xref: /netbsd-src/sys/netinet6/nd6.c (revision 1ca5c1b28139779176bd5c13ad7c5f25c0bcd5f8)
1 /*	$NetBSD: nd6.c,v 1.56 2001/12/18 03:04:04 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/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.56 2001/12/18 03:04:04 itojun Exp $");
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/ioctl.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 
59 #include <net/if.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/if_atm.h>
63 #include <net/if_ieee1394.h>
64 #include <net/route.h>
65 
66 #include <netinet/in.h>
67 #include <net/if_ether.h>
68 #include <netinet/if_inarp.h>
69 #include <net/if_fddi.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/in6_prefix.h>
75 #include <netinet/icmp6.h>
76 
77 #include "loop.h"
78 extern struct ifnet loif[NLOOP];
79 
80 #include <net/net_osdep.h>
81 
82 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
83 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
84 
85 #define SIN6(s) ((struct sockaddr_in6 *)s)
86 #define SDL(s) ((struct sockaddr_dl *)s)
87 
88 /* timer values */
89 int	nd6_prune	= 1;	/* walk list every 1 seconds */
90 int	nd6_delay	= 5;	/* delay first probe time 5 second */
91 int	nd6_umaxtries	= 3;	/* maximum unicast query */
92 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
93 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
94 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
95 
96 /* preventing too many loops in ND option parsing */
97 int nd6_maxndopt = 10;	/* max # of ND options allowed */
98 
99 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
100 
101 #ifdef ND6_DEBUG
102 int nd6_debug = 1;
103 #else
104 int nd6_debug = 0;
105 #endif
106 
107 /* for debugging? */
108 static int nd6_inuse, nd6_allocated;
109 
110 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
111 static size_t nd_ifinfo_indexlim = 8;
112 struct nd_ifinfo *nd_ifinfo = NULL;
113 struct nd_drhead nd_defrouter;
114 struct nd_prhead nd_prefix = { 0 };
115 
116 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
117 static struct sockaddr_in6 all1_sa;
118 
119 static void nd6_slowtimo __P((void *));
120 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
121 
122 struct callout nd6_slowtimo_ch;
123 struct callout nd6_timer_ch;
124 
125 void
126 nd6_init()
127 {
128 	static int nd6_init_done = 0;
129 	int i;
130 
131 	if (nd6_init_done) {
132 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
133 		return;
134 	}
135 
136 	all1_sa.sin6_family = AF_INET6;
137 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
138 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
139 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
140 
141 	/* initialization of the default router list */
142 	TAILQ_INIT(&nd_defrouter);
143 
144 	nd6_init_done = 1;
145 
146 	/* start timer */
147 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
148 	    nd6_slowtimo, NULL);
149 }
150 
151 void
152 nd6_ifattach(ifp)
153 	struct ifnet *ifp;
154 {
155 
156 	/*
157 	 * We have some arrays that should be indexed by if_index.
158 	 * since if_index will grow dynamically, they should grow too.
159 	 */
160 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
161 		size_t n;
162 		caddr_t q;
163 
164 		while (if_index >= nd_ifinfo_indexlim)
165 			nd_ifinfo_indexlim <<= 1;
166 
167 		/* grow nd_ifinfo */
168 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
169 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
170 		bzero(q, n);
171 		if (nd_ifinfo) {
172 			bcopy((caddr_t)nd_ifinfo, q, n/2);
173 			free((caddr_t)nd_ifinfo, M_IP6NDP);
174 		}
175 		nd_ifinfo = (struct nd_ifinfo *)q;
176 	}
177 
178 #define ND nd_ifinfo[ifp->if_index]
179 
180 	/*
181 	 * Don't initialize if called twice.
182 	 * XXX: to detect this, we should choose a member that is never set
183 	 * before initialization of the ND structure itself.  We formaly used
184 	 * the linkmtu member, which was not suitable because it could be
185 	 * initialized via "ifconfig mtu".
186 	 */
187 	if (ND.basereachable)
188 		return;
189 
190 #ifdef DIAGNOSTIC
191 	if (!ifindex2ifnet[ifp->if_index])
192 		panic("nd6_ifattach: ifindex2ifnet is NULL");
193 #endif
194 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
195 	ND.chlim = IPV6_DEFHLIM;
196 	ND.basereachable = REACHABLE_TIME;
197 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
198 	ND.retrans = RETRANS_TIMER;
199 	ND.receivedra = 0;
200 	ND.flags = ND6_IFF_PERFORMNUD;
201 	nd6_setmtu(ifp);
202 #undef ND
203 }
204 
205 /*
206  * Reset ND level link MTU. This function is called when the physical MTU
207  * changes, which means we might have to adjust the ND level MTU.
208  */
209 void
210 nd6_setmtu(ifp)
211 	struct ifnet *ifp;
212 {
213 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
214 	u_long oldmaxmtu = ndi->maxmtu;
215 	u_long oldlinkmtu = ndi->linkmtu;
216 
217 	switch (ifp->if_type) {
218 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
219 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
220 		break;
221 	case IFT_ETHER:
222 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
223 		break;
224 	case IFT_ATM:
225 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
226 		break;
227 	case IFT_IEEE1394:
228 		ndi->maxmtu = MIN(IEEE1394MTU, ifp->if_mtu);
229 		break;
230 	default:
231 		ndi->maxmtu = ifp->if_mtu;
232 		break;
233 	}
234 
235 	if (oldmaxmtu != ndi->maxmtu) {
236 		/*
237 		 * If the ND level MTU is not set yet, or if the maxmtu
238 		 * is reset to a smaller value than the ND level MTU,
239 		 * also reset the ND level MTU.
240 		 */
241 		if (ndi->linkmtu == 0 ||
242 		    ndi->maxmtu < ndi->linkmtu) {
243 			ndi->linkmtu = ndi->maxmtu;
244 			/* also adjust in6_maxmtu if necessary. */
245 			if (oldlinkmtu == 0) {
246 				/*
247 				 * XXX: the case analysis is grotty, but
248 				 * it is not efficient to call in6_setmaxmtu()
249 				 * here when we are during the initialization
250 				 * procedure.
251 				 */
252 				if (in6_maxmtu < ndi->linkmtu)
253 					in6_maxmtu = ndi->linkmtu;
254 			} else
255 				in6_setmaxmtu();
256 		}
257 	}
258 #undef MIN
259 }
260 
261 void
262 nd6_option_init(opt, icmp6len, ndopts)
263 	void *opt;
264 	int icmp6len;
265 	union nd_opts *ndopts;
266 {
267 	bzero(ndopts, sizeof(*ndopts));
268 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
269 	ndopts->nd_opts_last
270 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
271 
272 	if (icmp6len == 0) {
273 		ndopts->nd_opts_done = 1;
274 		ndopts->nd_opts_search = NULL;
275 	}
276 }
277 
278 /*
279  * Take one ND option.
280  */
281 struct nd_opt_hdr *
282 nd6_option(ndopts)
283 	union nd_opts *ndopts;
284 {
285 	struct nd_opt_hdr *nd_opt;
286 	int olen;
287 
288 	if (!ndopts)
289 		panic("ndopts == NULL in nd6_option\n");
290 	if (!ndopts->nd_opts_last)
291 		panic("uninitialized ndopts in nd6_option\n");
292 	if (!ndopts->nd_opts_search)
293 		return NULL;
294 	if (ndopts->nd_opts_done)
295 		return NULL;
296 
297 	nd_opt = ndopts->nd_opts_search;
298 
299 	/* make sure nd_opt_len is inside the buffer */
300 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
301 		bzero(ndopts, sizeof(*ndopts));
302 		return NULL;
303 	}
304 
305 	olen = nd_opt->nd_opt_len << 3;
306 	if (olen == 0) {
307 		/*
308 		 * Message validation requires that all included
309 		 * options have a length that is greater than zero.
310 		 */
311 		bzero(ndopts, sizeof(*ndopts));
312 		return NULL;
313 	}
314 
315 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
316 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
317 		/* option overruns the end of buffer, invalid */
318 		bzero(ndopts, sizeof(*ndopts));
319 		return NULL;
320 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
321 		/* reached the end of options chain */
322 		ndopts->nd_opts_done = 1;
323 		ndopts->nd_opts_search = NULL;
324 	}
325 	return nd_opt;
326 }
327 
328 /*
329  * Parse multiple ND options.
330  * This function is much easier to use, for ND routines that do not need
331  * multiple options of the same type.
332  */
333 int
334 nd6_options(ndopts)
335 	union nd_opts *ndopts;
336 {
337 	struct nd_opt_hdr *nd_opt;
338 	int i = 0;
339 
340 	if (!ndopts)
341 		panic("ndopts == NULL in nd6_options\n");
342 	if (!ndopts->nd_opts_last)
343 		panic("uninitialized ndopts in nd6_options\n");
344 	if (!ndopts->nd_opts_search)
345 		return 0;
346 
347 	while (1) {
348 		nd_opt = nd6_option(ndopts);
349 		if (!nd_opt && !ndopts->nd_opts_last) {
350 			/*
351 			 * Message validation requires that all included
352 			 * options have a length that is greater than zero.
353 			 */
354 			icmp6stat.icp6s_nd_badopt++;
355 			bzero(ndopts, sizeof(*ndopts));
356 			return -1;
357 		}
358 
359 		if (!nd_opt)
360 			goto skip1;
361 
362 		switch (nd_opt->nd_opt_type) {
363 		case ND_OPT_SOURCE_LINKADDR:
364 		case ND_OPT_TARGET_LINKADDR:
365 		case ND_OPT_MTU:
366 		case ND_OPT_REDIRECTED_HEADER:
367 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
368 				nd6log((LOG_INFO,
369 				    "duplicated ND6 option found (type=%d)\n",
370 				    nd_opt->nd_opt_type));
371 				/* XXX bark? */
372 			} else {
373 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
374 					= nd_opt;
375 			}
376 			break;
377 		case ND_OPT_PREFIX_INFORMATION:
378 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
379 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
380 					= nd_opt;
381 			}
382 			ndopts->nd_opts_pi_end =
383 				(struct nd_opt_prefix_info *)nd_opt;
384 			break;
385 		default:
386 			/*
387 			 * Unknown options must be silently ignored,
388 			 * to accomodate future extension to the protocol.
389 			 */
390 			nd6log((LOG_DEBUG,
391 			    "nd6_options: unsupported option %d - "
392 			    "option ignored\n", nd_opt->nd_opt_type));
393 		}
394 
395 skip1:
396 		i++;
397 		if (i > nd6_maxndopt) {
398 			icmp6stat.icp6s_nd_toomanyopt++;
399 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
400 			break;
401 		}
402 
403 		if (ndopts->nd_opts_done)
404 			break;
405 	}
406 
407 	return 0;
408 }
409 
410 /*
411  * ND6 timer routine to expire default route list and prefix list
412  */
413 void
414 nd6_timer(ignored_arg)
415 	void	*ignored_arg;
416 {
417 	int s;
418 	struct llinfo_nd6 *ln;
419 	struct nd_defrouter *dr;
420 	struct nd_prefix *pr;
421 	long time_second = time.tv_sec;
422 
423 	s = splsoftnet();
424 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
425 	    nd6_timer, NULL);
426 
427 	ln = llinfo_nd6.ln_next;
428 	while (ln && ln != &llinfo_nd6) {
429 		struct rtentry *rt;
430 		struct ifnet *ifp;
431 		struct sockaddr_in6 *dst;
432 		struct llinfo_nd6 *next = ln->ln_next;
433 		/* XXX: used for the DELAY case only: */
434 		struct nd_ifinfo *ndi = NULL;
435 
436 		if ((rt = ln->ln_rt) == NULL) {
437 			ln = next;
438 			continue;
439 		}
440 		if ((ifp = rt->rt_ifp) == NULL) {
441 			ln = next;
442 			continue;
443 		}
444 		ndi = &nd_ifinfo[ifp->if_index];
445 		dst = (struct sockaddr_in6 *)rt_key(rt);
446 
447 		if (ln->ln_expire > time_second) {
448 			ln = next;
449 			continue;
450 		}
451 
452 		/* sanity check */
453 		if (!rt)
454 			panic("rt=0 in nd6_timer(ln=%p)\n", ln);
455 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
456 			panic("rt_llinfo(%p) is not equal to ln(%p)\n",
457 			      rt->rt_llinfo, ln);
458 		if (!dst)
459 			panic("dst=0 in nd6_timer(ln=%p)\n", ln);
460 
461 		switch (ln->ln_state) {
462 		case ND6_LLINFO_INCOMPLETE:
463 			if (ln->ln_asked < nd6_mmaxtries) {
464 				ln->ln_asked++;
465 				ln->ln_expire = time_second +
466 					nd_ifinfo[ifp->if_index].retrans / 1000;
467 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
468 					ln, 0);
469 			} else {
470 				struct mbuf *m = ln->ln_hold;
471 				if (m) {
472 					if (rt->rt_ifp) {
473 						/*
474 						 * Fake rcvif to make ICMP error
475 						 * more helpful in diagnosing
476 						 * for the receiver.
477 						 * XXX: should we consider
478 						 * older rcvif?
479 						 */
480 						m->m_pkthdr.rcvif = rt->rt_ifp;
481 					}
482 					icmp6_error(m, ICMP6_DST_UNREACH,
483 						    ICMP6_DST_UNREACH_ADDR, 0);
484 					ln->ln_hold = NULL;
485 				}
486 				next = nd6_free(rt, 0);
487 			}
488 			break;
489 		case ND6_LLINFO_REACHABLE:
490 			if (ln->ln_expire) {
491 				ln->ln_state = ND6_LLINFO_STALE;
492 				ln->ln_expire = time_second + nd6_gctimer;
493 			}
494 			break;
495 
496 		case ND6_LLINFO_STALE:
497 			/* Garbage Collection(RFC 2461 5.3) */
498 			if (ln->ln_expire)
499 				next = nd6_free(rt, 1);
500 			break;
501 
502 		case ND6_LLINFO_DELAY:
503 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
504 				/* We need NUD */
505 				ln->ln_asked = 1;
506 				ln->ln_state = ND6_LLINFO_PROBE;
507 				ln->ln_expire = time_second +
508 					ndi->retrans / 1000;
509 				nd6_ns_output(ifp, &dst->sin6_addr,
510 					      &dst->sin6_addr,
511 					      ln, 0);
512 			} else {
513 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
514 				ln->ln_expire = time_second + nd6_gctimer;
515 			}
516 			break;
517 		case ND6_LLINFO_PROBE:
518 			if (ln->ln_asked < nd6_umaxtries) {
519 				ln->ln_asked++;
520 				ln->ln_expire = time_second +
521 					nd_ifinfo[ifp->if_index].retrans / 1000;
522 				nd6_ns_output(ifp, &dst->sin6_addr,
523 					       &dst->sin6_addr, ln, 0);
524 			} else {
525 				next = nd6_free(rt, 0);
526 			}
527 			break;
528 		}
529 		ln = next;
530 	}
531 
532 	/* expire default router list */
533 	dr = TAILQ_FIRST(&nd_defrouter);
534 	while (dr) {
535 		if (dr->expire && dr->expire < time_second) {
536 			struct nd_defrouter *t;
537 			t = TAILQ_NEXT(dr, dr_entry);
538 			defrtrlist_del(dr);
539 			dr = t;
540 		} else {
541 			dr = TAILQ_NEXT(dr, dr_entry);
542 		}
543 	}
544 	pr = nd_prefix.lh_first;
545 	while (pr) {
546 		struct in6_ifaddr *ia6;
547 		struct in6_addrlifetime *lt6;
548 
549 		if (IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
550 			ia6 = NULL;
551 		else
552 			ia6 = in6ifa_ifpwithaddr(pr->ndpr_ifp, &pr->ndpr_addr);
553 
554 		if (ia6) {
555 			/* check address lifetime */
556 			lt6 = &ia6->ia6_lifetime;
557 			if (lt6->ia6t_preferred && lt6->ia6t_preferred < time_second)
558 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
559 			if (lt6->ia6t_expire && lt6->ia6t_expire < time_second) {
560 				if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
561 					in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
562 				/* xxx ND_OPT_PI_FLAG_ONLINK processing */
563 			}
564 		}
565 
566 		/*
567 		 * check prefix lifetime.
568 		 * since pltime is just for autoconf, pltime processing for
569 		 * prefix is not necessary.
570 		 *
571 		 * we offset expire time by NDPR_KEEP_EXPIRE, so that we
572 		 * can use the old prefix information to validate the
573 		 * next prefix information to come.  See prelist_update()
574 		 * for actual validation.
575 		 */
576 		if (pr->ndpr_expire
577 		 && pr->ndpr_expire + NDPR_KEEP_EXPIRED < time_second) {
578 			struct nd_prefix *t;
579 			t = pr->ndpr_next;
580 
581 			/*
582 			 * address expiration and prefix expiration are
583 			 * separate.  NEVER perform in6_ifdel here.
584 			 */
585 
586 			prelist_remove(pr);
587 			pr = t;
588 		} else
589 			pr = pr->ndpr_next;
590 	}
591 	splx(s);
592 }
593 
594 /*
595  * Nuke neighbor cache/prefix/default router management table, right before
596  * ifp goes away.
597  */
598 void
599 nd6_purge(ifp)
600 	struct ifnet *ifp;
601 {
602 	struct llinfo_nd6 *ln, *nln;
603 	struct nd_defrouter *dr, *ndr, drany;
604 	struct nd_prefix *pr, *npr;
605 
606 	/* Nuke default router list entries toward ifp */
607 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
608 		/*
609 		 * The first entry of the list may be stored in
610 		 * the routing table, so we'll delete it later.
611 		 */
612 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
613 			ndr = TAILQ_NEXT(dr, dr_entry);
614 			if (dr->ifp == ifp)
615 				defrtrlist_del(dr);
616 		}
617 		dr = TAILQ_FIRST(&nd_defrouter);
618 		if (dr->ifp == ifp)
619 			defrtrlist_del(dr);
620 	}
621 
622 	/* Nuke prefix list entries toward ifp */
623 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
624 		npr = pr->ndpr_next;
625 		if (pr->ndpr_ifp == ifp) {
626 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
627 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
628 			prelist_remove(pr);
629 		}
630 	}
631 
632 	/* cancel default outgoing interface setting */
633 	if (nd6_defifindex == ifp->if_index)
634 		nd6_setdefaultiface(0);
635 
636 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
637 		/* refresh default router list */
638 		bzero(&drany, sizeof(drany));
639 		defrouter_delreq(&drany, 0);
640 		defrouter_select();
641 	}
642 
643 	/*
644 	 * Nuke neighbor cache entries for the ifp.
645 	 * Note that rt->rt_ifp may not be the same as ifp,
646 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
647 	 * nd6_rtrequest(), and ip6_input().
648 	 */
649 	ln = llinfo_nd6.ln_next;
650 	while (ln && ln != &llinfo_nd6) {
651 		struct rtentry *rt;
652 		struct sockaddr_dl *sdl;
653 
654 		nln = ln->ln_next;
655 		rt = ln->ln_rt;
656 		if (rt && rt->rt_gateway &&
657 		    rt->rt_gateway->sa_family == AF_LINK) {
658 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
659 			if (sdl->sdl_index == ifp->if_index)
660 				nln = nd6_free(rt, 0);
661 		}
662 		ln = nln;
663 	}
664 }
665 
666 struct rtentry *
667 nd6_lookup(addr6, create, ifp)
668 	struct in6_addr *addr6;
669 	int create;
670 	struct ifnet *ifp;
671 {
672 	struct rtentry *rt;
673 	struct sockaddr_in6 sin6;
674 
675 	bzero(&sin6, sizeof(sin6));
676 	sin6.sin6_len = sizeof(struct sockaddr_in6);
677 	sin6.sin6_family = AF_INET6;
678 	sin6.sin6_addr = *addr6;
679 	rt = rtalloc1((struct sockaddr *)&sin6, create);
680 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
681 		/*
682 		 * This is the case for the default route.
683 		 * If we want to create a neighbor cache for the address, we
684 		 * should free the route for the destination and allocate an
685 		 * interface route.
686 		 */
687 		if (create) {
688 			RTFREE(rt);
689 			rt = 0;
690 		}
691 	}
692 	if (!rt) {
693 		if (create && ifp) {
694 			int e;
695 
696 			/*
697 			 * If no route is available and create is set,
698 			 * we allocate a host route for the destination
699 			 * and treat it like an interface route.
700 			 * This hack is necessary for a neighbor which can't
701 			 * be covered by our own prefix.
702 			 */
703 			struct ifaddr *ifa =
704 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
705 			if (ifa == NULL)
706 				return(NULL);
707 
708 			/*
709 			 * Create a new route.  RTF_LLINFO is necessary
710 			 * to create a Neighbor Cache entry for the
711 			 * destination in nd6_rtrequest which will be
712 			 * called in rtrequest via ifa->ifa_rtrequest.
713 			 */
714 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
715 					   ifa->ifa_addr,
716 					   (struct sockaddr *)&all1_sa,
717 					   (ifa->ifa_flags |
718 					    RTF_HOST | RTF_LLINFO) &
719 					   ~RTF_CLONING,
720 					   &rt)) != 0) {
721 #if 0
722 				log(LOG_ERR,
723 				    "nd6_lookup: failed to add route for a "
724 				    "neighbor(%s), errno=%d\n",
725 				    ip6_sprintf(addr6), e);
726 #endif
727 				return(NULL);
728 			}
729 			if (rt == NULL)
730 				return(NULL);
731 			if (rt->rt_llinfo) {
732 				struct llinfo_nd6 *ln =
733 					(struct llinfo_nd6 *)rt->rt_llinfo;
734 				ln->ln_state = ND6_LLINFO_NOSTATE;
735 			}
736 		} else
737 			return(NULL);
738 	}
739 	rt->rt_refcnt--;
740 	/*
741 	 * Validation for the entry.
742 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
743 	 *      it might be the loopback interface if the entry is for our
744 	 *      own address on a non-loopback interface. Instead, we should
745 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL interface.
746 	 */
747 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
748 	    rt->rt_gateway->sa_family != AF_LINK ||
749 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
750 		if (create) {
751 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
752 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
753 		}
754 		return(0);
755 	}
756 	return(rt);
757 }
758 
759 /*
760  * Detect if a given IPv6 address identifies a neighbor on a given link.
761  * XXX: should take care of the destination of a p2p link?
762  */
763 int
764 nd6_is_addr_neighbor(addr, ifp)
765 	struct sockaddr_in6 *addr;
766 	struct ifnet *ifp;
767 {
768 	struct ifaddr *ifa;
769 	int i;
770 
771 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
772 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
773 
774 	/*
775 	 * A link-local address is always a neighbor.
776 	 * XXX: we should use the sin6_scope_id field rather than the embedded
777 	 * interface index.
778 	 * XXX: a link does not necessarily specify a single interface.
779 	 */
780 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
781 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
782 		return(1);
783 
784 	/*
785 	 * If the address matches one of our addresses,
786 	 * it should be a neighbor.
787 	 */
788 	for (ifa = ifp->if_addrlist.tqh_first;
789 	     ifa;
790 	     ifa = ifa->ifa_list.tqe_next)
791 	{
792 		if (ifa->ifa_addr->sa_family != AF_INET6)
793 			next: continue;
794 
795 		for (i = 0; i < 4; i++) {
796 			if ((IFADDR6(ifa).s6_addr32[i] ^
797 			     addr->sin6_addr.s6_addr32[i]) &
798 			    IFMASK6(ifa).s6_addr32[i])
799 				goto next;
800 		}
801 		return(1);
802 	}
803 
804 	/*
805 	 * Even if the address matches none of our addresses, it might be
806 	 * in the neighbor cache.
807 	 */
808 	if (nd6_lookup(&addr->sin6_addr, 0, ifp))
809 		return(1);
810 
811 	return(0);
812 #undef IFADDR6
813 #undef IFMASK6
814 }
815 
816 /*
817  * Free an nd6 llinfo entry.
818  * Since the function would cause significant changes in the kernel, DO NOT
819  * make it global, unless you have a strong reason for the change, and are sure
820  * that the change is safe.
821  */
822 static struct llinfo_nd6 *
823 nd6_free(rt, gc)
824 	struct rtentry *rt;
825 	int gc;
826 {
827 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
828 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
829 	struct nd_defrouter *dr;
830 
831 	/*
832 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
833 	 * even though it is not harmful, it was not really necessary.
834 	 */
835 
836 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
837 		int s;
838 		s = splsoftnet();
839 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
840 				      rt->rt_ifp);
841 
842 		if (dr != NULL && dr->expire &&
843 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
844 			/*
845 			 * If the reason for the deletion is just garbage
846 			 * collection, and the neighbor is an active default
847 			 * router, do not delete it.  Instead, reset the GC
848 			 * timer using the router's lifetime.
849 			 * Simply deleting the entry would affect default
850 			 * router selection, which is not necessarily a good
851 			 * thing, especially when we're using router preference
852 			 * values.
853 			 * XXX: the check for ln_state would be redundant,
854 			 *      but we intentionally keep it just in case.
855 			 */
856 			ln->ln_expire = dr->expire;
857 			splx(s);
858 			return(ln->ln_next);
859 		}
860 
861 		if (ln->ln_router || dr) {
862 			/*
863 			 * rt6_flush must be called whether or not the neighbor
864 			 * is in the Default Router List.
865 			 * See a corresponding comment in nd6_na_input().
866 			 */
867 			rt6_flush(&in6, rt->rt_ifp);
868 		}
869 
870 		if (dr) {
871 			/*
872 			 * Unreachablity of a router might affect the default
873 			 * router selection and on-link detection of advertised
874 			 * prefixes.
875 			 */
876 
877 			/*
878 			 * Temporarily fake the state to choose a new default
879 			 * router and to perform on-link determination of
880 			 * prefixes correctly.
881 			 * Below the state will be set correctly,
882 			 * or the entry itself will be deleted.
883 			 */
884 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
885 
886 			/*
887 			 * Since defrouter_select() does not affect the
888 			 * on-link determination and MIP6 needs the check
889 			 * before the default router selection, we perform
890 			 * the check now.
891 			 */
892 			pfxlist_onlink_check();
893 
894 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
895 				/*
896 				 * It is used as the current default router,
897 				 * so we have to move it to the end of the
898 				 * list and choose a new one.
899 				 * XXX: it is not very efficient if this is
900 				 *      the only router.
901 				 */
902 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
903 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
904 
905 				defrouter_select();
906 			}
907 		}
908 		splx(s);
909 	}
910 
911 	/*
912 	 * Before deleting the entry, remember the next entry as the
913 	 * return value.  We need this because pfxlist_onlink_check() above
914 	 * might have freed other entries (particularly the old next entry) as
915 	 * a side effect (XXX).
916 	 */
917 	next = ln->ln_next;
918 
919 	/*
920 	 * Detach the route from the routing tree and the list of neighbor
921 	 * caches, and disable the route entry not to be used in already
922 	 * cached routes.
923 	 */
924 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
925 		  rt_mask(rt), 0, (struct rtentry **)0);
926 
927 	return(next);
928 }
929 
930 /*
931  * Upper-layer reachability hint for Neighbor Unreachability Detection.
932  *
933  * XXX cost-effective metods?
934  */
935 void
936 nd6_nud_hint(rt, dst6, force)
937 	struct rtentry *rt;
938 	struct in6_addr *dst6;
939 	int force;
940 {
941 	struct llinfo_nd6 *ln;
942 	long time_second = time.tv_sec;
943 
944 	/*
945 	 * If the caller specified "rt", use that.  Otherwise, resolve the
946 	 * routing table by supplied "dst6".
947 	 */
948 	if (!rt) {
949 		if (!dst6)
950 			return;
951 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
952 			return;
953 	}
954 
955 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
956 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
957 	    !rt->rt_llinfo || !rt->rt_gateway ||
958 	    rt->rt_gateway->sa_family != AF_LINK) {
959 		/* This is not a host route. */
960 		return;
961 	}
962 
963 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
964 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
965 		return;
966 
967 	/*
968 	 * if we get upper-layer reachability confirmation many times,
969 	 * it is possible we have false information.
970 	 */
971 	if (!force) {
972 		ln->ln_byhint++;
973 		if (ln->ln_byhint > nd6_maxnudhint)
974 			return;
975 	}
976 
977 	ln->ln_state = ND6_LLINFO_REACHABLE;
978 	if (ln->ln_expire)
979 		ln->ln_expire = time_second +
980 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
981 }
982 
983 void
984 nd6_rtrequest(req, rt, info)
985 	int	req;
986 	struct rtentry *rt;
987 	struct rt_addrinfo *info; /* xxx unused */
988 {
989 	struct sockaddr *gate = rt->rt_gateway;
990 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
991 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
992 	struct ifnet *ifp = rt->rt_ifp;
993 	struct ifaddr *ifa;
994 	long time_second = time.tv_sec;
995 
996 	if ((rt->rt_flags & RTF_GATEWAY))
997 		return;
998 
999 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1000 		/*
1001 		 * This is probably an interface direct route for a link
1002 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1003 		 * We do not need special treatment below for such a route.
1004 		 * Moreover, the RTF_LLINFO flag which would be set below
1005 		 * would annoy the ndp(8) command.
1006 		 */
1007 		return;
1008 	}
1009 
1010 	switch (req) {
1011 	case RTM_ADD:
1012 		/*
1013 		 * There is no backward compatibility :)
1014 		 *
1015 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1016 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1017 		 *	   rt->rt_flags |= RTF_CLONING;
1018 		 */
1019 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1020 			/*
1021 			 * Case 1: This route should come from
1022 			 * a route to interface.  RTF_LLINFO flag is set
1023 			 * for a host route whose destination should be
1024 			 * treated as on-link.
1025 			 */
1026 			rt_setgate(rt, rt_key(rt),
1027 				   (struct sockaddr *)&null_sdl);
1028 			gate = rt->rt_gateway;
1029 			SDL(gate)->sdl_type = ifp->if_type;
1030 			SDL(gate)->sdl_index = ifp->if_index;
1031 			if (ln)
1032 				ln->ln_expire = time_second;
1033 #if 1
1034 			if (ln && ln->ln_expire == 0) {
1035 				/* kludge for desktops */
1036 #if 0
1037 				printf("nd6_rtequest: time.tv_sec is zero; "
1038 				       "treat it as 1\n");
1039 #endif
1040 				ln->ln_expire = 1;
1041 			}
1042 #endif
1043 			if ((rt->rt_flags & RTF_CLONING))
1044 				break;
1045 		}
1046 		/*
1047 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1048 		 * We don't do that here since llinfo is not ready yet.
1049 		 *
1050 		 * There are also couple of other things to be discussed:
1051 		 * - unsolicited NA code needs improvement beforehand
1052 		 * - RFC2461 says we MAY send multicast unsolicited NA
1053 		 *   (7.2.6 paragraph 4), however, it also says that we
1054 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1055 		 *   we don't have anything like it right now.
1056 		 *   note that the mechanism needs a mutual agreement
1057 		 *   between proxies, which means that we need to implement
1058 		 *   a new protocol, or a new kludge.
1059 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1060 		 *   we need to check ip6forwarding before sending it.
1061 		 *   (or should we allow proxy ND configuration only for
1062 		 *   routers?  there's no mention about proxy ND from hosts)
1063 		 */
1064 #if 0
1065 		/* XXX it does not work */
1066 		if (rt->rt_flags & RTF_ANNOUNCE)
1067 			nd6_na_output(ifp,
1068 			      &SIN6(rt_key(rt))->sin6_addr,
1069 			      &SIN6(rt_key(rt))->sin6_addr,
1070 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1071 			      1, NULL);
1072 #endif
1073 		/* FALLTHROUGH */
1074 	case RTM_RESOLVE:
1075 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
1076 			/*
1077 			 * Address resolution isn't necessary for a point to
1078 			 * point link, so we can skip this test for a p2p link.
1079 			 */
1080 			if (gate->sa_family != AF_LINK ||
1081 			    gate->sa_len < sizeof(null_sdl)) {
1082 				log(LOG_DEBUG,
1083 				    "nd6_rtrequest: bad gateway value: %s\n",
1084 				    if_name(ifp));
1085 				break;
1086 			}
1087 			SDL(gate)->sdl_type = ifp->if_type;
1088 			SDL(gate)->sdl_index = ifp->if_index;
1089 		}
1090 		if (ln != NULL)
1091 			break;	/* This happens on a route change */
1092 		/*
1093 		 * Case 2: This route may come from cloning, or a manual route
1094 		 * add with a LL address.
1095 		 */
1096 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1097 		rt->rt_llinfo = (caddr_t)ln;
1098 		if (!ln) {
1099 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1100 			break;
1101 		}
1102 		nd6_inuse++;
1103 		nd6_allocated++;
1104 		Bzero(ln, sizeof(*ln));
1105 		ln->ln_rt = rt;
1106 		/* this is required for "ndp" command. - shin */
1107 		if (req == RTM_ADD) {
1108 		        /*
1109 			 * gate should have some valid AF_LINK entry,
1110 			 * and ln->ln_expire should have some lifetime
1111 			 * which is specified by ndp command.
1112 			 */
1113 			ln->ln_state = ND6_LLINFO_REACHABLE;
1114 			ln->ln_byhint = 0;
1115 		} else {
1116 		        /*
1117 			 * When req == RTM_RESOLVE, rt is created and
1118 			 * initialized in rtrequest(), so rt_expire is 0.
1119 			 */
1120 			ln->ln_state = ND6_LLINFO_NOSTATE;
1121 			ln->ln_expire = time_second;
1122 		}
1123 		rt->rt_flags |= RTF_LLINFO;
1124 		ln->ln_next = llinfo_nd6.ln_next;
1125 		llinfo_nd6.ln_next = ln;
1126 		ln->ln_prev = &llinfo_nd6;
1127 		ln->ln_next->ln_prev = ln;
1128 
1129 		/*
1130 		 * check if rt_key(rt) is one of my address assigned
1131 		 * to the interface.
1132 		 */
1133 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1134 					  &SIN6(rt_key(rt))->sin6_addr);
1135 		if (ifa) {
1136 			caddr_t macp = nd6_ifptomac(ifp);
1137 			ln->ln_expire = 0;
1138 			ln->ln_state = ND6_LLINFO_REACHABLE;
1139 			ln->ln_byhint = 0;
1140 			if (macp) {
1141 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1142 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1143 			}
1144 			if (nd6_useloopback) {
1145 				rt->rt_ifp = &loif[0];	/* XXX */
1146 				/*
1147 				 * Make sure rt_ifa be equal to the ifaddr
1148 				 * corresponding to the address.
1149 				 * We need this because when we refer
1150 				 * rt_ifa->ia6_flags in ip6_input, we assume
1151 				 * that the rt_ifa points to the address instead
1152 				 * of the loopback address.
1153 				 */
1154 				if (ifa != rt->rt_ifa) {
1155 					IFAFREE(rt->rt_ifa);
1156 					IFAREF(ifa);
1157 					rt->rt_ifa = ifa;
1158 				}
1159 			}
1160 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1161 			ln->ln_expire = 0;
1162 			ln->ln_state = ND6_LLINFO_REACHABLE;
1163 			ln->ln_byhint = 0;
1164 
1165 			/* join solicited node multicast for proxy ND */
1166 			if (ifp->if_flags & IFF_MULTICAST) {
1167 				struct in6_addr llsol;
1168 				int error;
1169 
1170 				llsol = SIN6(rt_key(rt))->sin6_addr;
1171 				llsol.s6_addr16[0] = htons(0xff02);
1172 				llsol.s6_addr16[1] = htons(ifp->if_index);
1173 				llsol.s6_addr32[1] = 0;
1174 				llsol.s6_addr32[2] = htonl(1);
1175 				llsol.s6_addr8[12] = 0xff;
1176 
1177 				if (!in6_addmulti(&llsol, ifp, &error)) {
1178 					nd6log((LOG_ERR, "%s: failed to join "
1179 					    "%s (errno=%d)\n", if_name(ifp),
1180 					    ip6_sprintf(&llsol), error));
1181 				}
1182 			}
1183 		}
1184 		break;
1185 
1186 	case RTM_DELETE:
1187 		if (!ln)
1188 			break;
1189 		/* leave from solicited node multicast for proxy ND */
1190 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1191 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1192 			struct in6_addr llsol;
1193 			struct in6_multi *in6m;
1194 
1195 			llsol = SIN6(rt_key(rt))->sin6_addr;
1196 			llsol.s6_addr16[0] = htons(0xff02);
1197 			llsol.s6_addr16[1] = htons(ifp->if_index);
1198 			llsol.s6_addr32[1] = 0;
1199 			llsol.s6_addr32[2] = htonl(1);
1200 			llsol.s6_addr8[12] = 0xff;
1201 
1202 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1203 			if (in6m)
1204 				in6_delmulti(in6m);
1205 		}
1206 		nd6_inuse--;
1207 		ln->ln_next->ln_prev = ln->ln_prev;
1208 		ln->ln_prev->ln_next = ln->ln_next;
1209 		ln->ln_prev = NULL;
1210 		rt->rt_llinfo = 0;
1211 		rt->rt_flags &= ~RTF_LLINFO;
1212 		if (ln->ln_hold)
1213 			m_freem(ln->ln_hold);
1214 		Free((caddr_t)ln);
1215 	}
1216 }
1217 
1218 void
1219 nd6_p2p_rtrequest(req, rt, info)
1220 	int	req;
1221 	struct rtentry *rt;
1222 	struct rt_addrinfo *info; /* xxx unused */
1223 {
1224 	struct sockaddr *gate = rt->rt_gateway;
1225 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1226 	struct ifnet *ifp = rt->rt_ifp;
1227 	struct ifaddr *ifa;
1228 
1229 	if (rt->rt_flags & RTF_GATEWAY)
1230 		return;
1231 
1232 	switch (req) {
1233 	case RTM_ADD:
1234 		/*
1235 		 * There is no backward compatibility :)
1236 		 *
1237 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1238 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1239 		 *	   rt->rt_flags |= RTF_CLONING;
1240 		 */
1241 		if (rt->rt_flags & RTF_CLONING) {
1242 			/*
1243 			 * Case 1: This route should come from
1244 			 * a route to interface.
1245 			 */
1246 			rt_setgate(rt, rt_key(rt),
1247 				   (struct sockaddr *)&null_sdl);
1248 			gate = rt->rt_gateway;
1249 			SDL(gate)->sdl_type = ifp->if_type;
1250 			SDL(gate)->sdl_index = ifp->if_index;
1251 			break;
1252 		}
1253 		/* Announce a new entry if requested. */
1254 		if (rt->rt_flags & RTF_ANNOUNCE)
1255 			nd6_na_output(ifp,
1256 				      &SIN6(rt_key(rt))->sin6_addr,
1257 				      &SIN6(rt_key(rt))->sin6_addr,
1258 				      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1259 				      1, NULL);
1260 		/* FALLTHROUGH */
1261 	case RTM_RESOLVE:
1262 		/*
1263 		 * check if rt_key(rt) is one of my address assigned
1264 		 * to the interface.
1265 		 */
1266  		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1267 					  &SIN6(rt_key(rt))->sin6_addr);
1268 		if (ifa) {
1269 			if (nd6_useloopback) {
1270 				rt->rt_ifp = &loif[0];	/*XXX*/
1271 			}
1272 		}
1273 		break;
1274 	}
1275 }
1276 
1277 int
1278 nd6_ioctl(cmd, data, ifp)
1279 	u_long cmd;
1280 	caddr_t	data;
1281 	struct ifnet *ifp;
1282 {
1283 	struct in6_drlist *drl = (struct in6_drlist *)data;
1284 	struct in6_prlist *prl = (struct in6_prlist *)data;
1285 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1286 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1287 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1288 	struct nd_defrouter *dr, any;
1289 	struct nd_prefix *pr;
1290 	struct rtentry *rt;
1291 	int i = 0, error = 0;
1292 	int s;
1293 
1294 	switch (cmd) {
1295 	case SIOCGDRLST_IN6:
1296 		bzero(drl, sizeof(*drl));
1297 		s = splsoftnet();
1298 		dr = TAILQ_FIRST(&nd_defrouter);
1299 		while (dr && i < DRLSTSIZ) {
1300 			drl->defrouter[i].rtaddr = dr->rtaddr;
1301 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1302 				/* XXX: need to this hack for KAME stack */
1303 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1304 			} else
1305 				log(LOG_ERR,
1306 				    "default router list contains a "
1307 				    "non-linklocal address(%s)\n",
1308 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1309 
1310 			drl->defrouter[i].flags = dr->flags;
1311 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1312 			drl->defrouter[i].expire = dr->expire;
1313 			drl->defrouter[i].if_index = dr->ifp->if_index;
1314 			i++;
1315 			dr = TAILQ_NEXT(dr, dr_entry);
1316 		}
1317 		splx(s);
1318 		break;
1319 	case SIOCGPRLST_IN6:
1320 		/*
1321 		 * XXX meaning of fields, especialy "raflags", is very
1322 		 * differnet between RA prefix list and RR/static prefix list.
1323 		 * how about separating ioctls into two?
1324 		 */
1325 		bzero(prl, sizeof(*prl));
1326 		s = splsoftnet();
1327 		pr = nd_prefix.lh_first;
1328 		while (pr && i < PRLSTSIZ) {
1329 			struct nd_pfxrouter *pfr;
1330 			int j;
1331 
1332 			prl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1333 			prl->prefix[i].raflags = pr->ndpr_raf;
1334 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1335 			prl->prefix[i].vltime = pr->ndpr_vltime;
1336 			prl->prefix[i].pltime = pr->ndpr_pltime;
1337 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1338 			prl->prefix[i].expire = pr->ndpr_expire;
1339 
1340 			pfr = pr->ndpr_advrtrs.lh_first;
1341 			j = 0;
1342 			while (pfr) {
1343 				if (j < DRLSTSIZ) {
1344 #define RTRADDR prl->prefix[i].advrtr[j]
1345 					RTRADDR = pfr->router->rtaddr;
1346 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1347 						/* XXX: hack for KAME */
1348 						RTRADDR.s6_addr16[1] = 0;
1349 					} else
1350 						log(LOG_ERR,
1351 						    "a router(%s) advertises "
1352 						    "a prefix with "
1353 						    "non-link local address\n",
1354 						    ip6_sprintf(&RTRADDR));
1355 #undef RTRADDR
1356 				}
1357 				j++;
1358 				pfr = pfr->pfr_next;
1359 			}
1360 			prl->prefix[i].advrtrs = j;
1361 			prl->prefix[i].origin = PR_ORIG_RA;
1362 
1363 			i++;
1364 			pr = pr->ndpr_next;
1365 		}
1366 	      {
1367 		struct rr_prefix *rpp;
1368 
1369 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1370 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1371 			if (i >= PRLSTSIZ)
1372 				break;
1373 			prl->prefix[i].prefix = rpp->rp_prefix.sin6_addr;
1374 			prl->prefix[i].raflags = rpp->rp_raf;
1375 			prl->prefix[i].prefixlen = rpp->rp_plen;
1376 			prl->prefix[i].vltime = rpp->rp_vltime;
1377 			prl->prefix[i].pltime = rpp->rp_pltime;
1378 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1379 			prl->prefix[i].expire = rpp->rp_expire;
1380 			prl->prefix[i].advrtrs = 0;
1381 			prl->prefix[i].origin = rpp->rp_origin;
1382 			i++;
1383 		}
1384 	      }
1385 		splx(s);
1386 
1387 		break;
1388 	case SIOCGIFINFO_IN6:
1389 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1390 			error = EINVAL;
1391 			break;
1392 		}
1393 		ndi->ndi = nd_ifinfo[ifp->if_index];
1394 		break;
1395 	case SIOCSIFINFO_FLAGS:
1396 		/* XXX: almost all other fields of ndi->ndi is unused */
1397 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1398 			error = EINVAL;
1399 			break;
1400 		}
1401 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1402 		break;
1403 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1404 		/* flush default router list */
1405 		/*
1406 		 * xxx sumikawa: should not delete route if default
1407 		 * route equals to the top of default router list
1408 		 */
1409 		bzero(&any, sizeof(any));
1410 		defrouter_delreq(&any, 0);
1411 		defrouter_select();
1412 		/* xxx sumikawa: flush prefix list */
1413 		break;
1414 	case SIOCSPFXFLUSH_IN6:
1415 	    {
1416 		/* flush all the prefix advertised by routers */
1417 		struct nd_prefix *pr, *next;
1418 
1419 		s = splsoftnet();
1420 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1421 			next = pr->ndpr_next;
1422 			if (!IN6_IS_ADDR_UNSPECIFIED(&pr->ndpr_addr))
1423 				in6_ifdel(pr->ndpr_ifp, &pr->ndpr_addr);
1424 			prelist_remove(pr);
1425 		}
1426 		splx(s);
1427 		break;
1428 	    }
1429 	case SIOCSRTRFLUSH_IN6:
1430 	    {
1431 		/* flush all the default routers */
1432 		struct nd_defrouter *dr, *next;
1433 
1434 		s = splsoftnet();
1435 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1436 			/*
1437 			 * The first entry of the list may be stored in
1438 			 * the routing table, so we'll delete it later.
1439 			 */
1440 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1441 				next = TAILQ_NEXT(dr, dr_entry);
1442 				defrtrlist_del(dr);
1443 			}
1444 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1445 		}
1446 		splx(s);
1447 		break;
1448 	    }
1449 	case SIOCGNBRINFO_IN6:
1450 	    {
1451 		struct llinfo_nd6 *ln;
1452 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1453 
1454 		/*
1455 		 * XXX: KAME specific hack for scoped addresses
1456 		 *      XXXX: for other scopes than link-local?
1457 		 */
1458 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1459 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1460 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1461 
1462 			if (*idp == 0)
1463 				*idp = htons(ifp->if_index);
1464 		}
1465 
1466 		s = splsoftnet();
1467 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1468 			error = EINVAL;
1469 			splx(s);
1470 			break;
1471 		}
1472 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1473 		nbi->state = ln->ln_state;
1474 		nbi->asked = ln->ln_asked;
1475 		nbi->isrouter = ln->ln_router;
1476 		nbi->expire = ln->ln_expire;
1477 		splx(s);
1478 
1479 		break;
1480 	    }
1481 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1482 		ndif->ifindex = nd6_defifindex;
1483 		break;
1484 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1485 		return(nd6_setdefaultiface(ndif->ifindex));
1486 		break;
1487 	}
1488 	return(error);
1489 }
1490 
1491 /*
1492  * Create neighbor cache entry and cache link-layer address,
1493  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1494  */
1495 struct rtentry *
1496 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1497 	struct ifnet *ifp;
1498 	struct in6_addr *from;
1499 	char *lladdr;
1500 	int lladdrlen;
1501 	int type;	/* ICMP6 type */
1502 	int code;	/* type dependent information */
1503 {
1504 	struct rtentry *rt = NULL;
1505 	struct llinfo_nd6 *ln = NULL;
1506 	int is_newentry;
1507 	struct sockaddr_dl *sdl = NULL;
1508 	int do_update;
1509 	int olladdr;
1510 	int llchange;
1511 	int newstate = 0;
1512 	long time_second = time.tv_sec;
1513 
1514 	if (!ifp)
1515 		panic("ifp == NULL in nd6_cache_lladdr");
1516 	if (!from)
1517 		panic("from == NULL in nd6_cache_lladdr");
1518 
1519 	/* nothing must be updated for unspecified address */
1520 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1521 		return NULL;
1522 
1523 	/*
1524 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1525 	 * the caller.
1526 	 *
1527 	 * XXX If the link does not have link-layer adderss, what should
1528 	 * we do? (ifp->if_addrlen == 0)
1529 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1530 	 * description on it in NS section (RFC 2461 7.2.3).
1531 	 */
1532 
1533 	rt = nd6_lookup(from, 0, ifp);
1534 	if (!rt) {
1535 #if 0
1536 		/* nothing must be done if there's no lladdr */
1537 		if (!lladdr || !lladdrlen)
1538 			return NULL;
1539 #endif
1540 
1541 		rt = nd6_lookup(from, 1, ifp);
1542 		is_newentry = 1;
1543 	} else {
1544 		/* do nothing if static ndp is set */
1545 		if (rt->rt_flags & RTF_STATIC)
1546 			return NULL;
1547 		is_newentry = 0;
1548 	}
1549 
1550 	if (!rt)
1551 		return NULL;
1552 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1553 fail:
1554 		(void)nd6_free(rt, 0);
1555 		return NULL;
1556 	}
1557 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1558 	if (!ln)
1559 		goto fail;
1560 	if (!rt->rt_gateway)
1561 		goto fail;
1562 	if (rt->rt_gateway->sa_family != AF_LINK)
1563 		goto fail;
1564 	sdl = SDL(rt->rt_gateway);
1565 
1566 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1567 	if (olladdr && lladdr) {
1568 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1569 			llchange = 1;
1570 		else
1571 			llchange = 0;
1572 	} else
1573 		llchange = 0;
1574 
1575 	/*
1576 	 * newentry olladdr  lladdr  llchange	(*=record)
1577 	 *	0	n	n	--	(1)
1578 	 *	0	y	n	--	(2)
1579 	 *	0	n	y	--	(3) * STALE
1580 	 *	0	y	y	n	(4) *
1581 	 *	0	y	y	y	(5) * STALE
1582 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1583 	 *	1	--	y	--	(7) * STALE
1584 	 */
1585 
1586 	if (lladdr) {		/* (3-5) and (7) */
1587 		/*
1588 		 * Record source link-layer address
1589 		 * XXX is it dependent to ifp->if_type?
1590 		 */
1591 		sdl->sdl_alen = ifp->if_addrlen;
1592 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1593 	}
1594 
1595 	if (!is_newentry) {
1596 		if ((!olladdr && lladdr)		/* (3) */
1597 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1598 			do_update = 1;
1599 			newstate = ND6_LLINFO_STALE;
1600 		} else					/* (1-2,4) */
1601 			do_update = 0;
1602 	} else {
1603 		do_update = 1;
1604 		if (!lladdr)				/* (6) */
1605 			newstate = ND6_LLINFO_NOSTATE;
1606 		else					/* (7) */
1607 			newstate = ND6_LLINFO_STALE;
1608 	}
1609 
1610 	if (do_update) {
1611 		/*
1612 		 * Update the state of the neighbor cache.
1613 		 */
1614 		ln->ln_state = newstate;
1615 
1616 		if (ln->ln_state == ND6_LLINFO_STALE) {
1617 			/*
1618 			 * XXX: since nd6_output() below will cause
1619 			 * state tansition to DELAY and reset the timer,
1620 			 * we must set the timer now, although it is actually
1621 			 * meaningless.
1622 			 */
1623 			ln->ln_expire = time_second + nd6_gctimer;
1624 
1625 			if (ln->ln_hold) {
1626 				/*
1627 				 * we assume ifp is not a p2p here, so just
1628 				 * set the 2nd argument as the 1st one.
1629 				 */
1630 				nd6_output(ifp, ifp, ln->ln_hold,
1631 					   (struct sockaddr_in6 *)rt_key(rt),
1632 					   rt);
1633 				ln->ln_hold = NULL;
1634 			}
1635 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1636 			/* probe right away */
1637 			ln->ln_expire = time_second;
1638 		}
1639 	}
1640 
1641 	/*
1642 	 * ICMP6 type dependent behavior.
1643 	 *
1644 	 * NS: clear IsRouter if new entry
1645 	 * RS: clear IsRouter
1646 	 * RA: set IsRouter if there's lladdr
1647 	 * redir: clear IsRouter if new entry
1648 	 *
1649 	 * RA case, (1):
1650 	 * The spec says that we must set IsRouter in the following cases:
1651 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1652 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1653 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1654 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1655 	 * neighbor cache, this is similar to (6).
1656 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1657 	 *
1658 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1659 	 *							D R
1660 	 *	0	n	n	--	(1)	c   ?     s
1661 	 *	0	y	n	--	(2)	c   s     s
1662 	 *	0	n	y	--	(3)	c   s     s
1663 	 *	0	y	y	n	(4)	c   s     s
1664 	 *	0	y	y	y	(5)	c   s     s
1665 	 *	1	--	n	--	(6) c	c 	c s
1666 	 *	1	--	y	--	(7) c	c   s	c s
1667 	 *
1668 	 *					(c=clear s=set)
1669 	 */
1670 	switch (type & 0xff) {
1671 	case ND_NEIGHBOR_SOLICIT:
1672 		/*
1673 		 * New entry must have is_router flag cleared.
1674 		 */
1675 		if (is_newentry)	/* (6-7) */
1676 			ln->ln_router = 0;
1677 		break;
1678 	case ND_REDIRECT:
1679 		/*
1680 		 * If the icmp is a redirect to a better router, always set the
1681 		 * is_router flag.  Otherwise, if the entry is newly created,
1682 		 * clear the flag.  [RFC 2461, sec 8.3]
1683 		 */
1684 		if (code == ND_REDIRECT_ROUTER)
1685 			ln->ln_router = 1;
1686 		else if (is_newentry) /* (6-7) */
1687 			ln->ln_router = 0;
1688 		break;
1689 	case ND_ROUTER_SOLICIT:
1690 		/*
1691 		 * is_router flag must always be cleared.
1692 		 */
1693 		ln->ln_router = 0;
1694 		break;
1695 	case ND_ROUTER_ADVERT:
1696 		/*
1697 		 * Mark an entry with lladdr as a router.
1698 		 */
1699 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1700 		 || (is_newentry && lladdr)) {			/* (7) */
1701 			ln->ln_router = 1;
1702 		}
1703 		break;
1704 	}
1705 
1706 	/*
1707 	 * When the link-layer address of a router changes, select the
1708 	 * best router again.  In particular, when the neighbor entry is newly
1709 	 * created, it might affect the selection policy.
1710 	 * Question: can we restrict the first condition to the "is_newentry"
1711 	 * case?
1712 	 * XXX: when we hear an RA from a new router with the link-layer
1713 	 * address option, defrouter_select() is called twice, since
1714 	 * defrtrlist_update called the function as well.  However, I believe
1715 	 * we can compromise the overhead, since it only happens the first
1716 	 * time.
1717 	 * XXX: although defrouter_select() should not have a bad effect
1718 	 * for those are not autoconfigured hosts, we explicitly avoid such
1719 	 * cases for safety.
1720 	 */
1721 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1722 		defrouter_select();
1723 
1724 	return rt;
1725 }
1726 
1727 static void
1728 nd6_slowtimo(ignored_arg)
1729     void *ignored_arg;
1730 {
1731 	int s = splsoftnet();
1732 	int i;
1733 	struct nd_ifinfo *nd6if;
1734 
1735 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1736 	    nd6_slowtimo, NULL);
1737 	for (i = 1; i < if_index + 1; i++) {
1738 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1739 			continue;
1740 		nd6if = &nd_ifinfo[i];
1741 		if (nd6if->basereachable && /* already initialized */
1742 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1743 			/*
1744 			 * Since reachable time rarely changes by router
1745 			 * advertisements, we SHOULD insure that a new random
1746 			 * value gets recomputed at least once every few hours.
1747 			 * (RFC 2461, 6.3.4)
1748 			 */
1749 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1750 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1751 		}
1752 	}
1753 	splx(s);
1754 }
1755 
1756 #define senderr(e) { error = (e); goto bad;}
1757 int
1758 nd6_output(ifp, origifp, m0, dst, rt0)
1759 	struct ifnet *ifp;
1760 	struct ifnet *origifp;
1761 	struct mbuf *m0;
1762 	struct sockaddr_in6 *dst;
1763 	struct rtentry *rt0;
1764 {
1765 	struct mbuf *m = m0;
1766 	struct rtentry *rt = rt0;
1767 	struct sockaddr_in6 *gw6 = NULL;
1768 	struct llinfo_nd6 *ln = NULL;
1769 	int error = 0;
1770 	long time_second = time.tv_sec;
1771 
1772 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1773 		goto sendpkt;
1774 
1775 	if (nd6_need_cache(ifp) == 0)
1776 		goto sendpkt;
1777 
1778 	/*
1779 	 * next hop determination.  This routine is derived from ether_outpout.
1780 	 */
1781 	if (rt) {
1782 		if ((rt->rt_flags & RTF_UP) == 0) {
1783 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1)) !=
1784 				NULL)
1785 			{
1786 				rt->rt_refcnt--;
1787 				if (rt->rt_ifp != ifp) {
1788 					/* XXX: loop care? */
1789 					return nd6_output(ifp, origifp, m0,
1790 							  dst, rt);
1791 				}
1792 			} else
1793 				senderr(EHOSTUNREACH);
1794 		}
1795 
1796 		if (rt->rt_flags & RTF_GATEWAY) {
1797 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1798 
1799 			/*
1800 			 * We skip link-layer address resolution and NUD
1801 			 * if the gateway is not a neighbor from ND point
1802 			 * of view, regardless the value of nd_ifinfo.flags.
1803 			 * The second condition is a bit tricky; we skip
1804 			 * if the gateway is our own address, which is
1805 			 * sometimes used to install a route to a p2p link.
1806 			 */
1807 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1808 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1809 				/*
1810 				 * We allow this kind of tricky route only
1811 				 * when the outgoing interface is p2p.
1812 				 * XXX: we may need a more generic rule here.
1813 				 */
1814 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1815 					senderr(EHOSTUNREACH);
1816 
1817 				goto sendpkt;
1818 			}
1819 
1820 			if (rt->rt_gwroute == 0)
1821 				goto lookup;
1822 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1823 				rtfree(rt); rt = rt0;
1824 			lookup:
1825 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1);
1826 				if ((rt = rt->rt_gwroute) == 0)
1827 					senderr(EHOSTUNREACH);
1828 				/* the "G" test below also prevents rt == rt0 */
1829 				if ((rt->rt_flags & RTF_GATEWAY) ||
1830 				    (rt->rt_ifp != ifp)) {
1831 					rt->rt_refcnt--;
1832 					rt0->rt_gwroute = 0;
1833 					senderr(EHOSTUNREACH);
1834 				}
1835 			}
1836 		}
1837 	}
1838 
1839 	/*
1840 	 * Address resolution or Neighbor Unreachability Detection
1841 	 * for the next hop.
1842 	 * At this point, the destination of the packet must be a unicast
1843 	 * or an anycast address(i.e. not a multicast).
1844 	 */
1845 
1846 	/* Look up the neighbor cache for the nexthop */
1847 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1848 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1849 	else {
1850 		/*
1851 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1852 		 * the condition below is not very efficient.  But we believe
1853 		 * it is tolerable, because this should be a rare case.
1854 		 */
1855 		if (nd6_is_addr_neighbor(dst, ifp) &&
1856 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1857 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1858 	}
1859 	if (!ln || !rt) {
1860 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1861 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1862 			log(LOG_DEBUG,
1863 			    "nd6_output: can't allocate llinfo for %s "
1864 			    "(ln=%p, rt=%p)\n",
1865 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1866 			senderr(EIO);	/* XXX: good error? */
1867 		}
1868 
1869 		goto sendpkt;	/* send anyway */
1870 	}
1871 
1872 	/* We don't have to do link-layer address resolution on a p2p link. */
1873 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1874 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1875 		ln->ln_state = ND6_LLINFO_STALE;
1876 		ln->ln_expire = time_second + nd6_gctimer;
1877 	}
1878 
1879 	/*
1880 	 * The first time we send a packet to a neighbor whose entry is
1881 	 * STALE, we have to change the state to DELAY and a sets a timer to
1882 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1883 	 * neighbor unreachability detection on expiration.
1884 	 * (RFC 2461 7.3.3)
1885 	 */
1886 	if (ln->ln_state == ND6_LLINFO_STALE) {
1887 		ln->ln_asked = 0;
1888 		ln->ln_state = ND6_LLINFO_DELAY;
1889 		ln->ln_expire = time_second + nd6_delay;
1890 	}
1891 
1892 	/*
1893 	 * If the neighbor cache entry has a state other than INCOMPLETE
1894 	 * (i.e. its link-layer address is already resolved), just
1895 	 * send the packet.
1896 	 */
1897 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1898 		goto sendpkt;
1899 
1900 	/*
1901 	 * There is a neighbor cache entry, but no ethernet address
1902 	 * response yet.  Replace the held mbuf (if any) with this
1903 	 * latest one.
1904 	 * This code conforms to the rate-limiting rule described in Section
1905 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1906 	 * an NS below.
1907 	 */
1908 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1909 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1910 	if (ln->ln_hold)
1911 		m_freem(ln->ln_hold);
1912 	ln->ln_hold = m;
1913 	if (ln->ln_expire) {
1914 		if (ln->ln_asked < nd6_mmaxtries &&
1915 		    ln->ln_expire < time_second) {
1916 			ln->ln_asked++;
1917 			ln->ln_expire = time_second +
1918 				nd_ifinfo[ifp->if_index].retrans / 1000;
1919 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1920 		}
1921 	}
1922 	return(0);
1923 
1924   sendpkt:
1925 
1926 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1927 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1928 					 rt));
1929 	}
1930 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1931 
1932   bad:
1933 	if (m)
1934 		m_freem(m);
1935 	return (error);
1936 }
1937 #undef senderr
1938 
1939 int
1940 nd6_need_cache(ifp)
1941 	struct ifnet *ifp;
1942 {
1943 	/*
1944 	 * XXX: we currently do not make neighbor cache on any interface
1945 	 * other than ARCnet, Ethernet, FDDI and GIF.
1946 	 *
1947 	 * RFC2893 says:
1948 	 * - unidirectional tunnels needs no ND
1949 	 */
1950 	switch (ifp->if_type) {
1951 	case IFT_ARCNET:
1952 	case IFT_ETHER:
1953 	case IFT_FDDI:
1954 	case IFT_IEEE1394:
1955 	case IFT_GIF:		/* XXX need more cases? */
1956 		return(1);
1957 	default:
1958 		return(0);
1959 	}
1960 }
1961 
1962 int
1963 nd6_storelladdr(ifp, rt, m, dst, desten)
1964 	struct ifnet *ifp;
1965 	struct rtentry *rt;
1966 	struct mbuf *m;
1967 	struct sockaddr *dst;
1968 	u_char *desten;
1969 {
1970 	struct sockaddr_dl *sdl;
1971 
1972 	if (m->m_flags & M_MCAST) {
1973 		switch (ifp->if_type) {
1974 		case IFT_ETHER:
1975 		case IFT_FDDI:
1976 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
1977 						 desten);
1978 			return(1);
1979 		case IFT_IEEE1394:
1980 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
1981 			return(1);
1982 		case IFT_ARCNET:
1983 			*desten = 0;
1984 			return(1);
1985 		default:
1986 			m_freem(m);
1987 			return(0);
1988 		}
1989 	}
1990 
1991 	if (rt == NULL) {
1992 		/* this could happen, if we could not allocate memory */
1993 		m_freem(m);
1994 		return(0);
1995 	}
1996 	if (rt->rt_gateway->sa_family != AF_LINK) {
1997 		printf("nd6_storelladdr: something odd happens\n");
1998 		m_freem(m);
1999 		return(0);
2000 	}
2001 	sdl = SDL(rt->rt_gateway);
2002 	if (sdl->sdl_alen == 0) {
2003 		/* this should be impossible, but we bark here for debugging */
2004 		printf("nd6_storelladdr: sdl_alen == 0, dst=%s, if=%s\n",
2005 		       ip6_sprintf(&SIN6(dst)->sin6_addr), if_name(ifp));
2006 		m_freem(m);
2007 		return(0);
2008 	}
2009 
2010 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2011 	return(1);
2012 }
2013