xref: /netbsd-src/sys/netinet6/nd6.c (revision 2e2322c9c07009df921d11b1268f8506affbb8ba)
1 /*	$NetBSD: nd6.c,v 1.217 2016/12/14 04:05:11 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 1.217 2016/12/14 04:05:11 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 static struct nd_opt_hdr *nd6_option(union nd_opts *);
120 
121 static callout_t nd6_slowtimo_ch;
122 static callout_t nd6_timer_ch;
123 static struct workqueue	*nd6_timer_wq;
124 static struct work	nd6_timer_wk;
125 
126 static int fill_drlist(void *, size_t *, size_t);
127 static int fill_prlist(void *, size_t *, size_t);
128 
129 static struct ifnet *nd6_defifp;
130 static int nd6_defifindex;
131 
132 static int nd6_setdefaultiface(int);
133 
134 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
135 
136 void
137 nd6_init(void)
138 {
139 	int error;
140 
141 	/* initialization of the default router list */
142 	ND_DEFROUTER_LIST_INIT();
143 
144 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
145 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
146 
147 	error = workqueue_create(&nd6_timer_wq, "nd6_timer",
148 	    nd6_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
149 	if (error)
150 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
151 
152 	/* start timer */
153 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
154 	    nd6_slowtimo, NULL);
155 	callout_reset(&nd6_timer_ch, hz, nd6_timer, NULL);
156 }
157 
158 struct nd_ifinfo *
159 nd6_ifattach(struct ifnet *ifp)
160 {
161 	struct nd_ifinfo *nd;
162 
163 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
164 
165 	nd->initialized = 1;
166 
167 	nd->chlim = IPV6_DEFHLIM;
168 	nd->basereachable = REACHABLE_TIME;
169 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
170 	nd->retrans = RETRANS_TIMER;
171 
172 	nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV;
173 
174 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
175 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
176 	 * because one of its members should. */
177 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
178 	    (ifp->if_flags & IFF_LOOPBACK))
179 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
180 
181 	/* A loopback interface does not need to accept RTADV.
182 	 * A bridge interface should not accept RTADV
183 	 * because one of its members should. */
184 	if (ip6_accept_rtadv &&
185 	    !(ifp->if_flags & IFF_LOOPBACK) &&
186 	    !(ifp->if_type != IFT_BRIDGE))
187 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
188 
189 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
190 	nd6_setmtu0(ifp, nd);
191 
192 	return nd;
193 }
194 
195 void
196 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
197 {
198 
199 	nd6_purge(ifp, ext);
200 	free(ext->nd_ifinfo, M_IP6NDP);
201 }
202 
203 void
204 nd6_setmtu(struct ifnet *ifp)
205 {
206 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
207 }
208 
209 void
210 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
211 {
212 	u_int32_t omaxmtu;
213 
214 	omaxmtu = ndi->maxmtu;
215 
216 	switch (ifp->if_type) {
217 	case IFT_ARCNET:
218 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
219 		break;
220 	case IFT_FDDI:
221 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
222 		break;
223 	default:
224 		ndi->maxmtu = ifp->if_mtu;
225 		break;
226 	}
227 
228 	/*
229 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
230 	 * undesirable situation.  We thus notify the operator of the change
231 	 * explicitly.  The check for omaxmtu is necessary to restrict the
232 	 * log to the case of changing the MTU, not initializing it.
233 	 */
234 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
235 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
236 		    " small for IPv6 which needs %lu\n",
237 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
238 		    IPV6_MMTU);
239 	}
240 
241 	if (ndi->maxmtu > in6_maxmtu)
242 		in6_setmaxmtu(); /* check all interfaces just in case */
243 }
244 
245 void
246 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
247 {
248 
249 	memset(ndopts, 0, sizeof(*ndopts));
250 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
251 	ndopts->nd_opts_last
252 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
253 
254 	if (icmp6len == 0) {
255 		ndopts->nd_opts_done = 1;
256 		ndopts->nd_opts_search = NULL;
257 	}
258 }
259 
260 /*
261  * Take one ND option.
262  */
263 static struct nd_opt_hdr *
264 nd6_option(union nd_opts *ndopts)
265 {
266 	struct nd_opt_hdr *nd_opt;
267 	int olen;
268 
269 	KASSERT(ndopts != NULL);
270 	KASSERT(ndopts->nd_opts_last != NULL);
271 
272 	if (ndopts->nd_opts_search == NULL)
273 		return NULL;
274 	if (ndopts->nd_opts_done)
275 		return NULL;
276 
277 	nd_opt = ndopts->nd_opts_search;
278 
279 	/* make sure nd_opt_len is inside the buffer */
280 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
281 		memset(ndopts, 0, sizeof(*ndopts));
282 		return NULL;
283 	}
284 
285 	olen = nd_opt->nd_opt_len << 3;
286 	if (olen == 0) {
287 		/*
288 		 * Message validation requires that all included
289 		 * options have a length that is greater than zero.
290 		 */
291 		memset(ndopts, 0, sizeof(*ndopts));
292 		return NULL;
293 	}
294 
295 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
296 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
297 		/* option overruns the end of buffer, invalid */
298 		memset(ndopts, 0, sizeof(*ndopts));
299 		return NULL;
300 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
301 		/* reached the end of options chain */
302 		ndopts->nd_opts_done = 1;
303 		ndopts->nd_opts_search = NULL;
304 	}
305 	return nd_opt;
306 }
307 
308 /*
309  * Parse multiple ND options.
310  * This function is much easier to use, for ND routines that do not need
311  * multiple options of the same type.
312  */
313 int
314 nd6_options(union nd_opts *ndopts)
315 {
316 	struct nd_opt_hdr *nd_opt;
317 	int i = 0;
318 
319 	KASSERT(ndopts != NULL);
320 	KASSERT(ndopts->nd_opts_last != NULL);
321 
322 	if (ndopts->nd_opts_search == NULL)
323 		return 0;
324 
325 	while (1) {
326 		nd_opt = nd6_option(ndopts);
327 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
328 			/*
329 			 * Message validation requires that all included
330 			 * options have a length that is greater than zero.
331 			 */
332 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
333 			memset(ndopts, 0, sizeof(*ndopts));
334 			return -1;
335 		}
336 
337 		if (nd_opt == NULL)
338 			goto skip1;
339 
340 		switch (nd_opt->nd_opt_type) {
341 		case ND_OPT_SOURCE_LINKADDR:
342 		case ND_OPT_TARGET_LINKADDR:
343 		case ND_OPT_MTU:
344 		case ND_OPT_REDIRECTED_HEADER:
345 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
346 				nd6log(LOG_INFO,
347 				    "duplicated ND6 option found (type=%d)\n",
348 				    nd_opt->nd_opt_type);
349 				/* XXX bark? */
350 			} else {
351 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
352 					= nd_opt;
353 			}
354 			break;
355 		case ND_OPT_PREFIX_INFORMATION:
356 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
357 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
358 					= nd_opt;
359 			}
360 			ndopts->nd_opts_pi_end =
361 				(struct nd_opt_prefix_info *)nd_opt;
362 			break;
363 		default:
364 			/*
365 			 * Unknown options must be silently ignored,
366 			 * to accommodate future extension to the protocol.
367 			 */
368 			nd6log(LOG_DEBUG,
369 			    "nd6_options: unsupported option %d - "
370 			    "option ignored\n", nd_opt->nd_opt_type);
371 		}
372 
373 skip1:
374 		i++;
375 		if (i > nd6_maxndopt) {
376 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
377 			nd6log(LOG_INFO, "too many loop in nd opt\n");
378 			break;
379 		}
380 
381 		if (ndopts->nd_opts_done)
382 			break;
383 	}
384 
385 	return 0;
386 }
387 
388 /*
389  * ND6 timer routine to handle ND6 entries
390  */
391 void
392 nd6_llinfo_settimer(struct llentry *ln, time_t xtick)
393 {
394 
395 	CTASSERT(sizeof(time_t) > sizeof(int));
396 	LLE_WLOCK_ASSERT(ln);
397 
398 	if (xtick < 0) {
399 		ln->ln_expire = 0;
400 		ln->ln_ntick = 0;
401 		callout_halt(&ln->ln_timer_ch, &ln->lle_lock);
402 	} else {
403 		ln->ln_expire = time_uptime + xtick / hz;
404 		LLE_ADDREF(ln);
405 		if (xtick > INT_MAX) {
406 			ln->ln_ntick = xtick - INT_MAX;
407 			callout_reset(&ln->ln_timer_ch, INT_MAX,
408 			    nd6_llinfo_timer, ln);
409 		} else {
410 			ln->ln_ntick = 0;
411 			callout_reset(&ln->ln_timer_ch, xtick,
412 			    nd6_llinfo_timer, ln);
413 		}
414 	}
415 }
416 
417 /*
418  * Gets source address of the first packet in hold queue
419  * and stores it in @src.
420  * Returns pointer to @src (if hold queue is not empty) or NULL.
421  */
422 static struct in6_addr *
423 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
424 {
425 	struct ip6_hdr *hip6;
426 
427 	if (ln == NULL || ln->ln_hold == NULL)
428 		return NULL;
429 
430 	/*
431 	 * assuming every packet in ln_hold has the same IP header
432 	 */
433 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
434 	/* XXX pullup? */
435 	if (sizeof(*hip6) < ln->ln_hold->m_len)
436 		*src = hip6->ip6_src;
437 	else
438 		src = NULL;
439 
440 	return src;
441 }
442 
443 static void
444 nd6_llinfo_timer(void *arg)
445 {
446 	struct llentry *ln = arg;
447 	struct ifnet *ifp;
448 	struct nd_ifinfo *ndi = NULL;
449 	bool send_ns = false;
450 	const struct in6_addr *daddr6 = NULL;
451 
452 #ifndef NET_MPSAFE
453 	mutex_enter(softnet_lock);
454 	KERNEL_LOCK(1, NULL);
455 #endif
456 
457 	LLE_WLOCK(ln);
458 	if (ln->ln_ntick > 0) {
459 		nd6_llinfo_settimer(ln, ln->ln_ntick);
460 		goto out;
461 	}
462 
463 	if (callout_pending(&ln->la_timer)) {
464 		/*
465 		 * Here we are a bit odd here in the treatment of
466 		 * active/pending. If the pending bit is set, it got
467 		 * rescheduled before I ran. The active
468 		 * bit we ignore, since if it was stopped
469 		 * in ll_tablefree() and was currently running
470 		 * it would have return 0 so the code would
471 		 * not have deleted it since the callout could
472 		 * not be stopped so we want to go through
473 		 * with the delete here now. If the callout
474 		 * was restarted, the pending bit will be back on and
475 		 * we just want to bail since the callout_reset would
476 		 * return 1 and our reference would have been removed
477 		 * by nd6_llinfo_settimer above since canceled
478 		 * would have been 1.
479 		 */
480 		goto out;
481 	}
482 
483 	ifp = ln->lle_tbl->llt_ifp;
484 
485 	KASSERT(ifp != NULL);
486 
487 	ndi = ND_IFINFO(ifp);
488 
489 	switch (ln->ln_state) {
490 	case ND6_LLINFO_INCOMPLETE:
491 		if (ln->ln_asked < nd6_mmaxtries) {
492 			ln->ln_asked++;
493 			send_ns = true;
494 		} else {
495 			struct mbuf *m = ln->ln_hold;
496 			if (m) {
497 				struct mbuf *m0;
498 
499 				/*
500 				 * assuming every packet in ln_hold has
501 				 * the same IP header
502 				 */
503 				m0 = m->m_nextpkt;
504 				m->m_nextpkt = NULL;
505 				ln->ln_hold = m0;
506 				clear_llinfo_pqueue(ln);
507  			}
508 			nd6_free(ln, 0);
509 			ln = NULL;
510 			if (m != NULL)
511 				icmp6_error2(m, ICMP6_DST_UNREACH,
512 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
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 #ifndef NET_MPSAFE
570 	KERNEL_UNLOCK_ONE(NULL);
571 	mutex_exit(softnet_lock);
572 #endif
573 }
574 
575 /*
576  * ND6 timer routine to expire default route list and prefix list
577  */
578 static void
579 nd6_timer_work(struct work *wk, void *arg)
580 {
581 	struct nd_defrouter *next_dr, *dr;
582 	struct nd_prefix *next_pr, *pr;
583 	struct in6_ifaddr *ia6, *nia6;
584 	int s, bound;
585 	struct psref psref;
586 
587 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
588 	    nd6_timer, NULL);
589 
590 #ifndef NET_MPSAFE
591 	mutex_enter(softnet_lock);
592 	KERNEL_LOCK(1, NULL);
593 #endif
594 
595 	/* expire default router list */
596 
597 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, next_dr) {
598 		if (dr->expire && dr->expire < time_uptime) {
599 			nd6_defrtrlist_del(dr, NULL);
600 		}
601 	}
602 
603 	/*
604 	 * expire interface addresses.
605 	 * in the past the loop was inside prefix expiry processing.
606 	 * However, from a stricter speci-confrmance standpoint, we should
607 	 * rather separate address lifetimes and prefix lifetimes.
608 	 */
609 	bound = curlwp_bind();
610   addrloop:
611 	s = pserialize_read_enter();
612 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
613 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
614 
615 		ia6_acquire(ia6, &psref);
616 		pserialize_read_exit(s);
617 
618 		/* check address lifetime */
619 		if (IFA6_IS_INVALID(ia6)) {
620 			int regen = 0;
621 
622 			/*
623 			 * If the expiring address is temporary, try
624 			 * regenerating a new one.  This would be useful when
625 			 * we suspended a laptop PC, then turned it on after a
626 			 * period that could invalidate all temporary
627 			 * addresses.  Although we may have to restart the
628 			 * loop (see below), it must be after purging the
629 			 * address.  Otherwise, we'd see an infinite loop of
630 			 * regeneration.
631 			 */
632 			if (ip6_use_tempaddr &&
633 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
634 				if (regen_tmpaddr(ia6) == 0)
635 					regen = 1;
636 			}
637 
638 			ia6_release(ia6, &psref);
639  			in6_purgeaddr(&ia6->ia_ifa);
640 			ia6 = NULL;
641 
642 			if (regen)
643 				goto addrloop; /* XXX: see below */
644 		} else if (IFA6_IS_DEPRECATED(ia6)) {
645 			int oldflags = ia6->ia6_flags;
646 
647 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
648 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
649 				rt_newaddrmsg(RTM_NEWADDR,
650 				    (struct ifaddr *)ia6, 0, NULL);
651 			}
652 
653 			/*
654 			 * If a temporary address has just become deprecated,
655 			 * regenerate a new one if possible.
656 			 */
657 			if (ip6_use_tempaddr &&
658 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
659 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
660 
661 				if (regen_tmpaddr(ia6) == 0) {
662 					/*
663 					 * A new temporary address is
664 					 * generated.
665 					 * XXX: this means the address chain
666 					 * has changed while we are still in
667 					 * the loop.  Although the change
668 					 * would not cause disaster (because
669 					 * it's not a deletion, but an
670 					 * addition,) we'd rather restart the
671 					 * loop just for safety.  Or does this
672 					 * significantly reduce performance??
673 					 */
674 					ia6_release(ia6, &psref);
675 					goto addrloop;
676 				}
677 			}
678 		} else {
679 			/*
680 			 * A new RA might have made a deprecated address
681 			 * preferred.
682 			 */
683 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
684 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
685 				rt_newaddrmsg(RTM_NEWADDR,
686 				    (struct ifaddr *)ia6, 0, NULL);
687 			}
688 		}
689 		s = pserialize_read_enter();
690 		ia6_release(ia6, &psref);
691 	}
692 	pserialize_read_exit(s);
693 	curlwp_bindx(bound);
694 
695 	/* expire prefix list */
696 	ND_PREFIX_LIST_FOREACH_SAFE(pr, next_pr) {
697 		/*
698 		 * check prefix lifetime.
699 		 * since pltime is just for autoconf, pltime processing for
700 		 * prefix is not necessary.
701 		 */
702 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
703 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
704 
705 			/*
706 			 * address expiration and prefix expiration are
707 			 * separate.  NEVER perform in6_purgeaddr here.
708 			 */
709 
710 			nd6_prelist_remove(pr);
711 		}
712 	}
713 
714 #ifndef NET_MPSAFE
715 	KERNEL_UNLOCK_ONE(NULL);
716 	mutex_exit(softnet_lock);
717 #endif
718 }
719 
720 static void
721 nd6_timer(void *ignored_arg)
722 {
723 
724 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
725 }
726 
727 /* ia6: deprecated/invalidated temporary address */
728 static int
729 regen_tmpaddr(const struct in6_ifaddr *ia6)
730 {
731 	struct ifaddr *ifa;
732 	struct ifnet *ifp;
733 	struct in6_ifaddr *public_ifa6 = NULL;
734 	int s;
735 
736 	ifp = ia6->ia_ifa.ifa_ifp;
737 	s = pserialize_read_enter();
738 	IFADDR_READER_FOREACH(ifa, ifp) {
739 		struct in6_ifaddr *it6;
740 
741 		if (ifa->ifa_addr->sa_family != AF_INET6)
742 			continue;
743 
744 		it6 = (struct in6_ifaddr *)ifa;
745 
746 		/* ignore no autoconf addresses. */
747 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
748 			continue;
749 
750 		/* ignore autoconf addresses with different prefixes. */
751 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
752 			continue;
753 
754 		/*
755 		 * Now we are looking at an autoconf address with the same
756 		 * prefix as ours.  If the address is temporary and is still
757 		 * preferred, do not create another one.  It would be rare, but
758 		 * could happen, for example, when we resume a laptop PC after
759 		 * a long period.
760 		 */
761 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
762 		    !IFA6_IS_DEPRECATED(it6)) {
763 			public_ifa6 = NULL;
764 			break;
765 		}
766 
767 		/*
768 		 * This is a public autoconf address that has the same prefix
769 		 * as ours.  If it is preferred, keep it.  We can't break the
770 		 * loop here, because there may be a still-preferred temporary
771 		 * address with the prefix.
772 		 */
773 		if (!IFA6_IS_DEPRECATED(it6))
774 			public_ifa6 = it6;
775 	}
776 
777 	if (public_ifa6 != NULL) {
778 		int e;
779 		struct psref psref;
780 
781 		ia6_acquire(public_ifa6, &psref);
782 		pserialize_read_exit(s);
783 		/*
784 		 * Random factor is introduced in the preferred lifetime, so
785 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
786 		 */
787 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
788 			ia6_release(public_ifa6, &psref);
789 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
790 			    " tmp addr, errno=%d\n", e);
791 			return -1;
792 		}
793 		ia6_release(public_ifa6, &psref);
794 		return 0;
795 	}
796 	pserialize_read_exit(s);
797 
798 	return -1;
799 }
800 
801 bool
802 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
803 {
804 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
805 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
806 		return true;
807 	case ND6_IFF_ACCEPT_RTADV:
808 		return ip6_accept_rtadv != 0;
809 	case ND6_IFF_OVERRIDE_RTADV:
810 	case 0:
811 	default:
812 		return false;
813 	}
814 }
815 
816 /*
817  * Nuke neighbor cache/prefix/default router management table, right before
818  * ifp goes away.
819  */
820 void
821 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
822 {
823 	struct nd_defrouter *dr, *ndr;
824 	struct nd_prefix *pr, *npr;
825 
826 	/*
827 	 * During detach, the ND info might be already removed, but
828 	 * then is explitly passed as argument.
829 	 * Otherwise get it from ifp->if_afdata.
830 	 */
831 	if (ext == NULL)
832 		ext = ifp->if_afdata[AF_INET6];
833 	if (ext == NULL)
834 		return;
835 
836 	/*
837 	 * Nuke default router list entries toward ifp.
838 	 * We defer removal of default router list entries that is installed
839 	 * in the routing table, in order to keep additional side effects as
840 	 * small as possible.
841 	 */
842 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
843 		if (dr->installed)
844 			continue;
845 
846 		if (dr->ifp == ifp) {
847 			KASSERT(ext != NULL);
848 			nd6_defrtrlist_del(dr, ext);
849 		}
850 	}
851 
852 	ND_DEFROUTER_LIST_FOREACH_SAFE(dr, ndr) {
853 		if (!dr->installed)
854 			continue;
855 
856 		if (dr->ifp == ifp) {
857 			KASSERT(ext != NULL);
858 			nd6_defrtrlist_del(dr, ext);
859 		}
860 	}
861 
862 	/* Nuke prefix list entries toward ifp */
863 	ND_PREFIX_LIST_FOREACH_SAFE(pr, npr) {
864 		if (pr->ndpr_ifp == ifp) {
865 			/*
866 			 * Because if_detach() does *not* release prefixes
867 			 * while purging addresses the reference count will
868 			 * still be above zero. We therefore reset it to
869 			 * make sure that the prefix really gets purged.
870 			 */
871 			pr->ndpr_refcnt = 0;
872 			/*
873 			 * Previously, pr->ndpr_addr is removed as well,
874 			 * but I strongly believe we don't have to do it.
875 			 * nd6_purge() is only called from in6_ifdetach(),
876 			 * which removes all the associated interface addresses
877 			 * by itself.
878 			 * (jinmei@kame.net 20010129)
879 			 */
880 			nd6_prelist_remove(pr);
881 		}
882 	}
883 
884 	/* cancel default outgoing interface setting */
885 	if (nd6_defifindex == ifp->if_index)
886 		nd6_setdefaultiface(0);
887 
888 	/* XXX: too restrictive? */
889 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
890 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
891 		if (ndi && nd6_accepts_rtadv(ndi)) {
892 			/* refresh default router list */
893 			nd6_defrouter_select();
894 		}
895 	}
896 
897 	/*
898 	 * We may not need to nuke the neighbor cache entries here
899 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
900 	 * nd6_purge() is invoked by in6_ifdetach() which is called
901 	 * from if_detach() where everything gets purged. However
902 	 * in6_ifdetach is directly called from vlan(4), so we still
903 	 * need to purge entries here.
904 	 */
905 	if (ext->lltable != NULL)
906 		lltable_purge_entries(ext->lltable);
907 }
908 
909 struct llentry *
910 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
911 {
912 	struct sockaddr_in6 sin6;
913 	struct llentry *ln;
914 
915 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
916 
917 	IF_AFDATA_RLOCK(ifp);
918 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
919 	    sin6tosa(&sin6));
920 	IF_AFDATA_RUNLOCK(ifp);
921 
922 	return ln;
923 }
924 
925 struct llentry *
926 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
927 {
928 	struct sockaddr_in6 sin6;
929 	struct llentry *ln;
930 
931 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
932 
933 	IF_AFDATA_WLOCK(ifp);
934 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE,
935 	    sin6tosa(&sin6));
936 	IF_AFDATA_WUNLOCK(ifp);
937 
938 	if (ln != NULL)
939 		ln->ln_state = ND6_LLINFO_NOSTATE;
940 
941 	return ln;
942 }
943 
944 /*
945  * Test whether a given IPv6 address is a neighbor or not, ignoring
946  * the actual neighbor cache.  The neighbor cache is ignored in order
947  * to not reenter the routing code from within itself.
948  */
949 static int
950 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
951 {
952 	struct nd_prefix *pr;
953 	struct ifaddr *dstaddr;
954 	int s;
955 
956 	/*
957 	 * A link-local address is always a neighbor.
958 	 * XXX: a link does not necessarily specify a single interface.
959 	 */
960 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
961 		struct sockaddr_in6 sin6_copy;
962 		u_int32_t zone;
963 
964 		/*
965 		 * We need sin6_copy since sa6_recoverscope() may modify the
966 		 * content (XXX).
967 		 */
968 		sin6_copy = *addr;
969 		if (sa6_recoverscope(&sin6_copy))
970 			return 0; /* XXX: should be impossible */
971 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
972 			return 0;
973 		if (sin6_copy.sin6_scope_id == zone)
974 			return 1;
975 		else
976 			return 0;
977 	}
978 
979 	/*
980 	 * If the address matches one of our addresses,
981 	 * it should be a neighbor.
982 	 * If the address matches one of our on-link prefixes, it should be a
983 	 * neighbor.
984 	 */
985 	ND_PREFIX_LIST_FOREACH(pr) {
986 		if (pr->ndpr_ifp != ifp)
987 			continue;
988 
989 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
990 			struct rtentry *rt;
991 
992 			rt = rtalloc1(sin6tosa(&pr->ndpr_prefix), 0);
993 			if (rt == NULL)
994 				continue;
995 			/*
996 			 * This is the case where multiple interfaces
997 			 * have the same prefix, but only one is installed
998 			 * into the routing table and that prefix entry
999 			 * is not the one being examined here. In the case
1000 			 * where RADIX_MPATH is enabled, multiple route
1001 			 * entries (of the same rt_key value) will be
1002 			 * installed because the interface addresses all
1003 			 * differ.
1004 			 */
1005 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1006 			    &satocsin6(rt_getkey(rt))->sin6_addr)) {
1007 				rt_unref(rt);
1008 				continue;
1009 			}
1010 			rt_unref(rt);
1011 		}
1012 
1013 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1014 		    &addr->sin6_addr, &pr->ndpr_mask))
1015 			return 1;
1016 	}
1017 
1018 	/*
1019 	 * If the address is assigned on the node of the other side of
1020 	 * a p2p interface, the address should be a neighbor.
1021 	 */
1022 	s = pserialize_read_enter();
1023 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
1024 	if (dstaddr != NULL) {
1025 		if (dstaddr->ifa_ifp == ifp) {
1026 			pserialize_read_exit(s);
1027 			return 1;
1028 		}
1029 	}
1030 	pserialize_read_exit(s);
1031 
1032 	/*
1033 	 * If the default router list is empty, all addresses are regarded
1034 	 * as on-link, and thus, as a neighbor.
1035 	 */
1036 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1037 	    ND_DEFROUTER_LIST_EMPTY() &&
1038 	    nd6_defifindex == ifp->if_index) {
1039 		return 1;
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 /*
1046  * Detect if a given IPv6 address identifies a neighbor on a given link.
1047  * XXX: should take care of the destination of a p2p link?
1048  */
1049 int
1050 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1051 {
1052 	struct nd_prefix *pr;
1053 	struct llentry *ln;
1054 	struct rtentry *rt;
1055 
1056 	/*
1057 	 * A link-local address is always a neighbor.
1058 	 * XXX: a link does not necessarily specify a single interface.
1059 	 */
1060 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1061 		struct sockaddr_in6 sin6_copy;
1062 		u_int32_t zone;
1063 
1064 		/*
1065 		 * We need sin6_copy since sa6_recoverscope() may modify the
1066 		 * content (XXX).
1067 		 */
1068 		sin6_copy = *addr;
1069 		if (sa6_recoverscope(&sin6_copy))
1070 			return 0; /* XXX: should be impossible */
1071 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1072 			return 0;
1073 		if (sin6_copy.sin6_scope_id == zone)
1074 			return 1;
1075 		else
1076 			return 0;
1077 	}
1078 
1079 	/*
1080 	 * If the address matches one of our on-link prefixes, it should be a
1081 	 * neighbor.
1082 	 */
1083 	ND_PREFIX_LIST_FOREACH(pr) {
1084 		if (pr->ndpr_ifp != ifp)
1085 			continue;
1086 
1087 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1088 			continue;
1089 
1090 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1091 		    &addr->sin6_addr, &pr->ndpr_mask))
1092 			return 1;
1093 	}
1094 
1095 	/*
1096 	 * If the default router list is empty, all addresses are regarded
1097 	 * as on-link, and thus, as a neighbor.
1098 	 * XXX: we restrict the condition to hosts, because routers usually do
1099 	 * not have the "default router list".
1100 	 */
1101 	if (!ip6_forwarding && ND_DEFROUTER_LIST_EMPTY() &&
1102 	    nd6_defifindex == ifp->if_index) {
1103 		return 1;
1104 	}
1105 
1106 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1107 	if (nd6_is_new_addr_neighbor(addr, ifp))
1108 		return 1;
1109 
1110 	/*
1111 	 * Even if the address matches none of our addresses, it might be
1112 	 * in the neighbor cache or a connected route.
1113 	 */
1114 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
1115 	if (ln != NULL) {
1116 		LLE_RUNLOCK(ln);
1117 		return 1;
1118 	}
1119 
1120 	rt = rtalloc1(sin6tocsa(addr), 0);
1121 	if (rt == NULL)
1122 		return 0;
1123 
1124 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
1125 #if NBRIDGE > 0
1126 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
1127 #endif
1128 #if NCARP > 0
1129 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
1130 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
1131 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
1132 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
1133 #endif
1134 	    )) {
1135 		rt_unref(rt);
1136 		return 1;
1137 	}
1138 	rt_unref(rt);
1139 
1140 	return 0;
1141 }
1142 
1143 /*
1144  * Free an nd6 llinfo entry.
1145  * Since the function would cause significant changes in the kernel, DO NOT
1146  * make it global, unless you have a strong reason for the change, and are sure
1147  * that the change is safe.
1148  */
1149 static void
1150 nd6_free(struct llentry *ln, int gc)
1151 {
1152 	struct nd_defrouter *dr;
1153 	struct ifnet *ifp;
1154 	struct in6_addr *in6;
1155 
1156 	KASSERT(ln != NULL);
1157 	LLE_WLOCK_ASSERT(ln);
1158 
1159 	ifp = ln->lle_tbl->llt_ifp;
1160 	in6 = &ln->r_l3addr.addr6;
1161 	/*
1162 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1163 	 * even though it is not harmful, it was not really necessary.
1164 	 */
1165 
1166 	/* cancel timer */
1167 	nd6_llinfo_settimer(ln, -1);
1168 
1169 	if (!ip6_forwarding) {
1170 		int s;
1171 		s = splsoftnet();
1172 		dr = nd6_defrouter_lookup(in6, ifp);
1173 
1174 		if (dr != NULL && dr->expire &&
1175 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1176 			/*
1177 			 * If the reason for the deletion is just garbage
1178 			 * collection, and the neighbor is an active default
1179 			 * router, do not delete it.  Instead, reset the GC
1180 			 * timer using the router's lifetime.
1181 			 * Simply deleting the entry would affect default
1182 			 * router selection, which is not necessarily a good
1183 			 * thing, especially when we're using router preference
1184 			 * values.
1185 			 * XXX: the check for ln_state would be redundant,
1186 			 *      but we intentionally keep it just in case.
1187 			 */
1188 			if (dr->expire > time_uptime)
1189 				nd6_llinfo_settimer(ln,
1190 				    (dr->expire - time_uptime) * hz);
1191 			else
1192 				nd6_llinfo_settimer(ln, nd6_gctimer * hz);
1193 			splx(s);
1194 			LLE_WUNLOCK(ln);
1195 			return;
1196 		}
1197 
1198 		if (ln->ln_router || dr) {
1199 			/*
1200 			 * We need to unlock to avoid a LOR with nd6_rt_flush()
1201 			 * with the rnh and for the calls to
1202 			 * nd6_pfxlist_onlink_check() and nd6_defrouter_select() in the
1203 			 * block further down for calls into nd6_lookup().
1204 			 * We still hold a ref.
1205 			 */
1206 			LLE_WUNLOCK(ln);
1207 
1208 			/*
1209 			 * nd6_rt_flush must be called whether or not the neighbor
1210 			 * is in the Default Router List.
1211 			 * See a corresponding comment in nd6_na_input().
1212 			 */
1213 			nd6_rt_flush(in6, ifp);
1214 		}
1215 
1216 		if (dr) {
1217 			/*
1218 			 * Unreachablity of a router might affect the default
1219 			 * router selection and on-link detection of advertised
1220 			 * prefixes.
1221 			 */
1222 
1223 			/*
1224 			 * Temporarily fake the state to choose a new default
1225 			 * router and to perform on-link determination of
1226 			 * prefixes correctly.
1227 			 * Below the state will be set correctly,
1228 			 * or the entry itself will be deleted.
1229 			 */
1230 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1231 
1232 			/*
1233 			 * Since nd6_defrouter_select() does not affect the
1234 			 * on-link determination and MIP6 needs the check
1235 			 * before the default router selection, we perform
1236 			 * the check now.
1237 			 */
1238 			nd6_pfxlist_onlink_check();
1239 
1240 			/*
1241 			 * refresh default router list
1242 			 */
1243 			nd6_defrouter_select();
1244 		}
1245 
1246 #ifdef __FreeBSD__
1247 		/*
1248 		 * If this entry was added by an on-link redirect, remove the
1249 		 * corresponding host route.
1250 		 */
1251 		if (ln->la_flags & LLE_REDIRECT)
1252 			nd6_free_redirect(ln);
1253 #endif
1254 
1255 		if (ln->ln_router || dr)
1256 			LLE_WLOCK(ln);
1257 
1258 		splx(s);
1259 	}
1260 
1261 	/*
1262 	 * Save to unlock. We still hold an extra reference and will not
1263 	 * free(9) in llentry_free() if someone else holds one as well.
1264 	 */
1265 	LLE_WUNLOCK(ln);
1266 	IF_AFDATA_LOCK(ifp);
1267 	LLE_WLOCK(ln);
1268 
1269 	/* Guard against race with other llentry_free(). */
1270 	if (ln->la_flags & LLE_LINKED) {
1271 		LLE_REMREF(ln);
1272 		llentry_free(ln);
1273 	} else
1274 		LLE_FREE_LOCKED(ln);
1275 
1276 	IF_AFDATA_UNLOCK(ifp);
1277 }
1278 
1279 /*
1280  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1281  *
1282  * XXX cost-effective methods?
1283  */
1284 void
1285 nd6_nud_hint(struct rtentry *rt)
1286 {
1287 	struct llentry *ln;
1288 	struct ifnet *ifp;
1289 
1290 	if (rt == NULL)
1291 		return;
1292 
1293 	ifp = rt->rt_ifp;
1294 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
1295 	if (ln == NULL)
1296 		return;
1297 
1298 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1299 		goto done;
1300 
1301 	/*
1302 	 * if we get upper-layer reachability confirmation many times,
1303 	 * it is possible we have false information.
1304 	 */
1305 	ln->ln_byhint++;
1306 	if (ln->ln_byhint > nd6_maxnudhint)
1307 		goto done;
1308 
1309 	ln->ln_state = ND6_LLINFO_REACHABLE;
1310 	if (!ND6_LLINFO_PERMANENT(ln))
1311 		nd6_llinfo_settimer(ln, ND_IFINFO(rt->rt_ifp)->reachable * hz);
1312 
1313 done:
1314 	LLE_WUNLOCK(ln);
1315 
1316 	return;
1317 }
1318 
1319 struct gc_args {
1320 	int gc_entries;
1321 	const struct in6_addr *skip_in6;
1322 };
1323 
1324 static int
1325 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1326 {
1327 	struct gc_args *args = farg;
1328 	int *n = &args->gc_entries;
1329 	const struct in6_addr *skip_in6 = args->skip_in6;
1330 
1331 	if (*n <= 0)
1332 		return 0;
1333 
1334 	if (ND6_LLINFO_PERMANENT(ln))
1335 		return 0;
1336 
1337 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
1338 		return 0;
1339 
1340 	LLE_WLOCK(ln);
1341 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1342 		ln->ln_state = ND6_LLINFO_STALE;
1343 	else
1344 		ln->ln_state = ND6_LLINFO_PURGE;
1345 	nd6_llinfo_settimer(ln, 0);
1346 	LLE_WUNLOCK(ln);
1347 
1348 	(*n)--;
1349 	return 0;
1350 }
1351 
1352 static void
1353 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
1354 {
1355 
1356 	if (ip6_neighborgcthresh >= 0 &&
1357 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1358 		struct gc_args gc_args = {10, in6};
1359 		/*
1360 		 * XXX entries that are "less recently used" should be
1361 		 * freed first.
1362 		 */
1363 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
1364 	}
1365 }
1366 
1367 void
1368 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1369 {
1370 	struct sockaddr *gate = rt->rt_gateway;
1371 	struct ifnet *ifp = rt->rt_ifp;
1372 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1373 	struct ifaddr *ifa;
1374 
1375 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1376 
1377 	if (req == RTM_LLINFO_UPD) {
1378 		int rc;
1379 		struct in6_addr *in6;
1380 		struct in6_addr in6_all;
1381 		int anycast;
1382 
1383 		if ((ifa = info->rti_ifa) == NULL)
1384 			return;
1385 
1386 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1387 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1388 
1389 		in6_all = in6addr_linklocal_allnodes;
1390 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1391 			log(LOG_ERR, "%s: failed to set scope %s "
1392 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1393 			return;
1394 		}
1395 
1396 		/* XXX don't set Override for proxy addresses */
1397 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1398 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1399 #if 0
1400 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1401 #endif
1402 		    , 1, NULL);
1403 		return;
1404 	}
1405 
1406 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1407 		return;
1408 
1409 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1410 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1411 		/*
1412 		 * This is probably an interface direct route for a link
1413 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1414 		 * We do not need special treatment below for such a route.
1415 		 * Moreover, the RTF_LLINFO flag which would be set below
1416 		 * would annoy the ndp(8) command.
1417 		 */
1418 		return;
1419 	}
1420 
1421 	switch (req) {
1422 	case RTM_ADD: {
1423 		int s;
1424 
1425 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1426 		/*
1427 		 * There is no backward compatibility :)
1428 		 *
1429 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1430 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1431 		 *	   rt->rt_flags |= RTF_CLONING;
1432 		 */
1433 		/* XXX should move to route.c? */
1434 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
1435 			union {
1436 				struct sockaddr sa;
1437 				struct sockaddr_dl sdl;
1438 				struct sockaddr_storage ss;
1439 			} u;
1440 			/*
1441 			 * Case 1: This route should come from a route to
1442 			 * interface (RTF_CLONING case) or the route should be
1443 			 * treated as on-link but is currently not
1444 			 * (RTF_LLINFO && ln == NULL case).
1445 			 */
1446 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1447 			    ifp->if_index, ifp->if_type,
1448 			    NULL, namelen, NULL, addrlen) == NULL) {
1449 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1450 				    "failed on %s\n", __func__, __LINE__,
1451 				    sizeof(u.ss), if_name(ifp));
1452 			}
1453 			rt_setgate(rt, &u.sa);
1454 			gate = rt->rt_gateway;
1455 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1456 			if (gate == NULL) {
1457 				log(LOG_ERR,
1458 				    "%s: rt_setgate failed on %s\n", __func__,
1459 				    if_name(ifp));
1460 				break;
1461 			}
1462 
1463 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1464 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
1465 				break;
1466 		}
1467 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1468 		/*
1469 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1470 		 * We don't do that here since llinfo is not ready yet.
1471 		 *
1472 		 * There are also couple of other things to be discussed:
1473 		 * - unsolicited NA code needs improvement beforehand
1474 		 * - RFC2461 says we MAY send multicast unsolicited NA
1475 		 *   (7.2.6 paragraph 4), however, it also says that we
1476 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1477 		 *   we don't have anything like it right now.
1478 		 *   note that the mechanism needs a mutual agreement
1479 		 *   between proxies, which means that we need to implement
1480 		 *   a new protocol, or a new kludge.
1481 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1482 		 *   we need to check ip6forwarding before sending it.
1483 		 *   (or should we allow proxy ND configuration only for
1484 		 *   routers?  there's no mention about proxy ND from hosts)
1485 		 */
1486 #if 0
1487 		/* XXX it does not work */
1488 		if (rt->rt_flags & RTF_ANNOUNCE)
1489 			nd6_na_output(ifp,
1490 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1491 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1492 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1493 			      1, NULL);
1494 #endif
1495 
1496 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1497 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1498 			/*
1499 			 * Address resolution isn't necessary for a point to
1500 			 * point link, so we can skip this test for a p2p link.
1501 			 */
1502 			if (gate->sa_family != AF_LINK ||
1503 			    gate->sa_len <
1504 			    sockaddr_dl_measure(namelen, addrlen)) {
1505 				log(LOG_DEBUG,
1506 				    "nd6_rtrequest: bad gateway value: %s\n",
1507 				    if_name(ifp));
1508 				break;
1509 			}
1510 			satosdl(gate)->sdl_type = ifp->if_type;
1511 			satosdl(gate)->sdl_index = ifp->if_index;
1512 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1513 		}
1514 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1515 
1516 		/*
1517 		 * When called from rt_ifa_addlocal, we cannot depend on that
1518 		 * the address (rt_getkey(rt)) exits in the address list of the
1519 		 * interface. So check RTF_LOCAL instead.
1520 		 */
1521 		if (rt->rt_flags & RTF_LOCAL) {
1522 			if (nd6_useloopback)
1523 				rt->rt_ifp = lo0ifp;	/* XXX */
1524 			break;
1525 		}
1526 
1527 		/*
1528 		 * check if rt_getkey(rt) is an address assigned
1529 		 * to the interface.
1530 		 */
1531 		s = pserialize_read_enter();
1532 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1533 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1534 		if (ifa != NULL) {
1535 			if (nd6_useloopback) {
1536 				rt->rt_ifp = lo0ifp;	/* XXX */
1537 				/*
1538 				 * Make sure rt_ifa be equal to the ifaddr
1539 				 * corresponding to the address.
1540 				 * We need this because when we refer
1541 				 * rt_ifa->ia6_flags in ip6_input, we assume
1542 				 * that the rt_ifa points to the address instead
1543 				 * of the loopback address.
1544 				 */
1545 				if (ifa != rt->rt_ifa)
1546 					rt_replace_ifa(rt, ifa);
1547 			}
1548 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1549 			/* join solicited node multicast for proxy ND */
1550 			if (ifp->if_flags & IFF_MULTICAST) {
1551 				struct in6_addr llsol;
1552 				int error;
1553 
1554 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1555 				llsol.s6_addr32[0] = htonl(0xff020000);
1556 				llsol.s6_addr32[1] = 0;
1557 				llsol.s6_addr32[2] = htonl(1);
1558 				llsol.s6_addr8[12] = 0xff;
1559 				if (in6_setscope(&llsol, ifp, NULL))
1560 					goto out;
1561 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1562 					nd6log(LOG_ERR, "%s: failed to join "
1563 					    "%s (errno=%d)\n", if_name(ifp),
1564 					    ip6_sprintf(&llsol), error);
1565 				}
1566 			}
1567 		}
1568 	out:
1569 		pserialize_read_exit(s);
1570 		/*
1571 		 * If we have too many cache entries, initiate immediate
1572 		 * purging for some entries.
1573 		 */
1574 		if (rt->rt_ifp != NULL)
1575 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1576 		break;
1577 	    }
1578 
1579 	case RTM_DELETE:
1580 		/* leave from solicited node multicast for proxy ND */
1581 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1582 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1583 			struct in6_addr llsol;
1584 			struct in6_multi *in6m;
1585 
1586 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1587 			llsol.s6_addr32[0] = htonl(0xff020000);
1588 			llsol.s6_addr32[1] = 0;
1589 			llsol.s6_addr32[2] = htonl(1);
1590 			llsol.s6_addr8[12] = 0xff;
1591 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1592 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1593 				if (in6m)
1594 					in6_delmulti(in6m);
1595 			}
1596 		}
1597 		break;
1598 	}
1599 }
1600 
1601 int
1602 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1603 {
1604 	struct in6_drlist *drl = (struct in6_drlist *)data;
1605 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1606 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1607 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1608 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1609 	struct nd_defrouter *dr;
1610 	struct nd_prefix *pr;
1611 	int i = 0, error = 0;
1612 	int s;
1613 
1614 	switch (cmd) {
1615 	case SIOCGDRLST_IN6:
1616 		/*
1617 		 * obsolete API, use sysctl under net.inet6.icmp6
1618 		 */
1619 		memset(drl, 0, sizeof(*drl));
1620 		s = splsoftnet();
1621 		ND_DEFROUTER_LIST_FOREACH(dr) {
1622 			if (i >= DRLSTSIZ)
1623 				break;
1624 			drl->defrouter[i].rtaddr = dr->rtaddr;
1625 			in6_clearscope(&drl->defrouter[i].rtaddr);
1626 
1627 			drl->defrouter[i].flags = dr->flags;
1628 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1629 			drl->defrouter[i].expire = dr->expire ?
1630 			    time_mono_to_wall(dr->expire) : 0;
1631 			drl->defrouter[i].if_index = dr->ifp->if_index;
1632 			i++;
1633 		}
1634 		splx(s);
1635 		break;
1636 	case SIOCGPRLST_IN6:
1637 		/*
1638 		 * obsolete API, use sysctl under net.inet6.icmp6
1639 		 *
1640 		 * XXX the structure in6_prlist was changed in backward-
1641 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1642 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1643 		 */
1644 		/*
1645 		 * XXX meaning of fields, especialy "raflags", is very
1646 		 * differnet between RA prefix list and RR/static prefix list.
1647 		 * how about separating ioctls into two?
1648 		 */
1649 		memset(oprl, 0, sizeof(*oprl));
1650 		s = splsoftnet();
1651 		ND_PREFIX_LIST_FOREACH(pr) {
1652 			struct nd_pfxrouter *pfr;
1653 			int j;
1654 
1655 			if (i >= PRLSTSIZ)
1656 				break;
1657 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1658 			oprl->prefix[i].raflags = pr->ndpr_raf;
1659 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1660 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1661 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1662 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1663 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1664 				oprl->prefix[i].expire = 0;
1665 			else {
1666 				time_t maxexpire;
1667 
1668 				/* XXX: we assume time_t is signed. */
1669 				maxexpire = (-1) &
1670 				    ~((time_t)1 <<
1671 				    ((sizeof(maxexpire) * 8) - 1));
1672 				if (pr->ndpr_vltime <
1673 				    maxexpire - pr->ndpr_lastupdate) {
1674 					time_t expire;
1675 					expire = pr->ndpr_lastupdate +
1676 					    pr->ndpr_vltime;
1677 					oprl->prefix[i].expire = expire ?
1678 					    time_mono_to_wall(expire) : 0;
1679 				} else
1680 					oprl->prefix[i].expire = maxexpire;
1681 			}
1682 
1683 			j = 0;
1684 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1685 				if (j < DRLSTSIZ) {
1686 #define RTRADDR oprl->prefix[i].advrtr[j]
1687 					RTRADDR = pfr->router->rtaddr;
1688 					in6_clearscope(&RTRADDR);
1689 #undef RTRADDR
1690 				}
1691 				j++;
1692 			}
1693 			oprl->prefix[i].advrtrs = j;
1694 			oprl->prefix[i].origin = PR_ORIG_RA;
1695 
1696 			i++;
1697 		}
1698 		splx(s);
1699 
1700 		break;
1701 	case OSIOCGIFINFO_IN6:
1702 #define ND	ndi->ndi
1703 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1704 		memset(&ND, 0, sizeof(ND));
1705 		ND.linkmtu = IN6_LINKMTU(ifp);
1706 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1707 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1708 		ND.reachable = ND_IFINFO(ifp)->reachable;
1709 		ND.retrans = ND_IFINFO(ifp)->retrans;
1710 		ND.flags = ND_IFINFO(ifp)->flags;
1711 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1712 		ND.chlim = ND_IFINFO(ifp)->chlim;
1713 		break;
1714 	case SIOCGIFINFO_IN6:
1715 		ND = *ND_IFINFO(ifp);
1716 		break;
1717 	case SIOCSIFINFO_IN6:
1718 		/*
1719 		 * used to change host variables from userland.
1720 		 * intented for a use on router to reflect RA configurations.
1721 		 */
1722 		/* 0 means 'unspecified' */
1723 		if (ND.linkmtu != 0) {
1724 			if (ND.linkmtu < IPV6_MMTU ||
1725 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1726 				error = EINVAL;
1727 				break;
1728 			}
1729 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1730 		}
1731 
1732 		if (ND.basereachable != 0) {
1733 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1734 
1735 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1736 			if (ND.basereachable != obasereachable)
1737 				ND_IFINFO(ifp)->reachable =
1738 				    ND_COMPUTE_RTIME(ND.basereachable);
1739 		}
1740 		if (ND.retrans != 0)
1741 			ND_IFINFO(ifp)->retrans = ND.retrans;
1742 		if (ND.chlim != 0)
1743 			ND_IFINFO(ifp)->chlim = ND.chlim;
1744 		/* FALLTHROUGH */
1745 	case SIOCSIFINFO_FLAGS:
1746 	{
1747 		struct ifaddr *ifa;
1748 		struct in6_ifaddr *ia;
1749 
1750 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1751 		    !(ND.flags & ND6_IFF_IFDISABLED))
1752 		{
1753 			/*
1754 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1755 			 * has a link-local address with IN6_IFF_DUPLICATED,
1756 			 * do not clear ND6_IFF_IFDISABLED.
1757 			 * See RFC 4862, section 5.4.5.
1758 			 */
1759 			int duplicated_linklocal = 0;
1760 
1761 			s = pserialize_read_enter();
1762 			IFADDR_READER_FOREACH(ifa, ifp) {
1763 				if (ifa->ifa_addr->sa_family != AF_INET6)
1764 					continue;
1765 				ia = (struct in6_ifaddr *)ifa;
1766 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1767 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1768 				{
1769 					duplicated_linklocal = 1;
1770 					break;
1771 				}
1772 			}
1773 			pserialize_read_exit(s);
1774 
1775 			if (duplicated_linklocal) {
1776 				ND.flags |= ND6_IFF_IFDISABLED;
1777 				log(LOG_ERR, "Cannot enable an interface"
1778 				    " with a link-local address marked"
1779 				    " duplicate.\n");
1780 			} else {
1781 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1782 				if (ifp->if_flags & IFF_UP)
1783 					in6_if_up(ifp);
1784 			}
1785 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1786 		    (ND.flags & ND6_IFF_IFDISABLED)) {
1787 			int bound = curlwp_bind();
1788 			/* Mark all IPv6 addresses as tentative. */
1789 
1790 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1791 			s = pserialize_read_enter();
1792 			IFADDR_READER_FOREACH(ifa, ifp) {
1793 				struct psref psref;
1794 				if (ifa->ifa_addr->sa_family != AF_INET6)
1795 					continue;
1796 				ifa_acquire(ifa, &psref);
1797 				pserialize_read_exit(s);
1798 
1799 				nd6_dad_stop(ifa);
1800 
1801 				ia = (struct in6_ifaddr *)ifa;
1802 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1803 
1804 				s = pserialize_read_enter();
1805 				ifa_release(ifa, &psref);
1806 			}
1807 			pserialize_read_exit(s);
1808 			curlwp_bindx(bound);
1809 		}
1810 
1811 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1812 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1813 				/* auto_linklocal 0->1 transition */
1814 
1815 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1816 				in6_ifattach(ifp, NULL);
1817 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1818 			    ifp->if_flags & IFF_UP)
1819 			{
1820 				/*
1821 				 * When the IF already has
1822 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1823 				 * address is assigned, and IFF_UP, try to
1824 				 * assign one.
1825 				 */
1826 				int haslinklocal = 0;
1827 
1828 				s = pserialize_read_enter();
1829 				IFADDR_READER_FOREACH(ifa, ifp) {
1830 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1831 						continue;
1832 					ia = (struct in6_ifaddr *)ifa;
1833 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1834 						haslinklocal = 1;
1835 						break;
1836 					}
1837 				}
1838 				pserialize_read_exit(s);
1839 				if (!haslinklocal)
1840 					in6_ifattach(ifp, NULL);
1841 			}
1842 		}
1843 	}
1844 		ND_IFINFO(ifp)->flags = ND.flags;
1845 		break;
1846 #undef ND
1847 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1848 		/* sync kernel routing table with the default router list */
1849 		nd6_defrouter_reset();
1850 		nd6_defrouter_select();
1851 		break;
1852 	case SIOCSPFXFLUSH_IN6:
1853 	{
1854 		/* flush all the prefix advertised by routers */
1855 		struct nd_prefix *pfx, *next;
1856 
1857 		s = splsoftnet();
1858 		ND_PREFIX_LIST_FOREACH_SAFE(pfx, next) {
1859 			struct in6_ifaddr *ia, *ia_next;
1860 			int _s;
1861 
1862 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1863 				continue; /* XXX */
1864 
1865 			/* do we really have to remove addresses as well? */
1866 		restart:
1867 			_s = pserialize_read_enter();
1868 			for (ia = IN6_ADDRLIST_READER_FIRST(); ia;
1869 			     ia = ia_next) {
1870 				/* ia might be removed.  keep the next ptr. */
1871 				ia_next = IN6_ADDRLIST_READER_NEXT(ia);
1872 
1873 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1874 					continue;
1875 
1876 				if (ia->ia6_ndpr == pfx) {
1877 					pserialize_read_exit(_s);
1878 					/* XXX NOMPSAFE? */
1879 					in6_purgeaddr(&ia->ia_ifa);
1880 					goto restart;
1881 				}
1882 			}
1883 			pserialize_read_exit(_s);
1884 			nd6_prelist_remove(pfx);
1885 		}
1886 		splx(s);
1887 		break;
1888 	}
1889 	case SIOCSRTRFLUSH_IN6:
1890 	{
1891 		/* flush all the default routers */
1892 		struct nd_defrouter *drtr, *next;
1893 
1894 		s = splsoftnet();
1895 		nd6_defrouter_reset();
1896 		ND_DEFROUTER_LIST_FOREACH_SAFE(drtr, next) {
1897 			nd6_defrtrlist_del(drtr, NULL);
1898 		}
1899 		nd6_defrouter_select();
1900 		splx(s);
1901 		break;
1902 	}
1903 	case SIOCGNBRINFO_IN6:
1904 	{
1905 		struct llentry *ln;
1906 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1907 
1908 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1909 			return error;
1910 
1911 		ln = nd6_lookup(&nb_addr, ifp, false);
1912 		if (ln == NULL) {
1913 			error = EINVAL;
1914 			break;
1915 		}
1916 		nbi->state = ln->ln_state;
1917 		nbi->asked = ln->ln_asked;
1918 		nbi->isrouter = ln->ln_router;
1919 		nbi->expire = ln->ln_expire ?
1920 		    time_mono_to_wall(ln->ln_expire) : 0;
1921 		LLE_RUNLOCK(ln);
1922 
1923 		break;
1924 	}
1925 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1926 		ndif->ifindex = nd6_defifindex;
1927 		break;
1928 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1929 		return nd6_setdefaultiface(ndif->ifindex);
1930 	}
1931 	return error;
1932 }
1933 
1934 void
1935 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1936 {
1937 	struct mbuf *m_hold, *m_hold_next;
1938 	struct sockaddr_in6 sin6;
1939 
1940 	LLE_WLOCK_ASSERT(ln);
1941 
1942 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1943 
1944 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1945 
1946 	LLE_WUNLOCK(ln);
1947 	for (; m_hold != NULL; m_hold = m_hold_next) {
1948 		m_hold_next = m_hold->m_nextpkt;
1949 		m_hold->m_nextpkt = NULL;
1950 
1951 		/*
1952 		 * we assume ifp is not a p2p here, so
1953 		 * just set the 2nd argument as the
1954 		 * 1st one.
1955 		 */
1956 		nd6_output(ifp, ifp, m_hold, &sin6, NULL);
1957 	}
1958 	LLE_WLOCK(ln);
1959 }
1960 
1961 /*
1962  * Create neighbor cache entry and cache link-layer address,
1963  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1964  */
1965 void
1966 nd6_cache_lladdr(
1967     struct ifnet *ifp,
1968     struct in6_addr *from,
1969     char *lladdr,
1970     int lladdrlen,
1971     int type,	/* ICMP6 type */
1972     int code	/* type dependent information */
1973 )
1974 {
1975 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
1976 	struct llentry *ln = NULL;
1977 	int is_newentry;
1978 	int do_update;
1979 	int olladdr;
1980 	int llchange;
1981 	int newstate = 0;
1982 	uint16_t router = 0;
1983 
1984 	KASSERT(ifp != NULL);
1985 	KASSERT(from != NULL);
1986 
1987 	/* nothing must be updated for unspecified address */
1988 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1989 		return;
1990 
1991 	/*
1992 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1993 	 * the caller.
1994 	 *
1995 	 * XXX If the link does not have link-layer adderss, what should
1996 	 * we do? (ifp->if_addrlen == 0)
1997 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1998 	 * description on it in NS section (RFC 2461 7.2.3).
1999 	 */
2000 
2001 	ln = nd6_lookup(from, ifp, true);
2002 	if (ln == NULL) {
2003 #if 0
2004 		/* nothing must be done if there's no lladdr */
2005 		if (!lladdr || !lladdrlen)
2006 			return NULL;
2007 #endif
2008 
2009 		ln = nd6_create(from, ifp);
2010 		is_newentry = 1;
2011 	} else {
2012 		/* do nothing if static ndp is set */
2013 		if (ln->la_flags & LLE_STATIC) {
2014 			LLE_WUNLOCK(ln);
2015 			return;
2016 		}
2017 		is_newentry = 0;
2018 	}
2019 
2020 	if (ln == NULL)
2021 		return;
2022 
2023 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2024 	if (olladdr && lladdr) {
2025 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
2026 	} else
2027 		llchange = 0;
2028 
2029 	/*
2030 	 * newentry olladdr  lladdr  llchange	(*=record)
2031 	 *	0	n	n	--	(1)
2032 	 *	0	y	n	--	(2)
2033 	 *	0	n	y	--	(3) * STALE
2034 	 *	0	y	y	n	(4) *
2035 	 *	0	y	y	y	(5) * STALE
2036 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2037 	 *	1	--	y	--	(7) * STALE
2038 	 */
2039 
2040 	if (lladdr) {		/* (3-5) and (7) */
2041 		/*
2042 		 * Record source link-layer address
2043 		 * XXX is it dependent to ifp->if_type?
2044 		 */
2045 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
2046 		ln->la_flags |= LLE_VALID;
2047 	}
2048 
2049 	if (!is_newentry) {
2050 		if ((!olladdr && lladdr) ||		/* (3) */
2051 		    (olladdr && lladdr && llchange)) {	/* (5) */
2052 			do_update = 1;
2053 			newstate = ND6_LLINFO_STALE;
2054 		} else					/* (1-2,4) */
2055 			do_update = 0;
2056 	} else {
2057 		do_update = 1;
2058 		if (lladdr == NULL)			/* (6) */
2059 			newstate = ND6_LLINFO_NOSTATE;
2060 		else					/* (7) */
2061 			newstate = ND6_LLINFO_STALE;
2062 	}
2063 
2064 	if (do_update) {
2065 		/*
2066 		 * Update the state of the neighbor cache.
2067 		 */
2068 		ln->ln_state = newstate;
2069 
2070 		if (ln->ln_state == ND6_LLINFO_STALE) {
2071 			/*
2072 			 * XXX: since nd6_output() below will cause
2073 			 * state tansition to DELAY and reset the timer,
2074 			 * we must set the timer now, although it is actually
2075 			 * meaningless.
2076 			 */
2077 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2078 
2079 			nd6_llinfo_release_pkts(ln, ifp);
2080 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2081 			/* probe right away */
2082 			nd6_llinfo_settimer((void *)ln, 0);
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * ICMP6 type dependent behavior.
2088 	 *
2089 	 * NS: clear IsRouter if new entry
2090 	 * RS: clear IsRouter
2091 	 * RA: set IsRouter if there's lladdr
2092 	 * redir: clear IsRouter if new entry
2093 	 *
2094 	 * RA case, (1):
2095 	 * The spec says that we must set IsRouter in the following cases:
2096 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2097 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2098 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2099 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2100 	 * neighbor cache, this is similar to (6).
2101 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2102 	 *
2103 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2104 	 *							D R
2105 	 *	0	n	n	--	(1)	c   ?     s
2106 	 *	0	y	n	--	(2)	c   s     s
2107 	 *	0	n	y	--	(3)	c   s     s
2108 	 *	0	y	y	n	(4)	c   s     s
2109 	 *	0	y	y	y	(5)	c   s     s
2110 	 *	1	--	n	--	(6) c	c 	c s
2111 	 *	1	--	y	--	(7) c	c   s	c s
2112 	 *
2113 	 *					(c=clear s=set)
2114 	 */
2115 	switch (type & 0xff) {
2116 	case ND_NEIGHBOR_SOLICIT:
2117 		/*
2118 		 * New entry must have is_router flag cleared.
2119 		 */
2120 		if (is_newentry)	/* (6-7) */
2121 			ln->ln_router = 0;
2122 		break;
2123 	case ND_REDIRECT:
2124 		/*
2125 		 * If the icmp is a redirect to a better router, always set the
2126 		 * is_router flag.  Otherwise, if the entry is newly created,
2127 		 * clear the flag.  [RFC 2461, sec 8.3]
2128 		 */
2129 		if (code == ND_REDIRECT_ROUTER)
2130 			ln->ln_router = 1;
2131 		else if (is_newentry) /* (6-7) */
2132 			ln->ln_router = 0;
2133 		break;
2134 	case ND_ROUTER_SOLICIT:
2135 		/*
2136 		 * is_router flag must always be cleared.
2137 		 */
2138 		ln->ln_router = 0;
2139 		break;
2140 	case ND_ROUTER_ADVERT:
2141 		/*
2142 		 * Mark an entry with lladdr as a router.
2143 		 */
2144 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2145 		    (is_newentry && lladdr)) {			/* (7) */
2146 			ln->ln_router = 1;
2147 		}
2148 		break;
2149 	}
2150 
2151 #if 0
2152 	/* XXX should we send rtmsg as it used to be? */
2153 	if (do_update)
2154 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2155 #endif
2156 
2157 	if (ln != NULL) {
2158 		router = ln->ln_router;
2159 		LLE_WUNLOCK(ln);
2160 	}
2161 
2162 	/*
2163 	 * If we have too many cache entries, initiate immediate
2164 	 * purging for some entries.
2165 	 */
2166 	if (is_newentry)
2167 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
2168 
2169 	/*
2170 	 * When the link-layer address of a router changes, select the
2171 	 * best router again.  In particular, when the neighbor entry is newly
2172 	 * created, it might affect the selection policy.
2173 	 * Question: can we restrict the first condition to the "is_newentry"
2174 	 * case?
2175 	 * XXX: when we hear an RA from a new router with the link-layer
2176 	 * address option, nd6_defrouter_select() is called twice, since
2177 	 * defrtrlist_update called the function as well.  However, I believe
2178 	 * we can compromise the overhead, since it only happens the first
2179 	 * time.
2180 	 * XXX: although nd6_defrouter_select() should not have a bad effect
2181 	 * for those are not autoconfigured hosts, we explicitly avoid such
2182 	 * cases for safety.
2183 	 */
2184 	if (do_update && router && !ip6_forwarding &&
2185 	    nd6_accepts_rtadv(ndi))
2186 		nd6_defrouter_select();
2187 }
2188 
2189 static void
2190 nd6_slowtimo(void *ignored_arg)
2191 {
2192 	struct nd_ifinfo *nd6if;
2193 	struct ifnet *ifp;
2194 	int s;
2195 
2196 #ifndef NET_MPSAFE
2197 	mutex_enter(softnet_lock);
2198 	KERNEL_LOCK(1, NULL);
2199 #endif
2200 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2201 	    nd6_slowtimo, NULL);
2202 
2203 	s = pserialize_read_enter();
2204 	IFNET_READER_FOREACH(ifp) {
2205 		nd6if = ND_IFINFO(ifp);
2206 		if (nd6if->basereachable && /* already initialized */
2207 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2208 			/*
2209 			 * Since reachable time rarely changes by router
2210 			 * advertisements, we SHOULD insure that a new random
2211 			 * value gets recomputed at least once every few hours.
2212 			 * (RFC 2461, 6.3.4)
2213 			 */
2214 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2215 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2216 		}
2217 	}
2218 	pserialize_read_exit(s);
2219 
2220 #ifndef NET_MPSAFE
2221 	KERNEL_UNLOCK_ONE(NULL);
2222 	mutex_exit(softnet_lock);
2223 #endif
2224 }
2225 
2226 int
2227 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2228     const struct sockaddr_in6 *dst, struct rtentry *rt)
2229 {
2230 #define senderr(e) { error = (e); goto bad;}
2231 	struct llentry *ln = NULL;
2232 	int error = 0;
2233 	bool created = false;
2234 
2235 	if (rt != NULL) {
2236 		error = rt_check_reject_route(rt, ifp);
2237 		if (error != 0) {
2238 			m_freem(m);
2239 			return error;
2240 		}
2241 	}
2242 
2243 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2244 		goto sendpkt;
2245 
2246 	if (nd6_need_cache(ifp) == 0)
2247 		goto sendpkt;
2248 
2249 	if (rt != NULL && (rt->rt_flags & RTF_GATEWAY) != 0) {
2250 		struct sockaddr_in6 *gw6 = satosin6(rt->rt_gateway);
2251 		int s;
2252 
2253 		/* XXX remain the check to keep the original behavior. */
2254 		/*
2255 		 * We skip link-layer address resolution and NUD
2256 		 * if the gateway is not a neighbor from ND point
2257 		 * of view, regardless of the value of nd_ifinfo.flags.
2258 		 * The second condition is a bit tricky; we skip
2259 		 * if the gateway is our own address, which is
2260 		 * sometimes used to install a route to a p2p link.
2261 		 */
2262 		s = pserialize_read_enter();
2263 		if (!nd6_is_addr_neighbor(gw6, ifp) ||
2264 		    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
2265 			/*
2266 			 * We allow this kind of tricky route only
2267 			 * when the outgoing interface is p2p.
2268 			 * XXX: we may need a more generic rule here.
2269 			 */
2270 			if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
2271 				pserialize_read_exit(s);
2272 				senderr(EHOSTUNREACH);
2273 			}
2274 
2275 			pserialize_read_exit(s);
2276 			goto sendpkt;
2277 		}
2278 		pserialize_read_exit(s);
2279 	}
2280 
2281 	/*
2282 	 * Address resolution or Neighbor Unreachability Detection
2283 	 * for the next hop.
2284 	 * At this point, the destination of the packet must be a unicast
2285 	 * or an anycast address(i.e. not a multicast).
2286 	 */
2287 
2288 	/* Look up the neighbor cache for the nexthop */
2289 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
2290 	if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2291 		/*
2292 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2293 		 * the condition below is not very efficient.  But we believe
2294 		 * it is tolerable, because this should be a rare case.
2295 		 */
2296 		ln = nd6_create(&dst->sin6_addr, ifp);
2297 		if (ln != NULL)
2298 			created = true;
2299 	}
2300 
2301 	if (ln == NULL) {
2302 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2303 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2304 			log(LOG_DEBUG,
2305 			    "nd6_output: can't allocate llinfo for %s "
2306 			    "(ln=%p, rt=%p)\n",
2307 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2308 			senderr(EIO);	/* XXX: good error? */
2309 		}
2310 		goto sendpkt;	/* send anyway */
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 	 * If the neighbor cache entry has a state other than INCOMPLETE
2337 	 * (i.e. its link-layer address is already resolved), just
2338 	 * send the packet.
2339 	 */
2340 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2341 		goto sendpkt;
2342 
2343 	/*
2344 	 * There is a neighbor cache entry, but no ethernet address
2345 	 * response yet.  Append this latest packet to the end of the
2346 	 * packet queue in the mbuf, unless the number of the packet
2347 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2348 	 * the oldest packet in the queue will be removed.
2349 	 */
2350 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2351 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2352 	if (ln->ln_hold) {
2353 		struct mbuf *m_hold;
2354 		int i;
2355 
2356 		i = 0;
2357 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2358 			i++;
2359 			if (m_hold->m_nextpkt == NULL) {
2360 				m_hold->m_nextpkt = m;
2361 				break;
2362 			}
2363 		}
2364 		while (i >= nd6_maxqueuelen) {
2365 			m_hold = ln->ln_hold;
2366 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2367 			m_freem(m_hold);
2368 			i--;
2369 		}
2370 	} else {
2371 		ln->ln_hold = m;
2372 	}
2373 
2374 	/*
2375 	 * If there has been no NS for the neighbor after entering the
2376 	 * INCOMPLETE state, send the first solicitation.
2377 	 */
2378 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2379 		struct in6_addr src, *psrc;
2380 
2381 		ln->ln_asked++;
2382 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans * hz / 1000);
2383 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2384 		LLE_WUNLOCK(ln);
2385 		ln = NULL;
2386 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
2387 	} else {
2388 		/* We did the lookup so we need to do the unlock here. */
2389 		LLE_WUNLOCK(ln);
2390 	}
2391 
2392 	error = 0;
2393 	goto exit;
2394 
2395   sendpkt:
2396 	/* discard the packet if IPv6 operation is disabled on the interface */
2397 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2398 		error = ENETDOWN; /* better error? */
2399 		goto bad;
2400 	}
2401 
2402 	if (ln != NULL)
2403 		LLE_WUNLOCK(ln);
2404 
2405 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2406 		error = if_output_lock(ifp, origifp, m, sin6tocsa(dst), rt);
2407 	else
2408 		error = if_output_lock(ifp, ifp, m, sin6tocsa(dst), rt);
2409 	goto exit;
2410 
2411   bad:
2412 	if (m != NULL)
2413 		m_freem(m);
2414   exit:
2415 	if (created)
2416 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
2417 
2418 	return error;
2419 #undef senderr
2420 }
2421 
2422 int
2423 nd6_need_cache(struct ifnet *ifp)
2424 {
2425 	/*
2426 	 * XXX: we currently do not make neighbor cache on any interface
2427 	 * other than ARCnet, Ethernet, FDDI and GIF.
2428 	 *
2429 	 * RFC2893 says:
2430 	 * - unidirectional tunnels needs no ND
2431 	 */
2432 	switch (ifp->if_type) {
2433 	case IFT_ARCNET:
2434 	case IFT_ETHER:
2435 	case IFT_FDDI:
2436 	case IFT_IEEE1394:
2437 	case IFT_CARP:
2438 	case IFT_GIF:		/* XXX need more cases? */
2439 	case IFT_PPP:
2440 	case IFT_TUNNEL:
2441 		return 1;
2442 	default:
2443 		return 0;
2444 	}
2445 }
2446 
2447 /*
2448  * Add pernament ND6 link-layer record for given
2449  * interface address.
2450  *
2451  * Very similar to IPv4 arp_ifinit(), but:
2452  * 1) IPv6 DAD is performed in different place
2453  * 2) It is called by IPv6 protocol stack in contrast to
2454  * arp_ifinit() which is typically called in SIOCSIFADDR
2455  * driver ioctl handler.
2456  *
2457  */
2458 int
2459 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2460 {
2461 	struct ifnet *ifp;
2462 	struct llentry *ln;
2463 
2464 	ifp = ia->ia_ifa.ifa_ifp;
2465 	if (nd6_need_cache(ifp) == 0)
2466 		return 0;
2467 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2468 	ia->ia_ifa.ifa_flags |= RTF_CONNECTED;
2469 
2470 	IF_AFDATA_WLOCK(ifp);
2471 	ln = lla_create(LLTABLE6(ifp), LLE_IFADDR | LLE_EXCLUSIVE,
2472 	    sin6tosa(&ia->ia_addr));
2473 	IF_AFDATA_WUNLOCK(ifp);
2474 	if (ln == NULL)
2475 		return ENOBUFS;
2476 
2477 	ln->la_expire = 0;  /* for IPv6 this means permanent */
2478 	ln->ln_state = ND6_LLINFO_REACHABLE;
2479 
2480 	LLE_WUNLOCK(ln);
2481 	return 0;
2482 }
2483 
2484 /*
2485  * Removes ALL lle records for interface address prefix.
2486  * XXXME: That's probably not we really want to do, we need
2487  * to remove address record only and keep other records
2488  * until we determine if given prefix is really going
2489  * to be removed.
2490  */
2491 void
2492 nd6_rem_ifa_lle(struct in6_ifaddr *ia)
2493 {
2494 	struct sockaddr_in6 mask, addr;
2495 
2496 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2497 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2498 	lltable_prefix_free(AF_INET6, sin6tosa(&addr), sin6tosa(&mask),
2499 	    LLE_STATIC);
2500 }
2501 
2502 int
2503 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2504     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2505     size_t dstsize)
2506 {
2507 	struct llentry *ln;
2508 
2509 	if (m->m_flags & M_MCAST) {
2510 		switch (ifp->if_type) {
2511 		case IFT_ETHER:
2512 		case IFT_FDDI:
2513 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2514 			    lldst);
2515 			return 1;
2516 		case IFT_IEEE1394:
2517 			memcpy(lldst, ifp->if_broadcastaddr,
2518 			    MIN(dstsize, ifp->if_addrlen));
2519 			return 1;
2520 		case IFT_ARCNET:
2521 			*lldst = 0;
2522 			return 1;
2523 		default:
2524 			m_freem(m);
2525 			return 0;
2526 		}
2527 	}
2528 
2529 	/*
2530 	 * the entry should have been created in nd6_store_lladdr
2531 	 */
2532 	ln = nd6_lookup(&satocsin6(dst)->sin6_addr, ifp, false);
2533 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2534 		if (ln != NULL)
2535 			LLE_RUNLOCK(ln);
2536 		/* this could happen, if we could not allocate memory */
2537 		m_freem(m);
2538 		return 0;
2539 	}
2540 
2541 	/* XXX llentry should have addrlen? */
2542 #if 0
2543 	sdl = satocsdl(rt->rt_gateway);
2544 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2545 		char sbuf[INET6_ADDRSTRLEN];
2546 		char dbuf[LINK_ADDRSTRLEN];
2547 		/* this should be impossible, but we bark here for debugging */
2548 		printf("%s: sdl_alen == %" PRIu8 ", if=%s, dst=%s, sdl=%s\n",
2549 		    __func__, sdl->sdl_alen, if_name(ifp),
2550 		    IN6_PRINT(sbuf, &satocsin6(dst)->sin6_addr),
2551 		    DL_PRINT(dbuf, &sdl->sdl_addr));
2552 		m_freem(m);
2553 		return 0;
2554 	}
2555 #endif
2556 
2557 	memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
2558 
2559 	LLE_RUNLOCK(ln);
2560 
2561 	return 1;
2562 }
2563 
2564 static void
2565 clear_llinfo_pqueue(struct llentry *ln)
2566 {
2567 	struct mbuf *m_hold, *m_hold_next;
2568 
2569 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2570 		m_hold_next = m_hold->m_nextpkt;
2571 		m_hold->m_nextpkt = NULL;
2572 		m_freem(m_hold);
2573 	}
2574 
2575 	ln->ln_hold = NULL;
2576 	return;
2577 }
2578 
2579 int
2580 nd6_sysctl(
2581     int name,
2582     void *oldp,	/* syscall arg, need copyout */
2583     size_t *oldlenp,
2584     void *newp,	/* syscall arg, need copyin */
2585     size_t newlen
2586 )
2587 {
2588 	void *p;
2589 	size_t ol;
2590 	int error;
2591 
2592 	error = 0;
2593 
2594 	if (newp)
2595 		return EPERM;
2596 	if (oldp && !oldlenp)
2597 		return EINVAL;
2598 	ol = oldlenp ? *oldlenp : 0;
2599 
2600 	if (oldp) {
2601 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2602 		if (p == NULL)
2603 			return ENOMEM;
2604 	} else
2605 		p = NULL;
2606 	switch (name) {
2607 	case ICMPV6CTL_ND6_DRLIST:
2608 		error = fill_drlist(p, oldlenp, ol);
2609 		if (!error && p != NULL && oldp != NULL)
2610 			error = copyout(p, oldp, *oldlenp);
2611 		break;
2612 
2613 	case ICMPV6CTL_ND6_PRLIST:
2614 		error = fill_prlist(p, oldlenp, ol);
2615 		if (!error && p != NULL && oldp != NULL)
2616 			error = copyout(p, oldp, *oldlenp);
2617 		break;
2618 
2619 	case ICMPV6CTL_ND6_MAXQLEN:
2620 		break;
2621 
2622 	default:
2623 		error = ENOPROTOOPT;
2624 		break;
2625 	}
2626 	if (p)
2627 		free(p, M_TEMP);
2628 
2629 	return error;
2630 }
2631 
2632 static int
2633 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2634 {
2635 	int error = 0, s;
2636 	struct in6_defrouter *d = NULL, *de = NULL;
2637 	struct nd_defrouter *dr;
2638 	size_t l;
2639 
2640 	s = splsoftnet();
2641 
2642 	if (oldp) {
2643 		d = (struct in6_defrouter *)oldp;
2644 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2645 	}
2646 	l = 0;
2647 
2648 	ND_DEFROUTER_LIST_FOREACH(dr) {
2649 
2650 		if (oldp && d + 1 <= de) {
2651 			memset(d, 0, sizeof(*d));
2652 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2653 			if (sa6_recoverscope(&d->rtaddr)) {
2654 				log(LOG_ERR,
2655 				    "scope error in router list (%s)\n",
2656 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2657 				/* XXX: press on... */
2658 			}
2659 			d->flags = dr->flags;
2660 			d->rtlifetime = dr->rtlifetime;
2661 			d->expire = dr->expire ?
2662 			    time_mono_to_wall(dr->expire) : 0;
2663 			d->if_index = dr->ifp->if_index;
2664 		}
2665 
2666 		l += sizeof(*d);
2667 		if (d)
2668 			d++;
2669 	}
2670 
2671 	if (oldp) {
2672 		if (l > ol)
2673 			error = ENOMEM;
2674 	}
2675 	if (oldlenp)
2676 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2677 
2678 	splx(s);
2679 
2680 	return error;
2681 }
2682 
2683 static int
2684 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2685 {
2686 	int error = 0, s;
2687 	struct nd_prefix *pr;
2688 	uint8_t *p = NULL, *ps = NULL;
2689 	uint8_t *pe = NULL;
2690 	size_t l;
2691 
2692 	s = splsoftnet();
2693 
2694 	if (oldp) {
2695 		ps = p = (uint8_t*)oldp;
2696 		pe = (uint8_t*)oldp + *oldlenp;
2697 	}
2698 	l = 0;
2699 
2700 	ND_PREFIX_LIST_FOREACH(pr) {
2701 		u_short advrtrs;
2702 		struct sockaddr_in6 sin6;
2703 		struct nd_pfxrouter *pfr;
2704 		struct in6_prefix pfx;
2705 
2706 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2707 		{
2708 			memset(&pfx, 0, sizeof(pfx));
2709 			ps = p;
2710 			pfx.prefix = pr->ndpr_prefix;
2711 
2712 			if (sa6_recoverscope(&pfx.prefix)) {
2713 				log(LOG_ERR,
2714 				    "scope error in prefix list (%s)\n",
2715 				    ip6_sprintf(&pfx.prefix.sin6_addr));
2716 				/* XXX: press on... */
2717 			}
2718 			pfx.raflags = pr->ndpr_raf;
2719 			pfx.prefixlen = pr->ndpr_plen;
2720 			pfx.vltime = pr->ndpr_vltime;
2721 			pfx.pltime = pr->ndpr_pltime;
2722 			pfx.if_index = pr->ndpr_ifp->if_index;
2723 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2724 				pfx.expire = 0;
2725 			else {
2726 				time_t maxexpire;
2727 
2728 				/* XXX: we assume time_t is signed. */
2729 				maxexpire = (-1) &
2730 				    ~((time_t)1 <<
2731 				    ((sizeof(maxexpire) * 8) - 1));
2732 				if (pr->ndpr_vltime <
2733 				    maxexpire - pr->ndpr_lastupdate) {
2734 					pfx.expire = pr->ndpr_lastupdate +
2735 						pr->ndpr_vltime;
2736 				} else
2737 					pfx.expire = maxexpire;
2738 			}
2739 			pfx.refcnt = pr->ndpr_refcnt;
2740 			pfx.flags = pr->ndpr_stateflags;
2741 			pfx.origin = PR_ORIG_RA;
2742 
2743 			p += sizeof(pfx); l += sizeof(pfx);
2744 
2745 			advrtrs = 0;
2746 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2747 				if (p + sizeof(sin6) > pe) {
2748 					advrtrs++;
2749 					continue;
2750 				}
2751 
2752 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2753 				    0, 0, 0);
2754 				if (sa6_recoverscope(&sin6)) {
2755 					log(LOG_ERR,
2756 					    "scope error in "
2757 					    "prefix list (%s)\n",
2758 					    ip6_sprintf(&pfr->router->rtaddr));
2759 				}
2760 				advrtrs++;
2761 				memcpy(p, &sin6, sizeof(sin6));
2762 				p += sizeof(sin6);
2763 				l += sizeof(sin6);
2764 			}
2765 			pfx.advrtrs = advrtrs;
2766 			memcpy(ps, &pfx, sizeof(pfx));
2767 		}
2768 		else {
2769 			l += sizeof(pfx);
2770 			advrtrs = 0;
2771 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2772 				advrtrs++;
2773 				l += sizeof(sin6);
2774 			}
2775 		}
2776 	}
2777 
2778 	if (oldp) {
2779 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2780 		if (l > ol)
2781 			error = ENOMEM;
2782 	} else
2783 		*oldlenp = l;
2784 
2785 	splx(s);
2786 
2787 	return error;
2788 }
2789 
2790 static int
2791 nd6_setdefaultiface(int ifindex)
2792 {
2793 	ifnet_t *ifp;
2794 	int error = 0;
2795 	int s;
2796 
2797 	s = pserialize_read_enter();
2798 	ifp = if_byindex(ifindex);
2799 	if (ifp == NULL) {
2800 		pserialize_read_exit(s);
2801 		return EINVAL;
2802 	}
2803 	if (nd6_defifindex != ifindex) {
2804 		nd6_defifindex = ifindex;
2805 		nd6_defifp = nd6_defifindex > 0 ? ifp : NULL;
2806 
2807 		/*
2808 		 * Our current implementation assumes one-to-one maping between
2809 		 * interfaces and links, so it would be natural to use the
2810 		 * default interface as the default link.
2811 		 */
2812 		scope6_setdefault(nd6_defifp);
2813 	}
2814 	pserialize_read_exit(s);
2815 
2816 	return (error);
2817 }
2818