xref: /dflybsd-src/sys/netinet6/nd6.c (revision 7d69241457e99e4be5ba4af822ec46f8732fba73)
1 /*	$FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $	*/
2 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 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 "opt_inet.h"
34 #include "opt_inet6.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/callout.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/syslog.h>
48 #include <sys/queue.h>
49 #include <sys/sysctl.h>
50 #include <sys/mutex.h>
51 
52 #include <sys/thread2.h>
53 #include <sys/mutex2.h>
54 
55 #include <net/if.h>
56 #include <net/if_dl.h>
57 #include <net/if_types.h>
58 #include <net/route.h>
59 #include <net/netisr2.h>
60 #include <net/netmsg2.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69 
70 #include <net/net_osdep.h>
71 
72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
74 
75 #define SIN6(s) ((struct sockaddr_in6 *)s)
76 #define SDL(s) ((struct sockaddr_dl *)s)
77 
78 /*
79  * Note that the check for rt_llinfo is necessary because a cloned
80  * route from a parent route that has the L flag (e.g. the default
81  * route to a p2p interface) may have the flag, too, while the
82  * destination is not actually a neighbor.
83  * XXX: we can't use rt->rt_ifp to check for the interface, since
84  *      it might be the loopback interface if the entry is for our
85  *      own address on a non-loopback interface. Instead, we should
86  *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
87  *      interface.
88  */
89 #define ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)			\
90 	!(((rt)->rt_flags & RTF_GATEWAY) ||			\
91 	  ((rt)->rt_flags & RTF_LLINFO) == 0 ||			\
92 	  (rt)->rt_gateway->sa_family != AF_LINK ||		\
93 	  (rt)->rt_llinfo == NULL ||				\
94 	  ((ifp) != NULL && (rt)->rt_ifa->ifa_ifp != (ifp)))
95 
96 #define ND6_RTENTRY_IS_LLCLONING(rt)				\
97 	(((rt)->rt_flags & (RTF_PRCLONING | RTF_LLINFO)) ==	\
98 	 (RTF_PRCLONING | RTF_LLINFO) ||			\
99 	 ((rt)->rt_flags & RTF_CLONING))
100 
101 /* timer values */
102 int	nd6_prune	= 1;	/* walk list every 1 seconds */
103 int	nd6_delay	= 5;	/* delay first probe time 5 second */
104 int	nd6_umaxtries	= 3;	/* maximum unicast query */
105 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
106 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
107 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
108 
109 /* preventing too many loops in ND option parsing */
110 int nd6_maxndopt = 10;	/* max # of ND options allowed */
111 
112 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
113 
114 #ifdef ND6_DEBUG
115 int nd6_debug = 1;
116 #else
117 int nd6_debug = 0;
118 #endif
119 
120 /* for debugging? */
121 static int nd6_inuse, nd6_allocated;
122 
123 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
124 struct nd_drhead nd_defrouter;
125 struct nd_prhead nd_prefix = { 0 };
126 struct mtx nd6_mtx = MTX_INITIALIZER("nd6");
127 
128 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
129 static struct sockaddr_in6 all1_sa;
130 
131 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
132 static int regen_tmpaddr (struct in6_ifaddr *);
133 static void nd6_slowtimo(void *);
134 static void nd6_slowtimo_dispatch(netmsg_t);
135 static void nd6_timer(void *);
136 static void nd6_timer_dispatch(netmsg_t);
137 
138 static struct callout nd6_slowtimo_ch;
139 static struct netmsg_base nd6_slowtimo_netmsg;
140 
141 static struct callout nd6_timer_ch;
142 static struct netmsg_base nd6_timer_netmsg;
143 
144 void
145 nd6_init(void)
146 {
147 	static int nd6_init_done = 0;
148 	int i;
149 
150 	if (nd6_init_done) {
151 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
152 		return;
153 	}
154 
155 	all1_sa.sin6_family = AF_INET6;
156 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
157 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
158 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
159 
160 	/* initialization of the default router list */
161 	TAILQ_INIT(&nd_defrouter);
162 
163 	nd6_init_done = 1;
164 
165 	/* start timer */
166 	callout_init_mp(&nd6_slowtimo_ch);
167 	netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
168 	    MSGF_PRIORITY, nd6_slowtimo_dispatch);
169 	callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
170 	    nd6_slowtimo, NULL, 0);
171 }
172 
173 struct nd_ifinfo *
174 nd6_ifattach(struct ifnet *ifp)
175 {
176 	struct nd_ifinfo *nd;
177 
178 	nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
179 	    M_WAITOK | M_ZERO);
180 
181 	nd->initialized = 1;
182 
183 	nd->chlim = IPV6_DEFHLIM;
184 	nd->basereachable = REACHABLE_TIME;
185 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
186 	nd->retrans = RETRANS_TIMER;
187 
188 	/*
189 	 * Note that the default value of ip6_accept_rtadv is 0, which means
190 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
191 	 * here.
192 	 */
193 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
194 
195 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
196 	nd6_setmtu0(ifp, nd);
197 	return nd;
198 }
199 
200 void
201 nd6_ifdetach(struct nd_ifinfo *nd)
202 {
203 	kfree(nd, M_IP6NDP);
204 }
205 
206 /*
207  * Reset ND level link MTU. This function is called when the physical MTU
208  * changes, which means we might have to adjust the ND level MTU.
209  */
210 void
211 nd6_setmtu(struct ifnet *ifp)
212 {
213 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
214 }
215 
216 struct netmsg_nd6setmtu {
217 	struct netmsg_base	nmsg;
218 	struct ifnet		*ifp;
219 	struct nd_ifinfo	*ndi;
220 };
221 
222 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
223 static void
224 nd6_setmtu0_dispatch(netmsg_t msg)
225 {
226 	struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
227 	struct ifnet *ifp = nmsg->ifp;
228 	struct nd_ifinfo *ndi = nmsg->ndi;
229 	uint32_t omaxmtu;
230 
231 	omaxmtu = ndi->maxmtu;
232 
233 	switch (ifp->if_type) {
234 	case IFT_ETHER:
235 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
236 		break;
237 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
238 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
239 		break;
240 #ifdef IFT_IEEE80211
241 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
242 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
243 		break;
244 #endif
245 	default:
246 		ndi->maxmtu = ifp->if_mtu;
247 		break;
248 	}
249 
250 	/*
251 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
252 	 * undesirable situation.  We thus notify the operator of the change
253 	 * explicitly.  The check for omaxmtu is necessary to restrict the
254 	 * log to the case of changing the MTU, not initializing it.
255 	 */
256 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
257 		log(LOG_NOTICE, "nd6_setmtu0: "
258 		    "new link MTU on %s (%lu) is too small for IPv6\n",
259 		    if_name(ifp), (unsigned long)ndi->maxmtu);
260 	}
261 
262 	if (ndi->maxmtu > in6_maxmtu)
263 		in6_setmaxmtu(); /* check all interfaces just in case */
264 
265 	lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
266 }
267 
268 void
269 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
270 {
271 	struct netmsg_nd6setmtu nmsg;
272 
273 	netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
274 	    nd6_setmtu0_dispatch);
275 	nmsg.ifp = ifp;
276 	nmsg.ndi = ndi;
277 	lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
278 }
279 
280 void
281 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
282 {
283 	bzero(ndopts, sizeof(*ndopts));
284 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
285 	ndopts->nd_opts_last
286 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
287 
288 	if (icmp6len == 0) {
289 		ndopts->nd_opts_done = 1;
290 		ndopts->nd_opts_search = NULL;
291 	}
292 }
293 
294 /*
295  * Take one ND option.
296  */
297 struct nd_opt_hdr *
298 nd6_option(union nd_opts *ndopts)
299 {
300 	struct nd_opt_hdr *nd_opt;
301 	int olen;
302 
303 	if (!ndopts)
304 		panic("ndopts == NULL in nd6_option");
305 	if (!ndopts->nd_opts_last)
306 		panic("uninitialized ndopts in nd6_option");
307 	if (!ndopts->nd_opts_search)
308 		return NULL;
309 	if (ndopts->nd_opts_done)
310 		return NULL;
311 
312 	nd_opt = ndopts->nd_opts_search;
313 
314 	/* make sure nd_opt_len is inside the buffer */
315 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
316 		bzero(ndopts, sizeof(*ndopts));
317 		return NULL;
318 	}
319 
320 	olen = nd_opt->nd_opt_len << 3;
321 	if (olen == 0) {
322 		/*
323 		 * Message validation requires that all included
324 		 * options have a length that is greater than zero.
325 		 */
326 		bzero(ndopts, sizeof(*ndopts));
327 		return NULL;
328 	}
329 
330 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
331 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
332 		/* option overruns the end of buffer, invalid */
333 		bzero(ndopts, sizeof(*ndopts));
334 		return NULL;
335 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
336 		/* reached the end of options chain */
337 		ndopts->nd_opts_done = 1;
338 		ndopts->nd_opts_search = NULL;
339 	}
340 	return nd_opt;
341 }
342 
343 /*
344  * Parse multiple ND options.
345  * This function is much easier to use, for ND routines that do not need
346  * multiple options of the same type.
347  */
348 int
349 nd6_options(union nd_opts *ndopts)
350 {
351 	struct nd_opt_hdr *nd_opt;
352 	int i = 0;
353 
354 	if (!ndopts)
355 		panic("ndopts == NULL in nd6_options");
356 	if (!ndopts->nd_opts_last)
357 		panic("uninitialized ndopts in nd6_options");
358 	if (!ndopts->nd_opts_search)
359 		return 0;
360 
361 	while (1) {
362 		nd_opt = nd6_option(ndopts);
363 		if (!nd_opt && !ndopts->nd_opts_last) {
364 			/*
365 			 * Message validation requires that all included
366 			 * options have a length that is greater than zero.
367 			 */
368 			icmp6stat.icp6s_nd_badopt++;
369 			bzero(ndopts, sizeof(*ndopts));
370 			return -1;
371 		}
372 
373 		if (!nd_opt)
374 			goto skip1;
375 
376 		switch (nd_opt->nd_opt_type) {
377 		case ND_OPT_SOURCE_LINKADDR:
378 		case ND_OPT_TARGET_LINKADDR:
379 		case ND_OPT_MTU:
380 		case ND_OPT_REDIRECTED_HEADER:
381 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
382 				nd6log((LOG_INFO,
383 				    "duplicated ND6 option found (type=%d)\n",
384 				    nd_opt->nd_opt_type));
385 				/* XXX bark? */
386 			} else {
387 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
388 					= nd_opt;
389 			}
390 			break;
391 		case ND_OPT_PREFIX_INFORMATION:
392 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
393 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
394 					= nd_opt;
395 			}
396 			ndopts->nd_opts_pi_end =
397 				(struct nd_opt_prefix_info *)nd_opt;
398 			break;
399 		default:
400 			/*
401 			 * Unknown options must be silently ignored,
402 			 * to accomodate future extension to the protocol.
403 			 */
404 			nd6log((LOG_DEBUG,
405 			    "nd6_options: unsupported option %d - "
406 			    "option ignored\n", nd_opt->nd_opt_type));
407 		}
408 
409 skip1:
410 		i++;
411 		if (i > nd6_maxndopt) {
412 			icmp6stat.icp6s_nd_toomanyopt++;
413 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
414 			break;
415 		}
416 
417 		if (ndopts->nd_opts_done)
418 			break;
419 	}
420 
421 	return 0;
422 }
423 
424 /*
425  * ND6 timer routine to expire default route list and prefix list
426  */
427 static void
428 nd6_timer_dispatch(netmsg_t nmsg)
429 {
430 	struct llinfo_nd6 *ln;
431 	struct nd_defrouter *dr;
432 	struct nd_prefix *pr;
433 	struct ifnet *ifp;
434 	struct in6_ifaddr *ia6, *nia6;
435 
436 	ASSERT_NETISR0;
437 
438 	crit_enter();
439 	lwkt_replymsg(&nmsg->lmsg, 0);	/* reply ASAP */
440 	crit_exit();
441 
442 	mtx_lock(&nd6_mtx);
443 
444 	ln = llinfo_nd6.ln_next;
445 	while (ln && ln != &llinfo_nd6) {
446 		struct rtentry *rt;
447 		struct sockaddr_in6 *dst;
448 		struct llinfo_nd6 *next = ln->ln_next;
449 		/* XXX: used for the DELAY case only: */
450 		struct nd_ifinfo *ndi = NULL;
451 
452 		if ((rt = ln->ln_rt) == NULL) {
453 			ln = next;
454 			continue;
455 		}
456 		if ((ifp = rt->rt_ifp) == NULL) {
457 			ln = next;
458 			continue;
459 		}
460 		ndi = ND_IFINFO(ifp);
461 		dst = (struct sockaddr_in6 *)rt_key(rt);
462 
463 		if (ln->ln_expire > time_uptime) {
464 			ln = next;
465 			continue;
466 		}
467 
468 		/* sanity check */
469 		if (!rt)
470 			panic("rt=0 in nd6_timer(ln=%p)", ln);
471 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
472 			panic("rt_llinfo(%p) is not equal to ln(%p)",
473 			      rt->rt_llinfo, ln);
474 		if (!dst)
475 			panic("dst=0 in nd6_timer(ln=%p)", ln);
476 
477 		switch (ln->ln_state) {
478 		case ND6_LLINFO_WAITDELETE:
479 			next = nd6_free(rt);
480 			break;
481 		case ND6_LLINFO_INCOMPLETE:
482 			if (ln->ln_asked++ >= nd6_mmaxtries) {
483 				struct mbuf *m = ln->ln_hold;
484 				if (m) {
485 					if (rt->rt_ifp) {
486 						/*
487 						 * Fake rcvif to make ICMP error
488 						 * more helpful in diagnosing
489 						 * for the receiver.
490 						 * XXX: should we consider
491 						 * older rcvif?
492 						 */
493 						m->m_pkthdr.rcvif = rt->rt_ifp;
494 					}
495 					icmp6_error(m, ICMP6_DST_UNREACH,
496 						    ICMP6_DST_UNREACH_ADDR, 0);
497 					ln->ln_hold = NULL;
498 				}
499 				ln->ln_state = ND6_LLINFO_WAITDELETE;
500 				rt_rtmsg(RTM_MISS, rt, rt->rt_ifp, 0);
501 			}
502 			ln->ln_expire = time_uptime +
503 				ND_IFINFO(ifp)->retrans / 1000;
504 			nd6_ns_output(ifp, NULL, &dst->sin6_addr,
505 				ln, 0);
506 			break;
507 		case ND6_LLINFO_REACHABLE:
508 			if (ln->ln_expire) {
509 				ln->ln_state = ND6_LLINFO_STALE;
510 				ln->ln_expire = time_uptime + nd6_gctimer;
511 			}
512 			break;
513 
514 		case ND6_LLINFO_STALE:
515 			/* Garbage Collection(RFC 2461 5.3) */
516 			if (ln->ln_expire)
517 				next = nd6_free(rt);
518 			break;
519 
520 		case ND6_LLINFO_DELAY:
521 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
522 				/* We need NUD */
523 				ln->ln_asked = 1;
524 				ln->ln_state = ND6_LLINFO_PROBE;
525 				ln->ln_expire = time_uptime +
526 					ndi->retrans / 1000;
527 				nd6_ns_output(ifp, &dst->sin6_addr,
528 					      &dst->sin6_addr,
529 					      ln, 0);
530 			} else {
531 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
532 				ln->ln_expire = time_uptime + nd6_gctimer;
533 			}
534 			break;
535 		case ND6_LLINFO_PROBE:
536 			if (ln->ln_asked < nd6_umaxtries) {
537 				ln->ln_asked++;
538 				ln->ln_expire = time_uptime +
539 					ND_IFINFO(ifp)->retrans / 1000;
540 				nd6_ns_output(ifp, &dst->sin6_addr,
541 					       &dst->sin6_addr, ln, 0);
542 			} else {
543 				next = nd6_free(rt);
544 			}
545 			break;
546 		}
547 		ln = next;
548 	}
549 
550 	/* expire default router list */
551 	dr = TAILQ_FIRST(&nd_defrouter);
552 	while (dr) {
553 		if (dr->expire && dr->expire < time_uptime) {
554 			struct nd_defrouter *t;
555 			t = TAILQ_NEXT(dr, dr_entry);
556 			defrtrlist_del(dr);
557 			dr = t;
558 		} else {
559 			dr = TAILQ_NEXT(dr, dr_entry);
560 		}
561 	}
562 
563 	/*
564 	 * expire interface addresses.
565 	 * in the past the loop was inside prefix expiry processing.
566 	 * However, from a stricter speci-confrmance standpoint, we should
567 	 * rather separate address lifetimes and prefix lifetimes.
568 	 */
569 addrloop:
570 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
571 		nia6 = ia6->ia_next;
572 		/* check address lifetime */
573 		if (IFA6_IS_INVALID(ia6)) {
574 			int regen = 0;
575 
576 			/*
577 			 * If the expiring address is temporary, try
578 			 * regenerating a new one.  This would be useful when
579 			 * we suspended a laptop PC, then turned it on after a
580 			 * period that could invalidate all temporary
581 			 * addresses.  Although we may have to restart the
582 			 * loop (see below), it must be after purging the
583 			 * address.  Otherwise, we'd see an infinite loop of
584 			 * regeneration.
585 			 */
586 			if (ip6_use_tempaddr &&
587 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
588 				if (regen_tmpaddr(ia6) == 0)
589 					regen = 1;
590 			}
591 
592 			in6_purgeaddr(&ia6->ia_ifa);
593 
594 			if (regen)
595 				goto addrloop; /* XXX: see below */
596 		}
597 		if (IFA6_IS_DEPRECATED(ia6)) {
598 			int oldflags = ia6->ia6_flags;
599 
600 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
601 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
602 				in6_newaddrmsg((struct ifaddr *)ia6);
603 			}
604 
605 			/*
606 			 * If a temporary address has just become deprecated,
607 			 * regenerate a new one if possible.
608 			 */
609 			if (ip6_use_tempaddr &&
610 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
611 			    !(oldflags & IN6_IFF_DEPRECATED)) {
612 
613 				if (regen_tmpaddr(ia6) == 0) {
614 					/*
615 					 * A new temporary address is
616 					 * generated.
617 					 * XXX: this means the address chain
618 					 * has changed while we are still in
619 					 * the loop.  Although the change
620 					 * would not cause disaster (because
621 					 * it's not a deletion, but an
622 					 * addition,) we'd rather restart the
623 					 * loop just for safety.  Or does this
624 					 * significantly reduce performance??
625 					 */
626 					goto addrloop;
627 				}
628 			}
629 		} else {
630 			/*
631 			 * A new RA might have made a deprecated address
632 			 * preferred.
633 			 */
634 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
635 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
636 				in6_newaddrmsg((struct ifaddr *)ia6);
637 			}
638 		}
639 	}
640 
641 	/* expire prefix list */
642 	pr = nd_prefix.lh_first;
643 	while (pr) {
644 		/*
645 		 * check prefix lifetime.
646 		 * since pltime is just for autoconf, pltime processing for
647 		 * prefix is not necessary.
648 		 */
649 		if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
650 			struct nd_prefix *t;
651 			t = pr->ndpr_next;
652 
653 			/*
654 			 * address expiration and prefix expiration are
655 			 * separate.  NEVER perform in6_purgeaddr here.
656 			 */
657 
658 			prelist_remove(pr);
659 			pr = t;
660 		} else
661 			pr = pr->ndpr_next;
662 	}
663 
664 	mtx_unlock(&nd6_mtx);
665 
666 	callout_reset(&nd6_timer_ch, nd6_prune * hz, nd6_timer, NULL);
667 }
668 
669 static void
670 nd6_timer(void *arg __unused)
671 {
672 	struct lwkt_msg *lmsg = &nd6_timer_netmsg.lmsg;
673 
674 	KASSERT(mycpuid == 0, ("not on cpu0"));
675 	crit_enter();
676 	if (lmsg->ms_flags & MSGF_DONE)
677 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
678 	crit_exit();
679 }
680 
681 void
682 nd6_timer_init(void)
683 {
684 	callout_init_mp(&nd6_timer_ch);
685 	netmsg_init(&nd6_timer_netmsg, NULL, &netisr_adone_rport,
686 	    MSGF_PRIORITY, nd6_timer_dispatch);
687 	callout_reset_bycpu(&nd6_timer_ch, hz, nd6_timer, NULL, 0);
688 }
689 
690 static int
691 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
692 					 address */
693 {
694 	struct ifaddr_container *ifac;
695 	struct ifnet *ifp;
696 	struct in6_ifaddr *public_ifa6 = NULL;
697 
698 	ifp = ia6->ia_ifa.ifa_ifp;
699 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
700 		struct ifaddr *ifa = ifac->ifa;
701 		struct in6_ifaddr *it6;
702 
703 		if (ifa->ifa_addr->sa_family != AF_INET6)
704 			continue;
705 
706 		it6 = (struct in6_ifaddr *)ifa;
707 
708 		/* ignore no autoconf addresses. */
709 		if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
710 			continue;
711 
712 		/* ignore autoconf addresses with different prefixes. */
713 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
714 			continue;
715 
716 		/*
717 		 * Now we are looking at an autoconf address with the same
718 		 * prefix as ours.  If the address is temporary and is still
719 		 * preferred, do not create another one.  It would be rare, but
720 		 * could happen, for example, when we resume a laptop PC after
721 		 * a long period.
722 		 */
723 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
724 		    !IFA6_IS_DEPRECATED(it6)) {
725 			public_ifa6 = NULL;
726 			break;
727 		}
728 
729 		/*
730 		 * This is a public autoconf address that has the same prefix
731 		 * as ours.  If it is preferred, keep it.  We can't break the
732 		 * loop here, because there may be a still-preferred temporary
733 		 * address with the prefix.
734 		 */
735 		if (!IFA6_IS_DEPRECATED(it6))
736 		    public_ifa6 = it6;
737 	}
738 
739 	if (public_ifa6 != NULL) {
740 		int e;
741 
742 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
743 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
744 			    " tmp addr,errno=%d\n", e);
745 			return (-1);
746 		}
747 		return (0);
748 	}
749 
750 	return (-1);
751 }
752 
753 /*
754  * Nuke neighbor cache/prefix/default router management table, right before
755  * ifp goes away.
756  */
757 void
758 nd6_purge(struct ifnet *ifp)
759 {
760 	struct llinfo_nd6 *ln, *nln;
761 	struct nd_defrouter *dr, *ndr, drany;
762 	struct nd_prefix *pr, *npr;
763 
764 	/* Nuke default router list entries toward ifp */
765 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
766 		/*
767 		 * The first entry of the list may be stored in
768 		 * the routing table, so we'll delete it later.
769 		 */
770 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
771 			ndr = TAILQ_NEXT(dr, dr_entry);
772 			if (dr->ifp == ifp)
773 				defrtrlist_del(dr);
774 		}
775 		dr = TAILQ_FIRST(&nd_defrouter);
776 		if (dr->ifp == ifp)
777 			defrtrlist_del(dr);
778 	}
779 
780 	/* Nuke prefix list entries toward ifp */
781 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
782 		npr = pr->ndpr_next;
783 		if (pr->ndpr_ifp == ifp) {
784 			/*
785 			 * Previously, pr->ndpr_addr is removed as well,
786 			 * but I strongly believe we don't have to do it.
787 			 * nd6_purge() is only called from in6_ifdetach(),
788 			 * which removes all the associated interface addresses
789 			 * by itself.
790 			 * (jinmei@kame.net 20010129)
791 			 */
792 			prelist_remove(pr);
793 		}
794 	}
795 
796 	/* cancel default outgoing interface setting */
797 	if (nd6_defifindex == ifp->if_index)
798 		nd6_setdefaultiface(0);
799 
800 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
801 		/* refresh default router list */
802 		bzero(&drany, sizeof(drany));
803 		defrouter_delreq(&drany, 0);
804 		defrouter_select();
805 	}
806 
807 	/*
808 	 * Nuke neighbor cache entries for the ifp.
809 	 * Note that rt->rt_ifp may not be the same as ifp,
810 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
811 	 * nd6_rtrequest(), and ip6_input().
812 	 */
813 	ln = llinfo_nd6.ln_next;
814 	while (ln && ln != &llinfo_nd6) {
815 		struct rtentry *rt;
816 		struct sockaddr_dl *sdl;
817 
818 		nln = ln->ln_next;
819 		rt = ln->ln_rt;
820 		if (rt && rt->rt_gateway &&
821 		    rt->rt_gateway->sa_family == AF_LINK) {
822 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
823 			if (sdl->sdl_index == ifp->if_index)
824 				nln = nd6_free(rt);
825 		}
826 		ln = nln;
827 	}
828 }
829 
830 struct rtentry *
831 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
832 {
833 	struct rtentry *rt;
834 	struct sockaddr_in6 sin6;
835 
836 	bzero(&sin6, sizeof(sin6));
837 	sin6.sin6_len = sizeof(struct sockaddr_in6);
838 	sin6.sin6_family = AF_INET6;
839 	sin6.sin6_addr = *addr6;
840 
841 	if (create)
842 		rt = rtlookup((struct sockaddr *)&sin6);
843 	else
844 		rt = rtpurelookup((struct sockaddr *)&sin6);
845 	if (rt && !(rt->rt_flags & RTF_LLINFO)) {
846 		/*
847 		 * This is the case for the default route.
848 		 * If we want to create a neighbor cache for the address, we
849 		 * should free the route for the destination and allocate an
850 		 * interface route.
851 		 */
852 		if (create) {
853 			--rt->rt_refcnt;
854 			rt = NULL;
855 		}
856 	}
857 	if (!rt) {
858 		if (create && ifp) {
859 			int e;
860 
861 			/*
862 			 * If no route is available and create is set,
863 			 * we allocate a host route for the destination
864 			 * and treat it like an interface route.
865 			 * This hack is necessary for a neighbor which can't
866 			 * be covered by our own prefix.
867 			 */
868 			struct ifaddr *ifa;
869 
870 			ifa = ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
871 			if (ifa == NULL)
872 				return (NULL);
873 
874 			/*
875 			 * Create a new route.  RTF_LLINFO is necessary
876 			 * to create a Neighbor Cache entry for the
877 			 * destination in nd6_rtrequest which will be
878 			 * called in rtrequest via ifa->ifa_rtrequest.
879 			 */
880 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
881 			     ifa->ifa_addr, (struct sockaddr *)&all1_sa,
882 			     (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
883 			     ~RTF_CLONING, &rt)) != 0) {
884 				log(LOG_ERR,
885 				    "nd6_lookup: failed to add route for a "
886 				    "neighbor(%s), errno=%d\n",
887 				    ip6_sprintf(addr6), e);
888 			}
889 			if (rt == NULL)
890 				return (NULL);
891 			if (rt->rt_llinfo) {
892 				struct llinfo_nd6 *ln =
893 				    (struct llinfo_nd6 *)rt->rt_llinfo;
894 
895 				ln->ln_state = ND6_LLINFO_NOSTATE;
896 			}
897 		} else
898 			return (NULL);
899 	}
900 	rt->rt_refcnt--;
901 
902 	if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
903 		if (create) {
904 			log(LOG_DEBUG,
905 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
906 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
907 			/* xxx more logs... kazu */
908 		}
909 		return (NULL);
910 	}
911 	return (rt);
912 }
913 
914 static struct rtentry *
915 nd6_neighbor_lookup(struct in6_addr *addr6, struct ifnet *ifp)
916 {
917 	struct rtentry *rt;
918 	struct sockaddr_in6 sin6;
919 
920 	bzero(&sin6, sizeof(sin6));
921 	sin6.sin6_len = sizeof(struct sockaddr_in6);
922 	sin6.sin6_family = AF_INET6;
923 	sin6.sin6_addr = *addr6;
924 
925 	rt = rtpurelookup((struct sockaddr *)&sin6);
926 	if (rt == NULL)
927 		return (NULL);
928 	rt->rt_refcnt--;
929 
930 	if (!ND6_RTENTRY_IS_NEIGHBOR(rt, ifp)) {
931 		if (nd6_onlink_ns_rfc4861 &&
932 		    (ND6_RTENTRY_IS_LLCLONING(rt) ||	/* not cloned yet */
933 		     (rt->rt_parent != NULL &&		/* cloning */
934 		      ND6_RTENTRY_IS_LLCLONING(rt->rt_parent)))) {
935 			/*
936 			 * If cloning ever happened or is happening,
937 			 * rtentry for addr6 would or will become a
938 			 * neighbor cache.
939 			 */
940 		} else {
941 			rt = NULL;
942 		}
943 	}
944 	return (rt);
945 }
946 
947 /*
948  * Detect if a given IPv6 address identifies a neighbor on a given link.
949  * XXX: should take care of the destination of a p2p link?
950  */
951 int
952 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
953 {
954 	struct ifaddr_container *ifac;
955 	int i;
956 
957 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
958 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
959 
960 	/*
961 	 * A link-local address is always a neighbor.
962 	 * XXX: we should use the sin6_scope_id field rather than the embedded
963 	 * interface index.
964 	 */
965 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
966 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
967 		return (1);
968 
969 	/*
970 	 * If the address matches one of our addresses,
971 	 * it should be a neighbor.
972 	 */
973 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
974 		struct ifaddr *ifa = ifac->ifa;
975 
976 		if (ifa->ifa_addr->sa_family != AF_INET6)
977 			next: continue;
978 
979 		for (i = 0; i < 4; i++) {
980 			if ((IFADDR6(ifa).s6_addr32[i] ^
981 			     addr->sin6_addr.s6_addr32[i]) &
982 			    IFMASK6(ifa).s6_addr32[i])
983 				goto next;
984 		}
985 		return (1);
986 	}
987 
988 	/*
989 	 * Even if the address matches none of our addresses, it might be
990 	 * in the neighbor cache.
991 	 */
992 	if (nd6_neighbor_lookup(&addr->sin6_addr, ifp) != NULL)
993 		return (1);
994 
995 	return (0);
996 #undef IFADDR6
997 #undef IFMASK6
998 }
999 
1000 /*
1001  * Free an nd6 llinfo entry.
1002  */
1003 struct llinfo_nd6 *
1004 nd6_free(struct rtentry *rt)
1005 {
1006 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1007 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1008 	struct nd_defrouter *dr;
1009 	int error;
1010 	struct rtentry *nrt = NULL;
1011 
1012 	/*
1013 	 * we used to have kpfctlinput(PRC_HOSTDEAD) here.
1014 	 * even though it is not harmful, it was not really necessary.
1015 	 */
1016 
1017 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
1018 		mtx_lock(&nd6_mtx);
1019 		dr = defrouter_lookup(
1020 		    &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1021 		    rt->rt_ifp);
1022 
1023 		if (ln->ln_router || dr) {
1024 			/*
1025 			 * rt6_flush must be called whether or not the neighbor
1026 			 * is in the Default Router List.
1027 			 * See a corresponding comment in nd6_na_input().
1028 			 */
1029 			rt6_flush(&in6, rt->rt_ifp);
1030 		}
1031 
1032 		if (dr) {
1033 			/*
1034 			 * Unreachablity of a router might affect the default
1035 			 * router selection and on-link detection of advertised
1036 			 * prefixes.
1037 			 */
1038 
1039 			/*
1040 			 * Temporarily fake the state to choose a new default
1041 			 * router and to perform on-link determination of
1042 			 * prefixes correctly.
1043 			 * Below the state will be set correctly,
1044 			 * or the entry itself will be deleted.
1045 			 */
1046 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1047 
1048 			/*
1049 			 * Since defrouter_select() does not affect the
1050 			 * on-link determination and MIP6 needs the check
1051 			 * before the default router selection, we perform
1052 			 * the check now.
1053 			 */
1054 			pfxlist_onlink_check();
1055 
1056 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
1057 				/*
1058 				 * It is used as the current default router,
1059 				 * so we have to move it to the end of the
1060 				 * list and choose a new one.
1061 				 * XXX: it is not very efficient if this is
1062 				 *      the only router.
1063 				 */
1064 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1065 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1066 
1067 				defrouter_select();
1068 			}
1069 		}
1070 		mtx_unlock(&nd6_mtx);
1071 	}
1072 
1073 	/*
1074 	 * Before deleting the entry, remember the next entry as the
1075 	 * return value.  We need this because pfxlist_onlink_check() above
1076 	 * might have freed other entries (particularly the old next entry) as
1077 	 * a side effect (XXX).
1078 	 */
1079 	next = ln->ln_next;
1080 
1081 	/*
1082 	 * Detach the route from the routing tree and the list of neighbor
1083 	 * caches, and disable the route entry not to be used in already
1084 	 * cached routes.
1085 	 */
1086 	error = rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, &nrt);
1087 	if (error == 0 && nrt != NULL) {
1088 		struct sockaddr_dl *sdl;
1089 
1090 		sdl = (struct sockaddr_dl *)nrt->rt_gateway;
1091 		if (sdl->sdl_alen != 0)
1092 			rt_rtmsg(RTM_DELETE, nrt, nrt->rt_ifp, 0);
1093 		rtfree(nrt);
1094 	}
1095 
1096 	return (next);
1097 }
1098 
1099 /*
1100  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1101  *
1102  * XXX cost-effective metods?
1103  */
1104 void
1105 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1106 {
1107 	struct llinfo_nd6 *ln;
1108 
1109 	/*
1110 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1111 	 * routing table by supplied "dst6".
1112 	 */
1113 	if (!rt) {
1114 		if (!dst6)
1115 			return;
1116 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1117 			return;
1118 	}
1119 
1120 	if ((rt->rt_flags & RTF_GATEWAY) ||
1121 	    !(rt->rt_flags & RTF_LLINFO) ||
1122 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1123 	    rt->rt_gateway->sa_family != AF_LINK) {
1124 		/* This is not a host route. */
1125 		return;
1126 	}
1127 
1128 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1129 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1130 		return;
1131 
1132 	/*
1133 	 * if we get upper-layer reachability confirmation many times,
1134 	 * it is possible we have false information.
1135 	 */
1136 	if (!force) {
1137 		ln->ln_byhint++;
1138 		if (ln->ln_byhint > nd6_maxnudhint)
1139 			return;
1140 	}
1141 
1142 	ln->ln_state = ND6_LLINFO_REACHABLE;
1143 	if (ln->ln_expire)
1144 		ln->ln_expire = time_uptime +
1145 			ND_IFINFO(rt->rt_ifp)->reachable;
1146 }
1147 
1148 void
1149 nd6_rtrequest(int req, struct rtentry *rt)
1150 {
1151 	struct sockaddr *gate = rt->rt_gateway;
1152 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1153 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1154 	struct ifnet *ifp = rt->rt_ifp;
1155 	struct ifaddr *ifa;
1156 
1157 	if ((rt->rt_flags & RTF_GATEWAY))
1158 		return;
1159 
1160 	if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1161 		/*
1162 		 * This is probably an interface direct route for a link
1163 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1164 		 * We do not need special treatment below for such a route.
1165 		 * Moreover, the RTF_LLINFO flag which would be set below
1166 		 * would annoy the ndp(8) command.
1167 		 */
1168 		return;
1169 	}
1170 
1171 	if (req == RTM_RESOLVE &&
1172 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1173 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1174 		/*
1175 		 * FreeBSD and BSD/OS often make a cloned host route based
1176 		 * on a less-specific route (e.g. the default route).
1177 		 * If the less specific route does not have a "gateway"
1178 		 * (this is the case when the route just goes to a p2p or an
1179 		 * stf interface), we'll mistakenly make a neighbor cache for
1180 		 * the host route, and will see strange neighbor solicitation
1181 		 * for the corresponding destination.  In order to avoid the
1182 		 * confusion, we check if the destination of the route is
1183 		 * a neighbor in terms of neighbor discovery, and stop the
1184 		 * process if not.  Additionally, we remove the LLINFO flag
1185 		 * so that ndp(8) will not try to get the neighbor information
1186 		 * of the destination.
1187 		 */
1188 		rt->rt_flags &= ~RTF_LLINFO;
1189 		return;
1190 	}
1191 
1192 	switch (req) {
1193 	case RTM_ADD:
1194 		/*
1195 		 * There is no backward compatibility :)
1196 		 *
1197 		 * if (!(rt->rt_flags & RTF_HOST) &&
1198 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1199 		 *	   rt->rt_flags |= RTF_CLONING;
1200 		 */
1201 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1202 			/*
1203 			 * Case 1: This route should come from
1204 			 * a route to interface.  RTF_LLINFO flag is set
1205 			 * for a host route whose destination should be
1206 			 * treated as on-link.
1207 			 */
1208 			rt_setgate(rt, rt_key(rt),
1209 				   (struct sockaddr *)&null_sdl);
1210 			gate = rt->rt_gateway;
1211 			SDL(gate)->sdl_type = ifp->if_type;
1212 			SDL(gate)->sdl_index = ifp->if_index;
1213 			if (ln)
1214 				ln->ln_expire = time_uptime;
1215 			if (ln && ln->ln_expire == 0) {
1216 				/* kludge for desktops */
1217 				ln->ln_expire = 1;
1218 			}
1219 			if ((rt->rt_flags & RTF_CLONING))
1220 				break;
1221 		}
1222 		/*
1223 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1224 		 * We don't do that here since llinfo is not ready yet.
1225 		 *
1226 		 * There are also couple of other things to be discussed:
1227 		 * - unsolicited NA code needs improvement beforehand
1228 		 * - RFC2461 says we MAY send multicast unsolicited NA
1229 		 *   (7.2.6 paragraph 4), however, it also says that we
1230 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1231 		 *   we don't have anything like it right now.
1232 		 *   note that the mechanism needs a mutual agreement
1233 		 *   between proxies, which means that we need to implement
1234 		 *   a new protocol, or a new kludge.
1235 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1236 		 *   we need to check ip6forwarding before sending it.
1237 		 *   (or should we allow proxy ND configuration only for
1238 		 *   routers?  there's no mention about proxy ND from hosts)
1239 		 */
1240 #if 0
1241 		/* XXX it does not work */
1242 		if ((rt->rt_flags & RTF_ANNOUNCE) && mycpuid == 0) {
1243 			nd6_na_output(ifp,
1244 			      &SIN6(rt_key(rt))->sin6_addr,
1245 			      &SIN6(rt_key(rt))->sin6_addr,
1246 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1247 			      1, NULL);
1248 		}
1249 #endif
1250 		/* FALLTHROUGH */
1251 	case RTM_RESOLVE:
1252 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1253 			/*
1254 			 * Address resolution isn't necessary for a point to
1255 			 * point link, so we can skip this test for a p2p link.
1256 			 */
1257 			if (gate->sa_family != AF_LINK ||
1258 			    gate->sa_len < sizeof(null_sdl)) {
1259 				log(LOG_DEBUG,
1260 				    "nd6_rtrequest: bad gateway value: %s\n",
1261 				    if_name(ifp));
1262 				break;
1263 			}
1264 			SDL(gate)->sdl_type = ifp->if_type;
1265 			SDL(gate)->sdl_index = ifp->if_index;
1266 		}
1267 		if (ln != NULL)
1268 			break;	/* This happens on a route change */
1269 		/*
1270 		 * Case 2: This route may come from cloning, or a manual route
1271 		 * add with a LL address.
1272 		 */
1273 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1274 		rt->rt_llinfo = (caddr_t)ln;
1275 		if (!ln) {
1276 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1277 			break;
1278 		}
1279 		nd6_inuse++;
1280 		nd6_allocated++;
1281 		bzero(ln, sizeof(*ln));
1282 		ln->ln_rt = rt;
1283 		/* this is required for "ndp" command. - shin */
1284 		if (req == RTM_ADD) {
1285 		        /*
1286 			 * gate should have some valid AF_LINK entry,
1287 			 * and ln->ln_expire should have some lifetime
1288 			 * which is specified by ndp command.
1289 			 */
1290 			ln->ln_state = ND6_LLINFO_REACHABLE;
1291 			ln->ln_byhint = 0;
1292 		} else {
1293 		        /*
1294 			 * When req == RTM_RESOLVE, rt is created and
1295 			 * initialized in rtrequest(), so rt_expire is 0.
1296 			 */
1297 			ln->ln_state = ND6_LLINFO_NOSTATE;
1298 			ln->ln_expire = time_uptime;
1299 		}
1300 		rt->rt_flags |= RTF_LLINFO;
1301 		ln->ln_next = llinfo_nd6.ln_next;
1302 		llinfo_nd6.ln_next = ln;
1303 		ln->ln_prev = &llinfo_nd6;
1304 		ln->ln_next->ln_prev = ln;
1305 
1306 		/*
1307 		 * check if rt_key(rt) is one of my address assigned
1308 		 * to the interface.
1309 		 */
1310 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1311 		    &SIN6(rt_key(rt))->sin6_addr);
1312 		if (ifa) {
1313 			caddr_t macp = nd6_ifptomac(ifp);
1314 			ln->ln_expire = 0;
1315 			ln->ln_state = ND6_LLINFO_REACHABLE;
1316 			ln->ln_byhint = 0;
1317 			if (macp) {
1318 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1319 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1320 			}
1321 			if (nd6_useloopback) {
1322 				rt->rt_ifp = loif;	/* XXX */
1323 				/*
1324 				 * Make sure rt_ifa be equal to the ifaddr
1325 				 * corresponding to the address.
1326 				 * We need this because when we refer
1327 				 * rt_ifa->ia6_flags in ip6_input, we assume
1328 				 * that the rt_ifa points to the address instead
1329 				 * of the loopback address.
1330 				 */
1331 				if (ifa != rt->rt_ifa) {
1332 					IFAFREE(rt->rt_ifa);
1333 					IFAREF(ifa);
1334 					rt->rt_ifa = ifa;
1335 				}
1336 			}
1337 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1338 			ln->ln_expire = 0;
1339 			ln->ln_state = ND6_LLINFO_REACHABLE;
1340 			ln->ln_byhint = 0;
1341 
1342 			/*
1343 			 * Join solicited node multicast for proxy ND, and only
1344 			 * join it once on cpu0.
1345 			 */
1346 			if ((ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1347 				struct in6_addr llsol;
1348 				int error;
1349 
1350 				llsol = SIN6(rt_key(rt))->sin6_addr;
1351 				llsol.s6_addr16[0] = htons(0xff02);
1352 				llsol.s6_addr16[1] = htons(ifp->if_index);
1353 				llsol.s6_addr32[1] = 0;
1354 				llsol.s6_addr32[2] = htonl(1);
1355 				llsol.s6_addr8[12] = 0xff;
1356 
1357 				if (!in6_addmulti(&llsol, ifp, &error)) {
1358 					nd6log((LOG_ERR, "%s: failed to join "
1359 					    "%s (errno=%d)\n", if_name(ifp),
1360 					    ip6_sprintf(&llsol), error));
1361 				}
1362 			}
1363 		}
1364 		break;
1365 
1366 	case RTM_DELETE:
1367 		if (!ln)
1368 			break;
1369 		/*
1370 		 * Leave from solicited node multicast for proxy ND, and only
1371 		 * leave it once on cpu0 (since we joined it once on cpu0).
1372 		 */
1373 		if ((rt->rt_flags & RTF_ANNOUNCE) &&
1374 		    (ifp->if_flags & IFF_MULTICAST) && mycpuid == 0) {
1375 			struct in6_addr llsol;
1376 			struct in6_multi *in6m;
1377 
1378 			llsol = SIN6(rt_key(rt))->sin6_addr;
1379 			llsol.s6_addr16[0] = htons(0xff02);
1380 			llsol.s6_addr16[1] = htons(ifp->if_index);
1381 			llsol.s6_addr32[1] = 0;
1382 			llsol.s6_addr32[2] = htonl(1);
1383 			llsol.s6_addr8[12] = 0xff;
1384 
1385 			in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1386 			if (in6m)
1387 				in6_delmulti(in6m);
1388 		}
1389 		nd6_inuse--;
1390 		ln->ln_next->ln_prev = ln->ln_prev;
1391 		ln->ln_prev->ln_next = ln->ln_next;
1392 		ln->ln_prev = NULL;
1393 		rt->rt_llinfo = 0;
1394 		rt->rt_flags &= ~RTF_LLINFO;
1395 		if (ln->ln_hold)
1396 			m_freem(ln->ln_hold);
1397 		Free((caddr_t)ln);
1398 	}
1399 }
1400 
1401 int
1402 nd6_ioctl(u_long cmd, caddr_t	data, struct ifnet *ifp)
1403 {
1404 	struct in6_drlist *drl = (struct in6_drlist *)data;
1405 	struct in6_prlist *prl = (struct in6_prlist *)data;
1406 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1407 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1408 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1409 	struct nd_defrouter *dr, any;
1410 	struct nd_prefix *pr;
1411 	struct rtentry *rt;
1412 	int i = 0, error = 0;
1413 
1414 	switch (cmd) {
1415 	case SIOCGDRLST_IN6:
1416 		/*
1417 		 * obsolete API, use sysctl under net.inet6.icmp6
1418 		 */
1419 		bzero(drl, sizeof(*drl));
1420 		mtx_lock(&nd6_mtx);
1421 		dr = TAILQ_FIRST(&nd_defrouter);
1422 		while (dr && i < DRLSTSIZ) {
1423 			drl->defrouter[i].rtaddr = dr->rtaddr;
1424 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1425 				/* XXX: need to this hack for KAME stack */
1426 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1427 			} else
1428 				log(LOG_ERR,
1429 				    "default router list contains a "
1430 				    "non-linklocal address(%s)\n",
1431 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1432 
1433 			drl->defrouter[i].flags = dr->flags;
1434 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1435 			drl->defrouter[i].expire = dr->expire;
1436 			drl->defrouter[i].if_index = dr->ifp->if_index;
1437 			i++;
1438 			dr = TAILQ_NEXT(dr, dr_entry);
1439 		}
1440 		mtx_unlock(&nd6_mtx);
1441 		break;
1442 	case SIOCGPRLST_IN6:
1443 		/*
1444 		 * obsolete API, use sysctl under net.inet6.icmp6
1445 		 */
1446 		/*
1447 		 * XXX meaning of fields, especialy "raflags", is very
1448 		 * differnet between RA prefix list and RR/static prefix list.
1449 		 * how about separating ioctls into two?
1450 		 */
1451 		bzero(prl, sizeof(*prl));
1452 		mtx_lock(&nd6_mtx);
1453 		pr = nd_prefix.lh_first;
1454 		while (pr && i < PRLSTSIZ) {
1455 			struct nd_pfxrouter *pfr;
1456 			int j;
1457 
1458 			in6_embedscope(&prl->prefix[i].prefix,
1459 			    &pr->ndpr_prefix, NULL, NULL);
1460 			prl->prefix[i].raflags = pr->ndpr_raf;
1461 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1462 			prl->prefix[i].vltime = pr->ndpr_vltime;
1463 			prl->prefix[i].pltime = pr->ndpr_pltime;
1464 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1465 			prl->prefix[i].expire = pr->ndpr_expire;
1466 
1467 			pfr = pr->ndpr_advrtrs.lh_first;
1468 			j = 0;
1469 			while (pfr) {
1470 				if (j < DRLSTSIZ) {
1471 #define RTRADDR prl->prefix[i].advrtr[j]
1472 					RTRADDR = pfr->router->rtaddr;
1473 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1474 						/* XXX: hack for KAME */
1475 						RTRADDR.s6_addr16[1] = 0;
1476 					} else
1477 						log(LOG_ERR,
1478 						    "a router(%s) advertises "
1479 						    "a prefix with "
1480 						    "non-link local address\n",
1481 						    ip6_sprintf(&RTRADDR));
1482 #undef RTRADDR
1483 				}
1484 				j++;
1485 				pfr = pfr->pfr_next;
1486 			}
1487 			prl->prefix[i].advrtrs = j;
1488 			prl->prefix[i].origin = PR_ORIG_RA;
1489 
1490 			i++;
1491 			pr = pr->ndpr_next;
1492 		}
1493 		mtx_unlock(&nd6_mtx);
1494 
1495 		break;
1496 	case OSIOCGIFINFO_IN6:
1497 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1498 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1499 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1500 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1501 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1502 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1503 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1504 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1505 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1506 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1507 		break;
1508 	case SIOCGIFINFO_IN6:
1509 		ndi->ndi = *ND_IFINFO(ifp);
1510 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1511 		break;
1512 	case SIOCSIFINFO_FLAGS:
1513 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1514 		break;
1515 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1516 		/* flush default router list */
1517 		/*
1518 		 * xxx sumikawa: should not delete route if default
1519 		 * route equals to the top of default router list
1520 		 */
1521 		bzero(&any, sizeof(any));
1522 		defrouter_delreq(&any, 0);
1523 		defrouter_select();
1524 		/* xxx sumikawa: flush prefix list */
1525 		break;
1526 	case SIOCSPFXFLUSH_IN6:
1527 	{
1528 		/* flush all the prefix advertised by routers */
1529 		struct nd_prefix *pr, *next;
1530 
1531 		mtx_lock(&nd6_mtx);
1532 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1533 			struct in6_ifaddr *ia, *ia_next;
1534 
1535 			next = pr->ndpr_next;
1536 
1537 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1538 				continue; /* XXX */
1539 
1540 			/* do we really have to remove addresses as well? */
1541 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1542 				/* ia might be removed.  keep the next ptr. */
1543 				ia_next = ia->ia_next;
1544 
1545 				if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1546 					continue;
1547 
1548 				if (ia->ia6_ndpr == pr)
1549 					in6_purgeaddr(&ia->ia_ifa);
1550 			}
1551 			prelist_remove(pr);
1552 		}
1553 		mtx_unlock(&nd6_mtx);
1554 		break;
1555 	}
1556 	case SIOCSRTRFLUSH_IN6:
1557 	{
1558 		/* flush all the default routers */
1559 		struct nd_defrouter *dr, *next;
1560 
1561 		mtx_lock(&nd6_mtx);
1562 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1563 			/*
1564 			 * The first entry of the list may be stored in
1565 			 * the routing table, so we'll delete it later.
1566 			 */
1567 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1568 				next = TAILQ_NEXT(dr, dr_entry);
1569 				defrtrlist_del(dr);
1570 			}
1571 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1572 		}
1573 		mtx_unlock(&nd6_mtx);
1574 		break;
1575 	}
1576 	case SIOCGNBRINFO_IN6:
1577 	{
1578 		struct llinfo_nd6 *ln;
1579 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1580 
1581 		/*
1582 		 * XXX: KAME specific hack for scoped addresses
1583 		 *      XXXX: for other scopes than link-local?
1584 		 */
1585 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1586 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1587 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1588 
1589 			if (*idp == 0)
1590 				*idp = htons(ifp->if_index);
1591 		}
1592 
1593 		mtx_lock(&nd6_mtx);
1594 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1595 			error = EINVAL;
1596 			mtx_unlock(&nd6_mtx);
1597 			break;
1598 		}
1599 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1600 		nbi->state = ln->ln_state;
1601 		nbi->asked = ln->ln_asked;
1602 		nbi->isrouter = ln->ln_router;
1603 		nbi->expire = ln->ln_expire;
1604 		mtx_unlock(&nd6_mtx);
1605 
1606 		break;
1607 	}
1608 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1609 		ndif->ifindex = nd6_defifindex;
1610 		break;
1611 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1612 		return (nd6_setdefaultiface(ndif->ifindex));
1613 	}
1614 	return (error);
1615 }
1616 
1617 /*
1618  * Create neighbor cache entry and cache link-layer address,
1619  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1620  */
1621 struct rtentry *
1622 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1623 		 int lladdrlen,
1624 		 int type,	/* ICMP6 type */
1625 		 int code	/* type dependent information */)
1626 {
1627 	struct rtentry *rt = NULL;
1628 	struct llinfo_nd6 *ln = NULL;
1629 	int is_newentry;
1630 	struct sockaddr_dl *sdl = NULL;
1631 	int do_update;
1632 	int olladdr;
1633 	int llchange;
1634 	int newstate = 0;
1635 
1636 	if (!ifp)
1637 		panic("ifp == NULL in nd6_cache_lladdr");
1638 	if (!from)
1639 		panic("from == NULL in nd6_cache_lladdr");
1640 
1641 	/* nothing must be updated for unspecified address */
1642 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1643 		return NULL;
1644 
1645 	/*
1646 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1647 	 * the caller.
1648 	 *
1649 	 * XXX If the link does not have link-layer adderss, what should
1650 	 * we do? (ifp->if_addrlen == 0)
1651 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1652 	 * description on it in NS section (RFC 2461 7.2.3).
1653 	 */
1654 
1655 	rt = nd6_lookup(from, 0, ifp);
1656 	if (!rt) {
1657 #if 0
1658 		/* nothing must be done if there's no lladdr */
1659 		if (!lladdr || !lladdrlen)
1660 			return NULL;
1661 #endif
1662 
1663 		rt = nd6_lookup(from, 1, ifp);
1664 		is_newentry = 1;
1665 	} else {
1666 		/* do nothing if static ndp is set */
1667 		if (rt->rt_flags & RTF_STATIC)
1668 			return NULL;
1669 		is_newentry = 0;
1670 	}
1671 
1672 	if (!rt)
1673 		return NULL;
1674 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1675 fail:
1676 		nd6_free(rt);
1677 		return NULL;
1678 	}
1679 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1680 	if (!ln)
1681 		goto fail;
1682 	if (!rt->rt_gateway)
1683 		goto fail;
1684 	if (rt->rt_gateway->sa_family != AF_LINK)
1685 		goto fail;
1686 	sdl = SDL(rt->rt_gateway);
1687 
1688 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1689 	if (olladdr && lladdr) {
1690 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1691 			llchange = 1;
1692 		else
1693 			llchange = 0;
1694 	} else
1695 		llchange = 0;
1696 
1697 	/*
1698 	 * newentry olladdr  lladdr  llchange	(*=record)
1699 	 *	0	n	n	--	(1)
1700 	 *	0	y	n	--	(2)
1701 	 *	0	n	y	--	(3) * STALE
1702 	 *	0	y	y	n	(4) *
1703 	 *	0	y	y	y	(5) * STALE
1704 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1705 	 *	1	--	y	--	(7) * STALE
1706 	 */
1707 
1708 	if (lladdr) {		/* (3-5) and (7) */
1709 		/*
1710 		 * Record source link-layer address
1711 		 * XXX is it dependent to ifp->if_type?
1712 		 */
1713 		sdl->sdl_alen = ifp->if_addrlen;
1714 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1715 	}
1716 
1717 	if (!is_newentry) {
1718 		if ((!olladdr && lladdr) ||		/* (3) */
1719 		    (olladdr && lladdr && llchange)) {	/* (5) */
1720 			do_update = 1;
1721 			newstate = ND6_LLINFO_STALE;
1722 		} else {				/* (1-2,4) */
1723 			do_update = 0;
1724 		}
1725 	} else {
1726 		do_update = 1;
1727 		if (!lladdr)				/* (6) */
1728 			newstate = ND6_LLINFO_NOSTATE;
1729 		else					/* (7) */
1730 			newstate = ND6_LLINFO_STALE;
1731 	}
1732 
1733 	if (do_update) {
1734 		/*
1735 		 * Update the state of the neighbor cache.
1736 		 */
1737 		ln->ln_state = newstate;
1738 
1739 		if (ln->ln_state == ND6_LLINFO_STALE) {
1740 			/*
1741 			 * XXX: since nd6_output() below will cause
1742 			 * state tansition to DELAY and reset the timer,
1743 			 * we must set the timer now, although it is actually
1744 			 * meaningless.
1745 			 */
1746 			ln->ln_expire = time_uptime + nd6_gctimer;
1747 
1748 			if (ln->ln_hold) {
1749 				/*
1750 				 * we assume ifp is not a p2p here, so just
1751 				 * set the 2nd argument as the 1st one.
1752 				 */
1753 				nd6_output(ifp, ifp, ln->ln_hold,
1754 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1755 				ln->ln_hold = NULL;
1756 			}
1757 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1758 			/* probe right away */
1759 			ln->ln_expire = time_uptime;
1760 		}
1761 	}
1762 
1763 	/*
1764 	 * ICMP6 type dependent behavior.
1765 	 *
1766 	 * NS: clear IsRouter if new entry
1767 	 * RS: clear IsRouter
1768 	 * RA: set IsRouter if there's lladdr
1769 	 * redir: clear IsRouter if new entry
1770 	 *
1771 	 * RA case, (1):
1772 	 * The spec says that we must set IsRouter in the following cases:
1773 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1774 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1775 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1776 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1777 	 * neighbor cache, this is similar to (6).
1778 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1779 	 *
1780 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1781 	 *							D R
1782 	 *	0	n	n	--	(1)	c   ?     s
1783 	 *	0	y	n	--	(2)	c   s     s
1784 	 *	0	n	y	--	(3)	c   s     s
1785 	 *	0	y	y	n	(4)	c   s     s
1786 	 *	0	y	y	y	(5)	c   s     s
1787 	 *	1	--	n	--	(6) c	c 	c s
1788 	 *	1	--	y	--	(7) c	c   s	c s
1789 	 *
1790 	 *					(c=clear s=set)
1791 	 */
1792 	switch (type & 0xff) {
1793 	case ND_NEIGHBOR_SOLICIT:
1794 		/*
1795 		 * New entry must have is_router flag cleared.
1796 		 */
1797 		if (is_newentry)	/* (6-7) */
1798 			ln->ln_router = 0;
1799 		break;
1800 	case ND_REDIRECT:
1801 		/*
1802 		 * If the icmp is a redirect to a better router, always set the
1803 		 * is_router flag.  Otherwise, if the entry is newly created,
1804 		 * clear the flag.  [RFC 2461, sec 8.3]
1805 		 */
1806 		if (code == ND_REDIRECT_ROUTER)
1807 			ln->ln_router = 1;
1808 		else if (is_newentry) /* (6-7) */
1809 			ln->ln_router = 0;
1810 		break;
1811 	case ND_ROUTER_SOLICIT:
1812 		/*
1813 		 * is_router flag must always be cleared.
1814 		 */
1815 		ln->ln_router = 0;
1816 		break;
1817 	case ND_ROUTER_ADVERT:
1818 		/*
1819 		 * Mark an entry with lladdr as a router.
1820 		 */
1821 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1822 		    (is_newentry && lladdr)) {			/* (7) */
1823 			ln->ln_router = 1;
1824 		}
1825 		break;
1826 	}
1827 
1828 	if (llchange || lladdr)
1829 		rt_rtmsg(llchange ? RTM_CHANGE : RTM_ADD, rt, rt->rt_ifp, 0);
1830 
1831 	/*
1832 	 * When the link-layer address of a router changes, select the
1833 	 * best router again.  In particular, when the neighbor entry is newly
1834 	 * created, it might affect the selection policy.
1835 	 * Question: can we restrict the first condition to the "is_newentry"
1836 	 * case?
1837 	 * XXX: when we hear an RA from a new router with the link-layer
1838 	 * address option, defrouter_select() is called twice, since
1839 	 * defrtrlist_update called the function as well.  However, I believe
1840 	 * we can compromise the overhead, since it only happens the first
1841 	 * time.
1842 	 * XXX: although defrouter_select() should not have a bad effect
1843 	 * for those are not autoconfigured hosts, we explicitly avoid such
1844 	 * cases for safety.
1845 	 */
1846 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1847 		defrouter_select();
1848 
1849 	return rt;
1850 }
1851 
1852 static void
1853 nd6_slowtimo(void *arg __unused)
1854 {
1855 	struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1856 
1857 	KASSERT(mycpuid == 0, ("not on cpu0"));
1858 	crit_enter();
1859 	if (lmsg->ms_flags & MSGF_DONE)
1860 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1861 	crit_exit();
1862 }
1863 
1864 static void
1865 nd6_slowtimo_dispatch(netmsg_t nmsg)
1866 {
1867 	const struct ifnet_array *arr;
1868 	struct nd_ifinfo *nd6if;
1869 	int i;
1870 
1871 	ASSERT_NETISR0;
1872 
1873 	crit_enter();
1874 	lwkt_replymsg(&nmsg->lmsg, 0);	/* reply ASAP */
1875 	crit_exit();
1876 
1877 	arr = ifnet_array_get();
1878 
1879 	mtx_lock(&nd6_mtx);
1880 	for (i = 0; i < arr->ifnet_count; ++i) {
1881 		struct ifnet *ifp = arr->ifnet_arr[i];
1882 
1883 		if (ifp->if_afdata[AF_INET6] == NULL)
1884 			continue;
1885 		nd6if = ND_IFINFO(ifp);
1886 		if (nd6if->basereachable && /* already initialized */
1887 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1888 			/*
1889 			 * Since reachable time rarely changes by router
1890 			 * advertisements, we SHOULD insure that a new random
1891 			 * value gets recomputed at least once every few hours.
1892 			 * (RFC 2461, 6.3.4)
1893 			 */
1894 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1895 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1896 		}
1897 	}
1898 	mtx_unlock(&nd6_mtx);
1899 
1900 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1901 	    nd6_slowtimo, NULL);
1902 }
1903 
1904 int
1905 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1906 	   struct sockaddr_in6 *dst, struct rtentry *rt)
1907 {
1908 	int error;
1909 
1910 	if (ifp->if_flags & IFF_LOOPBACK)
1911 		error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1912 	else
1913 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1914 	return error;
1915 }
1916 
1917 int
1918 nd6_resolve(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1919     struct sockaddr *dst0, u_char *desten)
1920 {
1921 	struct sockaddr_in6 *dst = SIN6(dst0);
1922 	struct rtentry *rt = NULL;
1923 	struct llinfo_nd6 *ln = NULL;
1924 	int error;
1925 
1926 	if (m->m_flags & M_MCAST) {
1927 		switch (ifp->if_type) {
1928 		case IFT_ETHER:
1929 #ifdef IFT_L2VLAN
1930 		case IFT_L2VLAN:
1931 #endif
1932 #ifdef IFT_IEEE80211
1933 		case IFT_IEEE80211:
1934 #endif
1935 			ETHER_MAP_IPV6_MULTICAST(&dst->sin6_addr,
1936 						 desten);
1937 			return 0;
1938 		case IFT_IEEE1394:
1939 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1940 			return 0;
1941 		default:
1942 			error = EAFNOSUPPORT;
1943 			goto bad;
1944 		}
1945 	}
1946 
1947 	if (rt0 != NULL) {
1948 		error = rt_llroute(dst0, rt0, &rt);
1949 		if (error != 0)
1950 			goto bad;
1951 		ln = rt->rt_llinfo;
1952 	}
1953 
1954 	/*
1955 	 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1956 	 * the condition below is not very efficient.  But we believe
1957 	 * it is tolerable, because this should be a rare case.
1958 	 */
1959 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp)) {
1960 		rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
1961 		if (rt != NULL)
1962 			ln = rt->rt_llinfo;
1963 	}
1964 
1965 	if (ln == NULL || rt == NULL) {
1966 		if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1967 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1968 			log(LOG_DEBUG,
1969 			    "nd6_output: can't allocate llinfo for %s "
1970 			    "(ln=%p, rt=%p)\n",
1971 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1972 			error = ENOBUFS;
1973 			goto bad;
1974 		}
1975 		return 0;
1976 	}
1977 
1978 	/* We don't have to do link-layer address resolution on a p2p link. */
1979 	if ((ifp->if_flags & IFF_POINTOPOINT) &&
1980 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1981 		ln->ln_state = ND6_LLINFO_STALE;
1982 		ln->ln_expire = time_uptime + nd6_gctimer;
1983 	}
1984 
1985 	/*
1986 	 * The first time we send a packet to a neighbor whose entry is
1987 	 * STALE, we have to change the state to DELAY and a sets a timer to
1988 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1989 	 * neighbor unreachability detection on expiration.
1990 	 * (RFC 2461 7.3.3)
1991 	 */
1992 	if (ln->ln_state == ND6_LLINFO_STALE) {
1993 		ln->ln_asked = 0;
1994 		ln->ln_state = ND6_LLINFO_DELAY;
1995 		ln->ln_expire = time_uptime + nd6_delay;
1996 	}
1997 
1998 	/*
1999 	 * If the neighbor cache entry has a state other than INCOMPLETE
2000 	 * (i.e. its link-layer address is already resolved), return it.
2001 	 */
2002 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE) {
2003 		struct sockaddr_dl *sdl = SDL(rt->rt_gateway);
2004 
2005 		KKASSERT(sdl->sdl_family == AF_LINK && sdl->sdl_alen != 0);
2006 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2007 		return 0;
2008 	}
2009 
2010 	/*
2011 	 * There is a neighbor cache entry, but no ethernet address
2012 	 * response yet.  Replace the held mbuf (if any) with this
2013 	 * latest one.
2014 	 */
2015 	if (ln->ln_hold)
2016 		m_freem(ln->ln_hold);
2017 	ln->ln_hold = m;
2018 
2019 	/*
2020 	 * This code conforms to the rate-limiting rule described in Section
2021 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2022 	 * an NS below.
2023 	 */
2024 	if (ln->ln_state == ND6_LLINFO_NOSTATE ||
2025 	    ln->ln_state == ND6_LLINFO_WAITDELETE) {
2026 		/*
2027 		 * This neighbor cache entry was just created; change its
2028 		 * state to INCOMPLETE and start its life cycle.
2029 		 *
2030 		 * We force an NS output below by setting ln_expire to 1
2031 		 * (nd6_rtrequest() could set it to the current time_uptime)
2032 		 * and zeroing out ln_asked (XXX this may not be necessary).
2033 		 */
2034 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2035 		ln->ln_expire = 1;
2036 		ln->ln_asked = 0;
2037 	}
2038 	if (ln->ln_expire && ln->ln_expire < time_uptime && ln->ln_asked == 0) {
2039 		ln->ln_asked++;
2040 		ln->ln_expire = time_uptime + ND_IFINFO(ifp)->retrans / 1000;
2041 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2042 	}
2043 
2044 	if (ln->ln_asked >= nd6_mmaxtries)
2045 		return (rt != NULL && rt->rt_flags & RTF_GATEWAY) ?
2046 		    EHOSTUNREACH : EHOSTDOWN;
2047 	return EWOULDBLOCK;
2048 
2049 bad:
2050 	m_freem(m);
2051 	return error;
2052 }
2053 
2054 int
2055 nd6_need_cache(struct ifnet *ifp)
2056 {
2057 	/*
2058 	 * XXX: we currently do not make neighbor cache on any interface
2059 	 * other than Ethernet and GIF.
2060 	 *
2061 	 * RFC2893 says:
2062 	 * - unidirectional tunnels needs no ND
2063 	 */
2064 	switch (ifp->if_type) {
2065 	case IFT_ETHER:
2066 	case IFT_IEEE1394:
2067 #ifdef IFT_L2VLAN
2068 	case IFT_L2VLAN:
2069 #endif
2070 #ifdef IFT_IEEE80211
2071 	case IFT_IEEE80211:
2072 #endif
2073 #ifdef IFT_CARP
2074 	case IFT_CARP:
2075 #endif
2076 	case IFT_GIF:		/* XXX need more cases? */
2077 		return (1);
2078 	default:
2079 		return (0);
2080 	}
2081 }
2082 
2083 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2084 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2085 #ifdef SYSCTL_DECL
2086 SYSCTL_DECL(_net_inet6_icmp6);
2087 #endif
2088 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2089 	CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2090 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2091 	CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2092 
2093 static int
2094 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2095 {
2096 	int error;
2097 	char buf[1024];
2098 	struct in6_defrouter *d, *de;
2099 	struct nd_defrouter *dr;
2100 
2101 	if (req->newptr)
2102 		return EPERM;
2103 	error = 0;
2104 
2105 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2106 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2107 		d = (struct in6_defrouter *)buf;
2108 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2109 
2110 		if (d + 1 <= de) {
2111 			bzero(d, sizeof(*d));
2112 			d->rtaddr.sin6_family = AF_INET6;
2113 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2114 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2115 			    dr->ifp) != 0)
2116 				log(LOG_ERR,
2117 				    "scope error in "
2118 				    "default router list (%s)\n",
2119 				    ip6_sprintf(&dr->rtaddr));
2120 			d->flags = dr->flags;
2121 			d->rtlifetime = dr->rtlifetime;
2122 			d->expire = dr->expire;
2123 			d->if_index = dr->ifp->if_index;
2124 		} else
2125 			panic("buffer too short");
2126 
2127 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2128 		if (error)
2129 			break;
2130 	}
2131 	return error;
2132 }
2133 
2134 static int
2135 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2136 {
2137 	int error;
2138 	char buf[1024];
2139 	struct in6_prefix *p, *pe;
2140 	struct nd_prefix *pr;
2141 
2142 	if (req->newptr)
2143 		return EPERM;
2144 	error = 0;
2145 
2146 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2147 		u_short advrtrs;
2148 		size_t advance;
2149 		struct sockaddr_in6 *sin6, *s6;
2150 		struct nd_pfxrouter *pfr;
2151 
2152 		p = (struct in6_prefix *)buf;
2153 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2154 
2155 		if (p + 1 <= pe) {
2156 			bzero(p, sizeof(*p));
2157 			sin6 = (struct sockaddr_in6 *)(p + 1);
2158 
2159 			p->prefix = pr->ndpr_prefix;
2160 			if (in6_recoverscope(&p->prefix,
2161 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2162 				log(LOG_ERR,
2163 				    "scope error in prefix list (%s)\n",
2164 				    ip6_sprintf(&p->prefix.sin6_addr));
2165 			p->raflags = pr->ndpr_raf;
2166 			p->prefixlen = pr->ndpr_plen;
2167 			p->vltime = pr->ndpr_vltime;
2168 			p->pltime = pr->ndpr_pltime;
2169 			p->if_index = pr->ndpr_ifp->if_index;
2170 			p->expire = pr->ndpr_expire;
2171 			p->refcnt = pr->ndpr_refcnt;
2172 			p->flags = pr->ndpr_stateflags;
2173 			p->origin = PR_ORIG_RA;
2174 			advrtrs = 0;
2175 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2176 			     pfr = pfr->pfr_next) {
2177 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2178 					advrtrs++;
2179 					continue;
2180 				}
2181 				s6 = &sin6[advrtrs];
2182 				bzero(s6, sizeof(*s6));
2183 				s6->sin6_family = AF_INET6;
2184 				s6->sin6_len = sizeof(*sin6);
2185 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2186 				    pfr->router->ifp) != 0)
2187 					log(LOG_ERR,
2188 					    "scope error in "
2189 					    "prefix list (%s)\n",
2190 					    ip6_sprintf(&pfr->router->rtaddr));
2191 				advrtrs++;
2192 			}
2193 			p->advrtrs = advrtrs;
2194 		} else {
2195 			panic("buffer too short");
2196 		}
2197 
2198 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2199 		error = SYSCTL_OUT(req, buf, advance);
2200 		if (error)
2201 			break;
2202 	}
2203 	return error;
2204 }
2205