xref: /netbsd-src/sys/netinet6/nd6.c (revision f21b7d7f2cbdd5c14b3882c4e8a3d43580d460a6)
1 /*	$NetBSD: nd6.c,v 1.207 2016/09/02 07:15:14 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 1.207 2016/09/02 07:15:14 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/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/ioctl.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/cprng.h>
59 #include <sys/workqueue.h>
60 
61 #include <net/if.h>
62 #include <net/if_dl.h>
63 #include <net/if_llatbl.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/if_ether.h>
67 #include <net/if_fddi.h>
68 #include <net/if_arc.h>
69 
70 #include <netinet/in.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/icmp6_private.h>
79 
80 #include <net/net_osdep.h>
81 
82 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
83 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
84 
85 /* timer values */
86 int	nd6_prune	= 1;	/* walk list every 1 seconds */
87 int	nd6_delay	= 5;	/* delay first probe time 5 second */
88 int	nd6_umaxtries	= 3;	/* maximum unicast query */
89 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
90 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
91 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
92 
93 /* preventing too many loops in ND option parsing */
94 int nd6_maxndopt = 10;	/* max # of ND options allowed */
95 
96 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
97 
98 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
99 
100 #ifdef ND6_DEBUG
101 int nd6_debug = 1;
102 #else
103 int nd6_debug = 0;
104 #endif
105 
106 struct nd_drhead nd_defrouter;
107 struct nd_prhead nd_prefix = { 0 };
108 
109 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
110 
111 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
112 static void nd6_slowtimo(void *);
113 static int regen_tmpaddr(const struct in6_ifaddr *);
114 static void nd6_free(struct llentry *, int);
115 static void nd6_llinfo_timer(void *);
116 static void nd6_timer(void *);
117 static void nd6_timer_work(struct work *, void *);
118 static void clear_llinfo_pqueue(struct llentry *);
119 
120 static callout_t nd6_slowtimo_ch;
121 static callout_t nd6_timer_ch;
122 static struct workqueue	*nd6_timer_wq;
123 static struct work	nd6_timer_wk;
124 
125 static int fill_drlist(void *, size_t *, size_t);
126 static int fill_prlist(void *, size_t *, size_t);
127 
128 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
129 
130 void
131 nd6_init(void)
132 {
133 	int error;
134 
135 	/* initialization of the default router list */
136 	TAILQ_INIT(&nd_defrouter);
137 
138 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
139 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
140 
141 	error = workqueue_create(&nd6_timer_wq, "nd6_timer",
142 	    nd6_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
143 	if (error)
144 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
145 
146 	/* start timer */
147 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
148 	    nd6_slowtimo, NULL);
149 	callout_reset(&nd6_timer_ch, hz, nd6_timer, NULL);
150 }
151 
152 struct nd_ifinfo *
153 nd6_ifattach(struct ifnet *ifp)
154 {
155 	struct nd_ifinfo *nd;
156 
157 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
158 
159 	nd->initialized = 1;
160 
161 	nd->chlim = IPV6_DEFHLIM;
162 	nd->basereachable = REACHABLE_TIME;
163 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
164 	nd->retrans = RETRANS_TIMER;
165 
166 	nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV;
167 
168 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
169 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
170 	 * because one of its members should. */
171 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
172 	    (ifp->if_flags & IFF_LOOPBACK))
173 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
174 
175 	/* A loopback interface does not need to accept RTADV.
176 	 * A bridge interface should not accept RTADV
177 	 * because one of its members should. */
178 	if (ip6_accept_rtadv &&
179 	    !(ifp->if_flags & IFF_LOOPBACK) &&
180 	    !(ifp->if_type != IFT_BRIDGE))
181 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
182 
183 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
184 	nd6_setmtu0(ifp, nd);
185 
186 	return nd;
187 }
188 
189 void
190 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
191 {
192 
193 	nd6_purge(ifp, ext);
194 	free(ext->nd_ifinfo, M_IP6NDP);
195 }
196 
197 void
198 nd6_setmtu(struct ifnet *ifp)
199 {
200 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
201 }
202 
203 void
204 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
205 {
206 	u_int32_t omaxmtu;
207 
208 	omaxmtu = ndi->maxmtu;
209 
210 	switch (ifp->if_type) {
211 	case IFT_ARCNET:
212 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
213 		break;
214 	case IFT_FDDI:
215 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
216 		break;
217 	default:
218 		ndi->maxmtu = ifp->if_mtu;
219 		break;
220 	}
221 
222 	/*
223 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
224 	 * undesirable situation.  We thus notify the operator of the change
225 	 * explicitly.  The check for omaxmtu is necessary to restrict the
226 	 * log to the case of changing the MTU, not initializing it.
227 	 */
228 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
229 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
230 		    " small for IPv6 which needs %lu\n",
231 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
232 		    IPV6_MMTU);
233 	}
234 
235 	if (ndi->maxmtu > in6_maxmtu)
236 		in6_setmaxmtu(); /* check all interfaces just in case */
237 }
238 
239 void
240 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
241 {
242 
243 	memset(ndopts, 0, sizeof(*ndopts));
244 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
245 	ndopts->nd_opts_last
246 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
247 
248 	if (icmp6len == 0) {
249 		ndopts->nd_opts_done = 1;
250 		ndopts->nd_opts_search = NULL;
251 	}
252 }
253 
254 /*
255  * Take one ND option.
256  */
257 struct nd_opt_hdr *
258 nd6_option(union nd_opts *ndopts)
259 {
260 	struct nd_opt_hdr *nd_opt;
261 	int olen;
262 
263 	KASSERT(ndopts != NULL);
264 	KASSERT(ndopts->nd_opts_last != NULL);
265 
266 	if (ndopts->nd_opts_search == NULL)
267 		return NULL;
268 	if (ndopts->nd_opts_done)
269 		return NULL;
270 
271 	nd_opt = ndopts->nd_opts_search;
272 
273 	/* make sure nd_opt_len is inside the buffer */
274 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
275 		memset(ndopts, 0, sizeof(*ndopts));
276 		return NULL;
277 	}
278 
279 	olen = nd_opt->nd_opt_len << 3;
280 	if (olen == 0) {
281 		/*
282 		 * Message validation requires that all included
283 		 * options have a length that is greater than zero.
284 		 */
285 		memset(ndopts, 0, sizeof(*ndopts));
286 		return NULL;
287 	}
288 
289 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
290 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
291 		/* option overruns the end of buffer, invalid */
292 		memset(ndopts, 0, sizeof(*ndopts));
293 		return NULL;
294 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
295 		/* reached the end of options chain */
296 		ndopts->nd_opts_done = 1;
297 		ndopts->nd_opts_search = NULL;
298 	}
299 	return nd_opt;
300 }
301 
302 /*
303  * Parse multiple ND options.
304  * This function is much easier to use, for ND routines that do not need
305  * multiple options of the same type.
306  */
307 int
308 nd6_options(union nd_opts *ndopts)
309 {
310 	struct nd_opt_hdr *nd_opt;
311 	int i = 0;
312 
313 	KASSERT(ndopts != NULL);
314 	KASSERT(ndopts->nd_opts_last != NULL);
315 
316 	if (ndopts->nd_opts_search == NULL)
317 		return 0;
318 
319 	while (1) {
320 		nd_opt = nd6_option(ndopts);
321 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
322 			/*
323 			 * Message validation requires that all included
324 			 * options have a length that is greater than zero.
325 			 */
326 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
327 			memset(ndopts, 0, sizeof(*ndopts));
328 			return -1;
329 		}
330 
331 		if (nd_opt == NULL)
332 			goto skip1;
333 
334 		switch (nd_opt->nd_opt_type) {
335 		case ND_OPT_SOURCE_LINKADDR:
336 		case ND_OPT_TARGET_LINKADDR:
337 		case ND_OPT_MTU:
338 		case ND_OPT_REDIRECTED_HEADER:
339 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
340 				nd6log(LOG_INFO,
341 				    "duplicated ND6 option found (type=%d)\n",
342 				    nd_opt->nd_opt_type);
343 				/* XXX bark? */
344 			} else {
345 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
346 					= nd_opt;
347 			}
348 			break;
349 		case ND_OPT_PREFIX_INFORMATION:
350 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
351 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
352 					= nd_opt;
353 			}
354 			ndopts->nd_opts_pi_end =
355 				(struct nd_opt_prefix_info *)nd_opt;
356 			break;
357 		default:
358 			/*
359 			 * Unknown options must be silently ignored,
360 			 * to accommodate future extension to the protocol.
361 			 */
362 			nd6log(LOG_DEBUG,
363 			    "nd6_options: unsupported option %d - "
364 			    "option ignored\n", nd_opt->nd_opt_type);
365 		}
366 
367 skip1:
368 		i++;
369 		if (i > nd6_maxndopt) {
370 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
371 			nd6log(LOG_INFO, "too many loop in nd opt\n");
372 			break;
373 		}
374 
375 		if (ndopts->nd_opts_done)
376 			break;
377 	}
378 
379 	return 0;
380 }
381 
382 /*
383  * ND6 timer routine to handle ND6 entries
384  */
385 void
386 nd6_llinfo_settimer(struct llentry *ln, time_t xtick)
387 {
388 
389 	CTASSERT(sizeof(time_t) > sizeof(int));
390 	LLE_WLOCK_ASSERT(ln);
391 
392 	if (xtick < 0) {
393 		ln->ln_expire = 0;
394 		ln->ln_ntick = 0;
395 		callout_halt(&ln->ln_timer_ch, &ln->lle_lock);
396 	} else {
397 		ln->ln_expire = time_uptime + xtick / hz;
398 		LLE_ADDREF(ln);
399 		if (xtick > INT_MAX) {
400 			ln->ln_ntick = xtick - INT_MAX;
401 			callout_reset(&ln->ln_timer_ch, INT_MAX,
402 			    nd6_llinfo_timer, ln);
403 		} else {
404 			ln->ln_ntick = 0;
405 			callout_reset(&ln->ln_timer_ch, xtick,
406 			    nd6_llinfo_timer, ln);
407 		}
408 	}
409 }
410 
411 /*
412  * Gets source address of the first packet in hold queue
413  * and stores it in @src.
414  * Returns pointer to @src (if hold queue is not empty) or NULL.
415  */
416 static struct in6_addr *
417 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
418 {
419 	struct ip6_hdr *hip6;
420 
421 	if (ln == NULL || ln->ln_hold == NULL)
422 		return NULL;
423 
424 	/*
425 	 * assuming every packet in ln_hold has the same IP header
426 	 */
427 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
428 	/* XXX pullup? */
429 	if (sizeof(*hip6) < ln->ln_hold->m_len)
430 		*src = hip6->ip6_src;
431 	else
432 		src = NULL;
433 
434 	return src;
435 }
436 
437 static void
438 nd6_llinfo_timer(void *arg)
439 {
440 	struct llentry *ln = arg;
441 	struct ifnet *ifp;
442 	struct nd_ifinfo *ndi = NULL;
443 	bool send_ns = false;
444 	const struct in6_addr *daddr6 = NULL;
445 
446 	mutex_enter(softnet_lock);
447 	KERNEL_LOCK(1, NULL);
448 
449 	LLE_WLOCK(ln);
450 	if (ln->ln_ntick > 0) {
451 		nd6_llinfo_settimer(ln, ln->ln_ntick);
452 		goto out;
453 	}
454 
455 	if (callout_pending(&ln->la_timer)) {
456 		/*
457 		 * Here we are a bit odd here in the treatment of
458 		 * active/pending. If the pending bit is set, it got
459 		 * rescheduled before I ran. The active
460 		 * bit we ignore, since if it was stopped
461 		 * in ll_tablefree() and was currently running
462 		 * it would have return 0 so the code would
463 		 * not have deleted it since the callout could
464 		 * not be stopped so we want to go through
465 		 * with the delete here now. If the callout
466 		 * was restarted, the pending bit will be back on and
467 		 * we just want to bail since the callout_reset would
468 		 * return 1 and our reference would have been removed
469 		 * by nd6_llinfo_settimer above since canceled
470 		 * would have been 1.
471 		 */
472 		goto out;
473 	}
474 
475 	ifp = ln->lle_tbl->llt_ifp;
476 
477 	KASSERT(ifp != NULL);
478 
479 	ndi = ND_IFINFO(ifp);
480 
481 	switch (ln->ln_state) {
482 	case ND6_LLINFO_INCOMPLETE:
483 		if (ln->ln_asked < nd6_mmaxtries) {
484 			ln->ln_asked++;
485 			send_ns = true;
486 		} else {
487 			struct mbuf *m = ln->ln_hold;
488 			if (m) {
489 				struct mbuf *m0;
490 
491 				/*
492 				 * assuming every packet in ln_hold has
493 				 * the same IP header
494 				 */
495 				m0 = m->m_nextpkt;
496 				m->m_nextpkt = NULL;
497 				ln->ln_hold = m0;
498 				clear_llinfo_pqueue(ln);
499  			}
500 			nd6_free(ln, 0);
501 			ln = NULL;
502 			if (m != NULL)
503 				icmp6_error2(m, ICMP6_DST_UNREACH,
504 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
505 		}
506 		break;
507 	case ND6_LLINFO_REACHABLE:
508 		if (!ND6_LLINFO_PERMANENT(ln)) {
509 			ln->ln_state = ND6_LLINFO_STALE;
510 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
511 		}
512 		break;
513 
514 	case ND6_LLINFO_PURGE:
515 	case ND6_LLINFO_STALE:
516 		/* Garbage Collection(RFC 2461 5.3) */
517 		if (!ND6_LLINFO_PERMANENT(ln)) {
518 			nd6_free(ln, 1);
519 			ln = NULL;
520 		}
521 		break;
522 
523 	case ND6_LLINFO_DELAY:
524 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
525 			/* We need NUD */
526 			ln->ln_asked = 1;
527 			ln->ln_state = ND6_LLINFO_PROBE;
528 			daddr6 = &ln->r_l3addr.addr6;
529 			send_ns = true;
530 		} else {
531 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
532 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
533 		}
534 		break;
535 	case ND6_LLINFO_PROBE:
536 		if (ln->ln_asked < nd6_umaxtries) {
537 			ln->ln_asked++;
538 			daddr6 = &ln->r_l3addr.addr6;
539 			send_ns = true;
540 		} else {
541 			nd6_free(ln, 0);
542 			ln = NULL;
543 		}
544 		break;
545 	}
546 
547 	if (send_ns) {
548 		struct in6_addr src, *psrc;
549 		const struct in6_addr *taddr6 = &ln->r_l3addr.addr6;
550 
551 		nd6_llinfo_settimer(ln, ndi->retrans * hz / 1000);
552 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
553 		LLE_FREE_LOCKED(ln);
554 		ln = NULL;
555 		nd6_ns_output(ifp, daddr6, taddr6, psrc, 0);
556 	}
557 
558 out:
559 	if (ln != NULL)
560 		LLE_FREE_LOCKED(ln);
561 	KERNEL_UNLOCK_ONE(NULL);
562 	mutex_exit(softnet_lock);
563 }
564 
565 /*
566  * ND6 timer routine to expire default route list and prefix list
567  */
568 static void
569 nd6_timer_work(struct work *wk, void *arg)
570 {
571 	struct nd_defrouter *next_dr, *dr;
572 	struct nd_prefix *next_pr, *pr;
573 	struct in6_ifaddr *ia6, *nia6;
574 	int s, bound;
575 	struct psref psref;
576 
577 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
578 	    nd6_timer, NULL);
579 
580 	mutex_enter(softnet_lock);
581 	KERNEL_LOCK(1, NULL);
582 
583 	/* expire default router list */
584 
585 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, next_dr) {
586 		if (dr->expire && dr->expire < time_uptime) {
587 			defrtrlist_del(dr, NULL);
588 		}
589 	}
590 
591 	/*
592 	 * expire interface addresses.
593 	 * in the past the loop was inside prefix expiry processing.
594 	 * However, from a stricter speci-confrmance standpoint, we should
595 	 * rather separate address lifetimes and prefix lifetimes.
596 	 */
597 	bound = curlwp_bind();
598   addrloop:
599 	s = pserialize_read_enter();
600 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
601 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
602 
603 		ia6_acquire(ia6, &psref);
604 		pserialize_read_exit(s);
605 
606 		/* check address lifetime */
607 		if (IFA6_IS_INVALID(ia6)) {
608 			int regen = 0;
609 
610 			/*
611 			 * If the expiring address is temporary, try
612 			 * regenerating a new one.  This would be useful when
613 			 * we suspended a laptop PC, then turned it on after a
614 			 * period that could invalidate all temporary
615 			 * addresses.  Although we may have to restart the
616 			 * loop (see below), it must be after purging the
617 			 * address.  Otherwise, we'd see an infinite loop of
618 			 * regeneration.
619 			 */
620 			if (ip6_use_tempaddr &&
621 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
622 				if (regen_tmpaddr(ia6) == 0)
623 					regen = 1;
624 			}
625 
626 			ia6_release(ia6, &psref);
627  			in6_purgeaddr(&ia6->ia_ifa);
628 			ia6 = NULL;
629 
630 			if (regen)
631 				goto addrloop; /* XXX: see below */
632 		} else if (IFA6_IS_DEPRECATED(ia6)) {
633 			int oldflags = ia6->ia6_flags;
634 
635 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
636 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
637 				rt_newaddrmsg(RTM_NEWADDR,
638 				    (struct ifaddr *)ia6, 0, NULL);
639 			}
640 
641 			/*
642 			 * If a temporary address has just become deprecated,
643 			 * regenerate a new one if possible.
644 			 */
645 			if (ip6_use_tempaddr &&
646 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
647 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
648 
649 				if (regen_tmpaddr(ia6) == 0) {
650 					/*
651 					 * A new temporary address is
652 					 * generated.
653 					 * XXX: this means the address chain
654 					 * has changed while we are still in
655 					 * the loop.  Although the change
656 					 * would not cause disaster (because
657 					 * it's not a deletion, but an
658 					 * addition,) we'd rather restart the
659 					 * loop just for safety.  Or does this
660 					 * significantly reduce performance??
661 					 */
662 					ia6_release(ia6, &psref);
663 					goto addrloop;
664 				}
665 			}
666 		} else {
667 			/*
668 			 * A new RA might have made a deprecated address
669 			 * preferred.
670 			 */
671 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
672 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
673 				rt_newaddrmsg(RTM_NEWADDR,
674 				    (struct ifaddr *)ia6, 0, NULL);
675 			}
676 		}
677 		s = pserialize_read_enter();
678 		ia6_release(ia6, &psref);
679 	}
680 	pserialize_read_exit(s);
681 	curlwp_bindx(bound);
682 
683 	/* expire prefix list */
684 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, next_pr) {
685 		/*
686 		 * check prefix lifetime.
687 		 * since pltime is just for autoconf, pltime processing for
688 		 * prefix is not necessary.
689 		 */
690 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
691 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
692 
693 			/*
694 			 * address expiration and prefix expiration are
695 			 * separate.  NEVER perform in6_purgeaddr here.
696 			 */
697 
698 			prelist_remove(pr);
699 		}
700 	}
701 
702 	KERNEL_UNLOCK_ONE(NULL);
703 	mutex_exit(softnet_lock);
704 }
705 
706 static void
707 nd6_timer(void *ignored_arg)
708 {
709 
710 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
711 }
712 
713 /* ia6: deprecated/invalidated temporary address */
714 static int
715 regen_tmpaddr(const struct in6_ifaddr *ia6)
716 {
717 	struct ifaddr *ifa;
718 	struct ifnet *ifp;
719 	struct in6_ifaddr *public_ifa6 = NULL;
720 	int s;
721 
722 	ifp = ia6->ia_ifa.ifa_ifp;
723 	s = pserialize_read_enter();
724 	IFADDR_READER_FOREACH(ifa, ifp) {
725 		struct in6_ifaddr *it6;
726 
727 		if (ifa->ifa_addr->sa_family != AF_INET6)
728 			continue;
729 
730 		it6 = (struct in6_ifaddr *)ifa;
731 
732 		/* ignore no autoconf addresses. */
733 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
734 			continue;
735 
736 		/* ignore autoconf addresses with different prefixes. */
737 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
738 			continue;
739 
740 		/*
741 		 * Now we are looking at an autoconf address with the same
742 		 * prefix as ours.  If the address is temporary and is still
743 		 * preferred, do not create another one.  It would be rare, but
744 		 * could happen, for example, when we resume a laptop PC after
745 		 * a long period.
746 		 */
747 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
748 		    !IFA6_IS_DEPRECATED(it6)) {
749 			public_ifa6 = NULL;
750 			break;
751 		}
752 
753 		/*
754 		 * This is a public autoconf address that has the same prefix
755 		 * as ours.  If it is preferred, keep it.  We can't break the
756 		 * loop here, because there may be a still-preferred temporary
757 		 * address with the prefix.
758 		 */
759 		if (!IFA6_IS_DEPRECATED(it6))
760 			public_ifa6 = it6;
761 	}
762 
763 	if (public_ifa6 != NULL) {
764 		int e;
765 		struct psref psref;
766 
767 		ia6_acquire(public_ifa6, &psref);
768 		pserialize_read_exit(s);
769 		/*
770 		 * Random factor is introduced in the preferred lifetime, so
771 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
772 		 */
773 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
774 			ia6_release(public_ifa6, &psref);
775 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
776 			    " tmp addr, errno=%d\n", e);
777 			return -1;
778 		}
779 		ia6_release(public_ifa6, &psref);
780 		return 0;
781 	}
782 		pserialize_read_exit(s);
783 
784 	return -1;
785 }
786 
787 bool
788 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
789 {
790 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
791 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
792 		return true;
793 	case ND6_IFF_ACCEPT_RTADV:
794 		return ip6_accept_rtadv != 0;
795 	case ND6_IFF_OVERRIDE_RTADV:
796 	case 0:
797 	default:
798 		return false;
799 	}
800 }
801 
802 /*
803  * Nuke neighbor cache/prefix/default router management table, right before
804  * ifp goes away.
805  */
806 void
807 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
808 {
809 	struct nd_defrouter *dr, *ndr;
810 	struct nd_prefix *pr, *npr;
811 
812 	/*
813 	 * During detach, the ND info might be already removed, but
814 	 * then is explitly passed as argument.
815 	 * Otherwise get it from ifp->if_afdata.
816 	 */
817 	if (ext == NULL)
818 		ext = ifp->if_afdata[AF_INET6];
819 	if (ext == NULL)
820 		return;
821 
822 	/*
823 	 * Nuke default router list entries toward ifp.
824 	 * We defer removal of default router list entries that is installed
825 	 * in the routing table, in order to keep additional side effects as
826 	 * small as possible.
827 	 */
828 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
829 		if (dr->installed)
830 			continue;
831 
832 		if (dr->ifp == ifp) {
833 			KASSERT(ext != NULL);
834 			defrtrlist_del(dr, ext);
835 		}
836 	}
837 
838 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
839 		if (!dr->installed)
840 			continue;
841 
842 		if (dr->ifp == ifp) {
843 			KASSERT(ext != NULL);
844 			defrtrlist_del(dr, ext);
845 		}
846 	}
847 
848 	/* Nuke prefix list entries toward ifp */
849 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, npr) {
850 		if (pr->ndpr_ifp == ifp) {
851 			/*
852 			 * Because if_detach() does *not* release prefixes
853 			 * while purging addresses the reference count will
854 			 * still be above zero. We therefore reset it to
855 			 * make sure that the prefix really gets purged.
856 			 */
857 			pr->ndpr_refcnt = 0;
858 			/*
859 			 * Previously, pr->ndpr_addr is removed as well,
860 			 * but I strongly believe we don't have to do it.
861 			 * nd6_purge() is only called from in6_ifdetach(),
862 			 * which removes all the associated interface addresses
863 			 * by itself.
864 			 * (jinmei@kame.net 20010129)
865 			 */
866 			prelist_remove(pr);
867 		}
868 	}
869 
870 	/* cancel default outgoing interface setting */
871 	if (nd6_defifindex == ifp->if_index)
872 		nd6_setdefaultiface(0);
873 
874 	/* XXX: too restrictive? */
875 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
876 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
877 		if (ndi && nd6_accepts_rtadv(ndi)) {
878 			/* refresh default router list */
879 			defrouter_select();
880 		}
881 	}
882 
883 	/*
884 	 * We may not need to nuke the neighbor cache entries here
885 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
886 	 * nd6_purge() is invoked by in6_ifdetach() which is called
887 	 * from if_detach() where everything gets purged. However
888 	 * in6_ifdetach is directly called from vlan(4), so we still
889 	 * need to purge entries here.
890 	 */
891 	if (ext->lltable != NULL)
892 		lltable_purge_entries(ext->lltable);
893 }
894 
895 struct llentry *
896 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
897 {
898 	struct sockaddr_in6 sin6;
899 	struct llentry *ln;
900 
901 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
902 
903 	IF_AFDATA_RLOCK(ifp);
904 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
905 	    sin6tosa(&sin6));
906 	IF_AFDATA_RUNLOCK(ifp);
907 
908 	return ln;
909 }
910 
911 struct llentry *
912 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
913 {
914 	struct sockaddr_in6 sin6;
915 	struct llentry *ln;
916 
917 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
918 
919 	IF_AFDATA_WLOCK(ifp);
920 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE,
921 	    sin6tosa(&sin6));
922 	IF_AFDATA_WUNLOCK(ifp);
923 
924 	if (ln != NULL)
925 		ln->ln_state = ND6_LLINFO_NOSTATE;
926 
927 	return ln;
928 }
929 
930 /*
931  * Test whether a given IPv6 address is a neighbor or not, ignoring
932  * the actual neighbor cache.  The neighbor cache is ignored in order
933  * to not reenter the routing code from within itself.
934  */
935 static int
936 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
937 {
938 	struct nd_prefix *pr;
939 	struct ifaddr *dstaddr;
940 	int s;
941 
942 	/*
943 	 * A link-local address is always a neighbor.
944 	 * XXX: a link does not necessarily specify a single interface.
945 	 */
946 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
947 		struct sockaddr_in6 sin6_copy;
948 		u_int32_t zone;
949 
950 		/*
951 		 * We need sin6_copy since sa6_recoverscope() may modify the
952 		 * content (XXX).
953 		 */
954 		sin6_copy = *addr;
955 		if (sa6_recoverscope(&sin6_copy))
956 			return 0; /* XXX: should be impossible */
957 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
958 			return 0;
959 		if (sin6_copy.sin6_scope_id == zone)
960 			return 1;
961 		else
962 			return 0;
963 	}
964 
965 	/*
966 	 * If the address matches one of our addresses,
967 	 * it should be a neighbor.
968 	 * If the address matches one of our on-link prefixes, it should be a
969 	 * neighbor.
970 	 */
971 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
972 		if (pr->ndpr_ifp != ifp)
973 			continue;
974 
975 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
976 			struct rtentry *rt;
977 
978 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
979 			if (rt == NULL)
980 				continue;
981 			/*
982 			 * This is the case where multiple interfaces
983 			 * have the same prefix, but only one is installed
984 			 * into the routing table and that prefix entry
985 			 * is not the one being examined here. In the case
986 			 * where RADIX_MPATH is enabled, multiple route
987 			 * entries (of the same rt_key value) will be
988 			 * installed because the interface addresses all
989 			 * differ.
990 			 */
991 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
992 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
993 				continue;
994 			}
995 		}
996 
997 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
998 		    &addr->sin6_addr, &pr->ndpr_mask))
999 			return 1;
1000 	}
1001 
1002 	/*
1003 	 * If the address is assigned on the node of the other side of
1004 	 * a p2p interface, the address should be a neighbor.
1005 	 */
1006 	s = pserialize_read_enter();
1007 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1008 	if (dstaddr != NULL) {
1009 		if (dstaddr->ifa_ifp == ifp) {
1010 			pserialize_read_exit(s);
1011 			return 1;
1012 		}
1013 	}
1014 	pserialize_read_exit(s);
1015 
1016 	/*
1017 	 * If the default router list is empty, all addresses are regarded
1018 	 * as on-link, and thus, as a neighbor.
1019 	 */
1020 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1021 	    TAILQ_EMPTY(&nd_defrouter) &&
1022 	    nd6_defifindex == ifp->if_index) {
1023 		return 1;
1024 	}
1025 
1026 	return 0;
1027 }
1028 
1029 /*
1030  * Detect if a given IPv6 address identifies a neighbor on a given link.
1031  * XXX: should take care of the destination of a p2p link?
1032  */
1033 int
1034 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1035 {
1036 	struct nd_prefix *pr;
1037 	struct llentry *ln;
1038 	struct rtentry *rt;
1039 
1040 	/*
1041 	 * A link-local address is always a neighbor.
1042 	 * XXX: a link does not necessarily specify a single interface.
1043 	 */
1044 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1045 		struct sockaddr_in6 sin6_copy;
1046 		u_int32_t zone;
1047 
1048 		/*
1049 		 * We need sin6_copy since sa6_recoverscope() may modify the
1050 		 * content (XXX).
1051 		 */
1052 		sin6_copy = *addr;
1053 		if (sa6_recoverscope(&sin6_copy))
1054 			return 0; /* XXX: should be impossible */
1055 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1056 			return 0;
1057 		if (sin6_copy.sin6_scope_id == zone)
1058 			return 1;
1059 		else
1060 			return 0;
1061 	}
1062 
1063 	/*
1064 	 * If the address matches one of our on-link prefixes, it should be a
1065 	 * neighbor.
1066 	 */
1067 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1068 		if (pr->ndpr_ifp != ifp)
1069 			continue;
1070 
1071 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1072 			continue;
1073 
1074 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1075 		    &addr->sin6_addr, &pr->ndpr_mask))
1076 			return 1;
1077 	}
1078 
1079 	/*
1080 	 * If the default router list is empty, all addresses are regarded
1081 	 * as on-link, and thus, as a neighbor.
1082 	 * XXX: we restrict the condition to hosts, because routers usually do
1083 	 * not have the "default router list".
1084 	 */
1085 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
1086 	    nd6_defifindex == ifp->if_index) {
1087 		return 1;
1088 	}
1089 
1090 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1091 	if (nd6_is_new_addr_neighbor(addr, ifp))
1092 		return 1;
1093 
1094 	/*
1095 	 * Even if the address matches none of our addresses, it might be
1096 	 * in the neighbor cache or a connected route.
1097 	 */
1098 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1099 	if (ln != NULL) {
1100 		LLE_RUNLOCK(ln);
1101 		return 1;
1102 	}
1103 
1104 	rt = rtalloc1(sin6tocsa(addr), 0);
1105 	if (rt == NULL)
1106 		return 0;
1107 
1108 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1109 #if NBRIDGE > 0
1110 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1111 #endif
1112 #if NCARP > 0
1113 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1114 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1115 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1116 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1117 #endif
1118 	    )) {
1119 		rtfree(rt);
1120 		return 1;
1121 	}
1122 	rtfree(rt);
1123 
1124 	return 0;
1125 }
1126 
1127 /*
1128  * Free an nd6 llinfo entry.
1129  * Since the function would cause significant changes in the kernel, DO NOT
1130  * make it global, unless you have a strong reason for the change, and are sure
1131  * that the change is safe.
1132  */
1133 static void
1134 nd6_free(struct llentry *ln, int gc)
1135 {
1136 	struct nd_defrouter *dr;
1137 	struct ifnet *ifp;
1138 	struct in6_addr *in6;
1139 
1140 	KASSERT(ln != NULL);
1141 	LLE_WLOCK_ASSERT(ln);
1142 
1143 	ifp = ln->lle_tbl->llt_ifp;
1144 	in6 = &ln->r_l3addr.addr6;
1145 	/*
1146 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1147 	 * even though it is not harmful, it was not really necessary.
1148 	 */
1149 
1150 	/* cancel timer */
1151 	nd6_llinfo_settimer(ln, -1);
1152 
1153 	if (!ip6_forwarding) {
1154 		int s;
1155 		s = splsoftnet();
1156 		dr = defrouter_lookup(in6, ifp);
1157 
1158 		if (dr != NULL && dr->expire &&
1159 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1160 			/*
1161 			 * If the reason for the deletion is just garbage
1162 			 * collection, and the neighbor is an active default
1163 			 * router, do not delete it.  Instead, reset the GC
1164 			 * timer using the router's lifetime.
1165 			 * Simply deleting the entry would affect default
1166 			 * router selection, which is not necessarily a good
1167 			 * thing, especially when we're using router preference
1168 			 * values.
1169 			 * XXX: the check for ln_state would be redundant,
1170 			 *      but we intentionally keep it just in case.
1171 			 */
1172 			if (dr->expire > time_uptime)
1173 				nd6_llinfo_settimer(ln,
1174 				    (dr->expire - time_uptime) * hz);
1175 			else
1176 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1177 			splx(s);
1178 			LLE_WUNLOCK(ln);
1179 			return;
1180 		}
1181 
1182 		if (ln->ln_router || dr) {
1183 			/*
1184 			 * We need to unlock to avoid a LOR with rt6_flush()
1185 			 * with the rnh and for the calls to
1186 			 * pfxlist_onlink_check() and defrouter_select() in the
1187 			 * block further down for calls into nd6_lookup().
1188 			 * We still hold a ref.
1189 			 */
1190 			LLE_WUNLOCK(ln);
1191 
1192 			/*
1193 			 * rt6_flush must be called whether or not the neighbor
1194 			 * is in the Default Router List.
1195 			 * See a corresponding comment in nd6_na_input().
1196 			 */
1197 			rt6_flush(in6, ifp);
1198 		}
1199 
1200 		if (dr) {
1201 			/*
1202 			 * Unreachablity of a router might affect the default
1203 			 * router selection and on-link detection of advertised
1204 			 * prefixes.
1205 			 */
1206 
1207 			/*
1208 			 * Temporarily fake the state to choose a new default
1209 			 * router and to perform on-link determination of
1210 			 * prefixes correctly.
1211 			 * Below the state will be set correctly,
1212 			 * or the entry itself will be deleted.
1213 			 */
1214 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1215 
1216 			/*
1217 			 * Since defrouter_select() does not affect the
1218 			 * on-link determination and MIP6 needs the check
1219 			 * before the default router selection, we perform
1220 			 * the check now.
1221 			 */
1222 			pfxlist_onlink_check();
1223 
1224 			/*
1225 			 * refresh default router list
1226 			 */
1227 			defrouter_select();
1228 		}
1229 
1230 #ifdef __FreeBSD__
1231 		/*
1232 		 * If this entry was added by an on-link redirect, remove the
1233 		 * corresponding host route.
1234 		 */
1235 		if (ln->la_flags & LLE_REDIRECT)
1236 			nd6_free_redirect(ln);
1237 #endif
1238 
1239 		if (ln->ln_router || dr)
1240 			LLE_WLOCK(ln);
1241 
1242 		splx(s);
1243 	}
1244 
1245 	/*
1246 	 * Save to unlock. We still hold an extra reference and will not
1247 	 * free(9) in llentry_free() if someone else holds one as well.
1248 	 */
1249 	LLE_WUNLOCK(ln);
1250 	IF_AFDATA_LOCK(ifp);
1251 	LLE_WLOCK(ln);
1252 
1253 	/* Guard against race with other llentry_free(). */
1254 	if (ln->la_flags & LLE_LINKED) {
1255 		LLE_REMREF(ln);
1256 		llentry_free(ln);
1257 	} else
1258 		LLE_FREE_LOCKED(ln);
1259 
1260 	IF_AFDATA_UNLOCK(ifp);
1261 }
1262 
1263 /*
1264  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1265  *
1266  * XXX cost-effective methods?
1267  */
1268 void
1269 nd6_nud_hint(struct rtentry *rt)
1270 {
1271 	struct llentry *ln;
1272 	struct ifnet *ifp;
1273 
1274 	if (rt == NULL)
1275 		return;
1276 
1277 	ifp = rt->rt_ifp;
1278 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1279 	if (ln == NULL)
1280 		return;
1281 
1282 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1283 		goto done;
1284 
1285 	/*
1286 	 * if we get upper-layer reachability confirmation many times,
1287 	 * it is possible we have false information.
1288 	 */
1289 	ln->ln_byhint++;
1290 	if (ln->ln_byhint > nd6_maxnudhint)
1291 		goto done;
1292 
1293 	ln->ln_state = ND6_LLINFO_REACHABLE;
1294 	if (!ND6_LLINFO_PERMANENT(ln))
1295 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1296 
1297 done:
1298 	LLE_WUNLOCK(ln);
1299 
1300 	return;
1301 }
1302 
1303 struct gc_args {
1304 	int gc_entries;
1305 	const struct in6_addr *skip_in6;
1306 };
1307 
1308 static int
1309 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1310 {
1311 	struct gc_args *args = farg;
1312 	int *n = &args->gc_entries;
1313 	const struct in6_addr *skip_in6 = args->skip_in6;
1314 
1315 	if (*n <= 0)
1316 		return 0;
1317 
1318 	if (ND6_LLINFO_PERMANENT(ln))
1319 		return 0;
1320 
1321 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1322 		return 0;
1323 
1324 	LLE_WLOCK(ln);
1325 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1326 		ln->ln_state = ND6_LLINFO_STALE;
1327 	else
1328 		ln->ln_state = ND6_LLINFO_PURGE;
1329 	nd6_llinfo_settimer(ln, 0);
1330 	LLE_WUNLOCK(ln);
1331 
1332 	(*n)--;
1333 	return 0;
1334 }
1335 
1336 static void
1337 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1338 {
1339 
1340 	if (ip6_neighborgcthresh >= 0 &&
1341 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1342 		struct gc_args gc_args = {10, in6};
1343 		/*
1344 		 * XXX entries that are "less recently used" should be
1345 		 * freed first.
1346 		 */
1347 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1348 	}
1349 }
1350 
1351 void
1352 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1353 {
1354 	struct sockaddr *gate = rt->rt_gateway;
1355 	struct ifnet *ifp = rt->rt_ifp;
1356 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1357 	struct ifaddr *ifa;
1358 
1359 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1360 
1361 	if (req == RTM_LLINFO_UPD) {
1362 		int rc;
1363 		struct in6_addr *in6;
1364 		struct in6_addr in6_all;
1365 		int anycast;
1366 
1367 		if ((ifa = info->rti_ifa) == NULL)
1368 			return;
1369 
1370 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1371 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1372 
1373 		in6_all = in6addr_linklocal_allnodes;
1374 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1375 			log(LOG_ERR, "%s: failed to set scope %s "
1376 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1377 			return;
1378 		}
1379 
1380 		/* XXX don't set Override for proxy addresses */
1381 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1382 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1383 #if 0
1384 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1385 #endif
1386 		    , 1, NULL);
1387 		return;
1388 	}
1389 
1390 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1391 		return;
1392 
1393 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1394 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1395 		/*
1396 		 * This is probably an interface direct route for a link
1397 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1398 		 * We do not need special treatment below for such a route.
1399 		 * Moreover, the RTF_LLINFO flag which would be set below
1400 		 * would annoy the ndp(8) command.
1401 		 */
1402 		return;
1403 	}
1404 
1405 	switch (req) {
1406 	case RTM_ADD: {
1407 		int s;
1408 
1409 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1410 		/*
1411 		 * There is no backward compatibility :)
1412 		 *
1413 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1414 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1415 		 *	   rt->rt_flags |= RTF_CLONING;
1416 		 */
1417 		/* XXX should move to route.c? */
1418 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1419 			union {
1420 				struct sockaddr sa;
1421 				struct sockaddr_dl sdl;
1422 				struct sockaddr_storage ss;
1423 			} u;
1424 			/*
1425 			 * Case 1: This route should come from a route to
1426 			 * interface (RTF_CLONING case) or the route should be
1427 			 * treated as on-link but is currently not
1428 			 * (RTF_LLINFO && ln == NULL case).
1429 			 */
1430 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1431 			    ifp->if_index, ifp->if_type,
1432 			    NULL, namelen, NULL, addrlen) == NULL) {
1433 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1434 				    "failed on %s\n", __func__, __LINE__,
1435 				    sizeof(u.ss), if_name(ifp));
1436 			}
1437 			rt_setgate(rt, &u.sa);
1438 			gate = rt->rt_gateway;
1439 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1440 			if (gate == NULL) {
1441 				log(LOG_ERR,
1442 				    "%s: rt_setgate failed on %s\n", __func__,
1443 				    if_name(ifp));
1444 				break;
1445 			}
1446 
1447 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1448 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1449 				break;
1450 		}
1451 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1452 		/*
1453 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1454 		 * We don't do that here since llinfo is not ready yet.
1455 		 *
1456 		 * There are also couple of other things to be discussed:
1457 		 * - unsolicited NA code needs improvement beforehand
1458 		 * - RFC2461 says we MAY send multicast unsolicited NA
1459 		 *   (7.2.6 paragraph 4), however, it also says that we
1460 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1461 		 *   we don't have anything like it right now.
1462 		 *   note that the mechanism needs a mutual agreement
1463 		 *   between proxies, which means that we need to implement
1464 		 *   a new protocol, or a new kludge.
1465 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1466 		 *   we need to check ip6forwarding before sending it.
1467 		 *   (or should we allow proxy ND configuration only for
1468 		 *   routers?  there's no mention about proxy ND from hosts)
1469 		 */
1470 #if 0
1471 		/* XXX it does not work */
1472 		if (rt->rt_flags & RTF_ANNOUNCE)
1473 			nd6_na_output(ifp,
1474 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1475 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1476 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1477 			      1, NULL);
1478 #endif
1479 
1480 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1481 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1482 			/*
1483 			 * Address resolution isn't necessary for a point to
1484 			 * point link, so we can skip this test for a p2p link.
1485 			 */
1486 			if (gate->sa_family != AF_LINK ||
1487 			    gate->sa_len <
1488 			    sockaddr_dl_measure(namelen, addrlen)) {
1489 				log(LOG_DEBUG,
1490 				    "nd6_rtrequest: bad gateway value: %s\n",
1491 				    if_name(ifp));
1492 				break;
1493 			}
1494 			satosdl(gate)->sdl_type = ifp->if_type;
1495 			satosdl(gate)->sdl_index = ifp->if_index;
1496 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1497 		}
1498 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1499 
1500 		/*
1501 		 * When called from rt_ifa_addlocal, we cannot depend on that
1502 		 * the address (rt_getkey(rt)) exits in the address list of the
1503 		 * interface. So check RTF_LOCAL instead.
1504 		 */
1505 		if (rt->rt_flags & RTF_LOCAL) {
1506 			if (nd6_useloopback)
1507 				rt->rt_ifp = lo0ifp;	/* XXX */
1508 			break;
1509 		}
1510 
1511 		/*
1512 		 * check if rt_getkey(rt) is an address assigned
1513 		 * to the interface.
1514 		 */
1515 		s = pserialize_read_enter();
1516 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1517 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1518 		if (ifa != NULL) {
1519 			if (nd6_useloopback) {
1520 				rt->rt_ifp = lo0ifp;	/* XXX */
1521 				/*
1522 				 * Make sure rt_ifa be equal to the ifaddr
1523 				 * corresponding to the address.
1524 				 * We need this because when we refer
1525 				 * rt_ifa->ia6_flags in ip6_input, we assume
1526 				 * that the rt_ifa points to the address instead
1527 				 * of the loopback address.
1528 				 */
1529 				if (ifa != rt->rt_ifa)
1530 					rt_replace_ifa(rt, ifa);
1531 			}
1532 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1533 			/* join solicited node multicast for proxy ND */
1534 			if (ifp->if_flags & IFF_MULTICAST) {
1535 				struct in6_addr llsol;
1536 				int error;
1537 
1538 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1539 				llsol.s6_addr32[0] = htonl(0xff020000);
1540 				llsol.s6_addr32[1] = 0;
1541 				llsol.s6_addr32[2] = htonl(1);
1542 				llsol.s6_addr8[12] = 0xff;
1543 				if (in6_setscope(&llsol, ifp, NULL))
1544 					goto out;
1545 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1546 					nd6log(LOG_ERR, "%s: failed to join "
1547 					    "%s (errno=%d)\n", if_name(ifp),
1548 					    ip6_sprintf(&llsol), error);
1549 				}
1550 			}
1551 		}
1552 	out:
1553 		pserialize_read_exit(s);
1554 		/*
1555 		 * If we have too many cache entries, initiate immediate
1556 		 * purging for some entries.
1557 		 */
1558 		if (rt->rt_ifp != NULL)
1559 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1560 		break;
1561 	    }
1562 
1563 	case RTM_DELETE:
1564 		/* leave from solicited node multicast for proxy ND */
1565 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1566 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1567 			struct in6_addr llsol;
1568 			struct in6_multi *in6m;
1569 
1570 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1571 			llsol.s6_addr32[0] = htonl(0xff020000);
1572 			llsol.s6_addr32[1] = 0;
1573 			llsol.s6_addr32[2] = htonl(1);
1574 			llsol.s6_addr8[12] = 0xff;
1575 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1576 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1577 				if (in6m)
1578 					in6_delmulti(in6m);
1579 			}
1580 		}
1581 		break;
1582 	}
1583 }
1584 
1585 int
1586 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1587 {
1588 	struct in6_drlist *drl = (struct in6_drlist *)data;
1589 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1590 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1591 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1592 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1593 	struct nd_defrouter *dr;
1594 	struct nd_prefix *pr;
1595 	int i = 0, error = 0;
1596 	int s;
1597 
1598 	switch (cmd) {
1599 	case SIOCGDRLST_IN6:
1600 		/*
1601 		 * obsolete API, use sysctl under net.inet6.icmp6
1602 		 */
1603 		memset(drl, 0, sizeof(*drl));
1604 		s = splsoftnet();
1605 		TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
1606 			if (i >= DRLSTSIZ)
1607 				break;
1608 			drl->defrouter[i].rtaddr = dr->rtaddr;
1609 			in6_clearscope(&drl->defrouter[i].rtaddr);
1610 
1611 			drl->defrouter[i].flags = dr->flags;
1612 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1613 			drl->defrouter[i].expire = dr->expire ?
1614 			    time_mono_to_wall(dr->expire) : 0;
1615 			drl->defrouter[i].if_index = dr->ifp->if_index;
1616 			i++;
1617 		}
1618 		splx(s);
1619 		break;
1620 	case SIOCGPRLST_IN6:
1621 		/*
1622 		 * obsolete API, use sysctl under net.inet6.icmp6
1623 		 *
1624 		 * XXX the structure in6_prlist was changed in backward-
1625 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1626 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1627 		 */
1628 		/*
1629 		 * XXX meaning of fields, especialy "raflags", is very
1630 		 * differnet between RA prefix list and RR/static prefix list.
1631 		 * how about separating ioctls into two?
1632 		 */
1633 		memset(oprl, 0, sizeof(*oprl));
1634 		s = splsoftnet();
1635 		LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1636 			struct nd_pfxrouter *pfr;
1637 			int j;
1638 
1639 			if (i >= PRLSTSIZ)
1640 				break;
1641 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1642 			oprl->prefix[i].raflags = pr->ndpr_raf;
1643 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1644 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1645 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1646 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1647 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1648 				oprl->prefix[i].expire = 0;
1649 			else {
1650 				time_t maxexpire;
1651 
1652 				/* XXX: we assume time_t is signed. */
1653 				maxexpire = (-1) &
1654 				    ~((time_t)1 <<
1655 				    ((sizeof(maxexpire) * 8) - 1));
1656 				if (pr->ndpr_vltime <
1657 				    maxexpire - pr->ndpr_lastupdate) {
1658 					time_t expire;
1659 					expire = pr->ndpr_lastupdate +
1660 					    pr->ndpr_vltime;
1661 					oprl->prefix[i].expire = expire ?
1662 					    time_mono_to_wall(expire) : 0;
1663 				} else
1664 					oprl->prefix[i].expire = maxexpire;
1665 			}
1666 
1667 			j = 0;
1668 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1669 				if (j < DRLSTSIZ) {
1670 #define RTRADDR oprl->prefix[i].advrtr[j]
1671 					RTRADDR = pfr->router->rtaddr;
1672 					in6_clearscope(&RTRADDR);
1673 #undef RTRADDR
1674 				}
1675 				j++;
1676 			}
1677 			oprl->prefix[i].advrtrs = j;
1678 			oprl->prefix[i].origin = PR_ORIG_RA;
1679 
1680 			i++;
1681 		}
1682 		splx(s);
1683 
1684 		break;
1685 	case OSIOCGIFINFO_IN6:
1686 #define ND	ndi->ndi
1687 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1688 		memset(&ND, 0, sizeof(ND));
1689 		ND.linkmtu = IN6_LINKMTU(ifp);
1690 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1691 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1692 		ND.reachable = ND_IFINFO(ifp)->reachable;
1693 		ND.retrans = ND_IFINFO(ifp)->retrans;
1694 		ND.flags = ND_IFINFO(ifp)->flags;
1695 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1696 		ND.chlim = ND_IFINFO(ifp)->chlim;
1697 		break;
1698 	case SIOCGIFINFO_IN6:
1699 		ND = *ND_IFINFO(ifp);
1700 		break;
1701 	case SIOCSIFINFO_IN6:
1702 		/*
1703 		 * used to change host variables from userland.
1704 		 * intented for a use on router to reflect RA configurations.
1705 		 */
1706 		/* 0 means 'unspecified' */
1707 		if (ND.linkmtu != 0) {
1708 			if (ND.linkmtu < IPV6_MMTU ||
1709 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1710 				error = EINVAL;
1711 				break;
1712 			}
1713 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1714 		}
1715 
1716 		if (ND.basereachable != 0) {
1717 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1718 
1719 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1720 			if (ND.basereachable != obasereachable)
1721 				ND_IFINFO(ifp)->reachable =
1722 				    ND_COMPUTE_RTIME(ND.basereachable);
1723 		}
1724 		if (ND.retrans != 0)
1725 			ND_IFINFO(ifp)->retrans = ND.retrans;
1726 		if (ND.chlim != 0)
1727 			ND_IFINFO(ifp)->chlim = ND.chlim;
1728 		/* FALLTHROUGH */
1729 	case SIOCSIFINFO_FLAGS:
1730 	{
1731 		struct ifaddr *ifa;
1732 		struct in6_ifaddr *ia;
1733 
1734 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1735 		    !(ND.flags & ND6_IFF_IFDISABLED))
1736 		{
1737 			/*
1738 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1739 			 * has a link-local address with IN6_IFF_DUPLICATED,
1740 			 * do not clear ND6_IFF_IFDISABLED.
1741 			 * See RFC 4862, section 5.4.5.
1742 			 */
1743 			int duplicated_linklocal = 0;
1744 
1745 			s = pserialize_read_enter();
1746 			IFADDR_READER_FOREACH(ifa, ifp) {
1747 				if (ifa->ifa_addr->sa_family != AF_INET6)
1748 					continue;
1749 				ia = (struct in6_ifaddr *)ifa;
1750 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1751 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1752 				{
1753 					duplicated_linklocal = 1;
1754 					break;
1755 				}
1756 			}
1757 			pserialize_read_exit(s);
1758 
1759 			if (duplicated_linklocal) {
1760 				ND.flags |= ND6_IFF_IFDISABLED;
1761 				log(LOG_ERR, "Cannot enable an interface"
1762 				    " with a link-local address marked"
1763 				    " duplicate.\n");
1764 			} else {
1765 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1766 				if (ifp->if_flags & IFF_UP)
1767 					in6_if_up(ifp);
1768 			}
1769 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1770 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1771 			int bound = curlwp_bind();
1772 			/* Mark all IPv6 addresses as tentative. */
1773 
1774 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1775 			s = pserialize_read_enter();
1776 			IFADDR_READER_FOREACH(ifa, ifp) {
1777 				struct psref psref;
1778 				if (ifa->ifa_addr->sa_family != AF_INET6)
1779 					continue;
1780 				ifa_acquire(ifa, &psref);
1781 				pserialize_read_exit(s);
1782 
1783 				nd6_dad_stop(ifa);
1784 
1785 				ia = (struct in6_ifaddr *)ifa;
1786 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1787 
1788 				s = pserialize_read_enter();
1789 				ifa_release(ifa, &psref);
1790 			}
1791 			pserialize_read_exit(s);
1792 			curlwp_bindx(bound);
1793 		}
1794 
1795 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1796 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1797 				/* auto_linklocal 0->1 transition */
1798 
1799 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1800 				in6_ifattach(ifp, NULL);
1801 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1802 			    ifp->if_flags & IFF_UP)
1803 			{
1804 				/*
1805 				 * When the IF already has
1806 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1807 				 * address is assigned, and IFF_UP, try to
1808 				 * assign one.
1809 				 */
1810 				int haslinklocal = 0;
1811 
1812 				s = pserialize_read_enter();
1813 				IFADDR_READER_FOREACH(ifa, ifp) {
1814 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1815 						continue;
1816 					ia = (struct in6_ifaddr *)ifa;
1817 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1818 						haslinklocal = 1;
1819 						break;
1820 					}
1821 				}
1822 				pserialize_read_exit(s);
1823 				if (!haslinklocal)
1824 					in6_ifattach(ifp, NULL);
1825 			}
1826 		}
1827 	}
1828 		ND_IFINFO(ifp)->flags = ND.flags;
1829 		break;
1830 #undef ND
1831 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1832 		/* sync kernel routing table with the default router list */
1833 		defrouter_reset();
1834 		defrouter_select();
1835 		break;
1836 	case SIOCSPFXFLUSH_IN6:
1837 	{
1838 		/* flush all the prefix advertised by routers */
1839 		struct nd_prefix *pfx, *next;
1840 
1841 		s = splsoftnet();
1842 		LIST_FOREACH_SAFE(pfx, &nd_prefix, ndpr_entry, next) {
1843 			struct in6_ifaddr *ia, *ia_next;
1844 			int _s;
1845 
1846 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1847 				continue; /* XXX */
1848 
1849 			/* do we really have to remove addresses as well? */
1850 		restart:
1851 			_s = pserialize_read_enter();
1852 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1853 			     ia = ia_next) {
1854 				/* ia might be removed.  keep the next ptr. */
1855 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1856 
1857 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1858 					continue;
1859 
1860 				if (ia->ia6_ndpr == pfx) {
1861 					pserialize_read_exit(_s);
1862 					/* XXX NOMPSAFE? */
1863 					in6_purgeaddr(&ia->ia_ifa);
1864 					goto restart;
1865 				}
1866 			}
1867 			pserialize_read_exit(_s);
1868 			prelist_remove(pfx);
1869 		}
1870 		splx(s);
1871 		break;
1872 	}
1873 	case SIOCSRTRFLUSH_IN6:
1874 	{
1875 		/* flush all the default routers */
1876 		struct nd_defrouter *drtr, *next;
1877 
1878 		s = splsoftnet();
1879 		defrouter_reset();
1880 		TAILQ_FOREACH_SAFE(drtr, &nd_defrouter, dr_entry, next) {
1881 			defrtrlist_del(drtr, NULL);
1882 		}
1883 		defrouter_select();
1884 		splx(s);
1885 		break;
1886 	}
1887 	case SIOCGNBRINFO_IN6:
1888 	{
1889 		struct llentry *ln;
1890 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1891 
1892 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1893 			return error;
1894 
1895 		ln = nd6_lookup(&nb_addr, ifp, false);
1896 		if (ln == NULL) {
1897 			error = EINVAL;
1898 			break;
1899 		}
1900 		nbi->state = ln->ln_state;
1901 		nbi->asked = ln->ln_asked;
1902 		nbi->isrouter = ln->ln_router;
1903 		nbi->expire = ln->ln_expire ?
1904 		    time_mono_to_wall(ln->ln_expire) : 0;
1905 		LLE_RUNLOCK(ln);
1906 
1907 		break;
1908 	}
1909 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1910 		ndif->ifindex = nd6_defifindex;
1911 		break;
1912 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1913 		return nd6_setdefaultiface(ndif->ifindex);
1914 	}
1915 	return error;
1916 }
1917 
1918 void
1919 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1920 {
1921 	struct mbuf *m_hold, *m_hold_next;
1922 	struct sockaddr_in6 sin6;
1923 
1924 	LLE_WLOCK_ASSERT(ln);
1925 
1926 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1927 
1928 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1929 
1930 	LLE_WUNLOCK(ln);
1931 	for (; m_hold != NULL; m_hold = m_hold_next) {
1932 		m_hold_next = m_hold->m_nextpkt;
1933 		m_hold->m_nextpkt = NULL;
1934 
1935 		/*
1936 		 * we assume ifp is not a p2p here, so
1937 		 * just set the 2nd argument as the
1938 		 * 1st one.
1939 		 */
1940 		nd6_output(ifp, ifp, m_hold, &sin6, NULL);
1941 	}
1942 	LLE_WLOCK(ln);
1943 }
1944 
1945 /*
1946  * Create neighbor cache entry and cache link-layer address,
1947  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1948  */
1949 void
1950 nd6_cache_lladdr(
1951     struct ifnet *ifp,
1952     struct in6_addr *from,
1953     char *lladdr,
1954     int lladdrlen,
1955     int type,	/* ICMP6 type */
1956     int code	/* type dependent information */
1957 )
1958 {
1959 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
1960 	struct llentry *ln = NULL;
1961 	int is_newentry;
1962 	int do_update;
1963 	int olladdr;
1964 	int llchange;
1965 	int newstate = 0;
1966 	uint16_t router = 0;
1967 
1968 	KASSERT(ifp != NULL);
1969 	KASSERT(from != NULL);
1970 
1971 	/* nothing must be updated for unspecified address */
1972 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1973 		return;
1974 
1975 	/*
1976 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1977 	 * the caller.
1978 	 *
1979 	 * XXX If the link does not have link-layer adderss, what should
1980 	 * we do? (ifp->if_addrlen == 0)
1981 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1982 	 * description on it in NS section (RFC 2461 7.2.3).
1983 	 */
1984 
1985 	ln = nd6_lookup(from, ifp, true);
1986 	if (ln == NULL) {
1987 #if 0
1988 		/* nothing must be done if there's no lladdr */
1989 		if (!lladdr || !lladdrlen)
1990 			return NULL;
1991 #endif
1992 
1993 		ln = nd6_create(from, ifp);
1994 		is_newentry = 1;
1995 	} else {
1996 		/* do nothing if static ndp is set */
1997 		if (ln->la_flags & LLE_STATIC) {
1998 			LLE_WUNLOCK(ln);
1999 			return;
2000 		}
2001 		is_newentry = 0;
2002 	}
2003 
2004 	if (ln == NULL)
2005 		return;
2006 
2007 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2008 	if (olladdr && lladdr) {
2009 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2010 	} else
2011 		llchange = 0;
2012 
2013 	/*
2014 	 * newentry olladdr  lladdr  llchange	(*=record)
2015 	 *	0	n	n	--	(1)
2016 	 *	0	y	n	--	(2)
2017 	 *	0	n	y	--	(3) * STALE
2018 	 *	0	y	y	n	(4) *
2019 	 *	0	y	y	y	(5) * STALE
2020 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2021 	 *	1	--	y	--	(7) * STALE
2022 	 */
2023 
2024 	if (lladdr) {		/* (3-5) and (7) */
2025 		/*
2026 		 * Record source link-layer address
2027 		 * XXX is it dependent to ifp->if_type?
2028 		 */
2029 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2030 		ln->la_flags |= LLE_VALID;
2031 	}
2032 
2033 	if (!is_newentry) {
2034 		if ((!olladdr && lladdr) ||		/* (3) */
2035 		    (olladdr && lladdr && llchange)) {	/* (5) */
2036 			do_update = 1;
2037 			newstate = ND6_LLINFO_STALE;
2038 		} else					/* (1-2,4) */
2039 			do_update = 0;
2040 	} else {
2041 		do_update = 1;
2042 		if (lladdr == NULL)			/* (6) */
2043 			newstate = ND6_LLINFO_NOSTATE;
2044 		else					/* (7) */
2045 			newstate = ND6_LLINFO_STALE;
2046 	}
2047 
2048 	if (do_update) {
2049 		/*
2050 		 * Update the state of the neighbor cache.
2051 		 */
2052 		ln->ln_state = newstate;
2053 
2054 		if (ln->ln_state == ND6_LLINFO_STALE) {
2055 			/*
2056 			 * XXX: since nd6_output() below will cause
2057 			 * state tansition to DELAY and reset the timer,
2058 			 * we must set the timer now, although it is actually
2059 			 * meaningless.
2060 			 */
2061 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2062 
2063 			nd6_llinfo_release_pkts(ln, ifp);
2064 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2065 			/* probe right away */
2066 			nd6_llinfo_settimer((void *)ln, 0);
2067 		}
2068 	}
2069 
2070 	/*
2071 	 * ICMP6 type dependent behavior.
2072 	 *
2073 	 * NS: clear IsRouter if new entry
2074 	 * RS: clear IsRouter
2075 	 * RA: set IsRouter if there's lladdr
2076 	 * redir: clear IsRouter if new entry
2077 	 *
2078 	 * RA case, (1):
2079 	 * The spec says that we must set IsRouter in the following cases:
2080 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2081 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2082 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2083 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2084 	 * neighbor cache, this is similar to (6).
2085 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2086 	 *
2087 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2088 	 *							D R
2089 	 *	0	n	n	--	(1)	c   ?     s
2090 	 *	0	y	n	--	(2)	c   s     s
2091 	 *	0	n	y	--	(3)	c   s     s
2092 	 *	0	y	y	n	(4)	c   s     s
2093 	 *	0	y	y	y	(5)	c   s     s
2094 	 *	1	--	n	--	(6) c	c 	c s
2095 	 *	1	--	y	--	(7) c	c   s	c s
2096 	 *
2097 	 *					(c=clear s=set)
2098 	 */
2099 	switch (type & 0xff) {
2100 	case ND_NEIGHBOR_SOLICIT:
2101 		/*
2102 		 * New entry must have is_router flag cleared.
2103 		 */
2104 		if (is_newentry)	/* (6-7) */
2105 			ln->ln_router = 0;
2106 		break;
2107 	case ND_REDIRECT:
2108 		/*
2109 		 * If the icmp is a redirect to a better router, always set the
2110 		 * is_router flag.  Otherwise, if the entry is newly created,
2111 		 * clear the flag.  [RFC 2461, sec 8.3]
2112 		 */
2113 		if (code == ND_REDIRECT_ROUTER)
2114 			ln->ln_router = 1;
2115 		else if (is_newentry) /* (6-7) */
2116 			ln->ln_router = 0;
2117 		break;
2118 	case ND_ROUTER_SOLICIT:
2119 		/*
2120 		 * is_router flag must always be cleared.
2121 		 */
2122 		ln->ln_router = 0;
2123 		break;
2124 	case ND_ROUTER_ADVERT:
2125 		/*
2126 		 * Mark an entry with lladdr as a router.
2127 		 */
2128 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2129 		    (is_newentry && lladdr)) {			/* (7) */
2130 			ln->ln_router = 1;
2131 		}
2132 		break;
2133 	}
2134 
2135 #if 0
2136 	/* XXX should we send rtmsg as it used to be? */
2137 	if (do_update)
2138 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2139 #endif
2140 
2141 	if (ln != NULL) {
2142 		router = ln->ln_router;
2143 		LLE_WUNLOCK(ln);
2144 	}
2145 
2146 	/*
2147 	 * If we have too many cache entries, initiate immediate
2148 	 * purging for some entries.
2149 	 */
2150 	if (is_newentry)
2151 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2152 
2153 	/*
2154 	 * When the link-layer address of a router changes, select the
2155 	 * best router again.  In particular, when the neighbor entry is newly
2156 	 * created, it might affect the selection policy.
2157 	 * Question: can we restrict the first condition to the "is_newentry"
2158 	 * case?
2159 	 * XXX: when we hear an RA from a new router with the link-layer
2160 	 * address option, defrouter_select() is called twice, since
2161 	 * defrtrlist_update called the function as well.  However, I believe
2162 	 * we can compromise the overhead, since it only happens the first
2163 	 * time.
2164 	 * XXX: although defrouter_select() should not have a bad effect
2165 	 * for those are not autoconfigured hosts, we explicitly avoid such
2166 	 * cases for safety.
2167 	 */
2168 	if (do_update && router && !ip6_forwarding &&
2169 	    nd6_accepts_rtadv(ndi))
2170 		defrouter_select();
2171 }
2172 
2173 static void
2174 nd6_slowtimo(void *ignored_arg)
2175 {
2176 	struct nd_ifinfo *nd6if;
2177 	struct ifnet *ifp;
2178 	int s;
2179 
2180 	mutex_enter(softnet_lock);
2181 	KERNEL_LOCK(1, NULL);
2182 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2183 	    nd6_slowtimo, NULL);
2184 
2185 	s = pserialize_read_enter();
2186 	IFNET_READER_FOREACH(ifp) {
2187 		nd6if = ND_IFINFO(ifp);
2188 		if (nd6if->basereachable && /* already initialized */
2189 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2190 			/*
2191 			 * Since reachable time rarely changes by router
2192 			 * advertisements, we SHOULD insure that a new random
2193 			 * value gets recomputed at least once every few hours.
2194 			 * (RFC 2461, 6.3.4)
2195 			 */
2196 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2197 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2198 		}
2199 	}
2200 	pserialize_read_exit(s);
2201 
2202 	KERNEL_UNLOCK_ONE(NULL);
2203 	mutex_exit(softnet_lock);
2204 }
2205 
2206 int
2207 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2208     const struct sockaddr_in6 *dst, struct rtentry *rt)
2209 {
2210 #define senderr(e) { error = (e); goto bad;}
2211 	struct llentry *ln = NULL;
2212 	int error = 0;
2213 	bool created = false;
2214 
2215 	if (rt != NULL) {
2216 		error = rt_check_reject_route(rt, ifp);
2217 		if (error != 0) {
2218 			m_freem(m);
2219 			return error;
2220 		}
2221 	}
2222 
2223 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2224 		goto sendpkt;
2225 
2226 	if (nd6_need_cache(ifp) == 0)
2227 		goto sendpkt;
2228 
2229 	if (rt != NULL && (rt->rt_flags & RTF_GATEWAY) != 0) {
2230 		struct sockaddr_in6 *gw6 = satosin6(rt->rt_gateway);
2231 		int s;
2232 
2233 		/* XXX remain the check to keep the original behavior. */
2234 		/*
2235 		 * We skip link-layer address resolution and NUD
2236 		 * if the gateway is not a neighbor from ND point
2237 		 * of view, regardless of the value of nd_ifinfo.flags.
2238 		 * The second condition is a bit tricky; we skip
2239 		 * if the gateway is our own address, which is
2240 		 * sometimes used to install a route to a p2p link.
2241 		 */
2242 		s = pserialize_read_enter();
2243 		if (!nd6_is_addr_neighbor(gw6, ifp) ||
2244 		    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
2245 			/*
2246 			 * We allow this kind of tricky route only
2247 			 * when the outgoing interface is p2p.
2248 			 * XXX: we may need a more generic rule here.
2249 			 */
2250 			if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
2251 				senderr(EHOSTUNREACH);
2252 
2253 			pserialize_read_exit(s);
2254 			goto sendpkt;
2255 		}
2256 		pserialize_read_exit(s);
2257 	}
2258 
2259 	/*
2260 	 * Address resolution or Neighbor Unreachability Detection
2261 	 * for the next hop.
2262 	 * At this point, the destination of the packet must be a unicast
2263 	 * or an anycast address(i.e. not a multicast).
2264 	 */
2265 
2266 	/* Look up the neighbor cache for the nexthop */
2267 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2268 	if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2269 		/*
2270 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2271 		 * the condition below is not very efficient.  But we believe
2272 		 * it is tolerable, because this should be a rare case.
2273 		 */
2274 		ln = nd6_create(&dst->sin6_addr, ifp);
2275 		if (ln != NULL)
2276 			created = true;
2277 	}
2278 
2279 	if (ln == NULL) {
2280 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2281 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2282 			log(LOG_DEBUG,
2283 			    "nd6_output: can't allocate llinfo for %s "
2284 			    "(ln=%p, rt=%p)\n",
2285 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2286 			senderr(EIO);	/* XXX: good error? */
2287 		}
2288 		goto sendpkt;	/* send anyway */
2289 	}
2290 
2291 	LLE_WLOCK_ASSERT(ln);
2292 
2293 	/* We don't have to do link-layer address resolution on a p2p link. */
2294 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2295 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2296 		ln->ln_state = ND6_LLINFO_STALE;
2297 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2298 	}
2299 
2300 	/*
2301 	 * The first time we send a packet to a neighbor whose entry is
2302 	 * STALE, we have to change the state to DELAY and a sets a timer to
2303 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2304 	 * neighbor unreachability detection on expiration.
2305 	 * (RFC 2461 7.3.3)
2306 	 */
2307 	if (ln->ln_state == ND6_LLINFO_STALE) {
2308 		ln->ln_asked = 0;
2309 		ln->ln_state = ND6_LLINFO_DELAY;
2310 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2311 	}
2312 
2313 	/*
2314 	 * If the neighbor cache entry has a state other than INCOMPLETE
2315 	 * (i.e. its link-layer address is already resolved), just
2316 	 * send the packet.
2317 	 */
2318 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2319 		goto sendpkt;
2320 
2321 	/*
2322 	 * There is a neighbor cache entry, but no ethernet address
2323 	 * response yet.  Append this latest packet to the end of the
2324 	 * packet queue in the mbuf, unless the number of the packet
2325 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2326 	 * the oldest packet in the queue will be removed.
2327 	 */
2328 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2329 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2330 	if (ln->ln_hold) {
2331 		struct mbuf *m_hold;
2332 		int i;
2333 
2334 		i = 0;
2335 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2336 			i++;
2337 			if (m_hold->m_nextpkt == NULL) {
2338 				m_hold->m_nextpkt = m;
2339 				break;
2340 			}
2341 		}
2342 		while (i >= nd6_maxqueuelen) {
2343 			m_hold = ln->ln_hold;
2344 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2345 			m_freem(m_hold);
2346 			i--;
2347 		}
2348 	} else {
2349 		ln->ln_hold = m;
2350 	}
2351 
2352 	/*
2353 	 * If there has been no NS for the neighbor after entering the
2354 	 * INCOMPLETE state, send the first solicitation.
2355 	 */
2356 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2357 		struct in6_addr src, *psrc;
2358 
2359 		ln->ln_asked++;
2360 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2361 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2362 		LLE_WUNLOCK(ln);
2363 		ln = NULL;
2364 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
2365 	} else {
2366 		/* We did the lookup so we need to do the unlock here. */
2367 		LLE_WUNLOCK(ln);
2368 	}
2369 
2370 	error = 0;
2371 	goto exit;
2372 
2373   sendpkt:
2374 	/* discard the packet if IPv6 operation is disabled on the interface */
2375 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2376 		error = ENETDOWN; /* better error? */
2377 		goto bad;
2378 	}
2379 
2380 	if (ln != NULL)
2381 		LLE_WUNLOCK(ln);
2382 
2383 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2384 		error = if_output_lock(ifp, origifp, m, sin6tocsa(dst), rt);
2385 	else
2386 		error = if_output_lock(ifp, ifp, m, sin6tocsa(dst), rt);
2387 	goto exit;
2388 
2389   bad:
2390 	if (m != NULL)
2391 		m_freem(m);
2392   exit:
2393 	if (created)
2394 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2395 
2396 	return error;
2397 #undef senderr
2398 }
2399 
2400 int
2401 nd6_need_cache(struct ifnet *ifp)
2402 {
2403 	/*
2404 	 * XXX: we currently do not make neighbor cache on any interface
2405 	 * other than ARCnet, Ethernet, FDDI and GIF.
2406 	 *
2407 	 * RFC2893 says:
2408 	 * - unidirectional tunnels needs no ND
2409 	 */
2410 	switch (ifp->if_type) {
2411 	case IFT_ARCNET:
2412 	case IFT_ETHER:
2413 	case IFT_FDDI:
2414 	case IFT_IEEE1394:
2415 	case IFT_CARP:
2416 	case IFT_GIF:		/* XXX need more cases? */
2417 	case IFT_PPP:
2418 	case IFT_TUNNEL:
2419 		return 1;
2420 	default:
2421 		return 0;
2422 	}
2423 }
2424 
2425 /*
2426  * Add pernament ND6 link-layer record for given
2427  * interface address.
2428  *
2429  * Very similar to IPv4 arp_ifinit(), but:
2430  * 1) IPv6 DAD is performed in different place
2431  * 2) It is called by IPv6 protocol stack in contrast to
2432  * arp_ifinit() which is typically called in SIOCSIFADDR
2433  * driver ioctl handler.
2434  *
2435  */
2436 int
2437 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2438 {
2439 	struct ifnet *ifp;
2440 	struct llentry *ln;
2441 
2442 	ifp = ia->ia_ifa.ifa_ifp;
2443 	if (nd6_need_cache(ifp) == 0)
2444 		return 0;
2445 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2446 	ia->ia_ifa.ifa_flags |= RTF_CONNECTED;
2447 
2448 	IF_AFDATA_WLOCK(ifp);
2449 	ln = lla_create(LLTABLE6(ifp), LLE_IFADDR | LLE_EXCLUSIVE,
2450 	    sin6tosa(&ia->ia_addr));
2451 	IF_AFDATA_WUNLOCK(ifp);
2452 	if (ln == NULL)
2453 		return ENOBUFS;
2454 
2455 	ln->la_expire = 0;  /* for IPv6 this means permanent */
2456 	ln->ln_state = ND6_LLINFO_REACHABLE;
2457 
2458 	LLE_WUNLOCK(ln);
2459 	return 0;
2460 }
2461 
2462 /*
2463  * Removes ALL lle records for interface address prefix.
2464  * XXXME: That's probably not we really want to do, we need
2465  * to remove address record only and keep other records
2466  * until we determine if given prefix is really going
2467  * to be removed.
2468  */
2469 void
2470 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2471 {
2472 	struct sockaddr_in6 mask, addr;
2473 
2474 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2475 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2476 	lltable_prefix_free(AF_INET6, sin6tosa(&addr), sin6tosa(&mask),
2477 	    LLE_STATIC);
2478 }
2479 
2480 int
2481 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2482     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2483     size_t dstsize)
2484 {
2485 	struct llentry *ln;
2486 
2487 	if (m->m_flags & M_MCAST) {
2488 		switch (ifp->if_type) {
2489 		case IFT_ETHER:
2490 		case IFT_FDDI:
2491 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2492 			    lldst);
2493 			return 1;
2494 		case IFT_IEEE1394:
2495 			memcpy(lldst, ifp->if_broadcastaddr,
2496 			    MIN(dstsize, ifp->if_addrlen));
2497 			return 1;
2498 		case IFT_ARCNET:
2499 			*lldst = 0;
2500 			return 1;
2501 		default:
2502 			m_freem(m);
2503 			return 0;
2504 		}
2505 	}
2506 
2507 	/*
2508 	 * the entry should have been created in nd6_store_lladdr
2509 	 */
2510 	ln = nd6_lookup(&satocsin6(dst)->sin6_addr, ifp, false);
2511 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2512 		if (ln != NULL)
2513 			LLE_RUNLOCK(ln);
2514 		/* this could happen, if we could not allocate memory */
2515 		m_freem(m);
2516 		return 0;
2517 	}
2518 
2519 	/* XXX llentry should have addrlen? */
2520 #if 0
2521 	sdl = satocsdl(rt->rt_gateway);
2522 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2523 		char sbuf[INET6_ADDRSTRLEN];
2524 		char dbuf[LINK_ADDRSTRLEN];
2525 		/* this should be impossible, but we bark here for debugging */
2526 		printf("%s: sdl_alen == %" PRIu8 ", if=%s, dst=%s, sdl=%s\n",
2527 		    __func__, sdl->sdl_alen, if_name(ifp),
2528 		    IN6_PRINT(sbuf, &satocsin6(dst)->sin6_addr),
2529 		    DL_PRINT(dbuf, &sdl->sdl_addr));
2530 		m_freem(m);
2531 		return 0;
2532 	}
2533 #endif
2534 
2535 	memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2536 
2537 	LLE_RUNLOCK(ln);
2538 
2539 	return 1;
2540 }
2541 
2542 static void
2543 clear_llinfo_pqueue(struct llentry *ln)
2544 {
2545 	struct mbuf *m_hold, *m_hold_next;
2546 
2547 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2548 		m_hold_next = m_hold->m_nextpkt;
2549 		m_hold->m_nextpkt = NULL;
2550 		m_freem(m_hold);
2551 	}
2552 
2553 	ln->ln_hold = NULL;
2554 	return;
2555 }
2556 
2557 int
2558 nd6_sysctl(
2559     int name,
2560     void *oldp,	/* syscall arg, need copyout */
2561     size_t *oldlenp,
2562     void *newp,	/* syscall arg, need copyin */
2563     size_t newlen
2564 )
2565 {
2566 	void *p;
2567 	size_t ol;
2568 	int error;
2569 
2570 	error = 0;
2571 
2572 	if (newp)
2573 		return EPERM;
2574 	if (oldp && !oldlenp)
2575 		return EINVAL;
2576 	ol = oldlenp ? *oldlenp : 0;
2577 
2578 	if (oldp) {
2579 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2580 		if (p == NULL)
2581 			return ENOMEM;
2582 	} else
2583 		p = NULL;
2584 	switch (name) {
2585 	case ICMPV6CTL_ND6_DRLIST:
2586 		error = fill_drlist(p, oldlenp, ol);
2587 		if (!error && p != NULL && oldp != NULL)
2588 			error = copyout(p, oldp, *oldlenp);
2589 		break;
2590 
2591 	case ICMPV6CTL_ND6_PRLIST:
2592 		error = fill_prlist(p, oldlenp, ol);
2593 		if (!error && p != NULL && oldp != NULL)
2594 			error = copyout(p, oldp, *oldlenp);
2595 		break;
2596 
2597 	case ICMPV6CTL_ND6_MAXQLEN:
2598 		break;
2599 
2600 	default:
2601 		error = ENOPROTOOPT;
2602 		break;
2603 	}
2604 	if (p)
2605 		free(p, M_TEMP);
2606 
2607 	return error;
2608 }
2609 
2610 static int
2611 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2612 {
2613 	int error = 0, s;
2614 	struct in6_defrouter *d = NULL, *de = NULL;
2615 	struct nd_defrouter *dr;
2616 	size_t l;
2617 
2618 	s = splsoftnet();
2619 
2620 	if (oldp) {
2621 		d = (struct in6_defrouter *)oldp;
2622 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2623 	}
2624 	l = 0;
2625 
2626 	TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
2627 
2628 		if (oldp && d + 1 <= de) {
2629 			memset(d, 0, sizeof(*d));
2630 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2631 			if (sa6_recoverscope(&d->rtaddr)) {
2632 				log(LOG_ERR,
2633 				    "scope error in router list (%s)\n",
2634 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2635 				/* XXX: press on... */
2636 			}
2637 			d->flags = dr->flags;
2638 			d->rtlifetime = dr->rtlifetime;
2639 			d->expire = dr->expire ?
2640 			    time_mono_to_wall(dr->expire) : 0;
2641 			d->if_index = dr->ifp->if_index;
2642 		}
2643 
2644 		l += sizeof(*d);
2645 		if (d)
2646 			d++;
2647 	}
2648 
2649 	if (oldp) {
2650 		if (l > ol)
2651 			error = ENOMEM;
2652 	}
2653 	if (oldlenp)
2654 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2655 
2656 	splx(s);
2657 
2658 	return error;
2659 }
2660 
2661 static int
2662 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2663 {
2664 	int error = 0, s;
2665 	struct nd_prefix *pr;
2666 	uint8_t *p = NULL, *ps = NULL;
2667 	uint8_t *pe = NULL;
2668 	size_t l;
2669 
2670 	s = splsoftnet();
2671 
2672 	if (oldp) {
2673 		ps = p = (uint8_t*)oldp;
2674 		pe = (uint8_t*)oldp + *oldlenp;
2675 	}
2676 	l = 0;
2677 
2678 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2679 		u_short advrtrs;
2680 		struct sockaddr_in6 sin6;
2681 		struct nd_pfxrouter *pfr;
2682 		struct in6_prefix pfx;
2683 
2684 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2685 		{
2686 			memset(&pfx, 0, sizeof(pfx));
2687 			ps = p;
2688 			pfx.prefix = pr->ndpr_prefix;
2689 
2690 			if (sa6_recoverscope(&pfx.prefix)) {
2691 				log(LOG_ERR,
2692 				    "scope error in prefix list (%s)\n",
2693 				    ip6_sprintf(&pfx.prefix.sin6_addr));
2694 				/* XXX: press on... */
2695 			}
2696 			pfx.raflags = pr->ndpr_raf;
2697 			pfx.prefixlen = pr->ndpr_plen;
2698 			pfx.vltime = pr->ndpr_vltime;
2699 			pfx.pltime = pr->ndpr_pltime;
2700 			pfx.if_index = pr->ndpr_ifp->if_index;
2701 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2702 				pfx.expire = 0;
2703 			else {
2704 				time_t maxexpire;
2705 
2706 				/* XXX: we assume time_t is signed. */
2707 				maxexpire = (-1) &
2708 				    ~((time_t)1 <<
2709 				    ((sizeof(maxexpire) * 8) - 1));
2710 				if (pr->ndpr_vltime <
2711 				    maxexpire - pr->ndpr_lastupdate) {
2712 					pfx.expire = pr->ndpr_lastupdate +
2713 						pr->ndpr_vltime;
2714 				} else
2715 					pfx.expire = maxexpire;
2716 			}
2717 			pfx.refcnt = pr->ndpr_refcnt;
2718 			pfx.flags = pr->ndpr_stateflags;
2719 			pfx.origin = PR_ORIG_RA;
2720 
2721 			p += sizeof(pfx); l += sizeof(pfx);
2722 
2723 			advrtrs = 0;
2724 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2725 				if (p + sizeof(sin6) > pe) {
2726 					advrtrs++;
2727 					continue;
2728 				}
2729 
2730 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2731 				    0, 0, 0);
2732 				if (sa6_recoverscope(&sin6)) {
2733 					log(LOG_ERR,
2734 					    "scope error in "
2735 					    "prefix list (%s)\n",
2736 					    ip6_sprintf(&pfr->router->rtaddr));
2737 				}
2738 				advrtrs++;
2739 				memcpy(p, &sin6, sizeof(sin6));
2740 				p += sizeof(sin6);
2741 				l += sizeof(sin6);
2742 			}
2743 			pfx.advrtrs = advrtrs;
2744 			memcpy(ps, &pfx, sizeof(pfx));
2745 		}
2746 		else {
2747 			l += sizeof(pfx);
2748 			advrtrs = 0;
2749 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2750 				advrtrs++;
2751 				l += sizeof(sin6);
2752 			}
2753 		}
2754 	}
2755 
2756 	if (oldp) {
2757 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2758 		if (l > ol)
2759 			error = ENOMEM;
2760 	} else
2761 		*oldlenp = l;
2762 
2763 	splx(s);
2764 
2765 	return error;
2766 }
2767