xref: /netbsd-src/sys/netinet6/nd6.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: nd6.c,v 1.105 2006/10/12 01:32:39 christos 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 1.105 2006/10/12 01:32:39 christos Exp $");
35 
36 #include "opt_ipsec.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.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/ioctl.h>
50 #include <sys/syslog.h>
51 #include <sys/queue.h>
52 
53 #include <net/if.h>
54 #include <net/if_dl.h>
55 #include <net/if_types.h>
56 #include <net/route.h>
57 #include <net/if_ether.h>
58 #include <net/if_fddi.h>
59 #include <net/if_arc.h>
60 
61 #include <netinet/in.h>
62 #include <netinet6/in6_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet6/scope6_var.h>
66 #include <netinet6/nd6.h>
67 #include <netinet/icmp6.h>
68 
69 #ifdef IPSEC
70 #include <netinet6/ipsec.h>
71 #endif
72 
73 #include <net/net_osdep.h>
74 
75 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
76 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
77 
78 #define SIN6(s) ((struct sockaddr_in6 *)s)
79 #define SDL(s) ((struct sockaddr_dl *)s)
80 
81 /* timer values */
82 int	nd6_prune	= 1;	/* walk list every 1 seconds */
83 int	nd6_delay	= 5;	/* delay first probe time 5 second */
84 int	nd6_umaxtries	= 3;	/* maximum unicast query */
85 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
86 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
87 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
88 
89 /* preventing too many loops in ND option parsing */
90 int nd6_maxndopt = 10;	/* max # of ND options allowed */
91 
92 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
93 
94 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
95 
96 #ifdef ND6_DEBUG
97 int nd6_debug = 1;
98 #else
99 int nd6_debug = 0;
100 #endif
101 
102 /* for debugging? */
103 static int nd6_inuse, nd6_allocated;
104 
105 struct llinfo_nd6 llinfo_nd6 = {
106 	.ln_prev = &llinfo_nd6,
107 	.ln_next = &llinfo_nd6,
108 };
109 struct nd_drhead nd_defrouter;
110 struct nd_prhead nd_prefix = { 0 };
111 
112 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
113 static struct sockaddr_in6 all1_sa;
114 
115 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
116 static void nd6_slowtimo __P((void *));
117 static int regen_tmpaddr __P((struct in6_ifaddr *));
118 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
119 static void nd6_llinfo_timer __P((void *));
120 static void clear_llinfo_pqueue __P((struct llinfo_nd6 *));
121 
122 struct callout nd6_slowtimo_ch = CALLOUT_INITIALIZER;
123 struct callout nd6_timer_ch = CALLOUT_INITIALIZER;
124 extern struct callout in6_tmpaddrtimer_ch;
125 
126 static int fill_drlist __P((void *, size_t *, size_t));
127 static int fill_prlist __P((void *, size_t *, size_t));
128 
129 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
130 
131 void
132 nd6_init()
133 {
134 	static int nd6_init_done = 0;
135 	int i;
136 
137 	if (nd6_init_done) {
138 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
139 		return;
140 	}
141 
142 	all1_sa.sin6_family = AF_INET6;
143 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
144 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
145 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
146 
147 	/* initialization of the default router list */
148 	TAILQ_INIT(&nd_defrouter);
149 
150 	nd6_init_done = 1;
151 
152 	/* start timer */
153 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
154 	    nd6_slowtimo, NULL);
155 }
156 
157 struct nd_ifinfo *
158 nd6_ifattach(ifp)
159 	struct ifnet *ifp;
160 {
161 	struct nd_ifinfo *nd;
162 
163 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
164 	bzero(nd, sizeof(*nd));
165 
166 	nd->initialized = 1;
167 
168 	nd->chlim = IPV6_DEFHLIM;
169 	nd->basereachable = REACHABLE_TIME;
170 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
171 	nd->retrans = RETRANS_TIMER;
172 	/*
173 	 * Note that the default value of ip6_accept_rtadv is 0, which means
174 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
175 	 * here.
176 	 */
177 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
178 
179 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
180 	nd6_setmtu0(ifp, nd);
181 
182 	return nd;
183 }
184 
185 void
186 nd6_ifdetach(nd)
187 	struct nd_ifinfo *nd;
188 {
189 
190 	free(nd, M_IP6NDP);
191 }
192 
193 void
194 nd6_setmtu(ifp)
195 	struct ifnet *ifp;
196 {
197 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
198 }
199 
200 void
201 nd6_setmtu0(ifp, ndi)
202 	struct ifnet *ifp;
203 	struct nd_ifinfo *ndi;
204 {
205 	u_int32_t omaxmtu;
206 
207 	omaxmtu = ndi->maxmtu;
208 
209 	switch (ifp->if_type) {
210 	case IFT_ARCNET:
211 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
212 		break;
213 	case IFT_FDDI:
214 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
215 		break;
216 	default:
217 		ndi->maxmtu = ifp->if_mtu;
218 		break;
219 	}
220 
221 	/*
222 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
223 	 * undesirable situation.  We thus notify the operator of the change
224 	 * explicitly.  The check for omaxmtu is necessary to restrict the
225 	 * log to the case of changing the MTU, not initializing it.
226 	 */
227 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
228 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
229 		    " small for IPv6 which needs %lu\n",
230 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
231 		    IPV6_MMTU);
232 	}
233 
234 	if (ndi->maxmtu > in6_maxmtu)
235 		in6_setmaxmtu(); /* check all interfaces just in case */
236 }
237 
238 void
239 nd6_option_init(opt, icmp6len, ndopts)
240 	void *opt;
241 	int icmp6len;
242 	union nd_opts *ndopts;
243 {
244 
245 	bzero(ndopts, sizeof(*ndopts));
246 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
247 	ndopts->nd_opts_last
248 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
249 
250 	if (icmp6len == 0) {
251 		ndopts->nd_opts_done = 1;
252 		ndopts->nd_opts_search = NULL;
253 	}
254 }
255 
256 /*
257  * Take one ND option.
258  */
259 struct nd_opt_hdr *
260 nd6_option(ndopts)
261 	union nd_opts *ndopts;
262 {
263 	struct nd_opt_hdr *nd_opt;
264 	int olen;
265 
266 	if (ndopts == NULL)
267 		panic("ndopts == NULL in nd6_option");
268 	if (ndopts->nd_opts_last == NULL)
269 		panic("uninitialized ndopts in nd6_option");
270 	if (ndopts->nd_opts_search == NULL)
271 		return NULL;
272 	if (ndopts->nd_opts_done)
273 		return NULL;
274 
275 	nd_opt = ndopts->nd_opts_search;
276 
277 	/* make sure nd_opt_len is inside the buffer */
278 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
279 		bzero(ndopts, sizeof(*ndopts));
280 		return NULL;
281 	}
282 
283 	olen = nd_opt->nd_opt_len << 3;
284 	if (olen == 0) {
285 		/*
286 		 * Message validation requires that all included
287 		 * options have a length that is greater than zero.
288 		 */
289 		bzero(ndopts, sizeof(*ndopts));
290 		return NULL;
291 	}
292 
293 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
294 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
295 		/* option overruns the end of buffer, invalid */
296 		bzero(ndopts, sizeof(*ndopts));
297 		return NULL;
298 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
299 		/* reached the end of options chain */
300 		ndopts->nd_opts_done = 1;
301 		ndopts->nd_opts_search = NULL;
302 	}
303 	return nd_opt;
304 }
305 
306 /*
307  * Parse multiple ND options.
308  * This function is much easier to use, for ND routines that do not need
309  * multiple options of the same type.
310  */
311 int
312 nd6_options(ndopts)
313 	union nd_opts *ndopts;
314 {
315 	struct nd_opt_hdr *nd_opt;
316 	int i = 0;
317 
318 	if (ndopts == NULL)
319 		panic("ndopts == NULL in nd6_options");
320 	if (ndopts->nd_opts_last == NULL)
321 		panic("uninitialized ndopts in nd6_options");
322 	if (ndopts->nd_opts_search == NULL)
323 		return 0;
324 
325 	while (1) {
326 		nd_opt = nd6_option(ndopts);
327 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
328 			/*
329 			 * Message validation requires that all included
330 			 * options have a length that is greater than zero.
331 			 */
332 			icmp6stat.icp6s_nd_badopt++;
333 			bzero(ndopts, sizeof(*ndopts));
334 			return -1;
335 		}
336 
337 		if (nd_opt == NULL)
338 			goto skip1;
339 
340 		switch (nd_opt->nd_opt_type) {
341 		case ND_OPT_SOURCE_LINKADDR:
342 		case ND_OPT_TARGET_LINKADDR:
343 		case ND_OPT_MTU:
344 		case ND_OPT_REDIRECTED_HEADER:
345 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
346 				nd6log((LOG_INFO,
347 				    "duplicated ND6 option found (type=%d)\n",
348 				    nd_opt->nd_opt_type));
349 				/* XXX bark? */
350 			} else {
351 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
352 					= nd_opt;
353 			}
354 			break;
355 		case ND_OPT_PREFIX_INFORMATION:
356 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
357 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
358 					= nd_opt;
359 			}
360 			ndopts->nd_opts_pi_end =
361 				(struct nd_opt_prefix_info *)nd_opt;
362 			break;
363 		default:
364 			/*
365 			 * Unknown options must be silently ignored,
366 			 * to accomodate future extension to the protocol.
367 			 */
368 			nd6log((LOG_DEBUG,
369 			    "nd6_options: unsupported option %d - "
370 			    "option ignored\n", nd_opt->nd_opt_type));
371 		}
372 
373 skip1:
374 		i++;
375 		if (i > nd6_maxndopt) {
376 			icmp6stat.icp6s_nd_toomanyopt++;
377 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
378 			break;
379 		}
380 
381 		if (ndopts->nd_opts_done)
382 			break;
383 	}
384 
385 	return 0;
386 }
387 
388 /*
389  * ND6 timer routine to handle ND6 entries
390  */
391 void
392 nd6_llinfo_settimer(ln, xtick)
393 	struct llinfo_nd6 *ln;
394 	long xtick;
395 {
396 	int s;
397 
398 	s = splsoftnet();
399 
400 	if (xtick < 0) {
401 		ln->ln_expire = 0;
402 		ln->ln_ntick = 0;
403 		callout_stop(&ln->ln_timer_ch);
404 	} else {
405 		ln->ln_expire = time_second + xtick / hz;
406 		if (xtick > INT_MAX) {
407 			ln->ln_ntick = xtick - INT_MAX;
408 			callout_reset(&ln->ln_timer_ch, INT_MAX,
409 			    nd6_llinfo_timer, ln);
410 		} else {
411 			ln->ln_ntick = 0;
412 			callout_reset(&ln->ln_timer_ch, xtick,
413 			    nd6_llinfo_timer, ln);
414 		}
415 	}
416 
417 	splx(s);
418 }
419 
420 static void
421 nd6_llinfo_timer(arg)
422 	void *arg;
423 {
424 	int s;
425 	struct llinfo_nd6 *ln;
426 	struct rtentry *rt;
427 	const struct sockaddr_in6 *dst;
428 	struct ifnet *ifp;
429 	struct nd_ifinfo *ndi = NULL;
430 
431 	s = splsoftnet();
432 
433 	ln = (struct llinfo_nd6 *)arg;
434 
435 	if (ln->ln_ntick > 0) {
436 		nd6_llinfo_settimer(ln, ln->ln_ntick);
437 		splx(s);
438 		return;
439 	}
440 
441 	if ((rt = ln->ln_rt) == NULL)
442 		panic("ln->ln_rt == NULL");
443 	if ((ifp = rt->rt_ifp) == NULL)
444 		panic("ln->ln_rt->rt_ifp == NULL");
445 	ndi = ND_IFINFO(ifp);
446 	dst = (struct sockaddr_in6 *)rt_key(rt);
447 
448 	/* sanity check */
449 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
450 		panic("rt_llinfo(%p) is not equal to ln(%p)",
451 		      rt->rt_llinfo, ln);
452 	if (!dst)
453 		panic("dst=0 in nd6_timer(ln=%p)", ln);
454 
455 	switch (ln->ln_state) {
456 	case ND6_LLINFO_INCOMPLETE:
457 		if (ln->ln_asked < nd6_mmaxtries) {
458 			ln->ln_asked++;
459 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
460 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
461 		} else {
462 			struct mbuf *m = ln->ln_hold;
463 			if (m) {
464 				struct mbuf *m0;
465 
466 				/*
467 				 * assuming every packet in ln_hold has
468 				 * the same IP header
469 				 */
470 				m0 = m->m_nextpkt;
471 				m->m_nextpkt = NULL;
472 				icmp6_error2(m, ICMP6_DST_UNREACH,
473 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
474 
475 				ln->ln_hold = m0;
476 				clear_llinfo_pqueue(ln);
477  			}
478 			(void)nd6_free(rt, 0);
479 			ln = NULL;
480 		}
481 		break;
482 	case ND6_LLINFO_REACHABLE:
483 		if (!ND6_LLINFO_PERMANENT(ln)) {
484 			ln->ln_state = ND6_LLINFO_STALE;
485 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
486 		}
487 		break;
488 
489 	case ND6_LLINFO_STALE:
490 		/* Garbage Collection(RFC 2461 5.3) */
491 		if (!ND6_LLINFO_PERMANENT(ln)) {
492 			(void)nd6_free(rt, 1);
493 			ln = NULL;
494 		}
495 		break;
496 
497 	case ND6_LLINFO_DELAY:
498 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
499 			/* We need NUD */
500 			ln->ln_asked = 1;
501 			ln->ln_state = ND6_LLINFO_PROBE;
502 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
503 			nd6_ns_output(ifp, &dst->sin6_addr,
504 			    &dst->sin6_addr, ln, 0);
505 		} else {
506 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
507 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
508 		}
509 		break;
510 	case ND6_LLINFO_PROBE:
511 		if (ln->ln_asked < nd6_umaxtries) {
512 			ln->ln_asked++;
513 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
514 			nd6_ns_output(ifp, &dst->sin6_addr,
515 			    &dst->sin6_addr, ln, 0);
516 		} else {
517 			(void)nd6_free(rt, 0);
518 			ln = NULL;
519 		}
520 		break;
521 	}
522 
523 	splx(s);
524 }
525 
526 /*
527  * ND6 timer routine to expire default route list and prefix list
528  */
529 void
530 nd6_timer(void *ignored_arg __unused)
531 {
532 	int s;
533 	struct nd_defrouter *dr;
534 	struct nd_prefix *pr;
535 	struct in6_ifaddr *ia6, *nia6;
536 	struct in6_addrlifetime *lt6;
537 
538 	s = splsoftnet();
539 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
540 	    nd6_timer, NULL);
541 
542 	/* expire default router list */
543 	dr = TAILQ_FIRST(&nd_defrouter);
544 	while (dr) {
545 		if (dr->expire && dr->expire < time_second) {
546 			struct nd_defrouter *t;
547 			t = TAILQ_NEXT(dr, dr_entry);
548 			defrtrlist_del(dr);
549 			dr = t;
550 		} else {
551 			dr = TAILQ_NEXT(dr, dr_entry);
552 		}
553 	}
554 
555 	/*
556 	 * expire interface addresses.
557 	 * in the past the loop was inside prefix expiry processing.
558 	 * However, from a stricter speci-confrmance standpoint, we should
559 	 * rather separate address lifetimes and prefix lifetimes.
560 	 */
561   addrloop:
562 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
563 		nia6 = ia6->ia_next;
564 		/* check address lifetime */
565 		lt6 = &ia6->ia6_lifetime;
566 		if (IFA6_IS_INVALID(ia6)) {
567 			int regen = 0;
568 
569 			/*
570 			 * If the expiring address is temporary, try
571 			 * regenerating a new one.  This would be useful when
572 			 * we suspended a laptop PC, then turned it on after a
573 			 * period that could invalidate all temporary
574 			 * addresses.  Although we may have to restart the
575 			 * loop (see below), it must be after purging the
576 			 * address.  Otherwise, we'd see an infinite loop of
577 			 * regeneration.
578 			 */
579 			if (ip6_use_tempaddr &&
580 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
581 				if (regen_tmpaddr(ia6) == 0)
582 					regen = 1;
583 			}
584 
585  			in6_purgeaddr(&ia6->ia_ifa);
586 
587 			if (regen)
588 				goto addrloop; /* XXX: see below */
589 		} else if (IFA6_IS_DEPRECATED(ia6)) {
590 			int oldflags = ia6->ia6_flags;
591 
592  			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
593 
594 			/*
595 			 * If a temporary address has just become deprecated,
596 			 * regenerate a new one if possible.
597 			 */
598 			if (ip6_use_tempaddr &&
599 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
600 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
601 
602 				if (regen_tmpaddr(ia6) == 0) {
603 					/*
604 					 * A new temporary address is
605 					 * generated.
606 					 * XXX: this means the address chain
607 					 * has changed while we are still in
608 					 * the loop.  Although the change
609 					 * would not cause disaster (because
610 					 * it's not a deletion, but an
611 					 * addition,) we'd rather restart the
612 					 * loop just for safety.  Or does this
613 					 * significantly reduce performance??
614 					 */
615 					goto addrloop;
616 				}
617 			}
618 		} else {
619 			/*
620 			 * A new RA might have made a deprecated address
621 			 * preferred.
622 			 */
623 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
624 		}
625 	}
626 
627 	/* expire prefix list */
628 	pr = nd_prefix.lh_first;
629 	while (pr) {
630 		/*
631 		 * check prefix lifetime.
632 		 * since pltime is just for autoconf, pltime processing for
633 		 * prefix is not necessary.
634 		 */
635 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
636 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
637 			struct nd_prefix *t;
638 			t = pr->ndpr_next;
639 
640 			/*
641 			 * address expiration and prefix expiration are
642 			 * separate.  NEVER perform in6_purgeaddr here.
643 			 */
644 
645 			prelist_remove(pr);
646 			pr = t;
647 		} else
648 			pr = pr->ndpr_next;
649 	}
650 	splx(s);
651 }
652 
653 static int
654 regen_tmpaddr(ia6)
655 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
656 {
657 	struct ifaddr *ifa;
658 	struct ifnet *ifp;
659 	struct in6_ifaddr *public_ifa6 = NULL;
660 
661 	ifp = ia6->ia_ifa.ifa_ifp;
662 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
663 	     ifa = ifa->ifa_list.tqe_next) {
664 		struct in6_ifaddr *it6;
665 
666 		if (ifa->ifa_addr->sa_family != AF_INET6)
667 			continue;
668 
669 		it6 = (struct in6_ifaddr *)ifa;
670 
671 		/* ignore no autoconf addresses. */
672 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
673 			continue;
674 
675 		/* ignore autoconf addresses with different prefixes. */
676 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
677 			continue;
678 
679 		/*
680 		 * Now we are looking at an autoconf address with the same
681 		 * prefix as ours.  If the address is temporary and is still
682 		 * preferred, do not create another one.  It would be rare, but
683 		 * could happen, for example, when we resume a laptop PC after
684 		 * a long period.
685 		 */
686 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
687 		    !IFA6_IS_DEPRECATED(it6)) {
688 			public_ifa6 = NULL;
689 			break;
690 		}
691 
692 		/*
693 		 * This is a public autoconf address that has the same prefix
694 		 * as ours.  If it is preferred, keep it.  We can't break the
695 		 * loop here, because there may be a still-preferred temporary
696 		 * address with the prefix.
697 		 */
698 		if (!IFA6_IS_DEPRECATED(it6))
699 		    public_ifa6 = it6;
700 	}
701 
702 	if (public_ifa6 != NULL) {
703 		int e;
704 
705 		/*
706 		 * Random factor is introduced in the preferred lifetime, so
707 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
708 		 */
709 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
710 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
711 			    " tmp addr, errno=%d\n", e);
712 			return (-1);
713 		}
714 		return (0);
715 	}
716 
717 	return (-1);
718 }
719 
720 /*
721  * Nuke neighbor cache/prefix/default router management table, right before
722  * ifp goes away.
723  */
724 void
725 nd6_purge(ifp)
726 	struct ifnet *ifp;
727 {
728 	struct llinfo_nd6 *ln, *nln;
729 	struct nd_defrouter *dr, *ndr;
730 	struct nd_prefix *pr, *npr;
731 
732 	/*
733 	 * Nuke default router list entries toward ifp.
734 	 * We defer removal of default router list entries that is installed
735 	 * in the routing table, in order to keep additional side effects as
736 	 * small as possible.
737 	 */
738 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
739 		ndr = TAILQ_NEXT(dr, dr_entry);
740 		if (dr->installed)
741 			continue;
742 
743 		if (dr->ifp == ifp)
744 			defrtrlist_del(dr);
745 	}
746 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
747 		ndr = TAILQ_NEXT(dr, dr_entry);
748 		if (!dr->installed)
749 			continue;
750 
751 		if (dr->ifp == ifp)
752 			defrtrlist_del(dr);
753 	}
754 
755 	/* Nuke prefix list entries toward ifp */
756 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
757 		npr = pr->ndpr_next;
758 		if (pr->ndpr_ifp == ifp) {
759 			/*
760 			 * Because if_detach() does *not* release prefixes
761 			 * while purging addresses the reference count will
762 			 * still be above zero. We therefore reset it to
763 			 * make sure that the prefix really gets purged.
764 			 */
765 			pr->ndpr_refcnt = 0;
766 			/*
767 			 * Previously, pr->ndpr_addr is removed as well,
768 			 * but I strongly believe we don't have to do it.
769 			 * nd6_purge() is only called from in6_ifdetach(),
770 			 * which removes all the associated interface addresses
771 			 * by itself.
772 			 * (jinmei@kame.net 20010129)
773 			 */
774 			prelist_remove(pr);
775 		}
776 	}
777 
778 	/* cancel default outgoing interface setting */
779 	if (nd6_defifindex == ifp->if_index)
780 		nd6_setdefaultiface(0);
781 
782 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
783 		/* refresh default router list */
784 		defrouter_select();
785 	}
786 
787 	/*
788 	 * Nuke neighbor cache entries for the ifp.
789 	 * Note that rt->rt_ifp may not be the same as ifp,
790 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
791 	 * nd6_rtrequest(), and ip6_input().
792 	 */
793 	ln = llinfo_nd6.ln_next;
794 	while (ln && ln != &llinfo_nd6) {
795 		struct rtentry *rt;
796 		struct sockaddr_dl *sdl;
797 
798 		nln = ln->ln_next;
799 		rt = ln->ln_rt;
800 		if (rt && rt->rt_gateway &&
801 		    rt->rt_gateway->sa_family == AF_LINK) {
802 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
803 			if (sdl->sdl_index == ifp->if_index)
804 				nln = nd6_free(rt, 0);
805 		}
806 		ln = nln;
807 	}
808 }
809 
810 struct rtentry *
811 nd6_lookup(addr6, create, ifp)
812 	struct in6_addr *addr6;
813 	int create;
814 	struct ifnet *ifp;
815 {
816 	struct rtentry *rt;
817 	struct sockaddr_in6 sin6;
818 
819 	bzero(&sin6, sizeof(sin6));
820 	sin6.sin6_len = sizeof(struct sockaddr_in6);
821 	sin6.sin6_family = AF_INET6;
822 	sin6.sin6_addr = *addr6;
823 	rt = rtalloc1((struct sockaddr *)&sin6, create);
824 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
825 		/*
826 		 * This is the case for the default route.
827 		 * If we want to create a neighbor cache for the address, we
828 		 * should free the route for the destination and allocate an
829 		 * interface route.
830 		 */
831 		if (create) {
832 			RTFREE(rt);
833 			rt = NULL;
834 		}
835 	}
836 	if (rt == NULL) {
837 		if (create && ifp) {
838 			int e;
839 
840 			/*
841 			 * If no route is available and create is set,
842 			 * we allocate a host route for the destination
843 			 * and treat it like an interface route.
844 			 * This hack is necessary for a neighbor which can't
845 			 * be covered by our own prefix.
846 			 */
847 			struct ifaddr *ifa =
848 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
849 			if (ifa == NULL)
850 				return (NULL);
851 
852 			/*
853 			 * Create a new route.  RTF_LLINFO is necessary
854 			 * to create a Neighbor Cache entry for the
855 			 * destination in nd6_rtrequest which will be
856 			 * called in rtrequest via ifa->ifa_rtrequest.
857 			 */
858 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
859 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
860 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
861 			    ~RTF_CLONING, &rt)) != 0) {
862 #if 0
863 				log(LOG_ERR,
864 				    "nd6_lookup: failed to add route for a "
865 				    "neighbor(%s), errno=%d\n",
866 				    ip6_sprintf(addr6), e);
867 #endif
868 				return (NULL);
869 			}
870 			if (rt == NULL)
871 				return (NULL);
872 			if (rt->rt_llinfo) {
873 				struct llinfo_nd6 *ln =
874 				    (struct llinfo_nd6 *)rt->rt_llinfo;
875 				ln->ln_state = ND6_LLINFO_NOSTATE;
876 			}
877 		} else
878 			return (NULL);
879 	}
880 	rt->rt_refcnt--;
881 	/*
882 	 * Validation for the entry.
883 	 * Note that the check for rt_llinfo is necessary because a cloned
884 	 * route from a parent route that has the L flag (e.g. the default
885 	 * route to a p2p interface) may have the flag, too, while the
886 	 * destination is not actually a neighbor.
887 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
888 	 *      it might be the loopback interface if the entry is for our
889 	 *      own address on a non-loopback interface. Instead, we should
890 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
891 	 *	interface.
892 	 * Note also that ifa_ifp and ifp may differ when we connect two
893 	 * interfaces to a same link, install a link prefix to an interface,
894 	 * and try to install a neighbor cache on an interface that does not
895 	 * have a route to the prefix.
896 	 */
897 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
898 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
899 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
900 		if (create) {
901 			nd6log((LOG_DEBUG,
902 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
903 			    ip6_sprintf(addr6),
904 			    ifp ? if_name(ifp) : "unspec"));
905 		}
906 		return (NULL);
907 	}
908 	return (rt);
909 }
910 
911 /*
912  * Detect if a given IPv6 address identifies a neighbor on a given link.
913  * XXX: should take care of the destination of a p2p link?
914  */
915 int
916 nd6_is_addr_neighbor(addr, ifp)
917 	struct sockaddr_in6 *addr;
918 	struct ifnet *ifp;
919 {
920 	struct nd_prefix *pr;
921 
922 	/*
923 	 * A link-local address is always a neighbor.
924 	 * XXX: a link does not necessarily specify a single interface.
925 	 */
926 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
927 		struct sockaddr_in6 sin6_copy;
928 		u_int32_t zone;
929 
930 		/*
931 		 * We need sin6_copy since sa6_recoverscope() may modify the
932 		 * content (XXX).
933 		 */
934 		sin6_copy = *addr;
935 		if (sa6_recoverscope(&sin6_copy))
936 			return (0); /* XXX: should be impossible */
937 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
938 			return (0);
939 		if (sin6_copy.sin6_scope_id == zone)
940 			return (1);
941 		else
942 			return (0);
943 	}
944 
945 	/*
946 	 * If the address matches one of our on-link prefixes, it should be a
947 	 * neighbor.
948 	 */
949 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
950 		if (pr->ndpr_ifp != ifp)
951 			continue;
952 
953 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
954 			continue;
955 
956 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
957 		    &addr->sin6_addr, &pr->ndpr_mask))
958 			return (1);
959 	}
960 
961 	/*
962 	 * If the default router list is empty, all addresses are regarded
963 	 * as on-link, and thus, as a neighbor.
964 	 * XXX: we restrict the condition to hosts, because routers usually do
965 	 * not have the "default router list".
966 	 */
967 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
968 	    nd6_defifindex == ifp->if_index) {
969 		return (1);
970 	}
971 
972 	/*
973 	 * Even if the address matches none of our addresses, it might be
974 	 * in the neighbor cache.
975 	 */
976 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
977 		return (1);
978 
979 	return (0);
980 }
981 
982 /*
983  * Free an nd6 llinfo entry.
984  * Since the function would cause significant changes in the kernel, DO NOT
985  * make it global, unless you have a strong reason for the change, and are sure
986  * that the change is safe.
987  */
988 static struct llinfo_nd6 *
989 nd6_free(rt, gc)
990 	struct rtentry *rt;
991 	int gc;
992 {
993 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
994 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
995 	struct nd_defrouter *dr;
996 
997 	/*
998 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
999 	 * even though it is not harmful, it was not really necessary.
1000 	 */
1001 
1002 	/* cancel timer */
1003 	nd6_llinfo_settimer(ln, -1);
1004 
1005 	if (!ip6_forwarding) {
1006 		int s;
1007 		s = splsoftnet();
1008 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1009 		    rt->rt_ifp);
1010 
1011 		if (dr != NULL && dr->expire &&
1012 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1013 			/*
1014 			 * If the reason for the deletion is just garbage
1015 			 * collection, and the neighbor is an active default
1016 			 * router, do not delete it.  Instead, reset the GC
1017 			 * timer using the router's lifetime.
1018 			 * Simply deleting the entry would affect default
1019 			 * router selection, which is not necessarily a good
1020 			 * thing, especially when we're using router preference
1021 			 * values.
1022 			 * XXX: the check for ln_state would be redundant,
1023 			 *      but we intentionally keep it just in case.
1024 			 */
1025 			if (dr->expire > time_second)
1026 				nd6_llinfo_settimer(ln,
1027 				    (dr->expire - time_second) * hz);
1028 			else
1029 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1030 			splx(s);
1031 			return (ln->ln_next);
1032 		}
1033 
1034 		if (ln->ln_router || dr) {
1035 			/*
1036 			 * rt6_flush must be called whether or not the neighbor
1037 			 * is in the Default Router List.
1038 			 * See a corresponding comment in nd6_na_input().
1039 			 */
1040 			rt6_flush(&in6, rt->rt_ifp);
1041 		}
1042 
1043 		if (dr) {
1044 			/*
1045 			 * Unreachablity of a router might affect the default
1046 			 * router selection and on-link detection of advertised
1047 			 * prefixes.
1048 			 */
1049 
1050 			/*
1051 			 * Temporarily fake the state to choose a new default
1052 			 * router and to perform on-link determination of
1053 			 * prefixes correctly.
1054 			 * Below the state will be set correctly,
1055 			 * or the entry itself will be deleted.
1056 			 */
1057 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1058 
1059 			/*
1060 			 * Since defrouter_select() does not affect the
1061 			 * on-link determination and MIP6 needs the check
1062 			 * before the default router selection, we perform
1063 			 * the check now.
1064 			 */
1065 			pfxlist_onlink_check();
1066 
1067 			/*
1068 			 * refresh default router list
1069 			 */
1070 			defrouter_select();
1071 		}
1072 		splx(s);
1073 	}
1074 
1075 	/*
1076 	 * Before deleting the entry, remember the next entry as the
1077 	 * return value.  We need this because pfxlist_onlink_check() above
1078 	 * might have freed other entries (particularly the old next entry) as
1079 	 * a side effect (XXX).
1080 	 */
1081 	next = ln->ln_next;
1082 
1083 	/*
1084 	 * Detach the route from the routing tree and the list of neighbor
1085 	 * caches, and disable the route entry not to be used in already
1086 	 * cached routes.
1087 	 */
1088 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1089 	    rt_mask(rt), 0, (struct rtentry **)0);
1090 
1091 	return (next);
1092 }
1093 
1094 /*
1095  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1096  *
1097  * XXX cost-effective methods?
1098  */
1099 void
1100 nd6_nud_hint(rt, dst6, force)
1101 	struct rtentry *rt;
1102 	struct in6_addr *dst6;
1103 	int force;
1104 {
1105 	struct llinfo_nd6 *ln;
1106 
1107 	/*
1108 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1109 	 * routing table by supplied "dst6".
1110 	 */
1111 	if (rt == NULL) {
1112 		if (dst6 == NULL)
1113 			return;
1114 		if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1115 			return;
1116 	}
1117 
1118 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1119 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1120 	    !rt->rt_llinfo || !rt->rt_gateway ||
1121 	    rt->rt_gateway->sa_family != AF_LINK) {
1122 		/* This is not a host route. */
1123 		return;
1124 	}
1125 
1126 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1127 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1128 		return;
1129 
1130 	/*
1131 	 * if we get upper-layer reachability confirmation many times,
1132 	 * it is possible we have false information.
1133 	 */
1134 	if (!force) {
1135 		ln->ln_byhint++;
1136 		if (ln->ln_byhint > nd6_maxnudhint)
1137 			return;
1138 	}
1139 
1140 	ln->ln_state = ND6_LLINFO_REACHABLE;
1141 	if (!ND6_LLINFO_PERMANENT(ln)) {
1142 		nd6_llinfo_settimer(ln,
1143 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1144 	}
1145 }
1146 
1147 void
1148 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info __unused)
1149 {
1150 	struct sockaddr *gate = rt->rt_gateway;
1151 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1152 	static const struct sockaddr_dl null_sdl = {
1153 		.sdl_len = sizeof(null_sdl),
1154 		.sdl_family = AF_LINK,
1155 	};
1156 	struct ifnet *ifp = rt->rt_ifp;
1157 	struct ifaddr *ifa;
1158 
1159 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1160 		return;
1161 
1162 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1163 		/*
1164 		 * This is probably an interface direct route for a link
1165 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1166 		 * We do not need special treatment below for such a route.
1167 		 * Moreover, the RTF_LLINFO flag which would be set below
1168 		 * would annoy the ndp(8) command.
1169 		 */
1170 		return;
1171 	}
1172 
1173 	if (req == RTM_RESOLVE &&
1174 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1175 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1176 		/*
1177 		 * FreeBSD and BSD/OS often make a cloned host route based
1178 		 * on a less-specific route (e.g. the default route).
1179 		 * If the less specific route does not have a "gateway"
1180 		 * (this is the case when the route just goes to a p2p or an
1181 		 * stf interface), we'll mistakenly make a neighbor cache for
1182 		 * the host route, and will see strange neighbor solicitation
1183 		 * for the corresponding destination.  In order to avoid the
1184 		 * confusion, we check if the destination of the route is
1185 		 * a neighbor in terms of neighbor discovery, and stop the
1186 		 * process if not.  Additionally, we remove the LLINFO flag
1187 		 * so that ndp(8) will not try to get the neighbor information
1188 		 * of the destination.
1189 		 */
1190 		rt->rt_flags &= ~RTF_LLINFO;
1191 		return;
1192 	}
1193 
1194 	switch (req) {
1195 	case RTM_ADD:
1196 		/*
1197 		 * There is no backward compatibility :)
1198 		 *
1199 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1200 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1201 		 *	   rt->rt_flags |= RTF_CLONING;
1202 		 */
1203 		if ((rt->rt_flags & RTF_CLONING) ||
1204 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1205 			/*
1206 			 * Case 1: This route should come from a route to
1207 			 * interface (RTF_CLONING case) or the route should be
1208 			 * treated as on-link but is currently not
1209 			 * (RTF_LLINFO && !ln case).
1210 			 */
1211 			rt_setgate(rt, rt_key(rt),
1212 				   (const struct sockaddr *)&null_sdl);
1213 			gate = rt->rt_gateway;
1214 			SDL(gate)->sdl_type = ifp->if_type;
1215 			SDL(gate)->sdl_index = ifp->if_index;
1216 			if (ln)
1217 				nd6_llinfo_settimer(ln, 0);
1218 			if ((rt->rt_flags & RTF_CLONING) != 0)
1219 				break;
1220 		}
1221 		/*
1222 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1223 		 * We don't do that here since llinfo is not ready yet.
1224 		 *
1225 		 * There are also couple of other things to be discussed:
1226 		 * - unsolicited NA code needs improvement beforehand
1227 		 * - RFC2461 says we MAY send multicast unsolicited NA
1228 		 *   (7.2.6 paragraph 4), however, it also says that we
1229 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1230 		 *   we don't have anything like it right now.
1231 		 *   note that the mechanism needs a mutual agreement
1232 		 *   between proxies, which means that we need to implement
1233 		 *   a new protocol, or a new kludge.
1234 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1235 		 *   we need to check ip6forwarding before sending it.
1236 		 *   (or should we allow proxy ND configuration only for
1237 		 *   routers?  there's no mention about proxy ND from hosts)
1238 		 */
1239 #if 0
1240 		/* XXX it does not work */
1241 		if (rt->rt_flags & RTF_ANNOUNCE)
1242 			nd6_na_output(ifp,
1243 			      &SIN6(rt_key(rt))->sin6_addr,
1244 			      &SIN6(rt_key(rt))->sin6_addr,
1245 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1246 			      1, NULL);
1247 #endif
1248 		/* FALLTHROUGH */
1249 	case RTM_RESOLVE:
1250 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1251 			/*
1252 			 * Address resolution isn't necessary for a point to
1253 			 * point link, so we can skip this test for a p2p link.
1254 			 */
1255 			if (gate->sa_family != AF_LINK ||
1256 			    gate->sa_len < sizeof(null_sdl)) {
1257 				log(LOG_DEBUG,
1258 				    "nd6_rtrequest: bad gateway value: %s\n",
1259 				    if_name(ifp));
1260 				break;
1261 			}
1262 			SDL(gate)->sdl_type = ifp->if_type;
1263 			SDL(gate)->sdl_index = ifp->if_index;
1264 		}
1265 		if (ln != NULL)
1266 			break;	/* This happens on a route change */
1267 		/*
1268 		 * Case 2: This route may come from cloning, or a manual route
1269 		 * add with a LL address.
1270 		 */
1271 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1272 		rt->rt_llinfo = (caddr_t)ln;
1273 		if (ln == NULL) {
1274 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1275 			break;
1276 		}
1277 		nd6_inuse++;
1278 		nd6_allocated++;
1279 		bzero(ln, sizeof(*ln));
1280 		ln->ln_rt = rt;
1281 		callout_init(&ln->ln_timer_ch);
1282 		/* this is required for "ndp" command. - shin */
1283 		if (req == RTM_ADD) {
1284 		        /*
1285 			 * gate should have some valid AF_LINK entry,
1286 			 * and ln->ln_expire should have some lifetime
1287 			 * which is specified by ndp command.
1288 			 */
1289 			ln->ln_state = ND6_LLINFO_REACHABLE;
1290 			ln->ln_byhint = 0;
1291 		} else {
1292 		        /*
1293 			 * When req == RTM_RESOLVE, rt is created and
1294 			 * initialized in rtrequest(), so rt_expire is 0.
1295 			 */
1296 			ln->ln_state = ND6_LLINFO_NOSTATE;
1297 			nd6_llinfo_settimer(ln, 0);
1298 		}
1299 		rt->rt_flags |= RTF_LLINFO;
1300 		ln->ln_next = llinfo_nd6.ln_next;
1301 		llinfo_nd6.ln_next = ln;
1302 		ln->ln_prev = &llinfo_nd6;
1303 		ln->ln_next->ln_prev = ln;
1304 
1305 		/*
1306 		 * check if rt_key(rt) is one of my address assigned
1307 		 * to the interface.
1308 		 */
1309 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1310 		    &SIN6(rt_key(rt))->sin6_addr);
1311 		if (ifa) {
1312 			caddr_t macp = nd6_ifptomac(ifp);
1313 			nd6_llinfo_settimer(ln, -1);
1314 			ln->ln_state = ND6_LLINFO_REACHABLE;
1315 			ln->ln_byhint = 0;
1316 			if (macp) {
1317 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1318 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1319 			}
1320 			if (nd6_useloopback) {
1321 				rt->rt_ifp = lo0ifp;	/* XXX */
1322 				/*
1323 				 * Make sure rt_ifa be equal to the ifaddr
1324 				 * corresponding to the address.
1325 				 * We need this because when we refer
1326 				 * rt_ifa->ia6_flags in ip6_input, we assume
1327 				 * that the rt_ifa points to the address instead
1328 				 * of the loopback address.
1329 				 */
1330 				if (ifa != rt->rt_ifa) {
1331 					IFAFREE(rt->rt_ifa);
1332 					IFAREF(ifa);
1333 					rt->rt_ifa = ifa;
1334 				}
1335 			}
1336 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1337 			nd6_llinfo_settimer(ln, -1);
1338 			ln->ln_state = ND6_LLINFO_REACHABLE;
1339 			ln->ln_byhint = 0;
1340 
1341 			/* join solicited node multicast for proxy ND */
1342 			if (ifp->if_flags & IFF_MULTICAST) {
1343 				struct in6_addr llsol;
1344 				int error;
1345 
1346 				llsol = SIN6(rt_key(rt))->sin6_addr;
1347 				llsol.s6_addr32[0] = htonl(0xff020000);
1348 				llsol.s6_addr32[1] = 0;
1349 				llsol.s6_addr32[2] = htonl(1);
1350 				llsol.s6_addr8[12] = 0xff;
1351 				if (in6_setscope(&llsol, ifp, NULL))
1352 					break;
1353 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1354 					nd6log((LOG_ERR, "%s: failed to join "
1355 					    "%s (errno=%d)\n", if_name(ifp),
1356 					    ip6_sprintf(&llsol), error));
1357 				}
1358 			}
1359 		}
1360 		break;
1361 
1362 	case RTM_DELETE:
1363 		if (ln == NULL)
1364 			break;
1365 		/* leave from solicited node multicast for proxy ND */
1366 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1367 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1368 			struct in6_addr llsol;
1369 			struct in6_multi *in6m;
1370 
1371 			llsol = SIN6(rt_key(rt))->sin6_addr;
1372 			llsol.s6_addr32[0] = htonl(0xff020000);
1373 			llsol.s6_addr32[1] = 0;
1374 			llsol.s6_addr32[2] = htonl(1);
1375 			llsol.s6_addr8[12] = 0xff;
1376 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1377 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1378 				if (in6m)
1379 					in6_delmulti(in6m);
1380 			}
1381 		}
1382 		nd6_inuse--;
1383 		ln->ln_next->ln_prev = ln->ln_prev;
1384 		ln->ln_prev->ln_next = ln->ln_next;
1385 		ln->ln_prev = NULL;
1386 		nd6_llinfo_settimer(ln, -1);
1387 		rt->rt_llinfo = 0;
1388 		rt->rt_flags &= ~RTF_LLINFO;
1389 		clear_llinfo_pqueue(ln);
1390 		Free((caddr_t)ln);
1391 	}
1392 }
1393 
1394 int
1395 nd6_ioctl(cmd, data, ifp)
1396 	u_long cmd;
1397 	caddr_t	data;
1398 	struct ifnet *ifp;
1399 {
1400 	struct in6_drlist *drl = (struct in6_drlist *)data;
1401 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1402 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1403 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1404 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1405 	struct nd_defrouter *dr;
1406 	struct nd_prefix *pr;
1407 	struct rtentry *rt;
1408 	int i = 0, error = 0;
1409 	int s;
1410 
1411 	switch (cmd) {
1412 	case SIOCGDRLST_IN6:
1413 		/*
1414 		 * obsolete API, use sysctl under net.inet6.icmp6
1415 		 */
1416 		bzero(drl, sizeof(*drl));
1417 		s = splsoftnet();
1418 		dr = TAILQ_FIRST(&nd_defrouter);
1419 		while (dr && i < DRLSTSIZ) {
1420 			drl->defrouter[i].rtaddr = dr->rtaddr;
1421 			in6_clearscope(&drl->defrouter[i].rtaddr);
1422 
1423 			drl->defrouter[i].flags = dr->flags;
1424 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1425 			drl->defrouter[i].expire = dr->expire;
1426 			drl->defrouter[i].if_index = dr->ifp->if_index;
1427 			i++;
1428 			dr = TAILQ_NEXT(dr, dr_entry);
1429 		}
1430 		splx(s);
1431 		break;
1432 	case SIOCGPRLST_IN6:
1433 		/*
1434 		 * obsolete API, use sysctl under net.inet6.icmp6
1435 		 *
1436 		 * XXX the structure in6_prlist was changed in backward-
1437 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1438 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1439 		 */
1440 		/*
1441 		 * XXX meaning of fields, especialy "raflags", is very
1442 		 * differnet between RA prefix list and RR/static prefix list.
1443 		 * how about separating ioctls into two?
1444 		 */
1445 		bzero(oprl, sizeof(*oprl));
1446 		s = splsoftnet();
1447 		pr = nd_prefix.lh_first;
1448 		while (pr && i < PRLSTSIZ) {
1449 			struct nd_pfxrouter *pfr;
1450 			int j;
1451 
1452 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1453 			oprl->prefix[i].raflags = pr->ndpr_raf;
1454 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1455 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1456 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1457 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1458 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1459 				oprl->prefix[i].expire = 0;
1460 			else {
1461 				time_t maxexpire;
1462 
1463 				/* XXX: we assume time_t is signed. */
1464 				maxexpire = (-1) &
1465 				    ~((time_t)1 <<
1466 				    ((sizeof(maxexpire) * 8) - 1));
1467 				if (pr->ndpr_vltime <
1468 				    maxexpire - pr->ndpr_lastupdate) {
1469 					oprl->prefix[i].expire =
1470 						 pr->ndpr_lastupdate +
1471 						pr->ndpr_vltime;
1472 				} else
1473 					oprl->prefix[i].expire = maxexpire;
1474 			}
1475 
1476 			pfr = pr->ndpr_advrtrs.lh_first;
1477 			j = 0;
1478 			while (pfr) {
1479 				if (j < DRLSTSIZ) {
1480 #define RTRADDR oprl->prefix[i].advrtr[j]
1481 					RTRADDR = pfr->router->rtaddr;
1482 					in6_clearscope(&RTRADDR);
1483 #undef RTRADDR
1484 				}
1485 				j++;
1486 				pfr = pfr->pfr_next;
1487 			}
1488 			oprl->prefix[i].advrtrs = j;
1489 			oprl->prefix[i].origin = PR_ORIG_RA;
1490 
1491 			i++;
1492 			pr = pr->ndpr_next;
1493 		}
1494 		splx(s);
1495 
1496 		break;
1497 	case OSIOCGIFINFO_IN6:
1498 #define ND	ndi->ndi
1499 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1500 		memset(&ND, 0, sizeof(ND));
1501 		ND.linkmtu = IN6_LINKMTU(ifp);
1502 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1503 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1504 		ND.reachable = ND_IFINFO(ifp)->reachable;
1505 		ND.retrans = ND_IFINFO(ifp)->retrans;
1506 		ND.flags = ND_IFINFO(ifp)->flags;
1507 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1508 		ND.chlim = ND_IFINFO(ifp)->chlim;
1509 		break;
1510 	case SIOCGIFINFO_IN6:
1511 		ND = *ND_IFINFO(ifp);
1512 		break;
1513 	case SIOCSIFINFO_IN6:
1514 		/*
1515 		 * used to change host variables from userland.
1516 		 * intented for a use on router to reflect RA configurations.
1517 		 */
1518 		/* 0 means 'unspecified' */
1519 		if (ND.linkmtu != 0) {
1520 			if (ND.linkmtu < IPV6_MMTU ||
1521 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1522 				error = EINVAL;
1523 				break;
1524 			}
1525 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1526 		}
1527 
1528 		if (ND.basereachable != 0) {
1529 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1530 
1531 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1532 			if (ND.basereachable != obasereachable)
1533 				ND_IFINFO(ifp)->reachable =
1534 				    ND_COMPUTE_RTIME(ND.basereachable);
1535 		}
1536 		if (ND.retrans != 0)
1537 			ND_IFINFO(ifp)->retrans = ND.retrans;
1538 		if (ND.chlim != 0)
1539 			ND_IFINFO(ifp)->chlim = ND.chlim;
1540 		/* FALLTHROUGH */
1541 	case SIOCSIFINFO_FLAGS:
1542 		ND_IFINFO(ifp)->flags = ND.flags;
1543 		break;
1544 #undef ND
1545 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1546 		/* sync kernel routing table with the default router list */
1547 		defrouter_reset();
1548 		defrouter_select();
1549 		break;
1550 	case SIOCSPFXFLUSH_IN6:
1551 	{
1552 		/* flush all the prefix advertised by routers */
1553 		struct nd_prefix *pfx, *next;
1554 
1555 		s = splsoftnet();
1556 		for (pfx = nd_prefix.lh_first; pfx; pfx = next) {
1557 			struct in6_ifaddr *ia, *ia_next;
1558 
1559 			next = pfx->ndpr_next;
1560 
1561 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1562 				continue; /* XXX */
1563 
1564 			/* do we really have to remove addresses as well? */
1565 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1566 				/* ia might be removed.  keep the next ptr. */
1567 				ia_next = ia->ia_next;
1568 
1569 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1570 					continue;
1571 
1572 				if (ia->ia6_ndpr == pfx)
1573 					in6_purgeaddr(&ia->ia_ifa);
1574 			}
1575 			prelist_remove(pfx);
1576 		}
1577 		splx(s);
1578 		break;
1579 	}
1580 	case SIOCSRTRFLUSH_IN6:
1581 	{
1582 		/* flush all the default routers */
1583 		struct nd_defrouter *drtr, *next;
1584 
1585 		s = splsoftnet();
1586 		defrouter_reset();
1587 		for (drtr = TAILQ_FIRST(&nd_defrouter); drtr; drtr = next) {
1588 			next = TAILQ_NEXT(drtr, dr_entry);
1589 			defrtrlist_del(drtr);
1590 		}
1591 		defrouter_select();
1592 		splx(s);
1593 		break;
1594 	}
1595 	case SIOCGNBRINFO_IN6:
1596 	{
1597 		struct llinfo_nd6 *ln;
1598 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1599 
1600 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1601 			return (error);
1602 
1603 		s = splsoftnet();
1604 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1605 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1606 			error = EINVAL;
1607 			splx(s);
1608 			break;
1609 		}
1610 		nbi->state = ln->ln_state;
1611 		nbi->asked = ln->ln_asked;
1612 		nbi->isrouter = ln->ln_router;
1613 		nbi->expire = ln->ln_expire;
1614 		splx(s);
1615 
1616 		break;
1617 	}
1618 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1619 		ndif->ifindex = nd6_defifindex;
1620 		break;
1621 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1622 		return (nd6_setdefaultiface(ndif->ifindex));
1623 	}
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Create neighbor cache entry and cache link-layer address,
1629  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1630  */
1631 struct rtentry *
1632 nd6_cache_lladdr(
1633     struct ifnet *ifp,
1634     struct in6_addr *from,
1635     char *lladdr,
1636     int lladdrlen __unused,
1637     int type,	/* ICMP6 type */
1638     int code	/* type dependent information */
1639 )
1640 {
1641 	struct rtentry *rt = NULL;
1642 	struct llinfo_nd6 *ln = NULL;
1643 	int is_newentry;
1644 	struct sockaddr_dl *sdl = NULL;
1645 	int do_update;
1646 	int olladdr;
1647 	int llchange;
1648 	int newstate = 0;
1649 
1650 	if (ifp == NULL)
1651 		panic("ifp == NULL in nd6_cache_lladdr");
1652 	if (from == NULL)
1653 		panic("from == NULL in nd6_cache_lladdr");
1654 
1655 	/* nothing must be updated for unspecified address */
1656 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1657 		return NULL;
1658 
1659 	/*
1660 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1661 	 * the caller.
1662 	 *
1663 	 * XXX If the link does not have link-layer adderss, what should
1664 	 * we do? (ifp->if_addrlen == 0)
1665 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1666 	 * description on it in NS section (RFC 2461 7.2.3).
1667 	 */
1668 
1669 	rt = nd6_lookup(from, 0, ifp);
1670 	if (rt == NULL) {
1671 #if 0
1672 		/* nothing must be done if there's no lladdr */
1673 		if (!lladdr || !lladdrlen)
1674 			return NULL;
1675 #endif
1676 
1677 		rt = nd6_lookup(from, 1, ifp);
1678 		is_newentry = 1;
1679 	} else {
1680 		/* do nothing if static ndp is set */
1681 		if (rt->rt_flags & RTF_STATIC)
1682 			return NULL;
1683 		is_newentry = 0;
1684 	}
1685 
1686 	if (rt == NULL)
1687 		return NULL;
1688 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1689 fail:
1690 		(void)nd6_free(rt, 0);
1691 		return NULL;
1692 	}
1693 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1694 	if (ln == NULL)
1695 		goto fail;
1696 	if (rt->rt_gateway == NULL)
1697 		goto fail;
1698 	if (rt->rt_gateway->sa_family != AF_LINK)
1699 		goto fail;
1700 	sdl = SDL(rt->rt_gateway);
1701 
1702 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1703 	if (olladdr && lladdr) {
1704 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1705 			llchange = 1;
1706 		else
1707 			llchange = 0;
1708 	} else
1709 		llchange = 0;
1710 
1711 	/*
1712 	 * newentry olladdr  lladdr  llchange	(*=record)
1713 	 *	0	n	n	--	(1)
1714 	 *	0	y	n	--	(2)
1715 	 *	0	n	y	--	(3) * STALE
1716 	 *	0	y	y	n	(4) *
1717 	 *	0	y	y	y	(5) * STALE
1718 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1719 	 *	1	--	y	--	(7) * STALE
1720 	 */
1721 
1722 	if (lladdr) {		/* (3-5) and (7) */
1723 		/*
1724 		 * Record source link-layer address
1725 		 * XXX is it dependent to ifp->if_type?
1726 		 */
1727 		sdl->sdl_alen = ifp->if_addrlen;
1728 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1729 	}
1730 
1731 	if (!is_newentry) {
1732 		if ((!olladdr && lladdr) ||		/* (3) */
1733 		    (olladdr && lladdr && llchange)) {	/* (5) */
1734 			do_update = 1;
1735 			newstate = ND6_LLINFO_STALE;
1736 		} else					/* (1-2,4) */
1737 			do_update = 0;
1738 	} else {
1739 		do_update = 1;
1740 		if (lladdr == NULL)			/* (6) */
1741 			newstate = ND6_LLINFO_NOSTATE;
1742 		else					/* (7) */
1743 			newstate = ND6_LLINFO_STALE;
1744 	}
1745 
1746 	if (do_update) {
1747 		/*
1748 		 * Update the state of the neighbor cache.
1749 		 */
1750 		ln->ln_state = newstate;
1751 
1752 		if (ln->ln_state == ND6_LLINFO_STALE) {
1753 			/*
1754 			 * XXX: since nd6_output() below will cause
1755 			 * state tansition to DELAY and reset the timer,
1756 			 * we must set the timer now, although it is actually
1757 			 * meaningless.
1758 			 */
1759 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1760 
1761  			if (ln->ln_hold) {
1762 				struct mbuf *m_hold, *m_hold_next;
1763 				for (m_hold = ln->ln_hold; m_hold;
1764 				     m_hold = m_hold_next) {
1765 					struct mbuf *mpkt = NULL;
1766 
1767 					m_hold_next = m_hold->m_nextpkt;
1768 					mpkt = m_copym(m_hold, 0, M_COPYALL, M_DONTWAIT);
1769 					if (mpkt == NULL) {
1770 						m_freem(m_hold);
1771 						break;
1772 					}
1773 					mpkt->m_nextpkt = NULL;
1774 
1775 					/*
1776 					 * we assume ifp is not a p2p here, so
1777 					 * just set the 2nd argument as the
1778 					 * 1st one.
1779 					 */
1780 					nd6_output(ifp, ifp, mpkt,
1781 					     (struct sockaddr_in6 *)rt_key(rt),
1782 					     rt);
1783 				}
1784  				ln->ln_hold = NULL;
1785  			}
1786 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1787 			/* probe right away */
1788 			nd6_llinfo_settimer((void *)ln, 0);
1789 		}
1790 	}
1791 
1792 	/*
1793 	 * ICMP6 type dependent behavior.
1794 	 *
1795 	 * NS: clear IsRouter if new entry
1796 	 * RS: clear IsRouter
1797 	 * RA: set IsRouter if there's lladdr
1798 	 * redir: clear IsRouter if new entry
1799 	 *
1800 	 * RA case, (1):
1801 	 * The spec says that we must set IsRouter in the following cases:
1802 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1803 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1804 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1805 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1806 	 * neighbor cache, this is similar to (6).
1807 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1808 	 *
1809 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1810 	 *							D R
1811 	 *	0	n	n	--	(1)	c   ?     s
1812 	 *	0	y	n	--	(2)	c   s     s
1813 	 *	0	n	y	--	(3)	c   s     s
1814 	 *	0	y	y	n	(4)	c   s     s
1815 	 *	0	y	y	y	(5)	c   s     s
1816 	 *	1	--	n	--	(6) c	c 	c s
1817 	 *	1	--	y	--	(7) c	c   s	c s
1818 	 *
1819 	 *					(c=clear s=set)
1820 	 */
1821 	switch (type & 0xff) {
1822 	case ND_NEIGHBOR_SOLICIT:
1823 		/*
1824 		 * New entry must have is_router flag cleared.
1825 		 */
1826 		if (is_newentry)	/* (6-7) */
1827 			ln->ln_router = 0;
1828 		break;
1829 	case ND_REDIRECT:
1830 		/*
1831 		 * If the icmp is a redirect to a better router, always set the
1832 		 * is_router flag.  Otherwise, if the entry is newly created,
1833 		 * clear the flag.  [RFC 2461, sec 8.3]
1834 		 */
1835 		if (code == ND_REDIRECT_ROUTER)
1836 			ln->ln_router = 1;
1837 		else if (is_newentry) /* (6-7) */
1838 			ln->ln_router = 0;
1839 		break;
1840 	case ND_ROUTER_SOLICIT:
1841 		/*
1842 		 * is_router flag must always be cleared.
1843 		 */
1844 		ln->ln_router = 0;
1845 		break;
1846 	case ND_ROUTER_ADVERT:
1847 		/*
1848 		 * Mark an entry with lladdr as a router.
1849 		 */
1850 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1851 		    (is_newentry && lladdr)) {			/* (7) */
1852 			ln->ln_router = 1;
1853 		}
1854 		break;
1855 	}
1856 
1857 	/*
1858 	 * When the link-layer address of a router changes, select the
1859 	 * best router again.  In particular, when the neighbor entry is newly
1860 	 * created, it might affect the selection policy.
1861 	 * Question: can we restrict the first condition to the "is_newentry"
1862 	 * case?
1863 	 * XXX: when we hear an RA from a new router with the link-layer
1864 	 * address option, defrouter_select() is called twice, since
1865 	 * defrtrlist_update called the function as well.  However, I believe
1866 	 * we can compromise the overhead, since it only happens the first
1867 	 * time.
1868 	 * XXX: although defrouter_select() should not have a bad effect
1869 	 * for those are not autoconfigured hosts, we explicitly avoid such
1870 	 * cases for safety.
1871 	 */
1872 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1873 		defrouter_select();
1874 
1875 	return rt;
1876 }
1877 
1878 static void
1879 nd6_slowtimo(void *ignored_arg __unused)
1880 {
1881 	int s = splsoftnet();
1882 	struct nd_ifinfo *nd6if;
1883 	struct ifnet *ifp;
1884 
1885 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1886 	    nd6_slowtimo, NULL);
1887 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1888 	{
1889 		nd6if = ND_IFINFO(ifp);
1890 		if (nd6if->basereachable && /* already initialized */
1891 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1892 			/*
1893 			 * Since reachable time rarely changes by router
1894 			 * advertisements, we SHOULD insure that a new random
1895 			 * value gets recomputed at least once every few hours.
1896 			 * (RFC 2461, 6.3.4)
1897 			 */
1898 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1899 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1900 		}
1901 	}
1902 	splx(s);
1903 }
1904 
1905 #define senderr(e) { error = (e); goto bad;}
1906 int
1907 nd6_output(ifp, origifp, m0, dst, rt0)
1908 	struct ifnet *ifp;
1909 	struct ifnet *origifp;
1910 	struct mbuf *m0;
1911 	struct sockaddr_in6 *dst;
1912 	struct rtentry *rt0;
1913 {
1914 	struct mbuf *m = m0;
1915 	struct rtentry *rt = rt0;
1916 	struct sockaddr_in6 *gw6 = NULL;
1917 	struct llinfo_nd6 *ln = NULL;
1918 	int error = 0;
1919 
1920 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1921 		goto sendpkt;
1922 
1923 	if (nd6_need_cache(ifp) == 0)
1924 		goto sendpkt;
1925 
1926 	/*
1927 	 * next hop determination.  This routine is derived from ether_output.
1928 	 */
1929 	if (rt) {
1930 		if ((rt->rt_flags & RTF_UP) == 0) {
1931 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1932 			    1)) != NULL)
1933 			{
1934 				rt->rt_refcnt--;
1935 				if (rt->rt_ifp != ifp)
1936 					senderr(EHOSTUNREACH);
1937 			} else
1938 				senderr(EHOSTUNREACH);
1939 		}
1940 
1941 		if (rt->rt_flags & RTF_GATEWAY) {
1942 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1943 
1944 			/*
1945 			 * We skip link-layer address resolution and NUD
1946 			 * if the gateway is not a neighbor from ND point
1947 			 * of view, regardless of the value of nd_ifinfo.flags.
1948 			 * The second condition is a bit tricky; we skip
1949 			 * if the gateway is our own address, which is
1950 			 * sometimes used to install a route to a p2p link.
1951 			 */
1952 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1953 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1954 				/*
1955 				 * We allow this kind of tricky route only
1956 				 * when the outgoing interface is p2p.
1957 				 * XXX: we may need a more generic rule here.
1958 				 */
1959 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1960 					senderr(EHOSTUNREACH);
1961 
1962 				goto sendpkt;
1963 			}
1964 
1965 			if (rt->rt_gwroute == 0)
1966 				goto lookup;
1967 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1968 				rtfree(rt); rt = rt0;
1969 			lookup:
1970 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1971 				if ((rt = rt->rt_gwroute) == 0)
1972 					senderr(EHOSTUNREACH);
1973 				/* the "G" test below also prevents rt == rt0 */
1974 				if ((rt->rt_flags & RTF_GATEWAY) ||
1975 				    (rt->rt_ifp != ifp)) {
1976 					rt->rt_refcnt--;
1977 					rt0->rt_gwroute = 0;
1978 					senderr(EHOSTUNREACH);
1979 				}
1980 			}
1981 		}
1982 	}
1983 
1984 	/*
1985 	 * Address resolution or Neighbor Unreachability Detection
1986 	 * for the next hop.
1987 	 * At this point, the destination of the packet must be a unicast
1988 	 * or an anycast address(i.e. not a multicast).
1989 	 */
1990 
1991 	/* Look up the neighbor cache for the nexthop */
1992 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1993 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1994 	else {
1995 		/*
1996 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1997 		 * the condition below is not very efficient.  But we believe
1998 		 * it is tolerable, because this should be a rare case.
1999 		 */
2000 		if (nd6_is_addr_neighbor(dst, ifp) &&
2001 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
2002 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2003 	}
2004 	if (ln == NULL || rt == NULL) {
2005 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2006 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2007 			log(LOG_DEBUG,
2008 			    "nd6_output: can't allocate llinfo for %s "
2009 			    "(ln=%p, rt=%p)\n",
2010 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2011 			senderr(EIO);	/* XXX: good error? */
2012 		}
2013 
2014 		goto sendpkt;	/* send anyway */
2015 	}
2016 
2017 	/* We don't have to do link-layer address resolution on a p2p link. */
2018 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2019 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2020 		ln->ln_state = ND6_LLINFO_STALE;
2021 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2022 	}
2023 
2024 	/*
2025 	 * The first time we send a packet to a neighbor whose entry is
2026 	 * STALE, we have to change the state to DELAY and a sets a timer to
2027 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2028 	 * neighbor unreachability detection on expiration.
2029 	 * (RFC 2461 7.3.3)
2030 	 */
2031 	if (ln->ln_state == ND6_LLINFO_STALE) {
2032 		ln->ln_asked = 0;
2033 		ln->ln_state = ND6_LLINFO_DELAY;
2034 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2035 	}
2036 
2037 	/*
2038 	 * If the neighbor cache entry has a state other than INCOMPLETE
2039 	 * (i.e. its link-layer address is already resolved), just
2040 	 * send the packet.
2041 	 */
2042 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2043 		goto sendpkt;
2044 
2045 	/*
2046 	 * There is a neighbor cache entry, but no ethernet address
2047 	 * response yet.  Append this latest packet to the end of the
2048 	 * packet queue in the mbuf, unless the number of the packet
2049 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2050 	 * the oldest packet in the queue will be removed.
2051 	 */
2052 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2053 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2054 	if (ln->ln_hold) {
2055 		struct mbuf *m_hold;
2056 		int i;
2057 
2058 		i = 0;
2059 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2060 			i++;
2061 			if (m_hold->m_nextpkt == NULL) {
2062 				m_hold->m_nextpkt = m;
2063 				break;
2064 			}
2065 		}
2066 		while (i >= nd6_maxqueuelen) {
2067 			m_hold = ln->ln_hold;
2068 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2069 			m_freem(m_hold);
2070 			i--;
2071 		}
2072 	} else {
2073 		ln->ln_hold = m;
2074 	}
2075 
2076 	/*
2077 	 * If there has been no NS for the neighbor after entering the
2078 	 * INCOMPLETE state, send the first solicitation.
2079 	 */
2080 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2081 		ln->ln_asked++;
2082 		nd6_llinfo_settimer(ln,
2083 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2084 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2085 	}
2086 	return (0);
2087 
2088   sendpkt:
2089 	/* discard the packet if IPv6 operation is disabled on the interface */
2090 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2091 		error = ENETDOWN; /* better error? */
2092 		goto bad;
2093 	}
2094 
2095 #ifdef IPSEC
2096 	/* clean ipsec history once it goes out of the node */
2097 	ipsec_delaux(m);
2098 #endif
2099 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2100 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2101 					 rt));
2102 	}
2103 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2104 
2105   bad:
2106 	if (m)
2107 		m_freem(m);
2108 	return (error);
2109 }
2110 #undef senderr
2111 
2112 int
2113 nd6_need_cache(ifp)
2114 	struct ifnet *ifp;
2115 {
2116 	/*
2117 	 * XXX: we currently do not make neighbor cache on any interface
2118 	 * other than ARCnet, Ethernet, FDDI and GIF.
2119 	 *
2120 	 * RFC2893 says:
2121 	 * - unidirectional tunnels needs no ND
2122 	 */
2123 	switch (ifp->if_type) {
2124 	case IFT_ARCNET:
2125 	case IFT_ETHER:
2126 	case IFT_FDDI:
2127 	case IFT_IEEE1394:
2128 	case IFT_CARP:
2129 	case IFT_GIF:		/* XXX need more cases? */
2130 	case IFT_PPP:
2131 	case IFT_TUNNEL:
2132 		return (1);
2133 	default:
2134 		return (0);
2135 	}
2136 }
2137 
2138 int
2139 nd6_storelladdr(ifp, rt, m, dst, desten)
2140 	struct ifnet *ifp;
2141 	struct rtentry *rt;
2142 	struct mbuf *m;
2143 	struct sockaddr *dst;
2144 	u_char *desten;
2145 {
2146 	struct sockaddr_dl *sdl;
2147 
2148 	if (m->m_flags & M_MCAST) {
2149 		switch (ifp->if_type) {
2150 		case IFT_ETHER:
2151 		case IFT_FDDI:
2152 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2153 						 desten);
2154 			return (1);
2155 		case IFT_IEEE1394:
2156 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2157 			return (1);
2158 		case IFT_ARCNET:
2159 			*desten = 0;
2160 			return (1);
2161 		default:
2162 			m_freem(m);
2163 			return (0);
2164 		}
2165 	}
2166 
2167 	if (rt == NULL) {
2168 		/* this could happen, if we could not allocate memory */
2169 		m_freem(m);
2170 		return (0);
2171 	}
2172 	if (rt->rt_gateway->sa_family != AF_LINK) {
2173 		printf("nd6_storelladdr: something odd happens\n");
2174 		m_freem(m);
2175 		return (0);
2176 	}
2177 	sdl = SDL(rt->rt_gateway);
2178 	if (sdl->sdl_alen == 0) {
2179 		/* this should be impossible, but we bark here for debugging */
2180 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
2181 		    ip6_sprintf(&SIN6(dst)->sin6_addr), if_name(ifp));
2182 		m_freem(m);
2183 		return (0);
2184 	}
2185 
2186 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2187 	return (1);
2188 }
2189 
2190 static void
2191 clear_llinfo_pqueue(ln)
2192 	struct llinfo_nd6 *ln;
2193 {
2194 	struct mbuf *m_hold, *m_hold_next;
2195 
2196 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2197 		m_hold_next = m_hold->m_nextpkt;
2198 		m_hold->m_nextpkt = NULL;
2199 		m_freem(m_hold);
2200 	}
2201 
2202 	ln->ln_hold = NULL;
2203 	return;
2204 }
2205 
2206 int
2207 nd6_sysctl(
2208     int name,
2209     void *oldp,	/* syscall arg, need copyout */
2210     size_t *oldlenp,
2211     void *newp,	/* syscall arg, need copyin */
2212     size_t newlen __unused
2213 )
2214 {
2215 	void *p;
2216 	size_t ol;
2217 	int error;
2218 
2219 	error = 0;
2220 
2221 	if (newp)
2222 		return EPERM;
2223 	if (oldp && !oldlenp)
2224 		return EINVAL;
2225 	ol = oldlenp ? *oldlenp : 0;
2226 
2227 	if (oldp) {
2228 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2229 		if (p == NULL)
2230 			return ENOMEM;
2231 	} else
2232 		p = NULL;
2233 	switch (name) {
2234 	case ICMPV6CTL_ND6_DRLIST:
2235 		error = fill_drlist(p, oldlenp, ol);
2236 		if (!error && p != NULL && oldp != NULL)
2237 			error = copyout(p, oldp, *oldlenp);
2238 		break;
2239 
2240 	case ICMPV6CTL_ND6_PRLIST:
2241 		error = fill_prlist(p, oldlenp, ol);
2242 		if (!error && p != NULL && oldp != NULL)
2243 			error = copyout(p, oldp, *oldlenp);
2244 		break;
2245 
2246 	case ICMPV6CTL_ND6_MAXQLEN:
2247 		break;
2248 
2249 	default:
2250 		error = ENOPROTOOPT;
2251 		break;
2252 	}
2253 	if (p)
2254 		free(p, M_TEMP);
2255 
2256 	return (error);
2257 }
2258 
2259 static int
2260 fill_drlist(oldp, oldlenp, ol)
2261 	void *oldp;
2262 	size_t *oldlenp, ol;
2263 {
2264 	int error = 0, s;
2265 	struct in6_defrouter *d = NULL, *de = NULL;
2266 	struct nd_defrouter *dr;
2267 	size_t l;
2268 
2269 	s = splsoftnet();
2270 
2271 	if (oldp) {
2272 		d = (struct in6_defrouter *)oldp;
2273 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2274 	}
2275 	l = 0;
2276 
2277 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2278 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2279 
2280 		if (oldp && d + 1 <= de) {
2281 			bzero(d, sizeof(*d));
2282 			d->rtaddr.sin6_family = AF_INET6;
2283 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2284 			d->rtaddr.sin6_addr = dr->rtaddr;
2285 			if (sa6_recoverscope(&d->rtaddr)) {
2286 				log(LOG_ERR,
2287 				    "scope error in router list (%s)\n",
2288 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2289 				/* XXX: press on... */
2290 			}
2291 			d->flags = dr->flags;
2292 			d->rtlifetime = dr->rtlifetime;
2293 			d->expire = dr->expire;
2294 			d->if_index = dr->ifp->if_index;
2295 		}
2296 
2297 		l += sizeof(*d);
2298 		if (d)
2299 			d++;
2300 	}
2301 
2302 	if (oldp) {
2303 		if (l > ol)
2304 			error = ENOMEM;
2305 	}
2306 	if (oldlenp)
2307 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2308 
2309 	splx(s);
2310 
2311 	return (error);
2312 }
2313 
2314 static int
2315 fill_prlist(oldp, oldlenp, ol)
2316 	void *oldp;
2317 	size_t *oldlenp, ol;
2318 {
2319 	int error = 0, s;
2320 	struct nd_prefix *pr;
2321 	struct in6_prefix *p = NULL;
2322 	struct in6_prefix *pe = NULL;
2323 	size_t l;
2324 
2325 	s = splsoftnet();
2326 
2327 	if (oldp) {
2328 		p = (struct in6_prefix *)oldp;
2329 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2330 	}
2331 	l = 0;
2332 
2333 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2334 		u_short advrtrs;
2335 		size_t advance;
2336 		struct sockaddr_in6 *sin6;
2337 		struct sockaddr_in6 *s6;
2338 		struct nd_pfxrouter *pfr;
2339 
2340 		if (oldp && p + 1 <= pe)
2341 		{
2342 			bzero(p, sizeof(*p));
2343 			sin6 = (struct sockaddr_in6 *)(p + 1);
2344 
2345 			p->prefix = pr->ndpr_prefix;
2346 			if (sa6_recoverscope(&p->prefix)) {
2347 				log(LOG_ERR,
2348 				    "scope error in prefix list (%s)\n",
2349 				    ip6_sprintf(&p->prefix.sin6_addr));
2350 				/* XXX: press on... */
2351 			}
2352 			p->raflags = pr->ndpr_raf;
2353 			p->prefixlen = pr->ndpr_plen;
2354 			p->vltime = pr->ndpr_vltime;
2355 			p->pltime = pr->ndpr_pltime;
2356 			p->if_index = pr->ndpr_ifp->if_index;
2357 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2358 				p->expire = 0;
2359 			else {
2360 				time_t maxexpire;
2361 
2362 				/* XXX: we assume time_t is signed. */
2363 				maxexpire = (-1) &
2364 				    ~((time_t)1 <<
2365 				    ((sizeof(maxexpire) * 8) - 1));
2366 				if (pr->ndpr_vltime <
2367 				    maxexpire - pr->ndpr_lastupdate) {
2368 					p->expire = pr->ndpr_lastupdate +
2369 						pr->ndpr_vltime;
2370 				} else
2371 					p->expire = maxexpire;
2372 			}
2373 			p->refcnt = pr->ndpr_refcnt;
2374 			p->flags = pr->ndpr_stateflags;
2375 			p->origin = PR_ORIG_RA;
2376 			advrtrs = 0;
2377 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2378 			     pfr = pfr->pfr_next) {
2379 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2380 					advrtrs++;
2381 					continue;
2382 				}
2383 				s6 = &sin6[advrtrs];
2384 				s6->sin6_family = AF_INET6;
2385 				s6->sin6_len = sizeof(struct sockaddr_in6);
2386 				s6->sin6_addr = pfr->router->rtaddr;
2387 				s6->sin6_scope_id = 0;
2388 				if (sa6_recoverscope(s6)) {
2389 					log(LOG_ERR,
2390 					    "scope error in "
2391 					    "prefix list (%s)\n",
2392 					    ip6_sprintf(&pfr->router->rtaddr));
2393 				}
2394 				advrtrs++;
2395 			}
2396 			p->advrtrs = advrtrs;
2397 		}
2398 		else {
2399 			advrtrs = 0;
2400 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2401 			     pfr = pfr->pfr_next)
2402 				advrtrs++;
2403 		}
2404 
2405 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2406 		l += advance;
2407 		if (p)
2408 			p = (struct in6_prefix *)((caddr_t)p + advance);
2409 	}
2410 
2411 	if (oldp) {
2412 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2413 		if (l > ol)
2414 			error = ENOMEM;
2415 	} else
2416 		*oldlenp = l;
2417 
2418 	splx(s);
2419 
2420 	return (error);
2421 }
2422