1277 		}
1278 
1279 #ifdef __FreeBSD__
1280 		/*
1281 		 * If this entry was added by an on-link redirect, remove the
1282 		 * corresponding host route.
1283 		 */
1284 		if (ln->la_flags & LLE_REDIRECT)
1285 			nd6_free_redirect(ln);
1286 #endif
1287 		ND6_UNLOCK();
1288 
1289 		if (ln->ln_router || dr)
1290 			LLE_WLOCK(ln);
1291 	}
1292 
1293 	/*
1294 	 * Save to unlock. We still hold an extra reference and will not
1295 	 * free(9) in llentry_free() if someone else holds one as well.
1296 	 */
1297 	LLE_WUNLOCK(ln);
1298 	IF_AFDATA_LOCK(ifp);
1299 	LLE_WLOCK(ln);
1300 
1301 	lltable_free_entry(LLTABLE6(ifp), ln);
1302 
1303 	IF_AFDATA_UNLOCK(ifp);
1304 }
1305 
1306 /*
1307  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1308  *
1309  * XXX cost-effective methods?
1310  */
1311 void
1312 nd6_nud_hint(struct rtentry *rt)
1313 {
1314 	struct llentry *ln;
1315 	struct ifnet *ifp;
1316 
1317 	if (rt == NULL)
1318 		return;
1319 
1320 	ifp = rt->rt_ifp;
1321 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1322 	if (ln == NULL)
1323 		return;
1324 
1325 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1326 		goto done;
1327 
1328 	/*
1329 	 * if we get upper-layer reachability confirmation many times,
1330 	 * it is possible we have false information.
1331 	 */
1332 	ln->ln_byhint++;
1333 	if (ln->ln_byhint > nd6_maxnudhint)
1334 		goto done;
1335 
1336 	ln->ln_state = ND6_LLINFO_REACHABLE;
1337 	if (!ND6_LLINFO_PERMANENT(ln))
1338 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1339 
1340 done:
1341 	LLE_WUNLOCK(ln);
1342 
1343 	return;
1344 }
1345 
1346 struct gc_args {
1347 	int gc_entries;
1348 	const struct in6_addr *skip_in6;
1349 };
1350 
1351 static int
1352 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1353 {
1354 	struct gc_args *args = farg;
1355 	int *n = &args->gc_entries;
1356 	const struct in6_addr *skip_in6 = args->skip_in6;
1357 
1358 	if (*n <= 0)
1359 		return 0;
1360 
1361 	if (ND6_LLINFO_PERMANENT(ln))
1362 		return 0;
1363 
1364 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1365 		return 0;
1366 
1367 	LLE_WLOCK(ln);
1368 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1369 		ln->ln_state = ND6_LLINFO_STALE;
1370 	else
1371 		ln->ln_state = ND6_LLINFO_PURGE;
1372 	nd6_llinfo_settimer(ln, 0);
1373 	LLE_WUNLOCK(ln);
1374 
1375 	(*n)--;
1376 	return 0;
1377 }
1378 
1379 static void
1380 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1381 {
1382 
1383 	if (ip6_neighborgcthresh >= 0 &&
1384 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1385 		struct gc_args gc_args = {10, in6};
1386 		/*
1387 		 * XXX entries that are "less recently used" should be
1388 		 * freed first.
1389 		 */
1390 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1391 	}
1392 }
1393 
1394 void
1395 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1396 {
1397 	struct sockaddr *gate = rt->rt_gateway;
1398 	struct ifnet *ifp = rt->rt_ifp;
1399 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1400 	struct ifaddr *ifa;
1401 
1402 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1403 
1404 	if (req == RTM_LLINFO_UPD) {
1405 		int rc;
1406 		struct in6_addr *in6;
1407 		struct in6_addr in6_all;
1408 		int anycast;
1409 
1410 		if ((ifa = info->rti_ifa) == NULL)
1411 			return;
1412 
1413 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1414 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1415 
1416 		in6_all = in6addr_linklocal_allnodes;
1417 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1418 			log(LOG_ERR, "%s: failed to set scope %s "
1419 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1420 			return;
1421 		}
1422 
1423 		/* XXX don't set Override for proxy addresses */
1424 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1425 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1426 #if 0
1427 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1428 #endif
1429 		    , 1, NULL);
1430 		return;
1431 	}
1432 
1433 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1434 		return;
1435 
1436 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1437 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1438 		/*
1439 		 * This is probably an interface direct route for a link
1440 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1441 		 * We do not need special treatment below for such a route.
1442 		 * Moreover, the RTF_LLINFO flag which would be set below
1443 		 * would annoy the ndp(8) command.
1444 		 */
1445 		return;
1446 	}
1447 
1448 	switch (req) {
1449 	case RTM_ADD: {
1450 		struct psref psref;
1451 
1452 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1453 		/*
1454 		 * There is no backward compatibility :)
1455 		 *
1456 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1457 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1458 		 *	   rt->rt_flags |= RTF_CLONING;
1459 		 */
1460 		/* XXX should move to route.c? */
1461 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1462 			union {
1463 				struct sockaddr sa;
1464 				struct sockaddr_dl sdl;
1465 				struct sockaddr_storage ss;
1466 			} u;
1467 			/*
1468 			 * Case 1: This route should come from a route to
1469 			 * interface (RTF_CLONING case) or the route should be
1470 			 * treated as on-link but is currently not
1471 			 * (RTF_LLINFO && ln == NULL case).
1472 			 */
1473 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1474 			    ifp->if_index, ifp->if_type,
1475 			    NULL, namelen, NULL, addrlen) == NULL) {
1476 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1477 				    "failed on %s\n", __func__, __LINE__,
1478 				    sizeof(u.ss), if_name(ifp));
1479 			}
1480 			rt_setgate(rt, &u.sa);
1481 			gate = rt->rt_gateway;
1482 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1483 			if (gate == NULL) {
1484 				log(LOG_ERR,
1485 				    "%s: rt_setgate failed on %s\n", __func__,
1486 				    if_name(ifp));
1487 				break;
1488 			}
1489 
1490 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1491 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1492 				break;
1493 		}
1494 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1495 		/*
1496 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1497 		 * We don't do that here since llinfo is not ready yet.
1498 		 *
1499 		 * There are also couple of other things to be discussed:
1500 		 * - unsolicited NA code needs improvement beforehand
1501 		 * - RFC2461 says we MAY send multicast unsolicited NA
1502 		 *   (7.2.6 paragraph 4), however, it also says that we
1503 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1504 		 *   we don't have anything like it right now.
1505 		 *   note that the mechanism needs a mutual agreement
1506 		 *   between proxies, which means that we need to implement
1507 		 *   a new protocol, or a new kludge.
1508 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1509 		 *   we need to check ip6forwarding before sending it.
1510 		 *   (or should we allow proxy ND configuration only for
1511 		 *   routers?  there's no mention about proxy ND from hosts)
1512 		 */
1513 #if 0
1514 		/* XXX it does not work */
1515 		if (rt->rt_flags & RTF_ANNOUNCE)
1516 			nd6_na_output(ifp,
1517 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1518 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1519 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1520 			      1, NULL);
1521 #endif
1522 
1523 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1524 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1525 			/*
1526 			 * Address resolution isn't necessary for a point to
1527 			 * point link, so we can skip this test for a p2p link.
1528 			 */
1529 			if (gate->sa_family != AF_LINK ||
1530 			    gate->sa_len <
1531 			    sockaddr_dl_measure(namelen, addrlen)) {
1532 				log(LOG_DEBUG,
1533 				    "nd6_rtrequest: bad gateway value: %s\n",
1534 				    if_name(ifp));
1535 				break;
1536 			}
1537 			satosdl(gate)->sdl_type = ifp->if_type;
1538 			satosdl(gate)->sdl_index = ifp->if_index;
1539 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1540 		}
1541 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1542 
1543 		/*
1544 		 * When called from rt_ifa_addlocal, we cannot depend on that
1545 		 * the address (rt_getkey(rt)) exits in the address list of the
1546 		 * interface. So check RTF_LOCAL instead.
1547 		 */
1548 		if (rt->rt_flags & RTF_LOCAL) {
1549 			if (nd6_useloopback)
1550 				rt->rt_ifp = lo0ifp;	/* XXX */
1551 			break;
1552 		}
1553 
1554 		/*
1555 		 * check if rt_getkey(rt) is an address assigned
1556 		 * to the interface.
1557 		 */
1558 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp,
1559 		    &satocsin6(rt_getkey(rt))->sin6_addr, &psref);
1560 		if (ifa != NULL) {
1561 			if (nd6_useloopback) {
1562 				rt->rt_ifp = lo0ifp;	/* XXX */
1563 				/*
1564 				 * Make sure rt_ifa be equal to the ifaddr
1565 				 * corresponding to the address.
1566 				 * We need this because when we refer
1567 				 * rt_ifa->ia6_flags in ip6_input, we assume
1568 				 * that the rt_ifa points to the address instead
1569 				 * of the loopback address.
1570 				 */
1571 				if (ifa != rt->rt_ifa)
1572 					rt_replace_ifa(rt, ifa);
1573 			}
1574 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1575 			/* join solicited node multicast for proxy ND */
1576 			if (ifp->if_flags & IFF_MULTICAST) {
1577 				struct in6_addr llsol;
1578 				int error;
1579 
1580 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1581 				llsol.s6_addr32[0] = htonl(0xff020000);
1582 				llsol.s6_addr32[1] = 0;
1583 				llsol.s6_addr32[2] = htonl(1);
1584 				llsol.s6_addr8[12] = 0xff;
1585 				if (in6_setscope(&llsol, ifp, NULL))
1586 					goto out;
1587 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1588 					char ip6buf[INET6_ADDRSTRLEN];
1589 					nd6log(LOG_ERR, "%s: failed to join "
1590 					    "%s (errno=%d)\n", if_name(ifp),
1591 					    IN6_PRINT(ip6buf, &llsol), error);
1592 				}
1593 			}
1594 		}
1595 	out:
1596 		ifa_release(ifa, &psref);
1597 		/*
1598 		 * If we have too many cache entries, initiate immediate
1599 		 * purging for some entries.
1600 		 */
1601 		if (rt->rt_ifp != NULL)
1602 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1603 		break;
1604 	    }
1605 
1606 	case RTM_DELETE:
1607 		/* leave from solicited node multicast for proxy ND */
1608 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1609 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1610 			struct in6_addr llsol;
1611 			struct in6_multi *in6m;
1612 
1613 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1614 			llsol.s6_addr32[0] = htonl(0xff020000);
1615 			llsol.s6_addr32[1] = 0;
1616 			llsol.s6_addr32[2] = htonl(1);
1617 			llsol.s6_addr8[12] = 0xff;
1618 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1619 				in6m = in6_lookup_multi(&llsol, ifp);
1620 				if (in6m)
1621 					in6_delmulti(in6m);
1622 			}
1623 		}
1624 		break;
1625 	}
1626 }
1627 
1628 int
1629 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1630 {
1631 	struct in6_drlist *drl = (struct in6_drlist *)data;
1632 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1633 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1634 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1635 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1636 	struct nd_defrouter *dr;
1637 	struct nd_prefix *pr;
1638 	int i = 0, error = 0;
1639 
1640 	switch (cmd) {
1641 	case SIOCGDRLST_IN6:
1642 		/*
1643 		 * obsolete API, use sysctl under net.inet6.icmp6
1644 		 */
1645 		memset(drl, 0, sizeof(*drl));
1646 		ND6_RLOCK();
1647 		ND_DEFROUTER_LIST_FOREACH(dr) {
1648 			if (i >= DRLSTSIZ)
1649 				break;
1650 			drl->defrouter[i].rtaddr = dr->rtaddr;
1651 			in6_clearscope(&drl->defrouter[i].rtaddr);
1652 
1653 			drl->defrouter[i].flags = dr->flags;
1654 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1655 			drl->defrouter[i].expire = dr->expire ?
1656 			    time_mono_to_wall(dr->expire) : 0;
1657 			drl->defrouter[i].if_index = dr->ifp->if_index;
1658 			i++;
1659 		}
1660 		ND6_UNLOCK();
1661 		break;
1662 	case SIOCGPRLST_IN6:
1663 		/*
1664 		 * obsolete API, use sysctl under net.inet6.icmp6
1665 		 *
1666 		 * XXX the structure in6_prlist was changed in backward-
1667 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1668 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1669 		 */
1670 		/*
1671 		 * XXX meaning of fields, especialy "raflags", is very
1672 		 * differnet between RA prefix list and RR/static prefix list.
1673 		 * how about separating ioctls into two?
1674 		 */
1675 		memset(oprl, 0, sizeof(*oprl));
1676 		ND6_RLOCK();
1677 		ND_PREFIX_LIST_FOREACH(pr) {
1678 			struct nd_pfxrouter *pfr;
1679 			int j;
1680 
1681 			if (i >= PRLSTSIZ)
1682 				break;
1683 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1684 			oprl->prefix[i].raflags = pr->ndpr_raf;
1685 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1686 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1687 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1688 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1689 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1690 				oprl->prefix[i].expire = 0;
1691 			else {
1692 				time_t maxexpire;
1693 
1694 				/* XXX: we assume time_t is signed. */
1695 				maxexpire = (-1) &
1696 				    ~((time_t)1 <<
1697 				    ((sizeof(maxexpire) * 8) - 1));
1698 				if (pr->ndpr_vltime <
1699 				    maxexpire - pr->ndpr_lastupdate) {
1700 					time_t expire;
1701 					expire = pr->ndpr_lastupdate +
1702 					    pr->ndpr_vltime;
1703 					oprl->prefix[i].expire = expire ?
1704 					    time_mono_to_wall(expire) : 0;
1705 				} else
1706 					oprl->prefix[i].expire = maxexpire;
1707 			}
1708 
1709 			j = 0;
1710 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1711 				if (j < DRLSTSIZ) {
1712 #define RTRADDR oprl->prefix[i].advrtr[j]
1713 					RTRADDR = pfr->router->rtaddr;
1714 					in6_clearscope(&RTRADDR);
1715 #undef RTRADDR
1716 				}
1717 				j++;
1718 			}
1719 			oprl->prefix[i].advrtrs = j;
1720 			oprl->prefix[i].origin = PR_ORIG_RA;
1721 
1722 			i++;
1723 		}
1724 		ND6_UNLOCK();
1725 
1726 		break;
1727 	case OSIOCGIFINFO_IN6:
1728 #define ND	ndi->ndi
1729 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1730 		memset(&ND, 0, sizeof(ND));
1731 		ND.linkmtu = IN6_LINKMTU(ifp);
1732 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1733 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1734 		ND.reachable = ND_IFINFO(ifp)->reachable;
1735 		ND.retrans = ND_IFINFO(ifp)->retrans;
1736 		ND.flags = ND_IFINFO(ifp)->flags;
1737 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1738 		ND.chlim = ND_IFINFO(ifp)->chlim;
1739 		break;
1740 	case SIOCGIFINFO_IN6:
1741 		ND = *ND_IFINFO(ifp);
1742 		break;
1743 	case SIOCSIFINFO_IN6:
1744 		/*
1745 		 * used to change host variables from userland.
1746 		 * intented for a use on router to reflect RA configurations.
1747 		 */
1748 		/* 0 means 'unspecified' */
1749 		if (ND.linkmtu != 0) {
1750 			if (ND.linkmtu < IPV6_MMTU ||
1751 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1752 				error = EINVAL;
1753 				break;
1754 			}
1755 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1756 		}
1757 
1758 		if (ND.basereachable != 0) {
1759 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1760 
1761 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1762 			if (ND.basereachable != obasereachable)
1763 				ND_IFINFO(ifp)->reachable =
1764 				    ND_COMPUTE_RTIME(ND.basereachable);
1765 		}
1766 		if (ND.retrans != 0)
1767 			ND_IFINFO(ifp)->retrans = ND.retrans;
1768 		if (ND.chlim != 0)
1769 			ND_IFINFO(ifp)->chlim = ND.chlim;
1770 		/* FALLTHROUGH */
1771 	case SIOCSIFINFO_FLAGS:
1772 	{
1773 		struct ifaddr *ifa;
1774 		struct in6_ifaddr *ia;
1775 		int s;
1776 
1777 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1778 		    !(ND.flags & ND6_IFF_IFDISABLED))
1779 		{
1780 			/*
1781 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1782 			 * has a link-local address with IN6_IFF_DUPLICATED,
1783 			 * do not clear ND6_IFF_IFDISABLED.
1784 			 * See RFC 4862, section 5.4.5.
1785 			 */
1786 			int duplicated_linklocal = 0;
1787 
1788 			s = pserialize_read_enter();
1789 			IFADDR_READER_FOREACH(ifa, ifp) {
1790 				if (ifa->ifa_addr->sa_family != AF_INET6)
1791 					continue;
1792 				ia = (struct in6_ifaddr *)ifa;
1793 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1794 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1795 				{
1796 					duplicated_linklocal = 1;
1797 					break;
1798 				}
1799 			}
1800 			pserialize_read_exit(s);
1801 
1802 			if (duplicated_linklocal) {
1803 				ND.flags |= ND6_IFF_IFDISABLED;
1804 				log(LOG_ERR, "Cannot enable an interface"
1805 				    " with a link-local address marked"
1806 				    " duplicate.\n");
1807 			} else {
1808 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1809 				if (ifp->if_flags & IFF_UP)
1810 					in6_if_up(ifp);
1811 			}
1812 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1813 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1814 			int bound = curlwp_bind();
1815 			/* Mark all IPv6 addresses as tentative. */
1816 
1817 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1818 			s = pserialize_read_enter();
1819 			IFADDR_READER_FOREACH(ifa, ifp) {
1820 				struct psref psref;
1821 				if (ifa->ifa_addr->sa_family != AF_INET6)
1822 					continue;
1823 				ifa_acquire(ifa, &psref);
1824 				pserialize_read_exit(s);
1825 
1826 				nd6_dad_stop(ifa);
1827 
1828 				ia = (struct in6_ifaddr *)ifa;
1829 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1830 
1831 				s = pserialize_read_enter();
1832 				ifa_release(ifa, &psref);
1833 			}
1834 			pserialize_read_exit(s);
1835 			curlwp_bindx(bound);
1836 		}
1837 
1838 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1839 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1840 				/* auto_linklocal 0->1 transition */
1841 
1842 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1843 				in6_ifattach(ifp, NULL);
1844 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1845 			    ifp->if_flags & IFF_UP)
1846 			{
1847 				/*
1848 				 * When the IF already has
1849 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1850 				 * address is assigned, and IFF_UP, try to
1851 				 * assign one.
1852 				 */
1853 				int haslinklocal = 0;
1854 
1855 				s = pserialize_read_enter();
1856 				IFADDR_READER_FOREACH(ifa, ifp) {
1857 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1858 						continue;
1859 					ia = (struct in6_ifaddr *)ifa;
1860 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1861 						haslinklocal = 1;
1862 						break;
1863 					}
1864 				}
1865 				pserialize_read_exit(s);
1866 				if (!haslinklocal)
1867 					in6_ifattach(ifp, NULL);
1868 			}
1869 		}
1870 	}
1871 		ND_IFINFO(ifp)->flags = ND.flags;
1872 		break;
1873 #undef ND
1874 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1875 		/* sync kernel routing table with the default router list */
1876 		ND6_WLOCK();
1877 		nd6_defrouter_reset();
1878 		nd6_defrouter_select();
1879 		ND6_UNLOCK();
1880 		break;
1881 	case SIOCSPFXFLUSH_IN6:
1882 	{
1883 		/* flush all the prefix advertised by routers */
1884 		struct nd_prefix *pfx, *next;
1885 
1886 	restart:
1887 		ND6_WLOCK();
1888 		ND_PREFIX_LIST_FOREACH_SAFE(pfx, next) {
1889 			struct in6_ifaddr *ia, *ia_next;
1890 			int _s;
1891 
1892 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1893 				continue; /* XXX */
1894 
1895 			/* do we really have to remove addresses as well? */
1896 			_s = pserialize_read_enter();
1897 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1898 			     ia = ia_next) {
1899 				struct ifnet *ifa_ifp;
1900 				int bound;
1901 				struct psref psref;
1902 
1903 				/* ia might be removed.  keep the next ptr. */
1904 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1905 
1906 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1907 					continue;
1908 
1909 				if (ia->ia6_ndpr != pfx)
1910 					continue;
1911 
1912 				bound = curlwp_bind();
1913 				ia6_acquire(ia, &psref);
1914 				pserialize_read_exit(_s);
1915 				ND6_UNLOCK();
1916 
1917 				ifa_ifp = ia->ia_ifa.ifa_ifp;
1918 				if (ifa_ifp == ifp) {
1919 					/* Already have IFNET_LOCK(ifp) */
1920 					KASSERT(!if_is_deactivated(ifp));
1921 					ia6_release(ia, &psref);
1922 					in6_purgeaddr(&ia->ia_ifa);
1923 					curlwp_bindx(bound);
1924 					goto restart;
1925 				}
1926 				IFNET_LOCK(ifa_ifp);
1927 				/*
1928 				 * Need to take the lock first to prevent
1929 				 * if_detach from running in6_purgeaddr
1930 				 * concurrently.
1931 				 */
1932 				if (!if_is_deactivated(ifa_ifp)) {
1933 					ia6_release(ia, &psref);
1934 					in6_purgeaddr(&ia->ia_ifa);
1935 				} else {
1936 					/*
1937 					 * ifp is being destroyed, ia will be
1938 					 * destroyed by if_detach.
1939 					 */
1940 					ia6_release(ia, &psref);
1941 					/* XXX may cause busy loop */
1942 				}
1943 				IFNET_UNLOCK(ifa_ifp);
1944 				curlwp_bindx(bound);
1945 				goto restart;
1946 			}
1947 			pserialize_read_exit(_s);
1948 
1949 			KASSERT(pfx->ndpr_refcnt == 0);
1950 			nd6_prelist_remove(pfx);
1951 		}
1952 		ND6_UNLOCK();
1953 		break;
1954 	}
1955 	case SIOCSRTRFLUSH_IN6:
1956 	{
1957 		/* flush all the default routers */
1958 		struct nd_defrouter *drtr, *next;
1959 
1960 		ND6_WLOCK();
1961 		nd6_defrouter_reset();
1962 		ND_DEFROUTER_LIST_FOREACH_SAFE(drtr, next) {
1963 			nd6_defrtrlist_del(drtr, NULL);
1964 		}
1965 		nd6_defrouter_select();
1966 		ND6_UNLOCK();
1967 		break;
1968 	}
1969 	case SIOCGNBRINFO_IN6:
1970 	{
1971 		struct llentry *ln;
1972 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1973 
1974 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1975 			return error;
1976 
1977 		ln = nd6_lookup(&nb_addr, ifp, false);
1978 		if (ln == NULL) {
1979 			error = EINVAL;
1980 			break;
1981 		}
1982 		nbi->state = ln->ln_state;
1983 		nbi->asked = ln->ln_asked;
1984 		nbi->isrouter = ln->ln_router;
1985 		nbi->expire = ln->ln_expire ?
1986 		    time_mono_to_wall(ln->ln_expire) : 0;
1987 		LLE_RUNLOCK(ln);
1988 
1989 		break;
1990 	}
1991 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1992 		ndif->ifindex = nd6_defifindex;
1993 		break;
1994 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1995 		return nd6_setdefaultiface(ndif->ifindex);
1996 	}
1997 	return error;
1998 }
1999 
2000 void
2001 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
2002 {
2003 	struct mbuf *m_hold, *m_hold_next;
2004 	struct sockaddr_in6 sin6;
2005 
2006 	LLE_WLOCK_ASSERT(ln);
2007 
2008 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
2009 
2010 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
2011 
2012 	LLE_WUNLOCK(ln);
2013 	for (; m_hold != NULL; m_hold = m_hold_next) {
2014 		m_hold_next = m_hold->m_nextpkt;
2015 		m_hold->m_nextpkt = NULL;
2016 
2017 		/*
2018 		 * we assume ifp is not a p2p here, so
2019 		 * just set the 2nd argument as the
2020 		 * 1st one.
2021 		 */
2022 		ip6_if_output(ifp, ifp, m_hold, &sin6, NULL);
2023 	}
2024 	LLE_WLOCK(ln);
2025 }
2026 
2027 /*
2028  * Create neighbor cache entry and cache link-layer address,
2029  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
2030  */
2031 void
2032 nd6_cache_lladdr(
2033     struct ifnet *ifp,
2034     struct in6_addr *from,
2035     char *lladdr,
2036     int lladdrlen,
2037     int type,	/* ICMP6 type */
2038     int code	/* type dependent information */
2039 )
2040 {
2041 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
2042 	struct llentry *ln = NULL;
2043 	int is_newentry;
2044 	int do_update;
2045 	int olladdr;
2046 	int llchange;
2047 	int newstate = 0;
2048 	uint16_t router = 0;
2049 
2050 	KASSERT(ifp != NULL);
2051 	KASSERT(from != NULL);
2052 
2053 	/* nothing must be updated for unspecified address */
2054 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2055 		return;
2056 
2057 	/*
2058 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2059 	 * the caller.
2060 	 *
2061 	 * XXX If the link does not have link-layer adderss, what should
2062 	 * we do? (ifp->if_addrlen == 0)
2063 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2064 	 * description on it in NS section (RFC 2461 7.2.3).
2065 	 */
2066 
2067 	ln = nd6_lookup(from, ifp, true);
2068 	if (ln == NULL) {
2069 #if 0
2070 		/* nothing must be done if there's no lladdr */
2071 		if (!lladdr || !lladdrlen)
2072 			return NULL;
2073 #endif
2074 
2075 		ln = nd6_create(from, ifp);
2076 		is_newentry = 1;
2077 	} else {
2078 		/* do nothing if static ndp is set */
2079 		if (ln->la_flags & LLE_STATIC) {
2080 			LLE_WUNLOCK(ln);
2081 			return;
2082 		}
2083 		is_newentry = 0;
2084 	}
2085 
2086 	if (ln == NULL)
2087 		return;
2088 
2089 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2090 	if (olladdr && lladdr) {
2091 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2092 	} else
2093 		llchange = 0;
2094 
2095 	/*
2096 	 * newentry olladdr  lladdr  llchange	(*=record)
2097 	 *	0	n	n	--	(1)
2098 	 *	0	y	n	--	(2)
2099 	 *	0	n	y	--	(3) * STALE
2100 	 *	0	y	y	n	(4) *
2101 	 *	0	y	y	y	(5) * STALE
2102 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2103 	 *	1	--	y	--	(7) * STALE
2104 	 */
2105 
2106 	if (lladdr) {		/* (3-5) and (7) */
2107 		/*
2108 		 * Record source link-layer address
2109 		 * XXX is it dependent to ifp->if_type?
2110 		 */
2111 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2112 		ln->la_flags |= LLE_VALID;
2113 	}
2114 
2115 	if (!is_newentry) {
2116 		if ((!olladdr && lladdr) ||		/* (3) */
2117 		    (olladdr && lladdr && llchange)) {	/* (5) */
2118 			do_update = 1;
2119 			newstate = ND6_LLINFO_STALE;
2120 		} else					/* (1-2,4) */
2121 			do_update = 0;
2122 	} else {
2123 		do_update = 1;
2124 		if (lladdr == NULL)			/* (6) */
2125 			newstate = ND6_LLINFO_NOSTATE;
2126 		else					/* (7) */
2127 			newstate = ND6_LLINFO_STALE;
2128 	}
2129 
2130 	if (do_update) {
2131 		/*
2132 		 * Update the state of the neighbor cache.
2133 		 */
2134 		ln->ln_state = newstate;
2135 
2136 		if (ln->ln_state == ND6_LLINFO_STALE) {
2137 			/*
2138 			 * XXX: since nd6_output() below will cause
2139 			 * state tansition to DELAY and reset the timer,
2140 			 * we must set the timer now, although it is actually
2141 			 * meaningless.
2142 			 */
2143 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2144 
2145 			nd6_llinfo_release_pkts(ln, ifp);
2146 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2147 			/* probe right away */
2148 			nd6_llinfo_settimer((void *)ln, 0);
2149 		}
2150 	}
2151 
2152 	/*
2153 	 * ICMP6 type dependent behavior.
2154 	 *
2155 	 * NS: clear IsRouter if new entry
2156 	 * RS: clear IsRouter
2157 	 * RA: set IsRouter if there's lladdr
2158 	 * redir: clear IsRouter if new entry
2159 	 *
2160 	 * RA case, (1):
2161 	 * The spec says that we must set IsRouter in the following cases:
2162 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2163 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2164 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2165 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2166 	 * neighbor cache, this is similar to (6).
2167 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2168 	 *
2169 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2170 	 *							D R
2171 	 *	0	n	n	--	(1)	c   ?     s
2172 	 *	0	y	n	--	(2)	c   s     s
2173 	 *	0	n	y	--	(3)	c   s     s
2174 	 *	0	y	y	n	(4)	c   s     s
2175 	 *	0	y	y	y	(5)	c   s     s
2176 	 *	1	--	n	--	(6) c	c 	c s
2177 	 *	1	--	y	--	(7) c	c   s	c s
2178 	 *
2179 	 *					(c=clear s=set)
2180 	 */
2181 	switch (type & 0xff) {
2182 	case ND_NEIGHBOR_SOLICIT:
2183 		/*
2184 		 * New entry must have is_router flag cleared.
2185 		 */
2186 		if (is_newentry)	/* (6-7) */
2187 			ln->ln_router = 0;
2188 		break;
2189 	case ND_REDIRECT:
2190 		/*
2191 		 * If the icmp is a redirect to a better router, always set the
2192 		 * is_router flag.  Otherwise, if the entry is newly created,
2193 		 * clear the flag.  [RFC 2461, sec 8.3]
2194 		 */
2195 		if (code == ND_REDIRECT_ROUTER)
2196 			ln->ln_router = 1;
2197 		else if (is_newentry) /* (6-7) */
2198 			ln->ln_router = 0;
2199 		break;
2200 	case ND_ROUTER_SOLICIT:
2201 		/*
2202 		 * is_router flag must always be cleared.
2203 		 */
2204 		ln->ln_router = 0;
2205 		break;
2206 	case ND_ROUTER_ADVERT:
2207 		/*
2208 		 * Mark an entry with lladdr as a router.
2209 		 */
2210 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2211 		    (is_newentry && lladdr)) {			/* (7) */
2212 			ln->ln_router = 1;
2213 		}
2214 		break;
2215 	}
2216 
2217 #if 0
2218 	/* XXX should we send rtmsg as it used to be? */
2219 	if (do_update)
2220 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2221 #endif
2222 
2223 	if (ln != NULL) {
2224 		router = ln->ln_router;
2225 		LLE_WUNLOCK(ln);
2226 	}
2227 
2228 	/*
2229 	 * If we have too many cache entries, initiate immediate
2230 	 * purging for some entries.
2231 	 */
2232 	if (is_newentry)
2233 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2234 
2235 	/*
2236 	 * When the link-layer address of a router changes, select the
2237 	 * best router again.  In particular, when the neighbor entry is newly
2238 	 * created, it might affect the selection policy.
2239 	 * Question: can we restrict the first condition to the "is_newentry"
2240 	 * case?
2241 	 * XXX: when we hear an RA from a new router with the link-layer
2242 	 * address option, nd6_defrouter_select() is called twice, since
2243 	 * defrtrlist_update called the function as well.  However, I believe
2244 	 * we can compromise the overhead, since it only happens the first
2245 	 * time.
2246 	 * XXX: although nd6_defrouter_select() should not have a bad effect
2247 	 * for those are not autoconfigured hosts, we explicitly avoid such
2248 	 * cases for safety.
2249 	 */
2250 	if (do_update && router && !ip6_forwarding &&
2251 	    nd6_accepts_rtadv(ndi)) {
2252 		ND6_WLOCK();
2253 		nd6_defrouter_select();
2254 		ND6_UNLOCK();
2255 	}
2256 }
2257 
2258 static void
2259 nd6_slowtimo(void *ignored_arg)
2260 {
2261 	struct nd_ifinfo *nd6if;
2262 	struct ifnet *ifp;
2263 	int s;
2264 
2265 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2266 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2267 	    nd6_slowtimo, NULL);
2268 
2269 	s = pserialize_read_enter();
2270 	IFNET_READER_FOREACH(ifp) {
2271 		nd6if = ND_IFINFO(ifp);
2272 		if (nd6if->basereachable && /* already initialized */
2273 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2274 			/*
2275 			 * Since reachable time rarely changes by router
2276 			 * advertisements, we SHOULD insure that a new random
2277 			 * value gets recomputed at least once every few hours.
2278 			 * (RFC 2461, 6.3.4)
2279 			 */
2280 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2281 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2282 		}
2283 	}
2284 	pserialize_read_exit(s);
2285 
2286 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2287 }
2288 
2289 /*
2290  * Return 0 if a neighbor cache is found. Return EWOULDBLOCK if a cache is not
2291  * found and trying to resolve a neighbor; in this case the mbuf is queued in
2292  * the list. Otherwise return errno after freeing the mbuf.
2293  */
2294 int
2295 nd6_resolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
2296     const struct sockaddr *_dst, uint8_t *lldst, size_t dstsize)
2297 {
2298 	struct llentry *ln = NULL;
2299 	bool created = false;
2300 	const struct sockaddr_in6 *dst = satocsin6(_dst);
2301 
2302 	/* discard the packet if IPv6 operation is disabled on the interface */
2303 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2304 		m_freem(m);
2305 		return ENETDOWN; /* better error? */
2306 	}
2307 
2308 	/*
2309 	 * Address resolution or Neighbor Unreachability Detection
2310 	 * for the next hop.
2311 	 * At this point, the destination of the packet must be a unicast
2312 	 * or an anycast address(i.e. not a multicast).
2313 	 */
2314 
2315 	/* Look up the neighbor cache for the nexthop */
2316 	ln = nd6_lookup(&dst->sin6_addr, ifp, false);
2317 
2318 	if (ln != NULL && (ln->la_flags & LLE_VALID) != 0) {
2319 		KASSERT(ln->ln_state > ND6_LLINFO_INCOMPLETE);
2320 		/* Fast path */
2321 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2322 		LLE_RUNLOCK(ln);
2323 		return 0;
2324 	}
2325 	if (ln != NULL)
2326 		LLE_RUNLOCK(ln);
2327 
2328 	/* Slow path */
2329 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2330 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp))  {
2331 		struct sockaddr_in6 sin6;
2332 		/*
2333 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2334 		 * the condition below is not very efficient.  But we believe
2335 		 * it is tolerable, because this should be a rare case.
2336 		 */
2337 		ln = nd6_create(&dst->sin6_addr, ifp);
2338 		if (ln == NULL) {
2339 			char ip6buf[INET6_ADDRSTRLEN];
2340 			log(LOG_DEBUG,
2341 			    "%s: can't allocate llinfo for %s "
2342 			    "(ln=%p, rt=%p)\n", __func__,
2343 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
2344 			m_freem(m);
2345 			return ENOBUFS;
2346 		}
2347 
2348 		sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
2349 		rt_clonedmsg(sin6tosa(&sin6), ifp, rt);
2350 
2351 		created = true;
2352 	}
2353 
2354 	if (ln == NULL) {
2355 		m_freem(m);
2356 		return ENETDOWN; /* better error? */
2357 	}
2358 
2359 	LLE_WLOCK_ASSERT(ln);
2360 
2361 	/* We don't have to do link-layer address resolution on a p2p link. */
2362 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2363 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2364 		ln->ln_state = ND6_LLINFO_STALE;
2365 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2366 	}
2367 
2368 	/*
2369 	 * The first time we send a packet to a neighbor whose entry is
2370 	 * STALE, we have to change the state to DELAY and a sets a timer to
2371 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2372 	 * neighbor unreachability detection on expiration.
2373 	 * (RFC 2461 7.3.3)
2374 	 */
2375 	if (ln->ln_state == ND6_LLINFO_STALE) {
2376 		ln->ln_asked = 0;
2377 		ln->ln_state = ND6_LLINFO_DELAY;
2378 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2379 	}
2380 
2381 	/*
2382 	 * There is a neighbor cache entry, but no ethernet address
2383 	 * response yet.  Append this latest packet to the end of the
2384 	 * packet queue in the mbuf, unless the number of the packet
2385 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2386 	 * the oldest packet in the queue will be removed.
2387 	 */
2388 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2389 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2390 	if (ln->ln_hold) {
2391 		struct mbuf *m_hold;
2392 		int i;
2393 
2394 		i = 0;
2395 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2396 			i++;
2397 			if (m_hold->m_nextpkt == NULL) {
2398 				m_hold->m_nextpkt = m;
2399 				break;
2400 			}
2401 		}
2402 		while (i >= nd6_maxqueuelen) {
2403 			m_hold = ln->ln_hold;
2404 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2405 			m_freem(m_hold);
2406 			i--;
2407 		}
2408 	} else {
2409 		ln->ln_hold = m;
2410 	}
2411 
2412 	/*
2413 	 * If there has been no NS for the neighbor after entering the
2414 	 * INCOMPLETE state, send the first solicitation.
2415 	 */
2416 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2417 		struct in6_addr src, *psrc;
2418 
2419 		ln->ln_asked++;
2420 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2421 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2422 		LLE_WUNLOCK(ln);
2423 		ln = NULL;
2424 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, NULL);
2425 	} else {
2426 		/* We did the lookup so we need to do the unlock here. */
2427 		LLE_WUNLOCK(ln);
2428 	}
2429 
2430 	if (created)
2431 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2432 
2433 	return EWOULDBLOCK;
2434 }
2435 
2436 int
2437 nd6_need_cache(struct ifnet *ifp)
2438 {
2439 	/*
2440 	 * XXX: we currently do not make neighbor cache on any interface
2441 	 * other than ARCnet, Ethernet, FDDI and GIF.
2442 	 *
2443 	 * RFC2893 says:
2444 	 * - unidirectional tunnels needs no ND
2445 	 */
2446 	switch (ifp->if_type) {
2447 	case IFT_ARCNET:
2448 	case IFT_ETHER:
2449 	case IFT_FDDI:
2450 	case IFT_IEEE1394:
2451 	case IFT_CARP:
2452 	case IFT_GIF:		/* XXX need more cases? */
2453 	case IFT_PPP:
2454 	case IFT_TUNNEL:
2455 		return 1;
2456 	default:
2457 		return 0;
2458 	}
2459 }
2460 
2461 static void
2462 clear_llinfo_pqueue(struct llentry *ln)
2463 {
2464 	struct mbuf *m_hold, *m_hold_next;
2465 
2466 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2467 		m_hold_next = m_hold->m_nextpkt;
2468 		m_hold->m_nextpkt = NULL;
2469 		m_freem(m_hold);
2470 	}
2471 
2472 	ln->ln_hold = NULL;
2473 	return;
2474 }
2475 
2476 int
2477 nd6_sysctl(
2478     int name,
2479     void *oldp,	/* syscall arg, need copyout */
2480     size_t *oldlenp,
2481     void *newp,	/* syscall arg, need copyin */
2482     size_t newlen
2483 )
2484 {
2485 	int (*fill_func)(void *, size_t *);
2486 
2487 	if (newp)
2488 		return EPERM;
2489 
2490 	switch (name) {
2491 	case ICMPV6CTL_ND6_DRLIST:
2492 		fill_func = fill_drlist;
2493 		break;
2494 
2495 	case ICMPV6CTL_ND6_PRLIST:
2496 		fill_func = fill_prlist;
2497 		break;
2498 
2499 	case ICMPV6CTL_ND6_MAXQLEN:
2500 		return 0;
2501 
2502 	default:
2503 		return ENOPROTOOPT;
2504 	}
2505 
2506 	if (oldlenp == NULL)
2507 		return EINVAL;
2508 
2509 	size_t ol;
2510 	int error = (*fill_func)(NULL, &ol);	/* calc len needed */
2511 	if (error)
2512 		return error;
2513 
2514 	if (oldp == NULL) {
2515 		*oldlenp = ol;
2516 		return 0;
2517 	}
2518 
2519 	ol = *oldlenp = min(ol, *oldlenp);
2520 	if (ol == 0)
2521 		return 0;
2522 
2523 	void *p = kmem_alloc(ol, KM_SLEEP);
2524 	error = (*fill_func)(p, oldlenp);
2525 	if (!error)
2526 		error = copyout(p, oldp, *oldlenp);
2527 	kmem_free(p, ol);
2528 
2529 	return error;
2530 }
2531 
2532 static int
2533 fill_drlist(void *oldp, size_t *oldlenp)
2534 {
2535 	int error = 0;
2536 	struct in6_defrouter *d = NULL, *de = NULL;
2537 	struct nd_defrouter *dr;
2538 	size_t l;
2539 
2540 	if (oldp) {
2541 		d = (struct in6_defrouter *)oldp;
2542 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2543 	}
2544 	l = 0;
2545 
2546 	ND6_RLOCK();
2547 	ND_DEFROUTER_LIST_FOREACH(dr) {
2548 
2549 		if (oldp && d + 1 <= de) {
2550 			memset(d, 0, sizeof(*d));
2551 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2552 			if (sa6_recoverscope(&d->rtaddr)) {
2553 				char ip6buf[INET6_ADDRSTRLEN];
2554 				log(LOG_ERR,
2555 				    "scope error in router list (%s)\n",
2556 				    IN6_PRINT(ip6buf, &d->rtaddr.sin6_addr));
2557 				/* XXX: press on... */
2558 			}
2559 			d->flags = dr->flags;
2560 			d->rtlifetime = dr->rtlifetime;
2561 			d->expire = dr->expire ?
2562 			    time_mono_to_wall(dr->expire) : 0;
2563 			d->if_index = dr->ifp->if_index;
2564 		}
2565 
2566 		l += sizeof(*d);
2567 		if (d)
2568 			d++;
2569 	}
2570 	ND6_UNLOCK();
2571 
2572 	*oldlenp = l;	/* (void *)d - (void *)oldp */
2573 
2574 	return error;
2575 }
2576 
2577 static int
2578 fill_prlist(void *oldp, size_t *oldlenp)
2579 {
2580 	int error = 0;
2581 	struct nd_prefix *pr;
2582 	uint8_t *p = NULL, *ps = NULL;
2583 	uint8_t *pe = NULL;
2584 	size_t l;
2585 	char ip6buf[INET6_ADDRSTRLEN];
2586 
2587 	if (oldp) {
2588 		ps = p = (uint8_t*)oldp;
2589 		pe = (uint8_t*)oldp + *oldlenp;
2590 	}
2591 	l = 0;
2592 
2593 	ND6_RLOCK();
2594 	ND_PREFIX_LIST_FOREACH(pr) {
2595 		u_short advrtrs;
2596 		struct sockaddr_in6 sin6;
2597 		struct nd_pfxrouter *pfr;
2598 		struct in6_prefix pfx;
2599 
2600 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2601 		{
2602 			memset(&pfx, 0, sizeof(pfx));
2603 			ps = p;
2604 			pfx.prefix = pr->ndpr_prefix;
2605 
2606 			if (sa6_recoverscope(&pfx.prefix)) {
2607 				log(LOG_ERR,
2608 				    "scope error in prefix list (%s)\n",
2609 				    IN6_PRINT(ip6buf, &pfx.prefix.sin6_addr));
2610 				/* XXX: press on... */
2611 			}
2612 			pfx.raflags = pr->ndpr_raf;
2613 			pfx.prefixlen = pr->ndpr_plen;
2614 			pfx.vltime = pr->ndpr_vltime;
2615 			pfx.pltime = pr->ndpr_pltime;
2616 			pfx.if_index = pr->ndpr_ifp->if_index;
2617 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2618 				pfx.expire = 0;
2619 			else {
2620 				time_t maxexpire;
2621 
2622 				/* XXX: we assume time_t is signed. */
2623 				maxexpire = (-1) &
2624 				    ~((time_t)1 <<
2625 				    ((sizeof(maxexpire) * 8) - 1));
2626 				if (pr->ndpr_vltime <
2627 				    maxexpire - pr->ndpr_lastupdate) {
2628 					pfx.expire = pr->ndpr_lastupdate +
2629 						pr->ndpr_vltime;
2630 				} else
2631 					pfx.expire = maxexpire;
2632 			}
2633 			pfx.refcnt = pr->ndpr_refcnt;
2634 			pfx.flags = pr->ndpr_stateflags;
2635 			pfx.origin = PR_ORIG_RA;
2636 
2637 			p += sizeof(pfx); l += sizeof(pfx);
2638 
2639 			advrtrs = 0;
2640 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2641 				if (p + sizeof(sin6) > pe) {
2642 					advrtrs++;
2643 					continue;
2644 				}
2645 
2646 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2647 					    0, 0, 0);
2648 				if (sa6_recoverscope(&sin6)) {
2649 					log(LOG_ERR,
2650 					    "scope error in "
2651 					    "prefix list (%s)\n",
2652 					    IN6_PRINT(ip6buf,
2653 					    &pfr->router->rtaddr));
2654 				}
2655 				advrtrs++;
2656 				memcpy(p, &sin6, sizeof(sin6));
2657 				p += sizeof(sin6);
2658 				l += sizeof(sin6);
2659 			}
2660 			pfx.advrtrs = advrtrs;
2661 			memcpy(ps, &pfx, sizeof(pfx));
2662 		}
2663 		else {
2664 			l += sizeof(pfx);
2665 			advrtrs = 0;
2666 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2667 				advrtrs++;
2668 				l += sizeof(sin6);
2669 			}
2670 		}
2671 	}
2672 	ND6_UNLOCK();
2673 
2674 	*oldlenp = l;
2675 
2676 	return error;
2677 }
2678 
2679 static int
2680 nd6_setdefaultiface(int ifindex)
2681 {
2682 	ifnet_t *ifp;
2683 	int error = 0;
2684 	int s;
2685 
2686 	s = pserialize_read_enter();
2687 	ifp = if_byindex(ifindex);
2688 	if (ifp == NULL) {
2689 		pserialize_read_exit(s);
2690 		return EINVAL;
2691 	}
2692 	if (nd6_defifindex != ifindex) {
2693 		nd6_defifindex = ifindex;
2694 		nd6_defifp = nd6_defifindex > 0 ? ifp : NULL;
2695 
2696 		/*
2697 		 * Our current implementation assumes one-to-one maping between
2698 		 * interfaces and links, so it would be natural to use the
2699 		 * default interface as the default link.
2700 		 */
2701 		scope6_setdefault(nd6_defifp);
2702 	}
2703 	pserialize_read_exit(s);
2704 
2705 	return (error);
2706 }
2707