xref: /freebsd/sys/netinet6/nd6.c (revision 262e143bd46171a6415a5b28af260a5efa2a3db8)
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/iso88025.h>
58 #include <net/fddi.h>
59 #include <net/route.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/if_ether.h>
63 #include <netinet6/in6_var.h>
64 #include <netinet/ip6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet6/scope6_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet/icmp6.h>
69 
70 #include <sys/limits.h>
71 
72 #include <net/net_osdep.h>
73 
74 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
75 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
76 
77 #define SIN6(s) ((struct sockaddr_in6 *)s)
78 #define SDL(s) ((struct sockaddr_dl *)s)
79 
80 /* timer values */
81 int	nd6_prune	= 1;	/* walk list every 1 seconds */
82 int	nd6_delay	= 5;	/* delay first probe time 5 second */
83 int	nd6_umaxtries	= 3;	/* maximum unicast query */
84 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
85 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
86 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
87 
88 /* preventing too many loops in ND option parsing */
89 int nd6_maxndopt = 10;	/* max # of ND options allowed */
90 
91 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
92 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
93 
94 #ifdef ND6_DEBUG
95 int nd6_debug = 1;
96 #else
97 int nd6_debug = 0;
98 #endif
99 
100 /* for debugging? */
101 static int nd6_inuse, nd6_allocated;
102 
103 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6};
104 struct nd_drhead nd_defrouter;
105 struct nd_prhead nd_prefix = { 0 };
106 
107 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
108 static struct sockaddr_in6 all1_sa;
109 
110 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
111 	struct ifnet *));
112 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *));
113 static void nd6_slowtimo __P((void *));
114 static int regen_tmpaddr __P((struct in6_ifaddr *));
115 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int));
116 static void nd6_llinfo_timer __P((void *));
117 
118 struct callout nd6_slowtimo_ch;
119 struct callout nd6_timer_ch;
120 extern struct callout in6_tmpaddrtimer_ch;
121 
122 void
123 nd6_init()
124 {
125 	static int nd6_init_done = 0;
126 	int i;
127 
128 	if (nd6_init_done) {
129 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
130 		return;
131 	}
132 
133 	all1_sa.sin6_family = AF_INET6;
134 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
135 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
136 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
137 
138 	/* initialization of the default router list */
139 	TAILQ_INIT(&nd_defrouter);
140 
141 	nd6_init_done = 1;
142 
143 	/* start timer */
144 	callout_init(&nd6_slowtimo_ch, 0);
145 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
146 	    nd6_slowtimo, NULL);
147 }
148 
149 struct nd_ifinfo *
150 nd6_ifattach(ifp)
151 	struct ifnet *ifp;
152 {
153 	struct nd_ifinfo *nd;
154 
155 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
156 	bzero(nd, sizeof(*nd));
157 
158 	nd->initialized = 1;
159 
160 	nd->chlim = IPV6_DEFHLIM;
161 	nd->basereachable = REACHABLE_TIME;
162 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
163 	nd->retrans = RETRANS_TIMER;
164 	/*
165 	 * Note that the default value of ip6_accept_rtadv is 0, which means
166 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
167 	 * here.
168 	 */
169 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
170 
171 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
172 	nd6_setmtu0(ifp, nd);
173 
174 	return nd;
175 }
176 
177 void
178 nd6_ifdetach(nd)
179 	struct nd_ifinfo *nd;
180 {
181 
182 	free(nd, M_IP6NDP);
183 }
184 
185 /*
186  * Reset ND level link MTU. This function is called when the physical MTU
187  * changes, which means we might have to adjust the ND level MTU.
188  */
189 void
190 nd6_setmtu(ifp)
191 	struct ifnet *ifp;
192 {
193 
194 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
195 }
196 
197 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
198 void
199 nd6_setmtu0(ifp, ndi)
200 	struct ifnet *ifp;
201 	struct nd_ifinfo *ndi;
202 {
203 	u_int32_t omaxmtu;
204 
205 	omaxmtu = ndi->maxmtu;
206 
207 	switch (ifp->if_type) {
208 	case IFT_ARCNET:
209 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
210 		break;
211 	case IFT_FDDI:
212 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
213 		break;
214 	case IFT_ISO88025:
215 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
216 		 break;
217 	default:
218 		ndi->maxmtu = ifp->if_mtu;
219 		break;
220 	}
221 
222 	/*
223 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
224 	 * undesirable situation.  We thus notify the operator of the change
225 	 * explicitly.  The check for omaxmtu is necessary to restrict the
226 	 * log to the case of changing the MTU, not initializing it.
227 	 */
228 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
229 		log(LOG_NOTICE, "nd6_setmtu0: "
230 		    "new link MTU on %s (%lu) is too small for IPv6\n",
231 		    if_name(ifp), (unsigned long)ndi->maxmtu);
232 	}
233 
234 	if (ndi->maxmtu > in6_maxmtu)
235 		in6_setmaxmtu(); /* check all interfaces just in case */
236 
237 #undef MIN
238 }
239 
240 void
241 nd6_option_init(opt, icmp6len, ndopts)
242 	void *opt;
243 	int icmp6len;
244 	union nd_opts *ndopts;
245 {
246 
247 	bzero(ndopts, sizeof(*ndopts));
248 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
249 	ndopts->nd_opts_last
250 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
251 
252 	if (icmp6len == 0) {
253 		ndopts->nd_opts_done = 1;
254 		ndopts->nd_opts_search = NULL;
255 	}
256 }
257 
258 /*
259  * Take one ND option.
260  */
261 struct nd_opt_hdr *
262 nd6_option(ndopts)
263 	union nd_opts *ndopts;
264 {
265 	struct nd_opt_hdr *nd_opt;
266 	int olen;
267 
268 	if (ndopts == NULL)
269 		panic("ndopts == NULL in nd6_option");
270 	if (ndopts->nd_opts_last == NULL)
271 		panic("uninitialized ndopts in nd6_option");
272 	if (ndopts->nd_opts_search == NULL)
273 		return NULL;
274 	if (ndopts->nd_opts_done)
275 		return NULL;
276 
277 	nd_opt = ndopts->nd_opts_search;
278 
279 	/* make sure nd_opt_len is inside the buffer */
280 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
281 		bzero(ndopts, sizeof(*ndopts));
282 		return NULL;
283 	}
284 
285 	olen = nd_opt->nd_opt_len << 3;
286 	if (olen == 0) {
287 		/*
288 		 * Message validation requires that all included
289 		 * options have a length that is greater than zero.
290 		 */
291 		bzero(ndopts, sizeof(*ndopts));
292 		return NULL;
293 	}
294 
295 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
296 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
297 		/* option overruns the end of buffer, invalid */
298 		bzero(ndopts, sizeof(*ndopts));
299 		return NULL;
300 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
301 		/* reached the end of options chain */
302 		ndopts->nd_opts_done = 1;
303 		ndopts->nd_opts_search = NULL;
304 	}
305 	return nd_opt;
306 }
307 
308 /*
309  * Parse multiple ND options.
310  * This function is much easier to use, for ND routines that do not need
311  * multiple options of the same type.
312  */
313 int
314 nd6_options(ndopts)
315 	union nd_opts *ndopts;
316 {
317 	struct nd_opt_hdr *nd_opt;
318 	int i = 0;
319 
320 	if (ndopts == NULL)
321 		panic("ndopts == NULL in nd6_options");
322 	if (ndopts->nd_opts_last == NULL)
323 		panic("uninitialized ndopts in nd6_options");
324 	if (ndopts->nd_opts_search == NULL)
325 		return 0;
326 
327 	while (1) {
328 		nd_opt = nd6_option(ndopts);
329 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
330 			/*
331 			 * Message validation requires that all included
332 			 * options have a length that is greater than zero.
333 			 */
334 			icmp6stat.icp6s_nd_badopt++;
335 			bzero(ndopts, sizeof(*ndopts));
336 			return -1;
337 		}
338 
339 		if (nd_opt == NULL)
340 			goto skip1;
341 
342 		switch (nd_opt->nd_opt_type) {
343 		case ND_OPT_SOURCE_LINKADDR:
344 		case ND_OPT_TARGET_LINKADDR:
345 		case ND_OPT_MTU:
346 		case ND_OPT_REDIRECTED_HEADER:
347 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
348 				nd6log((LOG_INFO,
349 				    "duplicated ND6 option found (type=%d)\n",
350 				    nd_opt->nd_opt_type));
351 				/* XXX bark? */
352 			} else {
353 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
354 					= nd_opt;
355 			}
356 			break;
357 		case ND_OPT_PREFIX_INFORMATION:
358 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
359 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
360 					= nd_opt;
361 			}
362 			ndopts->nd_opts_pi_end =
363 				(struct nd_opt_prefix_info *)nd_opt;
364 			break;
365 		default:
366 			/*
367 			 * Unknown options must be silently ignored,
368 			 * to accomodate future extension to the protocol.
369 			 */
370 			nd6log((LOG_DEBUG,
371 			    "nd6_options: unsupported option %d - "
372 			    "option ignored\n", nd_opt->nd_opt_type));
373 		}
374 
375 skip1:
376 		i++;
377 		if (i > nd6_maxndopt) {
378 			icmp6stat.icp6s_nd_toomanyopt++;
379 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
380 			break;
381 		}
382 
383 		if (ndopts->nd_opts_done)
384 			break;
385 	}
386 
387 	return 0;
388 }
389 
390 /*
391  * ND6 timer routine to handle ND6 entries
392  */
393 void
394 nd6_llinfo_settimer(ln, tick)
395 	struct llinfo_nd6 *ln;
396 	long tick;
397 {
398 	if (tick < 0) {
399 		ln->ln_expire = 0;
400 		ln->ln_ntick = 0;
401 		callout_stop(&ln->ln_timer_ch);
402 	} else {
403 		ln->ln_expire = time_second + tick / hz;
404 		if (tick > INT_MAX) {
405 			ln->ln_ntick = tick - INT_MAX;
406 			callout_reset(&ln->ln_timer_ch, INT_MAX,
407 			    nd6_llinfo_timer, ln);
408 		} else {
409 			ln->ln_ntick = 0;
410 			callout_reset(&ln->ln_timer_ch, tick,
411 			    nd6_llinfo_timer, ln);
412 		}
413 	}
414 }
415 
416 static void
417 nd6_llinfo_timer(arg)
418 	void *arg;
419 {
420 	struct llinfo_nd6 *ln;
421 	struct rtentry *rt;
422 	struct in6_addr *dst;
423 	struct ifnet *ifp;
424 	struct nd_ifinfo *ndi = NULL;
425 
426 	ln = (struct llinfo_nd6 *)arg;
427 
428 	if (ln->ln_ntick > 0) {
429 		if (ln->ln_ntick > INT_MAX) {
430 			ln->ln_ntick -= INT_MAX;
431 			nd6_llinfo_settimer(ln, INT_MAX);
432 		} else {
433 			ln->ln_ntick = 0;
434 			nd6_llinfo_settimer(ln, ln->ln_ntick);
435 		}
436 		return;
437 	}
438 
439 	if ((rt = ln->ln_rt) == NULL)
440 		panic("ln->ln_rt == NULL");
441 	if ((ifp = rt->rt_ifp) == NULL)
442 		panic("ln->ln_rt->rt_ifp == NULL");
443 	ndi = ND_IFINFO(ifp);
444 
445 	/* sanity check */
446 	if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln)
447 		panic("rt_llinfo(%p) is not equal to ln(%p)",
448 		      rt->rt_llinfo, ln);
449 	if (rt_key(rt) == NULL)
450 		panic("rt key is NULL in nd6_timer(ln=%p)", ln);
451 
452 	dst = &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
453 
454 	switch (ln->ln_state) {
455 	case ND6_LLINFO_INCOMPLETE:
456 		if (ln->ln_asked < nd6_mmaxtries) {
457 			ln->ln_asked++;
458 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
459 			nd6_ns_output(ifp, NULL, dst, ln, 0);
460 		} else {
461 			struct mbuf *m = ln->ln_hold;
462 			if (m) {
463 				/*
464 				 * assuming every packet in ln_hold has the
465 				 * same IP header
466 				 */
467 				ln->ln_hold = NULL;
468 				icmp6_error2(m, ICMP6_DST_UNREACH,
469 				    ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp);
470 			}
471 			if (rt)
472 				(void)nd6_free(rt, 0);
473 			ln = NULL;
474 		}
475 		break;
476 	case ND6_LLINFO_REACHABLE:
477 		if (!ND6_LLINFO_PERMANENT(ln)) {
478 			ln->ln_state = ND6_LLINFO_STALE;
479 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
480 		}
481 		break;
482 
483 	case ND6_LLINFO_STALE:
484 		/* Garbage Collection(RFC 2461 5.3) */
485 		if (!ND6_LLINFO_PERMANENT(ln)) {
486 			(void)nd6_free(rt, 1);
487 			ln = NULL;
488 		}
489 		break;
490 
491 	case ND6_LLINFO_DELAY:
492 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
493 			/* We need NUD */
494 			ln->ln_asked = 1;
495 			ln->ln_state = ND6_LLINFO_PROBE;
496 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
497 			nd6_ns_output(ifp, dst, dst, ln, 0);
498 		} else {
499 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
500 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
501 		}
502 		break;
503 	case ND6_LLINFO_PROBE:
504 		if (ln->ln_asked < nd6_umaxtries) {
505 			ln->ln_asked++;
506 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
507 			nd6_ns_output(ifp, dst, dst, ln, 0);
508 		} else {
509 			(void)nd6_free(rt, 0);
510 			ln = NULL;
511 		}
512 		break;
513 	}
514 }
515 
516 
517 /*
518  * ND6 timer routine to expire default route list and prefix list
519  */
520 void
521 nd6_timer(ignored_arg)
522 	void	*ignored_arg;
523 {
524 	int s;
525 	struct nd_defrouter *dr;
526 	struct nd_prefix *pr;
527 	struct in6_ifaddr *ia6, *nia6;
528 	struct in6_addrlifetime *lt6;
529 
530 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
531 	    nd6_timer, NULL);
532 
533 	/* expire default router list */
534 	s = splnet();
535 	dr = TAILQ_FIRST(&nd_defrouter);
536 	while (dr) {
537 		if (dr->expire && dr->expire < time_second) {
538 			struct nd_defrouter *t;
539 			t = TAILQ_NEXT(dr, dr_entry);
540 			defrtrlist_del(dr);
541 			dr = t;
542 		} else {
543 			dr = TAILQ_NEXT(dr, dr_entry);
544 		}
545 	}
546 
547 	/*
548 	 * expire interface addresses.
549 	 * in the past the loop was inside prefix expiry processing.
550 	 * However, from a stricter speci-confrmance standpoint, we should
551 	 * rather separate address lifetimes and prefix lifetimes.
552 	 */
553   addrloop:
554 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
555 		nia6 = ia6->ia_next;
556 		/* check address lifetime */
557 		lt6 = &ia6->ia6_lifetime;
558 		if (IFA6_IS_INVALID(ia6)) {
559 			int regen = 0;
560 
561 			/*
562 			 * If the expiring address is temporary, try
563 			 * regenerating a new one.  This would be useful when
564 			 * we suspended a laptop PC, then turned it on after a
565 			 * period that could invalidate all temporary
566 			 * addresses.  Although we may have to restart the
567 			 * loop (see below), it must be after purging the
568 			 * address.  Otherwise, we'd see an infinite loop of
569 			 * regeneration.
570 			 */
571 			if (ip6_use_tempaddr &&
572 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
573 				if (regen_tmpaddr(ia6) == 0)
574 					regen = 1;
575 			}
576 
577 			in6_purgeaddr(&ia6->ia_ifa);
578 
579 			if (regen)
580 				goto addrloop; /* XXX: see below */
581 		}
582 		if (IFA6_IS_DEPRECATED(ia6)) {
583 			int oldflags = ia6->ia6_flags;
584 
585 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
586 
587 			/*
588 			 * If a temporary address has just become deprecated,
589 			 * regenerate a new one if possible.
590 			 */
591 			if (ip6_use_tempaddr &&
592 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
593 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
594 
595 				if (regen_tmpaddr(ia6) == 0) {
596 					/*
597 					 * A new temporary address is
598 					 * generated.
599 					 * XXX: this means the address chain
600 					 * has changed while we are still in
601 					 * the loop.  Although the change
602 					 * would not cause disaster (because
603 					 * it's not a deletion, but an
604 					 * addition,) we'd rather restart the
605 					 * loop just for safety.  Or does this
606 					 * significantly reduce performance??
607 					 */
608 					goto addrloop;
609 				}
610 			}
611 		} else {
612 			/*
613 			 * A new RA might have made a deprecated address
614 			 * preferred.
615 			 */
616 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
617 		}
618 	}
619 
620 	/* expire prefix list */
621 	pr = nd_prefix.lh_first;
622 	while (pr) {
623 		/*
624 		 * check prefix lifetime.
625 		 * since pltime is just for autoconf, pltime processing for
626 		 * prefix is not necessary.
627 		 */
628 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
629 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
630 			struct nd_prefix *t;
631 			t = pr->ndpr_next;
632 
633 			/*
634 			 * address expiration and prefix expiration are
635 			 * separate.  NEVER perform in6_purgeaddr here.
636 			 */
637 
638 			prelist_remove(pr);
639 			pr = t;
640 		} else
641 			pr = pr->ndpr_next;
642 	}
643 	splx(s);
644 }
645 
646 static int
647 regen_tmpaddr(ia6)
648 	struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */
649 {
650 	struct ifaddr *ifa;
651 	struct ifnet *ifp;
652 	struct in6_ifaddr *public_ifa6 = NULL;
653 
654 	ifp = ia6->ia_ifa.ifa_ifp;
655 	for (ifa = ifp->if_addrlist.tqh_first; ifa;
656 	     ifa = ifa->ifa_list.tqe_next) {
657 		struct in6_ifaddr *it6;
658 
659 		if (ifa->ifa_addr->sa_family != AF_INET6)
660 			continue;
661 
662 		it6 = (struct in6_ifaddr *)ifa;
663 
664 		/* ignore no autoconf addresses. */
665 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
666 			continue;
667 
668 		/* ignore autoconf addresses with different prefixes. */
669 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
670 			continue;
671 
672 		/*
673 		 * Now we are looking at an autoconf address with the same
674 		 * prefix as ours.  If the address is temporary and is still
675 		 * preferred, do not create another one.  It would be rare, but
676 		 * could happen, for example, when we resume a laptop PC after
677 		 * a long period.
678 		 */
679 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
680 		    !IFA6_IS_DEPRECATED(it6)) {
681 			public_ifa6 = NULL;
682 			break;
683 		}
684 
685 		/*
686 		 * This is a public autoconf address that has the same prefix
687 		 * as ours.  If it is preferred, keep it.  We can't break the
688 		 * loop here, because there may be a still-preferred temporary
689 		 * address with the prefix.
690 		 */
691 		if (!IFA6_IS_DEPRECATED(it6))
692 		    public_ifa6 = it6;
693 	}
694 
695 	if (public_ifa6 != NULL) {
696 		int e;
697 
698 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
699 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
700 			    " tmp addr,errno=%d\n", e);
701 			return (-1);
702 		}
703 		return (0);
704 	}
705 
706 	return (-1);
707 }
708 
709 /*
710  * Nuke neighbor cache/prefix/default router management table, right before
711  * ifp goes away.
712  */
713 void
714 nd6_purge(ifp)
715 	struct ifnet *ifp;
716 {
717 	struct llinfo_nd6 *ln, *nln;
718 	struct nd_defrouter *dr, *ndr;
719 	struct nd_prefix *pr, *npr;
720 
721 	/*
722 	 * Nuke default router list entries toward ifp.
723 	 * We defer removal of default router list entries that is installed
724 	 * in the routing table, in order to keep additional side effects as
725 	 * small as possible.
726 	 */
727 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
728 		ndr = TAILQ_NEXT(dr, dr_entry);
729 		if (dr->installed)
730 			continue;
731 
732 		if (dr->ifp == ifp)
733 			defrtrlist_del(dr);
734 	}
735 
736 	for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) {
737 		ndr = TAILQ_NEXT(dr, dr_entry);
738 		if (!dr->installed)
739 			continue;
740 
741 		if (dr->ifp == ifp)
742 			defrtrlist_del(dr);
743 	}
744 
745 	/* Nuke prefix list entries toward ifp */
746 	for (pr = nd_prefix.lh_first; pr; pr = npr) {
747 		npr = pr->ndpr_next;
748 		if (pr->ndpr_ifp == ifp) {
749 			/*
750 			 * Because if_detach() does *not* release prefixes
751 			 * while purging addresses the reference count will
752 			 * still be above zero. We therefore reset it to
753 			 * make sure that the prefix really gets purged.
754 			 */
755 			pr->ndpr_refcnt = 0;
756 
757 			/*
758 			 * Previously, pr->ndpr_addr is removed as well,
759 			 * but I strongly believe we don't have to do it.
760 			 * nd6_purge() is only called from in6_ifdetach(),
761 			 * which removes all the associated interface addresses
762 			 * by itself.
763 			 * (jinmei@kame.net 20010129)
764 			 */
765 			prelist_remove(pr);
766 		}
767 	}
768 
769 	/* cancel default outgoing interface setting */
770 	if (nd6_defifindex == ifp->if_index)
771 		nd6_setdefaultiface(0);
772 
773 	if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */
774 		/* refresh default router list */
775 		defrouter_select();
776 	}
777 
778 	/*
779 	 * Nuke neighbor cache entries for the ifp.
780 	 * Note that rt->rt_ifp may not be the same as ifp,
781 	 * due to KAME goto ours hack.  See RTM_RESOLVE case in
782 	 * nd6_rtrequest(), and ip6_input().
783 	 */
784 	ln = llinfo_nd6.ln_next;
785 	while (ln && ln != &llinfo_nd6) {
786 		struct rtentry *rt;
787 		struct sockaddr_dl *sdl;
788 
789 		nln = ln->ln_next;
790 		rt = ln->ln_rt;
791 		if (rt && rt->rt_gateway &&
792 		    rt->rt_gateway->sa_family == AF_LINK) {
793 			sdl = (struct sockaddr_dl *)rt->rt_gateway;
794 			if (sdl->sdl_index == ifp->if_index)
795 				nln = nd6_free(rt, 0);
796 		}
797 		ln = nln;
798 	}
799 }
800 
801 struct rtentry *
802 nd6_lookup(addr6, create, ifp)
803 	struct in6_addr *addr6;
804 	int create;
805 	struct ifnet *ifp;
806 {
807 	struct rtentry *rt;
808 	struct sockaddr_in6 sin6;
809 
810 	bzero(&sin6, sizeof(sin6));
811 	sin6.sin6_len = sizeof(struct sockaddr_in6);
812 	sin6.sin6_family = AF_INET6;
813 	sin6.sin6_addr = *addr6;
814 	rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL);
815 	if (rt) {
816 		if ((rt->rt_flags & RTF_LLINFO) == 0 && create) {
817 			/*
818 			 * This is the case for the default route.
819 			 * If we want to create a neighbor cache for the
820 			 * address, we should free the route for the
821 			 * destination and allocate an interface route.
822 			 */
823 			RTFREE_LOCKED(rt);
824 			rt = NULL;
825 		}
826 	}
827 	if (rt == NULL) {
828 		if (create && ifp) {
829 			int e;
830 
831 			/*
832 			 * If no route is available and create is set,
833 			 * we allocate a host route for the destination
834 			 * and treat it like an interface route.
835 			 * This hack is necessary for a neighbor which can't
836 			 * be covered by our own prefix.
837 			 */
838 			struct ifaddr *ifa =
839 			    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
840 			if (ifa == NULL)
841 				return (NULL);
842 
843 			/*
844 			 * Create a new route.  RTF_LLINFO is necessary
845 			 * to create a Neighbor Cache entry for the
846 			 * destination in nd6_rtrequest which will be
847 			 * called in rtrequest via ifa->ifa_rtrequest.
848 			 */
849 			if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6,
850 			    ifa->ifa_addr, (struct sockaddr *)&all1_sa,
851 			    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
852 			    ~RTF_CLONING, &rt)) != 0) {
853 				log(LOG_ERR,
854 				    "nd6_lookup: failed to add route for a "
855 				    "neighbor(%s), errno=%d\n",
856 				    ip6_sprintf(addr6), e);
857 			}
858 			if (rt == NULL)
859 				return (NULL);
860 			RT_LOCK(rt);
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_LOCK_ASSERT(rt);
870 	RT_REMREF(rt);
871 	/*
872 	 * Validation for the entry.
873 	 * Note that the check for rt_llinfo is necessary because a cloned
874 	 * route from a parent route that has the L flag (e.g. the default
875 	 * route to a p2p interface) may have the flag, too, while the
876 	 * destination is not actually a neighbor.
877 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
878 	 *      it might be the loopback interface if the entry is for our
879 	 *      own address on a non-loopback interface. Instead, we should
880 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
881 	 *	interface.
882 	 * Note also that ifa_ifp and ifp may differ when we connect two
883 	 * interfaces to a same link, install a link prefix to an interface,
884 	 * and try to install a neighbor cache on an interface that does not
885 	 * have a route to the prefix.
886 	 */
887 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
888 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
889 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
890 		if (create) {
891 			nd6log((LOG_DEBUG,
892 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
893 			    ip6_sprintf(addr6),
894 			    ifp ? if_name(ifp) : "unspec"));
895 		}
896 		RT_UNLOCK(rt);
897 		return (NULL);
898 	}
899 	RT_UNLOCK(rt);		/* XXX not ready to return rt locked */
900 	return (rt);
901 }
902 
903 /*
904  * Test whether a given IPv6 address is a neighbor or not, ignoring
905  * the actual neighbor cache.  The neighbor cache is ignored in order
906  * to not reenter the routing code from within itself.
907  */
908 static int
909 nd6_is_new_addr_neighbor(addr, ifp)
910 	struct sockaddr_in6 *addr;
911 	struct ifnet *ifp;
912 {
913 	struct nd_prefix *pr;
914 	struct ifaddr *dstaddr;
915 
916 	/*
917 	 * A link-local address is always a neighbor.
918 	 * XXX: a link does not necessarily specify a single interface.
919 	 */
920 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
921 		struct sockaddr_in6 sin6_copy;
922 		u_int32_t zone;
923 
924 		/*
925 		 * We need sin6_copy since sa6_recoverscope() may modify the
926 		 * content (XXX).
927 		 */
928 		sin6_copy = *addr;
929 		if (sa6_recoverscope(&sin6_copy))
930 			return (0); /* XXX: should be impossible */
931 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
932 			return (0);
933 		if (sin6_copy.sin6_scope_id == zone)
934 			return (1);
935 		else
936 			return (0);
937 	}
938 
939 	/*
940 	 * If the address matches one of our addresses,
941 	 * it should be a neighbor.
942 	 * If the address matches one of our on-link prefixes, it should be a
943 	 * neighbor.
944 	 */
945 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
946 		if (pr->ndpr_ifp != ifp)
947 			continue;
948 
949 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
950 			continue;
951 
952 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
953 		    &addr->sin6_addr, &pr->ndpr_mask))
954 			return (1);
955 	}
956 
957 	/*
958 	 * If the address is assigned on the node of the other side of
959 	 * a p2p interface, the address should be a neighbor.
960 	 */
961 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
962 	if ((dstaddr != NULL) && (dstaddr->ifa_ifp == ifp))
963 		return (1);
964 
965 	/*
966 	 * If the default router list is empty, all addresses are regarded
967 	 * as on-link, and thus, as a neighbor.
968 	 * XXX: we restrict the condition to hosts, because routers usually do
969 	 * not have the "default router list".
970 	 */
971 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
972 	    nd6_defifindex == ifp->if_index) {
973 		return (1);
974 	}
975 
976 	return (0);
977 }
978 
979 
980 /*
981  * Detect if a given IPv6 address identifies a neighbor on a given link.
982  * XXX: should take care of the destination of a p2p link?
983  */
984 int
985 nd6_is_addr_neighbor(addr, ifp)
986 	struct sockaddr_in6 *addr;
987 	struct ifnet *ifp;
988 {
989 
990 	if (nd6_is_new_addr_neighbor(addr, ifp))
991 		return (1);
992 
993 	/*
994 	 * Even if the address matches none of our addresses, it might be
995 	 * in the neighbor cache.
996 	 */
997 	if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL)
998 		return (1);
999 
1000 	return (0);
1001 }
1002 
1003 /*
1004  * Free an nd6 llinfo entry.
1005  * Since the function would cause significant changes in the kernel, DO NOT
1006  * make it global, unless you have a strong reason for the change, and are sure
1007  * that the change is safe.
1008  */
1009 static struct llinfo_nd6 *
1010 nd6_free(rt, gc)
1011 	struct rtentry *rt;
1012 	int gc;
1013 {
1014 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next;
1015 	struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr;
1016 	struct nd_defrouter *dr;
1017 
1018 	/*
1019 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1020 	 * even though it is not harmful, it was not really necessary.
1021 	 */
1022 
1023 	/* cancel timer */
1024 	nd6_llinfo_settimer(ln, -1);
1025 
1026 	if (!ip6_forwarding) {
1027 		int s;
1028 		s = splnet();
1029 		dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr,
1030 		    rt->rt_ifp);
1031 
1032 		if (dr != NULL && dr->expire &&
1033 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1034 			/*
1035 			 * If the reason for the deletion is just garbage
1036 			 * collection, and the neighbor is an active default
1037 			 * router, do not delete it.  Instead, reset the GC
1038 			 * timer using the router's lifetime.
1039 			 * Simply deleting the entry would affect default
1040 			 * router selection, which is not necessarily a good
1041 			 * thing, especially when we're using router preference
1042 			 * values.
1043 			 * XXX: the check for ln_state would be redundant,
1044 			 *      but we intentionally keep it just in case.
1045 			 */
1046 			if (dr->expire > time_second)
1047 				nd6_llinfo_settimer(ln,
1048 				    (dr->expire - time_second) * hz);
1049 			else
1050 				nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1051 			splx(s);
1052 			return (ln->ln_next);
1053 		}
1054 
1055 		if (ln->ln_router || dr) {
1056 			/*
1057 			 * rt6_flush must be called whether or not the neighbor
1058 			 * is in the Default Router List.
1059 			 * See a corresponding comment in nd6_na_input().
1060 			 */
1061 			rt6_flush(&in6, rt->rt_ifp);
1062 		}
1063 
1064 		if (dr) {
1065 			/*
1066 			 * Unreachablity of a router might affect the default
1067 			 * router selection and on-link detection of advertised
1068 			 * prefixes.
1069 			 */
1070 
1071 			/*
1072 			 * Temporarily fake the state to choose a new default
1073 			 * router and to perform on-link determination of
1074 			 * prefixes correctly.
1075 			 * Below the state will be set correctly,
1076 			 * or the entry itself will be deleted.
1077 			 */
1078 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1079 
1080 			/*
1081 			 * Since defrouter_select() does not affect the
1082 			 * on-link determination and MIP6 needs the check
1083 			 * before the default router selection, we perform
1084 			 * the check now.
1085 			 */
1086 			pfxlist_onlink_check();
1087 
1088 			/*
1089 			 * refresh default router list
1090 			 */
1091 			defrouter_select();
1092 		}
1093 		splx(s);
1094 	}
1095 
1096 	/*
1097 	 * Before deleting the entry, remember the next entry as the
1098 	 * return value.  We need this because pfxlist_onlink_check() above
1099 	 * might have freed other entries (particularly the old next entry) as
1100 	 * a side effect (XXX).
1101 	 */
1102 	next = ln->ln_next;
1103 
1104 	/*
1105 	 * Detach the route from the routing tree and the list of neighbor
1106 	 * caches, and disable the route entry not to be used in already
1107 	 * cached routes.
1108 	 */
1109 	rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0,
1110 	    rt_mask(rt), 0, (struct rtentry **)0);
1111 
1112 	return (next);
1113 }
1114 
1115 /*
1116  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1117  *
1118  * XXX cost-effective methods?
1119  */
1120 void
1121 nd6_nud_hint(rt, dst6, force)
1122 	struct rtentry *rt;
1123 	struct in6_addr *dst6;
1124 	int force;
1125 {
1126 	struct llinfo_nd6 *ln;
1127 
1128 	/*
1129 	 * If the caller specified "rt", use that.  Otherwise, resolve the
1130 	 * routing table by supplied "dst6".
1131 	 */
1132 	if (rt == NULL) {
1133 		if (dst6 == NULL)
1134 			return;
1135 		if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL)
1136 			return;
1137 	}
1138 
1139 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1140 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1141 	    rt->rt_llinfo == NULL || rt->rt_gateway == NULL ||
1142 	    rt->rt_gateway->sa_family != AF_LINK) {
1143 		/* This is not a host route. */
1144 		return;
1145 	}
1146 
1147 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1148 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1149 		return;
1150 
1151 	/*
1152 	 * if we get upper-layer reachability confirmation many times,
1153 	 * it is possible we have false information.
1154 	 */
1155 	if (!force) {
1156 		ln->ln_byhint++;
1157 		if (ln->ln_byhint > nd6_maxnudhint)
1158 			return;
1159 	}
1160 
1161 	ln->ln_state = ND6_LLINFO_REACHABLE;
1162 	if (!ND6_LLINFO_PERMANENT(ln)) {
1163 		nd6_llinfo_settimer(ln,
1164 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1165 	}
1166 }
1167 
1168 void
1169 nd6_rtrequest(req, rt, info)
1170 	int	req;
1171 	struct rtentry *rt;
1172 	struct rt_addrinfo *info; /* xxx unused */
1173 {
1174 	struct sockaddr *gate = rt->rt_gateway;
1175 	struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1176 	static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1177 	struct ifnet *ifp = rt->rt_ifp;
1178 	struct ifaddr *ifa;
1179 
1180 	RT_LOCK_ASSERT(rt);
1181 
1182 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1183 		return;
1184 
1185 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1186 		/*
1187 		 * This is probably an interface direct route for a link
1188 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1189 		 * We do not need special treatment below for such a route.
1190 		 * Moreover, the RTF_LLINFO flag which would be set below
1191 		 * would annoy the ndp(8) command.
1192 		 */
1193 		return;
1194 	}
1195 
1196 	if (req == RTM_RESOLVE &&
1197 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1198 	     !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt),
1199 	     ifp))) {
1200 		/*
1201 		 * FreeBSD and BSD/OS often make a cloned host route based
1202 		 * on a less-specific route (e.g. the default route).
1203 		 * If the less specific route does not have a "gateway"
1204 		 * (this is the case when the route just goes to a p2p or an
1205 		 * stf interface), we'll mistakenly make a neighbor cache for
1206 		 * the host route, and will see strange neighbor solicitation
1207 		 * for the corresponding destination.  In order to avoid the
1208 		 * confusion, we check if the destination of the route is
1209 		 * a neighbor in terms of neighbor discovery, and stop the
1210 		 * process if not.  Additionally, we remove the LLINFO flag
1211 		 * so that ndp(8) will not try to get the neighbor information
1212 		 * of the destination.
1213 		 */
1214 		rt->rt_flags &= ~RTF_LLINFO;
1215 		return;
1216 	}
1217 
1218 	switch (req) {
1219 	case RTM_ADD:
1220 		/*
1221 		 * There is no backward compatibility :)
1222 		 *
1223 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1224 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1225 		 *	   rt->rt_flags |= RTF_CLONING;
1226 		 */
1227 		if ((rt->rt_flags & RTF_CLONING) ||
1228 		    ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) {
1229 			/*
1230 			 * Case 1: This route should come from a route to
1231 			 * interface (RTF_CLONING case) or the route should be
1232 			 * treated as on-link but is currently not
1233 			 * (RTF_LLINFO && ln == NULL case).
1234 			 */
1235 			rt_setgate(rt, rt_key(rt),
1236 				   (struct sockaddr *)&null_sdl);
1237 			gate = rt->rt_gateway;
1238 			SDL(gate)->sdl_type = ifp->if_type;
1239 			SDL(gate)->sdl_index = ifp->if_index;
1240 			if (ln)
1241 				nd6_llinfo_settimer(ln, 0);
1242 			if ((rt->rt_flags & RTF_CLONING) != 0)
1243 				break;
1244 		}
1245 		/*
1246 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1247 		 * We don't do that here since llinfo is not ready yet.
1248 		 *
1249 		 * There are also couple of other things to be discussed:
1250 		 * - unsolicited NA code needs improvement beforehand
1251 		 * - RFC2461 says we MAY send multicast unsolicited NA
1252 		 *   (7.2.6 paragraph 4), however, it also says that we
1253 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1254 		 *   we don't have anything like it right now.
1255 		 *   note that the mechanism needs a mutual agreement
1256 		 *   between proxies, which means that we need to implement
1257 		 *   a new protocol, or a new kludge.
1258 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1259 		 *   we need to check ip6forwarding before sending it.
1260 		 *   (or should we allow proxy ND configuration only for
1261 		 *   routers?  there's no mention about proxy ND from hosts)
1262 		 */
1263 		/* FALLTHROUGH */
1264 	case RTM_RESOLVE:
1265 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1266 			/*
1267 			 * Address resolution isn't necessary for a point to
1268 			 * point link, so we can skip this test for a p2p link.
1269 			 */
1270 			if (gate->sa_family != AF_LINK ||
1271 			    gate->sa_len < sizeof(null_sdl)) {
1272 				log(LOG_DEBUG,
1273 				    "nd6_rtrequest: bad gateway value: %s\n",
1274 				    if_name(ifp));
1275 				break;
1276 			}
1277 			SDL(gate)->sdl_type = ifp->if_type;
1278 			SDL(gate)->sdl_index = ifp->if_index;
1279 		}
1280 		if (ln != NULL)
1281 			break;	/* This happens on a route change */
1282 		/*
1283 		 * Case 2: This route may come from cloning, or a manual route
1284 		 * add with a LL address.
1285 		 */
1286 		R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln));
1287 		rt->rt_llinfo = (caddr_t)ln;
1288 		if (ln == NULL) {
1289 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1290 			break;
1291 		}
1292 		nd6_inuse++;
1293 		nd6_allocated++;
1294 		bzero(ln, sizeof(*ln));
1295 		ln->ln_rt = rt;
1296 		callout_init(&ln->ln_timer_ch, 0);
1297 
1298 		/* this is required for "ndp" command. - shin */
1299 		if (req == RTM_ADD) {
1300 		        /*
1301 			 * gate should have some valid AF_LINK entry,
1302 			 * and ln->ln_expire should have some lifetime
1303 			 * which is specified by ndp command.
1304 			 */
1305 			ln->ln_state = ND6_LLINFO_REACHABLE;
1306 			ln->ln_byhint = 0;
1307 		} else {
1308 		        /*
1309 			 * When req == RTM_RESOLVE, rt is created and
1310 			 * initialized in rtrequest(), so rt_expire is 0.
1311 			 */
1312 			ln->ln_state = ND6_LLINFO_NOSTATE;
1313 			nd6_llinfo_settimer(ln, 0);
1314 		}
1315 		rt->rt_flags |= RTF_LLINFO;
1316 		ln->ln_next = llinfo_nd6.ln_next;
1317 		llinfo_nd6.ln_next = ln;
1318 		ln->ln_prev = &llinfo_nd6;
1319 		ln->ln_next->ln_prev = ln;
1320 
1321 		/*
1322 		 * check if rt_key(rt) is one of my address assigned
1323 		 * to the interface.
1324 		 */
1325 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp,
1326 		    &SIN6(rt_key(rt))->sin6_addr);
1327 		if (ifa) {
1328 			caddr_t macp = nd6_ifptomac(ifp);
1329 			nd6_llinfo_settimer(ln, -1);
1330 			ln->ln_state = ND6_LLINFO_REACHABLE;
1331 			ln->ln_byhint = 0;
1332 			if (macp) {
1333 				bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen);
1334 				SDL(gate)->sdl_alen = ifp->if_addrlen;
1335 			}
1336 			if (nd6_useloopback) {
1337 				rt->rt_ifp = &loif[0];	/* XXX */
1338 				/*
1339 				 * Make sure rt_ifa be equal to the ifaddr
1340 				 * corresponding to the address.
1341 				 * We need this because when we refer
1342 				 * rt_ifa->ia6_flags in ip6_input, we assume
1343 				 * that the rt_ifa points to the address instead
1344 				 * of the loopback address.
1345 				 */
1346 				if (ifa != rt->rt_ifa) {
1347 					IFAFREE(rt->rt_ifa);
1348 					IFAREF(ifa);
1349 					rt->rt_ifa = ifa;
1350 				}
1351 			}
1352 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1353 			nd6_llinfo_settimer(ln, -1);
1354 			ln->ln_state = ND6_LLINFO_REACHABLE;
1355 			ln->ln_byhint = 0;
1356 
1357 			/* join solicited node multicast for proxy ND */
1358 			if (ifp->if_flags & IFF_MULTICAST) {
1359 				struct in6_addr llsol;
1360 				int error;
1361 
1362 				llsol = SIN6(rt_key(rt))->sin6_addr;
1363 				llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
1364 				llsol.s6_addr32[1] = 0;
1365 				llsol.s6_addr32[2] = htonl(1);
1366 				llsol.s6_addr8[12] = 0xff;
1367 				if (in6_setscope(&llsol, ifp, NULL))
1368 					break;
1369 				if (in6_addmulti(&llsol, ifp,
1370 				    &error, 0) == NULL) {
1371 					nd6log((LOG_ERR, "%s: failed to join "
1372 					    "%s (errno=%d)\n", if_name(ifp),
1373 					    ip6_sprintf(&llsol), error));
1374 				}
1375 			}
1376 		}
1377 		break;
1378 
1379 	case RTM_DELETE:
1380 		if (ln == NULL)
1381 			break;
1382 		/* leave from solicited node multicast for proxy ND */
1383 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1384 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1385 			struct in6_addr llsol;
1386 			struct in6_multi *in6m;
1387 
1388 			llsol = SIN6(rt_key(rt))->sin6_addr;
1389 			llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL;
1390 			llsol.s6_addr32[1] = 0;
1391 			llsol.s6_addr32[2] = htonl(1);
1392 			llsol.s6_addr8[12] = 0xff;
1393 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1394 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1395 				if (in6m)
1396 					in6_delmulti(in6m);
1397 			} else
1398 				; /* XXX: should not happen. bark here? */
1399 		}
1400 		nd6_inuse--;
1401 		ln->ln_next->ln_prev = ln->ln_prev;
1402 		ln->ln_prev->ln_next = ln->ln_next;
1403 		ln->ln_prev = NULL;
1404 		nd6_llinfo_settimer(ln, -1);
1405 		rt->rt_llinfo = 0;
1406 		rt->rt_flags &= ~RTF_LLINFO;
1407 		if (ln->ln_hold)
1408 			m_freem(ln->ln_hold);
1409 		Free((caddr_t)ln);
1410 	}
1411 }
1412 
1413 int
1414 nd6_ioctl(cmd, data, ifp)
1415 	u_long cmd;
1416 	caddr_t	data;
1417 	struct ifnet *ifp;
1418 {
1419 	struct in6_drlist *drl = (struct in6_drlist *)data;
1420 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1421 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1422 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1423 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1424 	struct nd_defrouter *dr;
1425 	struct nd_prefix *pr;
1426 	struct rtentry *rt;
1427 	int i = 0, error = 0;
1428 	int s;
1429 
1430 	switch (cmd) {
1431 	case SIOCGDRLST_IN6:
1432 		/*
1433 		 * obsolete API, use sysctl under net.inet6.icmp6
1434 		 */
1435 		bzero(drl, sizeof(*drl));
1436 		s = splnet();
1437 		dr = TAILQ_FIRST(&nd_defrouter);
1438 		while (dr && i < DRLSTSIZ) {
1439 			drl->defrouter[i].rtaddr = dr->rtaddr;
1440 			in6_clearscope(&drl->defrouter[i].rtaddr);
1441 
1442 			drl->defrouter[i].flags = dr->flags;
1443 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1444 			drl->defrouter[i].expire = dr->expire;
1445 			drl->defrouter[i].if_index = dr->ifp->if_index;
1446 			i++;
1447 			dr = TAILQ_NEXT(dr, dr_entry);
1448 		}
1449 		splx(s);
1450 		break;
1451 	case SIOCGPRLST_IN6:
1452 		/*
1453 		 * obsolete API, use sysctl under net.inet6.icmp6
1454 		 *
1455 		 * XXX the structure in6_prlist was changed in backward-
1456 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1457 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1458 		 */
1459 		/*
1460 		 * XXX meaning of fields, especialy "raflags", is very
1461 		 * differnet between RA prefix list and RR/static prefix list.
1462 		 * how about separating ioctls into two?
1463 		 */
1464 		bzero(oprl, sizeof(*oprl));
1465 		s = splnet();
1466 		pr = nd_prefix.lh_first;
1467 		while (pr && i < PRLSTSIZ) {
1468 			struct nd_pfxrouter *pfr;
1469 			int j;
1470 
1471 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1472 			oprl->prefix[i].raflags = pr->ndpr_raf;
1473 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1474 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1475 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1476 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1477 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1478 				oprl->prefix[i].expire = 0;
1479 			else {
1480 				time_t maxexpire;
1481 
1482 				/* XXX: we assume time_t is signed. */
1483 				maxexpire = (-1) &
1484 				    ~((time_t)1 <<
1485 				    ((sizeof(maxexpire) * 8) - 1));
1486 				if (pr->ndpr_vltime <
1487 				    maxexpire - pr->ndpr_lastupdate) {
1488 					oprl->prefix[i].expire =
1489 					    pr->ndpr_lastupdate +
1490 					    pr->ndpr_vltime;
1491 				} else
1492 					oprl->prefix[i].expire = maxexpire;
1493 			}
1494 
1495 			pfr = pr->ndpr_advrtrs.lh_first;
1496 			j = 0;
1497 			while (pfr) {
1498 				if (j < DRLSTSIZ) {
1499 #define RTRADDR oprl->prefix[i].advrtr[j]
1500 					RTRADDR = pfr->router->rtaddr;
1501 					in6_clearscope(&RTRADDR);
1502 #undef RTRADDR
1503 				}
1504 				j++;
1505 				pfr = pfr->pfr_next;
1506 			}
1507 			oprl->prefix[i].advrtrs = j;
1508 			oprl->prefix[i].origin = PR_ORIG_RA;
1509 
1510 			i++;
1511 			pr = pr->ndpr_next;
1512 		}
1513 		splx(s);
1514 
1515 		break;
1516 	case OSIOCGIFINFO_IN6:
1517 #define ND	ndi->ndi
1518 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1519 		bzero(&ND, sizeof(ND));
1520 		ND.linkmtu = IN6_LINKMTU(ifp);
1521 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1522 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1523 		ND.reachable = ND_IFINFO(ifp)->reachable;
1524 		ND.retrans = ND_IFINFO(ifp)->retrans;
1525 		ND.flags = ND_IFINFO(ifp)->flags;
1526 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1527 		ND.chlim = ND_IFINFO(ifp)->chlim;
1528 		break;
1529 	case SIOCGIFINFO_IN6:
1530 		ND = *ND_IFINFO(ifp);
1531 		break;
1532 	case SIOCSIFINFO_IN6:
1533 		/*
1534 		 * used to change host variables from userland.
1535 		 * intented for a use on router to reflect RA configurations.
1536 		 */
1537 		/* 0 means 'unspecified' */
1538 		if (ND.linkmtu != 0) {
1539 			if (ND.linkmtu < IPV6_MMTU ||
1540 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1541 				error = EINVAL;
1542 				break;
1543 			}
1544 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1545 		}
1546 
1547 		if (ND.basereachable != 0) {
1548 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1549 
1550 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1551 			if (ND.basereachable != obasereachable)
1552 				ND_IFINFO(ifp)->reachable =
1553 				    ND_COMPUTE_RTIME(ND.basereachable);
1554 		}
1555 		if (ND.retrans != 0)
1556 			ND_IFINFO(ifp)->retrans = ND.retrans;
1557 		if (ND.chlim != 0)
1558 			ND_IFINFO(ifp)->chlim = ND.chlim;
1559 		/* FALLTHROUGH */
1560 	case SIOCSIFINFO_FLAGS:
1561 		ND_IFINFO(ifp)->flags = ND.flags;
1562 		break;
1563 #undef ND
1564 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1565 		/* sync kernel routing table with the default router list */
1566 		defrouter_reset();
1567 		defrouter_select();
1568 		break;
1569 	case SIOCSPFXFLUSH_IN6:
1570 	{
1571 		/* flush all the prefix advertised by routers */
1572 		struct nd_prefix *pr, *next;
1573 
1574 		s = splnet();
1575 		for (pr = nd_prefix.lh_first; pr; pr = next) {
1576 			struct in6_ifaddr *ia, *ia_next;
1577 
1578 			next = pr->ndpr_next;
1579 
1580 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1581 				continue; /* XXX */
1582 
1583 			/* do we really have to remove addresses as well? */
1584 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1585 				/* ia might be removed.  keep the next ptr. */
1586 				ia_next = ia->ia_next;
1587 
1588 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1589 					continue;
1590 
1591 				if (ia->ia6_ndpr == pr)
1592 					in6_purgeaddr(&ia->ia_ifa);
1593 			}
1594 			prelist_remove(pr);
1595 		}
1596 		splx(s);
1597 		break;
1598 	}
1599 	case SIOCSRTRFLUSH_IN6:
1600 	{
1601 		/* flush all the default routers */
1602 		struct nd_defrouter *dr, *next;
1603 
1604 		s = splnet();
1605 		defrouter_reset();
1606 		for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) {
1607 			next = TAILQ_NEXT(dr, dr_entry);
1608 			defrtrlist_del(dr);
1609 		}
1610 		defrouter_select();
1611 		splx(s);
1612 		break;
1613 	}
1614 	case SIOCGNBRINFO_IN6:
1615 	{
1616 		struct llinfo_nd6 *ln;
1617 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1618 
1619 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1620 			return (error);
1621 
1622 		s = splnet();
1623 		if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) {
1624 			error = EINVAL;
1625 			splx(s);
1626 			break;
1627 		}
1628 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1629 		nbi->state = ln->ln_state;
1630 		nbi->asked = ln->ln_asked;
1631 		nbi->isrouter = ln->ln_router;
1632 		nbi->expire = ln->ln_expire;
1633 		splx(s);
1634 
1635 		break;
1636 	}
1637 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1638 		ndif->ifindex = nd6_defifindex;
1639 		break;
1640 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1641 		return (nd6_setdefaultiface(ndif->ifindex));
1642 	}
1643 	return (error);
1644 }
1645 
1646 /*
1647  * Create neighbor cache entry and cache link-layer address,
1648  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1649  */
1650 struct rtentry *
1651 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code)
1652 	struct ifnet *ifp;
1653 	struct in6_addr *from;
1654 	char *lladdr;
1655 	int lladdrlen;
1656 	int type;	/* ICMP6 type */
1657 	int code;	/* type dependent information */
1658 {
1659 	struct rtentry *rt = NULL;
1660 	struct llinfo_nd6 *ln = NULL;
1661 	int is_newentry;
1662 	struct sockaddr_dl *sdl = NULL;
1663 	int do_update;
1664 	int olladdr;
1665 	int llchange;
1666 	int newstate = 0;
1667 
1668 	if (ifp == NULL)
1669 		panic("ifp == NULL in nd6_cache_lladdr");
1670 	if (from == NULL)
1671 		panic("from == NULL in nd6_cache_lladdr");
1672 
1673 	/* nothing must be updated for unspecified address */
1674 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1675 		return NULL;
1676 
1677 	/*
1678 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1679 	 * the caller.
1680 	 *
1681 	 * XXX If the link does not have link-layer adderss, what should
1682 	 * we do? (ifp->if_addrlen == 0)
1683 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1684 	 * description on it in NS section (RFC 2461 7.2.3).
1685 	 */
1686 
1687 	rt = nd6_lookup(from, 0, ifp);
1688 	if (rt == NULL) {
1689 		rt = nd6_lookup(from, 1, ifp);
1690 		is_newentry = 1;
1691 	} else {
1692 		/* do nothing if static ndp is set */
1693 		if (rt->rt_flags & RTF_STATIC)
1694 			return NULL;
1695 		is_newentry = 0;
1696 	}
1697 
1698 	if (rt == NULL)
1699 		return NULL;
1700 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
1701 fail:
1702 		(void)nd6_free(rt, 0);
1703 		return NULL;
1704 	}
1705 	ln = (struct llinfo_nd6 *)rt->rt_llinfo;
1706 	if (ln == NULL)
1707 		goto fail;
1708 	if (rt->rt_gateway == NULL)
1709 		goto fail;
1710 	if (rt->rt_gateway->sa_family != AF_LINK)
1711 		goto fail;
1712 	sdl = SDL(rt->rt_gateway);
1713 
1714 	olladdr = (sdl->sdl_alen) ? 1 : 0;
1715 	if (olladdr && lladdr) {
1716 		if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen))
1717 			llchange = 1;
1718 		else
1719 			llchange = 0;
1720 	} else
1721 		llchange = 0;
1722 
1723 	/*
1724 	 * newentry olladdr  lladdr  llchange	(*=record)
1725 	 *	0	n	n	--	(1)
1726 	 *	0	y	n	--	(2)
1727 	 *	0	n	y	--	(3) * STALE
1728 	 *	0	y	y	n	(4) *
1729 	 *	0	y	y	y	(5) * STALE
1730 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1731 	 *	1	--	y	--	(7) * STALE
1732 	 */
1733 
1734 	if (lladdr) {		/* (3-5) and (7) */
1735 		/*
1736 		 * Record source link-layer address
1737 		 * XXX is it dependent to ifp->if_type?
1738 		 */
1739 		sdl->sdl_alen = ifp->if_addrlen;
1740 		bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen);
1741 	}
1742 
1743 	if (!is_newentry) {
1744 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1745 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1746 			do_update = 1;
1747 			newstate = ND6_LLINFO_STALE;
1748 		} else					/* (1-2,4) */
1749 			do_update = 0;
1750 	} else {
1751 		do_update = 1;
1752 		if (lladdr == NULL)			/* (6) */
1753 			newstate = ND6_LLINFO_NOSTATE;
1754 		else					/* (7) */
1755 			newstate = ND6_LLINFO_STALE;
1756 	}
1757 
1758 	if (do_update) {
1759 		/*
1760 		 * Update the state of the neighbor cache.
1761 		 */
1762 		ln->ln_state = newstate;
1763 
1764 		if (ln->ln_state == ND6_LLINFO_STALE) {
1765 			/*
1766 			 * XXX: since nd6_output() below will cause
1767 			 * state tansition to DELAY and reset the timer,
1768 			 * we must set the timer now, although it is actually
1769 			 * meaningless.
1770 			 */
1771 			nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
1772 
1773 			if (ln->ln_hold) {
1774 				struct mbuf *m_hold, *m_hold_next;
1775 				for (m_hold = ln->ln_hold; m_hold;
1776 				     m_hold = m_hold_next) {
1777 					struct mbuf *mpkt = NULL;
1778 
1779 					m_hold_next = m_hold->m_nextpkt;
1780 					mpkt = m_copym(m_hold, 0, M_COPYALL, M_DONTWAIT);
1781 					if (mpkt == NULL) {
1782 						m_freem(m_hold);
1783 						break;
1784 					}
1785 					mpkt->m_nextpkt = NULL;
1786 
1787 					/*
1788 					 * we assume ifp is not a p2p here, so
1789 					 * just set the 2nd argument as the
1790 					 * 1st one.
1791 					 */
1792 					nd6_output(ifp, ifp, mpkt,
1793 					     (struct sockaddr_in6 *)rt_key(rt),
1794 					     rt);
1795 				}
1796 				ln->ln_hold = NULL;
1797 			}
1798 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1799 			/* probe right away */
1800 			nd6_llinfo_settimer((void *)ln, 0);
1801 		}
1802 	}
1803 
1804 	/*
1805 	 * ICMP6 type dependent behavior.
1806 	 *
1807 	 * NS: clear IsRouter if new entry
1808 	 * RS: clear IsRouter
1809 	 * RA: set IsRouter if there's lladdr
1810 	 * redir: clear IsRouter if new entry
1811 	 *
1812 	 * RA case, (1):
1813 	 * The spec says that we must set IsRouter in the following cases:
1814 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1815 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1816 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1817 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1818 	 * neighbor cache, this is similar to (6).
1819 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1820 	 *
1821 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1822 	 *							D R
1823 	 *	0	n	n	--	(1)	c   ?     s
1824 	 *	0	y	n	--	(2)	c   s     s
1825 	 *	0	n	y	--	(3)	c   s     s
1826 	 *	0	y	y	n	(4)	c   s     s
1827 	 *	0	y	y	y	(5)	c   s     s
1828 	 *	1	--	n	--	(6) c	c 	c s
1829 	 *	1	--	y	--	(7) c	c   s	c s
1830 	 *
1831 	 *					(c=clear s=set)
1832 	 */
1833 	switch (type & 0xff) {
1834 	case ND_NEIGHBOR_SOLICIT:
1835 		/*
1836 		 * New entry must have is_router flag cleared.
1837 		 */
1838 		if (is_newentry)	/* (6-7) */
1839 			ln->ln_router = 0;
1840 		break;
1841 	case ND_REDIRECT:
1842 		/*
1843 		 * If the icmp is a redirect to a better router, always set the
1844 		 * is_router flag.  Otherwise, if the entry is newly created,
1845 		 * clear the flag.  [RFC 2461, sec 8.3]
1846 		 */
1847 		if (code == ND_REDIRECT_ROUTER)
1848 			ln->ln_router = 1;
1849 		else if (is_newentry) /* (6-7) */
1850 			ln->ln_router = 0;
1851 		break;
1852 	case ND_ROUTER_SOLICIT:
1853 		/*
1854 		 * is_router flag must always be cleared.
1855 		 */
1856 		ln->ln_router = 0;
1857 		break;
1858 	case ND_ROUTER_ADVERT:
1859 		/*
1860 		 * Mark an entry with lladdr as a router.
1861 		 */
1862 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1863 		    (is_newentry && lladdr)) {			/* (7) */
1864 			ln->ln_router = 1;
1865 		}
1866 		break;
1867 	}
1868 
1869 	/*
1870 	 * When the link-layer address of a router changes, select the
1871 	 * best router again.  In particular, when the neighbor entry is newly
1872 	 * created, it might affect the selection policy.
1873 	 * Question: can we restrict the first condition to the "is_newentry"
1874 	 * case?
1875 	 * XXX: when we hear an RA from a new router with the link-layer
1876 	 * address option, defrouter_select() is called twice, since
1877 	 * defrtrlist_update called the function as well.  However, I believe
1878 	 * we can compromise the overhead, since it only happens the first
1879 	 * time.
1880 	 * XXX: although defrouter_select() should not have a bad effect
1881 	 * for those are not autoconfigured hosts, we explicitly avoid such
1882 	 * cases for safety.
1883 	 */
1884 	if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv)
1885 		defrouter_select();
1886 
1887 	return rt;
1888 }
1889 
1890 static void
1891 nd6_slowtimo(ignored_arg)
1892     void *ignored_arg;
1893 {
1894 	struct nd_ifinfo *nd6if;
1895 	struct ifnet *ifp;
1896 
1897 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1898 	    nd6_slowtimo, NULL);
1899 	IFNET_RLOCK();
1900 	for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) {
1901 		nd6if = ND_IFINFO(ifp);
1902 		if (nd6if->basereachable && /* already initialized */
1903 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1904 			/*
1905 			 * Since reachable time rarely changes by router
1906 			 * advertisements, we SHOULD insure that a new random
1907 			 * value gets recomputed at least once every few hours.
1908 			 * (RFC 2461, 6.3.4)
1909 			 */
1910 			nd6if->recalctm = nd6_recalc_reachtm_interval;
1911 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1912 		}
1913 	}
1914 	IFNET_RUNLOCK();
1915 }
1916 
1917 #define senderr(e) { error = (e); goto bad;}
1918 int
1919 nd6_output(ifp, origifp, m0, dst, rt0)
1920 	struct ifnet *ifp;
1921 	struct ifnet *origifp;
1922 	struct mbuf *m0;
1923 	struct sockaddr_in6 *dst;
1924 	struct rtentry *rt0;
1925 {
1926 	struct mbuf *m = m0;
1927 	struct rtentry *rt = rt0;
1928 	struct sockaddr_in6 *gw6 = NULL;
1929 	struct llinfo_nd6 *ln = NULL;
1930 	int error = 0;
1931 
1932 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1933 		goto sendpkt;
1934 
1935 	if (nd6_need_cache(ifp) == 0)
1936 		goto sendpkt;
1937 
1938 	/*
1939 	 * next hop determination.  This routine is derived from ether_output.
1940 	 */
1941 again:
1942 	if (rt) {
1943 		if ((rt->rt_flags & RTF_UP) == 0) {
1944 			rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL);
1945 			if (rt != NULL) {
1946 				RT_REMREF(rt);
1947 				RT_UNLOCK(rt);
1948 				if (rt->rt_ifp != ifp)
1949 					/*
1950 					 * XXX maybe we should update ifp too,
1951 					 * but the original code didn't and I
1952 					 * don't know what is correct here.
1953 					 */
1954 					goto again;
1955 			} else
1956 				senderr(EHOSTUNREACH);
1957 		}
1958 
1959 		if (rt->rt_flags & RTF_GATEWAY) {
1960 			gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
1961 
1962 			/*
1963 			 * We skip link-layer address resolution and NUD
1964 			 * if the gateway is not a neighbor from ND point
1965 			 * of view, regardless of the value of nd_ifinfo.flags.
1966 			 * The second condition is a bit tricky; we skip
1967 			 * if the gateway is our own address, which is
1968 			 * sometimes used to install a route to a p2p link.
1969 			 */
1970 			if (!nd6_is_addr_neighbor(gw6, ifp) ||
1971 			    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
1972 				/*
1973 				 * We allow this kind of tricky route only
1974 				 * when the outgoing interface is p2p.
1975 				 * XXX: we may need a more generic rule here.
1976 				 */
1977 				if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1978 					senderr(EHOSTUNREACH);
1979 
1980 				goto sendpkt;
1981 			}
1982 
1983 			if (rt->rt_gwroute == 0)
1984 				goto lookup;
1985 			if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) {
1986 				RT_LOCK(rt);
1987 				rtfree(rt); rt = rt0;
1988 			lookup:
1989 				rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL);
1990 				if ((rt = rt->rt_gwroute) == 0)
1991 					senderr(EHOSTUNREACH);
1992 				RT_UNLOCK(rt);
1993 			}
1994 		}
1995 	}
1996 
1997 	/*
1998 	 * Address resolution or Neighbor Unreachability Detection
1999 	 * for the next hop.
2000 	 * At this point, the destination of the packet must be a unicast
2001 	 * or an anycast address(i.e. not a multicast).
2002 	 */
2003 
2004 	/* Look up the neighbor cache for the nexthop */
2005 	if (rt && (rt->rt_flags & RTF_LLINFO) != 0)
2006 		ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2007 	else {
2008 		/*
2009 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2010 		 * the condition below is not very efficient.  But we believe
2011 		 * it is tolerable, because this should be a rare case.
2012 		 */
2013 		if (nd6_is_addr_neighbor(dst, ifp) &&
2014 		    (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL)
2015 			ln = (struct llinfo_nd6 *)rt->rt_llinfo;
2016 	}
2017 	if (ln == NULL || rt == NULL) {
2018 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2019 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2020 			log(LOG_DEBUG,
2021 			    "nd6_output: can't allocate llinfo for %s "
2022 			    "(ln=%p, rt=%p)\n",
2023 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2024 			senderr(EIO);	/* XXX: good error? */
2025 		}
2026 
2027 		goto sendpkt;	/* send anyway */
2028 	}
2029 
2030 	/* We don't have to do link-layer address resolution on a p2p link. */
2031 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2032 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2033 		ln->ln_state = ND6_LLINFO_STALE;
2034 		nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz);
2035 	}
2036 
2037 	/*
2038 	 * The first time we send a packet to a neighbor whose entry is
2039 	 * STALE, we have to change the state to DELAY and a sets a timer to
2040 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2041 	 * neighbor unreachability detection on expiration.
2042 	 * (RFC 2461 7.3.3)
2043 	 */
2044 	if (ln->ln_state == ND6_LLINFO_STALE) {
2045 		ln->ln_asked = 0;
2046 		ln->ln_state = ND6_LLINFO_DELAY;
2047 		nd6_llinfo_settimer(ln, (long)nd6_delay * hz);
2048 	}
2049 
2050 	/*
2051 	 * If the neighbor cache entry has a state other than INCOMPLETE
2052 	 * (i.e. its link-layer address is already resolved), just
2053 	 * send the packet.
2054 	 */
2055 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2056 		goto sendpkt;
2057 
2058 	/*
2059 	 * There is a neighbor cache entry, but no ethernet address
2060 	 * response yet.  Append this latest packet to the end of the
2061 	 * packet queue in the mbuf, unless the number of the packet
2062 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2063 	 * the oldest packet in the queue will be removed.
2064 	 */
2065 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2066 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2067 	if (ln->ln_hold) {
2068 		struct mbuf *m_hold;
2069 		int i;
2070 
2071 		i = 0;
2072 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2073 			i++;
2074 			if (m_hold->m_nextpkt == NULL) {
2075 				m_hold->m_nextpkt = m;
2076 				break;
2077 			}
2078 		}
2079 		while (i >= nd6_maxqueuelen) {
2080 			m_hold = ln->ln_hold;
2081 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2082 			m_free(m_hold);
2083 			i--;
2084 		}
2085 	} else {
2086 		ln->ln_hold = m;
2087 	}
2088 
2089 	/*
2090 	 * If there has been no NS for the neighbor after entering the
2091 	 * INCOMPLETE state, send the first solicitation.
2092 	 */
2093 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2094 		ln->ln_asked++;
2095 		nd6_llinfo_settimer(ln,
2096 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2097 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2098 	}
2099 	return (0);
2100 
2101   sendpkt:
2102 	/* discard the packet if IPv6 operation is disabled on the interface */
2103 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2104 		error = ENETDOWN; /* better error? */
2105 		goto bad;
2106 	}
2107 
2108 #ifdef IPSEC
2109 	/* clean ipsec history once it goes out of the node */
2110 	ipsec_delaux(m);
2111 #endif
2112 
2113 #ifdef MAC
2114 	mac_create_mbuf_linklayer(ifp, m);
2115 #endif
2116 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2117 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2118 		    rt));
2119 	}
2120 	return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt));
2121 
2122   bad:
2123 	if (m)
2124 		m_freem(m);
2125 	return (error);
2126 }
2127 #undef senderr
2128 
2129 int
2130 nd6_need_cache(ifp)
2131 	struct ifnet *ifp;
2132 {
2133 	/*
2134 	 * XXX: we currently do not make neighbor cache on any interface
2135 	 * other than ARCnet, Ethernet, FDDI and GIF.
2136 	 *
2137 	 * RFC2893 says:
2138 	 * - unidirectional tunnels needs no ND
2139 	 */
2140 	switch (ifp->if_type) {
2141 	case IFT_ARCNET:
2142 	case IFT_ETHER:
2143 	case IFT_FDDI:
2144 	case IFT_IEEE1394:
2145 #ifdef IFT_L2VLAN
2146 	case IFT_L2VLAN:
2147 #endif
2148 #ifdef IFT_IEEE80211
2149 	case IFT_IEEE80211:
2150 #endif
2151 #ifdef IFT_CARP
2152 	case IFT_CARP:
2153 #endif
2154 	case IFT_GIF:		/* XXX need more cases? */
2155 	case IFT_PPP:
2156 	case IFT_TUNNEL:
2157 	case IFT_BRIDGE:
2158 		return (1);
2159 	default:
2160 		return (0);
2161 	}
2162 }
2163 
2164 int
2165 nd6_storelladdr(ifp, rt0, m, dst, desten)
2166 	struct ifnet *ifp;
2167 	struct rtentry *rt0;
2168 	struct mbuf *m;
2169 	struct sockaddr *dst;
2170 	u_char *desten;
2171 {
2172 	struct sockaddr_dl *sdl;
2173 	struct rtentry *rt;
2174 	int error;
2175 
2176 	if (m->m_flags & M_MCAST) {
2177 		int i;
2178 
2179 		switch (ifp->if_type) {
2180 		case IFT_ETHER:
2181 		case IFT_FDDI:
2182 #ifdef IFT_L2VLAN
2183 		case IFT_L2VLAN:
2184 #endif
2185 #ifdef IFT_IEEE80211
2186 		case IFT_IEEE80211:
2187 #endif
2188 		case IFT_BRIDGE:
2189 		case IFT_ISO88025:
2190 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2191 						 desten);
2192 			return (0);
2193 		case IFT_IEEE1394:
2194 			/*
2195 			 * netbsd can use if_broadcastaddr, but we don't do so
2196 			 * to reduce # of ifdef.
2197 			 */
2198 			for (i = 0; i < ifp->if_addrlen; i++)
2199 				desten[i] = ~0;
2200 			return (0);
2201 		case IFT_ARCNET:
2202 			*desten = 0;
2203 			return (0);
2204 		default:
2205 			m_freem(m);
2206 			return (EAFNOSUPPORT);
2207 		}
2208 	}
2209 
2210 	if (rt0 == NULL) {
2211 		/* this could happen, if we could not allocate memory */
2212 		m_freem(m);
2213 		return (ENOMEM);
2214 	}
2215 
2216 	error = rt_check(&rt, &rt0, dst);
2217 	if (error) {
2218 		m_freem(m);
2219 		return (error);
2220 	}
2221 	RT_UNLOCK(rt);
2222 
2223 	if (rt->rt_gateway->sa_family != AF_LINK) {
2224 		printf("nd6_storelladdr: something odd happens\n");
2225 		m_freem(m);
2226 		return (EINVAL);
2227 	}
2228 	sdl = SDL(rt->rt_gateway);
2229 	if (sdl->sdl_alen == 0) {
2230 		/* this should be impossible, but we bark here for debugging */
2231 		printf("nd6_storelladdr: sdl_alen == 0\n");
2232 		m_freem(m);
2233 		return (EINVAL);
2234 	}
2235 
2236 	bcopy(LLADDR(sdl), desten, sdl->sdl_alen);
2237 	return (0);
2238 }
2239 
2240 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2241 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2242 #ifdef SYSCTL_DECL
2243 SYSCTL_DECL(_net_inet6_icmp6);
2244 #endif
2245 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2246 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2247 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2248 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2249 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2250 	CTLFLAG_RW, &nd6_maxqueuelen, 1, "");
2251 
2252 static int
2253 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2254 {
2255 	int error;
2256 	char buf[1024];
2257 	struct in6_defrouter *d, *de;
2258 	struct nd_defrouter *dr;
2259 
2260 	if (req->newptr)
2261 		return EPERM;
2262 	error = 0;
2263 
2264 	for (dr = TAILQ_FIRST(&nd_defrouter); dr;
2265 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2266 		d = (struct in6_defrouter *)buf;
2267 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2268 
2269 		if (d + 1 <= de) {
2270 			bzero(d, sizeof(*d));
2271 			d->rtaddr.sin6_family = AF_INET6;
2272 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2273 			d->rtaddr.sin6_addr = dr->rtaddr;
2274 			sa6_recoverscope(&d->rtaddr);
2275 			d->flags = dr->flags;
2276 			d->rtlifetime = dr->rtlifetime;
2277 			d->expire = dr->expire;
2278 			d->if_index = dr->ifp->if_index;
2279 		} else
2280 			panic("buffer too short");
2281 
2282 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2283 		if (error)
2284 			break;
2285 	}
2286 
2287 	return (error);
2288 }
2289 
2290 static int
2291 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2292 {
2293 	int error;
2294 	char buf[1024];
2295 	struct in6_prefix *p, *pe;
2296 	struct nd_prefix *pr;
2297 
2298 	if (req->newptr)
2299 		return EPERM;
2300 	error = 0;
2301 
2302 	for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2303 		u_short advrtrs;
2304 		size_t advance;
2305 		struct sockaddr_in6 *sin6, *s6;
2306 		struct nd_pfxrouter *pfr;
2307 
2308 		p = (struct in6_prefix *)buf;
2309 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2310 
2311 		if (p + 1 <= pe) {
2312 			bzero(p, sizeof(*p));
2313 			sin6 = (struct sockaddr_in6 *)(p + 1);
2314 
2315 			p->prefix = pr->ndpr_prefix;
2316 			if (sa6_recoverscope(&p->prefix)) {
2317 				log(LOG_ERR,
2318 				    "scope error in prefix list (%s)\n",
2319 				    ip6_sprintf(&p->prefix.sin6_addr));
2320 				/* XXX: press on... */
2321 			}
2322 			p->raflags = pr->ndpr_raf;
2323 			p->prefixlen = pr->ndpr_plen;
2324 			p->vltime = pr->ndpr_vltime;
2325 			p->pltime = pr->ndpr_pltime;
2326 			p->if_index = pr->ndpr_ifp->if_index;
2327 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2328 				p->expire = 0;
2329 			else {
2330 				time_t maxexpire;
2331 
2332 				/* XXX: we assume time_t is signed. */
2333 				maxexpire = (-1) &
2334 				    ~((time_t)1 <<
2335 				    ((sizeof(maxexpire) * 8) - 1));
2336 				if (pr->ndpr_vltime <
2337 				    maxexpire - pr->ndpr_lastupdate) {
2338 				    p->expire = pr->ndpr_lastupdate +
2339 				        pr->ndpr_vltime;
2340 				} else
2341 					p->expire = maxexpire;
2342 			}
2343 			p->refcnt = pr->ndpr_refcnt;
2344 			p->flags = pr->ndpr_stateflags;
2345 			p->origin = PR_ORIG_RA;
2346 			advrtrs = 0;
2347 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2348 			     pfr = pfr->pfr_next) {
2349 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2350 					advrtrs++;
2351 					continue;
2352 				}
2353 				s6 = &sin6[advrtrs];
2354 				bzero(s6, sizeof(*s6));
2355 				s6->sin6_family = AF_INET6;
2356 				s6->sin6_len = sizeof(*sin6);
2357 				s6->sin6_addr = pfr->router->rtaddr;
2358 				if (sa6_recoverscope(s6)) {
2359 					log(LOG_ERR,
2360 					    "scope error in "
2361 					    "prefix list (%s)\n",
2362 					    ip6_sprintf(&pfr->router->rtaddr));
2363 				}
2364 				advrtrs++;
2365 			}
2366 			p->advrtrs = advrtrs;
2367 		} else
2368 			panic("buffer too short");
2369 
2370 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2371 		error = SYSCTL_OUT(req, buf, advance);
2372 		if (error)
2373 			break;
2374 	}
2375 
2376 	return (error);
2377 }
2378