xref: /netbsd-src/sys/netinet6/nd6.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: nd6.c,v 1.31 2000/07/06 12:36:19 itojun Exp $	*/
2 /*	$KAME: nd6.c,v 1.68 2000/07/02 14:48:02 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 /*
34  * XXX
35  * KAME 970409 note:
36  * BSD/OS version heavily modifies this code, related to llinfo.
37  * Since we don't have BSD/OS version of net/route.c in our hand,
38  * I left the code mostly as it was in 970310.  -- itojun
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/protosw.h>
51 #include <sys/errno.h>
52 #include <sys/ioctl.h>
53 #include <sys/syslog.h>
54 #include <sys/queue.h>
55 
56 #include <net/if.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/if_atm.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <net/if_ether.h>
64 #include <netinet/if_inarp.h>
65 #include <net/if_fddi.h>
66 #include <netinet6/in6_var.h>
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
69 #include <netinet6/nd6.h>
70 #include <netinet6/in6_prefix.h>
71 #include <netinet/icmp6.h>
72 
73 #include "loop.h"
74 extern struct ifnet loif[NLOOP];
75 
76 #include <net/net_osdep.h>
77 
78 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
79 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
80 
81 #define SIN6(s) ((struct sockaddr_in6 *)s)
82 #define SDL(s) ((struct sockaddr_dl *)s)
83 
84 /* timer values */
85 int	nd6_prune	= 1;	/* walk list every 1 seconds */
86 int	nd6_delay	= 5;	/* delay first probe time 5 second */
87 int	nd6_umaxtries	= 3;	/* maximum unicast query */
88 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
89 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
90 
91 /* preventing too many loops in ND option parsing */
92 int nd6_maxndopt = 10;	/* max # of ND options allowed */
93 
94 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
95 
96 /* for debugging? */
97 static int nd6_inuse, nd6_allocated;
98 
99 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
100 static size_t nd_ifinfo_indexlim = 8;
101 struct nd_ifinfo *nd_ifinfo = NULL;
102 struct nd_drhead nd_defrouter;
103 struct nd_prhead nd_prefix = { 0 };
104 
105 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
106 static struct sockaddr_in6 all1_sa;
107 
108 static void nd6_slowtimo __P((void *));
109 
110 struct callout nd6_slowtimo_ch;
111 struct callout nd6_timer_ch;
112 
113 void
114 nd6_init()
115 {
116 	static int nd6_init_done = 0;
117 	int i;
118 
119 	if (nd6_init_done) {
120 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
121 		return;
122 	}
123 
124 	all1_sa.sin6_family = AF_INET6;
125 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
126 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
127 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
128 
129 	/* initialization of the default router list */
130 	TAILQ_INIT(&nd_defrouter);
131 
132 	nd6_init_done = 1;
133 
134 	/* start timer */
135 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
136 	    nd6_slowtimo, NULL);
137 }
138 
139 void
140 nd6_ifattach(ifp)
141 	struct ifnet *ifp;
142 {
143 
144 	/*
145 	 * We have some arrays that should be indexed by if_index.
146 	 * since if_index will grow dynamically, they should grow too.
147 	 */
148 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
149 		size_t n;
150 		caddr_t q;
151 
152 		while (if_index >= nd_ifinfo_indexlim)
153 			nd_ifinfo_indexlim <<= 1;
154 
155 		/* grow nd_ifinfo */
156 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
157 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
158 		bzero(q, n);
159 		if (nd_ifinfo) {
160 			bcopy((caddr_t)nd_ifinfo, q, n/2);
161 			free((caddr_t)nd_ifinfo, M_IP6NDP);
162 		}
163 		nd_ifinfo = (struct nd_ifinfo *)q;
164 	}
165 
166 #define ND nd_ifinfo[ifp->if_index]
167 
168 	/* don't initialize if called twice */
169 	if (ND.linkmtu)
170 		return;
171 
172 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
173 	ND.chlim = IPV6_DEFHLIM;
174 	ND.basereachable = REACHABLE_TIME;
175 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
176 	ND.retrans = RETRANS_TIMER;
177 	ND.receivedra = 0;
178 	ND.flags = ND6_IFF_PERFORMNUD;
179 	nd6_setmtu(ifp);
180 #undef ND
181 }
182 
183 /*
184  * Reset ND level link MTU. This function is called when the physical MTU
185  * changes, which means we might have to adjust the ND level MTU.
186  */
187 void
188 nd6_setmtu(ifp)
189 	struct ifnet *ifp;
190 {
191 #define MIN(a,b) ((a) < (b) ? (a) : (b))
192 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
193 	u_long oldmaxmtu = ndi->maxmtu;
194 	u_long oldlinkmtu = ndi->linkmtu;
195 
196 	switch(ifp->if_type) {
197 	 case IFT_ARCNET:	/* XXX MTU handling needs more work */
198 		 ndi->maxmtu = MIN(60480, ifp->if_mtu);
199 		 break;
200 	 case IFT_ETHER:
201 		 ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
202 		 break;
203 	 case IFT_ATM:
204 		 ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
205 		 break;
206 	 default:
207 		 ndi->maxmtu = ifp->if_mtu;
208 		 break;
209 	}
210 
211 	if (oldmaxmtu != ndi->maxmtu) {
212 		/*
213 		 * If the ND level MTU is not set yet, or if the maxmtu
214 		 * is reset to a smaller value than the ND level MTU,
215 		 * also reset the ND level MTU.
216 		 */
217 		if (ndi->linkmtu == 0 ||
218 		    ndi->maxmtu < ndi->linkmtu) {
219 			ndi->linkmtu = ndi->maxmtu;
220 			/* also adjust in6_maxmtu if necessary. */
221 			if (oldlinkmtu == 0) {
222 				/*
223 				 * XXX: the case analysis is grotty, but
224 				 * it is not efficient to call in6_setmaxmtu()
225 				 * here when we are during the initialization
226 				 * procedure.
227 				 */
228 				if (in6_maxmtu < ndi->linkmtu)
229 					in6_maxmtu = ndi->linkmtu;
230 			} else
231 				in6_setmaxmtu();
232 		}
233 	}
234 #undef MIN
235 }
236 
237 void
238 nd6_option_init(opt, icmp6len, ndopts)
239 	void *opt;
240 	int icmp6len;
241 	union nd_opts *ndopts;
242 {
243 	bzero(ndopts, sizeof(*ndopts));
244 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
245 	ndopts->nd_opts_last
246 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
247 
248 	if (icmp6len == 0) {
249 		ndopts->nd_opts_done = 1;
250 		ndopts->nd_opts_search = NULL;
251 	}
252 }
253 
254 /*
255  * Take one ND option.
256  */
257 struct nd_opt_hdr *
258 nd6_option(ndopts)
259 	union nd_opts *ndopts;
260 {
261 	struct nd_opt_hdr *nd_opt;
262 	int olen;
263 
264 	if (!ndopts)
265 		panic("ndopts == NULL in nd6_option\n");
266 	if (!ndopts->nd_opts_last)
267 		panic("uninitialized ndopts in nd6_option\n");
268 	if (!ndopts->nd_opts_search)
269 		return NULL;
270 	if (ndopts->nd_opts_done)
271 		return NULL;
272 
273 	nd_opt = ndopts->nd_opts_search;
274 
275 	olen = nd_opt->nd_opt_len << 3;
276 	if (olen == 0) {
277 		/*
278 		 * Message validation requires that all included
279 		 * options have a length that is greater than zero.
280 		 */
281 		bzero(ndopts, sizeof(*ndopts));
282 		return NULL;
283 	}
284 
285 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
286 	if (!(ndopts->nd_opts_search < ndopts->nd_opts_last)) {
287 		ndopts->nd_opts_done = 1;
288 		ndopts->nd_opts_search = NULL;
289 	}
290 	return nd_opt;
291 }
292 
293 /*
294  * Parse multiple ND options.
295  * This function is much easier to use, for ND routines that do not need
296  * multiple options of the same type.
297  */
298 int
299 nd6_options(ndopts)
300 	union nd_opts *ndopts;
301 {
302 	struct nd_opt_hdr *nd_opt;
303 	int i = 0;
304 
305 	if (!ndopts)
306 		panic("ndopts == NULL in nd6_options\n");
307 	if (!ndopts->nd_opts_last)
308 		panic("uninitialized ndopts in nd6_options\n");
309 	if (!ndopts->nd_opts_search)
310 		return 0;
311 
312 	while (1) {
313 		nd_opt = nd6_option(ndopts);
314 		if (!nd_opt && !ndopts->nd_opts_last) {
315 			/*
316 			 * Message validation requires that all included
317 			 * options have a length that is greater than zero.
318 			 */
319 			bzero(ndopts, sizeof(*ndopts));
320 			return -1;
321 		}
322 
323 		if (!nd_opt)
324 			goto skip1;
325 
326 		switch (nd_opt->nd_opt_type) {
327 		case ND_OPT_SOURCE_LINKADDR:
328 		case ND_OPT_TARGET_LINKADDR:
329 		case ND_OPT_MTU:
330 		case ND_OPT_REDIRECTED_HEADER:
331 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
332 				printf("duplicated ND6 option found "
333 					"(type=%d)\n", nd_opt->nd_opt_type);
334 				/* XXX bark? */
335 			} else {
336 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
337 					= nd_opt;
338 			}
339 			break;
340 		case ND_OPT_PREFIX_INFORMATION:
341 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
342 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
343 					= nd_opt;
344 			}
345 			ndopts->nd_opts_pi_end =
346 				(struct nd_opt_prefix_info *)nd_opt;
347 			break;
348 		default:
349 			/*
350 			 * Unknown options must be silently ignored,
351 			 * to accomodate future extension to the protocol.
352 			 */
353 			log(LOG_DEBUG,
354 			    "nd6_options: unsupported option %d - "
355 			    "option ignored\n", nd_opt->nd_opt_type);
356 		}
357 
358 skip1:
359 		i++;
360 		if (i > nd6_maxndopt) {
361 			icmp6stat.icp6s_nd_toomanyopt++;
362 			printf("too many loop in nd opt\n");
363 			break;
364 		}
365 
366 		if (ndopts->nd_opts_done)
367 			break;
368 	}
369 
370 	return 0;
371 }
372 
373 /*
374  * ND6 timer routine to expire default route list and prefix list
375  */
376 void
377 nd6_timer(ignored_arg)
378 	void	*ignored_arg;
379 {
380 	int s;
381 	register struct llinfo_nd6 *ln;
382 	register struct nd_defrouter *dr;
383 	register struct nd_prefix *pr;
384 	long time_second = time.tv_sec;
385 
386 	s = splsoftnet();
387 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
388 	    nd6_timer, NULL);
389 
390 	ln = llinfo_nd6.ln_next;
391 	/* XXX BSD/OS separates this code -- itojun */
392 	while (ln && ln != &llinfo_nd6) {
393 		struct rtentry *rt;
394 		struct ifnet *ifp;
395 		struct sockaddr_in6 *dst;
396 		struct llinfo_nd6 *next = ln->ln_next;
397 		/* XXX: used for the DELAY case only: */
398 		struct nd_ifinfo *ndi = NULL;
399 
400 		if ((rt = ln->ln_rt) == NULL) {
401 			ln = next;
402 			continue;
403 		}
404 		if ((ifp = rt->rt_ifp) == NULL) {
405 			ln = next;
406 			continue;
407 		}
408 		ndi = &nd_ifinfo[ifp->if_index];
409 		dst = (struct sockaddr_in6 *)rt_key(rt);
410 
411 		if (ln->ln_expire > time_second) {
412 			ln = next;
413 			continue;
414 		}
415 
416 		/* sanity check */
417 		if (!rt)
418 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
419 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
420 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
421 			      rt->rt_llinfo, ln);
422 		if (!dst)
423 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
424 
425 		switch (ln->ln_state) {
426 		case ND6_LLINFO_INCOMPLETE:
427 			if (ln->ln_asked < nd6_mmaxtries) {
428 				ln->ln_asked++;
429 				ln->ln_expire = time_second +
430 					nd_ifinfo[ifp->if_index].retrans / 1000;
431 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
432 					ln, 0);
433 			} else {
434 				struct mbuf *m = ln->ln_hold;
435 				if (m) {
436 					if (rt->rt_ifp) {
437 						/*
438 						 * Fake rcvif to make ICMP error
439 						 * more helpful in diagnosing
440 						 * for the receiver.
441 						 * XXX: should we consider
442 						 * older rcvif?
443 						 */
444 						m->m_pkthdr.rcvif = rt->rt_ifp;
445 					}
446 					icmp6_error(m, ICMP6_DST_UNREACH,
447 						    ICMP6_DST_UNREACH_ADDR, 0);
448 					ln->ln_hold = NULL;
449 				}
450 				nd6_free(rt);
451 			}
452 			break;
453 		case ND6_LLINFO_REACHABLE:
454 			if (ln->ln_expire)
455 				ln->ln_state = ND6_LLINFO_STALE;
456 			break;
457 		/*
458 		 * ND6_LLINFO_STALE state requires nothing for timer
459 		 * routine.
460 		 */
461 		case ND6_LLINFO_DELAY:
462 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
463 				/* We need NUD */
464 				ln->ln_asked = 1;
465 				ln->ln_state = ND6_LLINFO_PROBE;
466 				ln->ln_expire = time_second +
467 					ndi->retrans / 1000;
468 				nd6_ns_output(ifp, &dst->sin6_addr,
469 					      &dst->sin6_addr,
470 					      ln, 0);
471 			} else
472 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
473 			break;
474 		case ND6_LLINFO_PROBE:
475 			if (ln->ln_asked < nd6_umaxtries) {
476 				ln->ln_asked++;
477 				ln->ln_expire = time_second +
478 					nd_ifinfo[ifp->if_index].retrans / 1000;
479 				nd6_ns_output(ifp, &dst->sin6_addr,
480 					       &dst->sin6_addr, ln, 0);
481 			} else {
482 				nd6_free(rt);
483 			}
484 			break;
485 		case ND6_LLINFO_WAITDELETE:
486 			nd6_free(rt);
487 			break;
488 		}
489 		ln = next;
490 	}
491 
492 	/* expire */
493 	dr = TAILQ_FIRST(&nd_defrouter);
494 	while (dr) {
495 		if (dr->expire && dr->expire < time_second) {
496 			struct nd_defrouter *t;
497 			t = TAILQ_NEXT(dr, dr_entry);
498 			defrtrlist_del(dr);
499 			dr = t;
500 		} else {
501 			dr = TAILQ_NEXT(dr, dr_entry);
502 		}
503 	}
504 	pr = nd_prefix.lh_first;
505 	while (pr) {
506 		struct in6_ifaddr *ia6;
507 		struct in6_addrlifetime *lt6;
508 
509 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
510 			ia6 = NULL;
511 		else
512 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
513 
514 		if (ia6) {
515 			/* check address lifetime */
516 			lt6 = &ia6->ia6_lifetime;
517 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
518 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
519 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
520 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
521 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
522 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
523 			}
524 		}
525 
526 		/*
527 		 * check prefix lifetime.
528 		 * since pltime is just for autoconf, pltime processing for
529 		 * prefix is not necessary.
530 		 *
531 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
532 		 * can use the old prefix information to validate the
533 		 * next prefix information to come.  See prelist_update()
534 		 * for actual validation.
535 		 */
536 		if (pr->ndpr_expire
537 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
538 			struct nd_prefix *t;
539 			t = pr->ndpr_next;
540 
541 			/*
542 			 * address expiration and prefix expiration are
543 			 * separate.  NEVER perform in6_ifdel here.
544 			 */
545 
546 			prelist_remove(pr);
547 			pr = t;
548 		} else
549 			pr = pr->ndpr_next;
550 	}
551 	splx(s);
552 }
553 
554 /*
555  * Nuke neighbor cache/prefix/default router management table, right before
556  * ifp goes away.
557  */
558 void
559 nd6_purge(ifp)
560 	struct ifnet *ifp;
561 {
562 	struct llinfo_nd6 *ln, *nln;
563 	struct nd_defrouter *dr, *ndr, drany;
564 	struct nd_prefix *pr, *npr;
565 
566 	/* Nuke default router list entries toward ifp */
567 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
568 		/*
569 		 * The first entry of the list may be stored in
570 		 * the routing table, so we'll delete it later.
571 		 */
572 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
573 			ndr = TAILQ_NEXT(dr, dr_entry);
574 			if (dr->ifp == ifp)
575 				defrtrlist_del(dr);
576 		}
577 		dr = TAILQ_FIRST(&nd_defrouter);
578 		if (dr->ifp == ifp)
579 			defrtrlist_del(dr);
580 	}
581 
582 	/* Nuke prefix list entries toward ifp */
583 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
584 		npr = pr->ndpr_next;
585 		if (pr->ndpr_ifp == ifp) {
586 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
587 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
588 			prelist_remove(pr);
589 		}
590 	}
591 
592 	/* cancel default outgoing interface setting */
593 	if (nd6_defifindex == ifp->if_index)
594 		nd6_setdefaultiface(0);
595 
596 	/* refresh default router list */
597 	bzero(&drany, sizeof(drany));
598 	defrouter_delreq(&drany, 0);
599 	defrouter_select();
600 
601 	/*
602 	 * Nuke neighbor cache entries for the ifp.
603 	 * Note that rt->rt_ifp may not be the same as ifp,
604 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
605 	 * nd6_rtrequest(), and ip6_input().
606 	 */
607 	ln = llinfo_nd6.ln_next;
608 	while (ln && ln != &llinfo_nd6) {
609 		struct rtentry *rt;
610 		struct sockaddr_dl *sdl;
611 
612 		nln = ln->ln_next;
613 		rt = ln->ln_rt;
614 		if (rt && rt->rt_gateway &&
615 		    rt->rt_gateway->sa_family == AF_LINK) {
616 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
617 			if (sdl->sdl_index == ifp->if_index)
618 				nd6_free(rt);
619 		}
620 		ln = nln;
621 	}
622 
623 	/*
624 	 * Neighbor cache entry for interface route will be retained
625 	 * with ND6_LLINFO_WAITDELETE state, by nd6_free().  Nuke it.
626 	 */
627 	ln = llinfo_nd6.ln_next;
628 	while (ln && ln != &llinfo_nd6) {
629 		struct rtentry *rt;
630 		struct sockaddr_dl *sdl;
631 
632 		nln = ln->ln_next;
633 		rt = ln->ln_rt;
634 		if (rt && rt->rt_gateway &&
635 		    rt->rt_gateway->sa_family == AF_LINK) {
636 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
637 			if (sdl->sdl_index == ifp->if_index) {
638 				rtrequest(RTM_DELETE, rt_key(rt),
639 				    (struct sockaddr *)0, rt_mask(rt), 0,
640 				    (struct rtentry **)0);
641 			}
642 		}
643 		ln = nln;
644 	}
645 }
646 
647 struct rtentry *
648 nd6_lookup(addr6, create, ifp)
649 	struct in6_addr *addr6;
650 	int create;
651 	struct ifnet *ifp;
652 {
653 	struct rtentry *rt;
654 	struct sockaddr_in6 sin6;
655 
656 	bzero(&sin6, sizeof(sin6));
657 	sin6.sin6_len = sizeof(struct sockaddr_in6);
658 	sin6.sin6_family = AF_INET6;
659 	sin6.sin6_addr = *addr6;
660 	rt = rtalloc1((struct sockaddr *)&sin6, create);
661 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
662 		/*
663 		 * This is the case for the default route.
664 		 * If we want to create a neighbor cache for the address, we
665 		 * should free the route for the destination and allocate an
666 		 * interface route.
667 		 */
668 		if (create) {
669 			RTFREE(rt);
670 			rt = 0;
671 		}
672 	}
673 	if (!rt) {
674 		if (create && ifp) {
675 			int e;
676 
677 			/*
678 			 * If no route is available and create is set,
679 			 * we allocate a host route for the destination
680 			 * and treat it like an interface route.
681 			 * This hack is necessary for a neighbor which can't
682 			 * be covered by our own prefix.
683 			 */
684 			struct ifaddr *ifa =
685 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
686 			if (ifa == NULL)
687 				return(NULL);
688 
689 			/*
690 			 * Create a new route. RTF_LLINFO is necessary
691 			 * to create a Neighbor Cache entry for the
692 			 * destination in nd6_rtrequest which will be
693 			 * called in rtequest via ifa->ifa_rtrequest.
694 			 */
695 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
696 					   ifa->ifa_addr,
697 					   (struct sockaddr *)&all1_sa,
698 					   (ifa->ifa_flags |
699 					    RTF_HOST | RTF_LLINFO) &
700 					   ~RTF_CLONING,
701 					   &rt)) != 0)
702 				log(LOG_ERR,
703 				    "nd6_lookup: failed to add route for a "
704 				    "neighbor(%s), errno=%d\n",
705 				    ip6_sprintf(addr6), e);
706 			if (rt == NULL)
707 				return(NULL);
708 			if (rt->rt_llinfo) {
709 				struct llinfo_nd6 *ln =
710 					(struct llinfo_nd6 *)rt->rt_llinfo;
711 				ln->ln_state = ND6_LLINFO_NOSTATE;
712 			}
713 		} else
714 			return(NULL);
715 	}
716 	rt->rt_refcnt--;
717 	/*
718 	 * Validation for the entry.
719 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
720 	 *      it might be the loopback interface if the entry is for our
721 	 *      own address on a non-loopback interface. Instead, we should
722 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
723 	 */
724 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
725 	    rt->rt_gateway->sa_family != AF_LINK ||
726 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
727 		if (create) {
728 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
729 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
730 			/* xxx more logs... kazu */
731 		}
732 		return(0);
733 	}
734 	return(rt);
735 }
736 
737 /*
738  * Detect if a given IPv6 address identifies a neighbor on a given link.
739  * XXX: should take care of the destination of a p2p link?
740  */
741 int
742 nd6_is_addr_neighbor(addr, ifp)
743 	struct sockaddr_in6 *addr;
744 	struct ifnet *ifp;
745 {
746 	register struct ifaddr *ifa;
747 	int i;
748 
749 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
750 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
751 
752 	/*
753 	 * A link-local address is always a neighbor.
754 	 * XXX: we should use the sin6_scope_id field rather than the embedded
755 	 * interface index.
756 	 */
757 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
758 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
759 		return(1);
760 
761 	/*
762 	 * If the address matches one of our addresses,
763 	 * it should be a neighbor.
764 	 */
765 	for (ifa = ifp->if_addrlist.tqh_first;
766 	     ifa;
767 	     ifa = ifa->ifa_list.tqe_next)
768 	{
769 		if (ifa->ifa_addr->sa_family != AF_INET6)
770 			next: continue;
771 
772 		for (i = 0; i < 4; i++) {
773 			if ((IFADDR6(ifa).s6_addr32[i] ^
774 			     addr->sin6_addr.s6_addr32[i]) &
775 			    IFMASK6(ifa).s6_addr32[i])
776 				goto next;
777 		}
778 		return(1);
779 	}
780 
781 	/*
782 	 * Even if the address matches none of our addresses, it might be
783 	 * in the neighbor cache.
784 	 */
785 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
786 		return(1);
787 
788 	return(0);
789 #undef IFADDR6
790 #undef IFMASK6
791 }
792 
793 /*
794  * Free an nd6 llinfo entry.
795  */
796 void
797 nd6_free(rt)
798 	struct rtentry *rt;
799 {
800 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
801 	struct sockaddr_dl *sdl;
802 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
803 	struct nd_defrouter *dr;
804 
805 	/*
806 	 * Clear all destination cache entries for the neighbor.
807 	 * XXX: is it better to restrict this to hosts?
808 	 */
809 	pfctlinput(PRC_HOSTDEAD, rt_key(rt));
810 
811 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
812 		int s;
813 		s = splsoftnet();
814 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
815 				      rt->rt_ifp);
816 		if (ln->ln_router || dr) {
817 			/*
818 			 * rt6_flush must be called whether or not the neighbor
819 			 * is in the Default Router List.
820 			 * See a corresponding comment in nd6_na_input().
821 			 */
822 			rt6_flush(&in6, rt->rt_ifp);
823 		}
824 
825 		if (dr) {
826 			/*
827 			 * Unreachablity of a router might affect the default
828 			 * router selection and on-link detection of advertised
829 			 * prefixes.
830 			 */
831 
832 			/*
833 			 * Temporarily fake the state to choose a new default
834 			 * router and to perform on-link determination of
835 			 * prefixes coreectly.
836 			 * Below the state will be set correctly,
837 			 * or the entry itself will be deleted.
838 			 */
839 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
840 
841 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
842 				/*
843 				 * It is used as the current default router,
844 				 * so we have to move it to the end of the
845 				 * list and choose a new one.
846 				 * XXX: it is not very efficient if this is
847 				 *      the only router.
848 				 */
849 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
850 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
851 
852 				defrouter_select();
853 			}
854 			pfxlist_onlink_check();
855 		}
856 		splx(s);
857 	}
858 
859 	if (rt->rt_refcnt > 0 && (sdl = SDL(rt->rt_gateway)) &&
860 	    sdl->sdl_family == AF_LINK) {
861 		sdl->sdl_alen = 0;
862 		ln->ln_state = ND6_LLINFO_WAITDELETE;
863 		ln->ln_asked = 0;
864 		rt->rt_flags &= ~RTF_REJECT;
865 		return;
866 	}
867 
868 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
869 		  rt_mask(rt), 0, (struct rtentry **)0);
870 }
871 
872 /*
873  * Upper-layer reachability hint for Neighbor Unreachability Detection.
874  *
875  * XXX cost-effective metods?
876  */
877 void
878 nd6_nud_hint(rt, dst6, force)
879 	struct rtentry *rt;
880 	struct in6_addr *dst6;
881 	int force;
882 {
883 	struct llinfo_nd6 *ln;
884 	long time_second = time.tv_sec;
885 
886 	/*
887 	 * If the caller specified "rt", use that.  Otherwise, resolve the
888 	 * routing table by supplied "dst6".
889 	 */
890 	if (!rt) {
891 		if (!dst6)
892 			return;
893 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
894 			return;
895 	}
896 
897 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
898 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
899 	    !rt->rt_llinfo || !rt->rt_gateway ||
900 	    rt->rt_gateway->sa_family != AF_LINK) {
901 		/* This is not a host route. */
902 		return;
903 	}
904 
905 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
906 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
907 		return;
908 
909 	/*
910 	 * if we get upper-layer reachability confirmation many times,
911 	 * it is possible we have false information.
912 	 */
913 	if (!force) {
914 		ln->ln_byhint++;
915 		if (ln->ln_byhint > nd6_maxnudhint)
916 			return;
917 	}
918 
919 	ln->ln_state = ND6_LLINFO_REACHABLE;
920 	if (ln->ln_expire)
921 		ln->ln_expire = time_second +
922 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
923 }
924 
925 #ifdef OLDIP6OUTPUT
926 /*
927  * Resolve an IP6 address into an ethernet address. If success,
928  * desten is filled in. If there is no entry in ndptab,
929  * set one up and multicast a solicitation for the IP6 address.
930  * Hold onto this mbuf and resend it once the address
931  * is finally resolved. A return value of 1 indicates
932  * that desten has been filled in and the packet should be sent
933  * normally; a 0 return indicates that the packet has been
934  * taken over here, either now or for later transmission.
935  */
936 int
937 nd6_resolve(ifp, rt, m, dst, desten)
938 	struct ifnet *ifp;
939 	struct rtentry *rt;
940 	struct mbuf *m;
941 	struct sockaddr *dst;
942 	u_char *desten;
943 {
944 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)NULL;
945 	struct sockaddr_dl *sdl;
946 	long time_second = time.tv_sec;
947 
948 	if (m->m_flags & M_MCAST) {
949 		switch (ifp->if_type) {
950 		case IFT_ETHER:
951 		case IFT_FDDI:
952 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
953 						 desten);
954 			return(1);
955 			break;
956 		case IFT_ARCNET:
957 			*desten = 0;
958 			return(1);
959 			break;
960 		default:
961 			return(0);
962 		}
963 	}
964 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
965 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
966 	else {
967 		if ((rt = nd6_lookup(&(SIN6(dst)->sin6_addr), 1, ifp)) != NULL)
968 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
969 	}
970 	if (!ln || !rt) {
971 		log(LOG_DEBUG, "nd6_resolve: can't allocate llinfo for %s\n",
972 			ip6_sprintf(&(SIN6(dst)->sin6_addr)));
973 		m_freem(m);
974 		return(0);
975 	}
976 	sdl = SDL(rt->rt_gateway);
977 	/*
978 	 * Ckeck the address family and length is valid, the address
979 	 * is resolved; otherwise, try to resolve.
980 	 */
981 	if (ln->ln_state >= ND6_LLINFO_REACHABLE
982 	   && sdl->sdl_family == AF_LINK
983 	   && sdl->sdl_alen != 0) {
984 		bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
985 		if (ln->ln_state == ND6_LLINFO_STALE) {
986 			ln->ln_asked = 0;
987 			ln->ln_state = ND6_LLINFO_DELAY;
988 			ln->ln_expire = time_second + nd6_delay;
989 		}
990 		return(1);
991 	}
992 	/*
993 	 * There is an ndp entry, but no ethernet address
994 	 * response yet. Replace the held mbuf with this
995 	 * latest one.
996 	 *
997 	 * XXX Does the code conform to rate-limiting rule?
998 	 * (RFC 2461 7.2.2)
999 	 */
1000 	if (ln->ln_state == ND6_LLINFO_WAITDELETE ||
1001 	    ln->ln_state == ND6_LLINFO_NOSTATE)
1002 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1003 	if (ln->ln_hold)
1004 		m_freem(ln->ln_hold);
1005 	ln->ln_hold = m;
1006 	if (ln->ln_expire) {
1007 		rt->rt_flags &= ~RTF_REJECT;
1008 		if (ln->ln_asked < nd6_mmaxtries &&
1009 		    ln->ln_expire < time_second) {
1010 			ln->ln_asked++;
1011 			ln->ln_expire = time_second +
1012 				nd_ifinfo[ifp->if_index].retrans / 1000;
1013 			nd6_ns_output(ifp, NULL, &(SIN6(dst)->sin6_addr),
1014 				ln, 0);
1015 		}
1016 	}
1017 	return(0);
1018 }
1019 #endif /* OLDIP6OUTPUT */
1020 
1021 void
1022 nd6_rtrequest(req, rt, sa)
1023 	int	req;
1024 	struct rtentry *rt;
1025 	struct sockaddr *sa; /* xxx unused */
1026 {
1027 	struct sockaddr *gate = rt->rt_gateway;
1028 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1029 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1030 	struct ifnet *ifp = rt->rt_ifp;
1031 	struct ifaddr *ifa;
1032 	long time_second = time.tv_sec;
1033 
1034 	if (rt->rt_flags & RTF_GATEWAY)
1035 		return;
1036 
1037 	switch (req) {
1038 	case RTM_ADD:
1039 		/*
1040 		 * There is no backward compatibility :)
1041 		 *
1042 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1043 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1044 		 *	   rt->rt_flags |= RTF_CLONING;
1045 		 */
1046 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1047 			/*
1048 			 * Case 1: This route should come from
1049 			 * a route to interface. RTF_LLINFO flag is set
1050 			 * for a host route whose destination should be
1051 			 * treated as on-link.
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 				ln->ln_expire = time_second;
1060 #if 1
1061 			if (ln && ln->ln_expire == 0) {
1062 				/* cludge for desktops */
1063 #if 0
1064 				printf("nd6_request: time.tv_sec is zero; "
1065 				       "treat it as 1\n");
1066 #endif
1067 				ln->ln_expire = 1;
1068 			}
1069 #endif
1070 			if (rt->rt_flags & RTF_CLONING)
1071 				break;
1072 		}
1073 		/*
1074 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1075 		 * We don't do that here since llinfo is not ready yet.
1076 		 *
1077 		 * There are also couple of other things to be discussed:
1078 		 * - unsolicited NA code needs improvement beforehand
1079 		 * - RFC2461 says we MAY send multicast unsolicited NA
1080 		 *   (7.2.6 paragraph 4), however, it also says that we
1081 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1082 		 *   we don't have anything like it right now.
1083 		 *   note that the mechanism need a mutual agreement
1084 		 *   between proxies, which means that we need to implement
1085 		 *   a new protocol, or new kludge.
1086 		 * - from RFC2461 6.2.4, host MUST NOT send unsolicited NA.
1087 		 *   we need to check ip6forwarding before sending it.
1088 		 *   (or should we allow proxy ND configuration only for
1089 		 *   routers?  there's no mention about proxy ND from hosts)
1090 		 */
1091 #if 0
1092 		/* XXX it does not work */
1093 		if (rt->rt_flags & RTF_ANNOUNCE)
1094 			nd6_na_output(ifp,
1095 			      &SIN6(rt_key(rt))->sin6_addr,
1096 			      &SIN6(rt_key(rt))->sin6_addr,
1097 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1098 			      1, NULL);
1099 #endif
1100 		/* FALLTHROUGH */
1101 	case RTM_RESOLVE:
1102 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
1103 			/*
1104 			 * Address resolution isn't necessary for a point to
1105 			 * point link, so we can skip this test for a p2p link.
1106 			 */
1107 			if (gate->sa_family != AF_LINK ||
1108 			    gate->sa_len < sizeof(null_sdl)) {
1109 				log(LOG_DEBUG,
1110 				    "nd6_rtrequest: bad gateway value\n");
1111 				break;
1112 			}
1113 			SDL(gate)->sdl_type = ifp->if_type;
1114 			SDL(gate)->sdl_index = ifp->if_index;
1115 		}
1116 		if (ln != NULL)
1117 			break;	/* This happens on a route change */
1118 		/*
1119 		 * Case 2: This route may come from cloning, or a manual route
1120 		 * add with a LL address.
1121 		 */
1122 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1123 		rt->rt_llinfo = (caddr_t)ln;
1124 		if (!ln) {
1125 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1126 			break;
1127 		}
1128 		nd6_inuse++;
1129 		nd6_allocated++;
1130 		Bzero(ln, sizeof(*ln));
1131 		ln->ln_rt = rt;
1132 		/* this is required for "ndp" command. - shin */
1133 		if (req == RTM_ADD) {
1134 		        /*
1135 			 * gate should have some valid AF_LINK entry,
1136 			 * and ln->ln_expire should have some lifetime
1137 			 * which is specified by ndp command.
1138 			 */
1139 			ln->ln_state = ND6_LLINFO_REACHABLE;
1140 			ln->ln_byhint = 0;
1141 		} else {
1142 		        /*
1143 			 * When req == RTM_RESOLVE, rt is created and
1144 			 * initialized in rtrequest(), so rt_expire is 0.
1145 			 */
1146 			ln->ln_state = ND6_LLINFO_NOSTATE;
1147 			ln->ln_expire = time_second;
1148 		}
1149 		rt->rt_flags |= RTF_LLINFO;
1150 		ln->ln_next = llinfo_nd6.ln_next;
1151 		llinfo_nd6.ln_next = ln;
1152 		ln->ln_prev = &llinfo_nd6;
1153 		ln->ln_next->ln_prev = ln;
1154 
1155 		/*
1156 		 * check if rt_key(rt) is one of my address assigned
1157 		 * to the interface.
1158 		 */
1159 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1160 					  &SIN6(rt_key(rt))->sin6_addr);
1161 		if (ifa) {
1162 			caddr_t macp = nd6_ifptomac(ifp);
1163 			ln->ln_expire = 0;
1164 			ln->ln_state = ND6_LLINFO_REACHABLE;
1165 			ln->ln_byhint = 0;
1166 			if (macp) {
1167 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1168 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1169 			}
1170 			if (nd6_useloopback) {
1171 				rt->rt_ifp = &loif[0];	/*XXX*/
1172 				/*
1173 				 * Make sure rt_ifa be equal to the ifaddr
1174 				 * corresponding to the address.
1175 				 * We need this because when we refer
1176 				 * rt_ifa->ia6_flags in ip6_input, we assume
1177 				 * that the rt_ifa points to the address instead
1178 				 * of the loopback address.
1179 				 */
1180 				if (ifa != rt->rt_ifa) {
1181 					IFAFREE(rt->rt_ifa);
1182 					IFAREF(ifa);
1183 					rt->rt_ifa = ifa;
1184 				}
1185 			}
1186 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1187 			ln->ln_expire = 0;
1188 			ln->ln_state = ND6_LLINFO_REACHABLE;
1189 			ln->ln_byhint = 0;
1190 
1191 			/* join solicited node multicast for proxy ND */
1192 			if (ifp->if_flags & IFF_MULTICAST) {
1193 				struct in6_addr llsol;
1194 				int error;
1195 
1196 				llsol = SIN6(rt_key(rt))->sin6_addr;
1197 				llsol.s6_addr16[0] = htons(0xff02);
1198 				llsol.s6_addr16[1] = htons(ifp->if_index);
1199 				llsol.s6_addr32[1] = 0;
1200 				llsol.s6_addr32[2] = htonl(1);
1201 				llsol.s6_addr8[12] = 0xff;
1202 
1203 				(void)in6_addmulti(&llsol, ifp, &error);
1204 				if (error)
1205 					printf(
1206 "nd6_rtrequest: could not join solicited node multicast (errno=%d)\n", error);
1207 			}
1208 		}
1209 		break;
1210 
1211 	case RTM_DELETE:
1212 		if (!ln)
1213 			break;
1214 		/* leave from solicited node multicast for proxy ND */
1215 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1216 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1217 			struct in6_addr llsol;
1218 			struct in6_multi *in6m;
1219 
1220 			llsol = SIN6(rt_key(rt))->sin6_addr;
1221 			llsol.s6_addr16[0] = htons(0xff02);
1222 			llsol.s6_addr16[1] = htons(ifp->if_index);
1223 			llsol.s6_addr32[1] = 0;
1224 			llsol.s6_addr32[2] = htonl(1);
1225 			llsol.s6_addr8[12] = 0xff;
1226 
1227 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1228 			if (in6m)
1229 				in6_delmulti(in6m);
1230 		}
1231 		nd6_inuse--;
1232 		ln->ln_next->ln_prev = ln->ln_prev;
1233 		ln->ln_prev->ln_next = ln->ln_next;
1234 		ln->ln_prev = NULL;
1235 		rt->rt_llinfo = 0;
1236 		rt->rt_flags &= ~RTF_LLINFO;
1237 		if (ln->ln_hold)
1238 			m_freem(ln->ln_hold);
1239 		Free((caddr_t)ln);
1240 	}
1241 }
1242 
1243 void
1244 nd6_p2p_rtrequest(req, rt, sa)
1245 	int	req;
1246 	struct rtentry *rt;
1247 	struct sockaddr *sa; /* xxx unused */
1248 {
1249 	struct sockaddr *gate = rt->rt_gateway;
1250 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1251 	struct ifnet *ifp = rt->rt_ifp;
1252 	struct ifaddr *ifa;
1253 
1254 	if (rt->rt_flags & RTF_GATEWAY)
1255 		return;
1256 
1257 	switch (req) {
1258 	case RTM_ADD:
1259 		/*
1260 		 * There is no backward compatibility :)
1261 		 *
1262 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1263 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1264 		 *	   rt->rt_flags |= RTF_CLONING;
1265 		 */
1266 		if (rt->rt_flags & RTF_CLONING) {
1267 			/*
1268 			 * Case 1: This route should come from
1269 			 * a route to interface.
1270 			 */
1271 			rt_setgate(rt, rt_key(rt),
1272 				   (struct sockaddr *)&null_sdl);
1273 			gate = rt->rt_gateway;
1274 			SDL(gate)->sdl_type = ifp->if_type;
1275 			SDL(gate)->sdl_index = ifp->if_index;
1276 			break;
1277 		}
1278 		/* Announce a new entry if requested. */
1279 		if (rt->rt_flags & RTF_ANNOUNCE)
1280 			nd6_na_output(ifp,
1281 				      &SIN6(rt_key(rt))->sin6_addr,
1282 				      &SIN6(rt_key(rt))->sin6_addr,
1283 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1284 				      1, NULL);
1285 		/* FALLTHROUGH */
1286 	case RTM_RESOLVE:
1287 		/*
1288 		 * check if rt_key(rt) is one of my address assigned
1289 		 * to the interface.
1290 		 */
1291  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1292 					  &SIN6(rt_key(rt))->sin6_addr);
1293 		if (ifa) {
1294 			if (nd6_useloopback) {
1295 				rt->rt_ifp = &loif[0];	/*XXX*/
1296 			}
1297 		}
1298 		break;
1299 	}
1300 }
1301 
1302 int
1303 nd6_ioctl(cmd, data, ifp)
1304 	u_long cmd;
1305 	caddr_t	data;
1306 	struct ifnet *ifp;
1307 {
1308 	struct in6_drlist *drl = (struct in6_drlist *)data;
1309 	struct in6_prlist *prl = (struct in6_prlist *)data;
1310 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1311 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1312 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1313 	struct nd_defrouter *dr, any;
1314 	struct nd_prefix *pr;
1315 	struct rtentry *rt;
1316 	int i = 0, error = 0;
1317 	int s;
1318 
1319 	switch (cmd) {
1320 	case SIOCGDRLST_IN6:
1321 		bzero(drl, sizeof(*drl));
1322 		s = splsoftnet();
1323 		dr = TAILQ_FIRST(&nd_defrouter);
1324 		while (dr && i < DRLSTSIZ) {
1325 			drl->defrouter[i].rtaddr = dr->rtaddr;
1326 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1327 				/* XXX: need to this hack for KAME stack */
1328 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1329 			} else
1330 				log(LOG_ERR,
1331 				    "default router list contains a "
1332 				    "non-linklocal address(%s)\n",
1333 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1334 
1335 			drl->defrouter[i].flags = dr->flags;
1336 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1337 			drl->defrouter[i].expire = dr->expire;
1338 			drl->defrouter[i].if_index = dr->ifp->if_index;
1339 			i++;
1340 			dr = TAILQ_NEXT(dr, dr_entry);
1341 		}
1342 		splx(s);
1343 		break;
1344 	case SIOCGPRLST_IN6:
1345 		/*
1346 		 * XXX meaning of fields, especialy "raflags", is very
1347 		 * differnet between RA prefix list and RR/static prefix list.
1348 		 * how about separating ioctls into two?
1349 		 */
1350 		bzero(prl, sizeof(*prl));
1351 		s = splsoftnet();
1352 		pr = nd_prefix.lh_first;
1353 		while (pr && i < PRLSTSIZ) {
1354 			struct nd_pfxrouter *pfr;
1355 			int j;
1356 
1357 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1358 			prl->prefix[i].raflags = pr->ndpr_raf;
1359 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1360 			prl->prefix[i].vltime = pr->ndpr_vltime;
1361 			prl->prefix[i].pltime = pr->ndpr_pltime;
1362 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1363 			prl->prefix[i].expire = pr->ndpr_expire;
1364 
1365 			pfr = pr->ndpr_advrtrs.lh_first;
1366 			j = 0;
1367 			while(pfr) {
1368 				if (j < DRLSTSIZ) {
1369 #define RTRADDR prl->prefix[i].advrtr[j]
1370 					RTRADDR = pfr->router->rtaddr;
1371 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1372 						/* XXX: hack for KAME */
1373 						RTRADDR.s6_addr16[1] = 0;
1374 					} else
1375 						log(LOG_ERR,
1376 						    "a router(%s) advertises "
1377 						    "a prefix with "
1378 						    "non-link local address\n",
1379 						    ip6_sprintf(&RTRADDR));
1380 #undef RTRADDR
1381 				}
1382 				j++;
1383 				pfr = pfr->pfr_next;
1384 			}
1385 			prl->prefix[i].advrtrs = j;
1386 			prl->prefix[i].origin = PR_ORIG_RA;
1387 
1388 			i++;
1389 			pr = pr->ndpr_next;
1390 		}
1391 	      {
1392 		struct rr_prefix *rpp;
1393 
1394 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1395 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1396 			if (i >= PRLSTSIZ)
1397 				break;
1398 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
1399 			prl->prefix[i].raflags = rpp->rp_raf;
1400 			prl->prefix[i].prefixlen = rpp->rp_plen;
1401 			prl->prefix[i].vltime = rpp->rp_vltime;
1402 			prl->prefix[i].pltime = rpp->rp_pltime;
1403 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1404 			prl->prefix[i].expire = rpp->rp_expire;
1405 			prl->prefix[i].advrtrs = 0;
1406 			prl->prefix[i].origin = rpp->rp_origin;
1407 			i++;
1408 		}
1409 	      }
1410 		splx(s);
1411 
1412 		break;
1413 	case SIOCGIFINFO_IN6:
1414 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1415 			error = EINVAL;
1416 			break;
1417 		}
1418 		ndi->ndi = nd_ifinfo[ifp->if_index];
1419 		break;
1420 	case SIOCSIFINFO_FLAGS:
1421 		/* XXX: almost all other fields of ndi->ndi is unused */
1422 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1423 			error = EINVAL;
1424 			break;
1425 		}
1426 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1427 		break;
1428 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1429 		/* flush default router list */
1430 		/*
1431 		 * xxx sumikawa: should not delete route if default
1432 		 * route equals to the top of default router list
1433 		 */
1434 		bzero(&any, sizeof(any));
1435 		defrouter_delreq(&any, 0);
1436 		defrouter_select();
1437 		/* xxx sumikawa: flush prefix list */
1438 		break;
1439 	case SIOCSPFXFLUSH_IN6:
1440 	    {
1441 		/* flush all the prefix advertised by routers */
1442 		struct nd_prefix *pr, *next;
1443 
1444 		s = splsoftnet();
1445 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1446 			next = pr->ndpr_next;
1447 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
1448 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
1449 			prelist_remove(pr);
1450 		}
1451 		splx(s);
1452 		break;
1453 	    }
1454 	case SIOCSRTRFLUSH_IN6:
1455 	    {
1456 		/* flush all the default routers */
1457 		struct nd_defrouter *dr, *next;
1458 
1459 		s = splsoftnet();
1460 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1461 			/*
1462 			 * The first entry of the list may be stored in
1463 			 * the routing table, so we'll delete it later.
1464 			 */
1465 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1466 				next = TAILQ_NEXT(dr, dr_entry);
1467 				defrtrlist_del(dr);
1468 			}
1469 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1470 		}
1471 		splx(s);
1472 		break;
1473 	    }
1474 	case SIOCGNBRINFO_IN6:
1475 	    {
1476 		struct llinfo_nd6 *ln;
1477 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1478 
1479 		/*
1480 		 * XXX: KAME specific hack for scoped addresses
1481 		 *      XXXX: for other scopes than link-local?
1482 		 */
1483 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1484 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1485 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1486 
1487 			if (*idp == 0)
1488 				*idp = htons(ifp->if_index);
1489 		}
1490 
1491 		s = splsoftnet();
1492 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1493 			error = EINVAL;
1494 			splx(s);
1495 			break;
1496 		}
1497 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1498 		nbi->state = ln->ln_state;
1499 		nbi->asked = ln->ln_asked;
1500 		nbi->isrouter = ln->ln_router;
1501 		nbi->expire = ln->ln_expire;
1502 		splx(s);
1503 
1504 		break;
1505 	    }
1506 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1507 		ndif->ifindex = nd6_defifindex;
1508 		break;
1509 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1510 		return(nd6_setdefaultiface(ndif->ifindex));
1511 		break;
1512 	}
1513 	return(error);
1514 }
1515 
1516 /*
1517  * Create neighbor cache entry and cache link-layer address,
1518  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1519  */
1520 struct rtentry *
1521 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1522 	struct ifnet *ifp;
1523 	struct in6_addr *from;
1524 	char *lladdr;
1525 	int lladdrlen;
1526 	int type;	/* ICMP6 type */
1527 	int code;	/* type dependent information */
1528 {
1529 	struct rtentry *rt = NULL;
1530 	struct llinfo_nd6 *ln = NULL;
1531 	int is_newentry;
1532 	struct sockaddr_dl *sdl = NULL;
1533 	int do_update;
1534 	int olladdr;
1535 	int llchange;
1536 	int newstate = 0;
1537 	long time_second = time.tv_sec;
1538 
1539 	if (!ifp)
1540 		panic("ifp == NULL in nd6_cache_lladdr");
1541 	if (!from)
1542 		panic("from == NULL in nd6_cache_lladdr");
1543 
1544 	/* nothing must be updated for unspecified address */
1545 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1546 		return NULL;
1547 
1548 	/*
1549 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1550 	 * the caller.
1551 	 *
1552 	 * XXX If the link does not have link-layer adderss, what should
1553 	 * we do? (ifp->if_addrlen == 0)
1554 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1555 	 * description on it in NS section (RFC 2461 7.2.3).
1556 	 */
1557 
1558 	rt = nd6_lookup(from, 0, ifp);
1559 	if (!rt) {
1560 #if 0
1561 		/* nothing must be done if there's no lladdr */
1562 		if (!lladdr || !lladdrlen)
1563 			return NULL;
1564 #endif
1565 
1566 		rt = nd6_lookup(from, 1, ifp);
1567 		is_newentry = 1;
1568 	} else
1569 		is_newentry = 0;
1570 
1571 	if (!rt)
1572 		return NULL;
1573 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1574 fail:
1575 		nd6_free(rt);
1576 		return NULL;
1577 	}
1578 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1579 	if (!ln)
1580 		goto fail;
1581 	if (!rt->rt_gateway)
1582 		goto fail;
1583 	if (rt->rt_gateway->sa_family != AF_LINK)
1584 		goto fail;
1585 	sdl = SDL(rt->rt_gateway);
1586 
1587 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1588 	if (olladdr && lladdr) {
1589 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1590 			llchange = 1;
1591 		else
1592 			llchange = 0;
1593 	} else
1594 		llchange = 0;
1595 
1596 	/*
1597 	 * newentry olladdr  lladdr  llchange	(*=record)
1598 	 *	0	n	n	--	(1)
1599 	 *	0	y	n	--	(2)
1600 	 *	0	n	y	--	(3) * STALE
1601 	 *	0	y	y	n	(4) *
1602 	 *	0	y	y	y	(5) * STALE
1603 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1604 	 *	1	--	y	--	(7) * STALE
1605 	 */
1606 
1607 	if (lladdr) {		/*(3-5) and (7)*/
1608 		/*
1609 		 * Record source link-layer address
1610 		 * XXX is it dependent to ifp->if_type?
1611 		 */
1612 		sdl->sdl_alen = ifp->if_addrlen;
1613 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1614 	}
1615 
1616 	if (!is_newentry) {
1617 		if ((!olladdr && lladdr)		/*(3)*/
1618 		 || (olladdr && lladdr && llchange)) {	/*(5)*/
1619 			do_update = 1;
1620 			newstate = ND6_LLINFO_STALE;
1621 		} else					/*(1-2,4)*/
1622 			do_update = 0;
1623 	} else {
1624 		do_update = 1;
1625 		if (!lladdr)				/*(6)*/
1626 			newstate = ND6_LLINFO_NOSTATE;
1627 		else					/*(7)*/
1628 			newstate = ND6_LLINFO_STALE;
1629 	}
1630 
1631 	if (do_update) {
1632 		/*
1633 		 * Update the state of the neighbor cache.
1634 		 */
1635 		ln->ln_state = newstate;
1636 
1637 		if (ln->ln_state == ND6_LLINFO_STALE) {
1638 			rt->rt_flags &= ~RTF_REJECT;
1639 			if (ln->ln_hold) {
1640 #ifdef OLDIP6OUTPUT
1641 				(*ifp->if_output)(ifp, ln->ln_hold,
1642 						  rt_key(rt), rt);
1643 #else
1644 				/*
1645 				 * we assume ifp is not a p2p here, so just
1646 				 * set the 2nd argument as the 1st one.
1647 				 */
1648 				nd6_output(ifp, ifp, ln->ln_hold,
1649 					   (struct sockaddr_in6 *)rt_key(rt),
1650 					   rt);
1651 #endif
1652 				ln->ln_hold = 0;
1653 			}
1654 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1655 			/* probe right away */
1656 			ln->ln_expire = time_second;
1657 		}
1658 	}
1659 
1660 	/*
1661 	 * ICMP6 type dependent behavior.
1662 	 *
1663 	 * NS: clear IsRouter if new entry
1664 	 * RS: clear IsRouter
1665 	 * RA: set IsRouter if there's lladdr
1666 	 * redir: clear IsRouter if new entry
1667 	 *
1668 	 * RA case, (1):
1669 	 * The spec says that we must set IsRouter in the following cases:
1670 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1671 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1672 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1673 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1674 	 * neighbor cache, this is similar to (6).
1675 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1676 	 *
1677 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1678 	 *							D R
1679 	 *	0	n	n	--	(1)	c   ?     s
1680 	 *	0	y	n	--	(2)	c   s     s
1681 	 *	0	n	y	--	(3)	c   s     s
1682 	 *	0	y	y	n	(4)	c   s     s
1683 	 *	0	y	y	y	(5)	c   s     s
1684 	 *	1	--	n	--	(6) c	c 	c s
1685 	 *	1	--	y	--	(7) c	c   s	c s
1686 	 *
1687 	 *					(c=clear s=set)
1688 	 */
1689 	switch (type & 0xff) {
1690 	case ND_NEIGHBOR_SOLICIT:
1691 		/*
1692 		 * New entry must have is_router flag cleared.
1693 		 */
1694 		if (is_newentry)	/*(6-7)*/
1695 			ln->ln_router = 0;
1696 		break;
1697 	case ND_REDIRECT:
1698 		/*
1699 		 * If the icmp is a redirect to a better router, always set the
1700 		 * is_router flag. Otherwise, if the entry is newly created,
1701 		 * clear the flag. [RFC 2461, sec 8.3]
1702 		 */
1703 		if (code == ND_REDIRECT_ROUTER)
1704 			ln->ln_router = 1;
1705 		else if (is_newentry) /*(6-7)*/
1706 			ln->ln_router = 0;
1707 		break;
1708 	case ND_ROUTER_SOLICIT:
1709 		/*
1710 		 * is_router flag must always be cleared.
1711 		 */
1712 		ln->ln_router = 0;
1713 		break;
1714 	case ND_ROUTER_ADVERT:
1715 		/*
1716 		 * Mark an entry with lladdr as a router.
1717 		 */
1718 		if ((!is_newentry && (olladdr || lladdr))	/*(2-5)*/
1719 		 || (is_newentry && lladdr)) {			/*(7)*/
1720 			ln->ln_router = 1;
1721 		}
1722 		break;
1723 	}
1724 
1725 	return rt;
1726 }
1727 
1728 static void
1729 nd6_slowtimo(ignored_arg)
1730     void *ignored_arg;
1731 {
1732 	int s = splsoftnet();
1733 	register int i;
1734 	register struct nd_ifinfo *nd6if;
1735 
1736 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1737 	    nd6_slowtimo, NULL);
1738 	for (i = 1; i < if_index + 1; i++) {
1739 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1740 			continue;
1741 		nd6if = &nd_ifinfo[i];
1742 		if (nd6if->basereachable && /* already initialized */
1743 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1744 			/*
1745 			 * Since reachable time rarely changes by router
1746 			 * advertisements, we SHOULD insure that a new random
1747 			 * value gets recomputed at least once every few hours.
1748 			 * (RFC 2461, 6.3.4)
1749 			 */
1750 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1751 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1752 		}
1753 	}
1754 	splx(s);
1755 }
1756 
1757 #define senderr(e) { error = (e); goto bad;}
1758 int
1759 nd6_output(ifp, origifp, m0, dst, rt0)
1760 	register struct ifnet *ifp;
1761 	struct ifnet *origifp;
1762 	struct mbuf *m0;
1763 	struct sockaddr_in6 *dst;
1764 	struct rtentry *rt0;
1765 {
1766 	register struct mbuf *m = m0;
1767 	register struct rtentry *rt = rt0;
1768 	struct sockaddr_in6 *gw6 = NULL;
1769 	struct llinfo_nd6 *ln = NULL;
1770 	int error = 0;
1771 	long time_second = time.tv_sec;
1772 
1773 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1774 		goto sendpkt;
1775 
1776 	/*
1777 	 * XXX: we currently do not make neighbor cache on any interface
1778 	 * other than ARCnet, Ethernet, FDDI and GIF.
1779 	 *
1780 	 * draft-ietf-ngtrans-mech-06.txt says:
1781 	 * - unidirectional tunnels needs no ND
1782 	 */
1783 	switch (ifp->if_type) {
1784 	case IFT_ARCNET:
1785 	case IFT_ETHER:
1786 	case IFT_FDDI:
1787 	case IFT_GIF:		/* XXX need more cases? */
1788 		break;
1789 	default:
1790 		goto sendpkt;
1791 	}
1792 
1793 	/*
1794 	 * next hop determination. This routine is derived from ether_outpout.
1795 	 */
1796 	if (rt) {
1797 		if ((rt->rt_flags & RTF_UP) == 0) {
1798 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
1799 				NULL)
1800 			{
1801 				rt->rt_refcnt--;
1802 				if (rt->rt_ifp != ifp) {
1803 					/* XXX: loop care? */
1804 					return nd6_output(ifp, origifp, m0,
1805 							  dst, rt);
1806 				}
1807 			} else
1808 				senderr(EHOSTUNREACH);
1809 		}
1810 
1811 		if (rt->rt_flags & RTF_GATEWAY) {
1812 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1813 
1814 			/*
1815 			 * We skip link-layer address resolution and NUD
1816 			 * if the gateway is not a neighbor from ND point
1817 			 * of view, regardless the value of the value of
1818 			 * nd_ifinfo.flags.
1819 			 * The second condition is a bit tricky: we skip
1820 			 * if the gateway is our own address, which is
1821 			 * sometimes used to install a route to a p2p link.
1822 			 */
1823 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1824 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1825 				if (rt->rt_flags & RTF_REJECT)
1826 					senderr(EHOSTDOWN);
1827 
1828 				/*
1829 				 * We allow this kind of tricky route only
1830 				 * when the outgoing interface is p2p.
1831 				 * XXX: we may need a more generic rule here.
1832 				 */
1833 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1834 					senderr(EHOSTUNREACH);
1835 
1836 				goto sendpkt;
1837 			}
1838 
1839 			if (rt->rt_gwroute == 0)
1840 				goto lookup;
1841 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1842 				rtfree(rt); rt = rt0;
1843 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1844 				if ((rt = rt->rt_gwroute) == 0)
1845 					senderr(EHOSTUNREACH);
1846 			}
1847 		}
1848 		if (rt->rt_flags & RTF_REJECT)
1849 			senderr(rt == rt0 ? EHOSTDOWN : EHOSTUNREACH);
1850 	}
1851 
1852 	/*
1853 	 * Address resolution or Neighbor Unreachability Detection
1854 	 * for the next hop.
1855 	 * At this point, the destination of the packet must be a unicast
1856 	 * or an anycast address(i.e. not a multicast).
1857 	 */
1858 
1859 	/* Look up the neighbor cache for the nexthop */
1860 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1861 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1862 	else {
1863 		/*
1864 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1865 		 * the condition below is not very efficient. But we believe
1866 		 * it is tolerable, because this should be a rare case.
1867 		 */
1868 		if (nd6_is_addr_neighbor(dst, ifp) &&
1869 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1870 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1871 	}
1872 	if (!ln || !rt) {
1873 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1874 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1875 			log(LOG_DEBUG,
1876 			    "nd6_output: can't allocate llinfo for %s "
1877 			    "(ln=%p, rt=%p)\n",
1878 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1879 			senderr(EIO);	/* XXX: good error? */
1880 		}
1881 
1882 		goto sendpkt;	/* send anyway */
1883 	}
1884 
1885 	/* We don't have to do link-layer address resolution on a p2p link. */
1886 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1887 	    ln->ln_state < ND6_LLINFO_REACHABLE)
1888 		ln->ln_state = ND6_LLINFO_STALE;
1889 
1890 	/*
1891 	 * The first time we send a packet to a neighbor whose entry is
1892 	 * STALE, we have to change the state to DELAY and a sets a timer to
1893 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1894 	 * neighbor unreachability detection on expiration.
1895 	 * (RFC 2461 7.3.3)
1896 	 */
1897 	if (ln->ln_state == ND6_LLINFO_STALE) {
1898 		ln->ln_asked = 0;
1899 		ln->ln_state = ND6_LLINFO_DELAY;
1900 		ln->ln_expire = time_second + nd6_delay;
1901 	}
1902 
1903 	/*
1904 	 * If the neighbor cache entry has a state other than INCOMPLETE
1905 	 * (i.e. its link-layer address is already reloved), just
1906 	 * send the packet.
1907 	 */
1908 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1909 		goto sendpkt;
1910 
1911 	/*
1912 	 * There is a neighbor cache entry, but no ethernet address
1913 	 * response yet. Replace the held mbuf (if any) with this
1914 	 * latest one.
1915 	 *
1916 	 * XXX Does the code conform to rate-limiting rule?
1917 	 * (RFC 2461 7.2.2)
1918 	 */
1919 	if (ln->ln_state == ND6_LLINFO_WAITDELETE ||
1920 	    ln->ln_state == ND6_LLINFO_NOSTATE)
1921 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1922 	if (ln->ln_hold)
1923 		m_freem(ln->ln_hold);
1924 	ln->ln_hold = m;
1925 	if (ln->ln_expire) {
1926 		rt->rt_flags &= ~RTF_REJECT;
1927 		if (ln->ln_asked < nd6_mmaxtries &&
1928 		    ln->ln_expire < time_second) {
1929 			ln->ln_asked++;
1930 			ln->ln_expire = time_second +
1931 				nd_ifinfo[ifp->if_index].retrans / 1000;
1932 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1933 		}
1934 	}
1935 	return(0);
1936 
1937   sendpkt:
1938 
1939 #ifdef FAKE_LOOPBACK_IF
1940 	if (ifp->if_flags & IFF_LOOPBACK) {
1941 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1942 					 rt));
1943 	}
1944 #endif
1945 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1946 
1947   bad:
1948 	if (m)
1949 		m_freem(m);
1950 	return (error);
1951 }
1952 #undef senderr
1953 
1954 int
1955 nd6_storelladdr(ifp, rt, m, dst, desten)
1956 	struct ifnet *ifp;
1957 	struct rtentry *rt;
1958 	struct mbuf *m;
1959 	struct sockaddr *dst;
1960 	u_char *desten;
1961 {
1962 	struct sockaddr_dl *sdl;
1963 
1964 	if (m->m_flags & M_MCAST) {
1965 		switch (ifp->if_type) {
1966 		case IFT_ETHER:
1967 		case IFT_FDDI:
1968 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1969 						 desten);
1970 			return(1);
1971 			break;
1972 		case IFT_ARCNET:
1973 			*desten = 0;
1974 			return(1);
1975 		default:
1976 			return(0);
1977 		}
1978 	}
1979 
1980 	if (rt == NULL ||
1981 	    rt->rt_gateway->sa_family != AF_LINK) {
1982 		printf("nd6_storelladdr: something odd happens\n");
1983 		return(0);
1984 	}
1985 	sdl = SDL(rt->rt_gateway);
1986 	if (sdl->sdl_alen == 0) {
1987 		/* this should be impossible, but we bark here for debugging */
1988 		printf("nd6_storelladdr: sdl_alen == 0\n");
1989 		return(0);
1990 	}
1991 
1992 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
1993 	return(1);
1994 }
1995