xref: /netbsd-src/sys/netinet6/nd6.c (revision 796c32c94f6e154afc9de0f63da35c91bb739b45)
1 /*	$NetBSD: nd6.c,v 1.239 2017/11/17 07:37:12 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 1.239 2017/11/17 07:37:12 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 #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 = 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 			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 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
454 	LLE_WLOCK(ln);
455 	if ((ln->la_flags & LLE_LINKED) == 0)
456 		goto out;
457 	if (ln->ln_ntick > 0) {
458 		nd6_llinfo_settimer(ln, ln->ln_ntick);
459 		goto out;
460 	}
461 
462 	if (callout_pending(&ln->la_timer)) {
463 		/*
464 		 * Here we are a bit odd here in the treatment of
465 		 * active/pending. If the pending bit is set, it got
466 		 * rescheduled before I ran. The active
467 		 * bit we ignore, since if it was stopped
468 		 * in ll_tablefree() and was currently running
469 		 * it would have return 0 so the code would
470 		 * not have deleted it since the callout could
471 		 * not be stopped so we want to go through
472 		 * with the delete here now. If the callout
473 		 * was restarted, the pending bit will be back on and
474 		 * we just want to bail since the callout_reset would
475 		 * return 1 and our reference would have been removed
476 		 * by nd6_llinfo_settimer above since canceled
477 		 * would have been 1.
478 		 */
479 		goto out;
480 	}
481 
482 	ifp = ln->lle_tbl->llt_ifp;
483 
484 	KASSERT(ifp != NULL);
485 
486 	ndi = ND_IFINFO(ifp);
487 
488 	switch (ln->ln_state) {
489 	case ND6_LLINFO_INCOMPLETE:
490 		if (ln->ln_asked < nd6_mmaxtries) {
491 			ln->ln_asked++;
492 			send_ns = true;
493 		} else {
494 			struct mbuf *m = ln->ln_hold;
495 			if (m) {
496 				struct mbuf *m0;
497 
498 				/*
499 				 * assuming every packet in ln_hold has
500 				 * the same IP header
501 				 */
502 				m0 = m->m_nextpkt;
503 				m->m_nextpkt = NULL;
504 				ln->ln_hold = m0;
505 				clear_llinfo_pqueue(ln);
506  			}
507 			nd6_free(ln, 0);
508 			ln = NULL;
509 			if (m != NULL) {
510 				icmp6_error2(m, ICMP6_DST_UNREACH,
511 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
512 			}
513 		}
514 		break;
515 	case ND6_LLINFO_REACHABLE:
516 		if (!ND6_LLINFO_PERMANENT(ln)) {
517 			ln->ln_state = ND6_LLINFO_STALE;
518 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
519 		}
520 		break;
521 
522 	case ND6_LLINFO_PURGE:
523 	case ND6_LLINFO_STALE:
524 		/* Garbage Collection(RFC 2461 5.3) */
525 		if (!ND6_LLINFO_PERMANENT(ln)) {
526 			nd6_free(ln, 1);
527 			ln = NULL;
528 		}
529 		break;
530 
531 	case ND6_LLINFO_DELAY:
532 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
533 			/* We need NUD */
534 			ln->ln_asked = 1;
535 			ln->ln_state = ND6_LLINFO_PROBE;
536 			daddr6 = &ln->r_l3addr.addr6;
537 			send_ns = true;
538 		} else {
539 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
540 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
541 		}
542 		break;
543 	case ND6_LLINFO_PROBE:
544 		if (ln->ln_asked < nd6_umaxtries) {
545 			ln->ln_asked++;
546 			daddr6 = &ln->r_l3addr.addr6;
547 			send_ns = true;
548 		} else {
549 			nd6_free(ln, 0);
550 			ln = NULL;
551 		}
552 		break;
553 	}
554 
555 	if (send_ns) {
556 		struct in6_addr src, *psrc;
557 		const struct in6_addr *taddr6 = &ln->r_l3addr.addr6;
558 
559 		nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
560 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
561 		LLE_FREE_LOCKED(ln);
562 		ln = NULL;
563 		nd6_ns_output(ifp, daddr6, taddr6, psrc, 0);
564 	}
565 
566 out:
567 	if (ln != NULL)
568 		LLE_FREE_LOCKED(ln);
569 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
570 }
571 
572 /*
573  * ND6 timer routine to expire default route list and prefix list
574  */
575 static void
576 nd6_timer_work(struct work *wk, void *arg)
577 {
578 	struct nd_defrouter *next_dr, *dr;
579 	struct nd_prefix *next_pr, *pr;
580 	struct in6_ifaddr *ia6, *nia6;
581 	int s, bound;
582 	struct psref psref;
583 
584 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
585 	    nd6_timer, NULL);
586 
587 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
588 
589 	/* expire default router list */
590 
591 	ND6_WLOCK();
592 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, next_dr) {
593 		if (dr->expire && dr->expire < time_uptime) {
594 			nd6_defrtrlist_del(dr, NULL);
595 		}
596 	}
597 	ND6_UNLOCK();
598 
599 	/*
600 	 * expire interface addresses.
601 	 * in the past the loop was inside prefix expiry processing.
602 	 * However, from a stricter speci-confrmance standpoint, we should
603 	 * rather separate address lifetimes and prefix lifetimes.
604 	 */
605 	bound = curlwp_bind();
606   addrloop:
607 	s = pserialize_read_enter();
608 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
609 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
610 
611 		ia6_acquire(ia6, &psref);
612 		pserialize_read_exit(s);
613 
614 		/* check address lifetime */
615 		if (IFA6_IS_INVALID(ia6)) {
616 			int regen = 0;
617 
618 			/*
619 			 * If the expiring address is temporary, try
620 			 * regenerating a new one.  This would be useful when
621 			 * we suspended a laptop PC, then turned it on after a
622 			 * period that could invalidate all temporary
623 			 * addresses.  Although we may have to restart the
624 			 * loop (see below), it must be after purging the
625 			 * address.  Otherwise, we'd see an infinite loop of
626 			 * regeneration.
627 			 */
628 			if (ip6_use_tempaddr &&
629 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
630 				if (regen_tmpaddr(ia6) == 0)
631 					regen = 1;
632 			}
633 
634 			ia6_release(ia6, &psref);
635  			in6_purgeaddr(&ia6->ia_ifa);
636 			ia6 = NULL;
637 
638 			if (regen)
639 				goto addrloop; /* XXX: see below */
640 		} else if (IFA6_IS_DEPRECATED(ia6)) {
641 			int oldflags = ia6->ia6_flags;
642 
643 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
644 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
645 				rt_newaddrmsg(RTM_NEWADDR,
646 				    (struct ifaddr *)ia6, 0, NULL);
647 			}
648 
649 			/*
650 			 * If a temporary address has just become deprecated,
651 			 * regenerate a new one if possible.
652 			 */
653 			if (ip6_use_tempaddr &&
654 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
655 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
656 
657 				if (regen_tmpaddr(ia6) == 0) {
658 					/*
659 					 * A new temporary address is
660 					 * generated.
661 					 * XXX: this means the address chain
662 					 * has changed while we are still in
663 					 * the loop.  Although the change
664 					 * would not cause disaster (because
665 					 * it's not a deletion, but an
666 					 * addition,) we'd rather restart the
667 					 * loop just for safety.  Or does this
668 					 * significantly reduce performance??
669 					 */
670 					ia6_release(ia6, &psref);
671 					goto addrloop;
672 				}
673 			}
674 		} else {
675 			/*
676 			 * A new RA might have made a deprecated address
677 			 * preferred.
678 			 */
679 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
680 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
681 				rt_newaddrmsg(RTM_NEWADDR,
682 				    (struct ifaddr *)ia6, 0, NULL);
683 			}
684 		}
685 		s = pserialize_read_enter();
686 		ia6_release(ia6, &psref);
687 	}
688 	pserialize_read_exit(s);
689 	curlwp_bindx(bound);
690 
691 	/* expire prefix list */
692 	ND6_WLOCK();
693 	ND_PREFIX_LIST_FOREACH_SAFE(pr, next_pr) {
694 		/*
695 		 * check prefix lifetime.
696 		 * since pltime is just for autoconf, pltime processing for
697 		 * prefix is not necessary.
698 		 */
699 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
700 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
701 			/*
702 			 * Just invalidate the prefix here. Removing it
703 			 * will be done when purging an associated address.
704 			 */
705 			KASSERT(pr->ndpr_refcnt > 0);
706 			nd6_invalidate_prefix(pr);
707 		}
708 	}
709 	ND6_UNLOCK();
710 
711 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
712 }
713 
714 static void
715 nd6_timer(void *ignored_arg)
716 {
717 
718 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
719 }
720 
721 /* ia6: deprecated/invalidated temporary address */
722 static int
723 regen_tmpaddr(const struct in6_ifaddr *ia6)
724 {
725 	struct ifaddr *ifa;
726 	struct ifnet *ifp;
727 	struct in6_ifaddr *public_ifa6 = NULL;
728 	int s;
729 
730 	ifp = ia6->ia_ifa.ifa_ifp;
731 	s = pserialize_read_enter();
732 	IFADDR_READER_FOREACH(ifa, ifp) {
733 		struct in6_ifaddr *it6;
734 
735 		if (ifa->ifa_addr->sa_family != AF_INET6)
736 			continue;
737 
738 		it6 = (struct in6_ifaddr *)ifa;
739 
740 		/* ignore no autoconf addresses. */
741 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
742 			continue;
743 
744 		/* ignore autoconf addresses with different prefixes. */
745 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
746 			continue;
747 
748 		/*
749 		 * Now we are looking at an autoconf address with the same
750 		 * prefix as ours.  If the address is temporary and is still
751 		 * preferred, do not create another one.  It would be rare, but
752 		 * could happen, for example, when we resume a laptop PC after
753 		 * a long period.
754 		 */
755 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
756 		    !IFA6_IS_DEPRECATED(it6)) {
757 			public_ifa6 = NULL;
758 			break;
759 		}
760 
761 		/*
762 		 * This is a public autoconf address that has the same prefix
763 		 * as ours.  If it is preferred, keep it.  We can't break the
764 		 * loop here, because there may be a still-preferred temporary
765 		 * address with the prefix.
766 		 */
767 		if (!IFA6_IS_DEPRECATED(it6))
768 			public_ifa6 = it6;
769 	}
770 
771 	if (public_ifa6 != NULL) {
772 		int e;
773 		struct psref psref;
774 
775 		ia6_acquire(public_ifa6, &psref);
776 		pserialize_read_exit(s);
777 		/*
778 		 * Random factor is introduced in the preferred lifetime, so
779 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
780 		 */
781 		ND6_WLOCK();
782 		e = in6_tmpifadd(public_ifa6, 0, 0);
783 		ND6_UNLOCK();
784 		if (e != 0) {
785 			ia6_release(public_ifa6, &psref);
786 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
787 			    " tmp addr, errno=%d\n", e);
788 			return -1;
789 		}
790 		ia6_release(public_ifa6, &psref);
791 		return 0;
792 	}
793 	pserialize_read_exit(s);
794 
795 	return -1;
796 }
797 
798 bool
799 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
800 {
801 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
802 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
803 		return true;
804 	case ND6_IFF_ACCEPT_RTADV:
805 		return ip6_accept_rtadv != 0;
806 	case ND6_IFF_OVERRIDE_RTADV:
807 	case 0:
808 	default:
809 		return false;
810 	}
811 }
812 
813 /*
814  * Nuke neighbor cache/prefix/default router management table, right before
815  * ifp goes away.
816  */
817 void
818 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
819 {
820 	struct nd_defrouter *dr, *ndr;
821 	struct nd_prefix *pr, *npr;
822 
823 	/*
824 	 * During detach, the ND info might be already removed, but
825 	 * then is explitly passed as argument.
826 	 * Otherwise get it from ifp->if_afdata.
827 	 */
828 	if (ext == NULL)
829 		ext = ifp->if_afdata[AF_INET6];
830 	if (ext == NULL)
831 		return;
832 
833 	ND6_WLOCK();
834 	/*
835 	 * Nuke default router list entries toward ifp.
836 	 * We defer removal of default router list entries that is installed
837 	 * in the routing table, in order to keep additional side effects as
838 	 * small as possible.
839 	 */
840 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
841 		if (dr->installed)
842 			continue;
843 
844 		if (dr->ifp == ifp) {
845 			KASSERT(ext != NULL);
846 			nd6_defrtrlist_del(dr, ext);
847 		}
848 	}
849 
850 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
851 		if (!dr->installed)
852 			continue;
853 
854 		if (dr->ifp == ifp) {
855 			KASSERT(ext != NULL);
856 			nd6_defrtrlist_del(dr, ext);
857 		}
858 	}
859 
860 	/* Nuke prefix list entries toward ifp */
861 	ND_PREFIX_LIST_FOREACH_SAFE(pr, npr) {
862 		if (pr->ndpr_ifp == ifp) {
863 			/*
864 			 * All addresses referencing pr should be already freed.
865 			 */
866 			KASSERT(pr->ndpr_refcnt == 0);
867 			nd6_prelist_remove(pr);
868 		}
869 	}
870 
871 	/* cancel default outgoing interface setting */
872 	if (nd6_defifindex == ifp->if_index)
873 		nd6_setdefaultiface(0);
874 
875 	/* XXX: too restrictive? */
876 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
877 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
878 		if (ndi && nd6_accepts_rtadv(ndi)) {
879 			/* refresh default router list */
880 			nd6_defrouter_select();
881 		}
882 	}
883 	ND6_UNLOCK();
884 
885 	/*
886 	 * We may not need to nuke the neighbor cache entries here
887 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
888 	 * nd6_purge() is invoked by in6_ifdetach() which is called
889 	 * from if_detach() where everything gets purged. However
890 	 * in6_ifdetach is directly called from vlan(4), so we still
891 	 * need to purge entries here.
892 	 */
893 	if (ext->lltable != NULL)
894 		lltable_purge_entries(ext->lltable);
895 }
896 
897 void
898 nd6_assert_purged(struct ifnet *ifp)
899 {
900 	struct nd_defrouter *dr;
901 	struct nd_prefix *pr;
902 	char ip6buf[INET6_ADDRSTRLEN] __diagused;
903 
904 	ND6_RLOCK();
905 	ND_DEFROUTER_LIST_FOREACH(dr) {
906 		KASSERTMSG(dr->ifp != ifp,
907 		    "defrouter %s remains on %s",
908 		    IN6_PRINT(ip6buf, &dr->rtaddr), ifp->if_xname);
909 	}
910 
911 	ND_PREFIX_LIST_FOREACH(pr) {
912 		KASSERTMSG(pr->ndpr_ifp != ifp,
913 		    "prefix %s/%d remains on %s",
914 		    IN6_PRINT(ip6buf, &pr->ndpr_prefix.sin6_addr),
915 		    pr->ndpr_plen, ifp->if_xname);
916 	}
917 	ND6_UNLOCK();
918 }
919 
920 struct llentry *
921 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
922 {
923 	struct sockaddr_in6 sin6;
924 	struct llentry *ln;
925 
926 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
927 
928 	IF_AFDATA_RLOCK(ifp);
929 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
930 	    sin6tosa(&sin6));
931 	IF_AFDATA_RUNLOCK(ifp);
932 
933 	return ln;
934 }
935 
936 struct llentry *
937 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
938 {
939 	struct sockaddr_in6 sin6;
940 	struct llentry *ln;
941 	struct rtentry *rt;
942 
943 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
944 	rt = rtalloc1(sin6tosa(&sin6), 0);
945 
946 	IF_AFDATA_WLOCK(ifp);
947 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE, sin6tosa(&sin6), rt);
948 	IF_AFDATA_WUNLOCK(ifp);
949 
950 	if (rt != NULL)
951 		rt_unref(rt);
952 	if (ln != NULL)
953 		ln->ln_state = ND6_LLINFO_NOSTATE;
954 
955 	return ln;
956 }
957 
958 /*
959  * Test whether a given IPv6 address is a neighbor or not, ignoring
960  * the actual neighbor cache.  The neighbor cache is ignored in order
961  * to not reenter the routing code from within itself.
962  */
963 static int
964 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
965 {
966 	struct nd_prefix *pr;
967 	struct ifaddr *dstaddr;
968 	int s;
969 
970 	/*
971 	 * A link-local address is always a neighbor.
972 	 * XXX: a link does not necessarily specify a single interface.
973 	 */
974 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
975 		struct sockaddr_in6 sin6_copy;
976 		u_int32_t zone;
977 
978 		/*
979 		 * We need sin6_copy since sa6_recoverscope() may modify the
980 		 * content (XXX).
981 		 */
982 		sin6_copy = *addr;
983 		if (sa6_recoverscope(&sin6_copy))
984 			return 0; /* XXX: should be impossible */
985 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
986 			return 0;
987 		if (sin6_copy.sin6_scope_id == zone)
988 			return 1;
989 		else
990 			return 0;
991 	}
992 
993 	/*
994 	 * If the address matches one of our addresses,
995 	 * it should be a neighbor.
996 	 * If the address matches one of our on-link prefixes, it should be a
997 	 * neighbor.
998 	 */
999 	ND6_RLOCK();
1000 	ND_PREFIX_LIST_FOREACH(pr) {
1001 		if (pr->ndpr_ifp != ifp)
1002 			continue;
1003 
1004 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1005 			struct rtentry *rt;
1006 
1007 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
1008 			if (rt == NULL)
1009 				continue;
1010 			/*
1011 			 * This is the case where multiple interfaces
1012 			 * have the same prefix, but only one is installed
1013 			 * into the routing table and that prefix entry
1014 			 * is not the one being examined here. In the case
1015 			 * where RADIX_MPATH is enabled, multiple route
1016 			 * entries (of the same rt_key value) will be
1017 			 * installed because the interface addresses all
1018 			 * differ.
1019 			 */
1020 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1021 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
1022 				rt_unref(rt);
1023 				continue;
1024 			}
1025 			rt_unref(rt);
1026 		}
1027 
1028 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1029 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1030 			ND6_UNLOCK();
1031 			return 1;
1032 		}
1033 	}
1034 	ND6_UNLOCK();
1035 
1036 	/*
1037 	 * If the address is assigned on the node of the other side of
1038 	 * a p2p interface, the address should be a neighbor.
1039 	 */
1040 	s = pserialize_read_enter();
1041 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1042 	if (dstaddr != NULL) {
1043 		if (dstaddr->ifa_ifp == ifp) {
1044 			pserialize_read_exit(s);
1045 			return 1;
1046 		}
1047 	}
1048 	pserialize_read_exit(s);
1049 
1050 	/*
1051 	 * If the default router list is empty, all addresses are regarded
1052 	 * as on-link, and thus, as a neighbor.
1053 	 */
1054 	ND6_RLOCK();
1055 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1056 	    ND_DEFROUTER_LIST_EMPTY() && nd6_defifindex == ifp->if_index) {
1057 		ND6_UNLOCK();
1058 		return 1;
1059 	}
1060 	ND6_UNLOCK();
1061 
1062 	return 0;
1063 }
1064 
1065 /*
1066  * Detect if a given IPv6 address identifies a neighbor on a given link.
1067  * XXX: should take care of the destination of a p2p link?
1068  */
1069 int
1070 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1071 {
1072 	struct nd_prefix *pr;
1073 	struct llentry *ln;
1074 	struct rtentry *rt;
1075 
1076 	/*
1077 	 * A link-local address is always a neighbor.
1078 	 * XXX: a link does not necessarily specify a single interface.
1079 	 */
1080 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1081 		struct sockaddr_in6 sin6_copy;
1082 		u_int32_t zone;
1083 
1084 		/*
1085 		 * We need sin6_copy since sa6_recoverscope() may modify the
1086 		 * content (XXX).
1087 		 */
1088 		sin6_copy = *addr;
1089 		if (sa6_recoverscope(&sin6_copy))
1090 			return 0; /* XXX: should be impossible */
1091 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1092 			return 0;
1093 		if (sin6_copy.sin6_scope_id == zone)
1094 			return 1;
1095 		else
1096 			return 0;
1097 	}
1098 
1099 	/*
1100 	 * If the address matches one of our on-link prefixes, it should be a
1101 	 * neighbor.
1102 	 */
1103 	ND6_RLOCK();
1104 	ND_PREFIX_LIST_FOREACH(pr) {
1105 		if (pr->ndpr_ifp != ifp)
1106 			continue;
1107 
1108 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1109 			continue;
1110 
1111 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1112 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1113 			ND6_UNLOCK();
1114 			return 1;
1115 		}
1116 	}
1117 
1118 	/*
1119 	 * If the default router list is empty, all addresses are regarded
1120 	 * as on-link, and thus, as a neighbor.
1121 	 * XXX: we restrict the condition to hosts, because routers usually do
1122 	 * not have the "default router list".
1123 	 */
1124 	if (!ip6_forwarding && ND_DEFROUTER_LIST_EMPTY() &&
1125 	    nd6_defifindex == ifp->if_index) {
1126 		ND6_UNLOCK();
1127 		return 1;
1128 	}
1129 	ND6_UNLOCK();
1130 
1131 	if (nd6_is_new_addr_neighbor(addr, ifp))
1132 		return 1;
1133 
1134 	/*
1135 	 * Even if the address matches none of our addresses, it might be
1136 	 * in the neighbor cache or a connected route.
1137 	 */
1138 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1139 	if (ln != NULL) {
1140 		LLE_RUNLOCK(ln);
1141 		return 1;
1142 	}
1143 
1144 	rt = rtalloc1(sin6tocsa(addr), 0);
1145 	if (rt == NULL)
1146 		return 0;
1147 
1148 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1149 #if NBRIDGE > 0
1150 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1151 #endif
1152 #if NCARP > 0
1153 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1154 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1155 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1156 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1157 #endif
1158 	    )) {
1159 		rt_unref(rt);
1160 		return 1;
1161 	}
1162 	rt_unref(rt);
1163 
1164 	return 0;
1165 }
1166 
1167 /*
1168  * Free an nd6 llinfo entry.
1169  * Since the function would cause significant changes in the kernel, DO NOT
1170  * make it global, unless you have a strong reason for the change, and are sure
1171  * that the change is safe.
1172  */
1173 static void
1174 nd6_free(struct llentry *ln, int gc)
1175 {
1176 	struct nd_defrouter *dr;
1177 	struct ifnet *ifp;
1178 	struct in6_addr *in6;
1179 
1180 	KASSERT(ln != NULL);
1181 	LLE_WLOCK_ASSERT(ln);
1182 
1183 	ifp = ln->lle_tbl->llt_ifp;
1184 	in6 = &ln->r_l3addr.addr6;
1185 	/*
1186 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1187 	 * even though it is not harmful, it was not really necessary.
1188 	 */
1189 
1190 	if (!ip6_forwarding) {
1191 		ND6_WLOCK();
1192 		dr = nd6_defrouter_lookup(in6, ifp);
1193 
1194 		if (dr != NULL && dr->expire &&
1195 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1196 			/*
1197 			 * If the reason for the deletion is just garbage
1198 			 * collection, and the neighbor is an active default
1199 			 * router, do not delete it.  Instead, reset the GC
1200 			 * timer using the router's lifetime.
1201 			 * Simply deleting the entry would affect default
1202 			 * router selection, which is not necessarily a good
1203 			 * thing, especially when we're using router preference
1204 			 * values.
1205 			 * XXX: the check for ln_state would be redundant,
1206 			 *      but we intentionally keep it just in case.
1207 			 */
1208 			if (dr->expire > time_uptime)
1209 				nd6_llinfo_settimer(ln,
1210 				    (dr->expire - time_uptime) * hz);
1211 			else
1212 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1213 			ND6_UNLOCK();
1214 			LLE_WUNLOCK(ln);
1215 			return;
1216 		}
1217 
1218 		if (ln->ln_router || dr) {
1219 			/*
1220 			 * We need to unlock to avoid a LOR with nd6_rt_flush()
1221 			 * with the rnh and for the calls to
1222 			 * nd6_pfxlist_onlink_check() and nd6_defrouter_select() in the
1223 			 * block further down for calls into nd6_lookup().
1224 			 * We still hold a ref.
1225 			 */
1226 			LLE_WUNLOCK(ln);
1227 
1228 			/*
1229 			 * nd6_rt_flush must be called whether or not the neighbor
1230 			 * is in the Default Router List.
1231 			 * See a corresponding comment in nd6_na_input().
1232 			 */
1233 			nd6_rt_flush(in6, ifp);
1234 		}
1235 
1236 		if (dr) {
1237 			/*
1238 			 * Unreachablity of a router might affect the default
1239 			 * router selection and on-link detection of advertised
1240 			 * prefixes.
1241 			 */
1242 
1243 			/*
1244 			 * Temporarily fake the state to choose a new default
1245 			 * router and to perform on-link determination of
1246 			 * prefixes correctly.
1247 			 * Below the state will be set correctly,
1248 			 * or the entry itself will be deleted.
1249 			 */
1250 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1251 
1252 			/*
1253 			 * Since nd6_defrouter_select() does not affect the
1254 			 * on-link determination and MIP6 needs the check
1255 			 * before the default router selection, we perform
1256 			 * the check now.
1257 			 */
1258 			nd6_pfxlist_onlink_check();
1259 
1260 			/*
1261 			 * refresh default router list
1262 			 */
1263 			nd6_defrouter_select();
1264 		}
1265 
1266 #ifdef __FreeBSD__
1267 		/*
1268 		 * If this entry was added by an on-link redirect, remove the
1269 		 * corresponding host route.
1270 		 */
1271 		if (ln->la_flags & LLE_REDIRECT)
1272 			nd6_free_redirect(ln);
1273 #endif
1274 		ND6_UNLOCK();
1275 
1276 		if (ln->ln_router || dr)
1277 			LLE_WLOCK(ln);
1278 	}
1279 
1280 	/*
1281 	 * Save to unlock. We still hold an extra reference and will not
1282 	 * free(9) in llentry_free() if someone else holds one as well.
1283 	 */
1284 	LLE_WUNLOCK(ln);
1285 	IF_AFDATA_LOCK(ifp);
1286 	LLE_WLOCK(ln);
1287 
1288 	lltable_free_entry(LLTABLE6(ifp), ln);
1289 
1290 	IF_AFDATA_UNLOCK(ifp);
1291 }
1292 
1293 /*
1294  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1295  *
1296  * XXX cost-effective methods?
1297  */
1298 void
1299 nd6_nud_hint(struct rtentry *rt)
1300 {
1301 	struct llentry *ln;
1302 	struct ifnet *ifp;
1303 
1304 	if (rt == NULL)
1305 		return;
1306 
1307 	ifp = rt->rt_ifp;
1308 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1309 	if (ln == NULL)
1310 		return;
1311 
1312 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1313 		goto done;
1314 
1315 	/*
1316 	 * if we get upper-layer reachability confirmation many times,
1317 	 * it is possible we have false information.
1318 	 */
1319 	ln->ln_byhint++;
1320 	if (ln->ln_byhint > nd6_maxnudhint)
1321 		goto done;
1322 
1323 	ln->ln_state = ND6_LLINFO_REACHABLE;
1324 	if (!ND6_LLINFO_PERMANENT(ln))
1325 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1326 
1327 done:
1328 	LLE_WUNLOCK(ln);
1329 
1330 	return;
1331 }
1332 
1333 struct gc_args {
1334 	int gc_entries;
1335 	const struct in6_addr *skip_in6;
1336 };
1337 
1338 static int
1339 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1340 {
1341 	struct gc_args *args = farg;
1342 	int *n = &args->gc_entries;
1343 	const struct in6_addr *skip_in6 = args->skip_in6;
1344 
1345 	if (*n <= 0)
1346 		return 0;
1347 
1348 	if (ND6_LLINFO_PERMANENT(ln))
1349 		return 0;
1350 
1351 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1352 		return 0;
1353 
1354 	LLE_WLOCK(ln);
1355 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1356 		ln->ln_state = ND6_LLINFO_STALE;
1357 	else
1358 		ln->ln_state = ND6_LLINFO_PURGE;
1359 	nd6_llinfo_settimer(ln, 0);
1360 	LLE_WUNLOCK(ln);
1361 
1362 	(*n)--;
1363 	return 0;
1364 }
1365 
1366 static void
1367 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1368 {
1369 
1370 	if (ip6_neighborgcthresh >= 0 &&
1371 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1372 		struct gc_args gc_args = {10, in6};
1373 		/*
1374 		 * XXX entries that are "less recently used" should be
1375 		 * freed first.
1376 		 */
1377 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1378 	}
1379 }
1380 
1381 void
1382 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1383 {
1384 	struct sockaddr *gate = rt->rt_gateway;
1385 	struct ifnet *ifp = rt->rt_ifp;
1386 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1387 	struct ifaddr *ifa;
1388 
1389 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1390 
1391 	if (req == RTM_LLINFO_UPD) {
1392 		int rc;
1393 		struct in6_addr *in6;
1394 		struct in6_addr in6_all;
1395 		int anycast;
1396 
1397 		if ((ifa = info->rti_ifa) == NULL)
1398 			return;
1399 
1400 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1401 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1402 
1403 		in6_all = in6addr_linklocal_allnodes;
1404 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1405 			log(LOG_ERR, "%s: failed to set scope %s "
1406 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1407 			return;
1408 		}
1409 
1410 		/* XXX don't set Override for proxy addresses */
1411 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1412 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1413 #if 0
1414 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1415 #endif
1416 		    , 1, NULL);
1417 		return;
1418 	}
1419 
1420 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1421 		return;
1422 
1423 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1424 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1425 		/*
1426 		 * This is probably an interface direct route for a link
1427 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1428 		 * We do not need special treatment below for such a route.
1429 		 * Moreover, the RTF_LLINFO flag which would be set below
1430 		 * would annoy the ndp(8) command.
1431 		 */
1432 		return;
1433 	}
1434 
1435 	switch (req) {
1436 	case RTM_ADD: {
1437 		struct psref psref;
1438 
1439 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1440 		/*
1441 		 * There is no backward compatibility :)
1442 		 *
1443 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1444 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1445 		 *	   rt->rt_flags |= RTF_CLONING;
1446 		 */
1447 		/* XXX should move to route.c? */
1448 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1449 			union {
1450 				struct sockaddr sa;
1451 				struct sockaddr_dl sdl;
1452 				struct sockaddr_storage ss;
1453 			} u;
1454 			/*
1455 			 * Case 1: This route should come from a route to
1456 			 * interface (RTF_CLONING case) or the route should be
1457 			 * treated as on-link but is currently not
1458 			 * (RTF_LLINFO && ln == NULL case).
1459 			 */
1460 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1461 			    ifp->if_index, ifp->if_type,
1462 			    NULL, namelen, NULL, addrlen) == NULL) {
1463 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1464 				    "failed on %s\n", __func__, __LINE__,
1465 				    sizeof(u.ss), if_name(ifp));
1466 			}
1467 			rt_setgate(rt, &u.sa);
1468 			gate = rt->rt_gateway;
1469 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1470 			if (gate == NULL) {
1471 				log(LOG_ERR,
1472 				    "%s: rt_setgate failed on %s\n", __func__,
1473 				    if_name(ifp));
1474 				break;
1475 			}
1476 
1477 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1478 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1479 				break;
1480 		}
1481 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1482 		/*
1483 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1484 		 * We don't do that here since llinfo is not ready yet.
1485 		 *
1486 		 * There are also couple of other things to be discussed:
1487 		 * - unsolicited NA code needs improvement beforehand
1488 		 * - RFC2461 says we MAY send multicast unsolicited NA
1489 		 *   (7.2.6 paragraph 4), however, it also says that we
1490 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1491 		 *   we don't have anything like it right now.
1492 		 *   note that the mechanism needs a mutual agreement
1493 		 *   between proxies, which means that we need to implement
1494 		 *   a new protocol, or a new kludge.
1495 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1496 		 *   we need to check ip6forwarding before sending it.
1497 		 *   (or should we allow proxy ND configuration only for
1498 		 *   routers?  there's no mention about proxy ND from hosts)
1499 		 */
1500 #if 0
1501 		/* XXX it does not work */
1502 		if (rt->rt_flags & RTF_ANNOUNCE)
1503 			nd6_na_output(ifp,
1504 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1505 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1506 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1507 			      1, NULL);
1508 #endif
1509 
1510 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1511 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1512 			/*
1513 			 * Address resolution isn't necessary for a point to
1514 			 * point link, so we can skip this test for a p2p link.
1515 			 */
1516 			if (gate->sa_family != AF_LINK ||
1517 			    gate->sa_len <
1518 			    sockaddr_dl_measure(namelen, addrlen)) {
1519 				log(LOG_DEBUG,
1520 				    "nd6_rtrequest: bad gateway value: %s\n",
1521 				    if_name(ifp));
1522 				break;
1523 			}
1524 			satosdl(gate)->sdl_type = ifp->if_type;
1525 			satosdl(gate)->sdl_index = ifp->if_index;
1526 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1527 		}
1528 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1529 
1530 		/*
1531 		 * When called from rt_ifa_addlocal, we cannot depend on that
1532 		 * the address (rt_getkey(rt)) exits in the address list of the
1533 		 * interface. So check RTF_LOCAL instead.
1534 		 */
1535 		if (rt->rt_flags & RTF_LOCAL) {
1536 			if (nd6_useloopback)
1537 				rt->rt_ifp = lo0ifp;	/* XXX */
1538 			break;
1539 		}
1540 
1541 		/*
1542 		 * check if rt_getkey(rt) is an address assigned
1543 		 * to the interface.
1544 		 */
1545 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp,
1546 		    &satocsin6(rt_getkey(rt))->sin6_addr, &psref);
1547 		if (ifa != NULL) {
1548 			if (nd6_useloopback) {
1549 				rt->rt_ifp = lo0ifp;	/* XXX */
1550 				/*
1551 				 * Make sure rt_ifa be equal to the ifaddr
1552 				 * corresponding to the address.
1553 				 * We need this because when we refer
1554 				 * rt_ifa->ia6_flags in ip6_input, we assume
1555 				 * that the rt_ifa points to the address instead
1556 				 * of the loopback address.
1557 				 */
1558 				if (ifa != rt->rt_ifa)
1559 					rt_replace_ifa(rt, ifa);
1560 			}
1561 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1562 			/* join solicited node multicast for proxy ND */
1563 			if (ifp->if_flags & IFF_MULTICAST) {
1564 				struct in6_addr llsol;
1565 				int error;
1566 
1567 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1568 				llsol.s6_addr32[0] = htonl(0xff020000);
1569 				llsol.s6_addr32[1] = 0;
1570 				llsol.s6_addr32[2] = htonl(1);
1571 				llsol.s6_addr8[12] = 0xff;
1572 				if (in6_setscope(&llsol, ifp, NULL))
1573 					goto out;
1574 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1575 					char ip6buf[INET6_ADDRSTRLEN];
1576 					nd6log(LOG_ERR, "%s: failed to join "
1577 					    "%s (errno=%d)\n", if_name(ifp),
1578 					    IN6_PRINT(ip6buf, &llsol), error);
1579 				}
1580 			}
1581 		}
1582 	out:
1583 		ifa_release(ifa, &psref);
1584 		/*
1585 		 * If we have too many cache entries, initiate immediate
1586 		 * purging for some entries.
1587 		 */
1588 		if (rt->rt_ifp != NULL)
1589 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1590 		break;
1591 	    }
1592 
1593 	case RTM_DELETE:
1594 		/* leave from solicited node multicast for proxy ND */
1595 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1596 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1597 			struct in6_addr llsol;
1598 			struct in6_multi *in6m;
1599 
1600 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1601 			llsol.s6_addr32[0] = htonl(0xff020000);
1602 			llsol.s6_addr32[1] = 0;
1603 			llsol.s6_addr32[2] = htonl(1);
1604 			llsol.s6_addr8[12] = 0xff;
1605 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1606 				in6m = in6_lookup_multi(&llsol, ifp);
1607 				if (in6m)
1608 					in6_delmulti(in6m);
1609 			}
1610 		}
1611 		break;
1612 	}
1613 }
1614 
1615 int
1616 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1617 {
1618 	struct in6_drlist *drl = (struct in6_drlist *)data;
1619 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1620 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1621 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1622 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1623 	struct nd_defrouter *dr;
1624 	struct nd_prefix *pr;
1625 	int i = 0, error = 0;
1626 
1627 	switch (cmd) {
1628 	case SIOCGDRLST_IN6:
1629 		/*
1630 		 * obsolete API, use sysctl under net.inet6.icmp6
1631 		 */
1632 		memset(drl, 0, sizeof(*drl));
1633 		ND6_RLOCK();
1634 		ND_DEFROUTER_LIST_FOREACH(dr) {
1635 			if (i >= DRLSTSIZ)
1636 				break;
1637 			drl->defrouter[i].rtaddr = dr->rtaddr;
1638 			in6_clearscope(&drl->defrouter[i].rtaddr);
1639 
1640 			drl->defrouter[i].flags = dr->flags;
1641 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1642 			drl->defrouter[i].expire = dr->expire ?
1643 			    time_mono_to_wall(dr->expire) : 0;
1644 			drl->defrouter[i].if_index = dr->ifp->if_index;
1645 			i++;
1646 		}
1647 		ND6_UNLOCK();
1648 		break;
1649 	case SIOCGPRLST_IN6:
1650 		/*
1651 		 * obsolete API, use sysctl under net.inet6.icmp6
1652 		 *
1653 		 * XXX the structure in6_prlist was changed in backward-
1654 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1655 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1656 		 */
1657 		/*
1658 		 * XXX meaning of fields, especialy "raflags", is very
1659 		 * differnet between RA prefix list and RR/static prefix list.
1660 		 * how about separating ioctls into two?
1661 		 */
1662 		memset(oprl, 0, sizeof(*oprl));
1663 		ND6_RLOCK();
1664 		ND_PREFIX_LIST_FOREACH(pr) {
1665 			struct nd_pfxrouter *pfr;
1666 			int j;
1667 
1668 			if (i >= PRLSTSIZ)
1669 				break;
1670 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1671 			oprl->prefix[i].raflags = pr->ndpr_raf;
1672 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1673 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1674 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1675 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1676 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1677 				oprl->prefix[i].expire = 0;
1678 			else {
1679 				time_t maxexpire;
1680 
1681 				/* XXX: we assume time_t is signed. */
1682 				maxexpire = (-1) &
1683 				    ~((time_t)1 <<
1684 				    ((sizeof(maxexpire) * 8) - 1));
1685 				if (pr->ndpr_vltime <
1686 				    maxexpire - pr->ndpr_lastupdate) {
1687 					time_t expire;
1688 					expire = pr->ndpr_lastupdate +
1689 					    pr->ndpr_vltime;
1690 					oprl->prefix[i].expire = expire ?
1691 					    time_mono_to_wall(expire) : 0;
1692 				} else
1693 					oprl->prefix[i].expire = maxexpire;
1694 			}
1695 
1696 			j = 0;
1697 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1698 				if (j < DRLSTSIZ) {
1699 #define RTRADDR oprl->prefix[i].advrtr[j]
1700 					RTRADDR = pfr->router->rtaddr;
1701 					in6_clearscope(&RTRADDR);
1702 #undef RTRADDR
1703 				}
1704 				j++;
1705 			}
1706 			oprl->prefix[i].advrtrs = j;
1707 			oprl->prefix[i].origin = PR_ORIG_RA;
1708 
1709 			i++;
1710 		}
1711 		ND6_UNLOCK();
1712 
1713 		break;
1714 	case OSIOCGIFINFO_IN6:
1715 #define ND	ndi->ndi
1716 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1717 		memset(&ND, 0, sizeof(ND));
1718 		ND.linkmtu = IN6_LINKMTU(ifp);
1719 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1720 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1721 		ND.reachable = ND_IFINFO(ifp)->reachable;
1722 		ND.retrans = ND_IFINFO(ifp)->retrans;
1723 		ND.flags = ND_IFINFO(ifp)->flags;
1724 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1725 		ND.chlim = ND_IFINFO(ifp)->chlim;
1726 		break;
1727 	case SIOCGIFINFO_IN6:
1728 		ND = *ND_IFINFO(ifp);
1729 		break;
1730 	case SIOCSIFINFO_IN6:
1731 		/*
1732 		 * used to change host variables from userland.
1733 		 * intented for a use on router to reflect RA configurations.
1734 		 */
1735 		/* 0 means 'unspecified' */
1736 		if (ND.linkmtu != 0) {
1737 			if (ND.linkmtu < IPV6_MMTU ||
1738 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1739 				error = EINVAL;
1740 				break;
1741 			}
1742 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1743 		}
1744 
1745 		if (ND.basereachable != 0) {
1746 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1747 
1748 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1749 			if (ND.basereachable != obasereachable)
1750 				ND_IFINFO(ifp)->reachable =
1751 				    ND_COMPUTE_RTIME(ND.basereachable);
1752 		}
1753 		if (ND.retrans != 0)
1754 			ND_IFINFO(ifp)->retrans = ND.retrans;
1755 		if (ND.chlim != 0)
1756 			ND_IFINFO(ifp)->chlim = ND.chlim;
1757 		/* FALLTHROUGH */
1758 	case SIOCSIFINFO_FLAGS:
1759 	{
1760 		struct ifaddr *ifa;
1761 		struct in6_ifaddr *ia;
1762 		int s;
1763 
1764 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1765 		    !(ND.flags & ND6_IFF_IFDISABLED))
1766 		{
1767 			/*
1768 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1769 			 * has a link-local address with IN6_IFF_DUPLICATED,
1770 			 * do not clear ND6_IFF_IFDISABLED.
1771 			 * See RFC 4862, section 5.4.5.
1772 			 */
1773 			int duplicated_linklocal = 0;
1774 
1775 			s = pserialize_read_enter();
1776 			IFADDR_READER_FOREACH(ifa, ifp) {
1777 				if (ifa->ifa_addr->sa_family != AF_INET6)
1778 					continue;
1779 				ia = (struct in6_ifaddr *)ifa;
1780 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1781 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1782 				{
1783 					duplicated_linklocal = 1;
1784 					break;
1785 				}
1786 			}
1787 			pserialize_read_exit(s);
1788 
1789 			if (duplicated_linklocal) {
1790 				ND.flags |= ND6_IFF_IFDISABLED;
1791 				log(LOG_ERR, "Cannot enable an interface"
1792 				    " with a link-local address marked"
1793 				    " duplicate.\n");
1794 			} else {
1795 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1796 				if (ifp->if_flags & IFF_UP)
1797 					in6_if_up(ifp);
1798 			}
1799 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1800 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1801 			int bound = curlwp_bind();
1802 			/* Mark all IPv6 addresses as tentative. */
1803 
1804 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1805 			s = pserialize_read_enter();
1806 			IFADDR_READER_FOREACH(ifa, ifp) {
1807 				struct psref psref;
1808 				if (ifa->ifa_addr->sa_family != AF_INET6)
1809 					continue;
1810 				ifa_acquire(ifa, &psref);
1811 				pserialize_read_exit(s);
1812 
1813 				nd6_dad_stop(ifa);
1814 
1815 				ia = (struct in6_ifaddr *)ifa;
1816 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1817 
1818 				s = pserialize_read_enter();
1819 				ifa_release(ifa, &psref);
1820 			}
1821 			pserialize_read_exit(s);
1822 			curlwp_bindx(bound);
1823 		}
1824 
1825 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1826 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1827 				/* auto_linklocal 0->1 transition */
1828 
1829 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1830 				in6_ifattach(ifp, NULL);
1831 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1832 			    ifp->if_flags & IFF_UP)
1833 			{
1834 				/*
1835 				 * When the IF already has
1836 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1837 				 * address is assigned, and IFF_UP, try to
1838 				 * assign one.
1839 				 */
1840 				int haslinklocal = 0;
1841 
1842 				s = pserialize_read_enter();
1843 				IFADDR_READER_FOREACH(ifa, ifp) {
1844 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1845 						continue;
1846 					ia = (struct in6_ifaddr *)ifa;
1847 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1848 						haslinklocal = 1;
1849 						break;
1850 					}
1851 				}
1852 				pserialize_read_exit(s);
1853 				if (!haslinklocal)
1854 					in6_ifattach(ifp, NULL);
1855 			}
1856 		}
1857 	}
1858 		ND_IFINFO(ifp)->flags = ND.flags;
1859 		break;
1860 #undef ND
1861 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1862 		/* sync kernel routing table with the default router list */
1863 		ND6_WLOCK();
1864 		nd6_defrouter_reset();
1865 		nd6_defrouter_select();
1866 		ND6_UNLOCK();
1867 		break;
1868 	case SIOCSPFXFLUSH_IN6:
1869 	{
1870 		/* flush all the prefix advertised by routers */
1871 		struct nd_prefix *pfx, *next;
1872 
1873 	restart:
1874 		ND6_WLOCK();
1875 		ND_PREFIX_LIST_FOREACH_SAFE(pfx, next) {
1876 			struct in6_ifaddr *ia, *ia_next;
1877 			int _s;
1878 
1879 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1880 				continue; /* XXX */
1881 
1882 			/* do we really have to remove addresses as well? */
1883 			_s = pserialize_read_enter();
1884 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1885 			     ia = ia_next) {
1886 				/* ia might be removed.  keep the next ptr. */
1887 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1888 
1889 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1890 					continue;
1891 
1892 				if (ia->ia6_ndpr == pfx) {
1893 					pserialize_read_exit(_s);
1894 					ND6_UNLOCK();
1895 					/* XXX NOMPSAFE? */
1896 					/* in6_purgeaddr may destroy pfx. */
1897 					in6_purgeaddr(&ia->ia_ifa);
1898 					goto restart;
1899 				}
1900 			}
1901 			pserialize_read_exit(_s);
1902 
1903 			KASSERT(pfx->ndpr_refcnt == 0);
1904 			nd6_prelist_remove(pfx);
1905 		}
1906 		ND6_UNLOCK();
1907 		break;
1908 	}
1909 	case SIOCSRTRFLUSH_IN6:
1910 	{
1911 		/* flush all the default routers */
1912 		struct nd_defrouter *drtr, *next;
1913 
1914 		ND6_WLOCK();
1915 		nd6_defrouter_reset();
1916 		ND_DEFROUTER_LIST_FOREACH_SAFE(drtr, next) {
1917 			nd6_defrtrlist_del(drtr, NULL);
1918 		}
1919 		nd6_defrouter_select();
1920 		ND6_UNLOCK();
1921 		break;
1922 	}
1923 	case SIOCGNBRINFO_IN6:
1924 	{
1925 		struct llentry *ln;
1926 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1927 
1928 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1929 			return error;
1930 
1931 		ln = nd6_lookup(&nb_addr, ifp, false);
1932 		if (ln == NULL) {
1933 			error = EINVAL;
1934 			break;
1935 		}
1936 		nbi->state = ln->ln_state;
1937 		nbi->asked = ln->ln_asked;
1938 		nbi->isrouter = ln->ln_router;
1939 		nbi->expire = ln->ln_expire ?
1940 		    time_mono_to_wall(ln->ln_expire) : 0;
1941 		LLE_RUNLOCK(ln);
1942 
1943 		break;
1944 	}
1945 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1946 		ndif->ifindex = nd6_defifindex;
1947 		break;
1948 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1949 		return nd6_setdefaultiface(ndif->ifindex);
1950 	}
1951 	return error;
1952 }
1953 
1954 void
1955 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1956 {
1957 	struct mbuf *m_hold, *m_hold_next;
1958 	struct sockaddr_in6 sin6;
1959 
1960 	LLE_WLOCK_ASSERT(ln);
1961 
1962 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1963 
1964 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1965 
1966 	LLE_WUNLOCK(ln);
1967 	for (; m_hold != NULL; m_hold = m_hold_next) {
1968 		m_hold_next = m_hold->m_nextpkt;
1969 		m_hold->m_nextpkt = NULL;
1970 
1971 		/*
1972 		 * we assume ifp is not a p2p here, so
1973 		 * just set the 2nd argument as the
1974 		 * 1st one.
1975 		 */
1976 		ip6_if_output(ifp, ifp, m_hold, &sin6, NULL);
1977 	}
1978 	LLE_WLOCK(ln);
1979 }
1980 
1981 /*
1982  * Create neighbor cache entry and cache link-layer address,
1983  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1984  */
1985 void
1986 nd6_cache_lladdr(
1987     struct ifnet *ifp,
1988     struct in6_addr *from,
1989     char *lladdr,
1990     int lladdrlen,
1991     int type,	/* ICMP6 type */
1992     int code	/* type dependent information */
1993 )
1994 {
1995 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
1996 	struct llentry *ln = NULL;
1997 	int is_newentry;
1998 	int do_update;
1999 	int olladdr;
2000 	int llchange;
2001 	int newstate = 0;
2002 	uint16_t router = 0;
2003 
2004 	KASSERT(ifp != NULL);
2005 	KASSERT(from != NULL);
2006 
2007 	/* nothing must be updated for unspecified address */
2008 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2009 		return;
2010 
2011 	/*
2012 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2013 	 * the caller.
2014 	 *
2015 	 * XXX If the link does not have link-layer adderss, what should
2016 	 * we do? (ifp->if_addrlen == 0)
2017 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2018 	 * description on it in NS section (RFC 2461 7.2.3).
2019 	 */
2020 
2021 	ln = nd6_lookup(from, ifp, true);
2022 	if (ln == NULL) {
2023 #if 0
2024 		/* nothing must be done if there's no lladdr */
2025 		if (!lladdr || !lladdrlen)
2026 			return NULL;
2027 #endif
2028 
2029 		ln = nd6_create(from, ifp);
2030 		is_newentry = 1;
2031 	} else {
2032 		/* do nothing if static ndp is set */
2033 		if (ln->la_flags & LLE_STATIC) {
2034 			LLE_WUNLOCK(ln);
2035 			return;
2036 		}
2037 		is_newentry = 0;
2038 	}
2039 
2040 	if (ln == NULL)
2041 		return;
2042 
2043 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2044 	if (olladdr && lladdr) {
2045 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2046 	} else
2047 		llchange = 0;
2048 
2049 	/*
2050 	 * newentry olladdr  lladdr  llchange	(*=record)
2051 	 *	0	n	n	--	(1)
2052 	 *	0	y	n	--	(2)
2053 	 *	0	n	y	--	(3) * STALE
2054 	 *	0	y	y	n	(4) *
2055 	 *	0	y	y	y	(5) * STALE
2056 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2057 	 *	1	--	y	--	(7) * STALE
2058 	 */
2059 
2060 	if (lladdr) {		/* (3-5) and (7) */
2061 		/*
2062 		 * Record source link-layer address
2063 		 * XXX is it dependent to ifp->if_type?
2064 		 */
2065 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2066 		ln->la_flags |= LLE_VALID;
2067 	}
2068 
2069 	if (!is_newentry) {
2070 		if ((!olladdr && lladdr) ||		/* (3) */
2071 		    (olladdr && lladdr && llchange)) {	/* (5) */
2072 			do_update = 1;
2073 			newstate = ND6_LLINFO_STALE;
2074 		} else					/* (1-2,4) */
2075 			do_update = 0;
2076 	} else {
2077 		do_update = 1;
2078 		if (lladdr == NULL)			/* (6) */
2079 			newstate = ND6_LLINFO_NOSTATE;
2080 		else					/* (7) */
2081 			newstate = ND6_LLINFO_STALE;
2082 	}
2083 
2084 	if (do_update) {
2085 		/*
2086 		 * Update the state of the neighbor cache.
2087 		 */
2088 		ln->ln_state = newstate;
2089 
2090 		if (ln->ln_state == ND6_LLINFO_STALE) {
2091 			/*
2092 			 * XXX: since nd6_output() below will cause
2093 			 * state tansition to DELAY and reset the timer,
2094 			 * we must set the timer now, although it is actually
2095 			 * meaningless.
2096 			 */
2097 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2098 
2099 			nd6_llinfo_release_pkts(ln, ifp);
2100 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2101 			/* probe right away */
2102 			nd6_llinfo_settimer((void *)ln, 0);
2103 		}
2104 	}
2105 
2106 	/*
2107 	 * ICMP6 type dependent behavior.
2108 	 *
2109 	 * NS: clear IsRouter if new entry
2110 	 * RS: clear IsRouter
2111 	 * RA: set IsRouter if there's lladdr
2112 	 * redir: clear IsRouter if new entry
2113 	 *
2114 	 * RA case, (1):
2115 	 * The spec says that we must set IsRouter in the following cases:
2116 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2117 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2118 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2119 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2120 	 * neighbor cache, this is similar to (6).
2121 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2122 	 *
2123 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2124 	 *							D R
2125 	 *	0	n	n	--	(1)	c   ?     s
2126 	 *	0	y	n	--	(2)	c   s     s
2127 	 *	0	n	y	--	(3)	c   s     s
2128 	 *	0	y	y	n	(4)	c   s     s
2129 	 *	0	y	y	y	(5)	c   s     s
2130 	 *	1	--	n	--	(6) c	c 	c s
2131 	 *	1	--	y	--	(7) c	c   s	c s
2132 	 *
2133 	 *					(c=clear s=set)
2134 	 */
2135 	switch (type & 0xff) {
2136 	case ND_NEIGHBOR_SOLICIT:
2137 		/*
2138 		 * New entry must have is_router flag cleared.
2139 		 */
2140 		if (is_newentry)	/* (6-7) */
2141 			ln->ln_router = 0;
2142 		break;
2143 	case ND_REDIRECT:
2144 		/*
2145 		 * If the icmp is a redirect to a better router, always set the
2146 		 * is_router flag.  Otherwise, if the entry is newly created,
2147 		 * clear the flag.  [RFC 2461, sec 8.3]
2148 		 */
2149 		if (code == ND_REDIRECT_ROUTER)
2150 			ln->ln_router = 1;
2151 		else if (is_newentry) /* (6-7) */
2152 			ln->ln_router = 0;
2153 		break;
2154 	case ND_ROUTER_SOLICIT:
2155 		/*
2156 		 * is_router flag must always be cleared.
2157 		 */
2158 		ln->ln_router = 0;
2159 		break;
2160 	case ND_ROUTER_ADVERT:
2161 		/*
2162 		 * Mark an entry with lladdr as a router.
2163 		 */
2164 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2165 		    (is_newentry && lladdr)) {			/* (7) */
2166 			ln->ln_router = 1;
2167 		}
2168 		break;
2169 	}
2170 
2171 #if 0
2172 	/* XXX should we send rtmsg as it used to be? */
2173 	if (do_update)
2174 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2175 #endif
2176 
2177 	if (ln != NULL) {
2178 		router = ln->ln_router;
2179 		LLE_WUNLOCK(ln);
2180 	}
2181 
2182 	/*
2183 	 * If we have too many cache entries, initiate immediate
2184 	 * purging for some entries.
2185 	 */
2186 	if (is_newentry)
2187 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2188 
2189 	/*
2190 	 * When the link-layer address of a router changes, select the
2191 	 * best router again.  In particular, when the neighbor entry is newly
2192 	 * created, it might affect the selection policy.
2193 	 * Question: can we restrict the first condition to the "is_newentry"
2194 	 * case?
2195 	 * XXX: when we hear an RA from a new router with the link-layer
2196 	 * address option, nd6_defrouter_select() is called twice, since
2197 	 * defrtrlist_update called the function as well.  However, I believe
2198 	 * we can compromise the overhead, since it only happens the first
2199 	 * time.
2200 	 * XXX: although nd6_defrouter_select() should not have a bad effect
2201 	 * for those are not autoconfigured hosts, we explicitly avoid such
2202 	 * cases for safety.
2203 	 */
2204 	if (do_update && router && !ip6_forwarding &&
2205 	    nd6_accepts_rtadv(ndi)) {
2206 		ND6_WLOCK();
2207 		nd6_defrouter_select();
2208 		ND6_UNLOCK();
2209 	}
2210 }
2211 
2212 static void
2213 nd6_slowtimo(void *ignored_arg)
2214 {
2215 	struct nd_ifinfo *nd6if;
2216 	struct ifnet *ifp;
2217 	int s;
2218 
2219 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
2220 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2221 	    nd6_slowtimo, NULL);
2222 
2223 	s = pserialize_read_enter();
2224 	IFNET_READER_FOREACH(ifp) {
2225 		nd6if = ND_IFINFO(ifp);
2226 		if (nd6if->basereachable && /* already initialized */
2227 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2228 			/*
2229 			 * Since reachable time rarely changes by router
2230 			 * advertisements, we SHOULD insure that a new random
2231 			 * value gets recomputed at least once every few hours.
2232 			 * (RFC 2461, 6.3.4)
2233 			 */
2234 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2235 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2236 		}
2237 	}
2238 	pserialize_read_exit(s);
2239 
2240 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
2241 }
2242 
2243 /*
2244  * Return 0 if a neighbor cache is found. Return EWOULDBLOCK if a cache is not
2245  * found and trying to resolve a neighbor; in this case the mbuf is queued in
2246  * the list. Otherwise return errno after freeing the mbuf.
2247  */
2248 int
2249 nd6_resolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
2250     const struct sockaddr *_dst, uint8_t *lldst, size_t dstsize)
2251 {
2252 	struct llentry *ln = NULL;
2253 	bool created = false;
2254 	const struct sockaddr_in6 *dst = satocsin6(_dst);
2255 
2256 	/* discard the packet if IPv6 operation is disabled on the interface */
2257 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2258 		m_freem(m);
2259 		return ENETDOWN; /* better error? */
2260 	}
2261 
2262 	/*
2263 	 * Address resolution or Neighbor Unreachability Detection
2264 	 * for the next hop.
2265 	 * At this point, the destination of the packet must be a unicast
2266 	 * or an anycast address(i.e. not a multicast).
2267 	 */
2268 
2269 	/* Look up the neighbor cache for the nexthop */
2270 	ln = nd6_lookup(&dst->sin6_addr, ifp, false);
2271 
2272 	if (ln != NULL && (ln->la_flags & LLE_VALID) != 0) {
2273 		KASSERT(ln->ln_state > ND6_LLINFO_INCOMPLETE);
2274 		/* Fast path */
2275 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2276 		LLE_RUNLOCK(ln);
2277 		return 0;
2278 	}
2279 	if (ln != NULL)
2280 		LLE_RUNLOCK(ln);
2281 
2282 	/* Slow path */
2283 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2284 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp))  {
2285 		struct sockaddr_in6 sin6;
2286 		/*
2287 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2288 		 * the condition below is not very efficient.  But we believe
2289 		 * it is tolerable, because this should be a rare case.
2290 		 */
2291 		ln = nd6_create(&dst->sin6_addr, ifp);
2292 		if (ln == NULL) {
2293 			char ip6buf[INET6_ADDRSTRLEN];
2294 			log(LOG_DEBUG,
2295 			    "%s: can't allocate llinfo for %s "
2296 			    "(ln=%p, rt=%p)\n", __func__,
2297 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
2298 			m_freem(m);
2299 			return ENOBUFS;
2300 		}
2301 
2302 		sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
2303 		rt_clonedmsg(sin6tosa(&sin6), ifp, rt);
2304 
2305 		created = true;
2306 	}
2307 
2308 	if (ln == NULL) {
2309 		m_freem(m);
2310 		return ENETDOWN; /* better error? */
2311 	}
2312 
2313 	LLE_WLOCK_ASSERT(ln);
2314 
2315 	/* We don't have to do link-layer address resolution on a p2p link. */
2316 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2317 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2318 		ln->ln_state = ND6_LLINFO_STALE;
2319 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2320 	}
2321 
2322 	/*
2323 	 * The first time we send a packet to a neighbor whose entry is
2324 	 * STALE, we have to change the state to DELAY and a sets a timer to
2325 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2326 	 * neighbor unreachability detection on expiration.
2327 	 * (RFC 2461 7.3.3)
2328 	 */
2329 	if (ln->ln_state == ND6_LLINFO_STALE) {
2330 		ln->ln_asked = 0;
2331 		ln->ln_state = ND6_LLINFO_DELAY;
2332 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2333 	}
2334 
2335 	/*
2336 	 * There is a neighbor cache entry, but no ethernet address
2337 	 * response yet.  Append this latest packet to the end of the
2338 	 * packet queue in the mbuf, unless the number of the packet
2339 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2340 	 * the oldest packet in the queue will be removed.
2341 	 */
2342 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2343 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2344 	if (ln->ln_hold) {
2345 		struct mbuf *m_hold;
2346 		int i;
2347 
2348 		i = 0;
2349 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2350 			i++;
2351 			if (m_hold->m_nextpkt == NULL) {
2352 				m_hold->m_nextpkt = m;
2353 				break;
2354 			}
2355 		}
2356 		while (i >= nd6_maxqueuelen) {
2357 			m_hold = ln->ln_hold;
2358 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2359 			m_freem(m_hold);
2360 			i--;
2361 		}
2362 	} else {
2363 		ln->ln_hold = m;
2364 	}
2365 
2366 	/*
2367 	 * If there has been no NS for the neighbor after entering the
2368 	 * INCOMPLETE state, send the first solicitation.
2369 	 */
2370 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2371 		struct in6_addr src, *psrc;
2372 
2373 		ln->ln_asked++;
2374 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2375 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2376 		LLE_WUNLOCK(ln);
2377 		ln = NULL;
2378 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
2379 	} else {
2380 		/* We did the lookup so we need to do the unlock here. */
2381 		LLE_WUNLOCK(ln);
2382 	}
2383 
2384 	if (created)
2385 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2386 
2387 	return EWOULDBLOCK;
2388 }
2389 
2390 int
2391 nd6_need_cache(struct ifnet *ifp)
2392 {
2393 	/*
2394 	 * XXX: we currently do not make neighbor cache on any interface
2395 	 * other than ARCnet, Ethernet, FDDI and GIF.
2396 	 *
2397 	 * RFC2893 says:
2398 	 * - unidirectional tunnels needs no ND
2399 	 */
2400 	switch (ifp->if_type) {
2401 	case IFT_ARCNET:
2402 	case IFT_ETHER:
2403 	case IFT_FDDI:
2404 	case IFT_IEEE1394:
2405 	case IFT_CARP:
2406 	case IFT_GIF:		/* XXX need more cases? */
2407 	case IFT_PPP:
2408 	case IFT_TUNNEL:
2409 		return 1;
2410 	default:
2411 		return 0;
2412 	}
2413 }
2414 
2415 static void
2416 clear_llinfo_pqueue(struct llentry *ln)
2417 {
2418 	struct mbuf *m_hold, *m_hold_next;
2419 
2420 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2421 		m_hold_next = m_hold->m_nextpkt;
2422 		m_hold->m_nextpkt = NULL;
2423 		m_freem(m_hold);
2424 	}
2425 
2426 	ln->ln_hold = NULL;
2427 	return;
2428 }
2429 
2430 int
2431 nd6_sysctl(
2432     int name,
2433     void *oldp,	/* syscall arg, need copyout */
2434     size_t *oldlenp,
2435     void *newp,	/* syscall arg, need copyin */
2436     size_t newlen
2437 )
2438 {
2439 	void *p;
2440 	size_t ol;
2441 	int error;
2442 	size_t bufsize = 0;
2443 
2444 	error = 0;
2445 
2446 	if (newp)
2447 		return EPERM;
2448 	if (oldp && !oldlenp)
2449 		return EINVAL;
2450 	ol = oldlenp ? *oldlenp : 0;
2451 
2452 	if (oldp && *oldlenp > 0) {
2453 		p = kmem_alloc(*oldlenp, KM_SLEEP);
2454 		bufsize = *oldlenp;
2455 	} else
2456 		p = NULL;
2457 	switch (name) {
2458 	case ICMPV6CTL_ND6_DRLIST:
2459 		error = fill_drlist(p, oldlenp, ol);
2460 		if (!error && p != NULL && oldp != NULL)
2461 			error = copyout(p, oldp, *oldlenp);
2462 		break;
2463 
2464 	case ICMPV6CTL_ND6_PRLIST:
2465 		error = fill_prlist(p, oldlenp, ol);
2466 		if (!error && p != NULL && oldp != NULL)
2467 			error = copyout(p, oldp, *oldlenp);
2468 		break;
2469 
2470 	case ICMPV6CTL_ND6_MAXQLEN:
2471 		break;
2472 
2473 	default:
2474 		error = ENOPROTOOPT;
2475 		break;
2476 	}
2477 	if (p)
2478 		kmem_free(p, bufsize);
2479 
2480 	return error;
2481 }
2482 
2483 static int
2484 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2485 {
2486 	int error = 0;
2487 	struct in6_defrouter *d = NULL, *de = NULL;
2488 	struct nd_defrouter *dr;
2489 	size_t l;
2490 
2491 	if (oldp) {
2492 		d = (struct in6_defrouter *)oldp;
2493 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2494 	}
2495 	l = 0;
2496 
2497 	ND6_RLOCK();
2498 	ND_DEFROUTER_LIST_FOREACH(dr) {
2499 
2500 		if (oldp && d + 1 <= de) {
2501 			memset(d, 0, sizeof(*d));
2502 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2503 			if (sa6_recoverscope(&d->rtaddr)) {
2504 				char ip6buf[INET6_ADDRSTRLEN];
2505 				log(LOG_ERR,
2506 				    "scope error in router list (%s)\n",
2507 				    IN6_PRINT(ip6buf, &d->rtaddr.sin6_addr));
2508 				/* XXX: press on... */
2509 			}
2510 			d->flags = dr->flags;
2511 			d->rtlifetime = dr->rtlifetime;
2512 			d->expire = dr->expire ?
2513 			    time_mono_to_wall(dr->expire) : 0;
2514 			d->if_index = dr->ifp->if_index;
2515 		}
2516 
2517 		l += sizeof(*d);
2518 		if (d)
2519 			d++;
2520 	}
2521 	ND6_UNLOCK();
2522 
2523 	if (oldp) {
2524 		if (l > ol)
2525 			error = ENOMEM;
2526 	}
2527 	if (oldlenp)
2528 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2529 
2530 	return error;
2531 }
2532 
2533 static int
2534 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2535 {
2536 	int error = 0;
2537 	struct nd_prefix *pr;
2538 	uint8_t *p = NULL, *ps = NULL;
2539 	uint8_t *pe = NULL;
2540 	size_t l;
2541 	char ip6buf[INET6_ADDRSTRLEN];
2542 
2543 	if (oldp) {
2544 		ps = p = (uint8_t*)oldp;
2545 		pe = (uint8_t*)oldp + *oldlenp;
2546 	}
2547 	l = 0;
2548 
2549 	ND6_RLOCK();
2550 	ND_PREFIX_LIST_FOREACH(pr) {
2551 		u_short advrtrs;
2552 		struct sockaddr_in6 sin6;
2553 		struct nd_pfxrouter *pfr;
2554 		struct in6_prefix pfx;
2555 
2556 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2557 		{
2558 			memset(&pfx, 0, sizeof(pfx));
2559 			ps = p;
2560 			pfx.prefix = pr->ndpr_prefix;
2561 
2562 			if (sa6_recoverscope(&pfx.prefix)) {
2563 				log(LOG_ERR,
2564 				    "scope error in prefix list (%s)\n",
2565 				    IN6_PRINT(ip6buf, &pfx.prefix.sin6_addr));
2566 				/* XXX: press on... */
2567 			}
2568 			pfx.raflags = pr->ndpr_raf;
2569 			pfx.prefixlen = pr->ndpr_plen;
2570 			pfx.vltime = pr->ndpr_vltime;
2571 			pfx.pltime = pr->ndpr_pltime;
2572 			pfx.if_index = pr->ndpr_ifp->if_index;
2573 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2574 				pfx.expire = 0;
2575 			else {
2576 				time_t maxexpire;
2577 
2578 				/* XXX: we assume time_t is signed. */
2579 				maxexpire = (-1) &
2580 				    ~((time_t)1 <<
2581 				    ((sizeof(maxexpire) * 8) - 1));
2582 				if (pr->ndpr_vltime <
2583 				    maxexpire - pr->ndpr_lastupdate) {
2584 					pfx.expire = pr->ndpr_lastupdate +
2585 						pr->ndpr_vltime;
2586 				} else
2587 					pfx.expire = maxexpire;
2588 			}
2589 			pfx.refcnt = pr->ndpr_refcnt;
2590 			pfx.flags = pr->ndpr_stateflags;
2591 			pfx.origin = PR_ORIG_RA;
2592 
2593 			p += sizeof(pfx); l += sizeof(pfx);
2594 
2595 			advrtrs = 0;
2596 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2597 				if (p + sizeof(sin6) > pe) {
2598 					advrtrs++;
2599 					continue;
2600 				}
2601 
2602 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2603 					    0, 0, 0);
2604 				if (sa6_recoverscope(&sin6)) {
2605 					log(LOG_ERR,
2606 					    "scope error in "
2607 					    "prefix list (%s)\n",
2608 					    IN6_PRINT(ip6buf,
2609 					    &pfr->router->rtaddr));
2610 				}
2611 				advrtrs++;
2612 				memcpy(p, &sin6, sizeof(sin6));
2613 				p += sizeof(sin6);
2614 				l += sizeof(sin6);
2615 			}
2616 			pfx.advrtrs = advrtrs;
2617 			memcpy(ps, &pfx, sizeof(pfx));
2618 		}
2619 		else {
2620 			l += sizeof(pfx);
2621 			advrtrs = 0;
2622 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2623 				advrtrs++;
2624 				l += sizeof(sin6);
2625 			}
2626 		}
2627 	}
2628 	ND6_UNLOCK();
2629 
2630 	if (oldp) {
2631 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2632 		if (l > ol)
2633 			error = ENOMEM;
2634 	} else
2635 		*oldlenp = l;
2636 
2637 	return error;
2638 }
2639 
2640 static int
2641 nd6_setdefaultiface(int ifindex)
2642 {
2643 	ifnet_t *ifp;
2644 	int error = 0;
2645 	int s;
2646 
2647 	s = pserialize_read_enter();
2648 	ifp = if_byindex(ifindex);
2649 	if (ifp == NULL) {
2650 		pserialize_read_exit(s);
2651 		return EINVAL;
2652 	}
2653 	if (nd6_defifindex != ifindex) {
2654 		nd6_defifindex = ifindex;
2655 		nd6_defifp = nd6_defifindex > 0 ? ifp : NULL;
2656 
2657 		/*
2658 		 * Our current implementation assumes one-to-one maping between
2659 		 * interfaces and links, so it would be natural to use the
2660 		 * default interface as the default link.
2661 		 */
2662 		scope6_setdefault(nd6_defifp);
2663 	}
2664 	pserialize_read_exit(s);
2665 
2666 	return (error);
2667 }
2668