xref: /openbsd-src/sys/netinet6/nd6.c (revision cba26e98faa2b48aa4705f205ed876af460243a2)
1 /*	$OpenBSD: nd6.c,v 1.234 2021/01/06 08:10:15 florian Exp $	*/
2 /*	$KAME: nd6.c,v 1.280 2002/06/08 19:52:07 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/param.h>
34 #include <sys/systm.h>
35 #include <sys/timeout.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/socket.h>
39 #include <sys/sockio.h>
40 #include <sys/time.h>
41 #include <sys/kernel.h>
42 #include <sys/pool.h>
43 #include <sys/protosw.h>
44 #include <sys/errno.h>
45 #include <sys/ioctl.h>
46 #include <sys/syslog.h>
47 #include <sys/queue.h>
48 #include <sys/stdint.h>
49 #include <sys/task.h>
50 
51 #include <net/if.h>
52 #include <net/if_dl.h>
53 #include <net/if_types.h>
54 #include <net/route.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/if_ether.h>
58 #include <netinet/ip_ipsp.h>
59 
60 #include <netinet6/in6_var.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet6/nd6.h>
64 #include <netinet/icmp6.h>
65 
66 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
67 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
68 
69 /* timer values */
70 int	nd6_timer_next	= -1;	/* at which uptime nd6_timer runs */
71 time_t	nd6_expire_next	= -1;	/* at which uptime nd6_expire runs */
72 int	nd6_delay	= 5;	/* delay first probe time 5 second */
73 int	nd6_umaxtries	= 3;	/* maximum unicast query */
74 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
75 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
76 
77 /* preventing too many loops in ND option parsing */
78 int nd6_maxndopt = 10;	/* max # of ND options allowed */
79 
80 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
81 
82 #ifdef ND6_DEBUG
83 int nd6_debug = 1;
84 #else
85 int nd6_debug = 0;
86 #endif
87 
88 TAILQ_HEAD(llinfo_nd6_head, llinfo_nd6) nd6_list;
89 struct	pool nd6_pool;		/* pool for llinfo_nd6 structures */
90 int	nd6_inuse, nd6_allocated;
91 
92 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
93 
94 void nd6_timer(void *);
95 void nd6_slowtimo(void *);
96 void nd6_expire(void *);
97 void nd6_expire_timer(void *);
98 void nd6_invalidate(struct rtentry *);
99 void nd6_free(struct rtentry *);
100 int nd6_llinfo_timer(struct rtentry *);
101 
102 struct timeout nd6_timer_to;
103 struct timeout nd6_slowtimo_ch;
104 struct timeout nd6_expire_timeout;
105 struct task nd6_expire_task;
106 
107 void
108 nd6_init(void)
109 {
110 	static int nd6_init_done = 0;
111 
112 	if (nd6_init_done) {
113 		log(LOG_NOTICE, "%s called more than once\n", __func__);
114 		return;
115 	}
116 
117 	TAILQ_INIT(&nd6_list);
118 	pool_init(&nd6_pool, sizeof(struct llinfo_nd6), 0,
119 	    IPL_SOFTNET, 0, "nd6", NULL);
120 
121 	task_set(&nd6_expire_task, nd6_expire, NULL);
122 
123 	nd6_init_done = 1;
124 
125 	/* start timer */
126 	timeout_set_proc(&nd6_timer_to, nd6_timer, &nd6_timer_to);
127 	timeout_set_proc(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
128 	timeout_add_sec(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL);
129 	timeout_set(&nd6_expire_timeout, nd6_expire_timer, NULL);
130 }
131 
132 struct nd_ifinfo *
133 nd6_ifattach(struct ifnet *ifp)
134 {
135 	struct nd_ifinfo *nd;
136 
137 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
138 
139 	nd->initialized = 1;
140 
141 	nd->basereachable = REACHABLE_TIME;
142 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
143 	nd->retrans = RETRANS_TIMER;
144 	/* per-interface IFXF_AUTOCONF6 needs to be set too to accept RAs */
145 
146 	return nd;
147 }
148 
149 void
150 nd6_ifdetach(struct nd_ifinfo *nd)
151 {
152 
153 	free(nd, M_IP6NDP, sizeof(*nd));
154 }
155 
156 void
157 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
158 {
159 	bzero(ndopts, sizeof(*ndopts));
160 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
161 	ndopts->nd_opts_last
162 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
163 
164 	if (icmp6len == 0) {
165 		ndopts->nd_opts_done = 1;
166 		ndopts->nd_opts_search = NULL;
167 	}
168 }
169 
170 /*
171  * Take one ND option.
172  */
173 struct nd_opt_hdr *
174 nd6_option(union nd_opts *ndopts)
175 {
176 	struct nd_opt_hdr *nd_opt;
177 	int olen;
178 
179 	if (!ndopts)
180 		panic("%s: ndopts == NULL", __func__);
181 	if (!ndopts->nd_opts_last)
182 		panic("%s: uninitialized ndopts", __func__);
183 	if (!ndopts->nd_opts_search)
184 		return NULL;
185 	if (ndopts->nd_opts_done)
186 		return NULL;
187 
188 	nd_opt = ndopts->nd_opts_search;
189 
190 	/* make sure nd_opt_len is inside the buffer */
191 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
192 		bzero(ndopts, sizeof(*ndopts));
193 		return NULL;
194 	}
195 
196 	olen = nd_opt->nd_opt_len << 3;
197 	if (olen == 0) {
198 		/*
199 		 * Message validation requires that all included
200 		 * options have a length that is greater than zero.
201 		 */
202 		bzero(ndopts, sizeof(*ndopts));
203 		return NULL;
204 	}
205 
206 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
207 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
208 		/* option overruns the end of buffer, invalid */
209 		bzero(ndopts, sizeof(*ndopts));
210 		return NULL;
211 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
212 		/* reached the end of options chain */
213 		ndopts->nd_opts_done = 1;
214 		ndopts->nd_opts_search = NULL;
215 	}
216 	return nd_opt;
217 }
218 
219 /*
220  * Parse multiple ND options.
221  * This function is much easier to use, for ND routines that do not need
222  * multiple options of the same type.
223  */
224 int
225 nd6_options(union nd_opts *ndopts)
226 {
227 	struct nd_opt_hdr *nd_opt;
228 	int i = 0;
229 
230 	if (!ndopts)
231 		panic("%s: ndopts == NULL", __func__);
232 	if (!ndopts->nd_opts_last)
233 		panic("%s: uninitialized ndopts", __func__);
234 	if (!ndopts->nd_opts_search)
235 		return 0;
236 
237 	while (1) {
238 		nd_opt = nd6_option(ndopts);
239 		if (!nd_opt && !ndopts->nd_opts_last) {
240 			/*
241 			 * Message validation requires that all included
242 			 * options have a length that is greater than zero.
243 			 */
244 			icmp6stat_inc(icp6s_nd_badopt);
245 			bzero(ndopts, sizeof(*ndopts));
246 			return -1;
247 		}
248 
249 		if (!nd_opt)
250 			goto skip1;
251 
252 		switch (nd_opt->nd_opt_type) {
253 		case ND_OPT_SOURCE_LINKADDR:
254 		case ND_OPT_TARGET_LINKADDR:
255 		case ND_OPT_MTU:
256 		case ND_OPT_REDIRECTED_HEADER:
257 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
258 				nd6log((LOG_INFO,
259 				    "duplicated ND6 option found (type=%d)\n",
260 				    nd_opt->nd_opt_type));
261 				/* XXX bark? */
262 			} else {
263 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
264 					= nd_opt;
265 			}
266 			break;
267 		case ND_OPT_PREFIX_INFORMATION:
268 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
269 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
270 					= nd_opt;
271 			}
272 			ndopts->nd_opts_pi_end =
273 				(struct nd_opt_prefix_info *)nd_opt;
274 			break;
275 		default:
276 			/*
277 			 * Unknown options must be silently ignored,
278 			 * to accommodate future extension to the protocol.
279 			 */
280 			nd6log((LOG_DEBUG,
281 			    "nd6_options: unsupported option %d - "
282 			    "option ignored\n", nd_opt->nd_opt_type));
283 		}
284 
285 skip1:
286 		i++;
287 		if (i > nd6_maxndopt) {
288 			icmp6stat_inc(icp6s_nd_toomanyopt);
289 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
290 			break;
291 		}
292 
293 		if (ndopts->nd_opts_done)
294 			break;
295 	}
296 
297 	return 0;
298 }
299 
300 /*
301  * ND6 timer routine to handle ND6 entries
302  */
303 void
304 nd6_llinfo_settimer(struct llinfo_nd6 *ln, unsigned int secs)
305 {
306 	time_t expire = getuptime() + secs;
307 
308 	NET_ASSERT_LOCKED();
309 	KASSERT(!ISSET(ln->ln_rt->rt_flags, RTF_LOCAL));
310 
311 	ln->ln_rt->rt_expire = expire;
312 	if (!timeout_pending(&nd6_timer_to) || expire < nd6_timer_next) {
313 		nd6_timer_next = expire;
314 		timeout_add_sec(&nd6_timer_to, secs);
315 	}
316 }
317 
318 void
319 nd6_timer(void *arg)
320 {
321 	struct llinfo_nd6 *ln, *nln;
322 	time_t expire = getuptime() + nd6_gctimer;
323 	int secs;
324 
325 	NET_LOCK();
326 	TAILQ_FOREACH_SAFE(ln, &nd6_list, ln_list, nln) {
327 		struct rtentry *rt = ln->ln_rt;
328 
329 		if (rt->rt_expire && rt->rt_expire <= getuptime())
330 			if (nd6_llinfo_timer(rt))
331 				continue;
332 
333 		if (rt->rt_expire && rt->rt_expire < expire)
334 			expire = rt->rt_expire;
335 	}
336 
337 	secs = expire - getuptime();
338 	if (secs < 0)
339 		secs = 0;
340 	if (!TAILQ_EMPTY(&nd6_list)) {
341 		nd6_timer_next = getuptime() + secs;
342 		timeout_add_sec(&nd6_timer_to, secs);
343 	}
344 
345 	NET_UNLOCK();
346 }
347 
348 /*
349  * ND timer state handling.
350  *
351  * Returns 1 if `rt' should no longer be used, 0 otherwise.
352  */
353 int
354 nd6_llinfo_timer(struct rtentry *rt)
355 {
356 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
357 	struct sockaddr_in6 *dst = satosin6(rt_key(rt));
358 	struct ifnet *ifp;
359 	struct nd_ifinfo *ndi = NULL;
360 
361 	NET_ASSERT_LOCKED();
362 
363 	if ((ifp = if_get(rt->rt_ifidx)) == NULL)
364 		return 1;
365 
366 	ndi = ND_IFINFO(ifp);
367 
368 	switch (ln->ln_state) {
369 	case ND6_LLINFO_INCOMPLETE:
370 		if (ln->ln_asked < nd6_mmaxtries) {
371 			ln->ln_asked++;
372 			nd6_llinfo_settimer(ln, ndi->retrans / 1000);
373 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
374 		} else {
375 			struct mbuf *m = ln->ln_hold;
376 			if (m) {
377 				ln->ln_hold = NULL;
378 				/*
379 				 * Fake rcvif to make the ICMP error
380 				 * more helpful in diagnosing for the
381 				 * receiver.
382 				 * XXX: should we consider
383 				 * older rcvif?
384 				 */
385 				m->m_pkthdr.ph_ifidx = rt->rt_ifidx;
386 
387 				icmp6_error(m, ICMP6_DST_UNREACH,
388 				    ICMP6_DST_UNREACH_ADDR, 0);
389 				if (ln->ln_hold == m) {
390 					/* m is back in ln_hold. Discard. */
391 					m_freem(ln->ln_hold);
392 					ln->ln_hold = NULL;
393 				}
394 			}
395 			nd6_free(rt);
396 			ln = NULL;
397 		}
398 		break;
399 	case ND6_LLINFO_REACHABLE:
400 		if (!ND6_LLINFO_PERMANENT(ln)) {
401 			ln->ln_state = ND6_LLINFO_STALE;
402 			nd6_llinfo_settimer(ln, nd6_gctimer);
403 		}
404 		break;
405 
406 	case ND6_LLINFO_STALE:
407 	case ND6_LLINFO_PURGE:
408 		/* Garbage Collection(RFC 2461 5.3) */
409 		if (!ND6_LLINFO_PERMANENT(ln)) {
410 			nd6_free(rt);
411 			ln = NULL;
412 		}
413 		break;
414 
415 	case ND6_LLINFO_DELAY:
416 		if (ndi) {
417 			/* We need NUD */
418 			ln->ln_asked = 1;
419 			ln->ln_state = ND6_LLINFO_PROBE;
420 			nd6_llinfo_settimer(ln, ndi->retrans / 1000);
421 			nd6_ns_output(ifp, &dst->sin6_addr,
422 			    &dst->sin6_addr, ln, 0);
423 		}
424 		break;
425 	case ND6_LLINFO_PROBE:
426 		if (ln->ln_asked < nd6_umaxtries) {
427 			ln->ln_asked++;
428 			nd6_llinfo_settimer(ln, ndi->retrans / 1000);
429 			nd6_ns_output(ifp, &dst->sin6_addr,
430 			    &dst->sin6_addr, ln, 0);
431 		} else {
432 			nd6_free(rt);
433 			ln = NULL;
434 		}
435 		break;
436 	}
437 
438 	if_put(ifp);
439 
440 	return (ln == NULL);
441 }
442 
443 void
444 nd6_expire_timer_update(struct in6_ifaddr *ia6)
445 {
446 	time_t expire_time = INT64_MAX;
447 	int secs;
448 
449 	KERNEL_ASSERT_LOCKED();
450 
451 	if (ia6->ia6_lifetime.ia6t_vltime != ND6_INFINITE_LIFETIME)
452 		expire_time = ia6->ia6_lifetime.ia6t_expire;
453 
454 	if (!(ia6->ia6_flags & IN6_IFF_DEPRECATED) &&
455 	    ia6->ia6_lifetime.ia6t_pltime != ND6_INFINITE_LIFETIME &&
456 	    expire_time > ia6->ia6_lifetime.ia6t_preferred)
457 		expire_time = ia6->ia6_lifetime.ia6t_preferred;
458 
459 	if (expire_time == INT64_MAX)
460 		return;
461 
462 	/*
463 	 * IFA6_IS_INVALID() and IFA6_IS_DEPRECATED() check for uptime
464 	 * greater than ia6t_expire or ia6t_preferred, not greater or equal.
465 	 * Schedule timeout one second later so that either IFA6_IS_INVALID()
466 	 * or IFA6_IS_DEPRECATED() is true.
467 	 */
468 	expire_time++;
469 
470 	if (!timeout_pending(&nd6_expire_timeout) ||
471 	    nd6_expire_next > expire_time) {
472 		secs = expire_time - getuptime();
473 		if (secs < 0)
474 			secs = 0;
475 
476 		timeout_add_sec(&nd6_expire_timeout, secs);
477 		nd6_expire_next = expire_time;
478 	}
479 }
480 
481 /*
482  * Expire interface addresses.
483  */
484 void
485 nd6_expire(void *unused)
486 {
487 	struct ifnet *ifp;
488 
489 	KERNEL_LOCK();
490 	NET_LOCK();
491 
492 	TAILQ_FOREACH(ifp, &ifnet, if_list) {
493 		struct ifaddr *ifa, *nifa;
494 		struct in6_ifaddr *ia6;
495 
496 		TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrlist, ifa_list, nifa) {
497 			if (ifa->ifa_addr->sa_family != AF_INET6)
498 				continue;
499 			ia6 = ifatoia6(ifa);
500 			/* check address lifetime */
501 			if (IFA6_IS_INVALID(ia6)) {
502 				in6_purgeaddr(&ia6->ia_ifa);
503 			} else {
504 				if (IFA6_IS_DEPRECATED(ia6))
505 					ia6->ia6_flags |= IN6_IFF_DEPRECATED;
506 				nd6_expire_timer_update(ia6);
507 			}
508 		}
509 	}
510 
511 	NET_UNLOCK();
512 	KERNEL_UNLOCK();
513 }
514 
515 void
516 nd6_expire_timer(void *unused)
517 {
518 	task_add(net_tq(0), &nd6_expire_task);
519 }
520 
521 /*
522  * Nuke neighbor cache/prefix/default router management table, right before
523  * ifp goes away.
524  */
525 void
526 nd6_purge(struct ifnet *ifp)
527 {
528 	struct llinfo_nd6 *ln, *nln;
529 
530 	NET_ASSERT_LOCKED();
531 
532 	/*
533 	 * Nuke neighbor cache entries for the ifp.
534 	 */
535 	TAILQ_FOREACH_SAFE(ln, &nd6_list, ln_list, nln) {
536 		struct rtentry *rt;
537 		struct sockaddr_dl *sdl;
538 
539 		rt = ln->ln_rt;
540 		if (rt != NULL && rt->rt_gateway != NULL &&
541 		    rt->rt_gateway->sa_family == AF_LINK) {
542 			sdl = satosdl(rt->rt_gateway);
543 			if (sdl->sdl_index == ifp->if_index)
544 				nd6_free(rt);
545 		}
546 	}
547 }
548 
549 struct rtentry *
550 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp,
551     u_int rtableid)
552 {
553 	struct rtentry *rt;
554 	struct sockaddr_in6 sin6;
555 	int flags;
556 
557 	bzero(&sin6, sizeof(sin6));
558 	sin6.sin6_len = sizeof(struct sockaddr_in6);
559 	sin6.sin6_family = AF_INET6;
560 	sin6.sin6_addr = *addr6;
561 	flags = (create) ? RT_RESOLVE : 0;
562 
563 	rt = rtalloc(sin6tosa(&sin6), flags, rtableid);
564 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) == 0) {
565 		/*
566 		 * This is the case for the default route.
567 		 * If we want to create a neighbor cache for the address, we
568 		 * should free the route for the destination and allocate an
569 		 * interface route.
570 		 */
571 		if (create) {
572 			rtfree(rt);
573 			rt = NULL;
574 		}
575 	}
576 	if (rt == NULL) {
577 		if (create && ifp) {
578 			struct rt_addrinfo info;
579 			struct ifaddr *ifa;
580 			int error;
581 
582 			/*
583 			 * If no route is available and create is set,
584 			 * we allocate a host route for the destination
585 			 * and treat it like an interface route.
586 			 * This hack is necessary for a neighbor which can't
587 			 * be covered by our own prefix.
588 			 */
589 			ifa = ifaof_ifpforaddr(sin6tosa(&sin6), ifp);
590 			if (ifa == NULL)
591 				return (NULL);
592 
593 			/*
594 			 * Create a new route.  RTF_LLINFO is necessary
595 			 * to create a Neighbor Cache entry for the
596 			 * destination in nd6_rtrequest which will be
597 			 * called in rtrequest.
598 			 */
599 			bzero(&info, sizeof(info));
600 			info.rti_ifa = ifa;
601 			info.rti_flags = RTF_HOST | RTF_LLINFO;
602 			info.rti_info[RTAX_DST] = sin6tosa(&sin6);
603 			info.rti_info[RTAX_GATEWAY] = sdltosa(ifp->if_sadl);
604 			error = rtrequest(RTM_ADD, &info, RTP_CONNECTED, &rt,
605 			    rtableid);
606 			if (error)
607 				return (NULL);
608 			if (rt->rt_llinfo != NULL) {
609 				struct llinfo_nd6 *ln =
610 				    (struct llinfo_nd6 *)rt->rt_llinfo;
611 				ln->ln_state = ND6_LLINFO_NOSTATE;
612 			}
613 		} else
614 			return (NULL);
615 	}
616 	/*
617 	 * Validation for the entry.
618 	 * Note that the check for rt_llinfo is necessary because a cloned
619 	 * route from a parent route that has the L flag (e.g. the default
620 	 * route to a p2p interface) may have the flag, too, while the
621 	 * destination is not actually a neighbor.
622 	 */
623 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
624 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
625 	    (ifp != NULL && rt->rt_ifidx != ifp->if_index)) {
626 		if (create) {
627 			char addr[INET6_ADDRSTRLEN];
628 			nd6log((LOG_DEBUG, "%s: failed to lookup %s (if=%s)\n",
629 			    __func__,
630 			    inet_ntop(AF_INET6, addr6, addr, sizeof(addr)),
631 			    ifp ? ifp->if_xname : "unspec"));
632 		}
633 		rtfree(rt);
634 		return (NULL);
635 	}
636 	return (rt);
637 }
638 
639 /*
640  * Detect if a given IPv6 address identifies a neighbor on a given link.
641  * XXX: should take care of the destination of a p2p link?
642  */
643 int
644 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
645 {
646 	struct in6_ifaddr *ia6;
647 	struct ifaddr *ifa;
648 	struct rtentry *rt;
649 
650 	/*
651 	 * A link-local address is always a neighbor.
652 	 * XXX: we should use the sin6_scope_id field rather than the embedded
653 	 * interface index.
654 	 * XXX: a link does not necessarily specify a single interface.
655 	 */
656 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
657 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
658 		return (1);
659 
660 	TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
661 		if (ifa->ifa_addr->sa_family != AF_INET6)
662 			continue;
663 
664 		ia6 = ifatoia6(ifa);
665 
666 		/* Prefix check down below. */
667 		if (ia6->ia6_flags & IN6_IFF_AUTOCONF)
668 			continue;
669 
670 		if (IN6_ARE_MASKED_ADDR_EQUAL(&addr->sin6_addr,
671 		    &ia6->ia_addr.sin6_addr,
672 		    &ia6->ia_prefixmask.sin6_addr))
673 			return (1);
674 	}
675 
676 	/*
677 	 * Even if the address matches none of our addresses, it might be
678 	 * in the neighbor cache.
679 	 */
680 	rt = nd6_lookup(&addr->sin6_addr, 0, ifp, ifp->if_rdomain);
681 	if (rt != NULL) {
682 		rtfree(rt);
683 		return (1);
684 	}
685 
686 	return (0);
687 }
688 
689 void
690 nd6_invalidate(struct rtentry *rt)
691 {
692 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
693 	struct sockaddr_dl *sdl = satosdl(rt->rt_gateway);
694 
695 	m_freem(ln->ln_hold);
696 	sdl->sdl_alen = 0;
697 	ln->ln_hold = NULL;
698 	ln->ln_state = ND6_LLINFO_INCOMPLETE;
699 	ln->ln_asked = 0;
700 }
701 
702 /*
703  * Free an nd6 llinfo entry.
704  * Since the function would cause significant changes in the kernel, DO NOT
705  * make it global, unless you have a strong reason for the change, and are sure
706  * that the change is safe.
707  */
708 void
709 nd6_free(struct rtentry *rt)
710 {
711 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
712 	struct in6_addr in6 = satosin6(rt_key(rt))->sin6_addr;
713 	struct ifnet *ifp;
714 
715 	NET_ASSERT_LOCKED();
716 
717 	ifp = if_get(rt->rt_ifidx);
718 
719 	if (!ip6_forwarding) {
720 		if (ln->ln_router) {
721 			/*
722 			 * rt6_flush must be called whether or not the neighbor
723 			 * is in the Default Router List.
724 			 * See a corresponding comment in nd6_na_input().
725 			 */
726 			rt6_flush(&in6, ifp);
727 		}
728 	}
729 
730 	KASSERT(!ISSET(rt->rt_flags, RTF_LOCAL));
731 	nd6_invalidate(rt);
732 
733 	/*
734 	 * Detach the route from the routing tree and the list of neighbor
735 	 * caches, and disable the route entry not to be used in already
736 	 * cached routes.
737 	 */
738 	if (!ISSET(rt->rt_flags, RTF_STATIC|RTF_CACHED))
739 		rtdeletemsg(rt, ifp, ifp->if_rdomain);
740 
741 	if_put(ifp);
742 }
743 
744 /*
745  * Upper-layer reachability hint for Neighbor Unreachability Detection.
746  *
747  * XXX cost-effective methods?
748  */
749 void
750 nd6_nud_hint(struct rtentry *rt)
751 {
752 	struct llinfo_nd6 *ln;
753 	struct ifnet *ifp;
754 
755 	ifp = if_get(rt->rt_ifidx);
756 	if (ifp == NULL)
757 		return;
758 
759 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
760 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
761 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
762 	    rt->rt_gateway->sa_family != AF_LINK) {
763 		/* This is not a host route. */
764 		goto out;
765 	}
766 
767 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
768 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
769 		goto out;
770 
771 	/*
772 	 * if we get upper-layer reachability confirmation many times,
773 	 * it is possible we have false information.
774 	 */
775 	ln->ln_byhint++;
776 	if (ln->ln_byhint > nd6_maxnudhint)
777 		goto out;
778 
779 	ln->ln_state = ND6_LLINFO_REACHABLE;
780 	if (!ND6_LLINFO_PERMANENT(ln))
781 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->reachable);
782 out:
783 	if_put(ifp);
784 }
785 
786 void
787 nd6_rtrequest(struct ifnet *ifp, int req, struct rtentry *rt)
788 {
789 	struct sockaddr *gate = rt->rt_gateway;
790 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
791 	struct ifaddr *ifa;
792 
793 	if (ISSET(rt->rt_flags, RTF_GATEWAY|RTF_MULTICAST|RTF_MPLS))
794 		return;
795 
796 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
797 		/*
798 		 * This is probably an interface direct route for a link
799 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
800 		 * We do not need special treatment below for such a route.
801 		 * Moreover, the RTF_LLINFO flag which would be set below
802 		 * would annoy the ndp(8) command.
803 		 */
804 		return;
805 	}
806 
807 	if (req == RTM_RESOLVE && nd6_need_cache(ifp) == 0) {
808 		/*
809 		 * For routing daemons like ospf6d we allow neighbor discovery
810 		 * based on the cloning route only.  This allows us to sent
811 		 * packets directly into a network without having an address
812 		 * with matching prefix on the interface.  If the cloning
813 		 * route is used for an stf interface, we would mistakenly
814 		 * make a neighbor cache for the host route, and would see
815 		 * strange neighbor solicitation for the corresponding
816 		 * destination.  In order to avoid confusion, we check if the
817 		 * interface is suitable for neighbor discovery, and stop the
818 		 * process if not.  Additionally, we remove the LLINFO flag
819 		 * so that ndp(8) will not try to get the neighbor information
820 		 * of the destination.
821 		 */
822 		rt->rt_flags &= ~RTF_LLINFO;
823 		return;
824 	}
825 
826 	switch (req) {
827 	case RTM_ADD:
828 		if ((rt->rt_flags & RTF_CLONING) ||
829 		    ((rt->rt_flags & (RTF_LLINFO | RTF_LOCAL)) && ln == NULL)) {
830 			if (ln != NULL)
831 				nd6_llinfo_settimer(ln, 0);
832 			if ((rt->rt_flags & RTF_CLONING) != 0)
833 				break;
834 		}
835 		/*
836 		 * In IPv4 code, we try to announce new RTF_ANNOUNCE entry here.
837 		 * We don't do that here since llinfo is not ready yet.
838 		 *
839 		 * There are also couple of other things to be discussed:
840 		 * - unsolicited NA code needs improvement beforehand
841 		 * - RFC2461 says we MAY send multicast unsolicited NA
842 		 *   (7.2.6 paragraph 4), however, it also says that we
843 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
844 		 *   we don't have anything like it right now.
845 		 *   note that the mechanism needs a mutual agreement
846 		 *   between proxies, which means that we need to implement
847 		 *   a new protocol, or a new kludge.
848 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
849 		 *   we need to check ip6forwarding before sending it.
850 		 *   (or should we allow proxy ND configuration only for
851 		 *   routers?  there's no mention about proxy ND from hosts)
852 		 */
853 #if 0
854 		/* XXX it does not work */
855 		if (rt->rt_flags & RTF_ANNOUNCE)
856 			nd6_na_output(ifp,
857 			      &satosin6(rt_key(rt))->sin6_addr,
858 			      &satosin6(rt_key(rt))->sin6_addr,
859 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
860 			      1, NULL);
861 #endif
862 		/* FALLTHROUGH */
863 	case RTM_RESOLVE:
864 		if (gate->sa_family != AF_LINK ||
865 		    gate->sa_len < sizeof(struct sockaddr_dl)) {
866 			log(LOG_DEBUG, "%s: bad gateway value: %s\n",
867 			    __func__, ifp->if_xname);
868 			break;
869 		}
870 		satosdl(gate)->sdl_type = ifp->if_type;
871 		satosdl(gate)->sdl_index = ifp->if_index;
872 		if (ln != NULL)
873 			break;	/* This happens on a route change */
874 		/*
875 		 * Case 2: This route may come from cloning, or a manual route
876 		 * add with a LL address.
877 		 */
878 		ln = pool_get(&nd6_pool, PR_NOWAIT | PR_ZERO);
879 		rt->rt_llinfo = (caddr_t)ln;
880 		if (ln == NULL) {
881 			log(LOG_DEBUG, "%s: pool get failed\n", __func__);
882 			break;
883 		}
884 		nd6_inuse++;
885 		nd6_allocated++;
886 		ln->ln_rt = rt;
887 		/* this is required for "ndp" command. - shin */
888 		if (req == RTM_ADD) {
889 		        /*
890 			 * gate should have some valid AF_LINK entry,
891 			 * and ln expire should have some lifetime
892 			 * which is specified by ndp command.
893 			 */
894 			ln->ln_state = ND6_LLINFO_REACHABLE;
895 			ln->ln_byhint = 0;
896 		} else {
897 		        /*
898 			 * When req == RTM_RESOLVE, rt is created and
899 			 * initialized in rtrequest(), so rt_expire is 0.
900 			 */
901 			ln->ln_state = ND6_LLINFO_NOSTATE;
902 			nd6_llinfo_settimer(ln, 0);
903 		}
904 		rt->rt_flags |= RTF_LLINFO;
905 		TAILQ_INSERT_HEAD(&nd6_list, ln, ln_list);
906 
907 		/*
908 		 * If we have too many cache entries, initiate immediate
909 		 * purging for some "less recently used" entries.  Note that
910 		 * we cannot directly call nd6_free() here because it would
911 		 * cause re-entering rtable related routines triggering an LOR
912 		 * problem for FreeBSD.
913 		 */
914 		if (ip6_neighborgcthresh >= 0 &&
915 		    nd6_inuse >= ip6_neighborgcthresh) {
916 			int i;
917 
918 			for (i = 0; i < 10; i++) {
919 				struct llinfo_nd6 *ln_end;
920 
921 				ln_end = TAILQ_LAST(&nd6_list, llinfo_nd6_head);
922 				if (ln_end == ln)
923 					break;
924 
925 				/* Move this entry to the head */
926 				TAILQ_REMOVE(&nd6_list, ln_end, ln_list);
927 				TAILQ_INSERT_HEAD(&nd6_list, ln_end, ln_list);
928 
929 				if (ND6_LLINFO_PERMANENT(ln_end))
930 					continue;
931 
932 				if (ln_end->ln_state > ND6_LLINFO_INCOMPLETE)
933 					ln_end->ln_state = ND6_LLINFO_STALE;
934 				else
935 					ln_end->ln_state = ND6_LLINFO_PURGE;
936 				nd6_llinfo_settimer(ln_end, 0);
937 			}
938 		}
939 
940 		/*
941 		 * check if rt_key(rt) is one of my address assigned
942 		 * to the interface.
943 		 */
944 		ifa = &in6ifa_ifpwithaddr(ifp,
945 		    &satosin6(rt_key(rt))->sin6_addr)->ia_ifa;
946 		if (ifa) {
947 			ln->ln_state = ND6_LLINFO_REACHABLE;
948 			ln->ln_byhint = 0;
949 			rt->rt_expire = 0;
950 			KASSERT(ifa == rt->rt_ifa);
951 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
952 			ln->ln_state = ND6_LLINFO_REACHABLE;
953 			ln->ln_byhint = 0;
954 			rt->rt_expire = 0;
955 
956 			/* join solicited node multicast for proxy ND */
957 			if (ifp->if_flags & IFF_MULTICAST) {
958 				struct in6_addr llsol;
959 				int error;
960 
961 				llsol = satosin6(rt_key(rt))->sin6_addr;
962 				llsol.s6_addr16[0] = htons(0xff02);
963 				llsol.s6_addr16[1] = htons(ifp->if_index);
964 				llsol.s6_addr32[1] = 0;
965 				llsol.s6_addr32[2] = htonl(1);
966 				llsol.s6_addr8[12] = 0xff;
967 
968 				if (in6_addmulti(&llsol, ifp, &error)) {
969 					char addr[INET6_ADDRSTRLEN];
970 					nd6log((LOG_ERR, "%s: failed to join "
971 					    "%s (errno=%d)\n", ifp->if_xname,
972 					    inet_ntop(AF_INET6, &llsol,
973 						addr, sizeof(addr)),
974 					    error));
975 				}
976 			}
977 		}
978 		break;
979 
980 	case RTM_DELETE:
981 		if (ln == NULL)
982 			break;
983 		/* leave from solicited node multicast for proxy ND */
984 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
985 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
986 			struct in6_addr llsol;
987 			struct in6_multi *in6m;
988 
989 			llsol = satosin6(rt_key(rt))->sin6_addr;
990 			llsol.s6_addr16[0] = htons(0xff02);
991 			llsol.s6_addr16[1] = htons(ifp->if_index);
992 			llsol.s6_addr32[1] = 0;
993 			llsol.s6_addr32[2] = htonl(1);
994 			llsol.s6_addr8[12] = 0xff;
995 
996 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
997 			if (in6m)
998 				in6_delmulti(in6m);
999 		}
1000 		nd6_inuse--;
1001 		TAILQ_REMOVE(&nd6_list, ln, ln_list);
1002 		rt->rt_expire = 0;
1003 		rt->rt_llinfo = NULL;
1004 		rt->rt_flags &= ~RTF_LLINFO;
1005 		m_freem(ln->ln_hold);
1006 		pool_put(&nd6_pool, ln);
1007 		break;
1008 
1009 	case RTM_INVALIDATE:
1010 		if (ln == NULL)
1011 			break;
1012 		if (!ISSET(rt->rt_flags, RTF_LOCAL))
1013 			nd6_invalidate(rt);
1014 		break;
1015 	}
1016 }
1017 
1018 int
1019 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1020 {
1021 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1022 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1023 	struct rtentry *rt;
1024 
1025 	switch (cmd) {
1026 	case SIOCGIFINFO_IN6:
1027 		NET_RLOCK_IN_IOCTL();
1028 		ndi->ndi = *ND_IFINFO(ifp);
1029 		NET_RUNLOCK_IN_IOCTL();
1030 		return (0);
1031 	case SIOCGNBRINFO_IN6:
1032 	{
1033 		struct llinfo_nd6 *ln;
1034 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1035 		time_t expire;
1036 
1037 		NET_RLOCK_IN_IOCTL();
1038 		/*
1039 		 * XXX: KAME specific hack for scoped addresses
1040 		 *      XXXX: for other scopes than link-local?
1041 		 */
1042 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1043 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1044 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1045 
1046 			if (*idp == 0)
1047 				*idp = htons(ifp->if_index);
1048 		}
1049 
1050 		rt = nd6_lookup(&nb_addr, 0, ifp, ifp->if_rdomain);
1051 		if (rt == NULL ||
1052 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1053 			rtfree(rt);
1054 			NET_RUNLOCK_IN_IOCTL();
1055 			return (EINVAL);
1056 		}
1057 		expire = ln->ln_rt->rt_expire;
1058 		if (expire != 0) {
1059 			expire -= getuptime();
1060 			expire += gettime();
1061 		}
1062 
1063 		nbi->state = ln->ln_state;
1064 		nbi->asked = ln->ln_asked;
1065 		nbi->isrouter = ln->ln_router;
1066 		nbi->expire = expire;
1067 
1068 		rtfree(rt);
1069 		NET_RUNLOCK_IN_IOCTL();
1070 		return (0);
1071 	}
1072 	}
1073 	return (0);
1074 }
1075 
1076 /*
1077  * Create neighbor cache entry and cache link-layer address,
1078  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1079  *
1080  * type - ICMP6 type
1081  * code - type dependent information
1082  */
1083 void
1084 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1085     int lladdrlen, int type, int code)
1086 {
1087 	struct rtentry *rt = NULL;
1088 	struct llinfo_nd6 *ln = NULL;
1089 	int is_newentry;
1090 	struct sockaddr_dl *sdl = NULL;
1091 	int do_update;
1092 	int olladdr;
1093 	int llchange;
1094 	int newstate = 0;
1095 
1096 	if (!ifp)
1097 		panic("%s: ifp == NULL", __func__);
1098 	if (!from)
1099 		panic("%s: from == NULL", __func__);
1100 
1101 	/* nothing must be updated for unspecified address */
1102 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1103 		return;
1104 
1105 	/*
1106 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1107 	 * the caller.
1108 	 *
1109 	 * XXX If the link does not have link-layer address, what should
1110 	 * we do? (ifp->if_addrlen == 0)
1111 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1112 	 * description on it in NS section (RFC 2461 7.2.3).
1113 	 */
1114 
1115 	rt = nd6_lookup(from, 0, ifp, ifp->if_rdomain);
1116 	if (rt == NULL) {
1117 		rt = nd6_lookup(from, 1, ifp, ifp->if_rdomain);
1118 		is_newentry = 1;
1119 	} else {
1120 		/* do not overwrite local or static entry */
1121 		if (ISSET(rt->rt_flags, RTF_STATIC|RTF_LOCAL)) {
1122 			rtfree(rt);
1123 			return;
1124 		}
1125 		is_newentry = 0;
1126 	}
1127 
1128 	if (!rt)
1129 		return;
1130 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1131 fail:
1132 		nd6_free(rt);
1133 		rtfree(rt);
1134 		return;
1135 	}
1136 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1137 	if (ln == NULL)
1138 		goto fail;
1139 	if (rt->rt_gateway == NULL)
1140 		goto fail;
1141 	if (rt->rt_gateway->sa_family != AF_LINK)
1142 		goto fail;
1143 	sdl = satosdl(rt->rt_gateway);
1144 
1145 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1146 	if (olladdr && lladdr) {
1147 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1148 			llchange = 1;
1149 		else
1150 			llchange = 0;
1151 	} else
1152 		llchange = 0;
1153 
1154 	/*
1155 	 * newentry olladdr  lladdr  llchange	(*=record)
1156 	 *	0	n	n	--	(1)
1157 	 *	0	y	n	--	(2)
1158 	 *	0	n	y	--	(3) * STALE
1159 	 *	0	y	y	n	(4) *
1160 	 *	0	y	y	y	(5) * STALE
1161 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1162 	 *	1	--	y	--	(7) * STALE
1163 	 */
1164 
1165 	if (llchange) {
1166 		char addr[INET6_ADDRSTRLEN];
1167 		log(LOG_INFO, "ndp info overwritten for %s by %s on %s\n",
1168 		    inet_ntop(AF_INET6, from, addr, sizeof(addr)),
1169 		    ether_sprintf(lladdr), ifp->if_xname);
1170 	}
1171 	if (lladdr) {		/* (3-5) and (7) */
1172 		/*
1173 		 * Record source link-layer address
1174 		 * XXX is it dependent to ifp->if_type?
1175 		 */
1176 		sdl->sdl_alen = ifp->if_addrlen;
1177 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1178 	}
1179 
1180 	if (!is_newentry) {
1181 		if ((!olladdr && lladdr) ||		/* (3) */
1182 		    (olladdr && lladdr && llchange)) {	/* (5) */
1183 			do_update = 1;
1184 			newstate = ND6_LLINFO_STALE;
1185 		} else					/* (1-2,4) */
1186 			do_update = 0;
1187 	} else {
1188 		do_update = 1;
1189 		if (!lladdr)				/* (6) */
1190 			newstate = ND6_LLINFO_NOSTATE;
1191 		else					/* (7) */
1192 			newstate = ND6_LLINFO_STALE;
1193 	}
1194 
1195 	if (do_update) {
1196 		/*
1197 		 * Update the state of the neighbor cache.
1198 		 */
1199 		ln->ln_state = newstate;
1200 
1201 		if (ln->ln_state == ND6_LLINFO_STALE) {
1202 			/*
1203 			 * Since nd6_resolve() in ifp->if_output() will cause
1204 			 * state transition to DELAY and reset the timer,
1205 			 * we must set the timer now, although it is actually
1206 			 * meaningless.
1207 			 */
1208 			nd6_llinfo_settimer(ln, nd6_gctimer);
1209 
1210 			if (ln->ln_hold) {
1211 				struct mbuf *n = ln->ln_hold;
1212 				ln->ln_hold = NULL;
1213 				/*
1214 				 * we assume ifp is not a p2p here, so just
1215 				 * set the 2nd argument as the 1st one.
1216 				 */
1217 				ifp->if_output(ifp, n, rt_key(rt), rt);
1218 				if (ln->ln_hold == n) {
1219 					/* n is back in ln_hold. Discard. */
1220 					m_freem(ln->ln_hold);
1221 					ln->ln_hold = NULL;
1222 				}
1223 			}
1224 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1225 			/* probe right away */
1226 			nd6_llinfo_settimer(ln, 0);
1227 		}
1228 	}
1229 
1230 	/*
1231 	 * ICMP6 type dependent behavior.
1232 	 *
1233 	 * NS: clear IsRouter if new entry
1234 	 * RS: clear IsRouter
1235 	 * RA: set IsRouter if there's lladdr
1236 	 * redir: clear IsRouter if new entry
1237 	 *
1238 	 * RA case, (1):
1239 	 * The spec says that we must set IsRouter in the following cases:
1240 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1241 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1242 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1243 	 * A question arises for (1) case.  (1) case has no lladdr in the
1244 	 * neighbor cache, this is similar to (6).
1245 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1246 	 *
1247 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1248 	 *							D R
1249 	 *	0	n	n	--	(1)	c   ?     s
1250 	 *	0	y	n	--	(2)	c   s     s
1251 	 *	0	n	y	--	(3)	c   s     s
1252 	 *	0	y	y	n	(4)	c   s     s
1253 	 *	0	y	y	y	(5)	c   s     s
1254 	 *	1	--	n	--	(6) c	c	c s
1255 	 *	1	--	y	--	(7) c	c   s	c s
1256 	 *
1257 	 *					(c=clear s=set)
1258 	 */
1259 	switch (type & 0xff) {
1260 	case ND_NEIGHBOR_SOLICIT:
1261 		/*
1262 		 * New entry must have is_router flag cleared.
1263 		 */
1264 		if (is_newentry)	/* (6-7) */
1265 			ln->ln_router = 0;
1266 		break;
1267 	case ND_REDIRECT:
1268 		/*
1269 		 * If the icmp is a redirect to a better router, always set the
1270 		 * is_router flag.  Otherwise, if the entry is newly created,
1271 		 * clear the flag.  [RFC 2461, sec 8.3]
1272 		 */
1273 		if (code == ND_REDIRECT_ROUTER)
1274 			ln->ln_router = 1;
1275 		else if (is_newentry) /* (6-7) */
1276 			ln->ln_router = 0;
1277 		break;
1278 	case ND_ROUTER_SOLICIT:
1279 		/*
1280 		 * is_router flag must always be cleared.
1281 		 */
1282 		ln->ln_router = 0;
1283 		break;
1284 	case ND_ROUTER_ADVERT:
1285 		/*
1286 		 * Mark an entry with lladdr as a router.
1287 		 */
1288 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1289 		    (is_newentry && lladdr)) {			/* (7) */
1290 			ln->ln_router = 1;
1291 		}
1292 		break;
1293 	}
1294 
1295 	rtfree(rt);
1296 }
1297 
1298 void
1299 nd6_slowtimo(void *ignored_arg)
1300 {
1301 	struct nd_ifinfo *nd6if;
1302 	struct ifnet *ifp;
1303 
1304 	NET_LOCK();
1305 
1306 	timeout_add_sec(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL);
1307 
1308 	TAILQ_FOREACH(ifp, &ifnet, if_list) {
1309 		nd6if = ND_IFINFO(ifp);
1310 		if (nd6if->basereachable && /* already initialized */
1311 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1312 			/*
1313 			 * Since reachable time rarely changes by router
1314 			 * advertisements, we SHOULD insure that a new random
1315 			 * value gets recomputed at least once every few hours.
1316 			 * (RFC 2461, 6.3.4)
1317 			 */
1318 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1319 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1320 		}
1321 	}
1322 	NET_UNLOCK();
1323 }
1324 
1325 int
1326 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1327     struct sockaddr *dst, u_char *desten)
1328 {
1329 	struct sockaddr_dl *sdl;
1330 	struct rtentry *rt;
1331 	struct llinfo_nd6 *ln = NULL;
1332 
1333 	if (m->m_flags & M_MCAST) {
1334 		ETHER_MAP_IPV6_MULTICAST(&satosin6(dst)->sin6_addr, desten);
1335 		return (0);
1336 	}
1337 
1338 	rt = rt_getll(rt0);
1339 
1340 	if (ISSET(rt->rt_flags, RTF_REJECT) &&
1341 	    (rt->rt_expire == 0 || getuptime() < rt->rt_expire)) {
1342 		m_freem(m);
1343 		return (rt == rt0 ? EHOSTDOWN : EHOSTUNREACH);
1344 	}
1345 
1346 	/*
1347 	 * Address resolution or Neighbor Unreachability Detection
1348 	 * for the next hop.
1349 	 * At this point, the destination of the packet must be a unicast
1350 	 * or an anycast address(i.e. not a multicast).
1351 	 */
1352 	if (!ISSET(rt->rt_flags, RTF_LLINFO)) {
1353 		char addr[INET6_ADDRSTRLEN];
1354 		log(LOG_DEBUG, "%s: %s: route contains no ND information\n",
1355 		    __func__, inet_ntop(AF_INET6,
1356 		    &satosin6(rt_key(rt))->sin6_addr, addr, sizeof(addr)));
1357 		m_freem(m);
1358 		return (EINVAL);
1359 	}
1360 
1361 	if (rt->rt_gateway->sa_family != AF_LINK) {
1362 		printf("%s: something odd happens\n", __func__);
1363 		m_freem(m);
1364 		return (EINVAL);
1365 	}
1366 
1367 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1368 	KASSERT(ln != NULL);
1369 
1370 	/*
1371 	 * Move this entry to the head of the queue so that it is less likely
1372 	 * for this entry to be a target of forced garbage collection (see
1373 	 * nd6_rtrequest()).
1374 	 */
1375 	TAILQ_REMOVE(&nd6_list, ln, ln_list);
1376 	TAILQ_INSERT_HEAD(&nd6_list, ln, ln_list);
1377 
1378 	/*
1379 	 * The first time we send a packet to a neighbor whose entry is
1380 	 * STALE, we have to change the state to DELAY and a sets a timer to
1381 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1382 	 * neighbor unreachability detection on expiration.
1383 	 * (RFC 2461 7.3.3)
1384 	 */
1385 	if (ln->ln_state == ND6_LLINFO_STALE) {
1386 		ln->ln_asked = 0;
1387 		ln->ln_state = ND6_LLINFO_DELAY;
1388 		nd6_llinfo_settimer(ln, nd6_delay);
1389 	}
1390 
1391 	/*
1392 	 * If the neighbor cache entry has a state other than INCOMPLETE
1393 	 * (i.e. its link-layer address is already resolved), just
1394 	 * send the packet.
1395 	 */
1396 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
1397 		sdl = satosdl(rt->rt_gateway);
1398 		if (sdl->sdl_alen != ETHER_ADDR_LEN) {
1399 			char addr[INET6_ADDRSTRLEN];
1400 			log(LOG_DEBUG, "%s: %s: incorrect nd6 information\n",
1401 			    __func__,
1402 			    inet_ntop(AF_INET6, &satosin6(dst)->sin6_addr,
1403 				addr, sizeof(addr)));
1404 			m_freem(m);
1405 			return (EINVAL);
1406 		}
1407 
1408 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
1409 		return (0);
1410 	}
1411 
1412 	/*
1413 	 * There is a neighbor cache entry, but no ethernet address
1414 	 * response yet.  Replace the held mbuf (if any) with this
1415 	 * latest one.
1416 	 */
1417 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1418 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1419 	m_freem(ln->ln_hold);
1420 	ln->ln_hold = m;
1421 
1422 	/*
1423 	 * If there has been no NS for the neighbor after entering the
1424 	 * INCOMPLETE state, send the first solicitation.
1425 	 */
1426 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
1427 		ln->ln_asked++;
1428 		nd6_llinfo_settimer(ln, ND_IFINFO(ifp)->retrans / 1000);
1429 		nd6_ns_output(ifp, NULL, &satosin6(dst)->sin6_addr, ln, 0);
1430 	}
1431 	return (EAGAIN);
1432 }
1433 
1434 int
1435 nd6_need_cache(struct ifnet *ifp)
1436 {
1437 	/*
1438 	 * RFC2893 says:
1439 	 * - unidirectional tunnels needs no ND
1440 	 */
1441 	switch (ifp->if_type) {
1442 	case IFT_ETHER:
1443 	case IFT_IEEE80211:
1444 	case IFT_CARP:
1445 		return (1);
1446 	default:
1447 		return (0);
1448 	}
1449 }
1450