xref: /netbsd-src/sys/netinet6/nd6.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: nd6.c,v 1.86 2003/06/27 08:41:08 itojun Exp $	*/
2 /*	$KAME: nd6.c,v 1.279 2002/06/08 11:16:51 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.86 2003/06/27 08:41:08 itojun Exp $");
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/callout.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/time.h>
44 #include <sys/kernel.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/ioctl.h>
48 #include <sys/syslog.h>
49 #include <sys/queue.h>
50 
51 #include <net/if.h>
52 #include <net/if_dl.h>
53 #include <net/if_types.h>
54 #include <net/route.h>
55 #include <net/if_ether.h>
56 #include <net/if_fddi.h>
57 #include <net/if_arc.h>
58 
59 #include <netinet/in.h>
60 #include <netinet6/in6_var.h>
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet6/nd6.h>
64 #include <netinet/icmp6.h>
65 
66 #include "loop.h"
67 extern struct ifnet loif[NLOOP];
68 
69 #include <net/net_osdep.h>
70 
71 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
72 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
73 
74 #define SIN6(s) ((struct sockaddr_in6 *)s)
75 #define SDL(s) ((struct sockaddr_dl *)s)
76 
77 /* timer values */
78 int	nd6_prune	= 1;	/* walk list every 1 seconds */
79 int	nd6_delay	= 5;	/* delay first probe time 5 second */
80 int	nd6_umaxtries	= 3;	/* maximum unicast query */
81 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
82 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
83 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
84 
85 /* preventing too many loops in ND option parsing */
86 int nd6_maxndopt = 10;	/* max # of ND options allowed */
87 
88 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
89 
90 #ifdef ND6_DEBUG
91 int nd6_debug = 1;
92 #else
93 int nd6_debug = 0;
94 #endif
95 
96 /* for debugging? */
97 static int nd6_inuse, nd6_allocated;
98 
99 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
100 struct nd_drhead nd_defrouter;
101 struct nd_prhead nd_prefix = { 0 };
102 
103 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
104 static struct sockaddr_in6 all1_sa;
105 
106 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
107 static void nd6_slowtimo __P((void *));
108 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
109 static void nd6_llinfo_timer __P((void *));
110 
111 struct callout nd6_slowtimo_ch = CALLOUT_INITIALIZER;
112 struct callout nd6_timer_ch = CALLOUT_INITIALIZER;
113 
114 static int fill_drlist __P((void *, size_t *, size_t));
115 static int fill_prlist __P((void *, size_t *, size_t));
116 
117 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
118 
119 void
120 nd6_init()
121 {
122 	static int nd6_init_done = 0;
123 	int i;
124 
125 	if (nd6_init_done) {
126 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
127 		return;
128 	}
129 
130 	all1_sa.sin6_family = AF_INET6;
131 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
132 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
133 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
134 
135 	/* initialization of the default router list */
136 	TAILQ_INIT(&nd_defrouter);
137 
138 	nd6_init_done = 1;
139 
140 	/* start timer */
141 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
142 	    nd6_slowtimo, NULL);
143 }
144 
145 struct nd_ifinfo *
146 nd6_ifattach(ifp)
147 	struct ifnet *ifp;
148 {
149 	struct nd_ifinfo *nd;
150 
151 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
152 	bzero(nd, sizeof(*nd));
153 
154 	nd->initialized = 1;
155 
156 	nd->chlim = IPV6_DEFHLIM;
157 	nd->basereachable = REACHABLE_TIME;
158 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
159 	nd->retrans = RETRANS_TIMER;
160 	/*
161 	 * Note that the default value of ip6_accept_rtadv is 0, which means
162 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
163 	 * here.
164 	 */
165 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
166 
167 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
168 	nd6_setmtu0(ifp, nd);
169 
170 	return nd;
171 }
172 
173 void
174 nd6_ifdetach(nd)
175 	struct nd_ifinfo *nd;
176 {
177 
178 	free(nd, M_IP6NDP);
179 }
180 
181 void
182 nd6_setmtu(ifp)
183 	struct ifnet *ifp;
184 {
185 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
186 }
187 
188 void
189 nd6_setmtu0(ifp, ndi)
190 	struct ifnet *ifp;
191 	struct nd_ifinfo *ndi;
192 {
193 	u_int32_t omaxmtu;
194 
195 	omaxmtu = ndi->maxmtu;
196 
197 	switch (ifp->if_type) {
198 	case IFT_ARCNET:
199 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
200 		break;
201 	case IFT_FDDI:
202 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
203 		break;
204 	default:
205 		ndi->maxmtu = ifp->if_mtu;
206 		break;
207 	}
208 
209 	/*
210 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
211 	 * undesirable situation.  We thus notify the operator of the change
212 	 * explicitly.  The check for omaxmtu is necessary to restrict the
213 	 * log to the case of changing the MTU, not initializing it.
214 	 */
215 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
216 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
217 		    " small for IPv6 which needs %lu\n",
218 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
219 		    IPV6_MMTU);
220 	}
221 
222 	if (ndi->maxmtu > in6_maxmtu)
223 		in6_setmaxmtu(); /* check all interfaces just in case */
224 }
225 
226 void
227 nd6_option_init(opt, icmp6len, ndopts)
228 	void *opt;
229 	int icmp6len;
230 	union nd_opts *ndopts;
231 {
232 
233 	bzero(ndopts, sizeof(*ndopts));
234 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
235 	ndopts->nd_opts_last
236 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
237 
238 	if (icmp6len == 0) {
239 		ndopts->nd_opts_done = 1;
240 		ndopts->nd_opts_search = NULL;
241 	}
242 }
243 
244 /*
245  * Take one ND option.
246  */
247 struct nd_opt_hdr *
248 nd6_option(ndopts)
249 	union nd_opts *ndopts;
250 {
251 	struct nd_opt_hdr *nd_opt;
252 	int olen;
253 
254 	if (!ndopts)
255 		panic("ndopts == NULL in nd6_option");
256 	if (!ndopts->nd_opts_last)
257 		panic("uninitialized ndopts in nd6_option");
258 	if (!ndopts->nd_opts_search)
259 		return NULL;
260 	if (ndopts->nd_opts_done)
261 		return NULL;
262 
263 	nd_opt = ndopts->nd_opts_search;
264 
265 	/* make sure nd_opt_len is inside the buffer */
266 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
267 		bzero(ndopts, sizeof(*ndopts));
268 		return NULL;
269 	}
270 
271 	olen = nd_opt->nd_opt_len << 3;
272 	if (olen == 0) {
273 		/*
274 		 * Message validation requires that all included
275 		 * options have a length that is greater than zero.
276 		 */
277 		bzero(ndopts, sizeof(*ndopts));
278 		return NULL;
279 	}
280 
281 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
282 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
283 		/* option overruns the end of buffer, invalid */
284 		bzero(ndopts, sizeof(*ndopts));
285 		return NULL;
286 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
287 		/* reached the end of options chain */
288 		ndopts->nd_opts_done = 1;
289 		ndopts->nd_opts_search = NULL;
290 	}
291 	return nd_opt;
292 }
293 
294 /*
295  * Parse multiple ND options.
296  * This function is much easier to use, for ND routines that do not need
297  * multiple options of the same type.
298  */
299 int
300 nd6_options(ndopts)
301 	union nd_opts *ndopts;
302 {
303 	struct nd_opt_hdr *nd_opt;
304 	int i = 0;
305 
306 	if (!ndopts)
307 		panic("ndopts == NULL in nd6_options");
308 	if (!ndopts->nd_opts_last)
309 		panic("uninitialized ndopts in nd6_options");
310 	if (!ndopts->nd_opts_search)
311 		return 0;
312 
313 	while (1) {
314 		nd_opt = nd6_option(ndopts);
315 		if (!nd_opt && !ndopts->nd_opts_last) {
316 			/*
317 			 * Message validation requires that all included
318 			 * options have a length that is greater than zero.
319 			 */
320 			icmp6stat.icp6s_nd_badopt++;
321 			bzero(ndopts, sizeof(*ndopts));
322 			return -1;
323 		}
324 
325 		if (!nd_opt)
326 			goto skip1;
327 
328 		switch (nd_opt->nd_opt_type) {
329 		case ND_OPT_SOURCE_LINKADDR:
330 		case ND_OPT_TARGET_LINKADDR:
331 		case ND_OPT_MTU:
332 		case ND_OPT_REDIRECTED_HEADER:
333 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
334 				nd6log((LOG_INFO,
335 				    "duplicated ND6 option found (type=%d)\n",
336 				    nd_opt->nd_opt_type));
337 				/* XXX bark? */
338 			} else {
339 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
340 					= nd_opt;
341 			}
342 			break;
343 		case ND_OPT_PREFIX_INFORMATION:
344 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
345 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
346 					= nd_opt;
347 			}
348 			ndopts->nd_opts_pi_end =
349 				(struct nd_opt_prefix_info *)nd_opt;
350 			break;
351 		default:
352 			/*
353 			 * Unknown options must be silently ignored,
354 			 * to accomodate future extension to the protocol.
355 			 */
356 			nd6log((LOG_DEBUG,
357 			    "nd6_options: unsupported option %d - "
358 			    "option ignored\n", nd_opt->nd_opt_type));
359 		}
360 
361 skip1:
362 		i++;
363 		if (i > nd6_maxndopt) {
364 			icmp6stat.icp6s_nd_toomanyopt++;
365 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
366 			break;
367 		}
368 
369 		if (ndopts->nd_opts_done)
370 			break;
371 	}
372 
373 	return 0;
374 }
375 
376 /*
377  * ND6 timer routine to handle ND6 entries
378  */
379 void
380 nd6_llinfo_settimer(ln, tick)
381 	struct llinfo_nd6 *ln;
382 	long tick;
383 {
384 	int s;
385 
386 	s = splsoftnet();
387 
388 	if (tick < 0) {
389 		ln->ln_expire = 0;
390 		ln->ln_ntick = 0;
391 		callout_stop(&ln->ln_timer_ch);
392 	} else {
393 		ln->ln_expire = time.tv_sec + tick / hz;
394 		if (tick > INT_MAX) {
395 			ln->ln_ntick = tick - INT_MAX;
396 			callout_reset(&ln->ln_timer_ch, INT_MAX,
397 			    nd6_llinfo_timer, ln);
398 		} else {
399 			ln->ln_ntick = 0;
400 			callout_reset(&ln->ln_timer_ch, tick,
401 			    nd6_llinfo_timer, ln);
402 		}
403 	}
404 
405 	splx(s);
406 }
407 
408 static void
409 nd6_llinfo_timer(arg)
410 	void *arg;
411 {
412 	int s;
413 	struct llinfo_nd6 *ln;
414 	struct rtentry *rt;
415 	const struct sockaddr_in6 *dst;
416 	struct ifnet *ifp;
417 	struct nd_ifinfo *ndi = NULL;
418 
419 	s = splsoftnet();
420 
421 	ln = (struct llinfo_nd6 *)arg;
422 
423 	if (ln->ln_ntick > 0) {
424 		if (ln->ln_ntick > INT_MAX) {
425 			ln->ln_ntick -= INT_MAX;
426 			nd6_llinfo_settimer(ln, INT_MAX);
427 		} else {
428 			ln->ln_ntick = 0;
429 			nd6_llinfo_settimer(ln, ln->ln_ntick);
430 		}
431 		splx(s);
432 		return;
433 	}
434 
435 	if ((rt = ln->ln_rt) == NULL)
436 		panic("ln->ln_rt == NULL");
437 	if ((ifp = rt->rt_ifp) == NULL)
438 		panic("ln->ln_rt->rt_ifp == NULL");
439 	ndi = ND_IFINFO(ifp);
440 	dst = (struct sockaddr_in6 *)rt_key(rt);
441 
442 	/* sanity check */
443 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
444 		panic("rt_llinfo(%p) is not equal to ln(%p)",
445 		      rt->rt_llinfo, ln);
446 	if (!dst)
447 		panic("dst=0 in nd6_timer(ln=%p)", ln);
448 
449 	switch (ln->ln_state) {
450 	case ND6_LLINFO_INCOMPLETE:
451 		if (ln->ln_asked < nd6_mmaxtries) {
452 			ln->ln_asked++;
453 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
454 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
455 		} else {
456 			struct mbuf *m = ln->ln_hold;
457 			if (m) {
458 				ln->ln_hold = NULL;
459 				/*
460 				 * Fake rcvif to make the ICMP error
461 				 * more helpful in diagnosing for the
462 				 * receiver.
463 				 * XXX: should we consider
464 				 * older rcvif?
465 				 */
466 				m->m_pkthdr.rcvif = rt->rt_ifp;
467 
468 				icmp6_error(m, ICMP6_DST_UNREACH,
469 				    ICMP6_DST_UNREACH_ADDR, 0);
470 			}
471 			(void)nd6_free(rt, 0);
472 			ln = NULL;
473 		}
474 		break;
475 	case ND6_LLINFO_REACHABLE:
476 		if (!ND6_LLINFO_PERMANENT(ln)) {
477 			ln->ln_state = ND6_LLINFO_STALE;
478 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
479 		}
480 		break;
481 
482 	case ND6_LLINFO_STALE:
483 		/* Garbage Collection(RFC 2461 5.3) */
484 		if (!ND6_LLINFO_PERMANENT(ln)) {
485 			(void)nd6_free(rt, 1);
486 			ln = NULL;
487 		}
488 		break;
489 
490 	case ND6_LLINFO_DELAY:
491 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
492 			/* We need NUD */
493 			ln->ln_asked = 1;
494 			ln->ln_state = ND6_LLINFO_PROBE;
495 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
496 			nd6_ns_output(ifp, &dst->sin6_addr,
497 			    &dst->sin6_addr, ln, 0);
498 		} else {
499 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
500 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
501 		}
502 		break;
503 	case ND6_LLINFO_PROBE:
504 		if (ln->ln_asked < nd6_umaxtries) {
505 			ln->ln_asked++;
506 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
507 			nd6_ns_output(ifp, &dst->sin6_addr,
508 			    &dst->sin6_addr, ln, 0);
509 		} else {
510 			(void)nd6_free(rt, 0);
511 			ln = NULL;
512 		}
513 		break;
514 	}
515 
516 	splx(s);
517 }
518 
519 /*
520  * ND6 timer routine to expire default route list and prefix list
521  */
522 void
523 nd6_timer(ignored_arg)
524 	void	*ignored_arg;
525 {
526 	int s;
527 	struct nd_defrouter *dr;
528 	struct nd_prefix *pr;
529 	struct in6_ifaddr *ia6, *nia6;
530 	struct in6_addrlifetime *lt6;
531 
532 	s = splsoftnet();
533 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
534 	    nd6_timer, NULL);
535 
536 	/* expire default router list */
537 	dr = TAILQ_FIRST(&nd_defrouter);
538 	while (dr) {
539 		if (dr->expire && dr->expire < time.tv_sec) {
540 			struct nd_defrouter *t;
541 			t = TAILQ_NEXT(dr, dr_entry);
542 			defrtrlist_del(dr);
543 			dr = t;
544 		} else {
545 			dr = TAILQ_NEXT(dr, dr_entry);
546 		}
547 	}
548 
549 	/*
550 	 * expire interface addresses.
551 	 * in the past the loop was inside prefix expiry processing.
552 	 * However, from a stricter speci-confrmance standpoint, we should
553 	 * rather separate address lifetimes and prefix lifetimes.
554 	 */
555 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
556 		nia6 = ia6->ia_next;
557 		/* check address lifetime */
558 		lt6 = &ia6->ia6_lifetime;
559 		if (IFA6_IS_INVALID(ia6)) {
560 			in6_purgeaddr(&ia6->ia_ifa);
561 		}
562 		if (IFA6_IS_DEPRECATED(ia6)) {
563 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
564 		} else {
565 			/*
566 			 * A new RA might have made a deprecated address
567 			 * preferred.
568 			 */
569 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
570 		}
571 	}
572 
573 	/* expire prefix list */
574 	pr = nd_prefix.lh_first;
575 	while (pr) {
576 		/*
577 		 * check prefix lifetime.
578 		 * since pltime is just for autoconf, pltime processing for
579 		 * prefix is not necessary.
580 		 */
581 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
582 		    time.tv_sec - pr->ndpr_lastupdate > pr->ndpr_vltime) {
583 			struct nd_prefix *t;
584 			t = pr->ndpr_next;
585 
586 			/*
587 			 * address expiration and prefix expiration are
588 			 * separate.  NEVER perform in6_purgeaddr here.
589 			 */
590 
591 			prelist_remove(pr);
592 			pr = t;
593 		} else
594 			pr = pr->ndpr_next;
595 	}
596 	splx(s);
597 }
598 
599 /*
600  * Nuke neighbor cache/prefix/default router management table, right before
601  * ifp goes away.
602  */
603 void
604 nd6_purge(ifp)
605 	struct ifnet *ifp;
606 {
607 	struct llinfo_nd6 *ln, *nln;
608 	struct nd_defrouter *dr, *ndr;
609 	struct nd_prefix *pr, *npr;
610 
611 	/*
612 	 * Nuke default router list entries toward ifp.
613 	 * We defer removal of default router list entries that is installed
614 	 * in the routing table, in order to keep additional side effects as
615 	 * small as possible.
616 	 */
617 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
618 		ndr = TAILQ_NEXT(dr, dr_entry);
619 		if (dr->installed)
620 			continue;
621 
622 		if (dr->ifp == ifp)
623 			defrtrlist_del(dr);
624 	}
625 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
626 		ndr = TAILQ_NEXT(dr, dr_entry);
627 		if (!dr->installed)
628 			continue;
629 
630 		if (dr->ifp == ifp)
631 			defrtrlist_del(dr);
632 	}
633 
634 	/* Nuke prefix list entries toward ifp */
635 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
636 		npr = pr->ndpr_next;
637 		if (pr->ndpr_ifp == ifp) {
638 			/*
639 			 * Previously, pr->ndpr_addr is removed as well,
640 			 * but I strongly believe we don't have to do it.
641 			 * nd6_purge() is only called from in6_ifdetach(),
642 			 * which removes all the associated interface addresses
643 			 * by itself.
644 			 * (jinmei@kame.net 20010129)
645 			 */
646 			prelist_remove(pr);
647 		}
648 	}
649 
650 	/* cancel default outgoing interface setting */
651 	if (nd6_defifindex == ifp->if_index)
652 		nd6_setdefaultiface(0);
653 
654 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
655 		/* refresh default router list */
656 		defrouter_select();
657 	}
658 
659 	/*
660 	 * Nuke neighbor cache entries for the ifp.
661 	 * Note that rt->rt_ifp may not be the same as ifp,
662 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
663 	 * nd6_rtrequest(), and ip6_input().
664 	 */
665 	ln = llinfo_nd6.ln_next;
666 	while (ln && ln != &llinfo_nd6) {
667 		struct rtentry *rt;
668 		struct sockaddr_dl *sdl;
669 
670 		nln = ln->ln_next;
671 		rt = ln->ln_rt;
672 		if (rt && rt->rt_gateway &&
673 		    rt->rt_gateway->sa_family == AF_LINK) {
674 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
675 			if (sdl->sdl_index == ifp->if_index)
676 				nln = nd6_free(rt, 0);
677 		}
678 		ln = nln;
679 	}
680 }
681 
682 struct rtentry *
683 nd6_lookup(addr6, create, ifp)
684 	struct in6_addr *addr6;
685 	int create;
686 	struct ifnet *ifp;
687 {
688 	struct rtentry *rt;
689 	struct sockaddr_in6 sin6;
690 
691 	bzero(&sin6, sizeof(sin6));
692 	sin6.sin6_len = sizeof(struct sockaddr_in6);
693 	sin6.sin6_family = AF_INET6;
694 	sin6.sin6_addr = *addr6;
695 	rt = rtalloc1((struct sockaddr *)&sin6, create);
696 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
697 		/*
698 		 * This is the case for the default route.
699 		 * If we want to create a neighbor cache for the address, we
700 		 * should free the route for the destination and allocate an
701 		 * interface route.
702 		 */
703 		if (create) {
704 			RTFREE(rt);
705 			rt = 0;
706 		}
707 	}
708 	if (!rt) {
709 		if (create && ifp) {
710 			int e;
711 
712 			/*
713 			 * If no route is available and create is set,
714 			 * we allocate a host route for the destination
715 			 * and treat it like an interface route.
716 			 * This hack is necessary for a neighbor which can't
717 			 * be covered by our own prefix.
718 			 */
719 			struct ifaddr *ifa =
720 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
721 			if (ifa == NULL)
722 				return (NULL);
723 
724 			/*
725 			 * Create a new route.  RTF_LLINFO is necessary
726 			 * to create a Neighbor Cache entry for the
727 			 * destination in nd6_rtrequest which will be
728 			 * called in rtrequest via ifa->ifa_rtrequest.
729 			 */
730 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
731 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
732 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
733 			    ~RTF_CLONING, &rt)) != 0) {
734 #if 0
735 				log(LOG_ERR,
736 				    "nd6_lookup: failed to add route for a "
737 				    "neighbor(%s), errno=%d\n",
738 				    ip6_sprintf(addr6), e);
739 #endif
740 				return (NULL);
741 			}
742 			if (rt == NULL)
743 				return (NULL);
744 			if (rt->rt_llinfo) {
745 				struct llinfo_nd6 *ln =
746 				    (struct llinfo_nd6 *)rt->rt_llinfo;
747 				ln->ln_state = ND6_LLINFO_NOSTATE;
748 			}
749 		} else
750 			return (NULL);
751 	}
752 	rt->rt_refcnt--;
753 	/*
754 	 * Validation for the entry.
755 	 * Note that the check for rt_llinfo is necessary because a cloned
756 	 * route from a parent route that has the L flag (e.g. the default
757 	 * route to a p2p interface) may have the flag, too, while the
758 	 * destination is not actually a neighbor.
759 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
760 	 *      it might be the loopback interface if the entry is for our
761 	 *      own address on a non-loopback interface. Instead, we should
762 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
763 	 *	interface.
764 	 */
765 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
766 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
767 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
768 		if (create) {
769 			nd6log((LOG_DEBUG,
770 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
771 			    ip6_sprintf(addr6),
772 			    ifp ? if_name(ifp) : "unspec"));
773 		}
774 		return (NULL);
775 	}
776 	return (rt);
777 }
778 
779 /*
780  * Detect if a given IPv6 address identifies a neighbor on a given link.
781  * XXX: should take care of the destination of a p2p link?
782  */
783 int
784 nd6_is_addr_neighbor(addr, ifp)
785 	struct sockaddr_in6 *addr;
786 	struct ifnet *ifp;
787 {
788 	struct nd_prefix *pr;
789 	struct rtentry *rt;
790 
791 	/*
792 	 * A link-local address is always a neighbor.
793 	 * XXX: we should use the sin6_scope_id field rather than the embedded
794 	 * interface index.
795 	 * XXX: a link does not necessarily specify a single interface.
796 	 */
797 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
798 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
799 		return (1);
800 
801 	/*
802 	 * If the address matches one of our on-link prefixes, it should be a
803 	 * neighbor.
804 	 */
805 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
806 		if (pr->ndpr_ifp != ifp)
807 			continue;
808 
809 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
810 			continue;
811 
812 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
813 		    &addr->sin6_addr, &pr->ndpr_mask))
814 			return (1);
815 	}
816 
817 	/*
818 	 * If the default router list is empty, all addresses are regarded
819 	 * as on-link, and thus, as a neighbor.
820 	 * XXX: we restrict the condition to hosts, because routers usually do
821 	 * not have the "default router list".
822 	 */
823 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
824 	    nd6_defifindex == ifp->if_index) {
825 		return (1);
826 	}
827 
828 	/*
829 	 * Even if the address matches none of our addresses, it might be
830 	 * in the neighbor cache.
831 	 */
832 	if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL)
833 		return (1);
834 
835 	return (0);
836 }
837 
838 /*
839  * Free an nd6 llinfo entry.
840  * Since the function would cause significant changes in the kernel, DO NOT
841  * make it global, unless you have a strong reason for the change, and are sure
842  * that the change is safe.
843  */
844 static struct llinfo_nd6 *
845 nd6_free(rt, gc)
846 	struct rtentry *rt;
847 	int gc;
848 {
849 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
850 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
851 	struct nd_defrouter *dr;
852 
853 	/*
854 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
855 	 * even though it is not harmful, it was not really necessary.
856 	 */
857 
858 	/* cancel timer */
859 	nd6_llinfo_settimer(ln, -1);
860 
861 	if (!ip6_forwarding) {
862 		int s;
863 		s = splsoftnet();
864 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
865 		    rt->rt_ifp);
866 
867 		if (dr != NULL && dr->expire &&
868 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
869 			/*
870 			 * If the reason for the deletion is just garbage
871 			 * collection, and the neighbor is an active default
872 			 * router, do not delete it.  Instead, reset the GC
873 			 * timer using the router's lifetime.
874 			 * Simply deleting the entry would affect default
875 			 * router selection, which is not necessarily a good
876 			 * thing, especially when we're using router preference
877 			 * values.
878 			 * XXX: the check for ln_state would be redundant,
879 			 *      but we intentionally keep it just in case.
880 			 */
881 			if (dr->expire > time.tv_sec * hz)
882 				nd6_llinfo_settimer(ln,
883 				    dr->expire - time.tv_sec * hz);
884 			else
885 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
886 			splx(s);
887 			return (ln->ln_next);
888 		}
889 
890 		if (ln->ln_router || dr) {
891 			/*
892 			 * rt6_flush must be called whether or not the neighbor
893 			 * is in the Default Router List.
894 			 * See a corresponding comment in nd6_na_input().
895 			 */
896 			rt6_flush(&in6, rt->rt_ifp);
897 		}
898 
899 		if (dr) {
900 			/*
901 			 * Unreachablity of a router might affect the default
902 			 * router selection and on-link detection of advertised
903 			 * prefixes.
904 			 */
905 
906 			/*
907 			 * Temporarily fake the state to choose a new default
908 			 * router and to perform on-link determination of
909 			 * prefixes correctly.
910 			 * Below the state will be set correctly,
911 			 * or the entry itself will be deleted.
912 			 */
913 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
914 
915 			/*
916 			 * Since defrouter_select() does not affect the
917 			 * on-link determination and MIP6 needs the check
918 			 * before the default router selection, we perform
919 			 * the check now.
920 			 */
921 			pfxlist_onlink_check();
922 
923 			/*
924 			 * refresh default router list
925 			 */
926 			defrouter_select();
927 		}
928 		splx(s);
929 	}
930 
931 	/*
932 	 * Before deleting the entry, remember the next entry as the
933 	 * return value.  We need this because pfxlist_onlink_check() above
934 	 * might have freed other entries (particularly the old next entry) as
935 	 * a side effect (XXX).
936 	 */
937 	next = ln->ln_next;
938 
939 	/*
940 	 * Detach the route from the routing tree and the list of neighbor
941 	 * caches, and disable the route entry not to be used in already
942 	 * cached routes.
943 	 */
944 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
945 	    rt_mask(rt), 0, (struct rtentry **)0);
946 
947 	return (next);
948 }
949 
950 /*
951  * Upper-layer reachability hint for Neighbor Unreachability Detection.
952  *
953  * XXX cost-effective metods?
954  */
955 void
956 nd6_nud_hint(rt, dst6, force)
957 	struct rtentry *rt;
958 	struct in6_addr *dst6;
959 	int force;
960 {
961 	struct llinfo_nd6 *ln;
962 
963 	/*
964 	 * If the caller specified "rt", use that.  Otherwise, resolve the
965 	 * routing table by supplied "dst6".
966 	 */
967 	if (!rt) {
968 		if (!dst6)
969 			return;
970 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
971 			return;
972 	}
973 
974 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
975 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
976 	    !rt->rt_llinfo || !rt->rt_gateway ||
977 	    rt->rt_gateway->sa_family != AF_LINK) {
978 		/* This is not a host route. */
979 		return;
980 	}
981 
982 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
983 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
984 		return;
985 
986 	/*
987 	 * if we get upper-layer reachability confirmation many times,
988 	 * it is possible we have false information.
989 	 */
990 	if (!force) {
991 		ln->ln_byhint++;
992 		if (ln->ln_byhint > nd6_maxnudhint)
993 			return;
994 	}
995 
996 	ln->ln_state = ND6_LLINFO_REACHABLE;
997 	if (!ND6_LLINFO_PERMANENT(ln)) {
998 		nd6_llinfo_settimer(ln,
999 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1000 	}
1001 }
1002 
1003 void
1004 nd6_rtrequest(req, rt, info)
1005 	int	req;
1006 	struct rtentry *rt;
1007 	struct rt_addrinfo *info; /* xxx unused */
1008 {
1009 	struct sockaddr *gate = rt->rt_gateway;
1010 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1011 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1012 	struct ifnet *ifp = rt->rt_ifp;
1013 	struct ifaddr *ifa;
1014 	int mine = 0;
1015 
1016 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1017 		return;
1018 
1019 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1020 		/*
1021 		 * This is probably an interface direct route for a link
1022 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1023 		 * We do not need special treatment below for such a route.
1024 		 * Moreover, the RTF_LLINFO flag which would be set below
1025 		 * would annoy the ndp(8) command.
1026 		 */
1027 		return;
1028 	}
1029 
1030 	if (req == RTM_RESOLVE &&
1031 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1032 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1033 		/*
1034 		 * FreeBSD and BSD/OS often make a cloned host route based
1035 		 * on a less-specific route (e.g. the default route).
1036 		 * If the less specific route does not have a "gateway"
1037 		 * (this is the case when the route just goes to a p2p or an
1038 		 * stf interface), we'll mistakenly make a neighbor cache for
1039 		 * the host route, and will see strange neighbor solicitation
1040 		 * for the corresponding destination.  In order to avoid the
1041 		 * confusion, we check if the destination of the route is
1042 		 * a neighbor in terms of neighbor discovery, and stop the
1043 		 * process if not.  Additionally, we remove the LLINFO flag
1044 		 * so that ndp(8) will not try to get the neighbor information
1045 		 * of the destination.
1046 		 */
1047 		rt->rt_flags &= ~RTF_LLINFO;
1048 		return;
1049 	}
1050 
1051 	switch (req) {
1052 	case RTM_ADD:
1053 		/*
1054 		 * There is no backward compatibility :)
1055 		 *
1056 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1057 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1058 		 *	   rt->rt_flags |= RTF_CLONING;
1059 		 */
1060 		if ((rt->rt_flags & RTF_CLONING) ||
1061 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1062 			/*
1063 			 * Case 1: This route should come from a route to
1064 			 * interface (RTF_CLONING case) or the route should be
1065 			 * treated as on-link but is currently not
1066 			 * (RTF_LLINFO && !ln case).
1067 			 */
1068 			rt_setgate(rt, rt_key(rt),
1069 				   (struct sockaddr *)&null_sdl);
1070 			gate = rt->rt_gateway;
1071 			SDL(gate)->sdl_type = ifp->if_type;
1072 			SDL(gate)->sdl_index = ifp->if_index;
1073 			if (ln)
1074 				nd6_llinfo_settimer(ln, 0);
1075 			if ((rt->rt_flags & RTF_CLONING) != 0)
1076 				break;
1077 		}
1078 		/*
1079 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1080 		 * We don't do that here since llinfo is not ready yet.
1081 		 *
1082 		 * There are also couple of other things to be discussed:
1083 		 * - unsolicited NA code needs improvement beforehand
1084 		 * - RFC2461 says we MAY send multicast unsolicited NA
1085 		 *   (7.2.6 paragraph 4), however, it also says that we
1086 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1087 		 *   we don't have anything like it right now.
1088 		 *   note that the mechanism needs a mutual agreement
1089 		 *   between proxies, which means that we need to implement
1090 		 *   a new protocol, or a new kludge.
1091 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1092 		 *   we need to check ip6forwarding before sending it.
1093 		 *   (or should we allow proxy ND configuration only for
1094 		 *   routers?  there's no mention about proxy ND from hosts)
1095 		 */
1096 #if 0
1097 		/* XXX it does not work */
1098 		if (rt->rt_flags & RTF_ANNOUNCE)
1099 			nd6_na_output(ifp,
1100 			      &SIN6(rt_key(rt))->sin6_addr,
1101 			      &SIN6(rt_key(rt))->sin6_addr,
1102 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1103 			      1, NULL);
1104 #endif
1105 		/* FALLTHROUGH */
1106 	case RTM_RESOLVE:
1107 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1108 			/*
1109 			 * Address resolution isn't necessary for a point to
1110 			 * point link, so we can skip this test for a p2p link.
1111 			 */
1112 			if (gate->sa_family != AF_LINK ||
1113 			    gate->sa_len < sizeof(null_sdl)) {
1114 				log(LOG_DEBUG,
1115 				    "nd6_rtrequest: bad gateway value: %s\n",
1116 				    if_name(ifp));
1117 				break;
1118 			}
1119 			SDL(gate)->sdl_type = ifp->if_type;
1120 			SDL(gate)->sdl_index = ifp->if_index;
1121 		}
1122 		if (ln != NULL)
1123 			break;	/* This happens on a route change */
1124 		/*
1125 		 * Case 2: This route may come from cloning, or a manual route
1126 		 * add with a LL address.
1127 		 */
1128 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1129 		rt->rt_llinfo = (caddr_t)ln;
1130 		if (!ln) {
1131 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1132 			break;
1133 		}
1134 		nd6_inuse++;
1135 		nd6_allocated++;
1136 		Bzero(ln, sizeof(*ln));
1137 		ln->ln_rt = rt;
1138 		callout_init(&ln->ln_timer_ch);
1139 		/* this is required for "ndp" command. - shin */
1140 		if (req == RTM_ADD) {
1141 		        /*
1142 			 * gate should have some valid AF_LINK entry,
1143 			 * and ln->ln_expire should have some lifetime
1144 			 * which is specified by ndp command.
1145 			 */
1146 			ln->ln_state = ND6_LLINFO_REACHABLE;
1147 			ln->ln_byhint = 0;
1148 		} else {
1149 		        /*
1150 			 * When req == RTM_RESOLVE, rt is created and
1151 			 * initialized in rtrequest(), so rt_expire is 0.
1152 			 */
1153 			ln->ln_state = ND6_LLINFO_NOSTATE;
1154 			nd6_llinfo_settimer(ln, 0);
1155 		}
1156 		rt->rt_flags |= RTF_LLINFO;
1157 		ln->ln_next = llinfo_nd6.ln_next;
1158 		llinfo_nd6.ln_next = ln;
1159 		ln->ln_prev = &llinfo_nd6;
1160 		ln->ln_next->ln_prev = ln;
1161 
1162 		/*
1163 		 * check if rt_key(rt) is one of my address assigned
1164 		 * to the interface.
1165 		 */
1166 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1167 		    &SIN6(rt_key(rt))->sin6_addr);
1168 		if (ifa) {
1169 			caddr_t macp = nd6_ifptomac(ifp);
1170 			nd6_llinfo_settimer(ln, -1);
1171 			ln->ln_state = ND6_LLINFO_REACHABLE;
1172 			ln->ln_byhint = 0;
1173 			mine = 1;
1174 			if (macp) {
1175 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1176 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1177 			}
1178 			if (nd6_useloopback) {
1179 				rt->rt_ifp = &loif[0];	/* XXX */
1180 				/*
1181 				 * Make sure rt_ifa be equal to the ifaddr
1182 				 * corresponding to the address.
1183 				 * We need this because when we refer
1184 				 * rt_ifa->ia6_flags in ip6_input, we assume
1185 				 * that the rt_ifa points to the address instead
1186 				 * of the loopback address.
1187 				 */
1188 				if (ifa != rt->rt_ifa) {
1189 					IFAFREE(rt->rt_ifa);
1190 					IFAREF(ifa);
1191 					rt->rt_ifa = ifa;
1192 				}
1193 			}
1194 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1195 			nd6_llinfo_settimer(ln, -1);
1196 			ln->ln_state = ND6_LLINFO_REACHABLE;
1197 			ln->ln_byhint = 0;
1198 
1199 			/* join solicited node multicast for proxy ND */
1200 			if (ifp->if_flags & IFF_MULTICAST) {
1201 				struct in6_addr llsol;
1202 				int error;
1203 
1204 				llsol = SIN6(rt_key(rt))->sin6_addr;
1205 				llsol.s6_addr16[0] = htons(0xff02);
1206 				llsol.s6_addr16[1] = htons(ifp->if_index);
1207 				llsol.s6_addr32[1] = 0;
1208 				llsol.s6_addr32[2] = htonl(1);
1209 				llsol.s6_addr8[12] = 0xff;
1210 
1211 				if (!in6_addmulti(&llsol, ifp, &error)) {
1212 					nd6log((LOG_ERR, "%s: failed to join "
1213 					    "%s (errno=%d)\n", if_name(ifp),
1214 					    ip6_sprintf(&llsol), error));
1215 				}
1216 			}
1217 		}
1218 		break;
1219 
1220 	case RTM_DELETE:
1221 		if (!ln)
1222 			break;
1223 		/* leave from solicited node multicast for proxy ND */
1224 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1225 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1226 			struct in6_addr llsol;
1227 			struct in6_multi *in6m;
1228 
1229 			llsol = SIN6(rt_key(rt))->sin6_addr;
1230 			llsol.s6_addr16[0] = htons(0xff02);
1231 			llsol.s6_addr16[1] = htons(ifp->if_index);
1232 			llsol.s6_addr32[1] = 0;
1233 			llsol.s6_addr32[2] = htonl(1);
1234 			llsol.s6_addr8[12] = 0xff;
1235 
1236 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1237 			if (in6m)
1238 				in6_delmulti(in6m);
1239 		}
1240 		nd6_inuse--;
1241 		ln->ln_next->ln_prev = ln->ln_prev;
1242 		ln->ln_prev->ln_next = ln->ln_next;
1243 		ln->ln_prev = NULL;
1244 		nd6_llinfo_settimer(ln, -1);
1245 		rt->rt_llinfo = 0;
1246 		rt->rt_flags &= ~RTF_LLINFO;
1247 		if (ln->ln_hold)
1248 			m_freem(ln->ln_hold);
1249 		Free((caddr_t)ln);
1250 	}
1251 }
1252 
1253 int
1254 nd6_ioctl(cmd, data, ifp)
1255 	u_long cmd;
1256 	caddr_t	data;
1257 	struct ifnet *ifp;
1258 {
1259 	struct in6_drlist *drl = (struct in6_drlist *)data;
1260 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1261 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1262 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1263 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1264 	struct nd_defrouter *dr;
1265 	struct nd_prefix *pr;
1266 	struct rtentry *rt;
1267 	int i = 0, error = 0;
1268 	int s;
1269 
1270 	switch (cmd) {
1271 	case SIOCGDRLST_IN6:
1272 		/*
1273 		 * obsolete API, use sysctl under net.inet6.icmp6
1274 		 */
1275 		bzero(drl, sizeof(*drl));
1276 		s = splsoftnet();
1277 		dr = TAILQ_FIRST(&nd_defrouter);
1278 		while (dr && i < DRLSTSIZ) {
1279 			drl->defrouter[i].rtaddr = dr->rtaddr;
1280 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1281 				/* XXX: need to this hack for KAME stack */
1282 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1283 			} else
1284 				log(LOG_ERR,
1285 				    "default router list contains a "
1286 				    "non-linklocal address(%s)\n",
1287 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1288 
1289 			drl->defrouter[i].flags = dr->flags;
1290 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1291 			drl->defrouter[i].expire = dr->expire;
1292 			drl->defrouter[i].if_index = dr->ifp->if_index;
1293 			i++;
1294 			dr = TAILQ_NEXT(dr, dr_entry);
1295 		}
1296 		splx(s);
1297 		break;
1298 	case SIOCGPRLST_IN6:
1299 		/*
1300 		 * obsolete API, use sysctl under net.inet6.icmp6
1301 		 *
1302 		 * XXX the structure in6_prlist was changed in backward-
1303 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1304 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1305 		 */
1306 		/*
1307 		 * XXX meaning of fields, especialy "raflags", is very
1308 		 * differnet between RA prefix list and RR/static prefix list.
1309 		 * how about separating ioctls into two?
1310 		 */
1311 		bzero(oprl, sizeof(*oprl));
1312 		s = splsoftnet();
1313 		pr = nd_prefix.lh_first;
1314 		while (pr && i < PRLSTSIZ) {
1315 			struct nd_pfxrouter *pfr;
1316 			int j;
1317 
1318 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1319 			oprl->prefix[i].raflags = pr->ndpr_raf;
1320 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1321 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1322 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1323 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1324 			oprl->prefix[i].expire = pr->ndpr_expire;
1325 
1326 			pfr = pr->ndpr_advrtrs.lh_first;
1327 			j = 0;
1328 			while (pfr) {
1329 				if (j < DRLSTSIZ) {
1330 #define RTRADDR oprl->prefix[i].advrtr[j]
1331 					RTRADDR = pfr->router->rtaddr;
1332 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1333 						/* XXX: hack for KAME */
1334 						RTRADDR.s6_addr16[1] = 0;
1335 					} else
1336 						log(LOG_ERR,
1337 						    "a router(%s) advertises "
1338 						    "a prefix with "
1339 						    "non-link local address\n",
1340 						    ip6_sprintf(&RTRADDR));
1341 #undef RTRADDR
1342 				}
1343 				j++;
1344 				pfr = pfr->pfr_next;
1345 			}
1346 			oprl->prefix[i].advrtrs = j;
1347 			oprl->prefix[i].origin = PR_ORIG_RA;
1348 
1349 			i++;
1350 			pr = pr->ndpr_next;
1351 		}
1352 		splx(s);
1353 
1354 		break;
1355 	case OSIOCGIFINFO_IN6:
1356 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1357 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1358 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1359 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1360 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1361 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1362 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1363 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1364 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1365 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1366 		break;
1367 	case SIOCGIFINFO_IN6:
1368 		ndi->ndi = *ND_IFINFO(ifp);
1369 		break;
1370 	case SIOCSIFINFO_FLAGS:
1371 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1372 		break;
1373 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1374 		/* sync kernel routing table with the default router list */
1375 		defrouter_reset();
1376 		defrouter_select();
1377 		break;
1378 	case SIOCSPFXFLUSH_IN6:
1379 	{
1380 		/* flush all the prefix advertised by routers */
1381 		struct nd_prefix *pr, *next;
1382 
1383 		s = splsoftnet();
1384 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1385 			struct in6_ifaddr *ia, *ia_next;
1386 
1387 			next = pr->ndpr_next;
1388 
1389 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1390 				continue; /* XXX */
1391 
1392 			/* do we really have to remove addresses as well? */
1393 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1394 				/* ia might be removed.  keep the next ptr. */
1395 				ia_next = ia->ia_next;
1396 
1397 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1398 					continue;
1399 
1400 				if (ia->ia6_ndpr == pr)
1401 					in6_purgeaddr(&ia->ia_ifa);
1402 			}
1403 			prelist_remove(pr);
1404 		}
1405 		splx(s);
1406 		break;
1407 	}
1408 	case SIOCSRTRFLUSH_IN6:
1409 	{
1410 		/* flush all the default routers */
1411 		struct nd_defrouter *dr, *next;
1412 
1413 		s = splsoftnet();
1414 		defrouter_reset();
1415 		for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) {
1416 			next = TAILQ_NEXT(dr, dr_entry);
1417 			defrtrlist_del(dr);
1418 		}
1419 		defrouter_select();
1420 		splx(s);
1421 		break;
1422 	}
1423 	case SIOCGNBRINFO_IN6:
1424 	    {
1425 		struct llinfo_nd6 *ln;
1426 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1427 
1428 		/*
1429 		 * XXX: KAME specific hack for scoped addresses
1430 		 *      XXXX: for other scopes than link-local?
1431 		 */
1432 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1433 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1434 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1435 
1436 			if (*idp == 0)
1437 				*idp = htons(ifp->if_index);
1438 		}
1439 
1440 		s = splsoftnet();
1441 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1442 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1443 			error = EINVAL;
1444 			splx(s);
1445 			break;
1446 		}
1447 		nbi->state = ln->ln_state;
1448 		nbi->asked = ln->ln_asked;
1449 		nbi->isrouter = ln->ln_router;
1450 		nbi->expire = ln->ln_expire;
1451 		splx(s);
1452 
1453 		break;
1454 	}
1455 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1456 		ndif->ifindex = nd6_defifindex;
1457 		break;
1458 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1459 		return (nd6_setdefaultiface(ndif->ifindex));
1460 	}
1461 	return (error);
1462 }
1463 
1464 /*
1465  * Create neighbor cache entry and cache link-layer address,
1466  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1467  */
1468 struct rtentry *
1469 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1470 	struct ifnet *ifp;
1471 	struct in6_addr *from;
1472 	char *lladdr;
1473 	int lladdrlen;
1474 	int type;	/* ICMP6 type */
1475 	int code;	/* type dependent information */
1476 {
1477 	struct rtentry *rt = NULL;
1478 	struct llinfo_nd6 *ln = NULL;
1479 	int is_newentry;
1480 	struct sockaddr_dl *sdl = NULL;
1481 	int do_update;
1482 	int olladdr;
1483 	int llchange;
1484 	int newstate = 0;
1485 
1486 	if (!ifp)
1487 		panic("ifp == NULL in nd6_cache_lladdr");
1488 	if (!from)
1489 		panic("from == NULL in nd6_cache_lladdr");
1490 
1491 	/* nothing must be updated for unspecified address */
1492 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1493 		return NULL;
1494 
1495 	/*
1496 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1497 	 * the caller.
1498 	 *
1499 	 * XXX If the link does not have link-layer adderss, what should
1500 	 * we do? (ifp->if_addrlen == 0)
1501 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1502 	 * description on it in NS section (RFC 2461 7.2.3).
1503 	 */
1504 
1505 	rt = nd6_lookup(from, 0, ifp);
1506 	if (!rt) {
1507 #if 0
1508 		/* nothing must be done if there's no lladdr */
1509 		if (!lladdr || !lladdrlen)
1510 			return NULL;
1511 #endif
1512 
1513 		rt = nd6_lookup(from, 1, ifp);
1514 		is_newentry = 1;
1515 	} else {
1516 		/* do nothing if static ndp is set */
1517 		if (rt->rt_flags & RTF_STATIC)
1518 			return NULL;
1519 		is_newentry = 0;
1520 	}
1521 
1522 	if (!rt)
1523 		return NULL;
1524 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1525 fail:
1526 		(void)nd6_free(rt, 0);
1527 		return NULL;
1528 	}
1529 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1530 	if (!ln)
1531 		goto fail;
1532 	if (!rt->rt_gateway)
1533 		goto fail;
1534 	if (rt->rt_gateway->sa_family != AF_LINK)
1535 		goto fail;
1536 	sdl = SDL(rt->rt_gateway);
1537 
1538 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1539 	if (olladdr && lladdr) {
1540 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1541 			llchange = 1;
1542 		else
1543 			llchange = 0;
1544 	} else
1545 		llchange = 0;
1546 
1547 	/*
1548 	 * newentry olladdr  lladdr  llchange	(*=record)
1549 	 *	0	n	n	--	(1)
1550 	 *	0	y	n	--	(2)
1551 	 *	0	n	y	--	(3) * STALE
1552 	 *	0	y	y	n	(4) *
1553 	 *	0	y	y	y	(5) * STALE
1554 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1555 	 *	1	--	y	--	(7) * STALE
1556 	 */
1557 
1558 	if (lladdr) {		/* (3-5) and (7) */
1559 		/*
1560 		 * Record source link-layer address
1561 		 * XXX is it dependent to ifp->if_type?
1562 		 */
1563 		sdl->sdl_alen = ifp->if_addrlen;
1564 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1565 	}
1566 
1567 	if (!is_newentry) {
1568 		if ((!olladdr && lladdr) ||		/* (3) */
1569 		    (olladdr && lladdr && llchange)) {	/* (5) */
1570 			do_update = 1;
1571 			newstate = ND6_LLINFO_STALE;
1572 		} else					/* (1-2,4) */
1573 			do_update = 0;
1574 	} else {
1575 		do_update = 1;
1576 		if (!lladdr)				/* (6) */
1577 			newstate = ND6_LLINFO_NOSTATE;
1578 		else					/* (7) */
1579 			newstate = ND6_LLINFO_STALE;
1580 	}
1581 
1582 	if (do_update) {
1583 		/*
1584 		 * Update the state of the neighbor cache.
1585 		 */
1586 		ln->ln_state = newstate;
1587 
1588 		if (ln->ln_state == ND6_LLINFO_STALE) {
1589 			/*
1590 			 * XXX: since nd6_output() below will cause
1591 			 * state tansition to DELAY and reset the timer,
1592 			 * we must set the timer now, although it is actually
1593 			 * meaningless.
1594 			 */
1595 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1596 
1597 			if (ln->ln_hold) {
1598 				/*
1599 				 * we assume ifp is not a p2p here, so just
1600 				 * set the 2nd argument as the 1st one.
1601 				 */
1602 				nd6_output(ifp, ifp, ln->ln_hold,
1603 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1604 				ln->ln_hold = NULL;
1605 			}
1606 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1607 			/* probe right away */
1608 			nd6_llinfo_settimer((void *)ln, 0);
1609 		}
1610 	}
1611 
1612 	/*
1613 	 * ICMP6 type dependent behavior.
1614 	 *
1615 	 * NS: clear IsRouter if new entry
1616 	 * RS: clear IsRouter
1617 	 * RA: set IsRouter if there's lladdr
1618 	 * redir: clear IsRouter if new entry
1619 	 *
1620 	 * RA case, (1):
1621 	 * The spec says that we must set IsRouter in the following cases:
1622 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1623 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1624 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1625 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1626 	 * neighbor cache, this is similar to (6).
1627 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1628 	 *
1629 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1630 	 *							D R
1631 	 *	0	n	n	--	(1)	c   ?     s
1632 	 *	0	y	n	--	(2)	c   s     s
1633 	 *	0	n	y	--	(3)	c   s     s
1634 	 *	0	y	y	n	(4)	c   s     s
1635 	 *	0	y	y	y	(5)	c   s     s
1636 	 *	1	--	n	--	(6) c	c 	c s
1637 	 *	1	--	y	--	(7) c	c   s	c s
1638 	 *
1639 	 *					(c=clear s=set)
1640 	 */
1641 	switch (type & 0xff) {
1642 	case ND_NEIGHBOR_SOLICIT:
1643 		/*
1644 		 * New entry must have is_router flag cleared.
1645 		 */
1646 		if (is_newentry)	/* (6-7) */
1647 			ln->ln_router = 0;
1648 		break;
1649 	case ND_REDIRECT:
1650 		/*
1651 		 * If the icmp is a redirect to a better router, always set the
1652 		 * is_router flag.  Otherwise, if the entry is newly created,
1653 		 * clear the flag.  [RFC 2461, sec 8.3]
1654 		 */
1655 		if (code == ND_REDIRECT_ROUTER)
1656 			ln->ln_router = 1;
1657 		else if (is_newentry) /* (6-7) */
1658 			ln->ln_router = 0;
1659 		break;
1660 	case ND_ROUTER_SOLICIT:
1661 		/*
1662 		 * is_router flag must always be cleared.
1663 		 */
1664 		ln->ln_router = 0;
1665 		break;
1666 	case ND_ROUTER_ADVERT:
1667 		/*
1668 		 * Mark an entry with lladdr as a router.
1669 		 */
1670 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1671 		    (is_newentry && lladdr)) {			/* (7) */
1672 			ln->ln_router = 1;
1673 		}
1674 		break;
1675 	}
1676 
1677 	/*
1678 	 * When the link-layer address of a router changes, select the
1679 	 * best router again.  In particular, when the neighbor entry is newly
1680 	 * created, it might affect the selection policy.
1681 	 * Question: can we restrict the first condition to the "is_newentry"
1682 	 * case?
1683 	 * XXX: when we hear an RA from a new router with the link-layer
1684 	 * address option, defrouter_select() is called twice, since
1685 	 * defrtrlist_update called the function as well.  However, I believe
1686 	 * we can compromise the overhead, since it only happens the first
1687 	 * time.
1688 	 * XXX: although defrouter_select() should not have a bad effect
1689 	 * for those are not autoconfigured hosts, we explicitly avoid such
1690 	 * cases for safety.
1691 	 */
1692 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1693 		defrouter_select();
1694 
1695 	return rt;
1696 }
1697 
1698 static void
1699 nd6_slowtimo(ignored_arg)
1700     void *ignored_arg;
1701 {
1702 	int s = splsoftnet();
1703 	struct nd_ifinfo *nd6if;
1704 	struct ifnet *ifp;
1705 
1706 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1707 	    nd6_slowtimo, NULL);
1708 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1709 	{
1710 		nd6if = ND_IFINFO(ifp);
1711 		if (nd6if->basereachable && /* already initialized */
1712 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1713 			/*
1714 			 * Since reachable time rarely changes by router
1715 			 * advertisements, we SHOULD insure that a new random
1716 			 * value gets recomputed at least once every few hours.
1717 			 * (RFC 2461, 6.3.4)
1718 			 */
1719 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1720 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1721 		}
1722 	}
1723 	splx(s);
1724 }
1725 
1726 #define senderr(e) { error = (e); goto bad;}
1727 int
1728 nd6_output(ifp, origifp, m0, dst, rt0)
1729 	struct ifnet *ifp;
1730 	struct ifnet *origifp;
1731 	struct mbuf *m0;
1732 	struct sockaddr_in6 *dst;
1733 	struct rtentry *rt0;
1734 {
1735 	struct mbuf *m = m0;
1736 	struct rtentry *rt = rt0;
1737 	struct sockaddr_in6 *gw6 = NULL;
1738 	struct llinfo_nd6 *ln = NULL;
1739 	int error = 0;
1740 
1741 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1742 		goto sendpkt;
1743 
1744 	if (nd6_need_cache(ifp) == 0)
1745 		goto sendpkt;
1746 
1747 	/*
1748 	 * next hop determination.  This routine is derived from ether_outpout.
1749 	 */
1750 	if (rt) {
1751 		if ((rt->rt_flags & RTF_UP) == 0) {
1752 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1753 			    1)) != NULL)
1754 			{
1755 				rt->rt_refcnt--;
1756 				if (rt->rt_ifp != ifp) {
1757 					/* XXX: loop care? */
1758 					return nd6_output(ifp, origifp, m0,
1759 					    dst, rt);
1760 				}
1761 			} else
1762 				senderr(EHOSTUNREACH);
1763 		}
1764 
1765 		if (rt->rt_flags & RTF_GATEWAY) {
1766 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1767 
1768 			/*
1769 			 * We skip link-layer address resolution and NUD
1770 			 * if the gateway is not a neighbor from ND point
1771 			 * of view, regardless of the value of nd_ifinfo.flags.
1772 			 * The second condition is a bit tricky; we skip
1773 			 * if the gateway is our own address, which is
1774 			 * sometimes used to install a route to a p2p link.
1775 			 */
1776 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1777 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1778 				/*
1779 				 * We allow this kind of tricky route only
1780 				 * when the outgoing interface is p2p.
1781 				 * XXX: we may need a more generic rule here.
1782 				 */
1783 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1784 					senderr(EHOSTUNREACH);
1785 
1786 				goto sendpkt;
1787 			}
1788 
1789 			if (rt->rt_gwroute == 0)
1790 				goto lookup;
1791 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1792 				rtfree(rt); rt = rt0;
1793 			lookup:
1794 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1795 				if ((rt = rt->rt_gwroute) == 0)
1796 					senderr(EHOSTUNREACH);
1797 				/* the "G" test below also prevents rt == rt0 */
1798 				if ((rt->rt_flags & RTF_GATEWAY) ||
1799 				    (rt->rt_ifp != ifp)) {
1800 					rt->rt_refcnt--;
1801 					rt0->rt_gwroute = 0;
1802 					senderr(EHOSTUNREACH);
1803 				}
1804 			}
1805 		}
1806 	}
1807 
1808 	/*
1809 	 * Address resolution or Neighbor Unreachability Detection
1810 	 * for the next hop.
1811 	 * At this point, the destination of the packet must be a unicast
1812 	 * or an anycast address(i.e. not a multicast).
1813 	 */
1814 
1815 	/* Look up the neighbor cache for the nexthop */
1816 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1817 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1818 	else {
1819 		/*
1820 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1821 		 * the condition below is not very efficient.  But we believe
1822 		 * it is tolerable, because this should be a rare case.
1823 		 */
1824 		if (nd6_is_addr_neighbor(dst, ifp) &&
1825 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1826 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1827 	}
1828 	if (!ln || !rt) {
1829 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1830 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1831 			log(LOG_DEBUG,
1832 			    "nd6_output: can't allocate llinfo for %s "
1833 			    "(ln=%p, rt=%p)\n",
1834 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1835 			senderr(EIO);	/* XXX: good error? */
1836 		}
1837 
1838 		goto sendpkt;	/* send anyway */
1839 	}
1840 
1841 	/* We don't have to do link-layer address resolution on a p2p link. */
1842 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1843 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1844 		ln->ln_state = ND6_LLINFO_STALE;
1845 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1846 	}
1847 
1848 	/*
1849 	 * The first time we send a packet to a neighbor whose entry is
1850 	 * STALE, we have to change the state to DELAY and a sets a timer to
1851 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1852 	 * neighbor unreachability detection on expiration.
1853 	 * (RFC 2461 7.3.3)
1854 	 */
1855 	if (ln->ln_state == ND6_LLINFO_STALE) {
1856 		ln->ln_asked = 0;
1857 		ln->ln_state = ND6_LLINFO_DELAY;
1858 		nd6_llinfo_settimer(ln, nd6_delay * hz);
1859 	}
1860 
1861 	/*
1862 	 * If the neighbor cache entry has a state other than INCOMPLETE
1863 	 * (i.e. its link-layer address is already resolved), just
1864 	 * send the packet.
1865 	 */
1866 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1867 		goto sendpkt;
1868 
1869 	/*
1870 	 * There is a neighbor cache entry, but no ethernet address
1871 	 * response yet.  Replace the held mbuf (if any) with this
1872 	 * latest one.
1873 	 */
1874 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1875 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1876 	if (ln->ln_hold)
1877 		m_freem(ln->ln_hold);
1878 	ln->ln_hold = m;
1879 	/*
1880 	 * If there has been no NS for the neighbor after entering the
1881 	 * INCOMPLETE state, send the first solicitation.
1882 	 */
1883 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
1884 		ln->ln_asked++;
1885 		nd6_llinfo_settimer(ln,
1886 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1887 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1888 	}
1889 	return (0);
1890 
1891   sendpkt:
1892 
1893 #ifdef IPSEC
1894 	/* clean ipsec history once it goes out of the node */
1895 	ipsec_delaux(m);
1896 #endif
1897 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1898 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1899 					 rt));
1900 	}
1901 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1902 
1903   bad:
1904 	if (m)
1905 		m_freem(m);
1906 	return (error);
1907 }
1908 #undef senderr
1909 
1910 int
1911 nd6_need_cache(ifp)
1912 	struct ifnet *ifp;
1913 {
1914 	/*
1915 	 * XXX: we currently do not make neighbor cache on any interface
1916 	 * other than ARCnet, Ethernet, FDDI and GIF.
1917 	 *
1918 	 * RFC2893 says:
1919 	 * - unidirectional tunnels needs no ND
1920 	 */
1921 	switch (ifp->if_type) {
1922 	case IFT_ARCNET:
1923 	case IFT_ETHER:
1924 	case IFT_FDDI:
1925 	case IFT_IEEE1394:
1926 	case IFT_GIF:		/* XXX need more cases? */
1927 		return (1);
1928 	default:
1929 		return (0);
1930 	}
1931 }
1932 
1933 int
1934 nd6_storelladdr(ifp, rt, m, dst, desten)
1935 	struct ifnet *ifp;
1936 	struct rtentry *rt;
1937 	struct mbuf *m;
1938 	struct sockaddr *dst;
1939 	u_char *desten;
1940 {
1941 	struct sockaddr_dl *sdl;
1942 
1943 	if (m->m_flags & M_MCAST) {
1944 		switch (ifp->if_type) {
1945 		case IFT_ETHER:
1946 		case IFT_FDDI:
1947 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1948 						 desten);
1949 			return (1);
1950 		case IFT_IEEE1394:
1951 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1952 			return (1);
1953 		case IFT_ARCNET:
1954 			*desten = 0;
1955 			return (1);
1956 		default:
1957 			m_freem(m);
1958 			return (0);
1959 		}
1960 	}
1961 
1962 	if (rt == NULL) {
1963 		/* this could happen, if we could not allocate memory */
1964 		m_freem(m);
1965 		return (0);
1966 	}
1967 	if (rt->rt_gateway->sa_family != AF_LINK) {
1968 		printf("nd6_storelladdr: something odd happens\n");
1969 		m_freem(m);
1970 		return (0);
1971 	}
1972 	sdl = SDL(rt->rt_gateway);
1973 	if (sdl->sdl_alen == 0) {
1974 		/* this should be impossible, but we bark here for debugging */
1975 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
1976 		    ip6_sprintf(&SIN6(dst)->sin6_addr), if_name(ifp));
1977 		m_freem(m);
1978 		return (0);
1979 	}
1980 
1981 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
1982 	return (1);
1983 }
1984 
1985 int
1986 nd6_sysctl(name, oldp, oldlenp, newp, newlen)
1987 	int name;
1988 	void *oldp;	/* syscall arg, need copyout */
1989 	size_t *oldlenp;
1990 	void *newp;	/* syscall arg, need copyin */
1991 	size_t newlen;
1992 {
1993 	void *p;
1994 	size_t ol, l;
1995 	int error;
1996 
1997 	error = 0;
1998 	l = 0;
1999 
2000 	if (newp)
2001 		return EPERM;
2002 	if (oldp && !oldlenp)
2003 		return EINVAL;
2004 	ol = oldlenp ? *oldlenp : 0;
2005 
2006 	if (oldp) {
2007 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2008 		if (!p)
2009 			return ENOMEM;
2010 	} else
2011 		p = NULL;
2012 	switch (name) {
2013 	case ICMPV6CTL_ND6_DRLIST:
2014 		error = fill_drlist(p, oldlenp, ol);
2015 		if (!error && p && oldp)
2016 			error = copyout(p, oldp, *oldlenp);
2017 		break;
2018 
2019 	case ICMPV6CTL_ND6_PRLIST:
2020 		error = fill_prlist(p, oldlenp, ol);
2021 		if (!error && p && oldp)
2022 			error = copyout(p, oldp, *oldlenp);
2023 		break;
2024 
2025 	default:
2026 		error = ENOPROTOOPT;
2027 		break;
2028 	}
2029 	if (p)
2030 		free(p, M_TEMP);
2031 
2032 	return (error);
2033 }
2034 
2035 static int
2036 fill_drlist(oldp, oldlenp, ol)
2037 	void *oldp;
2038 	size_t *oldlenp, ol;
2039 {
2040 	int error = 0, s;
2041 	struct in6_defrouter *d = NULL, *de = NULL;
2042 	struct nd_defrouter *dr;
2043 	size_t l;
2044 
2045 	s = splsoftnet();
2046 
2047 	if (oldp) {
2048 		d = (struct in6_defrouter *)oldp;
2049 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2050 	}
2051 	l = 0;
2052 
2053 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2054 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2055 
2056 		if (oldp && d + 1 <= de) {
2057 			bzero(d, sizeof(*d));
2058 			d->rtaddr.sin6_family = AF_INET6;
2059 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2060 			d->rtaddr.sin6_addr = dr->rtaddr;
2061 			in6_recoverscope(&d->rtaddr, &d->rtaddr.sin6_addr,
2062 			    dr->ifp);
2063 			d->flags = dr->flags;
2064 			d->rtlifetime = dr->rtlifetime;
2065 			d->expire = dr->expire;
2066 			d->if_index = dr->ifp->if_index;
2067 		}
2068 
2069 		l += sizeof(*d);
2070 		if (d)
2071 			d++;
2072 	}
2073 
2074 	if (oldp) {
2075 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2076 		if (l > ol)
2077 			error = ENOMEM;
2078 	} else
2079 		*oldlenp = l;
2080 
2081 	splx(s);
2082 
2083 	return (error);
2084 }
2085 
2086 static int
2087 fill_prlist(oldp, oldlenp, ol)
2088 	void *oldp;
2089 	size_t *oldlenp, ol;
2090 {
2091 	int error = 0, s;
2092 	struct nd_prefix *pr;
2093 	struct in6_prefix *p = NULL;
2094 	struct in6_prefix *pe = NULL;
2095 	size_t l;
2096 
2097 	s = splsoftnet();
2098 
2099 	if (oldp) {
2100 		p = (struct in6_prefix *)oldp;
2101 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2102 	}
2103 	l = 0;
2104 
2105 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2106 		u_short advrtrs;
2107 		size_t advance;
2108 		struct sockaddr_in6 *sin6;
2109 		struct sockaddr_in6 *s6;
2110 		struct nd_pfxrouter *pfr;
2111 
2112 		if (oldp && p + 1 <= pe)
2113 		{
2114 			bzero(p, sizeof(*p));
2115 			sin6 = (struct sockaddr_in6 *)(p + 1);
2116 
2117 			p->prefix = pr->ndpr_prefix;
2118 			if (in6_recoverscope(&p->prefix,
2119 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2120 				log(LOG_ERR,
2121 				    "scope error in prefix list (%s)\n",
2122 				    ip6_sprintf(&p->prefix.sin6_addr));
2123 			p->raflags = pr->ndpr_raf;
2124 			p->prefixlen = pr->ndpr_plen;
2125 			p->vltime = pr->ndpr_vltime;
2126 			p->pltime = pr->ndpr_pltime;
2127 			p->if_index = pr->ndpr_ifp->if_index;
2128 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2129 				p->expire = 0;
2130 			else {
2131 				time_t maxexpire;
2132 
2133 				/* XXX: we assume time_t is signed. */
2134 				maxexpire = (-1) &
2135 					~(1 << ((sizeof(maxexpire) * 8) - 1));
2136 				if (pr->ndpr_vltime <
2137 				    maxexpire - pr->ndpr_lastupdate) {
2138 					p->expire = pr->ndpr_lastupdate +
2139 						pr->ndpr_vltime;
2140 				} else
2141 					p->expire = maxexpire;
2142 			}
2143 			p->refcnt = pr->ndpr_refcnt;
2144 			p->flags = pr->ndpr_stateflags;
2145 			p->origin = PR_ORIG_RA;
2146 			advrtrs = 0;
2147 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2148 			     pfr = pfr->pfr_next) {
2149 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2150 					advrtrs++;
2151 					continue;
2152 				}
2153 				s6 = &sin6[advrtrs];
2154 				s6->sin6_family = AF_INET6;
2155 				s6->sin6_len = sizeof(struct sockaddr_in6);
2156 				s6->sin6_addr = pfr->router->rtaddr;
2157 				in6_recoverscope(s6, &s6->sin6_addr,
2158 				    pfr->router->ifp);
2159 				advrtrs++;
2160 			}
2161 			p->advrtrs = advrtrs;
2162 		}
2163 		else {
2164 			advrtrs = 0;
2165 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2166 			     pfr = pfr->pfr_next)
2167 				advrtrs++;
2168 		}
2169 
2170 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2171 		l += advance;
2172 		if (p)
2173 			p = (struct in6_prefix *)((caddr_t)p + advance);
2174 	}
2175 
2176 	if (oldp) {
2177 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2178 		if (l > ol)
2179 			error = ENOMEM;
2180 	} else
2181 		*oldlenp = l;
2182 
2183 	splx(s);
2184 
2185 	return (error);
2186 }
2187