xref: /freebsd/sys/netinet6/nd6.c (revision 7660b554bc59a07be0431c17e0e33815818baa69)
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 /*
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 #include "opt_mac.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/callout.h>
48 #include <sys/mac.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #include <sys/socket.h>
52 #include <sys/sockio.h>
53 #include <sys/time.h>
54 #include <sys/kernel.h>
55 #include <sys/protosw.h>
56 #include <sys/errno.h>
57 #include <sys/syslog.h>
58 #include <sys/queue.h>
59 #include <sys/sysctl.h>
60 
61 #include <net/if.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/if_atm.h>
65 #include <net/iso88025.h>
66 #include <net/fddi.h>
67 #include <net/route.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_prefix.h>
76 #include <netinet/icmp6.h>
77 
78 #include <net/net_osdep.h>
79 
80 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
81 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
82 
83 #define SIN6(s) ((struct sockaddr_in6 *)s)
84 #define SDL(s) ((struct sockaddr_dl *)s)
85 
86 /* timer values */
87 int	nd6_prune	= 1;	/* walk list every 1 seconds */
88 int	nd6_delay	= 5;	/* delay first probe time 5 second */
89 int	nd6_umaxtries	= 3;	/* maximum unicast query */
90 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
91 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
92 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
93 
94 /* preventing too many loops in ND option parsing */
95 int nd6_maxndopt = 10;	/* max # of ND options allowed */
96 
97 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
98 
99 #ifdef ND6_DEBUG
100 int nd6_debug = 1;
101 #else
102 int nd6_debug = 0;
103 #endif
104 
105 /* for debugging? */
106 static int nd6_inuse, nd6_allocated;
107 
108 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
109 static size_t nd_ifinfo_indexlim = 8;
110 struct nd_ifinfo *nd_ifinfo = NULL;
111 struct nd_drhead nd_defrouter;
112 struct nd_prhead nd_prefix = { 0 };
113 
114 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
115 static struct sockaddr_in6 all1_sa;
116 
117 static void nd6_slowtimo __P((void *));
118 static int regen_tmpaddr __P((struct in6_ifaddr *));
119 
120 struct callout nd6_slowtimo_ch;
121 struct callout nd6_timer_ch;
122 extern struct callout in6_tmpaddrtimer_ch;
123 
124 void
125 nd6_init()
126 {
127 	static int nd6_init_done = 0;
128 	int i;
129 
130 	if (nd6_init_done) {
131 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
132 		return;
133 	}
134 
135 	all1_sa.sin6_family = AF_INET6;
136 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
137 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
138 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
139 
140 	/* initialization of the default router list */
141 	TAILQ_INIT(&nd_defrouter);
142 
143 	nd6_init_done = 1;
144 
145 	/* start timer */
146 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
147 	    nd6_slowtimo, NULL);
148 }
149 
150 void
151 nd6_ifattach(ifp)
152 	struct ifnet *ifp;
153 {
154 
155 	/*
156 	 * We have some arrays that should be indexed by if_index.
157 	 * since if_index will grow dynamically, they should grow too.
158 	 */
159 	if (nd_ifinfo == NULL || if_index >= nd_ifinfo_indexlim) {
160 		size_t n;
161 		caddr_t q;
162 
163 		while (if_index >= nd_ifinfo_indexlim)
164 			nd_ifinfo_indexlim <<= 1;
165 
166 		/* grow nd_ifinfo */
167 		n = nd_ifinfo_indexlim * sizeof(struct nd_ifinfo);
168 		q = (caddr_t)malloc(n, M_IP6NDP, M_WAITOK);
169 		bzero(q, n);
170 		if (nd_ifinfo) {
171 			bcopy((caddr_t)nd_ifinfo, q, n/2);
172 			free((caddr_t)nd_ifinfo, M_IP6NDP);
173 		}
174 		nd_ifinfo = (struct nd_ifinfo *)q;
175 	}
176 
177 #define ND nd_ifinfo[ifp->if_index]
178 
179 	/*
180 	 * Don't initialize if called twice.
181 	 * XXX: to detect this, we should choose a member that is never set
182 	 * before initialization of the ND structure itself.  We formaly used
183 	 * the linkmtu member, which was not suitable because it could be
184 	 * initialized via "ifconfig mtu".
185 	 */
186 	if (ND.basereachable)
187 		return;
188 
189 	ND.linkmtu = ifnet_byindex(ifp->if_index)->if_mtu;
190 	ND.chlim = IPV6_DEFHLIM;
191 	ND.basereachable = REACHABLE_TIME;
192 	ND.reachable = ND_COMPUTE_RTIME(ND.basereachable);
193 	ND.retrans = RETRANS_TIMER;
194 	ND.receivedra = 0;
195 	/*
196 	 * Note that the default value of ip6_accept_rtadv is 0, which means
197 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
198 	 * here.
199 	 */
200 	ND.flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
201 	nd6_setmtu(ifp);
202 #undef ND
203 }
204 
205 /*
206  * Reset ND level link MTU. This function is called when the physical MTU
207  * changes, which means we might have to adjust the ND level MTU.
208  */
209 void
210 nd6_setmtu(ifp)
211 	struct ifnet *ifp;
212 {
213 	struct nd_ifinfo *ndi = &nd_ifinfo[ifp->if_index];
214 	u_long oldmaxmtu = ndi->maxmtu;
215 	u_long oldlinkmtu = ndi->linkmtu;
216 
217 	switch (ifp->if_type) {
218 	case IFT_ARCNET:	/* XXX MTU handling needs more work */
219 		ndi->maxmtu = MIN(60480, ifp->if_mtu);
220 		break;
221 	case IFT_ETHER:
222 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
223 		break;
224 	case IFT_FDDI:
225 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
226 		break;
227 	case IFT_ATM:
228 		ndi->maxmtu = MIN(ATMMTU, ifp->if_mtu);
229 		break;
230 	case IFT_IEEE1394:	/* XXX should be IEEE1394MTU(1500) */
231 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
232 		break;
233 #ifdef IFT_IEEE80211
234 	case IFT_IEEE80211:	/* XXX should be IEEE80211MTU(1500) */
235 		ndi->maxmtu = MIN(ETHERMTU, ifp->if_mtu);
236 		break;
237 #endif
238 	 case IFT_ISO88025:
239 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
240 		 break;
241 	default:
242 		ndi->maxmtu = ifp->if_mtu;
243 		break;
244 	}
245 
246 	if (oldmaxmtu != ndi->maxmtu) {
247 		/*
248 		 * If the ND level MTU is not set yet, or if the maxmtu
249 		 * is reset to a smaller value than the ND level MTU,
250 		 * also reset the ND level MTU.
251 		 */
252 		if (ndi->linkmtu == 0 ||
253 		    ndi->maxmtu < ndi->linkmtu) {
254 			ndi->linkmtu = ndi->maxmtu;
255 			/* also adjust in6_maxmtu if necessary. */
256 			if (oldlinkmtu == 0) {
257 				/*
258 				 * XXX: the case analysis is grotty, but
259 				 * it is not efficient to call in6_setmaxmtu()
260 				 * here when we are during the initialization
261 				 * procedure.
262 				 */
263 				if (in6_maxmtu < ndi->linkmtu)
264 					in6_maxmtu = ndi->linkmtu;
265 			} else
266 				in6_setmaxmtu();
267 		}
268 	}
269 #undef MIN
270 }
271 
272 void
273 nd6_option_init(opt, icmp6len, ndopts)
274 	void *opt;
275 	int icmp6len;
276 	union nd_opts *ndopts;
277 {
278 	bzero(ndopts, sizeof(*ndopts));
279 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
280 	ndopts->nd_opts_last
281 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
282 
283 	if (icmp6len == 0) {
284 		ndopts->nd_opts_done = 1;
285 		ndopts->nd_opts_search = NULL;
286 	}
287 }
288 
289 /*
290  * Take one ND option.
291  */
292 struct nd_opt_hdr *
293 nd6_option(ndopts)
294 	union nd_opts *ndopts;
295 {
296 	struct nd_opt_hdr *nd_opt;
297 	int olen;
298 
299 	if (!ndopts)
300 		panic("ndopts == NULL in nd6_option");
301 	if (!ndopts->nd_opts_last)
302 		panic("uninitialized ndopts in nd6_option");
303 	if (!ndopts->nd_opts_search)
304 		return NULL;
305 	if (ndopts->nd_opts_done)
306 		return NULL;
307 
308 	nd_opt = ndopts->nd_opts_search;
309 
310 	/* make sure nd_opt_len is inside the buffer */
311 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
312 		bzero(ndopts, sizeof(*ndopts));
313 		return NULL;
314 	}
315 
316 	olen = nd_opt->nd_opt_len << 3;
317 	if (olen == 0) {
318 		/*
319 		 * Message validation requires that all included
320 		 * options have a length that is greater than zero.
321 		 */
322 		bzero(ndopts, sizeof(*ndopts));
323 		return NULL;
324 	}
325 
326 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
327 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
328 		/* option overruns the end of buffer, invalid */
329 		bzero(ndopts, sizeof(*ndopts));
330 		return NULL;
331 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
332 		/* reached the end of options chain */
333 		ndopts->nd_opts_done = 1;
334 		ndopts->nd_opts_search = NULL;
335 	}
336 	return nd_opt;
337 }
338 
339 /*
340  * Parse multiple ND options.
341  * This function is much easier to use, for ND routines that do not need
342  * multiple options of the same type.
343  */
344 int
345 nd6_options(ndopts)
346 	union nd_opts *ndopts;
347 {
348 	struct nd_opt_hdr *nd_opt;
349 	int i = 0;
350 
351 	if (!ndopts)
352 		panic("ndopts == NULL in nd6_options");
353 	if (!ndopts->nd_opts_last)
354 		panic("uninitialized ndopts in nd6_options");
355 	if (!ndopts->nd_opts_search)
356 		return 0;
357 
358 	while (1) {
359 		nd_opt = nd6_option(ndopts);
360 		if (!nd_opt && !ndopts->nd_opts_last) {
361 			/*
362 			 * Message validation requires that all included
363 			 * options have a length that is greater than zero.
364 			 */
365 			icmp6stat.icp6s_nd_badopt++;
366 			bzero(ndopts, sizeof(*ndopts));
367 			return -1;
368 		}
369 
370 		if (!nd_opt)
371 			goto skip1;
372 
373 		switch (nd_opt->nd_opt_type) {
374 		case ND_OPT_SOURCE_LINKADDR:
375 		case ND_OPT_TARGET_LINKADDR:
376 		case ND_OPT_MTU:
377 		case ND_OPT_REDIRECTED_HEADER:
378 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
379 				nd6log((LOG_INFO,
380 				    "duplicated ND6 option found (type=%d)\n",
381 				    nd_opt->nd_opt_type));
382 				/* XXX bark? */
383 			} else {
384 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
385 					= nd_opt;
386 			}
387 			break;
388 		case ND_OPT_PREFIX_INFORMATION:
389 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
390 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
391 					= nd_opt;
392 			}
393 			ndopts->nd_opts_pi_end =
394 				(struct nd_opt_prefix_info *)nd_opt;
395 			break;
396 		default:
397 			/*
398 			 * Unknown options must be silently ignored,
399 			 * to accomodate future extension to the protocol.
400 			 */
401 			nd6log((LOG_DEBUG,
402 			    "nd6_options: unsupported option %d - "
403 			    "option ignored\n", nd_opt->nd_opt_type));
404 		}
405 
406 skip1:
407 		i++;
408 		if (i > nd6_maxndopt) {
409 			icmp6stat.icp6s_nd_toomanyopt++;
410 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
411 			break;
412 		}
413 
414 		if (ndopts->nd_opts_done)
415 			break;
416 	}
417 
418 	return 0;
419 }
420 
421 /*
422  * ND6 timer routine to expire default route list and prefix list
423  */
424 void
425 nd6_timer(ignored_arg)
426 	void	*ignored_arg;
427 {
428 	int s;
429 	struct llinfo_nd6 *ln;
430 	struct nd_defrouter *dr;
431 	struct nd_prefix *pr;
432 	struct ifnet *ifp;
433 	struct in6_ifaddr *ia6, *nia6;
434 	struct in6_addrlifetime *lt6;
435 
436 	s = splnet();
437 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
438 		      nd6_timer, NULL);
439 
440 	ln = llinfo_nd6.ln_next;
441 	while (ln && ln != &llinfo_nd6) {
442 		struct rtentry *rt;
443 		struct sockaddr_in6 *dst;
444 		struct llinfo_nd6 *next = ln->ln_next;
445 		/* XXX: used for the DELAY case only: */
446 		struct nd_ifinfo *ndi = NULL;
447 
448 		if ((rt = ln->ln_rt) == NULL) {
449 			ln = next;
450 			continue;
451 		}
452 		if ((ifp = rt->rt_ifp) == NULL) {
453 			ln = next;
454 			continue;
455 		}
456 		ndi = &nd_ifinfo[ifp->if_index];
457 		dst = (struct sockaddr_in6 *)rt_key(rt);
458 
459 		if (ln->ln_expire > time_second) {
460 			ln = next;
461 			continue;
462 		}
463 
464 		/* sanity check */
465 		if (!rt)
466 			panic("rt=0 in nd6_timer(ln=%p)", ln);
467 		if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
468 			panic("rt_llinfo(%p) is not equal to ln(%p)",
469 			      rt->rt_llinfo, ln);
470 		if (!dst)
471 			panic("dst=0 in nd6_timer(ln=%p)", ln);
472 
473 		switch (ln->ln_state) {
474 		case ND6_LLINFO_INCOMPLETE:
475 			if (ln->ln_asked < nd6_mmaxtries) {
476 				ln->ln_asked++;
477 				ln->ln_expire = time_second +
478 					nd_ifinfo[ifp->if_index].retrans / 1000;
479 				nd6_ns_output(ifp, NULL, &dst->sin6_addr,
480 					ln, 0);
481 			} else {
482 				struct mbuf *m = ln->ln_hold;
483 				if (m) {
484 					if (rt->rt_ifp) {
485 						/*
486 						 * Fake rcvif to make ICMP error
487 						 * more helpful in diagnosing
488 						 * for the receiver.
489 						 * XXX: should we consider
490 						 * older rcvif?
491 						 */
492 						m->m_pkthdr.rcvif = rt->rt_ifp;
493 					}
494 					icmp6_error(m, ICMP6_DST_UNREACH,
495 						    ICMP6_DST_UNREACH_ADDR, 0);
496 					ln->ln_hold = NULL;
497 				}
498 				next = nd6_free(rt);
499 			}
500 			break;
501 		case ND6_LLINFO_REACHABLE:
502 			if (ln->ln_expire) {
503 				ln->ln_state = ND6_LLINFO_STALE;
504 				ln->ln_expire = time_second + nd6_gctimer;
505 			}
506 			break;
507 
508 		case ND6_LLINFO_STALE:
509 			/* Garbage Collection(RFC 2461 5.3) */
510 			if (ln->ln_expire)
511 				next = nd6_free(rt);
512 			break;
513 
514 		case ND6_LLINFO_DELAY:
515 			if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
516 				/* We need NUD */
517 				ln->ln_asked = 1;
518 				ln->ln_state = ND6_LLINFO_PROBE;
519 				ln->ln_expire = time_second +
520 					ndi->retrans / 1000;
521 				nd6_ns_output(ifp, &dst->sin6_addr,
522 					      &dst->sin6_addr,
523 					      ln, 0);
524 			} else {
525 				ln->ln_state = ND6_LLINFO_STALE; /* XXX */
526 				ln->ln_expire = time_second + nd6_gctimer;
527 			}
528 			break;
529 		case ND6_LLINFO_PROBE:
530 			if (ln->ln_asked < nd6_umaxtries) {
531 				ln->ln_asked++;
532 				ln->ln_expire = time_second +
533 					nd_ifinfo[ifp->if_index].retrans / 1000;
534 				nd6_ns_output(ifp, &dst->sin6_addr,
535 					       &dst->sin6_addr, ln, 0);
536 			} else {
537 				next = nd6_free(rt);
538 			}
539 			break;
540 		}
541 		ln = next;
542 	}
543 
544 	/* expire default router list */
545 	dr = TAILQ_FIRST(&nd_defrouter);
546 	while (dr) {
547 		if (dr->expire && dr->expire < time_second) {
548 			struct nd_defrouter *t;
549 			t = TAILQ_NEXT(dr, dr_entry);
550 			defrtrlist_del(dr);
551 			dr = t;
552 		} else {
553 			dr = TAILQ_NEXT(dr, dr_entry);
554 		}
555 	}
556 
557 	/*
558 	 * expire interface addresses.
559 	 * in the past the loop was inside prefix expiry processing.
560 	 * However, from a stricter speci-confrmance standpoint, we should
561 	 * rather separate address lifetimes and prefix lifetimes.
562 	 */
563   addrloop:
564 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
565 		nia6 = ia6->ia_next;
566 		/* check address lifetime */
567 		lt6 = &ia6->ia6_lifetime;
568 		if (IFA6_IS_INVALID(ia6)) {
569 			int regen = 0;
570 
571 			/*
572 			 * If the expiring address is temporary, try
573 			 * regenerating a new one.  This would be useful when
574 			 * we suspended a laptop PC, then turned it on after a
575 			 * period that could invalidate all temporary
576 			 * addresses.  Although we may have to restart the
577 			 * loop (see below), it must be after purging the
578 			 * address.  Otherwise, we'd see an infinite loop of
579 			 * regeneration.
580 			 */
581 			if (ip6_use_tempaddr &&
582 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
583 				if (regen_tmpaddr(ia6) == 0)
584 					regen = 1;
585 			}
586 
587 			in6_purgeaddr(&ia6->ia_ifa);
588 
589 			if (regen)
590 				goto addrloop; /* XXX: see below */
591 		}
592 		if (IFA6_IS_DEPRECATED(ia6)) {
593 			int oldflags = ia6->ia6_flags;
594 
595 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
596 
597 			/*
598 			 * If a temporary address has just become deprecated,
599 			 * regenerate a new one if possible.
600 			 */
601 			if (ip6_use_tempaddr &&
602 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
603 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
604 
605 				if (regen_tmpaddr(ia6) == 0) {
606 					/*
607 					 * A new temporary address is
608 					 * generated.
609 					 * XXX: this means the address chain
610 					 * has changed while we are still in
611 					 * the loop.  Although the change
612 					 * would not cause disaster (because
613 					 * it's not a deletion, but an
614 					 * addition,) we'd rather restart the
615 					 * loop just for safety.  Or does this
616 					 * significantly reduce performance??
617 					 */
618 					goto addrloop;
619 				}
620 			}
621 		} else {
622 			/*
623 			 * A new RA might have made a deprecated address
624 			 * preferred.
625 			 */
626 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
627 		}
628 	}
629 
630 	/* expire prefix list */
631 	pr = nd_prefix.lh_first;
632 	while (pr) {
633 		/*
634 		 * check prefix lifetime.
635 		 * since pltime is just for autoconf, pltime processing for
636 		 * prefix is not necessary.
637 		 */
638 		if (pr->ndpr_expire && pr->ndpr_expire < time_second) {
639 			struct nd_prefix *t;
640 			t = pr->ndpr_next;
641 
642 			/*
643 			 * address expiration and prefix expiration are
644 			 * separate.  NEVER perform in6_purgeaddr here.
645 			 */
646 
647 			prelist_remove(pr);
648 			pr = t;
649 		} else
650 			pr = pr->ndpr_next;
651 	}
652 	splx(s);
653 }
654 
655 static int
656 regen_tmpaddr(ia6)
657 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
658 {
659 	struct ifaddr *ifa;
660 	struct ifnet *ifp;
661 	struct in6_ifaddr *public_ifa6 = NULL;
662 
663 	ifp = ia6->ia_ifa.ifa_ifp;
664 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
665 	     ifa = ifa->ifa_list.tqe_next)
666 	{
667 		struct in6_ifaddr *it6;
668 
669 		if (ifa->ifa_addr->sa_family != AF_INET6)
670 			continue;
671 
672 		it6 = (struct in6_ifaddr *)ifa;
673 
674 		/* ignore no autoconf addresses. */
675 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
676 			continue;
677 
678 		/* ignore autoconf addresses with different prefixes. */
679 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
680 			continue;
681 
682 		/*
683 		 * Now we are looking at an autoconf address with the same
684 		 * prefix as ours.  If the address is temporary and is still
685 		 * preferred, do not create another one.  It would be rare, but
686 		 * could happen, for example, when we resume a laptop PC after
687 		 * a long period.
688 		 */
689 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
690 		    !IFA6_IS_DEPRECATED(it6)) {
691 			public_ifa6 = NULL;
692 			break;
693 		}
694 
695 		/*
696 		 * This is a public autoconf address that has the same prefix
697 		 * as ours.  If it is preferred, keep it.  We can't break the
698 		 * loop here, because there may be a still-preferred temporary
699 		 * address with the prefix.
700 		 */
701 		if (!IFA6_IS_DEPRECATED(it6))
702 		    public_ifa6 = it6;
703 	}
704 
705 	if (public_ifa6 != NULL) {
706 		int e;
707 
708 		if ((e = in6_tmpifadd(public_ifa6, 0)) != 0) {
709 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
710 			    " tmp addr,errno=%d\n", e);
711 			return(-1);
712 		}
713 		return(0);
714 	}
715 
716 	return(-1);
717 }
718 
719 /*
720  * Nuke neighbor cache/prefix/default router management table, right before
721  * ifp goes away.
722  */
723 void
724 nd6_purge(ifp)
725 	struct ifnet *ifp;
726 {
727 	struct llinfo_nd6 *ln, *nln;
728 	struct nd_defrouter *dr, *ndr, drany;
729 	struct nd_prefix *pr, *npr;
730 
731 	/* Nuke default router list entries toward ifp */
732 	if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
733 		/*
734 		 * The first entry of the list may be stored in
735 		 * the routing table, so we'll delete it later.
736 		 */
737 		for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = ndr) {
738 			ndr = TAILQ_NEXT(dr, dr_entry);
739 			if (dr->ifp == ifp)
740 				defrtrlist_del(dr);
741 		}
742 		dr = TAILQ_FIRST(&nd_defrouter);
743 		if (dr->ifp == ifp)
744 			defrtrlist_del(dr);
745 	}
746 
747 	/* Nuke prefix list entries toward ifp */
748 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
749 		npr = pr->ndpr_next;
750 		if (pr->ndpr_ifp == ifp) {
751 			/*
752 			 * Previously, pr->ndpr_addr is removed as well,
753 			 * but I strongly believe we don't have to do it.
754 			 * nd6_purge() is only called from in6_ifdetach(),
755 			 * which removes all the associated interface addresses
756 			 * by itself.
757 			 * (jinmei@kame.net 20010129)
758 			 */
759 			prelist_remove(pr);
760 		}
761 	}
762 
763 	/* cancel default outgoing interface setting */
764 	if (nd6_defifindex == ifp->if_index)
765 		nd6_setdefaultiface(0);
766 
767 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
768 		/* refresh default router list */
769 		bzero(&drany, sizeof(drany));
770 		defrouter_delreq(&drany, 0);
771 		defrouter_select();
772 	}
773 
774 	/*
775 	 * Nuke neighbor cache entries for the ifp.
776 	 * Note that rt->rt_ifp may not be the same as ifp,
777 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
778 	 * nd6_rtrequest(), and ip6_input().
779 	 */
780 	ln = llinfo_nd6.ln_next;
781 	while (ln && ln != &llinfo_nd6) {
782 		struct rtentry *rt;
783 		struct sockaddr_dl *sdl;
784 
785 		nln = ln->ln_next;
786 		rt = ln->ln_rt;
787 		if (rt && rt->rt_gateway &&
788 		    rt->rt_gateway->sa_family == AF_LINK) {
789 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
790 			if (sdl->sdl_index == ifp->if_index)
791 				nln = nd6_free(rt);
792 		}
793 		ln = nln;
794 	}
795 }
796 
797 struct rtentry *
798 nd6_lookup(addr6, create, ifp)
799 	struct in6_addr *addr6;
800 	int create;
801 	struct ifnet *ifp;
802 {
803 	struct rtentry *rt;
804 	struct sockaddr_in6 sin6;
805 
806 	bzero(&sin6, sizeof(sin6));
807 	sin6.sin6_len = sizeof(struct sockaddr_in6);
808 	sin6.sin6_family = AF_INET6;
809 	sin6.sin6_addr = *addr6;
810 #ifdef SCOPEDROUTING
811 	sin6.sin6_scope_id = in6_addr2scopeid(ifp, addr6);
812 #endif
813 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
814 	if (rt && (rt->rt_flags & RTF_LLINFO) == 0) {
815 		/*
816 		 * This is the case for the default route.
817 		 * If we want to create a neighbor cache for the address, we
818 		 * should free the route for the destination and allocate an
819 		 * interface route.
820 		 */
821 		if (create) {
822 			RTFREE(rt);
823 			rt = 0;
824 		}
825 	}
826 	if (!rt) {
827 		if (create && ifp) {
828 			int e;
829 
830 			/*
831 			 * If no route is available and create is set,
832 			 * we allocate a host route for the destination
833 			 * and treat it like an interface route.
834 			 * This hack is necessary for a neighbor which can't
835 			 * be covered by our own prefix.
836 			 */
837 			struct ifaddr *ifa =
838 				ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
839 			if (ifa == NULL)
840 				return(NULL);
841 
842 			/*
843 			 * Create a new route.  RTF_LLINFO is necessary
844 			 * to create a Neighbor Cache entry for the
845 			 * destination in nd6_rtrequest which will be
846 			 * called in rtrequest via ifa->ifa_rtrequest.
847 			 */
848 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
849 					   ifa->ifa_addr,
850 					   (struct sockaddr *)&all1_sa,
851 					   (ifa->ifa_flags |
852 					    RTF_HOST | RTF_LLINFO) &
853 					   ~RTF_CLONING,
854 					   &rt)) != 0)
855 				log(LOG_ERR,
856 				    "nd6_lookup: failed to add route for a "
857 				    "neighbor(%s), errno=%d\n",
858 				    ip6_sprintf(addr6), e);
859 			if (rt == NULL)
860 				return(NULL);
861 			if (rt->rt_llinfo) {
862 				struct llinfo_nd6 *ln =
863 					(struct llinfo_nd6 *)rt->rt_llinfo;
864 				ln->ln_state = ND6_LLINFO_NOSTATE;
865 			}
866 		} else
867 			return(NULL);
868 	}
869 	rt->rt_refcnt--;
870 	/*
871 	 * Validation for the entry.
872 	 * Note that the check for rt_llinfo is necessary because a cloned
873 	 * route from a parent route that has the L flag (e.g. the default
874 	 * route to a p2p interface) may have the flag, too, while the
875 	 * destination is not actually a neighbor.
876 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
877 	 *      it might be the loopback interface if the entry is for our
878 	 *      own address on a non-loopback interface. Instead, we should
879 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
880 	 *      interface.
881 	 */
882 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
883 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
884 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
885 		if (create) {
886 			log(LOG_DEBUG, "nd6_lookup: failed to lookup %s (if = %s)\n",
887 			    ip6_sprintf(addr6), ifp ? if_name(ifp) : "unspec");
888 			/* xxx more logs... kazu */
889 		}
890 		return(NULL);
891 	}
892 	return(rt);
893 }
894 
895 /*
896  * Detect if a given IPv6 address identifies a neighbor on a given link.
897  * XXX: should take care of the destination of a p2p link?
898  */
899 int
900 nd6_is_addr_neighbor(addr, ifp)
901 	struct sockaddr_in6 *addr;
902 	struct ifnet *ifp;
903 {
904 	struct ifaddr *ifa;
905 	int i;
906 
907 #define IFADDR6(a) ((((struct in6_ifaddr *)(a))->ia_addr).sin6_addr)
908 #define IFMASK6(a) ((((struct in6_ifaddr *)(a))->ia_prefixmask).sin6_addr)
909 
910 	/*
911 	 * A link-local address is always a neighbor.
912 	 * XXX: we should use the sin6_scope_id field rather than the embedded
913 	 * interface index.
914 	 */
915 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr) &&
916 	    ntohs(*(u_int16_t *)&addr->sin6_addr.s6_addr[2]) == ifp->if_index)
917 		return(1);
918 
919 	/*
920 	 * If the address matches one of our addresses,
921 	 * it should be a neighbor.
922 	 */
923 	for (ifa = ifp->if_addrlist.tqh_first;
924 	     ifa;
925 	     ifa = ifa->ifa_list.tqe_next)
926 	{
927 		if (ifa->ifa_addr->sa_family != AF_INET6)
928 			next: continue;
929 
930 		for (i = 0; i < 4; i++) {
931 			if ((IFADDR6(ifa).s6_addr32[i] ^
932 			     addr->sin6_addr.s6_addr32[i]) &
933 			    IFMASK6(ifa).s6_addr32[i])
934 				goto next;
935 		}
936 		return(1);
937 	}
938 
939 	/*
940 	 * Even if the address matches none of our addresses, it might be
941 	 * in the neighbor cache.
942 	 */
943 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
944 		return(1);
945 
946 	return(0);
947 #undef IFADDR6
948 #undef IFMASK6
949 }
950 
951 /*
952  * Free an nd6 llinfo entry.
953  */
954 struct llinfo_nd6 *
955 nd6_free(rt)
956 	struct rtentry *rt;
957 {
958 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
959 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
960 	struct nd_defrouter *dr;
961 
962 	/*
963 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
964 	 * even though it is not harmful, it was not really necessary.
965 	 */
966 
967 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
968 		int s;
969 		s = splnet();
970 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
971 				      rt->rt_ifp);
972 
973 		if (ln->ln_router || dr) {
974 			/*
975 			 * rt6_flush must be called whether or not the neighbor
976 			 * is in the Default Router List.
977 			 * See a corresponding comment in nd6_na_input().
978 			 */
979 			rt6_flush(&in6, rt->rt_ifp);
980 		}
981 
982 		if (dr) {
983 			/*
984 			 * Unreachablity of a router might affect the default
985 			 * router selection and on-link detection of advertised
986 			 * prefixes.
987 			 */
988 
989 			/*
990 			 * Temporarily fake the state to choose a new default
991 			 * router and to perform on-link determination of
992 			 * prefixes correctly.
993 			 * Below the state will be set correctly,
994 			 * or the entry itself will be deleted.
995 			 */
996 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
997 
998 			/*
999 			 * Since defrouter_select() does not affect the
1000 			 * on-link determination and MIP6 needs the check
1001 			 * before the default router selection, we perform
1002 			 * the check now.
1003 			 */
1004 			pfxlist_onlink_check();
1005 
1006 			if (dr == TAILQ_FIRST(&nd_defrouter)) {
1007 				/*
1008 				 * It is used as the current default router,
1009 				 * so we have to move it to the end of the
1010 				 * list and choose a new one.
1011 				 * XXX: it is not very efficient if this is
1012 				 *      the only router.
1013 				 */
1014 				TAILQ_REMOVE(&nd_defrouter, dr, dr_entry);
1015 				TAILQ_INSERT_TAIL(&nd_defrouter, dr, dr_entry);
1016 
1017 				defrouter_select();
1018 			}
1019 		}
1020 		splx(s);
1021 	}
1022 
1023 	/*
1024 	 * Before deleting the entry, remember the next entry as the
1025 	 * return value.  We need this because pfxlist_onlink_check() above
1026 	 * might have freed other entries (particularly the old next entry) as
1027 	 * a side effect (XXX).
1028 	 */
1029 	next = ln->ln_next;
1030 
1031 	/*
1032 	 * Detach the route from the routing tree and the list of neighbor
1033 	 * caches, and disable the route entry not to be used in already
1034 	 * cached routes.
1035 	 */
1036 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1037 		  rt_mask(rt), 0, (struct rtentry **)0);
1038 
1039 	return(next);
1040 }
1041 
1042 /*
1043  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1044  *
1045  * XXX cost-effective metods?
1046  */
1047 void
1048 nd6_nud_hint(rt, dst6, force)
1049 	struct rtentry *rt;
1050 	struct in6_addr *dst6;
1051 	int force;
1052 {
1053 	struct llinfo_nd6 *ln;
1054 
1055 	/*
1056 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1057 	 * routing table by supplied "dst6".
1058 	 */
1059 	if (!rt) {
1060 		if (!dst6)
1061 			return;
1062 		if (!(rt = nd6_lookup(dst6, 0, NULL)))
1063 			return;
1064 	}
1065 
1066 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1067 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1068 	    !rt->rt_llinfo || !rt->rt_gateway ||
1069 	    rt->rt_gateway->sa_family != AF_LINK) {
1070 		/* This is not a host route. */
1071 		return;
1072 	}
1073 
1074 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1075 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1076 		return;
1077 
1078 	/*
1079 	 * if we get upper-layer reachability confirmation many times,
1080 	 * it is possible we have false information.
1081 	 */
1082 	if (!force) {
1083 		ln->ln_byhint++;
1084 		if (ln->ln_byhint > nd6_maxnudhint)
1085 			return;
1086 	}
1087 
1088 	ln->ln_state = ND6_LLINFO_REACHABLE;
1089 	if (ln->ln_expire)
1090 		ln->ln_expire = time_second +
1091 			nd_ifinfo[rt->rt_ifp->if_index].reachable;
1092 }
1093 
1094 void
1095 nd6_rtrequest(req, rt, info)
1096 	int	req;
1097 	struct rtentry *rt;
1098 	struct rt_addrinfo *info; /* xxx unused */
1099 {
1100 	struct sockaddr *gate = rt->rt_gateway;
1101 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1102 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1103 	struct ifnet *ifp = rt->rt_ifp;
1104 	struct ifaddr *ifa;
1105 
1106 	if ((rt->rt_flags & RTF_GATEWAY))
1107 		return;
1108 
1109 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1110 		/*
1111 		 * This is probably an interface direct route for a link
1112 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1113 		 * We do not need special treatment below for such a route.
1114 		 * Moreover, the RTF_LLINFO flag which would be set below
1115 		 * would annoy the ndp(8) command.
1116 		 */
1117 		return;
1118 	}
1119 
1120 	if (req == RTM_RESOLVE &&
1121 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1122 	     !nd6_is_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), ifp))) {
1123 		/*
1124 		 * FreeBSD and BSD/OS often make a cloned host route based
1125 		 * on a less-specific route (e.g. the default route).
1126 		 * If the less specific route does not have a "gateway"
1127 		 * (this is the case when the route just goes to a p2p or an
1128 		 * stf interface), we'll mistakenly make a neighbor cache for
1129 		 * the host route, and will see strange neighbor solicitation
1130 		 * for the corresponding destination.  In order to avoid the
1131 		 * confusion, we check if the destination of the route is
1132 		 * a neighbor in terms of neighbor discovery, and stop the
1133 		 * process if not.  Additionally, we remove the LLINFO flag
1134 		 * so that ndp(8) will not try to get the neighbor information
1135 		 * of the destination.
1136 		 */
1137 		rt->rt_flags &= ~RTF_LLINFO;
1138 		return;
1139 	}
1140 
1141 	switch (req) {
1142 	case RTM_ADD:
1143 		/*
1144 		 * There is no backward compatibility :)
1145 		 *
1146 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1147 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1148 		 *	   rt->rt_flags |= RTF_CLONING;
1149 		 */
1150 		if (rt->rt_flags & (RTF_CLONING | RTF_LLINFO)) {
1151 			/*
1152 			 * Case 1: This route should come from
1153 			 * a route to interface.  RTF_LLINFO flag is set
1154 			 * for a host route whose destination should be
1155 			 * treated as on-link.
1156 			 */
1157 			rt_setgate(rt, rt_key(rt),
1158 				   (struct sockaddr *)&null_sdl);
1159 			gate = rt->rt_gateway;
1160 			SDL(gate)->sdl_type = ifp->if_type;
1161 			SDL(gate)->sdl_index = ifp->if_index;
1162 			if (ln)
1163 				ln->ln_expire = time_second;
1164 #if 1
1165 			if (ln && ln->ln_expire == 0) {
1166 				/* kludge for desktops */
1167 #if 0
1168 				printf("nd6_rtequest: time.tv_sec is zero; "
1169 				       "treat it as 1\n");
1170 #endif
1171 				ln->ln_expire = 1;
1172 			}
1173 #endif
1174 			if ((rt->rt_flags & RTF_CLONING))
1175 				break;
1176 		}
1177 		/*
1178 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1179 		 * We don't do that here since llinfo is not ready yet.
1180 		 *
1181 		 * There are also couple of other things to be discussed:
1182 		 * - unsolicited NA code needs improvement beforehand
1183 		 * - RFC2461 says we MAY send multicast unsolicited NA
1184 		 *   (7.2.6 paragraph 4), however, it also says that we
1185 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1186 		 *   we don't have anything like it right now.
1187 		 *   note that the mechanism needs a mutual agreement
1188 		 *   between proxies, which means that we need to implement
1189 		 *   a new protocol, or a new kludge.
1190 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1191 		 *   we need to check ip6forwarding before sending it.
1192 		 *   (or should we allow proxy ND configuration only for
1193 		 *   routers?  there's no mention about proxy ND from hosts)
1194 		 */
1195 #if 0
1196 		/* XXX it does not work */
1197 		if (rt->rt_flags & RTF_ANNOUNCE)
1198 			nd6_na_output(ifp,
1199 			      &SIN6(rt_key(rt))->sin6_addr,
1200 			      &SIN6(rt_key(rt))->sin6_addr,
1201 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1202 			      1, NULL);
1203 #endif
1204 		/* FALLTHROUGH */
1205 	case RTM_RESOLVE:
1206 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1207 			/*
1208 			 * Address resolution isn't necessary for a point to
1209 			 * point link, so we can skip this test for a p2p link.
1210 			 */
1211 			if (gate->sa_family != AF_LINK ||
1212 			    gate->sa_len < sizeof(null_sdl)) {
1213 				log(LOG_DEBUG,
1214 				    "nd6_rtrequest: bad gateway value: %s\n",
1215 				    if_name(ifp));
1216 				break;
1217 			}
1218 			SDL(gate)->sdl_type = ifp->if_type;
1219 			SDL(gate)->sdl_index = ifp->if_index;
1220 		}
1221 		if (ln != NULL)
1222 			break;	/* This happens on a route change */
1223 		/*
1224 		 * Case 2: This route may come from cloning, or a manual route
1225 		 * add with a LL address.
1226 		 */
1227 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1228 		rt->rt_llinfo = (caddr_t)ln;
1229 		if (!ln) {
1230 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1231 			break;
1232 		}
1233 		nd6_inuse++;
1234 		nd6_allocated++;
1235 		Bzero(ln, sizeof(*ln));
1236 		ln->ln_rt = rt;
1237 		/* this is required for "ndp" command. - shin */
1238 		if (req == RTM_ADD) {
1239 		        /*
1240 			 * gate should have some valid AF_LINK entry,
1241 			 * and ln->ln_expire should have some lifetime
1242 			 * which is specified by ndp command.
1243 			 */
1244 			ln->ln_state = ND6_LLINFO_REACHABLE;
1245 			ln->ln_byhint = 0;
1246 		} else {
1247 		        /*
1248 			 * When req == RTM_RESOLVE, rt is created and
1249 			 * initialized in rtrequest(), so rt_expire is 0.
1250 			 */
1251 			ln->ln_state = ND6_LLINFO_NOSTATE;
1252 			ln->ln_expire = time_second;
1253 		}
1254 		rt->rt_flags |= RTF_LLINFO;
1255 		ln->ln_next = llinfo_nd6.ln_next;
1256 		llinfo_nd6.ln_next = ln;
1257 		ln->ln_prev = &llinfo_nd6;
1258 		ln->ln_next->ln_prev = ln;
1259 
1260 		/*
1261 		 * check if rt_key(rt) is one of my address assigned
1262 		 * to the interface.
1263 		 */
1264 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1265 					  &SIN6(rt_key(rt))->sin6_addr);
1266 		if (ifa) {
1267 			caddr_t macp = nd6_ifptomac(ifp);
1268 			ln->ln_expire = 0;
1269 			ln->ln_state = ND6_LLINFO_REACHABLE;
1270 			ln->ln_byhint = 0;
1271 			if (macp) {
1272 				Bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1273 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1274 			}
1275 			if (nd6_useloopback) {
1276 				rt->rt_ifp = &loif[0];	/* XXX */
1277 				/*
1278 				 * Make sure rt_ifa be equal to the ifaddr
1279 				 * corresponding to the address.
1280 				 * We need this because when we refer
1281 				 * rt_ifa->ia6_flags in ip6_input, we assume
1282 				 * that the rt_ifa points to the address instead
1283 				 * of the loopback address.
1284 				 */
1285 				if (ifa != rt->rt_ifa) {
1286 					IFAFREE(rt->rt_ifa);
1287 					IFAREF(ifa);
1288 					rt->rt_ifa = ifa;
1289 				}
1290 			}
1291 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1292 			ln->ln_expire = 0;
1293 			ln->ln_state = ND6_LLINFO_REACHABLE;
1294 			ln->ln_byhint = 0;
1295 
1296 			/* join solicited node multicast for proxy ND */
1297 			if (ifp->if_flags & IFF_MULTICAST) {
1298 				struct in6_addr llsol;
1299 				int error;
1300 
1301 				llsol = SIN6(rt_key(rt))->sin6_addr;
1302 				llsol.s6_addr16[0] = htons(0xff02);
1303 				llsol.s6_addr16[1] = htons(ifp->if_index);
1304 				llsol.s6_addr32[1] = 0;
1305 				llsol.s6_addr32[2] = htonl(1);
1306 				llsol.s6_addr8[12] = 0xff;
1307 
1308 				if (!in6_addmulti(&llsol, ifp, &error)) {
1309 					nd6log((LOG_ERR, "%s: failed to join "
1310 					    "%s (errno=%d)\n", if_name(ifp),
1311 					    ip6_sprintf(&llsol), error));
1312 				}
1313 			}
1314 		}
1315 		break;
1316 
1317 	case RTM_DELETE:
1318 		if (!ln)
1319 			break;
1320 		/* leave from solicited node multicast for proxy ND */
1321 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1322 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1323 			struct in6_addr llsol;
1324 			struct in6_multi *in6m;
1325 
1326 			llsol = SIN6(rt_key(rt))->sin6_addr;
1327 			llsol.s6_addr16[0] = htons(0xff02);
1328 			llsol.s6_addr16[1] = htons(ifp->if_index);
1329 			llsol.s6_addr32[1] = 0;
1330 			llsol.s6_addr32[2] = htonl(1);
1331 			llsol.s6_addr8[12] = 0xff;
1332 
1333 			IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1334 			if (in6m)
1335 				in6_delmulti(in6m);
1336 		}
1337 		nd6_inuse--;
1338 		ln->ln_next->ln_prev = ln->ln_prev;
1339 		ln->ln_prev->ln_next = ln->ln_next;
1340 		ln->ln_prev = NULL;
1341 		rt->rt_llinfo = 0;
1342 		rt->rt_flags &= ~RTF_LLINFO;
1343 		if (ln->ln_hold)
1344 			m_freem(ln->ln_hold);
1345 		Free((caddr_t)ln);
1346 	}
1347 }
1348 
1349 int
1350 nd6_ioctl(cmd, data, ifp)
1351 	u_long cmd;
1352 	caddr_t	data;
1353 	struct ifnet *ifp;
1354 {
1355 	struct in6_drlist *drl = (struct in6_drlist *)data;
1356 	struct in6_prlist *prl = (struct in6_prlist *)data;
1357 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1358 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1359 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1360 	struct nd_defrouter *dr, any;
1361 	struct nd_prefix *pr;
1362 	struct rtentry *rt;
1363 	int i = 0, error = 0;
1364 	int s;
1365 
1366 	switch (cmd) {
1367 	case SIOCGDRLST_IN6:
1368 		/*
1369 		 * obsolete API, use sysctl under net.inet6.icmp6
1370 		 */
1371 		bzero(drl, sizeof(*drl));
1372 		s = splnet();
1373 		dr = TAILQ_FIRST(&nd_defrouter);
1374 		while (dr && i < DRLSTSIZ) {
1375 			drl->defrouter[i].rtaddr = dr->rtaddr;
1376 			if (IN6_IS_ADDR_LINKLOCAL(&drl->defrouter[i].rtaddr)) {
1377 				/* XXX: need to this hack for KAME stack */
1378 				drl->defrouter[i].rtaddr.s6_addr16[1] = 0;
1379 			} else
1380 				log(LOG_ERR,
1381 				    "default router list contains a "
1382 				    "non-linklocal address(%s)\n",
1383 				    ip6_sprintf(&drl->defrouter[i].rtaddr));
1384 
1385 			drl->defrouter[i].flags = dr->flags;
1386 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1387 			drl->defrouter[i].expire = dr->expire;
1388 			drl->defrouter[i].if_index = dr->ifp->if_index;
1389 			i++;
1390 			dr = TAILQ_NEXT(dr, dr_entry);
1391 		}
1392 		splx(s);
1393 		break;
1394 	case SIOCGPRLST_IN6:
1395 		/*
1396 		 * obsolete API, use sysctl under net.inet6.icmp6
1397 		 */
1398 		/*
1399 		 * XXX meaning of fields, especialy "raflags", is very
1400 		 * differnet between RA prefix list and RR/static prefix list.
1401 		 * how about separating ioctls into two?
1402 		 */
1403 		bzero(prl, sizeof(*prl));
1404 		s = splnet();
1405 		pr = nd_prefix.lh_first;
1406 		while (pr && i < PRLSTSIZ) {
1407 			struct nd_pfxrouter *pfr;
1408 			int j;
1409 
1410 			(void)in6_embedscope(&prl->prefix[i].prefix,
1411 			    &pr->ndpr_prefix, NULL, NULL);
1412 			prl->prefix[i].raflags = pr->ndpr_raf;
1413 			prl->prefix[i].prefixlen = pr->ndpr_plen;
1414 			prl->prefix[i].vltime = pr->ndpr_vltime;
1415 			prl->prefix[i].pltime = pr->ndpr_pltime;
1416 			prl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1417 			prl->prefix[i].expire = pr->ndpr_expire;
1418 
1419 			pfr = pr->ndpr_advrtrs.lh_first;
1420 			j = 0;
1421 			while (pfr) {
1422 				if (j < DRLSTSIZ) {
1423 #define RTRADDR prl->prefix[i].advrtr[j]
1424 					RTRADDR = pfr->router->rtaddr;
1425 					if (IN6_IS_ADDR_LINKLOCAL(&RTRADDR)) {
1426 						/* XXX: hack for KAME */
1427 						RTRADDR.s6_addr16[1] = 0;
1428 					} else
1429 						log(LOG_ERR,
1430 						    "a router(%s) advertises "
1431 						    "a prefix with "
1432 						    "non-link local address\n",
1433 						    ip6_sprintf(&RTRADDR));
1434 #undef RTRADDR
1435 				}
1436 				j++;
1437 				pfr = pfr->pfr_next;
1438 			}
1439 			prl->prefix[i].advrtrs = j;
1440 			prl->prefix[i].origin = PR_ORIG_RA;
1441 
1442 			i++;
1443 			pr = pr->ndpr_next;
1444 		}
1445 	      {
1446 		struct rr_prefix *rpp;
1447 
1448 		for (rpp = LIST_FIRST(&rr_prefix); rpp;
1449 		     rpp = LIST_NEXT(rpp, rp_entry)) {
1450 			if (i >= PRLSTSIZ)
1451 				break;
1452 			(void)in6_embedscope(&prl->prefix[i].prefix,
1453 			    &pr->ndpr_prefix, NULL, NULL);
1454 			prl->prefix[i].raflags = rpp->rp_raf;
1455 			prl->prefix[i].prefixlen = rpp->rp_plen;
1456 			prl->prefix[i].vltime = rpp->rp_vltime;
1457 			prl->prefix[i].pltime = rpp->rp_pltime;
1458 			prl->prefix[i].if_index = rpp->rp_ifp->if_index;
1459 			prl->prefix[i].expire = rpp->rp_expire;
1460 			prl->prefix[i].advrtrs = 0;
1461 			prl->prefix[i].origin = rpp->rp_origin;
1462 			i++;
1463 		}
1464 	      }
1465 		splx(s);
1466 
1467 		break;
1468 	case OSIOCGIFINFO_IN6:
1469 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1470 			error = EINVAL;
1471 			break;
1472 		}
1473 		ndi->ndi.linkmtu = nd_ifinfo[ifp->if_index].linkmtu;
1474 		ndi->ndi.maxmtu = nd_ifinfo[ifp->if_index].maxmtu;
1475 		ndi->ndi.basereachable =
1476 		    nd_ifinfo[ifp->if_index].basereachable;
1477 		ndi->ndi.reachable = nd_ifinfo[ifp->if_index].reachable;
1478 		ndi->ndi.retrans = nd_ifinfo[ifp->if_index].retrans;
1479 		ndi->ndi.flags = nd_ifinfo[ifp->if_index].flags;
1480 		ndi->ndi.recalctm = nd_ifinfo[ifp->if_index].recalctm;
1481 		ndi->ndi.chlim = nd_ifinfo[ifp->if_index].chlim;
1482 		ndi->ndi.receivedra = nd_ifinfo[ifp->if_index].receivedra;
1483 		break;
1484 	case SIOCGIFINFO_IN6:
1485 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1486 			error = EINVAL;
1487 			break;
1488 		}
1489 		ndi->ndi = nd_ifinfo[ifp->if_index];
1490 		break;
1491 	case SIOCSIFINFO_FLAGS:
1492 		/* XXX: almost all other fields of ndi->ndi is unused */
1493 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim) {
1494 			error = EINVAL;
1495 			break;
1496 		}
1497 		nd_ifinfo[ifp->if_index].flags = ndi->ndi.flags;
1498 		break;
1499 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1500 		/* flush default router list */
1501 		/*
1502 		 * xxx sumikawa: should not delete route if default
1503 		 * route equals to the top of default router list
1504 		 */
1505 		bzero(&any, sizeof(any));
1506 		defrouter_delreq(&any, 0);
1507 		defrouter_select();
1508 		/* xxx sumikawa: flush prefix list */
1509 		break;
1510 	case SIOCSPFXFLUSH_IN6:
1511 	    {
1512 		/* flush all the prefix advertised by routers */
1513 		struct nd_prefix *pr, *next;
1514 
1515 		s = splnet();
1516 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1517 			struct in6_ifaddr *ia, *ia_next;
1518 
1519 			next = pr->ndpr_next;
1520 
1521 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1522 				continue; /* XXX */
1523 
1524 			/* do we really have to remove addresses as well? */
1525 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1526 				/* ia might be removed.  keep the next ptr. */
1527 				ia_next = ia->ia_next;
1528 
1529 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1530 					continue;
1531 
1532 				if (ia->ia6_ndpr == pr)
1533 					in6_purgeaddr(&ia->ia_ifa);
1534 			}
1535 			prelist_remove(pr);
1536 		}
1537 		splx(s);
1538 		break;
1539 	    }
1540 	case SIOCSRTRFLUSH_IN6:
1541 	    {
1542 		/* flush all the default routers */
1543 		struct nd_defrouter *dr, *next;
1544 
1545 		s = splnet();
1546 		if ((dr = TAILQ_FIRST(&nd_defrouter)) != NULL) {
1547 			/*
1548 			 * The first entry of the list may be stored in
1549 			 * the routing table, so we'll delete it later.
1550 			 */
1551 			for (dr = TAILQ_NEXT(dr, dr_entry); dr; dr = next) {
1552 				next = TAILQ_NEXT(dr, dr_entry);
1553 				defrtrlist_del(dr);
1554 			}
1555 			defrtrlist_del(TAILQ_FIRST(&nd_defrouter));
1556 		}
1557 		splx(s);
1558 		break;
1559 	    }
1560 	case SIOCGNBRINFO_IN6:
1561 	    {
1562 		struct llinfo_nd6 *ln;
1563 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1564 
1565 		/*
1566 		 * XXX: KAME specific hack for scoped addresses
1567 		 *      XXXX: for other scopes than link-local?
1568 		 */
1569 		if (IN6_IS_ADDR_LINKLOCAL(&nbi->addr) ||
1570 		    IN6_IS_ADDR_MC_LINKLOCAL(&nbi->addr)) {
1571 			u_int16_t *idp = (u_int16_t *)&nb_addr.s6_addr[2];
1572 
1573 			if (*idp == 0)
1574 				*idp = htons(ifp->if_index);
1575 		}
1576 
1577 		s = splnet();
1578 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1579 			error = EINVAL;
1580 			splx(s);
1581 			break;
1582 		}
1583 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1584 		nbi->state = ln->ln_state;
1585 		nbi->asked = ln->ln_asked;
1586 		nbi->isrouter = ln->ln_router;
1587 		nbi->expire = ln->ln_expire;
1588 		splx(s);
1589 
1590 		break;
1591 	    }
1592 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1593 		ndif->ifindex = nd6_defifindex;
1594 		break;
1595 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1596 		return(nd6_setdefaultiface(ndif->ifindex));
1597 		break;
1598 	}
1599 	return(error);
1600 }
1601 
1602 /*
1603  * Create neighbor cache entry and cache link-layer address,
1604  * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1605  */
1606 struct rtentry *
1607 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1608 	struct ifnet *ifp;
1609 	struct in6_addr *from;
1610 	char *lladdr;
1611 	int lladdrlen;
1612 	int type;	/* ICMP6 type */
1613 	int code;	/* type dependent information */
1614 {
1615 	struct rtentry *rt = NULL;
1616 	struct llinfo_nd6 *ln = NULL;
1617 	int is_newentry;
1618 	struct sockaddr_dl *sdl = NULL;
1619 	int do_update;
1620 	int olladdr;
1621 	int llchange;
1622 	int newstate = 0;
1623 
1624 	if (!ifp)
1625 		panic("ifp == NULL in nd6_cache_lladdr");
1626 	if (!from)
1627 		panic("from == NULL in nd6_cache_lladdr");
1628 
1629 	/* nothing must be updated for unspecified address */
1630 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1631 		return NULL;
1632 
1633 	/*
1634 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1635 	 * the caller.
1636 	 *
1637 	 * XXX If the link does not have link-layer adderss, what should
1638 	 * we do? (ifp->if_addrlen == 0)
1639 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1640 	 * description on it in NS section (RFC 2461 7.2.3).
1641 	 */
1642 
1643 	rt = nd6_lookup(from, 0, ifp);
1644 	if (!rt) {
1645 #if 0
1646 		/* nothing must be done if there's no lladdr */
1647 		if (!lladdr || !lladdrlen)
1648 			return NULL;
1649 #endif
1650 
1651 		rt = nd6_lookup(from, 1, ifp);
1652 		is_newentry = 1;
1653 	} else {
1654 		/* do nothing if static ndp is set */
1655 		if (rt->rt_flags & RTF_STATIC)
1656 			return NULL;
1657 		is_newentry = 0;
1658 	}
1659 
1660 	if (!rt)
1661 		return NULL;
1662 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1663 fail:
1664 		(void)nd6_free(rt);
1665 		return NULL;
1666 	}
1667 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1668 	if (!ln)
1669 		goto fail;
1670 	if (!rt->rt_gateway)
1671 		goto fail;
1672 	if (rt->rt_gateway->sa_family != AF_LINK)
1673 		goto fail;
1674 	sdl = SDL(rt->rt_gateway);
1675 
1676 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1677 	if (olladdr && lladdr) {
1678 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1679 			llchange = 1;
1680 		else
1681 			llchange = 0;
1682 	} else
1683 		llchange = 0;
1684 
1685 	/*
1686 	 * newentry olladdr  lladdr  llchange	(*=record)
1687 	 *	0	n	n	--	(1)
1688 	 *	0	y	n	--	(2)
1689 	 *	0	n	y	--	(3) * STALE
1690 	 *	0	y	y	n	(4) *
1691 	 *	0	y	y	y	(5) * STALE
1692 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1693 	 *	1	--	y	--	(7) * STALE
1694 	 */
1695 
1696 	if (lladdr) {		/* (3-5) and (7) */
1697 		/*
1698 		 * Record source link-layer address
1699 		 * XXX is it dependent to ifp->if_type?
1700 		 */
1701 		sdl->sdl_alen = ifp->if_addrlen;
1702 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1703 	}
1704 
1705 	if (!is_newentry) {
1706 		if ((!olladdr && lladdr)		/* (3) */
1707 		 || (olladdr && lladdr && llchange)) {	/* (5) */
1708 			do_update = 1;
1709 			newstate = ND6_LLINFO_STALE;
1710 		} else					/* (1-2,4) */
1711 			do_update = 0;
1712 	} else {
1713 		do_update = 1;
1714 		if (!lladdr)				/* (6) */
1715 			newstate = ND6_LLINFO_NOSTATE;
1716 		else					/* (7) */
1717 			newstate = ND6_LLINFO_STALE;
1718 	}
1719 
1720 	if (do_update) {
1721 		/*
1722 		 * Update the state of the neighbor cache.
1723 		 */
1724 		ln->ln_state = newstate;
1725 
1726 		if (ln->ln_state == ND6_LLINFO_STALE) {
1727 			/*
1728 			 * XXX: since nd6_output() below will cause
1729 			 * state tansition to DELAY and reset the timer,
1730 			 * we must set the timer now, although it is actually
1731 			 * meaningless.
1732 			 */
1733 			ln->ln_expire = time_second + nd6_gctimer;
1734 
1735 			if (ln->ln_hold) {
1736 				/*
1737 				 * we assume ifp is not a p2p here, so just
1738 				 * set the 2nd argument as the 1st one.
1739 				 */
1740 				nd6_output(ifp, ifp, ln->ln_hold,
1741 					   (struct sockaddr_in6 *)rt_key(rt),
1742 					   rt);
1743 				ln->ln_hold = NULL;
1744 			}
1745 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1746 			/* probe right away */
1747 			ln->ln_expire = time_second;
1748 		}
1749 	}
1750 
1751 	/*
1752 	 * ICMP6 type dependent behavior.
1753 	 *
1754 	 * NS: clear IsRouter if new entry
1755 	 * RS: clear IsRouter
1756 	 * RA: set IsRouter if there's lladdr
1757 	 * redir: clear IsRouter if new entry
1758 	 *
1759 	 * RA case, (1):
1760 	 * The spec says that we must set IsRouter in the following cases:
1761 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1762 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1763 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1764 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1765 	 * neighbor cache, this is similar to (6).
1766 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1767 	 *
1768 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1769 	 *							D R
1770 	 *	0	n	n	--	(1)	c   ?     s
1771 	 *	0	y	n	--	(2)	c   s     s
1772 	 *	0	n	y	--	(3)	c   s     s
1773 	 *	0	y	y	n	(4)	c   s     s
1774 	 *	0	y	y	y	(5)	c   s     s
1775 	 *	1	--	n	--	(6) c	c 	c s
1776 	 *	1	--	y	--	(7) c	c   s	c s
1777 	 *
1778 	 *					(c=clear s=set)
1779 	 */
1780 	switch (type & 0xff) {
1781 	case ND_NEIGHBOR_SOLICIT:
1782 		/*
1783 		 * New entry must have is_router flag cleared.
1784 		 */
1785 		if (is_newentry)	/* (6-7) */
1786 			ln->ln_router = 0;
1787 		break;
1788 	case ND_REDIRECT:
1789 		/*
1790 		 * If the icmp is a redirect to a better router, always set the
1791 		 * is_router flag. Otherwise, if the entry is newly created,
1792 		 * clear the flag. [RFC 2461, sec 8.3]
1793 		 */
1794 		if (code == ND_REDIRECT_ROUTER)
1795 			ln->ln_router = 1;
1796 		else if (is_newentry) /* (6-7) */
1797 			ln->ln_router = 0;
1798 		break;
1799 	case ND_ROUTER_SOLICIT:
1800 		/*
1801 		 * is_router flag must always be cleared.
1802 		 */
1803 		ln->ln_router = 0;
1804 		break;
1805 	case ND_ROUTER_ADVERT:
1806 		/*
1807 		 * Mark an entry with lladdr as a router.
1808 		 */
1809 		if ((!is_newentry && (olladdr || lladdr))	/* (2-5) */
1810 		 || (is_newentry && lladdr)) {			/* (7) */
1811 			ln->ln_router = 1;
1812 		}
1813 		break;
1814 	}
1815 
1816 	/*
1817 	 * When the link-layer address of a router changes, select the
1818 	 * best router again.  In particular, when the neighbor entry is newly
1819 	 * created, it might affect the selection policy.
1820 	 * Question: can we restrict the first condition to the "is_newentry"
1821 	 * case?
1822 	 * XXX: when we hear an RA from a new router with the link-layer
1823 	 * address option, defrouter_select() is called twice, since
1824 	 * defrtrlist_update called the function as well.  However, I believe
1825 	 * we can compromise the overhead, since it only happens the first
1826 	 * time.
1827 	 * XXX: although defrouter_select() should not have a bad effect
1828 	 * for those are not autoconfigured hosts, we explicitly avoid such
1829 	 * cases for safety.
1830 	 */
1831 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1832 		defrouter_select();
1833 
1834 	return rt;
1835 }
1836 
1837 static void
1838 nd6_slowtimo(ignored_arg)
1839     void *ignored_arg;
1840 {
1841 	int s = splnet();
1842 	int i;
1843 	struct nd_ifinfo *nd6if;
1844 
1845 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1846 	    nd6_slowtimo, NULL);
1847 	for (i = 1; i < if_index + 1; i++) {
1848 		if (!nd_ifinfo || i >= nd_ifinfo_indexlim)
1849 			continue;
1850 		nd6if = &nd_ifinfo[i];
1851 		if (nd6if->basereachable && /* already initialized */
1852 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1853 			/*
1854 			 * Since reachable time rarely changes by router
1855 			 * advertisements, we SHOULD insure that a new random
1856 			 * value gets recomputed at least once every few hours.
1857 			 * (RFC 2461, 6.3.4)
1858 			 */
1859 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1860 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1861 		}
1862 	}
1863 	splx(s);
1864 }
1865 
1866 #define senderr(e) { error = (e); goto bad;}
1867 int
1868 nd6_output(ifp, origifp, m0, dst, rt0)
1869 	struct ifnet *ifp;
1870 	struct ifnet *origifp;
1871 	struct mbuf *m0;
1872 	struct sockaddr_in6 *dst;
1873 	struct rtentry *rt0;
1874 {
1875 	struct mbuf *m = m0;
1876 	struct rtentry *rt = rt0;
1877 	struct sockaddr_in6 *gw6 = NULL;
1878 	struct llinfo_nd6 *ln = NULL;
1879 	int error = 0;
1880 
1881 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1882 		goto sendpkt;
1883 
1884 	if (nd6_need_cache(ifp) == 0)
1885 		goto sendpkt;
1886 
1887 	/*
1888 	 * next hop determination.  This routine is derived from ether_outpout.
1889 	 */
1890 	if (rt) {
1891 		if ((rt->rt_flags & RTF_UP) == 0) {
1892 			if ((rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL)) !=
1893 				NULL)
1894 			{
1895 				rt->rt_refcnt--;
1896 				if (rt->rt_ifp != ifp) {
1897 					/* XXX: loop care? */
1898 					return nd6_output(ifp, origifp, m0,
1899 							  dst, rt);
1900 				}
1901 			} else
1902 				senderr(EHOSTUNREACH);
1903 		}
1904 
1905 		if (rt->rt_flags & RTF_GATEWAY) {
1906 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1907 
1908 			/*
1909 			 * We skip link-layer address resolution and NUD
1910 			 * if the gateway is not a neighbor from ND point
1911 			 * of view, regardless of the value of nd_ifinfo.flags.
1912 			 * The second condition is a bit tricky; we skip
1913 			 * if the gateway is our own address, which is
1914 			 * sometimes used to install a route to a p2p link.
1915 			 */
1916 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1917 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1918 				/*
1919 				 * We allow this kind of tricky route only
1920 				 * when the outgoing interface is p2p.
1921 				 * XXX: we may need a more generic rule here.
1922 				 */
1923 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1924 					senderr(EHOSTUNREACH);
1925 
1926 				goto sendpkt;
1927 			}
1928 
1929 			if (rt->rt_gwroute == 0)
1930 				goto lookup;
1931 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1932 				rtfree(rt); rt = rt0;
1933 			lookup: rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1934 				if ((rt = rt->rt_gwroute) == 0)
1935 					senderr(EHOSTUNREACH);
1936 			}
1937 		}
1938 	}
1939 
1940 	/*
1941 	 * Address resolution or Neighbor Unreachability Detection
1942 	 * for the next hop.
1943 	 * At this point, the destination of the packet must be a unicast
1944 	 * or an anycast address(i.e. not a multicast).
1945 	 */
1946 
1947 	/* Look up the neighbor cache for the nexthop */
1948 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
1949 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1950 	else {
1951 		/*
1952 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1953 		 * the condition below is not very efficient.  But we believe
1954 		 * it is tolerable, because this should be a rare case.
1955 		 */
1956 		if (nd6_is_addr_neighbor(dst, ifp) &&
1957 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
1958 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1959 	}
1960 	if (!ln || !rt) {
1961 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1962 		    !(nd_ifinfo[ifp->if_index].flags & ND6_IFF_PERFORMNUD)) {
1963 			log(LOG_DEBUG,
1964 			    "nd6_output: can't allocate llinfo for %s "
1965 			    "(ln=%p, rt=%p)\n",
1966 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
1967 			senderr(EIO);	/* XXX: good error? */
1968 		}
1969 
1970 		goto sendpkt;	/* send anyway */
1971 	}
1972 
1973 	/* We don't have to do link-layer address resolution on a p2p link. */
1974 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1975 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1976 		ln->ln_state = ND6_LLINFO_STALE;
1977 		ln->ln_expire = time_second + nd6_gctimer;
1978 	}
1979 
1980 	/*
1981 	 * The first time we send a packet to a neighbor whose entry is
1982 	 * STALE, we have to change the state to DELAY and a sets a timer to
1983 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1984 	 * neighbor unreachability detection on expiration.
1985 	 * (RFC 2461 7.3.3)
1986 	 */
1987 	if (ln->ln_state == ND6_LLINFO_STALE) {
1988 		ln->ln_asked = 0;
1989 		ln->ln_state = ND6_LLINFO_DELAY;
1990 		ln->ln_expire = time_second + nd6_delay;
1991 	}
1992 
1993 	/*
1994 	 * If the neighbor cache entry has a state other than INCOMPLETE
1995 	 * (i.e. its link-layer address is already resolved), just
1996 	 * send the packet.
1997 	 */
1998 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1999 		goto sendpkt;
2000 
2001 	/*
2002 	 * There is a neighbor cache entry, but no ethernet address
2003 	 * response yet.  Replace the held mbuf (if any) with this
2004 	 * latest one.
2005 	 *
2006 	 * This code conforms to the rate-limiting rule described in Section
2007 	 * 7.2.2 of RFC 2461, because the timer is set correctly after sending
2008 	 * an NS below.
2009 	 */
2010 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2011 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2012 	if (ln->ln_hold)
2013 		m_freem(ln->ln_hold);
2014 	ln->ln_hold = m;
2015 	if (ln->ln_expire) {
2016 		if (ln->ln_asked < nd6_mmaxtries &&
2017 		    ln->ln_expire < time_second) {
2018 			ln->ln_asked++;
2019 			ln->ln_expire = time_second +
2020 				nd_ifinfo[ifp->if_index].retrans / 1000;
2021 			nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2022 		}
2023 	}
2024 	return(0);
2025 
2026   sendpkt:
2027 
2028 #ifdef MAC
2029 	mac_create_mbuf_linklayer(ifp, m);
2030 #endif
2031 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2032 		return((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2033 					 rt));
2034 	}
2035 	return((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2036 
2037   bad:
2038 	if (m)
2039 		m_freem(m);
2040 	return (error);
2041 }
2042 #undef senderr
2043 
2044 int
2045 nd6_need_cache(ifp)
2046 	struct ifnet *ifp;
2047 {
2048 	/*
2049 	 * XXX: we currently do not make neighbor cache on any interface
2050 	 * other than ARCnet, Ethernet, FDDI and GIF.
2051 	 *
2052 	 * RFC2893 says:
2053 	 * - unidirectional tunnels needs no ND
2054 	 */
2055 	switch (ifp->if_type) {
2056 	case IFT_ARCNET:
2057 	case IFT_ETHER:
2058 	case IFT_FDDI:
2059 	case IFT_IEEE1394:
2060 #ifdef IFT_L2VLAN
2061 	case IFT_L2VLAN:
2062 #endif
2063 #ifdef IFT_IEEE80211
2064 	case IFT_IEEE80211:
2065 #endif
2066 	case IFT_GIF:		/* XXX need more cases? */
2067 		return(1);
2068 	default:
2069 		return(0);
2070 	}
2071 }
2072 
2073 int
2074 nd6_storelladdr(ifp, rt, m, dst, desten)
2075 	struct ifnet *ifp;
2076 	struct rtentry *rt;
2077 	struct mbuf *m;
2078 	struct sockaddr *dst;
2079 	u_char *desten;
2080 {
2081 	int i;
2082 	struct sockaddr_dl *sdl;
2083 
2084 	if (m->m_flags & M_MCAST) {
2085 		switch (ifp->if_type) {
2086 		case IFT_ETHER:
2087 		case IFT_FDDI:
2088 #ifdef IFT_L2VLAN
2089 	case IFT_L2VLAN:
2090 #endif
2091 #ifdef IFT_IEEE80211
2092 		case IFT_IEEE80211:
2093 #endif
2094 		case IFT_ISO88025:
2095 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2096 						 desten);
2097 			return(1);
2098 		case IFT_IEEE1394:
2099 			/*
2100 			 * netbsd can use if_broadcastaddr, but we don't do so
2101 			 * to reduce # of ifdef.
2102 			 */
2103 			for (i = 0; i < ifp->if_addrlen; i++)
2104 				desten[i] = ~0;
2105 			return(1);
2106 		case IFT_ARCNET:
2107 			*desten = 0;
2108 			return(1);
2109 		default:
2110 			m_freem(m);
2111 			return(0);
2112 		}
2113 	}
2114 
2115 	if (rt == NULL) {
2116 		/* this could happen, if we could not allocate memory */
2117 		m_freem(m);
2118 		return(0);
2119 	}
2120 	if (rt->rt_gateway->sa_family != AF_LINK) {
2121 		printf("nd6_storelladdr: something odd happens\n");
2122 		m_freem(m);
2123 		return(0);
2124 	}
2125 	sdl = SDL(rt->rt_gateway);
2126 	if (sdl->sdl_alen == 0) {
2127 		/* this should be impossible, but we bark here for debugging */
2128 		printf("nd6_storelladdr: sdl_alen == 0\n");
2129 		m_freem(m);
2130 		return(0);
2131 	}
2132 
2133 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2134 	return(1);
2135 }
2136 
2137 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2138 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2139 #ifdef SYSCTL_DECL
2140 SYSCTL_DECL(_net_inet6_icmp6);
2141 #endif
2142 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2143 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2144 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2145 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2146 
2147 static int
2148 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2149 {
2150 	int error;
2151 	char buf[1024];
2152 	struct in6_defrouter *d, *de;
2153 	struct nd_defrouter *dr;
2154 
2155 	if (req->newptr)
2156 		return EPERM;
2157 	error = 0;
2158 
2159 	for (dr = TAILQ_FIRST(&nd_defrouter);
2160 	     dr;
2161 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2162 		d = (struct in6_defrouter *)buf;
2163 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2164 
2165 		if (d + 1 <= de) {
2166 			bzero(d, sizeof(*d));
2167 			d->rtaddr.sin6_family = AF_INET6;
2168 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2169 			if (in6_recoverscope(&d->rtaddr, &dr->rtaddr,
2170 			    dr->ifp) != 0)
2171 				log(LOG_ERR,
2172 				    "scope error in "
2173 				    "default router list (%s)\n",
2174 				    ip6_sprintf(&dr->rtaddr));
2175 			d->flags = dr->flags;
2176 			d->rtlifetime = dr->rtlifetime;
2177 			d->expire = dr->expire;
2178 			d->if_index = dr->ifp->if_index;
2179 		} else
2180 			panic("buffer too short");
2181 
2182 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2183 		if (error)
2184 			break;
2185 	}
2186 	return error;
2187 }
2188 
2189 static int
2190 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2191 {
2192 	int error;
2193 	char buf[1024];
2194 	struct in6_prefix *p, *pe;
2195 	struct nd_prefix *pr;
2196 
2197 	if (req->newptr)
2198 		return EPERM;
2199 	error = 0;
2200 
2201 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2202 		u_short advrtrs;
2203 		size_t advance;
2204 		struct sockaddr_in6 *sin6, *s6;
2205 		struct nd_pfxrouter *pfr;
2206 
2207 		p = (struct in6_prefix *)buf;
2208 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2209 
2210 		if (p + 1 <= pe) {
2211 			bzero(p, sizeof(*p));
2212 			sin6 = (struct sockaddr_in6 *)(p + 1);
2213 
2214 			p->prefix = pr->ndpr_prefix;
2215 			if (in6_recoverscope(&p->prefix,
2216 			    &p->prefix.sin6_addr, pr->ndpr_ifp) != 0)
2217 				log(LOG_ERR,
2218 				    "scope error in prefix list (%s)\n",
2219 				    ip6_sprintf(&p->prefix.sin6_addr));
2220 			p->raflags = pr->ndpr_raf;
2221 			p->prefixlen = pr->ndpr_plen;
2222 			p->vltime = pr->ndpr_vltime;
2223 			p->pltime = pr->ndpr_pltime;
2224 			p->if_index = pr->ndpr_ifp->if_index;
2225 			p->expire = pr->ndpr_expire;
2226 			p->refcnt = pr->ndpr_refcnt;
2227 			p->flags = pr->ndpr_stateflags;
2228 			p->origin = PR_ORIG_RA;
2229 			advrtrs = 0;
2230 			for (pfr = pr->ndpr_advrtrs.lh_first;
2231 			     pfr;
2232 			     pfr = pfr->pfr_next) {
2233 				if ((void *)&sin6[advrtrs + 1] >
2234 				    (void *)pe) {
2235 					advrtrs++;
2236 					continue;
2237 				}
2238 				s6 = &sin6[advrtrs];
2239 				bzero(s6, sizeof(*s6));
2240 				s6->sin6_family = AF_INET6;
2241 				s6->sin6_len = sizeof(*sin6);
2242 				if (in6_recoverscope(s6,
2243 				    &pfr->router->rtaddr,
2244 				    pfr->router->ifp) != 0)
2245 					log(LOG_ERR,
2246 					    "scope error in "
2247 					    "prefix list (%s)\n",
2248 					    ip6_sprintf(&pfr->router->rtaddr));
2249 				advrtrs++;
2250 			}
2251 			p->advrtrs = advrtrs;
2252 		} else
2253 			panic("buffer too short");
2254 
2255 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2256 		error = SYSCTL_OUT(req, buf, advance);
2257 		if (error)
2258 			break;
2259 	}
2260 	return error;
2261 }
2262