xref: /freebsd/sys/netinet6/nd6.c (revision 6af83ee0d2941d18880b6aaa2b4facd1d30c6106)
1 /*	$FreeBSD$	*/
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 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/callout.h>
40 #include <sys/mac.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/queue.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/if.h>
54 #include <net/if_arc.h>
55 #include <net/if_dl.h>
56 #include <net/if_types.h>
57 #include <net/if_atm.h>
58 #include <net/iso88025.h>
59 #include <net/fddi.h>
60 #include <net/route.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 /* timer values */
79 int	nd6_prune	= 1;	/* walk list every 1 seconds */
80 int	nd6_delay	= 5;	/* delay first probe time 5 second */
81 int	nd6_umaxtries	= 3;	/* maximum unicast query */
82 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
83 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
84 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
85 
86 /* preventing too many loops in ND option parsing */
87 int nd6_maxndopt = 10;	/* max # of ND options allowed */
88 
89 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
90 
91 #ifdef ND6_DEBUG
92 int nd6_debug = 1;
93 #else
94 int nd6_debug = 0;
95 #endif
96 
97 /* for debugging? */
98 static int nd6_inuse, nd6_allocated;
99 
100 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
101 struct nd_drhead nd_defrouter;
102 struct nd_prhead nd_prefix = { 0 };
103 
104 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
105 static struct sockaddr_in6 all1_sa;
106 
107 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
108 	struct ifnet *));
109 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
110 static void nd6_slowtimo __P((void *));
111 static int regen_tmpaddr __P((struct in6_ifaddr *));
112 
113 struct callout nd6_slowtimo_ch;
114 struct callout nd6_timer_ch;
115 extern struct callout in6_tmpaddrtimer_ch;
116 
117 void
118 nd6_init()
119 {
120 	static int nd6_init_done = 0;
121 	int i;
122 
123 	if (nd6_init_done) {
124 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
125 		return;
126 	}
127 
128 	all1_sa.sin6_family = AF_INET6;
129 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
130 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
131 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
132 
133 	/* initialization of the default router list */
134 	TAILQ_INIT(&nd_defrouter);
135 
136 	nd6_init_done = 1;
137 
138 	/* start timer */
139 	callout_init(&nd6_slowtimo_ch, 0);
140 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
141 	    nd6_slowtimo, NULL);
142 }
143 
144 struct nd_ifinfo *
145 nd6_ifattach(ifp)
146 	struct ifnet *ifp;
147 {
148 	struct nd_ifinfo *nd;
149 
150 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
151 	bzero(nd, sizeof(*nd));
152 
153 	nd->initialized = 1;
154 
155 	nd->chlim = IPV6_DEFHLIM;
156 	nd->basereachable = REACHABLE_TIME;
157 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
158 	nd->retrans = RETRANS_TIMER;
159 	/*
160 	 * Note that the default value of ip6_accept_rtadv is 0, which means
161 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
162 	 * here.
163 	 */
164 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
165 
166 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
167 	nd6_setmtu0(ifp, nd);
168 
169 	return nd;
170 }
171 
172 void
173 nd6_ifdetach(nd)
174 	struct nd_ifinfo *nd;
175 {
176 
177 	free(nd, M_IP6NDP);
178 }
179 
180 /*
181  * Reset ND level link MTU. This function is called when the physical MTU
182  * changes, which means we might have to adjust the ND level MTU.
183  */
184 void
185 nd6_setmtu(ifp)
186 	struct ifnet *ifp;
187 {
188 
189 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
190 }
191 
192 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
193 void
194 nd6_setmtu0(ifp, ndi)
195 	struct ifnet *ifp;
196 	struct nd_ifinfo *ndi;
197 {
198 	u_int32_t omaxmtu;
199 
200 	omaxmtu = ndi->maxmtu;
201 
202 	switch (ifp->if_type) {
203 	case IFT_ARCNET:
204 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
205 		break;
206 	case IFT_ETHER:
207 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
208 		break;
209 	case IFT_FDDI:
210 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
211 		break;
212 	case IFT_ATM:
213 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
214 		break;
215 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
216 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
217 		break;
218 #ifdef IFT_IEEE80211
219 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
220 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
221 		break;
222 #endif
223 	 case IFT_ISO88025:
224 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
225 		 break;
226 	default:
227 		ndi->maxmtu = ifp->if_mtu;
228 		break;
229 	}
230 
231 	/*
232 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
233 	 * undesirable situation.  We thus notify the operator of the change
234 	 * explicitly.  The check for omaxmtu is necessary to restrict the
235 	 * log to the case of changing the MTU, not initializing it.
236 	 */
237 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
238 		log(LOG_NOTICE, "nd6_setmtu0: "
239 		    "new link MTU on %s (%lu) is too small for IPv6\n",
240 		    if_name(ifp), (unsigned long)ndi->maxmtu);
241 	}
242 
243 	if (ndi->maxmtu > in6_maxmtu)
244 		in6_setmaxmtu(); /* check all interfaces just in case */
245 
246 #undef MIN
247 }
248 
249 void
250 nd6_option_init(opt, icmp6len, ndopts)
251 	void *opt;
252 	int icmp6len;
253 	union nd_opts *ndopts;
254 {
255 
256 	bzero(ndopts, sizeof(*ndopts));
257 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
258 	ndopts->nd_opts_last
259 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
260 
261 	if (icmp6len == 0) {
262 		ndopts->nd_opts_done = 1;
263 		ndopts->nd_opts_search = NULL;
264 	}
265 }
266 
267 /*
268  * Take one ND option.
269  */
270 struct nd_opt_hdr *
271 nd6_option(ndopts)
272 	union nd_opts *ndopts;
273 {
274 	struct nd_opt_hdr *nd_opt;
275 	int olen;
276 
277 	if (!ndopts)
278 		panic("ndopts == NULL in nd6_option");
279 	if (!ndopts->nd_opts_last)
280 		panic("uninitialized ndopts in nd6_option");
281 	if (!ndopts->nd_opts_search)
282 		return NULL;
283 	if (ndopts->nd_opts_done)
284 		return NULL;
285 
286 	nd_opt = ndopts->nd_opts_search;
287 
288 	/* make sure nd_opt_len is inside the buffer */
289 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
290 		bzero(ndopts, sizeof(*ndopts));
291 		return NULL;
292 	}
293 
294 	olen = nd_opt->nd_opt_len << 3;
295 	if (olen == 0) {
296 		/*
297 		 * Message validation requires that all included
298 		 * options have a length that is greater than zero.
299 		 */
300 		bzero(ndopts, sizeof(*ndopts));
301 		return NULL;
302 	}
303 
304 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
305 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
306 		/* option overruns the end of buffer, invalid */
307 		bzero(ndopts, sizeof(*ndopts));
308 		return NULL;
309 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
310 		/* reached the end of options chain */
311 		ndopts->nd_opts_done = 1;
312 		ndopts->nd_opts_search = NULL;
313 	}
314 	return nd_opt;
315 }
316 
317 /*
318  * Parse multiple ND options.
319  * This function is much easier to use, for ND routines that do not need
320  * multiple options of the same type.
321  */
322 int
323 nd6_options(ndopts)
324 	union nd_opts *ndopts;
325 {
326 	struct nd_opt_hdr *nd_opt;
327 	int i = 0;
328 
329 	if (!ndopts)
330 		panic("ndopts == NULL in nd6_options");
331 	if (!ndopts->nd_opts_last)
332 		panic("uninitialized ndopts in nd6_options");
333 	if (!ndopts->nd_opts_search)
334 		return 0;
335 
336 	while (1) {
337 		nd_opt = nd6_option(ndopts);
338 		if (!nd_opt && !ndopts->nd_opts_last) {
339 			/*
340 			 * Message validation requires that all included
341 			 * options have a length that is greater than zero.
342 			 */
343 			icmp6stat.icp6s_nd_badopt++;
344 			bzero(ndopts, sizeof(*ndopts));
345 			return -1;
346 		}
347 
348 		if (!nd_opt)
349 			goto skip1;
350 
351 		switch (nd_opt->nd_opt_type) {
352 		case ND_OPT_SOURCE_LINKADDR:
353 		case ND_OPT_TARGET_LINKADDR:
354 		case ND_OPT_MTU:
355 		case ND_OPT_REDIRECTED_HEADER:
356 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
357 				nd6log((LOG_INFO,
358 				    "duplicated ND6 option found (type=%d)\n",
359 				    nd_opt->nd_opt_type));
360 				/* XXX bark? */
361 			} else {
362 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
363 					= nd_opt;
364 			}
365 			break;
366 		case ND_OPT_PREFIX_INFORMATION:
367 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
368 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
369 					= nd_opt;
370 			}
371 			ndopts->nd_opts_pi_end =
372 				(struct nd_opt_prefix_info *)nd_opt;
373 			break;
374 		default:
375 			/*
376 			 * Unknown options must be silently ignored,
377 			 * to accomodate future extension to the protocol.
378 			 */
379 			nd6log((LOG_DEBUG,
380 			    "nd6_options: unsupported option %d - "
381 			    "option ignored\n", nd_opt->nd_opt_type));
382 		}
383 
384 skip1:
385 		i++;
386 		if (i > nd6_maxndopt) {
387 			icmp6stat.icp6s_nd_toomanyopt++;
388 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
389 			break;
390 		}
391 
392 		if (ndopts->nd_opts_done)
393 			break;
394 	}
395 
396 	return 0;
397 }
398 
399 /*
400  * ND6 timer routine to expire default route list and prefix list
401  */
402 void
403 nd6_timer(ignored_arg)
404 	void	*ignored_arg;
405 {
406 	int s;
407 	struct llinfo_nd6 *ln;
408 	struct nd_defrouter *dr;
409 	struct nd_prefix *pr;
410 	struct ifnet *ifp;
411 	struct in6_ifaddr *ia6, *nia6;
412 	struct in6_addrlifetime *lt6;
413 
414 	s = splnet();
415 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
416 	    nd6_timer, NULL);
417 
418 	ln = llinfo_nd6.ln_next;
419 	while (ln && ln != &llinfo_nd6) {
420 		struct rtentry *rt;
421 		struct sockaddr_in6 *dst;
422 		struct llinfo_nd6 *next = ln->ln_next;
423 		/* XXX: used for the DELAY case only: */
424 		struct nd_ifinfo *ndi = NULL;
425 
426 		if ((rt = ln->ln_rt) == NULL) {
427 			ln = next;
428 			continue;
429 		}
430 		if ((ifp = rt->rt_ifp) == NULL) {
431 			ln = next;
432 			continue;
433 		}
434 		ndi = ND_IFINFO(ifp);
435 		dst = (struct sockaddr_in6 *)rt_key(rt);
436 
437 		if (ln->ln_expire > time_second) {
438 			ln = next;
439 			continue;
440 		}
441 
442 		/* sanity check */
443 		if (!rt)
444 			panic("rt=0 in nd6_timer(ln=%p)", ln);
445 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
446 			panic("rt_llinfo(%p) is not equal to ln(%p)",
447 			      rt->rt_llinfo, ln);
448 		if (!dst)
449 			panic("dst=0 in nd6_timer(ln=%p)", ln);
450 
451 		switch (ln->ln_state) {
452 		case ND6_LLINFO_INCOMPLETE:
453 			if (ln->ln_asked < nd6_mmaxtries) {
454 				ln->ln_asked++;
455 				ln->ln_expire = time_second +
456 					ND_IFINFO(ifp)->retrans / 1000;
457 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
458 					ln, 0);
459 			} else {
460 				struct mbuf *m = ln->ln_hold;
461 				if (m) {
462 					if (rt->rt_ifp) {
463 						/*
464 						 * Fake rcvif to make ICMP error
465 						 * more helpful in diagnosing
466 						 * for the receiver.
467 						 * XXX: should we consider
468 						 * older rcvif?
469 						 */
470 						m->m_pkthdr.rcvif = rt->rt_ifp;
471 					}
472 					icmp6_error(m, ICMP6_DST_UNREACH,
473 						    ICMP6_DST_UNREACH_ADDR, 0);
474 					ln->ln_hold = NULL;
475 				}
476 				next = nd6_free(rt);
477 			}
478 			break;
479 		case ND6_LLINFO_REACHABLE:
480 			if (ln->ln_expire) {
481 				ln->ln_state = ND6_LLINFO_STALE;
482 				ln->ln_expire = time_second + nd6_gctimer;
483 			}
484 			break;
485 
486 		case ND6_LLINFO_STALE:
487 			/* Garbage Collection(RFC 2461 5.3) */
488 			if (ln->ln_expire)
489 				next = nd6_free(rt);
490 			break;
491 
492 		case ND6_LLINFO_DELAY:
493 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
494 				/* We need NUD */
495 				ln->ln_asked = 1;
496 				ln->ln_state = ND6_LLINFO_PROBE;
497 				ln->ln_expire = time_second +
498 					ndi->retrans / 1000;
499 				nd6_ns_output(ifp, &dst->sin6_addr,
500 					      &dst->sin6_addr,
501 					      ln, 0);
502 			} else {
503 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
504 				ln->ln_expire = time_second + nd6_gctimer;
505 			}
506 			break;
507 		case ND6_LLINFO_PROBE:
508 			if (ln->ln_asked < nd6_umaxtries) {
509 				ln->ln_asked++;
510 				ln->ln_expire = time_second +
511 					ND_IFINFO(ifp)->retrans / 1000;
512 				nd6_ns_output(ifp, &dst->sin6_addr,
513 					       &dst->sin6_addr, ln, 0);
514 			} else {
515 				next = nd6_free(rt);
516 			}
517 			break;
518 		}
519 		ln = next;
520 	}
521 
522 	/* expire default router list */
523 	dr = TAILQ_FIRST(&nd_defrouter);
524 	while (dr) {
525 		if (dr->expire && dr->expire < time_second) {
526 			struct nd_defrouter *t;
527 			t = TAILQ_NEXT(dr, dr_entry);
528 			defrtrlist_del(dr);
529 			dr = t;
530 		} else {
531 			dr = TAILQ_NEXT(dr, dr_entry);
532 		}
533 	}
534 
535 	/*
536 	 * expire interface addresses.
537 	 * in the past the loop was inside prefix expiry processing.
538 	 * However, from a stricter speci-confrmance standpoint, we should
539 	 * rather separate address lifetimes and prefix lifetimes.
540 	 */
541   addrloop:
542 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
543 		nia6 = ia6->ia_next;
544 		/* check address lifetime */
545 		lt6 = &ia6->ia6_lifetime;
546 		if (IFA6_IS_INVALID(ia6)) {
547 			int regen = 0;
548 
549 			/*
550 			 * If the expiring address is temporary, try
551 			 * regenerating a new one.  This would be useful when
552 			 * we suspended a laptop PC, then turned it on after a
553 			 * period that could invalidate all temporary
554 			 * addresses.  Although we may have to restart the
555 			 * loop (see below), it must be after purging the
556 			 * address.  Otherwise, we'd see an infinite loop of
557 			 * regeneration.
558 			 */
559 			if (ip6_use_tempaddr &&
560 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
561 				if (regen_tmpaddr(ia6) == 0)
562 					regen = 1;
563 			}
564 
565 			in6_purgeaddr(&ia6->ia_ifa);
566 
567 			if (regen)
568 				goto addrloop; /* XXX: see below */
569 		}
570 		if (IFA6_IS_DEPRECATED(ia6)) {
571 			int oldflags = ia6->ia6_flags;
572 
573 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
574 
575 			/*
576 			 * If a temporary address has just become deprecated,
577 			 * regenerate a new one if possible.
578 			 */
579 			if (ip6_use_tempaddr &&
580 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
581 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
582 
583 				if (regen_tmpaddr(ia6) == 0) {
584 					/*
585 					 * A new temporary address is
586 					 * generated.
587 					 * XXX: this means the address chain
588 					 * has changed while we are still in
589 					 * the loop.  Although the change
590 					 * would not cause disaster (because
591 					 * it's not a deletion, but an
592 					 * addition,) we'd rather restart the
593 					 * loop just for safety.  Or does this
594 					 * significantly reduce performance??
595 					 */
596 					goto addrloop;
597 				}
598 			}
599 		} else {
600 			/*
601 			 * A new RA might have made a deprecated address
602 			 * preferred.
603 			 */
604 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
605 		}
606 	}
607 
608 	/* expire prefix list */
609 	pr = nd_prefix.lh_first;
610 	while (pr) {
611 		/*
612 		 * check prefix lifetime.
613 		 * since pltime is just for autoconf, pltime processing for
614 		 * prefix is not necessary.
615 		 */
616 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
617 			struct nd_prefix *t;
618 			t = pr->ndpr_next;
619 
620 			/*
621 			 * address expiration and prefix expiration are
622 			 * separate.  NEVER perform in6_purgeaddr here.
623 			 */
624 
625 			prelist_remove(pr);
626 			pr = t;
627 		} else
628 			pr = pr->ndpr_next;
629 	}
630 	splx(s);
631 }
632 
633 static int
634 regen_tmpaddr(ia6)
635 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
636 {
637 	struct ifaddr *ifa;
638 	struct ifnet *ifp;
639 	struct in6_ifaddr *public_ifa6 = NULL;
640 
641 	ifp = ia6->ia_ifa.ifa_ifp;
642 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
643 	     ifa = ifa->ifa_list.tqe_next) {
644 		struct in6_ifaddr *it6;
645 
646 		if (ifa->ifa_addr->sa_family != AF_INET6)
647 			continue;
648 
649 		it6 = (struct in6_ifaddr *)ifa;
650 
651 		/* ignore no autoconf addresses. */
652 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
653 			continue;
654 
655 		/* ignore autoconf addresses with different prefixes. */
656 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
657 			continue;
658 
659 		/*
660 		 * Now we are looking at an autoconf address with the same
661 		 * prefix as ours.  If the address is temporary and is still
662 		 * preferred, do not create another one.  It would be rare, but
663 		 * could happen, for example, when we resume a laptop PC after
664 		 * a long period.
665 		 */
666 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
667 		    !IFA6_IS_DEPRECATED(it6)) {
668 			public_ifa6 = NULL;
669 			break;
670 		}
671 
672 		/*
673 		 * This is a public autoconf address that has the same prefix
674 		 * as ours.  If it is preferred, keep it.  We can't break the
675 		 * loop here, because there may be a still-preferred temporary
676 		 * address with the prefix.
677 		 */
678 		if (!IFA6_IS_DEPRECATED(it6))
679 		    public_ifa6 = it6;
680 	}
681 
682 	if (public_ifa6 != NULL) {
683 		int e;
684 
685 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
686 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
687 			    " tmp addr,errno=%d\n", e);
688 			return (-1);
689 		}
690 		return (0);
691 	}
692 
693 	return (-1);
694 }
695 
696 /*
697  * Nuke neighbor cache/prefix/default router management table, right before
698  * ifp goes away.
699  */
700 void
701 nd6_purge(ifp)
702 	struct ifnet *ifp;
703 {
704 	struct llinfo_nd6 *ln, *nln;
705 	struct nd_defrouter *dr, *ndr, drany;
706 	struct nd_prefix *pr, *npr;
707 
708 	/* Nuke default router list entries toward ifp */
709 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
710 		/*
711 		 * The first entry of the list may be stored in
712 		 * the routing table, so we'll delete it later.
713 		 */
714 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
715 			ndr = TAILQ_NEXT(dr, dr_entry);
716 			if (dr->ifp == ifp)
717 				defrtrlist_del(dr);
718 		}
719 		dr = TAILQ_FIRST(&nd_defrouter);
720 		if (dr->ifp == ifp)
721 			defrtrlist_del(dr);
722 	}
723 
724 	/* Nuke prefix list entries toward ifp */
725 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
726 		npr = pr->ndpr_next;
727 		if (pr->ndpr_ifp == ifp) {
728 			/*
729 			 * Previously, pr->ndpr_addr is removed as well,
730 			 * but I strongly believe we don't have to do it.
731 			 * nd6_purge() is only called from in6_ifdetach(),
732 			 * which removes all the associated interface addresses
733 			 * by itself.
734 			 * (jinmei@kame.net 20010129)
735 			 */
736 			prelist_remove(pr);
737 		}
738 	}
739 
740 	/* cancel default outgoing interface setting */
741 	if (nd6_defifindex == ifp->if_index)
742 		nd6_setdefaultiface(0);
743 
744 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
745 		/* refresh default router list */
746 		bzero(&drany, sizeof(drany));
747 		defrouter_delreq(&drany, 0);
748 		defrouter_select();
749 	}
750 
751 	/*
752 	 * Nuke neighbor cache entries for the ifp.
753 	 * Note that rt->rt_ifp may not be the same as ifp,
754 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
755 	 * nd6_rtrequest(), and ip6_input().
756 	 */
757 	ln = llinfo_nd6.ln_next;
758 	while (ln && ln != &llinfo_nd6) {
759 		struct rtentry *rt;
760 		struct sockaddr_dl *sdl;
761 
762 		nln = ln->ln_next;
763 		rt = ln->ln_rt;
764 		if (rt && rt->rt_gateway &&
765 		    rt->rt_gateway->sa_family == AF_LINK) {
766 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
767 			if (sdl->sdl_index == ifp->if_index)
768 				nln = nd6_free(rt);
769 		}
770 		ln = nln;
771 	}
772 }
773 
774 struct rtentry *
775 nd6_lookup(addr6, create, ifp)
776 	struct in6_addr *addr6;
777 	int create;
778 	struct ifnet *ifp;
779 {
780 	struct rtentry *rt;
781 	struct sockaddr_in6 sin6;
782 
783 	bzero(&sin6, sizeof(sin6));
784 	sin6.sin6_len = sizeof(struct sockaddr_in6);
785 	sin6.sin6_family = AF_INET6;
786 	sin6.sin6_addr = *addr6;
787 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
788 	if (rt) {
789 		if ((rt->rt_flags & RTF_LLINFO) == 0 && create) {
790 			/*
791 			 * This is the case for the default route.
792 			 * If we want to create a neighbor cache for the
793 			 * address, we should free the route for the
794 			 * destination and allocate an interface route.
795 			 */
796 			RTFREE_LOCKED(rt);
797 			rt = 0;
798 		}
799 	}
800 	if (!rt) {
801 		if (create && ifp) {
802 			int e;
803 
804 			/*
805 			 * If no route is available and create is set,
806 			 * we allocate a host route for the destination
807 			 * and treat it like an interface route.
808 			 * This hack is necessary for a neighbor which can't
809 			 * be covered by our own prefix.
810 			 */
811 			struct ifaddr *ifa =
812 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
813 			if (ifa == NULL)
814 				return (NULL);
815 
816 			/*
817 			 * Create a new route.  RTF_LLINFO is necessary
818 			 * to create a Neighbor Cache entry for the
819 			 * destination in nd6_rtrequest which will be
820 			 * called in rtrequest via ifa->ifa_rtrequest.
821 			 */
822 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
823 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
824 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
825 			    ~RTF_CLONING, &rt)) != 0) {
826 				log(LOG_ERR,
827 				    "nd6_lookup: failed to add route for a "
828 				    "neighbor(%s), errno=%d\n",
829 				    ip6_sprintf(addr6), e);
830 			}
831 			if (rt == NULL)
832 				return (NULL);
833 			RT_LOCK(rt);
834 			if (rt->rt_llinfo) {
835 				struct llinfo_nd6 *ln =
836 				    (struct llinfo_nd6 *)rt->rt_llinfo;
837 				ln->ln_state = ND6_LLINFO_NOSTATE;
838 			}
839 		} else
840 			return (NULL);
841 	}
842 	RT_LOCK_ASSERT(rt);
843 	RT_REMREF(rt);
844 	/*
845 	 * Validation for the entry.
846 	 * Note that the check for rt_llinfo is necessary because a cloned
847 	 * route from a parent route that has the L flag (e.g. the default
848 	 * route to a p2p interface) may have the flag, too, while the
849 	 * destination is not actually a neighbor.
850 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
851 	 *      it might be the loopback interface if the entry is for our
852 	 *      own address on a non-loopback interface. Instead, we should
853 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
854 	 *	interface.
855 	 */
856 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
857 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
858 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
859 		if (create) {
860 			log(LOG_DEBUG,
861 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
862 			    ip6_sprintf(addr6),
863 			    ifp ? if_name(ifp) : "unspec");
864 			/* xxx more logs... kazu */
865 		}
866 		RT_UNLOCK(rt);
867 		return (NULL);
868 	}
869 	RT_UNLOCK(rt);		/* XXX not ready to return rt locked */
870 	return (rt);
871 }
872 
873 /*
874  * Test whether a given IPv6 address is a neighbor or not, ignoring
875  * the actual neighbor cache.  The neighbor cache is ignored in order
876  * to not reenter the routing code from within itself.
877  */
878 static int
879 nd6_is_new_addr_neighbor(addr, ifp)
880 	struct sockaddr_in6 *addr;
881 	struct ifnet *ifp;
882 {
883 	struct nd_prefix *pr;
884 
885 	/*
886 	 * A link-local address is always a neighbor.
887 	 * XXX: we should use the sin6_scope_id field rather than the embedded
888 	 * interface index.
889 	 * XXX: a link does not necessarily specify a single interface.
890 	 */
891 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
892 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
893 		return (1);
894 
895 	/*
896 	 * If the address matches one of our addresses,
897 	 * it should be a neighbor.
898 	 * If the address matches one of our on-link prefixes, it should be a
899 	 * neighbor.
900 	 */
901 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
902 		if (pr->ndpr_ifp != ifp)
903 			continue;
904 
905 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
906 			continue;
907 
908 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
909 		    &addr->sin6_addr, &pr->ndpr_mask))
910 			return (1);
911 	}
912 
913 	/*
914 	 * If the default router list is empty, all addresses are regarded
915 	 * as on-link, and thus, as a neighbor.
916 	 * XXX: we restrict the condition to hosts, because routers usually do
917 	 * not have the "default router list".
918 	 */
919 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
920 	    nd6_defifindex == ifp->if_index) {
921 		return (1);
922 	}
923 
924 	return (0);
925 }
926 
927 
928 /*
929  * Detect if a given IPv6 address identifies a neighbor on a given link.
930  * XXX: should take care of the destination of a p2p link?
931  */
932 int
933 nd6_is_addr_neighbor(addr, ifp)
934 	struct sockaddr_in6 *addr;
935 	struct ifnet *ifp;
936 {
937 
938 	if (nd6_is_new_addr_neighbor(addr, ifp))
939 		return (1);
940 
941 	/*
942 	 * Even if the address matches none of our addresses, it might be
943 	 * in the neighbor cache.
944 	 */
945 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
946 		return (1);
947 
948 	return (0);
949 }
950 
951 /*
952  * Free an nd6 llinfo entry.
953  */
954 struct llinfo_nd6 *
955 nd6_free(rt)
956 	struct rtentry *rt;
957 {
958 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
959 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
960 	struct nd_defrouter *dr;
961 
962 	/*
963 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
964 	 * even though it is not harmful, it was not really necessary.
965 	 */
966 
967 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
968 		int s;
969 		s = splnet();
970 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
971 		    rt->rt_ifp);
972 
973 		if (ln->ln_router || dr) {
974 			/*
975 			 * rt6_flush must be called whether or not the neighbor
976 			 * is in the Default Router List.
977 			 * See a corresponding comment in nd6_na_input().
978 			 */
979 			rt6_flush(&in6, rt->rt_ifp);
980 		}
981 
982 		if (dr) {
983 			/*
984 			 * Unreachablity of a router might affect the default
985 			 * router selection and on-link detection of advertised
986 			 * prefixes.
987 			 */
988 
989 			/*
990 			 * Temporarily fake the state to choose a new default
991 			 * router and to perform on-link determination of
992 			 * prefixes correctly.
993 			 * Below the state will be set correctly,
994 			 * or the entry itself will be deleted.
995 			 */
996 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
997 
998 			/*
999 			 * Since defrouter_select() does not affect the
1000 			 * on-link determination and MIP6 needs the check
1001 			 * before the default router selection, we perform
1002 			 * the check now.
1003 			 */
1004 			pfxlist_onlink_check();
1005 
1006 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
1007 				/*
1008 				 * It is used as the current default router,
1009 				 * so we have to move it to the end of the
1010 				 * list and choose a new one.
1011 				 * XXX: it is not very efficient if this is
1012 				 *      the only router.
1013 				 */
1014 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1015 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1016 
1017 				defrouter_select();
1018 			}
1019 		}
1020 		splx(s);
1021 	}
1022 
1023 	/*
1024 	 * Before deleting the entry, remember the next entry as the
1025 	 * return value.  We need this because pfxlist_onlink_check() above
1026 	 * might have freed other entries (particularly the old next entry) as
1027 	 * a side effect (XXX).
1028 	 */
1029 	next = ln->ln_next;
1030 
1031 	/*
1032 	 * Detach the route from the routing tree and the list of neighbor
1033 	 * caches, and disable the route entry not to be used in already
1034 	 * cached routes.
1035 	 */
1036 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1037 	    rt_mask(rt), 0, (struct rtentry **)0);
1038 
1039 	return (next);
1040 }
1041 
1042 /*
1043  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1044  *
1045  * XXX cost-effective metods?
1046  */
1047 void
1048 nd6_nud_hint(rt, dst6, force)
1049 	struct rtentry *rt;
1050 	struct in6_addr *dst6;
1051 	int force;
1052 {
1053 	struct llinfo_nd6 *ln;
1054 
1055 	/*
1056 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1057 	 * routing table by supplied "dst6".
1058 	 */
1059 	if (!rt) {
1060 		if (!dst6)
1061 			return;
1062 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1063 			return;
1064 	}
1065 
1066 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1067 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1068 	    !rt->rt_llinfo || !rt->rt_gateway ||
1069 	    rt->rt_gateway->sa_family != AF_LINK) {
1070 		/* This is not a host route. */
1071 		return;
1072 	}
1073 
1074 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1075 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1076 		return;
1077 
1078 	/*
1079 	 * if we get upper-layer reachability confirmation many times,
1080 	 * it is possible we have false information.
1081 	 */
1082 	if (!force) {
1083 		ln->ln_byhint++;
1084 		if (ln->ln_byhint > nd6_maxnudhint)
1085 			return;
1086 	}
1087 
1088 	ln->ln_state = ND6_LLINFO_REACHABLE;
1089 	if (ln->ln_expire)
1090 		ln->ln_expire = time_second +
1091 			ND_IFINFO(rt->rt_ifp)->reachable;
1092 }
1093 
1094 void
1095 nd6_rtrequest(req, rt, info)
1096 	int	req;
1097 	struct rtentry *rt;
1098 	struct rt_addrinfo *info; /* xxx unused */
1099 {
1100 	struct sockaddr *gate = rt->rt_gateway;
1101 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1102 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1103 	struct ifnet *ifp = rt->rt_ifp;
1104 	struct ifaddr *ifa;
1105 
1106 	RT_LOCK_ASSERT(rt);
1107 
1108 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1109 		return;
1110 
1111 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1112 		/*
1113 		 * This is probably an interface direct route for a link
1114 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1115 		 * We do not need special treatment below for such a route.
1116 		 * Moreover, the RTF_LLINFO flag which would be set below
1117 		 * would annoy the ndp(8) command.
1118 		 */
1119 		return;
1120 	}
1121 
1122 	if (req == RTM_RESOLVE &&
1123 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1124 	     !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt),
1125 	     ifp))) {
1126 		/*
1127 		 * FreeBSD and BSD/OS often make a cloned host route based
1128 		 * on a less-specific route (e.g. the default route).
1129 		 * If the less specific route does not have a "gateway"
1130 		 * (this is the case when the route just goes to a p2p or an
1131 		 * stf interface), we'll mistakenly make a neighbor cache for
1132 		 * the host route, and will see strange neighbor solicitation
1133 		 * for the corresponding destination.  In order to avoid the
1134 		 * confusion, we check if the destination of the route is
1135 		 * a neighbor in terms of neighbor discovery, and stop the
1136 		 * process if not.  Additionally, we remove the LLINFO flag
1137 		 * so that ndp(8) will not try to get the neighbor information
1138 		 * of the destination.
1139 		 */
1140 		rt->rt_flags &= ~RTF_LLINFO;
1141 		return;
1142 	}
1143 
1144 	switch (req) {
1145 	case RTM_ADD:
1146 		/*
1147 		 * There is no backward compatibility :)
1148 		 *
1149 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1150 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1151 		 *	   rt->rt_flags |= RTF_CLONING;
1152 		 */
1153 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1154 			/*
1155 			 * Case 1: This route should come from
1156 			 * a route to interface.  RTF_LLINFO flag is set
1157 			 * for a host route whose destination should be
1158 			 * treated as on-link.
1159 			 */
1160 			rt_setgate(rt, rt_key(rt),
1161 				   (struct sockaddr *)&null_sdl);
1162 			gate = rt->rt_gateway;
1163 			SDL(gate)->sdl_type = ifp->if_type;
1164 			SDL(gate)->sdl_index = ifp->if_index;
1165 			if (ln)
1166 				ln->ln_expire = time_second;
1167 			if (ln && ln->ln_expire == 0) {
1168 				/* kludge for desktops */
1169 				ln->ln_expire = 1;
1170 			}
1171 			if ((rt->rt_flags & RTF_CLONING) != 0)
1172 				break;
1173 		}
1174 		/*
1175 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1176 		 * We don't do that here since llinfo is not ready yet.
1177 		 *
1178 		 * There are also couple of other things to be discussed:
1179 		 * - unsolicited NA code needs improvement beforehand
1180 		 * - RFC2461 says we MAY send multicast unsolicited NA
1181 		 *   (7.2.6 paragraph 4), however, it also says that we
1182 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1183 		 *   we don't have anything like it right now.
1184 		 *   note that the mechanism needs a mutual agreement
1185 		 *   between proxies, which means that we need to implement
1186 		 *   a new protocol, or a new kludge.
1187 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1188 		 *   we need to check ip6forwarding before sending it.
1189 		 *   (or should we allow proxy ND configuration only for
1190 		 *   routers?  there's no mention about proxy ND from hosts)
1191 		 */
1192 #if 0
1193 		/* XXX it does not work */
1194 		if (rt->rt_flags & RTF_ANNOUNCE)
1195 			nd6_na_output(ifp,
1196 			      &SIN6(rt_key(rt))->sin6_addr,
1197 			      &SIN6(rt_key(rt))->sin6_addr,
1198 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1199 			      1, NULL);
1200 #endif
1201 		/* FALLTHROUGH */
1202 	case RTM_RESOLVE:
1203 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1204 			/*
1205 			 * Address resolution isn't necessary for a point to
1206 			 * point link, so we can skip this test for a p2p link.
1207 			 */
1208 			if (gate->sa_family != AF_LINK ||
1209 			    gate->sa_len < sizeof(null_sdl)) {
1210 				log(LOG_DEBUG,
1211 				    "nd6_rtrequest: bad gateway value: %s\n",
1212 				    if_name(ifp));
1213 				break;
1214 			}
1215 			SDL(gate)->sdl_type = ifp->if_type;
1216 			SDL(gate)->sdl_index = ifp->if_index;
1217 		}
1218 		if (ln != NULL)
1219 			break;	/* This happens on a route change */
1220 		/*
1221 		 * Case 2: This route may come from cloning, or a manual route
1222 		 * add with a LL address.
1223 		 */
1224 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1225 		rt->rt_llinfo = (caddr_t)ln;
1226 		if (!ln) {
1227 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1228 			break;
1229 		}
1230 		nd6_inuse++;
1231 		nd6_allocated++;
1232 		bzero(ln, sizeof(*ln));
1233 		ln->ln_rt = rt;
1234 		/* this is required for "ndp" command. - shin */
1235 		if (req == RTM_ADD) {
1236 		        /*
1237 			 * gate should have some valid AF_LINK entry,
1238 			 * and ln->ln_expire should have some lifetime
1239 			 * which is specified by ndp command.
1240 			 */
1241 			ln->ln_state = ND6_LLINFO_REACHABLE;
1242 			ln->ln_byhint = 0;
1243 		} else {
1244 		        /*
1245 			 * When req == RTM_RESOLVE, rt is created and
1246 			 * initialized in rtrequest(), so rt_expire is 0.
1247 			 */
1248 			ln->ln_state = ND6_LLINFO_NOSTATE;
1249 			ln->ln_expire = time_second;
1250 		}
1251 		rt->rt_flags |= RTF_LLINFO;
1252 		ln->ln_next = llinfo_nd6.ln_next;
1253 		llinfo_nd6.ln_next = ln;
1254 		ln->ln_prev = &llinfo_nd6;
1255 		ln->ln_next->ln_prev = ln;
1256 
1257 		/*
1258 		 * check if rt_key(rt) is one of my address assigned
1259 		 * to the interface.
1260 		 */
1261 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1262 		    &SIN6(rt_key(rt))->sin6_addr);
1263 		if (ifa) {
1264 			caddr_t macp = nd6_ifptomac(ifp);
1265 			ln->ln_expire = 0;
1266 			ln->ln_state = ND6_LLINFO_REACHABLE;
1267 			ln->ln_byhint = 0;
1268 			if (macp) {
1269 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1270 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1271 			}
1272 			if (nd6_useloopback) {
1273 				rt->rt_ifp = &loif[0];	/* XXX */
1274 				/*
1275 				 * Make sure rt_ifa be equal to the ifaddr
1276 				 * corresponding to the address.
1277 				 * We need this because when we refer
1278 				 * rt_ifa->ia6_flags in ip6_input, we assume
1279 				 * that the rt_ifa points to the address instead
1280 				 * of the loopback address.
1281 				 */
1282 				if (ifa != rt->rt_ifa) {
1283 					IFAFREE(rt->rt_ifa);
1284 					IFAREF(ifa);
1285 					rt->rt_ifa = ifa;
1286 				}
1287 			}
1288 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1289 			ln->ln_expire = 0;
1290 			ln->ln_state = ND6_LLINFO_REACHABLE;
1291 			ln->ln_byhint = 0;
1292 
1293 			/* join solicited node multicast for proxy ND */
1294 			if (ifp->if_flags & IFF_MULTICAST) {
1295 				struct in6_addr llsol;
1296 				int error;
1297 
1298 				llsol = SIN6(rt_key(rt))->sin6_addr;
1299 				llsol.s6_addr16[0] = htons(0xff02);
1300 				llsol.s6_addr16[1] = htons(ifp->if_index);
1301 				llsol.s6_addr32[1] = 0;
1302 				llsol.s6_addr32[2] = htonl(1);
1303 				llsol.s6_addr8[12] = 0xff;
1304 
1305 				if (!in6_addmulti(&llsol, ifp, &error)) {
1306 					nd6log((LOG_ERR, "%s: failed to join "
1307 					    "%s (errno=%d)\n", if_name(ifp),
1308 					    ip6_sprintf(&llsol), error));
1309 				}
1310 			}
1311 		}
1312 		break;
1313 
1314 	case RTM_DELETE:
1315 		if (!ln)
1316 			break;
1317 		/* leave from solicited node multicast for proxy ND */
1318 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1319 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1320 			struct in6_addr llsol;
1321 			struct in6_multi *in6m;
1322 
1323 			llsol = SIN6(rt_key(rt))->sin6_addr;
1324 			llsol.s6_addr16[0] = htons(0xff02);
1325 			llsol.s6_addr16[1] = htons(ifp->if_index);
1326 			llsol.s6_addr32[1] = 0;
1327 			llsol.s6_addr32[2] = htonl(1);
1328 			llsol.s6_addr8[12] = 0xff;
1329 
1330 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1331 			if (in6m)
1332 				in6_delmulti(in6m);
1333 		}
1334 		nd6_inuse--;
1335 		ln->ln_next->ln_prev = ln->ln_prev;
1336 		ln->ln_prev->ln_next = ln->ln_next;
1337 		ln->ln_prev = NULL;
1338 		rt->rt_llinfo = 0;
1339 		rt->rt_flags &= ~RTF_LLINFO;
1340 		if (ln->ln_hold)
1341 			m_freem(ln->ln_hold);
1342 		Free((caddr_t)ln);
1343 	}
1344 }
1345 
1346 int
1347 nd6_ioctl(cmd, data, ifp)
1348 	u_long cmd;
1349 	caddr_t	data;
1350 	struct ifnet *ifp;
1351 {
1352 	struct in6_drlist *drl = (struct in6_drlist *)data;
1353 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1354 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1355 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1356 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1357 	struct nd_defrouter *dr, any;
1358 	struct nd_prefix *pr;
1359 	struct rtentry *rt;
1360 	int i = 0, error = 0;
1361 	int s;
1362 
1363 	switch (cmd) {
1364 	case SIOCGDRLST_IN6:
1365 		/*
1366 		 * obsolete API, use sysctl under net.inet6.icmp6
1367 		 */
1368 		bzero(drl, sizeof(*drl));
1369 		s = splnet();
1370 		dr = TAILQ_FIRST(&nd_defrouter);
1371 		while (dr && i < DRLSTSIZ) {
1372 			drl->defrouter[i].rtaddr = dr->rtaddr;
1373 			in6_clearscope(&drl->defrouter[i].rtaddr);
1374 
1375 			drl->defrouter[i].flags = dr->flags;
1376 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1377 			drl->defrouter[i].expire = dr->expire;
1378 			drl->defrouter[i].if_index = dr->ifp->if_index;
1379 			i++;
1380 			dr = TAILQ_NEXT(dr, dr_entry);
1381 		}
1382 		splx(s);
1383 		break;
1384 	case SIOCGPRLST_IN6:
1385 		/*
1386 		 * obsolete API, use sysctl under net.inet6.icmp6
1387 		 *
1388 		 * XXX the structure in6_prlist was changed in backward-
1389 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1390 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1391 		 */
1392 		/*
1393 		 * XXX meaning of fields, especialy "raflags", is very
1394 		 * differnet between RA prefix list and RR/static prefix list.
1395 		 * how about separating ioctls into two?
1396 		 */
1397 		bzero(oprl, sizeof(*oprl));
1398 		s = splnet();
1399 		pr = nd_prefix.lh_first;
1400 		while (pr && i < PRLSTSIZ) {
1401 			struct nd_pfxrouter *pfr;
1402 			int j;
1403 
1404 			(void)in6_embedscope(&oprl->prefix[i].prefix,
1405 			    &pr->ndpr_prefix, NULL, NULL);
1406 			oprl->prefix[i].raflags = pr->ndpr_raf;
1407 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1408 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1409 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1410 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1411 			oprl->prefix[i].expire = pr->ndpr_expire;
1412 
1413 			pfr = pr->ndpr_advrtrs.lh_first;
1414 			j = 0;
1415 			while (pfr) {
1416 				if (j < DRLSTSIZ) {
1417 #define RTRADDR oprl->prefix[i].advrtr[j]
1418 					RTRADDR = pfr->router->rtaddr;
1419 					in6_clearscope(&RTRADDR);
1420 #undef RTRADDR
1421 				}
1422 				j++;
1423 				pfr = pfr->pfr_next;
1424 			}
1425 			oprl->prefix[i].advrtrs = j;
1426 			oprl->prefix[i].origin = PR_ORIG_RA;
1427 
1428 			i++;
1429 			pr = pr->ndpr_next;
1430 		}
1431 		splx(s);
1432 
1433 		break;
1434 	case OSIOCGIFINFO_IN6:
1435 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1436 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1437 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1438 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1439 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1440 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1441 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1442 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1443 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1444 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1445 		break;
1446 	case SIOCGIFINFO_IN6:
1447 		ndi->ndi = *ND_IFINFO(ifp);
1448 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1449 		break;
1450 	case SIOCSIFINFO_FLAGS:
1451 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1452 		break;
1453 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1454 		/* flush default router list */
1455 		/*
1456 		 * xxx sumikawa: should not delete route if default
1457 		 * route equals to the top of default router list
1458 		 */
1459 		bzero(&any, sizeof(any));
1460 		defrouter_delreq(&any, 0);
1461 		defrouter_select();
1462 		/* xxx sumikawa: flush prefix list */
1463 		break;
1464 	case SIOCSPFXFLUSH_IN6:
1465 	{
1466 		/* flush all the prefix advertised by routers */
1467 		struct nd_prefix *pr, *next;
1468 
1469 		s = splnet();
1470 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1471 			struct in6_ifaddr *ia, *ia_next;
1472 
1473 			next = pr->ndpr_next;
1474 
1475 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1476 				continue; /* XXX */
1477 
1478 			/* do we really have to remove addresses as well? */
1479 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1480 				/* ia might be removed.  keep the next ptr. */
1481 				ia_next = ia->ia_next;
1482 
1483 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1484 					continue;
1485 
1486 				if (ia->ia6_ndpr == pr)
1487 					in6_purgeaddr(&ia->ia_ifa);
1488 			}
1489 			prelist_remove(pr);
1490 		}
1491 		splx(s);
1492 		break;
1493 	}
1494 	case SIOCSRTRFLUSH_IN6:
1495 	{
1496 		/* flush all the default routers */
1497 		struct nd_defrouter *dr, *next;
1498 
1499 		s = splnet();
1500 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1501 			/*
1502 			 * The first entry of the list may be stored in
1503 			 * the routing table, so we'll delete it later.
1504 			 */
1505 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1506 				next = TAILQ_NEXT(dr, dr_entry);
1507 				defrtrlist_del(dr);
1508 			}
1509 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1510 		}
1511 		splx(s);
1512 		break;
1513 	}
1514 	case SIOCGNBRINFO_IN6:
1515 	{
1516 		struct llinfo_nd6 *ln;
1517 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1518 
1519 		/*
1520 		 * XXX: KAME specific hack for scoped addresses
1521 		 *      XXXX: for other scopes than link-local?
1522 		 */
1523 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1524 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1525 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1526 
1527 			if (*idp == 0)
1528 				*idp = htons(ifp->if_index);
1529 		}
1530 
1531 		s = splnet();
1532 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1533 			error = EINVAL;
1534 			splx(s);
1535 			break;
1536 		}
1537 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1538 		nbi->state = ln->ln_state;
1539 		nbi->asked = ln->ln_asked;
1540 		nbi->isrouter = ln->ln_router;
1541 		nbi->expire = ln->ln_expire;
1542 		splx(s);
1543 
1544 		break;
1545 	}
1546 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1547 		ndif->ifindex = nd6_defifindex;
1548 		break;
1549 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1550 		return (nd6_setdefaultiface(ndif->ifindex));
1551 	}
1552 	return (error);
1553 }
1554 
1555 /*
1556  * Create neighbor cache entry and cache link-layer address,
1557  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1558  */
1559 struct rtentry *
1560 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1561 	struct ifnet *ifp;
1562 	struct in6_addr *from;
1563 	char *lladdr;
1564 	int lladdrlen;
1565 	int type;	/* ICMP6 type */
1566 	int code;	/* type dependent information */
1567 {
1568 	struct rtentry *rt = NULL;
1569 	struct llinfo_nd6 *ln = NULL;
1570 	int is_newentry;
1571 	struct sockaddr_dl *sdl = NULL;
1572 	int do_update;
1573 	int olladdr;
1574 	int llchange;
1575 	int newstate = 0;
1576 
1577 	if (!ifp)
1578 		panic("ifp == NULL in nd6_cache_lladdr");
1579 	if (!from)
1580 		panic("from == NULL in nd6_cache_lladdr");
1581 
1582 	/* nothing must be updated for unspecified address */
1583 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1584 		return NULL;
1585 
1586 	/*
1587 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1588 	 * the caller.
1589 	 *
1590 	 * XXX If the link does not have link-layer adderss, what should
1591 	 * we do? (ifp->if_addrlen == 0)
1592 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1593 	 * description on it in NS section (RFC 2461 7.2.3).
1594 	 */
1595 
1596 	rt = nd6_lookup(from, 0, ifp);
1597 	if (!rt) {
1598 #if 0
1599 		/* nothing must be done if there's no lladdr */
1600 		if (!lladdr || !lladdrlen)
1601 			return NULL;
1602 #endif
1603 
1604 		rt = nd6_lookup(from, 1, ifp);
1605 		is_newentry = 1;
1606 	} else {
1607 		/* do nothing if static ndp is set */
1608 		if (rt->rt_flags & RTF_STATIC)
1609 			return NULL;
1610 		is_newentry = 0;
1611 	}
1612 
1613 	if (!rt)
1614 		return NULL;
1615 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1616 fail:
1617 		(void)nd6_free(rt);
1618 		return NULL;
1619 	}
1620 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1621 	if (!ln)
1622 		goto fail;
1623 	if (!rt->rt_gateway)
1624 		goto fail;
1625 	if (rt->rt_gateway->sa_family != AF_LINK)
1626 		goto fail;
1627 	sdl = SDL(rt->rt_gateway);
1628 
1629 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1630 	if (olladdr && lladdr) {
1631 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1632 			llchange = 1;
1633 		else
1634 			llchange = 0;
1635 	} else
1636 		llchange = 0;
1637 
1638 	/*
1639 	 * newentry olladdr  lladdr  llchange	(*=record)
1640 	 *	0	n	n	--	(1)
1641 	 *	0	y	n	--	(2)
1642 	 *	0	n	y	--	(3) * STALE
1643 	 *	0	y	y	n	(4) *
1644 	 *	0	y	y	y	(5) * STALE
1645 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1646 	 *	1	--	y	--	(7) * STALE
1647 	 */
1648 
1649 	if (lladdr) {		/* (3-5) and (7) */
1650 		/*
1651 		 * Record source link-layer address
1652 		 * XXX is it dependent to ifp->if_type?
1653 		 */
1654 		sdl->sdl_alen = ifp->if_addrlen;
1655 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1656 	}
1657 
1658 	if (!is_newentry) {
1659 		if ((!olladdr && lladdr) ||		/* (3) */
1660 		    (olladdr && lladdr && llchange)) {	/* (5) */
1661 			do_update = 1;
1662 			newstate = ND6_LLINFO_STALE;
1663 		} else					/* (1-2,4) */
1664 			do_update = 0;
1665 	} else {
1666 		do_update = 1;
1667 		if (!lladdr)				/* (6) */
1668 			newstate = ND6_LLINFO_NOSTATE;
1669 		else					/* (7) */
1670 			newstate = ND6_LLINFO_STALE;
1671 	}
1672 
1673 	if (do_update) {
1674 		/*
1675 		 * Update the state of the neighbor cache.
1676 		 */
1677 		ln->ln_state = newstate;
1678 
1679 		if (ln->ln_state == ND6_LLINFO_STALE) {
1680 			/*
1681 			 * XXX: since nd6_output() below will cause
1682 			 * state tansition to DELAY and reset the timer,
1683 			 * we must set the timer now, although it is actually
1684 			 * meaningless.
1685 			 */
1686 			ln->ln_expire = time_second + nd6_gctimer;
1687 
1688 			if (ln->ln_hold) {
1689 				/*
1690 				 * we assume ifp is not a p2p here, so just
1691 				 * set the 2nd argument as the 1st one.
1692 				 */
1693 				nd6_output(ifp, ifp, ln->ln_hold,
1694 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1695 				ln->ln_hold = NULL;
1696 			}
1697 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1698 			/* probe right away */
1699 			ln->ln_expire = time_second;
1700 		}
1701 	}
1702 
1703 	/*
1704 	 * ICMP6 type dependent behavior.
1705 	 *
1706 	 * NS: clear IsRouter if new entry
1707 	 * RS: clear IsRouter
1708 	 * RA: set IsRouter if there's lladdr
1709 	 * redir: clear IsRouter if new entry
1710 	 *
1711 	 * RA case, (1):
1712 	 * The spec says that we must set IsRouter in the following cases:
1713 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1714 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1715 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1716 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1717 	 * neighbor cache, this is similar to (6).
1718 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1719 	 *
1720 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1721 	 *							D R
1722 	 *	0	n	n	--	(1)	c   ?     s
1723 	 *	0	y	n	--	(2)	c   s     s
1724 	 *	0	n	y	--	(3)	c   s     s
1725 	 *	0	y	y	n	(4)	c   s     s
1726 	 *	0	y	y	y	(5)	c   s     s
1727 	 *	1	--	n	--	(6) c	c 	c s
1728 	 *	1	--	y	--	(7) c	c   s	c s
1729 	 *
1730 	 *					(c=clear s=set)
1731 	 */
1732 	switch (type & 0xff) {
1733 	case ND_NEIGHBOR_SOLICIT:
1734 		/*
1735 		 * New entry must have is_router flag cleared.
1736 		 */
1737 		if (is_newentry)	/* (6-7) */
1738 			ln->ln_router = 0;
1739 		break;
1740 	case ND_REDIRECT:
1741 		/*
1742 		 * If the icmp is a redirect to a better router, always set the
1743 		 * is_router flag.  Otherwise, if the entry is newly created,
1744 		 * clear the flag.  [RFC 2461, sec 8.3]
1745 		 */
1746 		if (code == ND_REDIRECT_ROUTER)
1747 			ln->ln_router = 1;
1748 		else if (is_newentry) /* (6-7) */
1749 			ln->ln_router = 0;
1750 		break;
1751 	case ND_ROUTER_SOLICIT:
1752 		/*
1753 		 * is_router flag must always be cleared.
1754 		 */
1755 		ln->ln_router = 0;
1756 		break;
1757 	case ND_ROUTER_ADVERT:
1758 		/*
1759 		 * Mark an entry with lladdr as a router.
1760 		 */
1761 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1762 		    (is_newentry && lladdr)) {			/* (7) */
1763 			ln->ln_router = 1;
1764 		}
1765 		break;
1766 	}
1767 
1768 	/*
1769 	 * When the link-layer address of a router changes, select the
1770 	 * best router again.  In particular, when the neighbor entry is newly
1771 	 * created, it might affect the selection policy.
1772 	 * Question: can we restrict the first condition to the "is_newentry"
1773 	 * case?
1774 	 * XXX: when we hear an RA from a new router with the link-layer
1775 	 * address option, defrouter_select() is called twice, since
1776 	 * defrtrlist_update called the function as well.  However, I believe
1777 	 * we can compromise the overhead, since it only happens the first
1778 	 * time.
1779 	 * XXX: although defrouter_select() should not have a bad effect
1780 	 * for those are not autoconfigured hosts, we explicitly avoid such
1781 	 * cases for safety.
1782 	 */
1783 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1784 		defrouter_select();
1785 
1786 	return rt;
1787 }
1788 
1789 static void
1790 nd6_slowtimo(ignored_arg)
1791     void *ignored_arg;
1792 {
1793 	int s = splnet();
1794 	struct nd_ifinfo *nd6if;
1795 	struct ifnet *ifp;
1796 
1797 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1798 	    nd6_slowtimo, NULL);
1799 	IFNET_RLOCK();
1800 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1801 		nd6if = ND_IFINFO(ifp);
1802 		if (nd6if->basereachable && /* already initialized */
1803 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1804 			/*
1805 			 * Since reachable time rarely changes by router
1806 			 * advertisements, we SHOULD insure that a new random
1807 			 * value gets recomputed at least once every few hours.
1808 			 * (RFC 2461, 6.3.4)
1809 			 */
1810 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1811 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1812 		}
1813 	}
1814 	IFNET_RUNLOCK();
1815 	splx(s);
1816 }
1817 
1818 #define senderr(e) { error = (e); goto bad;}
1819 int
1820 nd6_output(ifp, origifp, m0, dst, rt0)
1821 	struct ifnet *ifp;
1822 	struct ifnet *origifp;
1823 	struct mbuf *m0;
1824 	struct sockaddr_in6 *dst;
1825 	struct rtentry *rt0;
1826 {
1827 	struct mbuf *m = m0;
1828 	struct rtentry *rt = rt0;
1829 	struct sockaddr_in6 *gw6 = NULL;
1830 	struct llinfo_nd6 *ln = NULL;
1831 	int error = 0;
1832 
1833 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1834 		goto sendpkt;
1835 
1836 	if (nd6_need_cache(ifp) == 0)
1837 		goto sendpkt;
1838 
1839 	/*
1840 	 * next hop determination.  This routine is derived from ether_outpout.
1841 	 */
1842 again:
1843 	if (rt) {
1844 		if ((rt->rt_flags & RTF_UP) == 0) {
1845 			rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1846 			if (rt != NULL) {
1847 				RT_REMREF(rt);
1848 				RT_UNLOCK(rt);
1849 				if (rt->rt_ifp != ifp)
1850 					/*
1851 					 * XXX maybe we should update ifp too,
1852 					 * but the original code didn't and I
1853 					 * don't know what is correct here.
1854 					 */
1855 					goto again;
1856 			} else
1857 				senderr(EHOSTUNREACH);
1858 		}
1859 
1860 		if (rt->rt_flags & RTF_GATEWAY) {
1861 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1862 
1863 			/*
1864 			 * We skip link-layer address resolution and NUD
1865 			 * if the gateway is not a neighbor from ND point
1866 			 * of view, regardless of the value of nd_ifinfo.flags.
1867 			 * The second condition is a bit tricky; we skip
1868 			 * if the gateway is our own address, which is
1869 			 * sometimes used to install a route to a p2p link.
1870 			 */
1871 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1872 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1873 				/*
1874 				 * We allow this kind of tricky route only
1875 				 * when the outgoing interface is p2p.
1876 				 * XXX: we may need a more generic rule here.
1877 				 */
1878 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1879 					senderr(EHOSTUNREACH);
1880 
1881 				goto sendpkt;
1882 			}
1883 
1884 			if (rt->rt_gwroute == 0)
1885 				goto lookup;
1886 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1887 				RT_LOCK(rt);
1888 				rtfree(rt); rt = rt0;
1889 			lookup:
1890 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1891 				if ((rt = rt->rt_gwroute) == 0)
1892 					senderr(EHOSTUNREACH);
1893 				RT_UNLOCK(rt);
1894 			}
1895 		}
1896 	}
1897 
1898 	/*
1899 	 * Address resolution or Neighbor Unreachability Detection
1900 	 * for the next hop.
1901 	 * At this point, the destination of the packet must be a unicast
1902 	 * or an anycast address(i.e. not a multicast).
1903 	 */
1904 
1905 	/* Look up the neighbor cache for the nexthop */
1906 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1907 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1908 	else {
1909 		/*
1910 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1911 		 * the condition below is not very efficient.  But we believe
1912 		 * it is tolerable, because this should be a rare case.
1913 		 */
1914 		if (nd6_is_addr_neighbor(dst, ifp) &&
1915 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1916 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1917 	}
1918 	if (!ln || !rt) {
1919 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1920 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1921 			log(LOG_DEBUG,
1922 			    "nd6_output: can't allocate llinfo for %s "
1923 			    "(ln=%p, rt=%p)\n",
1924 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1925 			senderr(EIO);	/* XXX: good error? */
1926 		}
1927 
1928 		goto sendpkt;	/* send anyway */
1929 	}
1930 
1931 	/* We don't have to do link-layer address resolution on a p2p link. */
1932 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1933 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1934 		ln->ln_state = ND6_LLINFO_STALE;
1935 		ln->ln_expire = time_second + nd6_gctimer;
1936 	}
1937 
1938 	/*
1939 	 * The first time we send a packet to a neighbor whose entry is
1940 	 * STALE, we have to change the state to DELAY and a sets a timer to
1941 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1942 	 * neighbor unreachability detection on expiration.
1943 	 * (RFC 2461 7.3.3)
1944 	 */
1945 	if (ln->ln_state == ND6_LLINFO_STALE) {
1946 		ln->ln_asked = 0;
1947 		ln->ln_state = ND6_LLINFO_DELAY;
1948 		ln->ln_expire = time_second + nd6_delay;
1949 	}
1950 
1951 	/*
1952 	 * If the neighbor cache entry has a state other than INCOMPLETE
1953 	 * (i.e. its link-layer address is already resolved), just
1954 	 * send the packet.
1955 	 */
1956 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1957 		goto sendpkt;
1958 
1959 	/*
1960 	 * There is a neighbor cache entry, but no ethernet address
1961 	 * response yet.  Replace the held mbuf (if any) with this
1962 	 * latest one.
1963 	 *
1964 	 * This code conforms to the rate-limiting rule described in Section
1965 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1966 	 * an NS below.
1967 	 */
1968 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1969 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1970 	if (ln->ln_hold)
1971 		m_freem(ln->ln_hold);
1972 	ln->ln_hold = m;
1973 	if (ln->ln_expire) {
1974 		if (ln->ln_asked < nd6_mmaxtries &&
1975 		    ln->ln_expire < time_second) {
1976 			ln->ln_asked++;
1977 			ln->ln_expire = time_second +
1978 				ND_IFINFO(ifp)->retrans / 1000;
1979 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1980 		}
1981 	}
1982 	return (0);
1983 
1984   sendpkt:
1985 #ifdef IPSEC
1986 	/* clean ipsec history once it goes out of the node */
1987 	ipsec_delaux(m);
1988 #endif
1989 
1990 #ifdef MAC
1991 	mac_create_mbuf_linklayer(ifp, m);
1992 #endif
1993 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1994 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1995 		    rt));
1996 	}
1997 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1998 
1999   bad:
2000 	if (m)
2001 		m_freem(m);
2002 	return (error);
2003 }
2004 #undef senderr
2005 
2006 int
2007 nd6_need_cache(ifp)
2008 	struct ifnet *ifp;
2009 {
2010 	/*
2011 	 * XXX: we currently do not make neighbor cache on any interface
2012 	 * other than ARCnet, Ethernet, FDDI and GIF.
2013 	 *
2014 	 * RFC2893 says:
2015 	 * - unidirectional tunnels needs no ND
2016 	 */
2017 	switch (ifp->if_type) {
2018 	case IFT_ARCNET:
2019 	case IFT_ETHER:
2020 	case IFT_FDDI:
2021 	case IFT_IEEE1394:
2022 #ifdef IFT_L2VLAN
2023 	case IFT_L2VLAN:
2024 #endif
2025 #ifdef IFT_IEEE80211
2026 	case IFT_IEEE80211:
2027 #endif
2028 	case IFT_GIF:		/* XXX need more cases? */
2029 		return (1);
2030 	default:
2031 		return (0);
2032 	}
2033 }
2034 
2035 int
2036 nd6_storelladdr(ifp, rt0, m, dst, desten)
2037 	struct ifnet *ifp;
2038 	struct rtentry *rt0;
2039 	struct mbuf *m;
2040 	struct sockaddr *dst;
2041 	u_char *desten;
2042 {
2043 	int i;
2044 	struct sockaddr_dl *sdl;
2045 	struct rtentry *rt;
2046 
2047 	if (m->m_flags & M_MCAST) {
2048 		switch (ifp->if_type) {
2049 		case IFT_ETHER:
2050 		case IFT_FDDI:
2051 #ifdef IFT_L2VLAN
2052 		case IFT_L2VLAN:
2053 #endif
2054 #ifdef IFT_IEEE80211
2055 		case IFT_IEEE80211:
2056 #endif
2057 		case IFT_ISO88025:
2058 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2059 						 desten);
2060 			return (0);
2061 		case IFT_IEEE1394:
2062 			/*
2063 			 * netbsd can use if_broadcastaddr, but we don't do so
2064 			 * to reduce # of ifdef.
2065 			 */
2066 			for (i = 0; i < ifp->if_addrlen; i++)
2067 				desten[i] = ~0;
2068 			return (0);
2069 		case IFT_ARCNET:
2070 			*desten = 0;
2071 			return (0);
2072 		default:
2073 			m_freem(m);
2074 			return (EAFNOSUPPORT);
2075 		}
2076 	}
2077 
2078 	i = rt_check(&rt, &rt0, dst);
2079 	if (i) {
2080 		m_freem(m);
2081 		return i;
2082 	}
2083 
2084 	if (rt == NULL) {
2085 		/* this could happen, if we could not allocate memory */
2086 		m_freem(m);
2087 		return (ENOMEM);
2088 	}
2089 	if (rt->rt_gateway->sa_family != AF_LINK) {
2090 		printf("nd6_storelladdr: something odd happens\n");
2091 		m_freem(m);
2092 		return (EINVAL);
2093 	}
2094 	sdl = SDL(rt->rt_gateway);
2095 	if (sdl->sdl_alen == 0) {
2096 		/* this should be impossible, but we bark here for debugging */
2097 		printf("nd6_storelladdr: sdl_alen == 0\n");
2098 		m_freem(m);
2099 		return (EINVAL);
2100 	}
2101 
2102 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2103 	return (0);
2104 }
2105 
2106 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2107 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2108 #ifdef SYSCTL_DECL
2109 SYSCTL_DECL(_net_inet6_icmp6);
2110 #endif
2111 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2112 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2113 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2114 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2115 
2116 static int
2117 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2118 {
2119 	int error;
2120 	char buf[1024];
2121 	struct in6_defrouter *d, *de;
2122 	struct nd_defrouter *dr;
2123 
2124 	if (req->newptr)
2125 		return EPERM;
2126 	error = 0;
2127 
2128 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2129 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2130 		d = (struct in6_defrouter *)buf;
2131 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2132 
2133 		if (d + 1 <= de) {
2134 			bzero(d, sizeof(*d));
2135 			d->rtaddr.sin6_family = AF_INET6;
2136 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2137 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2138 			    dr->ifp) != 0)
2139 				log(LOG_ERR,
2140 				    "scope error in "
2141 				    "default router list (%s)\n",
2142 				    ip6_sprintf(&dr->rtaddr));
2143 			d->flags = dr->flags;
2144 			d->rtlifetime = dr->rtlifetime;
2145 			d->expire = dr->expire;
2146 			d->if_index = dr->ifp->if_index;
2147 		} else
2148 			panic("buffer too short");
2149 
2150 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2151 		if (error)
2152 			break;
2153 	}
2154 
2155 	return (error);
2156 }
2157 
2158 static int
2159 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2160 {
2161 	int error;
2162 	char buf[1024];
2163 	struct in6_prefix *p, *pe;
2164 	struct nd_prefix *pr;
2165 
2166 	if (req->newptr)
2167 		return EPERM;
2168 	error = 0;
2169 
2170 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2171 		u_short advrtrs;
2172 		size_t advance;
2173 		struct sockaddr_in6 *sin6, *s6;
2174 		struct nd_pfxrouter *pfr;
2175 
2176 		p = (struct in6_prefix *)buf;
2177 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2178 
2179 		if (p + 1 <= pe) {
2180 			bzero(p, sizeof(*p));
2181 			sin6 = (struct sockaddr_in6 *)(p + 1);
2182 
2183 			p->prefix = pr->ndpr_prefix;
2184 			if (in6_recoverscope(&p->prefix,
2185 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2186 				log(LOG_ERR,
2187 				    "scope error in prefix list (%s)\n",
2188 				    ip6_sprintf(&p->prefix.sin6_addr));
2189 			p->raflags = pr->ndpr_raf;
2190 			p->prefixlen = pr->ndpr_plen;
2191 			p->vltime = pr->ndpr_vltime;
2192 			p->pltime = pr->ndpr_pltime;
2193 			p->if_index = pr->ndpr_ifp->if_index;
2194 			p->expire = pr->ndpr_expire;
2195 			p->refcnt = pr->ndpr_refcnt;
2196 			p->flags = pr->ndpr_stateflags;
2197 			p->origin = PR_ORIG_RA;
2198 			advrtrs = 0;
2199 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2200 			     pfr = pfr->pfr_next) {
2201 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2202 					advrtrs++;
2203 					continue;
2204 				}
2205 				s6 = &sin6[advrtrs];
2206 				bzero(s6, sizeof(*s6));
2207 				s6->sin6_family = AF_INET6;
2208 				s6->sin6_len = sizeof(*sin6);
2209 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2210 				    pfr->router->ifp) != 0)
2211 					log(LOG_ERR,
2212 					    "scope error in "
2213 					    "prefix list (%s)\n",
2214 					    ip6_sprintf(&pfr->router->rtaddr));
2215 				advrtrs++;
2216 			}
2217 			p->advrtrs = advrtrs;
2218 		} else
2219 			panic("buffer too short");
2220 
2221 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2222 		error = SYSCTL_OUT(req, buf, advance);
2223 		if (error)
2224 			break;
2225 	}
2226 
2227 	return (error);
2228 }
2229