xref: /netbsd-src/sys/netinet6/nd6.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: nd6.c,v 1.226 2017/01/16 15:44:47 christos 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 1.226 2017/01/16 15:44:47 christos Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_net_mpsafe.h"
38 #endif
39 
40 #include "bridge.h"
41 #include "carp.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/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 *, size_t);
128 static int fill_prlist(void *, size_t *, 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 = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
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 	free(ext->nd_ifinfo, M_IP6NDP);
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 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
351 				nd6log(LOG_INFO,
352 				    "duplicated ND6 option found (type=%d)\n",
353 				    nd_opt->nd_opt_type);
354 				/* XXX bark? */
355 			} else {
356 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
357 					= nd_opt;
358 			}
359 			break;
360 		case ND_OPT_PREFIX_INFORMATION:
361 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
362 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
363 					= nd_opt;
364 			}
365 			ndopts->nd_opts_pi_end =
366 				(struct nd_opt_prefix_info *)nd_opt;
367 			break;
368 		default:
369 			/*
370 			 * Unknown options must be silently ignored,
371 			 * to accommodate future extension to the protocol.
372 			 */
373 			nd6log(LOG_DEBUG,
374 			    "nd6_options: unsupported option %d - "
375 			    "option ignored\n", nd_opt->nd_opt_type);
376 		}
377 
378 skip1:
379 		i++;
380 		if (i > nd6_maxndopt) {
381 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
382 			nd6log(LOG_INFO, "too many loop in nd opt\n");
383 			break;
384 		}
385 
386 		if (ndopts->nd_opts_done)
387 			break;
388 	}
389 
390 	return 0;
391 }
392 
393 /*
394  * ND6 timer routine to handle ND6 entries
395  */
396 void
397 nd6_llinfo_settimer(struct llentry *ln, time_t xtick)
398 {
399 
400 	CTASSERT(sizeof(time_t) > sizeof(int));
401 	LLE_WLOCK_ASSERT(ln);
402 
403 	KASSERT(xtick >= 0);
404 
405 	ln->ln_expire = time_uptime + xtick / hz;
406 	LLE_ADDREF(ln);
407 	if (xtick > INT_MAX) {
408 		ln->ln_ntick = xtick - INT_MAX;
409 		callout_reset(&ln->ln_timer_ch, INT_MAX,
410 		    nd6_llinfo_timer, ln);
411 	} else {
412 		ln->ln_ntick = 0;
413 		callout_reset(&ln->ln_timer_ch, xtick,
414 		    nd6_llinfo_timer, ln);
415 	}
416 }
417 
418 /*
419  * Gets source address of the first packet in hold queue
420  * and stores it in @src.
421  * Returns pointer to @src (if hold queue is not empty) or NULL.
422  */
423 static struct in6_addr *
424 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
425 {
426 	struct ip6_hdr *hip6;
427 
428 	if (ln == NULL || ln->ln_hold == NULL)
429 		return NULL;
430 
431 	/*
432 	 * assuming every packet in ln_hold has the same IP header
433 	 */
434 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
435 	/* XXX pullup? */
436 	if (sizeof(*hip6) < ln->ln_hold->m_len)
437 		*src = hip6->ip6_src;
438 	else
439 		src = NULL;
440 
441 	return src;
442 }
443 
444 static void
445 nd6_llinfo_timer(void *arg)
446 {
447 	struct llentry *ln = arg;
448 	struct ifnet *ifp;
449 	struct nd_ifinfo *ndi = NULL;
450 	bool send_ns = false;
451 	const struct in6_addr *daddr6 = NULL;
452 
453 	LLE_WLOCK(ln);
454 	if ((ln->la_flags & LLE_LINKED) == 0)
455 		goto out;
456 	if (ln->ln_ntick > 0) {
457 		nd6_llinfo_settimer(ln, ln->ln_ntick);
458 		goto out;
459 	}
460 
461 	if (callout_pending(&ln->la_timer)) {
462 		/*
463 		 * Here we are a bit odd here in the treatment of
464 		 * active/pending. If the pending bit is set, it got
465 		 * rescheduled before I ran. The active
466 		 * bit we ignore, since if it was stopped
467 		 * in ll_tablefree() and was currently running
468 		 * it would have return 0 so the code would
469 		 * not have deleted it since the callout could
470 		 * not be stopped so we want to go through
471 		 * with the delete here now. If the callout
472 		 * was restarted, the pending bit will be back on and
473 		 * we just want to bail since the callout_reset would
474 		 * return 1 and our reference would have been removed
475 		 * by nd6_llinfo_settimer above since canceled
476 		 * would have been 1.
477 		 */
478 		goto out;
479 	}
480 
481 	ifp = ln->lle_tbl->llt_ifp;
482 
483 	KASSERT(ifp != NULL);
484 
485 	ndi = ND_IFINFO(ifp);
486 
487 	switch (ln->ln_state) {
488 	case ND6_LLINFO_INCOMPLETE:
489 		if (ln->ln_asked < nd6_mmaxtries) {
490 			ln->ln_asked++;
491 			send_ns = true;
492 		} else {
493 			struct mbuf *m = ln->ln_hold;
494 			if (m) {
495 				struct mbuf *m0;
496 
497 				/*
498 				 * assuming every packet in ln_hold has
499 				 * the same IP header
500 				 */
501 				m0 = m->m_nextpkt;
502 				m->m_nextpkt = NULL;
503 				ln->ln_hold = m0;
504 				clear_llinfo_pqueue(ln);
505  			}
506 			nd6_free(ln, 0);
507 			ln = NULL;
508 			if (m != NULL) {
509 #ifndef NET_MPSAFE
510 				mutex_enter(softnet_lock);
511 				KERNEL_LOCK(1, NULL);
512 #endif
513 				icmp6_error2(m, ICMP6_DST_UNREACH,
514 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
515 #ifndef NET_MPSAFE
516 				KERNEL_UNLOCK_ONE(NULL);
517 				mutex_exit(softnet_lock);
518 #endif
519 			}
520 		}
521 		break;
522 	case ND6_LLINFO_REACHABLE:
523 		if (!ND6_LLINFO_PERMANENT(ln)) {
524 			ln->ln_state = ND6_LLINFO_STALE;
525 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
526 		}
527 		break;
528 
529 	case ND6_LLINFO_PURGE:
530 	case ND6_LLINFO_STALE:
531 		/* Garbage Collection(RFC 2461 5.3) */
532 		if (!ND6_LLINFO_PERMANENT(ln)) {
533 			nd6_free(ln, 1);
534 			ln = NULL;
535 		}
536 		break;
537 
538 	case ND6_LLINFO_DELAY:
539 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
540 			/* We need NUD */
541 			ln->ln_asked = 1;
542 			ln->ln_state = ND6_LLINFO_PROBE;
543 			daddr6 = &ln->r_l3addr.addr6;
544 			send_ns = true;
545 		} else {
546 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
547 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
548 		}
549 		break;
550 	case ND6_LLINFO_PROBE:
551 		if (ln->ln_asked < nd6_umaxtries) {
552 			ln->ln_asked++;
553 			daddr6 = &ln->r_l3addr.addr6;
554 			send_ns = true;
555 		} else {
556 			nd6_free(ln, 0);
557 			ln = NULL;
558 		}
559 		break;
560 	}
561 
562 	if (send_ns) {
563 		struct in6_addr src, *psrc;
564 		const struct in6_addr *taddr6 = &ln->r_l3addr.addr6;
565 
566 		nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
567 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
568 		LLE_FREE_LOCKED(ln);
569 		ln = NULL;
570 #ifndef NET_MPSAFE
571 		mutex_enter(softnet_lock);
572 		KERNEL_LOCK(1, NULL);
573 #endif
574 		nd6_ns_output(ifp, daddr6, taddr6, psrc, 0);
575 #ifndef NET_MPSAFE
576 		KERNEL_UNLOCK_ONE(NULL);
577 		mutex_exit(softnet_lock);
578 #endif
579 	}
580 
581 out:
582 	if (ln != NULL)
583 		LLE_FREE_LOCKED(ln);
584 }
585 
586 /*
587  * ND6 timer routine to expire default route list and prefix list
588  */
589 static void
590 nd6_timer_work(struct work *wk, void *arg)
591 {
592 	struct nd_defrouter *next_dr, *dr;
593 	struct nd_prefix *next_pr, *pr;
594 	struct in6_ifaddr *ia6, *nia6;
595 	int s, bound;
596 	struct psref psref;
597 
598 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
599 	    nd6_timer, NULL);
600 
601 #ifndef NET_MPSAFE
602 	mutex_enter(softnet_lock);
603 	KERNEL_LOCK(1, NULL);
604 #endif
605 
606 	/* expire default router list */
607 
608 	ND6_WLOCK();
609 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, next_dr) {
610 		if (dr->expire && dr->expire < time_uptime) {
611 			nd6_defrtrlist_del(dr, NULL);
612 		}
613 	}
614 	ND6_UNLOCK();
615 
616 	/*
617 	 * expire interface addresses.
618 	 * in the past the loop was inside prefix expiry processing.
619 	 * However, from a stricter speci-confrmance standpoint, we should
620 	 * rather separate address lifetimes and prefix lifetimes.
621 	 */
622 	bound = curlwp_bind();
623   addrloop:
624 	s = pserialize_read_enter();
625 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
626 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
627 
628 		ia6_acquire(ia6, &psref);
629 		pserialize_read_exit(s);
630 
631 		/* check address lifetime */
632 		if (IFA6_IS_INVALID(ia6)) {
633 			int regen = 0;
634 
635 			/*
636 			 * If the expiring address is temporary, try
637 			 * regenerating a new one.  This would be useful when
638 			 * we suspended a laptop PC, then turned it on after a
639 			 * period that could invalidate all temporary
640 			 * addresses.  Although we may have to restart the
641 			 * loop (see below), it must be after purging the
642 			 * address.  Otherwise, we'd see an infinite loop of
643 			 * regeneration.
644 			 */
645 			if (ip6_use_tempaddr &&
646 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
647 				if (regen_tmpaddr(ia6) == 0)
648 					regen = 1;
649 			}
650 
651 			ia6_release(ia6, &psref);
652  			in6_purgeaddr(&ia6->ia_ifa);
653 			ia6 = NULL;
654 
655 			if (regen)
656 				goto addrloop; /* XXX: see below */
657 		} else if (IFA6_IS_DEPRECATED(ia6)) {
658 			int oldflags = ia6->ia6_flags;
659 
660 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
661 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
662 				rt_newaddrmsg(RTM_NEWADDR,
663 				    (struct ifaddr *)ia6, 0, NULL);
664 			}
665 
666 			/*
667 			 * If a temporary address has just become deprecated,
668 			 * regenerate a new one if possible.
669 			 */
670 			if (ip6_use_tempaddr &&
671 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
672 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
673 
674 				if (regen_tmpaddr(ia6) == 0) {
675 					/*
676 					 * A new temporary address is
677 					 * generated.
678 					 * XXX: this means the address chain
679 					 * has changed while we are still in
680 					 * the loop.  Although the change
681 					 * would not cause disaster (because
682 					 * it's not a deletion, but an
683 					 * addition,) we'd rather restart the
684 					 * loop just for safety.  Or does this
685 					 * significantly reduce performance??
686 					 */
687 					ia6_release(ia6, &psref);
688 					goto addrloop;
689 				}
690 			}
691 		} else {
692 			/*
693 			 * A new RA might have made a deprecated address
694 			 * preferred.
695 			 */
696 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
697 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
698 				rt_newaddrmsg(RTM_NEWADDR,
699 				    (struct ifaddr *)ia6, 0, NULL);
700 			}
701 		}
702 		s = pserialize_read_enter();
703 		ia6_release(ia6, &psref);
704 	}
705 	pserialize_read_exit(s);
706 	curlwp_bindx(bound);
707 
708 	/* expire prefix list */
709 	ND6_WLOCK();
710 	ND_PREFIX_LIST_FOREACH_SAFE(pr, next_pr) {
711 		/*
712 		 * check prefix lifetime.
713 		 * since pltime is just for autoconf, pltime processing for
714 		 * prefix is not necessary.
715 		 */
716 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
717 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
718 
719 			/*
720 			 * address expiration and prefix expiration are
721 			 * separate.  NEVER perform in6_purgeaddr here.
722 			 */
723 
724 			nd6_prelist_remove(pr);
725 		}
726 	}
727 	ND6_UNLOCK();
728 
729 #ifndef NET_MPSAFE
730 	KERNEL_UNLOCK_ONE(NULL);
731 	mutex_exit(softnet_lock);
732 #endif
733 }
734 
735 static void
736 nd6_timer(void *ignored_arg)
737 {
738 
739 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
740 }
741 
742 /* ia6: deprecated/invalidated temporary address */
743 static int
744 regen_tmpaddr(const struct in6_ifaddr *ia6)
745 {
746 	struct ifaddr *ifa;
747 	struct ifnet *ifp;
748 	struct in6_ifaddr *public_ifa6 = NULL;
749 	int s;
750 
751 	ifp = ia6->ia_ifa.ifa_ifp;
752 	s = pserialize_read_enter();
753 	IFADDR_READER_FOREACH(ifa, ifp) {
754 		struct in6_ifaddr *it6;
755 
756 		if (ifa->ifa_addr->sa_family != AF_INET6)
757 			continue;
758 
759 		it6 = (struct in6_ifaddr *)ifa;
760 
761 		/* ignore no autoconf addresses. */
762 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
763 			continue;
764 
765 		/* ignore autoconf addresses with different prefixes. */
766 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
767 			continue;
768 
769 		/*
770 		 * Now we are looking at an autoconf address with the same
771 		 * prefix as ours.  If the address is temporary and is still
772 		 * preferred, do not create another one.  It would be rare, but
773 		 * could happen, for example, when we resume a laptop PC after
774 		 * a long period.
775 		 */
776 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
777 		    !IFA6_IS_DEPRECATED(it6)) {
778 			public_ifa6 = NULL;
779 			break;
780 		}
781 
782 		/*
783 		 * This is a public autoconf address that has the same prefix
784 		 * as ours.  If it is preferred, keep it.  We can't break the
785 		 * loop here, because there may be a still-preferred temporary
786 		 * address with the prefix.
787 		 */
788 		if (!IFA6_IS_DEPRECATED(it6))
789 			public_ifa6 = it6;
790 	}
791 
792 	if (public_ifa6 != NULL) {
793 		int e;
794 		struct psref psref;
795 
796 		ia6_acquire(public_ifa6, &psref);
797 		pserialize_read_exit(s);
798 		/*
799 		 * Random factor is introduced in the preferred lifetime, so
800 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
801 		 */
802 		ND6_WLOCK();
803 		e = in6_tmpifadd(public_ifa6, 0, 0);
804 		ND6_UNLOCK();
805 		if (e != 0) {
806 			ia6_release(public_ifa6, &psref);
807 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
808 			    " tmp addr, errno=%d\n", e);
809 			return -1;
810 		}
811 		ia6_release(public_ifa6, &psref);
812 		return 0;
813 	}
814 	pserialize_read_exit(s);
815 
816 	return -1;
817 }
818 
819 bool
820 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
821 {
822 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
823 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
824 		return true;
825 	case ND6_IFF_ACCEPT_RTADV:
826 		return ip6_accept_rtadv != 0;
827 	case ND6_IFF_OVERRIDE_RTADV:
828 	case 0:
829 	default:
830 		return false;
831 	}
832 }
833 
834 /*
835  * Nuke neighbor cache/prefix/default router management table, right before
836  * ifp goes away.
837  */
838 void
839 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
840 {
841 	struct nd_defrouter *dr, *ndr;
842 	struct nd_prefix *pr, *npr;
843 
844 	/*
845 	 * During detach, the ND info might be already removed, but
846 	 * then is explitly passed as argument.
847 	 * Otherwise get it from ifp->if_afdata.
848 	 */
849 	if (ext == NULL)
850 		ext = ifp->if_afdata[AF_INET6];
851 	if (ext == NULL)
852 		return;
853 
854 	ND6_WLOCK();
855 	/*
856 	 * Nuke default router list entries toward ifp.
857 	 * We defer removal of default router list entries that is installed
858 	 * in the routing table, in order to keep additional side effects as
859 	 * small as possible.
860 	 */
861 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
862 		if (dr->installed)
863 			continue;
864 
865 		if (dr->ifp == ifp) {
866 			KASSERT(ext != NULL);
867 			nd6_defrtrlist_del(dr, ext);
868 		}
869 	}
870 
871 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
872 		if (!dr->installed)
873 			continue;
874 
875 		if (dr->ifp == ifp) {
876 			KASSERT(ext != NULL);
877 			nd6_defrtrlist_del(dr, ext);
878 		}
879 	}
880 
881 	/* Nuke prefix list entries toward ifp */
882 	ND_PREFIX_LIST_FOREACH_SAFE(pr, npr) {
883 		if (pr->ndpr_ifp == ifp) {
884 			/*
885 			 * All addresses referencing pr should be already freed.
886 			 */
887 			KASSERT(pr->ndpr_refcnt == 0);
888 			nd6_prelist_remove(pr);
889 		}
890 	}
891 
892 	/* cancel default outgoing interface setting */
893 	if (nd6_defifindex == ifp->if_index)
894 		nd6_setdefaultiface(0);
895 
896 	/* XXX: too restrictive? */
897 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
898 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
899 		if (ndi && nd6_accepts_rtadv(ndi)) {
900 			/* refresh default router list */
901 			nd6_defrouter_select();
902 		}
903 	}
904 	ND6_UNLOCK();
905 
906 	/*
907 	 * We may not need to nuke the neighbor cache entries here
908 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
909 	 * nd6_purge() is invoked by in6_ifdetach() which is called
910 	 * from if_detach() where everything gets purged. However
911 	 * in6_ifdetach is directly called from vlan(4), so we still
912 	 * need to purge entries here.
913 	 */
914 	if (ext->lltable != NULL)
915 		lltable_purge_entries(ext->lltable);
916 }
917 
918 void
919 nd6_assert_purged(struct ifnet *ifp)
920 {
921 	struct nd_defrouter *dr;
922 	struct nd_prefix *pr;
923 	char ip6buf[INET6_ADDRSTRLEN] __diagused;
924 
925 	ND6_RLOCK();
926 	ND_DEFROUTER_LIST_FOREACH(dr) {
927 		KASSERTMSG(dr->ifp != ifp,
928 		    "defrouter %s remains on %s",
929 		    IN6_PRINT(ip6buf, &dr->rtaddr), ifp->if_xname);
930 	}
931 
932 	ND_PREFIX_LIST_FOREACH(pr) {
933 		KASSERTMSG(pr->ndpr_ifp != ifp,
934 		    "prefix %s/%d remains on %s",
935 		    IN6_PRINT(ip6buf, &pr->ndpr_prefix.sin6_addr),
936 		    pr->ndpr_plen, ifp->if_xname);
937 	}
938 	ND6_UNLOCK();
939 }
940 
941 struct llentry *
942 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
943 {
944 	struct sockaddr_in6 sin6;
945 	struct llentry *ln;
946 
947 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
948 
949 	IF_AFDATA_RLOCK(ifp);
950 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
951 	    sin6tosa(&sin6));
952 	IF_AFDATA_RUNLOCK(ifp);
953 
954 	return ln;
955 }
956 
957 struct llentry *
958 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
959 {
960 	struct sockaddr_in6 sin6;
961 	struct llentry *ln;
962 
963 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
964 
965 	IF_AFDATA_WLOCK(ifp);
966 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE,
967 	    sin6tosa(&sin6));
968 	IF_AFDATA_WUNLOCK(ifp);
969 
970 	if (ln != NULL)
971 		ln->ln_state = ND6_LLINFO_NOSTATE;
972 
973 	return ln;
974 }
975 
976 /*
977  * Test whether a given IPv6 address is a neighbor or not, ignoring
978  * the actual neighbor cache.  The neighbor cache is ignored in order
979  * to not reenter the routing code from within itself.
980  */
981 static int
982 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
983 {
984 	struct nd_prefix *pr;
985 	struct ifaddr *dstaddr;
986 	int s;
987 
988 	/*
989 	 * A link-local address is always a neighbor.
990 	 * XXX: a link does not necessarily specify a single interface.
991 	 */
992 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
993 		struct sockaddr_in6 sin6_copy;
994 		u_int32_t zone;
995 
996 		/*
997 		 * We need sin6_copy since sa6_recoverscope() may modify the
998 		 * content (XXX).
999 		 */
1000 		sin6_copy = *addr;
1001 		if (sa6_recoverscope(&sin6_copy))
1002 			return 0; /* XXX: should be impossible */
1003 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1004 			return 0;
1005 		if (sin6_copy.sin6_scope_id == zone)
1006 			return 1;
1007 		else
1008 			return 0;
1009 	}
1010 
1011 	/*
1012 	 * If the address matches one of our addresses,
1013 	 * it should be a neighbor.
1014 	 * If the address matches one of our on-link prefixes, it should be a
1015 	 * neighbor.
1016 	 */
1017 	ND6_RLOCK();
1018 	ND_PREFIX_LIST_FOREACH(pr) {
1019 		if (pr->ndpr_ifp != ifp)
1020 			continue;
1021 
1022 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1023 			struct rtentry *rt;
1024 
1025 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
1026 			if (rt == NULL)
1027 				continue;
1028 			/*
1029 			 * This is the case where multiple interfaces
1030 			 * have the same prefix, but only one is installed
1031 			 * into the routing table and that prefix entry
1032 			 * is not the one being examined here. In the case
1033 			 * where RADIX_MPATH is enabled, multiple route
1034 			 * entries (of the same rt_key value) will be
1035 			 * installed because the interface addresses all
1036 			 * differ.
1037 			 */
1038 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1039 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
1040 				rt_unref(rt);
1041 				continue;
1042 			}
1043 			rt_unref(rt);
1044 		}
1045 
1046 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1047 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1048 			ND6_UNLOCK();
1049 			return 1;
1050 		}
1051 	}
1052 	ND6_UNLOCK();
1053 
1054 	/*
1055 	 * If the address is assigned on the node of the other side of
1056 	 * a p2p interface, the address should be a neighbor.
1057 	 */
1058 	s = pserialize_read_enter();
1059 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1060 	if (dstaddr != NULL) {
1061 		if (dstaddr->ifa_ifp == ifp) {
1062 			pserialize_read_exit(s);
1063 			return 1;
1064 		}
1065 	}
1066 	pserialize_read_exit(s);
1067 
1068 	/*
1069 	 * If the default router list is empty, all addresses are regarded
1070 	 * as on-link, and thus, as a neighbor.
1071 	 */
1072 	ND6_RLOCK();
1073 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1074 	    ND_DEFROUTER_LIST_EMPTY() && nd6_defifindex == ifp->if_index) {
1075 		ND6_UNLOCK();
1076 		return 1;
1077 	}
1078 	ND6_UNLOCK();
1079 
1080 	return 0;
1081 }
1082 
1083 /*
1084  * Detect if a given IPv6 address identifies a neighbor on a given link.
1085  * XXX: should take care of the destination of a p2p link?
1086  */
1087 int
1088 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1089 {
1090 	struct nd_prefix *pr;
1091 	struct llentry *ln;
1092 	struct rtentry *rt;
1093 
1094 	/*
1095 	 * A link-local address is always a neighbor.
1096 	 * XXX: a link does not necessarily specify a single interface.
1097 	 */
1098 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1099 		struct sockaddr_in6 sin6_copy;
1100 		u_int32_t zone;
1101 
1102 		/*
1103 		 * We need sin6_copy since sa6_recoverscope() may modify the
1104 		 * content (XXX).
1105 		 */
1106 		sin6_copy = *addr;
1107 		if (sa6_recoverscope(&sin6_copy))
1108 			return 0; /* XXX: should be impossible */
1109 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1110 			return 0;
1111 		if (sin6_copy.sin6_scope_id == zone)
1112 			return 1;
1113 		else
1114 			return 0;
1115 	}
1116 
1117 	/*
1118 	 * If the address matches one of our on-link prefixes, it should be a
1119 	 * neighbor.
1120 	 */
1121 	ND6_RLOCK();
1122 	ND_PREFIX_LIST_FOREACH(pr) {
1123 		if (pr->ndpr_ifp != ifp)
1124 			continue;
1125 
1126 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1127 			continue;
1128 
1129 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1130 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1131 			ND6_UNLOCK();
1132 			return 1;
1133 		}
1134 	}
1135 
1136 	/*
1137 	 * If the default router list is empty, all addresses are regarded
1138 	 * as on-link, and thus, as a neighbor.
1139 	 * XXX: we restrict the condition to hosts, because routers usually do
1140 	 * not have the "default router list".
1141 	 */
1142 	if (!ip6_forwarding && ND_DEFROUTER_LIST_EMPTY() &&
1143 	    nd6_defifindex == ifp->if_index) {
1144 		ND6_UNLOCK();
1145 		return 1;
1146 	}
1147 	ND6_UNLOCK();
1148 
1149 	if (nd6_is_new_addr_neighbor(addr, ifp))
1150 		return 1;
1151 
1152 	/*
1153 	 * Even if the address matches none of our addresses, it might be
1154 	 * in the neighbor cache or a connected route.
1155 	 */
1156 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1157 	if (ln != NULL) {
1158 		LLE_RUNLOCK(ln);
1159 		return 1;
1160 	}
1161 
1162 	rt = rtalloc1(sin6tocsa(addr), 0);
1163 	if (rt == NULL)
1164 		return 0;
1165 
1166 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1167 #if NBRIDGE > 0
1168 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1169 #endif
1170 #if NCARP > 0
1171 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1172 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1173 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1174 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1175 #endif
1176 	    )) {
1177 		rt_unref(rt);
1178 		return 1;
1179 	}
1180 	rt_unref(rt);
1181 
1182 	return 0;
1183 }
1184 
1185 /*
1186  * Free an nd6 llinfo entry.
1187  * Since the function would cause significant changes in the kernel, DO NOT
1188  * make it global, unless you have a strong reason for the change, and are sure
1189  * that the change is safe.
1190  */
1191 static void
1192 nd6_free(struct llentry *ln, int gc)
1193 {
1194 	struct nd_defrouter *dr;
1195 	struct ifnet *ifp;
1196 	struct in6_addr *in6;
1197 
1198 	KASSERT(ln != NULL);
1199 	LLE_WLOCK_ASSERT(ln);
1200 
1201 	ifp = ln->lle_tbl->llt_ifp;
1202 	in6 = &ln->r_l3addr.addr6;
1203 	/*
1204 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1205 	 * even though it is not harmful, it was not really necessary.
1206 	 */
1207 
1208 	if (!ip6_forwarding) {
1209 		ND6_WLOCK();
1210 		dr = nd6_defrouter_lookup(in6, ifp);
1211 
1212 		if (dr != NULL && dr->expire &&
1213 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1214 			/*
1215 			 * If the reason for the deletion is just garbage
1216 			 * collection, and the neighbor is an active default
1217 			 * router, do not delete it.  Instead, reset the GC
1218 			 * timer using the router's lifetime.
1219 			 * Simply deleting the entry would affect default
1220 			 * router selection, which is not necessarily a good
1221 			 * thing, especially when we're using router preference
1222 			 * values.
1223 			 * XXX: the check for ln_state would be redundant,
1224 			 *      but we intentionally keep it just in case.
1225 			 */
1226 			if (dr->expire > time_uptime)
1227 				nd6_llinfo_settimer(ln,
1228 				    (dr->expire - time_uptime) * hz);
1229 			else
1230 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1231 			ND6_UNLOCK();
1232 			LLE_WUNLOCK(ln);
1233 			return;
1234 		}
1235 
1236 		if (ln->ln_router || dr) {
1237 			/*
1238 			 * We need to unlock to avoid a LOR with nd6_rt_flush()
1239 			 * with the rnh and for the calls to
1240 			 * nd6_pfxlist_onlink_check() and nd6_defrouter_select() in the
1241 			 * block further down for calls into nd6_lookup().
1242 			 * We still hold a ref.
1243 			 */
1244 			LLE_WUNLOCK(ln);
1245 
1246 			/*
1247 			 * nd6_rt_flush must be called whether or not the neighbor
1248 			 * is in the Default Router List.
1249 			 * See a corresponding comment in nd6_na_input().
1250 			 */
1251 			nd6_rt_flush(in6, ifp);
1252 		}
1253 
1254 		if (dr) {
1255 			/*
1256 			 * Unreachablity of a router might affect the default
1257 			 * router selection and on-link detection of advertised
1258 			 * prefixes.
1259 			 */
1260 
1261 			/*
1262 			 * Temporarily fake the state to choose a new default
1263 			 * router and to perform on-link determination of
1264 			 * prefixes correctly.
1265 			 * Below the state will be set correctly,
1266 			 * or the entry itself will be deleted.
1267 			 */
1268 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1269 
1270 			/*
1271 			 * Since nd6_defrouter_select() does not affect the
1272 			 * on-link determination and MIP6 needs the check
1273 			 * before the default router selection, we perform
1274 			 * the check now.
1275 			 */
1276 			nd6_pfxlist_onlink_check();
1277 
1278 			/*
1279 			 * refresh default router list
1280 			 */
1281 			nd6_defrouter_select();
1282 		}
1283 
1284 #ifdef __FreeBSD__
1285 		/*
1286 		 * If this entry was added by an on-link redirect, remove the
1287 		 * corresponding host route.
1288 		 */
1289 		if (ln->la_flags & LLE_REDIRECT)
1290 			nd6_free_redirect(ln);
1291 #endif
1292 		ND6_UNLOCK();
1293 
1294 		if (ln->ln_router || dr)
1295 			LLE_WLOCK(ln);
1296 	}
1297 
1298 	/*
1299 	 * Save to unlock. We still hold an extra reference and will not
1300 	 * free(9) in llentry_free() if someone else holds one as well.
1301 	 */
1302 	LLE_WUNLOCK(ln);
1303 	IF_AFDATA_LOCK(ifp);
1304 	LLE_WLOCK(ln);
1305 
1306 	lltable_free_entry(LLTABLE6(ifp), ln);
1307 
1308 	IF_AFDATA_UNLOCK(ifp);
1309 }
1310 
1311 /*
1312  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1313  *
1314  * XXX cost-effective methods?
1315  */
1316 void
1317 nd6_nud_hint(struct rtentry *rt)
1318 {
1319 	struct llentry *ln;
1320 	struct ifnet *ifp;
1321 
1322 	if (rt == NULL)
1323 		return;
1324 
1325 	ifp = rt->rt_ifp;
1326 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1327 	if (ln == NULL)
1328 		return;
1329 
1330 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1331 		goto done;
1332 
1333 	/*
1334 	 * if we get upper-layer reachability confirmation many times,
1335 	 * it is possible we have false information.
1336 	 */
1337 	ln->ln_byhint++;
1338 	if (ln->ln_byhint > nd6_maxnudhint)
1339 		goto done;
1340 
1341 	ln->ln_state = ND6_LLINFO_REACHABLE;
1342 	if (!ND6_LLINFO_PERMANENT(ln))
1343 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1344 
1345 done:
1346 	LLE_WUNLOCK(ln);
1347 
1348 	return;
1349 }
1350 
1351 struct gc_args {
1352 	int gc_entries;
1353 	const struct in6_addr *skip_in6;
1354 };
1355 
1356 static int
1357 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1358 {
1359 	struct gc_args *args = farg;
1360 	int *n = &args->gc_entries;
1361 	const struct in6_addr *skip_in6 = args->skip_in6;
1362 
1363 	if (*n <= 0)
1364 		return 0;
1365 
1366 	if (ND6_LLINFO_PERMANENT(ln))
1367 		return 0;
1368 
1369 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1370 		return 0;
1371 
1372 	LLE_WLOCK(ln);
1373 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1374 		ln->ln_state = ND6_LLINFO_STALE;
1375 	else
1376 		ln->ln_state = ND6_LLINFO_PURGE;
1377 	nd6_llinfo_settimer(ln, 0);
1378 	LLE_WUNLOCK(ln);
1379 
1380 	(*n)--;
1381 	return 0;
1382 }
1383 
1384 static void
1385 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1386 {
1387 
1388 	if (ip6_neighborgcthresh >= 0 &&
1389 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1390 		struct gc_args gc_args = {10, in6};
1391 		/*
1392 		 * XXX entries that are "less recently used" should be
1393 		 * freed first.
1394 		 */
1395 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1396 	}
1397 }
1398 
1399 void
1400 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1401 {
1402 	struct sockaddr *gate = rt->rt_gateway;
1403 	struct ifnet *ifp = rt->rt_ifp;
1404 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1405 	struct ifaddr *ifa;
1406 
1407 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1408 
1409 	if (req == RTM_LLINFO_UPD) {
1410 		int rc;
1411 		struct in6_addr *in6;
1412 		struct in6_addr in6_all;
1413 		int anycast;
1414 
1415 		if ((ifa = info->rti_ifa) == NULL)
1416 			return;
1417 
1418 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1419 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1420 
1421 		in6_all = in6addr_linklocal_allnodes;
1422 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1423 			log(LOG_ERR, "%s: failed to set scope %s "
1424 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1425 			return;
1426 		}
1427 
1428 		/* XXX don't set Override for proxy addresses */
1429 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1430 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1431 #if 0
1432 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1433 #endif
1434 		    , 1, NULL);
1435 		return;
1436 	}
1437 
1438 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1439 		return;
1440 
1441 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1442 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1443 		/*
1444 		 * This is probably an interface direct route for a link
1445 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1446 		 * We do not need special treatment below for such a route.
1447 		 * Moreover, the RTF_LLINFO flag which would be set below
1448 		 * would annoy the ndp(8) command.
1449 		 */
1450 		return;
1451 	}
1452 
1453 	switch (req) {
1454 	case RTM_ADD: {
1455 		int s;
1456 
1457 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1458 		/*
1459 		 * There is no backward compatibility :)
1460 		 *
1461 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1462 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1463 		 *	   rt->rt_flags |= RTF_CLONING;
1464 		 */
1465 		/* XXX should move to route.c? */
1466 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1467 			union {
1468 				struct sockaddr sa;
1469 				struct sockaddr_dl sdl;
1470 				struct sockaddr_storage ss;
1471 			} u;
1472 			/*
1473 			 * Case 1: This route should come from a route to
1474 			 * interface (RTF_CLONING case) or the route should be
1475 			 * treated as on-link but is currently not
1476 			 * (RTF_LLINFO && ln == NULL case).
1477 			 */
1478 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1479 			    ifp->if_index, ifp->if_type,
1480 			    NULL, namelen, NULL, addrlen) == NULL) {
1481 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1482 				    "failed on %s\n", __func__, __LINE__,
1483 				    sizeof(u.ss), if_name(ifp));
1484 			}
1485 			rt_setgate(rt, &u.sa);
1486 			gate = rt->rt_gateway;
1487 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1488 			if (gate == NULL) {
1489 				log(LOG_ERR,
1490 				    "%s: rt_setgate failed on %s\n", __func__,
1491 				    if_name(ifp));
1492 				break;
1493 			}
1494 
1495 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1496 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1497 				break;
1498 		}
1499 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1500 		/*
1501 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1502 		 * We don't do that here since llinfo is not ready yet.
1503 		 *
1504 		 * There are also couple of other things to be discussed:
1505 		 * - unsolicited NA code needs improvement beforehand
1506 		 * - RFC2461 says we MAY send multicast unsolicited NA
1507 		 *   (7.2.6 paragraph 4), however, it also says that we
1508 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1509 		 *   we don't have anything like it right now.
1510 		 *   note that the mechanism needs a mutual agreement
1511 		 *   between proxies, which means that we need to implement
1512 		 *   a new protocol, or a new kludge.
1513 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1514 		 *   we need to check ip6forwarding before sending it.
1515 		 *   (or should we allow proxy ND configuration only for
1516 		 *   routers?  there's no mention about proxy ND from hosts)
1517 		 */
1518 #if 0
1519 		/* XXX it does not work */
1520 		if (rt->rt_flags & RTF_ANNOUNCE)
1521 			nd6_na_output(ifp,
1522 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1523 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1524 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1525 			      1, NULL);
1526 #endif
1527 
1528 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1529 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1530 			/*
1531 			 * Address resolution isn't necessary for a point to
1532 			 * point link, so we can skip this test for a p2p link.
1533 			 */
1534 			if (gate->sa_family != AF_LINK ||
1535 			    gate->sa_len <
1536 			    sockaddr_dl_measure(namelen, addrlen)) {
1537 				log(LOG_DEBUG,
1538 				    "nd6_rtrequest: bad gateway value: %s\n",
1539 				    if_name(ifp));
1540 				break;
1541 			}
1542 			satosdl(gate)->sdl_type = ifp->if_type;
1543 			satosdl(gate)->sdl_index = ifp->if_index;
1544 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1545 		}
1546 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1547 
1548 		/*
1549 		 * When called from rt_ifa_addlocal, we cannot depend on that
1550 		 * the address (rt_getkey(rt)) exits in the address list of the
1551 		 * interface. So check RTF_LOCAL instead.
1552 		 */
1553 		if (rt->rt_flags & RTF_LOCAL) {
1554 			if (nd6_useloopback)
1555 				rt->rt_ifp = lo0ifp;	/* XXX */
1556 			break;
1557 		}
1558 
1559 		/*
1560 		 * check if rt_getkey(rt) is an address assigned
1561 		 * to the interface.
1562 		 */
1563 		s = pserialize_read_enter();
1564 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1565 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1566 		if (ifa != NULL) {
1567 			if (nd6_useloopback) {
1568 				rt->rt_ifp = lo0ifp;	/* XXX */
1569 				/*
1570 				 * Make sure rt_ifa be equal to the ifaddr
1571 				 * corresponding to the address.
1572 				 * We need this because when we refer
1573 				 * rt_ifa->ia6_flags in ip6_input, we assume
1574 				 * that the rt_ifa points to the address instead
1575 				 * of the loopback address.
1576 				 */
1577 				if (ifa != rt->rt_ifa)
1578 					rt_replace_ifa(rt, ifa);
1579 			}
1580 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1581 			/* join solicited node multicast for proxy ND */
1582 			if (ifp->if_flags & IFF_MULTICAST) {
1583 				struct in6_addr llsol;
1584 				int error;
1585 
1586 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1587 				llsol.s6_addr32[0] = htonl(0xff020000);
1588 				llsol.s6_addr32[1] = 0;
1589 				llsol.s6_addr32[2] = htonl(1);
1590 				llsol.s6_addr8[12] = 0xff;
1591 				if (in6_setscope(&llsol, ifp, NULL))
1592 					goto out;
1593 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1594 					char ip6buf[INET6_ADDRSTRLEN];
1595 					nd6log(LOG_ERR, "%s: failed to join "
1596 					    "%s (errno=%d)\n", if_name(ifp),
1597 					    IN6_PRINT(ip6buf, &llsol), error);
1598 				}
1599 			}
1600 		}
1601 	out:
1602 		pserialize_read_exit(s);
1603 		/*
1604 		 * If we have too many cache entries, initiate immediate
1605 		 * purging for some entries.
1606 		 */
1607 		if (rt->rt_ifp != NULL)
1608 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1609 		break;
1610 	    }
1611 
1612 	case RTM_DELETE:
1613 		/* leave from solicited node multicast for proxy ND */
1614 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1615 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1616 			struct in6_addr llsol;
1617 			struct in6_multi *in6m;
1618 
1619 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1620 			llsol.s6_addr32[0] = htonl(0xff020000);
1621 			llsol.s6_addr32[1] = 0;
1622 			llsol.s6_addr32[2] = htonl(1);
1623 			llsol.s6_addr8[12] = 0xff;
1624 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1625 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1626 				if (in6m)
1627 					in6_delmulti(in6m);
1628 			}
1629 		}
1630 		break;
1631 	}
1632 }
1633 
1634 int
1635 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1636 {
1637 	struct in6_drlist *drl = (struct in6_drlist *)data;
1638 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1639 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1640 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1641 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1642 	struct nd_defrouter *dr;
1643 	struct nd_prefix *pr;
1644 	int i = 0, error = 0;
1645 
1646 	switch (cmd) {
1647 	case SIOCGDRLST_IN6:
1648 		/*
1649 		 * obsolete API, use sysctl under net.inet6.icmp6
1650 		 */
1651 		memset(drl, 0, sizeof(*drl));
1652 		ND6_RLOCK();
1653 		ND_DEFROUTER_LIST_FOREACH(dr) {
1654 			if (i >= DRLSTSIZ)
1655 				break;
1656 			drl->defrouter[i].rtaddr = dr->rtaddr;
1657 			in6_clearscope(&drl->defrouter[i].rtaddr);
1658 
1659 			drl->defrouter[i].flags = dr->flags;
1660 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1661 			drl->defrouter[i].expire = dr->expire ?
1662 			    time_mono_to_wall(dr->expire) : 0;
1663 			drl->defrouter[i].if_index = dr->ifp->if_index;
1664 			i++;
1665 		}
1666 		ND6_UNLOCK();
1667 		break;
1668 	case SIOCGPRLST_IN6:
1669 		/*
1670 		 * obsolete API, use sysctl under net.inet6.icmp6
1671 		 *
1672 		 * XXX the structure in6_prlist was changed in backward-
1673 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1674 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1675 		 */
1676 		/*
1677 		 * XXX meaning of fields, especialy "raflags", is very
1678 		 * differnet between RA prefix list and RR/static prefix list.
1679 		 * how about separating ioctls into two?
1680 		 */
1681 		memset(oprl, 0, sizeof(*oprl));
1682 		ND6_RLOCK();
1683 		ND_PREFIX_LIST_FOREACH(pr) {
1684 			struct nd_pfxrouter *pfr;
1685 			int j;
1686 
1687 			if (i >= PRLSTSIZ)
1688 				break;
1689 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1690 			oprl->prefix[i].raflags = pr->ndpr_raf;
1691 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1692 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1693 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1694 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1695 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1696 				oprl->prefix[i].expire = 0;
1697 			else {
1698 				time_t maxexpire;
1699 
1700 				/* XXX: we assume time_t is signed. */
1701 				maxexpire = (-1) &
1702 				    ~((time_t)1 <<
1703 				    ((sizeof(maxexpire) * 8) - 1));
1704 				if (pr->ndpr_vltime <
1705 				    maxexpire - pr->ndpr_lastupdate) {
1706 					time_t expire;
1707 					expire = pr->ndpr_lastupdate +
1708 					    pr->ndpr_vltime;
1709 					oprl->prefix[i].expire = expire ?
1710 					    time_mono_to_wall(expire) : 0;
1711 				} else
1712 					oprl->prefix[i].expire = maxexpire;
1713 			}
1714 
1715 			j = 0;
1716 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1717 				if (j < DRLSTSIZ) {
1718 #define RTRADDR oprl->prefix[i].advrtr[j]
1719 					RTRADDR = pfr->router->rtaddr;
1720 					in6_clearscope(&RTRADDR);
1721 #undef RTRADDR
1722 				}
1723 				j++;
1724 			}
1725 			oprl->prefix[i].advrtrs = j;
1726 			oprl->prefix[i].origin = PR_ORIG_RA;
1727 
1728 			i++;
1729 		}
1730 		ND6_UNLOCK();
1731 
1732 		break;
1733 	case OSIOCGIFINFO_IN6:
1734 #define ND	ndi->ndi
1735 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1736 		memset(&ND, 0, sizeof(ND));
1737 		ND.linkmtu = IN6_LINKMTU(ifp);
1738 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1739 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1740 		ND.reachable = ND_IFINFO(ifp)->reachable;
1741 		ND.retrans = ND_IFINFO(ifp)->retrans;
1742 		ND.flags = ND_IFINFO(ifp)->flags;
1743 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1744 		ND.chlim = ND_IFINFO(ifp)->chlim;
1745 		break;
1746 	case SIOCGIFINFO_IN6:
1747 		ND = *ND_IFINFO(ifp);
1748 		break;
1749 	case SIOCSIFINFO_IN6:
1750 		/*
1751 		 * used to change host variables from userland.
1752 		 * intented for a use on router to reflect RA configurations.
1753 		 */
1754 		/* 0 means 'unspecified' */
1755 		if (ND.linkmtu != 0) {
1756 			if (ND.linkmtu < IPV6_MMTU ||
1757 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1758 				error = EINVAL;
1759 				break;
1760 			}
1761 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1762 		}
1763 
1764 		if (ND.basereachable != 0) {
1765 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1766 
1767 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1768 			if (ND.basereachable != obasereachable)
1769 				ND_IFINFO(ifp)->reachable =
1770 				    ND_COMPUTE_RTIME(ND.basereachable);
1771 		}
1772 		if (ND.retrans != 0)
1773 			ND_IFINFO(ifp)->retrans = ND.retrans;
1774 		if (ND.chlim != 0)
1775 			ND_IFINFO(ifp)->chlim = ND.chlim;
1776 		/* FALLTHROUGH */
1777 	case SIOCSIFINFO_FLAGS:
1778 	{
1779 		struct ifaddr *ifa;
1780 		struct in6_ifaddr *ia;
1781 		int s;
1782 
1783 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1784 		    !(ND.flags & ND6_IFF_IFDISABLED))
1785 		{
1786 			/*
1787 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1788 			 * has a link-local address with IN6_IFF_DUPLICATED,
1789 			 * do not clear ND6_IFF_IFDISABLED.
1790 			 * See RFC 4862, section 5.4.5.
1791 			 */
1792 			int duplicated_linklocal = 0;
1793 
1794 			s = pserialize_read_enter();
1795 			IFADDR_READER_FOREACH(ifa, ifp) {
1796 				if (ifa->ifa_addr->sa_family != AF_INET6)
1797 					continue;
1798 				ia = (struct in6_ifaddr *)ifa;
1799 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1800 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1801 				{
1802 					duplicated_linklocal = 1;
1803 					break;
1804 				}
1805 			}
1806 			pserialize_read_exit(s);
1807 
1808 			if (duplicated_linklocal) {
1809 				ND.flags |= ND6_IFF_IFDISABLED;
1810 				log(LOG_ERR, "Cannot enable an interface"
1811 				    " with a link-local address marked"
1812 				    " duplicate.\n");
1813 			} else {
1814 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1815 				if (ifp->if_flags & IFF_UP)
1816 					in6_if_up(ifp);
1817 			}
1818 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1819 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1820 			int bound = curlwp_bind();
1821 			/* Mark all IPv6 addresses as tentative. */
1822 
1823 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1824 			s = pserialize_read_enter();
1825 			IFADDR_READER_FOREACH(ifa, ifp) {
1826 				struct psref psref;
1827 				if (ifa->ifa_addr->sa_family != AF_INET6)
1828 					continue;
1829 				ifa_acquire(ifa, &psref);
1830 				pserialize_read_exit(s);
1831 
1832 				nd6_dad_stop(ifa);
1833 
1834 				ia = (struct in6_ifaddr *)ifa;
1835 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1836 
1837 				s = pserialize_read_enter();
1838 				ifa_release(ifa, &psref);
1839 			}
1840 			pserialize_read_exit(s);
1841 			curlwp_bindx(bound);
1842 		}
1843 
1844 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1845 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1846 				/* auto_linklocal 0->1 transition */
1847 
1848 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1849 				in6_ifattach(ifp, NULL);
1850 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1851 			    ifp->if_flags & IFF_UP)
1852 			{
1853 				/*
1854 				 * When the IF already has
1855 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1856 				 * address is assigned, and IFF_UP, try to
1857 				 * assign one.
1858 				 */
1859 				int haslinklocal = 0;
1860 
1861 				s = pserialize_read_enter();
1862 				IFADDR_READER_FOREACH(ifa, ifp) {
1863 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1864 						continue;
1865 					ia = (struct in6_ifaddr *)ifa;
1866 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1867 						haslinklocal = 1;
1868 						break;
1869 					}
1870 				}
1871 				pserialize_read_exit(s);
1872 				if (!haslinklocal)
1873 					in6_ifattach(ifp, NULL);
1874 			}
1875 		}
1876 	}
1877 		ND_IFINFO(ifp)->flags = ND.flags;
1878 		break;
1879 #undef ND
1880 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1881 		/* sync kernel routing table with the default router list */
1882 		ND6_WLOCK();
1883 		nd6_defrouter_reset();
1884 		nd6_defrouter_select();
1885 		ND6_UNLOCK();
1886 		break;
1887 	case SIOCSPFXFLUSH_IN6:
1888 	{
1889 		/* flush all the prefix advertised by routers */
1890 		struct nd_prefix *pfx, *next;
1891 
1892 	restart:
1893 		ND6_WLOCK();
1894 		ND_PREFIX_LIST_FOREACH_SAFE(pfx, next) {
1895 			struct in6_ifaddr *ia, *ia_next;
1896 			int _s;
1897 
1898 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1899 				continue; /* XXX */
1900 
1901 			/* do we really have to remove addresses as well? */
1902 			_s = pserialize_read_enter();
1903 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1904 			     ia = ia_next) {
1905 				/* ia might be removed.  keep the next ptr. */
1906 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1907 
1908 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1909 					continue;
1910 
1911 				if (ia->ia6_ndpr == pfx) {
1912 					pserialize_read_exit(_s);
1913 					ND6_UNLOCK();
1914 					/* XXX NOMPSAFE? */
1915 					/* in6_purgeaddr may destroy pfx. */
1916 					in6_purgeaddr(&ia->ia_ifa);
1917 					goto restart;
1918 				}
1919 			}
1920 			pserialize_read_exit(_s);
1921 
1922 			KASSERT(pfx->ndpr_refcnt == 0);
1923 			nd6_prelist_remove(pfx);
1924 		}
1925 		ND6_UNLOCK();
1926 		break;
1927 	}
1928 	case SIOCSRTRFLUSH_IN6:
1929 	{
1930 		/* flush all the default routers */
1931 		struct nd_defrouter *drtr, *next;
1932 
1933 		ND6_WLOCK();
1934 		nd6_defrouter_reset();
1935 		ND_DEFROUTER_LIST_FOREACH_SAFE(drtr, next) {
1936 			nd6_defrtrlist_del(drtr, NULL);
1937 		}
1938 		nd6_defrouter_select();
1939 		ND6_UNLOCK();
1940 		break;
1941 	}
1942 	case SIOCGNBRINFO_IN6:
1943 	{
1944 		struct llentry *ln;
1945 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1946 
1947 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1948 			return error;
1949 
1950 		ln = nd6_lookup(&nb_addr, ifp, false);
1951 		if (ln == NULL) {
1952 			error = EINVAL;
1953 			break;
1954 		}
1955 		nbi->state = ln->ln_state;
1956 		nbi->asked = ln->ln_asked;
1957 		nbi->isrouter = ln->ln_router;
1958 		nbi->expire = ln->ln_expire ?
1959 		    time_mono_to_wall(ln->ln_expire) : 0;
1960 		LLE_RUNLOCK(ln);
1961 
1962 		break;
1963 	}
1964 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1965 		ndif->ifindex = nd6_defifindex;
1966 		break;
1967 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1968 		return nd6_setdefaultiface(ndif->ifindex);
1969 	}
1970 	return error;
1971 }
1972 
1973 void
1974 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1975 {
1976 	struct mbuf *m_hold, *m_hold_next;
1977 	struct sockaddr_in6 sin6;
1978 
1979 	LLE_WLOCK_ASSERT(ln);
1980 
1981 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1982 
1983 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1984 
1985 	LLE_WUNLOCK(ln);
1986 	for (; m_hold != NULL; m_hold = m_hold_next) {
1987 		m_hold_next = m_hold->m_nextpkt;
1988 		m_hold->m_nextpkt = NULL;
1989 
1990 		/*
1991 		 * we assume ifp is not a p2p here, so
1992 		 * just set the 2nd argument as the
1993 		 * 1st one.
1994 		 */
1995 		nd6_output(ifp, ifp, m_hold, &sin6, NULL);
1996 	}
1997 	LLE_WLOCK(ln);
1998 }
1999 
2000 /*
2001  * Create neighbor cache entry and cache link-layer address,
2002  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
2003  */
2004 void
2005 nd6_cache_lladdr(
2006     struct ifnet *ifp,
2007     struct in6_addr *from,
2008     char *lladdr,
2009     int lladdrlen,
2010     int type,	/* ICMP6 type */
2011     int code	/* type dependent information */
2012 )
2013 {
2014 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
2015 	struct llentry *ln = NULL;
2016 	int is_newentry;
2017 	int do_update;
2018 	int olladdr;
2019 	int llchange;
2020 	int newstate = 0;
2021 	uint16_t router = 0;
2022 
2023 	KASSERT(ifp != NULL);
2024 	KASSERT(from != NULL);
2025 
2026 	/* nothing must be updated for unspecified address */
2027 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2028 		return;
2029 
2030 	/*
2031 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2032 	 * the caller.
2033 	 *
2034 	 * XXX If the link does not have link-layer adderss, what should
2035 	 * we do? (ifp->if_addrlen == 0)
2036 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2037 	 * description on it in NS section (RFC 2461 7.2.3).
2038 	 */
2039 
2040 	ln = nd6_lookup(from, ifp, true);
2041 	if (ln == NULL) {
2042 #if 0
2043 		/* nothing must be done if there's no lladdr */
2044 		if (!lladdr || !lladdrlen)
2045 			return NULL;
2046 #endif
2047 
2048 		ln = nd6_create(from, ifp);
2049 		is_newentry = 1;
2050 	} else {
2051 		/* do nothing if static ndp is set */
2052 		if (ln->la_flags & LLE_STATIC) {
2053 			LLE_WUNLOCK(ln);
2054 			return;
2055 		}
2056 		is_newentry = 0;
2057 	}
2058 
2059 	if (ln == NULL)
2060 		return;
2061 
2062 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2063 	if (olladdr && lladdr) {
2064 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2065 	} else
2066 		llchange = 0;
2067 
2068 	/*
2069 	 * newentry olladdr  lladdr  llchange	(*=record)
2070 	 *	0	n	n	--	(1)
2071 	 *	0	y	n	--	(2)
2072 	 *	0	n	y	--	(3) * STALE
2073 	 *	0	y	y	n	(4) *
2074 	 *	0	y	y	y	(5) * STALE
2075 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2076 	 *	1	--	y	--	(7) * STALE
2077 	 */
2078 
2079 	if (lladdr) {		/* (3-5) and (7) */
2080 		/*
2081 		 * Record source link-layer address
2082 		 * XXX is it dependent to ifp->if_type?
2083 		 */
2084 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2085 		ln->la_flags |= LLE_VALID;
2086 	}
2087 
2088 	if (!is_newentry) {
2089 		if ((!olladdr && lladdr) ||		/* (3) */
2090 		    (olladdr && lladdr && llchange)) {	/* (5) */
2091 			do_update = 1;
2092 			newstate = ND6_LLINFO_STALE;
2093 		} else					/* (1-2,4) */
2094 			do_update = 0;
2095 	} else {
2096 		do_update = 1;
2097 		if (lladdr == NULL)			/* (6) */
2098 			newstate = ND6_LLINFO_NOSTATE;
2099 		else					/* (7) */
2100 			newstate = ND6_LLINFO_STALE;
2101 	}
2102 
2103 	if (do_update) {
2104 		/*
2105 		 * Update the state of the neighbor cache.
2106 		 */
2107 		ln->ln_state = newstate;
2108 
2109 		if (ln->ln_state == ND6_LLINFO_STALE) {
2110 			/*
2111 			 * XXX: since nd6_output() below will cause
2112 			 * state tansition to DELAY and reset the timer,
2113 			 * we must set the timer now, although it is actually
2114 			 * meaningless.
2115 			 */
2116 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2117 
2118 			nd6_llinfo_release_pkts(ln, ifp);
2119 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2120 			/* probe right away */
2121 			nd6_llinfo_settimer((void *)ln, 0);
2122 		}
2123 	}
2124 
2125 	/*
2126 	 * ICMP6 type dependent behavior.
2127 	 *
2128 	 * NS: clear IsRouter if new entry
2129 	 * RS: clear IsRouter
2130 	 * RA: set IsRouter if there's lladdr
2131 	 * redir: clear IsRouter if new entry
2132 	 *
2133 	 * RA case, (1):
2134 	 * The spec says that we must set IsRouter in the following cases:
2135 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2136 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2137 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2138 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2139 	 * neighbor cache, this is similar to (6).
2140 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2141 	 *
2142 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2143 	 *							D R
2144 	 *	0	n	n	--	(1)	c   ?     s
2145 	 *	0	y	n	--	(2)	c   s     s
2146 	 *	0	n	y	--	(3)	c   s     s
2147 	 *	0	y	y	n	(4)	c   s     s
2148 	 *	0	y	y	y	(5)	c   s     s
2149 	 *	1	--	n	--	(6) c	c 	c s
2150 	 *	1	--	y	--	(7) c	c   s	c s
2151 	 *
2152 	 *					(c=clear s=set)
2153 	 */
2154 	switch (type & 0xff) {
2155 	case ND_NEIGHBOR_SOLICIT:
2156 		/*
2157 		 * New entry must have is_router flag cleared.
2158 		 */
2159 		if (is_newentry)	/* (6-7) */
2160 			ln->ln_router = 0;
2161 		break;
2162 	case ND_REDIRECT:
2163 		/*
2164 		 * If the icmp is a redirect to a better router, always set the
2165 		 * is_router flag.  Otherwise, if the entry is newly created,
2166 		 * clear the flag.  [RFC 2461, sec 8.3]
2167 		 */
2168 		if (code == ND_REDIRECT_ROUTER)
2169 			ln->ln_router = 1;
2170 		else if (is_newentry) /* (6-7) */
2171 			ln->ln_router = 0;
2172 		break;
2173 	case ND_ROUTER_SOLICIT:
2174 		/*
2175 		 * is_router flag must always be cleared.
2176 		 */
2177 		ln->ln_router = 0;
2178 		break;
2179 	case ND_ROUTER_ADVERT:
2180 		/*
2181 		 * Mark an entry with lladdr as a router.
2182 		 */
2183 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2184 		    (is_newentry && lladdr)) {			/* (7) */
2185 			ln->ln_router = 1;
2186 		}
2187 		break;
2188 	}
2189 
2190 #if 0
2191 	/* XXX should we send rtmsg as it used to be? */
2192 	if (do_update)
2193 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2194 #endif
2195 
2196 	if (ln != NULL) {
2197 		router = ln->ln_router;
2198 		LLE_WUNLOCK(ln);
2199 	}
2200 
2201 	/*
2202 	 * If we have too many cache entries, initiate immediate
2203 	 * purging for some entries.
2204 	 */
2205 	if (is_newentry)
2206 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2207 
2208 	/*
2209 	 * When the link-layer address of a router changes, select the
2210 	 * best router again.  In particular, when the neighbor entry is newly
2211 	 * created, it might affect the selection policy.
2212 	 * Question: can we restrict the first condition to the "is_newentry"
2213 	 * case?
2214 	 * XXX: when we hear an RA from a new router with the link-layer
2215 	 * address option, nd6_defrouter_select() is called twice, since
2216 	 * defrtrlist_update called the function as well.  However, I believe
2217 	 * we can compromise the overhead, since it only happens the first
2218 	 * time.
2219 	 * XXX: although nd6_defrouter_select() should not have a bad effect
2220 	 * for those are not autoconfigured hosts, we explicitly avoid such
2221 	 * cases for safety.
2222 	 */
2223 	if (do_update && router && !ip6_forwarding &&
2224 	    nd6_accepts_rtadv(ndi)) {
2225 		ND6_WLOCK();
2226 		nd6_defrouter_select();
2227 		ND6_UNLOCK();
2228 	}
2229 }
2230 
2231 static void
2232 nd6_slowtimo(void *ignored_arg)
2233 {
2234 	struct nd_ifinfo *nd6if;
2235 	struct ifnet *ifp;
2236 	int s;
2237 
2238 #ifndef NET_MPSAFE
2239 	mutex_enter(softnet_lock);
2240 	KERNEL_LOCK(1, NULL);
2241 #endif
2242 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2243 	    nd6_slowtimo, NULL);
2244 
2245 	s = pserialize_read_enter();
2246 	IFNET_READER_FOREACH(ifp) {
2247 		nd6if = ND_IFINFO(ifp);
2248 		if (nd6if->basereachable && /* already initialized */
2249 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2250 			/*
2251 			 * Since reachable time rarely changes by router
2252 			 * advertisements, we SHOULD insure that a new random
2253 			 * value gets recomputed at least once every few hours.
2254 			 * (RFC 2461, 6.3.4)
2255 			 */
2256 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2257 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2258 		}
2259 	}
2260 	pserialize_read_exit(s);
2261 
2262 #ifndef NET_MPSAFE
2263 	KERNEL_UNLOCK_ONE(NULL);
2264 	mutex_exit(softnet_lock);
2265 #endif
2266 }
2267 
2268 int
2269 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2270     const struct sockaddr_in6 *dst, struct rtentry *rt)
2271 {
2272 #define senderr(e) { error = (e); goto bad;}
2273 	struct llentry *ln = NULL;
2274 	int error = 0;
2275 	bool created = false;
2276 
2277 	if (rt != NULL) {
2278 		error = rt_check_reject_route(rt, ifp);
2279 		if (error != 0) {
2280 			m_freem(m);
2281 			return error;
2282 		}
2283 	}
2284 
2285 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2286 		goto sendpkt;
2287 
2288 	if (nd6_need_cache(ifp) == 0)
2289 		goto sendpkt;
2290 
2291 	if (rt != NULL && (rt->rt_flags & RTF_GATEWAY) != 0) {
2292 		struct sockaddr_in6 *gw6 = satosin6(rt->rt_gateway);
2293 		int s;
2294 
2295 		/* XXX remain the check to keep the original behavior. */
2296 		/*
2297 		 * We skip link-layer address resolution and NUD
2298 		 * if the gateway is not a neighbor from ND point
2299 		 * of view, regardless of the value of nd_ifinfo.flags.
2300 		 * The second condition is a bit tricky; we skip
2301 		 * if the gateway is our own address, which is
2302 		 * sometimes used to install a route to a p2p link.
2303 		 */
2304 		s = pserialize_read_enter();
2305 		if (!nd6_is_addr_neighbor(gw6, ifp) ||
2306 		    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
2307 			/*
2308 			 * We allow this kind of tricky route only
2309 			 * when the outgoing interface is p2p.
2310 			 * XXX: we may need a more generic rule here.
2311 			 */
2312 			if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
2313 				pserialize_read_exit(s);
2314 				senderr(EHOSTUNREACH);
2315 			}
2316 
2317 			pserialize_read_exit(s);
2318 			goto sendpkt;
2319 		}
2320 		pserialize_read_exit(s);
2321 	}
2322 
2323 	/*
2324 	 * Address resolution or Neighbor Unreachability Detection
2325 	 * for the next hop.
2326 	 * At this point, the destination of the packet must be a unicast
2327 	 * or an anycast address(i.e. not a multicast).
2328 	 */
2329 
2330 	/* Look up the neighbor cache for the nexthop */
2331 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2332 	if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2333 		/*
2334 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2335 		 * the condition below is not very efficient.  But we believe
2336 		 * it is tolerable, because this should be a rare case.
2337 		 */
2338 		ln = nd6_create(&dst->sin6_addr, ifp);
2339 		if (ln != NULL)
2340 			created = true;
2341 	}
2342 
2343 	if (ln == NULL) {
2344 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2345 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2346 			char ip6buf[INET6_ADDRSTRLEN];
2347 			log(LOG_DEBUG,
2348 			    "nd6_output: can't allocate llinfo for %s "
2349 			    "(ln=%p, rt=%p)\n",
2350 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
2351 			senderr(EIO);	/* XXX: good error? */
2352 		}
2353 		goto sendpkt;	/* send anyway */
2354 	}
2355 
2356 	LLE_WLOCK_ASSERT(ln);
2357 
2358 	/* We don't have to do link-layer address resolution on a p2p link. */
2359 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2360 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2361 		ln->ln_state = ND6_LLINFO_STALE;
2362 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2363 	}
2364 
2365 	/*
2366 	 * The first time we send a packet to a neighbor whose entry is
2367 	 * STALE, we have to change the state to DELAY and a sets a timer to
2368 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2369 	 * neighbor unreachability detection on expiration.
2370 	 * (RFC 2461 7.3.3)
2371 	 */
2372 	if (ln->ln_state == ND6_LLINFO_STALE) {
2373 		ln->ln_asked = 0;
2374 		ln->ln_state = ND6_LLINFO_DELAY;
2375 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2376 	}
2377 
2378 	/*
2379 	 * If the neighbor cache entry has a state other than INCOMPLETE
2380 	 * (i.e. its link-layer address is already resolved), just
2381 	 * send the packet.
2382 	 */
2383 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2384 		goto sendpkt;
2385 
2386 	/*
2387 	 * There is a neighbor cache entry, but no ethernet address
2388 	 * response yet.  Append this latest packet to the end of the
2389 	 * packet queue in the mbuf, unless the number of the packet
2390 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2391 	 * the oldest packet in the queue will be removed.
2392 	 */
2393 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2394 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2395 	if (ln->ln_hold) {
2396 		struct mbuf *m_hold;
2397 		int i;
2398 
2399 		i = 0;
2400 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2401 			i++;
2402 			if (m_hold->m_nextpkt == NULL) {
2403 				m_hold->m_nextpkt = m;
2404 				break;
2405 			}
2406 		}
2407 		while (i >= nd6_maxqueuelen) {
2408 			m_hold = ln->ln_hold;
2409 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2410 			m_freem(m_hold);
2411 			i--;
2412 		}
2413 	} else {
2414 		ln->ln_hold = m;
2415 	}
2416 
2417 	/*
2418 	 * If there has been no NS for the neighbor after entering the
2419 	 * INCOMPLETE state, send the first solicitation.
2420 	 */
2421 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2422 		struct in6_addr src, *psrc;
2423 
2424 		ln->ln_asked++;
2425 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2426 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2427 		LLE_WUNLOCK(ln);
2428 		ln = NULL;
2429 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
2430 	} else {
2431 		/* We did the lookup so we need to do the unlock here. */
2432 		LLE_WUNLOCK(ln);
2433 	}
2434 
2435 	error = 0;
2436 	goto exit;
2437 
2438   sendpkt:
2439 	/* discard the packet if IPv6 operation is disabled on the interface */
2440 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2441 		error = ENETDOWN; /* better error? */
2442 		goto bad;
2443 	}
2444 
2445 	if (ln != NULL)
2446 		LLE_WUNLOCK(ln);
2447 
2448 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2449 		error = if_output_lock(ifp, origifp, m, sin6tocsa(dst), rt);
2450 	else
2451 		error = if_output_lock(ifp, ifp, m, sin6tocsa(dst), rt);
2452 	goto exit;
2453 
2454   bad:
2455 	if (m != NULL)
2456 		m_freem(m);
2457   exit:
2458 	if (created)
2459 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2460 
2461 	return error;
2462 #undef senderr
2463 }
2464 
2465 int
2466 nd6_need_cache(struct ifnet *ifp)
2467 {
2468 	/*
2469 	 * XXX: we currently do not make neighbor cache on any interface
2470 	 * other than ARCnet, Ethernet, FDDI and GIF.
2471 	 *
2472 	 * RFC2893 says:
2473 	 * - unidirectional tunnels needs no ND
2474 	 */
2475 	switch (ifp->if_type) {
2476 	case IFT_ARCNET:
2477 	case IFT_ETHER:
2478 	case IFT_FDDI:
2479 	case IFT_IEEE1394:
2480 	case IFT_CARP:
2481 	case IFT_GIF:		/* XXX need more cases? */
2482 	case IFT_PPP:
2483 	case IFT_TUNNEL:
2484 		return 1;
2485 	default:
2486 		return 0;
2487 	}
2488 }
2489 
2490 /*
2491  * Add pernament ND6 link-layer record for given
2492  * interface address.
2493  *
2494  * Very similar to IPv4 arp_ifinit(), but:
2495  * 1) IPv6 DAD is performed in different place
2496  * 2) It is called by IPv6 protocol stack in contrast to
2497  * arp_ifinit() which is typically called in SIOCSIFADDR
2498  * driver ioctl handler.
2499  *
2500  */
2501 int
2502 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2503 {
2504 	struct ifnet *ifp;
2505 	struct llentry *ln;
2506 
2507 	ifp = ia->ia_ifa.ifa_ifp;
2508 	if (nd6_need_cache(ifp) == 0)
2509 		return 0;
2510 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2511 	ia->ia_ifa.ifa_flags |= RTF_CONNECTED;
2512 
2513 	IF_AFDATA_WLOCK(ifp);
2514 	ln = lla_create(LLTABLE6(ifp), LLE_IFADDR | LLE_EXCLUSIVE,
2515 	    sin6tosa(&ia->ia_addr));
2516 	IF_AFDATA_WUNLOCK(ifp);
2517 	if (ln == NULL)
2518 		return ENOBUFS;
2519 
2520 	ln->la_expire = 0;  /* for IPv6 this means permanent */
2521 	ln->ln_state = ND6_LLINFO_REACHABLE;
2522 
2523 	LLE_WUNLOCK(ln);
2524 	return 0;
2525 }
2526 
2527 /*
2528  * Removes ALL lle records for interface address prefix.
2529  * XXXME: That's probably not we really want to do, we need
2530  * to remove address record only and keep other records
2531  * until we determine if given prefix is really going
2532  * to be removed.
2533  */
2534 void
2535 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2536 {
2537 	struct sockaddr_in6 mask, addr;
2538 
2539 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2540 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2541 	lltable_prefix_free(AF_INET6, sin6tosa(&addr), sin6tosa(&mask),
2542 	    LLE_STATIC);
2543 }
2544 
2545 int
2546 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2547     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2548     size_t dstsize)
2549 {
2550 	struct llentry *ln;
2551 
2552 	if (m->m_flags & M_MCAST) {
2553 		switch (ifp->if_type) {
2554 		case IFT_ETHER:
2555 		case IFT_FDDI:
2556 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2557 			    lldst);
2558 			return 1;
2559 		case IFT_IEEE1394:
2560 			memcpy(lldst, ifp->if_broadcastaddr,
2561 			    MIN(dstsize, ifp->if_addrlen));
2562 			return 1;
2563 		case IFT_ARCNET:
2564 			*lldst = 0;
2565 			return 1;
2566 		default:
2567 			m_freem(m);
2568 			return 0;
2569 		}
2570 	}
2571 
2572 	/*
2573 	 * the entry should have been created in nd6_store_lladdr
2574 	 */
2575 	ln = nd6_lookup(&satocsin6(dst)->sin6_addr, ifp, false);
2576 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2577 		if (ln != NULL)
2578 			LLE_RUNLOCK(ln);
2579 		/* this could happen, if we could not allocate memory */
2580 		m_freem(m);
2581 		return 0;
2582 	}
2583 
2584 	/* XXX llentry should have addrlen? */
2585 #if 0
2586 	sdl = satocsdl(rt->rt_gateway);
2587 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2588 		char sbuf[INET6_ADDRSTRLEN];
2589 		char dbuf[LINK_ADDRSTRLEN];
2590 		/* this should be impossible, but we bark here for debugging */
2591 		printf("%s: sdl_alen == %" PRIu8 ", if=%s, dst=%s, sdl=%s\n",
2592 		    __func__, sdl->sdl_alen, if_name(ifp),
2593 		    IN6_PRINT(sbuf, &satocsin6(dst)->sin6_addr),
2594 		    DL_PRINT(dbuf, &sdl->sdl_addr));
2595 		m_freem(m);
2596 		return 0;
2597 	}
2598 #endif
2599 
2600 	memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2601 
2602 	LLE_RUNLOCK(ln);
2603 
2604 	return 1;
2605 }
2606 
2607 static void
2608 clear_llinfo_pqueue(struct llentry *ln)
2609 {
2610 	struct mbuf *m_hold, *m_hold_next;
2611 
2612 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2613 		m_hold_next = m_hold->m_nextpkt;
2614 		m_hold->m_nextpkt = NULL;
2615 		m_freem(m_hold);
2616 	}
2617 
2618 	ln->ln_hold = NULL;
2619 	return;
2620 }
2621 
2622 int
2623 nd6_sysctl(
2624     int name,
2625     void *oldp,	/* syscall arg, need copyout */
2626     size_t *oldlenp,
2627     void *newp,	/* syscall arg, need copyin */
2628     size_t newlen
2629 )
2630 {
2631 	void *p;
2632 	size_t ol;
2633 	int error;
2634 
2635 	error = 0;
2636 
2637 	if (newp)
2638 		return EPERM;
2639 	if (oldp && !oldlenp)
2640 		return EINVAL;
2641 	ol = oldlenp ? *oldlenp : 0;
2642 
2643 	if (oldp) {
2644 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2645 		if (p == NULL)
2646 			return ENOMEM;
2647 	} else
2648 		p = NULL;
2649 	switch (name) {
2650 	case ICMPV6CTL_ND6_DRLIST:
2651 		error = fill_drlist(p, oldlenp, ol);
2652 		if (!error && p != NULL && oldp != NULL)
2653 			error = copyout(p, oldp, *oldlenp);
2654 		break;
2655 
2656 	case ICMPV6CTL_ND6_PRLIST:
2657 		error = fill_prlist(p, oldlenp, ol);
2658 		if (!error && p != NULL && oldp != NULL)
2659 			error = copyout(p, oldp, *oldlenp);
2660 		break;
2661 
2662 	case ICMPV6CTL_ND6_MAXQLEN:
2663 		break;
2664 
2665 	default:
2666 		error = ENOPROTOOPT;
2667 		break;
2668 	}
2669 	if (p)
2670 		free(p, M_TEMP);
2671 
2672 	return error;
2673 }
2674 
2675 static int
2676 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2677 {
2678 	int error = 0;
2679 	struct in6_defrouter *d = NULL, *de = NULL;
2680 	struct nd_defrouter *dr;
2681 	size_t l;
2682 
2683 	if (oldp) {
2684 		d = (struct in6_defrouter *)oldp;
2685 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2686 	}
2687 	l = 0;
2688 
2689 	ND6_RLOCK();
2690 	ND_DEFROUTER_LIST_FOREACH(dr) {
2691 
2692 		if (oldp && d + 1 <= de) {
2693 			memset(d, 0, sizeof(*d));
2694 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2695 			if (sa6_recoverscope(&d->rtaddr)) {
2696 				char ip6buf[INET6_ADDRSTRLEN];
2697 				log(LOG_ERR,
2698 				    "scope error in router list (%s)\n",
2699 				    IN6_PRINT(ip6buf, &d->rtaddr.sin6_addr));
2700 				/* XXX: press on... */
2701 			}
2702 			d->flags = dr->flags;
2703 			d->rtlifetime = dr->rtlifetime;
2704 			d->expire = dr->expire ?
2705 			    time_mono_to_wall(dr->expire) : 0;
2706 			d->if_index = dr->ifp->if_index;
2707 		}
2708 
2709 		l += sizeof(*d);
2710 		if (d)
2711 			d++;
2712 	}
2713 	ND6_UNLOCK();
2714 
2715 	if (oldp) {
2716 		if (l > ol)
2717 			error = ENOMEM;
2718 	}
2719 	if (oldlenp)
2720 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2721 
2722 	return error;
2723 }
2724 
2725 static int
2726 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2727 {
2728 	int error = 0;
2729 	struct nd_prefix *pr;
2730 	uint8_t *p = NULL, *ps = NULL;
2731 	uint8_t *pe = NULL;
2732 	size_t l;
2733 	char ip6buf[INET6_ADDRSTRLEN];
2734 
2735 	if (oldp) {
2736 		ps = p = (uint8_t*)oldp;
2737 		pe = (uint8_t*)oldp + *oldlenp;
2738 	}
2739 	l = 0;
2740 
2741 	ND6_RLOCK();
2742 	ND_PREFIX_LIST_FOREACH(pr) {
2743 		u_short advrtrs;
2744 		struct sockaddr_in6 sin6;
2745 		struct nd_pfxrouter *pfr;
2746 		struct in6_prefix pfx;
2747 
2748 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2749 		{
2750 			memset(&pfx, 0, sizeof(pfx));
2751 			ps = p;
2752 			pfx.prefix = pr->ndpr_prefix;
2753 
2754 			if (sa6_recoverscope(&pfx.prefix)) {
2755 				log(LOG_ERR,
2756 				    "scope error in prefix list (%s)\n",
2757 				    IN6_PRINT(ip6buf, &pfx.prefix.sin6_addr));
2758 				/* XXX: press on... */
2759 			}
2760 			pfx.raflags = pr->ndpr_raf;
2761 			pfx.prefixlen = pr->ndpr_plen;
2762 			pfx.vltime = pr->ndpr_vltime;
2763 			pfx.pltime = pr->ndpr_pltime;
2764 			pfx.if_index = pr->ndpr_ifp->if_index;
2765 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2766 				pfx.expire = 0;
2767 			else {
2768 				time_t maxexpire;
2769 
2770 				/* XXX: we assume time_t is signed. */
2771 				maxexpire = (-1) &
2772 				    ~((time_t)1 <<
2773 				    ((sizeof(maxexpire) * 8) - 1));
2774 				if (pr->ndpr_vltime <
2775 				    maxexpire - pr->ndpr_lastupdate) {
2776 					pfx.expire = pr->ndpr_lastupdate +
2777 						pr->ndpr_vltime;
2778 				} else
2779 					pfx.expire = maxexpire;
2780 			}
2781 			pfx.refcnt = pr->ndpr_refcnt;
2782 			pfx.flags = pr->ndpr_stateflags;
2783 			pfx.origin = PR_ORIG_RA;
2784 
2785 			p += sizeof(pfx); l += sizeof(pfx);
2786 
2787 			advrtrs = 0;
2788 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2789 				if (p + sizeof(sin6) > pe) {
2790 					advrtrs++;
2791 					continue;
2792 				}
2793 
2794 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2795 					    0, 0, 0);
2796 				if (sa6_recoverscope(&sin6)) {
2797 					log(LOG_ERR,
2798 					    "scope error in "
2799 					    "prefix list (%s)\n",
2800 					    IN6_PRINT(ip6buf,
2801 					    &pfr->router->rtaddr));
2802 				}
2803 				advrtrs++;
2804 				memcpy(p, &sin6, sizeof(sin6));
2805 				p += sizeof(sin6);
2806 				l += sizeof(sin6);
2807 			}
2808 			pfx.advrtrs = advrtrs;
2809 			memcpy(ps, &pfx, sizeof(pfx));
2810 		}
2811 		else {
2812 			l += sizeof(pfx);
2813 			advrtrs = 0;
2814 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2815 				advrtrs++;
2816 				l += sizeof(sin6);
2817 			}
2818 		}
2819 	}
2820 	ND6_UNLOCK();
2821 
2822 	if (oldp) {
2823 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2824 		if (l > ol)
2825 			error = ENOMEM;
2826 	} else
2827 		*oldlenp = l;
2828 
2829 	return error;
2830 }
2831 
2832 static int
2833 nd6_setdefaultiface(int ifindex)
2834 {
2835 	ifnet_t *ifp;
2836 	int error = 0;
2837 	int s;
2838 
2839 	s = pserialize_read_enter();
2840 	ifp = if_byindex(ifindex);
2841 	if (ifp == NULL) {
2842 		pserialize_read_exit(s);
2843 		return EINVAL;
2844 	}
2845 	if (nd6_defifindex != ifindex) {
2846 		nd6_defifindex = ifindex;
2847 		nd6_defifp = nd6_defifindex > 0 ? ifp : NULL;
2848 
2849 		/*
2850 		 * Our current implementation assumes one-to-one maping between
2851 		 * interfaces and links, so it would be natural to use the
2852 		 * default interface as the default link.
2853 		 */
2854 		scope6_setdefault(nd6_defifp);
2855 	}
2856 	pserialize_read_exit(s);
2857 
2858 	return (error);
2859 }
2860