xref: /netbsd-src/sys/netinet6/nd6.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: nd6.c,v 1.184 2016/01/08 08:50:07 ozaki-r Exp $	*/
2 /*	$KAME: nd6.c,v 1.279 2002/06/08 11:16:51 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 <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.184 2016/01/08 08:50:07 ozaki-r Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_net_mpsafe.h"
38 #endif
39 
40 #include "bridge.h"
41 #include "carp.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/ioctl.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/cprng.h>
59 
60 #include <net/if.h>
61 #include <net/if_dl.h>
62 #include <net/if_llatbl.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/if_ether.h>
66 #include <net/if_fddi.h>
67 #include <net/if_arc.h>
68 
69 #include <netinet/in.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/icmp6_private.h>
78 
79 #include <net/net_osdep.h>
80 
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
83 
84 /* timer values */
85 int	nd6_prune	= 1;	/* walk list every 1 seconds */
86 int	nd6_delay	= 5;	/* delay first probe time 5 second */
87 int	nd6_umaxtries	= 3;	/* maximum unicast query */
88 int	nd6_mmaxtries	= 3;	/* maximum multicast query */
89 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
90 int	nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
91 
92 /* preventing too many loops in ND option parsing */
93 int nd6_maxndopt = 10;	/* max # of ND options allowed */
94 
95 int nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
96 
97 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */
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 nd_drhead nd_defrouter;
109 struct nd_prhead nd_prefix = { 0 };
110 
111 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
112 static const struct sockaddr_in6 all1_sa = {
113 	  .sin6_family = AF_INET6
114 	, .sin6_len = sizeof(struct sockaddr_in6)
115 	, .sin6_addr = {.s6_addr = {0xff, 0xff, 0xff, 0xff,
116 				    0xff, 0xff, 0xff, 0xff,
117 				    0xff, 0xff, 0xff, 0xff,
118 				    0xff, 0xff, 0xff, 0xff}}
119 };
120 
121 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
122 static void nd6_slowtimo(void *);
123 static int regen_tmpaddr(struct in6_ifaddr *);
124 static void nd6_free(struct rtentry *, struct llentry *, int);
125 static void nd6_llinfo_timer(void *);
126 static void clear_llinfo_pqueue(struct llentry *);
127 
128 callout_t nd6_slowtimo_ch;
129 callout_t nd6_timer_ch;
130 extern callout_t in6_tmpaddrtimer_ch;
131 
132 static int fill_drlist(void *, size_t *, size_t);
133 static int fill_prlist(void *, size_t *, size_t);
134 
135 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
136 
137 void
138 nd6_init(void)
139 {
140 	static int nd6_init_done = 0;
141 
142 	if (nd6_init_done) {
143 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
144 		return;
145 	}
146 
147 	/* initialization of the default router list */
148 	TAILQ_INIT(&nd_defrouter);
149 
150 	nd6_init_done = 1;
151 
152 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
153 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
154 
155 	/* start timer */
156 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
157 	    nd6_slowtimo, NULL);
158 }
159 
160 struct nd_ifinfo *
161 nd6_ifattach(struct ifnet *ifp)
162 {
163 	struct nd_ifinfo *nd;
164 
165 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
166 
167 	nd->initialized = 1;
168 
169 	nd->chlim = IPV6_DEFHLIM;
170 	nd->basereachable = REACHABLE_TIME;
171 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
172 	nd->retrans = RETRANS_TIMER;
173 
174 	nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV;
175 
176 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
177 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
178 	 * because one of its members should. */
179 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
180 	    (ifp->if_flags & IFF_LOOPBACK))
181 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
182 
183 	/* A loopback interface does not need to accept RTADV.
184 	 * A bridge interface should not accept RTADV
185 	 * because one of its members should. */
186 	if (ip6_accept_rtadv &&
187 	    !(ifp->if_flags & IFF_LOOPBACK) &&
188 	    !(ifp->if_type != IFT_BRIDGE))
189 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
190 
191 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
192 	nd6_setmtu0(ifp, nd);
193 
194 	return nd;
195 }
196 
197 void
198 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
199 {
200 
201 	nd6_purge(ifp, ext);
202 	free(ext->nd_ifinfo, M_IP6NDP);
203 }
204 
205 void
206 nd6_setmtu(struct ifnet *ifp)
207 {
208 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
209 }
210 
211 void
212 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
213 {
214 	u_int32_t omaxmtu;
215 
216 	omaxmtu = ndi->maxmtu;
217 
218 	switch (ifp->if_type) {
219 	case IFT_ARCNET:
220 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
221 		break;
222 	case IFT_FDDI:
223 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu);
224 		break;
225 	default:
226 		ndi->maxmtu = ifp->if_mtu;
227 		break;
228 	}
229 
230 	/*
231 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
232 	 * undesirable situation.  We thus notify the operator of the change
233 	 * explicitly.  The check for omaxmtu is necessary to restrict the
234 	 * log to the case of changing the MTU, not initializing it.
235 	 */
236 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
237 		log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too"
238 		    " small for IPv6 which needs %lu\n",
239 		    if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long)
240 		    IPV6_MMTU);
241 	}
242 
243 	if (ndi->maxmtu > in6_maxmtu)
244 		in6_setmaxmtu(); /* check all interfaces just in case */
245 }
246 
247 void
248 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
249 {
250 
251 	memset(ndopts, 0, sizeof(*ndopts));
252 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
253 	ndopts->nd_opts_last
254 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
255 
256 	if (icmp6len == 0) {
257 		ndopts->nd_opts_done = 1;
258 		ndopts->nd_opts_search = NULL;
259 	}
260 }
261 
262 /*
263  * Take one ND option.
264  */
265 struct nd_opt_hdr *
266 nd6_option(union nd_opts *ndopts)
267 {
268 	struct nd_opt_hdr *nd_opt;
269 	int olen;
270 
271 	KASSERT(ndopts != NULL);
272 	KASSERT(ndopts->nd_opts_last != NULL);
273 
274 	if (ndopts->nd_opts_search == NULL)
275 		return NULL;
276 	if (ndopts->nd_opts_done)
277 		return NULL;
278 
279 	nd_opt = ndopts->nd_opts_search;
280 
281 	/* make sure nd_opt_len is inside the buffer */
282 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
283 		memset(ndopts, 0, sizeof(*ndopts));
284 		return NULL;
285 	}
286 
287 	olen = nd_opt->nd_opt_len << 3;
288 	if (olen == 0) {
289 		/*
290 		 * Message validation requires that all included
291 		 * options have a length that is greater than zero.
292 		 */
293 		memset(ndopts, 0, sizeof(*ndopts));
294 		return NULL;
295 	}
296 
297 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
298 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
299 		/* option overruns the end of buffer, invalid */
300 		memset(ndopts, 0, sizeof(*ndopts));
301 		return NULL;
302 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
303 		/* reached the end of options chain */
304 		ndopts->nd_opts_done = 1;
305 		ndopts->nd_opts_search = NULL;
306 	}
307 	return nd_opt;
308 }
309 
310 /*
311  * Parse multiple ND options.
312  * This function is much easier to use, for ND routines that do not need
313  * multiple options of the same type.
314  */
315 int
316 nd6_options(union nd_opts *ndopts)
317 {
318 	struct nd_opt_hdr *nd_opt;
319 	int i = 0;
320 
321 	KASSERT(ndopts != NULL);
322 	KASSERT(ndopts->nd_opts_last != NULL);
323 
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 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
335 			memset(ndopts, 0, 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 accommodate 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 			ICMP6_STATINC(ICMP6_STAT_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_locked(struct llentry *ln, time_t xtick)
395 {
396 
397 	CTASSERT(sizeof(time_t) > sizeof(int));
398 	LLE_WLOCK_ASSERT(ln);
399 
400 	if (xtick < 0) {
401 		ln->ln_expire = 0;
402 		ln->ln_ntick = 0;
403 		callout_halt(&ln->ln_timer_ch, &ln->lle_lock);
404 	} else {
405 		ln->ln_expire = time_uptime + xtick / hz;
406 		LLE_ADDREF(ln);
407 		if (xtick > INT_MAX) {
408 			ln->ln_ntick = xtick - INT_MAX;
409 			callout_reset(&ln->ln_timer_ch, INT_MAX,
410 			    nd6_llinfo_timer, ln);
411 		} else {
412 			ln->ln_ntick = 0;
413 			callout_reset(&ln->ln_timer_ch, xtick,
414 			    nd6_llinfo_timer, ln);
415 		}
416 	}
417 }
418 
419 void
420 nd6_llinfo_settimer(struct llentry *ln, time_t xtick)
421 {
422 
423 	LLE_WLOCK(ln);
424 	nd6_llinfo_settimer_locked(ln, xtick);
425 	LLE_WUNLOCK(ln);
426 }
427 
428 /*
429  * Gets source address of the first packet in hold queue
430  * and stores it in @src.
431  * Returns pointer to @src (if hold queue is not empty) or NULL.
432  */
433 static struct in6_addr *
434 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
435 {
436 	struct ip6_hdr *hip6;
437 
438 	if (ln == NULL || ln->ln_hold == NULL)
439 		return NULL;
440 
441 	/*
442 	 * assuming every packet in ln_hold has the same IP header
443 	 */
444 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
445 	/* XXX pullup? */
446 	if (sizeof(*hip6) < ln->ln_hold->m_len)
447 		*src = hip6->ip6_src;
448 	else
449 		src = NULL;
450 
451 	return src;
452 }
453 
454 static void
455 nd6_llinfo_timer(void *arg)
456 {
457 	struct llentry *ln = arg;
458 	struct rtentry *rt;
459 	const struct sockaddr_in6 *dst;
460 	struct ifnet *ifp;
461 	struct nd_ifinfo *ndi = NULL;
462 	bool send_ns = false;
463 	const struct in6_addr *daddr6 = NULL;
464 
465 	mutex_enter(softnet_lock);
466 	KERNEL_LOCK(1, NULL);
467 
468 	LLE_WLOCK(ln);
469 	if (ln->ln_ntick > 0) {
470 		nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
471 		goto out;
472 	}
473 
474 	if (callout_pending(&ln->la_timer)) {
475 		/*
476 		 * Here we are a bit odd here in the treatment of
477 		 * active/pending. If the pending bit is set, it got
478 		 * rescheduled before I ran. The active
479 		 * bit we ignore, since if it was stopped
480 		 * in ll_tablefree() and was currently running
481 		 * it would have return 0 so the code would
482 		 * not have deleted it since the callout could
483 		 * not be stopped so we want to go through
484 		 * with the delete here now. If the callout
485 		 * was restarted, the pending bit will be back on and
486 		 * we just want to bail since the callout_reset would
487 		 * return 1 and our reference would have been removed
488 		 * by nd6_llinfo_settimer_locked above since canceled
489 		 * would have been 1.
490 		 */
491 		goto out;
492 	}
493 
494 	ifp = ln->lle_tbl->llt_ifp;
495 	rt = ln->ln_rt;
496 
497 	KASSERT(rt != NULL);
498 	KASSERT(ifp != NULL);
499 
500 	ndi = ND_IFINFO(ifp);
501 	dst = satocsin6(rt_getkey(rt));
502 
503 	/* sanity check */
504 	if (rt->rt_llinfo && (struct llentry *)rt->rt_llinfo != ln)
505 		panic("rt_llinfo(%p) is not equal to ln(%p)",
506 		      rt->rt_llinfo, ln);
507 	if (!dst)
508 		panic("dst=0 in nd6_timer(ln=%p)", ln);
509 
510 	switch (ln->ln_state) {
511 	case ND6_LLINFO_INCOMPLETE:
512 		if (ln->ln_asked < nd6_mmaxtries) {
513 			ln->ln_asked++;
514 			send_ns = true;
515 		} else {
516 			struct mbuf *m = ln->ln_hold;
517 			if (m) {
518 				struct mbuf *m0;
519 
520 				/*
521 				 * assuming every packet in ln_hold has
522 				 * the same IP header
523 				 */
524 				m0 = m->m_nextpkt;
525 				m->m_nextpkt = NULL;
526 				ln->ln_hold = m0;
527 				clear_llinfo_pqueue(ln);
528  			}
529 			nd6_free(rt, ln, 0);
530 			ln = NULL;
531 			if (m != NULL)
532 				icmp6_error2(m, ICMP6_DST_UNREACH,
533 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
534 		}
535 		break;
536 	case ND6_LLINFO_REACHABLE:
537 		if (!ND6_LLINFO_PERMANENT(ln)) {
538 			ln->ln_state = ND6_LLINFO_STALE;
539 			nd6_llinfo_settimer_locked(ln, nd6_gctimer * hz);
540 		}
541 		break;
542 
543 	case ND6_LLINFO_PURGE:
544 	case ND6_LLINFO_STALE:
545 		/* Garbage Collection(RFC 2461 5.3) */
546 		if (!ND6_LLINFO_PERMANENT(ln)) {
547 			nd6_free(rt, ln, 1);
548 			ln = NULL;
549 		}
550 		break;
551 
552 	case ND6_LLINFO_DELAY:
553 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
554 			/* We need NUD */
555 			ln->ln_asked = 1;
556 			ln->ln_state = ND6_LLINFO_PROBE;
557 			daddr6 = &dst->sin6_addr;
558 			send_ns = true;
559 		} else {
560 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
561 			nd6_llinfo_settimer_locked(ln, nd6_gctimer * hz);
562 		}
563 		break;
564 	case ND6_LLINFO_PROBE:
565 		if (ln->ln_asked < nd6_umaxtries) {
566 			ln->ln_asked++;
567 			daddr6 = &dst->sin6_addr;
568 			send_ns = true;
569 		} else {
570 			nd6_free(rt, ln, 0);
571 			ln = NULL;
572 		}
573 		break;
574 	}
575 
576 	if (send_ns) {
577 		struct in6_addr src, *psrc;
578 
579 		nd6_llinfo_settimer_locked(ln, ndi->retrans * hz / 1000);
580 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
581 		LLE_FREE_LOCKED(ln);
582 		ln = NULL;
583 		nd6_ns_output(ifp, daddr6, &dst->sin6_addr, psrc, 0);
584 	}
585 
586 out:
587 	if (ln != NULL)
588 		LLE_FREE_LOCKED(ln);
589 	KERNEL_UNLOCK_ONE(NULL);
590 	mutex_exit(softnet_lock);
591 }
592 
593 /*
594  * ND6 timer routine to expire default route list and prefix list
595  */
596 void
597 nd6_timer(void *ignored_arg)
598 {
599 	struct nd_defrouter *next_dr, *dr;
600 	struct nd_prefix *next_pr, *pr;
601 	struct in6_ifaddr *ia6, *nia6;
602 
603 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
604 	    nd6_timer, NULL);
605 
606 	mutex_enter(softnet_lock);
607 	KERNEL_LOCK(1, NULL);
608 
609 	/* expire default router list */
610 
611 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, next_dr) {
612 		if (dr->expire && dr->expire < time_uptime) {
613 			defrtrlist_del(dr, NULL);
614 		}
615 	}
616 
617 	/*
618 	 * expire interface addresses.
619 	 * in the past the loop was inside prefix expiry processing.
620 	 * However, from a stricter speci-confrmance standpoint, we should
621 	 * rather separate address lifetimes and prefix lifetimes.
622 	 */
623   addrloop:
624 	for (ia6 = in6_ifaddr; ia6; ia6 = nia6) {
625 		nia6 = ia6->ia_next;
626 		/* check address lifetime */
627 		if (IFA6_IS_INVALID(ia6)) {
628 			int regen = 0;
629 
630 			/*
631 			 * If the expiring address is temporary, try
632 			 * regenerating a new one.  This would be useful when
633 			 * we suspended a laptop PC, then turned it on after a
634 			 * period that could invalidate all temporary
635 			 * addresses.  Although we may have to restart the
636 			 * loop (see below), it must be after purging the
637 			 * address.  Otherwise, we'd see an infinite loop of
638 			 * regeneration.
639 			 */
640 			if (ip6_use_tempaddr &&
641 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
642 				if (regen_tmpaddr(ia6) == 0)
643 					regen = 1;
644 			}
645 
646  			in6_purgeaddr(&ia6->ia_ifa);
647 
648 			if (regen)
649 				goto addrloop; /* XXX: see below */
650 		} else if (IFA6_IS_DEPRECATED(ia6)) {
651 			int oldflags = ia6->ia6_flags;
652 
653 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
654 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
655 				rt_newaddrmsg(RTM_NEWADDR,
656 				    (struct ifaddr *)ia6, 0, NULL);
657 			}
658 
659 			/*
660 			 * If a temporary address has just become deprecated,
661 			 * regenerate a new one if possible.
662 			 */
663 			if (ip6_use_tempaddr &&
664 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
665 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
666 
667 				if (regen_tmpaddr(ia6) == 0) {
668 					/*
669 					 * A new temporary address is
670 					 * generated.
671 					 * XXX: this means the address chain
672 					 * has changed while we are still in
673 					 * the loop.  Although the change
674 					 * would not cause disaster (because
675 					 * it's not a deletion, but an
676 					 * addition,) we'd rather restart the
677 					 * loop just for safety.  Or does this
678 					 * significantly reduce performance??
679 					 */
680 					goto addrloop;
681 				}
682 			}
683 		} else {
684 			/*
685 			 * A new RA might have made a deprecated address
686 			 * preferred.
687 			 */
688 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
689 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
690 				rt_newaddrmsg(RTM_NEWADDR,
691 				    (struct ifaddr *)ia6, 0, NULL);
692 			}
693 		}
694 	}
695 
696 	/* expire prefix list */
697 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, next_pr) {
698 		/*
699 		 * check prefix lifetime.
700 		 * since pltime is just for autoconf, pltime processing for
701 		 * prefix is not necessary.
702 		 */
703 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
704 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
705 
706 			/*
707 			 * address expiration and prefix expiration are
708 			 * separate.  NEVER perform in6_purgeaddr here.
709 			 */
710 
711 			prelist_remove(pr);
712 		}
713 	}
714 
715 	KERNEL_UNLOCK_ONE(NULL);
716 	mutex_exit(softnet_lock);
717 }
718 
719 /* ia6: deprecated/invalidated temporary address */
720 static int
721 regen_tmpaddr(struct in6_ifaddr *ia6)
722 {
723 	struct ifaddr *ifa;
724 	struct ifnet *ifp;
725 	struct in6_ifaddr *public_ifa6 = NULL;
726 
727 	ifp = ia6->ia_ifa.ifa_ifp;
728 	IFADDR_FOREACH(ifa, ifp) {
729 		struct in6_ifaddr *it6;
730 
731 		if (ifa->ifa_addr->sa_family != AF_INET6)
732 			continue;
733 
734 		it6 = (struct in6_ifaddr *)ifa;
735 
736 		/* ignore no autoconf addresses. */
737 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
738 			continue;
739 
740 		/* ignore autoconf addresses with different prefixes. */
741 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
742 			continue;
743 
744 		/*
745 		 * Now we are looking at an autoconf address with the same
746 		 * prefix as ours.  If the address is temporary and is still
747 		 * preferred, do not create another one.  It would be rare, but
748 		 * could happen, for example, when we resume a laptop PC after
749 		 * a long period.
750 		 */
751 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
752 		    !IFA6_IS_DEPRECATED(it6)) {
753 			public_ifa6 = NULL;
754 			break;
755 		}
756 
757 		/*
758 		 * This is a public autoconf address that has the same prefix
759 		 * as ours.  If it is preferred, keep it.  We can't break the
760 		 * loop here, because there may be a still-preferred temporary
761 		 * address with the prefix.
762 		 */
763 		if (!IFA6_IS_DEPRECATED(it6))
764 		    public_ifa6 = it6;
765 	}
766 
767 	if (public_ifa6 != NULL) {
768 		int e;
769 
770 		/*
771 		 * Random factor is introduced in the preferred lifetime, so
772 		 * we do not need additional delay (3rd arg to in6_tmpifadd).
773 		 */
774 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
775 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
776 			    " tmp addr, errno=%d\n", e);
777 			return -1;
778 		}
779 		return 0;
780 	}
781 
782 	return -1;
783 }
784 
785 bool
786 nd6_accepts_rtadv(const struct nd_ifinfo *ndi)
787 {
788 	switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) {
789 	case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV:
790 		return true;
791 	case ND6_IFF_ACCEPT_RTADV:
792 		return ip6_accept_rtadv != 0;
793 	case ND6_IFF_OVERRIDE_RTADV:
794 	case 0:
795 	default:
796 		return false;
797 	}
798 }
799 
800 /*
801  * Nuke neighbor cache/prefix/default router management table, right before
802  * ifp goes away.
803  */
804 void
805 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
806 {
807 	struct nd_defrouter *dr, *ndr;
808 	struct nd_prefix *pr, *npr;
809 
810 	/*
811 	 * During detach, the ND info might be already removed, but
812 	 * then is explitly passed as argument.
813 	 * Otherwise get it from ifp->if_afdata.
814 	 */
815 	if (ext == NULL)
816 		ext = ifp->if_afdata[AF_INET6];
817 	if (ext == NULL)
818 		return;
819 
820 	/*
821 	 * Nuke default router list entries toward ifp.
822 	 * We defer removal of default router list entries that is installed
823 	 * in the routing table, in order to keep additional side effects as
824 	 * small as possible.
825 	 */
826 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
827 		if (dr->installed)
828 			continue;
829 
830 		if (dr->ifp == ifp) {
831 			KASSERT(ext != NULL);
832 			defrtrlist_del(dr, ext);
833 		}
834 	}
835 
836 	TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) {
837 		if (!dr->installed)
838 			continue;
839 
840 		if (dr->ifp == ifp) {
841 			KASSERT(ext != NULL);
842 			defrtrlist_del(dr, ext);
843 		}
844 	}
845 
846 	/* Nuke prefix list entries toward ifp */
847 	LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, npr) {
848 		if (pr->ndpr_ifp == ifp) {
849 			/*
850 			 * Because if_detach() does *not* release prefixes
851 			 * while purging addresses the reference count will
852 			 * still be above zero. We therefore reset it to
853 			 * make sure that the prefix really gets purged.
854 			 */
855 			pr->ndpr_refcnt = 0;
856 			/*
857 			 * Previously, pr->ndpr_addr is removed as well,
858 			 * but I strongly believe we don't have to do it.
859 			 * nd6_purge() is only called from in6_ifdetach(),
860 			 * which removes all the associated interface addresses
861 			 * by itself.
862 			 * (jinmei@kame.net 20010129)
863 			 */
864 			prelist_remove(pr);
865 		}
866 	}
867 
868 	/* cancel default outgoing interface setting */
869 	if (nd6_defifindex == ifp->if_index)
870 		nd6_setdefaultiface(0);
871 
872 	/* XXX: too restrictive? */
873 	if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) {
874 		struct nd_ifinfo *ndi = ND_IFINFO(ifp);
875 		if (ndi && nd6_accepts_rtadv(ndi)) {
876 			/* refresh default router list */
877 			defrouter_select();
878 		}
879 	}
880 
881 	/*
882 	 * We may not need to nuke the neighbor cache entries here
883 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
884 	 * nd6_purge() is invoked by in6_ifdetach() which is called
885 	 * from if_detach() where everything gets purged. However
886 	 * in6_ifdetach is directly called from vlan(4), so we still
887 	 * need to purge entries here.
888 	 */
889 	if (ext->lltable != NULL)
890 		lltable_purge_entries(ext->lltable);
891 }
892 
893 static struct rtentry *
894 nd6_lookup1(const struct in6_addr *addr6, int create, struct ifnet *ifp,
895     int cloning)
896 {
897 	struct rtentry *rt;
898 	struct sockaddr_in6 sin6;
899 
900 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
901 	rt = rtalloc1((struct sockaddr *)&sin6, create);
902 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) == 0) {
903 		/*
904 		 * This is the case for the default route.
905 		 * If we want to create a neighbor cache for the address, we
906 		 * should free the route for the destination and allocate an
907 		 * interface route.
908 		 */
909 		if (create) {
910 			rtfree(rt);
911 			rt = NULL;
912 		}
913 	}
914 	if (rt != NULL)
915 		;
916 	else if (create && ifp) {
917 		int e;
918 
919 		/*
920 		 * If no route is available and create is set,
921 		 * we allocate a host route for the destination
922 		 * and treat it like an interface route.
923 		 * This hack is necessary for a neighbor which can't
924 		 * be covered by our own prefix.
925 		 */
926 		struct ifaddr *ifa =
927 		    ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp);
928 		if (ifa == NULL)
929 			return NULL;
930 
931 		/*
932 		 * Create a new route.  RTF_LLINFO is necessary
933 		 * to create a Neighbor Cache entry for the
934 		 * destination in nd6_rtrequest which will be
935 		 * called in rtrequest via ifa->ifa_rtrequest.
936 		 */
937 		if ((e = rtrequest(RTM_ADD, (const struct sockaddr *)&sin6,
938 		    ifa->ifa_addr, (const struct sockaddr *)&all1_sa,
939 		    (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) &
940 		    ~RTF_CLONING, &rt)) != 0) {
941 #if 0
942 			log(LOG_ERR,
943 			    "nd6_lookup: failed to add route for a "
944 			    "neighbor(%s), errno=%d\n",
945 			    ip6_sprintf(addr6), e);
946 #endif
947 			return NULL;
948 		}
949 		if (rt == NULL)
950 			return NULL;
951 		if (rt->rt_llinfo) {
952 			struct llentry *ln = rt->rt_llinfo;
953 			ln->ln_state = ND6_LLINFO_NOSTATE;
954 		}
955 	} else
956 		return NULL;
957 
958 	/*
959 	 * Check for a cloning route to match the address.
960 	 * This should only be set from in6_is_addr_neighbor so we avoid
961 	 * a potentially expensive second call to rtalloc1.
962 	 */
963 	if (cloning &&
964 	    rt->rt_flags & (RTF_CLONING | RTF_CLONED) &&
965 	    (rt->rt_ifp == ifp
966 #if NBRIDGE > 0
967 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
968 #endif
969 #if NCARP > 0
970 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
971 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
972 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
973 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
974 #endif
975 	    ))
976 		return rt;
977 
978 	/*
979 	 * Validation for the entry.
980 	 * Note that the check for rt_llinfo is necessary because a cloned
981 	 * route from a parent route that has the L flag (e.g. the default
982 	 * route to a p2p interface) may have the flag, too, while the
983 	 * destination is not actually a neighbor.
984 	 * XXX: we can't use rt->rt_ifp to check for the interface, since
985 	 *      it might be the loopback interface if the entry is for our
986 	 *      own address on a non-loopback interface. Instead, we should
987 	 *      use rt->rt_ifa->ifa_ifp, which would specify the REAL
988 	 *	interface.
989 	 * Note also that ifa_ifp and ifp may differ when we connect two
990 	 * interfaces to a same link, install a link prefix to an interface,
991 	 * and try to install a neighbor cache on an interface that does not
992 	 * have a route to the prefix.
993 	 */
994 	if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 ||
995 	    rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL ||
996 	    (ifp && rt->rt_ifa->ifa_ifp != ifp)) {
997 		if (create) {
998 			nd6log((LOG_DEBUG,
999 			    "nd6_lookup: failed to lookup %s (if = %s)\n",
1000 			    ip6_sprintf(addr6),
1001 			    ifp ? if_name(ifp) : "unspec"));
1002 		}
1003 		rtfree(rt);
1004 		return NULL;
1005 	}
1006 	return rt;
1007 }
1008 
1009 struct rtentry *
1010 nd6_lookup(const struct in6_addr *addr6, int create, struct ifnet *ifp)
1011 {
1012 
1013 	return nd6_lookup1(addr6, create, ifp, 0);
1014 }
1015 
1016 /*
1017  * Detect if a given IPv6 address identifies a neighbor on a given link.
1018  * XXX: should take care of the destination of a p2p link?
1019  */
1020 int
1021 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1022 {
1023 	struct nd_prefix *pr;
1024 	struct rtentry *rt;
1025 
1026 	/*
1027 	 * A link-local address is always a neighbor.
1028 	 * XXX: a link does not necessarily specify a single interface.
1029 	 */
1030 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1031 		struct sockaddr_in6 sin6_copy;
1032 		u_int32_t zone;
1033 
1034 		/*
1035 		 * We need sin6_copy since sa6_recoverscope() may modify the
1036 		 * content (XXX).
1037 		 */
1038 		sin6_copy = *addr;
1039 		if (sa6_recoverscope(&sin6_copy))
1040 			return 0; /* XXX: should be impossible */
1041 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1042 			return 0;
1043 		if (sin6_copy.sin6_scope_id == zone)
1044 			return 1;
1045 		else
1046 			return 0;
1047 	}
1048 
1049 	/*
1050 	 * If the address matches one of our on-link prefixes, it should be a
1051 	 * neighbor.
1052 	 */
1053 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1054 		if (pr->ndpr_ifp != ifp)
1055 			continue;
1056 
1057 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
1058 			continue;
1059 
1060 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1061 		    &addr->sin6_addr, &pr->ndpr_mask))
1062 			return 1;
1063 	}
1064 
1065 	/*
1066 	 * If the default router list is empty, all addresses are regarded
1067 	 * as on-link, and thus, as a neighbor.
1068 	 * XXX: we restrict the condition to hosts, because routers usually do
1069 	 * not have the "default router list".
1070 	 */
1071 	if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL &&
1072 	    nd6_defifindex == ifp->if_index) {
1073 		return 1;
1074 	}
1075 
1076 	/*
1077 	 * Even if the address matches none of our addresses, it might match
1078 	 * a cloning route or be in the neighbor cache.
1079 	 */
1080 	rt = nd6_lookup1(&addr->sin6_addr, 0, ifp, 1);
1081 	if (rt != NULL) {
1082 		rtfree(rt);
1083 		return 1;
1084 	}
1085 
1086 	return 0;
1087 }
1088 
1089 /*
1090  * Free an nd6 llinfo entry.
1091  * Since the function would cause significant changes in the kernel, DO NOT
1092  * make it global, unless you have a strong reason for the change, and are sure
1093  * that the change is safe.
1094  */
1095 static void
1096 nd6_free(struct rtentry *rt, struct llentry *ln, int gc)
1097 {
1098 	struct in6_addr in6 = satocsin6(rt_getkey(rt))->sin6_addr;
1099 	struct nd_defrouter *dr;
1100 	int error;
1101 
1102 	KASSERT(ln != NULL);
1103 	KASSERT(ln == rt->rt_llinfo);
1104 	LLE_WLOCK_ASSERT(ln);
1105 
1106 	/*
1107 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1108 	 * even though it is not harmful, it was not really necessary.
1109 	 */
1110 
1111 	/* cancel timer */
1112 	nd6_llinfo_settimer_locked(ln, -1);
1113 
1114 	if (!ip6_forwarding) {
1115 		int s;
1116 		s = splsoftnet();
1117 		dr = defrouter_lookup(&satocsin6(rt_getkey(rt))->sin6_addr,
1118 		    rt->rt_ifp);
1119 
1120 		if (dr != NULL && dr->expire &&
1121 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1122 			/*
1123 			 * If the reason for the deletion is just garbage
1124 			 * collection, and the neighbor is an active default
1125 			 * router, do not delete it.  Instead, reset the GC
1126 			 * timer using the router's lifetime.
1127 			 * Simply deleting the entry would affect default
1128 			 * router selection, which is not necessarily a good
1129 			 * thing, especially when we're using router preference
1130 			 * values.
1131 			 * XXX: the check for ln_state would be redundant,
1132 			 *      but we intentionally keep it just in case.
1133 			 */
1134 			if (dr->expire > time_uptime)
1135 				nd6_llinfo_settimer_locked(ln,
1136 				    (dr->expire - time_uptime) * hz);
1137 			else
1138 				nd6_llinfo_settimer_locked(ln,
1139 				    nd6_gctimer * hz);
1140 			splx(s);
1141 			LLE_WUNLOCK(ln);
1142 			return;
1143 		}
1144 
1145 		if (ln->ln_router || dr) {
1146 			/*
1147 			 * rt6_flush must be called whether or not the neighbor
1148 			 * is in the Default Router List.
1149 			 * See a corresponding comment in nd6_na_input().
1150 			 */
1151 			rt6_flush(&in6, rt->rt_ifp);
1152 		}
1153 
1154 		if (dr) {
1155 			/*
1156 			 * Unreachablity of a router might affect the default
1157 			 * router selection and on-link detection of advertised
1158 			 * prefixes.
1159 			 */
1160 
1161 			/*
1162 			 * Temporarily fake the state to choose a new default
1163 			 * router and to perform on-link determination of
1164 			 * prefixes correctly.
1165 			 * Below the state will be set correctly,
1166 			 * or the entry itself will be deleted.
1167 			 */
1168 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1169 
1170 			/*
1171 			 * Since defrouter_select() does not affect the
1172 			 * on-link determination and MIP6 needs the check
1173 			 * before the default router selection, we perform
1174 			 * the check now.
1175 			 */
1176 			pfxlist_onlink_check();
1177 
1178 			/*
1179 			 * refresh default router list
1180 			 */
1181 			defrouter_select();
1182 		}
1183 		splx(s);
1184 	}
1185 
1186 	LLE_WUNLOCK(ln);
1187 	/*
1188 	 * Detach the route from the routing tree and the list of neighbor
1189 	 * caches, and disable the route entry not to be used in already
1190 	 * cached routes.
1191 	 */
1192 	error = rtrequest_newmsg(RTM_DELETE, rt_getkey(rt), NULL,
1193 	    rt_mask(rt), 0);
1194 	if (error != 0) {
1195 		/* XXX need error message? */;
1196 	}
1197 }
1198 
1199 /*
1200  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1201  *
1202  * XXX cost-effective methods?
1203  */
1204 void
1205 nd6_nud_hint(struct rtentry *rt)
1206 {
1207 	struct llentry *ln;
1208 
1209 	if (rt == NULL)
1210 		return;
1211 
1212 	if ((rt->rt_flags & RTF_GATEWAY) != 0 ||
1213 	    (rt->rt_flags & RTF_LLINFO) == 0 ||
1214 	    !rt->rt_llinfo || !rt->rt_gateway ||
1215 	    rt->rt_gateway->sa_family != AF_LINK) {
1216 		/* This is not a host route. */
1217 		return;
1218 	}
1219 
1220 	ln = rt->rt_llinfo;
1221 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1222 		return;
1223 
1224 	/*
1225 	 * if we get upper-layer reachability confirmation many times,
1226 	 * it is possible we have false information.
1227 	 */
1228 	ln->ln_byhint++;
1229 	if (ln->ln_byhint > nd6_maxnudhint)
1230 		return;
1231 
1232 	ln->ln_state = ND6_LLINFO_REACHABLE;
1233 	if (!ND6_LLINFO_PERMANENT(ln)) {
1234 		nd6_llinfo_settimer(ln,
1235 		    ND_IFINFO(rt->rt_ifp)->reachable * hz);
1236 	}
1237 
1238 	return;
1239 }
1240 
1241 static int
1242 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
1243 {
1244 	int *n = farg;
1245 
1246 	if (*n <= 0)
1247 		return 0;
1248 
1249 	if (ND6_LLINFO_PERMANENT(ln))
1250 		return 0;
1251 
1252 	LLE_WLOCK(ln);
1253 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1254 		ln->ln_state = ND6_LLINFO_STALE;
1255 	else
1256 		ln->ln_state = ND6_LLINFO_PURGE;
1257 	nd6_llinfo_settimer_locked(ln, 0);
1258 	LLE_WUNLOCK(ln);
1259 
1260 	(*n)--;
1261 	return 0;
1262 }
1263 
1264 static void
1265 nd6_gc_neighbors(struct lltable *llt)
1266 {
1267 	int max_gc_entries = 10;
1268 
1269 	if (ip6_neighborgcthresh >= 0 &&
1270 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
1271 		/*
1272 		 * XXX entries that are "less recently used" should be
1273 		 * freed first.
1274 		 */
1275 		lltable_foreach_lle(llt, nd6_purge_entry, &max_gc_entries);
1276 	}
1277 }
1278 
1279 void
1280 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
1281 {
1282 	struct sockaddr *gate = rt->rt_gateway;
1283 	struct llentry *ln;
1284 	struct ifnet *ifp = rt->rt_ifp;
1285 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
1286 	struct ifaddr *ifa;
1287 	int flags = 0;
1288 	bool use_lo0ifp = false;
1289 
1290 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1291 
1292 	if (req == RTM_LLINFO_UPD) {
1293 		int rc;
1294 		struct in6_addr *in6;
1295 		struct in6_addr in6_all;
1296 		int anycast;
1297 
1298 		if ((ifa = info->rti_ifa) == NULL)
1299 			return;
1300 
1301 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1302 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
1303 
1304 		in6_all = in6addr_linklocal_allnodes;
1305 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
1306 			log(LOG_ERR, "%s: failed to set scope %s "
1307 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
1308 			return;
1309 		}
1310 
1311 		/* XXX don't set Override for proxy addresses */
1312 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
1313 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
1314 #if 0
1315 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
1316 #endif
1317 		    , 1, NULL);
1318 		return;
1319 	}
1320 
1321 	if ((rt->rt_flags & RTF_GATEWAY) != 0)
1322 		return;
1323 
1324 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
1325 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1326 		/*
1327 		 * This is probably an interface direct route for a link
1328 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
1329 		 * We do not need special treatment below for such a route.
1330 		 * Moreover, the RTF_LLINFO flag which would be set below
1331 		 * would annoy the ndp(8) command.
1332 		 */
1333 		return;
1334 	}
1335 
1336 	IF_AFDATA_RLOCK(ifp);
1337 	ln = lla_lookup(LLTABLE6(ifp), flags, rt_getkey(rt));
1338 	IF_AFDATA_RUNLOCK(ifp);
1339 
1340 	if (req == RTM_RESOLVE &&
1341 	    (nd6_need_cache(ifp) == 0 || /* stf case */
1342 	     !nd6_is_addr_neighbor(satocsin6(rt_getkey(rt)), ifp))) {
1343 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1344 		/*
1345 		 * FreeBSD and BSD/OS often make a cloned host route based
1346 		 * on a less-specific route (e.g. the default route).
1347 		 * If the less specific route does not have a "gateway"
1348 		 * (this is the case when the route just goes to a p2p or an
1349 		 * stf interface), we'll mistakenly make a neighbor cache for
1350 		 * the host route, and will see strange neighbor solicitation
1351 		 * for the corresponding destination.  In order to avoid the
1352 		 * confusion, we check if the destination of the route is
1353 		 * a neighbor in terms of neighbor discovery, and stop the
1354 		 * process if not.  Additionally, we remove the LLINFO flag
1355 		 * so that ndp(8) will not try to get the neighbor information
1356 		 * of the destination.
1357 		 */
1358 		rt->rt_flags &= ~RTF_LLINFO;
1359 		return;
1360 	}
1361 
1362 	switch (req) {
1363 	case RTM_ADD:
1364 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1365 		/*
1366 		 * There is no backward compatibility :)
1367 		 *
1368 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
1369 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
1370 		 *	   rt->rt_flags |= RTF_CLONING;
1371 		 */
1372 		if ((rt->rt_flags & RTF_CLONING) ||
1373 		    ((rt->rt_flags & (RTF_LLINFO | RTF_LOCAL)) && ln == NULL)) {
1374 			union {
1375 				struct sockaddr sa;
1376 				struct sockaddr_dl sdl;
1377 				struct sockaddr_storage ss;
1378 			} u;
1379 			/*
1380 			 * Case 1: This route should come from a route to
1381 			 * interface (RTF_CLONING case) or the route should be
1382 			 * treated as on-link but is currently not
1383 			 * (RTF_LLINFO && ln == NULL case).
1384 			 */
1385 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
1386 			    ifp->if_index, ifp->if_type,
1387 			    NULL, namelen, NULL, addrlen) == NULL) {
1388 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
1389 				    "failed on %s\n", __func__, __LINE__,
1390 				    sizeof(u.ss), if_name(ifp));
1391 			}
1392 			rt_setgate(rt, &u.sa);
1393 			gate = rt->rt_gateway;
1394 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1395 			if (ln != NULL)
1396 				nd6_llinfo_settimer_locked(ln, 0);
1397 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1398 			if ((rt->rt_flags & RTF_CLONING) != 0)
1399 				break;
1400 		}
1401 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1402 		/*
1403 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
1404 		 * We don't do that here since llinfo is not ready yet.
1405 		 *
1406 		 * There are also couple of other things to be discussed:
1407 		 * - unsolicited NA code needs improvement beforehand
1408 		 * - RFC2461 says we MAY send multicast unsolicited NA
1409 		 *   (7.2.6 paragraph 4), however, it also says that we
1410 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
1411 		 *   we don't have anything like it right now.
1412 		 *   note that the mechanism needs a mutual agreement
1413 		 *   between proxies, which means that we need to implement
1414 		 *   a new protocol, or a new kludge.
1415 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
1416 		 *   we need to check ip6forwarding before sending it.
1417 		 *   (or should we allow proxy ND configuration only for
1418 		 *   routers?  there's no mention about proxy ND from hosts)
1419 		 */
1420 #if 0
1421 		/* XXX it does not work */
1422 		if (rt->rt_flags & RTF_ANNOUNCE)
1423 			nd6_na_output(ifp,
1424 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1425 			      &satocsin6(rt_getkey(rt))->sin6_addr,
1426 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
1427 			      1, NULL);
1428 #endif
1429 		/* FALLTHROUGH */
1430 	case RTM_RESOLVE:
1431 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
1432 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1433 			/*
1434 			 * Address resolution isn't necessary for a point to
1435 			 * point link, so we can skip this test for a p2p link.
1436 			 */
1437 			if (gate->sa_family != AF_LINK ||
1438 			    gate->sa_len <
1439 			    sockaddr_dl_measure(namelen, addrlen)) {
1440 				log(LOG_DEBUG,
1441 				    "nd6_rtrequest: bad gateway value: %s\n",
1442 				    if_name(ifp));
1443 				break;
1444 			}
1445 			satosdl(gate)->sdl_type = ifp->if_type;
1446 			satosdl(gate)->sdl_index = ifp->if_index;
1447 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1448 		}
1449 		if (ln != NULL)
1450 			break;	/* This happens on a route change */
1451 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1452 
1453 		/* Determine to use lo0ifp or not before lla_create */
1454 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp,
1455 		    &satocsin6(rt_getkey(rt))->sin6_addr);
1456 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1457 		if (ifa != NULL && nd6_useloopback)
1458 			use_lo0ifp = true;
1459 
1460 		/*
1461 		 * Case 2: This route may come from cloning, or a manual route
1462 		 * add with a LL address.
1463 		 */
1464 		flags = LLE_EXCLUSIVE;
1465 		if ((rt->rt_flags & RTF_CLONED) == 0)
1466 			flags |= LLE_IFADDR;
1467 
1468 #define	_IFP()	(use_lo0ifp ? lo0ifp : ifp)
1469 		IF_AFDATA_WLOCK(_IFP());
1470 		ln = lla_create(LLTABLE6(_IFP()), flags, rt_getkey(rt));
1471 		IF_AFDATA_WUNLOCK(_IFP());
1472 
1473 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1474 		if (ln == NULL) {
1475 			log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n");
1476 			break;
1477 		}
1478 
1479 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1480 		nd6_inuse++;
1481 		nd6_allocated++;
1482 		ln->ln_rt = rt;
1483 		rt->rt_refcnt++;
1484 		rt->rt_llinfo = ln;
1485 		LLE_ADDREF(ln);
1486 		rt->rt_flags |= RTF_LLINFO;
1487 		switch (_IFP()->if_type) {
1488 #if NTOKEN > 0
1489 		case IFT_ISO88025:
1490 			ln->la_opaque = kmem_alloc(sizeof(struct token_rif),
1491 			    KM_SLEEP);
1492 			break;
1493 #endif /* NTOKEN > 0 */
1494 		default:
1495 			break;
1496 		}
1497 #undef _IFP
1498 
1499 		/* this is required for "ndp" command. - shin */
1500 		if (req == RTM_ADD) {
1501 		        /*
1502 			 * gate should have some valid AF_LINK entry,
1503 			 * and ln->ln_expire should have some lifetime
1504 			 * which is specified by ndp command.
1505 			 */
1506 			ln->ln_state = ND6_LLINFO_REACHABLE;
1507 			ln->ln_byhint = 0;
1508 		} else {
1509 		        /*
1510 			 * When req == RTM_RESOLVE, rt is created and
1511 			 * initialized in rtrequest(), so rt_expire is 0.
1512 			 */
1513 			ln->ln_state = ND6_LLINFO_NOSTATE;
1514 			nd6_llinfo_settimer_locked(ln, 0);
1515 		}
1516 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
1517 
1518 		/*
1519 		 * check if rt_getkey(rt) is an address assigned
1520 		 * to the interface.
1521 		 */
1522 		if (ifa != NULL) {
1523 			const void *mac;
1524 			nd6_llinfo_settimer_locked(ln, -1);
1525 			ln->ln_state = ND6_LLINFO_REACHABLE;
1526 			ln->ln_byhint = 0;
1527 			if ((mac = nd6_ifptomac(ifp)) != NULL) {
1528 				/* XXX check for error */
1529 				if (sockaddr_dl_setaddr(satosdl(gate),
1530 				    gate->sa_len, mac,
1531 				    ifp->if_addrlen) == NULL) {
1532 					printf("%s.%d: "
1533 					    "sockaddr_dl_setaddr(, %d, ) "
1534 					    "failed on %s\n", __func__,
1535 					    __LINE__, gate->sa_len,
1536 					    if_name(ifp));
1537 				}
1538 			}
1539 			if (nd6_useloopback) {
1540 				ifp = rt->rt_ifp = lo0ifp;	/* XXX */
1541 				/*
1542 				 * Make sure rt_ifa be equal to the ifaddr
1543 				 * corresponding to the address.
1544 				 * We need this because when we refer
1545 				 * rt_ifa->ia6_flags in ip6_input, we assume
1546 				 * that the rt_ifa points to the address instead
1547 				 * of the loopback address.
1548 				 */
1549 				if (ifa != rt->rt_ifa)
1550 					rt_replace_ifa(rt, ifa);
1551 				rt->rt_rmx.rmx_mtu = 0;
1552 				rt->rt_flags &= ~RTF_CLONED;
1553 			}
1554 			rt->rt_flags |= RTF_LOCAL;
1555 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1556 			nd6_llinfo_settimer_locked(ln, -1);
1557 			ln->ln_state = ND6_LLINFO_REACHABLE;
1558 			ln->ln_byhint = 0;
1559 
1560 			/* join solicited node multicast for proxy ND */
1561 			if (ifp->if_flags & IFF_MULTICAST) {
1562 				struct in6_addr llsol;
1563 				int error;
1564 
1565 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1566 				llsol.s6_addr32[0] = htonl(0xff020000);
1567 				llsol.s6_addr32[1] = 0;
1568 				llsol.s6_addr32[2] = htonl(1);
1569 				llsol.s6_addr8[12] = 0xff;
1570 				if (in6_setscope(&llsol, ifp, NULL))
1571 					break;
1572 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1573 					nd6log((LOG_ERR, "%s: failed to join "
1574 					    "%s (errno=%d)\n", if_name(ifp),
1575 					    ip6_sprintf(&llsol), error));
1576 				}
1577 			}
1578 		}
1579 		LLE_WUNLOCK(ln);
1580 		/*
1581 		 * If we have too many cache entries, initiate immediate
1582 		 * purging for some entries.
1583 		 */
1584 		nd6_gc_neighbors(ln->lle_tbl);
1585 		ln = NULL;
1586 
1587 		break;
1588 
1589 	case RTM_DELETE:
1590 		if (ln == NULL)
1591 			break;
1592 		/* leave from solicited node multicast for proxy ND */
1593 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1594 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1595 			struct in6_addr llsol;
1596 			struct in6_multi *in6m;
1597 
1598 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1599 			llsol.s6_addr32[0] = htonl(0xff020000);
1600 			llsol.s6_addr32[1] = 0;
1601 			llsol.s6_addr32[2] = htonl(1);
1602 			llsol.s6_addr8[12] = 0xff;
1603 			if (in6_setscope(&llsol, ifp, NULL) == 0) {
1604 				IN6_LOOKUP_MULTI(llsol, ifp, in6m);
1605 				if (in6m)
1606 					in6_delmulti(in6m);
1607 			}
1608 		}
1609 		nd6_inuse--;
1610 		rt->rt_llinfo = NULL;
1611 		rt->rt_flags &= ~RTF_LLINFO;
1612 
1613 		/* Have to do before IF_AFDATA_WLOCK to avoid deadlock */
1614 		callout_halt(&ln->la_timer, &ln->lle_lock);
1615 		/* XXX: LOR avoidance. We still have ref on lle. */
1616 		LLE_RUNLOCK(ln);
1617 
1618 		IF_AFDATA_WLOCK(ifp);
1619 		LLE_WLOCK(ln);
1620 
1621 		clear_llinfo_pqueue(ln);
1622 		if (ln->la_opaque != NULL) {
1623 			switch (ifp->if_type) {
1624 #if NTOKEN > 0
1625 			case IFT_ISO88025:
1626 				kmem_free(ln->la_opaque,
1627 				    sizeof(struct token_rif));
1628 				break;
1629 #endif /* NTOKEN > 0 */
1630 			default:
1631 				break;
1632 			}
1633 		}
1634 
1635 		if (ln->la_rt != NULL) {
1636 			/*
1637 			 * Don't rtfree (may actually free objects) here.
1638 			 * Leave it to rtrequest1.
1639 			 */
1640 			ln->la_rt->rt_refcnt--;
1641 			ln->la_rt = NULL;
1642 		}
1643 		/* Guard against race with other llentry_free(). */
1644 		if (ln->la_flags & LLE_LINKED) {
1645 			LLE_REMREF(ln);
1646 			llentry_free(ln);
1647 		} else {
1648 			LLE_FREE_LOCKED(ln);
1649 		}
1650 
1651 		IF_AFDATA_WUNLOCK(ifp);
1652 		ln = NULL;
1653 	}
1654 
1655 	if (ln != NULL) {
1656 		if (flags & LLE_EXCLUSIVE)
1657 			LLE_WUNLOCK(ln);
1658 		else
1659 			LLE_RUNLOCK(ln);
1660 	}
1661 }
1662 
1663 int
1664 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1665 {
1666 	struct in6_drlist *drl = (struct in6_drlist *)data;
1667 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1668 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1669 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1670 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1671 	struct nd_defrouter *dr;
1672 	struct nd_prefix *pr;
1673 	int i = 0, error = 0;
1674 	int s;
1675 
1676 	switch (cmd) {
1677 	case SIOCGDRLST_IN6:
1678 		/*
1679 		 * obsolete API, use sysctl under net.inet6.icmp6
1680 		 */
1681 		memset(drl, 0, sizeof(*drl));
1682 		s = splsoftnet();
1683 		TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
1684 			if (i >= DRLSTSIZ)
1685 				break;
1686 			drl->defrouter[i].rtaddr = dr->rtaddr;
1687 			in6_clearscope(&drl->defrouter[i].rtaddr);
1688 
1689 			drl->defrouter[i].flags = dr->flags;
1690 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1691 			drl->defrouter[i].expire = dr->expire ?
1692 			    time_mono_to_wall(dr->expire) : 0;
1693 			drl->defrouter[i].if_index = dr->ifp->if_index;
1694 			i++;
1695 		}
1696 		splx(s);
1697 		break;
1698 	case SIOCGPRLST_IN6:
1699 		/*
1700 		 * obsolete API, use sysctl under net.inet6.icmp6
1701 		 *
1702 		 * XXX the structure in6_prlist was changed in backward-
1703 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1704 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1705 		 */
1706 		/*
1707 		 * XXX meaning of fields, especialy "raflags", is very
1708 		 * differnet between RA prefix list and RR/static prefix list.
1709 		 * how about separating ioctls into two?
1710 		 */
1711 		memset(oprl, 0, sizeof(*oprl));
1712 		s = splsoftnet();
1713 		LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
1714 			struct nd_pfxrouter *pfr;
1715 			int j;
1716 
1717 			if (i >= PRLSTSIZ)
1718 				break;
1719 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1720 			oprl->prefix[i].raflags = pr->ndpr_raf;
1721 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1722 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1723 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1724 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1725 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1726 				oprl->prefix[i].expire = 0;
1727 			else {
1728 				time_t maxexpire;
1729 
1730 				/* XXX: we assume time_t is signed. */
1731 				maxexpire = (-1) &
1732 				    ~((time_t)1 <<
1733 				    ((sizeof(maxexpire) * 8) - 1));
1734 				if (pr->ndpr_vltime <
1735 				    maxexpire - pr->ndpr_lastupdate) {
1736 					time_t expire;
1737 					expire = pr->ndpr_lastupdate +
1738 					    pr->ndpr_vltime;
1739 					oprl->prefix[i].expire = expire ?
1740 					    time_mono_to_wall(expire) : 0;
1741 				} else
1742 					oprl->prefix[i].expire = maxexpire;
1743 			}
1744 
1745 			j = 0;
1746 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1747 				if (j < DRLSTSIZ) {
1748 #define RTRADDR oprl->prefix[i].advrtr[j]
1749 					RTRADDR = pfr->router->rtaddr;
1750 					in6_clearscope(&RTRADDR);
1751 #undef RTRADDR
1752 				}
1753 				j++;
1754 			}
1755 			oprl->prefix[i].advrtrs = j;
1756 			oprl->prefix[i].origin = PR_ORIG_RA;
1757 
1758 			i++;
1759 		}
1760 		splx(s);
1761 
1762 		break;
1763 	case OSIOCGIFINFO_IN6:
1764 #define ND	ndi->ndi
1765 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1766 		memset(&ND, 0, sizeof(ND));
1767 		ND.linkmtu = IN6_LINKMTU(ifp);
1768 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1769 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1770 		ND.reachable = ND_IFINFO(ifp)->reachable;
1771 		ND.retrans = ND_IFINFO(ifp)->retrans;
1772 		ND.flags = ND_IFINFO(ifp)->flags;
1773 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1774 		ND.chlim = ND_IFINFO(ifp)->chlim;
1775 		break;
1776 	case SIOCGIFINFO_IN6:
1777 		ND = *ND_IFINFO(ifp);
1778 		break;
1779 	case SIOCSIFINFO_IN6:
1780 		/*
1781 		 * used to change host variables from userland.
1782 		 * intented for a use on router to reflect RA configurations.
1783 		 */
1784 		/* 0 means 'unspecified' */
1785 		if (ND.linkmtu != 0) {
1786 			if (ND.linkmtu < IPV6_MMTU ||
1787 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1788 				error = EINVAL;
1789 				break;
1790 			}
1791 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1792 		}
1793 
1794 		if (ND.basereachable != 0) {
1795 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1796 
1797 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1798 			if (ND.basereachable != obasereachable)
1799 				ND_IFINFO(ifp)->reachable =
1800 				    ND_COMPUTE_RTIME(ND.basereachable);
1801 		}
1802 		if (ND.retrans != 0)
1803 			ND_IFINFO(ifp)->retrans = ND.retrans;
1804 		if (ND.chlim != 0)
1805 			ND_IFINFO(ifp)->chlim = ND.chlim;
1806 		/* FALLTHROUGH */
1807 	case SIOCSIFINFO_FLAGS:
1808 	{
1809 		struct ifaddr *ifa;
1810 		struct in6_ifaddr *ia;
1811 
1812 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1813 		    !(ND.flags & ND6_IFF_IFDISABLED))
1814 		{
1815 			/*
1816 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1817 			 * has a link-local address with IN6_IFF_DUPLICATED,
1818 			 * do not clear ND6_IFF_IFDISABLED.
1819 			 * See RFC 4862, section 5.4.5.
1820 			 */
1821 			int duplicated_linklocal = 0;
1822 
1823 			IFADDR_FOREACH(ifa, ifp) {
1824 				if (ifa->ifa_addr->sa_family != AF_INET6)
1825 					continue;
1826 				ia = (struct in6_ifaddr *)ifa;
1827 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1828 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1829 				{
1830 					duplicated_linklocal = 1;
1831 					break;
1832 				}
1833 			}
1834 
1835 			if (duplicated_linklocal) {
1836 				ND.flags |= ND6_IFF_IFDISABLED;
1837 				log(LOG_ERR, "Cannot enable an interface"
1838 				    " with a link-local address marked"
1839 				    " duplicate.\n");
1840 			} else {
1841 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1842 				if (ifp->if_flags & IFF_UP)
1843 					in6_if_up(ifp);
1844 			}
1845 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1846 		    (ND.flags & ND6_IFF_IFDISABLED))
1847 		{
1848 			/* Mark all IPv6 addresses as tentative. */
1849 
1850 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1851 			IFADDR_FOREACH(ifa, ifp) {
1852 				if (ifa->ifa_addr->sa_family != AF_INET6)
1853 					continue;
1854 				nd6_dad_stop(ifa);
1855 				ia = (struct in6_ifaddr *)ifa;
1856 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1857 			}
1858 		}
1859 
1860 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1861 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1862 				/* auto_linklocal 0->1 transition */
1863 
1864 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1865 				in6_ifattach(ifp, NULL);
1866 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1867 			    ifp->if_flags & IFF_UP)
1868 			{
1869 				/*
1870 				 * When the IF already has
1871 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1872 				 * address is assigned, and IFF_UP, try to
1873 				 * assign one.
1874 				 */
1875 				 int haslinklocal = 0;
1876 
1877 				 IFADDR_FOREACH(ifa, ifp) {
1878 					if (ifa->ifa_addr->sa_family !=AF_INET6)
1879 						continue;
1880 					ia = (struct in6_ifaddr *)ifa;
1881 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1882 						haslinklocal = 1;
1883 						break;
1884 					}
1885 				 }
1886 				 if (!haslinklocal)
1887 					in6_ifattach(ifp, NULL);
1888 			}
1889 		}
1890 	}
1891 		ND_IFINFO(ifp)->flags = ND.flags;
1892 		break;
1893 #undef ND
1894 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1895 		/* sync kernel routing table with the default router list */
1896 		defrouter_reset();
1897 		defrouter_select();
1898 		break;
1899 	case SIOCSPFXFLUSH_IN6:
1900 	{
1901 		/* flush all the prefix advertised by routers */
1902 		struct nd_prefix *pfx, *next;
1903 
1904 		s = splsoftnet();
1905 		LIST_FOREACH_SAFE(pfx, &nd_prefix, ndpr_entry, next) {
1906 			struct in6_ifaddr *ia, *ia_next;
1907 
1908 			if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr))
1909 				continue; /* XXX */
1910 
1911 			/* do we really have to remove addresses as well? */
1912 			for (ia = in6_ifaddr; ia; ia = ia_next) {
1913 				/* ia might be removed.  keep the next ptr. */
1914 				ia_next = ia->ia_next;
1915 
1916 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1917 					continue;
1918 
1919 				if (ia->ia6_ndpr == pfx)
1920 					in6_purgeaddr(&ia->ia_ifa);
1921 			}
1922 			prelist_remove(pfx);
1923 		}
1924 		splx(s);
1925 		break;
1926 	}
1927 	case SIOCSRTRFLUSH_IN6:
1928 	{
1929 		/* flush all the default routers */
1930 		struct nd_defrouter *drtr, *next;
1931 
1932 		s = splsoftnet();
1933 		defrouter_reset();
1934 		TAILQ_FOREACH_SAFE(drtr, &nd_defrouter, dr_entry, next) {
1935 			defrtrlist_del(drtr, NULL);
1936 		}
1937 		defrouter_select();
1938 		splx(s);
1939 		break;
1940 	}
1941 	case SIOCGNBRINFO_IN6:
1942 	{
1943 		struct llentry *ln;
1944 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1945 		struct rtentry *rt;
1946 
1947 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1948 			return error;
1949 
1950 		s = splsoftnet();
1951 		rt = nd6_lookup(&nb_addr, 0, ifp);
1952 		if (rt == NULL) {
1953 			error = EINVAL;
1954 			splx(s);
1955 			break;
1956 		}
1957 
1958 		ln = rt->rt_llinfo;
1959 		rtfree(rt);
1960 		if (ln == NULL) {
1961 			error = EINVAL;
1962 			splx(s);
1963 			break;
1964 		}
1965 		nbi->state = ln->ln_state;
1966 		nbi->asked = ln->ln_asked;
1967 		nbi->isrouter = ln->ln_router;
1968 		nbi->expire = ln->ln_expire ?
1969 		    time_mono_to_wall(ln->ln_expire) : 0;
1970 		splx(s);
1971 
1972 		break;
1973 	}
1974 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1975 		ndif->ifindex = nd6_defifindex;
1976 		break;
1977 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1978 		return nd6_setdefaultiface(ndif->ifindex);
1979 	}
1980 	return error;
1981 }
1982 
1983 void
1984 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp,
1985     struct rtentry *rt)
1986 {
1987 	struct mbuf *m_hold, *m_hold_next;
1988 
1989 	for (m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1990 	     m_hold != NULL;
1991 	     m_hold = m_hold_next) {
1992 		m_hold_next = m_hold->m_nextpkt;
1993 		m_hold->m_nextpkt = NULL;
1994 
1995 		/*
1996 		 * we assume ifp is not a p2p here, so
1997 		 * just set the 2nd argument as the
1998 		 * 1st one.
1999 		 */
2000 		nd6_output(ifp, ifp, m_hold, satocsin6(rt_getkey(rt)), rt);
2001 	}
2002 }
2003 
2004 /*
2005  * Create neighbor cache entry and cache link-layer address,
2006  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
2007  */
2008 void
2009 nd6_cache_lladdr(
2010     struct ifnet *ifp,
2011     struct in6_addr *from,
2012     char *lladdr,
2013     int lladdrlen,
2014     int type,	/* ICMP6 type */
2015     int code	/* type dependent information */
2016 )
2017 {
2018 	struct nd_ifinfo *ndi = ND_IFINFO(ifp);
2019 	struct rtentry *rt = NULL;
2020 	struct llentry *ln = NULL;
2021 	int is_newentry;
2022 	struct sockaddr_dl *sdl = NULL;
2023 	int do_update;
2024 	int olladdr;
2025 	int llchange;
2026 	int newstate = 0;
2027 
2028 	KASSERT(ifp != NULL);
2029 	KASSERT(from != NULL);
2030 
2031 	/* nothing must be updated for unspecified address */
2032 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2033 		return;
2034 
2035 	/*
2036 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2037 	 * the caller.
2038 	 *
2039 	 * XXX If the link does not have link-layer adderss, what should
2040 	 * we do? (ifp->if_addrlen == 0)
2041 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2042 	 * description on it in NS section (RFC 2461 7.2.3).
2043 	 */
2044 
2045 	rt = nd6_lookup(from, 0, ifp);
2046 	if (rt == NULL) {
2047 #if 0
2048 		/* nothing must be done if there's no lladdr */
2049 		if (!lladdr || !lladdrlen)
2050 			return NULL;
2051 #endif
2052 
2053 		rt = nd6_lookup(from, 1, ifp);
2054 		is_newentry = 1;
2055 	} else {
2056 		/* do nothing if static ndp is set */
2057 		if (rt->rt_flags & RTF_STATIC) {
2058 			rtfree(rt);
2059 			return;
2060 		}
2061 		is_newentry = 0;
2062 	}
2063 
2064 	if (rt == NULL)
2065 		return;
2066 	if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) {
2067 fail:
2068 		if (rt->rt_llinfo != NULL)
2069 			LLE_WLOCK((struct llentry *)rt->rt_llinfo);
2070 		nd6_free(rt, rt->rt_llinfo, 0);
2071 		rtfree(rt);
2072 		return;
2073 	}
2074 	ln = rt->rt_llinfo;
2075 	if (ln == NULL)
2076 		goto fail;
2077 	if (rt->rt_gateway == NULL)
2078 		goto fail;
2079 	if (rt->rt_gateway->sa_family != AF_LINK)
2080 		goto fail;
2081 	sdl = satosdl(rt->rt_gateway);
2082 
2083 	olladdr = (sdl->sdl_alen) ? 1 : 0;
2084 	if (olladdr && lladdr) {
2085 		if (memcmp(lladdr, CLLADDR(sdl), ifp->if_addrlen))
2086 			llchange = 1;
2087 		else
2088 			llchange = 0;
2089 	} else
2090 		llchange = 0;
2091 
2092 	/*
2093 	 * newentry olladdr  lladdr  llchange	(*=record)
2094 	 *	0	n	n	--	(1)
2095 	 *	0	y	n	--	(2)
2096 	 *	0	n	y	--	(3) * STALE
2097 	 *	0	y	y	n	(4) *
2098 	 *	0	y	y	y	(5) * STALE
2099 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2100 	 *	1	--	y	--	(7) * STALE
2101 	 */
2102 
2103 	if (lladdr) {		/* (3-5) and (7) */
2104 		/*
2105 		 * Record source link-layer address
2106 		 * XXX is it dependent to ifp->if_type?
2107 		 */
2108 		/* XXX check for error */
2109 		if (sockaddr_dl_setaddr(sdl, sdl->sdl_len, lladdr,
2110 		    ifp->if_addrlen) == NULL) {
2111 			printf("%s.%d: sockaddr_dl_setaddr(, %d, ) "
2112 			    "failed on %s\n", __func__, __LINE__,
2113 			    sdl->sdl_len, if_name(ifp));
2114 		}
2115 	}
2116 
2117 	if (!is_newentry) {
2118 		if ((!olladdr && lladdr) ||		/* (3) */
2119 		    (olladdr && lladdr && llchange)) {	/* (5) */
2120 			do_update = 1;
2121 			newstate = ND6_LLINFO_STALE;
2122 		} else					/* (1-2,4) */
2123 			do_update = 0;
2124 	} else {
2125 		do_update = 1;
2126 		if (lladdr == NULL)			/* (6) */
2127 			newstate = ND6_LLINFO_NOSTATE;
2128 		else					/* (7) */
2129 			newstate = ND6_LLINFO_STALE;
2130 	}
2131 
2132 	if (do_update) {
2133 		/*
2134 		 * Update the state of the neighbor cache.
2135 		 */
2136 		ln->ln_state = newstate;
2137 
2138 		if (ln->ln_state == ND6_LLINFO_STALE) {
2139 			/*
2140 			 * XXX: since nd6_output() below will cause
2141 			 * state tansition to DELAY and reset the timer,
2142 			 * we must set the timer now, although it is actually
2143 			 * meaningless.
2144 			 */
2145 			nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2146 
2147 			nd6_llinfo_release_pkts(ln, ifp, rt);
2148 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
2149 			/* probe right away */
2150 			nd6_llinfo_settimer((void *)ln, 0);
2151 		}
2152 	}
2153 
2154 	/*
2155 	 * ICMP6 type dependent behavior.
2156 	 *
2157 	 * NS: clear IsRouter if new entry
2158 	 * RS: clear IsRouter
2159 	 * RA: set IsRouter if there's lladdr
2160 	 * redir: clear IsRouter if new entry
2161 	 *
2162 	 * RA case, (1):
2163 	 * The spec says that we must set IsRouter in the following cases:
2164 	 * - If lladdr exist, set IsRouter.  This means (1-5).
2165 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
2166 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
2167 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
2168 	 * neighbor cache, this is similar to (6).
2169 	 * This case is rare but we figured that we MUST NOT set IsRouter.
2170 	 *
2171 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
2172 	 *							D R
2173 	 *	0	n	n	--	(1)	c   ?     s
2174 	 *	0	y	n	--	(2)	c   s     s
2175 	 *	0	n	y	--	(3)	c   s     s
2176 	 *	0	y	y	n	(4)	c   s     s
2177 	 *	0	y	y	y	(5)	c   s     s
2178 	 *	1	--	n	--	(6) c	c 	c s
2179 	 *	1	--	y	--	(7) c	c   s	c s
2180 	 *
2181 	 *					(c=clear s=set)
2182 	 */
2183 	switch (type & 0xff) {
2184 	case ND_NEIGHBOR_SOLICIT:
2185 		/*
2186 		 * New entry must have is_router flag cleared.
2187 		 */
2188 		if (is_newentry)	/* (6-7) */
2189 			ln->ln_router = 0;
2190 		break;
2191 	case ND_REDIRECT:
2192 		/*
2193 		 * If the icmp is a redirect to a better router, always set the
2194 		 * is_router flag.  Otherwise, if the entry is newly created,
2195 		 * clear the flag.  [RFC 2461, sec 8.3]
2196 		 */
2197 		if (code == ND_REDIRECT_ROUTER)
2198 			ln->ln_router = 1;
2199 		else if (is_newentry) /* (6-7) */
2200 			ln->ln_router = 0;
2201 		break;
2202 	case ND_ROUTER_SOLICIT:
2203 		/*
2204 		 * is_router flag must always be cleared.
2205 		 */
2206 		ln->ln_router = 0;
2207 		break;
2208 	case ND_ROUTER_ADVERT:
2209 		/*
2210 		 * Mark an entry with lladdr as a router.
2211 		 */
2212 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
2213 		    (is_newentry && lladdr)) {			/* (7) */
2214 			ln->ln_router = 1;
2215 		}
2216 		break;
2217 	}
2218 
2219 	if (do_update)
2220 		rt_newmsg(RTM_CHANGE, rt);  /* tell user process */
2221 
2222 	/*
2223 	 * When the link-layer address of a router changes, select the
2224 	 * best router again.  In particular, when the neighbor entry is newly
2225 	 * created, it might affect the selection policy.
2226 	 * Question: can we restrict the first condition to the "is_newentry"
2227 	 * case?
2228 	 * XXX: when we hear an RA from a new router with the link-layer
2229 	 * address option, defrouter_select() is called twice, since
2230 	 * defrtrlist_update called the function as well.  However, I believe
2231 	 * we can compromise the overhead, since it only happens the first
2232 	 * time.
2233 	 * XXX: although defrouter_select() should not have a bad effect
2234 	 * for those are not autoconfigured hosts, we explicitly avoid such
2235 	 * cases for safety.
2236 	 */
2237 	if (do_update && ln->ln_router && !ip6_forwarding &&
2238 	    nd6_accepts_rtadv(ndi))
2239 		defrouter_select();
2240 
2241 	rtfree(rt);
2242 }
2243 
2244 static void
2245 nd6_slowtimo(void *ignored_arg)
2246 {
2247 	struct nd_ifinfo *nd6if;
2248 	struct ifnet *ifp;
2249 
2250 	mutex_enter(softnet_lock);
2251 	KERNEL_LOCK(1, NULL);
2252 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2253 	    nd6_slowtimo, NULL);
2254 	IFNET_FOREACH(ifp) {
2255 		nd6if = ND_IFINFO(ifp);
2256 		if (nd6if->basereachable && /* already initialized */
2257 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2258 			/*
2259 			 * Since reachable time rarely changes by router
2260 			 * advertisements, we SHOULD insure that a new random
2261 			 * value gets recomputed at least once every few hours.
2262 			 * (RFC 2461, 6.3.4)
2263 			 */
2264 			nd6if->recalctm = nd6_recalc_reachtm_interval;
2265 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2266 		}
2267 	}
2268 	KERNEL_UNLOCK_ONE(NULL);
2269 	mutex_exit(softnet_lock);
2270 }
2271 
2272 /*
2273  * Next hop determination.  This routine was derived from ether_output.
2274  */
2275 static int
2276 nd6_determine_nexthop(struct ifnet *ifp, const struct sockaddr_in6 *dst,
2277     struct rtentry *rt00, struct rtentry **ret_rt, bool *sendpkt)
2278 {
2279 	struct rtentry *rt, *rt0;
2280 	struct rtentry *gwrt;
2281 	struct sockaddr_in6 *gw6;
2282 
2283 #define RTFREE_IF_NEEDED(_rt) \
2284 	if ((_rt) != NULL && (_rt) != rt00) \
2285 		rtfree((_rt));
2286 
2287 	KASSERT(rt00 != NULL);
2288 
2289 	rt = rt0 = rt00;
2290 
2291 	if ((rt->rt_flags & RTF_UP) == 0) {
2292 		rt0 = rt = rtalloc1(sin6tocsa(dst), 1);
2293 		if (rt == NULL)
2294 			goto hostunreach;
2295 		if (rt->rt_ifp != ifp)
2296 			goto hostunreach;
2297 	}
2298 
2299 	if ((rt->rt_flags & RTF_GATEWAY) == 0)
2300 		goto out;
2301 
2302 	gw6 = (struct sockaddr_in6 *)rt->rt_gateway;
2303 
2304 	/*
2305 	 * We skip link-layer address resolution and NUD
2306 	 * if the gateway is not a neighbor from ND point
2307 	 * of view, regardless of the value of nd_ifinfo.flags.
2308 	 * The second condition is a bit tricky; we skip
2309 	 * if the gateway is our own address, which is
2310 	 * sometimes used to install a route to a p2p link.
2311 	 */
2312 	if (!nd6_is_addr_neighbor(gw6, ifp) ||
2313 	    in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) {
2314 		/*
2315 		 * We allow this kind of tricky route only
2316 		 * when the outgoing interface is p2p.
2317 		 * XXX: we may need a more generic rule here.
2318 		 */
2319 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
2320 			goto hostunreach;
2321 
2322 		*sendpkt = true;
2323 		goto out;
2324 	}
2325 
2326 	/* Try to use a cached nexthop route (gwroute) if exists */
2327 	gwrt = rt_get_gwroute(rt);
2328 	if (gwrt == NULL)
2329 		goto lookup;
2330 
2331 	RTFREE_IF_NEEDED(rt);
2332 	rt = gwrt;
2333 	if ((rt->rt_flags & RTF_UP) == 0) {
2334 		RTFREE_IF_NEEDED(rt);
2335 		rt = rt0;
2336 	lookup:
2337 		/* Look up a nexthop route */
2338 		gwrt = rtalloc1(rt->rt_gateway, 1);
2339 		rt_set_gwroute(rt, gwrt);
2340 		RTFREE_IF_NEEDED(rt);
2341 		rt = gwrt;
2342 		if (rt == NULL)
2343 			goto hostunreach;
2344 		/* the "G" test below also prevents rt == rt0 */
2345 		if ((rt->rt_flags & RTF_GATEWAY) ||
2346 		    (rt->rt_ifp != ifp)) {
2347 			if (rt0->rt_gwroute != NULL)
2348 				rtfree(rt0->rt_gwroute);
2349 			rt0->rt_gwroute = NULL;
2350 			goto hostunreach;
2351 		}
2352 	}
2353 
2354 out:
2355 	*ret_rt = rt;
2356 	return 0;
2357 
2358 hostunreach:
2359 	RTFREE_IF_NEEDED(rt);
2360 
2361 	return EHOSTUNREACH;
2362 #undef RTFREE_IF_NEEDED
2363 }
2364 
2365 #define senderr(e) { error = (e); goto bad;}
2366 int
2367 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
2368     const struct sockaddr_in6 *dst, struct rtentry *rt0)
2369 {
2370 	struct mbuf *m = m0;
2371 	struct rtentry *rt = rt0;
2372 	struct llentry *ln = NULL;
2373 	int error = 0;
2374 
2375 #define RTFREE_IF_NEEDED(_rt) \
2376 	if ((_rt) != NULL && (_rt) != rt0) \
2377 		rtfree((_rt));
2378 
2379 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
2380 		goto sendpkt;
2381 
2382 	if (nd6_need_cache(ifp) == 0)
2383 		goto sendpkt;
2384 
2385 	if (rt) {
2386 		struct rtentry *nexthop = NULL;
2387 		bool sendpkt = false;
2388 
2389 		error = nd6_determine_nexthop(ifp, dst, rt, &nexthop, &sendpkt);
2390 		if (error != 0)
2391 			senderr(error);
2392 		rt = nexthop;
2393 		if (sendpkt)
2394 			goto sendpkt;
2395 	}
2396 
2397 	/*
2398 	 * Address resolution or Neighbor Unreachability Detection
2399 	 * for the next hop.
2400 	 * At this point, the destination of the packet must be a unicast
2401 	 * or an anycast address(i.e. not a multicast).
2402 	 */
2403 
2404 	/* Look up the neighbor cache for the nexthop */
2405 	if (rt != NULL && (rt->rt_flags & RTF_LLINFO) != 0)
2406 		ln = rt->rt_llinfo;
2407 	else {
2408 		/*
2409 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2410 		 * the condition below is not very efficient.  But we believe
2411 		 * it is tolerable, because this should be a rare case.
2412 		 */
2413 		if (nd6_is_addr_neighbor(dst, ifp)) {
2414 			RTFREE_IF_NEEDED(rt);
2415 			rt = nd6_lookup(&dst->sin6_addr, 1, ifp);
2416 			if (rt != NULL)
2417 				ln = rt->rt_llinfo;
2418 		}
2419 	}
2420 	if (ln == NULL || rt == NULL) {
2421 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
2422 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2423 			log(LOG_DEBUG,
2424 			    "nd6_output: can't allocate llinfo for %s "
2425 			    "(ln=%p, rt=%p)\n",
2426 			    ip6_sprintf(&dst->sin6_addr), ln, rt);
2427 			senderr(EIO);	/* XXX: good error? */
2428 		}
2429 
2430 		goto sendpkt;	/* send anyway */
2431 	}
2432 
2433 	/* We don't have to do link-layer address resolution on a p2p link. */
2434 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
2435 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
2436 		ln->ln_state = ND6_LLINFO_STALE;
2437 		nd6_llinfo_settimer(ln, nd6_gctimer * hz);
2438 	}
2439 
2440 	/*
2441 	 * The first time we send a packet to a neighbor whose entry is
2442 	 * STALE, we have to change the state to DELAY and a sets a timer to
2443 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2444 	 * neighbor unreachability detection on expiration.
2445 	 * (RFC 2461 7.3.3)
2446 	 */
2447 	if (ln->ln_state == ND6_LLINFO_STALE) {
2448 		ln->ln_asked = 0;
2449 		ln->ln_state = ND6_LLINFO_DELAY;
2450 		nd6_llinfo_settimer(ln, nd6_delay * hz);
2451 	}
2452 
2453 	/*
2454 	 * If the neighbor cache entry has a state other than INCOMPLETE
2455 	 * (i.e. its link-layer address is already resolved), just
2456 	 * send the packet.
2457 	 */
2458 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
2459 		goto sendpkt;
2460 
2461 	/*
2462 	 * There is a neighbor cache entry, but no ethernet address
2463 	 * response yet.  Append this latest packet to the end of the
2464 	 * packet queue in the mbuf, unless the number of the packet
2465 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2466 	 * the oldest packet in the queue will be removed.
2467 	 */
2468 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
2469 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
2470 	if (ln->ln_hold) {
2471 		struct mbuf *m_hold;
2472 		int i;
2473 
2474 		i = 0;
2475 		for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) {
2476 			i++;
2477 			if (m_hold->m_nextpkt == NULL) {
2478 				m_hold->m_nextpkt = m;
2479 				break;
2480 			}
2481 		}
2482 		while (i >= nd6_maxqueuelen) {
2483 			m_hold = ln->ln_hold;
2484 			ln->ln_hold = ln->ln_hold->m_nextpkt;
2485 			m_freem(m_hold);
2486 			i--;
2487 		}
2488 	} else {
2489 		ln->ln_hold = m;
2490 	}
2491 
2492 	/*
2493 	 * If there has been no NS for the neighbor after entering the
2494 	 * INCOMPLETE state, send the first solicitation.
2495 	 */
2496 	if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) {
2497 		struct in6_addr src, *psrc;
2498 
2499 		ln->ln_asked++;
2500 		nd6_llinfo_settimer(ln,
2501 		    ND_IFINFO(ifp)->retrans * hz / 1000);
2502 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
2503 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, psrc, 0);
2504 	}
2505 	error = 0;
2506 	goto exit;
2507 
2508   sendpkt:
2509 	/* discard the packet if IPv6 operation is disabled on the interface */
2510 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2511 		error = ENETDOWN; /* better error? */
2512 		goto bad;
2513 	}
2514 
2515 #ifndef NET_MPSAFE
2516 	KERNEL_LOCK(1, NULL);
2517 #endif
2518 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2519 		error = (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt);
2520 	else
2521 		error = (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt);
2522 #ifndef NET_MPSAFE
2523 	KERNEL_UNLOCK_ONE(NULL);
2524 #endif
2525 	goto exit;
2526 
2527   bad:
2528 	if (m != NULL)
2529 		m_freem(m);
2530   exit:
2531 	RTFREE_IF_NEEDED(rt);
2532 
2533 	return error;
2534 #undef RTFREE_IF_NEEDED
2535 }
2536 #undef senderr
2537 
2538 int
2539 nd6_need_cache(struct ifnet *ifp)
2540 {
2541 	/*
2542 	 * XXX: we currently do not make neighbor cache on any interface
2543 	 * other than ARCnet, Ethernet, FDDI and GIF.
2544 	 *
2545 	 * RFC2893 says:
2546 	 * - unidirectional tunnels needs no ND
2547 	 */
2548 	switch (ifp->if_type) {
2549 	case IFT_ARCNET:
2550 	case IFT_ETHER:
2551 	case IFT_FDDI:
2552 	case IFT_IEEE1394:
2553 	case IFT_CARP:
2554 	case IFT_GIF:		/* XXX need more cases? */
2555 	case IFT_PPP:
2556 	case IFT_TUNNEL:
2557 		return 1;
2558 	default:
2559 		return 0;
2560 	}
2561 }
2562 
2563 int
2564 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt,
2565     struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst,
2566     size_t dstsize)
2567 {
2568 	const struct sockaddr_dl *sdl;
2569 
2570 	if (m->m_flags & M_MCAST) {
2571 		switch (ifp->if_type) {
2572 		case IFT_ETHER:
2573 		case IFT_FDDI:
2574 			ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr,
2575 			    lldst);
2576 			return 1;
2577 		case IFT_IEEE1394:
2578 			memcpy(lldst, ifp->if_broadcastaddr,
2579 			    MIN(dstsize, ifp->if_addrlen));
2580 			return 1;
2581 		case IFT_ARCNET:
2582 			*lldst = 0;
2583 			return 1;
2584 		default:
2585 			m_freem(m);
2586 			return 0;
2587 		}
2588 	}
2589 
2590 	if (rt == NULL) {
2591 		/* this could happen, if we could not allocate memory */
2592 		m_freem(m);
2593 		return 0;
2594 	}
2595 	if (rt->rt_gateway->sa_family != AF_LINK) {
2596 		char gbuf[256];
2597 		char dbuf[LINK_ADDRSTRLEN];
2598 		sockaddr_format(rt->rt_gateway, gbuf, sizeof(gbuf));
2599 		printf("%s: bad gateway address type %s for dst %s"
2600 		    " through interface %s\n", __func__, gbuf,
2601 		    IN6_PRINT(dbuf, &satocsin6(dst)->sin6_addr),
2602 		    if_name(ifp));
2603 		m_freem(m);
2604 		return 0;
2605 	}
2606 	sdl = satocsdl(rt->rt_gateway);
2607 	if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) {
2608 		char sbuf[INET6_ADDRSTRLEN];
2609 		char dbuf[LINK_ADDRSTRLEN];
2610 		/* this should be impossible, but we bark here for debugging */
2611 		printf("%s: sdl_alen == %" PRIu8 ", if=%s, dst=%s, sdl=%s\n",
2612 		    __func__, sdl->sdl_alen, if_name(ifp),
2613 		    IN6_PRINT(sbuf, &satocsin6(dst)->sin6_addr),
2614 		    DL_PRINT(dbuf, &sdl->sdl_addr));
2615 		m_freem(m);
2616 		return 0;
2617 	}
2618 
2619 	memcpy(lldst, CLLADDR(sdl), MIN(dstsize, sdl->sdl_alen));
2620 	return 1;
2621 }
2622 
2623 static void
2624 clear_llinfo_pqueue(struct llentry *ln)
2625 {
2626 	struct mbuf *m_hold, *m_hold_next;
2627 
2628 	for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) {
2629 		m_hold_next = m_hold->m_nextpkt;
2630 		m_hold->m_nextpkt = NULL;
2631 		m_freem(m_hold);
2632 	}
2633 
2634 	ln->ln_hold = NULL;
2635 	return;
2636 }
2637 
2638 int
2639 nd6_sysctl(
2640     int name,
2641     void *oldp,	/* syscall arg, need copyout */
2642     size_t *oldlenp,
2643     void *newp,	/* syscall arg, need copyin */
2644     size_t newlen
2645 )
2646 {
2647 	void *p;
2648 	size_t ol;
2649 	int error;
2650 
2651 	error = 0;
2652 
2653 	if (newp)
2654 		return EPERM;
2655 	if (oldp && !oldlenp)
2656 		return EINVAL;
2657 	ol = oldlenp ? *oldlenp : 0;
2658 
2659 	if (oldp) {
2660 		p = malloc(*oldlenp, M_TEMP, M_WAITOK);
2661 		if (p == NULL)
2662 			return ENOMEM;
2663 	} else
2664 		p = NULL;
2665 	switch (name) {
2666 	case ICMPV6CTL_ND6_DRLIST:
2667 		error = fill_drlist(p, oldlenp, ol);
2668 		if (!error && p != NULL && oldp != NULL)
2669 			error = copyout(p, oldp, *oldlenp);
2670 		break;
2671 
2672 	case ICMPV6CTL_ND6_PRLIST:
2673 		error = fill_prlist(p, oldlenp, ol);
2674 		if (!error && p != NULL && oldp != NULL)
2675 			error = copyout(p, oldp, *oldlenp);
2676 		break;
2677 
2678 	case ICMPV6CTL_ND6_MAXQLEN:
2679 		break;
2680 
2681 	default:
2682 		error = ENOPROTOOPT;
2683 		break;
2684 	}
2685 	if (p)
2686 		free(p, M_TEMP);
2687 
2688 	return error;
2689 }
2690 
2691 static int
2692 fill_drlist(void *oldp, size_t *oldlenp, size_t ol)
2693 {
2694 	int error = 0, s;
2695 	struct in6_defrouter *d = NULL, *de = NULL;
2696 	struct nd_defrouter *dr;
2697 	size_t l;
2698 
2699 	s = splsoftnet();
2700 
2701 	if (oldp) {
2702 		d = (struct in6_defrouter *)oldp;
2703 		de = (struct in6_defrouter *)((char *)oldp + *oldlenp);
2704 	}
2705 	l = 0;
2706 
2707 	TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) {
2708 
2709 		if (oldp && d + 1 <= de) {
2710 			memset(d, 0, sizeof(*d));
2711 			sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0);
2712 			if (sa6_recoverscope(&d->rtaddr)) {
2713 				log(LOG_ERR,
2714 				    "scope error in router list (%s)\n",
2715 				    ip6_sprintf(&d->rtaddr.sin6_addr));
2716 				/* XXX: press on... */
2717 			}
2718 			d->flags = dr->flags;
2719 			d->rtlifetime = dr->rtlifetime;
2720 			d->expire = dr->expire ?
2721 			    time_mono_to_wall(dr->expire) : 0;
2722 			d->if_index = dr->ifp->if_index;
2723 		}
2724 
2725 		l += sizeof(*d);
2726 		if (d)
2727 			d++;
2728 	}
2729 
2730 	if (oldp) {
2731 		if (l > ol)
2732 			error = ENOMEM;
2733 	}
2734 	if (oldlenp)
2735 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2736 
2737 	splx(s);
2738 
2739 	return error;
2740 }
2741 
2742 static int
2743 fill_prlist(void *oldp, size_t *oldlenp, size_t ol)
2744 {
2745 	int error = 0, s;
2746 	struct nd_prefix *pr;
2747 	uint8_t *p = NULL, *ps = NULL;
2748 	uint8_t *pe = NULL;
2749 	size_t l;
2750 
2751 	s = splsoftnet();
2752 
2753 	if (oldp) {
2754 		ps = p = (uint8_t*)oldp;
2755 		pe = (uint8_t*)oldp + *oldlenp;
2756 	}
2757 	l = 0;
2758 
2759 	LIST_FOREACH(pr, &nd_prefix, ndpr_entry) {
2760 		u_short advrtrs;
2761 		struct sockaddr_in6 sin6;
2762 		struct nd_pfxrouter *pfr;
2763 		struct in6_prefix pfx;
2764 
2765 		if (oldp && p + sizeof(struct in6_prefix) <= pe)
2766 		{
2767 			memset(&pfx, 0, sizeof(pfx));
2768 			ps = p;
2769 			pfx.prefix = pr->ndpr_prefix;
2770 
2771 			if (sa6_recoverscope(&pfx.prefix)) {
2772 				log(LOG_ERR,
2773 				    "scope error in prefix list (%s)\n",
2774 				    ip6_sprintf(&pfx.prefix.sin6_addr));
2775 				/* XXX: press on... */
2776 			}
2777 			pfx.raflags = pr->ndpr_raf;
2778 			pfx.prefixlen = pr->ndpr_plen;
2779 			pfx.vltime = pr->ndpr_vltime;
2780 			pfx.pltime = pr->ndpr_pltime;
2781 			pfx.if_index = pr->ndpr_ifp->if_index;
2782 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2783 				pfx.expire = 0;
2784 			else {
2785 				time_t maxexpire;
2786 
2787 				/* XXX: we assume time_t is signed. */
2788 				maxexpire = (-1) &
2789 				    ~((time_t)1 <<
2790 				    ((sizeof(maxexpire) * 8) - 1));
2791 				if (pr->ndpr_vltime <
2792 				    maxexpire - pr->ndpr_lastupdate) {
2793 					pfx.expire = pr->ndpr_lastupdate +
2794 						pr->ndpr_vltime;
2795 				} else
2796 					pfx.expire = maxexpire;
2797 			}
2798 			pfx.refcnt = pr->ndpr_refcnt;
2799 			pfx.flags = pr->ndpr_stateflags;
2800 			pfx.origin = PR_ORIG_RA;
2801 
2802 			p += sizeof(pfx); l += sizeof(pfx);
2803 
2804 			advrtrs = 0;
2805 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2806 				if (p + sizeof(sin6) > pe) {
2807 					advrtrs++;
2808 					continue;
2809 				}
2810 
2811 				sockaddr_in6_init(&sin6, &pfr->router->rtaddr,
2812 				    0, 0, 0);
2813 				if (sa6_recoverscope(&sin6)) {
2814 					log(LOG_ERR,
2815 					    "scope error in "
2816 					    "prefix list (%s)\n",
2817 					    ip6_sprintf(&pfr->router->rtaddr));
2818 				}
2819 				advrtrs++;
2820 				memcpy(p, &sin6, sizeof(sin6));
2821 				p += sizeof(sin6);
2822 				l += sizeof(sin6);
2823 			}
2824 			pfx.advrtrs = advrtrs;
2825 			memcpy(ps, &pfx, sizeof(pfx));
2826 		}
2827 		else {
2828 			l += sizeof(pfx);
2829 			advrtrs = 0;
2830 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2831 				advrtrs++;
2832 				l += sizeof(sin6);
2833 			}
2834 		}
2835 	}
2836 
2837 	if (oldp) {
2838 		*oldlenp = l;	/* (void *)d - (void *)oldp */
2839 		if (l > ol)
2840 			error = ENOMEM;
2841 	} else
2842 		*oldlenp = l;
2843 
2844 	splx(s);
2845 
2846 	return error;
2847 }
2848