xref: /openbsd-src/sys/netinet6/nd6.c (revision 8500990981f885cbe5e6a4958549cacc238b5ae6)
1 /*	$OpenBSD: nd6.c,v 1.65 2003/06/27 22:47:32 itojun Exp $	*/
2 /*	$KAME: nd6.c,v 1.280 2002/06/08 19:52:07 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/param.h>
34 #include <sys/systm.h>
35 #include <sys/timeout.h>
36 #include <sys/malloc.h>
37 #include <sys/mbuf.h>
38 #include <sys/socket.h>
39 #include <sys/sockio.h>
40 #include <sys/time.h>
41 #include <sys/kernel.h>
42 #include <sys/protosw.h>
43 #include <sys/errno.h>
44 #include <sys/ioctl.h>
45 #include <sys/syslog.h>
46 #include <sys/queue.h>
47 #include <dev/rndvar.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/if_types.h>
52 #include <net/if_fddi.h>
53 #include <net/route.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/if_ether.h>
57 #include <netinet/ip_ipsp.h>
58 
59 #include <netinet6/in6_var.h>
60 #include <netinet/ip6.h>
61 #include <netinet6/ip6_var.h>
62 #include <netinet6/nd6.h>
63 #include <netinet/icmp6.h>
64 
65 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
66 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
67 
68 #define SIN6(s) ((struct sockaddr_in6 *)s)
69 #define SDL(s) ((struct sockaddr_dl *)s)
70 
71 /* timer values */
72 int	nd6_prune	= 1;	/* walk list every 1 seconds */
73 int	nd6_delay	= 5;	/* delay first probe time 5 second */
74 int	nd6_umaxtries	= 3;	/* maximum unicast query */
75 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
76 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
77 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
78 
79 /* preventing too many loops in ND option parsing */
80 int nd6_maxndopt = 10;	/* max # of ND options allowed */
81 
82 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
83 
84 #ifdef ND6_DEBUG
85 int nd6_debug = 1;
86 #else
87 int nd6_debug = 0;
88 #endif
89 
90 /* for debugging? */
91 static int nd6_inuse, nd6_allocated;
92 
93 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
94 struct nd_drhead nd_defrouter;
95 struct nd_prhead nd_prefix = { 0 };
96 
97 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
98 static struct sockaddr_in6 all1_sa;
99 
100 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
101 static void nd6_slowtimo(void *);
102 static struct llinfo_nd6 *nd6_free(struct rtentry *, int);
103 static void nd6_llinfo_timer(void *);
104 
105 struct timeout nd6_slowtimo_ch;
106 struct timeout nd6_timer_ch;
107 extern struct timeout in6_tmpaddrtimer_ch;
108 
109 static int fill_drlist(void *, size_t *, size_t);
110 static int fill_prlist(void *, size_t *, size_t);
111 
112 void
113 nd6_init()
114 {
115 	static int nd6_init_done = 0;
116 	int i;
117 
118 	if (nd6_init_done) {
119 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
120 		return;
121 	}
122 
123 	all1_sa.sin6_family = AF_INET6;
124 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
125 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
126 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
127 
128 	/* initialization of the default router list */
129 	TAILQ_INIT(&nd_defrouter);
130 
131 	nd6_init_done = 1;
132 
133 	/* start timer */
134 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
135 	timeout_add(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz);
136 }
137 
138 struct nd_ifinfo *
139 nd6_ifattach(ifp)
140 	struct ifnet *ifp;
141 {
142 	struct nd_ifinfo *nd;
143 
144 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
145 	bzero(nd, sizeof(*nd));
146 
147 	nd->initialized = 1;
148 
149 	nd->chlim = IPV6_DEFHLIM;
150 	nd->basereachable = REACHABLE_TIME;
151 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
152 	nd->retrans = RETRANS_TIMER;
153 	/*
154 	 * Note that the default value of ip6_accept_rtadv is 0, which means
155 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
156 	 * here.
157 	 */
158 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
159 
160 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
161 	nd6_setmtu0(ifp, nd);
162 
163 	return nd;
164 }
165 
166 void
167 nd6_ifdetach(nd)
168 	struct nd_ifinfo *nd;
169 {
170 
171 	free(nd, M_IP6NDP);
172 }
173 
174 void
175 nd6_setmtu(ifp)
176 	struct ifnet *ifp;
177 {
178 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
179 }
180 
181 void
182 nd6_setmtu0(ifp, ndi)
183 	struct ifnet *ifp;
184 	struct nd_ifinfo *ndi;
185 {
186 	u_int32_t omaxmtu;
187 
188 	omaxmtu = ndi->maxmtu;
189 
190 	switch (ifp->if_type) {
191 	case IFT_ARCNET:
192 		ndi->maxmtu = MIN(60480, ifp->if_mtu); /* RFC2497 */
193 		break;
194 	case IFT_FDDI:
195 		ndi->maxmtu = MIN(FDDIMTU, ifp->if_mtu);
196 		break;
197 	default:
198 		ndi->maxmtu = ifp->if_mtu;
199 		break;
200 	}
201 
202 	/*
203 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
204 	 * undesirable situation.  We thus notify the operator of the change
205 	 * explicitly.  The check for omaxmtu is necessary to restrict the
206 	 * log to the case of changing the MTU, not initializing it.
207 	 */
208 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
209 		log(LOG_NOTICE, "nd6_setmtu0: "
210 		    "new link MTU on %s (%lu) is too small for IPv6\n",
211 		    ifp->if_xname, (unsigned long)ndi->maxmtu);
212 	}
213 
214 	if (ndi->maxmtu > in6_maxmtu)
215 		in6_setmaxmtu(); /* check all interfaces just in case */
216 }
217 
218 void
219 nd6_option_init(opt, icmp6len, ndopts)
220 	void *opt;
221 	int icmp6len;
222 	union nd_opts *ndopts;
223 {
224 
225 	bzero(ndopts, sizeof(*ndopts));
226 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
227 	ndopts->nd_opts_last
228 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
229 
230 	if (icmp6len == 0) {
231 		ndopts->nd_opts_done = 1;
232 		ndopts->nd_opts_search = NULL;
233 	}
234 }
235 
236 /*
237  * Take one ND option.
238  */
239 struct nd_opt_hdr *
240 nd6_option(ndopts)
241 	union nd_opts *ndopts;
242 {
243 	struct nd_opt_hdr *nd_opt;
244 	int olen;
245 
246 	if (!ndopts)
247 		panic("ndopts == NULL in nd6_option");
248 	if (!ndopts->nd_opts_last)
249 		panic("uninitialized ndopts in nd6_option");
250 	if (!ndopts->nd_opts_search)
251 		return NULL;
252 	if (ndopts->nd_opts_done)
253 		return NULL;
254 
255 	nd_opt = ndopts->nd_opts_search;
256 
257 	/* make sure nd_opt_len is inside the buffer */
258 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
259 		bzero(ndopts, sizeof(*ndopts));
260 		return NULL;
261 	}
262 
263 	olen = nd_opt->nd_opt_len << 3;
264 	if (olen == 0) {
265 		/*
266 		 * Message validation requires that all included
267 		 * options have a length that is greater than zero.
268 		 */
269 		bzero(ndopts, sizeof(*ndopts));
270 		return NULL;
271 	}
272 
273 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
274 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
275 		/* option overruns the end of buffer, invalid */
276 		bzero(ndopts, sizeof(*ndopts));
277 		return NULL;
278 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
279 		/* reached the end of options chain */
280 		ndopts->nd_opts_done = 1;
281 		ndopts->nd_opts_search = NULL;
282 	}
283 	return nd_opt;
284 }
285 
286 /*
287  * Parse multiple ND options.
288  * This function is much easier to use, for ND routines that do not need
289  * multiple options of the same type.
290  */
291 int
292 nd6_options(ndopts)
293 	union nd_opts *ndopts;
294 {
295 	struct nd_opt_hdr *nd_opt;
296 	int i = 0;
297 
298 	if (!ndopts)
299 		panic("ndopts == NULL in nd6_options");
300 	if (!ndopts->nd_opts_last)
301 		panic("uninitialized ndopts in nd6_options");
302 	if (!ndopts->nd_opts_search)
303 		return 0;
304 
305 	while (1) {
306 		nd_opt = nd6_option(ndopts);
307 		if (!nd_opt && !ndopts->nd_opts_last) {
308 			/*
309 			 * Message validation requires that all included
310 			 * options have a length that is greater than zero.
311 			 */
312 			icmp6stat.icp6s_nd_badopt++;
313 			bzero(ndopts, sizeof(*ndopts));
314 			return -1;
315 		}
316 
317 		if (!nd_opt)
318 			goto skip1;
319 
320 		switch (nd_opt->nd_opt_type) {
321 		case ND_OPT_SOURCE_LINKADDR:
322 		case ND_OPT_TARGET_LINKADDR:
323 		case ND_OPT_MTU:
324 		case ND_OPT_REDIRECTED_HEADER:
325 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
326 				nd6log((LOG_INFO,
327 				    "duplicated ND6 option found (type=%d)\n",
328 				    nd_opt->nd_opt_type));
329 				/* XXX bark? */
330 			} else {
331 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
332 					= nd_opt;
333 			}
334 			break;
335 		case ND_OPT_PREFIX_INFORMATION:
336 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
337 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
338 					= nd_opt;
339 			}
340 			ndopts->nd_opts_pi_end =
341 				(struct nd_opt_prefix_info *)nd_opt;
342 			break;
343 		default:
344 			/*
345 			 * Unknown options must be silently ignored,
346 			 * to accommodate future extension to the protocol.
347 			 */
348 			nd6log((LOG_DEBUG,
349 			    "nd6_options: unsupported option %d - "
350 			    "option ignored\n", nd_opt->nd_opt_type));
351 		}
352 
353 skip1:
354 		i++;
355 		if (i > nd6_maxndopt) {
356 			icmp6stat.icp6s_nd_toomanyopt++;
357 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
358 			break;
359 		}
360 
361 		if (ndopts->nd_opts_done)
362 			break;
363 	}
364 
365 	return 0;
366 }
367 
368 /*
369  * ND6 timer routine to handle ND6 entries
370  */
371 void
372 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long tick)
373 {
374 	int s;
375 
376 	s = splsoftnet();
377 
378 	if (tick < 0) {
379 		ln->ln_expire = 0;
380 		ln->ln_ntick = 0;
381 		timeout_del(&ln->ln_timer_ch);
382 	} else {
383 		ln->ln_expire = time.tv_sec + tick / hz;
384 		if (tick > INT_MAX) {
385 			ln->ln_ntick = tick - INT_MAX;
386 			timeout_add(&ln->ln_timer_ch, INT_MAX);
387 		} else {
388 			ln->ln_ntick = 0;
389 			timeout_add(&ln->ln_timer_ch, tick);
390 		}
391 	}
392 
393 	splx(s);
394 }
395 
396 static void
397 nd6_llinfo_timer(void *arg)
398 {
399 	int s;
400 	struct llinfo_nd6 *ln;
401 	struct rtentry *rt;
402 	struct sockaddr_in6 *dst;
403 	struct ifnet *ifp;
404 	struct nd_ifinfo *ndi = NULL;
405 
406 	s = splsoftnet();
407 
408 	ln = (struct llinfo_nd6 *)arg;
409 
410 	if (ln->ln_ntick > 0) {
411 		if (ln->ln_ntick > INT_MAX) {
412 			ln->ln_ntick -= INT_MAX;
413 			nd6_llinfo_settimer(ln, INT_MAX);
414 		} else {
415 			ln->ln_ntick = 0;
416 			nd6_llinfo_settimer(ln, ln->ln_ntick);
417 		}
418 		splx(s);
419 		return;
420 	}
421 
422 	if ((rt = ln->ln_rt) == NULL)
423 		panic("ln->ln_rt == NULL");
424 	if ((ifp = rt->rt_ifp) == NULL)
425 		panic("ln->ln_rt->rt_ifp == NULL");
426 	ndi = ND_IFINFO(ifp);
427 	dst = (struct sockaddr_in6 *)rt_key(rt);
428 
429 	/* sanity check */
430 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
431 		panic("rt_llinfo(%p) is not equal to ln(%p)",
432 		      rt->rt_llinfo, ln);
433 	if (!dst)
434 		panic("dst=0 in nd6_timer(ln=%p)", ln);
435 
436 	switch (ln->ln_state) {
437 	case ND6_LLINFO_INCOMPLETE:
438 		if (ln->ln_asked < nd6_mmaxtries) {
439 			ln->ln_asked++;
440 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
441 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
442 		} else {
443 			struct mbuf *m = ln->ln_hold;
444 			if (m) {
445 				ln->ln_hold = NULL;
446 				/*
447 				 * Fake rcvif to make the ICMP error
448 				 * more helpful in diagnosing for the
449 				 * receiver.
450 				 * XXX: should we consider
451 				 * older rcvif?
452 				 */
453 				m->m_pkthdr.rcvif = rt->rt_ifp;
454 
455 				icmp6_error(m, ICMP6_DST_UNREACH,
456 				    ICMP6_DST_UNREACH_ADDR, 0);
457 			}
458 			(void)nd6_free(rt, 0);
459 			ln = NULL;
460 		}
461 		break;
462 	case ND6_LLINFO_REACHABLE:
463 		if (!ND6_LLINFO_PERMANENT(ln)) {
464 			ln->ln_state = ND6_LLINFO_STALE;
465 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
466 		}
467 		break;
468 
469 	case ND6_LLINFO_STALE:
470 		/* Garbage Collection(RFC 2461 5.3) */
471 		if (!ND6_LLINFO_PERMANENT(ln)) {
472 			(void)nd6_free(rt, 1);
473 			ln = NULL;
474 		}
475 		break;
476 
477 	case ND6_LLINFO_DELAY:
478 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
479 			/* We need NUD */
480 			ln->ln_asked = 1;
481 			ln->ln_state = ND6_LLINFO_PROBE;
482 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
483 			nd6_ns_output(ifp, &dst->sin6_addr,
484 			    &dst->sin6_addr, ln, 0);
485 		} else {
486 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
487 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
488 		}
489 		break;
490 	case ND6_LLINFO_PROBE:
491 		if (ln->ln_asked < nd6_umaxtries) {
492 			ln->ln_asked++;
493 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
494 			nd6_ns_output(ifp, &dst->sin6_addr,
495 			    &dst->sin6_addr, ln, 0);
496 		} else {
497 			(void)nd6_free(rt, 0);
498 			ln = NULL;
499 		}
500 		break;
501 	}
502 
503 	splx(s);
504 }
505 
506 /*
507  * ND6 timer routine to expire default route list and prefix list
508  */
509 void
510 nd6_timer(ignored_arg)
511 	void	*ignored_arg;
512 {
513 	int s;
514 	struct nd_defrouter *dr;
515 	struct nd_prefix *pr;
516 	struct in6_ifaddr *ia6, *nia6;
517 	struct in6_addrlifetime *lt6;
518 
519 	s = splsoftnet();
520 	timeout_set(&nd6_timer_ch, nd6_timer, NULL);
521 	timeout_add(&nd6_timer_ch, nd6_prune * hz);
522 
523 	/* expire default router list */
524 	dr = TAILQ_FIRST(&nd_defrouter);
525 	while (dr) {
526 		if (dr->expire && dr->expire < time.tv_sec) {
527 			struct nd_defrouter *t;
528 			t = TAILQ_NEXT(dr, dr_entry);
529 			defrtrlist_del(dr);
530 			dr = t;
531 		} else {
532 			dr = TAILQ_NEXT(dr, dr_entry);
533 		}
534 	}
535 
536 	/*
537 	 * expire interface addresses.
538 	 * in the past the loop was inside prefix expiry processing.
539 	 * However, from a stricter speci-confrmance standpoint, we should
540 	 * rather separate address lifetimes and prefix lifetimes.
541 	 */
542 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
543 		nia6 = ia6->ia_next;
544 		/* check address lifetime */
545 		lt6 = &ia6->ia6_lifetime;
546 		if (IFA6_IS_INVALID(ia6)) {
547 			in6_purgeaddr(&ia6->ia_ifa);
548 		}
549 		if (IFA6_IS_DEPRECATED(ia6)) {
550 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
551 		} else {
552 			/*
553 			 * A new RA might have made a deprecated address
554 			 * preferred.
555 			 */
556 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
557 		}
558 	}
559 
560 	/* expire prefix list */
561 	pr = nd_prefix.lh_first;
562 	while (pr) {
563 		/*
564 		 * check prefix lifetime.
565 		 * since pltime is just for autoconf, pltime processing for
566 		 * prefix is not necessary.
567 		 */
568 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
569 		    time.tv_sec - pr->ndpr_lastupdate > pr->ndpr_vltime) {
570 			struct nd_prefix *t;
571 			t = pr->ndpr_next;
572 
573 			/*
574 			 * address expiration and prefix expiration are
575 			 * separate.  NEVER perform in6_purgeaddr here.
576 			 */
577 
578 			prelist_remove(pr);
579 			pr = t;
580 		} else
581 			pr = pr->ndpr_next;
582 	}
583 	splx(s);
584 }
585 
586 /*
587  * Nuke neighbor cache/prefix/default router management table, right before
588  * ifp goes away.
589  */
590 void
591 nd6_purge(ifp)
592 	struct ifnet *ifp;
593 {
594 	struct llinfo_nd6 *ln, *nln;
595 	struct nd_defrouter *dr, *ndr;
596 	struct nd_prefix *pr, *npr;
597 
598 	/*
599 	 * Nuke default router list entries toward ifp.
600 	 * We defer removal of default router list entries that is installed
601 	 * in the routing table, in order to keep additional side effects as
602 	 * small as possible.
603 	 */
604 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
605 		ndr = TAILQ_NEXT(dr, dr_entry);
606 		if (dr->installed)
607 			continue;
608 
609 		if (dr->ifp == ifp)
610 			defrtrlist_del(dr);
611 	}
612 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
613 		ndr = TAILQ_NEXT(dr, dr_entry);
614 		if (!dr->installed)
615 			continue;
616 
617 		if (dr->ifp == ifp)
618 			defrtrlist_del(dr);
619 	}
620 
621 	/* Nuke prefix list entries toward ifp */
622 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
623 		npr = pr->ndpr_next;
624 		if (pr->ndpr_ifp == ifp) {
625 			/*
626 			 * Previously, pr->ndpr_addr is removed as well,
627 			 * but I strongly believe we don't have to do it.
628 			 * nd6_purge() is only called from in6_ifdetach(),
629 			 * which removes all the associated interface addresses
630 			 * by itself.
631 			 * (jinmei@kame.net 20010129)
632 			 */
633 			prelist_remove(pr);
634 		}
635 	}
636 
637 	/* cancel default outgoing interface setting */
638 	if (nd6_defifindex == ifp->if_index)
639 		nd6_setdefaultiface(0);
640 
641 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
642 		/* refresh default router list */
643 		defrouter_select();
644 	}
645 
646 	/*
647 	 * Nuke neighbor cache entries for the ifp.
648 	 * Note that rt->rt_ifp may not be the same as ifp,
649 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
650 	 * nd6_rtrequest(), and ip6_input().
651 	 */
652 	ln = llinfo_nd6.ln_next;
653 	while (ln && ln != &llinfo_nd6) {
654 		struct rtentry *rt;
655 		struct sockaddr_dl *sdl;
656 
657 		nln = ln->ln_next;
658 		rt = ln->ln_rt;
659 		if (rt && rt->rt_gateway &&
660 		    rt->rt_gateway->sa_family == AF_LINK) {
661 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
662 			if (sdl->sdl_index == ifp->if_index)
663 				nln = nd6_free(rt, 0);
664 		}
665 		ln = nln;
666 	}
667 }
668 
669 struct rtentry *
670 nd6_lookup(addr6, create, ifp)
671 	struct in6_addr *addr6;
672 	int create;
673 	struct ifnet *ifp;
674 {
675 	struct rtentry *rt;
676 	struct sockaddr_in6 sin6;
677 
678 	bzero(&sin6, sizeof(sin6));
679 	sin6.sin6_len = sizeof(struct sockaddr_in6);
680 	sin6.sin6_family = AF_INET6;
681 	sin6.sin6_addr = *addr6;
682 
683 	rt = rtalloc1((struct sockaddr *)&sin6, create);
684 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
685 		/*
686 		 * This is the case for the default route.
687 		 * If we want to create a neighbor cache for the address, we
688 		 * should free the route for the destination and allocate an
689 		 * interface route.
690 		 */
691 		if (create) {
692 			RTFREE(rt);
693 			rt = 0;
694 		}
695 	}
696 	if (!rt) {
697 		if (create && ifp) {
698 			int e;
699 
700 			/*
701 			 * If no route is available and create is set,
702 			 * we allocate a host route for the destination
703 			 * and treat it like an interface route.
704 			 * This hack is necessary for a neighbor which can't
705 			 * be covered by our own prefix.
706 			 */
707 			struct ifaddr *ifa =
708 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
709 			if (ifa == NULL)
710 				return (NULL);
711 
712 			/*
713 			 * Create a new route.  RTF_LLINFO is necessary
714 			 * to create a Neighbor Cache entry for the
715 			 * destination in nd6_rtrequest which will be
716 			 * called in rtrequest via ifa->ifa_rtrequest.
717 			 */
718 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
719 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
720 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
721 			    ~RTF_CLONING, &rt)) != 0) {
722 #if 0
723 				log(LOG_ERR,
724 				    "nd6_lookup: failed to add route for a "
725 				    "neighbor(%s), errno=%d\n",
726 				    ip6_sprintf(addr6), e);
727 #endif
728 				return (NULL);
729 			}
730 			if (rt == NULL)
731 				return (NULL);
732 			if (rt->rt_llinfo) {
733 				struct llinfo_nd6 *ln =
734 				    (struct llinfo_nd6 *)rt->rt_llinfo;
735 				ln->ln_state = ND6_LLINFO_NOSTATE;
736 			}
737 		} else
738 			return (NULL);
739 	}
740 	rt->rt_refcnt--;
741 	/*
742 	 * Validation for the entry.
743 	 * Note that the check for rt_llinfo is necessary because a cloned
744 	 * route from a parent route that has the L flag (e.g. the default
745 	 * route to a p2p interface) may have the flag, too, while the
746 	 * destination is not actually a neighbor.
747 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
748 	 *      it might be the loopback interface if the entry is for our
749 	 *      own address on a non-loopback interface. Instead, we should
750 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
751 	 *	interface.
752 	 */
753 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
754 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
755 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
756 		if (create) {
757 			nd6log((LOG_DEBUG,
758 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
759 			    ip6_sprintf(addr6),
760 			    ifp ? ifp->if_xname : "unspec"));
761 		}
762 		return (NULL);
763 	}
764 	return (rt);
765 }
766 
767 /*
768  * Detect if a given IPv6 address identifies a neighbor on a given link.
769  * XXX: should take care of the destination of a p2p link?
770  */
771 int
772 nd6_is_addr_neighbor(addr, ifp)
773 	struct sockaddr_in6 *addr;
774 	struct ifnet *ifp;
775 {
776 	struct nd_prefix *pr;
777 	struct rtentry *rt;
778 
779 	/*
780 	 * A link-local address is always a neighbor.
781 	 * XXX: we should use the sin6_scope_id field rather than the embedded
782 	 * interface index.
783 	 * XXX: a link does not necessarily specify a single interface.
784 	 */
785 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
786 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
787 		return (1);
788 
789 	/*
790 	 * If the address matches one of our on-link prefixes, it should be a
791 	 * neighbor.
792 	 */
793 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
794 		if (pr->ndpr_ifp != ifp)
795 			continue;
796 
797 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
798 			continue;
799 
800 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
801 		    &addr->sin6_addr, &pr->ndpr_mask))
802 			return (1);
803 	}
804 
805 	/*
806 	 * If the default router list is empty, all addresses are regarded
807 	 * as on-link, and thus, as a neighbor.
808 	 * XXX: we restrict the condition to hosts, because routers usually do
809 	 * not have the "default router list".
810 	 */
811 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
812 	    nd6_defifindex == ifp->if_index) {
813 		return (1);
814 	}
815 
816 	/*
817 	 * Even if the address matches none of our addresses, it might be
818 	 * in the neighbor cache.
819 	 */
820 	if ((rt = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL)
821 		return (1);
822 
823 	return (0);
824 }
825 
826 /*
827  * Free an nd6 llinfo entry.
828  * Since the function would cause significant changes in the kernel, DO NOT
829  * make it global, unless you have a strong reason for the change, and are sure
830  * that the change is safe.
831  */
832 static struct llinfo_nd6 *
833 nd6_free(rt, gc)
834 	struct rtentry *rt;
835 	int gc;
836 {
837 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
838 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
839 	struct nd_defrouter *dr;
840 
841 	/*
842 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
843 	 * even though it is not harmful, it was not really necessary.
844 	 */
845 
846 	if (!ip6_forwarding) {
847 		int s;
848 		s = splsoftnet();
849 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
850 		    rt->rt_ifp);
851 
852 		if (dr != NULL && dr->expire &&
853 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
854 			/*
855 			 * If the reason for the deletion is just garbage
856 			 * collection, and the neighbor is an active default
857 			 * router, do not delete it.  Instead, reset the GC
858 			 * timer using the router's lifetime.
859 			 * Simply deleting the entry would affect default
860 			 * router selection, which is not necessarily a good
861 			 * thing, especially when we're using router preference
862 			 * values.
863 			 * XXX: the check for ln_state would be redundant,
864 			 *      but we intentionally keep it just in case.
865 			 */
866 			if (dr->expire > time.tv_sec * hz) {
867 				nd6_llinfo_settimer(ln,
868 				    dr->expire - time.tv_sec * hz);
869 			} else
870 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
871 			splx(s);
872 			return (ln->ln_next);
873 		}
874 
875 		if (ln->ln_router || dr) {
876 			/*
877 			 * rt6_flush must be called whether or not the neighbor
878 			 * is in the Default Router List.
879 			 * See a corresponding comment in nd6_na_input().
880 			 */
881 			rt6_flush(&in6, rt->rt_ifp);
882 		}
883 
884 		if (dr) {
885 			/*
886 			 * Unreachablity of a router might affect the default
887 			 * router selection and on-link detection of advertised
888 			 * prefixes.
889 			 */
890 
891 			/*
892 			 * Temporarily fake the state to choose a new default
893 			 * router and to perform on-link determination of
894 			 * prefixes correctly.
895 			 * Below the state will be set correctly,
896 			 * or the entry itself will be deleted.
897 			 */
898 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
899 
900 			/*
901 			 * Since defrouter_select() does not affect the
902 			 * on-link determination and MIP6 needs the check
903 			 * before the default router selection, we perform
904 			 * the check now.
905 			 */
906 			pfxlist_onlink_check();
907 
908 			/*
909 			 * refresh default router list
910 			 */
911 			defrouter_select();
912 		}
913 		splx(s);
914 	}
915 
916 	/*
917 	 * Before deleting the entry, remember the next entry as the
918 	 * return value.  We need this because pfxlist_onlink_check() above
919 	 * might have freed other entries (particularly the old next entry) as
920 	 * a side effect (XXX).
921 	 */
922 	next = ln->ln_next;
923 
924 	/*
925 	 * Detach the route from the routing tree and the list of neighbor
926 	 * caches, and disable the route entry not to be used in already
927 	 * cached routes.
928 	 */
929 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
930 	    rt_mask(rt), 0, (struct rtentry **)0);
931 
932 	return (next);
933 }
934 
935 /*
936  * Upper-layer reachability hint for Neighbor Unreachability Detection.
937  *
938  * XXX cost-effective metods?
939  */
940 void
941 nd6_nud_hint(rt, dst6, force)
942 	struct rtentry *rt;
943 	struct in6_addr *dst6;
944 	int force;
945 {
946 	struct llinfo_nd6 *ln;
947 
948 	/*
949 	 * If the caller specified "rt", use that.  Otherwise, resolve the
950 	 * routing table by supplied "dst6".
951 	 */
952 	if (!rt) {
953 		if (!dst6)
954 			return;
955 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
956 			return;
957 	}
958 
959 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
960 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
961 	    !rt->rt_llinfo || !rt->rt_gateway ||
962 	    rt->rt_gateway->sa_family != AF_LINK) {
963 		/* This is not a host route. */
964 		return;
965 	}
966 
967 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
968 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
969 		return;
970 
971 	/*
972 	 * if we get upper-layer reachability confirmation many times,
973 	 * it is possible we have false information.
974 	 */
975 	if (!force) {
976 		ln->ln_byhint++;
977 		if (ln->ln_byhint > nd6_maxnudhint)
978 			return;
979 	}
980 
981 	ln->ln_state = ND6_LLINFO_REACHABLE;
982 	if (!ND6_LLINFO_PERMANENT(ln)) {
983 		nd6_llinfo_settimer(ln,
984 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
985 	}
986 }
987 
988 void
989 nd6_rtrequest(req, rt, info)
990 	int	req;
991 	struct rtentry *rt;
992 	struct rt_addrinfo *info; /* xxx unused */
993 {
994 	struct sockaddr *gate = rt->rt_gateway;
995 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
996 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
997 	struct ifnet *ifp = rt->rt_ifp;
998 	struct ifaddr *ifa;
999 	int mine = 0;
1000 
1001 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1002 		return;
1003 
1004 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1005 		/*
1006 		 * This is probably an interface direct route for a link
1007 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1008 		 * We do not need special treatment below for such a route.
1009 		 * Moreover, the RTF_LLINFO flag which would be set below
1010 		 * would annoy the ndp(8) command.
1011 		 */
1012 		return;
1013 	}
1014 
1015 	if (req == RTM_RESOLVE &&
1016 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1017 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1018 		/*
1019 		 * FreeBSD and BSD/OS often make a cloned host route based
1020 		 * on a less-specific route (e.g. the default route).
1021 		 * If the less specific route does not have a "gateway"
1022 		 * (this is the case when the route just goes to a p2p or an
1023 		 * stf interface), we'll mistakenly make a neighbor cache for
1024 		 * the host route, and will see strange neighbor solicitation
1025 		 * for the corresponding destination.  In order to avoid the
1026 		 * confusion, we check if the destination of the route is
1027 		 * a neighbor in terms of neighbor discovery, and stop the
1028 		 * process if not.  Additionally, we remove the LLINFO flag
1029 		 * so that ndp(8) will not try to get the neighbor information
1030 		 * of the destination.
1031 		 */
1032 		rt->rt_flags &= ~RTF_LLINFO;
1033 		return;
1034 	}
1035 
1036 	switch (req) {
1037 	case RTM_ADD:
1038 		/*
1039 		 * There is no backward compatibility :)
1040 		 *
1041 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1042 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1043 		 *	   rt->rt_flags |= RTF_CLONING;
1044 		 */
1045 		if ((rt->rt_flags & RTF_CLONING) ||
1046 		    ((rt->rt_flags & RTF_LLINFO) && !ln)) {
1047 			/*
1048 			 * Case 1: This route should come from a route to
1049 			 * interface (RTF_CLONING case) or the route should be
1050 			 * treated as on-link but is currently not
1051 			 * (RTF_LLINFO && !ln case).
1052 			 */
1053 			rt_setgate(rt, rt_key(rt),
1054 				   (struct sockaddr *)&null_sdl);
1055 			gate = rt->rt_gateway;
1056 			SDL(gate)->sdl_type = ifp->if_type;
1057 			SDL(gate)->sdl_index = ifp->if_index;
1058 			if (ln)
1059 				nd6_llinfo_settimer(ln, 0);
1060 			if ((rt->rt_flags & RTF_CLONING) != 0)
1061 				break;
1062 		}
1063 		/*
1064 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1065 		 * We don't do that here since llinfo is not ready yet.
1066 		 *
1067 		 * There are also couple of other things to be discussed:
1068 		 * - unsolicited NA code needs improvement beforehand
1069 		 * - RFC2461 says we MAY send multicast unsolicited NA
1070 		 *   (7.2.6 paragraph 4), however, it also says that we
1071 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1072 		 *   we don't have anything like it right now.
1073 		 *   note that the mechanism needs a mutual agreement
1074 		 *   between proxies, which means that we need to implement
1075 		 *   a new protocol, or a new kludge.
1076 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1077 		 *   we need to check ip6forwarding before sending it.
1078 		 *   (or should we allow proxy ND configuration only for
1079 		 *   routers?  there's no mention about proxy ND from hosts)
1080 		 */
1081 #if 0
1082 		/* XXX it does not work */
1083 		if (rt->rt_flags & RTF_ANNOUNCE)
1084 			nd6_na_output(ifp,
1085 			      &SIN6(rt_key(rt))->sin6_addr,
1086 			      &SIN6(rt_key(rt))->sin6_addr,
1087 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1088 			      1, NULL);
1089 #endif
1090 		/* FALLTHROUGH */
1091 	case RTM_RESOLVE:
1092 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1093 			/*
1094 			 * Address resolution isn't necessary for a point to
1095 			 * point link, so we can skip this test for a p2p link.
1096 			 */
1097 			if (gate->sa_family != AF_LINK ||
1098 			    gate->sa_len < sizeof(null_sdl)) {
1099 				log(LOG_DEBUG,
1100 				    "nd6_rtrequest: bad gateway value: %s\n",
1101 				    ifp->if_xname);
1102 				break;
1103 			}
1104 			SDL(gate)->sdl_type = ifp->if_type;
1105 			SDL(gate)->sdl_index = ifp->if_index;
1106 		}
1107 		if (ln != NULL)
1108 			break;	/* This happens on a route change */
1109 		/*
1110 		 * Case 2: This route may come from cloning, or a manual route
1111 		 * add with a LL address.
1112 		 */
1113 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1114 		rt->rt_llinfo = (caddr_t)ln;
1115 		if (!ln) {
1116 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1117 			break;
1118 		}
1119 		nd6_inuse++;
1120 		nd6_allocated++;
1121 		Bzero(ln, sizeof(*ln));
1122 		ln->ln_rt = rt;
1123 		timeout_set(&ln->ln_timer_ch, nd6_llinfo_timer, ln);
1124 		/* this is required for "ndp" command. - shin */
1125 		if (req == RTM_ADD) {
1126 		        /*
1127 			 * gate should have some valid AF_LINK entry,
1128 			 * and ln->ln_expire should have some lifetime
1129 			 * which is specified by ndp command.
1130 			 */
1131 			ln->ln_state = ND6_LLINFO_REACHABLE;
1132 			ln->ln_byhint = 0;
1133 		} else {
1134 		        /*
1135 			 * When req == RTM_RESOLVE, rt is created and
1136 			 * initialized in rtrequest(), so rt_expire is 0.
1137 			 */
1138 			ln->ln_state = ND6_LLINFO_NOSTATE;
1139 			nd6_llinfo_settimer(ln, 0);
1140 		}
1141 		rt->rt_flags |= RTF_LLINFO;
1142 		ln->ln_next = llinfo_nd6.ln_next;
1143 		llinfo_nd6.ln_next = ln;
1144 		ln->ln_prev = &llinfo_nd6;
1145 		ln->ln_next->ln_prev = ln;
1146 
1147 		/*
1148 		 * check if rt_key(rt) is one of my address assigned
1149 		 * to the interface.
1150 		 */
1151 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1152 		    &SIN6(rt_key(rt))->sin6_addr);
1153 		if (ifa) {
1154 			caddr_t macp = nd6_ifptomac(ifp);
1155 			nd6_llinfo_settimer(ln, -1);
1156 			ln->ln_state = ND6_LLINFO_REACHABLE;
1157 			ln->ln_byhint = 0;
1158 			mine = 1;
1159 			if (macp) {
1160 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1161 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1162 			}
1163 			if (nd6_useloopback) {
1164 				rt->rt_ifp = lo0ifp;	/*XXX*/
1165 				/*
1166 				 * Make sure rt_ifa be equal to the ifaddr
1167 				 * corresponding to the address.
1168 				 * We need this because when we refer
1169 				 * rt_ifa->ia6_flags in ip6_input, we assume
1170 				 * that the rt_ifa points to the address instead
1171 				 * of the loopback address.
1172 				 */
1173 				if (ifa != rt->rt_ifa) {
1174 					IFAFREE(rt->rt_ifa);
1175 					ifa->ifa_refcnt++;
1176 					rt->rt_ifa = ifa;
1177 				}
1178 			}
1179 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1180 			nd6_llinfo_settimer(ln, -1);
1181 			ln->ln_state = ND6_LLINFO_REACHABLE;
1182 			ln->ln_byhint = 0;
1183 
1184 			/* join solicited node multicast for proxy ND */
1185 			if (ifp->if_flags & IFF_MULTICAST) {
1186 				struct in6_addr llsol;
1187 				int error;
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 				if (in6_addmulti(&llsol, ifp, &error)) {
1197 					nd6log((LOG_ERR, "%s: failed to join "
1198 					    "%s (errno=%d)\n", ifp->if_xname,
1199 					    ip6_sprintf(&llsol), error));
1200 				}
1201 			}
1202 		}
1203 		break;
1204 
1205 	case RTM_DELETE:
1206 		if (!ln)
1207 			break;
1208 		/* leave from solicited node multicast for proxy ND */
1209 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1210 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1211 			struct in6_addr llsol;
1212 			struct in6_multi *in6m;
1213 
1214 			llsol = SIN6(rt_key(rt))->sin6_addr;
1215 			llsol.s6_addr16[0] = htons(0xff02);
1216 			llsol.s6_addr16[1] = htons(ifp->if_index);
1217 			llsol.s6_addr32[1] = 0;
1218 			llsol.s6_addr32[2] = htonl(1);
1219 			llsol.s6_addr8[12] = 0xff;
1220 
1221 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1222 			if (in6m)
1223 				in6_delmulti(in6m);
1224 		}
1225 		nd6_inuse--;
1226 		ln->ln_next->ln_prev = ln->ln_prev;
1227 		ln->ln_prev->ln_next = ln->ln_next;
1228 		ln->ln_prev = NULL;
1229 		nd6_llinfo_settimer(ln, -1);
1230 		rt->rt_llinfo = 0;
1231 		rt->rt_flags &= ~RTF_LLINFO;
1232 		if (ln->ln_hold)
1233 			m_freem(ln->ln_hold);
1234 		Free((caddr_t)ln);
1235 	}
1236 }
1237 
1238 int
1239 nd6_ioctl(cmd, data, ifp)
1240 	u_long cmd;
1241 	caddr_t	data;
1242 	struct ifnet *ifp;
1243 {
1244 	struct in6_drlist *drl = (struct in6_drlist *)data;
1245 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1246 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1247 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1248 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1249 	struct nd_defrouter *dr;
1250 	struct nd_prefix *pr;
1251 	struct rtentry *rt;
1252 	int i = 0, error = 0;
1253 	int s;
1254 
1255 	switch (cmd) {
1256 	case SIOCGDRLST_IN6:
1257 		/*
1258 		 * obsolete API, use sysctl under net.inet6.icmp6
1259 		 */
1260 		bzero(drl, sizeof(*drl));
1261 		s = splsoftnet();
1262 		dr = TAILQ_FIRST(&nd_defrouter);
1263 		while (dr && i < DRLSTSIZ) {
1264 			drl->defrouter[i].rtaddr = dr->rtaddr;
1265 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1266 				/* XXX: need to this hack for KAME stack */
1267 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1268 			} else
1269 				log(LOG_ERR,
1270 				    "default router list contains a "
1271 				    "non-linklocal address(%s)\n",
1272 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1273 
1274 			drl->defrouter[i].flags = dr->flags;
1275 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1276 			drl->defrouter[i].expire = dr->expire;
1277 			drl->defrouter[i].if_index = dr->ifp->if_index;
1278 			i++;
1279 			dr = TAILQ_NEXT(dr, dr_entry);
1280 		}
1281 		splx(s);
1282 		break;
1283 	case SIOCGPRLST_IN6:
1284 		/*
1285 		 * obsolete API, use sysctl under net.inet6.icmp6
1286 		 *
1287 		 * XXX the structure in6_prlist was changed in backward-
1288 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1289 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1290 		 */
1291 		/*
1292 		 * XXX meaning of fields, especialy "raflags", is very
1293 		 * differnet between RA prefix list and RR/static prefix list.
1294 		 * how about separating ioctls into two?
1295 		 */
1296 		bzero(oprl, sizeof(*oprl));
1297 		s = splsoftnet();
1298 		pr = nd_prefix.lh_first;
1299 		while (pr && i < PRLSTSIZ) {
1300 			struct nd_pfxrouter *pfr;
1301 			int j;
1302 
1303 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1304 			oprl->prefix[i].raflags = pr->ndpr_raf;
1305 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1306 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1307 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1308 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1309 			oprl->prefix[i].expire = pr->ndpr_expire;
1310 
1311 			pfr = pr->ndpr_advrtrs.lh_first;
1312 			j = 0;
1313 			while(pfr) {
1314 				if (j < DRLSTSIZ) {
1315 #define RTRADDR oprl->prefix[i].advrtr[j]
1316 					RTRADDR = pfr->router->rtaddr;
1317 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1318 						/* XXX: hack for KAME */
1319 						RTRADDR.s6_addr16[1] = 0;
1320 					} else
1321 						log(LOG_ERR,
1322 						    "a router(%s) advertises "
1323 						    "a prefix with "
1324 						    "non-link local address\n",
1325 						    ip6_sprintf(&RTRADDR));
1326 #undef RTRADDR
1327 				}
1328 				j++;
1329 				pfr = pfr->pfr_next;
1330 			}
1331 			oprl->prefix[i].advrtrs = j;
1332 			oprl->prefix[i].origin = PR_ORIG_RA;
1333 
1334 			i++;
1335 			pr = pr->ndpr_next;
1336 		}
1337 		splx(s);
1338 
1339 		break;
1340 	case OSIOCGIFINFO_IN6:
1341 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1342 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1343 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1344 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1345 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1346 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1347 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1348 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1349 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1350 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1351 		break;
1352 	case SIOCGIFINFO_IN6:
1353 		ndi->ndi = *ND_IFINFO(ifp);
1354 		break;
1355 	case SIOCSIFINFO_FLAGS:
1356 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1357 		break;
1358 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1359 		/* sync kernel routing table with the default router list */
1360 		defrouter_reset();
1361 		defrouter_select();
1362 		break;
1363 	case SIOCSPFXFLUSH_IN6:
1364 	{
1365 		/* flush all the prefix advertised by routers */
1366 		struct nd_prefix *pr, *next;
1367 
1368 		s = splsoftnet();
1369 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1370 			struct in6_ifaddr *ia, *ia_next;
1371 
1372 			next = pr->ndpr_next;
1373 
1374 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1375 				continue; /* XXX */
1376 
1377 			/* do we really have to remove addresses as well? */
1378 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1379 				/* ia might be removed.  keep the next ptr. */
1380 				ia_next = ia->ia_next;
1381 
1382 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1383 					continue;
1384 
1385 				if (ia->ia6_ndpr == pr)
1386 					in6_purgeaddr(&ia->ia_ifa);
1387 			}
1388 			prelist_remove(pr);
1389 		}
1390 		splx(s);
1391 		break;
1392 	}
1393 	case SIOCSRTRFLUSH_IN6:
1394 	{
1395 		/* flush all the default routers */
1396 		struct nd_defrouter *dr, *next;
1397 
1398 		s = splsoftnet();
1399 		defrouter_reset();
1400 		for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) {
1401 			next = TAILQ_NEXT(dr, dr_entry);
1402 			defrtrlist_del(dr);
1403 		}
1404 		defrouter_select();
1405 		splx(s);
1406 		break;
1407 	}
1408 	case SIOCGNBRINFO_IN6:
1409 	{
1410 		struct llinfo_nd6 *ln;
1411 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1412 
1413 		/*
1414 		 * XXX: KAME specific hack for scoped addresses
1415 		 *      XXXX: for other scopes than link-local?
1416 		 */
1417 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1418 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1419 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1420 
1421 			if (*idp == 0)
1422 				*idp = htons(ifp->if_index);
1423 		}
1424 
1425 		s = splsoftnet();
1426 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL ||
1427 		    (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) {
1428 			error = EINVAL;
1429 			splx(s);
1430 			break;
1431 		}
1432 		nbi->state = ln->ln_state;
1433 		nbi->asked = ln->ln_asked;
1434 		nbi->isrouter = ln->ln_router;
1435 		nbi->expire = ln->ln_expire;
1436 		splx(s);
1437 
1438 		break;
1439 	}
1440 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1441 		ndif->ifindex = nd6_defifindex;
1442 		break;
1443 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1444 		return (nd6_setdefaultiface(ndif->ifindex));
1445 		break;
1446 	}
1447 	return (error);
1448 }
1449 
1450 /*
1451  * Create neighbor cache entry and cache link-layer address,
1452  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1453  */
1454 struct rtentry *
1455 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1456 	struct ifnet *ifp;
1457 	struct in6_addr *from;
1458 	char *lladdr;
1459 	int lladdrlen;
1460 	int type;	/* ICMP6 type */
1461 	int code;	/* type dependent information */
1462 {
1463 	struct rtentry *rt = NULL;
1464 	struct llinfo_nd6 *ln = NULL;
1465 	int is_newentry;
1466 	struct sockaddr_dl *sdl = NULL;
1467 	int do_update;
1468 	int olladdr;
1469 	int llchange;
1470 	int newstate = 0;
1471 
1472 	if (!ifp)
1473 		panic("ifp == NULL in nd6_cache_lladdr");
1474 	if (!from)
1475 		panic("from == NULL in nd6_cache_lladdr");
1476 
1477 	/* nothing must be updated for unspecified address */
1478 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1479 		return NULL;
1480 
1481 	/*
1482 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1483 	 * the caller.
1484 	 *
1485 	 * XXX If the link does not have link-layer adderss, what should
1486 	 * we do? (ifp->if_addrlen == 0)
1487 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1488 	 * description on it in NS section (RFC 2461 7.2.3).
1489 	 */
1490 
1491 	rt = nd6_lookup(from, 0, ifp);
1492 	if (!rt) {
1493 #if 0
1494 		/* nothing must be done if there's no lladdr */
1495 		if (!lladdr || !lladdrlen)
1496 			return NULL;
1497 #endif
1498 
1499 		rt = nd6_lookup(from, 1, ifp);
1500 		is_newentry = 1;
1501 	} else {
1502 		/* do nothing if static ndp is set */
1503 		if (rt->rt_flags & RTF_STATIC)
1504 			return NULL;
1505 		is_newentry = 0;
1506 	}
1507 
1508 	if (!rt)
1509 		return NULL;
1510 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1511 fail:
1512 		(void)nd6_free(rt, 0);
1513 		return NULL;
1514 	}
1515 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1516 	if (!ln)
1517 		goto fail;
1518 	if (!rt->rt_gateway)
1519 		goto fail;
1520 	if (rt->rt_gateway->sa_family != AF_LINK)
1521 		goto fail;
1522 	sdl = SDL(rt->rt_gateway);
1523 
1524 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1525 	if (olladdr && lladdr) {
1526 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1527 			llchange = 1;
1528 		else
1529 			llchange = 0;
1530 	} else
1531 		llchange = 0;
1532 
1533 	/*
1534 	 * newentry olladdr  lladdr  llchange	(*=record)
1535 	 *	0	n	n	--	(1)
1536 	 *	0	y	n	--	(2)
1537 	 *	0	n	y	--	(3) * STALE
1538 	 *	0	y	y	n	(4) *
1539 	 *	0	y	y	y	(5) * STALE
1540 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1541 	 *	1	--	y	--	(7) * STALE
1542 	 */
1543 
1544 	if (lladdr) {		/* (3-5) and (7) */
1545 		/*
1546 		 * Record source link-layer address
1547 		 * XXX is it dependent to ifp->if_type?
1548 		 */
1549 		sdl->sdl_alen = ifp->if_addrlen;
1550 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1551 	}
1552 
1553 	if (!is_newentry) {
1554 		if ((!olladdr && lladdr) ||		/* (3) */
1555 		    (olladdr && lladdr && llchange)) {	/* (5) */
1556 			do_update = 1;
1557 			newstate = ND6_LLINFO_STALE;
1558 		} else					/* (1-2,4) */
1559 			do_update = 0;
1560 	} else {
1561 		do_update = 1;
1562 		if (!lladdr)				/* (6) */
1563 			newstate = ND6_LLINFO_NOSTATE;
1564 		else					/* (7) */
1565 			newstate = ND6_LLINFO_STALE;
1566 	}
1567 
1568 	if (do_update) {
1569 		/*
1570 		 * Update the state of the neighbor cache.
1571 		 */
1572 		ln->ln_state = newstate;
1573 
1574 		if (ln->ln_state == ND6_LLINFO_STALE) {
1575 			/*
1576 			 * XXX: since nd6_output() below will cause
1577 			 * state tansition to DELAY and reset the timer,
1578 			 * we must set the timer now, although it is actually
1579 			 * meaningless.
1580 			 */
1581 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1582 
1583 			if (ln->ln_hold) {
1584 				/*
1585 				 * we assume ifp is not a p2p here, so just
1586 				 * set the 2nd argument as the 1st one.
1587 				 */
1588 				nd6_output(ifp, ifp, ln->ln_hold,
1589 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1590 				ln->ln_hold = NULL;
1591 			}
1592 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1593 			/* probe right away */
1594 			nd6_llinfo_settimer((void *)ln, 0);
1595 		}
1596 	}
1597 
1598 	/*
1599 	 * ICMP6 type dependent behavior.
1600 	 *
1601 	 * NS: clear IsRouter if new entry
1602 	 * RS: clear IsRouter
1603 	 * RA: set IsRouter if there's lladdr
1604 	 * redir: clear IsRouter if new entry
1605 	 *
1606 	 * RA case, (1):
1607 	 * The spec says that we must set IsRouter in the following cases:
1608 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1609 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1610 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1611 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1612 	 * neighbor cache, this is similar to (6).
1613 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1614 	 *
1615 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1616 	 *							D R
1617 	 *	0	n	n	--	(1)	c   ?     s
1618 	 *	0	y	n	--	(2)	c   s     s
1619 	 *	0	n	y	--	(3)	c   s     s
1620 	 *	0	y	y	n	(4)	c   s     s
1621 	 *	0	y	y	y	(5)	c   s     s
1622 	 *	1	--	n	--	(6) c	c 	c s
1623 	 *	1	--	y	--	(7) c	c   s	c s
1624 	 *
1625 	 *					(c=clear s=set)
1626 	 */
1627 	switch (type & 0xff) {
1628 	case ND_NEIGHBOR_SOLICIT:
1629 		/*
1630 		 * New entry must have is_router flag cleared.
1631 		 */
1632 		if (is_newentry)	/* (6-7) */
1633 			ln->ln_router = 0;
1634 		break;
1635 	case ND_REDIRECT:
1636 		/*
1637 		 * If the icmp is a redirect to a better router, always set the
1638 		 * is_router flag.  Otherwise, if the entry is newly created,
1639 		 * clear the flag.  [RFC 2461, sec 8.3]
1640 		 */
1641 		if (code == ND_REDIRECT_ROUTER)
1642 			ln->ln_router = 1;
1643 		else if (is_newentry) /* (6-7) */
1644 			ln->ln_router = 0;
1645 		break;
1646 	case ND_ROUTER_SOLICIT:
1647 		/*
1648 		 * is_router flag must always be cleared.
1649 		 */
1650 		ln->ln_router = 0;
1651 		break;
1652 	case ND_ROUTER_ADVERT:
1653 		/*
1654 		 * Mark an entry with lladdr as a router.
1655 		 */
1656 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1657 		    (is_newentry && lladdr)) {			/* (7) */
1658 			ln->ln_router = 1;
1659 		}
1660 		break;
1661 	}
1662 
1663 	/*
1664 	 * When the link-layer address of a router changes, select the
1665 	 * best router again.  In particular, when the neighbor entry is newly
1666 	 * created, it might affect the selection policy.
1667 	 * Question: can we restrict the first condition to the "is_newentry"
1668 	 * case?
1669 	 * XXX: when we hear an RA from a new router with the link-layer
1670 	 * address option, defrouter_select() is called twice, since
1671 	 * defrtrlist_update called the function as well.  However, I believe
1672 	 * we can compromise the overhead, since it only happens the first
1673 	 * time.
1674 	 * XXX: although defrouter_select() should not have a bad effect
1675 	 * for those are not autoconfigured hosts, we explicitly avoid such
1676 	 * cases for safety.
1677 	 */
1678 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1679 		defrouter_select();
1680 
1681 	return rt;
1682 }
1683 
1684 static void
1685 nd6_slowtimo(ignored_arg)
1686     void *ignored_arg;
1687 {
1688 	int s = splsoftnet();
1689 	struct nd_ifinfo *nd6if;
1690 	struct ifnet *ifp;
1691 
1692 	timeout_set(&nd6_slowtimo_ch, nd6_slowtimo, NULL);
1693 	timeout_add(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz);
1694 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list))
1695 	{
1696 		nd6if = ND_IFINFO(ifp);
1697 		if (nd6if->basereachable && /* already initialized */
1698 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1699 			/*
1700 			 * Since reachable time rarely changes by router
1701 			 * advertisements, we SHOULD insure that a new random
1702 			 * value gets recomputed at least once every few hours.
1703 			 * (RFC 2461, 6.3.4)
1704 			 */
1705 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1706 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1707 		}
1708 	}
1709 	splx(s);
1710 }
1711 
1712 #define senderr(e) { error = (e); goto bad;}
1713 int
1714 nd6_output(ifp, origifp, m0, dst, rt0)
1715 	struct ifnet *ifp;
1716 	struct ifnet *origifp;
1717 	struct mbuf *m0;
1718 	struct sockaddr_in6 *dst;
1719 	struct rtentry *rt0;
1720 {
1721 	struct mbuf *m = m0;
1722 	struct rtentry *rt = rt0;
1723 	struct sockaddr_in6 *gw6 = NULL;
1724 	struct llinfo_nd6 *ln = NULL;
1725 	int error = 0;
1726 #ifdef IPSEC
1727 	struct m_tag *mtag;
1728 #endif /* IPSEC */
1729 
1730 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1731 		goto sendpkt;
1732 
1733 	if (nd6_need_cache(ifp) == 0)
1734 		goto sendpkt;
1735 
1736 	/*
1737 	 * next hop determination.  This routine is derived from ether_outpout.
1738 	 */
1739 	if (rt) {
1740 		if ((rt->rt_flags & RTF_UP) == 0) {
1741 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst,
1742 			    1)) != NULL)
1743 			{
1744 				rt->rt_refcnt--;
1745 				if (rt->rt_ifp != ifp) {
1746 					/* XXX: loop care? */
1747 					return nd6_output(ifp, origifp, m0,
1748 					    dst, rt);
1749 				}
1750 			} else
1751 				senderr(EHOSTUNREACH);
1752 		}
1753 
1754 		if (rt->rt_flags & RTF_GATEWAY) {
1755 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1756 
1757 			/*
1758 			 * We skip link-layer address resolution and NUD
1759 			 * if the gateway is not a neighbor from ND point
1760 			 * of view, regardless of the value of nd_ifinfo.flags.
1761 			 * The second condition is a bit tricky; we skip
1762 			 * if the gateway is our own address, which is
1763 			 * sometimes used to install a route to a p2p link.
1764 			 */
1765 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1766 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1767 				/*
1768 				 * We allow this kind of tricky route only
1769 				 * when the outgoing interface is p2p.
1770 				 * XXX: we may need a more generic rule here.
1771 				 */
1772 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1773 					senderr(EHOSTUNREACH);
1774 
1775 				goto sendpkt;
1776 			}
1777 
1778 			if (rt->rt_gwroute == 0)
1779 				goto lookup;
1780 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1781 				rtfree(rt); rt = rt0;
1782 			lookup:
1783 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1784 				if ((rt = rt->rt_gwroute) == 0)
1785 					senderr(EHOSTUNREACH);
1786 			}
1787 		}
1788 	}
1789 
1790 	/*
1791 	 * Address resolution or Neighbor Unreachability Detection
1792 	 * for the next hop.
1793 	 * At this point, the destination of the packet must be a unicast
1794 	 * or an anycast address(i.e. not a multicast).
1795 	 */
1796 
1797 	/* Look up the neighbor cache for the nexthop */
1798 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1799 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1800 	else {
1801 		/*
1802 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1803 		 * the condition below is not very efficient.  But we believe
1804 		 * it is tolerable, because this should be a rare case.
1805 		 */
1806 		if (nd6_is_addr_neighbor(dst, ifp) &&
1807 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1808 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1809 	}
1810 	if (!ln || !rt) {
1811 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1812 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1813 			log(LOG_DEBUG,
1814 			    "nd6_output: can't allocate llinfo for %s "
1815 			    "(ln=%p, rt=%p)\n",
1816 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1817 			senderr(EIO);	/* XXX: good error? */
1818 		}
1819 
1820 		goto sendpkt;	/* send anyway */
1821 	}
1822 
1823 	/* We don't have to do link-layer address resolution on a p2p link. */
1824 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1825 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1826 		ln->ln_state = ND6_LLINFO_STALE;
1827 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1828 	}
1829 
1830 	/*
1831 	 * The first time we send a packet to a neighbor whose entry is
1832 	 * STALE, we have to change the state to DELAY and a sets a timer to
1833 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1834 	 * neighbor unreachability detection on expiration.
1835 	 * (RFC 2461 7.3.3)
1836 	 */
1837 	if (ln->ln_state == ND6_LLINFO_STALE) {
1838 		ln->ln_asked = 0;
1839 		ln->ln_state = ND6_LLINFO_DELAY;
1840 		nd6_llinfo_settimer(ln, nd6_delay * hz);
1841 	}
1842 
1843 	/*
1844 	 * If the neighbor cache entry has a state other than INCOMPLETE
1845 	 * (i.e. its link-layer address is already resolved), just
1846 	 * send the packet.
1847 	 */
1848 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1849 		goto sendpkt;
1850 
1851 	/*
1852 	 * There is a neighbor cache entry, but no ethernet address
1853 	 * response yet.  Replace the held mbuf (if any) with this
1854 	 * latest one.
1855 	 */
1856 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1857 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1858 	if (ln->ln_hold)
1859 		m_freem(ln->ln_hold);
1860 	ln->ln_hold = m;
1861 	/*
1862 	 * If there has been no NS for the neighbor after entering the
1863 	 * INCOMPLETE state, send the first solicitation.
1864 	 */
1865 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
1866 		ln->ln_asked++;
1867 		nd6_llinfo_settimer(ln,
1868 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1869 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1870 	}
1871 	return (0);
1872 
1873   sendpkt:
1874 #ifdef IPSEC
1875 	/*
1876 	 * If the packet needs outgoing IPsec crypto processing and the
1877 	 * interface doesn't support it, drop it.
1878 	 */
1879 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
1880 #endif /* IPSEC */
1881 
1882 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1883 #ifdef IPSEC
1884 		if (mtag != NULL &&
1885 		    (origifp->if_capabilities & IFCAP_IPSEC) == 0) {
1886 			/* Tell IPsec to do its own crypto. */
1887 			ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
1888 			error = EACCES;
1889 			goto bad;
1890 		}
1891 #endif /* IPSEC */
1892 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1893 		    rt));
1894 	}
1895 #ifdef IPSEC
1896 	if (mtag != NULL &&
1897 	    (ifp->if_capabilities & IFCAP_IPSEC) == 0) {
1898 		/* Tell IPsec to do its own crypto. */
1899 		ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1));
1900 		error = EACCES;
1901 		goto bad;
1902 	}
1903 #endif /* IPSEC */
1904 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1905 
1906   bad:
1907 	if (m)
1908 		m_freem(m);
1909 	return (error);
1910 }
1911 #undef senderr
1912 
1913 int
1914 nd6_need_cache(ifp)
1915 	struct ifnet *ifp;
1916 {
1917 	/*
1918 	 * XXX: we currently do not make neighbor cache on any interface
1919 	 * other than ARCnet, Ethernet, FDDI and GIF.
1920 	 *
1921 	 * RFC2893 says:
1922 	 * - unidirectional tunnels needs no ND
1923 	 */
1924 	switch (ifp->if_type) {
1925 	case IFT_ARCNET:
1926 	case IFT_ETHER:
1927 	case IFT_FDDI:
1928 	case IFT_IEEE1394:
1929 	case IFT_PROPVIRTUAL:
1930 	case IFT_L2VLAN:
1931 	case IFT_IEEE80211:
1932 	case IFT_GIF:		/* XXX need more cases? */
1933 		return (1);
1934 	default:
1935 		return (0);
1936 	}
1937 }
1938 
1939 int
1940 nd6_storelladdr(ifp, rt, m, dst, desten)
1941 	struct ifnet *ifp;
1942 	struct rtentry *rt;
1943 	struct mbuf *m;
1944 	struct sockaddr *dst;
1945 	u_char *desten;
1946 {
1947 	struct sockaddr_dl *sdl;
1948 
1949 	if (m->m_flags & M_MCAST) {
1950 		switch (ifp->if_type) {
1951 		case IFT_ETHER:
1952 		case IFT_FDDI:
1953 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1954 						 desten);
1955 			return (1);
1956 			break;
1957 		case IFT_ARCNET:
1958 			*desten = 0;
1959 			return (1);
1960 		default:
1961 			m_freem(m);
1962 			return (0);
1963 		}
1964 	}
1965 
1966 	if (rt == NULL) {
1967 		/* this could happen, if we could not allocate memory */
1968 		m_freem(m);
1969 		return (0);
1970 	}
1971 	if (rt->rt_gateway->sa_family != AF_LINK) {
1972 		printf("nd6_storelladdr: something odd happens\n");
1973 		m_freem(m);
1974 		return (0);
1975 	}
1976 	sdl = SDL(rt->rt_gateway);
1977 	if (sdl->sdl_alen == 0) {
1978 		/* this should be impossible, but we bark here for debugging */
1979 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
1980 		    ip6_sprintf(&SIN6(dst)->sin6_addr), ifp->if_xname);
1981 		m_freem(m);
1982 		return (0);
1983 	}
1984 
1985 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
1986 	return (1);
1987 }
1988 
1989 int
1990 nd6_sysctl(name, oldp, oldlenp, newp, newlen)
1991 	int name;
1992 	void *oldp;	/* syscall arg, need copyout */
1993 	size_t *oldlenp;
1994 	void *newp;	/* syscall arg, need copyin */
1995 	size_t newlen;
1996 {
1997 	void *p;
1998 	size_t ol, l;
1999 	int error;
2000 
2001 	error = 0;
2002 	l = 0;
2003 
2004 	if (newp)
2005 		return EPERM;
2006 	if (oldp && !oldlenp)
2007 		return EINVAL;
2008 	ol = oldlenp ? *oldlenp : 0;
2009 
2010 	if (oldp) {
2011 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2012 		if (!p)
2013 			return ENOMEM;
2014 	} else
2015 		p = NULL;
2016 	switch (name) {
2017 	case ICMPV6CTL_ND6_DRLIST:
2018 		error = fill_drlist(p, oldlenp, ol);
2019 		if (!error && p && oldp)
2020 			error = copyout(p, oldp, *oldlenp);
2021 		break;
2022 
2023 	case ICMPV6CTL_ND6_PRLIST:
2024 		error = fill_prlist(p, oldlenp, ol);
2025 		if (!error && p && oldp)
2026 			error = copyout(p, oldp, *oldlenp);
2027 		break;
2028 
2029 	default:
2030 		error = ENOPROTOOPT;
2031 		break;
2032 	}
2033 	if (p)
2034 		free(p, M_TEMP);
2035 
2036 	return (error);
2037 }
2038 
2039 static int
2040 fill_drlist(oldp, oldlenp, ol)
2041 	void *oldp;
2042 	size_t *oldlenp, ol;
2043 {
2044 	int error = 0, s;
2045 	struct in6_defrouter *d = NULL, *de = NULL;
2046 	struct nd_defrouter *dr;
2047 	size_t l;
2048 
2049 	s = splsoftnet();
2050 
2051 	if (oldp) {
2052 		d = (struct in6_defrouter *)oldp;
2053 		de = (struct in6_defrouter *)((caddr_t)oldp + *oldlenp);
2054 	}
2055 	l = 0;
2056 
2057 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2058 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2059 
2060 		if (oldp && d + 1 <= de) {
2061 			bzero(d, sizeof(*d));
2062 			d->rtaddr.sin6_family = AF_INET6;
2063 			d->rtaddr.sin6_len = sizeof(struct sockaddr_in6);
2064 			d->rtaddr.sin6_addr = dr->rtaddr;
2065 			in6_recoverscope(&d->rtaddr, &d->rtaddr.sin6_addr,
2066 			    dr->ifp);
2067 			d->flags = dr->flags;
2068 			d->rtlifetime = dr->rtlifetime;
2069 			d->expire = dr->expire;
2070 			d->if_index = dr->ifp->if_index;
2071 		}
2072 
2073 		l += sizeof(*d);
2074 		if (d)
2075 			d++;
2076 	}
2077 
2078 	if (oldp) {
2079 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2080 		if (l > ol)
2081 			error = ENOMEM;
2082 	} else
2083 		*oldlenp = l;
2084 
2085 	splx(s);
2086 
2087 	return (error);
2088 }
2089 
2090 static int
2091 fill_prlist(oldp, oldlenp, ol)
2092 	void *oldp;
2093 	size_t *oldlenp, ol;
2094 {
2095 	int error = 0, s;
2096 	struct nd_prefix *pr;
2097 	struct in6_prefix *p = NULL;
2098 	struct in6_prefix *pe = NULL;
2099 	size_t l;
2100 
2101 	s = splsoftnet();
2102 
2103 	if (oldp) {
2104 		p = (struct in6_prefix *)oldp;
2105 		pe = (struct in6_prefix *)((caddr_t)oldp + *oldlenp);
2106 	}
2107 	l = 0;
2108 
2109 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2110 		u_short advrtrs;
2111 		size_t advance;
2112 		struct sockaddr_in6 *sin6;
2113 		struct sockaddr_in6 *s6;
2114 		struct nd_pfxrouter *pfr;
2115 
2116 		if (oldp && p + 1 <= pe)
2117 		{
2118 			bzero(p, sizeof(*p));
2119 			sin6 = (struct sockaddr_in6 *)(p + 1);
2120 
2121 			p->prefix = pr->ndpr_prefix;
2122 			if (in6_recoverscope(&p->prefix,
2123 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2124 				log(LOG_ERR,
2125 				    "scope error in prefix list (%s)\n",
2126 				    ip6_sprintf(&p->prefix.sin6_addr));
2127 			p->raflags = pr->ndpr_raf;
2128 			p->prefixlen = pr->ndpr_plen;
2129 			p->vltime = pr->ndpr_vltime;
2130 			p->pltime = pr->ndpr_pltime;
2131 			p->if_index = pr->ndpr_ifp->if_index;
2132 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2133 				p->expire = 0;
2134 			else {
2135 				time_t maxexpire;
2136 
2137 				/* XXX: we assume time_t is signed. */
2138 				maxexpire = (-1) &
2139 					~(1 << ((sizeof(maxexpire) * 8) - 1));
2140 				if (pr->ndpr_vltime <
2141 				    maxexpire - pr->ndpr_lastupdate) {
2142 					p->expire = pr->ndpr_lastupdate +
2143 						pr->ndpr_vltime;
2144 				} else
2145 					p->expire = maxexpire;
2146 			}
2147 			p->refcnt = pr->ndpr_refcnt;
2148 			p->flags = pr->ndpr_stateflags;
2149 			p->origin = PR_ORIG_RA;
2150 			advrtrs = 0;
2151 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2152 			     pfr = pfr->pfr_next) {
2153 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2154 					advrtrs++;
2155 					continue;
2156 				}
2157 				s6 = &sin6[advrtrs];
2158 				s6->sin6_family = AF_INET6;
2159 				s6->sin6_len = sizeof(struct sockaddr_in6);
2160 				s6->sin6_addr = pfr->router->rtaddr;
2161 				in6_recoverscope(s6, &pfr->router->rtaddr,
2162 				    pfr->router->ifp);
2163 				advrtrs++;
2164 			}
2165 			p->advrtrs = advrtrs;
2166 		}
2167 		else {
2168 			advrtrs = 0;
2169 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2170 			     pfr = pfr->pfr_next)
2171 				advrtrs++;
2172 		}
2173 
2174 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2175 		l += advance;
2176 		if (p)
2177 			p = (struct in6_prefix *)((caddr_t)p + advance);
2178 	}
2179 
2180 	if (oldp) {
2181 		*oldlenp = l;	/* (caddr_t)d - (caddr_t)oldp */
2182 		if (l > ol)
2183 			error = ENOMEM;
2184 	} else
2185 		*oldlenp = l;
2186 
2187 	splx(s);
2188 
2189 	return (error);
2190 }
2191