xref: /dragonfly/sys/netinet6/nd6.c (revision b58f1e66)
1 /*	$FreeBSD: src/sys/netinet6/nd6.c,v 1.2.2.15 2003/05/06 06:46:58 suz Exp $	*/
2 /*	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
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 "opt_inet.h"
42 #include "opt_inet6.h"
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/syslog.h>
56 #include <sys/queue.h>
57 #include <sys/sysctl.h>
58 #include <sys/thread2.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 <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/in6_prefix.h>
73 #include <netinet/icmp6.h>
74 
75 #include <net/net_osdep.h>
76 
77 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
78 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
79 
80 #define SIN6(s) ((struct sockaddr_in6 *)s)
81 #define SDL(s) ((struct sockaddr_dl *)s)
82 
83 /* timer values */
84 int	nd6_prune	= 1;	/* walk list every 1 seconds */
85 int	nd6_delay	= 5;	/* delay first probe time 5 second */
86 int	nd6_umaxtries	= 3;	/* maximum unicast query */
87 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
88 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
89 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
90 
91 /* preventing too many loops in ND option parsing */
92 int nd6_maxndopt = 10;	/* max # of ND options allowed */
93 
94 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
95 
96 #ifdef ND6_DEBUG
97 int nd6_debug = 1;
98 #else
99 int nd6_debug = 0;
100 #endif
101 
102 /* for debugging? */
103 static int nd6_inuse, nd6_allocated;
104 
105 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
106 struct nd_drhead nd_defrouter;
107 struct nd_prhead nd_prefix = { 0 };
108 
109 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
110 static struct sockaddr_in6 all1_sa;
111 
112 static void nd6_setmtu0 (struct ifnet *, struct nd_ifinfo *);
113 static void nd6_slowtimo (void *);
114 static int regen_tmpaddr (struct in6_ifaddr *);
115 
116 struct callout nd6_slowtimo_ch;
117 struct callout nd6_timer_ch;
118 extern struct callout in6_tmpaddrtimer_ch;
119 
120 void
121 nd6_init(void)
122 {
123 	static int nd6_init_done = 0;
124 	int i;
125 
126 	if (nd6_init_done) {
127 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
128 		return;
129 	}
130 
131 	all1_sa.sin6_family = AF_INET6;
132 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
133 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
134 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
135 
136 	/* initialization of the default router list */
137 	TAILQ_INIT(&nd_defrouter);
138 
139 	nd6_init_done = 1;
140 
141 	/* start timer */
142 	callout_init(&nd6_slowtimo_ch);
143 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
144 	    nd6_slowtimo, NULL);
145 }
146 
147 struct nd_ifinfo *
148 nd6_ifattach(struct ifnet *ifp)
149 {
150 	struct nd_ifinfo *nd;
151 
152 	nd = (struct nd_ifinfo *)kmalloc(sizeof(*nd), M_IP6NDP,
153 	    M_WAITOK | M_ZERO);
154 
155 	nd->initialized = 1;
156 
157 	nd->linkmtu = ifindex2ifnet[ifp->if_index]->if_mtu;
158 	nd->chlim = IPV6_DEFHLIM;
159 	nd->basereachable = REACHABLE_TIME;
160 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
161 	nd->retrans = RETRANS_TIMER;
162 	nd->receivedra = 0;
163 
164 	/*
165 	 * Note that the default value of ip6_accept_rtadv is 0, which means
166 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
167 	 * here.
168 	 */
169 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
170 
171 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
172 	nd6_setmtu0(ifp, nd);
173 	return nd;
174 }
175 
176 void
177 nd6_ifdetach(struct nd_ifinfo *nd)
178 {
179 	kfree(nd, M_IP6NDP);
180 }
181 
182 /*
183  * Reset ND level link MTU. This function is called when the physical MTU
184  * changes, which means we might have to adjust the ND level MTU.
185  */
186 void
187 nd6_setmtu(struct ifnet *ifp)
188 {
189 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
190 }
191 
192 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
193 void
194 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
195 {
196 	u_long oldmaxmtu;
197 	u_long oldlinkmtu;
198 
199 	oldmaxmtu = ndi->maxmtu;
200 	oldlinkmtu = ndi->linkmtu;
201 
202 	switch (ifp->if_type) {
203 	case IFT_ETHER:
204 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
205 		break;
206 	case IFT_ATM:
207 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
208 		break;
209 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
210 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
211 		break;
212 #ifdef IFT_IEEE80211
213 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
214 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
215 		break;
216 #endif
217 	default:
218 		ndi->maxmtu = ifp->if_mtu;
219 		break;
220 	}
221 
222 	if (oldmaxmtu != ndi->maxmtu) {
223 		/*
224 		 * If the ND level MTU is not set yet, or if the maxmtu
225 		 * is reset to a smaller value than the ND level MTU,
226 		 * also reset the ND level MTU.
227 		 */
228 		if (ndi->linkmtu == 0 ||
229 		    ndi->maxmtu < ndi->linkmtu) {
230 			ndi->linkmtu = ndi->maxmtu;
231 			/* also adjust in6_maxmtu if necessary. */
232 			if (oldlinkmtu == 0) {
233 				/*
234 				 * XXX: the case analysis is grotty, but
235 				 * it is not efficient to call in6_setmaxmtu()
236 				 * here when we are during the initialization
237 				 * procedure.
238 				 */
239 				if (in6_maxmtu < ndi->linkmtu)
240 					in6_maxmtu = ndi->linkmtu;
241 			} else
242 				in6_setmaxmtu();
243 		}
244 	}
245 #undef MIN
246 }
247 
248 void
249 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
250 {
251 	bzero(ndopts, sizeof(*ndopts));
252 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
253 	ndopts->nd_opts_last
254 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
255 
256 	if (icmp6len == 0) {
257 		ndopts->nd_opts_done = 1;
258 		ndopts->nd_opts_search = NULL;
259 	}
260 }
261 
262 /*
263  * Take one ND option.
264  */
265 struct nd_opt_hdr *
266 nd6_option(union nd_opts *ndopts)
267 {
268 	struct nd_opt_hdr *nd_opt;
269 	int olen;
270 
271 	if (!ndopts)
272 		panic("ndopts == NULL in nd6_option");
273 	if (!ndopts->nd_opts_last)
274 		panic("uninitialized ndopts in nd6_option");
275 	if (!ndopts->nd_opts_search)
276 		return NULL;
277 	if (ndopts->nd_opts_done)
278 		return NULL;
279 
280 	nd_opt = ndopts->nd_opts_search;
281 
282 	/* make sure nd_opt_len is inside the buffer */
283 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
284 		bzero(ndopts, sizeof(*ndopts));
285 		return NULL;
286 	}
287 
288 	olen = nd_opt->nd_opt_len << 3;
289 	if (olen == 0) {
290 		/*
291 		 * Message validation requires that all included
292 		 * options have a length that is greater than zero.
293 		 */
294 		bzero(ndopts, sizeof(*ndopts));
295 		return NULL;
296 	}
297 
298 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
299 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
300 		/* option overruns the end of buffer, invalid */
301 		bzero(ndopts, sizeof(*ndopts));
302 		return NULL;
303 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
304 		/* reached the end of options chain */
305 		ndopts->nd_opts_done = 1;
306 		ndopts->nd_opts_search = NULL;
307 	}
308 	return nd_opt;
309 }
310 
311 /*
312  * Parse multiple ND options.
313  * This function is much easier to use, for ND routines that do not need
314  * multiple options of the same type.
315  */
316 int
317 nd6_options(union nd_opts *ndopts)
318 {
319 	struct nd_opt_hdr *nd_opt;
320 	int i = 0;
321 
322 	if (!ndopts)
323 		panic("ndopts == NULL in nd6_options");
324 	if (!ndopts->nd_opts_last)
325 		panic("uninitialized ndopts in nd6_options");
326 	if (!ndopts->nd_opts_search)
327 		return 0;
328 
329 	while (1) {
330 		nd_opt = nd6_option(ndopts);
331 		if (!nd_opt && !ndopts->nd_opts_last) {
332 			/*
333 			 * Message validation requires that all included
334 			 * options have a length that is greater than zero.
335 			 */
336 			icmp6stat.icp6s_nd_badopt++;
337 			bzero(ndopts, sizeof(*ndopts));
338 			return -1;
339 		}
340 
341 		if (!nd_opt)
342 			goto skip1;
343 
344 		switch (nd_opt->nd_opt_type) {
345 		case ND_OPT_SOURCE_LINKADDR:
346 		case ND_OPT_TARGET_LINKADDR:
347 		case ND_OPT_MTU:
348 		case ND_OPT_REDIRECTED_HEADER:
349 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
350 				nd6log((LOG_INFO,
351 				    "duplicated ND6 option found (type=%d)\n",
352 				    nd_opt->nd_opt_type));
353 				/* XXX bark? */
354 			} else {
355 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
356 					= nd_opt;
357 			}
358 			break;
359 		case ND_OPT_PREFIX_INFORMATION:
360 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
361 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
362 					= nd_opt;
363 			}
364 			ndopts->nd_opts_pi_end =
365 				(struct nd_opt_prefix_info *)nd_opt;
366 			break;
367 		default:
368 			/*
369 			 * Unknown options must be silently ignored,
370 			 * to accomodate future extension to the protocol.
371 			 */
372 			nd6log((LOG_DEBUG,
373 			    "nd6_options: unsupported option %d - "
374 			    "option ignored\n", nd_opt->nd_opt_type));
375 		}
376 
377 skip1:
378 		i++;
379 		if (i > nd6_maxndopt) {
380 			icmp6stat.icp6s_nd_toomanyopt++;
381 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
382 			break;
383 		}
384 
385 		if (ndopts->nd_opts_done)
386 			break;
387 	}
388 
389 	return 0;
390 }
391 
392 /*
393  * ND6 timer routine to expire default route list and prefix list
394  */
395 void
396 nd6_timer(void *ignored_arg)
397 {
398 	struct llinfo_nd6 *ln;
399 	struct nd_defrouter *dr;
400 	struct nd_prefix *pr;
401 	struct ifnet *ifp;
402 	struct in6_ifaddr *ia6, *nia6;
403 	struct in6_addrlifetime *lt6;
404 
405 	crit_enter();
406 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
407 		      nd6_timer, NULL);
408 
409 	ln = llinfo_nd6.ln_next;
410 	while (ln && ln != &llinfo_nd6) {
411 		struct rtentry *rt;
412 		struct sockaddr_in6 *dst;
413 		struct llinfo_nd6 *next = ln->ln_next;
414 		/* XXX: used for the DELAY case only: */
415 		struct nd_ifinfo *ndi = NULL;
416 
417 		if ((rt = ln->ln_rt) == NULL) {
418 			ln = next;
419 			continue;
420 		}
421 		if ((ifp = rt->rt_ifp) == NULL) {
422 			ln = next;
423 			continue;
424 		}
425 		ndi = ND_IFINFO(ifp);
426 		dst = (struct sockaddr_in6 *)rt_key(rt);
427 
428 		if (ln->ln_expire > time_second) {
429 			ln = next;
430 			continue;
431 		}
432 
433 		/* sanity check */
434 		if (!rt)
435 			panic("rt=0 in nd6_timer(ln=%p)", ln);
436 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
437 			panic("rt_llinfo(%p) is not equal to ln(%p)",
438 			      rt->rt_llinfo, ln);
439 		if (!dst)
440 			panic("dst=0 in nd6_timer(ln=%p)", ln);
441 
442 		switch (ln->ln_state) {
443 		case ND6_LLINFO_INCOMPLETE:
444 			if (ln->ln_asked < nd6_mmaxtries) {
445 				ln->ln_asked++;
446 				ln->ln_expire = time_second +
447 					ND_IFINFO(ifp)->retrans / 1000;
448 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
449 					ln, 0);
450 			} else {
451 				struct mbuf *m = ln->ln_hold;
452 				if (m) {
453 					if (rt->rt_ifp) {
454 						/*
455 						 * Fake rcvif to make ICMP error
456 						 * more helpful in diagnosing
457 						 * for the receiver.
458 						 * XXX: should we consider
459 						 * older rcvif?
460 						 */
461 						m->m_pkthdr.rcvif = rt->rt_ifp;
462 					}
463 					icmp6_error(m, ICMP6_DST_UNREACH,
464 						    ICMP6_DST_UNREACH_ADDR, 0);
465 					ln->ln_hold = NULL;
466 				}
467 				next = nd6_free(rt);
468 			}
469 			break;
470 		case ND6_LLINFO_REACHABLE:
471 			if (ln->ln_expire) {
472 				ln->ln_state = ND6_LLINFO_STALE;
473 				ln->ln_expire = time_second + nd6_gctimer;
474 			}
475 			break;
476 
477 		case ND6_LLINFO_STALE:
478 			/* Garbage Collection(RFC 2461 5.3) */
479 			if (ln->ln_expire)
480 				next = nd6_free(rt);
481 			break;
482 
483 		case ND6_LLINFO_DELAY:
484 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD)) {
485 				/* We need NUD */
486 				ln->ln_asked = 1;
487 				ln->ln_state = ND6_LLINFO_PROBE;
488 				ln->ln_expire = time_second +
489 					ndi->retrans / 1000;
490 				nd6_ns_output(ifp, &dst->sin6_addr,
491 					      &dst->sin6_addr,
492 					      ln, 0);
493 			} else {
494 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
495 				ln->ln_expire = time_second + nd6_gctimer;
496 			}
497 			break;
498 		case ND6_LLINFO_PROBE:
499 			if (ln->ln_asked < nd6_umaxtries) {
500 				ln->ln_asked++;
501 				ln->ln_expire = time_second +
502 					ND_IFINFO(ifp)->retrans / 1000;
503 				nd6_ns_output(ifp, &dst->sin6_addr,
504 					       &dst->sin6_addr, ln, 0);
505 			} else {
506 				next = nd6_free(rt);
507 			}
508 			break;
509 		}
510 		ln = next;
511 	}
512 
513 	/* expire default router list */
514 	dr = TAILQ_FIRST(&nd_defrouter);
515 	while (dr) {
516 		if (dr->expire && dr->expire < time_second) {
517 			struct nd_defrouter *t;
518 			t = TAILQ_NEXT(dr, dr_entry);
519 			defrtrlist_del(dr);
520 			dr = t;
521 		} else {
522 			dr = TAILQ_NEXT(dr, dr_entry);
523 		}
524 	}
525 
526 	/*
527 	 * expire interface addresses.
528 	 * in the past the loop was inside prefix expiry processing.
529 	 * However, from a stricter speci-confrmance standpoint, we should
530 	 * rather separate address lifetimes and prefix lifetimes.
531 	 */
532 addrloop:
533 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
534 		nia6 = ia6->ia_next;
535 		/* check address lifetime */
536 		lt6 = &ia6->ia6_lifetime;
537 		if (IFA6_IS_INVALID(ia6)) {
538 			int regen = 0;
539 
540 			/*
541 			 * If the expiring address is temporary, try
542 			 * regenerating a new one.  This would be useful when
543 			 * we suspended a laptop PC, then turned it on after a
544 			 * period that could invalidate all temporary
545 			 * addresses.  Although we may have to restart the
546 			 * loop (see below), it must be after purging the
547 			 * address.  Otherwise, we'd see an infinite loop of
548 			 * regeneration.
549 			 */
550 			if (ip6_use_tempaddr &&
551 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY)) {
552 				if (regen_tmpaddr(ia6) == 0)
553 					regen = 1;
554 			}
555 
556 			in6_purgeaddr(&ia6->ia_ifa);
557 
558 			if (regen)
559 				goto addrloop; /* XXX: see below */
560 		}
561 		if (IFA6_IS_DEPRECATED(ia6)) {
562 			int oldflags = ia6->ia6_flags;
563 
564 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
565 
566 			/*
567 			 * If a temporary address has just become deprecated,
568 			 * regenerate a new one if possible.
569 			 */
570 			if (ip6_use_tempaddr &&
571 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) &&
572 			    !(oldflags & IN6_IFF_DEPRECATED)) {
573 
574 				if (regen_tmpaddr(ia6) == 0) {
575 					/*
576 					 * A new temporary address is
577 					 * generated.
578 					 * XXX: this means the address chain
579 					 * has changed while we are still in
580 					 * the loop.  Although the change
581 					 * would not cause disaster (because
582 					 * it's not a deletion, but an
583 					 * addition,) we'd rather restart the
584 					 * loop just for safety.  Or does this
585 					 * significantly reduce performance??
586 					 */
587 					goto addrloop;
588 				}
589 			}
590 		} else {
591 			/*
592 			 * A new RA might have made a deprecated address
593 			 * preferred.
594 			 */
595 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
596 		}
597 	}
598 
599 	/* expire prefix list */
600 	pr = nd_prefix.lh_first;
601 	while (pr) {
602 		/*
603 		 * check prefix lifetime.
604 		 * since pltime is just for autoconf, pltime processing for
605 		 * prefix is not necessary.
606 		 */
607 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
608 			struct nd_prefix *t;
609 			t = pr->ndpr_next;
610 
611 			/*
612 			 * address expiration and prefix expiration are
613 			 * separate.  NEVER perform in6_purgeaddr here.
614 			 */
615 
616 			prelist_remove(pr);
617 			pr = t;
618 		} else
619 			pr = pr->ndpr_next;
620 	}
621 	crit_exit();
622 }
623 
624 static int
625 regen_tmpaddr(struct in6_ifaddr *ia6) /* deprecated/invalidated temporary
626 					 address */
627 {
628 	struct ifaddr_container *ifac;
629 	struct ifnet *ifp;
630 	struct in6_ifaddr *public_ifa6 = NULL;
631 
632 	ifp = ia6->ia_ifa.ifa_ifp;
633 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
634 		struct ifaddr *ifa = ifac->ifa;
635 		struct in6_ifaddr *it6;
636 
637 		if (ifa->ifa_addr->sa_family != AF_INET6)
638 			continue;
639 
640 		it6 = (struct in6_ifaddr *)ifa;
641 
642 		/* ignore no autoconf addresses. */
643 		if (!(it6->ia6_flags & IN6_IFF_AUTOCONF))
644 			continue;
645 
646 		/* ignore autoconf addresses with different prefixes. */
647 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
648 			continue;
649 
650 		/*
651 		 * Now we are looking at an autoconf address with the same
652 		 * prefix as ours.  If the address is temporary and is still
653 		 * preferred, do not create another one.  It would be rare, but
654 		 * could happen, for example, when we resume a laptop PC after
655 		 * a long period.
656 		 */
657 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) &&
658 		    !IFA6_IS_DEPRECATED(it6)) {
659 			public_ifa6 = NULL;
660 			break;
661 		}
662 
663 		/*
664 		 * This is a public autoconf address that has the same prefix
665 		 * as ours.  If it is preferred, keep it.  We can't break the
666 		 * loop here, because there may be a still-preferred temporary
667 		 * address with the prefix.
668 		 */
669 		if (!IFA6_IS_DEPRECATED(it6))
670 		    public_ifa6 = it6;
671 	}
672 
673 	if (public_ifa6 != NULL) {
674 		int e;
675 
676 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
677 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
678 			    " tmp addr,errno=%d\n", e);
679 			return (-1);
680 		}
681 		return (0);
682 	}
683 
684 	return (-1);
685 }
686 
687 /*
688  * Nuke neighbor cache/prefix/default router management table, right before
689  * ifp goes away.
690  */
691 void
692 nd6_purge(struct ifnet *ifp)
693 {
694 	struct llinfo_nd6 *ln, *nln;
695 	struct nd_defrouter *dr, *ndr, drany;
696 	struct nd_prefix *pr, *npr;
697 
698 	/* Nuke default router list entries toward ifp */
699 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
700 		/*
701 		 * The first entry of the list may be stored in
702 		 * the routing table, so we'll delete it later.
703 		 */
704 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
705 			ndr = TAILQ_NEXT(dr, dr_entry);
706 			if (dr->ifp == ifp)
707 				defrtrlist_del(dr);
708 		}
709 		dr = TAILQ_FIRST(&nd_defrouter);
710 		if (dr->ifp == ifp)
711 			defrtrlist_del(dr);
712 	}
713 
714 	/* Nuke prefix list entries toward ifp */
715 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
716 		npr = pr->ndpr_next;
717 		if (pr->ndpr_ifp == ifp) {
718 			/*
719 			 * Previously, pr->ndpr_addr is removed as well,
720 			 * but I strongly believe we don't have to do it.
721 			 * nd6_purge() is only called from in6_ifdetach(),
722 			 * which removes all the associated interface addresses
723 			 * by itself.
724 			 * (jinmei@kame.net 20010129)
725 			 */
726 			prelist_remove(pr);
727 		}
728 	}
729 
730 	/* cancel default outgoing interface setting */
731 	if (nd6_defifindex == ifp->if_index)
732 		nd6_setdefaultiface(0);
733 
734 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
735 		/* refresh default router list */
736 		bzero(&drany, sizeof(drany));
737 		defrouter_delreq(&drany, 0);
738 		defrouter_select();
739 	}
740 
741 	/*
742 	 * Nuke neighbor cache entries for the ifp.
743 	 * Note that rt->rt_ifp may not be the same as ifp,
744 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
745 	 * nd6_rtrequest(), and ip6_input().
746 	 */
747 	ln = llinfo_nd6.ln_next;
748 	while (ln && ln != &llinfo_nd6) {
749 		struct rtentry *rt;
750 		struct sockaddr_dl *sdl;
751 
752 		nln = ln->ln_next;
753 		rt = ln->ln_rt;
754 		if (rt && rt->rt_gateway &&
755 		    rt->rt_gateway->sa_family == AF_LINK) {
756 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
757 			if (sdl->sdl_index == ifp->if_index)
758 				nln = nd6_free(rt);
759 		}
760 		ln = nln;
761 	}
762 }
763 
764 struct rtentry *
765 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp)
766 {
767 	struct rtentry *rt;
768 	struct sockaddr_in6 sin6;
769 
770 	bzero(&sin6, sizeof(sin6));
771 	sin6.sin6_len = sizeof(struct sockaddr_in6);
772 	sin6.sin6_family = AF_INET6;
773 	sin6.sin6_addr = *addr6;
774 
775 	if (create)
776 		rt = rtlookup((struct sockaddr *)&sin6);
777 	else
778 		rt = rtpurelookup((struct sockaddr *)&sin6);
779 	if (rt && !(rt->rt_flags & RTF_LLINFO)) {
780 		/*
781 		 * This is the case for the default route.
782 		 * If we want to create a neighbor cache for the address, we
783 		 * should free the route for the destination and allocate an
784 		 * interface route.
785 		 */
786 		if (create) {
787 			--rt->rt_refcnt;
788 			rt = NULL;
789 		}
790 	}
791 	if (!rt) {
792 		if (create && ifp) {
793 			int e;
794 
795 			/*
796 			 * If no route is available and create is set,
797 			 * we allocate a host route for the destination
798 			 * and treat it like an interface route.
799 			 * This hack is necessary for a neighbor which can't
800 			 * be covered by our own prefix.
801 			 */
802 			struct ifaddr *ifa =
803 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
804 			if (ifa == NULL)
805 				return (NULL);
806 
807 			/*
808 			 * Create a new route.  RTF_LLINFO is necessary
809 			 * to create a Neighbor Cache entry for the
810 			 * destination in nd6_rtrequest which will be
811 			 * called in rtrequest via ifa->ifa_rtrequest.
812 			 */
813 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
814 					   ifa->ifa_addr,
815 					   (struct sockaddr *)&all1_sa,
816 					   (ifa->ifa_flags |
817 					    RTF_HOST | RTF_LLINFO) &
818 					   ~RTF_CLONING,
819 					   &rt)) != 0)
820 				log(LOG_ERR,
821 				    "nd6_lookup: failed to add route for a "
822 				    "neighbor(%s), errno=%d\n",
823 				    ip6_sprintf(addr6), e);
824 			if (rt == NULL)
825 				return (NULL);
826 			if (rt->rt_llinfo) {
827 				struct llinfo_nd6 *ln =
828 					(struct llinfo_nd6 *)rt->rt_llinfo;
829 				ln->ln_state = ND6_LLINFO_NOSTATE;
830 			}
831 		} else
832 			return (NULL);
833 	}
834 	rt->rt_refcnt--;
835 	/*
836 	 * Validation for the entry.
837 	 * Note that the check for rt_llinfo is necessary because a cloned
838 	 * route from a parent route that has the L flag (e.g. the default
839 	 * route to a p2p interface) may have the flag, too, while the
840 	 * destination is not actually a neighbor.
841 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
842 	 *      it might be the loopback interface if the entry is for our
843 	 *      own address on a non-loopback interface. Instead, we should
844 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
845 	 *      interface.
846 	 */
847 	if ((rt->rt_flags & RTF_GATEWAY) || !(rt->rt_flags & RTF_LLINFO) ||
848 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
849 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
850 		if (create) {
851 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
852 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
853 			/* xxx more logs... kazu */
854 		}
855 		return (NULL);
856 	}
857 	return (rt);
858 }
859 
860 /*
861  * Detect if a given IPv6 address identifies a neighbor on a given link.
862  * XXX: should take care of the destination of a p2p link?
863  */
864 int
865 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
866 {
867 	struct ifaddr_container *ifac;
868 	int i;
869 
870 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
871 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
872 
873 	/*
874 	 * A link-local address is always a neighbor.
875 	 * XXX: we should use the sin6_scope_id field rather than the embedded
876 	 * interface index.
877 	 */
878 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
879 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
880 		return (1);
881 
882 	/*
883 	 * If the address matches one of our addresses,
884 	 * it should be a neighbor.
885 	 */
886 	TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
887 		struct ifaddr *ifa = ifac->ifa;
888 
889 		if (ifa->ifa_addr->sa_family != AF_INET6)
890 			next: continue;
891 
892 		for (i = 0; i < 4; i++) {
893 			if ((IFADDR6(ifa).s6_addr32[i] ^
894 			     addr->sin6_addr.s6_addr32[i]) &
895 			    IFMASK6(ifa).s6_addr32[i])
896 				goto next;
897 		}
898 		return (1);
899 	}
900 
901 	/*
902 	 * Even if the address matches none of our addresses, it might be
903 	 * in the neighbor cache.
904 	 */
905 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
906 		return (1);
907 
908 	return (0);
909 #undef IFADDR6
910 #undef IFMASK6
911 }
912 
913 /*
914  * Free an nd6 llinfo entry.
915  */
916 struct llinfo_nd6 *
917 nd6_free(struct rtentry *rt)
918 {
919 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
920 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
921 	struct nd_defrouter *dr;
922 
923 	/*
924 	 * we used to have kpfctlinput(PRC_HOSTDEAD) here.
925 	 * even though it is not harmful, it was not really necessary.
926 	 */
927 
928 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
929 		crit_enter();
930 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
931 				      rt->rt_ifp);
932 
933 		if (ln->ln_router || dr) {
934 			/*
935 			 * rt6_flush must be called whether or not the neighbor
936 			 * is in the Default Router List.
937 			 * See a corresponding comment in nd6_na_input().
938 			 */
939 			rt6_flush(&in6, rt->rt_ifp);
940 		}
941 
942 		if (dr) {
943 			/*
944 			 * Unreachablity of a router might affect the default
945 			 * router selection and on-link detection of advertised
946 			 * prefixes.
947 			 */
948 
949 			/*
950 			 * Temporarily fake the state to choose a new default
951 			 * router and to perform on-link determination of
952 			 * prefixes correctly.
953 			 * Below the state will be set correctly,
954 			 * or the entry itself will be deleted.
955 			 */
956 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
957 
958 			/*
959 			 * Since defrouter_select() does not affect the
960 			 * on-link determination and MIP6 needs the check
961 			 * before the default router selection, we perform
962 			 * the check now.
963 			 */
964 			pfxlist_onlink_check();
965 
966 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
967 				/*
968 				 * It is used as the current default router,
969 				 * so we have to move it to the end of the
970 				 * list and choose a new one.
971 				 * XXX: it is not very efficient if this is
972 				 *      the only router.
973 				 */
974 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
975 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
976 
977 				defrouter_select();
978 			}
979 		}
980 		crit_exit();
981 	}
982 
983 	/*
984 	 * Before deleting the entry, remember the next entry as the
985 	 * return value.  We need this because pfxlist_onlink_check() above
986 	 * might have freed other entries (particularly the old next entry) as
987 	 * a side effect (XXX).
988 	 */
989 	next = ln->ln_next;
990 
991 	/*
992 	 * Detach the route from the routing tree and the list of neighbor
993 	 * caches, and disable the route entry not to be used in already
994 	 * cached routes.
995 	 */
996 	rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
997 
998 	return (next);
999 }
1000 
1001 /*
1002  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1003  *
1004  * XXX cost-effective metods?
1005  */
1006 void
1007 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1008 {
1009 	struct llinfo_nd6 *ln;
1010 
1011 	/*
1012 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1013 	 * routing table by supplied "dst6".
1014 	 */
1015 	if (!rt) {
1016 		if (!dst6)
1017 			return;
1018 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1019 			return;
1020 	}
1021 
1022 	if ((rt->rt_flags & RTF_GATEWAY) ||
1023 	    !(rt->rt_flags & RTF_LLINFO) ||
1024 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1025 	    rt->rt_gateway->sa_family != AF_LINK) {
1026 		/* This is not a host route. */
1027 		return;
1028 	}
1029 
1030 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1031 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1032 		return;
1033 
1034 	/*
1035 	 * if we get upper-layer reachability confirmation many times,
1036 	 * it is possible we have false information.
1037 	 */
1038 	if (!force) {
1039 		ln->ln_byhint++;
1040 		if (ln->ln_byhint > nd6_maxnudhint)
1041 			return;
1042 	}
1043 
1044 	ln->ln_state = ND6_LLINFO_REACHABLE;
1045 	if (ln->ln_expire)
1046 		ln->ln_expire = time_second +
1047 			ND_IFINFO(rt->rt_ifp)->reachable;
1048 }
1049 
1050 void
1051 nd6_rtrequest(int req, struct rtentry *rt,
1052 	      struct rt_addrinfo *info) /* xxx unused */
1053 {
1054 	struct sockaddr *gate = rt->rt_gateway;
1055 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1056 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1057 	struct ifnet *ifp = rt->rt_ifp;
1058 	struct ifaddr *ifa;
1059 
1060 	if ((rt->rt_flags & RTF_GATEWAY))
1061 		return;
1062 
1063 	if (nd6_need_cache(ifp) == 0 && !(rt->rt_flags & RTF_HOST)) {
1064 		/*
1065 		 * This is probably an interface direct route for a link
1066 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1067 		 * We do not need special treatment below for such a route.
1068 		 * Moreover, the RTF_LLINFO flag which would be set below
1069 		 * would annoy the ndp(8) command.
1070 		 */
1071 		return;
1072 	}
1073 
1074 	if (req == RTM_RESOLVE &&
1075 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1076 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1077 		/*
1078 		 * FreeBSD and BSD/OS often make a cloned host route based
1079 		 * on a less-specific route (e.g. the default route).
1080 		 * If the less specific route does not have a "gateway"
1081 		 * (this is the case when the route just goes to a p2p or an
1082 		 * stf interface), we'll mistakenly make a neighbor cache for
1083 		 * the host route, and will see strange neighbor solicitation
1084 		 * for the corresponding destination.  In order to avoid the
1085 		 * confusion, we check if the destination of the route is
1086 		 * a neighbor in terms of neighbor discovery, and stop the
1087 		 * process if not.  Additionally, we remove the LLINFO flag
1088 		 * so that ndp(8) will not try to get the neighbor information
1089 		 * of the destination.
1090 		 */
1091 		rt->rt_flags &= ~RTF_LLINFO;
1092 		return;
1093 	}
1094 
1095 	switch (req) {
1096 	case RTM_ADD:
1097 		/*
1098 		 * There is no backward compatibility :)
1099 		 *
1100 		 * if (!(rt->rt_flags & RTF_HOST) &&
1101 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1102 		 *	   rt->rt_flags |= RTF_CLONING;
1103 		 */
1104 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1105 			/*
1106 			 * Case 1: This route should come from
1107 			 * a route to interface.  RTF_LLINFO flag is set
1108 			 * for a host route whose destination should be
1109 			 * treated as on-link.
1110 			 */
1111 			rt_setgate(rt, rt_key(rt),
1112 				   (struct sockaddr *)&null_sdl,
1113 				   RTL_DONTREPORT);
1114 			gate = rt->rt_gateway;
1115 			SDL(gate)->sdl_type = ifp->if_type;
1116 			SDL(gate)->sdl_index = ifp->if_index;
1117 			if (ln)
1118 				ln->ln_expire = time_second;
1119 #if 1
1120 			if (ln && ln->ln_expire == 0) {
1121 				/* kludge for desktops */
1122 #if 0
1123 				kprintf("nd6_rtequest: time.tv_sec is zero; "
1124 				       "treat it as 1\n");
1125 #endif
1126 				ln->ln_expire = 1;
1127 			}
1128 #endif
1129 			if ((rt->rt_flags & RTF_CLONING))
1130 				break;
1131 		}
1132 		/*
1133 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1134 		 * We don't do that here since llinfo is not ready yet.
1135 		 *
1136 		 * There are also couple of other things to be discussed:
1137 		 * - unsolicited NA code needs improvement beforehand
1138 		 * - RFC2461 says we MAY send multicast unsolicited NA
1139 		 *   (7.2.6 paragraph 4), however, it also says that we
1140 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1141 		 *   we don't have anything like it right now.
1142 		 *   note that the mechanism needs a mutual agreement
1143 		 *   between proxies, which means that we need to implement
1144 		 *   a new protocol, or a new kludge.
1145 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1146 		 *   we need to check ip6forwarding before sending it.
1147 		 *   (or should we allow proxy ND configuration only for
1148 		 *   routers?  there's no mention about proxy ND from hosts)
1149 		 */
1150 #if 0
1151 		/* XXX it does not work */
1152 		if (rt->rt_flags & RTF_ANNOUNCE)
1153 			nd6_na_output(ifp,
1154 			      &SIN6(rt_key(rt))->sin6_addr,
1155 			      &SIN6(rt_key(rt))->sin6_addr,
1156 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1157 			      1, NULL);
1158 #endif
1159 		/* FALLTHROUGH */
1160 	case RTM_RESOLVE:
1161 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1162 			/*
1163 			 * Address resolution isn't necessary for a point to
1164 			 * point link, so we can skip this test for a p2p link.
1165 			 */
1166 			if (gate->sa_family != AF_LINK ||
1167 			    gate->sa_len < sizeof(null_sdl)) {
1168 				log(LOG_DEBUG,
1169 				    "nd6_rtrequest: bad gateway value: %s\n",
1170 				    if_name(ifp));
1171 				break;
1172 			}
1173 			SDL(gate)->sdl_type = ifp->if_type;
1174 			SDL(gate)->sdl_index = ifp->if_index;
1175 		}
1176 		if (ln != NULL)
1177 			break;	/* This happens on a route change */
1178 		/*
1179 		 * Case 2: This route may come from cloning, or a manual route
1180 		 * add with a LL address.
1181 		 */
1182 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1183 		rt->rt_llinfo = (caddr_t)ln;
1184 		if (!ln) {
1185 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1186 			break;
1187 		}
1188 		nd6_inuse++;
1189 		nd6_allocated++;
1190 		bzero(ln, sizeof(*ln));
1191 		ln->ln_rt = rt;
1192 		/* this is required for "ndp" command. - shin */
1193 		if (req == RTM_ADD) {
1194 		        /*
1195 			 * gate should have some valid AF_LINK entry,
1196 			 * and ln->ln_expire should have some lifetime
1197 			 * which is specified by ndp command.
1198 			 */
1199 			ln->ln_state = ND6_LLINFO_REACHABLE;
1200 			ln->ln_byhint = 0;
1201 		} else {
1202 		        /*
1203 			 * When req == RTM_RESOLVE, rt is created and
1204 			 * initialized in rtrequest(), so rt_expire is 0.
1205 			 */
1206 			ln->ln_state = ND6_LLINFO_NOSTATE;
1207 			ln->ln_expire = time_second;
1208 		}
1209 		rt->rt_flags |= RTF_LLINFO;
1210 		ln->ln_next = llinfo_nd6.ln_next;
1211 		llinfo_nd6.ln_next = ln;
1212 		ln->ln_prev = &llinfo_nd6;
1213 		ln->ln_next->ln_prev = ln;
1214 
1215 		/*
1216 		 * check if rt_key(rt) is one of my address assigned
1217 		 * to the interface.
1218 		 */
1219 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1220 					  &SIN6(rt_key(rt))->sin6_addr);
1221 		if (ifa) {
1222 			caddr_t macp = nd6_ifptomac(ifp);
1223 			ln->ln_expire = 0;
1224 			ln->ln_state = ND6_LLINFO_REACHABLE;
1225 			ln->ln_byhint = 0;
1226 			if (macp) {
1227 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1228 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1229 			}
1230 			if (nd6_useloopback) {
1231 				rt->rt_ifp = &loif[0];	/* XXX */
1232 				/*
1233 				 * Make sure rt_ifa be equal to the ifaddr
1234 				 * corresponding to the address.
1235 				 * We need this because when we refer
1236 				 * rt_ifa->ia6_flags in ip6_input, we assume
1237 				 * that the rt_ifa points to the address instead
1238 				 * of the loopback address.
1239 				 */
1240 				if (ifa != rt->rt_ifa) {
1241 					IFAFREE(rt->rt_ifa);
1242 					IFAREF(ifa);
1243 					rt->rt_ifa = ifa;
1244 				}
1245 			}
1246 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1247 			ln->ln_expire = 0;
1248 			ln->ln_state = ND6_LLINFO_REACHABLE;
1249 			ln->ln_byhint = 0;
1250 
1251 			/* join solicited node multicast for proxy ND */
1252 			if (ifp->if_flags & IFF_MULTICAST) {
1253 				struct in6_addr llsol;
1254 				int error;
1255 
1256 				llsol = SIN6(rt_key(rt))->sin6_addr;
1257 				llsol.s6_addr16[0] = htons(0xff02);
1258 				llsol.s6_addr16[1] = htons(ifp->if_index);
1259 				llsol.s6_addr32[1] = 0;
1260 				llsol.s6_addr32[2] = htonl(1);
1261 				llsol.s6_addr8[12] = 0xff;
1262 
1263 				if (!in6_addmulti(&llsol, ifp, &error)) {
1264 					nd6log((LOG_ERR, "%s: failed to join "
1265 					    "%s (errno=%d)\n", if_name(ifp),
1266 					    ip6_sprintf(&llsol), error));
1267 				}
1268 			}
1269 		}
1270 		break;
1271 
1272 	case RTM_DELETE:
1273 		if (!ln)
1274 			break;
1275 		/* leave from solicited node multicast for proxy ND */
1276 		if ((rt->rt_flags & RTF_ANNOUNCE) &&
1277 		    (ifp->if_flags & IFF_MULTICAST)) {
1278 			struct in6_addr llsol;
1279 			struct in6_multi *in6m;
1280 
1281 			llsol = SIN6(rt_key(rt))->sin6_addr;
1282 			llsol.s6_addr16[0] = htons(0xff02);
1283 			llsol.s6_addr16[1] = htons(ifp->if_index);
1284 			llsol.s6_addr32[1] = 0;
1285 			llsol.s6_addr32[2] = htonl(1);
1286 			llsol.s6_addr8[12] = 0xff;
1287 
1288 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1289 			if (in6m)
1290 				in6_delmulti(in6m);
1291 		}
1292 		nd6_inuse--;
1293 		ln->ln_next->ln_prev = ln->ln_prev;
1294 		ln->ln_prev->ln_next = ln->ln_next;
1295 		ln->ln_prev = NULL;
1296 		rt->rt_llinfo = 0;
1297 		rt->rt_flags &= ~RTF_LLINFO;
1298 		if (ln->ln_hold)
1299 			m_freem(ln->ln_hold);
1300 		Free((caddr_t)ln);
1301 	}
1302 }
1303 
1304 int
1305 nd6_ioctl(u_long cmd, caddr_t	data, struct ifnet *ifp)
1306 {
1307 	struct in6_drlist *drl = (struct in6_drlist *)data;
1308 	struct in6_prlist *prl = (struct in6_prlist *)data;
1309 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1310 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1311 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1312 	struct nd_defrouter *dr, any;
1313 	struct nd_prefix *pr;
1314 	struct rtentry *rt;
1315 	int i = 0, error = 0;
1316 
1317 	switch (cmd) {
1318 	case SIOCGDRLST_IN6:
1319 		/*
1320 		 * obsolete API, use sysctl under net.inet6.icmp6
1321 		 */
1322 		bzero(drl, sizeof(*drl));
1323 		crit_enter();
1324 		dr = TAILQ_FIRST(&nd_defrouter);
1325 		while (dr && i < DRLSTSIZ) {
1326 			drl->defrouter[i].rtaddr = dr->rtaddr;
1327 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1328 				/* XXX: need to this hack for KAME stack */
1329 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1330 			} else
1331 				log(LOG_ERR,
1332 				    "default router list contains a "
1333 				    "non-linklocal address(%s)\n",
1334 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1335 
1336 			drl->defrouter[i].flags = dr->flags;
1337 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1338 			drl->defrouter[i].expire = dr->expire;
1339 			drl->defrouter[i].if_index = dr->ifp->if_index;
1340 			i++;
1341 			dr = TAILQ_NEXT(dr, dr_entry);
1342 		}
1343 		crit_exit();
1344 		break;
1345 	case SIOCGPRLST_IN6:
1346 		/*
1347 		 * obsolete API, use sysctl under net.inet6.icmp6
1348 		 */
1349 		/*
1350 		 * XXX meaning of fields, especialy "raflags", is very
1351 		 * differnet between RA prefix list and RR/static prefix list.
1352 		 * how about separating ioctls into two?
1353 		 */
1354 		bzero(prl, sizeof(*prl));
1355 		crit_enter();
1356 		pr = nd_prefix.lh_first;
1357 		while (pr && i < PRLSTSIZ) {
1358 			struct nd_pfxrouter *pfr;
1359 			int j;
1360 
1361 			in6_embedscope(&prl->prefix[i].prefix,
1362 			    &pr->ndpr_prefix, NULL, NULL);
1363 			prl->prefix[i].raflags = pr->ndpr_raf;
1364 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1365 			prl->prefix[i].vltime = pr->ndpr_vltime;
1366 			prl->prefix[i].pltime = pr->ndpr_pltime;
1367 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1368 			prl->prefix[i].expire = pr->ndpr_expire;
1369 
1370 			pfr = pr->ndpr_advrtrs.lh_first;
1371 			j = 0;
1372 			while (pfr) {
1373 				if (j < DRLSTSIZ) {
1374 #define RTRADDR prl->prefix[i].advrtr[j]
1375 					RTRADDR = pfr->router->rtaddr;
1376 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1377 						/* XXX: hack for KAME */
1378 						RTRADDR.s6_addr16[1] = 0;
1379 					} else
1380 						log(LOG_ERR,
1381 						    "a router(%s) advertises "
1382 						    "a prefix with "
1383 						    "non-link local address\n",
1384 						    ip6_sprintf(&RTRADDR));
1385 #undef RTRADDR
1386 				}
1387 				j++;
1388 				pfr = pfr->pfr_next;
1389 			}
1390 			prl->prefix[i].advrtrs = j;
1391 			prl->prefix[i].origin = PR_ORIG_RA;
1392 
1393 			i++;
1394 			pr = pr->ndpr_next;
1395 		}
1396 	      {
1397 		struct rr_prefix *rpp;
1398 
1399 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1400 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1401 			if (i >= PRLSTSIZ)
1402 				break;
1403 			in6_embedscope(&prl->prefix[i].prefix,
1404 			    &pr->ndpr_prefix, NULL, NULL);
1405 			prl->prefix[i].raflags = rpp->rp_raf;
1406 			prl->prefix[i].prefixlen = rpp->rp_plen;
1407 			prl->prefix[i].vltime = rpp->rp_vltime;
1408 			prl->prefix[i].pltime = rpp->rp_pltime;
1409 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1410 			prl->prefix[i].expire = rpp->rp_expire;
1411 			prl->prefix[i].advrtrs = 0;
1412 			prl->prefix[i].origin = rpp->rp_origin;
1413 			i++;
1414 		}
1415 	      }
1416 		crit_exit();
1417 
1418 		break;
1419 	case OSIOCGIFINFO_IN6:
1420 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1421 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1422 		ndi->ndi.linkmtu = ND_IFINFO(ifp)->linkmtu;
1423 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1424 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1425 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1426 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1427 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1428 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1429 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1430 		ndi->ndi.receivedra = ND_IFINFO(ifp)->receivedra;
1431 		break;
1432 	case SIOCGIFINFO_IN6:
1433 		ndi->ndi = *ND_IFINFO(ifp);
1434 		break;
1435 	case SIOCSIFINFO_FLAGS:
1436 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1437 		break;
1438 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1439 		/* flush default router list */
1440 		/*
1441 		 * xxx sumikawa: should not delete route if default
1442 		 * route equals to the top of default router list
1443 		 */
1444 		bzero(&any, sizeof(any));
1445 		defrouter_delreq(&any, 0);
1446 		defrouter_select();
1447 		/* xxx sumikawa: flush prefix list */
1448 		break;
1449 	case SIOCSPFXFLUSH_IN6:
1450 	    {
1451 		/* flush all the prefix advertised by routers */
1452 		struct nd_prefix *pr, *next;
1453 
1454 		crit_enter();
1455 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1456 			struct in6_ifaddr *ia, *ia_next;
1457 
1458 			next = pr->ndpr_next;
1459 
1460 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1461 				continue; /* XXX */
1462 
1463 			/* do we really have to remove addresses as well? */
1464 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1465 				/* ia might be removed.  keep the next ptr. */
1466 				ia_next = ia->ia_next;
1467 
1468 				if (!(ia->ia6_flags & IN6_IFF_AUTOCONF))
1469 					continue;
1470 
1471 				if (ia->ia6_ndpr == pr)
1472 					in6_purgeaddr(&ia->ia_ifa);
1473 			}
1474 			prelist_remove(pr);
1475 		}
1476 		crit_exit();
1477 		break;
1478 	    }
1479 	case SIOCSRTRFLUSH_IN6:
1480 	    {
1481 		/* flush all the default routers */
1482 		struct nd_defrouter *dr, *next;
1483 
1484 		crit_enter();
1485 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1486 			/*
1487 			 * The first entry of the list may be stored in
1488 			 * the routing table, so we'll delete it later.
1489 			 */
1490 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1491 				next = TAILQ_NEXT(dr, dr_entry);
1492 				defrtrlist_del(dr);
1493 			}
1494 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1495 		}
1496 		crit_exit();
1497 		break;
1498 	    }
1499 	case SIOCGNBRINFO_IN6:
1500 	    {
1501 		struct llinfo_nd6 *ln;
1502 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1503 
1504 		/*
1505 		 * XXX: KAME specific hack for scoped addresses
1506 		 *      XXXX: for other scopes than link-local?
1507 		 */
1508 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1509 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1510 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1511 
1512 			if (*idp == 0)
1513 				*idp = htons(ifp->if_index);
1514 		}
1515 
1516 		crit_enter();
1517 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1518 			error = EINVAL;
1519 			crit_exit();
1520 			break;
1521 		}
1522 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1523 		nbi->state = ln->ln_state;
1524 		nbi->asked = ln->ln_asked;
1525 		nbi->isrouter = ln->ln_router;
1526 		nbi->expire = ln->ln_expire;
1527 		crit_exit();
1528 
1529 		break;
1530 	    }
1531 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1532 		ndif->ifindex = nd6_defifindex;
1533 		break;
1534 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1535 		return (nd6_setdefaultiface(ndif->ifindex));
1536 		break;
1537 	}
1538 	return (error);
1539 }
1540 
1541 /*
1542  * Create neighbor cache entry and cache link-layer address,
1543  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1544  */
1545 struct rtentry *
1546 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1547 		 int lladdrlen,
1548 		 int type,	/* ICMP6 type */
1549 		 int code	/* type dependent information */)
1550 {
1551 	struct rtentry *rt = NULL;
1552 	struct llinfo_nd6 *ln = NULL;
1553 	int is_newentry;
1554 	struct sockaddr_dl *sdl = NULL;
1555 	int do_update;
1556 	int olladdr;
1557 	int llchange;
1558 	int newstate = 0;
1559 
1560 	if (!ifp)
1561 		panic("ifp == NULL in nd6_cache_lladdr");
1562 	if (!from)
1563 		panic("from == NULL in nd6_cache_lladdr");
1564 
1565 	/* nothing must be updated for unspecified address */
1566 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1567 		return NULL;
1568 
1569 	/*
1570 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1571 	 * the caller.
1572 	 *
1573 	 * XXX If the link does not have link-layer adderss, what should
1574 	 * we do? (ifp->if_addrlen == 0)
1575 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1576 	 * description on it in NS section (RFC 2461 7.2.3).
1577 	 */
1578 
1579 	rt = nd6_lookup(from, 0, ifp);
1580 	if (!rt) {
1581 #if 0
1582 		/* nothing must be done if there's no lladdr */
1583 		if (!lladdr || !lladdrlen)
1584 			return NULL;
1585 #endif
1586 
1587 		rt = nd6_lookup(from, 1, ifp);
1588 		is_newentry = 1;
1589 	} else {
1590 		/* do nothing if static ndp is set */
1591 		if (rt->rt_flags & RTF_STATIC)
1592 			return NULL;
1593 		is_newentry = 0;
1594 	}
1595 
1596 	if (!rt)
1597 		return NULL;
1598 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1599 fail:
1600 		nd6_free(rt);
1601 		return NULL;
1602 	}
1603 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1604 	if (!ln)
1605 		goto fail;
1606 	if (!rt->rt_gateway)
1607 		goto fail;
1608 	if (rt->rt_gateway->sa_family != AF_LINK)
1609 		goto fail;
1610 	sdl = SDL(rt->rt_gateway);
1611 
1612 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1613 	if (olladdr && lladdr) {
1614 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1615 			llchange = 1;
1616 		else
1617 			llchange = 0;
1618 	} else
1619 		llchange = 0;
1620 
1621 	/*
1622 	 * newentry olladdr  lladdr  llchange	(*=record)
1623 	 *	0	n	n	--	(1)
1624 	 *	0	y	n	--	(2)
1625 	 *	0	n	y	--	(3) * STALE
1626 	 *	0	y	y	n	(4) *
1627 	 *	0	y	y	y	(5) * STALE
1628 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1629 	 *	1	--	y	--	(7) * STALE
1630 	 */
1631 
1632 	if (lladdr) {		/* (3-5) and (7) */
1633 		/*
1634 		 * Record source link-layer address
1635 		 * XXX is it dependent to ifp->if_type?
1636 		 */
1637 		sdl->sdl_alen = ifp->if_addrlen;
1638 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1639 	}
1640 
1641 	if (!is_newentry) {
1642 		if ((!olladdr && lladdr)		/* (3) */
1643 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1644 			do_update = 1;
1645 			newstate = ND6_LLINFO_STALE;
1646 		} else					/* (1-2,4) */
1647 			do_update = 0;
1648 	} else {
1649 		do_update = 1;
1650 		if (!lladdr)				/* (6) */
1651 			newstate = ND6_LLINFO_NOSTATE;
1652 		else					/* (7) */
1653 			newstate = ND6_LLINFO_STALE;
1654 	}
1655 
1656 	if (do_update) {
1657 		/*
1658 		 * Update the state of the neighbor cache.
1659 		 */
1660 		ln->ln_state = newstate;
1661 
1662 		if (ln->ln_state == ND6_LLINFO_STALE) {
1663 			/*
1664 			 * XXX: since nd6_output() below will cause
1665 			 * state tansition to DELAY and reset the timer,
1666 			 * we must set the timer now, although it is actually
1667 			 * meaningless.
1668 			 */
1669 			ln->ln_expire = time_second + nd6_gctimer;
1670 
1671 			if (ln->ln_hold) {
1672 				/*
1673 				 * we assume ifp is not a p2p here, so just
1674 				 * set the 2nd argument as the 1st one.
1675 				 */
1676 				nd6_output(ifp, ifp, ln->ln_hold,
1677 					   (struct sockaddr_in6 *)rt_key(rt),
1678 					   rt);
1679 				ln->ln_hold = NULL;
1680 			}
1681 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1682 			/* probe right away */
1683 			ln->ln_expire = time_second;
1684 		}
1685 	}
1686 
1687 	/*
1688 	 * ICMP6 type dependent behavior.
1689 	 *
1690 	 * NS: clear IsRouter if new entry
1691 	 * RS: clear IsRouter
1692 	 * RA: set IsRouter if there's lladdr
1693 	 * redir: clear IsRouter if new entry
1694 	 *
1695 	 * RA case, (1):
1696 	 * The spec says that we must set IsRouter in the following cases:
1697 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1698 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1699 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1700 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1701 	 * neighbor cache, this is similar to (6).
1702 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1703 	 *
1704 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1705 	 *							D R
1706 	 *	0	n	n	--	(1)	c   ?     s
1707 	 *	0	y	n	--	(2)	c   s     s
1708 	 *	0	n	y	--	(3)	c   s     s
1709 	 *	0	y	y	n	(4)	c   s     s
1710 	 *	0	y	y	y	(5)	c   s     s
1711 	 *	1	--	n	--	(6) c	c 	c s
1712 	 *	1	--	y	--	(7) c	c   s	c s
1713 	 *
1714 	 *					(c=clear s=set)
1715 	 */
1716 	switch (type & 0xff) {
1717 	case ND_NEIGHBOR_SOLICIT:
1718 		/*
1719 		 * New entry must have is_router flag cleared.
1720 		 */
1721 		if (is_newentry)	/* (6-7) */
1722 			ln->ln_router = 0;
1723 		break;
1724 	case ND_REDIRECT:
1725 		/*
1726 		 * If the icmp is a redirect to a better router, always set the
1727 		 * is_router flag. Otherwise, if the entry is newly created,
1728 		 * clear the flag. [RFC 2461, sec 8.3]
1729 		 */
1730 		if (code == ND_REDIRECT_ROUTER)
1731 			ln->ln_router = 1;
1732 		else if (is_newentry) /* (6-7) */
1733 			ln->ln_router = 0;
1734 		break;
1735 	case ND_ROUTER_SOLICIT:
1736 		/*
1737 		 * is_router flag must always be cleared.
1738 		 */
1739 		ln->ln_router = 0;
1740 		break;
1741 	case ND_ROUTER_ADVERT:
1742 		/*
1743 		 * Mark an entry with lladdr as a router.
1744 		 */
1745 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1746 		 || (is_newentry && lladdr)) {			/* (7) */
1747 			ln->ln_router = 1;
1748 		}
1749 		break;
1750 	}
1751 
1752 	/*
1753 	 * When the link-layer address of a router changes, select the
1754 	 * best router again.  In particular, when the neighbor entry is newly
1755 	 * created, it might affect the selection policy.
1756 	 * Question: can we restrict the first condition to the "is_newentry"
1757 	 * case?
1758 	 * XXX: when we hear an RA from a new router with the link-layer
1759 	 * address option, defrouter_select() is called twice, since
1760 	 * defrtrlist_update called the function as well.  However, I believe
1761 	 * we can compromise the overhead, since it only happens the first
1762 	 * time.
1763 	 * XXX: although defrouter_select() should not have a bad effect
1764 	 * for those are not autoconfigured hosts, we explicitly avoid such
1765 	 * cases for safety.
1766 	 */
1767 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1768 		defrouter_select();
1769 
1770 	return rt;
1771 }
1772 
1773 static void
1774 nd6_slowtimo(void *ignored_arg)
1775 {
1776 	struct nd_ifinfo *nd6if;
1777 	struct ifnet *ifp;
1778 
1779 	crit_enter();
1780 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1781 	    nd6_slowtimo, NULL);
1782 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1783 		nd6if = ND_IFINFO(ifp);
1784 		if (nd6if->basereachable && /* already initialized */
1785 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1786 			/*
1787 			 * Since reachable time rarely changes by router
1788 			 * advertisements, we SHOULD insure that a new random
1789 			 * value gets recomputed at least once every few hours.
1790 			 * (RFC 2461, 6.3.4)
1791 			 */
1792 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1793 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1794 		}
1795 	}
1796 	crit_exit();
1797 }
1798 
1799 #define gotoerr(e) { error = (e); goto bad;}
1800 
1801 int
1802 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
1803 	   struct sockaddr_in6 *dst, struct rtentry *rt)
1804 {
1805 	struct llinfo_nd6 *ln = NULL;
1806 	int error = 0;
1807 
1808 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1809 		goto sendpkt;
1810 
1811 	if (nd6_need_cache(ifp) == 0)
1812 		goto sendpkt;
1813 
1814 	/*
1815 	 * next hop determination.  This routine is derived from ether_outpout.
1816 	 */
1817 	if (rt != NULL) {
1818 		if (!(rt->rt_flags & RTF_UP)) {
1819 			rt = rtlookup((struct sockaddr *)dst);
1820 			if (rt == NULL)
1821 				gotoerr(EHOSTUNREACH);
1822 			rt->rt_refcnt--;
1823 			if (rt->rt_ifp != ifp) {
1824 				/* XXX: loop care? */
1825 				return nd6_output(ifp, origifp, m, dst, rt);
1826 			}
1827 		}
1828 		if (rt->rt_flags & RTF_GATEWAY) {
1829 			struct sockaddr_in6 *gw6;
1830 
1831 			/*
1832 			 * We skip link-layer address resolution and NUD
1833 			 * if the gateway is not a neighbor from ND point
1834 			 * of view, regardless of the value of nd_ifinfo.flags.
1835 			 * The second condition is a bit tricky; we skip
1836 			 * if the gateway is our own address, which is
1837 			 * sometimes used to install a route to a p2p link.
1838 			 */
1839 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1840 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1841 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1842 				/*
1843 				 * We allow this kind of tricky route only
1844 				 * when the outgoing interface is p2p.
1845 				 * XXX: we may need a more generic rule here.
1846 				 */
1847 				if (!(ifp->if_flags & IFF_POINTOPOINT))
1848 					gotoerr(EHOSTUNREACH);
1849 
1850 				goto sendpkt;
1851 			}
1852 
1853 			if (rt->rt_gwroute == NULL) {
1854 				rt->rt_gwroute = rtlookup(rt->rt_gateway);
1855 				if (rt->rt_gwroute == NULL)
1856 					gotoerr(EHOSTUNREACH);
1857 			} else if (!(rt->rt_gwroute->rt_flags & RTF_UP)) {
1858 				rtfree(rt->rt_gwroute);
1859 				rt->rt_gwroute = rtlookup(rt->rt_gateway);
1860 				if (rt->rt_gwroute == NULL)
1861 					gotoerr(EHOSTUNREACH);
1862 			}
1863 		}
1864 	}
1865 
1866 	/*
1867 	 * Address resolution or Neighbor Unreachability Detection
1868 	 * for the next hop.
1869 	 * At this point, the destination of the packet must be a unicast
1870 	 * or an anycast address(i.e. not a multicast).
1871 	 */
1872 
1873 	/* Look up the neighbor cache for the nexthop */
1874 	if (rt && (rt->rt_flags & RTF_LLINFO))
1875 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1876 	else {
1877 		/*
1878 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1879 		 * the condition below is not very efficient.  But we believe
1880 		 * it is tolerable, because this should be a rare case.
1881 		 */
1882 		if (nd6_is_addr_neighbor(dst, ifp) &&
1883 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1884 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1885 	}
1886 	if (!ln || !rt) {
1887 		if (!(ifp->if_flags & IFF_POINTOPOINT) &&
1888 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1889 			log(LOG_DEBUG,
1890 			    "nd6_output: can't allocate llinfo for %s "
1891 			    "(ln=%p, rt=%p)\n",
1892 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1893 			gotoerr(EIO);	/* XXX: good error? */
1894 		}
1895 
1896 		goto sendpkt;	/* send anyway */
1897 	}
1898 
1899 	/* We don't have to do link-layer address resolution on a p2p link. */
1900 	if ((ifp->if_flags & IFF_POINTOPOINT) &&
1901 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1902 		ln->ln_state = ND6_LLINFO_STALE;
1903 		ln->ln_expire = time_second + nd6_gctimer;
1904 	}
1905 
1906 	/*
1907 	 * The first time we send a packet to a neighbor whose entry is
1908 	 * STALE, we have to change the state to DELAY and a sets a timer to
1909 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1910 	 * neighbor unreachability detection on expiration.
1911 	 * (RFC 2461 7.3.3)
1912 	 */
1913 	if (ln->ln_state == ND6_LLINFO_STALE) {
1914 		ln->ln_asked = 0;
1915 		ln->ln_state = ND6_LLINFO_DELAY;
1916 		ln->ln_expire = time_second + nd6_delay;
1917 	}
1918 
1919 	/*
1920 	 * If the neighbor cache entry has a state other than INCOMPLETE
1921 	 * (i.e. its link-layer address is already resolved), just
1922 	 * send the packet.
1923 	 */
1924 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1925 		goto sendpkt;
1926 
1927 	/*
1928 	 * There is a neighbor cache entry, but no ethernet address
1929 	 * response yet.  Replace the held mbuf (if any) with this
1930 	 * latest one.
1931 	 *
1932 	 * This code conforms to the rate-limiting rule described in Section
1933 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1934 	 * an NS below.
1935 	 */
1936 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1937 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1938 	if (ln->ln_hold)
1939 		m_freem(ln->ln_hold);
1940 	ln->ln_hold = m;
1941 	if (ln->ln_expire) {
1942 		if (ln->ln_asked < nd6_mmaxtries &&
1943 		    ln->ln_expire < time_second) {
1944 			ln->ln_asked++;
1945 			ln->ln_expire = time_second +
1946 				ND_IFINFO(ifp)->retrans / 1000;
1947 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1948 		}
1949 	}
1950 	return (0);
1951 
1952 sendpkt:
1953 	if (ifp->if_flags & IFF_LOOPBACK)
1954 		error = ifp->if_output(origifp, m, (struct sockaddr *)dst, rt);
1955 	else
1956 		error = ifp->if_output(ifp, m, (struct sockaddr *)dst, rt);
1957 	return (error);
1958 
1959 bad:
1960 	m_freem(m);
1961 	return (error);
1962 }
1963 #undef gotoerr
1964 
1965 int
1966 nd6_need_cache(struct ifnet *ifp)
1967 {
1968 	/*
1969 	 * XXX: we currently do not make neighbor cache on any interface
1970 	 * other than Ethernet and GIF.
1971 	 *
1972 	 * RFC2893 says:
1973 	 * - unidirectional tunnels needs no ND
1974 	 */
1975 	switch (ifp->if_type) {
1976 	case IFT_ETHER:
1977 	case IFT_IEEE1394:
1978 #ifdef IFT_L2VLAN
1979 	case IFT_L2VLAN:
1980 #endif
1981 #ifdef IFT_IEEE80211
1982 	case IFT_IEEE80211:
1983 #endif
1984 #ifdef IFT_CARP
1985 	case IFT_CARP:
1986 #endif
1987 	case IFT_GIF:		/* XXX need more cases? */
1988 		return (1);
1989 	default:
1990 		return (0);
1991 	}
1992 }
1993 
1994 int
1995 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m,
1996 		struct sockaddr *dst, u_char *desten)
1997 {
1998 	struct sockaddr_dl *sdl;
1999 	struct rtentry *rt;
2000 
2001 
2002 	if (m->m_flags & M_MCAST) {
2003 		switch (ifp->if_type) {
2004 		case IFT_ETHER:
2005 #ifdef IFT_L2VLAN
2006 	case IFT_L2VLAN:
2007 #endif
2008 #ifdef IFT_IEEE80211
2009 		case IFT_IEEE80211:
2010 #endif
2011 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2012 						 desten);
2013 			return (1);
2014 		case IFT_IEEE1394:
2015 			bcopy(ifp->if_broadcastaddr, desten, ifp->if_addrlen);
2016 			return (1);
2017 		default:
2018 			m_freem(m);
2019 			return (0);
2020 		}
2021 	}
2022 	if (rt0 == NULL) {
2023 		/* this could happen, if we could not allocate memory */
2024 		m_freem(m);
2025 		return (0);
2026 	}
2027 	if (rt_llroute(dst, rt0, &rt) != 0) {
2028 		m_freem(m);
2029 		return (0);
2030 	}
2031 	if (rt->rt_gateway->sa_family != AF_LINK) {
2032 		kprintf("nd6_storelladdr: something odd happens\n");
2033 		m_freem(m);
2034 		return (0);
2035 	}
2036 	sdl = SDL(rt->rt_gateway);
2037 	if (sdl->sdl_alen == 0) {
2038 		/* this should be impossible, but we bark here for debugging */
2039 		kprintf("nd6_storelladdr: sdl_alen == 0\n");
2040 		m_freem(m);
2041 		return (0);
2042 	}
2043 
2044 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2045 	return (1);
2046 }
2047 
2048 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2049 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2050 #ifdef SYSCTL_DECL
2051 SYSCTL_DECL(_net_inet6_icmp6);
2052 #endif
2053 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2054 	CTLFLAG_RD, nd6_sysctl_drlist, "List default routers");
2055 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2056 	CTLFLAG_RD, nd6_sysctl_prlist, "List prefixes");
2057 
2058 static int
2059 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2060 {
2061 	int error;
2062 	char buf[1024];
2063 	struct in6_defrouter *d, *de;
2064 	struct nd_defrouter *dr;
2065 
2066 	if (req->newptr)
2067 		return EPERM;
2068 	error = 0;
2069 
2070 	for (dr = TAILQ_FIRST(&nd_defrouter);
2071 	     dr;
2072 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2073 		d = (struct in6_defrouter *)buf;
2074 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2075 
2076 		if (d + 1 <= de) {
2077 			bzero(d, sizeof(*d));
2078 			d->rtaddr.sin6_family = AF_INET6;
2079 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2080 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2081 			    dr->ifp) != 0)
2082 				log(LOG_ERR,
2083 				    "scope error in "
2084 				    "default router list (%s)\n",
2085 				    ip6_sprintf(&dr->rtaddr));
2086 			d->flags = dr->flags;
2087 			d->rtlifetime = dr->rtlifetime;
2088 			d->expire = dr->expire;
2089 			d->if_index = dr->ifp->if_index;
2090 		} else
2091 			panic("buffer too short");
2092 
2093 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2094 		if (error)
2095 			break;
2096 	}
2097 	return error;
2098 }
2099 
2100 static int
2101 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2102 {
2103 	int error;
2104 	char buf[1024];
2105 	struct in6_prefix *p, *pe;
2106 	struct nd_prefix *pr;
2107 
2108 	if (req->newptr)
2109 		return EPERM;
2110 	error = 0;
2111 
2112 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2113 		u_short advrtrs;
2114 		size_t advance;
2115 		struct sockaddr_in6 *sin6, *s6;
2116 		struct nd_pfxrouter *pfr;
2117 
2118 		p = (struct in6_prefix *)buf;
2119 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2120 
2121 		if (p + 1 <= pe) {
2122 			bzero(p, sizeof(*p));
2123 			sin6 = (struct sockaddr_in6 *)(p + 1);
2124 
2125 			p->prefix = pr->ndpr_prefix;
2126 			if (in6_recoverscope(&p->prefix,
2127 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2128 				log(LOG_ERR,
2129 				    "scope error in prefix list (%s)\n",
2130 				    ip6_sprintf(&p->prefix.sin6_addr));
2131 			p->raflags = pr->ndpr_raf;
2132 			p->prefixlen = pr->ndpr_plen;
2133 			p->vltime = pr->ndpr_vltime;
2134 			p->pltime = pr->ndpr_pltime;
2135 			p->if_index = pr->ndpr_ifp->if_index;
2136 			p->expire = pr->ndpr_expire;
2137 			p->refcnt = pr->ndpr_refcnt;
2138 			p->flags = pr->ndpr_stateflags;
2139 			p->origin = PR_ORIG_RA;
2140 			advrtrs = 0;
2141 			for (pfr = pr->ndpr_advrtrs.lh_first;
2142 			     pfr;
2143 			     pfr = pfr->pfr_next) {
2144 				if ((void *)&sin6[advrtrs + 1] >
2145 				    (void *)pe) {
2146 					advrtrs++;
2147 					continue;
2148 				}
2149 				s6 = &sin6[advrtrs];
2150 				bzero(s6, sizeof(*s6));
2151 				s6->sin6_family = AF_INET6;
2152 				s6->sin6_len = sizeof(*sin6);
2153 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2154 						     pfr->router->ifp) != 0)
2155 					log(LOG_ERR,
2156 					    "scope error in "
2157 					    "prefix list (%s)\n",
2158 					    ip6_sprintf(&pfr->router->rtaddr));
2159 				advrtrs++;
2160 			}
2161 			p->advrtrs = advrtrs;
2162 		} else
2163 			panic("buffer too short");
2164 
2165 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2166 		error = SYSCTL_OUT(req, buf, advance);
2167 		if (error)
2168 			break;
2169 	}
2170 	return error;
2171 }
2172