xref: /freebsd/sys/netinet6/nd6.c (revision 74bf4e164ba5851606a27d4feff27717452583e5)
1 /*	$FreeBSD$	*/
2 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/callout.h>
40 #include <sys/mac.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/queue.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/if.h>
54 #include <net/if_arc.h>
55 #include <net/if_dl.h>
56 #include <net/if_types.h>
57 #include <net/if_atm.h>
58 #include <net/iso88025.h>
59 #include <net/fddi.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69 
70 #include <net/net_osdep.h>
71 
72 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
73 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
74 
75 #define SIN6(s) ((struct sockaddr_in6 *)s)
76 #define SDL(s) ((struct sockaddr_dl *)s)
77 
78 /* timer values */
79 int	nd6_prune	= 1;	/* walk list every 1 seconds */
80 int	nd6_delay	= 5;	/* delay first probe time 5 second */
81 int	nd6_umaxtries	= 3;	/* maximum unicast query */
82 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
83 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
84 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
85 
86 /* preventing too many loops in ND option parsing */
87 int nd6_maxndopt = 10;	/* max # of ND options allowed */
88 
89 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
90 
91 #ifdef ND6_DEBUG
92 int nd6_debug = 1;
93 #else
94 int nd6_debug = 0;
95 #endif
96 
97 /* for debugging? */
98 static int nd6_inuse, nd6_allocated;
99 
100 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
101 struct nd_drhead nd_defrouter;
102 struct nd_prhead nd_prefix = { 0 };
103 
104 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
105 static struct sockaddr_in6 all1_sa;
106 
107 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
108 static void nd6_slowtimo __P((void *));
109 static int regen_tmpaddr __P((struct in6_ifaddr *));
110 
111 struct callout nd6_slowtimo_ch;
112 struct callout nd6_timer_ch;
113 extern struct callout in6_tmpaddrtimer_ch;
114 
115 void
116 nd6_init()
117 {
118 	static int nd6_init_done = 0;
119 	int i;
120 
121 	if (nd6_init_done) {
122 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
123 		return;
124 	}
125 
126 	all1_sa.sin6_family = AF_INET6;
127 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
128 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
129 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
130 
131 	/* initialization of the default router list */
132 	TAILQ_INIT(&nd_defrouter);
133 
134 	nd6_init_done = 1;
135 
136 	/* start timer */
137 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
138 	    nd6_slowtimo, NULL);
139 }
140 
141 struct nd_ifinfo *
142 nd6_ifattach(ifp)
143 	struct ifnet *ifp;
144 {
145 	struct nd_ifinfo *nd;
146 
147 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
148 	bzero(nd, sizeof(*nd));
149 
150 	nd->initialized = 1;
151 
152 	nd->chlim = IPV6_DEFHLIM;
153 	nd->basereachable = REACHABLE_TIME;
154 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
155 	nd->retrans = RETRANS_TIMER;
156 	/*
157 	 * Note that the default value of ip6_accept_rtadv is 0, which means
158 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
159 	 * here.
160 	 */
161 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
162 
163 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
164 	nd6_setmtu0(ifp, nd);
165 
166 	return nd;
167 }
168 
169 void
170 nd6_ifdetach(nd)
171 	struct nd_ifinfo *nd;
172 {
173 
174 	free(nd, M_IP6NDP);
175 }
176 
177 /*
178  * Reset ND level link MTU. This function is called when the physical MTU
179  * changes, which means we might have to adjust the ND level MTU.
180  */
181 void
182 nd6_setmtu(ifp)
183 	struct ifnet *ifp;
184 {
185 
186 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
187 }
188 
189 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
190 void
191 nd6_setmtu0(ifp, ndi)
192 	struct ifnet *ifp;
193 	struct nd_ifinfo *ndi;
194 {
195 	u_int32_t omaxmtu;
196 
197 	omaxmtu = ndi->maxmtu;
198 
199 	switch (ifp->if_type) {
200 	case IFT_ARCNET:
201 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
202 		break;
203 	case IFT_ETHER:
204 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
205 		break;
206 	case IFT_FDDI:
207 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
208 		break;
209 	case IFT_ATM:
210 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
211 		break;
212 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
213 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
214 		break;
215 #ifdef IFT_IEEE80211
216 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
217 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
218 		break;
219 #endif
220 	 case IFT_ISO88025:
221 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
222 		 break;
223 	default:
224 		ndi->maxmtu = ifp->if_mtu;
225 		break;
226 	}
227 
228 	/*
229 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
230 	 * undesirable situation.  We thus notify the operator of the change
231 	 * explicitly.  The check for omaxmtu is necessary to restrict the
232 	 * log to the case of changing the MTU, not initializing it.
233 	 */
234 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
235 		log(LOG_NOTICE, "nd6_setmtu0: "
236 		    "new link MTU on %s (%lu) is too small for IPv6\n",
237 		    if_name(ifp), (unsigned long)ndi->maxmtu);
238 	}
239 
240 	if (ndi->maxmtu > in6_maxmtu)
241 		in6_setmaxmtu(); /* check all interfaces just in case */
242 
243 #undef MIN
244 }
245 
246 void
247 nd6_option_init(opt, icmp6len, ndopts)
248 	void *opt;
249 	int icmp6len;
250 	union nd_opts *ndopts;
251 {
252 
253 	bzero(ndopts, sizeof(*ndopts));
254 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
255 	ndopts->nd_opts_last
256 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
257 
258 	if (icmp6len == 0) {
259 		ndopts->nd_opts_done = 1;
260 		ndopts->nd_opts_search = NULL;
261 	}
262 }
263 
264 /*
265  * Take one ND option.
266  */
267 struct nd_opt_hdr *
268 nd6_option(ndopts)
269 	union nd_opts *ndopts;
270 {
271 	struct nd_opt_hdr *nd_opt;
272 	int olen;
273 
274 	if (!ndopts)
275 		panic("ndopts == NULL in nd6_option");
276 	if (!ndopts->nd_opts_last)
277 		panic("uninitialized ndopts in nd6_option");
278 	if (!ndopts->nd_opts_search)
279 		return NULL;
280 	if (ndopts->nd_opts_done)
281 		return NULL;
282 
283 	nd_opt = ndopts->nd_opts_search;
284 
285 	/* make sure nd_opt_len is inside the buffer */
286 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
287 		bzero(ndopts, sizeof(*ndopts));
288 		return NULL;
289 	}
290 
291 	olen = nd_opt->nd_opt_len << 3;
292 	if (olen == 0) {
293 		/*
294 		 * Message validation requires that all included
295 		 * options have a length that is greater than zero.
296 		 */
297 		bzero(ndopts, sizeof(*ndopts));
298 		return NULL;
299 	}
300 
301 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
302 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
303 		/* option overruns the end of buffer, invalid */
304 		bzero(ndopts, sizeof(*ndopts));
305 		return NULL;
306 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
307 		/* reached the end of options chain */
308 		ndopts->nd_opts_done = 1;
309 		ndopts->nd_opts_search = NULL;
310 	}
311 	return nd_opt;
312 }
313 
314 /*
315  * Parse multiple ND options.
316  * This function is much easier to use, for ND routines that do not need
317  * multiple options of the same type.
318  */
319 int
320 nd6_options(ndopts)
321 	union nd_opts *ndopts;
322 {
323 	struct nd_opt_hdr *nd_opt;
324 	int i = 0;
325 
326 	if (!ndopts)
327 		panic("ndopts == NULL in nd6_options");
328 	if (!ndopts->nd_opts_last)
329 		panic("uninitialized ndopts in nd6_options");
330 	if (!ndopts->nd_opts_search)
331 		return 0;
332 
333 	while (1) {
334 		nd_opt = nd6_option(ndopts);
335 		if (!nd_opt && !ndopts->nd_opts_last) {
336 			/*
337 			 * Message validation requires that all included
338 			 * options have a length that is greater than zero.
339 			 */
340 			icmp6stat.icp6s_nd_badopt++;
341 			bzero(ndopts, sizeof(*ndopts));
342 			return -1;
343 		}
344 
345 		if (!nd_opt)
346 			goto skip1;
347 
348 		switch (nd_opt->nd_opt_type) {
349 		case ND_OPT_SOURCE_LINKADDR:
350 		case ND_OPT_TARGET_LINKADDR:
351 		case ND_OPT_MTU:
352 		case ND_OPT_REDIRECTED_HEADER:
353 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
354 				nd6log((LOG_INFO,
355 				    "duplicated ND6 option found (type=%d)\n",
356 				    nd_opt->nd_opt_type));
357 				/* XXX bark? */
358 			} else {
359 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
360 					= nd_opt;
361 			}
362 			break;
363 		case ND_OPT_PREFIX_INFORMATION:
364 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
365 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
366 					= nd_opt;
367 			}
368 			ndopts->nd_opts_pi_end =
369 				(struct nd_opt_prefix_info *)nd_opt;
370 			break;
371 		default:
372 			/*
373 			 * Unknown options must be silently ignored,
374 			 * to accomodate future extension to the protocol.
375 			 */
376 			nd6log((LOG_DEBUG,
377 			    "nd6_options: unsupported option %d - "
378 			    "option ignored\n", nd_opt->nd_opt_type));
379 		}
380 
381 skip1:
382 		i++;
383 		if (i > nd6_maxndopt) {
384 			icmp6stat.icp6s_nd_toomanyopt++;
385 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
386 			break;
387 		}
388 
389 		if (ndopts->nd_opts_done)
390 			break;
391 	}
392 
393 	return 0;
394 }
395 
396 /*
397  * ND6 timer routine to expire default route list and prefix list
398  */
399 void
400 nd6_timer(ignored_arg)
401 	void	*ignored_arg;
402 {
403 	int s;
404 	struct llinfo_nd6 *ln;
405 	struct nd_defrouter *dr;
406 	struct nd_prefix *pr;
407 	struct ifnet *ifp;
408 	struct in6_ifaddr *ia6, *nia6;
409 	struct in6_addrlifetime *lt6;
410 
411 	s = splnet();
412 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
413 	    nd6_timer, NULL);
414 
415 	ln = llinfo_nd6.ln_next;
416 	while (ln && ln != &llinfo_nd6) {
417 		struct rtentry *rt;
418 		struct sockaddr_in6 *dst;
419 		struct llinfo_nd6 *next = ln->ln_next;
420 		/* XXX: used for the DELAY case only: */
421 		struct nd_ifinfo *ndi = NULL;
422 
423 		if ((rt = ln->ln_rt) == NULL) {
424 			ln = next;
425 			continue;
426 		}
427 		if ((ifp = rt->rt_ifp) == NULL) {
428 			ln = next;
429 			continue;
430 		}
431 		ndi = ND_IFINFO(ifp);
432 		dst = (struct sockaddr_in6 *)rt_key(rt);
433 
434 		if (ln->ln_expire > time_second) {
435 			ln = next;
436 			continue;
437 		}
438 
439 		/* sanity check */
440 		if (!rt)
441 			panic("rt=0 in nd6_timer(ln=%p)", ln);
442 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
443 			panic("rt_llinfo(%p) is not equal to ln(%p)",
444 			      rt->rt_llinfo, ln);
445 		if (!dst)
446 			panic("dst=0 in nd6_timer(ln=%p)", ln);
447 
448 		switch (ln->ln_state) {
449 		case ND6_LLINFO_INCOMPLETE:
450 			if (ln->ln_asked < nd6_mmaxtries) {
451 				ln->ln_asked++;
452 				ln->ln_expire = time_second +
453 					ND_IFINFO(ifp)->retrans / 1000;
454 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
455 					ln, 0);
456 			} else {
457 				struct mbuf *m = ln->ln_hold;
458 				if (m) {
459 					if (rt->rt_ifp) {
460 						/*
461 						 * Fake rcvif to make ICMP error
462 						 * more helpful in diagnosing
463 						 * for the receiver.
464 						 * XXX: should we consider
465 						 * older rcvif?
466 						 */
467 						m->m_pkthdr.rcvif = rt->rt_ifp;
468 					}
469 					icmp6_error(m, ICMP6_DST_UNREACH,
470 						    ICMP6_DST_UNREACH_ADDR, 0);
471 					ln->ln_hold = NULL;
472 				}
473 				next = nd6_free(rt);
474 			}
475 			break;
476 		case ND6_LLINFO_REACHABLE:
477 			if (ln->ln_expire) {
478 				ln->ln_state = ND6_LLINFO_STALE;
479 				ln->ln_expire = time_second + nd6_gctimer;
480 			}
481 			break;
482 
483 		case ND6_LLINFO_STALE:
484 			/* Garbage Collection(RFC 2461 5.3) */
485 			if (ln->ln_expire)
486 				next = nd6_free(rt);
487 			break;
488 
489 		case ND6_LLINFO_DELAY:
490 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
491 				/* We need NUD */
492 				ln->ln_asked = 1;
493 				ln->ln_state = ND6_LLINFO_PROBE;
494 				ln->ln_expire = time_second +
495 					ndi->retrans / 1000;
496 				nd6_ns_output(ifp, &dst->sin6_addr,
497 					      &dst->sin6_addr,
498 					      ln, 0);
499 			} else {
500 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
501 				ln->ln_expire = time_second + nd6_gctimer;
502 			}
503 			break;
504 		case ND6_LLINFO_PROBE:
505 			if (ln->ln_asked < nd6_umaxtries) {
506 				ln->ln_asked++;
507 				ln->ln_expire = time_second +
508 					ND_IFINFO(ifp)->retrans / 1000;
509 				nd6_ns_output(ifp, &dst->sin6_addr,
510 					       &dst->sin6_addr, ln, 0);
511 			} else {
512 				next = nd6_free(rt);
513 			}
514 			break;
515 		}
516 		ln = next;
517 	}
518 
519 	/* expire default router list */
520 	dr = TAILQ_FIRST(&nd_defrouter);
521 	while (dr) {
522 		if (dr->expire && dr->expire < time_second) {
523 			struct nd_defrouter *t;
524 			t = TAILQ_NEXT(dr, dr_entry);
525 			defrtrlist_del(dr);
526 			dr = t;
527 		} else {
528 			dr = TAILQ_NEXT(dr, dr_entry);
529 		}
530 	}
531 
532 	/*
533 	 * expire interface addresses.
534 	 * in the past the loop was inside prefix expiry processing.
535 	 * However, from a stricter speci-confrmance standpoint, we should
536 	 * rather separate address lifetimes and prefix lifetimes.
537 	 */
538   addrloop:
539 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
540 		nia6 = ia6->ia_next;
541 		/* check address lifetime */
542 		lt6 = &ia6->ia6_lifetime;
543 		if (IFA6_IS_INVALID(ia6)) {
544 			int regen = 0;
545 
546 			/*
547 			 * If the expiring address is temporary, try
548 			 * regenerating a new one.  This would be useful when
549 			 * we suspended a laptop PC, then turned it on after a
550 			 * period that could invalidate all temporary
551 			 * addresses.  Although we may have to restart the
552 			 * loop (see below), it must be after purging the
553 			 * address.  Otherwise, we'd see an infinite loop of
554 			 * regeneration.
555 			 */
556 			if (ip6_use_tempaddr &&
557 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
558 				if (regen_tmpaddr(ia6) == 0)
559 					regen = 1;
560 			}
561 
562 			in6_purgeaddr(&ia6->ia_ifa);
563 
564 			if (regen)
565 				goto addrloop; /* XXX: see below */
566 		}
567 		if (IFA6_IS_DEPRECATED(ia6)) {
568 			int oldflags = ia6->ia6_flags;
569 
570 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
571 
572 			/*
573 			 * If a temporary address has just become deprecated,
574 			 * regenerate a new one if possible.
575 			 */
576 			if (ip6_use_tempaddr &&
577 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
578 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
579 
580 				if (regen_tmpaddr(ia6) == 0) {
581 					/*
582 					 * A new temporary address is
583 					 * generated.
584 					 * XXX: this means the address chain
585 					 * has changed while we are still in
586 					 * the loop.  Although the change
587 					 * would not cause disaster (because
588 					 * it's not a deletion, but an
589 					 * addition,) we'd rather restart the
590 					 * loop just for safety.  Or does this
591 					 * significantly reduce performance??
592 					 */
593 					goto addrloop;
594 				}
595 			}
596 		} else {
597 			/*
598 			 * A new RA might have made a deprecated address
599 			 * preferred.
600 			 */
601 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
602 		}
603 	}
604 
605 	/* expire prefix list */
606 	pr = nd_prefix.lh_first;
607 	while (pr) {
608 		/*
609 		 * check prefix lifetime.
610 		 * since pltime is just for autoconf, pltime processing for
611 		 * prefix is not necessary.
612 		 */
613 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
614 			struct nd_prefix *t;
615 			t = pr->ndpr_next;
616 
617 			/*
618 			 * address expiration and prefix expiration are
619 			 * separate.  NEVER perform in6_purgeaddr here.
620 			 */
621 
622 			prelist_remove(pr);
623 			pr = t;
624 		} else
625 			pr = pr->ndpr_next;
626 	}
627 	splx(s);
628 }
629 
630 static int
631 regen_tmpaddr(ia6)
632 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
633 {
634 	struct ifaddr *ifa;
635 	struct ifnet *ifp;
636 	struct in6_ifaddr *public_ifa6 = NULL;
637 
638 	ifp = ia6->ia_ifa.ifa_ifp;
639 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
640 	     ifa = ifa->ifa_list.tqe_next) {
641 		struct in6_ifaddr *it6;
642 
643 		if (ifa->ifa_addr->sa_family != AF_INET6)
644 			continue;
645 
646 		it6 = (struct in6_ifaddr *)ifa;
647 
648 		/* ignore no autoconf addresses. */
649 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
650 			continue;
651 
652 		/* ignore autoconf addresses with different prefixes. */
653 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
654 			continue;
655 
656 		/*
657 		 * Now we are looking at an autoconf address with the same
658 		 * prefix as ours.  If the address is temporary and is still
659 		 * preferred, do not create another one.  It would be rare, but
660 		 * could happen, for example, when we resume a laptop PC after
661 		 * a long period.
662 		 */
663 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
664 		    !IFA6_IS_DEPRECATED(it6)) {
665 			public_ifa6 = NULL;
666 			break;
667 		}
668 
669 		/*
670 		 * This is a public autoconf address that has the same prefix
671 		 * as ours.  If it is preferred, keep it.  We can't break the
672 		 * loop here, because there may be a still-preferred temporary
673 		 * address with the prefix.
674 		 */
675 		if (!IFA6_IS_DEPRECATED(it6))
676 		    public_ifa6 = it6;
677 	}
678 
679 	if (public_ifa6 != NULL) {
680 		int e;
681 
682 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
683 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
684 			    " tmp addr,errno=%d\n", e);
685 			return (-1);
686 		}
687 		return (0);
688 	}
689 
690 	return (-1);
691 }
692 
693 /*
694  * Nuke neighbor cache/prefix/default router management table, right before
695  * ifp goes away.
696  */
697 void
698 nd6_purge(ifp)
699 	struct ifnet *ifp;
700 {
701 	struct llinfo_nd6 *ln, *nln;
702 	struct nd_defrouter *dr, *ndr, drany;
703 	struct nd_prefix *pr, *npr;
704 
705 	/* Nuke default router list entries toward ifp */
706 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
707 		/*
708 		 * The first entry of the list may be stored in
709 		 * the routing table, so we'll delete it later.
710 		 */
711 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
712 			ndr = TAILQ_NEXT(dr, dr_entry);
713 			if (dr->ifp == ifp)
714 				defrtrlist_del(dr);
715 		}
716 		dr = TAILQ_FIRST(&nd_defrouter);
717 		if (dr->ifp == ifp)
718 			defrtrlist_del(dr);
719 	}
720 
721 	/* Nuke prefix list entries toward ifp */
722 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
723 		npr = pr->ndpr_next;
724 		if (pr->ndpr_ifp == ifp) {
725 			/*
726 			 * Previously, pr->ndpr_addr is removed as well,
727 			 * but I strongly believe we don't have to do it.
728 			 * nd6_purge() is only called from in6_ifdetach(),
729 			 * which removes all the associated interface addresses
730 			 * by itself.
731 			 * (jinmei@kame.net 20010129)
732 			 */
733 			prelist_remove(pr);
734 		}
735 	}
736 
737 	/* cancel default outgoing interface setting */
738 	if (nd6_defifindex == ifp->if_index)
739 		nd6_setdefaultiface(0);
740 
741 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
742 		/* refresh default router list */
743 		bzero(&drany, sizeof(drany));
744 		defrouter_delreq(&drany, 0);
745 		defrouter_select();
746 	}
747 
748 	/*
749 	 * Nuke neighbor cache entries for the ifp.
750 	 * Note that rt->rt_ifp may not be the same as ifp,
751 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
752 	 * nd6_rtrequest(), and ip6_input().
753 	 */
754 	ln = llinfo_nd6.ln_next;
755 	while (ln && ln != &llinfo_nd6) {
756 		struct rtentry *rt;
757 		struct sockaddr_dl *sdl;
758 
759 		nln = ln->ln_next;
760 		rt = ln->ln_rt;
761 		if (rt && rt->rt_gateway &&
762 		    rt->rt_gateway->sa_family == AF_LINK) {
763 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
764 			if (sdl->sdl_index == ifp->if_index)
765 				nln = nd6_free(rt);
766 		}
767 		ln = nln;
768 	}
769 }
770 
771 struct rtentry *
772 nd6_lookup(addr6, create, ifp)
773 	struct in6_addr *addr6;
774 	int create;
775 	struct ifnet *ifp;
776 {
777 	struct rtentry *rt;
778 	struct sockaddr_in6 sin6;
779 
780 	bzero(&sin6, sizeof(sin6));
781 	sin6.sin6_len = sizeof(struct sockaddr_in6);
782 	sin6.sin6_family = AF_INET6;
783 	sin6.sin6_addr = *addr6;
784 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
785 	if (rt) {
786 		if ((rt->rt_flags & RTF_LLINFO) == 0 && create) {
787 			/*
788 			 * This is the case for the default route.
789 			 * If we want to create a neighbor cache for the
790 			 * address, we should free the route for the
791 			 * destination and allocate an interface route.
792 			 */
793 			RTFREE_LOCKED(rt);
794 			rt = 0;
795 		}
796 	}
797 	if (!rt) {
798 		if (create && ifp) {
799 			int e;
800 
801 			/*
802 			 * If no route is available and create is set,
803 			 * we allocate a host route for the destination
804 			 * and treat it like an interface route.
805 			 * This hack is necessary for a neighbor which can't
806 			 * be covered by our own prefix.
807 			 */
808 			struct ifaddr *ifa =
809 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
810 			if (ifa == NULL)
811 				return (NULL);
812 
813 			/*
814 			 * Create a new route.  RTF_LLINFO is necessary
815 			 * to create a Neighbor Cache entry for the
816 			 * destination in nd6_rtrequest which will be
817 			 * called in rtrequest via ifa->ifa_rtrequest.
818 			 */
819 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
820 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
821 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
822 			    ~RTF_CLONING, &rt)) != 0) {
823 				log(LOG_ERR,
824 				    "nd6_lookup: failed to add route for a "
825 				    "neighbor(%s), errno=%d\n",
826 				    ip6_sprintf(addr6), e);
827 			}
828 			if (rt == NULL)
829 				return (NULL);
830 			RT_LOCK(rt);
831 			if (rt->rt_llinfo) {
832 				struct llinfo_nd6 *ln =
833 				    (struct llinfo_nd6 *)rt->rt_llinfo;
834 				ln->ln_state = ND6_LLINFO_NOSTATE;
835 			}
836 		} else
837 			return (NULL);
838 	}
839 	RT_LOCK_ASSERT(rt);
840 	RT_REMREF(rt);
841 	/*
842 	 * Validation for the entry.
843 	 * Note that the check for rt_llinfo is necessary because a cloned
844 	 * route from a parent route that has the L flag (e.g. the default
845 	 * route to a p2p interface) may have the flag, too, while the
846 	 * destination is not actually a neighbor.
847 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
848 	 *      it might be the loopback interface if the entry is for our
849 	 *      own address on a non-loopback interface. Instead, we should
850 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
851 	 *	interface.
852 	 */
853 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
854 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
855 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
856 		if (create) {
857 			log(LOG_DEBUG,
858 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
859 			    ip6_sprintf(addr6),
860 			    ifp ? if_name(ifp) : "unspec");
861 			/* xxx more logs... kazu */
862 		}
863 		RT_UNLOCK(rt);
864 		return (NULL);
865 	}
866 	RT_UNLOCK(rt);		/* XXX not ready to return rt locked */
867 	return (rt);
868 }
869 
870 /*
871  * Detect if a given IPv6 address identifies a neighbor on a given link.
872  * XXX: should take care of the destination of a p2p link?
873  */
874 int
875 nd6_is_addr_neighbor(addr, ifp)
876 	struct sockaddr_in6 *addr;
877 	struct ifnet *ifp;
878 {
879 	struct nd_prefix *pr;
880 
881 	/*
882 	 * A link-local address is always a neighbor.
883 	 * XXX: we should use the sin6_scope_id field rather than the embedded
884 	 * interface index.
885 	 * XXX: a link does not necessarily specify a single interface.
886 	 */
887 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
888 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
889 		return (1);
890 
891 	/*
892 	 * If the address matches one of our addresses,
893 	 * it should be a neighbor.
894 	 * If the address matches one of our on-link prefixes, it should be a
895 	 * neighbor.
896 	 */
897 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
898 		if (pr->ndpr_ifp != ifp)
899 			continue;
900 
901 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
902 			continue;
903 
904 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
905 		    &addr->sin6_addr, &pr->ndpr_mask))
906 			return (1);
907 	}
908 
909 	/*
910 	 * If the default router list is empty, all addresses are regarded
911 	 * as on-link, and thus, as a neighbor.
912 	 * XXX: we restrict the condition to hosts, because routers usually do
913 	 * not have the "default router list".
914 	 */
915 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
916 	    nd6_defifindex == ifp->if_index) {
917 		return (1);
918 	}
919 
920 	/*
921 	 * Even if the address matches none of our addresses, it might be
922 	 * in the neighbor cache.
923 	 */
924 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
925 		return (1);
926 
927 	return (0);
928 }
929 
930 /*
931  * Free an nd6 llinfo entry.
932  */
933 struct llinfo_nd6 *
934 nd6_free(rt)
935 	struct rtentry *rt;
936 {
937 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
938 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
939 	struct nd_defrouter *dr;
940 
941 	/*
942 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
943 	 * even though it is not harmful, it was not really necessary.
944 	 */
945 
946 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
947 		int s;
948 		s = splnet();
949 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
950 		    rt->rt_ifp);
951 
952 		if (ln->ln_router || dr) {
953 			/*
954 			 * rt6_flush must be called whether or not the neighbor
955 			 * is in the Default Router List.
956 			 * See a corresponding comment in nd6_na_input().
957 			 */
958 			rt6_flush(&in6, rt->rt_ifp);
959 		}
960 
961 		if (dr) {
962 			/*
963 			 * Unreachablity of a router might affect the default
964 			 * router selection and on-link detection of advertised
965 			 * prefixes.
966 			 */
967 
968 			/*
969 			 * Temporarily fake the state to choose a new default
970 			 * router and to perform on-link determination of
971 			 * prefixes correctly.
972 			 * Below the state will be set correctly,
973 			 * or the entry itself will be deleted.
974 			 */
975 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
976 
977 			/*
978 			 * Since defrouter_select() does not affect the
979 			 * on-link determination and MIP6 needs the check
980 			 * before the default router selection, we perform
981 			 * the check now.
982 			 */
983 			pfxlist_onlink_check();
984 
985 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
986 				/*
987 				 * It is used as the current default router,
988 				 * so we have to move it to the end of the
989 				 * list and choose a new one.
990 				 * XXX: it is not very efficient if this is
991 				 *      the only router.
992 				 */
993 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
994 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
995 
996 				defrouter_select();
997 			}
998 		}
999 		splx(s);
1000 	}
1001 
1002 	/*
1003 	 * Before deleting the entry, remember the next entry as the
1004 	 * return value.  We need this because pfxlist_onlink_check() above
1005 	 * might have freed other entries (particularly the old next entry) as
1006 	 * a side effect (XXX).
1007 	 */
1008 	next = ln->ln_next;
1009 
1010 	/*
1011 	 * Detach the route from the routing tree and the list of neighbor
1012 	 * caches, and disable the route entry not to be used in already
1013 	 * cached routes.
1014 	 */
1015 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1016 	    rt_mask(rt), 0, (struct rtentry **)0);
1017 
1018 	return (next);
1019 }
1020 
1021 /*
1022  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1023  *
1024  * XXX cost-effective metods?
1025  */
1026 void
1027 nd6_nud_hint(rt, dst6, force)
1028 	struct rtentry *rt;
1029 	struct in6_addr *dst6;
1030 	int force;
1031 {
1032 	struct llinfo_nd6 *ln;
1033 
1034 	/*
1035 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1036 	 * routing table by supplied "dst6".
1037 	 */
1038 	if (!rt) {
1039 		if (!dst6)
1040 			return;
1041 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1042 			return;
1043 	}
1044 
1045 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1046 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1047 	    !rt->rt_llinfo || !rt->rt_gateway ||
1048 	    rt->rt_gateway->sa_family != AF_LINK) {
1049 		/* This is not a host route. */
1050 		return;
1051 	}
1052 
1053 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1054 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1055 		return;
1056 
1057 	/*
1058 	 * if we get upper-layer reachability confirmation many times,
1059 	 * it is possible we have false information.
1060 	 */
1061 	if (!force) {
1062 		ln->ln_byhint++;
1063 		if (ln->ln_byhint > nd6_maxnudhint)
1064 			return;
1065 	}
1066 
1067 	ln->ln_state = ND6_LLINFO_REACHABLE;
1068 	if (ln->ln_expire)
1069 		ln->ln_expire = time_second +
1070 			ND_IFINFO(rt->rt_ifp)->reachable;
1071 }
1072 
1073 void
1074 nd6_rtrequest(req, rt, info)
1075 	int	req;
1076 	struct rtentry *rt;
1077 	struct rt_addrinfo *info; /* xxx unused */
1078 {
1079 	struct sockaddr *gate = rt->rt_gateway;
1080 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1081 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1082 	struct ifnet *ifp = rt->rt_ifp;
1083 	struct ifaddr *ifa;
1084 
1085 	RT_LOCK_ASSERT(rt);
1086 
1087 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1088 		return;
1089 
1090 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1091 		/*
1092 		 * This is probably an interface direct route for a link
1093 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1094 		 * We do not need special treatment below for such a route.
1095 		 * Moreover, the RTF_LLINFO flag which would be set below
1096 		 * would annoy the ndp(8) command.
1097 		 */
1098 		return;
1099 	}
1100 
1101 	if (req == RTM_RESOLVE &&
1102 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1103 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1104 		/*
1105 		 * FreeBSD and BSD/OS often make a cloned host route based
1106 		 * on a less-specific route (e.g. the default route).
1107 		 * If the less specific route does not have a "gateway"
1108 		 * (this is the case when the route just goes to a p2p or an
1109 		 * stf interface), we'll mistakenly make a neighbor cache for
1110 		 * the host route, and will see strange neighbor solicitation
1111 		 * for the corresponding destination.  In order to avoid the
1112 		 * confusion, we check if the destination of the route is
1113 		 * a neighbor in terms of neighbor discovery, and stop the
1114 		 * process if not.  Additionally, we remove the LLINFO flag
1115 		 * so that ndp(8) will not try to get the neighbor information
1116 		 * of the destination.
1117 		 */
1118 		rt->rt_flags &= ~RTF_LLINFO;
1119 		return;
1120 	}
1121 
1122 	switch (req) {
1123 	case RTM_ADD:
1124 		/*
1125 		 * There is no backward compatibility :)
1126 		 *
1127 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1128 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1129 		 *	   rt->rt_flags |= RTF_CLONING;
1130 		 */
1131 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1132 			/*
1133 			 * Case 1: This route should come from
1134 			 * a route to interface.  RTF_LLINFO flag is set
1135 			 * for a host route whose destination should be
1136 			 * treated as on-link.
1137 			 */
1138 			rt_setgate(rt, rt_key(rt),
1139 				   (struct sockaddr *)&null_sdl);
1140 			gate = rt->rt_gateway;
1141 			SDL(gate)->sdl_type = ifp->if_type;
1142 			SDL(gate)->sdl_index = ifp->if_index;
1143 			if (ln)
1144 				ln->ln_expire = time_second;
1145 			if (ln && ln->ln_expire == 0) {
1146 				/* kludge for desktops */
1147 				ln->ln_expire = 1;
1148 			}
1149 			if ((rt->rt_flags & RTF_CLONING) != 0)
1150 				break;
1151 		}
1152 		/*
1153 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1154 		 * We don't do that here since llinfo is not ready yet.
1155 		 *
1156 		 * There are also couple of other things to be discussed:
1157 		 * - unsolicited NA code needs improvement beforehand
1158 		 * - RFC2461 says we MAY send multicast unsolicited NA
1159 		 *   (7.2.6 paragraph 4), however, it also says that we
1160 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1161 		 *   we don't have anything like it right now.
1162 		 *   note that the mechanism needs a mutual agreement
1163 		 *   between proxies, which means that we need to implement
1164 		 *   a new protocol, or a new kludge.
1165 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1166 		 *   we need to check ip6forwarding before sending it.
1167 		 *   (or should we allow proxy ND configuration only for
1168 		 *   routers?  there's no mention about proxy ND from hosts)
1169 		 */
1170 #if 0
1171 		/* XXX it does not work */
1172 		if (rt->rt_flags & RTF_ANNOUNCE)
1173 			nd6_na_output(ifp,
1174 			      &SIN6(rt_key(rt))->sin6_addr,
1175 			      &SIN6(rt_key(rt))->sin6_addr,
1176 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1177 			      1, NULL);
1178 #endif
1179 		/* FALLTHROUGH */
1180 	case RTM_RESOLVE:
1181 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1182 			/*
1183 			 * Address resolution isn't necessary for a point to
1184 			 * point link, so we can skip this test for a p2p link.
1185 			 */
1186 			if (gate->sa_family != AF_LINK ||
1187 			    gate->sa_len < sizeof(null_sdl)) {
1188 				log(LOG_DEBUG,
1189 				    "nd6_rtrequest: bad gateway value: %s\n",
1190 				    if_name(ifp));
1191 				break;
1192 			}
1193 			SDL(gate)->sdl_type = ifp->if_type;
1194 			SDL(gate)->sdl_index = ifp->if_index;
1195 		}
1196 		if (ln != NULL)
1197 			break;	/* This happens on a route change */
1198 		/*
1199 		 * Case 2: This route may come from cloning, or a manual route
1200 		 * add with a LL address.
1201 		 */
1202 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1203 		rt->rt_llinfo = (caddr_t)ln;
1204 		if (!ln) {
1205 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1206 			break;
1207 		}
1208 		nd6_inuse++;
1209 		nd6_allocated++;
1210 		bzero(ln, sizeof(*ln));
1211 		ln->ln_rt = rt;
1212 		/* this is required for "ndp" command. - shin */
1213 		if (req == RTM_ADD) {
1214 		        /*
1215 			 * gate should have some valid AF_LINK entry,
1216 			 * and ln->ln_expire should have some lifetime
1217 			 * which is specified by ndp command.
1218 			 */
1219 			ln->ln_state = ND6_LLINFO_REACHABLE;
1220 			ln->ln_byhint = 0;
1221 		} else {
1222 		        /*
1223 			 * When req == RTM_RESOLVE, rt is created and
1224 			 * initialized in rtrequest(), so rt_expire is 0.
1225 			 */
1226 			ln->ln_state = ND6_LLINFO_NOSTATE;
1227 			ln->ln_expire = time_second;
1228 		}
1229 		rt->rt_flags |= RTF_LLINFO;
1230 		ln->ln_next = llinfo_nd6.ln_next;
1231 		llinfo_nd6.ln_next = ln;
1232 		ln->ln_prev = &llinfo_nd6;
1233 		ln->ln_next->ln_prev = ln;
1234 
1235 		/*
1236 		 * check if rt_key(rt) is one of my address assigned
1237 		 * to the interface.
1238 		 */
1239 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1240 		    &SIN6(rt_key(rt))->sin6_addr);
1241 		if (ifa) {
1242 			caddr_t macp = nd6_ifptomac(ifp);
1243 			ln->ln_expire = 0;
1244 			ln->ln_state = ND6_LLINFO_REACHABLE;
1245 			ln->ln_byhint = 0;
1246 			if (macp) {
1247 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1248 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1249 			}
1250 			if (nd6_useloopback) {
1251 				rt->rt_ifp = &loif[0];	/* XXX */
1252 				/*
1253 				 * Make sure rt_ifa be equal to the ifaddr
1254 				 * corresponding to the address.
1255 				 * We need this because when we refer
1256 				 * rt_ifa->ia6_flags in ip6_input, we assume
1257 				 * that the rt_ifa points to the address instead
1258 				 * of the loopback address.
1259 				 */
1260 				if (ifa != rt->rt_ifa) {
1261 					IFAFREE(rt->rt_ifa);
1262 					IFAREF(ifa);
1263 					rt->rt_ifa = ifa;
1264 				}
1265 			}
1266 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1267 			ln->ln_expire = 0;
1268 			ln->ln_state = ND6_LLINFO_REACHABLE;
1269 			ln->ln_byhint = 0;
1270 
1271 			/* join solicited node multicast for proxy ND */
1272 			if (ifp->if_flags & IFF_MULTICAST) {
1273 				struct in6_addr llsol;
1274 				int error;
1275 
1276 				llsol = SIN6(rt_key(rt))->sin6_addr;
1277 				llsol.s6_addr16[0] = htons(0xff02);
1278 				llsol.s6_addr16[1] = htons(ifp->if_index);
1279 				llsol.s6_addr32[1] = 0;
1280 				llsol.s6_addr32[2] = htonl(1);
1281 				llsol.s6_addr8[12] = 0xff;
1282 
1283 				if (!in6_addmulti(&llsol, ifp, &error)) {
1284 					nd6log((LOG_ERR, "%s: failed to join "
1285 					    "%s (errno=%d)\n", if_name(ifp),
1286 					    ip6_sprintf(&llsol), error));
1287 				}
1288 			}
1289 		}
1290 		break;
1291 
1292 	case RTM_DELETE:
1293 		if (!ln)
1294 			break;
1295 		/* leave from solicited node multicast for proxy ND */
1296 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1297 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1298 			struct in6_addr llsol;
1299 			struct in6_multi *in6m;
1300 
1301 			llsol = SIN6(rt_key(rt))->sin6_addr;
1302 			llsol.s6_addr16[0] = htons(0xff02);
1303 			llsol.s6_addr16[1] = htons(ifp->if_index);
1304 			llsol.s6_addr32[1] = 0;
1305 			llsol.s6_addr32[2] = htonl(1);
1306 			llsol.s6_addr8[12] = 0xff;
1307 
1308 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1309 			if (in6m)
1310 				in6_delmulti(in6m);
1311 		}
1312 		nd6_inuse--;
1313 		ln->ln_next->ln_prev = ln->ln_prev;
1314 		ln->ln_prev->ln_next = ln->ln_next;
1315 		ln->ln_prev = NULL;
1316 		rt->rt_llinfo = 0;
1317 		rt->rt_flags &= ~RTF_LLINFO;
1318 		if (ln->ln_hold)
1319 			m_freem(ln->ln_hold);
1320 		Free((caddr_t)ln);
1321 	}
1322 }
1323 
1324 int
1325 nd6_ioctl(cmd, data, ifp)
1326 	u_long cmd;
1327 	caddr_t	data;
1328 	struct ifnet *ifp;
1329 {
1330 	struct in6_drlist *drl = (struct in6_drlist *)data;
1331 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1332 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1333 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1334 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1335 	struct nd_defrouter *dr, any;
1336 	struct nd_prefix *pr;
1337 	struct rtentry *rt;
1338 	int i = 0, error = 0;
1339 	int s;
1340 
1341 	switch (cmd) {
1342 	case SIOCGDRLST_IN6:
1343 		/*
1344 		 * obsolete API, use sysctl under net.inet6.icmp6
1345 		 */
1346 		bzero(drl, sizeof(*drl));
1347 		s = splnet();
1348 		dr = TAILQ_FIRST(&nd_defrouter);
1349 		while (dr && i < DRLSTSIZ) {
1350 			drl->defrouter[i].rtaddr = dr->rtaddr;
1351 			in6_clearscope(&drl->defrouter[i].rtaddr);
1352 
1353 			drl->defrouter[i].flags = dr->flags;
1354 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1355 			drl->defrouter[i].expire = dr->expire;
1356 			drl->defrouter[i].if_index = dr->ifp->if_index;
1357 			i++;
1358 			dr = TAILQ_NEXT(dr, dr_entry);
1359 		}
1360 		splx(s);
1361 		break;
1362 	case SIOCGPRLST_IN6:
1363 		/*
1364 		 * obsolete API, use sysctl under net.inet6.icmp6
1365 		 *
1366 		 * XXX the structure in6_prlist was changed in backward-
1367 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1368 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1369 		 */
1370 		/*
1371 		 * XXX meaning of fields, especialy "raflags", is very
1372 		 * differnet between RA prefix list and RR/static prefix list.
1373 		 * how about separating ioctls into two?
1374 		 */
1375 		bzero(oprl, sizeof(*oprl));
1376 		s = splnet();
1377 		pr = nd_prefix.lh_first;
1378 		while (pr && i < PRLSTSIZ) {
1379 			struct nd_pfxrouter *pfr;
1380 			int j;
1381 
1382 			(void)in6_embedscope(&oprl->prefix[i].prefix,
1383 			    &pr->ndpr_prefix, NULL, NULL);
1384 			oprl->prefix[i].raflags = pr->ndpr_raf;
1385 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1386 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1387 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1388 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1389 			oprl->prefix[i].expire = pr->ndpr_expire;
1390 
1391 			pfr = pr->ndpr_advrtrs.lh_first;
1392 			j = 0;
1393 			while (pfr) {
1394 				if (j < DRLSTSIZ) {
1395 #define RTRADDR oprl->prefix[i].advrtr[j]
1396 					RTRADDR = pfr->router->rtaddr;
1397 					in6_clearscope(&RTRADDR);
1398 #undef RTRADDR
1399 				}
1400 				j++;
1401 				pfr = pfr->pfr_next;
1402 			}
1403 			oprl->prefix[i].advrtrs = j;
1404 			oprl->prefix[i].origin = PR_ORIG_RA;
1405 
1406 			i++;
1407 			pr = pr->ndpr_next;
1408 		}
1409 		splx(s);
1410 
1411 		break;
1412 	case OSIOCGIFINFO_IN6:
1413 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1414 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1415 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1416 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1417 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1418 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1419 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1420 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1421 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1422 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1423 		break;
1424 	case SIOCGIFINFO_IN6:
1425 		ndi->ndi = *ND_IFINFO(ifp);
1426 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1427 		break;
1428 	case SIOCSIFINFO_FLAGS:
1429 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1430 		break;
1431 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1432 		/* flush default router list */
1433 		/*
1434 		 * xxx sumikawa: should not delete route if default
1435 		 * route equals to the top of default router list
1436 		 */
1437 		bzero(&any, sizeof(any));
1438 		defrouter_delreq(&any, 0);
1439 		defrouter_select();
1440 		/* xxx sumikawa: flush prefix list */
1441 		break;
1442 	case SIOCSPFXFLUSH_IN6:
1443 	{
1444 		/* flush all the prefix advertised by routers */
1445 		struct nd_prefix *pr, *next;
1446 
1447 		s = splnet();
1448 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1449 			struct in6_ifaddr *ia, *ia_next;
1450 
1451 			next = pr->ndpr_next;
1452 
1453 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1454 				continue; /* XXX */
1455 
1456 			/* do we really have to remove addresses as well? */
1457 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1458 				/* ia might be removed.  keep the next ptr. */
1459 				ia_next = ia->ia_next;
1460 
1461 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1462 					continue;
1463 
1464 				if (ia->ia6_ndpr == pr)
1465 					in6_purgeaddr(&ia->ia_ifa);
1466 			}
1467 			prelist_remove(pr);
1468 		}
1469 		splx(s);
1470 		break;
1471 	}
1472 	case SIOCSRTRFLUSH_IN6:
1473 	{
1474 		/* flush all the default routers */
1475 		struct nd_defrouter *dr, *next;
1476 
1477 		s = splnet();
1478 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1479 			/*
1480 			 * The first entry of the list may be stored in
1481 			 * the routing table, so we'll delete it later.
1482 			 */
1483 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1484 				next = TAILQ_NEXT(dr, dr_entry);
1485 				defrtrlist_del(dr);
1486 			}
1487 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1488 		}
1489 		splx(s);
1490 		break;
1491 	}
1492 	case SIOCGNBRINFO_IN6:
1493 	{
1494 		struct llinfo_nd6 *ln;
1495 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1496 
1497 		/*
1498 		 * XXX: KAME specific hack for scoped addresses
1499 		 *      XXXX: for other scopes than link-local?
1500 		 */
1501 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1502 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1503 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1504 
1505 			if (*idp == 0)
1506 				*idp = htons(ifp->if_index);
1507 		}
1508 
1509 		s = splnet();
1510 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1511 			error = EINVAL;
1512 			splx(s);
1513 			break;
1514 		}
1515 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1516 		nbi->state = ln->ln_state;
1517 		nbi->asked = ln->ln_asked;
1518 		nbi->isrouter = ln->ln_router;
1519 		nbi->expire = ln->ln_expire;
1520 		splx(s);
1521 
1522 		break;
1523 	}
1524 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1525 		ndif->ifindex = nd6_defifindex;
1526 		break;
1527 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1528 		return (nd6_setdefaultiface(ndif->ifindex));
1529 	}
1530 	return (error);
1531 }
1532 
1533 /*
1534  * Create neighbor cache entry and cache link-layer address,
1535  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1536  */
1537 struct rtentry *
1538 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1539 	struct ifnet *ifp;
1540 	struct in6_addr *from;
1541 	char *lladdr;
1542 	int lladdrlen;
1543 	int type;	/* ICMP6 type */
1544 	int code;	/* type dependent information */
1545 {
1546 	struct rtentry *rt = NULL;
1547 	struct llinfo_nd6 *ln = NULL;
1548 	int is_newentry;
1549 	struct sockaddr_dl *sdl = NULL;
1550 	int do_update;
1551 	int olladdr;
1552 	int llchange;
1553 	int newstate = 0;
1554 
1555 	if (!ifp)
1556 		panic("ifp == NULL in nd6_cache_lladdr");
1557 	if (!from)
1558 		panic("from == NULL in nd6_cache_lladdr");
1559 
1560 	/* nothing must be updated for unspecified address */
1561 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1562 		return NULL;
1563 
1564 	/*
1565 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1566 	 * the caller.
1567 	 *
1568 	 * XXX If the link does not have link-layer adderss, what should
1569 	 * we do? (ifp->if_addrlen == 0)
1570 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1571 	 * description on it in NS section (RFC 2461 7.2.3).
1572 	 */
1573 
1574 	rt = nd6_lookup(from, 0, ifp);
1575 	if (!rt) {
1576 #if 0
1577 		/* nothing must be done if there's no lladdr */
1578 		if (!lladdr || !lladdrlen)
1579 			return NULL;
1580 #endif
1581 
1582 		rt = nd6_lookup(from, 1, ifp);
1583 		is_newentry = 1;
1584 	} else {
1585 		/* do nothing if static ndp is set */
1586 		if (rt->rt_flags & RTF_STATIC)
1587 			return NULL;
1588 		is_newentry = 0;
1589 	}
1590 
1591 	if (!rt)
1592 		return NULL;
1593 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1594 fail:
1595 		(void)nd6_free(rt);
1596 		return NULL;
1597 	}
1598 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1599 	if (!ln)
1600 		goto fail;
1601 	if (!rt->rt_gateway)
1602 		goto fail;
1603 	if (rt->rt_gateway->sa_family != AF_LINK)
1604 		goto fail;
1605 	sdl = SDL(rt->rt_gateway);
1606 
1607 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1608 	if (olladdr && lladdr) {
1609 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1610 			llchange = 1;
1611 		else
1612 			llchange = 0;
1613 	} else
1614 		llchange = 0;
1615 
1616 	/*
1617 	 * newentry olladdr  lladdr  llchange	(*=record)
1618 	 *	0	n	n	--	(1)
1619 	 *	0	y	n	--	(2)
1620 	 *	0	n	y	--	(3) * STALE
1621 	 *	0	y	y	n	(4) *
1622 	 *	0	y	y	y	(5) * STALE
1623 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1624 	 *	1	--	y	--	(7) * STALE
1625 	 */
1626 
1627 	if (lladdr) {		/* (3-5) and (7) */
1628 		/*
1629 		 * Record source link-layer address
1630 		 * XXX is it dependent to ifp->if_type?
1631 		 */
1632 		sdl->sdl_alen = ifp->if_addrlen;
1633 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1634 	}
1635 
1636 	if (!is_newentry) {
1637 		if ((!olladdr && lladdr) ||		/* (3) */
1638 		    (olladdr && lladdr && llchange)) {	/* (5) */
1639 			do_update = 1;
1640 			newstate = ND6_LLINFO_STALE;
1641 		} else					/* (1-2,4) */
1642 			do_update = 0;
1643 	} else {
1644 		do_update = 1;
1645 		if (!lladdr)				/* (6) */
1646 			newstate = ND6_LLINFO_NOSTATE;
1647 		else					/* (7) */
1648 			newstate = ND6_LLINFO_STALE;
1649 	}
1650 
1651 	if (do_update) {
1652 		/*
1653 		 * Update the state of the neighbor cache.
1654 		 */
1655 		ln->ln_state = newstate;
1656 
1657 		if (ln->ln_state == ND6_LLINFO_STALE) {
1658 			/*
1659 			 * XXX: since nd6_output() below will cause
1660 			 * state tansition to DELAY and reset the timer,
1661 			 * we must set the timer now, although it is actually
1662 			 * meaningless.
1663 			 */
1664 			ln->ln_expire = time_second + nd6_gctimer;
1665 
1666 			if (ln->ln_hold) {
1667 				/*
1668 				 * we assume ifp is not a p2p here, so just
1669 				 * set the 2nd argument as the 1st one.
1670 				 */
1671 				nd6_output(ifp, ifp, ln->ln_hold,
1672 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1673 				ln->ln_hold = NULL;
1674 			}
1675 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1676 			/* probe right away */
1677 			ln->ln_expire = time_second;
1678 		}
1679 	}
1680 
1681 	/*
1682 	 * ICMP6 type dependent behavior.
1683 	 *
1684 	 * NS: clear IsRouter if new entry
1685 	 * RS: clear IsRouter
1686 	 * RA: set IsRouter if there's lladdr
1687 	 * redir: clear IsRouter if new entry
1688 	 *
1689 	 * RA case, (1):
1690 	 * The spec says that we must set IsRouter in the following cases:
1691 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1692 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1693 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1694 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1695 	 * neighbor cache, this is similar to (6).
1696 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1697 	 *
1698 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1699 	 *							D R
1700 	 *	0	n	n	--	(1)	c   ?     s
1701 	 *	0	y	n	--	(2)	c   s     s
1702 	 *	0	n	y	--	(3)	c   s     s
1703 	 *	0	y	y	n	(4)	c   s     s
1704 	 *	0	y	y	y	(5)	c   s     s
1705 	 *	1	--	n	--	(6) c	c 	c s
1706 	 *	1	--	y	--	(7) c	c   s	c s
1707 	 *
1708 	 *					(c=clear s=set)
1709 	 */
1710 	switch (type & 0xff) {
1711 	case ND_NEIGHBOR_SOLICIT:
1712 		/*
1713 		 * New entry must have is_router flag cleared.
1714 		 */
1715 		if (is_newentry)	/* (6-7) */
1716 			ln->ln_router = 0;
1717 		break;
1718 	case ND_REDIRECT:
1719 		/*
1720 		 * If the icmp is a redirect to a better router, always set the
1721 		 * is_router flag.  Otherwise, if the entry is newly created,
1722 		 * clear the flag.  [RFC 2461, sec 8.3]
1723 		 */
1724 		if (code == ND_REDIRECT_ROUTER)
1725 			ln->ln_router = 1;
1726 		else if (is_newentry) /* (6-7) */
1727 			ln->ln_router = 0;
1728 		break;
1729 	case ND_ROUTER_SOLICIT:
1730 		/*
1731 		 * is_router flag must always be cleared.
1732 		 */
1733 		ln->ln_router = 0;
1734 		break;
1735 	case ND_ROUTER_ADVERT:
1736 		/*
1737 		 * Mark an entry with lladdr as a router.
1738 		 */
1739 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1740 		    (is_newentry && lladdr)) {			/* (7) */
1741 			ln->ln_router = 1;
1742 		}
1743 		break;
1744 	}
1745 
1746 	/*
1747 	 * When the link-layer address of a router changes, select the
1748 	 * best router again.  In particular, when the neighbor entry is newly
1749 	 * created, it might affect the selection policy.
1750 	 * Question: can we restrict the first condition to the "is_newentry"
1751 	 * case?
1752 	 * XXX: when we hear an RA from a new router with the link-layer
1753 	 * address option, defrouter_select() is called twice, since
1754 	 * defrtrlist_update called the function as well.  However, I believe
1755 	 * we can compromise the overhead, since it only happens the first
1756 	 * time.
1757 	 * XXX: although defrouter_select() should not have a bad effect
1758 	 * for those are not autoconfigured hosts, we explicitly avoid such
1759 	 * cases for safety.
1760 	 */
1761 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1762 		defrouter_select();
1763 
1764 	return rt;
1765 }
1766 
1767 static void
1768 nd6_slowtimo(ignored_arg)
1769     void *ignored_arg;
1770 {
1771 	int s = splnet();
1772 	struct nd_ifinfo *nd6if;
1773 	struct ifnet *ifp;
1774 
1775 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1776 	    nd6_slowtimo, NULL);
1777 	IFNET_RLOCK();
1778 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1779 		nd6if = ND_IFINFO(ifp);
1780 		if (nd6if->basereachable && /* already initialized */
1781 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1782 			/*
1783 			 * Since reachable time rarely changes by router
1784 			 * advertisements, we SHOULD insure that a new random
1785 			 * value gets recomputed at least once every few hours.
1786 			 * (RFC 2461, 6.3.4)
1787 			 */
1788 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1789 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1790 		}
1791 	}
1792 	IFNET_RUNLOCK();
1793 	splx(s);
1794 }
1795 
1796 #define senderr(e) { error = (e); goto bad;}
1797 int
1798 nd6_output(ifp, origifp, m0, dst, rt0)
1799 	struct ifnet *ifp;
1800 	struct ifnet *origifp;
1801 	struct mbuf *m0;
1802 	struct sockaddr_in6 *dst;
1803 	struct rtentry *rt0;
1804 {
1805 	struct mbuf *m = m0;
1806 	struct rtentry *rt = rt0;
1807 	struct sockaddr_in6 *gw6 = NULL;
1808 	struct llinfo_nd6 *ln = NULL;
1809 	int error = 0;
1810 
1811 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1812 		goto sendpkt;
1813 
1814 	if (nd6_need_cache(ifp) == 0)
1815 		goto sendpkt;
1816 
1817 	/*
1818 	 * next hop determination.  This routine is derived from ether_outpout.
1819 	 */
1820 again:
1821 	if (rt) {
1822 		if ((rt->rt_flags & RTF_UP) == 0) {
1823 			rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1824 			if (rt != NULL) {
1825 				RT_REMREF(rt);
1826 				RT_UNLOCK(rt);
1827 				if (rt->rt_ifp != ifp)
1828 					/*
1829 					 * XXX maybe we should update ifp too,
1830 					 * but the original code didn't and I
1831 					 * don't know what is correct here.
1832 					 */
1833 					goto again;
1834 			} else
1835 				senderr(EHOSTUNREACH);
1836 		}
1837 
1838 		if (rt->rt_flags & RTF_GATEWAY) {
1839 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1840 
1841 			/*
1842 			 * We skip link-layer address resolution and NUD
1843 			 * if the gateway is not a neighbor from ND point
1844 			 * of view, regardless of the value of nd_ifinfo.flags.
1845 			 * The second condition is a bit tricky; we skip
1846 			 * if the gateway is our own address, which is
1847 			 * sometimes used to install a route to a p2p link.
1848 			 */
1849 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1850 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1851 				/*
1852 				 * We allow this kind of tricky route only
1853 				 * when the outgoing interface is p2p.
1854 				 * XXX: we may need a more generic rule here.
1855 				 */
1856 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1857 					senderr(EHOSTUNREACH);
1858 
1859 				goto sendpkt;
1860 			}
1861 
1862 			if (rt->rt_gwroute == 0)
1863 				goto lookup;
1864 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1865 				RT_LOCK(rt);
1866 				rtfree(rt); rt = rt0;
1867 			lookup:
1868 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1869 				if ((rt = rt->rt_gwroute) == 0)
1870 					senderr(EHOSTUNREACH);
1871 				RT_UNLOCK(rt);
1872 			}
1873 		}
1874 	}
1875 
1876 	/*
1877 	 * Address resolution or Neighbor Unreachability Detection
1878 	 * for the next hop.
1879 	 * At this point, the destination of the packet must be a unicast
1880 	 * or an anycast address(i.e. not a multicast).
1881 	 */
1882 
1883 	/* Look up the neighbor cache for the nexthop */
1884 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1885 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1886 	else {
1887 		/*
1888 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1889 		 * the condition below is not very efficient.  But we believe
1890 		 * it is tolerable, because this should be a rare case.
1891 		 */
1892 		if (nd6_is_addr_neighbor(dst, ifp) &&
1893 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1894 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1895 	}
1896 	if (!ln || !rt) {
1897 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1898 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1899 			log(LOG_DEBUG,
1900 			    "nd6_output: can't allocate llinfo for %s "
1901 			    "(ln=%p, rt=%p)\n",
1902 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1903 			senderr(EIO);	/* XXX: good error? */
1904 		}
1905 
1906 		goto sendpkt;	/* send anyway */
1907 	}
1908 
1909 	/* We don't have to do link-layer address resolution on a p2p link. */
1910 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1911 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1912 		ln->ln_state = ND6_LLINFO_STALE;
1913 		ln->ln_expire = time_second + nd6_gctimer;
1914 	}
1915 
1916 	/*
1917 	 * The first time we send a packet to a neighbor whose entry is
1918 	 * STALE, we have to change the state to DELAY and a sets a timer to
1919 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1920 	 * neighbor unreachability detection on expiration.
1921 	 * (RFC 2461 7.3.3)
1922 	 */
1923 	if (ln->ln_state == ND6_LLINFO_STALE) {
1924 		ln->ln_asked = 0;
1925 		ln->ln_state = ND6_LLINFO_DELAY;
1926 		ln->ln_expire = time_second + nd6_delay;
1927 	}
1928 
1929 	/*
1930 	 * If the neighbor cache entry has a state other than INCOMPLETE
1931 	 * (i.e. its link-layer address is already resolved), just
1932 	 * send the packet.
1933 	 */
1934 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1935 		goto sendpkt;
1936 
1937 	/*
1938 	 * There is a neighbor cache entry, but no ethernet address
1939 	 * response yet.  Replace the held mbuf (if any) with this
1940 	 * latest one.
1941 	 *
1942 	 * This code conforms to the rate-limiting rule described in Section
1943 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1944 	 * an NS below.
1945 	 */
1946 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1947 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1948 	if (ln->ln_hold)
1949 		m_freem(ln->ln_hold);
1950 	ln->ln_hold = m;
1951 	if (ln->ln_expire) {
1952 		if (ln->ln_asked < nd6_mmaxtries &&
1953 		    ln->ln_expire < time_second) {
1954 			ln->ln_asked++;
1955 			ln->ln_expire = time_second +
1956 				ND_IFINFO(ifp)->retrans / 1000;
1957 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1958 		}
1959 	}
1960 	return (0);
1961 
1962   sendpkt:
1963 #ifdef IPSEC
1964 	/* clean ipsec history once it goes out of the node */
1965 	ipsec_delaux(m);
1966 #endif
1967 
1968 #ifdef MAC
1969 	mac_create_mbuf_linklayer(ifp, m);
1970 #endif
1971 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1972 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1973 		    rt));
1974 	}
1975 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1976 
1977   bad:
1978 	if (m)
1979 		m_freem(m);
1980 	return (error);
1981 }
1982 #undef senderr
1983 
1984 int
1985 nd6_need_cache(ifp)
1986 	struct ifnet *ifp;
1987 {
1988 	/*
1989 	 * XXX: we currently do not make neighbor cache on any interface
1990 	 * other than ARCnet, Ethernet, FDDI and GIF.
1991 	 *
1992 	 * RFC2893 says:
1993 	 * - unidirectional tunnels needs no ND
1994 	 */
1995 	switch (ifp->if_type) {
1996 	case IFT_ARCNET:
1997 	case IFT_ETHER:
1998 	case IFT_FDDI:
1999 	case IFT_IEEE1394:
2000 #ifdef IFT_L2VLAN
2001 	case IFT_L2VLAN:
2002 #endif
2003 #ifdef IFT_IEEE80211
2004 	case IFT_IEEE80211:
2005 #endif
2006 	case IFT_GIF:		/* XXX need more cases? */
2007 		return (1);
2008 	default:
2009 		return (0);
2010 	}
2011 }
2012 
2013 int
2014 nd6_storelladdr(ifp, rt0, m, dst, desten)
2015 	struct ifnet *ifp;
2016 	struct rtentry *rt0;
2017 	struct mbuf *m;
2018 	struct sockaddr *dst;
2019 	u_char *desten;
2020 {
2021 	int i;
2022 	struct sockaddr_dl *sdl;
2023 	struct rtentry *rt;
2024 
2025 	if (m->m_flags & M_MCAST) {
2026 		switch (ifp->if_type) {
2027 		case IFT_ETHER:
2028 		case IFT_FDDI:
2029 #ifdef IFT_L2VLAN
2030 		case IFT_L2VLAN:
2031 #endif
2032 #ifdef IFT_IEEE80211
2033 		case IFT_IEEE80211:
2034 #endif
2035 		case IFT_ISO88025:
2036 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2037 						 desten);
2038 			return (0);
2039 		case IFT_IEEE1394:
2040 			/*
2041 			 * netbsd can use if_broadcastaddr, but we don't do so
2042 			 * to reduce # of ifdef.
2043 			 */
2044 			for (i = 0; i < ifp->if_addrlen; i++)
2045 				desten[i] = ~0;
2046 			return (0);
2047 		case IFT_ARCNET:
2048 			*desten = 0;
2049 			return (0);
2050 		default:
2051 			m_freem(m);
2052 			return (EAFNOSUPPORT);
2053 		}
2054 	}
2055 
2056 	i = rt_check(&rt, &rt0, dst);
2057 	if (i) {
2058 		m_freem(m);
2059 		return i;
2060 	}
2061 
2062 	if (rt == NULL) {
2063 		/* this could happen, if we could not allocate memory */
2064 		m_freem(m);
2065 		return (ENOMEM);
2066 	}
2067 	if (rt->rt_gateway->sa_family != AF_LINK) {
2068 		printf("nd6_storelladdr: something odd happens\n");
2069 		m_freem(m);
2070 		return (EINVAL);
2071 	}
2072 	sdl = SDL(rt->rt_gateway);
2073 	if (sdl->sdl_alen == 0) {
2074 		/* this should be impossible, but we bark here for debugging */
2075 		printf("nd6_storelladdr: sdl_alen == 0\n");
2076 		m_freem(m);
2077 		return (EINVAL);
2078 	}
2079 
2080 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2081 	return (0);
2082 }
2083 
2084 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2085 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2086 #ifdef SYSCTL_DECL
2087 SYSCTL_DECL(_net_inet6_icmp6);
2088 #endif
2089 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2090 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2091 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2092 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2093 
2094 static int
2095 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2096 {
2097 	int error;
2098 	char buf[1024];
2099 	struct in6_defrouter *d, *de;
2100 	struct nd_defrouter *dr;
2101 
2102 	if (req->newptr)
2103 		return EPERM;
2104 	error = 0;
2105 
2106 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2107 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2108 		d = (struct in6_defrouter *)buf;
2109 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2110 
2111 		if (d + 1 <= de) {
2112 			bzero(d, sizeof(*d));
2113 			d->rtaddr.sin6_family = AF_INET6;
2114 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2115 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2116 			    dr->ifp) != 0)
2117 				log(LOG_ERR,
2118 				    "scope error in "
2119 				    "default router list (%s)\n",
2120 				    ip6_sprintf(&dr->rtaddr));
2121 			d->flags = dr->flags;
2122 			d->rtlifetime = dr->rtlifetime;
2123 			d->expire = dr->expire;
2124 			d->if_index = dr->ifp->if_index;
2125 		} else
2126 			panic("buffer too short");
2127 
2128 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2129 		if (error)
2130 			break;
2131 	}
2132 
2133 	return (error);
2134 }
2135 
2136 static int
2137 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2138 {
2139 	int error;
2140 	char buf[1024];
2141 	struct in6_prefix *p, *pe;
2142 	struct nd_prefix *pr;
2143 
2144 	if (req->newptr)
2145 		return EPERM;
2146 	error = 0;
2147 
2148 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2149 		u_short advrtrs;
2150 		size_t advance;
2151 		struct sockaddr_in6 *sin6, *s6;
2152 		struct nd_pfxrouter *pfr;
2153 
2154 		p = (struct in6_prefix *)buf;
2155 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2156 
2157 		if (p + 1 <= pe) {
2158 			bzero(p, sizeof(*p));
2159 			sin6 = (struct sockaddr_in6 *)(p + 1);
2160 
2161 			p->prefix = pr->ndpr_prefix;
2162 			if (in6_recoverscope(&p->prefix,
2163 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2164 				log(LOG_ERR,
2165 				    "scope error in prefix list (%s)\n",
2166 				    ip6_sprintf(&p->prefix.sin6_addr));
2167 			p->raflags = pr->ndpr_raf;
2168 			p->prefixlen = pr->ndpr_plen;
2169 			p->vltime = pr->ndpr_vltime;
2170 			p->pltime = pr->ndpr_pltime;
2171 			p->if_index = pr->ndpr_ifp->if_index;
2172 			p->expire = pr->ndpr_expire;
2173 			p->refcnt = pr->ndpr_refcnt;
2174 			p->flags = pr->ndpr_stateflags;
2175 			p->origin = PR_ORIG_RA;
2176 			advrtrs = 0;
2177 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2178 			     pfr = pfr->pfr_next) {
2179 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2180 					advrtrs++;
2181 					continue;
2182 				}
2183 				s6 = &sin6[advrtrs];
2184 				bzero(s6, sizeof(*s6));
2185 				s6->sin6_family = AF_INET6;
2186 				s6->sin6_len = sizeof(*sin6);
2187 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2188 				    pfr->router->ifp) != 0)
2189 					log(LOG_ERR,
2190 					    "scope error in "
2191 					    "prefix list (%s)\n",
2192 					    ip6_sprintf(&pfr->router->rtaddr));
2193 				advrtrs++;
2194 			}
2195 			p->advrtrs = advrtrs;
2196 		} else
2197 			panic("buffer too short");
2198 
2199 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2200 		error = SYSCTL_OUT(req, buf, advance);
2201 		if (error)
2202 			break;
2203 	}
2204 
2205 	return (error);
2206 }
2207