xref: /freebsd/sys/netinet6/nd6.c (revision 6b3455a7665208c366849f0b2b3bc916fb97516e)
1 /*	$FreeBSD$	*/
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 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/callout.h>
40 #include <sys/mac.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/queue.h>
51 #include <sys/sysctl.h>
52 
53 #include <net/if.h>
54 #include <net/if_arc.h>
55 #include <net/if_dl.h>
56 #include <net/if_types.h>
57 #include <net/if_atm.h>
58 #include <net/iso88025.h>
59 #include <net/fddi.h>
60 #include <net/route.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/if_ether.h>
64 #include <netinet6/in6_var.h>
65 #include <netinet/ip6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet6/in6_prefix.h>
69 #include <netinet/icmp6.h>
70 
71 #include <net/net_osdep.h>
72 
73 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
74 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
75 
76 #define SIN6(s) ((struct sockaddr_in6 *)s)
77 #define SDL(s) ((struct sockaddr_dl *)s)
78 
79 /* timer values */
80 int	nd6_prune	= 1;	/* walk list every 1 seconds */
81 int	nd6_delay	= 5;	/* delay first probe time 5 second */
82 int	nd6_umaxtries	= 3;	/* maximum unicast query */
83 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
84 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
85 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
86 
87 /* preventing too many loops in ND option parsing */
88 int nd6_maxndopt = 10;	/* max # of ND options allowed */
89 
90 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
91 
92 #ifdef ND6_DEBUG
93 int nd6_debug = 1;
94 #else
95 int nd6_debug = 0;
96 #endif
97 
98 /* for debugging? */
99 static int nd6_inuse, nd6_allocated;
100 
101 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
102 struct nd_drhead nd_defrouter;
103 struct nd_prhead nd_prefix = { 0 };
104 
105 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
106 static struct sockaddr_in6 all1_sa;
107 
108 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
109 static void nd6_slowtimo __P((void *));
110 static int regen_tmpaddr __P((struct in6_ifaddr *));
111 
112 struct callout nd6_slowtimo_ch;
113 struct callout nd6_timer_ch;
114 extern struct callout in6_tmpaddrtimer_ch;
115 
116 void
117 nd6_init()
118 {
119 	static int nd6_init_done = 0;
120 	int i;
121 
122 	if (nd6_init_done) {
123 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
124 		return;
125 	}
126 
127 	all1_sa.sin6_family = AF_INET6;
128 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
129 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
130 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
131 
132 	/* initialization of the default router list */
133 	TAILQ_INIT(&nd_defrouter);
134 
135 	nd6_init_done = 1;
136 
137 	/* start timer */
138 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
139 	    nd6_slowtimo, NULL);
140 }
141 
142 struct nd_ifinfo *
143 nd6_ifattach(ifp)
144 	struct ifnet *ifp;
145 {
146 	struct nd_ifinfo *nd;
147 
148 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
149 	bzero(nd, sizeof(*nd));
150 
151 	nd->initialized = 1;
152 
153 	nd->chlim = IPV6_DEFHLIM;
154 	nd->basereachable = REACHABLE_TIME;
155 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
156 	nd->retrans = RETRANS_TIMER;
157 	/*
158 	 * Note that the default value of ip6_accept_rtadv is 0, which means
159 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
160 	 * here.
161 	 */
162 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
163 
164 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
165 	nd6_setmtu0(ifp, nd);
166 
167 	return nd;
168 }
169 
170 void
171 nd6_ifdetach(nd)
172 	struct nd_ifinfo *nd;
173 {
174 
175 	free(nd, M_IP6NDP);
176 }
177 
178 /*
179  * Reset ND level link MTU. This function is called when the physical MTU
180  * changes, which means we might have to adjust the ND level MTU.
181  */
182 void
183 nd6_setmtu(ifp)
184 	struct ifnet *ifp;
185 {
186 
187 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
188 }
189 
190 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
191 void
192 nd6_setmtu0(ifp, ndi)
193 	struct ifnet *ifp;
194 	struct nd_ifinfo *ndi;
195 {
196 	u_int32_t omaxmtu;
197 
198 	omaxmtu = ndi->maxmtu;
199 
200 	switch (ifp->if_type) {
201 	case IFT_ARCNET:
202 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
203 		break;
204 	case IFT_ETHER:
205 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
206 		break;
207 	case IFT_FDDI:
208 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
209 		break;
210 	case IFT_ATM:
211 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
212 		break;
213 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
214 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
215 		break;
216 #ifdef IFT_IEEE80211
217 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
218 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
219 		break;
220 #endif
221 	 case IFT_ISO88025:
222 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
223 		 break;
224 	default:
225 		ndi->maxmtu = ifp->if_mtu;
226 		break;
227 	}
228 
229 	/*
230 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
231 	 * undesirable situation.  We thus notify the operator of the change
232 	 * explicitly.  The check for omaxmtu is necessary to restrict the
233 	 * log to the case of changing the MTU, not initializing it.
234 	 */
235 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
236 		log(LOG_NOTICE, "nd6_setmtu0: "
237 		    "new link MTU on %s (%lu) is too small for IPv6\n",
238 		    if_name(ifp), (unsigned long)ndi->maxmtu);
239 	}
240 
241 	if (ndi->maxmtu > in6_maxmtu)
242 		in6_setmaxmtu(); /* check all interfaces just in case */
243 
244 #undef MIN
245 }
246 
247 void
248 nd6_option_init(opt, icmp6len, ndopts)
249 	void *opt;
250 	int icmp6len;
251 	union nd_opts *ndopts;
252 {
253 
254 	bzero(ndopts, sizeof(*ndopts));
255 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
256 	ndopts->nd_opts_last
257 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
258 
259 	if (icmp6len == 0) {
260 		ndopts->nd_opts_done = 1;
261 		ndopts->nd_opts_search = NULL;
262 	}
263 }
264 
265 /*
266  * Take one ND option.
267  */
268 struct nd_opt_hdr *
269 nd6_option(ndopts)
270 	union nd_opts *ndopts;
271 {
272 	struct nd_opt_hdr *nd_opt;
273 	int olen;
274 
275 	if (!ndopts)
276 		panic("ndopts == NULL in nd6_option");
277 	if (!ndopts->nd_opts_last)
278 		panic("uninitialized ndopts in nd6_option");
279 	if (!ndopts->nd_opts_search)
280 		return NULL;
281 	if (ndopts->nd_opts_done)
282 		return NULL;
283 
284 	nd_opt = ndopts->nd_opts_search;
285 
286 	/* make sure nd_opt_len is inside the buffer */
287 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
288 		bzero(ndopts, sizeof(*ndopts));
289 		return NULL;
290 	}
291 
292 	olen = nd_opt->nd_opt_len << 3;
293 	if (olen == 0) {
294 		/*
295 		 * Message validation requires that all included
296 		 * options have a length that is greater than zero.
297 		 */
298 		bzero(ndopts, sizeof(*ndopts));
299 		return NULL;
300 	}
301 
302 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
303 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
304 		/* option overruns the end of buffer, invalid */
305 		bzero(ndopts, sizeof(*ndopts));
306 		return NULL;
307 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
308 		/* reached the end of options chain */
309 		ndopts->nd_opts_done = 1;
310 		ndopts->nd_opts_search = NULL;
311 	}
312 	return nd_opt;
313 }
314 
315 /*
316  * Parse multiple ND options.
317  * This function is much easier to use, for ND routines that do not need
318  * multiple options of the same type.
319  */
320 int
321 nd6_options(ndopts)
322 	union nd_opts *ndopts;
323 {
324 	struct nd_opt_hdr *nd_opt;
325 	int i = 0;
326 
327 	if (!ndopts)
328 		panic("ndopts == NULL in nd6_options");
329 	if (!ndopts->nd_opts_last)
330 		panic("uninitialized ndopts in nd6_options");
331 	if (!ndopts->nd_opts_search)
332 		return 0;
333 
334 	while (1) {
335 		nd_opt = nd6_option(ndopts);
336 		if (!nd_opt && !ndopts->nd_opts_last) {
337 			/*
338 			 * Message validation requires that all included
339 			 * options have a length that is greater than zero.
340 			 */
341 			icmp6stat.icp6s_nd_badopt++;
342 			bzero(ndopts, sizeof(*ndopts));
343 			return -1;
344 		}
345 
346 		if (!nd_opt)
347 			goto skip1;
348 
349 		switch (nd_opt->nd_opt_type) {
350 		case ND_OPT_SOURCE_LINKADDR:
351 		case ND_OPT_TARGET_LINKADDR:
352 		case ND_OPT_MTU:
353 		case ND_OPT_REDIRECTED_HEADER:
354 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
355 				nd6log((LOG_INFO,
356 				    "duplicated ND6 option found (type=%d)\n",
357 				    nd_opt->nd_opt_type));
358 				/* XXX bark? */
359 			} else {
360 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
361 					= nd_opt;
362 			}
363 			break;
364 		case ND_OPT_PREFIX_INFORMATION:
365 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
366 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
367 					= nd_opt;
368 			}
369 			ndopts->nd_opts_pi_end =
370 				(struct nd_opt_prefix_info *)nd_opt;
371 			break;
372 		default:
373 			/*
374 			 * Unknown options must be silently ignored,
375 			 * to accomodate future extension to the protocol.
376 			 */
377 			nd6log((LOG_DEBUG,
378 			    "nd6_options: unsupported option %d - "
379 			    "option ignored\n", nd_opt->nd_opt_type));
380 		}
381 
382 skip1:
383 		i++;
384 		if (i > nd6_maxndopt) {
385 			icmp6stat.icp6s_nd_toomanyopt++;
386 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
387 			break;
388 		}
389 
390 		if (ndopts->nd_opts_done)
391 			break;
392 	}
393 
394 	return 0;
395 }
396 
397 /*
398  * ND6 timer routine to expire default route list and prefix list
399  */
400 void
401 nd6_timer(ignored_arg)
402 	void	*ignored_arg;
403 {
404 	int s;
405 	struct llinfo_nd6 *ln;
406 	struct nd_defrouter *dr;
407 	struct nd_prefix *pr;
408 	struct ifnet *ifp;
409 	struct in6_ifaddr *ia6, *nia6;
410 	struct in6_addrlifetime *lt6;
411 
412 	s = splnet();
413 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
414 	    nd6_timer, NULL);
415 
416 	ln = llinfo_nd6.ln_next;
417 	while (ln && ln != &llinfo_nd6) {
418 		struct rtentry *rt;
419 		struct sockaddr_in6 *dst;
420 		struct llinfo_nd6 *next = ln->ln_next;
421 		/* XXX: used for the DELAY case only: */
422 		struct nd_ifinfo *ndi = NULL;
423 
424 		if ((rt = ln->ln_rt) == NULL) {
425 			ln = next;
426 			continue;
427 		}
428 		if ((ifp = rt->rt_ifp) == NULL) {
429 			ln = next;
430 			continue;
431 		}
432 		ndi = ND_IFINFO(ifp);
433 		dst = (struct sockaddr_in6 *)rt_key(rt);
434 
435 		if (ln->ln_expire > time_second) {
436 			ln = next;
437 			continue;
438 		}
439 
440 		/* sanity check */
441 		if (!rt)
442 			panic("rt=0 in nd6_timer(ln=%p)", ln);
443 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
444 			panic("rt_llinfo(%p) is not equal to ln(%p)",
445 			      rt->rt_llinfo, ln);
446 		if (!dst)
447 			panic("dst=0 in nd6_timer(ln=%p)", ln);
448 
449 		switch (ln->ln_state) {
450 		case ND6_LLINFO_INCOMPLETE:
451 			if (ln->ln_asked < nd6_mmaxtries) {
452 				ln->ln_asked++;
453 				ln->ln_expire = time_second +
454 					ND_IFINFO(ifp)->retrans / 1000;
455 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
456 					ln, 0);
457 			} else {
458 				struct mbuf *m = ln->ln_hold;
459 				if (m) {
460 					if (rt->rt_ifp) {
461 						/*
462 						 * Fake rcvif to make ICMP error
463 						 * more helpful in diagnosing
464 						 * for the receiver.
465 						 * XXX: should we consider
466 						 * older rcvif?
467 						 */
468 						m->m_pkthdr.rcvif = rt->rt_ifp;
469 					}
470 					icmp6_error(m, ICMP6_DST_UNREACH,
471 						    ICMP6_DST_UNREACH_ADDR, 0);
472 					ln->ln_hold = NULL;
473 				}
474 				next = nd6_free(rt);
475 			}
476 			break;
477 		case ND6_LLINFO_REACHABLE:
478 			if (ln->ln_expire) {
479 				ln->ln_state = ND6_LLINFO_STALE;
480 				ln->ln_expire = time_second + nd6_gctimer;
481 			}
482 			break;
483 
484 		case ND6_LLINFO_STALE:
485 			/* Garbage Collection(RFC 2461 5.3) */
486 			if (ln->ln_expire)
487 				next = nd6_free(rt);
488 			break;
489 
490 		case ND6_LLINFO_DELAY:
491 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
492 				/* We need NUD */
493 				ln->ln_asked = 1;
494 				ln->ln_state = ND6_LLINFO_PROBE;
495 				ln->ln_expire = time_second +
496 					ndi->retrans / 1000;
497 				nd6_ns_output(ifp, &dst->sin6_addr,
498 					      &dst->sin6_addr,
499 					      ln, 0);
500 			} else {
501 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
502 				ln->ln_expire = time_second + nd6_gctimer;
503 			}
504 			break;
505 		case ND6_LLINFO_PROBE:
506 			if (ln->ln_asked < nd6_umaxtries) {
507 				ln->ln_asked++;
508 				ln->ln_expire = time_second +
509 					ND_IFINFO(ifp)->retrans / 1000;
510 				nd6_ns_output(ifp, &dst->sin6_addr,
511 					       &dst->sin6_addr, ln, 0);
512 			} else {
513 				next = nd6_free(rt);
514 			}
515 			break;
516 		}
517 		ln = next;
518 	}
519 
520 	/* expire default router list */
521 	dr = TAILQ_FIRST(&nd_defrouter);
522 	while (dr) {
523 		if (dr->expire && dr->expire < time_second) {
524 			struct nd_defrouter *t;
525 			t = TAILQ_NEXT(dr, dr_entry);
526 			defrtrlist_del(dr);
527 			dr = t;
528 		} else {
529 			dr = TAILQ_NEXT(dr, dr_entry);
530 		}
531 	}
532 
533 	/*
534 	 * expire interface addresses.
535 	 * in the past the loop was inside prefix expiry processing.
536 	 * However, from a stricter speci-confrmance standpoint, we should
537 	 * rather separate address lifetimes and prefix lifetimes.
538 	 */
539   addrloop:
540 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
541 		nia6 = ia6->ia_next;
542 		/* check address lifetime */
543 		lt6 = &ia6->ia6_lifetime;
544 		if (IFA6_IS_INVALID(ia6)) {
545 			int regen = 0;
546 
547 			/*
548 			 * If the expiring address is temporary, try
549 			 * regenerating a new one.  This would be useful when
550 			 * we suspended a laptop PC, then turned it on after a
551 			 * period that could invalidate all temporary
552 			 * addresses.  Although we may have to restart the
553 			 * loop (see below), it must be after purging the
554 			 * address.  Otherwise, we'd see an infinite loop of
555 			 * regeneration.
556 			 */
557 			if (ip6_use_tempaddr &&
558 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
559 				if (regen_tmpaddr(ia6) == 0)
560 					regen = 1;
561 			}
562 
563 			in6_purgeaddr(&ia6->ia_ifa);
564 
565 			if (regen)
566 				goto addrloop; /* XXX: see below */
567 		}
568 		if (IFA6_IS_DEPRECATED(ia6)) {
569 			int oldflags = ia6->ia6_flags;
570 
571 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
572 
573 			/*
574 			 * If a temporary address has just become deprecated,
575 			 * regenerate a new one if possible.
576 			 */
577 			if (ip6_use_tempaddr &&
578 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
579 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
580 
581 				if (regen_tmpaddr(ia6) == 0) {
582 					/*
583 					 * A new temporary address is
584 					 * generated.
585 					 * XXX: this means the address chain
586 					 * has changed while we are still in
587 					 * the loop.  Although the change
588 					 * would not cause disaster (because
589 					 * it's not a deletion, but an
590 					 * addition,) we'd rather restart the
591 					 * loop just for safety.  Or does this
592 					 * significantly reduce performance??
593 					 */
594 					goto addrloop;
595 				}
596 			}
597 		} else {
598 			/*
599 			 * A new RA might have made a deprecated address
600 			 * preferred.
601 			 */
602 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
603 		}
604 	}
605 
606 	/* expire prefix list */
607 	pr = nd_prefix.lh_first;
608 	while (pr) {
609 		/*
610 		 * check prefix lifetime.
611 		 * since pltime is just for autoconf, pltime processing for
612 		 * prefix is not necessary.
613 		 */
614 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
615 			struct nd_prefix *t;
616 			t = pr->ndpr_next;
617 
618 			/*
619 			 * address expiration and prefix expiration are
620 			 * separate.  NEVER perform in6_purgeaddr here.
621 			 */
622 
623 			prelist_remove(pr);
624 			pr = t;
625 		} else
626 			pr = pr->ndpr_next;
627 	}
628 	splx(s);
629 }
630 
631 static int
632 regen_tmpaddr(ia6)
633 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
634 {
635 	struct ifaddr *ifa;
636 	struct ifnet *ifp;
637 	struct in6_ifaddr *public_ifa6 = NULL;
638 
639 	ifp = ia6->ia_ifa.ifa_ifp;
640 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
641 	     ifa = ifa->ifa_list.tqe_next) {
642 		struct in6_ifaddr *it6;
643 
644 		if (ifa->ifa_addr->sa_family != AF_INET6)
645 			continue;
646 
647 		it6 = (struct in6_ifaddr *)ifa;
648 
649 		/* ignore no autoconf addresses. */
650 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
651 			continue;
652 
653 		/* ignore autoconf addresses with different prefixes. */
654 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
655 			continue;
656 
657 		/*
658 		 * Now we are looking at an autoconf address with the same
659 		 * prefix as ours.  If the address is temporary and is still
660 		 * preferred, do not create another one.  It would be rare, but
661 		 * could happen, for example, when we resume a laptop PC after
662 		 * a long period.
663 		 */
664 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
665 		    !IFA6_IS_DEPRECATED(it6)) {
666 			public_ifa6 = NULL;
667 			break;
668 		}
669 
670 		/*
671 		 * This is a public autoconf address that has the same prefix
672 		 * as ours.  If it is preferred, keep it.  We can't break the
673 		 * loop here, because there may be a still-preferred temporary
674 		 * address with the prefix.
675 		 */
676 		if (!IFA6_IS_DEPRECATED(it6))
677 		    public_ifa6 = it6;
678 	}
679 
680 	if (public_ifa6 != NULL) {
681 		int e;
682 
683 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
684 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
685 			    " tmp addr,errno=%d\n", e);
686 			return (-1);
687 		}
688 		return (0);
689 	}
690 
691 	return (-1);
692 }
693 
694 /*
695  * Nuke neighbor cache/prefix/default router management table, right before
696  * ifp goes away.
697  */
698 void
699 nd6_purge(ifp)
700 	struct ifnet *ifp;
701 {
702 	struct llinfo_nd6 *ln, *nln;
703 	struct nd_defrouter *dr, *ndr, drany;
704 	struct nd_prefix *pr, *npr;
705 
706 	/* Nuke default router list entries toward ifp */
707 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
708 		/*
709 		 * The first entry of the list may be stored in
710 		 * the routing table, so we'll delete it later.
711 		 */
712 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
713 			ndr = TAILQ_NEXT(dr, dr_entry);
714 			if (dr->ifp == ifp)
715 				defrtrlist_del(dr);
716 		}
717 		dr = TAILQ_FIRST(&nd_defrouter);
718 		if (dr->ifp == ifp)
719 			defrtrlist_del(dr);
720 	}
721 
722 	/* Nuke prefix list entries toward ifp */
723 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
724 		npr = pr->ndpr_next;
725 		if (pr->ndpr_ifp == ifp) {
726 			/*
727 			 * Previously, pr->ndpr_addr is removed as well,
728 			 * but I strongly believe we don't have to do it.
729 			 * nd6_purge() is only called from in6_ifdetach(),
730 			 * which removes all the associated interface addresses
731 			 * by itself.
732 			 * (jinmei@kame.net 20010129)
733 			 */
734 			prelist_remove(pr);
735 		}
736 	}
737 
738 	/* cancel default outgoing interface setting */
739 	if (nd6_defifindex == ifp->if_index)
740 		nd6_setdefaultiface(0);
741 
742 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
743 		/* refresh default router list */
744 		bzero(&drany, sizeof(drany));
745 		defrouter_delreq(&drany, 0);
746 		defrouter_select();
747 	}
748 
749 	/*
750 	 * Nuke neighbor cache entries for the ifp.
751 	 * Note that rt->rt_ifp may not be the same as ifp,
752 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
753 	 * nd6_rtrequest(), and ip6_input().
754 	 */
755 	ln = llinfo_nd6.ln_next;
756 	while (ln && ln != &llinfo_nd6) {
757 		struct rtentry *rt;
758 		struct sockaddr_dl *sdl;
759 
760 		nln = ln->ln_next;
761 		rt = ln->ln_rt;
762 		if (rt && rt->rt_gateway &&
763 		    rt->rt_gateway->sa_family == AF_LINK) {
764 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
765 			if (sdl->sdl_index == ifp->if_index)
766 				nln = nd6_free(rt);
767 		}
768 		ln = nln;
769 	}
770 }
771 
772 struct rtentry *
773 nd6_lookup(addr6, create, ifp)
774 	struct in6_addr *addr6;
775 	int create;
776 	struct ifnet *ifp;
777 {
778 	struct rtentry *rt;
779 	struct sockaddr_in6 sin6;
780 
781 	bzero(&sin6, sizeof(sin6));
782 	sin6.sin6_len = sizeof(struct sockaddr_in6);
783 	sin6.sin6_family = AF_INET6;
784 	sin6.sin6_addr = *addr6;
785 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
786 	if (rt) {
787 		if ((rt->rt_flags & RTF_LLINFO) == 0 && create) {
788 			/*
789 			 * This is the case for the default route.
790 			 * If we want to create a neighbor cache for the
791 			 * address, we should free the route for the
792 			 * destination and allocate an interface route.
793 			 */
794 			RTFREE_LOCKED(rt);
795 			rt = 0;
796 		}
797 	}
798 	if (!rt) {
799 		if (create && ifp) {
800 			int e;
801 
802 			/*
803 			 * If no route is available and create is set,
804 			 * we allocate a host route for the destination
805 			 * and treat it like an interface route.
806 			 * This hack is necessary for a neighbor which can't
807 			 * be covered by our own prefix.
808 			 */
809 			struct ifaddr *ifa =
810 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
811 			if (ifa == NULL)
812 				return (NULL);
813 
814 			/*
815 			 * Create a new route.  RTF_LLINFO is necessary
816 			 * to create a Neighbor Cache entry for the
817 			 * destination in nd6_rtrequest which will be
818 			 * called in rtrequest via ifa->ifa_rtrequest.
819 			 */
820 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
821 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
822 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
823 			    ~RTF_CLONING, &rt)) != 0) {
824 				log(LOG_ERR,
825 				    "nd6_lookup: failed to add route for a "
826 				    "neighbor(%s), errno=%d\n",
827 				    ip6_sprintf(addr6), e);
828 			}
829 			if (rt == NULL)
830 				return (NULL);
831 			RT_LOCK(rt);
832 			if (rt->rt_llinfo) {
833 				struct llinfo_nd6 *ln =
834 				    (struct llinfo_nd6 *)rt->rt_llinfo;
835 				ln->ln_state = ND6_LLINFO_NOSTATE;
836 			}
837 		} else
838 			return (NULL);
839 	}
840 	RT_LOCK_ASSERT(rt);
841 	RT_REMREF(rt);
842 	/*
843 	 * Validation for the entry.
844 	 * Note that the check for rt_llinfo is necessary because a cloned
845 	 * route from a parent route that has the L flag (e.g. the default
846 	 * route to a p2p interface) may have the flag, too, while the
847 	 * destination is not actually a neighbor.
848 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
849 	 *      it might be the loopback interface if the entry is for our
850 	 *      own address on a non-loopback interface. Instead, we should
851 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
852 	 *	interface.
853 	 */
854 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
855 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
856 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
857 		if (create) {
858 			log(LOG_DEBUG,
859 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
860 			    ip6_sprintf(addr6),
861 			    ifp ? if_name(ifp) : "unspec");
862 			/* xxx more logs... kazu */
863 		}
864 		RT_UNLOCK(rt);
865 		return (NULL);
866 	}
867 	RT_UNLOCK(rt);		/* XXX not ready to return rt locked */
868 	return (rt);
869 }
870 
871 /*
872  * Detect if a given IPv6 address identifies a neighbor on a given link.
873  * XXX: should take care of the destination of a p2p link?
874  */
875 int
876 nd6_is_addr_neighbor(addr, ifp)
877 	struct sockaddr_in6 *addr;
878 	struct ifnet *ifp;
879 {
880 	struct nd_prefix *pr;
881 
882 	/*
883 	 * A link-local address is always a neighbor.
884 	 * XXX: we should use the sin6_scope_id field rather than the embedded
885 	 * interface index.
886 	 * XXX: a link does not necessarily specify a single interface.
887 	 */
888 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
889 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
890 		return (1);
891 
892 	/*
893 	 * If the address matches one of our addresses,
894 	 * it should be a neighbor.
895 	 * If the address matches one of our on-link prefixes, it should be a
896 	 * neighbor.
897 	 */
898 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
899 		if (pr->ndpr_ifp != ifp)
900 			continue;
901 
902 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
903 			continue;
904 
905 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
906 		    &addr->sin6_addr, &pr->ndpr_mask))
907 			return (1);
908 	}
909 
910 	/*
911 	 * If the default router list is empty, all addresses are regarded
912 	 * as on-link, and thus, as a neighbor.
913 	 * XXX: we restrict the condition to hosts, because routers usually do
914 	 * not have the "default router list".
915 	 */
916 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
917 	    nd6_defifindex == ifp->if_index) {
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 }
930 
931 /*
932  * Free an nd6 llinfo entry.
933  */
934 struct llinfo_nd6 *
935 nd6_free(rt)
936 	struct rtentry *rt;
937 {
938 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
939 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
940 	struct nd_defrouter *dr;
941 
942 	/*
943 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
944 	 * even though it is not harmful, it was not really necessary.
945 	 */
946 
947 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
948 		int s;
949 		s = splnet();
950 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
951 		    rt->rt_ifp);
952 
953 		if (ln->ln_router || dr) {
954 			/*
955 			 * rt6_flush must be called whether or not the neighbor
956 			 * is in the Default Router List.
957 			 * See a corresponding comment in nd6_na_input().
958 			 */
959 			rt6_flush(&in6, rt->rt_ifp);
960 		}
961 
962 		if (dr) {
963 			/*
964 			 * Unreachablity of a router might affect the default
965 			 * router selection and on-link detection of advertised
966 			 * prefixes.
967 			 */
968 
969 			/*
970 			 * Temporarily fake the state to choose a new default
971 			 * router and to perform on-link determination of
972 			 * prefixes correctly.
973 			 * Below the state will be set correctly,
974 			 * or the entry itself will be deleted.
975 			 */
976 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
977 
978 			/*
979 			 * Since defrouter_select() does not affect the
980 			 * on-link determination and MIP6 needs the check
981 			 * before the default router selection, we perform
982 			 * the check now.
983 			 */
984 			pfxlist_onlink_check();
985 
986 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
987 				/*
988 				 * It is used as the current default router,
989 				 * so we have to move it to the end of the
990 				 * list and choose a new one.
991 				 * XXX: it is not very efficient if this is
992 				 *      the only router.
993 				 */
994 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
995 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
996 
997 				defrouter_select();
998 			}
999 		}
1000 		splx(s);
1001 	}
1002 
1003 	/*
1004 	 * Before deleting the entry, remember the next entry as the
1005 	 * return value.  We need this because pfxlist_onlink_check() above
1006 	 * might have freed other entries (particularly the old next entry) as
1007 	 * a side effect (XXX).
1008 	 */
1009 	next = ln->ln_next;
1010 
1011 	/*
1012 	 * Detach the route from the routing tree and the list of neighbor
1013 	 * caches, and disable the route entry not to be used in already
1014 	 * cached routes.
1015 	 */
1016 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1017 	    rt_mask(rt), 0, (struct rtentry **)0);
1018 
1019 	return (next);
1020 }
1021 
1022 /*
1023  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1024  *
1025  * XXX cost-effective metods?
1026  */
1027 void
1028 nd6_nud_hint(rt, dst6, force)
1029 	struct rtentry *rt;
1030 	struct in6_addr *dst6;
1031 	int force;
1032 {
1033 	struct llinfo_nd6 *ln;
1034 
1035 	/*
1036 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1037 	 * routing table by supplied "dst6".
1038 	 */
1039 	if (!rt) {
1040 		if (!dst6)
1041 			return;
1042 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1043 			return;
1044 	}
1045 
1046 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1047 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1048 	    !rt->rt_llinfo || !rt->rt_gateway ||
1049 	    rt->rt_gateway->sa_family != AF_LINK) {
1050 		/* This is not a host route. */
1051 		return;
1052 	}
1053 
1054 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1055 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1056 		return;
1057 
1058 	/*
1059 	 * if we get upper-layer reachability confirmation many times,
1060 	 * it is possible we have false information.
1061 	 */
1062 	if (!force) {
1063 		ln->ln_byhint++;
1064 		if (ln->ln_byhint > nd6_maxnudhint)
1065 			return;
1066 	}
1067 
1068 	ln->ln_state = ND6_LLINFO_REACHABLE;
1069 	if (ln->ln_expire)
1070 		ln->ln_expire = time_second +
1071 			ND_IFINFO(rt->rt_ifp)->reachable;
1072 }
1073 
1074 void
1075 nd6_rtrequest(req, rt, info)
1076 	int	req;
1077 	struct rtentry *rt;
1078 	struct rt_addrinfo *info; /* xxx unused */
1079 {
1080 	struct sockaddr *gate = rt->rt_gateway;
1081 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1082 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1083 	struct ifnet *ifp = rt->rt_ifp;
1084 	struct ifaddr *ifa;
1085 
1086 	RT_LOCK_ASSERT(rt);
1087 
1088 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1089 		return;
1090 
1091 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1092 		/*
1093 		 * This is probably an interface direct route for a link
1094 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1095 		 * We do not need special treatment below for such a route.
1096 		 * Moreover, the RTF_LLINFO flag which would be set below
1097 		 * would annoy the ndp(8) command.
1098 		 */
1099 		return;
1100 	}
1101 
1102 	if (req == RTM_RESOLVE &&
1103 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1104 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1105 		/*
1106 		 * FreeBSD and BSD/OS often make a cloned host route based
1107 		 * on a less-specific route (e.g. the default route).
1108 		 * If the less specific route does not have a "gateway"
1109 		 * (this is the case when the route just goes to a p2p or an
1110 		 * stf interface), we'll mistakenly make a neighbor cache for
1111 		 * the host route, and will see strange neighbor solicitation
1112 		 * for the corresponding destination.  In order to avoid the
1113 		 * confusion, we check if the destination of the route is
1114 		 * a neighbor in terms of neighbor discovery, and stop the
1115 		 * process if not.  Additionally, we remove the LLINFO flag
1116 		 * so that ndp(8) will not try to get the neighbor information
1117 		 * of the destination.
1118 		 */
1119 		rt->rt_flags &= ~RTF_LLINFO;
1120 		return;
1121 	}
1122 
1123 	switch (req) {
1124 	case RTM_ADD:
1125 		/*
1126 		 * There is no backward compatibility :)
1127 		 *
1128 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1129 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1130 		 *	   rt->rt_flags |= RTF_CLONING;
1131 		 */
1132 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1133 			/*
1134 			 * Case 1: This route should come from
1135 			 * a route to interface.  RTF_LLINFO flag is set
1136 			 * for a host route whose destination should be
1137 			 * treated as on-link.
1138 			 */
1139 			rt_setgate(rt, rt_key(rt),
1140 				   (struct sockaddr *)&null_sdl);
1141 			gate = rt->rt_gateway;
1142 			SDL(gate)->sdl_type = ifp->if_type;
1143 			SDL(gate)->sdl_index = ifp->if_index;
1144 			if (ln)
1145 				ln->ln_expire = time_second;
1146 			if (ln && ln->ln_expire == 0) {
1147 				/* kludge for desktops */
1148 				ln->ln_expire = 1;
1149 			}
1150 			if ((rt->rt_flags & RTF_CLONING) != 0)
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(cmd, data, ifp)
1327 	u_long cmd;
1328 	caddr_t	data;
1329 	struct ifnet *ifp;
1330 {
1331 	struct in6_drlist *drl = (struct in6_drlist *)data;
1332 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1333 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1334 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1335 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1336 	struct nd_defrouter *dr, any;
1337 	struct nd_prefix *pr;
1338 	struct rtentry *rt;
1339 	int i = 0, error = 0;
1340 	int s;
1341 
1342 	switch (cmd) {
1343 	case SIOCGDRLST_IN6:
1344 		/*
1345 		 * obsolete API, use sysctl under net.inet6.icmp6
1346 		 */
1347 		bzero(drl, sizeof(*drl));
1348 		s = splnet();
1349 		dr = TAILQ_FIRST(&nd_defrouter);
1350 		while (dr && i < DRLSTSIZ) {
1351 			drl->defrouter[i].rtaddr = dr->rtaddr;
1352 			in6_clearscope(&drl->defrouter[i].rtaddr);
1353 
1354 			drl->defrouter[i].flags = dr->flags;
1355 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1356 			drl->defrouter[i].expire = dr->expire;
1357 			drl->defrouter[i].if_index = dr->ifp->if_index;
1358 			i++;
1359 			dr = TAILQ_NEXT(dr, dr_entry);
1360 		}
1361 		splx(s);
1362 		break;
1363 	case SIOCGPRLST_IN6:
1364 		/*
1365 		 * obsolete API, use sysctl under net.inet6.icmp6
1366 		 *
1367 		 * XXX the structure in6_prlist was changed in backward-
1368 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1369 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
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(oprl, sizeof(*oprl));
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(&oprl->prefix[i].prefix,
1384 			    &pr->ndpr_prefix, NULL, NULL);
1385 			oprl->prefix[i].raflags = pr->ndpr_raf;
1386 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1387 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1388 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1389 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1390 			oprl->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 oprl->prefix[i].advrtr[j]
1397 					RTRADDR = pfr->router->rtaddr;
1398 					in6_clearscope(&RTRADDR);
1399 #undef RTRADDR
1400 				}
1401 				j++;
1402 				pfr = pfr->pfr_next;
1403 			}
1404 			oprl->prefix[i].advrtrs = j;
1405 			oprl->prefix[i].origin = PR_ORIG_RA;
1406 
1407 			i++;
1408 			pr = pr->ndpr_next;
1409 		}
1410 	      {
1411 		struct rr_prefix *rpp;
1412 
1413 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1414 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1415 			if (i >= PRLSTSIZ)
1416 				break;
1417 			(void)in6_embedscope(&oprl->prefix[i].prefix,
1418 			    &pr->ndpr_prefix, NULL, NULL);
1419 			oprl->prefix[i].raflags = rpp->rp_raf;
1420 			oprl->prefix[i].prefixlen = rpp->rp_plen;
1421 			oprl->prefix[i].vltime = rpp->rp_vltime;
1422 			oprl->prefix[i].pltime = rpp->rp_pltime;
1423 			oprl->prefix[i].if_index = rpp->rp_ifp->if_index;
1424 			oprl->prefix[i].expire = rpp->rp_expire;
1425 			oprl->prefix[i].advrtrs = 0;
1426 			oprl->prefix[i].origin = rpp->rp_origin;
1427 			i++;
1428 		}
1429 	      }
1430 		splx(s);
1431 
1432 		break;
1433 	case OSIOCGIFINFO_IN6:
1434 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1435 		bzero(&ndi->ndi, sizeof(ndi->ndi));
1436 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1437 		ndi->ndi.maxmtu = ND_IFINFO(ifp)->maxmtu;
1438 		ndi->ndi.basereachable = ND_IFINFO(ifp)->basereachable;
1439 		ndi->ndi.reachable = ND_IFINFO(ifp)->reachable;
1440 		ndi->ndi.retrans = ND_IFINFO(ifp)->retrans;
1441 		ndi->ndi.flags = ND_IFINFO(ifp)->flags;
1442 		ndi->ndi.recalctm = ND_IFINFO(ifp)->recalctm;
1443 		ndi->ndi.chlim = ND_IFINFO(ifp)->chlim;
1444 		break;
1445 	case SIOCGIFINFO_IN6:
1446 		ndi->ndi = *ND_IFINFO(ifp);
1447 		ndi->ndi.linkmtu = IN6_LINKMTU(ifp);
1448 		break;
1449 	case SIOCSIFINFO_FLAGS:
1450 		ND_IFINFO(ifp)->flags = ndi->ndi.flags;
1451 		break;
1452 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1453 		/* flush default router list */
1454 		/*
1455 		 * xxx sumikawa: should not delete route if default
1456 		 * route equals to the top of default router list
1457 		 */
1458 		bzero(&any, sizeof(any));
1459 		defrouter_delreq(&any, 0);
1460 		defrouter_select();
1461 		/* xxx sumikawa: flush prefix list */
1462 		break;
1463 	case SIOCSPFXFLUSH_IN6:
1464 	{
1465 		/* flush all the prefix advertised by routers */
1466 		struct nd_prefix *pr, *next;
1467 
1468 		s = splnet();
1469 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1470 			struct in6_ifaddr *ia, *ia_next;
1471 
1472 			next = pr->ndpr_next;
1473 
1474 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1475 				continue; /* XXX */
1476 
1477 			/* do we really have to remove addresses as well? */
1478 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1479 				/* ia might be removed.  keep the next ptr. */
1480 				ia_next = ia->ia_next;
1481 
1482 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1483 					continue;
1484 
1485 				if (ia->ia6_ndpr == pr)
1486 					in6_purgeaddr(&ia->ia_ifa);
1487 			}
1488 			prelist_remove(pr);
1489 		}
1490 		splx(s);
1491 		break;
1492 	}
1493 	case SIOCSRTRFLUSH_IN6:
1494 	{
1495 		/* flush all the default routers */
1496 		struct nd_defrouter *dr, *next;
1497 
1498 		s = splnet();
1499 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1500 			/*
1501 			 * The first entry of the list may be stored in
1502 			 * the routing table, so we'll delete it later.
1503 			 */
1504 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1505 				next = TAILQ_NEXT(dr, dr_entry);
1506 				defrtrlist_del(dr);
1507 			}
1508 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1509 		}
1510 		splx(s);
1511 		break;
1512 	}
1513 	case SIOCGNBRINFO_IN6:
1514 	{
1515 		struct llinfo_nd6 *ln;
1516 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1517 
1518 		/*
1519 		 * XXX: KAME specific hack for scoped addresses
1520 		 *      XXXX: for other scopes than link-local?
1521 		 */
1522 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1523 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1524 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1525 
1526 			if (*idp == 0)
1527 				*idp = htons(ifp->if_index);
1528 		}
1529 
1530 		s = splnet();
1531 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1532 			error = EINVAL;
1533 			splx(s);
1534 			break;
1535 		}
1536 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1537 		nbi->state = ln->ln_state;
1538 		nbi->asked = ln->ln_asked;
1539 		nbi->isrouter = ln->ln_router;
1540 		nbi->expire = ln->ln_expire;
1541 		splx(s);
1542 
1543 		break;
1544 	}
1545 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1546 		ndif->ifindex = nd6_defifindex;
1547 		break;
1548 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1549 		return (nd6_setdefaultiface(ndif->ifindex));
1550 	}
1551 	return (error);
1552 }
1553 
1554 /*
1555  * Create neighbor cache entry and cache link-layer address,
1556  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1557  */
1558 struct rtentry *
1559 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1560 	struct ifnet *ifp;
1561 	struct in6_addr *from;
1562 	char *lladdr;
1563 	int lladdrlen;
1564 	int type;	/* ICMP6 type */
1565 	int code;	/* type dependent information */
1566 {
1567 	struct rtentry *rt = NULL;
1568 	struct llinfo_nd6 *ln = NULL;
1569 	int is_newentry;
1570 	struct sockaddr_dl *sdl = NULL;
1571 	int do_update;
1572 	int olladdr;
1573 	int llchange;
1574 	int newstate = 0;
1575 
1576 	if (!ifp)
1577 		panic("ifp == NULL in nd6_cache_lladdr");
1578 	if (!from)
1579 		panic("from == NULL in nd6_cache_lladdr");
1580 
1581 	/* nothing must be updated for unspecified address */
1582 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1583 		return NULL;
1584 
1585 	/*
1586 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1587 	 * the caller.
1588 	 *
1589 	 * XXX If the link does not have link-layer adderss, what should
1590 	 * we do? (ifp->if_addrlen == 0)
1591 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1592 	 * description on it in NS section (RFC 2461 7.2.3).
1593 	 */
1594 
1595 	rt = nd6_lookup(from, 0, ifp);
1596 	if (!rt) {
1597 #if 0
1598 		/* nothing must be done if there's no lladdr */
1599 		if (!lladdr || !lladdrlen)
1600 			return NULL;
1601 #endif
1602 
1603 		rt = nd6_lookup(from, 1, ifp);
1604 		is_newentry = 1;
1605 	} else {
1606 		/* do nothing if static ndp is set */
1607 		if (rt->rt_flags & RTF_STATIC)
1608 			return NULL;
1609 		is_newentry = 0;
1610 	}
1611 
1612 	if (!rt)
1613 		return NULL;
1614 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1615 fail:
1616 		(void)nd6_free(rt);
1617 		return NULL;
1618 	}
1619 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1620 	if (!ln)
1621 		goto fail;
1622 	if (!rt->rt_gateway)
1623 		goto fail;
1624 	if (rt->rt_gateway->sa_family != AF_LINK)
1625 		goto fail;
1626 	sdl = SDL(rt->rt_gateway);
1627 
1628 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1629 	if (olladdr && lladdr) {
1630 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1631 			llchange = 1;
1632 		else
1633 			llchange = 0;
1634 	} else
1635 		llchange = 0;
1636 
1637 	/*
1638 	 * newentry olladdr  lladdr  llchange	(*=record)
1639 	 *	0	n	n	--	(1)
1640 	 *	0	y	n	--	(2)
1641 	 *	0	n	y	--	(3) * STALE
1642 	 *	0	y	y	n	(4) *
1643 	 *	0	y	y	y	(5) * STALE
1644 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1645 	 *	1	--	y	--	(7) * STALE
1646 	 */
1647 
1648 	if (lladdr) {		/* (3-5) and (7) */
1649 		/*
1650 		 * Record source link-layer address
1651 		 * XXX is it dependent to ifp->if_type?
1652 		 */
1653 		sdl->sdl_alen = ifp->if_addrlen;
1654 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1655 	}
1656 
1657 	if (!is_newentry) {
1658 		if ((!olladdr && lladdr) ||		/* (3) */
1659 		    (olladdr && lladdr && llchange)) {	/* (5) */
1660 			do_update = 1;
1661 			newstate = ND6_LLINFO_STALE;
1662 		} else					/* (1-2,4) */
1663 			do_update = 0;
1664 	} else {
1665 		do_update = 1;
1666 		if (!lladdr)				/* (6) */
1667 			newstate = ND6_LLINFO_NOSTATE;
1668 		else					/* (7) */
1669 			newstate = ND6_LLINFO_STALE;
1670 	}
1671 
1672 	if (do_update) {
1673 		/*
1674 		 * Update the state of the neighbor cache.
1675 		 */
1676 		ln->ln_state = newstate;
1677 
1678 		if (ln->ln_state == ND6_LLINFO_STALE) {
1679 			/*
1680 			 * XXX: since nd6_output() below will cause
1681 			 * state tansition to DELAY and reset the timer,
1682 			 * we must set the timer now, although it is actually
1683 			 * meaningless.
1684 			 */
1685 			ln->ln_expire = time_second + nd6_gctimer;
1686 
1687 			if (ln->ln_hold) {
1688 				/*
1689 				 * we assume ifp is not a p2p here, so just
1690 				 * set the 2nd argument as the 1st one.
1691 				 */
1692 				nd6_output(ifp, ifp, ln->ln_hold,
1693 				    (struct sockaddr_in6 *)rt_key(rt), rt);
1694 				ln->ln_hold = NULL;
1695 			}
1696 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1697 			/* probe right away */
1698 			ln->ln_expire = time_second;
1699 		}
1700 	}
1701 
1702 	/*
1703 	 * ICMP6 type dependent behavior.
1704 	 *
1705 	 * NS: clear IsRouter if new entry
1706 	 * RS: clear IsRouter
1707 	 * RA: set IsRouter if there's lladdr
1708 	 * redir: clear IsRouter if new entry
1709 	 *
1710 	 * RA case, (1):
1711 	 * The spec says that we must set IsRouter in the following cases:
1712 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1713 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1714 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1715 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1716 	 * neighbor cache, this is similar to (6).
1717 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1718 	 *
1719 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1720 	 *							D R
1721 	 *	0	n	n	--	(1)	c   ?     s
1722 	 *	0	y	n	--	(2)	c   s     s
1723 	 *	0	n	y	--	(3)	c   s     s
1724 	 *	0	y	y	n	(4)	c   s     s
1725 	 *	0	y	y	y	(5)	c   s     s
1726 	 *	1	--	n	--	(6) c	c 	c s
1727 	 *	1	--	y	--	(7) c	c   s	c s
1728 	 *
1729 	 *					(c=clear s=set)
1730 	 */
1731 	switch (type & 0xff) {
1732 	case ND_NEIGHBOR_SOLICIT:
1733 		/*
1734 		 * New entry must have is_router flag cleared.
1735 		 */
1736 		if (is_newentry)	/* (6-7) */
1737 			ln->ln_router = 0;
1738 		break;
1739 	case ND_REDIRECT:
1740 		/*
1741 		 * If the icmp is a redirect to a better router, always set the
1742 		 * is_router flag.  Otherwise, if the entry is newly created,
1743 		 * clear the flag.  [RFC 2461, sec 8.3]
1744 		 */
1745 		if (code == ND_REDIRECT_ROUTER)
1746 			ln->ln_router = 1;
1747 		else if (is_newentry) /* (6-7) */
1748 			ln->ln_router = 0;
1749 		break;
1750 	case ND_ROUTER_SOLICIT:
1751 		/*
1752 		 * is_router flag must always be cleared.
1753 		 */
1754 		ln->ln_router = 0;
1755 		break;
1756 	case ND_ROUTER_ADVERT:
1757 		/*
1758 		 * Mark an entry with lladdr as a router.
1759 		 */
1760 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1761 		    (is_newentry && lladdr)) {			/* (7) */
1762 			ln->ln_router = 1;
1763 		}
1764 		break;
1765 	}
1766 
1767 	/*
1768 	 * When the link-layer address of a router changes, select the
1769 	 * best router again.  In particular, when the neighbor entry is newly
1770 	 * created, it might affect the selection policy.
1771 	 * Question: can we restrict the first condition to the "is_newentry"
1772 	 * case?
1773 	 * XXX: when we hear an RA from a new router with the link-layer
1774 	 * address option, defrouter_select() is called twice, since
1775 	 * defrtrlist_update called the function as well.  However, I believe
1776 	 * we can compromise the overhead, since it only happens the first
1777 	 * time.
1778 	 * XXX: although defrouter_select() should not have a bad effect
1779 	 * for those are not autoconfigured hosts, we explicitly avoid such
1780 	 * cases for safety.
1781 	 */
1782 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1783 		defrouter_select();
1784 
1785 	return rt;
1786 }
1787 
1788 static void
1789 nd6_slowtimo(ignored_arg)
1790     void *ignored_arg;
1791 {
1792 	int s = splnet();
1793 	struct nd_ifinfo *nd6if;
1794 	struct ifnet *ifp;
1795 
1796 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1797 	    nd6_slowtimo, NULL);
1798 	IFNET_RLOCK();
1799 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1800 		nd6if = ND_IFINFO(ifp);
1801 		if (nd6if->basereachable && /* already initialized */
1802 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1803 			/*
1804 			 * Since reachable time rarely changes by router
1805 			 * advertisements, we SHOULD insure that a new random
1806 			 * value gets recomputed at least once every few hours.
1807 			 * (RFC 2461, 6.3.4)
1808 			 */
1809 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1810 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1811 		}
1812 	}
1813 	IFNET_RUNLOCK();
1814 	splx(s);
1815 }
1816 
1817 #define senderr(e) { error = (e); goto bad;}
1818 int
1819 nd6_output(ifp, origifp, m0, dst, rt0)
1820 	struct ifnet *ifp;
1821 	struct ifnet *origifp;
1822 	struct mbuf *m0;
1823 	struct sockaddr_in6 *dst;
1824 	struct rtentry *rt0;
1825 {
1826 	struct mbuf *m = m0;
1827 	struct rtentry *rt = rt0;
1828 	struct sockaddr_in6 *gw6 = NULL;
1829 	struct llinfo_nd6 *ln = NULL;
1830 	int error = 0;
1831 
1832 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1833 		goto sendpkt;
1834 
1835 	if (nd6_need_cache(ifp) == 0)
1836 		goto sendpkt;
1837 
1838 	/*
1839 	 * next hop determination.  This routine is derived from ether_outpout.
1840 	 */
1841 again:
1842 	if (rt) {
1843 		if ((rt->rt_flags & RTF_UP) == 0) {
1844 			rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1845 			if (rt != NULL) {
1846 				RT_REMREF(rt);
1847 				RT_UNLOCK(rt);
1848 				if (rt->rt_ifp != ifp)
1849 					/*
1850 					 * XXX maybe we should update ifp too,
1851 					 * but the original code didn't and I
1852 					 * don't know what is correct here.
1853 					 */
1854 					goto again;
1855 			} else
1856 				senderr(EHOSTUNREACH);
1857 		}
1858 
1859 		if (rt->rt_flags & RTF_GATEWAY) {
1860 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1861 
1862 			/*
1863 			 * We skip link-layer address resolution and NUD
1864 			 * if the gateway is not a neighbor from ND point
1865 			 * of view, regardless of the value of nd_ifinfo.flags.
1866 			 * The second condition is a bit tricky; we skip
1867 			 * if the gateway is our own address, which is
1868 			 * sometimes used to install a route to a p2p link.
1869 			 */
1870 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1871 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1872 				/*
1873 				 * We allow this kind of tricky route only
1874 				 * when the outgoing interface is p2p.
1875 				 * XXX: we may need a more generic rule here.
1876 				 */
1877 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1878 					senderr(EHOSTUNREACH);
1879 
1880 				goto sendpkt;
1881 			}
1882 
1883 			if (rt->rt_gwroute == 0)
1884 				goto lookup;
1885 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1886 				RT_LOCK(rt);
1887 				rtfree(rt); rt = rt0;
1888 			lookup:
1889 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1890 				if ((rt = rt->rt_gwroute) == 0)
1891 					senderr(EHOSTUNREACH);
1892 				RT_UNLOCK(rt);
1893 			}
1894 		}
1895 	}
1896 
1897 	/*
1898 	 * Address resolution or Neighbor Unreachability Detection
1899 	 * for the next hop.
1900 	 * At this point, the destination of the packet must be a unicast
1901 	 * or an anycast address(i.e. not a multicast).
1902 	 */
1903 
1904 	/* Look up the neighbor cache for the nexthop */
1905 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1906 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1907 	else {
1908 		/*
1909 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1910 		 * the condition below is not very efficient.  But we believe
1911 		 * it is tolerable, because this should be a rare case.
1912 		 */
1913 		if (nd6_is_addr_neighbor(dst, ifp) &&
1914 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1915 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1916 	}
1917 	if (!ln || !rt) {
1918 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1919 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1920 			log(LOG_DEBUG,
1921 			    "nd6_output: can't allocate llinfo for %s "
1922 			    "(ln=%p, rt=%p)\n",
1923 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1924 			senderr(EIO);	/* XXX: good error? */
1925 		}
1926 
1927 		goto sendpkt;	/* send anyway */
1928 	}
1929 
1930 	/* We don't have to do link-layer address resolution on a p2p link. */
1931 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1932 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1933 		ln->ln_state = ND6_LLINFO_STALE;
1934 		ln->ln_expire = time_second + nd6_gctimer;
1935 	}
1936 
1937 	/*
1938 	 * The first time we send a packet to a neighbor whose entry is
1939 	 * STALE, we have to change the state to DELAY and a sets a timer to
1940 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1941 	 * neighbor unreachability detection on expiration.
1942 	 * (RFC 2461 7.3.3)
1943 	 */
1944 	if (ln->ln_state == ND6_LLINFO_STALE) {
1945 		ln->ln_asked = 0;
1946 		ln->ln_state = ND6_LLINFO_DELAY;
1947 		ln->ln_expire = time_second + nd6_delay;
1948 	}
1949 
1950 	/*
1951 	 * If the neighbor cache entry has a state other than INCOMPLETE
1952 	 * (i.e. its link-layer address is already resolved), just
1953 	 * send the packet.
1954 	 */
1955 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1956 		goto sendpkt;
1957 
1958 	/*
1959 	 * There is a neighbor cache entry, but no ethernet address
1960 	 * response yet.  Replace the held mbuf (if any) with this
1961 	 * latest one.
1962 	 *
1963 	 * This code conforms to the rate-limiting rule described in Section
1964 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
1965 	 * an NS below.
1966 	 */
1967 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1968 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1969 	if (ln->ln_hold)
1970 		m_freem(ln->ln_hold);
1971 	ln->ln_hold = m;
1972 	if (ln->ln_expire) {
1973 		if (ln->ln_asked < nd6_mmaxtries &&
1974 		    ln->ln_expire < time_second) {
1975 			ln->ln_asked++;
1976 			ln->ln_expire = time_second +
1977 				ND_IFINFO(ifp)->retrans / 1000;
1978 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1979 		}
1980 	}
1981 	return (0);
1982 
1983   sendpkt:
1984 #ifdef IPSEC
1985 	/* clean ipsec history once it goes out of the node */
1986 	ipsec_delaux(m);
1987 #endif
1988 
1989 #ifdef MAC
1990 	mac_create_mbuf_linklayer(ifp, m);
1991 #endif
1992 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1993 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1994 		    rt));
1995 	}
1996 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
1997 
1998   bad:
1999 	if (m)
2000 		m_freem(m);
2001 	return (error);
2002 }
2003 #undef senderr
2004 
2005 int
2006 nd6_need_cache(ifp)
2007 	struct ifnet *ifp;
2008 {
2009 	/*
2010 	 * XXX: we currently do not make neighbor cache on any interface
2011 	 * other than ARCnet, Ethernet, FDDI and GIF.
2012 	 *
2013 	 * RFC2893 says:
2014 	 * - unidirectional tunnels needs no ND
2015 	 */
2016 	switch (ifp->if_type) {
2017 	case IFT_ARCNET:
2018 	case IFT_ETHER:
2019 	case IFT_FDDI:
2020 	case IFT_IEEE1394:
2021 #ifdef IFT_L2VLAN
2022 	case IFT_L2VLAN:
2023 #endif
2024 #ifdef IFT_IEEE80211
2025 	case IFT_IEEE80211:
2026 #endif
2027 	case IFT_GIF:		/* XXX need more cases? */
2028 		return (1);
2029 	default:
2030 		return (0);
2031 	}
2032 }
2033 
2034 int
2035 nd6_storelladdr(ifp, rt0, m, dst, desten)
2036 	struct ifnet *ifp;
2037 	struct rtentry *rt0;
2038 	struct mbuf *m;
2039 	struct sockaddr *dst;
2040 	u_char *desten;
2041 {
2042 	int i;
2043 	struct sockaddr_dl *sdl;
2044 	struct rtentry *rt;
2045 
2046 	if (m->m_flags & M_MCAST) {
2047 		switch (ifp->if_type) {
2048 		case IFT_ETHER:
2049 		case IFT_FDDI:
2050 #ifdef IFT_L2VLAN
2051 		case IFT_L2VLAN:
2052 #endif
2053 #ifdef IFT_IEEE80211
2054 		case IFT_IEEE80211:
2055 #endif
2056 		case IFT_ISO88025:
2057 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2058 						 desten);
2059 			return (0);
2060 		case IFT_IEEE1394:
2061 			/*
2062 			 * netbsd can use if_broadcastaddr, but we don't do so
2063 			 * to reduce # of ifdef.
2064 			 */
2065 			for (i = 0; i < ifp->if_addrlen; i++)
2066 				desten[i] = ~0;
2067 			return (0);
2068 		case IFT_ARCNET:
2069 			*desten = 0;
2070 			return (0);
2071 		default:
2072 			m_freem(m);
2073 			return (EAFNOSUPPORT);
2074 		}
2075 	}
2076 
2077 	i = rt_check(&rt, &rt0, dst);
2078 	if (i) {
2079 		m_freem(m);
2080 		return i;
2081 	}
2082 
2083 	if (rt == NULL) {
2084 		/* this could happen, if we could not allocate memory */
2085 		m_freem(m);
2086 		return (ENOMEM);
2087 	}
2088 	if (rt->rt_gateway->sa_family != AF_LINK) {
2089 		printf("nd6_storelladdr: something odd happens\n");
2090 		m_freem(m);
2091 		return (EINVAL);
2092 	}
2093 	sdl = SDL(rt->rt_gateway);
2094 	if (sdl->sdl_alen == 0) {
2095 		/* this should be impossible, but we bark here for debugging */
2096 		printf("nd6_storelladdr: sdl_alen == 0\n");
2097 		m_freem(m);
2098 		return (EINVAL);
2099 	}
2100 
2101 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2102 	return (0);
2103 }
2104 
2105 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2106 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2107 #ifdef SYSCTL_DECL
2108 SYSCTL_DECL(_net_inet6_icmp6);
2109 #endif
2110 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2111 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2112 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2113 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2114 
2115 static int
2116 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2117 {
2118 	int error;
2119 	char buf[1024];
2120 	struct in6_defrouter *d, *de;
2121 	struct nd_defrouter *dr;
2122 
2123 	if (req->newptr)
2124 		return EPERM;
2125 	error = 0;
2126 
2127 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2128 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2129 		d = (struct in6_defrouter *)buf;
2130 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2131 
2132 		if (d + 1 <= de) {
2133 			bzero(d, sizeof(*d));
2134 			d->rtaddr.sin6_family = AF_INET6;
2135 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2136 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2137 			    dr->ifp) != 0)
2138 				log(LOG_ERR,
2139 				    "scope error in "
2140 				    "default router list (%s)\n",
2141 				    ip6_sprintf(&dr->rtaddr));
2142 			d->flags = dr->flags;
2143 			d->rtlifetime = dr->rtlifetime;
2144 			d->expire = dr->expire;
2145 			d->if_index = dr->ifp->if_index;
2146 		} else
2147 			panic("buffer too short");
2148 
2149 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2150 		if (error)
2151 			break;
2152 	}
2153 
2154 	return (error);
2155 }
2156 
2157 static int
2158 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2159 {
2160 	int error;
2161 	char buf[1024];
2162 	struct in6_prefix *p, *pe;
2163 	struct nd_prefix *pr;
2164 
2165 	if (req->newptr)
2166 		return EPERM;
2167 	error = 0;
2168 
2169 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2170 		u_short advrtrs;
2171 		size_t advance;
2172 		struct sockaddr_in6 *sin6, *s6;
2173 		struct nd_pfxrouter *pfr;
2174 
2175 		p = (struct in6_prefix *)buf;
2176 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2177 
2178 		if (p + 1 <= pe) {
2179 			bzero(p, sizeof(*p));
2180 			sin6 = (struct sockaddr_in6 *)(p + 1);
2181 
2182 			p->prefix = pr->ndpr_prefix;
2183 			if (in6_recoverscope(&p->prefix,
2184 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2185 				log(LOG_ERR,
2186 				    "scope error in prefix list (%s)\n",
2187 				    ip6_sprintf(&p->prefix.sin6_addr));
2188 			p->raflags = pr->ndpr_raf;
2189 			p->prefixlen = pr->ndpr_plen;
2190 			p->vltime = pr->ndpr_vltime;
2191 			p->pltime = pr->ndpr_pltime;
2192 			p->if_index = pr->ndpr_ifp->if_index;
2193 			p->expire = pr->ndpr_expire;
2194 			p->refcnt = pr->ndpr_refcnt;
2195 			p->flags = pr->ndpr_stateflags;
2196 			p->origin = PR_ORIG_RA;
2197 			advrtrs = 0;
2198 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2199 			     pfr = pfr->pfr_next) {
2200 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2201 					advrtrs++;
2202 					continue;
2203 				}
2204 				s6 = &sin6[advrtrs];
2205 				bzero(s6, sizeof(*s6));
2206 				s6->sin6_family = AF_INET6;
2207 				s6->sin6_len = sizeof(*sin6);
2208 				if (in6_recoverscope(s6, &pfr->router->rtaddr,
2209 				    pfr->router->ifp) != 0)
2210 					log(LOG_ERR,
2211 					    "scope error in "
2212 					    "prefix list (%s)\n",
2213 					    ip6_sprintf(&pfr->router->rtaddr));
2214 				advrtrs++;
2215 			}
2216 			p->advrtrs = advrtrs;
2217 		} else
2218 			panic("buffer too short");
2219 
2220 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2221 		error = SYSCTL_OUT(req, buf, advance);
2222 		if (error)
2223 			break;
2224 	}
2225 
2226 	return (error);
2227 }
2228