xref: /dflybsd-src/sys/netinet6/nd6.c (revision 31c068aaf635ad9fa72dbc4c65b32d890ff7544d)
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 /*
34  * XXX
35  * KAME 970409 note:
36  * BSD/OS version heavily modifies this code, related to llinfo.
37  * Since we don't have BSD/OS version of net/route.c in our hand,
38  * I left the code mostly as it was in 970310.  -- itojun
39  */
40 
41 #include "opt_inet.h"
42 #include "opt_inet6.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/sysctl.h>
58 #include <sys/mutex.h>
59 
60 #include <sys/thread2.h>
61 #include <sys/mutex2.h>
62 
63 #include <net/if.h>
64 #include <net/if_dl.h>
65 #include <net/if_types.h>
66 #include <net/route.h>
67 #include <net/netisr2.h>
68 #include <net/netmsg2.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/icmp6.h>
77 
78 #include <net/net_osdep.h>
79 
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 #define SDL(s) ((struct sockaddr_dl *)s)
85 
86 /* timer values */
87 int	nd6_prune	= 1;	/* walk list every 1 seconds */
88 int	nd6_delay	= 5;	/* delay first probe time 5 second */
89 int	nd6_umaxtries	= 3;	/* maximum unicast query */
90 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
91 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
92 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
93 
94 /* preventing too many loops in ND option parsing */
95 int nd6_maxndopt = 10;	/* max # of ND options allowed */
96 
97 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
98 
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104 
105 /* for debugging? */
106 static int nd6_inuse, nd6_allocated;
107 
108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
109 struct nd_drhead nd_defrouter;
110 struct nd_prhead nd_prefix = { 0 };
111 struct mtx nd6_mtx = MTX_INITIALIZER;
112 
113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
114 static struct sockaddr_in6 all1_sa;
115 
116 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
117 static int regen_tmpaddr (struct in6_ifaddr *);
118 static void nd6_slowtimo(void *);
119 static void nd6_slowtimo_dispatch(netmsg_t);
120 
121 static struct callout nd6_slowtimo_ch;
122 static struct netmsg_base nd6_slowtimo_netmsg;
123 
124 struct callout nd6_timer_ch;
125 
126 void
127 nd6_init(void)
128 {
129 	static int nd6_init_done = 0;
130 	int i;
131 
132 	if (nd6_init_done) {
133 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
134 		return;
135 	}
136 
137 	all1_sa.sin6_family = AF_INET6;
138 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
139 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
140 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
141 
142 	/* initialization of the default router list */
143 	TAILQ_INIT(&nd_defrouter);
144 
145 	nd6_init_done = 1;
146 
147 	/* start timer */
148 	callout_init_mp(&nd6_slowtimo_ch);
149 	netmsg_init(&nd6_slowtimo_netmsg, NULL, &netisr_adone_rport,
150 	    MSGF_PRIORITY, nd6_slowtimo_dispatch);
151 	callout_reset_bycpu(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
152 	    nd6_slowtimo, NULL, 0);
153 }
154 
155 struct nd_ifinfo *
156 nd6_ifattach(struct ifnet *ifp)
157 {
158 	struct nd_ifinfo *nd;
159 
160 	nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
161 	    M_WAITOK | M_ZERO);
162 
163 	nd->initialized = 1;
164 
165 	nd->linkmtu = ifp->if_mtu;
166 	nd->chlim = IPV6_DEFHLIM;
167 	nd->basereachable = REACHABLE_TIME;
168 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
169 	nd->retrans = RETRANS_TIMER;
170 	nd->receivedra = 0;
171 
172 	/*
173 	 * Note that the default value of ip6_accept_rtadv is 0, which means
174 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
175 	 * here.
176 	 */
177 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
178 
179 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
180 	nd6_setmtu0(ifp, nd);
181 	return nd;
182 }
183 
184 void
185 nd6_ifdetach(struct nd_ifinfo *nd)
186 {
187 	kfree(nd, M_IP6NDP);
188 }
189 
190 /*
191  * Reset ND level link MTU. This function is called when the physical MTU
192  * changes, which means we might have to adjust the ND level MTU.
193  */
194 void
195 nd6_setmtu(struct ifnet *ifp)
196 {
197 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
198 }
199 
200 struct netmsg_nd6setmtu {
201 	struct netmsg_base	nmsg;
202 	struct ifnet		*ifp;
203 	struct nd_ifinfo	*ndi;
204 };
205 
206 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
207 static void
208 nd6_setmtu0_dispatch(netmsg_t msg)
209 {
210 	struct netmsg_nd6setmtu *nmsg = (struct netmsg_nd6setmtu *)msg;
211 	struct ifnet *ifp = nmsg->ifp;
212 	struct nd_ifinfo *ndi = nmsg->ndi;
213 	u_long oldmaxmtu;
214 	u_long oldlinkmtu;
215 
216 	oldmaxmtu = ndi->maxmtu;
217 	oldlinkmtu = ndi->linkmtu;
218 
219 	switch (ifp->if_type) {
220 	case IFT_ETHER:
221 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
222 		break;
223 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
224 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
225 		break;
226 #ifdef IFT_IEEE80211
227 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
228 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
229 		break;
230 #endif
231 	default:
232 		ndi->maxmtu = ifp->if_mtu;
233 		break;
234 	}
235 
236 	if (oldmaxmtu != ndi->maxmtu) {
237 		/*
238 		 * If the ND level MTU is not set yet, or if the maxmtu
239 		 * is reset to a smaller value than the ND level MTU,
240 		 * also reset the ND level MTU.
241 		 */
242 		if (ndi->linkmtu == 0 ||
243 		    ndi->maxmtu < ndi->linkmtu) {
244 			ndi->linkmtu = ndi->maxmtu;
245 			/* also adjust in6_maxmtu if necessary. */
246 			if (oldlinkmtu == 0) {
247 				/*
248 				 * XXX: the case analysis is grotty, but
249 				 * it is not efficient to call in6_setmaxmtu()
250 				 * here when we are during the initialization
251 				 * procedure.
252 				 */
253 				if (in6_maxmtu < ndi->linkmtu)
254 					in6_maxmtu = ndi->linkmtu;
255 			} else
256 				in6_setmaxmtu();
257 		}
258 	}
259 #undef MIN
260 
261 	lwkt_replymsg(&nmsg->nmsg.lmsg, 0);
262 }
263 
264 void
265 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
266 {
267 	struct netmsg_nd6setmtu nmsg;
268 
269 	netmsg_init(&nmsg.nmsg, NULL, &curthread->td_msgport, 0,
270 	    nd6_setmtu0_dispatch);
271 	nmsg.ifp = ifp;
272 	nmsg.ndi = ndi;
273 	lwkt_domsg(netisr_cpuport(0), &nmsg.nmsg.lmsg, 0);
274 }
275 
276 void
277 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
278 {
279 	bzero(ndopts, sizeof(*ndopts));
280 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
281 	ndopts->nd_opts_last
282 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
283 
284 	if (icmp6len == 0) {
285 		ndopts->nd_opts_done = 1;
286 		ndopts->nd_opts_search = NULL;
287 	}
288 }
289 
290 /*
291  * Take one ND option.
292  */
293 struct nd_opt_hdr *
294 nd6_option(union nd_opts *ndopts)
295 {
296 	struct nd_opt_hdr *nd_opt;
297 	int olen;
298 
299 	if (!ndopts)
300 		panic("ndopts == NULL in nd6_option");
301 	if (!ndopts->nd_opts_last)
302 		panic("uninitialized ndopts in nd6_option");
303 	if (!ndopts->nd_opts_search)
304 		return NULL;
305 	if (ndopts->nd_opts_done)
306 		return NULL;
307 
308 	nd_opt = ndopts->nd_opts_search;
309 
310 	/* make sure nd_opt_len is inside the buffer */
311 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
312 		bzero(ndopts, sizeof(*ndopts));
313 		return NULL;
314 	}
315 
316 	olen = nd_opt->nd_opt_len << 3;
317 	if (olen == 0) {
318 		/*
319 		 * Message validation requires that all included
320 		 * options have a length that is greater than zero.
321 		 */
322 		bzero(ndopts, sizeof(*ndopts));
323 		return NULL;
324 	}
325 
326 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
327 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
328 		/* option overruns the end of buffer, invalid */
329 		bzero(ndopts, sizeof(*ndopts));
330 		return NULL;
331 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
332 		/* reached the end of options chain */
333 		ndopts->nd_opts_done = 1;
334 		ndopts->nd_opts_search = NULL;
335 	}
336 	return nd_opt;
337 }
338 
339 /*
340  * Parse multiple ND options.
341  * This function is much easier to use, for ND routines that do not need
342  * multiple options of the same type.
343  */
344 int
345 nd6_options(union nd_opts *ndopts)
346 {
347 	struct nd_opt_hdr *nd_opt;
348 	int i = 0;
349 
350 	if (!ndopts)
351 		panic("ndopts == NULL in nd6_options");
352 	if (!ndopts->nd_opts_last)
353 		panic("uninitialized ndopts in nd6_options");
354 	if (!ndopts->nd_opts_search)
355 		return 0;
356 
357 	while (1) {
358 		nd_opt = nd6_option(ndopts);
359 		if (!nd_opt && !ndopts->nd_opts_last) {
360 			/*
361 			 * Message validation requires that all included
362 			 * options have a length that is greater than zero.
363 			 */
364 			icmp6stat.icp6s_nd_badopt++;
365 			bzero(ndopts, sizeof(*ndopts));
366 			return -1;
367 		}
368 
369 		if (!nd_opt)
370 			goto skip1;
371 
372 		switch (nd_opt->nd_opt_type) {
373 		case ND_OPT_SOURCE_LINKADDR:
374 		case ND_OPT_TARGET_LINKADDR:
375 		case ND_OPT_MTU:
376 		case ND_OPT_REDIRECTED_HEADER:
377 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
378 				nd6log((LOG_INFO,
379 				    "duplicated ND6 option found (type=%d)\n",
380 				    nd_opt->nd_opt_type));
381 				/* XXX bark? */
382 			} else {
383 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
384 					= nd_opt;
385 			}
386 			break;
387 		case ND_OPT_PREFIX_INFORMATION:
388 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
389 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
390 					= nd_opt;
391 			}
392 			ndopts->nd_opts_pi_end =
393 				(struct nd_opt_prefix_info *)nd_opt;
394 			break;
395 		default:
396 			/*
397 			 * Unknown options must be silently ignored,
398 			 * to accomodate future extension to the protocol.
399 			 */
400 			nd6log((LOG_DEBUG,
401 			    "nd6_options: unsupported option %d - "
402 			    "option ignored\n", nd_opt->nd_opt_type));
403 		}
404 
405 skip1:
406 		i++;
407 		if (i > nd6_maxndopt) {
408 			icmp6stat.icp6s_nd_toomanyopt++;
409 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
410 			break;
411 		}
412 
413 		if (ndopts->nd_opts_done)
414 			break;
415 	}
416 
417 	return 0;
418 }
419 
420 /*
421  * ND6 timer routine to expire default route list and prefix list
422  */
423 void
424 nd6_timer(void *ignored_arg)
425 {
426 	struct llinfo_nd6 *ln;
427 	struct nd_defrouter *dr;
428 	struct nd_prefix *pr;
429 	struct ifnet *ifp;
430 	struct in6_ifaddr *ia6, *nia6;
431 
432 	mtx_lock(&nd6_mtx);
433 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
434 		      nd6_timer, NULL);
435 
436 	ln = llinfo_nd6.ln_next;
437 	while (ln && ln != &llinfo_nd6) {
438 		struct rtentry *rt;
439 		struct sockaddr_in6 *dst;
440 		struct llinfo_nd6 *next = ln->ln_next;
441 		/* XXX: used for the DELAY case only: */
442 		struct nd_ifinfo *ndi = NULL;
443 
444 		if ((rt = ln->ln_rt) == NULL) {
445 			ln = next;
446 			continue;
447 		}
448 		if ((ifp = rt->rt_ifp) == NULL) {
449 			ln = next;
450 			continue;
451 		}
452 		ndi = ND_IFINFO(ifp);
453 		dst = (struct sockaddr_in6 *)rt_key(rt);
454 
455 		if (ln->ln_expire > time_uptime) {
456 			ln = next;
457 			continue;
458 		}
459 
460 		/* sanity check */
461 		if (!rt)
462 			panic("rt=0 in nd6_timer(ln=%p)", ln);
463 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
464 			panic("rt_llinfo(%p) is not equal to ln(%p)",
465 			      rt->rt_llinfo, ln);
466 		if (!dst)
467 			panic("dst=0 in nd6_timer(ln=%p)", ln);
468 
469 		switch (ln->ln_state) {
470 		case ND6_LLINFO_INCOMPLETE:
471 			if (ln->ln_asked < nd6_mmaxtries) {
472 				ln->ln_asked++;
473 				ln->ln_expire = time_uptime +
474 					ND_IFINFO(ifp)->retrans / 1000;
475 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
476 					ln, 0);
477 			} else {
478 				struct mbuf *m = ln->ln_hold;
479 				if (m) {
480 					if (rt->rt_ifp) {
481 						/*
482 						 * Fake rcvif to make ICMP error
483 						 * more helpful in diagnosing
484 						 * for the receiver.
485 						 * XXX: should we consider
486 						 * older rcvif?
487 						 */
488 						m->m_pkthdr.rcvif = rt->rt_ifp;
489 					}
490 					icmp6_error(m, ICMP6_DST_UNREACH,
491 						    ICMP6_DST_UNREACH_ADDR, 0);
492 					ln->ln_hold = NULL;
493 				}
494 				next = nd6_free(rt);
495 			}
496 			break;
497 		case ND6_LLINFO_REACHABLE:
498 			if (ln->ln_expire) {
499 				ln->ln_state = ND6_LLINFO_STALE;
500 				ln->ln_expire = time_uptime + nd6_gctimer;
501 			}
502 			break;
503 
504 		case ND6_LLINFO_STALE:
505 			/* Garbage Collection(RFC 2461 5.3) */
506 			if (ln->ln_expire)
507 				next = nd6_free(rt);
508 			break;
509 
510 		case ND6_LLINFO_DELAY:
511 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
512 				/* We need NUD */
513 				ln->ln_asked = 1;
514 				ln->ln_state = ND6_LLINFO_PROBE;
515 				ln->ln_expire = time_uptime +
516 					ndi->retrans / 1000;
517 				nd6_ns_output(ifp, &dst->sin6_addr,
518 					      &dst->sin6_addr,
519 					      ln, 0);
520 			} else {
521 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
522 				ln->ln_expire = time_uptime + nd6_gctimer;
523 			}
524 			break;
525 		case ND6_LLINFO_PROBE:
526 			if (ln->ln_asked < nd6_umaxtries) {
527 				ln->ln_asked++;
528 				ln->ln_expire = time_uptime +
529 					ND_IFINFO(ifp)->retrans / 1000;
530 				nd6_ns_output(ifp, &dst->sin6_addr,
531 					       &dst->sin6_addr, ln, 0);
532 			} else {
533 				next = nd6_free(rt);
534 			}
535 			break;
536 		}
537 		ln = next;
538 	}
539 
540 	/* expire default router list */
541 	dr = TAILQ_FIRST(&nd_defrouter);
542 	while (dr) {
543 		if (dr->expire && dr->expire < time_uptime) {
544 			struct nd_defrouter *t;
545 			t = TAILQ_NEXT(dr, dr_entry);
546 			defrtrlist_del(dr);
547 			dr = t;
548 		} else {
549 			dr = TAILQ_NEXT(dr, dr_entry);
550 		}
551 	}
552 
553 	/*
554 	 * expire interface addresses.
555 	 * in the past the loop was inside prefix expiry processing.
556 	 * However, from a stricter speci-confrmance standpoint, we should
557 	 * rather separate address lifetimes and prefix lifetimes.
558 	 */
559 addrloop:
560 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
561 		nia6 = ia6->ia_next;
562 		/* check address lifetime */
563 		if (IFA6_IS_INVALID(ia6)) {
564 			int regen = 0;
565 
566 			/*
567 			 * If the expiring address is temporary, try
568 			 * regenerating a new one.  This would be useful when
569 			 * we suspended a laptop PC, then turned it on after a
570 			 * period that could invalidate all temporary
571 			 * addresses.  Although we may have to restart the
572 			 * loop (see below), it must be after purging the
573 			 * address.  Otherwise, we'd see an infinite loop of
574 			 * regeneration.
575 			 */
576 			if (ip6_use_tempaddr &&
577 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
578 				if (regen_tmpaddr(ia6) == 0)
579 					regen = 1;
580 			}
581 
582 			in6_purgeaddr(&ia6->ia_ifa);
583 
584 			if (regen)
585 				goto addrloop; /* XXX: see below */
586 		}
587 		if (IFA6_IS_DEPRECATED(ia6)) {
588 			int oldflags = ia6->ia6_flags;
589 
590 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
591 
592 			/*
593 			 * If a temporary address has just become deprecated,
594 			 * regenerate a new one if possible.
595 			 */
596 			if (ip6_use_tempaddr &&
597 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
598 			    !(oldflags & IN6_IFF_DEPRECATED)) {
599 
600 				if (regen_tmpaddr(ia6) == 0) {
601 					/*
602 					 * A new temporary address is
603 					 * generated.
604 					 * XXX: this means the address chain
605 					 * has changed while we are still in
606 					 * the loop.  Although the change
607 					 * would not cause disaster (because
608 					 * it's not a deletion, but an
609 					 * addition,) we'd rather restart the
610 					 * loop just for safety.  Or does this
611 					 * significantly reduce performance??
612 					 */
613 					goto addrloop;
614 				}
615 			}
616 		} else {
617 			/*
618 			 * A new RA might have made a deprecated address
619 			 * preferred.
620 			 */
621 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
622 		}
623 	}
624 
625 	/* expire prefix list */
626 	pr = nd_prefix.lh_first;
627 	while (pr) {
628 		/*
629 		 * check prefix lifetime.
630 		 * since pltime is just for autoconf, pltime processing for
631 		 * prefix is not necessary.
632 		 */
633 		if (pr->ndpr_expire && pr->ndpr_expire < time_uptime) {
634 			struct nd_prefix *t;
635 			t = pr->ndpr_next;
636 
637 			/*
638 			 * address expiration and prefix expiration are
639 			 * separate.  NEVER perform in6_purgeaddr here.
640 			 */
641 
642 			prelist_remove(pr);
643 			pr = t;
644 		} else
645 			pr = pr->ndpr_next;
646 	}
647 	mtx_unlock(&nd6_mtx);
648 }
649 
650 static int
651 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
652 					 address */
653 {
654 	struct ifaddr_container *ifac;
655 	struct ifnet *ifp;
656 	struct in6_ifaddr *public_ifa6 = NULL;
657 
658 	ifp = ia6->ia_ifa.ifa_ifp;
659 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
660 		struct ifaddr *ifa = ifac->ifa;
661 		struct in6_ifaddr *it6;
662 
663 		if (ifa->ifa_addr->sa_family != AF_INET6)
664 			continue;
665 
666 		it6 = (struct in6_ifaddr *)ifa;
667 
668 		/* ignore no autoconf addresses. */
669 		if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
670 			continue;
671 
672 		/* ignore autoconf addresses with different prefixes. */
673 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
674 			continue;
675 
676 		/*
677 		 * Now we are looking at an autoconf address with the same
678 		 * prefix as ours.  If the address is temporary and is still
679 		 * preferred, do not create another one.  It would be rare, but
680 		 * could happen, for example, when we resume a laptop PC after
681 		 * a long period.
682 		 */
683 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
684 		    !IFA6_IS_DEPRECATED(it6)) {
685 			public_ifa6 = NULL;
686 			break;
687 		}
688 
689 		/*
690 		 * This is a public autoconf address that has the same prefix
691 		 * as ours.  If it is preferred, keep it.  We can't break the
692 		 * loop here, because there may be a still-preferred temporary
693 		 * address with the prefix.
694 		 */
695 		if (!IFA6_IS_DEPRECATED(it6))
696 		    public_ifa6 = it6;
697 	}
698 
699 	if (public_ifa6 != NULL) {
700 		int e;
701 
702 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
703 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
704 			    " tmp addr,errno=%d\n", e);
705 			return (-1);
706 		}
707 		return (0);
708 	}
709 
710 	return (-1);
711 }
712 
713 /*
714  * Nuke neighbor cache/prefix/default router management table, right before
715  * ifp goes away.
716  */
717 void
718 nd6_purge(struct ifnet *ifp)
719 {
720 	struct llinfo_nd6 *ln, *nln;
721 	struct nd_defrouter *dr, *ndr, drany;
722 	struct nd_prefix *pr, *npr;
723 
724 	/* Nuke default router list entries toward ifp */
725 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
726 		/*
727 		 * The first entry of the list may be stored in
728 		 * the routing table, so we'll delete it later.
729 		 */
730 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
731 			ndr = TAILQ_NEXT(dr, dr_entry);
732 			if (dr->ifp == ifp)
733 				defrtrlist_del(dr);
734 		}
735 		dr = TAILQ_FIRST(&nd_defrouter);
736 		if (dr->ifp == ifp)
737 			defrtrlist_del(dr);
738 	}
739 
740 	/* Nuke prefix list entries toward ifp */
741 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
742 		npr = pr->ndpr_next;
743 		if (pr->ndpr_ifp == ifp) {
744 			/*
745 			 * Previously, pr->ndpr_addr is removed as well,
746 			 * but I strongly believe we don't have to do it.
747 			 * nd6_purge() is only called from in6_ifdetach(),
748 			 * which removes all the associated interface addresses
749 			 * by itself.
750 			 * (jinmei@kame.net 20010129)
751 			 */
752 			prelist_remove(pr);
753 		}
754 	}
755 
756 	/* cancel default outgoing interface setting */
757 	if (nd6_defifindex == ifp->if_index)
758 		nd6_setdefaultiface(0);
759 
760 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
761 		/* refresh default router list */
762 		bzero(&drany, sizeof(drany));
763 		defrouter_delreq(&drany, 0);
764 		defrouter_select();
765 	}
766 
767 	/*
768 	 * Nuke neighbor cache entries for the ifp.
769 	 * Note that rt->rt_ifp may not be the same as ifp,
770 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
771 	 * nd6_rtrequest(), and ip6_input().
772 	 */
773 	ln = llinfo_nd6.ln_next;
774 	while (ln && ln != &llinfo_nd6) {
775 		struct rtentry *rt;
776 		struct sockaddr_dl *sdl;
777 
778 		nln = ln->ln_next;
779 		rt = ln->ln_rt;
780 		if (rt && rt->rt_gateway &&
781 		    rt->rt_gateway->sa_family == AF_LINK) {
782 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
783 			if (sdl->sdl_index == ifp->if_index)
784 				nln = nd6_free(rt);
785 		}
786 		ln = nln;
787 	}
788 }
789 
790 struct rtentry *
791 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
792 {
793 	struct rtentry *rt;
794 	struct sockaddr_in6 sin6;
795 
796 	bzero(&sin6, sizeof(sin6));
797 	sin6.sin6_len = sizeof(struct sockaddr_in6);
798 	sin6.sin6_family = AF_INET6;
799 	sin6.sin6_addr = *addr6;
800 
801 	if (create)
802 		rt = rtlookup((struct sockaddr *)&sin6);
803 	else
804 		rt = rtpurelookup((struct sockaddr *)&sin6);
805 	if (rt && !(rt->rt_flags & RTF_LLINFO)) {
806 		/*
807 		 * This is the case for the default route.
808 		 * If we want to create a neighbor cache for the address, we
809 		 * should free the route for the destination and allocate an
810 		 * interface route.
811 		 */
812 		if (create) {
813 			--rt->rt_refcnt;
814 			rt = NULL;
815 		}
816 	}
817 	if (!rt) {
818 		if (create && ifp) {
819 			int e;
820 
821 			/*
822 			 * If no route is available and create is set,
823 			 * we allocate a host route for the destination
824 			 * and treat it like an interface route.
825 			 * This hack is necessary for a neighbor which can't
826 			 * be covered by our own prefix.
827 			 */
828 			struct ifaddr *ifa =
829 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
830 			if (ifa == NULL)
831 				return (NULL);
832 
833 			/*
834 			 * Create a new route.  RTF_LLINFO is necessary
835 			 * to create a Neighbor Cache entry for the
836 			 * destination in nd6_rtrequest which will be
837 			 * called in rtrequest via ifa->ifa_rtrequest.
838 			 */
839 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
840 					   ifa->ifa_addr,
841 					   (struct sockaddr *)&all1_sa,
842 					   (ifa->ifa_flags |
843 					    RTF_HOST | RTF_LLINFO) &
844 					   ~RTF_CLONING,
845 					   &rt)) != 0)
846 				log(LOG_ERR,
847 				    "nd6_lookup: failed to add route for a "
848 				    "neighbor(%s), errno=%d\n",
849 				    ip6_sprintf(addr6), e);
850 			if (rt == NULL)
851 				return (NULL);
852 			if (rt->rt_llinfo) {
853 				struct llinfo_nd6 *ln =
854 					(struct llinfo_nd6 *)rt->rt_llinfo;
855 				ln->ln_state = ND6_LLINFO_NOSTATE;
856 			}
857 		} else
858 			return (NULL);
859 	}
860 	rt->rt_refcnt--;
861 	/*
862 	 * Validation for the entry.
863 	 * Note that the check for rt_llinfo is necessary because a cloned
864 	 * route from a parent route that has the L flag (e.g. the default
865 	 * route to a p2p interface) may have the flag, too, while the
866 	 * destination is not actually a neighbor.
867 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
868 	 *      it might be the loopback interface if the entry is for our
869 	 *      own address on a non-loopback interface. Instead, we should
870 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
871 	 *      interface.
872 	 */
873 	if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
874 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
875 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
876 		if (create) {
877 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
878 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
879 			/* xxx more logs... kazu */
880 		}
881 		return (NULL);
882 	}
883 	return (rt);
884 }
885 
886 /*
887  * Detect if a given IPv6 address identifies a neighbor on a given link.
888  * XXX: should take care of the destination of a p2p link?
889  */
890 int
891 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
892 {
893 	struct ifaddr_container *ifac;
894 	int i;
895 
896 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
897 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
898 
899 	/*
900 	 * A link-local address is always a neighbor.
901 	 * XXX: we should use the sin6_scope_id field rather than the embedded
902 	 * interface index.
903 	 */
904 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
905 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
906 		return (1);
907 
908 	/*
909 	 * If the address matches one of our addresses,
910 	 * it should be a neighbor.
911 	 */
912 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
913 		struct ifaddr *ifa = ifac->ifa;
914 
915 		if (ifa->ifa_addr->sa_family != AF_INET6)
916 			next: continue;
917 
918 		for (i = 0; i < 4; i++) {
919 			if ((IFADDR6(ifa).s6_addr32[i] ^
920 			     addr->sin6_addr.s6_addr32[i]) &
921 			    IFMASK6(ifa).s6_addr32[i])
922 				goto next;
923 		}
924 		return (1);
925 	}
926 
927 	/*
928 	 * Even if the address matches none of our addresses, it might be
929 	 * in the neighbor cache.
930 	 */
931 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
932 		return (1);
933 
934 	return (0);
935 #undef IFADDR6
936 #undef IFMASK6
937 }
938 
939 /*
940  * Free an nd6 llinfo entry.
941  */
942 struct llinfo_nd6 *
943 nd6_free(struct rtentry *rt)
944 {
945 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
946 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
947 	struct nd_defrouter *dr;
948 
949 	/*
950 	 * we used to have kpfctlinput(PRC_HOSTDEAD) here.
951 	 * even though it is not harmful, it was not really necessary.
952 	 */
953 
954 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
955 		mtx_lock(&nd6_mtx);
956 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
957 				      rt->rt_ifp);
958 
959 		if (ln->ln_router || dr) {
960 			/*
961 			 * rt6_flush must be called whether or not the neighbor
962 			 * is in the Default Router List.
963 			 * See a corresponding comment in nd6_na_input().
964 			 */
965 			rt6_flush(&in6, rt->rt_ifp);
966 		}
967 
968 		if (dr) {
969 			/*
970 			 * Unreachablity of a router might affect the default
971 			 * router selection and on-link detection of advertised
972 			 * prefixes.
973 			 */
974 
975 			/*
976 			 * Temporarily fake the state to choose a new default
977 			 * router and to perform on-link determination of
978 			 * prefixes correctly.
979 			 * Below the state will be set correctly,
980 			 * or the entry itself will be deleted.
981 			 */
982 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
983 
984 			/*
985 			 * Since defrouter_select() does not affect the
986 			 * on-link determination and MIP6 needs the check
987 			 * before the default router selection, we perform
988 			 * the check now.
989 			 */
990 			pfxlist_onlink_check();
991 
992 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
993 				/*
994 				 * It is used as the current default router,
995 				 * so we have to move it to the end of the
996 				 * list and choose a new one.
997 				 * XXX: it is not very efficient if this is
998 				 *      the only router.
999 				 */
1000 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1001 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1002 
1003 				defrouter_select();
1004 			}
1005 		}
1006 		mtx_unlock(&nd6_mtx);
1007 	}
1008 
1009 	/*
1010 	 * Before deleting the entry, remember the next entry as the
1011 	 * return value.  We need this because pfxlist_onlink_check() above
1012 	 * might have freed other entries (particularly the old next entry) as
1013 	 * a side effect (XXX).
1014 	 */
1015 	next = ln->ln_next;
1016 
1017 	/*
1018 	 * Detach the route from the routing tree and the list of neighbor
1019 	 * caches, and disable the route entry not to be used in already
1020 	 * cached routes.
1021 	 */
1022 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
1023 
1024 	return (next);
1025 }
1026 
1027 /*
1028  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1029  *
1030  * XXX cost-effective metods?
1031  */
1032 void
1033 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1034 {
1035 	struct llinfo_nd6 *ln;
1036 
1037 	/*
1038 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1039 	 * routing table by supplied "dst6".
1040 	 */
1041 	if (!rt) {
1042 		if (!dst6)
1043 			return;
1044 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1045 			return;
1046 	}
1047 
1048 	if ((rt->rt_flags & RTF_GATEWAY) ||
1049 	    !(rt->rt_flags & RTF_LLINFO) ||
1050 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1051 	    rt->rt_gateway->sa_family != AF_LINK) {
1052 		/* This is not a host route. */
1053 		return;
1054 	}
1055 
1056 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1057 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1058 		return;
1059 
1060 	/*
1061 	 * if we get upper-layer reachability confirmation many times,
1062 	 * it is possible we have false information.
1063 	 */
1064 	if (!force) {
1065 		ln->ln_byhint++;
1066 		if (ln->ln_byhint > nd6_maxnudhint)
1067 			return;
1068 	}
1069 
1070 	ln->ln_state = ND6_LLINFO_REACHABLE;
1071 	if (ln->ln_expire)
1072 		ln->ln_expire = time_uptime +
1073 			ND_IFINFO(rt->rt_ifp)->reachable;
1074 }
1075 
1076 void
1077 nd6_rtrequest(int req, struct rtentry *rt)
1078 {
1079 	struct sockaddr *gate = rt->rt_gateway;
1080 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1081 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1082 	struct ifnet *ifp = rt->rt_ifp;
1083 	struct ifaddr *ifa;
1084 
1085 	if ((rt->rt_flags & RTF_GATEWAY))
1086 		return;
1087 
1088 	if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1089 		/*
1090 		 * This is probably an interface direct route for a link
1091 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1092 		 * We do not need special treatment below for such a route.
1093 		 * Moreover, the RTF_LLINFO flag which would be set below
1094 		 * would annoy the ndp(8) command.
1095 		 */
1096 		return;
1097 	}
1098 
1099 	if (req == RTM_RESOLVE &&
1100 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1101 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1102 		/*
1103 		 * FreeBSD and BSD/OS often make a cloned host route based
1104 		 * on a less-specific route (e.g. the default route).
1105 		 * If the less specific route does not have a "gateway"
1106 		 * (this is the case when the route just goes to a p2p or an
1107 		 * stf interface), we'll mistakenly make a neighbor cache for
1108 		 * the host route, and will see strange neighbor solicitation
1109 		 * for the corresponding destination.  In order to avoid the
1110 		 * confusion, we check if the destination of the route is
1111 		 * a neighbor in terms of neighbor discovery, and stop the
1112 		 * process if not.  Additionally, we remove the LLINFO flag
1113 		 * so that ndp(8) will not try to get the neighbor information
1114 		 * of the destination.
1115 		 */
1116 		rt->rt_flags &= ~RTF_LLINFO;
1117 		return;
1118 	}
1119 
1120 	switch (req) {
1121 	case RTM_ADD:
1122 		/*
1123 		 * There is no backward compatibility :)
1124 		 *
1125 		 * if (!(rt->rt_flags & RTF_HOST) &&
1126 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1127 		 *	   rt->rt_flags |= RTF_CLONING;
1128 		 */
1129 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1130 			/*
1131 			 * Case 1: This route should come from
1132 			 * a route to interface.  RTF_LLINFO flag is set
1133 			 * for a host route whose destination should be
1134 			 * treated as on-link.
1135 			 */
1136 			rt_setgate(rt, rt_key(rt),
1137 				   (struct sockaddr *)&null_sdl,
1138 				   RTL_DONTREPORT);
1139 			gate = rt->rt_gateway;
1140 			SDL(gate)->sdl_type = ifp->if_type;
1141 			SDL(gate)->sdl_index = ifp->if_index;
1142 			if (ln)
1143 				ln->ln_expire = time_uptime;
1144 #if 1
1145 			if (ln && ln->ln_expire == 0) {
1146 				/* kludge for desktops */
1147 #if 0
1148 				kprintf("nd6_rtequest: time.tv_sec is zero; "
1149 				       "treat it as 1\n");
1150 #endif
1151 				ln->ln_expire = 1;
1152 			}
1153 #endif
1154 			if ((rt->rt_flags & RTF_CLONING))
1155 				break;
1156 		}
1157 		/*
1158 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1159 		 * We don't do that here since llinfo is not ready yet.
1160 		 *
1161 		 * There are also couple of other things to be discussed:
1162 		 * - unsolicited NA code needs improvement beforehand
1163 		 * - RFC2461 says we MAY send multicast unsolicited NA
1164 		 *   (7.2.6 paragraph 4), however, it also says that we
1165 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1166 		 *   we don't have anything like it right now.
1167 		 *   note that the mechanism needs a mutual agreement
1168 		 *   between proxies, which means that we need to implement
1169 		 *   a new protocol, or a new kludge.
1170 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1171 		 *   we need to check ip6forwarding before sending it.
1172 		 *   (or should we allow proxy ND configuration only for
1173 		 *   routers?  there's no mention about proxy ND from hosts)
1174 		 */
1175 #if 0
1176 		/* XXX it does not work */
1177 		if (rt->rt_flags & RTF_ANNOUNCE)
1178 			nd6_na_output(ifp,
1179 			      &SIN6(rt_key(rt))->sin6_addr,
1180 			      &SIN6(rt_key(rt))->sin6_addr,
1181 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1182 			      1, NULL);
1183 #endif
1184 		/* FALLTHROUGH */
1185 	case RTM_RESOLVE:
1186 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1187 			/*
1188 			 * Address resolution isn't necessary for a point to
1189 			 * point link, so we can skip this test for a p2p link.
1190 			 */
1191 			if (gate->sa_family != AF_LINK ||
1192 			    gate->sa_len < sizeof(null_sdl)) {
1193 				log(LOG_DEBUG,
1194 				    "nd6_rtrequest: bad gateway value: %s\n",
1195 				    if_name(ifp));
1196 				break;
1197 			}
1198 			SDL(gate)->sdl_type = ifp->if_type;
1199 			SDL(gate)->sdl_index = ifp->if_index;
1200 		}
1201 		if (ln != NULL)
1202 			break;	/* This happens on a route change */
1203 		/*
1204 		 * Case 2: This route may come from cloning, or a manual route
1205 		 * add with a LL address.
1206 		 */
1207 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1208 		rt->rt_llinfo = (caddr_t)ln;
1209 		if (!ln) {
1210 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1211 			break;
1212 		}
1213 		nd6_inuse++;
1214 		nd6_allocated++;
1215 		bzero(ln, sizeof(*ln));
1216 		ln->ln_rt = rt;
1217 		/* this is required for "ndp" command. - shin */
1218 		if (req == RTM_ADD) {
1219 		        /*
1220 			 * gate should have some valid AF_LINK entry,
1221 			 * and ln->ln_expire should have some lifetime
1222 			 * which is specified by ndp command.
1223 			 */
1224 			ln->ln_state = ND6_LLINFO_REACHABLE;
1225 			ln->ln_byhint = 0;
1226 		} else {
1227 		        /*
1228 			 * When req == RTM_RESOLVE, rt is created and
1229 			 * initialized in rtrequest(), so rt_expire is 0.
1230 			 */
1231 			ln->ln_state = ND6_LLINFO_NOSTATE;
1232 			ln->ln_expire = time_uptime;
1233 		}
1234 		rt->rt_flags |= RTF_LLINFO;
1235 		ln->ln_next = llinfo_nd6.ln_next;
1236 		llinfo_nd6.ln_next = ln;
1237 		ln->ln_prev = &llinfo_nd6;
1238 		ln->ln_next->ln_prev = ln;
1239 
1240 		/*
1241 		 * check if rt_key(rt) is one of my address assigned
1242 		 * to the interface.
1243 		 */
1244 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1245 					  &SIN6(rt_key(rt))->sin6_addr);
1246 		if (ifa) {
1247 			caddr_t macp = nd6_ifptomac(ifp);
1248 			ln->ln_expire = 0;
1249 			ln->ln_state = ND6_LLINFO_REACHABLE;
1250 			ln->ln_byhint = 0;
1251 			if (macp) {
1252 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1253 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1254 			}
1255 			if (nd6_useloopback) {
1256 				rt->rt_ifp = &loif[0];	/* XXX */
1257 				/*
1258 				 * Make sure rt_ifa be equal to the ifaddr
1259 				 * corresponding to the address.
1260 				 * We need this because when we refer
1261 				 * rt_ifa->ia6_flags in ip6_input, we assume
1262 				 * that the rt_ifa points to the address instead
1263 				 * of the loopback address.
1264 				 */
1265 				if (ifa != rt->rt_ifa) {
1266 					IFAFREE(rt->rt_ifa);
1267 					IFAREF(ifa);
1268 					rt->rt_ifa = ifa;
1269 				}
1270 			}
1271 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1272 			ln->ln_expire = 0;
1273 			ln->ln_state = ND6_LLINFO_REACHABLE;
1274 			ln->ln_byhint = 0;
1275 
1276 			/* join solicited node multicast for proxy ND */
1277 			if (ifp->if_flags & IFF_MULTICAST) {
1278 				struct in6_addr llsol;
1279 				int error;
1280 
1281 				llsol = SIN6(rt_key(rt))->sin6_addr;
1282 				llsol.s6_addr16[0] = htons(0xff02);
1283 				llsol.s6_addr16[1] = htons(ifp->if_index);
1284 				llsol.s6_addr32[1] = 0;
1285 				llsol.s6_addr32[2] = htonl(1);
1286 				llsol.s6_addr8[12] = 0xff;
1287 
1288 				if (!in6_addmulti(&llsol, ifp, &error)) {
1289 					nd6log((LOG_ERR, "%s: failed to join "
1290 					    "%s (errno=%d)\n", if_name(ifp),
1291 					    ip6_sprintf(&llsol), error));
1292 				}
1293 			}
1294 		}
1295 		break;
1296 
1297 	case RTM_DELETE:
1298 		if (!ln)
1299 			break;
1300 		/* leave from solicited node multicast for proxy ND */
1301 		if ((rt->rt_flags & RTF_ANNOUNCE) &&
1302 		    (ifp->if_flags & IFF_MULTICAST)) {
1303 			struct in6_addr llsol;
1304 			struct in6_multi *in6m;
1305 
1306 			llsol = SIN6(rt_key(rt))->sin6_addr;
1307 			llsol.s6_addr16[0] = htons(0xff02);
1308 			llsol.s6_addr16[1] = htons(ifp->if_index);
1309 			llsol.s6_addr32[1] = 0;
1310 			llsol.s6_addr32[2] = htonl(1);
1311 			llsol.s6_addr8[12] = 0xff;
1312 
1313 			in6m = IN6_LOOKUP_MULTI(&llsol, ifp);
1314 			if (in6m)
1315 				in6_delmulti(in6m);
1316 		}
1317 		nd6_inuse--;
1318 		ln->ln_next->ln_prev = ln->ln_prev;
1319 		ln->ln_prev->ln_next = ln->ln_next;
1320 		ln->ln_prev = NULL;
1321 		rt->rt_llinfo = 0;
1322 		rt->rt_flags &= ~RTF_LLINFO;
1323 		if (ln->ln_hold)
1324 			m_freem(ln->ln_hold);
1325 		Free((caddr_t)ln);
1326 	}
1327 }
1328 
1329 int
1330 nd6_ioctl(u_long cmd, caddr_t	data, struct ifnet *ifp)
1331 {
1332 	struct in6_drlist *drl = (struct in6_drlist *)data;
1333 	struct in6_prlist *prl = (struct in6_prlist *)data;
1334 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1335 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1336 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1337 	struct nd_defrouter *dr, any;
1338 	struct nd_prefix *pr;
1339 	struct rtentry *rt;
1340 	int i = 0, error = 0;
1341 
1342 	switch (cmd) {
1343 	case SIOCGDRLST_IN6:
1344 		/*
1345 		 * obsolete API, use sysctl under net.inet6.icmp6
1346 		 */
1347 		bzero(drl, sizeof(*drl));
1348 		mtx_lock(&nd6_mtx);
1349 		dr = TAILQ_FIRST(&nd_defrouter);
1350 		while (dr && i < DRLSTSIZ) {
1351 			drl->defrouter[i].rtaddr = dr->rtaddr;
1352 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1353 				/* XXX: need to this hack for KAME stack */
1354 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1355 			} else
1356 				log(LOG_ERR,
1357 				    "default router list contains a "
1358 				    "non-linklocal address(%s)\n",
1359 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1360 
1361 			drl->defrouter[i].flags = dr->flags;
1362 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1363 			drl->defrouter[i].expire = dr->expire;
1364 			drl->defrouter[i].if_index = dr->ifp->if_index;
1365 			i++;
1366 			dr = TAILQ_NEXT(dr, dr_entry);
1367 		}
1368 		mtx_unlock(&nd6_mtx);
1369 		break;
1370 	case SIOCGPRLST_IN6:
1371 		/*
1372 		 * obsolete API, use sysctl under net.inet6.icmp6
1373 		 */
1374 		/*
1375 		 * XXX meaning of fields, especialy "raflags", is very
1376 		 * differnet between RA prefix list and RR/static prefix list.
1377 		 * how about separating ioctls into two?
1378 		 */
1379 		bzero(prl, sizeof(*prl));
1380 		mtx_lock(&nd6_mtx);
1381 		pr = nd_prefix.lh_first;
1382 		while (pr && i < PRLSTSIZ) {
1383 			struct nd_pfxrouter *pfr;
1384 			int j;
1385 
1386 			in6_embedscope(&prl->prefix[i].prefix,
1387 			    &pr->ndpr_prefix, NULL, NULL);
1388 			prl->prefix[i].raflags = pr->ndpr_raf;
1389 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1390 			prl->prefix[i].vltime = pr->ndpr_vltime;
1391 			prl->prefix[i].pltime = pr->ndpr_pltime;
1392 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1393 			prl->prefix[i].expire = pr->ndpr_expire;
1394 
1395 			pfr = pr->ndpr_advrtrs.lh_first;
1396 			j = 0;
1397 			while (pfr) {
1398 				if (j < DRLSTSIZ) {
1399 #define RTRADDR prl->prefix[i].advrtr[j]
1400 					RTRADDR = pfr->router->rtaddr;
1401 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1402 						/* XXX: hack for KAME */
1403 						RTRADDR.s6_addr16[1] = 0;
1404 					} else
1405 						log(LOG_ERR,
1406 						    "a router(%s) advertises "
1407 						    "a prefix with "
1408 						    "non-link local address\n",
1409 						    ip6_sprintf(&RTRADDR));
1410 #undef RTRADDR
1411 				}
1412 				j++;
1413 				pfr = pfr->pfr_next;
1414 			}
1415 			prl->prefix[i].advrtrs = j;
1416 			prl->prefix[i].origin = PR_ORIG_RA;
1417 
1418 			i++;
1419 			pr = pr->ndpr_next;
1420 		}
1421 		mtx_unlock(&nd6_mtx);
1422 
1423 		break;
1424 	case OSIOCGIFINFO_IN6:
1425 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1426 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1427 		ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1428 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1429 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1430 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1431 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1432 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1433 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1434 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1435 		ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1436 		break;
1437 	case SIOCGIFINFO_IN6:
1438 		ndi->ndi = *ND_IFINFO(ifp);
1439 		break;
1440 	case SIOCSIFINFO_FLAGS:
1441 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1442 		break;
1443 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1444 		/* flush default router list */
1445 		/*
1446 		 * xxx sumikawa: should not delete route if default
1447 		 * route equals to the top of default router list
1448 		 */
1449 		bzero(&any, sizeof(any));
1450 		defrouter_delreq(&any, 0);
1451 		defrouter_select();
1452 		/* xxx sumikawa: flush prefix list */
1453 		break;
1454 	case SIOCSPFXFLUSH_IN6:
1455 	    {
1456 		/* flush all the prefix advertised by routers */
1457 		struct nd_prefix *pr, *next;
1458 
1459 		mtx_lock(&nd6_mtx);
1460 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1461 			struct in6_ifaddr *ia, *ia_next;
1462 
1463 			next = pr->ndpr_next;
1464 
1465 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1466 				continue; /* XXX */
1467 
1468 			/* do we really have to remove addresses as well? */
1469 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1470 				/* ia might be removed.  keep the next ptr. */
1471 				ia_next = ia->ia_next;
1472 
1473 				if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1474 					continue;
1475 
1476 				if (ia->ia6_ndpr == pr)
1477 					in6_purgeaddr(&ia->ia_ifa);
1478 			}
1479 			prelist_remove(pr);
1480 		}
1481 		mtx_unlock(&nd6_mtx);
1482 		break;
1483 	    }
1484 	case SIOCSRTRFLUSH_IN6:
1485 	    {
1486 		/* flush all the default routers */
1487 		struct nd_defrouter *dr, *next;
1488 
1489 		mtx_lock(&nd6_mtx);
1490 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1491 			/*
1492 			 * The first entry of the list may be stored in
1493 			 * the routing table, so we'll delete it later.
1494 			 */
1495 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1496 				next = TAILQ_NEXT(dr, dr_entry);
1497 				defrtrlist_del(dr);
1498 			}
1499 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1500 		}
1501 		mtx_unlock(&nd6_mtx);
1502 		break;
1503 	    }
1504 	case SIOCGNBRINFO_IN6:
1505 	    {
1506 		struct llinfo_nd6 *ln;
1507 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1508 
1509 		/*
1510 		 * XXX: KAME specific hack for scoped addresses
1511 		 *      XXXX: for other scopes than link-local?
1512 		 */
1513 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1514 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1515 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1516 
1517 			if (*idp == 0)
1518 				*idp = htons(ifp->if_index);
1519 		}
1520 
1521 		mtx_lock(&nd6_mtx);
1522 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1523 			error = EINVAL;
1524 			mtx_unlock(&nd6_mtx);
1525 			break;
1526 		}
1527 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1528 		nbi->state = ln->ln_state;
1529 		nbi->asked = ln->ln_asked;
1530 		nbi->isrouter = ln->ln_router;
1531 		nbi->expire = ln->ln_expire;
1532 		mtx_unlock(&nd6_mtx);
1533 
1534 		break;
1535 	    }
1536 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1537 		ndif->ifindex = nd6_defifindex;
1538 		break;
1539 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1540 		return (nd6_setdefaultiface(ndif->ifindex));
1541 		break;
1542 	}
1543 	return (error);
1544 }
1545 
1546 /*
1547  * Create neighbor cache entry and cache link-layer address,
1548  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1549  */
1550 struct rtentry *
1551 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1552 		 int lladdrlen,
1553 		 int type,	/* ICMP6 type */
1554 		 int code	/* type dependent information */)
1555 {
1556 	struct rtentry *rt = NULL;
1557 	struct llinfo_nd6 *ln = NULL;
1558 	int is_newentry;
1559 	struct sockaddr_dl *sdl = NULL;
1560 	int do_update;
1561 	int olladdr;
1562 	int llchange;
1563 	int newstate = 0;
1564 
1565 	if (!ifp)
1566 		panic("ifp == NULL in nd6_cache_lladdr");
1567 	if (!from)
1568 		panic("from == NULL in nd6_cache_lladdr");
1569 
1570 	/* nothing must be updated for unspecified address */
1571 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1572 		return NULL;
1573 
1574 	/*
1575 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1576 	 * the caller.
1577 	 *
1578 	 * XXX If the link does not have link-layer adderss, what should
1579 	 * we do? (ifp->if_addrlen == 0)
1580 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1581 	 * description on it in NS section (RFC 2461 7.2.3).
1582 	 */
1583 
1584 	rt = nd6_lookup(from, 0, ifp);
1585 	if (!rt) {
1586 #if 0
1587 		/* nothing must be done if there's no lladdr */
1588 		if (!lladdr || !lladdrlen)
1589 			return NULL;
1590 #endif
1591 
1592 		rt = nd6_lookup(from, 1, ifp);
1593 		is_newentry = 1;
1594 	} else {
1595 		/* do nothing if static ndp is set */
1596 		if (rt->rt_flags & RTF_STATIC)
1597 			return NULL;
1598 		is_newentry = 0;
1599 	}
1600 
1601 	if (!rt)
1602 		return NULL;
1603 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1604 fail:
1605 		nd6_free(rt);
1606 		return NULL;
1607 	}
1608 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1609 	if (!ln)
1610 		goto fail;
1611 	if (!rt->rt_gateway)
1612 		goto fail;
1613 	if (rt->rt_gateway->sa_family != AF_LINK)
1614 		goto fail;
1615 	sdl = SDL(rt->rt_gateway);
1616 
1617 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1618 	if (olladdr && lladdr) {
1619 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1620 			llchange = 1;
1621 		else
1622 			llchange = 0;
1623 	} else
1624 		llchange = 0;
1625 
1626 	/*
1627 	 * newentry olladdr  lladdr  llchange	(*=record)
1628 	 *	0	n	n	--	(1)
1629 	 *	0	y	n	--	(2)
1630 	 *	0	n	y	--	(3) * STALE
1631 	 *	0	y	y	n	(4) *
1632 	 *	0	y	y	y	(5) * STALE
1633 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1634 	 *	1	--	y	--	(7) * STALE
1635 	 */
1636 
1637 	if (lladdr) {		/* (3-5) and (7) */
1638 		/*
1639 		 * Record source link-layer address
1640 		 * XXX is it dependent to ifp->if_type?
1641 		 */
1642 		sdl->sdl_alen = ifp->if_addrlen;
1643 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1644 	}
1645 
1646 	if (!is_newentry) {
1647 		if ((!olladdr && lladdr)		/* (3) */
1648 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1649 			do_update = 1;
1650 			newstate = ND6_LLINFO_STALE;
1651 		} else					/* (1-2,4) */
1652 			do_update = 0;
1653 	} else {
1654 		do_update = 1;
1655 		if (!lladdr)				/* (6) */
1656 			newstate = ND6_LLINFO_NOSTATE;
1657 		else					/* (7) */
1658 			newstate = ND6_LLINFO_STALE;
1659 	}
1660 
1661 	if (do_update) {
1662 		/*
1663 		 * Update the state of the neighbor cache.
1664 		 */
1665 		ln->ln_state = newstate;
1666 
1667 		if (ln->ln_state == ND6_LLINFO_STALE) {
1668 			/*
1669 			 * XXX: since nd6_output() below will cause
1670 			 * state tansition to DELAY and reset the timer,
1671 			 * we must set the timer now, although it is actually
1672 			 * meaningless.
1673 			 */
1674 			ln->ln_expire = time_uptime + nd6_gctimer;
1675 
1676 			if (ln->ln_hold) {
1677 				/*
1678 				 * we assume ifp is not a p2p here, so just
1679 				 * set the 2nd argument as the 1st one.
1680 				 */
1681 				nd6_output(ifp, ifp, ln->ln_hold,
1682 					   (struct sockaddr_in6 *)rt_key(rt),
1683 					   rt);
1684 				ln->ln_hold = NULL;
1685 			}
1686 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1687 			/* probe right away */
1688 			ln->ln_expire = time_uptime;
1689 		}
1690 	}
1691 
1692 	/*
1693 	 * ICMP6 type dependent behavior.
1694 	 *
1695 	 * NS: clear IsRouter if new entry
1696 	 * RS: clear IsRouter
1697 	 * RA: set IsRouter if there's lladdr
1698 	 * redir: clear IsRouter if new entry
1699 	 *
1700 	 * RA case, (1):
1701 	 * The spec says that we must set IsRouter in the following cases:
1702 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1703 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1704 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1705 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1706 	 * neighbor cache, this is similar to (6).
1707 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1708 	 *
1709 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1710 	 *							D R
1711 	 *	0	n	n	--	(1)	c   ?     s
1712 	 *	0	y	n	--	(2)	c   s     s
1713 	 *	0	n	y	--	(3)	c   s     s
1714 	 *	0	y	y	n	(4)	c   s     s
1715 	 *	0	y	y	y	(5)	c   s     s
1716 	 *	1	--	n	--	(6) c	c 	c s
1717 	 *	1	--	y	--	(7) c	c   s	c s
1718 	 *
1719 	 *					(c=clear s=set)
1720 	 */
1721 	switch (type & 0xff) {
1722 	case ND_NEIGHBOR_SOLICIT:
1723 		/*
1724 		 * New entry must have is_router flag cleared.
1725 		 */
1726 		if (is_newentry)	/* (6-7) */
1727 			ln->ln_router = 0;
1728 		break;
1729 	case ND_REDIRECT:
1730 		/*
1731 		 * If the icmp is a redirect to a better router, always set the
1732 		 * is_router flag. Otherwise, if the entry is newly created,
1733 		 * clear the flag. [RFC 2461, sec 8.3]
1734 		 */
1735 		if (code == ND_REDIRECT_ROUTER)
1736 			ln->ln_router = 1;
1737 		else if (is_newentry) /* (6-7) */
1738 			ln->ln_router = 0;
1739 		break;
1740 	case ND_ROUTER_SOLICIT:
1741 		/*
1742 		 * is_router flag must always be cleared.
1743 		 */
1744 		ln->ln_router = 0;
1745 		break;
1746 	case ND_ROUTER_ADVERT:
1747 		/*
1748 		 * Mark an entry with lladdr as a router.
1749 		 */
1750 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1751 		 || (is_newentry && lladdr)) {			/* (7) */
1752 			ln->ln_router = 1;
1753 		}
1754 		break;
1755 	}
1756 
1757 	/*
1758 	 * When the link-layer address of a router changes, select the
1759 	 * best router again.  In particular, when the neighbor entry is newly
1760 	 * created, it might affect the selection policy.
1761 	 * Question: can we restrict the first condition to the "is_newentry"
1762 	 * case?
1763 	 * XXX: when we hear an RA from a new router with the link-layer
1764 	 * address option, defrouter_select() is called twice, since
1765 	 * defrtrlist_update called the function as well.  However, I believe
1766 	 * we can compromise the overhead, since it only happens the first
1767 	 * time.
1768 	 * XXX: although defrouter_select() should not have a bad effect
1769 	 * for those are not autoconfigured hosts, we explicitly avoid such
1770 	 * cases for safety.
1771 	 */
1772 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1773 		defrouter_select();
1774 
1775 	return rt;
1776 }
1777 
1778 static void
1779 nd6_slowtimo(void *arg __unused)
1780 {
1781 	struct lwkt_msg *lmsg = &nd6_slowtimo_netmsg.lmsg;
1782 
1783 	KASSERT(mycpuid == 0, ("not on cpu0"));
1784 	crit_enter();
1785 	if (lmsg->ms_flags & MSGF_DONE)
1786 		lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1787 	crit_exit();
1788 }
1789 
1790 static void
1791 nd6_slowtimo_dispatch(netmsg_t nmsg)
1792 {
1793 	struct nd_ifinfo *nd6if;
1794 	struct ifnet *ifp;
1795 
1796 	KASSERT(&curthread->td_msgport == netisr_cpuport(0),
1797 	    ("not in netisr0"));
1798 
1799 	crit_enter();
1800 	lwkt_replymsg(&nmsg->lmsg, 0);	/* reply ASAP */
1801 	crit_exit();
1802 
1803 	mtx_lock(&nd6_mtx);
1804 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1805 		if (ifp->if_afdata[AF_INET6] == NULL)
1806 			continue;
1807 		nd6if = ND_IFINFO(ifp);
1808 		if (nd6if->basereachable && /* already initialized */
1809 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1810 			/*
1811 			 * Since reachable time rarely changes by router
1812 			 * advertisements, we SHOULD insure that a new random
1813 			 * value gets recomputed at least once every few hours.
1814 			 * (RFC 2461, 6.3.4)
1815 			 */
1816 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1817 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1818 		}
1819 	}
1820 	mtx_unlock(&nd6_mtx);
1821 
1822 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1823 	    nd6_slowtimo, NULL);
1824 }
1825 
1826 #define gotoerr(e) { error = (e); goto bad;}
1827 
1828 int
1829 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1830 	   struct sockaddr_in6 *dst, struct rtentry *rt)
1831 {
1832 	struct llinfo_nd6 *ln = NULL;
1833 	int error = 0;
1834 
1835 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1836 		goto sendpkt;
1837 
1838 	if (nd6_need_cache(ifp) == 0)
1839 		goto sendpkt;
1840 
1841 	/*
1842 	 * next hop determination.  This routine is derived from ether_outpout.
1843 	 */
1844 	if (rt != NULL) {
1845 		if (!(rt->rt_flags & RTF_UP)) {
1846 			rt = rtlookup((struct sockaddr *)dst);
1847 			if (rt == NULL)
1848 				gotoerr(EHOSTUNREACH);
1849 			rt->rt_refcnt--;
1850 			if (rt->rt_ifp != ifp) {
1851 				/* XXX: loop care? */
1852 				return nd6_output(ifp, origifp, m, dst, rt);
1853 			}
1854 		}
1855 		if (rt->rt_flags & RTF_GATEWAY) {
1856 			struct sockaddr_in6 *gw6;
1857 
1858 			/*
1859 			 * We skip link-layer address resolution and NUD
1860 			 * if the gateway is not a neighbor from ND point
1861 			 * of view, regardless of the value of nd_ifinfo.flags.
1862 			 * The second condition is a bit tricky; we skip
1863 			 * if the gateway is our own address, which is
1864 			 * sometimes used to install a route to a p2p link.
1865 			 */
1866 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1867 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1868 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1869 				/*
1870 				 * We allow this kind of tricky route only
1871 				 * when the outgoing interface is p2p.
1872 				 * XXX: we may need a more generic rule here.
1873 				 */
1874 				if (!(ifp->if_flags & IFF_POINTOPOINT))
1875 					gotoerr(EHOSTUNREACH);
1876 
1877 				goto sendpkt;
1878 			}
1879 
1880 			if (rt->rt_gwroute == NULL) {
1881 				rt->rt_gwroute = rtlookup(rt->rt_gateway);
1882 				if (rt->rt_gwroute == NULL)
1883 					gotoerr(EHOSTUNREACH);
1884 			} else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1885 				rtfree(rt->rt_gwroute);
1886 				rt->rt_gwroute = rtlookup(rt->rt_gateway);
1887 				if (rt->rt_gwroute == NULL)
1888 					gotoerr(EHOSTUNREACH);
1889 			}
1890 		}
1891 	}
1892 
1893 	/*
1894 	 * Address resolution or Neighbor Unreachability Detection
1895 	 * for the next hop.
1896 	 * At this point, the destination of the packet must be a unicast
1897 	 * or an anycast address(i.e. not a multicast).
1898 	 */
1899 
1900 	/* Look up the neighbor cache for the nexthop */
1901 	if (rt && (rt->rt_flags & RTF_LLINFO))
1902 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1903 	else {
1904 		/*
1905 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1906 		 * the condition below is not very efficient.  But we believe
1907 		 * it is tolerable, because this should be a rare case.
1908 		 */
1909 		if (nd6_is_addr_neighbor(dst, ifp) &&
1910 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1911 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1912 	}
1913 	if (!ln || !rt) {
1914 		if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1915 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1916 			log(LOG_DEBUG,
1917 			    "nd6_output: can't allocate llinfo for %s "
1918 			    "(ln=%p, rt=%p)\n",
1919 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1920 			gotoerr(EIO);	/* XXX: good error? */
1921 		}
1922 
1923 		goto sendpkt;	/* send anyway */
1924 	}
1925 
1926 	/* We don't have to do link-layer address resolution on a p2p link. */
1927 	if ((ifp->if_flags & IFF_POINTOPOINT) &&
1928 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1929 		ln->ln_state = ND6_LLINFO_STALE;
1930 		ln->ln_expire = time_uptime + nd6_gctimer;
1931 	}
1932 
1933 	/*
1934 	 * The first time we send a packet to a neighbor whose entry is
1935 	 * STALE, we have to change the state to DELAY and a sets a timer to
1936 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1937 	 * neighbor unreachability detection on expiration.
1938 	 * (RFC 2461 7.3.3)
1939 	 */
1940 	if (ln->ln_state == ND6_LLINFO_STALE) {
1941 		ln->ln_asked = 0;
1942 		ln->ln_state = ND6_LLINFO_DELAY;
1943 		ln->ln_expire = time_uptime + nd6_delay;
1944 	}
1945 
1946 	/*
1947 	 * If the neighbor cache entry has a state other than INCOMPLETE
1948 	 * (i.e. its link-layer address is already resolved), just
1949 	 * send the packet.
1950 	 */
1951 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1952 		goto sendpkt;
1953 
1954 	/*
1955 	 * There is a neighbor cache entry, but no ethernet address
1956 	 * response yet.  Replace the held mbuf (if any) with this
1957 	 * latest one.
1958 	 *
1959 	 * This code conforms to the rate-limiting rule described in Section
1960 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1961 	 * an NS below.
1962 	 */
1963 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1964 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1965 	if (ln->ln_hold)
1966 		m_freem(ln->ln_hold);
1967 	ln->ln_hold = m;
1968 	if (ln->ln_expire) {
1969 		if (ln->ln_asked < nd6_mmaxtries &&
1970 		    ln->ln_expire < time_uptime) {
1971 			ln->ln_asked++;
1972 			ln->ln_expire = time_uptime +
1973 				ND_IFINFO(ifp)->retrans / 1000;
1974 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1975 		}
1976 	}
1977 	return (0);
1978 
1979 sendpkt:
1980 	if (ifp->if_flags & IFF_LOOPBACK)
1981 		error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1982 	else
1983 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1984 	return (error);
1985 
1986 bad:
1987 	m_freem(m);
1988 	return (error);
1989 }
1990 #undef gotoerr
1991 
1992 int
1993 nd6_need_cache(struct ifnet *ifp)
1994 {
1995 	/*
1996 	 * XXX: we currently do not make neighbor cache on any interface
1997 	 * other than Ethernet and GIF.
1998 	 *
1999 	 * RFC2893 says:
2000 	 * - unidirectional tunnels needs no ND
2001 	 */
2002 	switch (ifp->if_type) {
2003 	case IFT_ETHER:
2004 	case IFT_IEEE1394:
2005 #ifdef IFT_L2VLAN
2006 	case IFT_L2VLAN:
2007 #endif
2008 #ifdef IFT_IEEE80211
2009 	case IFT_IEEE80211:
2010 #endif
2011 #ifdef IFT_CARP
2012 	case IFT_CARP:
2013 #endif
2014 	case IFT_GIF:		/* XXX need more cases? */
2015 		return (1);
2016 	default:
2017 		return (0);
2018 	}
2019 }
2020 
2021 int
2022 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2023 		struct sockaddr *dst, u_char *desten)
2024 {
2025 	struct sockaddr_dl *sdl;
2026 	struct rtentry *rt;
2027 
2028 
2029 	if (m->m_flags & M_MCAST) {
2030 		switch (ifp->if_type) {
2031 		case IFT_ETHER:
2032 #ifdef IFT_L2VLAN
2033 	case IFT_L2VLAN:
2034 #endif
2035 #ifdef IFT_IEEE80211
2036 		case IFT_IEEE80211:
2037 #endif
2038 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2039 						 desten);
2040 			return (1);
2041 		case IFT_IEEE1394:
2042 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2043 			return (1);
2044 		default:
2045 			m_freem(m);
2046 			return (0);
2047 		}
2048 	}
2049 	if (rt0 == NULL) {
2050 		/* this could happen, if we could not allocate memory */
2051 		m_freem(m);
2052 		return (0);
2053 	}
2054 	if (rt_llroute(dst, rt0, &rt) != 0) {
2055 		m_freem(m);
2056 		return (0);
2057 	}
2058 	if (rt->rt_gateway->sa_family != AF_LINK) {
2059 		kprintf("nd6_storelladdr: something odd happens\n");
2060 		m_freem(m);
2061 		return (0);
2062 	}
2063 	sdl = SDL(rt->rt_gateway);
2064 	if (sdl->sdl_alen == 0) {
2065 		/* this should be impossible, but we bark here for debugging */
2066 		kprintf("nd6_storelladdr: sdl_alen == 0\n");
2067 		m_freem(m);
2068 		return (0);
2069 	}
2070 
2071 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2072 	return (1);
2073 }
2074 
2075 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2076 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2077 #ifdef SYSCTL_DECL
2078 SYSCTL_DECL(_net_inet6_icmp6);
2079 #endif
2080 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2081 	CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2082 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2083 	CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2084 
2085 static int
2086 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2087 {
2088 	int error;
2089 	char buf[1024];
2090 	struct in6_defrouter *d, *de;
2091 	struct nd_defrouter *dr;
2092 
2093 	if (req->newptr)
2094 		return EPERM;
2095 	error = 0;
2096 
2097 	for (dr = TAILQ_FIRST(&nd_defrouter);
2098 	     dr;
2099 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2100 		d = (struct in6_defrouter *)buf;
2101 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2102 
2103 		if (d + 1 <= de) {
2104 			bzero(d, sizeof(*d));
2105 			d->rtaddr.sin6_family = AF_INET6;
2106 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2107 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2108 			    dr->ifp) != 0)
2109 				log(LOG_ERR,
2110 				    "scope error in "
2111 				    "default router list (%s)\n",
2112 				    ip6_sprintf(&dr->rtaddr));
2113 			d->flags = dr->flags;
2114 			d->rtlifetime = dr->rtlifetime;
2115 			d->expire = dr->expire;
2116 			d->if_index = dr->ifp->if_index;
2117 		} else
2118 			panic("buffer too short");
2119 
2120 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2121 		if (error)
2122 			break;
2123 	}
2124 	return error;
2125 }
2126 
2127 static int
2128 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2129 {
2130 	int error;
2131 	char buf[1024];
2132 	struct in6_prefix *p, *pe;
2133 	struct nd_prefix *pr;
2134 
2135 	if (req->newptr)
2136 		return EPERM;
2137 	error = 0;
2138 
2139 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2140 		u_short advrtrs;
2141 		size_t advance;
2142 		struct sockaddr_in6 *sin6, *s6;
2143 		struct nd_pfxrouter *pfr;
2144 
2145 		p = (struct in6_prefix *)buf;
2146 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2147 
2148 		if (p + 1 <= pe) {
2149 			bzero(p, sizeof(*p));
2150 			sin6 = (struct sockaddr_in6 *)(p + 1);
2151 
2152 			p->prefix = pr->ndpr_prefix;
2153 			if (in6_recoverscope(&p->prefix,
2154 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2155 				log(LOG_ERR,
2156 				    "scope error in prefix list (%s)\n",
2157 				    ip6_sprintf(&p->prefix.sin6_addr));
2158 			p->raflags = pr->ndpr_raf;
2159 			p->prefixlen = pr->ndpr_plen;
2160 			p->vltime = pr->ndpr_vltime;
2161 			p->pltime = pr->ndpr_pltime;
2162 			p->if_index = pr->ndpr_ifp->if_index;
2163 			p->expire = pr->ndpr_expire;
2164 			p->refcnt = pr->ndpr_refcnt;
2165 			p->flags = pr->ndpr_stateflags;
2166 			p->origin = PR_ORIG_RA;
2167 			advrtrs = 0;
2168 			for (pfr = pr->ndpr_advrtrs.lh_first;
2169 			     pfr;
2170 			     pfr = pfr->pfr_next) {
2171 				if ((void *)&sin6[advrtrs + 1] >
2172 				    (void *)pe) {
2173 					advrtrs++;
2174 					continue;
2175 				}
2176 				s6 = &sin6[advrtrs];
2177 				bzero(s6, sizeof(*s6));
2178 				s6->sin6_family = AF_INET6;
2179 				s6->sin6_len = sizeof(*sin6);
2180 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2181 						     pfr->router->ifp) != 0)
2182 					log(LOG_ERR,
2183 					    "scope error in "
2184 					    "prefix list (%s)\n",
2185 					    ip6_sprintf(&pfr->router->rtaddr));
2186 				advrtrs++;
2187 			}
2188 			p->advrtrs = advrtrs;
2189 		} else
2190 			panic("buffer too short");
2191 
2192 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2193 		error = SYSCTL_OUT(req, buf, advance);
2194 		if (error)
2195 			break;
2196 	}
2197 	return error;
2198 }
2199