xref: /netbsd-src/sys/netinet6/nd6.c (revision 7330f729ccf0bd976a06f95fad452fe774fc7fd1)
1 /*	$NetBSD: nd6.c,v 1.265 2019/09/25 09:53:38 ozaki-r 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 1.265 2019/09/25 09:53:38 ozaki-r 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 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
81 
82 /* timer values */
83 int	nd6_prune	= 1;	/* walk list every 1 seconds */
84 int	nd6_delay	= 5;	/* delay first probe time 5 second */
85 int	nd6_umaxtries	= 3;	/* maximum unicast query */
86 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
87 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
88 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
89 
90 /* preventing too many loops in ND option parsing */
91 int nd6_maxndopt = 10;	/* max # of ND options allowed */
92 
93 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
94 
95 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
96 
97 #ifdef ND6_DEBUG
98 int nd6_debug = 1;
99 #else
100 int nd6_debug = 0;
101 #endif
102 
103 krwlock_t nd6_lock __cacheline_aligned;
104 
105 struct nd_drhead nd_defrouter;
106 struct nd_prhead nd_prefix = { 0 };
107 
108 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
109 
110 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
111 static void nd6_slowtimo(void *);
112 static int regen_tmpaddr(const struct in6_ifaddr *);
113 static void nd6_free(struct llentry *, int);
114 static void nd6_llinfo_timer(void *);
115 static void nd6_timer(void *);
116 static void nd6_timer_work(struct work *, void *);
117 static void clear_llinfo_pqueue(struct llentry *);
118 static struct nd_opt_hdr *nd6_option(union nd_opts *);
119 
120 static callout_t nd6_slowtimo_ch;
121 static callout_t nd6_timer_ch;
122 static struct workqueue	*nd6_timer_wq;
123 static struct work	nd6_timer_wk;
124 
125 static int fill_drlist(void *, size_t *);
126 static int fill_prlist(void *, size_t *);
127 
128 static struct ifnet *nd6_defifp;
129 static int nd6_defifindex;
130 
131 static int nd6_setdefaultiface(int);
132 
133 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
134 
135 void
136 nd6_init(void)
137 {
138 	int error;
139 
140 	nd6_nbr_init();
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 	const struct in6_addr *taddr6 = &ln->r_l3addr.addr6;
467 	struct sockaddr_in6 sin6;
468 
469 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
470 
471 	LLE_WLOCK(ln);
472 	if ((ln->la_flags & LLE_LINKED) == 0)
473 		goto out;
474 	if (ln->ln_ntick > 0) {
475 		nd6_llinfo_settimer(ln, ln->ln_ntick);
476 		goto out;
477 	}
478 
479 	ifp = ln->lle_tbl->llt_ifp;
480 	KASSERT(ifp != NULL);
481 
482 	ndi = ND_IFINFO(ifp);
483 
484 	switch (ln->ln_state) {
485 	case ND6_LLINFO_WAITDELETE:
486 		LLE_REMREF(ln);
487 		nd6_free(ln, 0);
488 		ln = NULL;
489 		break;
490 
491 	case ND6_LLINFO_INCOMPLETE:
492 		if (ln->ln_asked++ < nd6_mmaxtries) {
493 			send_ns = true;
494 			break;
495 		}
496 
497 		if (ln->ln_hold) {
498 			struct mbuf *m = ln->ln_hold, *m0;
499 
500 			/*
501 			 * assuming every packet in ln_hold has
502 			 * the same IP header
503 			 */
504 			m0 = m->m_nextpkt;
505 			m->m_nextpkt = NULL;
506 			ln->ln_hold = m0;
507 			clear_llinfo_pqueue(ln);
508 
509 			icmp6_error2(m, ICMP6_DST_UNREACH,
510 			    ICMP6_DST_UNREACH_ADDR, 0, ifp);
511 		}
512 
513 		sockaddr_in6_init(&sin6, taddr6, 0, 0, 0);
514 		rt_clonedmsg(RTM_MISS, sin6tosa(&sin6), NULL, ifp);
515 
516 		/*
517 		 * Move to the ND6_LLINFO_WAITDELETE state for another
518 		 * interval at which point the llentry will be freed
519 		 * unless it's attempted to be used again and we'll
520 		 * resend NS again, rinse and repeat.
521 		 */
522 		ln->ln_state = ND6_LLINFO_WAITDELETE;
523 		if (ln->ln_asked == nd6_mmaxtries)
524 			nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
525 		else
526 			send_ns = true;
527 		break;
528 
529 	case ND6_LLINFO_REACHABLE:
530 		if (!ND6_LLINFO_PERMANENT(ln)) {
531 			ln->ln_state = ND6_LLINFO_STALE;
532 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
533 		}
534 		break;
535 
536 	case ND6_LLINFO_PURGE:
537 	case ND6_LLINFO_STALE:
538 		/* Garbage Collection(RFC 2461 5.3) */
539 		if (!ND6_LLINFO_PERMANENT(ln)) {
540 			LLE_REMREF(ln);
541 			nd6_free(ln, 1);
542 			ln = NULL;
543 		}
544 		break;
545 
546 	case ND6_LLINFO_DELAY:
547 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
548 			/* We need NUD */
549 			ln->ln_asked = 1;
550 			ln->ln_state = ND6_LLINFO_PROBE;
551 			daddr6 = &ln->r_l3addr.addr6;
552 			send_ns = true;
553 		} else {
554 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
555 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
556 		}
557 		break;
558 	case ND6_LLINFO_PROBE:
559 		if (ln->ln_asked < nd6_umaxtries) {
560 			ln->ln_asked++;
561 			daddr6 = &ln->r_l3addr.addr6;
562 			send_ns = true;
563 		} else {
564 			LLE_REMREF(ln);
565 			nd6_free(ln, 0);
566 			ln = NULL;
567 		}
568 		break;
569 	}
570 
571 	if (send_ns) {
572 		struct in6_addr src, *psrc;
573 
574 		nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
575 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
576 		LLE_FREE_LOCKED(ln);
577 		ln = NULL;
578 		nd6_ns_output(ifp, daddr6, taddr6, psrc, NULL);
579 	}
580 
581 out:
582 	if (ln != NULL)
583 		LLE_FREE_LOCKED(ln);
584 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
585 }
586 
587 /*
588  * ND6 timer routine to expire default route list and prefix list
589  */
590 static void
591 nd6_timer_work(struct work *wk, void *arg)
592 {
593 	struct nd_defrouter *next_dr, *dr;
594 	struct nd_prefix *next_pr, *pr;
595 	struct in6_ifaddr *ia6, *nia6;
596 	int s, bound;
597 	struct psref psref;
598 
599 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
600 	    nd6_timer, NULL);
601 
602 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
603 
604 	/* expire default router list */
605 
606 	ND6_WLOCK();
607 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, next_dr) {
608 		if (dr->expire && dr->expire < time_uptime) {
609 			nd6_defrtrlist_del(dr, NULL);
610 		}
611 	}
612 	ND6_UNLOCK();
613 
614 	/*
615 	 * expire interface addresses.
616 	 * in the past the loop was inside prefix expiry processing.
617 	 * However, from a stricter speci-confrmance standpoint, we should
618 	 * rather separate address lifetimes and prefix lifetimes.
619 	 */
620 	bound = curlwp_bind();
621   addrloop:
622 	s = pserialize_read_enter();
623 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
624 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
625 
626 		ia6_acquire(ia6, &psref);
627 		pserialize_read_exit(s);
628 
629 		/* check address lifetime */
630 		if (IFA6_IS_INVALID(ia6)) {
631 			int regen = 0;
632 			struct ifnet *ifp;
633 
634 			/*
635 			 * If the expiring address is temporary, try
636 			 * regenerating a new one.  This would be useful when
637 			 * we suspended a laptop PC, then turned it on after a
638 			 * period that could invalidate all temporary
639 			 * addresses.  Although we may have to restart the
640 			 * loop (see below), it must be after purging the
641 			 * address.  Otherwise, we'd see an infinite loop of
642 			 * regeneration.
643 			 */
644 			if (ip6_use_tempaddr &&
645 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
646 				IFNET_LOCK(ia6->ia_ifa.ifa_ifp);
647 				if (regen_tmpaddr(ia6) == 0)
648 					regen = 1;
649 				IFNET_UNLOCK(ia6->ia_ifa.ifa_ifp);
650 			}
651 
652 			ifp = ia6->ia_ifa.ifa_ifp;
653 			IFNET_LOCK(ifp);
654 			/*
655 			 * Need to take the lock first to prevent if_detach
656 			 * from running in6_purgeaddr concurrently.
657 			 */
658 			if (!if_is_deactivated(ifp)) {
659 				ia6_release(ia6, &psref);
660 				in6_purgeaddr(&ia6->ia_ifa);
661 			} else {
662 				/*
663 				 * ifp is being destroyed, ia6 will be destroyed
664 				 * by if_detach.
665 				 */
666 				ia6_release(ia6, &psref);
667 			}
668 			ia6 = NULL;
669 			IFNET_UNLOCK(ifp);
670 
671 			if (regen)
672 				goto addrloop; /* XXX: see below */
673 		} else if (IFA6_IS_DEPRECATED(ia6)) {
674 			int oldflags = ia6->ia6_flags;
675 
676 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
677 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
678 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
679 			}
680 
681 			/*
682 			 * If a temporary address has just become deprecated,
683 			 * regenerate a new one if possible.
684 			 */
685 			if (ip6_use_tempaddr &&
686 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
687 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
688 				int ret;
689 
690 				IFNET_LOCK(ia6->ia_ifa.ifa_ifp);
691 				ret = regen_tmpaddr(ia6);
692 				IFNET_UNLOCK(ia6->ia_ifa.ifa_ifp);
693 				if (ret == 0) {
694 					/*
695 					 * A new temporary address is
696 					 * generated.
697 					 * XXX: this means the address chain
698 					 * has changed while we are still in
699 					 * the loop.  Although the change
700 					 * would not cause disaster (because
701 					 * it's not a deletion, but an
702 					 * addition,) we'd rather restart the
703 					 * loop just for safety.  Or does this
704 					 * significantly reduce performance??
705 					 */
706 					ia6_release(ia6, &psref);
707 					goto addrloop;
708 				}
709 			}
710 		} else {
711 			/*
712 			 * A new RA might have made a deprecated address
713 			 * preferred.
714 			 */
715 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
716 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
717 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
718 			}
719 		}
720 		s = pserialize_read_enter();
721 		ia6_release(ia6, &psref);
722 	}
723 	pserialize_read_exit(s);
724 	curlwp_bindx(bound);
725 
726 	/* expire prefix list */
727 	ND6_WLOCK();
728 	ND_PREFIX_LIST_FOREACH_SAFE(pr, next_pr) {
729 		/*
730 		 * check prefix lifetime.
731 		 * since pltime is just for autoconf, pltime processing for
732 		 * prefix is not necessary.
733 		 */
734 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
735 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
736 			/*
737 			 * Just invalidate the prefix here. Removing it
738 			 * will be done when purging an associated address.
739 			 */
740 			KASSERTMSG(pr->ndpr_refcnt > 0, "ndpr_refcnt=%d",
741 			    pr->ndpr_refcnt);
742 			nd6_invalidate_prefix(pr);
743 		}
744 	}
745 	ND6_UNLOCK();
746 
747 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
748 }
749 
750 static void
751 nd6_timer(void *ignored_arg)
752 {
753 
754 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
755 }
756 
757 /* ia6: deprecated/invalidated temporary address */
758 static int
759 regen_tmpaddr(const struct in6_ifaddr *ia6)
760 {
761 	struct ifaddr *ifa;
762 	struct ifnet *ifp;
763 	struct in6_ifaddr *public_ifa6 = NULL;
764 	int s;
765 
766 	ifp = ia6->ia_ifa.ifa_ifp;
767 	s = pserialize_read_enter();
768 	IFADDR_READER_FOREACH(ifa, ifp) {
769 		struct in6_ifaddr *it6;
770 
771 		if (ifa->ifa_addr->sa_family != AF_INET6)
772 			continue;
773 
774 		it6 = (struct in6_ifaddr *)ifa;
775 
776 		/* ignore no autoconf addresses. */
777 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
778 			continue;
779 
780 		/* ignore autoconf addresses with different prefixes. */
781 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
782 			continue;
783 
784 		/*
785 		 * Now we are looking at an autoconf address with the same
786 		 * prefix as ours.  If the address is temporary and is still
787 		 * preferred, do not create another one.  It would be rare, but
788 		 * could happen, for example, when we resume a laptop PC after
789 		 * a long period.
790 		 */
791 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
792 		    !IFA6_IS_DEPRECATED(it6)) {
793 			public_ifa6 = NULL;
794 			break;
795 		}
796 
797 		/*
798 		 * This is a public autoconf address that has the same prefix
799 		 * as ours.  If it is preferred, keep it.  We can't break the
800 		 * loop here, because there may be a still-preferred temporary
801 		 * address with the prefix.
802 		 */
803 		if (!IFA6_IS_DEPRECATED(it6))
804 			public_ifa6 = it6;
805 	}
806 
807 	if (public_ifa6 != NULL) {
808 		int e;
809 		struct psref psref;
810 
811 		ia6_acquire(public_ifa6, &psref);
812 		pserialize_read_exit(s);
813 		/*
814 		 * Random factor is introduced in the preferred lifetime, so
815 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
816 		 */
817 		ND6_WLOCK();
818 		e = in6_tmpifadd(public_ifa6, 0, 0);
819 		ND6_UNLOCK();
820 		if (e != 0) {
821 			ia6_release(public_ifa6, &psref);
822 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
823 			    " tmp addr, errno=%d\n", e);
824 			return -1;
825 		}
826 		ia6_release(public_ifa6, &psref);
827 		return 0;
828 	}
829 	pserialize_read_exit(s);
830 
831 	return -1;
832 }
833 
834 bool
835 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
836 {
837 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
838 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
839 		return true;
840 	case ND6_IFF_ACCEPT_RTADV:
841 		return ip6_accept_rtadv != 0;
842 	case ND6_IFF_OVERRIDE_RTADV:
843 	case 0:
844 	default:
845 		return false;
846 	}
847 }
848 
849 /*
850  * Nuke neighbor cache/prefix/default router management table, right before
851  * ifp goes away.
852  */
853 void
854 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
855 {
856 	struct nd_defrouter *dr, *ndr;
857 	struct nd_prefix *pr, *npr;
858 
859 	/*
860 	 * During detach, the ND info might be already removed, but
861 	 * then is explitly passed as argument.
862 	 * Otherwise get it from ifp->if_afdata.
863 	 */
864 	if (ext == NULL)
865 		ext = ifp->if_afdata[AF_INET6];
866 	if (ext == NULL)
867 		return;
868 
869 	ND6_WLOCK();
870 	/*
871 	 * Nuke default router list entries toward ifp.
872 	 * We defer removal of default router list entries that is installed
873 	 * in the routing table, in order to keep additional side effects as
874 	 * small as possible.
875 	 */
876 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
877 		if (dr->installed)
878 			continue;
879 
880 		if (dr->ifp == ifp) {
881 			KASSERT(ext != NULL);
882 			nd6_defrtrlist_del(dr, ext);
883 		}
884 	}
885 
886 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
887 		if (!dr->installed)
888 			continue;
889 
890 		if (dr->ifp == ifp) {
891 			KASSERT(ext != NULL);
892 			nd6_defrtrlist_del(dr, ext);
893 		}
894 	}
895 
896 	/* Nuke prefix list entries toward ifp */
897 	ND_PREFIX_LIST_FOREACH_SAFE(pr, npr) {
898 		if (pr->ndpr_ifp == ifp) {
899 			/*
900 			 * All addresses referencing pr should be already freed.
901 			 */
902 			KASSERTMSG(pr->ndpr_refcnt == 0, "ndpr_refcnt=%d",
903 			    pr->ndpr_refcnt);
904 			nd6_prelist_remove(pr);
905 		}
906 	}
907 
908 	/* cancel default outgoing interface setting */
909 	if (nd6_defifindex == ifp->if_index)
910 		nd6_setdefaultiface(0);
911 
912 	/* XXX: too restrictive? */
913 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
914 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
915 		if (ndi && nd6_accepts_rtadv(ndi)) {
916 			/* refresh default router list */
917 			nd6_defrouter_select();
918 		}
919 	}
920 	ND6_UNLOCK();
921 
922 	/*
923 	 * We may not need to nuke the neighbor cache entries here
924 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
925 	 * nd6_purge() is invoked by in6_ifdetach() which is called
926 	 * from if_detach() where everything gets purged. However
927 	 * in6_ifdetach is directly called from vlan(4), so we still
928 	 * need to purge entries here.
929 	 */
930 	if (ext->lltable != NULL)
931 		lltable_purge_entries(ext->lltable);
932 }
933 
934 void
935 nd6_assert_purged(struct ifnet *ifp)
936 {
937 	struct nd_defrouter *dr;
938 	struct nd_prefix *pr;
939 	char ip6buf[INET6_ADDRSTRLEN] __diagused;
940 
941 	ND6_RLOCK();
942 	ND_DEFROUTER_LIST_FOREACH(dr) {
943 		KASSERTMSG(dr->ifp != ifp,
944 		    "defrouter %s remains on %s",
945 		    IN6_PRINT(ip6buf, &dr->rtaddr), ifp->if_xname);
946 	}
947 
948 	ND_PREFIX_LIST_FOREACH(pr) {
949 		KASSERTMSG(pr->ndpr_ifp != ifp,
950 		    "prefix %s/%d remains on %s",
951 		    IN6_PRINT(ip6buf, &pr->ndpr_prefix.sin6_addr),
952 		    pr->ndpr_plen, ifp->if_xname);
953 	}
954 	ND6_UNLOCK();
955 }
956 
957 struct llentry *
958 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
959 {
960 	struct sockaddr_in6 sin6;
961 	struct llentry *ln;
962 
963 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
964 
965 	IF_AFDATA_RLOCK(ifp);
966 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
967 	    sin6tosa(&sin6));
968 	IF_AFDATA_RUNLOCK(ifp);
969 
970 	return ln;
971 }
972 
973 struct llentry *
974 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
975 {
976 	struct sockaddr_in6 sin6;
977 	struct llentry *ln;
978 	struct rtentry *rt;
979 
980 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
981 	rt = rtalloc1(sin6tosa(&sin6), 0);
982 
983 	IF_AFDATA_WLOCK(ifp);
984 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE, sin6tosa(&sin6), rt);
985 	IF_AFDATA_WUNLOCK(ifp);
986 
987 	if (rt != NULL)
988 		rt_unref(rt);
989 	if (ln != NULL)
990 		ln->ln_state = ND6_LLINFO_NOSTATE;
991 
992 	return ln;
993 }
994 
995 /*
996  * Test whether a given IPv6 address is a neighbor or not, ignoring
997  * the actual neighbor cache.  The neighbor cache is ignored in order
998  * to not reenter the routing code from within itself.
999  */
1000 static int
1001 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1002 {
1003 	struct nd_prefix *pr;
1004 	struct ifaddr *dstaddr;
1005 	int s;
1006 
1007 	/*
1008 	 * A link-local address is always a neighbor.
1009 	 * XXX: a link does not necessarily specify a single interface.
1010 	 */
1011 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1012 		struct sockaddr_in6 sin6_copy;
1013 		u_int32_t zone;
1014 
1015 		/*
1016 		 * We need sin6_copy since sa6_recoverscope() may modify the
1017 		 * content (XXX).
1018 		 */
1019 		sin6_copy = *addr;
1020 		if (sa6_recoverscope(&sin6_copy))
1021 			return 0; /* XXX: should be impossible */
1022 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1023 			return 0;
1024 		if (sin6_copy.sin6_scope_id == zone)
1025 			return 1;
1026 		else
1027 			return 0;
1028 	}
1029 
1030 	/*
1031 	 * If the address matches one of our addresses,
1032 	 * it should be a neighbor.
1033 	 * If the address matches one of our on-link prefixes, it should be a
1034 	 * neighbor.
1035 	 */
1036 	ND6_RLOCK();
1037 	ND_PREFIX_LIST_FOREACH(pr) {
1038 		if (pr->ndpr_ifp != ifp)
1039 			continue;
1040 
1041 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1042 			struct rtentry *rt;
1043 
1044 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
1045 			if (rt == NULL)
1046 				continue;
1047 			/*
1048 			 * This is the case where multiple interfaces
1049 			 * have the same prefix, but only one is installed
1050 			 * into the routing table and that prefix entry
1051 			 * is not the one being examined here. In the case
1052 			 * where RADIX_MPATH is enabled, multiple route
1053 			 * entries (of the same rt_key value) will be
1054 			 * installed because the interface addresses all
1055 			 * differ.
1056 			 */
1057 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1058 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
1059 				rt_unref(rt);
1060 				continue;
1061 			}
1062 			rt_unref(rt);
1063 		}
1064 
1065 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1066 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1067 			ND6_UNLOCK();
1068 			return 1;
1069 		}
1070 	}
1071 	ND6_UNLOCK();
1072 
1073 	/*
1074 	 * If the address is assigned on the node of the other side of
1075 	 * a p2p interface, the address should be a neighbor.
1076 	 */
1077 	s = pserialize_read_enter();
1078 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1079 	if (dstaddr != NULL) {
1080 		if (dstaddr->ifa_ifp == ifp) {
1081 			pserialize_read_exit(s);
1082 			return 1;
1083 		}
1084 	}
1085 	pserialize_read_exit(s);
1086 
1087 	/*
1088 	 * If the default router list is empty, all addresses are regarded
1089 	 * as on-link, and thus, as a neighbor.
1090 	 */
1091 	ND6_RLOCK();
1092 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1093 	    ND_DEFROUTER_LIST_EMPTY() && nd6_defifindex == ifp->if_index) {
1094 		ND6_UNLOCK();
1095 		return 1;
1096 	}
1097 	ND6_UNLOCK();
1098 
1099 	return 0;
1100 }
1101 
1102 /*
1103  * Detect if a given IPv6 address identifies a neighbor on a given link.
1104  * XXX: should take care of the destination of a p2p link?
1105  */
1106 int
1107 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1108 {
1109 	struct nd_prefix *pr;
1110 	struct llentry *ln;
1111 	struct rtentry *rt;
1112 
1113 	/*
1114 	 * A link-local address is always a neighbor.
1115 	 * XXX: a link does not necessarily specify a single interface.
1116 	 */
1117 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1118 		struct sockaddr_in6 sin6_copy;
1119 		u_int32_t zone;
1120 
1121 		/*
1122 		 * We need sin6_copy since sa6_recoverscope() may modify the
1123 		 * content (XXX).
1124 		 */
1125 		sin6_copy = *addr;
1126 		if (sa6_recoverscope(&sin6_copy))
1127 			return 0; /* XXX: should be impossible */
1128 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1129 			return 0;
1130 		if (sin6_copy.sin6_scope_id == zone)
1131 			return 1;
1132 		else
1133 			return 0;
1134 	}
1135 
1136 	/*
1137 	 * If the address matches one of our on-link prefixes, it should be a
1138 	 * neighbor.
1139 	 */
1140 	ND6_RLOCK();
1141 	ND_PREFIX_LIST_FOREACH(pr) {
1142 		if (pr->ndpr_ifp != ifp)
1143 			continue;
1144 
1145 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1146 			continue;
1147 
1148 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1149 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1150 			ND6_UNLOCK();
1151 			return 1;
1152 		}
1153 	}
1154 
1155 	/*
1156 	 * If the default router list is empty, all addresses are regarded
1157 	 * as on-link, and thus, as a neighbor.
1158 	 * XXX: we restrict the condition to hosts, because routers usually do
1159 	 * not have the "default router list".
1160 	 */
1161 	if (!ip6_forwarding && ND_DEFROUTER_LIST_EMPTY() &&
1162 	    nd6_defifindex == ifp->if_index) {
1163 		ND6_UNLOCK();
1164 		return 1;
1165 	}
1166 	ND6_UNLOCK();
1167 
1168 	if (nd6_is_new_addr_neighbor(addr, ifp))
1169 		return 1;
1170 
1171 	/*
1172 	 * Even if the address matches none of our addresses, it might be
1173 	 * in the neighbor cache or a connected route.
1174 	 */
1175 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1176 	if (ln != NULL) {
1177 		LLE_RUNLOCK(ln);
1178 		return 1;
1179 	}
1180 
1181 	rt = rtalloc1(sin6tocsa(addr), 0);
1182 	if (rt == NULL)
1183 		return 0;
1184 
1185 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1186 #if NBRIDGE > 0
1187 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1188 #endif
1189 #if NCARP > 0
1190 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1191 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1192 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1193 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1194 #endif
1195 	    )) {
1196 		rt_unref(rt);
1197 		return 1;
1198 	}
1199 	rt_unref(rt);
1200 
1201 	return 0;
1202 }
1203 
1204 /*
1205  * Free an nd6 llinfo entry.
1206  * Since the function would cause significant changes in the kernel, DO NOT
1207  * make it global, unless you have a strong reason for the change, and are sure
1208  * that the change is safe.
1209  */
1210 static void
1211 nd6_free(struct llentry *ln, int gc)
1212 {
1213 	struct ifnet *ifp;
1214 	struct in6_addr *in6;
1215 
1216 	KASSERT(ln != NULL);
1217 	LLE_WLOCK_ASSERT(ln);
1218 
1219 	ifp = ln->lle_tbl->llt_ifp;
1220 	in6 = &ln->r_l3addr.addr6;
1221 	/*
1222 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1223 	 * even though it is not harmful, it was not really necessary.
1224 	 */
1225 
1226 	if (!ip6_forwarding && ln->ln_router) {
1227 		if (ln->ln_state == ND6_LLINFO_STALE && gc) {
1228 			/*
1229 			 * If the reason for the deletion is just garbage
1230 			 * collection, and the neighbor is an active
1231 			 * router, do not delete it.  Instead, reset the GC
1232 			 * timer using the router's lifetime.
1233 			 * Simply deleting the entry may affect default
1234 			 * router selection, which is not necessarily a good
1235 			 * thing, especially when we're using router preference
1236 			 * values.
1237 			 * XXX: the check for ln_state would be redundant,
1238 			 *      but we intentionally keep it just in case.
1239 			 */
1240 			if (ln->ln_expire > time_uptime)
1241 				nd6_llinfo_settimer(ln,
1242 				    (ln->ln_expire - time_uptime) * hz);
1243 			else
1244 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1245 			LLE_WUNLOCK(ln);
1246 			return;
1247 		}
1248 
1249 		ND6_WLOCK();
1250 
1251 		/*
1252 		 * We need to unlock to avoid a LOR with nd6_rt_flush()
1253 		 * with the rnh and for the calls to
1254 		 * nd6_pfxlist_onlink_check() and nd6_defrouter_select() in the
1255 		 * block further down for calls into nd6_lookup().
1256 		 * We still hold a ref.
1257 		 *
1258 		 * Temporarily fake the state to choose a new default
1259 		 * router and to perform on-link determination of
1260 		 * prefixes correctly.
1261 		 * Below the state will be set correctly,
1262 		 * or the entry itself will be deleted.
1263 		 */
1264 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1265 		LLE_WUNLOCK(ln);
1266 
1267 		/*
1268 		 * nd6_rt_flush must be called whether or not the neighbor
1269 		 * is in the Default Router List.
1270 		 * See a corresponding comment in nd6_na_input().
1271 		 */
1272 		nd6_rt_flush(in6, ifp);
1273 
1274 		/*
1275 		 * Unreachablity of a router might affect the default
1276 		 * router selection and on-link detection of advertised
1277 		 * prefixes.
1278 		 *
1279 		 * Since nd6_defrouter_select() does not affect the
1280 		 * on-link determination and MIP6 needs the check
1281 		 * before the default router selection, we perform
1282 		 * the check now.
1283 		 */
1284 		nd6_pfxlist_onlink_check();
1285 
1286 		/*
1287 		 * refresh default router list
1288 		 */
1289 		nd6_defrouter_select();
1290 
1291 #ifdef __FreeBSD__
1292 		/*
1293 		 * If this entry was added by an on-link redirect, remove the
1294 		 * corresponding host route.
1295 		 */
1296 		if (ln->la_flags & LLE_REDIRECT)
1297 			nd6_free_redirect(ln);
1298 #endif
1299 
1300 		ND6_UNLOCK();
1301 		LLE_WLOCK(ln);
1302 	}
1303 
1304 	if (ln->la_flags & LLE_VALID || gc) {
1305 		struct sockaddr_in6 sin6;
1306 		const char *lladdr;
1307 
1308 		sockaddr_in6_init(&sin6, in6, 0, 0, 0);
1309 		lladdr = ln->la_flags & LLE_VALID ?
1310 		    (const char *)&ln->ll_addr : NULL;
1311 		rt_clonedmsg(RTM_DELETE, sin6tosa(&sin6), lladdr, ifp);
1312 	}
1313 
1314 	/*
1315 	 * Save to unlock. We still hold an extra reference and will not
1316 	 * free(9) in llentry_free() if someone else holds one as well.
1317 	 */
1318 	LLE_WUNLOCK(ln);
1319 	IF_AFDATA_LOCK(ifp);
1320 	LLE_WLOCK(ln);
1321 
1322 	lltable_free_entry(LLTABLE6(ifp), ln);
1323 
1324 	IF_AFDATA_UNLOCK(ifp);
1325 }
1326 
1327 /*
1328  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1329  *
1330  * XXX cost-effective methods?
1331  */
1332 void
1333 nd6_nud_hint(struct rtentry *rt)
1334 {
1335 	struct llentry *ln;
1336 	struct ifnet *ifp;
1337 
1338 	if (rt == NULL)
1339 		return;
1340 
1341 	ifp = rt->rt_ifp;
1342 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1343 	if (ln == NULL)
1344 		return;
1345 
1346 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1347 		goto done;
1348 
1349 	/*
1350 	 * if we get upper-layer reachability confirmation many times,
1351 	 * it is possible we have false information.
1352 	 */
1353 	ln->ln_byhint++;
1354 	if (ln->ln_byhint > nd6_maxnudhint)
1355 		goto done;
1356 
1357 	ln->ln_state = ND6_LLINFO_REACHABLE;
1358 	if (!ND6_LLINFO_PERMANENT(ln))
1359 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1360 
1361 done:
1362 	LLE_WUNLOCK(ln);
1363 
1364 	return;
1365 }
1366 
1367 struct gc_args {
1368 	int gc_entries;
1369 	const struct in6_addr *skip_in6;
1370 };
1371 
1372 static int
1373 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1374 {
1375 	struct gc_args *args = farg;
1376 	int *n = &args->gc_entries;
1377 	const struct in6_addr *skip_in6 = args->skip_in6;
1378 
1379 	if (*n <= 0)
1380 		return 0;
1381 
1382 	if (ND6_LLINFO_PERMANENT(ln))
1383 		return 0;
1384 
1385 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1386 		return 0;
1387 
1388 	LLE_WLOCK(ln);
1389 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1390 		ln->ln_state = ND6_LLINFO_STALE;
1391 	else
1392 		ln->ln_state = ND6_LLINFO_PURGE;
1393 	nd6_llinfo_settimer(ln, 0);
1394 	LLE_WUNLOCK(ln);
1395 
1396 	(*n)--;
1397 	return 0;
1398 }
1399 
1400 static void
1401 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1402 {
1403 
1404 	if (ip6_neighborgcthresh >= 0 &&
1405 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1406 		struct gc_args gc_args = {10, in6};
1407 		/*
1408 		 * XXX entries that are "less recently used" should be
1409 		 * freed first.
1410 		 */
1411 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1412 	}
1413 }
1414 
1415 void
1416 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1417 {
1418 	struct sockaddr *gate = rt->rt_gateway;
1419 	struct ifnet *ifp = rt->rt_ifp;
1420 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1421 	struct ifaddr *ifa;
1422 
1423 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1424 
1425 	if (req == RTM_LLINFO_UPD) {
1426 		int rc;
1427 		struct in6_addr *in6;
1428 		struct in6_addr in6_all;
1429 		int anycast;
1430 
1431 		if ((ifa = info->rti_ifa) == NULL)
1432 			return;
1433 
1434 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1435 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1436 
1437 		in6_all = in6addr_linklocal_allnodes;
1438 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1439 			log(LOG_ERR, "%s: failed to set scope %s "
1440 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1441 			return;
1442 		}
1443 
1444 		/* XXX don't set Override for proxy addresses */
1445 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1446 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1447 #if 0
1448 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1449 #endif
1450 		    , 1, NULL);
1451 		return;
1452 	}
1453 
1454 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1455 		return;
1456 
1457 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1458 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1459 		/*
1460 		 * This is probably an interface direct route for a link
1461 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1462 		 * We do not need special treatment below for such a route.
1463 		 * Moreover, the RTF_LLINFO flag which would be set below
1464 		 * would annoy the ndp(8) command.
1465 		 */
1466 		return;
1467 	}
1468 
1469 	switch (req) {
1470 	case RTM_ADD: {
1471 		struct psref psref;
1472 
1473 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1474 		/*
1475 		 * There is no backward compatibility :)
1476 		 *
1477 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1478 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1479 		 *	   rt->rt_flags |= RTF_CLONING;
1480 		 */
1481 		/* XXX should move to route.c? */
1482 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1483 			union {
1484 				struct sockaddr sa;
1485 				struct sockaddr_dl sdl;
1486 				struct sockaddr_storage ss;
1487 			} u;
1488 			/*
1489 			 * Case 1: This route should come from a route to
1490 			 * interface (RTF_CLONING case) or the route should be
1491 			 * treated as on-link but is currently not
1492 			 * (RTF_LLINFO && ln == NULL case).
1493 			 */
1494 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1495 			    ifp->if_index, ifp->if_type,
1496 			    NULL, namelen, NULL, addrlen) == NULL) {
1497 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1498 				    "failed on %s\n", __func__, __LINE__,
1499 				    sizeof(u.ss), if_name(ifp));
1500 			}
1501 			rt_setgate(rt, &u.sa);
1502 			gate = rt->rt_gateway;
1503 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1504 			if (gate == NULL) {
1505 				log(LOG_ERR,
1506 				    "%s: rt_setgate failed on %s\n", __func__,
1507 				    if_name(ifp));
1508 				break;
1509 			}
1510 
1511 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1512 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1513 				break;
1514 		}
1515 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1516 		/*
1517 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1518 		 * We don't do that here since llinfo is not ready yet.
1519 		 *
1520 		 * There are also couple of other things to be discussed:
1521 		 * - unsolicited NA code needs improvement beforehand
1522 		 * - RFC2461 says we MAY send multicast unsolicited NA
1523 		 *   (7.2.6 paragraph 4), however, it also says that we
1524 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1525 		 *   we don't have anything like it right now.
1526 		 *   note that the mechanism needs a mutual agreement
1527 		 *   between proxies, which means that we need to implement
1528 		 *   a new protocol, or a new kludge.
1529 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1530 		 *   we need to check ip6forwarding before sending it.
1531 		 *   (or should we allow proxy ND configuration only for
1532 		 *   routers?  there's no mention about proxy ND from hosts)
1533 		 */
1534 #if 0
1535 		/* XXX it does not work */
1536 		if (rt->rt_flags & RTF_ANNOUNCE)
1537 			nd6_na_output(ifp,
1538 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1539 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1540 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1541 			      1, NULL);
1542 #endif
1543 
1544 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1545 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1546 			/*
1547 			 * Address resolution isn't necessary for a point to
1548 			 * point link, so we can skip this test for a p2p link.
1549 			 */
1550 			if (gate->sa_family != AF_LINK ||
1551 			    gate->sa_len <
1552 			    sockaddr_dl_measure(namelen, addrlen)) {
1553 				log(LOG_DEBUG,
1554 				    "nd6_rtrequest: bad gateway value: %s\n",
1555 				    if_name(ifp));
1556 				break;
1557 			}
1558 			satosdl(gate)->sdl_type = ifp->if_type;
1559 			satosdl(gate)->sdl_index = ifp->if_index;
1560 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1561 		}
1562 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1563 
1564 		/*
1565 		 * When called from rt_ifa_addlocal, we cannot depend on that
1566 		 * the address (rt_getkey(rt)) exits in the address list of the
1567 		 * interface. So check RTF_LOCAL instead.
1568 		 */
1569 		if (rt->rt_flags & RTF_LOCAL) {
1570 			if (nd6_useloopback)
1571 				rt->rt_ifp = lo0ifp;	/* XXX */
1572 			break;
1573 		}
1574 
1575 		/*
1576 		 * check if rt_getkey(rt) is an address assigned
1577 		 * to the interface.
1578 		 */
1579 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp,
1580 		    &satocsin6(rt_getkey(rt))->sin6_addr, &psref);
1581 		if (ifa != NULL) {
1582 			if (nd6_useloopback) {
1583 				rt->rt_ifp = lo0ifp;	/* XXX */
1584 				/*
1585 				 * Make sure rt_ifa be equal to the ifaddr
1586 				 * corresponding to the address.
1587 				 * We need this because when we refer
1588 				 * rt_ifa->ia6_flags in ip6_input, we assume
1589 				 * that the rt_ifa points to the address instead
1590 				 * of the loopback address.
1591 				 */
1592 				if (!ISSET(info->rti_flags, RTF_DONTCHANGEIFA)
1593 				    && ifa != rt->rt_ifa)
1594 					rt_replace_ifa(rt, ifa);
1595 			}
1596 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1597 			/* join solicited node multicast for proxy ND */
1598 			if (ifp->if_flags & IFF_MULTICAST) {
1599 				struct in6_addr llsol;
1600 				int error;
1601 
1602 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1603 				llsol.s6_addr32[0] = htonl(0xff020000);
1604 				llsol.s6_addr32[1] = 0;
1605 				llsol.s6_addr32[2] = htonl(1);
1606 				llsol.s6_addr8[12] = 0xff;
1607 				if (in6_setscope(&llsol, ifp, NULL))
1608 					goto out;
1609 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1610 					char ip6buf[INET6_ADDRSTRLEN];
1611 					nd6log(LOG_ERR, "%s: failed to join "
1612 					    "%s (errno=%d)\n", if_name(ifp),
1613 					    IN6_PRINT(ip6buf, &llsol), error);
1614 				}
1615 			}
1616 		}
1617 	out:
1618 		ifa_release(ifa, &psref);
1619 		/*
1620 		 * If we have too many cache entries, initiate immediate
1621 		 * purging for some entries.
1622 		 */
1623 		if (rt->rt_ifp != NULL)
1624 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1625 		break;
1626 	    }
1627 
1628 	case RTM_DELETE:
1629 		/* leave from solicited node multicast for proxy ND */
1630 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1631 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1632 			struct in6_addr llsol;
1633 
1634 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1635 			llsol.s6_addr32[0] = htonl(0xff020000);
1636 			llsol.s6_addr32[1] = 0;
1637 			llsol.s6_addr32[2] = htonl(1);
1638 			llsol.s6_addr8[12] = 0xff;
1639 			if (in6_setscope(&llsol, ifp, NULL) == 0)
1640 				in6_lookup_and_delete_multi(&llsol, ifp);
1641 		}
1642 		break;
1643 	}
1644 }
1645 
1646 int
1647 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1648 {
1649 	struct in6_drlist *drl = (struct in6_drlist *)data;
1650 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1651 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1652 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1653 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1654 	struct nd_defrouter *dr;
1655 	struct nd_prefix *pr;
1656 	int i = 0, error = 0;
1657 
1658 	switch (cmd) {
1659 	case SIOCGDRLST_IN6:
1660 		/*
1661 		 * obsolete API, use sysctl under net.inet6.icmp6
1662 		 */
1663 		memset(drl, 0, sizeof(*drl));
1664 		ND6_RLOCK();
1665 		ND_DEFROUTER_LIST_FOREACH(dr) {
1666 			if (i >= DRLSTSIZ)
1667 				break;
1668 			drl->defrouter[i].rtaddr = dr->rtaddr;
1669 			in6_clearscope(&drl->defrouter[i].rtaddr);
1670 
1671 			drl->defrouter[i].flags = dr->flags;
1672 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1673 			drl->defrouter[i].expire = dr->expire ?
1674 			    time_mono_to_wall(dr->expire) : 0;
1675 			drl->defrouter[i].if_index = dr->ifp->if_index;
1676 			i++;
1677 		}
1678 		ND6_UNLOCK();
1679 		break;
1680 	case SIOCGPRLST_IN6:
1681 		/*
1682 		 * obsolete API, use sysctl under net.inet6.icmp6
1683 		 *
1684 		 * XXX the structure in6_prlist was changed in backward-
1685 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1686 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1687 		 */
1688 		/*
1689 		 * XXX meaning of fields, especialy "raflags", is very
1690 		 * differnet between RA prefix list and RR/static prefix list.
1691 		 * how about separating ioctls into two?
1692 		 */
1693 		memset(oprl, 0, sizeof(*oprl));
1694 		ND6_RLOCK();
1695 		ND_PREFIX_LIST_FOREACH(pr) {
1696 			struct nd_pfxrouter *pfr;
1697 			int j;
1698 
1699 			if (i >= PRLSTSIZ)
1700 				break;
1701 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1702 			oprl->prefix[i].raflags = pr->ndpr_raf;
1703 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1704 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1705 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1706 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1707 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1708 				oprl->prefix[i].expire = 0;
1709 			else {
1710 				time_t maxexpire;
1711 
1712 				/* XXX: we assume time_t is signed. */
1713 				maxexpire = (-1) &
1714 				    ~((time_t)1 <<
1715 				    ((sizeof(maxexpire) * 8) - 1));
1716 				if (pr->ndpr_vltime <
1717 				    maxexpire - pr->ndpr_lastupdate) {
1718 					time_t expire;
1719 					expire = pr->ndpr_lastupdate +
1720 					    pr->ndpr_vltime;
1721 					oprl->prefix[i].expire = expire ?
1722 					    time_mono_to_wall(expire) : 0;
1723 				} else
1724 					oprl->prefix[i].expire = maxexpire;
1725 			}
1726 
1727 			j = 0;
1728 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1729 				if (j < DRLSTSIZ) {
1730 #define RTRADDR oprl->prefix[i].advrtr[j]
1731 					RTRADDR = pfr->router->rtaddr;
1732 					in6_clearscope(&RTRADDR);
1733 #undef RTRADDR
1734 				}
1735 				j++;
1736 			}
1737 			oprl->prefix[i].advrtrs = j;
1738 			oprl->prefix[i].origin = PR_ORIG_RA;
1739 
1740 			i++;
1741 		}
1742 		ND6_UNLOCK();
1743 
1744 		break;
1745 	case OSIOCGIFINFO_IN6:
1746 #define ND	ndi->ndi
1747 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1748 		memset(&ND, 0, sizeof(ND));
1749 		ND.linkmtu = IN6_LINKMTU(ifp);
1750 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1751 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1752 		ND.reachable = ND_IFINFO(ifp)->reachable;
1753 		ND.retrans = ND_IFINFO(ifp)->retrans;
1754 		ND.flags = ND_IFINFO(ifp)->flags;
1755 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1756 		ND.chlim = ND_IFINFO(ifp)->chlim;
1757 		break;
1758 	case SIOCGIFINFO_IN6:
1759 		ND = *ND_IFINFO(ifp);
1760 		break;
1761 	case SIOCSIFINFO_IN6:
1762 		/*
1763 		 * used to change host variables from userland.
1764 		 * intented for a use on router to reflect RA configurations.
1765 		 */
1766 		/* 0 means 'unspecified' */
1767 		if (ND.linkmtu != 0) {
1768 			if (ND.linkmtu < IPV6_MMTU ||
1769 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1770 				error = EINVAL;
1771 				break;
1772 			}
1773 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1774 		}
1775 
1776 		if (ND.basereachable != 0) {
1777 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1778 
1779 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1780 			if (ND.basereachable != obasereachable)
1781 				ND_IFINFO(ifp)->reachable =
1782 				    ND_COMPUTE_RTIME(ND.basereachable);
1783 		}
1784 		if (ND.retrans != 0)
1785 			ND_IFINFO(ifp)->retrans = ND.retrans;
1786 		if (ND.chlim != 0)
1787 			ND_IFINFO(ifp)->chlim = ND.chlim;
1788 		/* FALLTHROUGH */
1789 	case SIOCSIFINFO_FLAGS:
1790 	{
1791 		struct ifaddr *ifa;
1792 		struct in6_ifaddr *ia;
1793 		int s;
1794 
1795 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1796 		    !(ND.flags & ND6_IFF_IFDISABLED))
1797 		{
1798 			/*
1799 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1800 			 * has a link-local address with IN6_IFF_DUPLICATED,
1801 			 * do not clear ND6_IFF_IFDISABLED.
1802 			 * See RFC 4862, section 5.4.5.
1803 			 */
1804 			int duplicated_linklocal = 0;
1805 
1806 			s = pserialize_read_enter();
1807 			IFADDR_READER_FOREACH(ifa, ifp) {
1808 				if (ifa->ifa_addr->sa_family != AF_INET6)
1809 					continue;
1810 				ia = (struct in6_ifaddr *)ifa;
1811 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1812 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1813 				{
1814 					duplicated_linklocal = 1;
1815 					break;
1816 				}
1817 			}
1818 			pserialize_read_exit(s);
1819 
1820 			if (duplicated_linklocal) {
1821 				ND.flags |= ND6_IFF_IFDISABLED;
1822 				log(LOG_ERR, "%s: Cannot enable an interface"
1823 				    " with a link-local address marked"
1824 				    " duplicate.\n", if_name(ifp));
1825 			} else {
1826 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1827 				if (ifp->if_flags & IFF_UP)
1828 					in6_if_up(ifp);
1829 			}
1830 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1831 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1832 			int bound = curlwp_bind();
1833 			/* Mark all IPv6 addresses as tentative. */
1834 
1835 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1836 			s = pserialize_read_enter();
1837 			IFADDR_READER_FOREACH(ifa, ifp) {
1838 				struct psref psref;
1839 				if (ifa->ifa_addr->sa_family != AF_INET6)
1840 					continue;
1841 				ifa_acquire(ifa, &psref);
1842 				pserialize_read_exit(s);
1843 
1844 				nd6_dad_stop(ifa);
1845 
1846 				ia = (struct in6_ifaddr *)ifa;
1847 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1848 
1849 				s = pserialize_read_enter();
1850 				ifa_release(ifa, &psref);
1851 			}
1852 			pserialize_read_exit(s);
1853 			curlwp_bindx(bound);
1854 		}
1855 
1856 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1857 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1858 				/* auto_linklocal 0->1 transition */
1859 
1860 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1861 				in6_ifattach(ifp, NULL);
1862 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1863 			    ifp->if_flags & IFF_UP)
1864 			{
1865 				/*
1866 				 * When the IF already has
1867 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1868 				 * address is assigned, and IFF_UP, try to
1869 				 * assign one.
1870 				 */
1871 				int haslinklocal = 0;
1872 
1873 				s = pserialize_read_enter();
1874 				IFADDR_READER_FOREACH(ifa, ifp) {
1875 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1876 						continue;
1877 					ia = (struct in6_ifaddr *)ifa;
1878 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1879 						haslinklocal = 1;
1880 						break;
1881 					}
1882 				}
1883 				pserialize_read_exit(s);
1884 				if (!haslinklocal)
1885 					in6_ifattach(ifp, NULL);
1886 			}
1887 		}
1888 	}
1889 		ND_IFINFO(ifp)->flags = ND.flags;
1890 		break;
1891 #undef ND
1892 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1893 		/* sync kernel routing table with the default router list */
1894 		ND6_WLOCK();
1895 		nd6_defrouter_reset();
1896 		nd6_defrouter_select();
1897 		ND6_UNLOCK();
1898 		break;
1899 	case SIOCSPFXFLUSH_IN6:
1900 	{
1901 		/* flush all the prefix advertised by routers */
1902 		struct nd_prefix *pfx, *next;
1903 
1904 	restart:
1905 		ND6_WLOCK();
1906 		ND_PREFIX_LIST_FOREACH_SAFE(pfx, next) {
1907 			struct in6_ifaddr *ia, *ia_next;
1908 			int _s;
1909 
1910 			/* Only flush prefixes for the given interface. */
1911 			if (ifp != lo0ifp && ifp != pfx->ndpr_ifp)
1912 				continue;
1913 
1914 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1915 				continue; /* XXX */
1916 
1917 			/* do we really have to remove addresses as well? */
1918 			_s = pserialize_read_enter();
1919 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1920 			     ia = ia_next) {
1921 				struct ifnet *ifa_ifp;
1922 				int bound;
1923 				struct psref psref;
1924 
1925 				/* ia might be removed.  keep the next ptr. */
1926 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1927 
1928 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1929 					continue;
1930 
1931 				if (ia->ia6_ndpr != pfx)
1932 					continue;
1933 
1934 				bound = curlwp_bind();
1935 				ia6_acquire(ia, &psref);
1936 				pserialize_read_exit(_s);
1937 				ND6_UNLOCK();
1938 
1939 				ifa_ifp = ia->ia_ifa.ifa_ifp;
1940 				if (ifa_ifp == ifp) {
1941 					/* Already have IFNET_LOCK(ifp) */
1942 					KASSERT(!if_is_deactivated(ifp));
1943 					ia6_release(ia, &psref);
1944 					in6_purgeaddr(&ia->ia_ifa);
1945 					curlwp_bindx(bound);
1946 					goto restart;
1947 				}
1948 				IFNET_LOCK(ifa_ifp);
1949 				/*
1950 				 * Need to take the lock first to prevent
1951 				 * if_detach from running in6_purgeaddr
1952 				 * concurrently.
1953 				 */
1954 				if (!if_is_deactivated(ifa_ifp)) {
1955 					ia6_release(ia, &psref);
1956 					in6_purgeaddr(&ia->ia_ifa);
1957 				} else {
1958 					/*
1959 					 * ifp is being destroyed, ia will be
1960 					 * destroyed by if_detach.
1961 					 */
1962 					ia6_release(ia, &psref);
1963 					/* XXX may cause busy loop */
1964 				}
1965 				IFNET_UNLOCK(ifa_ifp);
1966 				curlwp_bindx(bound);
1967 				goto restart;
1968 			}
1969 			pserialize_read_exit(_s);
1970 
1971 			KASSERTMSG(pfx->ndpr_refcnt == 0, "ndpr_refcnt=%d",
1972 			    pfx->ndpr_refcnt);
1973 			nd6_prelist_remove(pfx);
1974 		}
1975 		ND6_UNLOCK();
1976 		break;
1977 	}
1978 	case SIOCSRTRFLUSH_IN6:
1979 	{
1980 		/* flush all the default routers */
1981 		struct nd_defrouter *drtr, *next;
1982 
1983 		ND6_WLOCK();
1984 #if 0
1985 		/* XXX Is this really needed? */
1986 		nd6_defrouter_reset();
1987 #endif
1988 		ND_DEFROUTER_LIST_FOREACH_SAFE(drtr, next) {
1989 			/* Only flush routers for the given interface. */
1990 			if (ifp != lo0ifp && ifp != drtr->ifp)
1991 				continue;
1992 
1993 			nd6_defrtrlist_del(drtr, NULL);
1994 		}
1995 		nd6_defrouter_select();
1996 		ND6_UNLOCK();
1997 		break;
1998 	}
1999 	case SIOCGNBRINFO_IN6:
2000 	{
2001 		struct llentry *ln;
2002 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
2003 
2004 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
2005 			return error;
2006 
2007 		ln = nd6_lookup(&nb_addr, ifp, false);
2008 		if (ln == NULL) {
2009 			error = EINVAL;
2010 			break;
2011 		}
2012 		nbi->state = ln->ln_state;
2013 		nbi->asked = ln->ln_asked;
2014 		nbi->isrouter = ln->ln_router;
2015 		nbi->expire = ln->ln_expire ?
2016 		    time_mono_to_wall(ln->ln_expire) : 0;
2017 		LLE_RUNLOCK(ln);
2018 
2019 		break;
2020 	}
2021 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
2022 		ndif->ifindex = nd6_defifindex;
2023 		break;
2024 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
2025 		return nd6_setdefaultiface(ndif->ifindex);
2026 	}
2027 	return error;
2028 }
2029 
2030 void
2031 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
2032 {
2033 	struct mbuf *m_hold, *m_hold_next;
2034 	struct sockaddr_in6 sin6;
2035 
2036 	LLE_WLOCK_ASSERT(ln);
2037 
2038 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
2039 
2040 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
2041 
2042 	LLE_WUNLOCK(ln);
2043 	for (; m_hold != NULL; m_hold = m_hold_next) {
2044 		m_hold_next = m_hold->m_nextpkt;
2045 		m_hold->m_nextpkt = NULL;
2046 
2047 		/*
2048 		 * we assume ifp is not a p2p here, so
2049 		 * just set the 2nd argument as the
2050 		 * 1st one.
2051 		 */
2052 		ip6_if_output(ifp, ifp, m_hold, &sin6, NULL);
2053 	}
2054 	LLE_WLOCK(ln);
2055 }
2056 
2057 /*
2058  * Create neighbor cache entry and cache link-layer address,
2059  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
2060  */
2061 void
2062 nd6_cache_lladdr(
2063     struct ifnet *ifp,
2064     struct in6_addr *from,
2065     char *lladdr,
2066     int lladdrlen,
2067     int type,	/* ICMP6 type */
2068     int code	/* type dependent information */
2069 )
2070 {
2071 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
2072 	struct llentry *ln = NULL;
2073 	int is_newentry;
2074 	int do_update;
2075 	int olladdr;
2076 	int llchange;
2077 	int newstate = 0;
2078 	uint16_t router = 0;
2079 
2080 	KASSERT(ifp != NULL);
2081 	KASSERT(from != NULL);
2082 
2083 	/* nothing must be updated for unspecified address */
2084 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2085 		return;
2086 
2087 	/*
2088 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2089 	 * the caller.
2090 	 *
2091 	 * XXX If the link does not have link-layer adderss, what should
2092 	 * we do? (ifp->if_addrlen == 0)
2093 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2094 	 * description on it in NS section (RFC 2461 7.2.3).
2095 	 */
2096 
2097 	ln = nd6_lookup(from, ifp, true);
2098 	if (ln == NULL) {
2099 #if 0
2100 		/* nothing must be done if there's no lladdr */
2101 		if (!lladdr || !lladdrlen)
2102 			return NULL;
2103 #endif
2104 
2105 		ln = nd6_create(from, ifp);
2106 		is_newentry = 1;
2107 	} else {
2108 		/* do nothing if static ndp is set */
2109 		if (ln->la_flags & LLE_STATIC) {
2110 			LLE_WUNLOCK(ln);
2111 			return;
2112 		}
2113 		is_newentry = 0;
2114 	}
2115 
2116 	if (ln == NULL)
2117 		return;
2118 
2119 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2120 	if (olladdr && lladdr) {
2121 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2122 	} else
2123 		llchange = 0;
2124 
2125 	/*
2126 	 * newentry olladdr  lladdr  llchange	(*=record)
2127 	 *	0	n	n	--	(1)
2128 	 *	0	y	n	--	(2)
2129 	 *	0	n	y	--	(3) * STALE
2130 	 *	0	y	y	n	(4) *
2131 	 *	0	y	y	y	(5) * STALE
2132 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2133 	 *	1	--	y	--	(7) * STALE
2134 	 */
2135 
2136 	if (lladdr) {		/* (3-5) and (7) */
2137 		/*
2138 		 * Record source link-layer address
2139 		 * XXX is it dependent to ifp->if_type?
2140 		 */
2141 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2142 		ln->la_flags |= LLE_VALID;
2143 	}
2144 
2145 	if (!is_newentry) {
2146 		if ((!olladdr && lladdr) ||		/* (3) */
2147 		    (olladdr && lladdr && llchange)) {	/* (5) */
2148 			do_update = 1;
2149 			newstate = ND6_LLINFO_STALE;
2150 		} else					/* (1-2,4) */
2151 			do_update = 0;
2152 	} else {
2153 		do_update = 1;
2154 		if (lladdr == NULL)			/* (6) */
2155 			newstate = ND6_LLINFO_NOSTATE;
2156 		else					/* (7) */
2157 			newstate = ND6_LLINFO_STALE;
2158 	}
2159 
2160 	if (do_update) {
2161 		/*
2162 		 * Update the state of the neighbor cache.
2163 		 */
2164 		ln->ln_state = newstate;
2165 
2166 		if (ln->ln_state == ND6_LLINFO_STALE) {
2167 			/*
2168 			 * XXX: since nd6_output() below will cause
2169 			 * state tansition to DELAY and reset the timer,
2170 			 * we must set the timer now, although it is actually
2171 			 * meaningless.
2172 			 */
2173 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2174 
2175 			nd6_llinfo_release_pkts(ln, ifp);
2176 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2177 			/* probe right away */
2178 			nd6_llinfo_settimer((void *)ln, 0);
2179 		}
2180 	}
2181 
2182 	/*
2183 	 * ICMP6 type dependent behavior.
2184 	 *
2185 	 * NS: clear IsRouter if new entry
2186 	 * RS: clear IsRouter
2187 	 * RA: set IsRouter if there's lladdr
2188 	 * redir: clear IsRouter if new entry
2189 	 *
2190 	 * RA case, (1):
2191 	 * The spec says that we must set IsRouter in the following cases:
2192 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2193 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2194 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2195 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2196 	 * neighbor cache, this is similar to (6).
2197 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2198 	 *
2199 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2200 	 *							D R
2201 	 *	0	n	n	--	(1)	c   ?     s
2202 	 *	0	y	n	--	(2)	c   s     s
2203 	 *	0	n	y	--	(3)	c   s     s
2204 	 *	0	y	y	n	(4)	c   s     s
2205 	 *	0	y	y	y	(5)	c   s     s
2206 	 *	1	--	n	--	(6) c	c 	c s
2207 	 *	1	--	y	--	(7) c	c   s	c s
2208 	 *
2209 	 *					(c=clear s=set)
2210 	 */
2211 	switch (type & 0xff) {
2212 	case ND_NEIGHBOR_SOLICIT:
2213 		/*
2214 		 * New entry must have is_router flag cleared.
2215 		 */
2216 		if (is_newentry)	/* (6-7) */
2217 			ln->ln_router = 0;
2218 		break;
2219 	case ND_REDIRECT:
2220 		/*
2221 		 * If the icmp is a redirect to a better router, always set the
2222 		 * is_router flag.  Otherwise, if the entry is newly created,
2223 		 * clear the flag.  [RFC 2461, sec 8.3]
2224 		 */
2225 		if (code == ND_REDIRECT_ROUTER)
2226 			ln->ln_router = 1;
2227 		else if (is_newentry) /* (6-7) */
2228 			ln->ln_router = 0;
2229 		break;
2230 	case ND_ROUTER_SOLICIT:
2231 		/*
2232 		 * is_router flag must always be cleared.
2233 		 */
2234 		ln->ln_router = 0;
2235 		break;
2236 	case ND_ROUTER_ADVERT:
2237 		/*
2238 		 * Mark an entry with lladdr as a router.
2239 		 */
2240 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2241 		    (is_newentry && lladdr)) {			/* (7) */
2242 			ln->ln_router = 1;
2243 		}
2244 		break;
2245 	}
2246 
2247 	if (do_update && lladdr != NULL) {
2248 		struct sockaddr_in6 sin6;
2249 
2250 		sockaddr_in6_init(&sin6, from, 0, 0, 0);
2251 		rt_clonedmsg(is_newentry ? RTM_ADD : RTM_CHANGE,
2252 		    sin6tosa(&sin6), lladdr, ifp);
2253 	}
2254 
2255 	if (ln != NULL) {
2256 		router = ln->ln_router;
2257 		LLE_WUNLOCK(ln);
2258 	}
2259 
2260 	/*
2261 	 * If we have too many cache entries, initiate immediate
2262 	 * purging for some entries.
2263 	 */
2264 	if (is_newentry)
2265 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2266 
2267 	/*
2268 	 * When the link-layer address of a router changes, select the
2269 	 * best router again.  In particular, when the neighbor entry is newly
2270 	 * created, it might affect the selection policy.
2271 	 * Question: can we restrict the first condition to the "is_newentry"
2272 	 * case?
2273 	 * XXX: when we hear an RA from a new router with the link-layer
2274 	 * address option, nd6_defrouter_select() is called twice, since
2275 	 * defrtrlist_update called the function as well.  However, I believe
2276 	 * we can compromise the overhead, since it only happens the first
2277 	 * time.
2278 	 * XXX: although nd6_defrouter_select() should not have a bad effect
2279 	 * for those are not autoconfigured hosts, we explicitly avoid such
2280 	 * cases for safety.
2281 	 */
2282 	if (do_update && router && !ip6_forwarding &&
2283 	    nd6_accepts_rtadv(ndi)) {
2284 		ND6_WLOCK();
2285 		nd6_defrouter_select();
2286 		ND6_UNLOCK();
2287 	}
2288 }
2289 
2290 static void
2291 nd6_slowtimo(void *ignored_arg)
2292 {
2293 	struct nd_ifinfo *nd6if;
2294 	struct ifnet *ifp;
2295 	int s;
2296 
2297 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2298 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2299 	    nd6_slowtimo, NULL);
2300 
2301 	s = pserialize_read_enter();
2302 	IFNET_READER_FOREACH(ifp) {
2303 		nd6if = ND_IFINFO(ifp);
2304 		if (nd6if->basereachable && /* already initialized */
2305 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2306 			/*
2307 			 * Since reachable time rarely changes by router
2308 			 * advertisements, we SHOULD insure that a new random
2309 			 * value gets recomputed at least once every few hours.
2310 			 * (RFC 2461, 6.3.4)
2311 			 */
2312 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2313 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2314 		}
2315 	}
2316 	pserialize_read_exit(s);
2317 
2318 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2319 }
2320 
2321 /*
2322  * Return 0 if a neighbor cache is found. Return EWOULDBLOCK if a cache is not
2323  * found and trying to resolve a neighbor; in this case the mbuf is queued in
2324  * the list. Otherwise return errno after freeing the mbuf.
2325  */
2326 int
2327 nd6_resolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
2328     const struct sockaddr *_dst, uint8_t *lldst, size_t dstsize)
2329 {
2330 	struct llentry *ln = NULL;
2331 	bool created = false;
2332 	const struct sockaddr_in6 *dst = satocsin6(_dst);
2333 	int error;
2334 
2335 	/* discard the packet if IPv6 operation is disabled on the interface */
2336 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2337 		m_freem(m);
2338 		return ENETDOWN; /* better error? */
2339 	}
2340 
2341 	/*
2342 	 * Address resolution or Neighbor Unreachability Detection
2343 	 * for the next hop.
2344 	 * At this point, the destination of the packet must be a unicast
2345 	 * or an anycast address(i.e. not a multicast).
2346 	 */
2347 
2348 	/* Look up the neighbor cache for the nexthop */
2349 	ln = nd6_lookup(&dst->sin6_addr, ifp, false);
2350 
2351 	if (ln != NULL && (ln->la_flags & LLE_VALID) != 0 &&
2352 	    ln->ln_state == ND6_LLINFO_REACHABLE) {
2353 		/* Fast path */
2354 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2355 		LLE_RUNLOCK(ln);
2356 		return 0;
2357 	}
2358 	if (ln != NULL)
2359 		LLE_RUNLOCK(ln);
2360 
2361 	/* Slow path */
2362 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2363 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp))  {
2364 		/*
2365 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2366 		 * the condition below is not very efficient.  But we believe
2367 		 * it is tolerable, because this should be a rare case.
2368 		 */
2369 		ln = nd6_create(&dst->sin6_addr, ifp);
2370 		if (ln == NULL) {
2371 			char ip6buf[INET6_ADDRSTRLEN];
2372 			log(LOG_DEBUG,
2373 			    "%s: can't allocate llinfo for %s "
2374 			    "(ln=%p, rt=%p)\n", __func__,
2375 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
2376 			m_freem(m);
2377 			return ENOBUFS;
2378 		}
2379 		created = true;
2380 	}
2381 
2382 	if (ln == NULL) {
2383 		m_freem(m);
2384 		return ENETDOWN; /* better error? */
2385 	}
2386 
2387 	LLE_WLOCK_ASSERT(ln);
2388 
2389 	/* We don't have to do link-layer address resolution on a p2p link. */
2390 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2391 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2392 		ln->ln_state = ND6_LLINFO_STALE;
2393 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2394 	}
2395 
2396 	/*
2397 	 * The first time we send a packet to a neighbor whose entry is
2398 	 * STALE, we have to change the state to DELAY and a sets a timer to
2399 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2400 	 * neighbor unreachability detection on expiration.
2401 	 * (RFC 2461 7.3.3)
2402 	 */
2403 	if (ln->ln_state == ND6_LLINFO_STALE) {
2404 		ln->ln_asked = 0;
2405 		ln->ln_state = ND6_LLINFO_DELAY;
2406 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2407 	}
2408 
2409 	/*
2410 	 * If the neighbor cache entry has a state other than INCOMPLETE
2411 	 * (i.e. its link-layer address is already resolved), just
2412 	 * send the packet.
2413 	 */
2414 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2415 		KASSERT((ln->la_flags & LLE_VALID) != 0);
2416 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2417 		LLE_WUNLOCK(ln);
2418 		return 0;
2419 	}
2420 
2421 	/*
2422 	 * There is a neighbor cache entry, but no ethernet address
2423 	 * response yet.  Append this latest packet to the end of the
2424 	 * packet queue in the mbuf, unless the number of the packet
2425 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2426 	 * the oldest packet in the queue will be removed.
2427 	 */
2428 	if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2429 	    ln->ln_state == ND6_LLINFO_WAITDELETE)
2430 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2431 	if (ln->ln_hold) {
2432 		struct mbuf *m_hold;
2433 		int i;
2434 
2435 		i = 0;
2436 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2437 			i++;
2438 			if (m_hold->m_nextpkt == NULL) {
2439 				m_hold->m_nextpkt = m;
2440 				break;
2441 			}
2442 		}
2443 		while (i >= nd6_maxqueuelen) {
2444 			m_hold = ln->ln_hold;
2445 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2446 			m_freem(m_hold);
2447 			i--;
2448 		}
2449 	} else {
2450 		ln->ln_hold = m;
2451 	}
2452 
2453 	if (ln->ln_asked >= nd6_mmaxtries)
2454 		error = (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2455 		    EHOSTUNREACH : EHOSTDOWN;
2456 	else
2457 		error = EWOULDBLOCK;
2458 
2459 	/*
2460 	 * If there has been no NS for the neighbor after entering the
2461 	 * INCOMPLETE state, send the first solicitation.
2462 	 */
2463 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2464 		struct in6_addr src, *psrc;
2465 
2466 		ln->ln_asked++;
2467 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2468 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2469 		LLE_WUNLOCK(ln);
2470 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, NULL);
2471 	} else
2472 		LLE_WUNLOCK(ln);
2473 
2474 	if (created)
2475 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2476 
2477 	return error;
2478 }
2479 
2480 int
2481 nd6_need_cache(struct ifnet *ifp)
2482 {
2483 	/*
2484 	 * XXX: we currently do not make neighbor cache on any interface
2485 	 * other than ARCnet, Ethernet, FDDI and GIF.
2486 	 *
2487 	 * RFC2893 says:
2488 	 * - unidirectional tunnels needs no ND
2489 	 */
2490 	switch (ifp->if_type) {
2491 	case IFT_ARCNET:
2492 	case IFT_ETHER:
2493 	case IFT_FDDI:
2494 	case IFT_IEEE1394:
2495 	case IFT_CARP:
2496 	case IFT_GIF:		/* XXX need more cases? */
2497 	case IFT_PPP:
2498 	case IFT_TUNNEL:
2499 		return 1;
2500 	default:
2501 		return 0;
2502 	}
2503 }
2504 
2505 static void
2506 clear_llinfo_pqueue(struct llentry *ln)
2507 {
2508 	struct mbuf *m_hold, *m_hold_next;
2509 
2510 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2511 		m_hold_next = m_hold->m_nextpkt;
2512 		m_hold->m_nextpkt = NULL;
2513 		m_freem(m_hold);
2514 	}
2515 
2516 	ln->ln_hold = NULL;
2517 	return;
2518 }
2519 
2520 int
2521 nd6_sysctl(
2522     int name,
2523     void *oldp,	/* syscall arg, need copyout */
2524     size_t *oldlenp,
2525     void *newp,	/* syscall arg, need copyin */
2526     size_t newlen
2527 )
2528 {
2529 	int (*fill_func)(void *, size_t *);
2530 
2531 	if (newp)
2532 		return EPERM;
2533 
2534 	switch (name) {
2535 	case ICMPV6CTL_ND6_DRLIST:
2536 		fill_func = fill_drlist;
2537 		break;
2538 
2539 	case ICMPV6CTL_ND6_PRLIST:
2540 		fill_func = fill_prlist;
2541 		break;
2542 
2543 	case ICMPV6CTL_ND6_MAXQLEN:
2544 		return 0;
2545 
2546 	default:
2547 		return ENOPROTOOPT;
2548 	}
2549 
2550 	if (oldlenp == NULL)
2551 		return EINVAL;
2552 
2553 	size_t ol;
2554 	int error = (*fill_func)(NULL, &ol);	/* calc len needed */
2555 	if (error)
2556 		return error;
2557 
2558 	if (oldp == NULL) {
2559 		*oldlenp = ol;
2560 		return 0;
2561 	}
2562 
2563 	ol = *oldlenp = uimin(ol, *oldlenp);
2564 	if (ol == 0)
2565 		return 0;
2566 
2567 	void *p = kmem_alloc(ol, KM_SLEEP);
2568 	error = (*fill_func)(p, oldlenp);
2569 	if (!error)
2570 		error = copyout(p, oldp, *oldlenp);
2571 	kmem_free(p, ol);
2572 
2573 	return error;
2574 }
2575 
2576 static int
2577 fill_drlist(void *oldp, size_t *oldlenp)
2578 {
2579 	int error = 0;
2580 	struct in6_defrouter *d = NULL, *de = NULL;
2581 	struct nd_defrouter *dr;
2582 	size_t l;
2583 
2584 	if (oldp) {
2585 		d = (struct in6_defrouter *)oldp;
2586 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2587 	}
2588 	l = 0;
2589 
2590 	ND6_RLOCK();
2591 	ND_DEFROUTER_LIST_FOREACH(dr) {
2592 
2593 		if (oldp && d + 1 <= de) {
2594 			memset(d, 0, sizeof(*d));
2595 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2596 			if (sa6_recoverscope(&d->rtaddr)) {
2597 				char ip6buf[INET6_ADDRSTRLEN];
2598 				log(LOG_ERR,
2599 				    "scope error in router list (%s)\n",
2600 				    IN6_PRINT(ip6buf, &d->rtaddr.sin6_addr));
2601 				/* XXX: press on... */
2602 			}
2603 			d->flags = dr->flags;
2604 			d->rtlifetime = dr->rtlifetime;
2605 			d->expire = dr->expire ?
2606 			    time_mono_to_wall(dr->expire) : 0;
2607 			d->if_index = dr->ifp->if_index;
2608 		}
2609 
2610 		l += sizeof(*d);
2611 		if (d)
2612 			d++;
2613 	}
2614 	ND6_UNLOCK();
2615 
2616 	*oldlenp = l;	/* (void *)d - (void *)oldp */
2617 
2618 	return error;
2619 }
2620 
2621 static int
2622 fill_prlist(void *oldp, size_t *oldlenp)
2623 {
2624 	int error = 0;
2625 	struct nd_prefix *pr;
2626 	uint8_t *p = NULL, *ps = NULL;
2627 	uint8_t *pe = NULL;
2628 	size_t l;
2629 	char ip6buf[INET6_ADDRSTRLEN];
2630 
2631 	if (oldp) {
2632 		ps = p = (uint8_t*)oldp;
2633 		pe = (uint8_t*)oldp + *oldlenp;
2634 	}
2635 	l = 0;
2636 
2637 	ND6_RLOCK();
2638 	ND_PREFIX_LIST_FOREACH(pr) {
2639 		u_short advrtrs;
2640 		struct sockaddr_in6 sin6;
2641 		struct nd_pfxrouter *pfr;
2642 		struct in6_prefix pfx;
2643 
2644 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2645 		{
2646 			memset(&pfx, 0, sizeof(pfx));
2647 			ps = p;
2648 			pfx.prefix = pr->ndpr_prefix;
2649 
2650 			if (sa6_recoverscope(&pfx.prefix)) {
2651 				log(LOG_ERR,
2652 				    "scope error in prefix list (%s)\n",
2653 				    IN6_PRINT(ip6buf, &pfx.prefix.sin6_addr));
2654 				/* XXX: press on... */
2655 			}
2656 			pfx.raflags = pr->ndpr_raf;
2657 			pfx.prefixlen = pr->ndpr_plen;
2658 			pfx.vltime = pr->ndpr_vltime;
2659 			pfx.pltime = pr->ndpr_pltime;
2660 			pfx.if_index = pr->ndpr_ifp->if_index;
2661 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2662 				pfx.expire = 0;
2663 			else {
2664 				time_t maxexpire;
2665 
2666 				/* XXX: we assume time_t is signed. */
2667 				maxexpire = (-1) &
2668 				    ~((time_t)1 <<
2669 				    ((sizeof(maxexpire) * 8) - 1));
2670 				if (pr->ndpr_vltime <
2671 				    maxexpire - pr->ndpr_lastupdate) {
2672 					pfx.expire = pr->ndpr_lastupdate +
2673 						pr->ndpr_vltime;
2674 				} else
2675 					pfx.expire = maxexpire;
2676 			}
2677 			pfx.refcnt = pr->ndpr_refcnt;
2678 			pfx.flags = pr->ndpr_stateflags;
2679 			pfx.origin = PR_ORIG_RA;
2680 
2681 			p += sizeof(pfx); l += sizeof(pfx);
2682 
2683 			advrtrs = 0;
2684 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2685 				if (p + sizeof(sin6) > pe) {
2686 					advrtrs++;
2687 					continue;
2688 				}
2689 
2690 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2691 					    0, 0, 0);
2692 				if (sa6_recoverscope(&sin6)) {
2693 					log(LOG_ERR,
2694 					    "scope error in "
2695 					    "prefix list (%s)\n",
2696 					    IN6_PRINT(ip6buf,
2697 					    &pfr->router->rtaddr));
2698 				}
2699 				advrtrs++;
2700 				memcpy(p, &sin6, sizeof(sin6));
2701 				p += sizeof(sin6);
2702 				l += sizeof(sin6);
2703 			}
2704 			pfx.advrtrs = advrtrs;
2705 			memcpy(ps, &pfx, sizeof(pfx));
2706 		}
2707 		else {
2708 			l += sizeof(pfx);
2709 			advrtrs = 0;
2710 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2711 				advrtrs++;
2712 				l += sizeof(sin6);
2713 			}
2714 		}
2715 	}
2716 	ND6_UNLOCK();
2717 
2718 	*oldlenp = l;
2719 
2720 	return error;
2721 }
2722 
2723 static int
2724 nd6_setdefaultiface(int ifindex)
2725 {
2726 	ifnet_t *ifp;
2727 	int error = 0;
2728 	int s;
2729 
2730 	s = pserialize_read_enter();
2731 	ifp = if_byindex(ifindex);
2732 	if (ifp == NULL) {
2733 		pserialize_read_exit(s);
2734 		return EINVAL;
2735 	}
2736 	if (nd6_defifindex != ifindex) {
2737 		nd6_defifindex = ifindex;
2738 		nd6_defifp = nd6_defifindex > 0 ? ifp : NULL;
2739 
2740 		/*
2741 		 * Our current implementation assumes one-to-one maping between
2742 		 * interfaces and links, so it would be natural to use the
2743 		 * default interface as the default link.
2744 		 */
2745 		scope6_setdefault(nd6_defifp);
2746 	}
2747 	pserialize_read_exit(s);
2748 
2749 	return (error);
2750 }
2751