xref: /netbsd-src/sys/netinet6/nd6.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: nd6.c,v 1.85 2003/06/24 07:54:48 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 1.85 2003/06/24 07:54:48 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 
110 struct callout nd6_slowtimo_ch = CALLOUT_INITIALIZER;
111 struct callout nd6_timer_ch = CALLOUT_INITIALIZER;
112 
113 static int fill_drlist __P((void *, size_t *, size_t));
114 static int fill_prlist __P((void *, size_t *, size_t));
115 
116 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
117 
118 void
119 nd6_init()
120 {
121 	static int nd6_init_done = 0;
122 	int i;
123 
124 	if (nd6_init_done) {
125 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
126 		return;
127 	}
128 
129 	all1_sa.sin6_family = AF_INET6;
130 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
131 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
132 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
133 
134 	/* initialization of the default router list */
135 	TAILQ_INIT(&nd_defrouter);
136 
137 	nd6_init_done = 1;
138 
139 	/* start timer */
140 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
141 	    nd6_slowtimo, NULL);
142 }
143 
144 struct nd_ifinfo *
145 nd6_ifattach(ifp)
146 	struct ifnet *ifp;
147 {
148 	struct nd_ifinfo *nd;
149 
150 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
151 	bzero(nd, sizeof(*nd));
152 
153 	nd->initialized = 1;
154 
155 	nd->chlim = IPV6_DEFHLIM;
156 	nd->basereachable = REACHABLE_TIME;
157 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
158 	nd->retrans = RETRANS_TIMER;
159 	/*
160 	 * Note that the default value of ip6_accept_rtadv is 0, which means
161 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
162 	 * here.
163 	 */
164 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
165 
166 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
167 	nd6_setmtu0(ifp, nd);
168 
169 	return nd;
170 }
171 
172 void
173 nd6_ifdetach(nd)
174 	struct nd_ifinfo *nd;
175 {
176 
177 	free(nd, M_IP6NDP);
178 }
179 
180 void
181 nd6_setmtu(ifp)
182 	struct ifnet *ifp;
183 {
184 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
185 }
186 
187 void
188 nd6_setmtu0(ifp, ndi)
189 	struct ifnet *ifp;
190 	struct nd_ifinfo *ndi;
191 {
192 	u_int32_t omaxmtu;
193 
194 	omaxmtu = ndi->maxmtu;
195 
196 	switch (ifp->if_type) {
197 	case IFT_ARCNET:
198 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
199 		break;
200 	case IFT_FDDI:
201 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
202 		break;
203 	default:
204 		ndi->maxmtu = ifp->if_mtu;
205 		break;
206 	}
207 
208 	/*
209 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
210 	 * undesirable situation.  We thus notify the operator of the change
211 	 * explicitly.  The check for omaxmtu is necessary to restrict the
212 	 * log to the case of changing the MTU, not initializing it.
213 	 */
214 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
215 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
216 		    " small for IPv6 which needs %lu\n",
217 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
218 		    IPV6_MMTU);
219 	}
220 
221 	if (ndi->maxmtu > in6_maxmtu)
222 		in6_setmaxmtu(); /* check all interfaces just in case */
223 }
224 
225 void
226 nd6_option_init(opt, icmp6len, ndopts)
227 	void *opt;
228 	int icmp6len;
229 	union nd_opts *ndopts;
230 {
231 
232 	bzero(ndopts, sizeof(*ndopts));
233 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
234 	ndopts->nd_opts_last
235 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
236 
237 	if (icmp6len == 0) {
238 		ndopts->nd_opts_done = 1;
239 		ndopts->nd_opts_search = NULL;
240 	}
241 }
242 
243 /*
244  * Take one ND option.
245  */
246 struct nd_opt_hdr *
247 nd6_option(ndopts)
248 	union nd_opts *ndopts;
249 {
250 	struct nd_opt_hdr *nd_opt;
251 	int olen;
252 
253 	if (!ndopts)
254 		panic("ndopts == NULL in nd6_option");
255 	if (!ndopts->nd_opts_last)
256 		panic("uninitialized ndopts in nd6_option");
257 	if (!ndopts->nd_opts_search)
258 		return NULL;
259 	if (ndopts->nd_opts_done)
260 		return NULL;
261 
262 	nd_opt = ndopts->nd_opts_search;
263 
264 	/* make sure nd_opt_len is inside the buffer */
265 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
266 		bzero(ndopts, sizeof(*ndopts));
267 		return NULL;
268 	}
269 
270 	olen = nd_opt->nd_opt_len << 3;
271 	if (olen == 0) {
272 		/*
273 		 * Message validation requires that all included
274 		 * options have a length that is greater than zero.
275 		 */
276 		bzero(ndopts, sizeof(*ndopts));
277 		return NULL;
278 	}
279 
280 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
281 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
282 		/* option overruns the end of buffer, invalid */
283 		bzero(ndopts, sizeof(*ndopts));
284 		return NULL;
285 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
286 		/* reached the end of options chain */
287 		ndopts->nd_opts_done = 1;
288 		ndopts->nd_opts_search = NULL;
289 	}
290 	return nd_opt;
291 }
292 
293 /*
294  * Parse multiple ND options.
295  * This function is much easier to use, for ND routines that do not need
296  * multiple options of the same type.
297  */
298 int
299 nd6_options(ndopts)
300 	union nd_opts *ndopts;
301 {
302 	struct nd_opt_hdr *nd_opt;
303 	int i = 0;
304 
305 	if (!ndopts)
306 		panic("ndopts == NULL in nd6_options");
307 	if (!ndopts->nd_opts_last)
308 		panic("uninitialized ndopts in nd6_options");
309 	if (!ndopts->nd_opts_search)
310 		return 0;
311 
312 	while (1) {
313 		nd_opt = nd6_option(ndopts);
314 		if (!nd_opt && !ndopts->nd_opts_last) {
315 			/*
316 			 * Message validation requires that all included
317 			 * options have a length that is greater than zero.
318 			 */
319 			icmp6stat.icp6s_nd_badopt++;
320 			bzero(ndopts, sizeof(*ndopts));
321 			return -1;
322 		}
323 
324 		if (!nd_opt)
325 			goto skip1;
326 
327 		switch (nd_opt->nd_opt_type) {
328 		case ND_OPT_SOURCE_LINKADDR:
329 		case ND_OPT_TARGET_LINKADDR:
330 		case ND_OPT_MTU:
331 		case ND_OPT_REDIRECTED_HEADER:
332 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
333 				nd6log((LOG_INFO,
334 				    "duplicated ND6 option found (type=%d)\n",
335 				    nd_opt->nd_opt_type));
336 				/* XXX bark? */
337 			} else {
338 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
339 					= nd_opt;
340 			}
341 			break;
342 		case ND_OPT_PREFIX_INFORMATION:
343 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
344 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
345 					= nd_opt;
346 			}
347 			ndopts->nd_opts_pi_end =
348 				(struct nd_opt_prefix_info *)nd_opt;
349 			break;
350 		default:
351 			/*
352 			 * Unknown options must be silently ignored,
353 			 * to accomodate future extension to the protocol.
354 			 */
355 			nd6log((LOG_DEBUG,
356 			    "nd6_options: unsupported option %d - "
357 			    "option ignored\n", nd_opt->nd_opt_type));
358 		}
359 
360 skip1:
361 		i++;
362 		if (i > nd6_maxndopt) {
363 			icmp6stat.icp6s_nd_toomanyopt++;
364 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
365 			break;
366 		}
367 
368 		if (ndopts->nd_opts_done)
369 			break;
370 	}
371 
372 	return 0;
373 }
374 
375 /*
376  * ND6 timer routine to expire default route list and prefix list
377  */
378 void
379 nd6_timer(ignored_arg)
380 	void	*ignored_arg;
381 {
382 	int s;
383 	struct llinfo_nd6 *ln;
384 	struct nd_defrouter *dr;
385 	struct nd_prefix *pr;
386 	struct ifnet *ifp;
387 	struct in6_ifaddr *ia6, *nia6;
388 	struct in6_addrlifetime *lt6;
389 
390 	s = splsoftnet();
391 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
392 	    nd6_timer, NULL);
393 
394 	ln = llinfo_nd6.ln_next;
395 	while (ln && ln != &llinfo_nd6) {
396 		struct rtentry *rt;
397 		struct sockaddr_in6 *dst;
398 		struct llinfo_nd6 *next = ln->ln_next;
399 		/* XXX: used for the DELAY case only: */
400 		struct nd_ifinfo *ndi = NULL;
401 
402 		if ((rt = ln->ln_rt) == NULL) {
403 			ln = next;
404 			continue;
405 		}
406 		if ((ifp = rt->rt_ifp) == NULL) {
407 			ln = next;
408 			continue;
409 		}
410 		ndi = ND_IFINFO(ifp);
411 		dst = (struct sockaddr_in6 *)rt_key(rt);
412 
413 		if (ln->ln_expire > time.tv_sec) {
414 			ln = next;
415 			continue;
416 		}
417 
418 		/* sanity check */
419 		if (!rt)
420 			panic("rt=0 in nd6_timer(ln=%p)", ln);
421 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
422 			panic("rt_llinfo(%p) is not equal to ln(%p)",
423 			      rt->rt_llinfo, ln);
424 		if (!dst)
425 			panic("dst=0 in nd6_timer(ln=%p)", ln);
426 
427 		switch (ln->ln_state) {
428 		case ND6_LLINFO_INCOMPLETE:
429 			if (ln->ln_asked < nd6_mmaxtries) {
430 				ln->ln_asked++;
431 				ln->ln_expire = time.tv_sec +
432 				    ND6_RETRANS_SEC(ND_IFINFO(ifp)->retrans);
433 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
434 				    ln, 0);
435 			} else {
436 				struct mbuf *m = ln->ln_hold;
437 				if (m) {
438 					ln->ln_hold = NULL;
439 					/*
440 					 * Fake rcvif to make the ICMP error
441 					 * more helpful in diagnosing for the
442 					 * receiver.
443 					 * XXX: should we consider
444 					 * older rcvif?
445 					 */
446 					m->m_pkthdr.rcvif = rt->rt_ifp;
447 
448 					icmp6_error(m, ICMP6_DST_UNREACH,
449 						    ICMP6_DST_UNREACH_ADDR, 0);
450 				}
451 				next = nd6_free(rt, 0);
452 			}
453 			break;
454 		case ND6_LLINFO_REACHABLE:
455 			if (ln->ln_expire) {
456 				ln->ln_state = ND6_LLINFO_STALE;
457 				ln->ln_expire = time.tv_sec + nd6_gctimer;
458 			}
459 			break;
460 
461 		case ND6_LLINFO_STALE:
462 			/* Garbage Collection(RFC 2461 5.3) */
463 			if (ln->ln_expire)
464 				next = nd6_free(rt, 1);
465 			break;
466 
467 		case ND6_LLINFO_DELAY:
468 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
469 				/* We need NUD */
470 				ln->ln_asked = 1;
471 				ln->ln_state = ND6_LLINFO_PROBE;
472 				ln->ln_expire = time.tv_sec +
473 				    ND6_RETRANS_SEC(ndi->retrans);
474 				nd6_ns_output(ifp, &dst->sin6_addr,
475 				    &dst->sin6_addr, ln, 0);
476 			} else {
477 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
478 				ln->ln_expire = time.tv_sec + nd6_gctimer;
479 			}
480 			break;
481 		case ND6_LLINFO_PROBE:
482 			if (ln->ln_asked < nd6_umaxtries) {
483 				ln->ln_asked++;
484 				ln->ln_expire = time.tv_sec +
485 				    ND6_RETRANS_SEC(ND_IFINFO(ifp)->retrans);
486 				nd6_ns_output(ifp, &dst->sin6_addr,
487 				    &dst->sin6_addr, ln, 0);
488 			} else {
489 				next = nd6_free(rt, 0);
490 			}
491 			break;
492 		}
493 		ln = next;
494 	}
495 
496 	/* expire default router list */
497 	dr = TAILQ_FIRST(&nd_defrouter);
498 	while (dr) {
499 		if (dr->expire && dr->expire < time.tv_sec) {
500 			struct nd_defrouter *t;
501 			t = TAILQ_NEXT(dr, dr_entry);
502 			defrtrlist_del(dr);
503 			dr = t;
504 		} else {
505 			dr = TAILQ_NEXT(dr, dr_entry);
506 		}
507 	}
508 
509 	/*
510 	 * expire interface addresses.
511 	 * in the past the loop was inside prefix expiry processing.
512 	 * However, from a stricter speci-confrmance standpoint, we should
513 	 * rather separate address lifetimes and prefix lifetimes.
514 	 */
515 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
516 		nia6 = ia6->ia_next;
517 		/* check address lifetime */
518 		lt6 = &ia6->ia6_lifetime;
519 		if (IFA6_IS_INVALID(ia6)) {
520 			in6_purgeaddr(&ia6->ia_ifa);
521 		}
522 		if (IFA6_IS_DEPRECATED(ia6)) {
523 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
524 		} else {
525 			/*
526 			 * A new RA might have made a deprecated address
527 			 * preferred.
528 			 */
529 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
530 		}
531 	}
532 
533 	/* expire prefix list */
534 	pr = nd_prefix.lh_first;
535 	while (pr) {
536 		/*
537 		 * check prefix lifetime.
538 		 * since pltime is just for autoconf, pltime processing for
539 		 * prefix is not necessary.
540 		 */
541 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
542 		    time.tv_sec - pr->ndpr_lastupdate > pr->ndpr_vltime) {
543 			struct nd_prefix *t;
544 			t = pr->ndpr_next;
545 
546 			/*
547 			 * address expiration and prefix expiration are
548 			 * separate.  NEVER perform in6_purgeaddr here.
549 			 */
550 
551 			prelist_remove(pr);
552 			pr = t;
553 		} else
554 			pr = pr->ndpr_next;
555 	}
556 	splx(s);
557 }
558 
559 /*
560  * Nuke neighbor cache/prefix/default router management table, right before
561  * ifp goes away.
562  */
563 void
564 nd6_purge(ifp)
565 	struct ifnet *ifp;
566 {
567 	struct llinfo_nd6 *ln, *nln;
568 	struct nd_defrouter *dr, *ndr;
569 	struct nd_prefix *pr, *npr;
570 
571 	/*
572 	 * Nuke default router list entries toward ifp.
573 	 * We defer removal of default router list entries that is installed
574 	 * in the routing table, in order to keep additional side effects as
575 	 * small as possible.
576 	 */
577 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
578 		ndr = TAILQ_NEXT(dr, dr_entry);
579 		if (dr->installed)
580 			continue;
581 
582 		if (dr->ifp == ifp)
583 			defrtrlist_del(dr);
584 	}
585 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
586 		ndr = TAILQ_NEXT(dr, dr_entry);
587 		if (!dr->installed)
588 			continue;
589 
590 		if (dr->ifp == ifp)
591 			defrtrlist_del(dr);
592 	}
593 
594 	/* Nuke prefix list entries toward ifp */
595 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
596 		npr = pr->ndpr_next;
597 		if (pr->ndpr_ifp == ifp) {
598 			/*
599 			 * Previously, pr->ndpr_addr is removed as well,
600 			 * but I strongly believe we don't have to do it.
601 			 * nd6_purge() is only called from in6_ifdetach(),
602 			 * which removes all the associated interface addresses
603 			 * by itself.
604 			 * (jinmei@kame.net 20010129)
605 			 */
606 			prelist_remove(pr);
607 		}
608 	}
609 
610 	/* cancel default outgoing interface setting */
611 	if (nd6_defifindex == ifp->if_index)
612 		nd6_setdefaultiface(0);
613 
614 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
615 		/* refresh default router list */
616 		defrouter_select();
617 	}
618 
619 	/*
620 	 * Nuke neighbor cache entries for the ifp.
621 	 * Note that rt->rt_ifp may not be the same as ifp,
622 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
623 	 * nd6_rtrequest(), and ip6_input().
624 	 */
625 	ln = llinfo_nd6.ln_next;
626 	while (ln && ln != &llinfo_nd6) {
627 		struct rtentry *rt;
628 		struct sockaddr_dl *sdl;
629 
630 		nln = ln->ln_next;
631 		rt = ln->ln_rt;
632 		if (rt && rt->rt_gateway &&
633 		    rt->rt_gateway->sa_family == AF_LINK) {
634 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
635 			if (sdl->sdl_index == ifp->if_index)
636 				nln = nd6_free(rt, 0);
637 		}
638 		ln = nln;
639 	}
640 }
641 
642 struct rtentry *
643 nd6_lookup(addr6, create, ifp)
644 	struct in6_addr *addr6;
645 	int create;
646 	struct ifnet *ifp;
647 {
648 	struct rtentry *rt;
649 	struct sockaddr_in6 sin6;
650 
651 	bzero(&sin6, sizeof(sin6));
652 	sin6.sin6_len = sizeof(struct sockaddr_in6);
653 	sin6.sin6_family = AF_INET6;
654 	sin6.sin6_addr = *addr6;
655 	rt = rtalloc1((struct sockaddr *)&sin6, create);
656 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
657 		/*
658 		 * This is the case for the default route.
659 		 * If we want to create a neighbor cache for the address, we
660 		 * should free the route for the destination and allocate an
661 		 * interface route.
662 		 */
663 		if (create) {
664 			RTFREE(rt);
665 			rt = 0;
666 		}
667 	}
668 	if (!rt) {
669 		if (create && ifp) {
670 			int e;
671 
672 			/*
673 			 * If no route is available and create is set,
674 			 * we allocate a host route for the destination
675 			 * and treat it like an interface route.
676 			 * This hack is necessary for a neighbor which can't
677 			 * be covered by our own prefix.
678 			 */
679 			struct ifaddr *ifa =
680 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
681 			if (ifa == NULL)
682 				return (NULL);
683 
684 			/*
685 			 * Create a new route.  RTF_LLINFO is necessary
686 			 * to create a Neighbor Cache entry for the
687 			 * destination in nd6_rtrequest which will be
688 			 * called in rtrequest via ifa->ifa_rtrequest.
689 			 */
690 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
691 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
692 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
693 			    ~RTF_CLONING, &rt)) != 0) {
694 #if 0
695 				log(LOG_ERR,
696 				    "nd6_lookup: failed to add route for a "
697 				    "neighbor(%s), errno=%d\n",
698 				    ip6_sprintf(addr6), e);
699 #endif
700 				return (NULL);
701 			}
702 			if (rt == NULL)
703 				return (NULL);
704 			if (rt->rt_llinfo) {
705 				struct llinfo_nd6 *ln =
706 				    (struct llinfo_nd6 *)rt->rt_llinfo;
707 				ln->ln_state = ND6_LLINFO_NOSTATE;
708 			}
709 		} else
710 			return (NULL);
711 	}
712 	rt->rt_refcnt--;
713 	/*
714 	 * Validation for the entry.
715 	 * Note that the check for rt_llinfo is necessary because a cloned
716 	 * route from a parent route that has the L flag (e.g. the default
717 	 * route to a p2p interface) may have the flag, too, while the
718 	 * destination is not actually a neighbor.
719 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
720 	 *      it might be the loopback interface if the entry is for our
721 	 *      own address on a non-loopback interface. Instead, we should
722 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
723 	 *	interface.
724 	 */
725 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
726 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
727 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
728 		if (create) {
729 			nd6log((LOG_DEBUG,
730 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
731 			    ip6_sprintf(addr6),
732 			    ifp ? if_name(ifp) : "unspec"));
733 		}
734 		return (NULL);
735 	}
736 	return (rt);
737 }
738 
739 /*
740  * Detect if a given IPv6 address identifies a neighbor on a given link.
741  * XXX: should take care of the destination of a p2p link?
742  */
743 int
744 nd6_is_addr_neighbor(addr, ifp)
745 	struct sockaddr_in6 *addr;
746 	struct ifnet *ifp;
747 {
748 	struct nd_prefix *pr;
749 	struct rtentry *rt;
750 
751 	/*
752 	 * A link-local address is always a neighbor.
753 	 * XXX: we should use the sin6_scope_id field rather than the embedded
754 	 * interface index.
755 	 * XXX: a link does not necessarily specify a single interface.
756 	 */
757 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
758 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
759 		return (1);
760 
761 	/*
762 	 * If the address matches one of our on-link prefixes, it should be a
763 	 * neighbor.
764 	 */
765 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
766 		if (pr->ndpr_ifp != ifp)
767 			continue;
768 
769 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
770 			continue;
771 
772 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
773 		    &addr->sin6_addr, &pr->ndpr_mask))
774 			return (1);
775 	}
776 
777 	/*
778 	 * If the default router list is empty, all addresses are regarded
779 	 * as on-link, and thus, as a neighbor.
780 	 * XXX: we restrict the condition to hosts, because routers usually do
781 	 * not have the "default router list".
782 	 */
783 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
784 	    nd6_defifindex == ifp->if_index) {
785 		return (1);
786 	}
787 
788 	/*
789 	 * Even if the address matches none of our addresses, it might be
790 	 * in the neighbor cache.
791 	 */
792 	if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL)
793 		return (1);
794 
795 	return (0);
796 }
797 
798 /*
799  * Free an nd6 llinfo entry.
800  * Since the function would cause significant changes in the kernel, DO NOT
801  * make it global, unless you have a strong reason for the change, and are sure
802  * that the change is safe.
803  */
804 static struct llinfo_nd6 *
805 nd6_free(rt, gc)
806 	struct rtentry *rt;
807 	int gc;
808 {
809 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
810 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
811 	struct nd_defrouter *dr;
812 
813 	/*
814 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
815 	 * even though it is not harmful, it was not really necessary.
816 	 */
817 
818 	if (!ip6_forwarding) {
819 		int s;
820 		s = splsoftnet();
821 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
822 		    rt->rt_ifp);
823 
824 		if (dr != NULL && dr->expire &&
825 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
826 			/*
827 			 * If the reason for the deletion is just garbage
828 			 * collection, and the neighbor is an active default
829 			 * router, do not delete it.  Instead, reset the GC
830 			 * timer using the router's lifetime.
831 			 * Simply deleting the entry would affect default
832 			 * router selection, which is not necessarily a good
833 			 * thing, especially when we're using router preference
834 			 * values.
835 			 * XXX: the check for ln_state would be redundant,
836 			 *      but we intentionally keep it just in case.
837 			 */
838 			ln->ln_expire = dr->expire;
839 			splx(s);
840 			return (ln->ln_next);
841 		}
842 
843 		if (ln->ln_router || dr) {
844 			/*
845 			 * rt6_flush must be called whether or not the neighbor
846 			 * is in the Default Router List.
847 			 * See a corresponding comment in nd6_na_input().
848 			 */
849 			rt6_flush(&in6, rt->rt_ifp);
850 		}
851 
852 		if (dr) {
853 			/*
854 			 * Unreachablity of a router might affect the default
855 			 * router selection and on-link detection of advertised
856 			 * prefixes.
857 			 */
858 
859 			/*
860 			 * Temporarily fake the state to choose a new default
861 			 * router and to perform on-link determination of
862 			 * prefixes correctly.
863 			 * Below the state will be set correctly,
864 			 * or the entry itself will be deleted.
865 			 */
866 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
867 
868 			/*
869 			 * Since defrouter_select() does not affect the
870 			 * on-link determination and MIP6 needs the check
871 			 * before the default router selection, we perform
872 			 * the check now.
873 			 */
874 			pfxlist_onlink_check();
875 
876 			/*
877 			 * refresh default router list
878 			 */
879 			defrouter_select();
880 		}
881 		splx(s);
882 	}
883 
884 	/*
885 	 * Before deleting the entry, remember the next entry as the
886 	 * return value.  We need this because pfxlist_onlink_check() above
887 	 * might have freed other entries (particularly the old next entry) as
888 	 * a side effect (XXX).
889 	 */
890 	next = ln->ln_next;
891 
892 	/*
893 	 * Detach the route from the routing tree and the list of neighbor
894 	 * caches, and disable the route entry not to be used in already
895 	 * cached routes.
896 	 */
897 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
898 	    rt_mask(rt), 0, (struct rtentry **)0);
899 
900 	return (next);
901 }
902 
903 /*
904  * Upper-layer reachability hint for Neighbor Unreachability Detection.
905  *
906  * XXX cost-effective metods?
907  */
908 void
909 nd6_nud_hint(rt, dst6, force)
910 	struct rtentry *rt;
911 	struct in6_addr *dst6;
912 	int force;
913 {
914 	struct llinfo_nd6 *ln;
915 
916 	/*
917 	 * If the caller specified "rt", use that.  Otherwise, resolve the
918 	 * routing table by supplied "dst6".
919 	 */
920 	if (!rt) {
921 		if (!dst6)
922 			return;
923 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
924 			return;
925 	}
926 
927 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
928 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
929 	    !rt->rt_llinfo || !rt->rt_gateway ||
930 	    rt->rt_gateway->sa_family != AF_LINK) {
931 		/* This is not a host route. */
932 		return;
933 	}
934 
935 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
936 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
937 		return;
938 
939 	/*
940 	 * if we get upper-layer reachability confirmation many times,
941 	 * it is possible we have false information.
942 	 */
943 	if (!force) {
944 		ln->ln_byhint++;
945 		if (ln->ln_byhint > nd6_maxnudhint)
946 			return;
947 	}
948 
949 	ln->ln_state = ND6_LLINFO_REACHABLE;
950 	if (ln->ln_expire)
951 		ln->ln_expire = time.tv_sec + ND_IFINFO(rt->rt_ifp)->reachable;
952 }
953 
954 void
955 nd6_rtrequest(req, rt, info)
956 	int	req;
957 	struct rtentry *rt;
958 	struct rt_addrinfo *info; /* xxx unused */
959 {
960 	struct sockaddr *gate = rt->rt_gateway;
961 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
962 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
963 	struct ifnet *ifp = rt->rt_ifp;
964 	struct ifaddr *ifa;
965 	int mine = 0;
966 
967 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
968 		return;
969 
970 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
971 		/*
972 		 * This is probably an interface direct route for a link
973 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
974 		 * We do not need special treatment below for such a route.
975 		 * Moreover, the RTF_LLINFO flag which would be set below
976 		 * would annoy the ndp(8) command.
977 		 */
978 		return;
979 	}
980 
981 	if (req == RTM_RESOLVE &&
982 	    (nd6_need_cache(ifp) == 0 || /* stf case */
983 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
984 		/*
985 		 * FreeBSD and BSD/OS often make a cloned host route based
986 		 * on a less-specific route (e.g. the default route).
987 		 * If the less specific route does not have a "gateway"
988 		 * (this is the case when the route just goes to a p2p or an
989 		 * stf interface), we'll mistakenly make a neighbor cache for
990 		 * the host route, and will see strange neighbor solicitation
991 		 * for the corresponding destination.  In order to avoid the
992 		 * confusion, we check if the destination of the route is
993 		 * a neighbor in terms of neighbor discovery, and stop the
994 		 * process if not.  Additionally, we remove the LLINFO flag
995 		 * so that ndp(8) will not try to get the neighbor information
996 		 * of the destination.
997 		 */
998 		rt->rt_flags &= ~RTF_LLINFO;
999 		return;
1000 	}
1001 
1002 	switch (req) {
1003 	case RTM_ADD:
1004 		/*
1005 		 * There is no backward compatibility :)
1006 		 *
1007 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1008 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1009 		 *	   rt->rt_flags |= RTF_CLONING;
1010 		 */
1011 		if ((rt->rt_flags & RTF_CLONING) ||
1012 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1013 			/*
1014 			 * Case 1: This route should come from a route to
1015 			 * interface (RTF_CLONING case) or the route should be
1016 			 * treated as on-link but is currently not
1017 			 * (RTF_LLINFO && !ln case).
1018 			 */
1019 			rt_setgate(rt, rt_key(rt),
1020 				   (struct sockaddr *)&null_sdl);
1021 			gate = rt->rt_gateway;
1022 			SDL(gate)->sdl_type = ifp->if_type;
1023 			SDL(gate)->sdl_index = ifp->if_index;
1024 			if (ln)
1025 				ln->ln_expire = time.tv_sec;
1026 #if 1
1027 			if (ln && ln->ln_expire == 0) {
1028 				/* kludge for desktops */
1029 #if 0
1030 				printf("nd6_rtrequest: time.tv_sec is zero; "
1031 				       "treat it as 1\n");
1032 #endif
1033 				ln->ln_expire = 1;
1034 			}
1035 #endif
1036 			if ((rt->rt_flags & RTF_CLONING) != 0)
1037 				break;
1038 		}
1039 		/*
1040 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1041 		 * We don't do that here since llinfo is not ready yet.
1042 		 *
1043 		 * There are also couple of other things to be discussed:
1044 		 * - unsolicited NA code needs improvement beforehand
1045 		 * - RFC2461 says we MAY send multicast unsolicited NA
1046 		 *   (7.2.6 paragraph 4), however, it also says that we
1047 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1048 		 *   we don't have anything like it right now.
1049 		 *   note that the mechanism needs a mutual agreement
1050 		 *   between proxies, which means that we need to implement
1051 		 *   a new protocol, or a new kludge.
1052 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1053 		 *   we need to check ip6forwarding before sending it.
1054 		 *   (or should we allow proxy ND configuration only for
1055 		 *   routers?  there's no mention about proxy ND from hosts)
1056 		 */
1057 #if 0
1058 		/* XXX it does not work */
1059 		if (rt->rt_flags & RTF_ANNOUNCE)
1060 			nd6_na_output(ifp,
1061 			      &SIN6(rt_key(rt))->sin6_addr,
1062 			      &SIN6(rt_key(rt))->sin6_addr,
1063 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1064 			      1, NULL);
1065 #endif
1066 		/* FALLTHROUGH */
1067 	case RTM_RESOLVE:
1068 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1069 			/*
1070 			 * Address resolution isn't necessary for a point to
1071 			 * point link, so we can skip this test for a p2p link.
1072 			 */
1073 			if (gate->sa_family != AF_LINK ||
1074 			    gate->sa_len < sizeof(null_sdl)) {
1075 				log(LOG_DEBUG,
1076 				    "nd6_rtrequest: bad gateway value: %s\n",
1077 				    if_name(ifp));
1078 				break;
1079 			}
1080 			SDL(gate)->sdl_type = ifp->if_type;
1081 			SDL(gate)->sdl_index = ifp->if_index;
1082 		}
1083 		if (ln != NULL)
1084 			break;	/* This happens on a route change */
1085 		/*
1086 		 * Case 2: This route may come from cloning, or a manual route
1087 		 * add with a LL address.
1088 		 */
1089 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1090 		rt->rt_llinfo = (caddr_t)ln;
1091 		if (!ln) {
1092 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1093 			break;
1094 		}
1095 		nd6_inuse++;
1096 		nd6_allocated++;
1097 		Bzero(ln, sizeof(*ln));
1098 		ln->ln_rt = rt;
1099 		/* this is required for "ndp" command. - shin */
1100 		if (req == RTM_ADD) {
1101 		        /*
1102 			 * gate should have some valid AF_LINK entry,
1103 			 * and ln->ln_expire should have some lifetime
1104 			 * which is specified by ndp command.
1105 			 */
1106 			ln->ln_state = ND6_LLINFO_REACHABLE;
1107 			ln->ln_byhint = 0;
1108 		} else {
1109 		        /*
1110 			 * When req == RTM_RESOLVE, rt is created and
1111 			 * initialized in rtrequest(), so rt_expire is 0.
1112 			 */
1113 			ln->ln_state = ND6_LLINFO_NOSTATE;
1114 			ln->ln_expire = time.tv_sec;
1115 		}
1116 		rt->rt_flags |= RTF_LLINFO;
1117 		ln->ln_next = llinfo_nd6.ln_next;
1118 		llinfo_nd6.ln_next = ln;
1119 		ln->ln_prev = &llinfo_nd6;
1120 		ln->ln_next->ln_prev = ln;
1121 
1122 		/*
1123 		 * check if rt_key(rt) is one of my address assigned
1124 		 * to the interface.
1125 		 */
1126 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1127 		    &SIN6(rt_key(rt))->sin6_addr);
1128 		if (ifa) {
1129 			caddr_t macp = nd6_ifptomac(ifp);
1130 			ln->ln_expire = 0;
1131 			ln->ln_state = ND6_LLINFO_REACHABLE;
1132 			ln->ln_byhint = 0;
1133 			mine = 1;
1134 			if (macp) {
1135 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1136 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1137 			}
1138 			if (nd6_useloopback) {
1139 				rt->rt_ifp = &loif[0];	/* XXX */
1140 				/*
1141 				 * Make sure rt_ifa be equal to the ifaddr
1142 				 * corresponding to the address.
1143 				 * We need this because when we refer
1144 				 * rt_ifa->ia6_flags in ip6_input, we assume
1145 				 * that the rt_ifa points to the address instead
1146 				 * of the loopback address.
1147 				 */
1148 				if (ifa != rt->rt_ifa) {
1149 					IFAFREE(rt->rt_ifa);
1150 					IFAREF(ifa);
1151 					rt->rt_ifa = ifa;
1152 				}
1153 			}
1154 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1155 			ln->ln_expire = 0;
1156 			ln->ln_state = ND6_LLINFO_REACHABLE;
1157 			ln->ln_byhint = 0;
1158 
1159 			/* join solicited node multicast for proxy ND */
1160 			if (ifp->if_flags & IFF_MULTICAST) {
1161 				struct in6_addr llsol;
1162 				int error;
1163 
1164 				llsol = SIN6(rt_key(rt))->sin6_addr;
1165 				llsol.s6_addr16[0] = htons(0xff02);
1166 				llsol.s6_addr16[1] = htons(ifp->if_index);
1167 				llsol.s6_addr32[1] = 0;
1168 				llsol.s6_addr32[2] = htonl(1);
1169 				llsol.s6_addr8[12] = 0xff;
1170 
1171 				if (!in6_addmulti(&llsol, ifp, &error)) {
1172 					nd6log((LOG_ERR, "%s: failed to join "
1173 					    "%s (errno=%d)\n", if_name(ifp),
1174 					    ip6_sprintf(&llsol), error));
1175 				}
1176 			}
1177 		}
1178 		break;
1179 
1180 	case RTM_DELETE:
1181 		if (!ln)
1182 			break;
1183 		/* leave from solicited node multicast for proxy ND */
1184 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1185 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1186 			struct in6_addr llsol;
1187 			struct in6_multi *in6m;
1188 
1189 			llsol = SIN6(rt_key(rt))->sin6_addr;
1190 			llsol.s6_addr16[0] = htons(0xff02);
1191 			llsol.s6_addr16[1] = htons(ifp->if_index);
1192 			llsol.s6_addr32[1] = 0;
1193 			llsol.s6_addr32[2] = htonl(1);
1194 			llsol.s6_addr8[12] = 0xff;
1195 
1196 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1197 			if (in6m)
1198 				in6_delmulti(in6m);
1199 		}
1200 		nd6_inuse--;
1201 		ln->ln_next->ln_prev = ln->ln_prev;
1202 		ln->ln_prev->ln_next = ln->ln_next;
1203 		ln->ln_prev = NULL;
1204 		rt->rt_llinfo = 0;
1205 		rt->rt_flags &= ~RTF_LLINFO;
1206 		if (ln->ln_hold)
1207 			m_freem(ln->ln_hold);
1208 		Free((caddr_t)ln);
1209 	}
1210 }
1211 
1212 int
1213 nd6_ioctl(cmd, data, ifp)
1214 	u_long cmd;
1215 	caddr_t	data;
1216 	struct ifnet *ifp;
1217 {
1218 	struct in6_drlist *drl = (struct in6_drlist *)data;
1219 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1220 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1221 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1222 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1223 	struct nd_defrouter *dr;
1224 	struct nd_prefix *pr;
1225 	struct rtentry *rt;
1226 	int i = 0, error = 0;
1227 	int s;
1228 
1229 	switch (cmd) {
1230 	case SIOCGDRLST_IN6:
1231 		/*
1232 		 * obsolete API, use sysctl under net.inet6.icmp6
1233 		 */
1234 		bzero(drl, sizeof(*drl));
1235 		s = splsoftnet();
1236 		dr = TAILQ_FIRST(&nd_defrouter);
1237 		while (dr && i < DRLSTSIZ) {
1238 			drl->defrouter[i].rtaddr = dr->rtaddr;
1239 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1240 				/* XXX: need to this hack for KAME stack */
1241 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1242 			} else
1243 				log(LOG_ERR,
1244 				    "default router list contains a "
1245 				    "non-linklocal address(%s)\n",
1246 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1247 
1248 			drl->defrouter[i].flags = dr->flags;
1249 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1250 			drl->defrouter[i].expire = dr->expire;
1251 			drl->defrouter[i].if_index = dr->ifp->if_index;
1252 			i++;
1253 			dr = TAILQ_NEXT(dr, dr_entry);
1254 		}
1255 		splx(s);
1256 		break;
1257 	case SIOCGPRLST_IN6:
1258 		/*
1259 		 * obsolete API, use sysctl under net.inet6.icmp6
1260 		 *
1261 		 * XXX the structure in6_prlist was changed in backward-
1262 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1263 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1264 		 */
1265 		/*
1266 		 * XXX meaning of fields, especialy "raflags", is very
1267 		 * differnet between RA prefix list and RR/static prefix list.
1268 		 * how about separating ioctls into two?
1269 		 */
1270 		bzero(oprl, sizeof(*oprl));
1271 		s = splsoftnet();
1272 		pr = nd_prefix.lh_first;
1273 		while (pr && i < PRLSTSIZ) {
1274 			struct nd_pfxrouter *pfr;
1275 			int j;
1276 
1277 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1278 			oprl->prefix[i].raflags = pr->ndpr_raf;
1279 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1280 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1281 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1282 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1283 			oprl->prefix[i].expire = pr->ndpr_expire;
1284 
1285 			pfr = pr->ndpr_advrtrs.lh_first;
1286 			j = 0;
1287 			while (pfr) {
1288 				if (j < DRLSTSIZ) {
1289 #define RTRADDR oprl->prefix[i].advrtr[j]
1290 					RTRADDR = pfr->router->rtaddr;
1291 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1292 						/* XXX: hack for KAME */
1293 						RTRADDR.s6_addr16[1] = 0;
1294 					} else
1295 						log(LOG_ERR,
1296 						    "a router(%s) advertises "
1297 						    "a prefix with "
1298 						    "non-link local address\n",
1299 						    ip6_sprintf(&RTRADDR));
1300 #undef RTRADDR
1301 				}
1302 				j++;
1303 				pfr = pfr->pfr_next;
1304 			}
1305 			oprl->prefix[i].advrtrs = j;
1306 			oprl->prefix[i].origin = PR_ORIG_RA;
1307 
1308 			i++;
1309 			pr = pr->ndpr_next;
1310 		}
1311 		splx(s);
1312 
1313 		break;
1314 	case OSIOCGIFINFO_IN6:
1315 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1316 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1317 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1318 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1319 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1320 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1321 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1322 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1323 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1324 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1325 		break;
1326 	case SIOCGIFINFO_IN6:
1327 		ndi->ndi = *ND_IFINFO(ifp);
1328 		break;
1329 	case SIOCSIFINFO_FLAGS:
1330 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1331 		break;
1332 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1333 		/* sync kernel routing table with the default router list */
1334 		defrouter_reset();
1335 		defrouter_select();
1336 		break;
1337 	case SIOCSPFXFLUSH_IN6:
1338 	{
1339 		/* flush all the prefix advertised by routers */
1340 		struct nd_prefix *pr, *next;
1341 
1342 		s = splsoftnet();
1343 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1344 			struct in6_ifaddr *ia, *ia_next;
1345 
1346 			next = pr->ndpr_next;
1347 
1348 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1349 				continue; /* XXX */
1350 
1351 			/* do we really have to remove addresses as well? */
1352 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1353 				/* ia might be removed.  keep the next ptr. */
1354 				ia_next = ia->ia_next;
1355 
1356 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1357 					continue;
1358 
1359 				if (ia->ia6_ndpr == pr)
1360 					in6_purgeaddr(&ia->ia_ifa);
1361 			}
1362 			prelist_remove(pr);
1363 		}
1364 		splx(s);
1365 		break;
1366 	}
1367 	case SIOCSRTRFLUSH_IN6:
1368 	{
1369 		/* flush all the default routers */
1370 		struct nd_defrouter *dr, *next;
1371 
1372 		s = splsoftnet();
1373 		defrouter_reset();
1374 		for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) {
1375 			next = TAILQ_NEXT(dr, dr_entry);
1376 			defrtrlist_del(dr);
1377 		}
1378 		defrouter_select();
1379 		splx(s);
1380 		break;
1381 	}
1382 	case SIOCGNBRINFO_IN6:
1383 	    {
1384 		struct llinfo_nd6 *ln;
1385 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1386 
1387 		/*
1388 		 * XXX: KAME specific hack for scoped addresses
1389 		 *      XXXX: for other scopes than link-local?
1390 		 */
1391 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1392 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1393 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1394 
1395 			if (*idp == 0)
1396 				*idp = htons(ifp->if_index);
1397 		}
1398 
1399 		s = splsoftnet();
1400 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1401 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1402 			error = EINVAL;
1403 			splx(s);
1404 			break;
1405 		}
1406 		nbi->state = ln->ln_state;
1407 		nbi->asked = ln->ln_asked;
1408 		nbi->isrouter = ln->ln_router;
1409 		nbi->expire = ln->ln_expire;
1410 		splx(s);
1411 
1412 		break;
1413 	}
1414 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1415 		ndif->ifindex = nd6_defifindex;
1416 		break;
1417 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1418 		return (nd6_setdefaultiface(ndif->ifindex));
1419 	}
1420 	return (error);
1421 }
1422 
1423 /*
1424  * Create neighbor cache entry and cache link-layer address,
1425  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1426  */
1427 struct rtentry *
1428 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1429 	struct ifnet *ifp;
1430 	struct in6_addr *from;
1431 	char *lladdr;
1432 	int lladdrlen;
1433 	int type;	/* ICMP6 type */
1434 	int code;	/* type dependent information */
1435 {
1436 	struct rtentry *rt = NULL;
1437 	struct llinfo_nd6 *ln = NULL;
1438 	int is_newentry;
1439 	struct sockaddr_dl *sdl = NULL;
1440 	int do_update;
1441 	int olladdr;
1442 	int llchange;
1443 	int newstate = 0;
1444 
1445 	if (!ifp)
1446 		panic("ifp == NULL in nd6_cache_lladdr");
1447 	if (!from)
1448 		panic("from == NULL in nd6_cache_lladdr");
1449 
1450 	/* nothing must be updated for unspecified address */
1451 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1452 		return NULL;
1453 
1454 	/*
1455 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1456 	 * the caller.
1457 	 *
1458 	 * XXX If the link does not have link-layer adderss, what should
1459 	 * we do? (ifp->if_addrlen == 0)
1460 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1461 	 * description on it in NS section (RFC 2461 7.2.3).
1462 	 */
1463 
1464 	rt = nd6_lookup(from, 0, ifp);
1465 	if (!rt) {
1466 #if 0
1467 		/* nothing must be done if there's no lladdr */
1468 		if (!lladdr || !lladdrlen)
1469 			return NULL;
1470 #endif
1471 
1472 		rt = nd6_lookup(from, 1, ifp);
1473 		is_newentry = 1;
1474 	} else {
1475 		/* do nothing if static ndp is set */
1476 		if (rt->rt_flags & RTF_STATIC)
1477 			return NULL;
1478 		is_newentry = 0;
1479 	}
1480 
1481 	if (!rt)
1482 		return NULL;
1483 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1484 fail:
1485 		(void)nd6_free(rt, 0);
1486 		return NULL;
1487 	}
1488 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1489 	if (!ln)
1490 		goto fail;
1491 	if (!rt->rt_gateway)
1492 		goto fail;
1493 	if (rt->rt_gateway->sa_family != AF_LINK)
1494 		goto fail;
1495 	sdl = SDL(rt->rt_gateway);
1496 
1497 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1498 	if (olladdr && lladdr) {
1499 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1500 			llchange = 1;
1501 		else
1502 			llchange = 0;
1503 	} else
1504 		llchange = 0;
1505 
1506 	/*
1507 	 * newentry olladdr  lladdr  llchange	(*=record)
1508 	 *	0	n	n	--	(1)
1509 	 *	0	y	n	--	(2)
1510 	 *	0	n	y	--	(3) * STALE
1511 	 *	0	y	y	n	(4) *
1512 	 *	0	y	y	y	(5) * STALE
1513 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1514 	 *	1	--	y	--	(7) * STALE
1515 	 */
1516 
1517 	if (lladdr) {		/* (3-5) and (7) */
1518 		/*
1519 		 * Record source link-layer address
1520 		 * XXX is it dependent to ifp->if_type?
1521 		 */
1522 		sdl->sdl_alen = ifp->if_addrlen;
1523 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1524 	}
1525 
1526 	if (!is_newentry) {
1527 		if ((!olladdr && lladdr) ||		/* (3) */
1528 		    (olladdr && lladdr && llchange)) {	/* (5) */
1529 			do_update = 1;
1530 			newstate = ND6_LLINFO_STALE;
1531 		} else					/* (1-2,4) */
1532 			do_update = 0;
1533 	} else {
1534 		do_update = 1;
1535 		if (!lladdr)				/* (6) */
1536 			newstate = ND6_LLINFO_NOSTATE;
1537 		else					/* (7) */
1538 			newstate = ND6_LLINFO_STALE;
1539 	}
1540 
1541 	if (do_update) {
1542 		/*
1543 		 * Update the state of the neighbor cache.
1544 		 */
1545 		ln->ln_state = newstate;
1546 
1547 		if (ln->ln_state == ND6_LLINFO_STALE) {
1548 			/*
1549 			 * XXX: since nd6_output() below will cause
1550 			 * state tansition to DELAY and reset the timer,
1551 			 * we must set the timer now, although it is actually
1552 			 * meaningless.
1553 			 */
1554 			ln->ln_expire = time.tv_sec + nd6_gctimer;
1555 
1556 			if (ln->ln_hold) {
1557 				/*
1558 				 * we assume ifp is not a p2p here, so just
1559 				 * set the 2nd argument as the 1st one.
1560 				 */
1561 				nd6_output(ifp, ifp, ln->ln_hold,
1562 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1563 				ln->ln_hold = NULL;
1564 			}
1565 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1566 			/* probe right away */
1567 			ln->ln_expire = time.tv_sec;
1568 		}
1569 	}
1570 
1571 	/*
1572 	 * ICMP6 type dependent behavior.
1573 	 *
1574 	 * NS: clear IsRouter if new entry
1575 	 * RS: clear IsRouter
1576 	 * RA: set IsRouter if there's lladdr
1577 	 * redir: clear IsRouter if new entry
1578 	 *
1579 	 * RA case, (1):
1580 	 * The spec says that we must set IsRouter in the following cases:
1581 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1582 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1583 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1584 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1585 	 * neighbor cache, this is similar to (6).
1586 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1587 	 *
1588 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1589 	 *							D R
1590 	 *	0	n	n	--	(1)	c   ?     s
1591 	 *	0	y	n	--	(2)	c   s     s
1592 	 *	0	n	y	--	(3)	c   s     s
1593 	 *	0	y	y	n	(4)	c   s     s
1594 	 *	0	y	y	y	(5)	c   s     s
1595 	 *	1	--	n	--	(6) c	c 	c s
1596 	 *	1	--	y	--	(7) c	c   s	c s
1597 	 *
1598 	 *					(c=clear s=set)
1599 	 */
1600 	switch (type & 0xff) {
1601 	case ND_NEIGHBOR_SOLICIT:
1602 		/*
1603 		 * New entry must have is_router flag cleared.
1604 		 */
1605 		if (is_newentry)	/* (6-7) */
1606 			ln->ln_router = 0;
1607 		break;
1608 	case ND_REDIRECT:
1609 		/*
1610 		 * If the icmp is a redirect to a better router, always set the
1611 		 * is_router flag.  Otherwise, if the entry is newly created,
1612 		 * clear the flag.  [RFC 2461, sec 8.3]
1613 		 */
1614 		if (code == ND_REDIRECT_ROUTER)
1615 			ln->ln_router = 1;
1616 		else if (is_newentry) /* (6-7) */
1617 			ln->ln_router = 0;
1618 		break;
1619 	case ND_ROUTER_SOLICIT:
1620 		/*
1621 		 * is_router flag must always be cleared.
1622 		 */
1623 		ln->ln_router = 0;
1624 		break;
1625 	case ND_ROUTER_ADVERT:
1626 		/*
1627 		 * Mark an entry with lladdr as a router.
1628 		 */
1629 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1630 		    (is_newentry && lladdr)) {			/* (7) */
1631 			ln->ln_router = 1;
1632 		}
1633 		break;
1634 	}
1635 
1636 	/*
1637 	 * When the link-layer address of a router changes, select the
1638 	 * best router again.  In particular, when the neighbor entry is newly
1639 	 * created, it might affect the selection policy.
1640 	 * Question: can we restrict the first condition to the "is_newentry"
1641 	 * case?
1642 	 * XXX: when we hear an RA from a new router with the link-layer
1643 	 * address option, defrouter_select() is called twice, since
1644 	 * defrtrlist_update called the function as well.  However, I believe
1645 	 * we can compromise the overhead, since it only happens the first
1646 	 * time.
1647 	 * XXX: although defrouter_select() should not have a bad effect
1648 	 * for those are not autoconfigured hosts, we explicitly avoid such
1649 	 * cases for safety.
1650 	 */
1651 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1652 		defrouter_select();
1653 
1654 	return rt;
1655 }
1656 
1657 static void
1658 nd6_slowtimo(ignored_arg)
1659     void *ignored_arg;
1660 {
1661 	int s = splsoftnet();
1662 	struct nd_ifinfo *nd6if;
1663 	struct ifnet *ifp;
1664 
1665 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1666 	    nd6_slowtimo, NULL);
1667 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1668 	{
1669 		nd6if = ND_IFINFO(ifp);
1670 		if (nd6if->basereachable && /* already initialized */
1671 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1672 			/*
1673 			 * Since reachable time rarely changes by router
1674 			 * advertisements, we SHOULD insure that a new random
1675 			 * value gets recomputed at least once every few hours.
1676 			 * (RFC 2461, 6.3.4)
1677 			 */
1678 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1679 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1680 		}
1681 	}
1682 	splx(s);
1683 }
1684 
1685 #define senderr(e) { error = (e); goto bad;}
1686 int
1687 nd6_output(ifp, origifp, m0, dst, rt0)
1688 	struct ifnet *ifp;
1689 	struct ifnet *origifp;
1690 	struct mbuf *m0;
1691 	struct sockaddr_in6 *dst;
1692 	struct rtentry *rt0;
1693 {
1694 	struct mbuf *m = m0;
1695 	struct rtentry *rt = rt0;
1696 	struct sockaddr_in6 *gw6 = NULL;
1697 	struct llinfo_nd6 *ln = NULL;
1698 	int error = 0;
1699 
1700 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1701 		goto sendpkt;
1702 
1703 	if (nd6_need_cache(ifp) == 0)
1704 		goto sendpkt;
1705 
1706 	/*
1707 	 * next hop determination.  This routine is derived from ether_outpout.
1708 	 */
1709 	if (rt) {
1710 		if ((rt->rt_flags & RTF_UP) == 0) {
1711 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1712 			    1)) != NULL)
1713 			{
1714 				rt->rt_refcnt--;
1715 				if (rt->rt_ifp != ifp) {
1716 					/* XXX: loop care? */
1717 					return nd6_output(ifp, origifp, m0,
1718 					    dst, rt);
1719 				}
1720 			} else
1721 				senderr(EHOSTUNREACH);
1722 		}
1723 
1724 		if (rt->rt_flags & RTF_GATEWAY) {
1725 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1726 
1727 			/*
1728 			 * We skip link-layer address resolution and NUD
1729 			 * if the gateway is not a neighbor from ND point
1730 			 * of view, regardless of the value of nd_ifinfo.flags.
1731 			 * The second condition is a bit tricky; we skip
1732 			 * if the gateway is our own address, which is
1733 			 * sometimes used to install a route to a p2p link.
1734 			 */
1735 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1736 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1737 				/*
1738 				 * We allow this kind of tricky route only
1739 				 * when the outgoing interface is p2p.
1740 				 * XXX: we may need a more generic rule here.
1741 				 */
1742 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1743 					senderr(EHOSTUNREACH);
1744 
1745 				goto sendpkt;
1746 			}
1747 
1748 			if (rt->rt_gwroute == 0)
1749 				goto lookup;
1750 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1751 				rtfree(rt); rt = rt0;
1752 			lookup:
1753 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1754 				if ((rt = rt->rt_gwroute) == 0)
1755 					senderr(EHOSTUNREACH);
1756 				/* the "G" test below also prevents rt == rt0 */
1757 				if ((rt->rt_flags & RTF_GATEWAY) ||
1758 				    (rt->rt_ifp != ifp)) {
1759 					rt->rt_refcnt--;
1760 					rt0->rt_gwroute = 0;
1761 					senderr(EHOSTUNREACH);
1762 				}
1763 			}
1764 		}
1765 	}
1766 
1767 	/*
1768 	 * Address resolution or Neighbor Unreachability Detection
1769 	 * for the next hop.
1770 	 * At this point, the destination of the packet must be a unicast
1771 	 * or an anycast address(i.e. not a multicast).
1772 	 */
1773 
1774 	/* Look up the neighbor cache for the nexthop */
1775 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1776 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1777 	else {
1778 		/*
1779 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1780 		 * the condition below is not very efficient.  But we believe
1781 		 * it is tolerable, because this should be a rare case.
1782 		 */
1783 		if (nd6_is_addr_neighbor(dst, ifp) &&
1784 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1785 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1786 	}
1787 	if (!ln || !rt) {
1788 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1789 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1790 			log(LOG_DEBUG,
1791 			    "nd6_output: can't allocate llinfo for %s "
1792 			    "(ln=%p, rt=%p)\n",
1793 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1794 			senderr(EIO);	/* XXX: good error? */
1795 		}
1796 
1797 		goto sendpkt;	/* send anyway */
1798 	}
1799 
1800 	/* We don't have to do link-layer address resolution on a p2p link. */
1801 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1802 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1803 		ln->ln_state = ND6_LLINFO_STALE;
1804 		ln->ln_expire = time.tv_sec + nd6_gctimer;
1805 	}
1806 
1807 	/*
1808 	 * The first time we send a packet to a neighbor whose entry is
1809 	 * STALE, we have to change the state to DELAY and a sets a timer to
1810 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1811 	 * neighbor unreachability detection on expiration.
1812 	 * (RFC 2461 7.3.3)
1813 	 */
1814 	if (ln->ln_state == ND6_LLINFO_STALE) {
1815 		ln->ln_asked = 0;
1816 		ln->ln_state = ND6_LLINFO_DELAY;
1817 		ln->ln_expire = time.tv_sec + nd6_delay;
1818 	}
1819 
1820 	/*
1821 	 * If the neighbor cache entry has a state other than INCOMPLETE
1822 	 * (i.e. its link-layer address is already resolved), just
1823 	 * send the packet.
1824 	 */
1825 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1826 		goto sendpkt;
1827 
1828 	/*
1829 	 * There is a neighbor cache entry, but no ethernet address
1830 	 * response yet.  Replace the held mbuf (if any) with this
1831 	 * latest one.
1832 	 */
1833 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1834 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1835 	if (ln->ln_hold)
1836 		m_freem(ln->ln_hold);
1837 	ln->ln_hold = m;
1838 	/*
1839 	 * If there has been no NS for the neighbor after entering the
1840 	 * INCOMPLETE state, send the first solicitation.
1841 	 * Technically this can be against the rate-limiting rule described in
1842 	 * Section 7.2.2 of RFC 2461 because the interval to the next scheduled
1843 	 * solicitation issued in nd6_timer() may be less than the specified
1844 	 * retransmission time.  This should not be a problem from a practical
1845 	 * point of view, because we'll typically see an immediate response
1846 	 * from the neighbor, which suppresses the succeeding solicitations.
1847 	 */
1848 	if (ln->ln_expire && ln->ln_asked == 0) {
1849 		ln->ln_asked++;
1850 		ln->ln_expire = time.tv_sec +
1851 		    ND6_RETRANS_SEC(ND_IFINFO(ifp)->retrans);
1852 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1853 	}
1854 	return (0);
1855 
1856   sendpkt:
1857 
1858 #ifdef IPSEC
1859 	/* clean ipsec history once it goes out of the node */
1860 	ipsec_delaux(m);
1861 #endif
1862 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1863 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1864 					 rt));
1865 	}
1866 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1867 
1868   bad:
1869 	if (m)
1870 		m_freem(m);
1871 	return (error);
1872 }
1873 #undef senderr
1874 
1875 int
1876 nd6_need_cache(ifp)
1877 	struct ifnet *ifp;
1878 {
1879 	/*
1880 	 * XXX: we currently do not make neighbor cache on any interface
1881 	 * other than ARCnet, Ethernet, FDDI and GIF.
1882 	 *
1883 	 * RFC2893 says:
1884 	 * - unidirectional tunnels needs no ND
1885 	 */
1886 	switch (ifp->if_type) {
1887 	case IFT_ARCNET:
1888 	case IFT_ETHER:
1889 	case IFT_FDDI:
1890 	case IFT_IEEE1394:
1891 	case IFT_GIF:		/* XXX need more cases? */
1892 		return (1);
1893 	default:
1894 		return (0);
1895 	}
1896 }
1897 
1898 int
1899 nd6_storelladdr(ifp, rt, m, dst, desten)
1900 	struct ifnet *ifp;
1901 	struct rtentry *rt;
1902 	struct mbuf *m;
1903 	struct sockaddr *dst;
1904 	u_char *desten;
1905 {
1906 	struct sockaddr_dl *sdl;
1907 
1908 	if (m->m_flags & M_MCAST) {
1909 		switch (ifp->if_type) {
1910 		case IFT_ETHER:
1911 		case IFT_FDDI:
1912 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1913 						 desten);
1914 			return (1);
1915 		case IFT_IEEE1394:
1916 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1917 			return (1);
1918 		case IFT_ARCNET:
1919 			*desten = 0;
1920 			return (1);
1921 		default:
1922 			m_freem(m);
1923 			return (0);
1924 		}
1925 	}
1926 
1927 	if (rt == NULL) {
1928 		/* this could happen, if we could not allocate memory */
1929 		m_freem(m);
1930 		return (0);
1931 	}
1932 	if (rt->rt_gateway->sa_family != AF_LINK) {
1933 		printf("nd6_storelladdr: something odd happens\n");
1934 		m_freem(m);
1935 		return (0);
1936 	}
1937 	sdl = SDL(rt->rt_gateway);
1938 	if (sdl->sdl_alen == 0) {
1939 		/* this should be impossible, but we bark here for debugging */
1940 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
1941 		    ip6_sprintf(&SIN6(dst)->sin6_addr), if_name(ifp));
1942 		m_freem(m);
1943 		return (0);
1944 	}
1945 
1946 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
1947 	return (1);
1948 }
1949 
1950 int
1951 nd6_sysctl(name, oldp, oldlenp, newp, newlen)
1952 	int name;
1953 	void *oldp;	/* syscall arg, need copyout */
1954 	size_t *oldlenp;
1955 	void *newp;	/* syscall arg, need copyin */
1956 	size_t newlen;
1957 {
1958 	void *p;
1959 	size_t ol, l;
1960 	int error;
1961 
1962 	error = 0;
1963 	l = 0;
1964 
1965 	if (newp)
1966 		return EPERM;
1967 	if (oldp && !oldlenp)
1968 		return EINVAL;
1969 	ol = oldlenp ? *oldlenp : 0;
1970 
1971 	if (oldp) {
1972 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
1973 		if (!p)
1974 			return ENOMEM;
1975 	} else
1976 		p = NULL;
1977 	switch (name) {
1978 	case ICMPV6CTL_ND6_DRLIST:
1979 		error = fill_drlist(p, oldlenp, ol);
1980 		if (!error && p && oldp)
1981 			error = copyout(p, oldp, *oldlenp);
1982 		break;
1983 
1984 	case ICMPV6CTL_ND6_PRLIST:
1985 		error = fill_prlist(p, oldlenp, ol);
1986 		if (!error && p && oldp)
1987 			error = copyout(p, oldp, *oldlenp);
1988 		break;
1989 
1990 	default:
1991 		error = ENOPROTOOPT;
1992 		break;
1993 	}
1994 	if (p)
1995 		free(p, M_TEMP);
1996 
1997 	return (error);
1998 }
1999 
2000 static int
2001 fill_drlist(oldp, oldlenp, ol)
2002 	void *oldp;
2003 	size_t *oldlenp, ol;
2004 {
2005 	int error = 0, s;
2006 	struct in6_defrouter *d = NULL, *de = NULL;
2007 	struct nd_defrouter *dr;
2008 	size_t l;
2009 
2010 	s = splsoftnet();
2011 
2012 	if (oldp) {
2013 		d = (struct in6_defrouter *)oldp;
2014 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2015 	}
2016 	l = 0;
2017 
2018 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2019 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2020 
2021 		if (oldp && d + 1 <= de) {
2022 			bzero(d, sizeof(*d));
2023 			d->rtaddr.sin6_family = AF_INET6;
2024 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2025 			d->rtaddr.sin6_addr = dr->rtaddr;
2026 			in6_recoverscope(&d->rtaddr, &d->rtaddr.sin6_addr,
2027 			    dr->ifp);
2028 			d->flags = dr->flags;
2029 			d->rtlifetime = dr->rtlifetime;
2030 			d->expire = dr->expire;
2031 			d->if_index = dr->ifp->if_index;
2032 		}
2033 
2034 		l += sizeof(*d);
2035 		if (d)
2036 			d++;
2037 	}
2038 
2039 	if (oldp) {
2040 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2041 		if (l > ol)
2042 			error = ENOMEM;
2043 	} else
2044 		*oldlenp = l;
2045 
2046 	splx(s);
2047 
2048 	return (error);
2049 }
2050 
2051 static int
2052 fill_prlist(oldp, oldlenp, ol)
2053 	void *oldp;
2054 	size_t *oldlenp, ol;
2055 {
2056 	int error = 0, s;
2057 	struct nd_prefix *pr;
2058 	struct in6_prefix *p = NULL;
2059 	struct in6_prefix *pe = NULL;
2060 	size_t l;
2061 
2062 	s = splsoftnet();
2063 
2064 	if (oldp) {
2065 		p = (struct in6_prefix *)oldp;
2066 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2067 	}
2068 	l = 0;
2069 
2070 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2071 		u_short advrtrs;
2072 		size_t advance;
2073 		struct sockaddr_in6 *sin6;
2074 		struct sockaddr_in6 *s6;
2075 		struct nd_pfxrouter *pfr;
2076 
2077 		if (oldp && p + 1 <= pe)
2078 		{
2079 			bzero(p, sizeof(*p));
2080 			sin6 = (struct sockaddr_in6 *)(p + 1);
2081 
2082 			p->prefix = pr->ndpr_prefix;
2083 			if (in6_recoverscope(&p->prefix,
2084 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2085 				log(LOG_ERR,
2086 				    "scope error in prefix list (%s)\n",
2087 				    ip6_sprintf(&p->prefix.sin6_addr));
2088 			p->raflags = pr->ndpr_raf;
2089 			p->prefixlen = pr->ndpr_plen;
2090 			p->vltime = pr->ndpr_vltime;
2091 			p->pltime = pr->ndpr_pltime;
2092 			p->if_index = pr->ndpr_ifp->if_index;
2093 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2094 				p->expire = 0;
2095 			else {
2096 				time_t maxexpire;
2097 
2098 				/* XXX: we assume time_t is signed. */
2099 				maxexpire = (-1) &
2100 					~(1 << ((sizeof(maxexpire) * 8) - 1));
2101 				if (pr->ndpr_vltime <
2102 				    maxexpire - pr->ndpr_lastupdate) {
2103 					p->expire = pr->ndpr_lastupdate +
2104 						pr->ndpr_vltime;
2105 				} else
2106 					p->expire = maxexpire;
2107 			}
2108 			p->refcnt = pr->ndpr_refcnt;
2109 			p->flags = pr->ndpr_stateflags;
2110 			p->origin = PR_ORIG_RA;
2111 			advrtrs = 0;
2112 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2113 			     pfr = pfr->pfr_next) {
2114 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2115 					advrtrs++;
2116 					continue;
2117 				}
2118 				s6 = &sin6[advrtrs];
2119 				s6->sin6_family = AF_INET6;
2120 				s6->sin6_len = sizeof(struct sockaddr_in6);
2121 				s6->sin6_addr = pfr->router->rtaddr;
2122 				in6_recoverscope(s6, &s6->sin6_addr,
2123 				    pfr->router->ifp);
2124 				advrtrs++;
2125 			}
2126 			p->advrtrs = advrtrs;
2127 		}
2128 		else {
2129 			advrtrs = 0;
2130 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2131 			     pfr = pfr->pfr_next)
2132 				advrtrs++;
2133 		}
2134 
2135 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2136 		l += advance;
2137 		if (p)
2138 			p = (struct in6_prefix *)((caddr_t)p + advance);
2139 	}
2140 
2141 	if (oldp) {
2142 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2143 		if (l > ol)
2144 			error = ENOMEM;
2145 	} else
2146 		*oldlenp = l;
2147 
2148 	splx(s);
2149 
2150 	return (error);
2151 }
2152