xref: /netbsd-src/sys/netinet6/nd6.c (revision 21a3d2f02241c56556f4b2305ef1b8036f268f70)
1 /*	$NetBSD: nd6.c,v 1.54 2001/10/17 10:55:09 itojun Exp $	*/
2 /*	$KAME: nd6.c,v 1.151 2001/06/19 14:24:41 sumikawa 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/if_ieee1394.h>
61 #include <net/route.h>
62 
63 #include <netinet/in.h>
64 #include <net/if_ether.h>
65 #include <netinet/if_inarp.h>
66 #include <net/if_fddi.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet/ip6.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet6/nd6.h>
71 #include <netinet6/in6_prefix.h>
72 #include <netinet/icmp6.h>
73 
74 #include "loop.h"
75 extern struct ifnet loif[NLOOP];
76 
77 #include <net/net_osdep.h>
78 
79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
81 
82 #define SIN6(s) ((struct sockaddr_in6 *)s)
83 #define SDL(s) ((struct sockaddr_dl *)s)
84 
85 /* timer values */
86 int	nd6_prune	= 1;	/* walk list every 1 seconds */
87 int	nd6_delay	= 5;	/* delay first probe time 5 second */
88 int	nd6_umaxtries	= 3;	/* maximum unicast query */
89 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
90 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
91 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
92 
93 /* preventing too many loops in ND option parsing */
94 int nd6_maxndopt = 10;	/* max # of ND options allowed */
95 
96 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
97 
98 #ifdef ND6_DEBUG
99 int nd6_debug = 1;
100 #else
101 int nd6_debug = 0;
102 #endif
103 
104 /* for debugging? */
105 static int nd6_inuse, nd6_allocated;
106 
107 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
108 static size_t nd_ifinfo_indexlim = 8;
109 struct nd_ifinfo *nd_ifinfo = NULL;
110 struct nd_drhead nd_defrouter;
111 struct nd_prhead nd_prefix = { 0 };
112 
113 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
114 static struct sockaddr_in6 all1_sa;
115 
116 static void nd6_slowtimo __P((void *));
117 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
118 
119 struct callout nd6_slowtimo_ch;
120 struct callout nd6_timer_ch;
121 
122 void
123 nd6_init()
124 {
125 	static int nd6_init_done = 0;
126 	int i;
127 
128 	if (nd6_init_done) {
129 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
130 		return;
131 	}
132 
133 	all1_sa.sin6_family = AF_INET6;
134 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
135 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
136 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
137 
138 	/* initialization of the default router list */
139 	TAILQ_INIT(&nd_defrouter);
140 
141 	nd6_init_done = 1;
142 
143 	/* start timer */
144 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
145 	    nd6_slowtimo, NULL);
146 }
147 
148 void
149 nd6_ifattach(ifp)
150 	struct ifnet *ifp;
151 {
152 
153 	/*
154 	 * We have some arrays that should be indexed by if_index.
155 	 * since if_index will grow dynamically, they should grow too.
156 	 */
157 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
158 		size_t n;
159 		caddr_t q;
160 
161 		while (if_index >= nd_ifinfo_indexlim)
162 			nd_ifinfo_indexlim <<= 1;
163 
164 		/* grow nd_ifinfo */
165 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
166 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
167 		bzero(q, n);
168 		if (nd_ifinfo) {
169 			bcopy((caddr_t)nd_ifinfo, q, n/2);
170 			free((caddr_t)nd_ifinfo, M_IP6NDP);
171 		}
172 		nd_ifinfo = (struct nd_ifinfo *)q;
173 	}
174 
175 #define ND nd_ifinfo[ifp->if_index]
176 
177 	/*
178 	 * Don't initialize if called twice.
179 	 * XXX: to detect this, we should choose a member that is never set
180 	 * before initialization of the ND structure itself.  We formaly used
181 	 * the linkmtu member, which was not suitable because it could be
182 	 * initialized via "ifconfig mtu".
183 	 */
184 	if (ND.basereachable)
185 		return;
186 
187 #ifdef DIAGNOSTIC
188 	if (!ifindex2ifnet[ifp->if_index])
189 		panic("nd6_ifattach: ifindex2ifnet is NULL");
190 #endif
191 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
192 	ND.chlim = IPV6_DEFHLIM;
193 	ND.basereachable = REACHABLE_TIME;
194 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
195 	ND.retrans = RETRANS_TIMER;
196 	ND.receivedra = 0;
197 	ND.flags = ND6_IFF_PERFORMNUD;
198 	nd6_setmtu(ifp);
199 #undef ND
200 }
201 
202 /*
203  * Reset ND level link MTU. This function is called when the physical MTU
204  * changes, which means we might have to adjust the ND level MTU.
205  */
206 void
207 nd6_setmtu(ifp)
208 	struct ifnet *ifp;
209 {
210 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
211 	u_long oldmaxmtu = ndi->maxmtu;
212 	u_long oldlinkmtu = ndi->linkmtu;
213 
214 	switch (ifp->if_type) {
215 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
216 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
217 		break;
218 	case IFT_ETHER:
219 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
220 		break;
221 	case IFT_ATM:
222 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
223 		break;
224 	case IFT_IEEE1394:
225 		ndi->maxmtu = MIN(IEEE1394MTU, ifp->if_mtu);
226 		break;
227 	default:
228 		ndi->maxmtu = ifp->if_mtu;
229 		break;
230 	}
231 
232 	if (oldmaxmtu != ndi->maxmtu) {
233 		/*
234 		 * If the ND level MTU is not set yet, or if the maxmtu
235 		 * is reset to a smaller value than the ND level MTU,
236 		 * also reset the ND level MTU.
237 		 */
238 		if (ndi->linkmtu == 0 ||
239 		    ndi->maxmtu < ndi->linkmtu) {
240 			ndi->linkmtu = ndi->maxmtu;
241 			/* also adjust in6_maxmtu if necessary. */
242 			if (oldlinkmtu == 0) {
243 				/*
244 				 * XXX: the case analysis is grotty, but
245 				 * it is not efficient to call in6_setmaxmtu()
246 				 * here when we are during the initialization
247 				 * procedure.
248 				 */
249 				if (in6_maxmtu < ndi->linkmtu)
250 					in6_maxmtu = ndi->linkmtu;
251 			} else
252 				in6_setmaxmtu();
253 		}
254 	}
255 #undef MIN
256 }
257 
258 void
259 nd6_option_init(opt, icmp6len, ndopts)
260 	void *opt;
261 	int icmp6len;
262 	union nd_opts *ndopts;
263 {
264 	bzero(ndopts, sizeof(*ndopts));
265 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
266 	ndopts->nd_opts_last
267 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
268 
269 	if (icmp6len == 0) {
270 		ndopts->nd_opts_done = 1;
271 		ndopts->nd_opts_search = NULL;
272 	}
273 }
274 
275 /*
276  * Take one ND option.
277  */
278 struct nd_opt_hdr *
279 nd6_option(ndopts)
280 	union nd_opts *ndopts;
281 {
282 	struct nd_opt_hdr *nd_opt;
283 	int olen;
284 
285 	if (!ndopts)
286 		panic("ndopts == NULL in nd6_option\n");
287 	if (!ndopts->nd_opts_last)
288 		panic("uninitialized ndopts in nd6_option\n");
289 	if (!ndopts->nd_opts_search)
290 		return NULL;
291 	if (ndopts->nd_opts_done)
292 		return NULL;
293 
294 	nd_opt = ndopts->nd_opts_search;
295 
296 	/* make sure nd_opt_len is inside the buffer */
297 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
298 		bzero(ndopts, sizeof(*ndopts));
299 		return NULL;
300 	}
301 
302 	olen = nd_opt->nd_opt_len << 3;
303 	if (olen == 0) {
304 		/*
305 		 * Message validation requires that all included
306 		 * options have a length that is greater than zero.
307 		 */
308 		bzero(ndopts, sizeof(*ndopts));
309 		return NULL;
310 	}
311 
312 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
313 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
314 		/* option overruns the end of buffer, invalid */
315 		bzero(ndopts, sizeof(*ndopts));
316 		return NULL;
317 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
318 		/* reached the end of options chain */
319 		ndopts->nd_opts_done = 1;
320 		ndopts->nd_opts_search = NULL;
321 	}
322 	return nd_opt;
323 }
324 
325 /*
326  * Parse multiple ND options.
327  * This function is much easier to use, for ND routines that do not need
328  * multiple options of the same type.
329  */
330 int
331 nd6_options(ndopts)
332 	union nd_opts *ndopts;
333 {
334 	struct nd_opt_hdr *nd_opt;
335 	int i = 0;
336 
337 	if (!ndopts)
338 		panic("ndopts == NULL in nd6_options\n");
339 	if (!ndopts->nd_opts_last)
340 		panic("uninitialized ndopts in nd6_options\n");
341 	if (!ndopts->nd_opts_search)
342 		return 0;
343 
344 	while (1) {
345 		nd_opt = nd6_option(ndopts);
346 		if (!nd_opt && !ndopts->nd_opts_last) {
347 			/*
348 			 * Message validation requires that all included
349 			 * options have a length that is greater than zero.
350 			 */
351 			icmp6stat.icp6s_nd_badopt++;
352 			bzero(ndopts, sizeof(*ndopts));
353 			return -1;
354 		}
355 
356 		if (!nd_opt)
357 			goto skip1;
358 
359 		switch (nd_opt->nd_opt_type) {
360 		case ND_OPT_SOURCE_LINKADDR:
361 		case ND_OPT_TARGET_LINKADDR:
362 		case ND_OPT_MTU:
363 		case ND_OPT_REDIRECTED_HEADER:
364 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
365 				nd6log((LOG_INFO,
366 				    "duplicated ND6 option found (type=%d)\n",
367 				    nd_opt->nd_opt_type));
368 				/* XXX bark? */
369 			} else {
370 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
371 					= nd_opt;
372 			}
373 			break;
374 		case ND_OPT_PREFIX_INFORMATION:
375 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
376 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
377 					= nd_opt;
378 			}
379 			ndopts->nd_opts_pi_end =
380 				(struct nd_opt_prefix_info *)nd_opt;
381 			break;
382 		default:
383 			/*
384 			 * Unknown options must be silently ignored,
385 			 * to accomodate future extension to the protocol.
386 			 */
387 			nd6log((LOG_DEBUG,
388 			    "nd6_options: unsupported option %d - "
389 			    "option ignored\n", nd_opt->nd_opt_type));
390 		}
391 
392 skip1:
393 		i++;
394 		if (i > nd6_maxndopt) {
395 			icmp6stat.icp6s_nd_toomanyopt++;
396 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
397 			break;
398 		}
399 
400 		if (ndopts->nd_opts_done)
401 			break;
402 	}
403 
404 	return 0;
405 }
406 
407 /*
408  * ND6 timer routine to expire default route list and prefix list
409  */
410 void
411 nd6_timer(ignored_arg)
412 	void	*ignored_arg;
413 {
414 	int s;
415 	struct llinfo_nd6 *ln;
416 	struct nd_defrouter *dr;
417 	struct nd_prefix *pr;
418 	long time_second = time.tv_sec;
419 
420 	s = splsoftnet();
421 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
422 	    nd6_timer, NULL);
423 
424 	ln = llinfo_nd6.ln_next;
425 	/* XXX BSD/OS separates this code -- itojun */
426 	while (ln && ln != &llinfo_nd6) {
427 		struct rtentry *rt;
428 		struct ifnet *ifp;
429 		struct sockaddr_in6 *dst;
430 		struct llinfo_nd6 *next = ln->ln_next;
431 		/* XXX: used for the DELAY case only: */
432 		struct nd_ifinfo *ndi = NULL;
433 
434 		if ((rt = ln->ln_rt) == NULL) {
435 			ln = next;
436 			continue;
437 		}
438 		if ((ifp = rt->rt_ifp) == NULL) {
439 			ln = next;
440 			continue;
441 		}
442 		ndi = &nd_ifinfo[ifp->if_index];
443 		dst = (struct sockaddr_in6 *)rt_key(rt);
444 
445 		if (ln->ln_expire > time_second) {
446 			ln = next;
447 			continue;
448 		}
449 
450 		/* sanity check */
451 		if (!rt)
452 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
453 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
454 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
455 			      rt->rt_llinfo, ln);
456 		if (!dst)
457 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
458 
459 		switch (ln->ln_state) {
460 		case ND6_LLINFO_INCOMPLETE:
461 			if (ln->ln_asked < nd6_mmaxtries) {
462 				ln->ln_asked++;
463 				ln->ln_expire = time_second +
464 					nd_ifinfo[ifp->if_index].retrans / 1000;
465 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
466 					ln, 0);
467 			} else {
468 				struct mbuf *m = ln->ln_hold;
469 				if (m) {
470 					if (rt->rt_ifp) {
471 						/*
472 						 * Fake rcvif to make ICMP error
473 						 * more helpful in diagnosing
474 						 * for the receiver.
475 						 * XXX: should we consider
476 						 * older rcvif?
477 						 */
478 						m->m_pkthdr.rcvif = rt->rt_ifp;
479 					}
480 					icmp6_error(m, ICMP6_DST_UNREACH,
481 						    ICMP6_DST_UNREACH_ADDR, 0);
482 					ln->ln_hold = NULL;
483 				}
484 				next = nd6_free(rt, 0);
485 			}
486 			break;
487 		case ND6_LLINFO_REACHABLE:
488 			if (ln->ln_expire) {
489 				ln->ln_state = ND6_LLINFO_STALE;
490 				ln->ln_expire = time_second + nd6_gctimer;
491 			}
492 			break;
493 
494 		case ND6_LLINFO_STALE:
495 			/* Garbage Collection(RFC 2461 5.3) */
496 			if (ln->ln_expire)
497 				next = nd6_free(rt, 1);
498 			break;
499 
500 		case ND6_LLINFO_DELAY:
501 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
502 				/* We need NUD */
503 				ln->ln_asked = 1;
504 				ln->ln_state = ND6_LLINFO_PROBE;
505 				ln->ln_expire = time_second +
506 					ndi->retrans / 1000;
507 				nd6_ns_output(ifp, &dst->sin6_addr,
508 					      &dst->sin6_addr,
509 					      ln, 0);
510 			} else {
511 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
512 				ln->ln_expire = time_second + nd6_gctimer;
513 			}
514 			break;
515 		case ND6_LLINFO_PROBE:
516 			if (ln->ln_asked < nd6_umaxtries) {
517 				ln->ln_asked++;
518 				ln->ln_expire = time_second +
519 					nd_ifinfo[ifp->if_index].retrans / 1000;
520 				nd6_ns_output(ifp, &dst->sin6_addr,
521 					       &dst->sin6_addr, ln, 0);
522 			} else {
523 				next = nd6_free(rt, 0);
524 			}
525 			break;
526 		}
527 		ln = next;
528 	}
529 
530 	/* expire default router list */
531 	dr = TAILQ_FIRST(&nd_defrouter);
532 	while (dr) {
533 		if (dr->expire && dr->expire < time_second) {
534 			struct nd_defrouter *t;
535 			t = TAILQ_NEXT(dr, dr_entry);
536 			defrtrlist_del(dr);
537 			dr = t;
538 		} else {
539 			dr = TAILQ_NEXT(dr, dr_entry);
540 		}
541 	}
542 	pr = nd_prefix.lh_first;
543 	while (pr) {
544 		struct in6_ifaddr *ia6;
545 		struct in6_addrlifetime *lt6;
546 
547 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
548 			ia6 = NULL;
549 		else
550 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
551 
552 		if (ia6) {
553 			/* check address lifetime */
554 			lt6 = &ia6->ia6_lifetime;
555 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
556 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
557 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
558 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
559 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
560 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
561 			}
562 		}
563 
564 		/*
565 		 * check prefix lifetime.
566 		 * since pltime is just for autoconf, pltime processing for
567 		 * prefix is not necessary.
568 		 *
569 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
570 		 * can use the old prefix information to validate the
571 		 * next prefix information to come.  See prelist_update()
572 		 * for actual validation.
573 		 */
574 		if (pr->ndpr_expire
575 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
576 			struct nd_prefix *t;
577 			t = pr->ndpr_next;
578 
579 			/*
580 			 * address expiration and prefix expiration are
581 			 * separate.  NEVER perform in6_ifdel here.
582 			 */
583 
584 			prelist_remove(pr);
585 			pr = t;
586 		} else
587 			pr = pr->ndpr_next;
588 	}
589 	splx(s);
590 }
591 
592 /*
593  * Nuke neighbor cache/prefix/default router management table, right before
594  * ifp goes away.
595  */
596 void
597 nd6_purge(ifp)
598 	struct ifnet *ifp;
599 {
600 	struct llinfo_nd6 *ln, *nln;
601 	struct nd_defrouter *dr, *ndr, drany;
602 	struct nd_prefix *pr, *npr;
603 
604 	/* Nuke default router list entries toward ifp */
605 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
606 		/*
607 		 * The first entry of the list may be stored in
608 		 * the routing table, so we'll delete it later.
609 		 */
610 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
611 			ndr = TAILQ_NEXT(dr, dr_entry);
612 			if (dr->ifp == ifp)
613 				defrtrlist_del(dr);
614 		}
615 		dr = TAILQ_FIRST(&nd_defrouter);
616 		if (dr->ifp == ifp)
617 			defrtrlist_del(dr);
618 	}
619 
620 	/* Nuke prefix list entries toward ifp */
621 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
622 		npr = pr->ndpr_next;
623 		if (pr->ndpr_ifp == ifp) {
624 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
625 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
626 			prelist_remove(pr);
627 		}
628 	}
629 
630 	/* cancel default outgoing interface setting */
631 	if (nd6_defifindex == ifp->if_index)
632 		nd6_setdefaultiface(0);
633 
634 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
635 		/* refresh default router list */
636 		bzero(&drany, sizeof(drany));
637 		defrouter_delreq(&drany, 0);
638 		defrouter_select();
639 	}
640 
641 	/*
642 	 * Nuke neighbor cache entries for the ifp.
643 	 * Note that rt->rt_ifp may not be the same as ifp,
644 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
645 	 * nd6_rtrequest(), and ip6_input().
646 	 */
647 	ln = llinfo_nd6.ln_next;
648 	while (ln && ln != &llinfo_nd6) {
649 		struct rtentry *rt;
650 		struct sockaddr_dl *sdl;
651 
652 		nln = ln->ln_next;
653 		rt = ln->ln_rt;
654 		if (rt && rt->rt_gateway &&
655 		    rt->rt_gateway->sa_family == AF_LINK) {
656 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
657 			if (sdl->sdl_index == ifp->if_index)
658 				nln = nd6_free(rt, 0);
659 		}
660 		ln = nln;
661 	}
662 }
663 
664 struct rtentry *
665 nd6_lookup(addr6, create, ifp)
666 	struct in6_addr *addr6;
667 	int create;
668 	struct ifnet *ifp;
669 {
670 	struct rtentry *rt;
671 	struct sockaddr_in6 sin6;
672 
673 	bzero(&sin6, sizeof(sin6));
674 	sin6.sin6_len = sizeof(struct sockaddr_in6);
675 	sin6.sin6_family = AF_INET6;
676 	sin6.sin6_addr = *addr6;
677 	rt = rtalloc1((struct sockaddr *)&sin6, create);
678 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
679 		/*
680 		 * This is the case for the default route.
681 		 * If we want to create a neighbor cache for the address, we
682 		 * should free the route for the destination and allocate an
683 		 * interface route.
684 		 */
685 		if (create) {
686 			RTFREE(rt);
687 			rt = 0;
688 		}
689 	}
690 	if (!rt) {
691 		if (create && ifp) {
692 			int e;
693 
694 			/*
695 			 * If no route is available and create is set,
696 			 * we allocate a host route for the destination
697 			 * and treat it like an interface route.
698 			 * This hack is necessary for a neighbor which can't
699 			 * be covered by our own prefix.
700 			 */
701 			struct ifaddr *ifa =
702 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
703 			if (ifa == NULL)
704 				return(NULL);
705 
706 			/*
707 			 * Create a new route.  RTF_LLINFO is necessary
708 			 * to create a Neighbor Cache entry for the
709 			 * destination in nd6_rtrequest which will be
710 			 * called in rtrequest via ifa->ifa_rtrequest.
711 			 */
712 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
713 					   ifa->ifa_addr,
714 					   (struct sockaddr *)&all1_sa,
715 					   (ifa->ifa_flags |
716 					    RTF_HOST | RTF_LLINFO) &
717 					   ~RTF_CLONING,
718 					   &rt)) != 0) {
719 #if 0
720 				log(LOG_ERR,
721 				    "nd6_lookup: failed to add route for a "
722 				    "neighbor(%s), errno=%d\n",
723 				    ip6_sprintf(addr6), e);
724 #endif
725 				return(NULL);
726 			}
727 			if (rt == NULL)
728 				return(NULL);
729 			if (rt->rt_llinfo) {
730 				struct llinfo_nd6 *ln =
731 					(struct llinfo_nd6 *)rt->rt_llinfo;
732 				ln->ln_state = ND6_LLINFO_NOSTATE;
733 			}
734 		} else
735 			return(NULL);
736 	}
737 	rt->rt_refcnt--;
738 	/*
739 	 * Validation for the entry.
740 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
741 	 *      it might be the loopback interface if the entry is for our
742 	 *      own address on a non-loopback interface. Instead, we should
743 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
744 	 */
745 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
746 	    rt->rt_gateway->sa_family != AF_LINK ||
747 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
748 		if (create) {
749 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
750 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
751 		}
752 		return(0);
753 	}
754 	return(rt);
755 }
756 
757 /*
758  * Detect if a given IPv6 address identifies a neighbor on a given link.
759  * XXX: should take care of the destination of a p2p link?
760  */
761 int
762 nd6_is_addr_neighbor(addr, ifp)
763 	struct sockaddr_in6 *addr;
764 	struct ifnet *ifp;
765 {
766 	struct ifaddr *ifa;
767 	int i;
768 
769 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
770 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
771 
772 	/*
773 	 * A link-local address is always a neighbor.
774 	 * XXX: we should use the sin6_scope_id field rather than the embedded
775 	 * interface index.
776 	 * XXX: a link does not necessarily specify a single interface.
777 	 */
778 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
779 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
780 		return(1);
781 
782 	/*
783 	 * If the address matches one of our addresses,
784 	 * it should be a neighbor.
785 	 */
786 	for (ifa = ifp->if_addrlist.tqh_first;
787 	     ifa;
788 	     ifa = ifa->ifa_list.tqe_next)
789 	{
790 		if (ifa->ifa_addr->sa_family != AF_INET6)
791 			next: continue;
792 
793 		for (i = 0; i < 4; i++) {
794 			if ((IFADDR6(ifa).s6_addr32[i] ^
795 			     addr->sin6_addr.s6_addr32[i]) &
796 			    IFMASK6(ifa).s6_addr32[i])
797 				goto next;
798 		}
799 		return(1);
800 	}
801 
802 	/*
803 	 * Even if the address matches none of our addresses, it might be
804 	 * in the neighbor cache.
805 	 */
806 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
807 		return(1);
808 
809 	return(0);
810 #undef IFADDR6
811 #undef IFMASK6
812 }
813 
814 /*
815  * Free an nd6 llinfo entry.
816  * Since the function would cause significant changes in the kernel, DO NOT
817  * make it global, unless you have a strong reason for the change, and are sure
818  * that the change is safe.
819  */
820 static struct llinfo_nd6 *
821 nd6_free(rt, gc)
822 	struct rtentry *rt;
823 	int gc;
824 {
825 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
826 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
827 	struct nd_defrouter *dr;
828 
829 	/*
830 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
831 	 * even though it is not harmful, it was not really necessary.
832 	 */
833 
834 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
835 		int s;
836 		s = splsoftnet();
837 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
838 				      rt->rt_ifp);
839 
840 		if (dr != NULL && dr->expire &&
841 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
842 			/*
843 			 * If the reason for the deletion is just garbage
844 			 * collection, and the neighbor is an active default
845 			 * router, do not delete it.  Instead, reset the GC
846 			 * timer using the router's lifetime.
847 			 * Simply deleting the entry would affect default
848 			 * router selection, which is not necessarily a good
849 			 * thing, especially when we're using router preference
850 			 * values.
851 			 * XXX: the check for ln_state would be redundant,
852 			 *      but we intentionally keep it just in case.
853 			 */
854 			ln->ln_expire = dr->expire;
855 			splx(s);
856 			return(ln->ln_next);
857 		}
858 
859 		if (ln->ln_router || dr) {
860 			/*
861 			 * rt6_flush must be called whether or not the neighbor
862 			 * is in the Default Router List.
863 			 * See a corresponding comment in nd6_na_input().
864 			 */
865 			rt6_flush(&in6, rt->rt_ifp);
866 		}
867 
868 		if (dr) {
869 			/*
870 			 * Unreachablity of a router might affect the default
871 			 * router selection and on-link detection of advertised
872 			 * prefixes.
873 			 */
874 
875 			/*
876 			 * Temporarily fake the state to choose a new default
877 			 * router and to perform on-link determination of
878 			 * prefixes correctly.
879 			 * Below the state will be set correctly,
880 			 * or the entry itself will be deleted.
881 			 */
882 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
883 
884 			/*
885 			 * Since defrouter_select() does not affect the
886 			 * on-link determination and MIP6 needs the check
887 			 * before the default router selection, we perform
888 			 * the check now.
889 			 */
890 			pfxlist_onlink_check();
891 
892 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
893 				/*
894 				 * It is used as the current default router,
895 				 * so we have to move it to the end of the
896 				 * list and choose a new one.
897 				 * XXX: it is not very efficient if this is
898 				 *      the only router.
899 				 */
900 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
901 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
902 
903 				defrouter_select();
904 			}
905 		}
906 		splx(s);
907 	}
908 
909 	/*
910 	 * Before deleting the entry, remember the next entry as the
911 	 * return value.  We need this because pfxlist_onlink_check() above
912 	 * might have freed other entries (particularly the old next entry) as
913 	 * a side effect (XXX).
914 	 */
915 	next = ln->ln_next;
916 
917 	/*
918 	 * Detach the route from the routing tree and the list of neighbor
919 	 * caches, and disable the route entry not to be used in already
920 	 * cached routes.
921 	 */
922 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
923 		  rt_mask(rt), 0, (struct rtentry **)0);
924 
925 	return(next);
926 }
927 
928 /*
929  * Upper-layer reachability hint for Neighbor Unreachability Detection.
930  *
931  * XXX cost-effective metods?
932  */
933 void
934 nd6_nud_hint(rt, dst6, force)
935 	struct rtentry *rt;
936 	struct in6_addr *dst6;
937 	int force;
938 {
939 	struct llinfo_nd6 *ln;
940 	long time_second = time.tv_sec;
941 
942 	/*
943 	 * If the caller specified "rt", use that.  Otherwise, resolve the
944 	 * routing table by supplied "dst6".
945 	 */
946 	if (!rt) {
947 		if (!dst6)
948 			return;
949 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
950 			return;
951 	}
952 
953 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
954 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
955 	    !rt->rt_llinfo || !rt->rt_gateway ||
956 	    rt->rt_gateway->sa_family != AF_LINK) {
957 		/* This is not a host route. */
958 		return;
959 	}
960 
961 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
962 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
963 		return;
964 
965 	/*
966 	 * if we get upper-layer reachability confirmation many times,
967 	 * it is possible we have false information.
968 	 */
969 	if (!force) {
970 		ln->ln_byhint++;
971 		if (ln->ln_byhint > nd6_maxnudhint)
972 			return;
973 	}
974 
975 	ln->ln_state = ND6_LLINFO_REACHABLE;
976 	if (ln->ln_expire)
977 		ln->ln_expire = time_second +
978 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
979 }
980 
981 void
982 nd6_rtrequest(req, rt, info)
983 	int	req;
984 	struct rtentry *rt;
985 	struct rt_addrinfo *info; /* xxx unused */
986 {
987 	struct sockaddr *gate = rt->rt_gateway;
988 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
989 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
990 	struct ifnet *ifp = rt->rt_ifp;
991 	struct ifaddr *ifa;
992 	long time_second = time.tv_sec;
993 
994 	if (rt->rt_flags & RTF_GATEWAY)
995 		return;
996 
997 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
998 		/*
999 		 * This is probably an interface direct route for a link
1000 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1001 		 * We do not need special treatment below for such a route.
1002 		 * Moreover, the RTF_LLINFO flag which would be set below
1003 		 * would annoy the ndp(8) command.
1004 		 */
1005 		return;
1006 	}
1007 
1008 	switch (req) {
1009 	case RTM_ADD:
1010 		/*
1011 		 * There is no backward compatibility :)
1012 		 *
1013 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1014 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1015 		 *	   rt->rt_flags |= RTF_CLONING;
1016 		 */
1017 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1018 			/*
1019 			 * Case 1: This route should come from
1020 			 * a route to interface.  RTF_LLINFO flag is set
1021 			 * for a host route whose destination should be
1022 			 * treated as on-link.
1023 			 */
1024 			rt_setgate(rt, rt_key(rt),
1025 				   (struct sockaddr *)&null_sdl);
1026 			gate = rt->rt_gateway;
1027 			SDL(gate)->sdl_type = ifp->if_type;
1028 			SDL(gate)->sdl_index = ifp->if_index;
1029 			if (ln)
1030 				ln->ln_expire = time_second;
1031 #if 1
1032 			if (ln && ln->ln_expire == 0) {
1033 				/* kludge for desktops */
1034 #if 0
1035 				printf("nd6_rtequest: time.tv_sec is zero; "
1036 				       "treat it as 1\n");
1037 #endif
1038 				ln->ln_expire = 1;
1039 			}
1040 #endif
1041 			if (rt->rt_flags & RTF_CLONING)
1042 				break;
1043 		}
1044 		/*
1045 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1046 		 * We don't do that here since llinfo is not ready yet.
1047 		 *
1048 		 * There are also couple of other things to be discussed:
1049 		 * - unsolicited NA code needs improvement beforehand
1050 		 * - RFC2461 says we MAY send multicast unsolicited NA
1051 		 *   (7.2.6 paragraph 4), however, it also says that we
1052 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1053 		 *   we don't have anything like it right now.
1054 		 *   note that the mechanism needs a mutual agreement
1055 		 *   between proxies, which means that we need to implement
1056 		 *   a new protocol, or a new kludge.
1057 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1058 		 *   we need to check ip6forwarding before sending it.
1059 		 *   (or should we allow proxy ND configuration only for
1060 		 *   routers?  there's no mention about proxy ND from hosts)
1061 		 */
1062 #if 0
1063 		/* XXX it does not work */
1064 		if (rt->rt_flags & RTF_ANNOUNCE)
1065 			nd6_na_output(ifp,
1066 			      &SIN6(rt_key(rt))->sin6_addr,
1067 			      &SIN6(rt_key(rt))->sin6_addr,
1068 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1069 			      1, NULL);
1070 #endif
1071 		/* FALLTHROUGH */
1072 	case RTM_RESOLVE:
1073 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
1074 			/*
1075 			 * Address resolution isn't necessary for a point to
1076 			 * point link, so we can skip this test for a p2p link.
1077 			 */
1078 			if (gate->sa_family != AF_LINK ||
1079 			    gate->sa_len < sizeof(null_sdl)) {
1080 				log(LOG_DEBUG,
1081 				    "nd6_rtrequest: bad gateway value: %s\n",
1082 				    if_name(ifp));
1083 				break;
1084 			}
1085 			SDL(gate)->sdl_type = ifp->if_type;
1086 			SDL(gate)->sdl_index = ifp->if_index;
1087 		}
1088 		if (ln != NULL)
1089 			break;	/* This happens on a route change */
1090 		/*
1091 		 * Case 2: This route may come from cloning, or a manual route
1092 		 * add with a LL address.
1093 		 */
1094 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1095 		rt->rt_llinfo = (caddr_t)ln;
1096 		if (!ln) {
1097 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1098 			break;
1099 		}
1100 		nd6_inuse++;
1101 		nd6_allocated++;
1102 		Bzero(ln, sizeof(*ln));
1103 		ln->ln_rt = rt;
1104 		/* this is required for "ndp" command. - shin */
1105 		if (req == RTM_ADD) {
1106 		        /*
1107 			 * gate should have some valid AF_LINK entry,
1108 			 * and ln->ln_expire should have some lifetime
1109 			 * which is specified by ndp command.
1110 			 */
1111 			ln->ln_state = ND6_LLINFO_REACHABLE;
1112 			ln->ln_byhint = 0;
1113 		} else {
1114 		        /*
1115 			 * When req == RTM_RESOLVE, rt is created and
1116 			 * initialized in rtrequest(), so rt_expire is 0.
1117 			 */
1118 			ln->ln_state = ND6_LLINFO_NOSTATE;
1119 			ln->ln_expire = time_second;
1120 		}
1121 		rt->rt_flags |= RTF_LLINFO;
1122 		ln->ln_next = llinfo_nd6.ln_next;
1123 		llinfo_nd6.ln_next = ln;
1124 		ln->ln_prev = &llinfo_nd6;
1125 		ln->ln_next->ln_prev = ln;
1126 
1127 		/*
1128 		 * check if rt_key(rt) is one of my address assigned
1129 		 * to the interface.
1130 		 */
1131 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1132 					  &SIN6(rt_key(rt))->sin6_addr);
1133 		if (ifa) {
1134 			caddr_t macp = nd6_ifptomac(ifp);
1135 			ln->ln_expire = 0;
1136 			ln->ln_state = ND6_LLINFO_REACHABLE;
1137 			ln->ln_byhint = 0;
1138 			if (macp) {
1139 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1140 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1141 			}
1142 			if (nd6_useloopback) {
1143 				rt->rt_ifp = &loif[0];	/* XXX */
1144 				/*
1145 				 * Make sure rt_ifa be equal to the ifaddr
1146 				 * corresponding to the address.
1147 				 * We need this because when we refer
1148 				 * rt_ifa->ia6_flags in ip6_input, we assume
1149 				 * that the rt_ifa points to the address instead
1150 				 * of the loopback address.
1151 				 */
1152 				if (ifa != rt->rt_ifa) {
1153 					IFAFREE(rt->rt_ifa);
1154 					IFAREF(ifa);
1155 					rt->rt_ifa = ifa;
1156 				}
1157 			}
1158 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1159 			ln->ln_expire = 0;
1160 			ln->ln_state = ND6_LLINFO_REACHABLE;
1161 			ln->ln_byhint = 0;
1162 
1163 			/* join solicited node multicast for proxy ND */
1164 			if (ifp->if_flags & IFF_MULTICAST) {
1165 				struct in6_addr llsol;
1166 				int error;
1167 
1168 				llsol = SIN6(rt_key(rt))->sin6_addr;
1169 				llsol.s6_addr16[0] = htons(0xff02);
1170 				llsol.s6_addr16[1] = htons(ifp->if_index);
1171 				llsol.s6_addr32[1] = 0;
1172 				llsol.s6_addr32[2] = htonl(1);
1173 				llsol.s6_addr8[12] = 0xff;
1174 
1175 				if (!in6_addmulti(&llsol, ifp, &error)) {
1176 					nd6log((LOG_ERR, "%s: failed to join "
1177 					    "%s (errno=%d)\n", if_name(ifp),
1178 					    ip6_sprintf(&llsol), error));
1179 				}
1180 			}
1181 		}
1182 		break;
1183 
1184 	case RTM_DELETE:
1185 		if (!ln)
1186 			break;
1187 		/* leave from solicited node multicast for proxy ND */
1188 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1189 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1190 			struct in6_addr llsol;
1191 			struct in6_multi *in6m;
1192 
1193 			llsol = SIN6(rt_key(rt))->sin6_addr;
1194 			llsol.s6_addr16[0] = htons(0xff02);
1195 			llsol.s6_addr16[1] = htons(ifp->if_index);
1196 			llsol.s6_addr32[1] = 0;
1197 			llsol.s6_addr32[2] = htonl(1);
1198 			llsol.s6_addr8[12] = 0xff;
1199 
1200 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1201 			if (in6m)
1202 				in6_delmulti(in6m);
1203 		}
1204 		nd6_inuse--;
1205 		ln->ln_next->ln_prev = ln->ln_prev;
1206 		ln->ln_prev->ln_next = ln->ln_next;
1207 		ln->ln_prev = NULL;
1208 		rt->rt_llinfo = 0;
1209 		rt->rt_flags &= ~RTF_LLINFO;
1210 		if (ln->ln_hold)
1211 			m_freem(ln->ln_hold);
1212 		Free((caddr_t)ln);
1213 	}
1214 }
1215 
1216 void
1217 nd6_p2p_rtrequest(req, rt, info)
1218 	int	req;
1219 	struct rtentry *rt;
1220 	struct rt_addrinfo *info; /* xxx unused */
1221 {
1222 	struct sockaddr *gate = rt->rt_gateway;
1223 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1224 	struct ifnet *ifp = rt->rt_ifp;
1225 	struct ifaddr *ifa;
1226 
1227 	if (rt->rt_flags & RTF_GATEWAY)
1228 		return;
1229 
1230 	switch (req) {
1231 	case RTM_ADD:
1232 		/*
1233 		 * There is no backward compatibility :)
1234 		 *
1235 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1236 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1237 		 *	   rt->rt_flags |= RTF_CLONING;
1238 		 */
1239 		if (rt->rt_flags & RTF_CLONING) {
1240 			/*
1241 			 * Case 1: This route should come from
1242 			 * a route to interface.
1243 			 */
1244 			rt_setgate(rt, rt_key(rt),
1245 				   (struct sockaddr *)&null_sdl);
1246 			gate = rt->rt_gateway;
1247 			SDL(gate)->sdl_type = ifp->if_type;
1248 			SDL(gate)->sdl_index = ifp->if_index;
1249 			break;
1250 		}
1251 		/* Announce a new entry if requested. */
1252 		if (rt->rt_flags & RTF_ANNOUNCE)
1253 			nd6_na_output(ifp,
1254 				      &SIN6(rt_key(rt))->sin6_addr,
1255 				      &SIN6(rt_key(rt))->sin6_addr,
1256 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1257 				      1, NULL);
1258 		/* FALLTHROUGH */
1259 	case RTM_RESOLVE:
1260 		/*
1261 		 * check if rt_key(rt) is one of my address assigned
1262 		 * to the interface.
1263 		 */
1264  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1265 					  &SIN6(rt_key(rt))->sin6_addr);
1266 		if (ifa) {
1267 			if (nd6_useloopback) {
1268 				rt->rt_ifp = &loif[0];	/*XXX*/
1269 			}
1270 		}
1271 		break;
1272 	}
1273 }
1274 
1275 int
1276 nd6_ioctl(cmd, data, ifp)
1277 	u_long cmd;
1278 	caddr_t	data;
1279 	struct ifnet *ifp;
1280 {
1281 	struct in6_drlist *drl = (struct in6_drlist *)data;
1282 	struct in6_prlist *prl = (struct in6_prlist *)data;
1283 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1284 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1285 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1286 	struct nd_defrouter *dr, any;
1287 	struct nd_prefix *pr;
1288 	struct rtentry *rt;
1289 	int i = 0, error = 0;
1290 	int s;
1291 
1292 	switch (cmd) {
1293 	case SIOCGDRLST_IN6:
1294 		bzero(drl, sizeof(*drl));
1295 		s = splsoftnet();
1296 		dr = TAILQ_FIRST(&nd_defrouter);
1297 		while (dr && i < DRLSTSIZ) {
1298 			drl->defrouter[i].rtaddr = dr->rtaddr;
1299 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1300 				/* XXX: need to this hack for KAME stack */
1301 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1302 			} else
1303 				log(LOG_ERR,
1304 				    "default router list contains a "
1305 				    "non-linklocal address(%s)\n",
1306 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1307 
1308 			drl->defrouter[i].flags = dr->flags;
1309 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1310 			drl->defrouter[i].expire = dr->expire;
1311 			drl->defrouter[i].if_index = dr->ifp->if_index;
1312 			i++;
1313 			dr = TAILQ_NEXT(dr, dr_entry);
1314 		}
1315 		splx(s);
1316 		break;
1317 	case SIOCGPRLST_IN6:
1318 		/*
1319 		 * XXX meaning of fields, especialy "raflags", is very
1320 		 * differnet between RA prefix list and RR/static prefix list.
1321 		 * how about separating ioctls into two?
1322 		 */
1323 		bzero(prl, sizeof(*prl));
1324 		s = splsoftnet();
1325 		pr = nd_prefix.lh_first;
1326 		while (pr && i < PRLSTSIZ) {
1327 			struct nd_pfxrouter *pfr;
1328 			int j;
1329 
1330 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1331 			prl->prefix[i].raflags = pr->ndpr_raf;
1332 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1333 			prl->prefix[i].vltime = pr->ndpr_vltime;
1334 			prl->prefix[i].pltime = pr->ndpr_pltime;
1335 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1336 			prl->prefix[i].expire = pr->ndpr_expire;
1337 
1338 			pfr = pr->ndpr_advrtrs.lh_first;
1339 			j = 0;
1340 			while (pfr) {
1341 				if (j < DRLSTSIZ) {
1342 #define RTRADDR prl->prefix[i].advrtr[j]
1343 					RTRADDR = pfr->router->rtaddr;
1344 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1345 						/* XXX: hack for KAME */
1346 						RTRADDR.s6_addr16[1] = 0;
1347 					} else
1348 						log(LOG_ERR,
1349 						    "a router(%s) advertises "
1350 						    "a prefix with "
1351 						    "non-link local address\n",
1352 						    ip6_sprintf(&RTRADDR));
1353 #undef RTRADDR
1354 				}
1355 				j++;
1356 				pfr = pfr->pfr_next;
1357 			}
1358 			prl->prefix[i].advrtrs = j;
1359 			prl->prefix[i].origin = PR_ORIG_RA;
1360 
1361 			i++;
1362 			pr = pr->ndpr_next;
1363 		}
1364 	      {
1365 		struct rr_prefix *rpp;
1366 
1367 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1368 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1369 			if (i >= PRLSTSIZ)
1370 				break;
1371 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
1372 			prl->prefix[i].raflags = rpp->rp_raf;
1373 			prl->prefix[i].prefixlen = rpp->rp_plen;
1374 			prl->prefix[i].vltime = rpp->rp_vltime;
1375 			prl->prefix[i].pltime = rpp->rp_pltime;
1376 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1377 			prl->prefix[i].expire = rpp->rp_expire;
1378 			prl->prefix[i].advrtrs = 0;
1379 			prl->prefix[i].origin = rpp->rp_origin;
1380 			i++;
1381 		}
1382 	      }
1383 		splx(s);
1384 
1385 		break;
1386 	case SIOCGIFINFO_IN6:
1387 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1388 			error = EINVAL;
1389 			break;
1390 		}
1391 		ndi->ndi = nd_ifinfo[ifp->if_index];
1392 		break;
1393 	case SIOCSIFINFO_FLAGS:
1394 		/* XXX: almost all other fields of ndi->ndi is unused */
1395 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1396 			error = EINVAL;
1397 			break;
1398 		}
1399 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1400 		break;
1401 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1402 		/* flush default router list */
1403 		/*
1404 		 * xxx sumikawa: should not delete route if default
1405 		 * route equals to the top of default router list
1406 		 */
1407 		bzero(&any, sizeof(any));
1408 		defrouter_delreq(&any, 0);
1409 		defrouter_select();
1410 		/* xxx sumikawa: flush prefix list */
1411 		break;
1412 	case SIOCSPFXFLUSH_IN6:
1413 	    {
1414 		/* flush all the prefix advertised by routers */
1415 		struct nd_prefix *pr, *next;
1416 
1417 		s = splsoftnet();
1418 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1419 			next = pr->ndpr_next;
1420 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
1421 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
1422 			prelist_remove(pr);
1423 		}
1424 		splx(s);
1425 		break;
1426 	    }
1427 	case SIOCSRTRFLUSH_IN6:
1428 	    {
1429 		/* flush all the default routers */
1430 		struct nd_defrouter *dr, *next;
1431 
1432 		s = splsoftnet();
1433 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1434 			/*
1435 			 * The first entry of the list may be stored in
1436 			 * the routing table, so we'll delete it later.
1437 			 */
1438 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1439 				next = TAILQ_NEXT(dr, dr_entry);
1440 				defrtrlist_del(dr);
1441 			}
1442 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1443 		}
1444 		splx(s);
1445 		break;
1446 	    }
1447 	case SIOCGNBRINFO_IN6:
1448 	    {
1449 		struct llinfo_nd6 *ln;
1450 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1451 
1452 		/*
1453 		 * XXX: KAME specific hack for scoped addresses
1454 		 *      XXXX: for other scopes than link-local?
1455 		 */
1456 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1457 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1458 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1459 
1460 			if (*idp == 0)
1461 				*idp = htons(ifp->if_index);
1462 		}
1463 
1464 		s = splsoftnet();
1465 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1466 			error = EINVAL;
1467 			splx(s);
1468 			break;
1469 		}
1470 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1471 		nbi->state = ln->ln_state;
1472 		nbi->asked = ln->ln_asked;
1473 		nbi->isrouter = ln->ln_router;
1474 		nbi->expire = ln->ln_expire;
1475 		splx(s);
1476 
1477 		break;
1478 	    }
1479 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1480 		ndif->ifindex = nd6_defifindex;
1481 		break;
1482 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1483 		return(nd6_setdefaultiface(ndif->ifindex));
1484 		break;
1485 	}
1486 	return(error);
1487 }
1488 
1489 /*
1490  * Create neighbor cache entry and cache link-layer address,
1491  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1492  */
1493 struct rtentry *
1494 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1495 	struct ifnet *ifp;
1496 	struct in6_addr *from;
1497 	char *lladdr;
1498 	int lladdrlen;
1499 	int type;	/* ICMP6 type */
1500 	int code;	/* type dependent information */
1501 {
1502 	struct rtentry *rt = NULL;
1503 	struct llinfo_nd6 *ln = NULL;
1504 	int is_newentry;
1505 	struct sockaddr_dl *sdl = NULL;
1506 	int do_update;
1507 	int olladdr;
1508 	int llchange;
1509 	int newstate = 0;
1510 	long time_second = time.tv_sec;
1511 
1512 	if (!ifp)
1513 		panic("ifp == NULL in nd6_cache_lladdr");
1514 	if (!from)
1515 		panic("from == NULL in nd6_cache_lladdr");
1516 
1517 	/* nothing must be updated for unspecified address */
1518 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1519 		return NULL;
1520 
1521 	/*
1522 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1523 	 * the caller.
1524 	 *
1525 	 * XXX If the link does not have link-layer adderss, what should
1526 	 * we do? (ifp->if_addrlen == 0)
1527 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1528 	 * description on it in NS section (RFC 2461 7.2.3).
1529 	 */
1530 
1531 	rt = nd6_lookup(from, 0, ifp);
1532 	if (!rt) {
1533 #if 0
1534 		/* nothing must be done if there's no lladdr */
1535 		if (!lladdr || !lladdrlen)
1536 			return NULL;
1537 #endif
1538 
1539 		rt = nd6_lookup(from, 1, ifp);
1540 		is_newentry = 1;
1541 	} else {
1542 		/* do nothing if static ndp is set */
1543 		if (rt->rt_flags & RTF_STATIC)
1544 			return NULL;
1545 		is_newentry = 0;
1546 	}
1547 
1548 	if (!rt)
1549 		return NULL;
1550 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1551 fail:
1552 		(void)nd6_free(rt, 0);
1553 		return NULL;
1554 	}
1555 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1556 	if (!ln)
1557 		goto fail;
1558 	if (!rt->rt_gateway)
1559 		goto fail;
1560 	if (rt->rt_gateway->sa_family != AF_LINK)
1561 		goto fail;
1562 	sdl = SDL(rt->rt_gateway);
1563 
1564 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1565 	if (olladdr && lladdr) {
1566 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1567 			llchange = 1;
1568 		else
1569 			llchange = 0;
1570 	} else
1571 		llchange = 0;
1572 
1573 	/*
1574 	 * newentry olladdr  lladdr  llchange	(*=record)
1575 	 *	0	n	n	--	(1)
1576 	 *	0	y	n	--	(2)
1577 	 *	0	n	y	--	(3) * STALE
1578 	 *	0	y	y	n	(4) *
1579 	 *	0	y	y	y	(5) * STALE
1580 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1581 	 *	1	--	y	--	(7) * STALE
1582 	 */
1583 
1584 	if (lladdr) {		/* (3-5) and (7) */
1585 		/*
1586 		 * Record source link-layer address
1587 		 * XXX is it dependent to ifp->if_type?
1588 		 */
1589 		sdl->sdl_alen = ifp->if_addrlen;
1590 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1591 	}
1592 
1593 	if (!is_newentry) {
1594 		if ((!olladdr && lladdr)		/* (3) */
1595 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1596 			do_update = 1;
1597 			newstate = ND6_LLINFO_STALE;
1598 		} else					/* (1-2,4) */
1599 			do_update = 0;
1600 	} else {
1601 		do_update = 1;
1602 		if (!lladdr)				/* (6) */
1603 			newstate = ND6_LLINFO_NOSTATE;
1604 		else					/* (7) */
1605 			newstate = ND6_LLINFO_STALE;
1606 	}
1607 
1608 	if (do_update) {
1609 		/*
1610 		 * Update the state of the neighbor cache.
1611 		 */
1612 		ln->ln_state = newstate;
1613 
1614 		if (ln->ln_state == ND6_LLINFO_STALE) {
1615 			/*
1616 			 * XXX: since nd6_output() below will cause
1617 			 * state tansition to DELAY and reset the timer,
1618 			 * we must set the timer now, although it is actually
1619 			 * meaningless.
1620 			 */
1621 			ln->ln_expire = time_second + nd6_gctimer;
1622 
1623 			if (ln->ln_hold) {
1624 				/*
1625 				 * we assume ifp is not a p2p here, so just
1626 				 * set the 2nd argument as the 1st one.
1627 				 */
1628 				nd6_output(ifp, ifp, ln->ln_hold,
1629 					   (struct sockaddr_in6 *)rt_key(rt),
1630 					   rt);
1631 				ln->ln_hold = NULL;
1632 			}
1633 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1634 			/* probe right away */
1635 			ln->ln_expire = time_second;
1636 		}
1637 	}
1638 
1639 	/*
1640 	 * ICMP6 type dependent behavior.
1641 	 *
1642 	 * NS: clear IsRouter if new entry
1643 	 * RS: clear IsRouter
1644 	 * RA: set IsRouter if there's lladdr
1645 	 * redir: clear IsRouter if new entry
1646 	 *
1647 	 * RA case, (1):
1648 	 * The spec says that we must set IsRouter in the following cases:
1649 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1650 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1651 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1652 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1653 	 * neighbor cache, this is similar to (6).
1654 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1655 	 *
1656 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1657 	 *							D R
1658 	 *	0	n	n	--	(1)	c   ?     s
1659 	 *	0	y	n	--	(2)	c   s     s
1660 	 *	0	n	y	--	(3)	c   s     s
1661 	 *	0	y	y	n	(4)	c   s     s
1662 	 *	0	y	y	y	(5)	c   s     s
1663 	 *	1	--	n	--	(6) c	c 	c s
1664 	 *	1	--	y	--	(7) c	c   s	c s
1665 	 *
1666 	 *					(c=clear s=set)
1667 	 */
1668 	switch (type & 0xff) {
1669 	case ND_NEIGHBOR_SOLICIT:
1670 		/*
1671 		 * New entry must have is_router flag cleared.
1672 		 */
1673 		if (is_newentry)	/* (6-7) */
1674 			ln->ln_router = 0;
1675 		break;
1676 	case ND_REDIRECT:
1677 		/*
1678 		 * If the icmp is a redirect to a better router, always set the
1679 		 * is_router flag.  Otherwise, if the entry is newly created,
1680 		 * clear the flag.  [RFC 2461, sec 8.3]
1681 		 */
1682 		if (code == ND_REDIRECT_ROUTER)
1683 			ln->ln_router = 1;
1684 		else if (is_newentry) /* (6-7) */
1685 			ln->ln_router = 0;
1686 		break;
1687 	case ND_ROUTER_SOLICIT:
1688 		/*
1689 		 * is_router flag must always be cleared.
1690 		 */
1691 		ln->ln_router = 0;
1692 		break;
1693 	case ND_ROUTER_ADVERT:
1694 		/*
1695 		 * Mark an entry with lladdr as a router.
1696 		 */
1697 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1698 		 || (is_newentry && lladdr)) {			/* (7) */
1699 			ln->ln_router = 1;
1700 		}
1701 		break;
1702 	}
1703 
1704 	/*
1705 	 * When the link-layer address of a router changes, select the
1706 	 * best router again.  In particular, when the neighbor entry is newly
1707 	 * created, it might affect the selection policy.
1708 	 * Question: can we restrict the first condition to the "is_newentry"
1709 	 * case?
1710 	 * XXX: when we hear an RA from a new router with the link-layer
1711 	 * address option, defrouter_select() is called twice, since
1712 	 * defrtrlist_update called the function as well.  However, I believe
1713 	 * we can compromise the overhead, since it only happens the first
1714 	 * time.
1715 	 * XXX: although defrouter_select() should not have a bad effect
1716 	 * for those are not autoconfigured hosts, we explicitly avoid such
1717 	 * cases for safety.
1718 	 */
1719 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1720 		defrouter_select();
1721 
1722 	return rt;
1723 }
1724 
1725 static void
1726 nd6_slowtimo(ignored_arg)
1727     void *ignored_arg;
1728 {
1729 	int s = splsoftnet();
1730 	int i;
1731 	struct nd_ifinfo *nd6if;
1732 
1733 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1734 	    nd6_slowtimo, NULL);
1735 	for (i = 1; i < if_index + 1; i++) {
1736 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1737 			continue;
1738 		nd6if = &nd_ifinfo[i];
1739 		if (nd6if->basereachable && /* already initialized */
1740 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1741 			/*
1742 			 * Since reachable time rarely changes by router
1743 			 * advertisements, we SHOULD insure that a new random
1744 			 * value gets recomputed at least once every few hours.
1745 			 * (RFC 2461, 6.3.4)
1746 			 */
1747 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1748 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1749 		}
1750 	}
1751 	splx(s);
1752 }
1753 
1754 #define senderr(e) { error = (e); goto bad;}
1755 int
1756 nd6_output(ifp, origifp, m0, dst, rt0)
1757 	struct ifnet *ifp;
1758 	struct ifnet *origifp;
1759 	struct mbuf *m0;
1760 	struct sockaddr_in6 *dst;
1761 	struct rtentry *rt0;
1762 {
1763 	struct mbuf *m = m0;
1764 	struct rtentry *rt = rt0;
1765 	struct sockaddr_in6 *gw6 = NULL;
1766 	struct llinfo_nd6 *ln = NULL;
1767 	int error = 0;
1768 	long time_second = time.tv_sec;
1769 
1770 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1771 		goto sendpkt;
1772 
1773 	if (nd6_need_cache(ifp) == 0)
1774 		goto sendpkt;
1775 
1776 	/*
1777 	 * next hop determination.  This routine is derived from ether_outpout.
1778 	 */
1779 	if (rt) {
1780 		if ((rt->rt_flags & RTF_UP) == 0) {
1781 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
1782 				NULL)
1783 			{
1784 				rt->rt_refcnt--;
1785 				if (rt->rt_ifp != ifp) {
1786 					/* XXX: loop care? */
1787 					return nd6_output(ifp, origifp, m0,
1788 							  dst, rt);
1789 				}
1790 			} else
1791 				senderr(EHOSTUNREACH);
1792 		}
1793 
1794 		if (rt->rt_flags & RTF_GATEWAY) {
1795 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1796 
1797 			/*
1798 			 * We skip link-layer address resolution and NUD
1799 			 * if the gateway is not a neighbor from ND point
1800 			 * of view, regardless the value of the
1801 			 * nd_ifinfo.flags.
1802 			 * The second condition is a bit tricky: we skip
1803 			 * if the gateway is our own address, which is
1804 			 * sometimes used to install a route to a p2p link.
1805 			 */
1806 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1807 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1808 				/*
1809 				 * We allow this kind of tricky route only
1810 				 * when the outgoing interface is p2p.
1811 				 * XXX: we may need a more generic rule here.
1812 				 */
1813 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1814 					senderr(EHOSTUNREACH);
1815 
1816 				goto sendpkt;
1817 			}
1818 
1819 			if (rt->rt_gwroute == 0)
1820 				goto lookup;
1821 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1822 				rtfree(rt); rt = rt0;
1823 			lookup:
1824 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1825 				if ((rt = rt->rt_gwroute) == 0)
1826 					senderr(EHOSTUNREACH);
1827 				/* the "G" test below also prevents rt == rt0 */
1828 				if ((rt->rt_flags & RTF_GATEWAY) ||
1829 				    (rt->rt_ifp != ifp)) {
1830 					rt->rt_refcnt--;
1831 					rt0->rt_gwroute = 0;
1832 					senderr(EHOSTUNREACH);
1833 				}
1834 			}
1835 		}
1836 	}
1837 
1838 	/*
1839 	 * Address resolution or Neighbor Unreachability Detection
1840 	 * for the next hop.
1841 	 * At this point, the destination of the packet must be a unicast
1842 	 * or an anycast address(i.e. not a multicast).
1843 	 */
1844 
1845 	/* Look up the neighbor cache for the nexthop */
1846 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1847 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1848 	else {
1849 		/*
1850 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1851 		 * the condition below is not very efficient.  But we believe
1852 		 * it is tolerable, because this should be a rare case.
1853 		 */
1854 		if (nd6_is_addr_neighbor(dst, ifp) &&
1855 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1856 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1857 	}
1858 	if (!ln || !rt) {
1859 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1860 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1861 			log(LOG_DEBUG,
1862 			    "nd6_output: can't allocate llinfo for %s "
1863 			    "(ln=%p, rt=%p)\n",
1864 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1865 			senderr(EIO);	/* XXX: good error? */
1866 		}
1867 
1868 		goto sendpkt;	/* send anyway */
1869 	}
1870 
1871 	/* We don't have to do link-layer address resolution on a p2p link. */
1872 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1873 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1874 		ln->ln_state = ND6_LLINFO_STALE;
1875 		ln->ln_expire = time_second + nd6_gctimer;
1876 	}
1877 
1878 	/*
1879 	 * The first time we send a packet to a neighbor whose entry is
1880 	 * STALE, we have to change the state to DELAY and a sets a timer to
1881 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1882 	 * neighbor unreachability detection on expiration.
1883 	 * (RFC 2461 7.3.3)
1884 	 */
1885 	if (ln->ln_state == ND6_LLINFO_STALE) {
1886 		ln->ln_asked = 0;
1887 		ln->ln_state = ND6_LLINFO_DELAY;
1888 		ln->ln_expire = time_second + nd6_delay;
1889 	}
1890 
1891 	/*
1892 	 * If the neighbor cache entry has a state other than INCOMPLETE
1893 	 * (i.e. its link-layer address is already resolved), just
1894 	 * send the packet.
1895 	 */
1896 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1897 		goto sendpkt;
1898 
1899 	/*
1900 	 * There is a neighbor cache entry, but no ethernet address
1901 	 * response yet.  Replace the held mbuf (if any) with this
1902 	 * latest one.
1903 	 * This code conforms to the rate-limiting rule described in Section
1904 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1905 	 * an NS below.
1906 	 */
1907 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1908 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1909 	if (ln->ln_hold)
1910 		m_freem(ln->ln_hold);
1911 	ln->ln_hold = m;
1912 	if (ln->ln_expire) {
1913 		if (ln->ln_asked < nd6_mmaxtries &&
1914 		    ln->ln_expire < time_second) {
1915 			ln->ln_asked++;
1916 			ln->ln_expire = time_second +
1917 				nd_ifinfo[ifp->if_index].retrans / 1000;
1918 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1919 		}
1920 	}
1921 	return(0);
1922 
1923   sendpkt:
1924 
1925 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1926 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1927 					 rt));
1928 	}
1929 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1930 
1931   bad:
1932 	if (m)
1933 		m_freem(m);
1934 	return (error);
1935 }
1936 #undef senderr
1937 
1938 int
1939 nd6_need_cache(ifp)
1940 	struct ifnet *ifp;
1941 {
1942 	/*
1943 	 * XXX: we currently do not make neighbor cache on any interface
1944 	 * other than ARCnet, Ethernet, FDDI and GIF.
1945 	 *
1946 	 * RFC2893 says:
1947 	 * - unidirectional tunnels needs no ND
1948 	 */
1949 	switch (ifp->if_type) {
1950 	case IFT_ARCNET:
1951 	case IFT_ETHER:
1952 	case IFT_FDDI:
1953 	case IFT_IEEE1394:
1954 	case IFT_GIF:		/* XXX need more cases? */
1955 		return(1);
1956 	default:
1957 		return(0);
1958 	}
1959 }
1960 
1961 int
1962 nd6_storelladdr(ifp, rt, m, dst, desten)
1963 	struct ifnet *ifp;
1964 	struct rtentry *rt;
1965 	struct mbuf *m;
1966 	struct sockaddr *dst;
1967 	u_char *desten;
1968 {
1969 	struct sockaddr_dl *sdl;
1970 
1971 	if (m->m_flags & M_MCAST) {
1972 		switch (ifp->if_type) {
1973 		case IFT_ETHER:
1974 		case IFT_FDDI:
1975 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1976 						 desten);
1977 			return(1);
1978 		case IFT_IEEE1394:
1979 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1980 			return(1);
1981 		case IFT_ARCNET:
1982 			*desten = 0;
1983 			return(1);
1984 		default:
1985 			m_freem(m);
1986 			return(0);
1987 		}
1988 	}
1989 
1990 	if (rt == NULL) {
1991 		/* this could happen, if we could not allocate memory */
1992 		m_freem(m);
1993 		return(0);
1994 	}
1995 	if (rt->rt_gateway->sa_family != AF_LINK) {
1996 		printf("nd6_storelladdr: something odd happens\n");
1997 		m_freem(m);
1998 		return(0);
1999 	}
2000 	sdl = SDL(rt->rt_gateway);
2001 	if (sdl->sdl_alen == 0) {
2002 		/* this should be impossible, but we bark here for debugging */
2003 		printf("nd6_storelladdr: sdl_alen == 0\n");
2004 		m_freem(m);
2005 		return(0);
2006 	}
2007 
2008 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2009 	return(1);
2010 }
2011