xref: /dflybsd-src/sys/netinet6/nd6.c (revision 17ea22213f86a5c5966c1e6bf8e95f022ebb92b9)
1 /*	$FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $	*/
2 /*	$DragonFly: src/sys/netinet6/nd6.c,v 1.14 2005/01/06 17:59:32 hsu Exp $	*/
3 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $	*/
4 
5 /*
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * XXX
36  * KAME 970409 note:
37  * BSD/OS version heavily modifies this code, related to llinfo.
38  * Since we don't have BSD/OS version of net/route.c in our hand,
39  * I left the code mostly as it was in 970310.  -- itojun
40  */
41 
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/callout.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/time.h>
53 #include <sys/kernel.h>
54 #include <sys/protosw.h>
55 #include <sys/errno.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/sysctl.h>
59 
60 #include <net/if.h>
61 #include <net/if_dl.h>
62 #include <net/if_types.h>
63 #include <net/if_atm.h>
64 #include <net/route.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/if_ether.h>
68 #include <netinet/if_fddi.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_prefix.h>
74 #include <netinet/icmp6.h>
75 
76 #include "use_loop.h"
77 
78 #include <net/net_osdep.h>
79 
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 #define SDL(s) ((struct sockaddr_dl *)s)
85 
86 /* timer values */
87 int	nd6_prune	= 1;	/* walk list every 1 seconds */
88 int	nd6_delay	= 5;	/* delay first probe time 5 second */
89 int	nd6_umaxtries	= 3;	/* maximum unicast query */
90 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
91 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
92 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
93 
94 /* preventing too many loops in ND option parsing */
95 int nd6_maxndopt = 10;	/* max # of ND options allowed */
96 
97 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
98 
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104 
105 /* for debugging? */
106 static int nd6_inuse, nd6_allocated;
107 
108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
109 static size_t nd_ifinfo_indexlim = 8;
110 struct nd_ifinfo *nd_ifinfo = NULL;
111 struct nd_drhead nd_defrouter;
112 struct nd_prhead nd_prefix = { 0 };
113 
114 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
115 static struct sockaddr_in6 all1_sa;
116 
117 static void nd6_slowtimo (void *);
118 static int regen_tmpaddr (struct in6_ifaddr *);
119 
120 struct callout nd6_slowtimo_ch;
121 struct callout nd6_timer_ch;
122 extern struct callout in6_tmpaddrtimer_ch;
123 
124 void
125 nd6_init(void)
126 {
127 	static int nd6_init_done = 0;
128 	int i;
129 
130 	if (nd6_init_done) {
131 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
132 		return;
133 	}
134 
135 	all1_sa.sin6_family = AF_INET6;
136 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
137 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
138 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
139 
140 	/* initialization of the default router list */
141 	TAILQ_INIT(&nd_defrouter);
142 
143 	nd6_init_done = 1;
144 
145 	/* start timer */
146 	callout_init(&nd6_slowtimo_ch);
147 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
148 	    nd6_slowtimo, NULL);
149 }
150 
151 void
152 nd6_ifattach(struct ifnet *ifp)
153 {
154 
155 	/*
156 	 * We have some arrays that should be indexed by if_index.
157 	 * since if_index will grow dynamically, they should grow too.
158 	 */
159 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
160 		size_t n;
161 		caddr_t q;
162 
163 		while (if_index >= nd_ifinfo_indexlim)
164 			nd_ifinfo_indexlim <<= 1;
165 
166 		/* grow nd_ifinfo */
167 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
168 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
169 		bzero(q, n);
170 		if (nd_ifinfo) {
171 			bcopy((caddr_t)nd_ifinfo, q, n/2);
172 			free((caddr_t)nd_ifinfo, M_IP6NDP);
173 		}
174 		nd_ifinfo = (struct nd_ifinfo *)q;
175 	}
176 
177 #define ND nd_ifinfo[ifp->if_index]
178 
179 	/*
180 	 * Don't initialize if called twice.
181 	 * XXX: to detect this, we should choose a member that is never set
182 	 * before initialization of the ND structure itself.  We formaly used
183 	 * the linkmtu member, which was not suitable because it could be
184 	 * initialized via "ifconfig mtu".
185 	 */
186 	if (ND.basereachable)
187 		return;
188 
189 	ND.linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
190 	ND.chlim = IPV6_DEFHLIM;
191 	ND.basereachable = REACHABLE_TIME;
192 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
193 	ND.retrans = RETRANS_TIMER;
194 	ND.receivedra = 0;
195 	/*
196 	 * Note that the default value of ip6_accept_rtadv is 0, which means
197 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
198 	 * here.
199 	 */
200 	ND.flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
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(struct ifnet *ifp)
211 {
212 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
213 	u_long oldmaxmtu = ndi->maxmtu;
214 	u_long oldlinkmtu = ndi->linkmtu;
215 
216 	switch (ifp->if_type) {
217 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
218 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
219 		break;
220 	case IFT_ETHER:
221 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
222 		break;
223 	case IFT_FDDI:
224 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
225 		break;
226 	case IFT_ATM:
227 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
228 		break;
229 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
230 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
231 		break;
232 #ifdef IFT_IEEE80211
233 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
234 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
235 		break;
236 #endif
237 	default:
238 		ndi->maxmtu = ifp->if_mtu;
239 		break;
240 	}
241 
242 	if (oldmaxmtu != ndi->maxmtu) {
243 		/*
244 		 * If the ND level MTU is not set yet, or if the maxmtu
245 		 * is reset to a smaller value than the ND level MTU,
246 		 * also reset the ND level MTU.
247 		 */
248 		if (ndi->linkmtu == 0 ||
249 		    ndi->maxmtu < ndi->linkmtu) {
250 			ndi->linkmtu = ndi->maxmtu;
251 			/* also adjust in6_maxmtu if necessary. */
252 			if (oldlinkmtu == 0) {
253 				/*
254 				 * XXX: the case analysis is grotty, but
255 				 * it is not efficient to call in6_setmaxmtu()
256 				 * here when we are during the initialization
257 				 * procedure.
258 				 */
259 				if (in6_maxmtu < ndi->linkmtu)
260 					in6_maxmtu = ndi->linkmtu;
261 			} else
262 				in6_setmaxmtu();
263 		}
264 	}
265 #undef MIN
266 }
267 
268 void
269 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
270 {
271 	bzero(ndopts, sizeof(*ndopts));
272 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
273 	ndopts->nd_opts_last
274 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
275 
276 	if (icmp6len == 0) {
277 		ndopts->nd_opts_done = 1;
278 		ndopts->nd_opts_search = NULL;
279 	}
280 }
281 
282 /*
283  * Take one ND option.
284  */
285 struct nd_opt_hdr *
286 nd6_option(union nd_opts *ndopts)
287 {
288 	struct nd_opt_hdr *nd_opt;
289 	int olen;
290 
291 	if (!ndopts)
292 		panic("ndopts == NULL in nd6_option");
293 	if (!ndopts->nd_opts_last)
294 		panic("uninitialized ndopts in nd6_option");
295 	if (!ndopts->nd_opts_search)
296 		return NULL;
297 	if (ndopts->nd_opts_done)
298 		return NULL;
299 
300 	nd_opt = ndopts->nd_opts_search;
301 
302 	/* make sure nd_opt_len is inside the buffer */
303 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
304 		bzero(ndopts, sizeof(*ndopts));
305 		return NULL;
306 	}
307 
308 	olen = nd_opt->nd_opt_len << 3;
309 	if (olen == 0) {
310 		/*
311 		 * Message validation requires that all included
312 		 * options have a length that is greater than zero.
313 		 */
314 		bzero(ndopts, sizeof(*ndopts));
315 		return NULL;
316 	}
317 
318 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
319 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
320 		/* option overruns the end of buffer, invalid */
321 		bzero(ndopts, sizeof(*ndopts));
322 		return NULL;
323 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
324 		/* reached the end of options chain */
325 		ndopts->nd_opts_done = 1;
326 		ndopts->nd_opts_search = NULL;
327 	}
328 	return nd_opt;
329 }
330 
331 /*
332  * Parse multiple ND options.
333  * This function is much easier to use, for ND routines that do not need
334  * multiple options of the same type.
335  */
336 int
337 nd6_options(union nd_opts *ndopts)
338 {
339 	struct nd_opt_hdr *nd_opt;
340 	int i = 0;
341 
342 	if (!ndopts)
343 		panic("ndopts == NULL in nd6_options");
344 	if (!ndopts->nd_opts_last)
345 		panic("uninitialized ndopts in nd6_options");
346 	if (!ndopts->nd_opts_search)
347 		return 0;
348 
349 	while (1) {
350 		nd_opt = nd6_option(ndopts);
351 		if (!nd_opt && !ndopts->nd_opts_last) {
352 			/*
353 			 * Message validation requires that all included
354 			 * options have a length that is greater than zero.
355 			 */
356 			icmp6stat.icp6s_nd_badopt++;
357 			bzero(ndopts, sizeof(*ndopts));
358 			return -1;
359 		}
360 
361 		if (!nd_opt)
362 			goto skip1;
363 
364 		switch (nd_opt->nd_opt_type) {
365 		case ND_OPT_SOURCE_LINKADDR:
366 		case ND_OPT_TARGET_LINKADDR:
367 		case ND_OPT_MTU:
368 		case ND_OPT_REDIRECTED_HEADER:
369 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
370 				nd6log((LOG_INFO,
371 				    "duplicated ND6 option found (type=%d)\n",
372 				    nd_opt->nd_opt_type));
373 				/* XXX bark? */
374 			} else {
375 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
376 					= nd_opt;
377 			}
378 			break;
379 		case ND_OPT_PREFIX_INFORMATION:
380 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
381 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
382 					= nd_opt;
383 			}
384 			ndopts->nd_opts_pi_end =
385 				(struct nd_opt_prefix_info *)nd_opt;
386 			break;
387 		default:
388 			/*
389 			 * Unknown options must be silently ignored,
390 			 * to accomodate future extension to the protocol.
391 			 */
392 			nd6log((LOG_DEBUG,
393 			    "nd6_options: unsupported option %d - "
394 			    "option ignored\n", nd_opt->nd_opt_type));
395 		}
396 
397 skip1:
398 		i++;
399 		if (i > nd6_maxndopt) {
400 			icmp6stat.icp6s_nd_toomanyopt++;
401 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
402 			break;
403 		}
404 
405 		if (ndopts->nd_opts_done)
406 			break;
407 	}
408 
409 	return 0;
410 }
411 
412 /*
413  * ND6 timer routine to expire default route list and prefix list
414  */
415 void
416 nd6_timer(void *ignored_arg)
417 {
418 	int s;
419 	struct llinfo_nd6 *ln;
420 	struct nd_defrouter *dr;
421 	struct nd_prefix *pr;
422 	struct ifnet *ifp;
423 	struct in6_ifaddr *ia6, *nia6;
424 	struct in6_addrlifetime *lt6;
425 
426 	s = splnet();
427 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
428 		      nd6_timer, NULL);
429 
430 	ln = llinfo_nd6.ln_next;
431 	while (ln && ln != &llinfo_nd6) {
432 		struct rtentry *rt;
433 		struct sockaddr_in6 *dst;
434 		struct llinfo_nd6 *next = ln->ln_next;
435 		/* XXX: used for the DELAY case only: */
436 		struct nd_ifinfo *ndi = NULL;
437 
438 		if ((rt = ln->ln_rt) == NULL) {
439 			ln = next;
440 			continue;
441 		}
442 		if ((ifp = rt->rt_ifp) == NULL) {
443 			ln = next;
444 			continue;
445 		}
446 		ndi = &nd_ifinfo[ifp->if_index];
447 		dst = (struct sockaddr_in6 *)rt_key(rt);
448 
449 		if (ln->ln_expire > time_second) {
450 			ln = next;
451 			continue;
452 		}
453 
454 		/* sanity check */
455 		if (!rt)
456 			panic("rt=0 in nd6_timer(ln=%p)", ln);
457 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
458 			panic("rt_llinfo(%p) is not equal to ln(%p)",
459 			      rt->rt_llinfo, ln);
460 		if (!dst)
461 			panic("dst=0 in nd6_timer(ln=%p)", ln);
462 
463 		switch (ln->ln_state) {
464 		case ND6_LLINFO_INCOMPLETE:
465 			if (ln->ln_asked < nd6_mmaxtries) {
466 				ln->ln_asked++;
467 				ln->ln_expire = time_second +
468 					nd_ifinfo[ifp->if_index].retrans / 1000;
469 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
470 					ln, 0);
471 			} else {
472 				struct mbuf *m = ln->ln_hold;
473 				if (m) {
474 					if (rt->rt_ifp) {
475 						/*
476 						 * Fake rcvif to make ICMP error
477 						 * more helpful in diagnosing
478 						 * for the receiver.
479 						 * XXX: should we consider
480 						 * older rcvif?
481 						 */
482 						m->m_pkthdr.rcvif = rt->rt_ifp;
483 					}
484 					icmp6_error(m, ICMP6_DST_UNREACH,
485 						    ICMP6_DST_UNREACH_ADDR, 0);
486 					ln->ln_hold = NULL;
487 				}
488 				next = nd6_free(rt);
489 			}
490 			break;
491 		case ND6_LLINFO_REACHABLE:
492 			if (ln->ln_expire) {
493 				ln->ln_state = ND6_LLINFO_STALE;
494 				ln->ln_expire = time_second + nd6_gctimer;
495 			}
496 			break;
497 
498 		case ND6_LLINFO_STALE:
499 			/* Garbage Collection(RFC 2461 5.3) */
500 			if (ln->ln_expire)
501 				next = nd6_free(rt);
502 			break;
503 
504 		case ND6_LLINFO_DELAY:
505 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
506 				/* We need NUD */
507 				ln->ln_asked = 1;
508 				ln->ln_state = ND6_LLINFO_PROBE;
509 				ln->ln_expire = time_second +
510 					ndi->retrans / 1000;
511 				nd6_ns_output(ifp, &dst->sin6_addr,
512 					      &dst->sin6_addr,
513 					      ln, 0);
514 			} else {
515 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
516 				ln->ln_expire = time_second + nd6_gctimer;
517 			}
518 			break;
519 		case ND6_LLINFO_PROBE:
520 			if (ln->ln_asked < nd6_umaxtries) {
521 				ln->ln_asked++;
522 				ln->ln_expire = time_second +
523 					nd_ifinfo[ifp->if_index].retrans / 1000;
524 				nd6_ns_output(ifp, &dst->sin6_addr,
525 					       &dst->sin6_addr, ln, 0);
526 			} else {
527 				next = nd6_free(rt);
528 			}
529 			break;
530 		}
531 		ln = next;
532 	}
533 
534 	/* expire default router list */
535 	dr = TAILQ_FIRST(&nd_defrouter);
536 	while (dr) {
537 		if (dr->expire && dr->expire < time_second) {
538 			struct nd_defrouter *t;
539 			t = TAILQ_NEXT(dr, dr_entry);
540 			defrtrlist_del(dr);
541 			dr = t;
542 		} else {
543 			dr = TAILQ_NEXT(dr, dr_entry);
544 		}
545 	}
546 
547 	/*
548 	 * expire interface addresses.
549 	 * in the past the loop was inside prefix expiry processing.
550 	 * However, from a stricter speci-confrmance standpoint, we should
551 	 * rather separate address lifetimes and prefix lifetimes.
552 	 */
553   addrloop:
554 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
555 		nia6 = ia6->ia_next;
556 		/* check address lifetime */
557 		lt6 = &ia6->ia6_lifetime;
558 		if (IFA6_IS_INVALID(ia6)) {
559 			int regen = 0;
560 
561 			/*
562 			 * If the expiring address is temporary, try
563 			 * regenerating a new one.  This would be useful when
564 			 * we suspended a laptop PC, then turned it on after a
565 			 * period that could invalidate all temporary
566 			 * addresses.  Although we may have to restart the
567 			 * loop (see below), it must be after purging the
568 			 * address.  Otherwise, we'd see an infinite loop of
569 			 * regeneration.
570 			 */
571 			if (ip6_use_tempaddr &&
572 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
573 				if (regen_tmpaddr(ia6) == 0)
574 					regen = 1;
575 			}
576 
577 			in6_purgeaddr(&ia6->ia_ifa);
578 
579 			if (regen)
580 				goto addrloop; /* XXX: see below */
581 		}
582 		if (IFA6_IS_DEPRECATED(ia6)) {
583 			int oldflags = ia6->ia6_flags;
584 
585 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
586 
587 			/*
588 			 * If a temporary address has just become deprecated,
589 			 * regenerate a new one if possible.
590 			 */
591 			if (ip6_use_tempaddr &&
592 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
593 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
594 
595 				if (regen_tmpaddr(ia6) == 0) {
596 					/*
597 					 * A new temporary address is
598 					 * generated.
599 					 * XXX: this means the address chain
600 					 * has changed while we are still in
601 					 * the loop.  Although the change
602 					 * would not cause disaster (because
603 					 * it's not a deletion, but an
604 					 * addition,) we'd rather restart the
605 					 * loop just for safety.  Or does this
606 					 * significantly reduce performance??
607 					 */
608 					goto addrloop;
609 				}
610 			}
611 		} else {
612 			/*
613 			 * A new RA might have made a deprecated address
614 			 * preferred.
615 			 */
616 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
617 		}
618 	}
619 
620 	/* expire prefix list */
621 	pr = nd_prefix.lh_first;
622 	while (pr) {
623 		/*
624 		 * check prefix lifetime.
625 		 * since pltime is just for autoconf, pltime processing for
626 		 * prefix is not necessary.
627 		 */
628 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
629 			struct nd_prefix *t;
630 			t = pr->ndpr_next;
631 
632 			/*
633 			 * address expiration and prefix expiration are
634 			 * separate.  NEVER perform in6_purgeaddr here.
635 			 */
636 
637 			prelist_remove(pr);
638 			pr = t;
639 		} else
640 			pr = pr->ndpr_next;
641 	}
642 	splx(s);
643 }
644 
645 static int
646 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
647 					 address */
648 {
649 	struct ifaddr *ifa;
650 	struct ifnet *ifp;
651 	struct in6_ifaddr *public_ifa6 = NULL;
652 
653 	ifp = ia6->ia_ifa.ifa_ifp;
654 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_list) {
655 		struct in6_ifaddr *it6;
656 
657 		if (ifa->ifa_addr->sa_family != AF_INET6)
658 			continue;
659 
660 		it6 = (struct in6_ifaddr *)ifa;
661 
662 		/* ignore no autoconf addresses. */
663 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
664 			continue;
665 
666 		/* ignore autoconf addresses with different prefixes. */
667 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
668 			continue;
669 
670 		/*
671 		 * Now we are looking at an autoconf address with the same
672 		 * prefix as ours.  If the address is temporary and is still
673 		 * preferred, do not create another one.  It would be rare, but
674 		 * could happen, for example, when we resume a laptop PC after
675 		 * a long period.
676 		 */
677 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
678 		    !IFA6_IS_DEPRECATED(it6)) {
679 			public_ifa6 = NULL;
680 			break;
681 		}
682 
683 		/*
684 		 * This is a public autoconf address that has the same prefix
685 		 * as ours.  If it is preferred, keep it.  We can't break the
686 		 * loop here, because there may be a still-preferred temporary
687 		 * address with the prefix.
688 		 */
689 		if (!IFA6_IS_DEPRECATED(it6))
690 		    public_ifa6 = it6;
691 	}
692 
693 	if (public_ifa6 != NULL) {
694 		int e;
695 
696 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
697 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
698 			    " tmp addr,errno=%d\n", e);
699 			return(-1);
700 		}
701 		return(0);
702 	}
703 
704 	return(-1);
705 }
706 
707 /*
708  * Nuke neighbor cache/prefix/default router management table, right before
709  * ifp goes away.
710  */
711 void
712 nd6_purge(struct ifnet *ifp)
713 {
714 	struct llinfo_nd6 *ln, *nln;
715 	struct nd_defrouter *dr, *ndr, drany;
716 	struct nd_prefix *pr, *npr;
717 
718 	/* Nuke default router list entries toward ifp */
719 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
720 		/*
721 		 * The first entry of the list may be stored in
722 		 * the routing table, so we'll delete it later.
723 		 */
724 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
725 			ndr = TAILQ_NEXT(dr, dr_entry);
726 			if (dr->ifp == ifp)
727 				defrtrlist_del(dr);
728 		}
729 		dr = TAILQ_FIRST(&nd_defrouter);
730 		if (dr->ifp == ifp)
731 			defrtrlist_del(dr);
732 	}
733 
734 	/* Nuke prefix list entries toward ifp */
735 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
736 		npr = pr->ndpr_next;
737 		if (pr->ndpr_ifp == ifp) {
738 			/*
739 			 * Previously, pr->ndpr_addr is removed as well,
740 			 * but I strongly believe we don't have to do it.
741 			 * nd6_purge() is only called from in6_ifdetach(),
742 			 * which removes all the associated interface addresses
743 			 * by itself.
744 			 * (jinmei@kame.net 20010129)
745 			 */
746 			prelist_remove(pr);
747 		}
748 	}
749 
750 	/* cancel default outgoing interface setting */
751 	if (nd6_defifindex == ifp->if_index)
752 		nd6_setdefaultiface(0);
753 
754 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
755 		/* refresh default router list */
756 		bzero(&drany, sizeof(drany));
757 		defrouter_delreq(&drany, 0);
758 		defrouter_select();
759 	}
760 
761 	/*
762 	 * Nuke neighbor cache entries for the ifp.
763 	 * Note that rt->rt_ifp may not be the same as ifp,
764 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
765 	 * nd6_rtrequest(), and ip6_input().
766 	 */
767 	ln = llinfo_nd6.ln_next;
768 	while (ln && ln != &llinfo_nd6) {
769 		struct rtentry *rt;
770 		struct sockaddr_dl *sdl;
771 
772 		nln = ln->ln_next;
773 		rt = ln->ln_rt;
774 		if (rt && rt->rt_gateway &&
775 		    rt->rt_gateway->sa_family == AF_LINK) {
776 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
777 			if (sdl->sdl_index == ifp->if_index)
778 				nln = nd6_free(rt);
779 		}
780 		ln = nln;
781 	}
782 }
783 
784 struct rtentry *
785 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
786 {
787 	struct rtentry *rt;
788 	struct sockaddr_in6 sin6;
789 
790 	bzero(&sin6, sizeof(sin6));
791 	sin6.sin6_len = sizeof(struct sockaddr_in6);
792 	sin6.sin6_family = AF_INET6;
793 	sin6.sin6_addr = *addr6;
794 #ifdef SCOPEDROUTING
795 	sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
796 #endif
797 	if (create)
798 		rt = rtlookup((struct sockaddr *)&sin6);
799 	else
800 		rt = rtpurelookup((struct sockaddr *)&sin6);
801 	if (rt && !(rt->rt_flags & RTF_LLINFO)) {
802 		/*
803 		 * This is the case for the default route.
804 		 * If we want to create a neighbor cache for the address, we
805 		 * should free the route for the destination and allocate an
806 		 * interface route.
807 		 */
808 		if (create) {
809 			--rt->rt_refcnt;
810 			rt = NULL;
811 		}
812 	}
813 	if (!rt) {
814 		if (create && ifp) {
815 			int e;
816 
817 			/*
818 			 * If no route is available and create is set,
819 			 * we allocate a host route for the destination
820 			 * and treat it like an interface route.
821 			 * This hack is necessary for a neighbor which can't
822 			 * be covered by our own prefix.
823 			 */
824 			struct ifaddr *ifa =
825 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
826 			if (ifa == NULL)
827 				return(NULL);
828 
829 			/*
830 			 * Create a new route.  RTF_LLINFO is necessary
831 			 * to create a Neighbor Cache entry for the
832 			 * destination in nd6_rtrequest which will be
833 			 * called in rtrequest via ifa->ifa_rtrequest.
834 			 */
835 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
836 					   ifa->ifa_addr,
837 					   (struct sockaddr *)&all1_sa,
838 					   (ifa->ifa_flags |
839 					    RTF_HOST | RTF_LLINFO) &
840 					   ~RTF_CLONING,
841 					   &rt)) != 0)
842 				log(LOG_ERR,
843 				    "nd6_lookup: failed to add route for a "
844 				    "neighbor(%s), errno=%d\n",
845 				    ip6_sprintf(addr6), e);
846 			if (rt == NULL)
847 				return(NULL);
848 			if (rt->rt_llinfo) {
849 				struct llinfo_nd6 *ln =
850 					(struct llinfo_nd6 *)rt->rt_llinfo;
851 				ln->ln_state = ND6_LLINFO_NOSTATE;
852 			}
853 		} else
854 			return(NULL);
855 	}
856 	rt->rt_refcnt--;
857 	/*
858 	 * Validation for the entry.
859 	 * Note that the check for rt_llinfo is necessary because a cloned
860 	 * route from a parent route that has the L flag (e.g. the default
861 	 * route to a p2p interface) may have the flag, too, while the
862 	 * destination is not actually a neighbor.
863 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
864 	 *      it might be the loopback interface if the entry is for our
865 	 *      own address on a non-loopback interface. Instead, we should
866 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
867 	 *      interface.
868 	 */
869 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
870 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
871 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
872 		if (create) {
873 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
874 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
875 			/* xxx more logs... kazu */
876 		}
877 		return(NULL);
878 	}
879 	return(rt);
880 }
881 
882 /*
883  * Detect if a given IPv6 address identifies a neighbor on a given link.
884  * XXX: should take care of the destination of a p2p link?
885  */
886 int
887 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
888 {
889 	struct ifaddr *ifa;
890 	int i;
891 
892 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
893 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
894 
895 	/*
896 	 * A link-local address is always a neighbor.
897 	 * XXX: we should use the sin6_scope_id field rather than the embedded
898 	 * interface index.
899 	 */
900 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
901 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
902 		return(1);
903 
904 	/*
905 	 * If the address matches one of our addresses,
906 	 * it should be a neighbor.
907 	 */
908 	TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
909 		if (ifa->ifa_addr->sa_family != AF_INET6)
910 			next: continue;
911 
912 		for (i = 0; i < 4; i++) {
913 			if ((IFADDR6(ifa).s6_addr32[i] ^
914 			     addr->sin6_addr.s6_addr32[i]) &
915 			    IFMASK6(ifa).s6_addr32[i])
916 				goto next;
917 		}
918 		return(1);
919 	}
920 
921 	/*
922 	 * Even if the address matches none of our addresses, it might be
923 	 * in the neighbor cache.
924 	 */
925 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
926 		return(1);
927 
928 	return(0);
929 #undef IFADDR6
930 #undef IFMASK6
931 }
932 
933 /*
934  * Free an nd6 llinfo entry.
935  */
936 struct llinfo_nd6 *
937 nd6_free(struct rtentry *rt)
938 {
939 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
940 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
941 	struct nd_defrouter *dr;
942 
943 	/*
944 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
945 	 * even though it is not harmful, it was not really necessary.
946 	 */
947 
948 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
949 		int s;
950 		s = splnet();
951 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
952 				      rt->rt_ifp);
953 
954 		if (ln->ln_router || dr) {
955 			/*
956 			 * rt6_flush must be called whether or not the neighbor
957 			 * is in the Default Router List.
958 			 * See a corresponding comment in nd6_na_input().
959 			 */
960 			rt6_flush(&in6, rt->rt_ifp);
961 		}
962 
963 		if (dr) {
964 			/*
965 			 * Unreachablity of a router might affect the default
966 			 * router selection and on-link detection of advertised
967 			 * prefixes.
968 			 */
969 
970 			/*
971 			 * Temporarily fake the state to choose a new default
972 			 * router and to perform on-link determination of
973 			 * prefixes correctly.
974 			 * Below the state will be set correctly,
975 			 * or the entry itself will be deleted.
976 			 */
977 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
978 
979 			/*
980 			 * Since defrouter_select() does not affect the
981 			 * on-link determination and MIP6 needs the check
982 			 * before the default router selection, we perform
983 			 * the check now.
984 			 */
985 			pfxlist_onlink_check();
986 
987 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
988 				/*
989 				 * It is used as the current default router,
990 				 * so we have to move it to the end of the
991 				 * list and choose a new one.
992 				 * XXX: it is not very efficient if this is
993 				 *      the only router.
994 				 */
995 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
996 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
997 
998 				defrouter_select();
999 			}
1000 		}
1001 		splx(s);
1002 	}
1003 
1004 	/*
1005 	 * Before deleting the entry, remember the next entry as the
1006 	 * return value.  We need this because pfxlist_onlink_check() above
1007 	 * might have freed other entries (particularly the old next entry) as
1008 	 * a side effect (XXX).
1009 	 */
1010 	next = ln->ln_next;
1011 
1012 	/*
1013 	 * Detach the route from the routing tree and the list of neighbor
1014 	 * caches, and disable the route entry not to be used in already
1015 	 * cached routes.
1016 	 */
1017 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1018 		  rt_mask(rt), 0, (struct rtentry **)0);
1019 
1020 	return(next);
1021 }
1022 
1023 /*
1024  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1025  *
1026  * XXX cost-effective metods?
1027  */
1028 void
1029 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1030 {
1031 	struct llinfo_nd6 *ln;
1032 
1033 	/*
1034 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1035 	 * routing table by supplied "dst6".
1036 	 */
1037 	if (!rt) {
1038 		if (!dst6)
1039 			return;
1040 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1041 			return;
1042 	}
1043 
1044 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1045 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1046 	    !rt->rt_llinfo || !rt->rt_gateway ||
1047 	    rt->rt_gateway->sa_family != AF_LINK) {
1048 		/* This is not a host route. */
1049 		return;
1050 	}
1051 
1052 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1053 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1054 		return;
1055 
1056 	/*
1057 	 * if we get upper-layer reachability confirmation many times,
1058 	 * it is possible we have false information.
1059 	 */
1060 	if (!force) {
1061 		ln->ln_byhint++;
1062 		if (ln->ln_byhint > nd6_maxnudhint)
1063 			return;
1064 	}
1065 
1066 	ln->ln_state = ND6_LLINFO_REACHABLE;
1067 	if (ln->ln_expire)
1068 		ln->ln_expire = time_second +
1069 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
1070 }
1071 
1072 void
1073 nd6_rtrequest(int req, struct rtentry *rt,
1074 	      struct rt_addrinfo *info) /* xxx unused */
1075 {
1076 	struct sockaddr *gate = rt->rt_gateway;
1077 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1078 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1079 	struct ifnet *ifp = rt->rt_ifp;
1080 	struct ifaddr *ifa;
1081 
1082 	if ((rt->rt_flags & RTF_GATEWAY))
1083 		return;
1084 
1085 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1086 		/*
1087 		 * This is probably an interface direct route for a link
1088 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1089 		 * We do not need special treatment below for such a route.
1090 		 * Moreover, the RTF_LLINFO flag which would be set below
1091 		 * would annoy the ndp(8) command.
1092 		 */
1093 		return;
1094 	}
1095 
1096 	if (req == RTM_RESOLVE &&
1097 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1098 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1099 		/*
1100 		 * FreeBSD and BSD/OS often make a cloned host route based
1101 		 * on a less-specific route (e.g. the default route).
1102 		 * If the less specific route does not have a "gateway"
1103 		 * (this is the case when the route just goes to a p2p or an
1104 		 * stf interface), we'll mistakenly make a neighbor cache for
1105 		 * the host route, and will see strange neighbor solicitation
1106 		 * for the corresponding destination.  In order to avoid the
1107 		 * confusion, we check if the destination of the route is
1108 		 * a neighbor in terms of neighbor discovery, and stop the
1109 		 * process if not.  Additionally, we remove the LLINFO flag
1110 		 * so that ndp(8) will not try to get the neighbor information
1111 		 * of the destination.
1112 		 */
1113 		rt->rt_flags &= ~RTF_LLINFO;
1114 		return;
1115 	}
1116 
1117 	switch (req) {
1118 	case RTM_ADD:
1119 		/*
1120 		 * There is no backward compatibility :)
1121 		 *
1122 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1123 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1124 		 *	   rt->rt_flags |= RTF_CLONING;
1125 		 */
1126 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1127 			/*
1128 			 * Case 1: This route should come from
1129 			 * a route to interface.  RTF_LLINFO flag is set
1130 			 * for a host route whose destination should be
1131 			 * treated as on-link.
1132 			 */
1133 			rt_setgate(rt, rt_key(rt),
1134 				   (struct sockaddr *)&null_sdl);
1135 			gate = rt->rt_gateway;
1136 			SDL(gate)->sdl_type = ifp->if_type;
1137 			SDL(gate)->sdl_index = ifp->if_index;
1138 			if (ln)
1139 				ln->ln_expire = time_second;
1140 #if 1
1141 			if (ln && ln->ln_expire == 0) {
1142 				/* kludge for desktops */
1143 #if 0
1144 				printf("nd6_rtequest: time.tv_sec is zero; "
1145 				       "treat it as 1\n");
1146 #endif
1147 				ln->ln_expire = 1;
1148 			}
1149 #endif
1150 			if ((rt->rt_flags & RTF_CLONING))
1151 				break;
1152 		}
1153 		/*
1154 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1155 		 * We don't do that here since llinfo is not ready yet.
1156 		 *
1157 		 * There are also couple of other things to be discussed:
1158 		 * - unsolicited NA code needs improvement beforehand
1159 		 * - RFC2461 says we MAY send multicast unsolicited NA
1160 		 *   (7.2.6 paragraph 4), however, it also says that we
1161 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1162 		 *   we don't have anything like it right now.
1163 		 *   note that the mechanism needs a mutual agreement
1164 		 *   between proxies, which means that we need to implement
1165 		 *   a new protocol, or a new kludge.
1166 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1167 		 *   we need to check ip6forwarding before sending it.
1168 		 *   (or should we allow proxy ND configuration only for
1169 		 *   routers?  there's no mention about proxy ND from hosts)
1170 		 */
1171 #if 0
1172 		/* XXX it does not work */
1173 		if (rt->rt_flags & RTF_ANNOUNCE)
1174 			nd6_na_output(ifp,
1175 			      &SIN6(rt_key(rt))->sin6_addr,
1176 			      &SIN6(rt_key(rt))->sin6_addr,
1177 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1178 			      1, NULL);
1179 #endif
1180 		/* FALLTHROUGH */
1181 	case RTM_RESOLVE:
1182 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1183 			/*
1184 			 * Address resolution isn't necessary for a point to
1185 			 * point link, so we can skip this test for a p2p link.
1186 			 */
1187 			if (gate->sa_family != AF_LINK ||
1188 			    gate->sa_len < sizeof(null_sdl)) {
1189 				log(LOG_DEBUG,
1190 				    "nd6_rtrequest: bad gateway value: %s\n",
1191 				    if_name(ifp));
1192 				break;
1193 			}
1194 			SDL(gate)->sdl_type = ifp->if_type;
1195 			SDL(gate)->sdl_index = ifp->if_index;
1196 		}
1197 		if (ln != NULL)
1198 			break;	/* This happens on a route change */
1199 		/*
1200 		 * Case 2: This route may come from cloning, or a manual route
1201 		 * add with a LL address.
1202 		 */
1203 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1204 		rt->rt_llinfo = (caddr_t)ln;
1205 		if (!ln) {
1206 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1207 			break;
1208 		}
1209 		nd6_inuse++;
1210 		nd6_allocated++;
1211 		bzero(ln, sizeof(*ln));
1212 		ln->ln_rt = rt;
1213 		/* this is required for "ndp" command. - shin */
1214 		if (req == RTM_ADD) {
1215 		        /*
1216 			 * gate should have some valid AF_LINK entry,
1217 			 * and ln->ln_expire should have some lifetime
1218 			 * which is specified by ndp command.
1219 			 */
1220 			ln->ln_state = ND6_LLINFO_REACHABLE;
1221 			ln->ln_byhint = 0;
1222 		} else {
1223 		        /*
1224 			 * When req == RTM_RESOLVE, rt is created and
1225 			 * initialized in rtrequest(), so rt_expire is 0.
1226 			 */
1227 			ln->ln_state = ND6_LLINFO_NOSTATE;
1228 			ln->ln_expire = time_second;
1229 		}
1230 		rt->rt_flags |= RTF_LLINFO;
1231 		ln->ln_next = llinfo_nd6.ln_next;
1232 		llinfo_nd6.ln_next = ln;
1233 		ln->ln_prev = &llinfo_nd6;
1234 		ln->ln_next->ln_prev = ln;
1235 
1236 		/*
1237 		 * check if rt_key(rt) is one of my address assigned
1238 		 * to the interface.
1239 		 */
1240 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1241 					  &SIN6(rt_key(rt))->sin6_addr);
1242 		if (ifa) {
1243 			caddr_t macp = nd6_ifptomac(ifp);
1244 			ln->ln_expire = 0;
1245 			ln->ln_state = ND6_LLINFO_REACHABLE;
1246 			ln->ln_byhint = 0;
1247 			if (macp) {
1248 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1249 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1250 			}
1251 			if (nd6_useloopback) {
1252 				rt->rt_ifp = &loif[0];	/* XXX */
1253 				/*
1254 				 * Make sure rt_ifa be equal to the ifaddr
1255 				 * corresponding to the address.
1256 				 * We need this because when we refer
1257 				 * rt_ifa->ia6_flags in ip6_input, we assume
1258 				 * that the rt_ifa points to the address instead
1259 				 * of the loopback address.
1260 				 */
1261 				if (ifa != rt->rt_ifa) {
1262 					IFAFREE(rt->rt_ifa);
1263 					IFAREF(ifa);
1264 					rt->rt_ifa = ifa;
1265 				}
1266 			}
1267 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1268 			ln->ln_expire = 0;
1269 			ln->ln_state = ND6_LLINFO_REACHABLE;
1270 			ln->ln_byhint = 0;
1271 
1272 			/* join solicited node multicast for proxy ND */
1273 			if (ifp->if_flags & IFF_MULTICAST) {
1274 				struct in6_addr llsol;
1275 				int error;
1276 
1277 				llsol = SIN6(rt_key(rt))->sin6_addr;
1278 				llsol.s6_addr16[0] = htons(0xff02);
1279 				llsol.s6_addr16[1] = htons(ifp->if_index);
1280 				llsol.s6_addr32[1] = 0;
1281 				llsol.s6_addr32[2] = htonl(1);
1282 				llsol.s6_addr8[12] = 0xff;
1283 
1284 				if (!in6_addmulti(&llsol, ifp, &error)) {
1285 					nd6log((LOG_ERR, "%s: failed to join "
1286 					    "%s (errno=%d)\n", if_name(ifp),
1287 					    ip6_sprintf(&llsol), error));
1288 				}
1289 			}
1290 		}
1291 		break;
1292 
1293 	case RTM_DELETE:
1294 		if (!ln)
1295 			break;
1296 		/* leave from solicited node multicast for proxy ND */
1297 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1298 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1299 			struct in6_addr llsol;
1300 			struct in6_multi *in6m;
1301 
1302 			llsol = SIN6(rt_key(rt))->sin6_addr;
1303 			llsol.s6_addr16[0] = htons(0xff02);
1304 			llsol.s6_addr16[1] = htons(ifp->if_index);
1305 			llsol.s6_addr32[1] = 0;
1306 			llsol.s6_addr32[2] = htonl(1);
1307 			llsol.s6_addr8[12] = 0xff;
1308 
1309 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1310 			if (in6m)
1311 				in6_delmulti(in6m);
1312 		}
1313 		nd6_inuse--;
1314 		ln->ln_next->ln_prev = ln->ln_prev;
1315 		ln->ln_prev->ln_next = ln->ln_next;
1316 		ln->ln_prev = NULL;
1317 		rt->rt_llinfo = 0;
1318 		rt->rt_flags &= ~RTF_LLINFO;
1319 		if (ln->ln_hold)
1320 			m_freem(ln->ln_hold);
1321 		Free((caddr_t)ln);
1322 	}
1323 }
1324 
1325 int
1326 nd6_ioctl(u_long cmd, caddr_t	data, struct ifnet *ifp)
1327 {
1328 	struct in6_drlist *drl = (struct in6_drlist *)data;
1329 	struct in6_prlist *prl = (struct in6_prlist *)data;
1330 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1331 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1332 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1333 	struct nd_defrouter *dr, any;
1334 	struct nd_prefix *pr;
1335 	struct rtentry *rt;
1336 	int i = 0, error = 0;
1337 	int s;
1338 
1339 	switch (cmd) {
1340 	case SIOCGDRLST_IN6:
1341 		/*
1342 		 * obsolete API, use sysctl under net.inet6.icmp6
1343 		 */
1344 		bzero(drl, sizeof(*drl));
1345 		s = splnet();
1346 		dr = TAILQ_FIRST(&nd_defrouter);
1347 		while (dr && i < DRLSTSIZ) {
1348 			drl->defrouter[i].rtaddr = dr->rtaddr;
1349 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1350 				/* XXX: need to this hack for KAME stack */
1351 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1352 			} else
1353 				log(LOG_ERR,
1354 				    "default router list contains a "
1355 				    "non-linklocal address(%s)\n",
1356 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1357 
1358 			drl->defrouter[i].flags = dr->flags;
1359 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1360 			drl->defrouter[i].expire = dr->expire;
1361 			drl->defrouter[i].if_index = dr->ifp->if_index;
1362 			i++;
1363 			dr = TAILQ_NEXT(dr, dr_entry);
1364 		}
1365 		splx(s);
1366 		break;
1367 	case SIOCGPRLST_IN6:
1368 		/*
1369 		 * obsolete API, use sysctl under net.inet6.icmp6
1370 		 */
1371 		/*
1372 		 * XXX meaning of fields, especialy "raflags", is very
1373 		 * differnet between RA prefix list and RR/static prefix list.
1374 		 * how about separating ioctls into two?
1375 		 */
1376 		bzero(prl, sizeof(*prl));
1377 		s = splnet();
1378 		pr = nd_prefix.lh_first;
1379 		while (pr && i < PRLSTSIZ) {
1380 			struct nd_pfxrouter *pfr;
1381 			int j;
1382 
1383 			(void)in6_embedscope(&prl->prefix[i].prefix,
1384 			    &pr->ndpr_prefix, NULL, NULL);
1385 			prl->prefix[i].raflags = pr->ndpr_raf;
1386 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1387 			prl->prefix[i].vltime = pr->ndpr_vltime;
1388 			prl->prefix[i].pltime = pr->ndpr_pltime;
1389 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1390 			prl->prefix[i].expire = pr->ndpr_expire;
1391 
1392 			pfr = pr->ndpr_advrtrs.lh_first;
1393 			j = 0;
1394 			while (pfr) {
1395 				if (j < DRLSTSIZ) {
1396 #define RTRADDR prl->prefix[i].advrtr[j]
1397 					RTRADDR = pfr->router->rtaddr;
1398 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1399 						/* XXX: hack for KAME */
1400 						RTRADDR.s6_addr16[1] = 0;
1401 					} else
1402 						log(LOG_ERR,
1403 						    "a router(%s) advertises "
1404 						    "a prefix with "
1405 						    "non-link local address\n",
1406 						    ip6_sprintf(&RTRADDR));
1407 #undef RTRADDR
1408 				}
1409 				j++;
1410 				pfr = pfr->pfr_next;
1411 			}
1412 			prl->prefix[i].advrtrs = j;
1413 			prl->prefix[i].origin = PR_ORIG_RA;
1414 
1415 			i++;
1416 			pr = pr->ndpr_next;
1417 		}
1418 	      {
1419 		struct rr_prefix *rpp;
1420 
1421 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1422 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1423 			if (i >= PRLSTSIZ)
1424 				break;
1425 			(void)in6_embedscope(&prl->prefix[i].prefix,
1426 			    &pr->ndpr_prefix, NULL, NULL);
1427 			prl->prefix[i].raflags = rpp->rp_raf;
1428 			prl->prefix[i].prefixlen = rpp->rp_plen;
1429 			prl->prefix[i].vltime = rpp->rp_vltime;
1430 			prl->prefix[i].pltime = rpp->rp_pltime;
1431 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1432 			prl->prefix[i].expire = rpp->rp_expire;
1433 			prl->prefix[i].advrtrs = 0;
1434 			prl->prefix[i].origin = rpp->rp_origin;
1435 			i++;
1436 		}
1437 	      }
1438 		splx(s);
1439 
1440 		break;
1441 	case OSIOCGIFINFO_IN6:
1442 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1443 			error = EINVAL;
1444 			break;
1445 		}
1446 		ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1447 		ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1448 		ndi->ndi.basereachable =
1449 		    nd_ifinfo[ifp->if_index].basereachable;
1450 		ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1451 		ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1452 		ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1453 		ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1454 		ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1455 		ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1456 		break;
1457 	case SIOCGIFINFO_IN6:
1458 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1459 			error = EINVAL;
1460 			break;
1461 		}
1462 		ndi->ndi = nd_ifinfo[ifp->if_index];
1463 		break;
1464 	case SIOCSIFINFO_FLAGS:
1465 		/* XXX: almost all other fields of ndi->ndi is unused */
1466 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1467 			error = EINVAL;
1468 			break;
1469 		}
1470 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1471 		break;
1472 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1473 		/* flush default router list */
1474 		/*
1475 		 * xxx sumikawa: should not delete route if default
1476 		 * route equals to the top of default router list
1477 		 */
1478 		bzero(&any, sizeof(any));
1479 		defrouter_delreq(&any, 0);
1480 		defrouter_select();
1481 		/* xxx sumikawa: flush prefix list */
1482 		break;
1483 	case SIOCSPFXFLUSH_IN6:
1484 	    {
1485 		/* flush all the prefix advertised by routers */
1486 		struct nd_prefix *pr, *next;
1487 
1488 		s = splnet();
1489 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1490 			struct in6_ifaddr *ia, *ia_next;
1491 
1492 			next = pr->ndpr_next;
1493 
1494 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1495 				continue; /* XXX */
1496 
1497 			/* do we really have to remove addresses as well? */
1498 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1499 				/* ia might be removed.  keep the next ptr. */
1500 				ia_next = ia->ia_next;
1501 
1502 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1503 					continue;
1504 
1505 				if (ia->ia6_ndpr == pr)
1506 					in6_purgeaddr(&ia->ia_ifa);
1507 			}
1508 			prelist_remove(pr);
1509 		}
1510 		splx(s);
1511 		break;
1512 	    }
1513 	case SIOCSRTRFLUSH_IN6:
1514 	    {
1515 		/* flush all the default routers */
1516 		struct nd_defrouter *dr, *next;
1517 
1518 		s = splnet();
1519 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1520 			/*
1521 			 * The first entry of the list may be stored in
1522 			 * the routing table, so we'll delete it later.
1523 			 */
1524 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1525 				next = TAILQ_NEXT(dr, dr_entry);
1526 				defrtrlist_del(dr);
1527 			}
1528 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1529 		}
1530 		splx(s);
1531 		break;
1532 	    }
1533 	case SIOCGNBRINFO_IN6:
1534 	    {
1535 		struct llinfo_nd6 *ln;
1536 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1537 
1538 		/*
1539 		 * XXX: KAME specific hack for scoped addresses
1540 		 *      XXXX: for other scopes than link-local?
1541 		 */
1542 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1543 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1544 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1545 
1546 			if (*idp == 0)
1547 				*idp = htons(ifp->if_index);
1548 		}
1549 
1550 		s = splnet();
1551 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1552 			error = EINVAL;
1553 			splx(s);
1554 			break;
1555 		}
1556 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1557 		nbi->state = ln->ln_state;
1558 		nbi->asked = ln->ln_asked;
1559 		nbi->isrouter = ln->ln_router;
1560 		nbi->expire = ln->ln_expire;
1561 		splx(s);
1562 
1563 		break;
1564 	    }
1565 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1566 		ndif->ifindex = nd6_defifindex;
1567 		break;
1568 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1569 		return(nd6_setdefaultiface(ndif->ifindex));
1570 		break;
1571 	}
1572 	return(error);
1573 }
1574 
1575 /*
1576  * Create neighbor cache entry and cache link-layer address,
1577  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1578  */
1579 struct rtentry *
1580 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1581 		 int lladdrlen,
1582 		 int type,	/* ICMP6 type */
1583 		 int code	/* type dependent information */)
1584 {
1585 	struct rtentry *rt = NULL;
1586 	struct llinfo_nd6 *ln = NULL;
1587 	int is_newentry;
1588 	struct sockaddr_dl *sdl = NULL;
1589 	int do_update;
1590 	int olladdr;
1591 	int llchange;
1592 	int newstate = 0;
1593 
1594 	if (!ifp)
1595 		panic("ifp == NULL in nd6_cache_lladdr");
1596 	if (!from)
1597 		panic("from == NULL in nd6_cache_lladdr");
1598 
1599 	/* nothing must be updated for unspecified address */
1600 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1601 		return NULL;
1602 
1603 	/*
1604 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1605 	 * the caller.
1606 	 *
1607 	 * XXX If the link does not have link-layer adderss, what should
1608 	 * we do? (ifp->if_addrlen == 0)
1609 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1610 	 * description on it in NS section (RFC 2461 7.2.3).
1611 	 */
1612 
1613 	rt = nd6_lookup(from, 0, ifp);
1614 	if (!rt) {
1615 #if 0
1616 		/* nothing must be done if there's no lladdr */
1617 		if (!lladdr || !lladdrlen)
1618 			return NULL;
1619 #endif
1620 
1621 		rt = nd6_lookup(from, 1, ifp);
1622 		is_newentry = 1;
1623 	} else {
1624 		/* do nothing if static ndp is set */
1625 		if (rt->rt_flags & RTF_STATIC)
1626 			return NULL;
1627 		is_newentry = 0;
1628 	}
1629 
1630 	if (!rt)
1631 		return NULL;
1632 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1633 fail:
1634 		(void)nd6_free(rt);
1635 		return NULL;
1636 	}
1637 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1638 	if (!ln)
1639 		goto fail;
1640 	if (!rt->rt_gateway)
1641 		goto fail;
1642 	if (rt->rt_gateway->sa_family != AF_LINK)
1643 		goto fail;
1644 	sdl = SDL(rt->rt_gateway);
1645 
1646 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1647 	if (olladdr && lladdr) {
1648 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1649 			llchange = 1;
1650 		else
1651 			llchange = 0;
1652 	} else
1653 		llchange = 0;
1654 
1655 	/*
1656 	 * newentry olladdr  lladdr  llchange	(*=record)
1657 	 *	0	n	n	--	(1)
1658 	 *	0	y	n	--	(2)
1659 	 *	0	n	y	--	(3) * STALE
1660 	 *	0	y	y	n	(4) *
1661 	 *	0	y	y	y	(5) * STALE
1662 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1663 	 *	1	--	y	--	(7) * STALE
1664 	 */
1665 
1666 	if (lladdr) {		/* (3-5) and (7) */
1667 		/*
1668 		 * Record source link-layer address
1669 		 * XXX is it dependent to ifp->if_type?
1670 		 */
1671 		sdl->sdl_alen = ifp->if_addrlen;
1672 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1673 	}
1674 
1675 	if (!is_newentry) {
1676 		if ((!olladdr && lladdr)		/* (3) */
1677 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1678 			do_update = 1;
1679 			newstate = ND6_LLINFO_STALE;
1680 		} else					/* (1-2,4) */
1681 			do_update = 0;
1682 	} else {
1683 		do_update = 1;
1684 		if (!lladdr)				/* (6) */
1685 			newstate = ND6_LLINFO_NOSTATE;
1686 		else					/* (7) */
1687 			newstate = ND6_LLINFO_STALE;
1688 	}
1689 
1690 	if (do_update) {
1691 		/*
1692 		 * Update the state of the neighbor cache.
1693 		 */
1694 		ln->ln_state = newstate;
1695 
1696 		if (ln->ln_state == ND6_LLINFO_STALE) {
1697 			/*
1698 			 * XXX: since nd6_output() below will cause
1699 			 * state tansition to DELAY and reset the timer,
1700 			 * we must set the timer now, although it is actually
1701 			 * meaningless.
1702 			 */
1703 			ln->ln_expire = time_second + nd6_gctimer;
1704 
1705 			if (ln->ln_hold) {
1706 				/*
1707 				 * we assume ifp is not a p2p here, so just
1708 				 * set the 2nd argument as the 1st one.
1709 				 */
1710 				nd6_output(ifp, ifp, ln->ln_hold,
1711 					   (struct sockaddr_in6 *)rt_key(rt),
1712 					   rt);
1713 				ln->ln_hold = NULL;
1714 			}
1715 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1716 			/* probe right away */
1717 			ln->ln_expire = time_second;
1718 		}
1719 	}
1720 
1721 	/*
1722 	 * ICMP6 type dependent behavior.
1723 	 *
1724 	 * NS: clear IsRouter if new entry
1725 	 * RS: clear IsRouter
1726 	 * RA: set IsRouter if there's lladdr
1727 	 * redir: clear IsRouter if new entry
1728 	 *
1729 	 * RA case, (1):
1730 	 * The spec says that we must set IsRouter in the following cases:
1731 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1732 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1733 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1734 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1735 	 * neighbor cache, this is similar to (6).
1736 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1737 	 *
1738 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1739 	 *							D R
1740 	 *	0	n	n	--	(1)	c   ?     s
1741 	 *	0	y	n	--	(2)	c   s     s
1742 	 *	0	n	y	--	(3)	c   s     s
1743 	 *	0	y	y	n	(4)	c   s     s
1744 	 *	0	y	y	y	(5)	c   s     s
1745 	 *	1	--	n	--	(6) c	c 	c s
1746 	 *	1	--	y	--	(7) c	c   s	c s
1747 	 *
1748 	 *					(c=clear s=set)
1749 	 */
1750 	switch (type & 0xff) {
1751 	case ND_NEIGHBOR_SOLICIT:
1752 		/*
1753 		 * New entry must have is_router flag cleared.
1754 		 */
1755 		if (is_newentry)	/* (6-7) */
1756 			ln->ln_router = 0;
1757 		break;
1758 	case ND_REDIRECT:
1759 		/*
1760 		 * If the icmp is a redirect to a better router, always set the
1761 		 * is_router flag. Otherwise, if the entry is newly created,
1762 		 * clear the flag. [RFC 2461, sec 8.3]
1763 		 */
1764 		if (code == ND_REDIRECT_ROUTER)
1765 			ln->ln_router = 1;
1766 		else if (is_newentry) /* (6-7) */
1767 			ln->ln_router = 0;
1768 		break;
1769 	case ND_ROUTER_SOLICIT:
1770 		/*
1771 		 * is_router flag must always be cleared.
1772 		 */
1773 		ln->ln_router = 0;
1774 		break;
1775 	case ND_ROUTER_ADVERT:
1776 		/*
1777 		 * Mark an entry with lladdr as a router.
1778 		 */
1779 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1780 		 || (is_newentry && lladdr)) {			/* (7) */
1781 			ln->ln_router = 1;
1782 		}
1783 		break;
1784 	}
1785 
1786 	/*
1787 	 * When the link-layer address of a router changes, select the
1788 	 * best router again.  In particular, when the neighbor entry is newly
1789 	 * created, it might affect the selection policy.
1790 	 * Question: can we restrict the first condition to the "is_newentry"
1791 	 * case?
1792 	 * XXX: when we hear an RA from a new router with the link-layer
1793 	 * address option, defrouter_select() is called twice, since
1794 	 * defrtrlist_update called the function as well.  However, I believe
1795 	 * we can compromise the overhead, since it only happens the first
1796 	 * time.
1797 	 * XXX: although defrouter_select() should not have a bad effect
1798 	 * for those are not autoconfigured hosts, we explicitly avoid such
1799 	 * cases for safety.
1800 	 */
1801 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1802 		defrouter_select();
1803 
1804 	return rt;
1805 }
1806 
1807 static void
1808 nd6_slowtimo(void *ignored_arg)
1809 {
1810 	int s = splnet();
1811 	int i;
1812 	struct nd_ifinfo *nd6if;
1813 
1814 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1815 	    nd6_slowtimo, NULL);
1816 	for (i = 1; i < if_index + 1; i++) {
1817 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1818 			continue;
1819 		nd6if = &nd_ifinfo[i];
1820 		if (nd6if->basereachable && /* already initialized */
1821 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1822 			/*
1823 			 * Since reachable time rarely changes by router
1824 			 * advertisements, we SHOULD insure that a new random
1825 			 * value gets recomputed at least once every few hours.
1826 			 * (RFC 2461, 6.3.4)
1827 			 */
1828 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1829 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1830 		}
1831 	}
1832 	splx(s);
1833 }
1834 
1835 #define senderr(e) { error = (e); goto bad;}
1836 int
1837 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1838 	   struct sockaddr_in6 *dst, struct rtentry *rt0)
1839 {
1840 	struct mbuf *m = m0;
1841 	struct rtentry *rt = rt0;
1842 	struct sockaddr_in6 *gw6 = NULL;
1843 	struct llinfo_nd6 *ln = NULL;
1844 	int error = 0;
1845 
1846 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1847 		goto sendpkt;
1848 
1849 	if (nd6_need_cache(ifp) == 0)
1850 		goto sendpkt;
1851 
1852 	/*
1853 	 * next hop determination.  This routine is derived from ether_outpout.
1854 	 */
1855 	if (rt) {
1856 		if (!(rt->rt_flags & RTF_UP)) {
1857 			if ((rt0 = rt = rtlookup((struct sockaddr *)dst))) {
1858 				rt->rt_refcnt--;
1859 				if (rt->rt_ifp != ifp) {
1860 					/* XXX: loop care? */
1861 					return nd6_output(ifp, origifp, m0,
1862 							  dst, rt);
1863 				}
1864 			} else
1865 				senderr(EHOSTUNREACH);
1866 		}
1867 
1868 		if (rt->rt_flags & RTF_GATEWAY) {
1869 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1870 
1871 			/*
1872 			 * We skip link-layer address resolution and NUD
1873 			 * if the gateway is not a neighbor from ND point
1874 			 * of view, regardless of the value of nd_ifinfo.flags.
1875 			 * The second condition is a bit tricky; we skip
1876 			 * if the gateway is our own address, which is
1877 			 * sometimes used to install a route to a p2p link.
1878 			 */
1879 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1880 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1881 				/*
1882 				 * We allow this kind of tricky route only
1883 				 * when the outgoing interface is p2p.
1884 				 * XXX: we may need a more generic rule here.
1885 				 */
1886 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1887 					senderr(EHOSTUNREACH);
1888 
1889 				goto sendpkt;
1890 			}
1891 
1892 			if (rt->rt_gwroute == NULL)
1893 				goto lookup;
1894 			if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1895 				rtfree(rt->rt_gwroute);
1896 lookup:				rt->rt_gwroute = rtlookup(rt->rt_gateway);
1897 				if (rt->rt_gwroute == NULL)
1898 					senderr(EHOSTUNREACH);
1899 			}
1900 		}
1901 	}
1902 
1903 	/*
1904 	 * Address resolution or Neighbor Unreachability Detection
1905 	 * for the next hop.
1906 	 * At this point, the destination of the packet must be a unicast
1907 	 * or an anycast address(i.e. not a multicast).
1908 	 */
1909 
1910 	/* Look up the neighbor cache for the nexthop */
1911 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1912 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1913 	else {
1914 		/*
1915 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1916 		 * the condition below is not very efficient.  But we believe
1917 		 * it is tolerable, because this should be a rare case.
1918 		 */
1919 		if (nd6_is_addr_neighbor(dst, ifp) &&
1920 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1921 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1922 	}
1923 	if (!ln || !rt) {
1924 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1925 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1926 			log(LOG_DEBUG,
1927 			    "nd6_output: can't allocate llinfo for %s "
1928 			    "(ln=%p, rt=%p)\n",
1929 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1930 			senderr(EIO);	/* XXX: good error? */
1931 		}
1932 
1933 		goto sendpkt;	/* send anyway */
1934 	}
1935 
1936 	/* We don't have to do link-layer address resolution on a p2p link. */
1937 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1938 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1939 		ln->ln_state = ND6_LLINFO_STALE;
1940 		ln->ln_expire = time_second + nd6_gctimer;
1941 	}
1942 
1943 	/*
1944 	 * The first time we send a packet to a neighbor whose entry is
1945 	 * STALE, we have to change the state to DELAY and a sets a timer to
1946 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1947 	 * neighbor unreachability detection on expiration.
1948 	 * (RFC 2461 7.3.3)
1949 	 */
1950 	if (ln->ln_state == ND6_LLINFO_STALE) {
1951 		ln->ln_asked = 0;
1952 		ln->ln_state = ND6_LLINFO_DELAY;
1953 		ln->ln_expire = time_second + nd6_delay;
1954 	}
1955 
1956 	/*
1957 	 * If the neighbor cache entry has a state other than INCOMPLETE
1958 	 * (i.e. its link-layer address is already resolved), just
1959 	 * send the packet.
1960 	 */
1961 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1962 		goto sendpkt;
1963 
1964 	/*
1965 	 * There is a neighbor cache entry, but no ethernet address
1966 	 * response yet.  Replace the held mbuf (if any) with this
1967 	 * latest one.
1968 	 *
1969 	 * This code conforms to the rate-limiting rule described in Section
1970 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1971 	 * an NS below.
1972 	 */
1973 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1974 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1975 	if (ln->ln_hold)
1976 		m_freem(ln->ln_hold);
1977 	ln->ln_hold = m;
1978 	if (ln->ln_expire) {
1979 		if (ln->ln_asked < nd6_mmaxtries &&
1980 		    ln->ln_expire < time_second) {
1981 			ln->ln_asked++;
1982 			ln->ln_expire = time_second +
1983 				nd_ifinfo[ifp->if_index].retrans / 1000;
1984 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1985 		}
1986 	}
1987 	return(0);
1988 
1989   sendpkt:
1990 
1991 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1992 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1993 					 rt));
1994 	}
1995 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1996 
1997   bad:
1998 	if (m)
1999 		m_freem(m);
2000 	return (error);
2001 }
2002 #undef senderr
2003 
2004 int
2005 nd6_need_cache(struct ifnet *ifp)
2006 {
2007 	/*
2008 	 * XXX: we currently do not make neighbor cache on any interface
2009 	 * other than ARCnet, Ethernet, FDDI and GIF.
2010 	 *
2011 	 * RFC2893 says:
2012 	 * - unidirectional tunnels needs no ND
2013 	 */
2014 	switch (ifp->if_type) {
2015 	case IFT_ARCNET:
2016 	case IFT_ETHER:
2017 	case IFT_FDDI:
2018 	case IFT_IEEE1394:
2019 #ifdef IFT_L2VLAN
2020 	case IFT_L2VLAN:
2021 #endif
2022 #ifdef IFT_IEEE80211
2023 	case IFT_IEEE80211:
2024 #endif
2025 	case IFT_GIF:		/* XXX need more cases? */
2026 		return(1);
2027 	default:
2028 		return(0);
2029 	}
2030 }
2031 
2032 int
2033 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
2034 		struct sockaddr *dst, u_char *desten)
2035 {
2036 	struct sockaddr_dl *sdl;
2037 	struct rtentry *rt;
2038 
2039 
2040 	if (m->m_flags & M_MCAST) {
2041 		switch (ifp->if_type) {
2042 		case IFT_ETHER:
2043 		case IFT_FDDI:
2044 #ifdef IFT_L2VLAN
2045 	case IFT_L2VLAN:
2046 #endif
2047 #ifdef IFT_IEEE80211
2048 		case IFT_IEEE80211:
2049 #endif
2050 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2051 						 desten);
2052 			return(1);
2053 		case IFT_IEEE1394:
2054 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2055 			return(1);
2056 		case IFT_ARCNET:
2057 			*desten = 0;
2058 			return(1);
2059 		default:
2060 			m_freem(m);
2061 			return(0);
2062 		}
2063 	}
2064 	if (rt0 == NULL) {
2065 		/* this could happen, if we could not allocate memory */
2066 		m_freem(m);
2067 		return(0);
2068 	}
2069 	if (rt_llroute(dst, rt0, &rt) != 0) {
2070 		m_freem(m);
2071 		return (0);
2072 	}
2073 	if (rt->rt_gateway->sa_family != AF_LINK) {
2074 		printf("nd6_storelladdr: something odd happens\n");
2075 		m_freem(m);
2076 		return(0);
2077 	}
2078 	sdl = SDL(rt->rt_gateway);
2079 	if (sdl->sdl_alen == 0) {
2080 		/* this should be impossible, but we bark here for debugging */
2081 		printf("nd6_storelladdr: sdl_alen == 0\n");
2082 		m_freem(m);
2083 		return(0);
2084 	}
2085 
2086 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2087 	return(1);
2088 }
2089 
2090 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2091 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2092 #ifdef SYSCTL_DECL
2093 SYSCTL_DECL(_net_inet6_icmp6);
2094 #endif
2095 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2096 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2097 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2098 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2099 
2100 static int
2101 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2102 {
2103 	int error;
2104 	char buf[1024];
2105 	struct in6_defrouter *d, *de;
2106 	struct nd_defrouter *dr;
2107 
2108 	if (req->newptr)
2109 		return EPERM;
2110 	error = 0;
2111 
2112 	for (dr = TAILQ_FIRST(&nd_defrouter);
2113 	     dr;
2114 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2115 		d = (struct in6_defrouter *)buf;
2116 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2117 
2118 		if (d + 1 <= de) {
2119 			bzero(d, sizeof(*d));
2120 			d->rtaddr.sin6_family = AF_INET6;
2121 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2122 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2123 			    dr->ifp) != 0)
2124 				log(LOG_ERR,
2125 				    "scope error in "
2126 				    "default router list (%s)\n",
2127 				    ip6_sprintf(&dr->rtaddr));
2128 			d->flags = dr->flags;
2129 			d->rtlifetime = dr->rtlifetime;
2130 			d->expire = dr->expire;
2131 			d->if_index = dr->ifp->if_index;
2132 		} else
2133 			panic("buffer too short");
2134 
2135 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2136 		if (error)
2137 			break;
2138 	}
2139 	return error;
2140 }
2141 
2142 static int
2143 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2144 {
2145 	int error;
2146 	char buf[1024];
2147 	struct in6_prefix *p, *pe;
2148 	struct nd_prefix *pr;
2149 
2150 	if (req->newptr)
2151 		return EPERM;
2152 	error = 0;
2153 
2154 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2155 		u_short advrtrs;
2156 		size_t advance;
2157 		struct sockaddr_in6 *sin6, *s6;
2158 		struct nd_pfxrouter *pfr;
2159 
2160 		p = (struct in6_prefix *)buf;
2161 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2162 
2163 		if (p + 1 <= pe) {
2164 			bzero(p, sizeof(*p));
2165 			sin6 = (struct sockaddr_in6 *)(p + 1);
2166 
2167 			p->prefix = pr->ndpr_prefix;
2168 			if (in6_recoverscope(&p->prefix,
2169 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2170 				log(LOG_ERR,
2171 				    "scope error in prefix list (%s)\n",
2172 				    ip6_sprintf(&p->prefix.sin6_addr));
2173 			p->raflags = pr->ndpr_raf;
2174 			p->prefixlen = pr->ndpr_plen;
2175 			p->vltime = pr->ndpr_vltime;
2176 			p->pltime = pr->ndpr_pltime;
2177 			p->if_index = pr->ndpr_ifp->if_index;
2178 			p->expire = pr->ndpr_expire;
2179 			p->refcnt = pr->ndpr_refcnt;
2180 			p->flags = pr->ndpr_stateflags;
2181 			p->origin = PR_ORIG_RA;
2182 			advrtrs = 0;
2183 			for (pfr = pr->ndpr_advrtrs.lh_first;
2184 			     pfr;
2185 			     pfr = pfr->pfr_next) {
2186 				if ((void *)&sin6[advrtrs + 1] >
2187 				    (void *)pe) {
2188 					advrtrs++;
2189 					continue;
2190 				}
2191 				s6 = &sin6[advrtrs];
2192 				bzero(s6, sizeof(*s6));
2193 				s6->sin6_family = AF_INET6;
2194 				s6->sin6_len = sizeof(*sin6);
2195 				if (in6_recoverscope(s6,
2196 				    &pfr->router->rtaddr,
2197 				    pfr->router->ifp) != 0)
2198 					log(LOG_ERR,
2199 					    "scope error in "
2200 					    "prefix list (%s)\n",
2201 					    ip6_sprintf(&pfr->router->rtaddr));
2202 				advrtrs++;
2203 			}
2204 			p->advrtrs = advrtrs;
2205 		} else
2206 			panic("buffer too short");
2207 
2208 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2209 		error = SYSCTL_OUT(req, buf, advance);
2210 		if (error)
2211 			break;
2212 	}
2213 	return error;
2214 }
2215