xref: /freebsd/sys/netinet6/nd6.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.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/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/if.h>
56 #include <net/if_arc.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/iso88025.h>
60 #include <net/fddi.h>
61 #include <net/route.h>
62 #include <net/vnet.h>
63 
64 #include <netinet/in.h>
65 #include <net/if_llatbl.h>
66 #define	L3_ADDR_SIN6(le)	((struct sockaddr_in6 *) L3_ADDR(le))
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_ifattach.h>
74 #include <netinet/icmp6.h>
75 #include <netinet6/send.h>
76 
77 #include <sys/limits.h>
78 
79 #include <security/mac/mac_framework.h>
80 
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
83 
84 #define SIN6(s) ((struct sockaddr_in6 *)s)
85 
86 /* timer values */
87 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
88 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
89 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
90 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
91 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
92 					 * local traffic */
93 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
94 					 * collection timer */
95 
96 /* preventing too many loops in ND option parsing */
97 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
98 
99 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
100 					 * layer hints */
101 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
102 					 * ND entries */
103 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
104 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
105 
106 #ifdef ND6_DEBUG
107 VNET_DEFINE(int, nd6_debug) = 1;
108 #else
109 VNET_DEFINE(int, nd6_debug) = 0;
110 #endif
111 
112 /* for debugging? */
113 #if 0
114 static int nd6_inuse, nd6_allocated;
115 #endif
116 
117 VNET_DEFINE(struct nd_drhead, nd_defrouter);
118 VNET_DEFINE(struct nd_prhead, nd_prefix);
119 
120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
121 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
122 
123 static struct sockaddr_in6 all1_sa;
124 
125 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
126 
127 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
128 	struct ifnet *));
129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
130 static void nd6_slowtimo(void *);
131 static int regen_tmpaddr(struct in6_ifaddr *);
132 static struct llentry *nd6_free(struct llentry *, int);
133 static void nd6_llinfo_timer(void *);
134 static void clear_llinfo_pqueue(struct llentry *);
135 
136 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
137 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
138 
139 VNET_DEFINE(struct callout, nd6_timer_ch);
140 
141 void
142 nd6_init(void)
143 {
144 	int i;
145 
146 	LIST_INIT(&V_nd_prefix);
147 
148 	all1_sa.sin6_family = AF_INET6;
149 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
150 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
151 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
152 
153 	/* initialization of the default router list */
154 	TAILQ_INIT(&V_nd_defrouter);
155 
156 	/* start timer */
157 	callout_init(&V_nd6_slowtimo_ch, 0);
158 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
159 	    nd6_slowtimo, curvnet);
160 }
161 
162 #ifdef VIMAGE
163 void
164 nd6_destroy()
165 {
166 
167 	callout_drain(&V_nd6_slowtimo_ch);
168 	callout_drain(&V_nd6_timer_ch);
169 }
170 #endif
171 
172 struct nd_ifinfo *
173 nd6_ifattach(struct ifnet *ifp)
174 {
175 	struct nd_ifinfo *nd;
176 
177 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
178 	bzero(nd, sizeof(*nd));
179 
180 	nd->initialized = 1;
181 
182 	nd->chlim = IPV6_DEFHLIM;
183 	nd->basereachable = REACHABLE_TIME;
184 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
185 	nd->retrans = RETRANS_TIMER;
186 
187 	nd->flags = ND6_IFF_PERFORMNUD;
188 
189 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
190 	if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
191 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
192 
193 	/* A loopback interface does not need to accept RTADV. */
194 	if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
195 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
196 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
197 		nd->flags |= ND6_IFF_NO_RADR;
198 
199 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
200 	nd6_setmtu0(ifp, nd);
201 
202 	return nd;
203 }
204 
205 void
206 nd6_ifdetach(struct nd_ifinfo *nd)
207 {
208 
209 	free(nd, M_IP6NDP);
210 }
211 
212 /*
213  * Reset ND level link MTU. This function is called when the physical MTU
214  * changes, which means we might have to adjust the ND level MTU.
215  */
216 void
217 nd6_setmtu(struct ifnet *ifp)
218 {
219 
220 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
221 }
222 
223 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
224 void
225 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
226 {
227 	u_int32_t omaxmtu;
228 
229 	omaxmtu = ndi->maxmtu;
230 
231 	switch (ifp->if_type) {
232 	case IFT_ARCNET:
233 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
234 		break;
235 	case IFT_FDDI:
236 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
237 		break;
238 	case IFT_ISO88025:
239 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
240 		 break;
241 	default:
242 		ndi->maxmtu = ifp->if_mtu;
243 		break;
244 	}
245 
246 	/*
247 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
248 	 * undesirable situation.  We thus notify the operator of the change
249 	 * explicitly.  The check for omaxmtu is necessary to restrict the
250 	 * log to the case of changing the MTU, not initializing it.
251 	 */
252 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
253 		log(LOG_NOTICE, "nd6_setmtu0: "
254 		    "new link MTU on %s (%lu) is too small for IPv6\n",
255 		    if_name(ifp), (unsigned long)ndi->maxmtu);
256 	}
257 
258 	if (ndi->maxmtu > V_in6_maxmtu)
259 		in6_setmaxmtu(); /* check all interfaces just in case */
260 
261 }
262 
263 void
264 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
265 {
266 
267 	bzero(ndopts, sizeof(*ndopts));
268 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
269 	ndopts->nd_opts_last
270 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
271 
272 	if (icmp6len == 0) {
273 		ndopts->nd_opts_done = 1;
274 		ndopts->nd_opts_search = NULL;
275 	}
276 }
277 
278 /*
279  * Take one ND option.
280  */
281 struct nd_opt_hdr *
282 nd6_option(union nd_opts *ndopts)
283 {
284 	struct nd_opt_hdr *nd_opt;
285 	int olen;
286 
287 	if (ndopts == NULL)
288 		panic("ndopts == NULL in nd6_option");
289 	if (ndopts->nd_opts_last == NULL)
290 		panic("uninitialized ndopts in nd6_option");
291 	if (ndopts->nd_opts_search == NULL)
292 		return NULL;
293 	if (ndopts->nd_opts_done)
294 		return NULL;
295 
296 	nd_opt = ndopts->nd_opts_search;
297 
298 	/* make sure nd_opt_len is inside the buffer */
299 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
300 		bzero(ndopts, sizeof(*ndopts));
301 		return NULL;
302 	}
303 
304 	olen = nd_opt->nd_opt_len << 3;
305 	if (olen == 0) {
306 		/*
307 		 * Message validation requires that all included
308 		 * options have a length that is greater than zero.
309 		 */
310 		bzero(ndopts, sizeof(*ndopts));
311 		return NULL;
312 	}
313 
314 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
315 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
316 		/* option overruns the end of buffer, invalid */
317 		bzero(ndopts, sizeof(*ndopts));
318 		return NULL;
319 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
320 		/* reached the end of options chain */
321 		ndopts->nd_opts_done = 1;
322 		ndopts->nd_opts_search = NULL;
323 	}
324 	return nd_opt;
325 }
326 
327 /*
328  * Parse multiple ND options.
329  * This function is much easier to use, for ND routines that do not need
330  * multiple options of the same type.
331  */
332 int
333 nd6_options(union nd_opts *ndopts)
334 {
335 	struct nd_opt_hdr *nd_opt;
336 	int i = 0;
337 
338 	if (ndopts == NULL)
339 		panic("ndopts == NULL in nd6_options");
340 	if (ndopts->nd_opts_last == NULL)
341 		panic("uninitialized ndopts in nd6_options");
342 	if (ndopts->nd_opts_search == NULL)
343 		return 0;
344 
345 	while (1) {
346 		nd_opt = nd6_option(ndopts);
347 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
348 			/*
349 			 * Message validation requires that all included
350 			 * options have a length that is greater than zero.
351 			 */
352 			ICMP6STAT_INC(icp6s_nd_badopt);
353 			bzero(ndopts, sizeof(*ndopts));
354 			return -1;
355 		}
356 
357 		if (nd_opt == NULL)
358 			goto skip1;
359 
360 		switch (nd_opt->nd_opt_type) {
361 		case ND_OPT_SOURCE_LINKADDR:
362 		case ND_OPT_TARGET_LINKADDR:
363 		case ND_OPT_MTU:
364 		case ND_OPT_REDIRECTED_HEADER:
365 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
366 				nd6log((LOG_INFO,
367 				    "duplicated ND6 option found (type=%d)\n",
368 				    nd_opt->nd_opt_type));
369 				/* XXX bark? */
370 			} else {
371 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
372 					= nd_opt;
373 			}
374 			break;
375 		case ND_OPT_PREFIX_INFORMATION:
376 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
377 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
378 					= nd_opt;
379 			}
380 			ndopts->nd_opts_pi_end =
381 				(struct nd_opt_prefix_info *)nd_opt;
382 			break;
383 		default:
384 			/*
385 			 * Unknown options must be silently ignored,
386 			 * to accomodate future extension to the protocol.
387 			 */
388 			nd6log((LOG_DEBUG,
389 			    "nd6_options: unsupported option %d - "
390 			    "option ignored\n", nd_opt->nd_opt_type));
391 		}
392 
393 skip1:
394 		i++;
395 		if (i > V_nd6_maxndopt) {
396 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
397 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
398 			break;
399 		}
400 
401 		if (ndopts->nd_opts_done)
402 			break;
403 	}
404 
405 	return 0;
406 }
407 
408 /*
409  * ND6 timer routine to handle ND6 entries
410  */
411 void
412 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
413 {
414 	int canceled;
415 
416 	LLE_WLOCK_ASSERT(ln);
417 
418 	if (tick < 0) {
419 		ln->la_expire = 0;
420 		ln->ln_ntick = 0;
421 		canceled = callout_stop(&ln->ln_timer_ch);
422 	} else {
423 		ln->la_expire = time_second + tick / hz;
424 		LLE_ADDREF(ln);
425 		if (tick > INT_MAX) {
426 			ln->ln_ntick = tick - INT_MAX;
427 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
428 			    nd6_llinfo_timer, ln);
429 		} else {
430 			ln->ln_ntick = 0;
431 			canceled = callout_reset(&ln->ln_timer_ch, tick,
432 			    nd6_llinfo_timer, ln);
433 		}
434 	}
435 	if (canceled)
436 		LLE_REMREF(ln);
437 }
438 
439 void
440 nd6_llinfo_settimer(struct llentry *ln, long tick)
441 {
442 
443 	LLE_WLOCK(ln);
444 	nd6_llinfo_settimer_locked(ln, tick);
445 	LLE_WUNLOCK(ln);
446 }
447 
448 static void
449 nd6_llinfo_timer(void *arg)
450 {
451 	struct llentry *ln;
452 	struct in6_addr *dst;
453 	struct ifnet *ifp;
454 	struct nd_ifinfo *ndi = NULL;
455 
456 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
457 	ln = (struct llentry *)arg;
458 	LLE_WLOCK_ASSERT(ln);
459 	ifp = ln->lle_tbl->llt_ifp;
460 
461 	CURVNET_SET(ifp->if_vnet);
462 
463 	if (ln->ln_ntick > 0) {
464 		if (ln->ln_ntick > INT_MAX) {
465 			ln->ln_ntick -= INT_MAX;
466 			nd6_llinfo_settimer_locked(ln, INT_MAX);
467 		} else {
468 			ln->ln_ntick = 0;
469 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
470 		}
471 		goto done;
472 	}
473 
474 	ndi = ND_IFINFO(ifp);
475 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
476 	if (ln->la_flags & LLE_STATIC) {
477 		goto done;
478 	}
479 
480 	if (ln->la_flags & LLE_DELETED) {
481 		(void)nd6_free(ln, 0);
482 		ln = NULL;
483 		goto done;
484 	}
485 
486 	switch (ln->ln_state) {
487 	case ND6_LLINFO_INCOMPLETE:
488 		if (ln->la_asked < V_nd6_mmaxtries) {
489 			ln->la_asked++;
490 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
491 			LLE_WUNLOCK(ln);
492 			nd6_ns_output(ifp, NULL, dst, ln, 0);
493 			LLE_WLOCK(ln);
494 		} else {
495 			struct mbuf *m = ln->la_hold;
496 			if (m) {
497 				struct mbuf *m0;
498 
499 				/*
500 				 * assuming every packet in la_hold has the
501 				 * same IP header.  Send error after unlock.
502 				 */
503 				m0 = m->m_nextpkt;
504 				m->m_nextpkt = NULL;
505 				ln->la_hold = m0;
506 				clear_llinfo_pqueue(ln);
507 			}
508 			(void)nd6_free(ln, 0);
509 			ln = NULL;
510 			if (m != NULL)
511 				icmp6_error2(m, ICMP6_DST_UNREACH,
512 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
513 		}
514 		break;
515 	case ND6_LLINFO_REACHABLE:
516 		if (!ND6_LLINFO_PERMANENT(ln)) {
517 			ln->ln_state = ND6_LLINFO_STALE;
518 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
519 		}
520 		break;
521 
522 	case ND6_LLINFO_STALE:
523 		/* Garbage Collection(RFC 2461 5.3) */
524 		if (!ND6_LLINFO_PERMANENT(ln)) {
525 			(void)nd6_free(ln, 1);
526 			ln = NULL;
527 		}
528 		break;
529 
530 	case ND6_LLINFO_DELAY:
531 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
532 			/* We need NUD */
533 			ln->la_asked = 1;
534 			ln->ln_state = ND6_LLINFO_PROBE;
535 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
536 			LLE_WUNLOCK(ln);
537 			nd6_ns_output(ifp, dst, dst, ln, 0);
538 			LLE_WLOCK(ln);
539 		} else {
540 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
541 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
542 		}
543 		break;
544 	case ND6_LLINFO_PROBE:
545 		if (ln->la_asked < V_nd6_umaxtries) {
546 			ln->la_asked++;
547 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
548 			LLE_WUNLOCK(ln);
549 			nd6_ns_output(ifp, dst, dst, ln, 0);
550 			LLE_WLOCK(ln);
551 		} else {
552 			(void)nd6_free(ln, 0);
553 			ln = NULL;
554 		}
555 		break;
556 	default:
557 		panic("%s: paths in a dark night can be confusing: %d",
558 		    __func__, ln->ln_state);
559 	}
560 done:
561 	if (ln != NULL)
562 		LLE_FREE_LOCKED(ln);
563 	CURVNET_RESTORE();
564 }
565 
566 
567 /*
568  * ND6 timer routine to expire default route list and prefix list
569  */
570 void
571 nd6_timer(void *arg)
572 {
573 	CURVNET_SET((struct vnet *) arg);
574 	int s;
575 	struct nd_defrouter *dr, *ndr;
576 	struct nd_prefix *pr, *npr;
577 	struct in6_ifaddr *ia6, *nia6;
578 	struct in6_addrlifetime *lt6;
579 
580 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
581 	    nd6_timer, curvnet);
582 
583 	/* expire default router list */
584 	s = splnet();
585 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
586 		if (dr->expire && dr->expire < time_second)
587 			defrtrlist_del(dr);
588 	}
589 
590 	/*
591 	 * expire interface addresses.
592 	 * in the past the loop was inside prefix expiry processing.
593 	 * However, from a stricter speci-confrmance standpoint, we should
594 	 * rather separate address lifetimes and prefix lifetimes.
595 	 *
596 	 * XXXRW: in6_ifaddrhead locking.
597 	 */
598   addrloop:
599 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
600 		/* check address lifetime */
601 		lt6 = &ia6->ia6_lifetime;
602 		if (IFA6_IS_INVALID(ia6)) {
603 			int regen = 0;
604 
605 			/*
606 			 * If the expiring address is temporary, try
607 			 * regenerating a new one.  This would be useful when
608 			 * we suspended a laptop PC, then turned it on after a
609 			 * period that could invalidate all temporary
610 			 * addresses.  Although we may have to restart the
611 			 * loop (see below), it must be after purging the
612 			 * address.  Otherwise, we'd see an infinite loop of
613 			 * regeneration.
614 			 */
615 			if (V_ip6_use_tempaddr &&
616 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
617 				if (regen_tmpaddr(ia6) == 0)
618 					regen = 1;
619 			}
620 
621 			in6_purgeaddr(&ia6->ia_ifa);
622 
623 			if (regen)
624 				goto addrloop; /* XXX: see below */
625 		} else if (IFA6_IS_DEPRECATED(ia6)) {
626 			int oldflags = ia6->ia6_flags;
627 
628 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
629 
630 			/*
631 			 * If a temporary address has just become deprecated,
632 			 * regenerate a new one if possible.
633 			 */
634 			if (V_ip6_use_tempaddr &&
635 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
636 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
637 
638 				if (regen_tmpaddr(ia6) == 0) {
639 					/*
640 					 * A new temporary address is
641 					 * generated.
642 					 * XXX: this means the address chain
643 					 * has changed while we are still in
644 					 * the loop.  Although the change
645 					 * would not cause disaster (because
646 					 * it's not a deletion, but an
647 					 * addition,) we'd rather restart the
648 					 * loop just for safety.  Or does this
649 					 * significantly reduce performance??
650 					 */
651 					goto addrloop;
652 				}
653 			}
654 		} else {
655 			/*
656 			 * A new RA might have made a deprecated address
657 			 * preferred.
658 			 */
659 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
660 		}
661 	}
662 
663 	/* expire prefix list */
664 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
665 		/*
666 		 * check prefix lifetime.
667 		 * since pltime is just for autoconf, pltime processing for
668 		 * prefix is not necessary.
669 		 */
670 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
671 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
672 
673 			/*
674 			 * address expiration and prefix expiration are
675 			 * separate.  NEVER perform in6_purgeaddr here.
676 			 */
677 			prelist_remove(pr);
678 		}
679 	}
680 	splx(s);
681 	CURVNET_RESTORE();
682 }
683 
684 /*
685  * ia6 - deprecated/invalidated temporary address
686  */
687 static int
688 regen_tmpaddr(struct in6_ifaddr *ia6)
689 {
690 	struct ifaddr *ifa;
691 	struct ifnet *ifp;
692 	struct in6_ifaddr *public_ifa6 = NULL;
693 
694 	ifp = ia6->ia_ifa.ifa_ifp;
695 	IF_ADDR_RLOCK(ifp);
696 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
697 		struct in6_ifaddr *it6;
698 
699 		if (ifa->ifa_addr->sa_family != AF_INET6)
700 			continue;
701 
702 		it6 = (struct in6_ifaddr *)ifa;
703 
704 		/* ignore no autoconf addresses. */
705 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
706 			continue;
707 
708 		/* ignore autoconf addresses with different prefixes. */
709 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
710 			continue;
711 
712 		/*
713 		 * Now we are looking at an autoconf address with the same
714 		 * prefix as ours.  If the address is temporary and is still
715 		 * preferred, do not create another one.  It would be rare, but
716 		 * could happen, for example, when we resume a laptop PC after
717 		 * a long period.
718 		 */
719 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
720 		    !IFA6_IS_DEPRECATED(it6)) {
721 			public_ifa6 = NULL;
722 			break;
723 		}
724 
725 		/*
726 		 * This is a public autoconf address that has the same prefix
727 		 * as ours.  If it is preferred, keep it.  We can't break the
728 		 * loop here, because there may be a still-preferred temporary
729 		 * address with the prefix.
730 		 */
731 		if (!IFA6_IS_DEPRECATED(it6))
732 		    public_ifa6 = it6;
733 
734 		if (public_ifa6 != NULL)
735 			ifa_ref(&public_ifa6->ia_ifa);
736 	}
737 	IF_ADDR_RUNLOCK(ifp);
738 
739 	if (public_ifa6 != NULL) {
740 		int e;
741 
742 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
743 			ifa_free(&public_ifa6->ia_ifa);
744 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
745 			    " tmp addr,errno=%d\n", e);
746 			return (-1);
747 		}
748 		ifa_free(&public_ifa6->ia_ifa);
749 		return (0);
750 	}
751 
752 	return (-1);
753 }
754 
755 /*
756  * Nuke neighbor cache/prefix/default router management table, right before
757  * ifp goes away.
758  */
759 void
760 nd6_purge(struct ifnet *ifp)
761 {
762 	struct nd_defrouter *dr, *ndr;
763 	struct nd_prefix *pr, *npr;
764 
765 	/*
766 	 * Nuke default router list entries toward ifp.
767 	 * We defer removal of default router list entries that is installed
768 	 * in the routing table, in order to keep additional side effects as
769 	 * small as possible.
770 	 */
771 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
772 		if (dr->installed)
773 			continue;
774 
775 		if (dr->ifp == ifp)
776 			defrtrlist_del(dr);
777 	}
778 
779 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
780 		if (!dr->installed)
781 			continue;
782 
783 		if (dr->ifp == ifp)
784 			defrtrlist_del(dr);
785 	}
786 
787 	/* Nuke prefix list entries toward ifp */
788 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
789 		if (pr->ndpr_ifp == ifp) {
790 			/*
791 			 * Because if_detach() does *not* release prefixes
792 			 * while purging addresses the reference count will
793 			 * still be above zero. We therefore reset it to
794 			 * make sure that the prefix really gets purged.
795 			 */
796 			pr->ndpr_refcnt = 0;
797 
798 			/*
799 			 * Previously, pr->ndpr_addr is removed as well,
800 			 * but I strongly believe we don't have to do it.
801 			 * nd6_purge() is only called from in6_ifdetach(),
802 			 * which removes all the associated interface addresses
803 			 * by itself.
804 			 * (jinmei@kame.net 20010129)
805 			 */
806 			prelist_remove(pr);
807 		}
808 	}
809 
810 	/* cancel default outgoing interface setting */
811 	if (V_nd6_defifindex == ifp->if_index)
812 		nd6_setdefaultiface(0);
813 
814 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
815 		/* Refresh default router list. */
816 		defrouter_select();
817 	}
818 
819 	/* XXXXX
820 	 * We do not nuke the neighbor cache entries here any more
821 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
822 	 * nd6_purge() is invoked by in6_ifdetach() which is called
823 	 * from if_detach() where everything gets purged. So let
824 	 * in6_domifdetach() do the actual L2 table purging work.
825 	 */
826 }
827 
828 /*
829  * the caller acquires and releases the lock on the lltbls
830  * Returns the llentry locked
831  */
832 struct llentry *
833 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
834 {
835 	struct sockaddr_in6 sin6;
836 	struct llentry *ln;
837 	int llflags;
838 
839 	bzero(&sin6, sizeof(sin6));
840 	sin6.sin6_len = sizeof(struct sockaddr_in6);
841 	sin6.sin6_family = AF_INET6;
842 	sin6.sin6_addr = *addr6;
843 
844 	IF_AFDATA_LOCK_ASSERT(ifp);
845 
846 	llflags = 0;
847 	if (flags & ND6_CREATE)
848 	    llflags |= LLE_CREATE;
849 	if (flags & ND6_EXCLUSIVE)
850 	    llflags |= LLE_EXCLUSIVE;
851 
852 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
853 	if ((ln != NULL) && (llflags & LLE_CREATE))
854 		ln->ln_state = ND6_LLINFO_NOSTATE;
855 
856 	return (ln);
857 }
858 
859 /*
860  * Test whether a given IPv6 address is a neighbor or not, ignoring
861  * the actual neighbor cache.  The neighbor cache is ignored in order
862  * to not reenter the routing code from within itself.
863  */
864 static int
865 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
866 {
867 	struct nd_prefix *pr;
868 	struct ifaddr *dstaddr;
869 
870 	/*
871 	 * A link-local address is always a neighbor.
872 	 * XXX: a link does not necessarily specify a single interface.
873 	 */
874 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
875 		struct sockaddr_in6 sin6_copy;
876 		u_int32_t zone;
877 
878 		/*
879 		 * We need sin6_copy since sa6_recoverscope() may modify the
880 		 * content (XXX).
881 		 */
882 		sin6_copy = *addr;
883 		if (sa6_recoverscope(&sin6_copy))
884 			return (0); /* XXX: should be impossible */
885 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
886 			return (0);
887 		if (sin6_copy.sin6_scope_id == zone)
888 			return (1);
889 		else
890 			return (0);
891 	}
892 
893 	/*
894 	 * If the address matches one of our addresses,
895 	 * it should be a neighbor.
896 	 * If the address matches one of our on-link prefixes, it should be a
897 	 * neighbor.
898 	 */
899 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
900 		if (pr->ndpr_ifp != ifp)
901 			continue;
902 
903 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
904 			struct rtentry *rt;
905 			rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
906 			if (rt == NULL)
907 				continue;
908 			/*
909 			 * This is the case where multiple interfaces
910 			 * have the same prefix, but only one is installed
911 			 * into the routing table and that prefix entry
912 			 * is not the one being examined here. In the case
913 			 * where RADIX_MPATH is enabled, multiple route
914 			 * entries (of the same rt_key value) will be
915 			 * installed because the interface addresses all
916 			 * differ.
917 			 */
918 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
919 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
920 				RTFREE_LOCKED(rt);
921 				continue;
922 			}
923 			RTFREE_LOCKED(rt);
924 		}
925 
926 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
927 		    &addr->sin6_addr, &pr->ndpr_mask))
928 			return (1);
929 	}
930 
931 	/*
932 	 * If the address is assigned on the node of the other side of
933 	 * a p2p interface, the address should be a neighbor.
934 	 */
935 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
936 	if (dstaddr != NULL) {
937 		if (dstaddr->ifa_ifp == ifp) {
938 			ifa_free(dstaddr);
939 			return (1);
940 		}
941 		ifa_free(dstaddr);
942 	}
943 
944 	/*
945 	 * If the default router list is empty, all addresses are regarded
946 	 * as on-link, and thus, as a neighbor.
947 	 */
948 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
949 	    TAILQ_EMPTY(&V_nd_defrouter) &&
950 	    V_nd6_defifindex == ifp->if_index) {
951 		return (1);
952 	}
953 
954 	return (0);
955 }
956 
957 
958 /*
959  * Detect if a given IPv6 address identifies a neighbor on a given link.
960  * XXX: should take care of the destination of a p2p link?
961  */
962 int
963 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
964 {
965 	struct llentry *lle;
966 	int rc = 0;
967 
968 	IF_AFDATA_UNLOCK_ASSERT(ifp);
969 	if (nd6_is_new_addr_neighbor(addr, ifp))
970 		return (1);
971 
972 	/*
973 	 * Even if the address matches none of our addresses, it might be
974 	 * in the neighbor cache.
975 	 */
976 	IF_AFDATA_LOCK(ifp);
977 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
978 		LLE_RUNLOCK(lle);
979 		rc = 1;
980 	}
981 	IF_AFDATA_UNLOCK(ifp);
982 	return (rc);
983 }
984 
985 /*
986  * Free an nd6 llinfo entry.
987  * Since the function would cause significant changes in the kernel, DO NOT
988  * make it global, unless you have a strong reason for the change, and are sure
989  * that the change is safe.
990  */
991 static struct llentry *
992 nd6_free(struct llentry *ln, int gc)
993 {
994         struct llentry *next;
995 	struct nd_defrouter *dr;
996 	struct ifnet *ifp;
997 
998 	LLE_WLOCK_ASSERT(ln);
999 
1000 	/*
1001 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1002 	 * even though it is not harmful, it was not really necessary.
1003 	 */
1004 
1005 	/* cancel timer */
1006 	nd6_llinfo_settimer_locked(ln, -1);
1007 
1008 	ifp = ln->lle_tbl->llt_ifp;
1009 
1010 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1011 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1012 
1013 		if (dr != NULL && dr->expire &&
1014 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1015 			/*
1016 			 * If the reason for the deletion is just garbage
1017 			 * collection, and the neighbor is an active default
1018 			 * router, do not delete it.  Instead, reset the GC
1019 			 * timer using the router's lifetime.
1020 			 * Simply deleting the entry would affect default
1021 			 * router selection, which is not necessarily a good
1022 			 * thing, especially when we're using router preference
1023 			 * values.
1024 			 * XXX: the check for ln_state would be redundant,
1025 			 *      but we intentionally keep it just in case.
1026 			 */
1027 			if (dr->expire > time_second)
1028 				nd6_llinfo_settimer_locked(ln,
1029 				    (dr->expire - time_second) * hz);
1030 			else
1031 				nd6_llinfo_settimer_locked(ln,
1032 				    (long)V_nd6_gctimer * hz);
1033 
1034 			next = LIST_NEXT(ln, lle_next);
1035 			LLE_REMREF(ln);
1036 			LLE_WUNLOCK(ln);
1037 			return (next);
1038 		}
1039 
1040 		if (dr) {
1041 			/*
1042 			 * Unreachablity of a router might affect the default
1043 			 * router selection and on-link detection of advertised
1044 			 * prefixes.
1045 			 */
1046 
1047 			/*
1048 			 * Temporarily fake the state to choose a new default
1049 			 * router and to perform on-link determination of
1050 			 * prefixes correctly.
1051 			 * Below the state will be set correctly,
1052 			 * or the entry itself will be deleted.
1053 			 */
1054 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1055 		}
1056 
1057 		if (ln->ln_router || dr) {
1058 
1059 			/*
1060 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1061 			 * rnh and for the calls to pfxlist_onlink_check() and
1062 			 * defrouter_select() in the block further down for calls
1063 			 * into nd6_lookup().  We still hold a ref.
1064 			 */
1065 			LLE_WUNLOCK(ln);
1066 
1067 			/*
1068 			 * rt6_flush must be called whether or not the neighbor
1069 			 * is in the Default Router List.
1070 			 * See a corresponding comment in nd6_na_input().
1071 			 */
1072 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1073 		}
1074 
1075 		if (dr) {
1076 			/*
1077 			 * Since defrouter_select() does not affect the
1078 			 * on-link determination and MIP6 needs the check
1079 			 * before the default router selection, we perform
1080 			 * the check now.
1081 			 */
1082 			pfxlist_onlink_check();
1083 
1084 			/*
1085 			 * Refresh default router list.
1086 			 */
1087 			defrouter_select();
1088 		}
1089 
1090 		if (ln->ln_router || dr)
1091 			LLE_WLOCK(ln);
1092 	}
1093 
1094 	/*
1095 	 * Before deleting the entry, remember the next entry as the
1096 	 * return value.  We need this because pfxlist_onlink_check() above
1097 	 * might have freed other entries (particularly the old next entry) as
1098 	 * a side effect (XXX).
1099 	 */
1100 	next = LIST_NEXT(ln, lle_next);
1101 
1102 	/*
1103 	 * Save to unlock. We still hold an extra reference and will not
1104 	 * free(9) in llentry_free() if someone else holds one as well.
1105 	 */
1106 	LLE_WUNLOCK(ln);
1107 	IF_AFDATA_LOCK(ifp);
1108 	LLE_WLOCK(ln);
1109 	LLE_REMREF(ln);
1110 	llentry_free(ln);
1111 	IF_AFDATA_UNLOCK(ifp);
1112 
1113 	return (next);
1114 }
1115 
1116 /*
1117  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1118  *
1119  * XXX cost-effective methods?
1120  */
1121 void
1122 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1123 {
1124 	struct llentry *ln;
1125 	struct ifnet *ifp;
1126 
1127 	if ((dst6 == NULL) || (rt == NULL))
1128 		return;
1129 
1130 	ifp = rt->rt_ifp;
1131 	IF_AFDATA_LOCK(ifp);
1132 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1133 	IF_AFDATA_UNLOCK(ifp);
1134 	if (ln == NULL)
1135 		return;
1136 
1137 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1138 		goto done;
1139 
1140 	/*
1141 	 * if we get upper-layer reachability confirmation many times,
1142 	 * it is possible we have false information.
1143 	 */
1144 	if (!force) {
1145 		ln->ln_byhint++;
1146 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1147 			goto done;
1148 		}
1149 	}
1150 
1151  	ln->ln_state = ND6_LLINFO_REACHABLE;
1152 	if (!ND6_LLINFO_PERMANENT(ln)) {
1153 		nd6_llinfo_settimer_locked(ln,
1154 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1155 	}
1156 done:
1157 	LLE_WUNLOCK(ln);
1158 }
1159 
1160 
1161 /*
1162  * Rejuvenate this function for routing operations related
1163  * processing.
1164  */
1165 void
1166 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1167 {
1168 	struct sockaddr_in6 *gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1169 	struct nd_defrouter *dr;
1170 	struct ifnet *ifp = rt->rt_ifp;
1171 
1172 	RT_LOCK_ASSERT(rt);
1173 
1174 	switch (req) {
1175 	case RTM_ADD:
1176 		break;
1177 
1178 	case RTM_DELETE:
1179 		if (!ifp)
1180 			return;
1181 		/*
1182 		 * Only indirect routes are interesting.
1183 		 */
1184 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1185 			return;
1186 		/*
1187 		 * check for default route
1188 		 */
1189 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1190 				       &SIN6(rt_key(rt))->sin6_addr)) {
1191 
1192 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1193 			if (dr != NULL)
1194 				dr->installed = 0;
1195 		}
1196 		break;
1197 	}
1198 }
1199 
1200 
1201 int
1202 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1203 {
1204 	struct in6_drlist *drl = (struct in6_drlist *)data;
1205 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1206 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1207 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1208 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1209 	struct nd_defrouter *dr;
1210 	struct nd_prefix *pr;
1211 	int i = 0, error = 0;
1212 	int s;
1213 
1214 	switch (cmd) {
1215 	case SIOCGDRLST_IN6:
1216 		/*
1217 		 * obsolete API, use sysctl under net.inet6.icmp6
1218 		 */
1219 		bzero(drl, sizeof(*drl));
1220 		s = splnet();
1221 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1222 			if (i >= DRLSTSIZ)
1223 				break;
1224 			drl->defrouter[i].rtaddr = dr->rtaddr;
1225 			in6_clearscope(&drl->defrouter[i].rtaddr);
1226 
1227 			drl->defrouter[i].flags = dr->flags;
1228 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1229 			drl->defrouter[i].expire = dr->expire;
1230 			drl->defrouter[i].if_index = dr->ifp->if_index;
1231 			i++;
1232 		}
1233 		splx(s);
1234 		break;
1235 	case SIOCGPRLST_IN6:
1236 		/*
1237 		 * obsolete API, use sysctl under net.inet6.icmp6
1238 		 *
1239 		 * XXX the structure in6_prlist was changed in backward-
1240 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1241 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1242 		 */
1243 		/*
1244 		 * XXX meaning of fields, especialy "raflags", is very
1245 		 * differnet between RA prefix list and RR/static prefix list.
1246 		 * how about separating ioctls into two?
1247 		 */
1248 		bzero(oprl, sizeof(*oprl));
1249 		s = splnet();
1250 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1251 			struct nd_pfxrouter *pfr;
1252 			int j;
1253 
1254 			if (i >= PRLSTSIZ)
1255 				break;
1256 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1257 			oprl->prefix[i].raflags = pr->ndpr_raf;
1258 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1259 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1260 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1261 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1262 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1263 				oprl->prefix[i].expire = 0;
1264 			else {
1265 				time_t maxexpire;
1266 
1267 				/* XXX: we assume time_t is signed. */
1268 				maxexpire = (-1) &
1269 				    ~((time_t)1 <<
1270 				    ((sizeof(maxexpire) * 8) - 1));
1271 				if (pr->ndpr_vltime <
1272 				    maxexpire - pr->ndpr_lastupdate) {
1273 					oprl->prefix[i].expire =
1274 					    pr->ndpr_lastupdate +
1275 					    pr->ndpr_vltime;
1276 				} else
1277 					oprl->prefix[i].expire = maxexpire;
1278 			}
1279 
1280 			j = 0;
1281 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1282 				if (j < DRLSTSIZ) {
1283 #define RTRADDR oprl->prefix[i].advrtr[j]
1284 					RTRADDR = pfr->router->rtaddr;
1285 					in6_clearscope(&RTRADDR);
1286 #undef RTRADDR
1287 				}
1288 				j++;
1289 			}
1290 			oprl->prefix[i].advrtrs = j;
1291 			oprl->prefix[i].origin = PR_ORIG_RA;
1292 
1293 			i++;
1294 		}
1295 		splx(s);
1296 
1297 		break;
1298 	case OSIOCGIFINFO_IN6:
1299 #define ND	ndi->ndi
1300 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1301 		bzero(&ND, sizeof(ND));
1302 		ND.linkmtu = IN6_LINKMTU(ifp);
1303 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1304 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1305 		ND.reachable = ND_IFINFO(ifp)->reachable;
1306 		ND.retrans = ND_IFINFO(ifp)->retrans;
1307 		ND.flags = ND_IFINFO(ifp)->flags;
1308 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1309 		ND.chlim = ND_IFINFO(ifp)->chlim;
1310 		break;
1311 	case SIOCGIFINFO_IN6:
1312 		ND = *ND_IFINFO(ifp);
1313 		break;
1314 	case SIOCSIFINFO_IN6:
1315 		/*
1316 		 * used to change host variables from userland.
1317 		 * intented for a use on router to reflect RA configurations.
1318 		 */
1319 		/* 0 means 'unspecified' */
1320 		if (ND.linkmtu != 0) {
1321 			if (ND.linkmtu < IPV6_MMTU ||
1322 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1323 				error = EINVAL;
1324 				break;
1325 			}
1326 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1327 		}
1328 
1329 		if (ND.basereachable != 0) {
1330 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1331 
1332 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1333 			if (ND.basereachable != obasereachable)
1334 				ND_IFINFO(ifp)->reachable =
1335 				    ND_COMPUTE_RTIME(ND.basereachable);
1336 		}
1337 		if (ND.retrans != 0)
1338 			ND_IFINFO(ifp)->retrans = ND.retrans;
1339 		if (ND.chlim != 0)
1340 			ND_IFINFO(ifp)->chlim = ND.chlim;
1341 		/* FALLTHROUGH */
1342 	case SIOCSIFINFO_FLAGS:
1343 	{
1344 		struct ifaddr *ifa;
1345 		struct in6_ifaddr *ia;
1346 
1347 		/*
1348 		 * Try to clear ifdisabled flag when enabling
1349 		 * accept_rtadv or auto_linklocal.
1350 		 */
1351 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1352 		    !(ND.flags & ND6_IFF_IFDISABLED) &&
1353 		    (ND.flags & (ND6_IFF_ACCEPT_RTADV |
1354 		    ND6_IFF_AUTO_LINKLOCAL)))
1355 			ND.flags &= ~ND6_IFF_IFDISABLED;
1356 
1357 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1358 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1359 			/* ifdisabled 1->0 transision */
1360 
1361 			/*
1362 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1363 			 * has an link-local address with IN6_IFF_DUPLICATED,
1364 			 * do not clear ND6_IFF_IFDISABLED.
1365 			 * See RFC 4862, Section 5.4.5.
1366 			 */
1367 			int duplicated_linklocal = 0;
1368 
1369 			IF_ADDR_RLOCK(ifp);
1370 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1371 				if (ifa->ifa_addr->sa_family != AF_INET6)
1372 					continue;
1373 				ia = (struct in6_ifaddr *)ifa;
1374 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1375 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1376 					duplicated_linklocal = 1;
1377 					break;
1378 				}
1379 			}
1380 			IF_ADDR_RUNLOCK(ifp);
1381 
1382 			if (duplicated_linklocal) {
1383 				ND.flags |= ND6_IFF_IFDISABLED;
1384 				log(LOG_ERR, "Cannot enable an interface"
1385 				    " with a link-local address marked"
1386 				    " duplicate.\n");
1387 			} else {
1388 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1389 				if (ifp->if_flags & IFF_UP)
1390 					in6_if_up(ifp);
1391 			}
1392 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1393 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1394 			/* ifdisabled 0->1 transision */
1395 			/* Mark all IPv6 address as tentative. */
1396 
1397 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1398 			IF_ADDR_RLOCK(ifp);
1399 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1400 				if (ifa->ifa_addr->sa_family != AF_INET6)
1401 					continue;
1402 				ia = (struct in6_ifaddr *)ifa;
1403 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1404 			}
1405 			IF_ADDR_RUNLOCK(ifp);
1406 		}
1407 
1408 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1409 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1410 				/* auto_linklocal 0->1 transision */
1411 
1412 				/* If no link-local address on ifp, configure */
1413 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1414 				in6_ifattach(ifp, NULL);
1415 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1416 			    ifp->if_flags & IFF_UP) {
1417 				/*
1418 				 * When the IF already has
1419 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1420 				 * address is assigned, and IFF_UP, try to
1421 				 * assign one.
1422 				 */
1423 				int haslinklocal = 0;
1424 
1425 				IF_ADDR_RLOCK(ifp);
1426 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1427 					if (ifa->ifa_addr->sa_family != AF_INET6)
1428 						continue;
1429 					ia = (struct in6_ifaddr *)ifa;
1430 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1431 						haslinklocal = 1;
1432 						break;
1433 					}
1434 				}
1435 				IF_ADDR_RUNLOCK(ifp);
1436 				if (!haslinklocal)
1437 					in6_ifattach(ifp, NULL);
1438 			}
1439 		}
1440 	}
1441 		ND_IFINFO(ifp)->flags = ND.flags;
1442 		break;
1443 #undef ND
1444 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1445 		/* sync kernel routing table with the default router list */
1446 		defrouter_reset();
1447 		defrouter_select();
1448 		break;
1449 	case SIOCSPFXFLUSH_IN6:
1450 	{
1451 		/* flush all the prefix advertised by routers */
1452 		struct nd_prefix *pr, *next;
1453 
1454 		s = splnet();
1455 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1456 			struct in6_ifaddr *ia, *ia_next;
1457 
1458 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1459 				continue; /* XXX */
1460 
1461 			/* do we really have to remove addresses as well? */
1462 			/* XXXRW: in6_ifaddrhead locking. */
1463 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1464 			    ia_next) {
1465 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1466 					continue;
1467 
1468 				if (ia->ia6_ndpr == pr)
1469 					in6_purgeaddr(&ia->ia_ifa);
1470 			}
1471 			prelist_remove(pr);
1472 		}
1473 		splx(s);
1474 		break;
1475 	}
1476 	case SIOCSRTRFLUSH_IN6:
1477 	{
1478 		/* flush all the default routers */
1479 		struct nd_defrouter *dr, *next;
1480 
1481 		s = splnet();
1482 		defrouter_reset();
1483 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1484 			defrtrlist_del(dr);
1485 		}
1486 		defrouter_select();
1487 		splx(s);
1488 		break;
1489 	}
1490 	case SIOCGNBRINFO_IN6:
1491 	{
1492 		struct llentry *ln;
1493 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1494 
1495 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1496 			return (error);
1497 
1498 		IF_AFDATA_LOCK(ifp);
1499 		ln = nd6_lookup(&nb_addr, 0, ifp);
1500 		IF_AFDATA_UNLOCK(ifp);
1501 
1502 		if (ln == NULL) {
1503 			error = EINVAL;
1504 			break;
1505 		}
1506 		nbi->state = ln->ln_state;
1507 		nbi->asked = ln->la_asked;
1508 		nbi->isrouter = ln->ln_router;
1509 		nbi->expire = ln->la_expire;
1510 		LLE_RUNLOCK(ln);
1511 		break;
1512 	}
1513 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1514 		ndif->ifindex = V_nd6_defifindex;
1515 		break;
1516 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1517 		return (nd6_setdefaultiface(ndif->ifindex));
1518 	}
1519 	return (error);
1520 }
1521 
1522 /*
1523  * Create neighbor cache entry and cache link-layer address,
1524  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1525  *
1526  * type - ICMP6 type
1527  * code - type dependent information
1528  *
1529  * XXXXX
1530  *  The caller of this function already acquired the ndp
1531  *  cache table lock because the cache entry is returned.
1532  */
1533 struct llentry *
1534 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1535     int lladdrlen, int type, int code)
1536 {
1537 	struct llentry *ln = NULL;
1538 	int is_newentry;
1539 	int do_update;
1540 	int olladdr;
1541 	int llchange;
1542 	int flags;
1543 	int newstate = 0;
1544 	uint16_t router = 0;
1545 	struct sockaddr_in6 sin6;
1546 	struct mbuf *chain = NULL;
1547 	int static_route = 0;
1548 
1549 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1550 
1551 	if (ifp == NULL)
1552 		panic("ifp == NULL in nd6_cache_lladdr");
1553 	if (from == NULL)
1554 		panic("from == NULL in nd6_cache_lladdr");
1555 
1556 	/* nothing must be updated for unspecified address */
1557 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1558 		return NULL;
1559 
1560 	/*
1561 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1562 	 * the caller.
1563 	 *
1564 	 * XXX If the link does not have link-layer adderss, what should
1565 	 * we do? (ifp->if_addrlen == 0)
1566 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1567 	 * description on it in NS section (RFC 2461 7.2.3).
1568 	 */
1569 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1570 	IF_AFDATA_LOCK(ifp);
1571 	ln = nd6_lookup(from, flags, ifp);
1572 
1573 	if (ln == NULL) {
1574 		flags |= ND6_EXCLUSIVE;
1575 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1576 		IF_AFDATA_UNLOCK(ifp);
1577 		is_newentry = 1;
1578 	} else {
1579 		IF_AFDATA_UNLOCK(ifp);
1580 		/* do nothing if static ndp is set */
1581 		if (ln->la_flags & LLE_STATIC) {
1582 			static_route = 1;
1583 			goto done;
1584 		}
1585 		is_newentry = 0;
1586 	}
1587 	if (ln == NULL)
1588 		return (NULL);
1589 
1590 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1591 	if (olladdr && lladdr) {
1592 		llchange = bcmp(lladdr, &ln->ll_addr,
1593 		    ifp->if_addrlen);
1594 	} else
1595 		llchange = 0;
1596 
1597 	/*
1598 	 * newentry olladdr  lladdr  llchange	(*=record)
1599 	 *	0	n	n	--	(1)
1600 	 *	0	y	n	--	(2)
1601 	 *	0	n	y	--	(3) * STALE
1602 	 *	0	y	y	n	(4) *
1603 	 *	0	y	y	y	(5) * STALE
1604 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1605 	 *	1	--	y	--	(7) * STALE
1606 	 */
1607 
1608 	if (lladdr) {		/* (3-5) and (7) */
1609 		/*
1610 		 * Record source link-layer address
1611 		 * XXX is it dependent to ifp->if_type?
1612 		 */
1613 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1614 		ln->la_flags |= LLE_VALID;
1615 	}
1616 
1617 	if (!is_newentry) {
1618 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1619 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1620 			do_update = 1;
1621 			newstate = ND6_LLINFO_STALE;
1622 		} else					/* (1-2,4) */
1623 			do_update = 0;
1624 	} else {
1625 		do_update = 1;
1626 		if (lladdr == NULL)			/* (6) */
1627 			newstate = ND6_LLINFO_NOSTATE;
1628 		else					/* (7) */
1629 			newstate = ND6_LLINFO_STALE;
1630 	}
1631 
1632 	if (do_update) {
1633 		/*
1634 		 * Update the state of the neighbor cache.
1635 		 */
1636 		ln->ln_state = newstate;
1637 
1638 		if (ln->ln_state == ND6_LLINFO_STALE) {
1639 			/*
1640 			 * XXX: since nd6_output() below will cause
1641 			 * state tansition to DELAY and reset the timer,
1642 			 * we must set the timer now, although it is actually
1643 			 * meaningless.
1644 			 */
1645 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1646 
1647 			if (ln->la_hold) {
1648 				struct mbuf *m_hold, *m_hold_next;
1649 
1650 				/*
1651 				 * reset the la_hold in advance, to explicitly
1652 				 * prevent a la_hold lookup in nd6_output()
1653 				 * (wouldn't happen, though...)
1654 				 */
1655 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1656 				    m_hold; m_hold = m_hold_next) {
1657 					m_hold_next = m_hold->m_nextpkt;
1658 					m_hold->m_nextpkt = NULL;
1659 
1660 					/*
1661 					 * we assume ifp is not a p2p here, so
1662 					 * just set the 2nd argument as the
1663 					 * 1st one.
1664 					 */
1665 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1666 				}
1667 				/*
1668 				 * If we have mbufs in the chain we need to do
1669 				 * deferred transmit. Copy the address from the
1670 				 * llentry before dropping the lock down below.
1671 				 */
1672 				if (chain != NULL)
1673 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1674 			}
1675 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1676 			/* probe right away */
1677 			nd6_llinfo_settimer_locked((void *)ln, 0);
1678 		}
1679 	}
1680 
1681 	/*
1682 	 * ICMP6 type dependent behavior.
1683 	 *
1684 	 * NS: clear IsRouter if new entry
1685 	 * RS: clear IsRouter
1686 	 * RA: set IsRouter if there's lladdr
1687 	 * redir: clear IsRouter if new entry
1688 	 *
1689 	 * RA case, (1):
1690 	 * The spec says that we must set IsRouter in the following cases:
1691 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1692 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1693 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1694 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1695 	 * neighbor cache, this is similar to (6).
1696 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1697 	 *
1698 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1699 	 *							D R
1700 	 *	0	n	n	--	(1)	c   ?     s
1701 	 *	0	y	n	--	(2)	c   s     s
1702 	 *	0	n	y	--	(3)	c   s     s
1703 	 *	0	y	y	n	(4)	c   s     s
1704 	 *	0	y	y	y	(5)	c   s     s
1705 	 *	1	--	n	--	(6) c	c	c s
1706 	 *	1	--	y	--	(7) c	c   s	c s
1707 	 *
1708 	 *					(c=clear s=set)
1709 	 */
1710 	switch (type & 0xff) {
1711 	case ND_NEIGHBOR_SOLICIT:
1712 		/*
1713 		 * New entry must have is_router flag cleared.
1714 		 */
1715 		if (is_newentry)	/* (6-7) */
1716 			ln->ln_router = 0;
1717 		break;
1718 	case ND_REDIRECT:
1719 		/*
1720 		 * If the icmp is a redirect to a better router, always set the
1721 		 * is_router flag.  Otherwise, if the entry is newly created,
1722 		 * clear the flag.  [RFC 2461, sec 8.3]
1723 		 */
1724 		if (code == ND_REDIRECT_ROUTER)
1725 			ln->ln_router = 1;
1726 		else if (is_newentry) /* (6-7) */
1727 			ln->ln_router = 0;
1728 		break;
1729 	case ND_ROUTER_SOLICIT:
1730 		/*
1731 		 * is_router flag must always be cleared.
1732 		 */
1733 		ln->ln_router = 0;
1734 		break;
1735 	case ND_ROUTER_ADVERT:
1736 		/*
1737 		 * Mark an entry with lladdr as a router.
1738 		 */
1739 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1740 		    (is_newentry && lladdr)) {			/* (7) */
1741 			ln->ln_router = 1;
1742 		}
1743 		break;
1744 	}
1745 
1746 	if (ln != NULL) {
1747 		static_route = (ln->la_flags & LLE_STATIC);
1748 		router = ln->ln_router;
1749 
1750 		if (flags & ND6_EXCLUSIVE)
1751 			LLE_WUNLOCK(ln);
1752 		else
1753 			LLE_RUNLOCK(ln);
1754 		if (static_route)
1755 			ln = NULL;
1756 	}
1757 	if (chain)
1758 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1759 
1760 	/*
1761 	 * When the link-layer address of a router changes, select the
1762 	 * best router again.  In particular, when the neighbor entry is newly
1763 	 * created, it might affect the selection policy.
1764 	 * Question: can we restrict the first condition to the "is_newentry"
1765 	 * case?
1766 	 * XXX: when we hear an RA from a new router with the link-layer
1767 	 * address option, defrouter_select() is called twice, since
1768 	 * defrtrlist_update called the function as well.  However, I believe
1769 	 * we can compromise the overhead, since it only happens the first
1770 	 * time.
1771 	 * XXX: although defrouter_select() should not have a bad effect
1772 	 * for those are not autoconfigured hosts, we explicitly avoid such
1773 	 * cases for safety.
1774 	 */
1775 	if (do_update && router &&
1776 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1777 		/*
1778 		 * guaranteed recursion
1779 		 */
1780 		defrouter_select();
1781 	}
1782 
1783 	return (ln);
1784 done:
1785 	if (ln != NULL) {
1786 		if (flags & ND6_EXCLUSIVE)
1787 			LLE_WUNLOCK(ln);
1788 		else
1789 			LLE_RUNLOCK(ln);
1790 		if (static_route)
1791 			ln = NULL;
1792 	}
1793 	return (ln);
1794 }
1795 
1796 static void
1797 nd6_slowtimo(void *arg)
1798 {
1799 	CURVNET_SET((struct vnet *) arg);
1800 	struct nd_ifinfo *nd6if;
1801 	struct ifnet *ifp;
1802 
1803 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1804 	    nd6_slowtimo, curvnet);
1805 	IFNET_RLOCK_NOSLEEP();
1806 	TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1807 		nd6if = ND_IFINFO(ifp);
1808 		if (nd6if->basereachable && /* already initialized */
1809 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1810 			/*
1811 			 * Since reachable time rarely changes by router
1812 			 * advertisements, we SHOULD insure that a new random
1813 			 * value gets recomputed at least once every few hours.
1814 			 * (RFC 2461, 6.3.4)
1815 			 */
1816 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1817 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1818 		}
1819 	}
1820 	IFNET_RUNLOCK_NOSLEEP();
1821 	CURVNET_RESTORE();
1822 }
1823 
1824 int
1825 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1826     struct sockaddr_in6 *dst, struct rtentry *rt0)
1827 {
1828 
1829 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1830 }
1831 
1832 
1833 /*
1834  * Note that I'm not enforcing any global serialization
1835  * lle state or asked changes here as the logic is too
1836  * complicated to avoid having to always acquire an exclusive
1837  * lock
1838  * KMM
1839  *
1840  */
1841 #define senderr(e) { error = (e); goto bad;}
1842 
1843 int
1844 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1845     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1846 	struct mbuf **chain)
1847 {
1848 	struct mbuf *m = m0;
1849 	struct m_tag *mtag;
1850 	struct llentry *ln = lle;
1851 	struct ip6_hdr *ip6;
1852 	int error = 0;
1853 	int flags = 0;
1854 	int ip6len;
1855 
1856 #ifdef INVARIANTS
1857 	if (lle != NULL) {
1858 
1859 		LLE_WLOCK_ASSERT(lle);
1860 
1861 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1862 	}
1863 #endif
1864 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1865 		goto sendpkt;
1866 
1867 	if (nd6_need_cache(ifp) == 0)
1868 		goto sendpkt;
1869 
1870 	/*
1871 	 * next hop determination.  This routine is derived from ether_output.
1872 	 */
1873 
1874 	/*
1875 	 * Address resolution or Neighbor Unreachability Detection
1876 	 * for the next hop.
1877 	 * At this point, the destination of the packet must be a unicast
1878 	 * or an anycast address(i.e. not a multicast).
1879 	 */
1880 
1881 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1882 	if (ln == NULL) {
1883 	retry:
1884 		IF_AFDATA_LOCK(ifp);
1885 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1886 		IF_AFDATA_UNLOCK(ifp);
1887 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1888 			/*
1889 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1890 			 * the condition below is not very efficient.  But we believe
1891 			 * it is tolerable, because this should be a rare case.
1892 			 */
1893 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1894 			IF_AFDATA_LOCK(ifp);
1895 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1896 			IF_AFDATA_UNLOCK(ifp);
1897 		}
1898 	}
1899 	if (ln == NULL) {
1900 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1901 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1902 			char ip6buf[INET6_ADDRSTRLEN];
1903 			log(LOG_DEBUG,
1904 			    "nd6_output: can't allocate llinfo for %s "
1905 			    "(ln=%p)\n",
1906 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1907 			senderr(EIO);	/* XXX: good error? */
1908 		}
1909 		goto sendpkt;	/* send anyway */
1910 	}
1911 
1912 	/* We don't have to do link-layer address resolution on a p2p link. */
1913 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1914 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1915 		if ((flags & LLE_EXCLUSIVE) == 0) {
1916 			flags |= LLE_EXCLUSIVE;
1917 			goto retry;
1918 		}
1919 		ln->ln_state = ND6_LLINFO_STALE;
1920 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1921 	}
1922 
1923 	/*
1924 	 * The first time we send a packet to a neighbor whose entry is
1925 	 * STALE, we have to change the state to DELAY and a sets a timer to
1926 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1927 	 * neighbor unreachability detection on expiration.
1928 	 * (RFC 2461 7.3.3)
1929 	 */
1930 	if (ln->ln_state == ND6_LLINFO_STALE) {
1931 		if ((flags & LLE_EXCLUSIVE) == 0) {
1932 			flags |= LLE_EXCLUSIVE;
1933 			LLE_RUNLOCK(ln);
1934 			goto retry;
1935 		}
1936 		ln->la_asked = 0;
1937 		ln->ln_state = ND6_LLINFO_DELAY;
1938 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1939 	}
1940 
1941 	/*
1942 	 * If the neighbor cache entry has a state other than INCOMPLETE
1943 	 * (i.e. its link-layer address is already resolved), just
1944 	 * send the packet.
1945 	 */
1946 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1947 		goto sendpkt;
1948 
1949 	/*
1950 	 * There is a neighbor cache entry, but no ethernet address
1951 	 * response yet.  Append this latest packet to the end of the
1952 	 * packet queue in the mbuf, unless the number of the packet
1953 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1954 	 * the oldest packet in the queue will be removed.
1955 	 */
1956 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1957 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1958 
1959 	if ((flags & LLE_EXCLUSIVE) == 0) {
1960 		flags |= LLE_EXCLUSIVE;
1961 		LLE_RUNLOCK(ln);
1962 		goto retry;
1963 	}
1964 
1965 	LLE_WLOCK_ASSERT(ln);
1966 
1967 	if (ln->la_hold) {
1968 		struct mbuf *m_hold;
1969 		int i;
1970 
1971 		i = 0;
1972 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1973 			i++;
1974 			if (m_hold->m_nextpkt == NULL) {
1975 				m_hold->m_nextpkt = m;
1976 				break;
1977 			}
1978 		}
1979 		while (i >= V_nd6_maxqueuelen) {
1980 			m_hold = ln->la_hold;
1981 			ln->la_hold = ln->la_hold->m_nextpkt;
1982 			m_freem(m_hold);
1983 			i--;
1984 		}
1985 	} else {
1986 		ln->la_hold = m;
1987 	}
1988 
1989 	/*
1990 	 * If there has been no NS for the neighbor after entering the
1991 	 * INCOMPLETE state, send the first solicitation.
1992 	 */
1993 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1994 		ln->la_asked++;
1995 
1996 		nd6_llinfo_settimer_locked(ln,
1997 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1998 		LLE_WUNLOCK(ln);
1999 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2000 		if (lle != NULL && ln == lle)
2001 			LLE_WLOCK(lle);
2002 
2003 	} else if (lle == NULL || ln != lle) {
2004 		/*
2005 		 * We did the lookup (no lle arg) so we
2006 		 * need to do the unlock here.
2007 		 */
2008 		LLE_WUNLOCK(ln);
2009 	}
2010 
2011 	return (0);
2012 
2013   sendpkt:
2014 	/* discard the packet if IPv6 operation is disabled on the interface */
2015 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2016 		error = ENETDOWN; /* better error? */
2017 		goto bad;
2018 	}
2019 	/*
2020 	 * ln is valid and the caller did not pass in
2021 	 * an llentry
2022 	 */
2023 	if ((ln != NULL) && (lle == NULL)) {
2024 		if (flags & LLE_EXCLUSIVE)
2025 			LLE_WUNLOCK(ln);
2026 		else
2027 			LLE_RUNLOCK(ln);
2028 	}
2029 
2030 #ifdef MAC
2031 	mac_netinet6_nd6_send(ifp, m);
2032 #endif
2033 
2034 	/*
2035 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2036 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2037 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2038 	 * to be diverted to user space.  When re-injected into the kernel,
2039 	 * send_output() will directly dispatch them to the outgoing interface.
2040 	 */
2041 	if (send_sendso_input_hook != NULL) {
2042 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2043 		if (mtag != NULL) {
2044 			ip6 = mtod(m, struct ip6_hdr *);
2045 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2046 			/* Use the SEND socket */
2047 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2048 			    ip6len);
2049 			/* -1 == no app on SEND socket */
2050 			if (error == 0 || error != -1)
2051 			    return (error);
2052 		}
2053 	}
2054 
2055 	/*
2056 	 * We were passed in a pointer to an lle with the lock held
2057 	 * this means that we can't call if_output as we will
2058 	 * recurse on the lle lock - so what we do is we create
2059 	 * a list of mbufs to send and transmit them in the caller
2060 	 * after the lock is dropped
2061 	 */
2062 	if (lle != NULL) {
2063 		if (*chain == NULL)
2064 			*chain = m;
2065 		else {
2066 			struct mbuf *mb;
2067 
2068 			/*
2069 			 * append mbuf to end of deferred chain
2070 			 */
2071 			mb = *chain;
2072 			while (mb->m_nextpkt != NULL)
2073 				mb = mb->m_nextpkt;
2074 			mb->m_nextpkt = m;
2075 		}
2076 		return (error);
2077 	}
2078 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2079 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2080 		    NULL));
2081 	}
2082 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2083 	return (error);
2084 
2085   bad:
2086 	/*
2087 	 * ln is valid and the caller did not pass in
2088 	 * an llentry
2089 	 */
2090 	if ((ln != NULL) && (lle == NULL)) {
2091 		if (flags & LLE_EXCLUSIVE)
2092 			LLE_WUNLOCK(ln);
2093 		else
2094 			LLE_RUNLOCK(ln);
2095 	}
2096 	if (m)
2097 		m_freem(m);
2098 	return (error);
2099 }
2100 #undef senderr
2101 
2102 
2103 int
2104 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2105     struct sockaddr_in6 *dst, struct route *ro)
2106 {
2107 	struct mbuf *m, *m_head;
2108 	struct ifnet *outifp;
2109 	int error = 0;
2110 
2111 	m_head = chain;
2112 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2113 		outifp = origifp;
2114 	else
2115 		outifp = ifp;
2116 
2117 	while (m_head) {
2118 		m = m_head;
2119 		m_head = m_head->m_nextpkt;
2120 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2121 	}
2122 
2123 	/*
2124 	 * XXX
2125 	 * note that intermediate errors are blindly ignored - but this is
2126 	 * the same convention as used with nd6_output when called by
2127 	 * nd6_cache_lladdr
2128 	 */
2129 	return (error);
2130 }
2131 
2132 
2133 int
2134 nd6_need_cache(struct ifnet *ifp)
2135 {
2136 	/*
2137 	 * XXX: we currently do not make neighbor cache on any interface
2138 	 * other than ARCnet, Ethernet, FDDI and GIF.
2139 	 *
2140 	 * RFC2893 says:
2141 	 * - unidirectional tunnels needs no ND
2142 	 */
2143 	switch (ifp->if_type) {
2144 	case IFT_ARCNET:
2145 	case IFT_ETHER:
2146 	case IFT_FDDI:
2147 	case IFT_IEEE1394:
2148 #ifdef IFT_L2VLAN
2149 	case IFT_L2VLAN:
2150 #endif
2151 #ifdef IFT_IEEE80211
2152 	case IFT_IEEE80211:
2153 #endif
2154 	case IFT_INFINIBAND:
2155 	case IFT_GIF:		/* XXX need more cases? */
2156 	case IFT_PPP:
2157 	case IFT_TUNNEL:
2158 	case IFT_BRIDGE:
2159 	case IFT_PROPVIRTUAL:
2160 		return (1);
2161 	default:
2162 		return (0);
2163 	}
2164 }
2165 
2166 /*
2167  * the callers of this function need to be re-worked to drop
2168  * the lle lock, drop here for now
2169  */
2170 int
2171 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2172     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2173 {
2174 	struct llentry *ln;
2175 
2176 	*lle = NULL;
2177 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2178 	if (m->m_flags & M_MCAST) {
2179 		int i;
2180 
2181 		switch (ifp->if_type) {
2182 		case IFT_ETHER:
2183 		case IFT_FDDI:
2184 #ifdef IFT_L2VLAN
2185 		case IFT_L2VLAN:
2186 #endif
2187 #ifdef IFT_IEEE80211
2188 		case IFT_IEEE80211:
2189 #endif
2190 		case IFT_BRIDGE:
2191 		case IFT_ISO88025:
2192 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2193 						 desten);
2194 			return (0);
2195 		case IFT_IEEE1394:
2196 			/*
2197 			 * netbsd can use if_broadcastaddr, but we don't do so
2198 			 * to reduce # of ifdef.
2199 			 */
2200 			for (i = 0; i < ifp->if_addrlen; i++)
2201 				desten[i] = ~0;
2202 			return (0);
2203 		case IFT_ARCNET:
2204 			*desten = 0;
2205 			return (0);
2206 		default:
2207 			m_freem(m);
2208 			return (EAFNOSUPPORT);
2209 		}
2210 	}
2211 
2212 
2213 	/*
2214 	 * the entry should have been created in nd6_store_lladdr
2215 	 */
2216 	IF_AFDATA_LOCK(ifp);
2217 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2218 	IF_AFDATA_UNLOCK(ifp);
2219 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2220 		if (ln != NULL)
2221 			LLE_RUNLOCK(ln);
2222 		/* this could happen, if we could not allocate memory */
2223 		m_freem(m);
2224 		return (1);
2225 	}
2226 
2227 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2228 	*lle = ln;
2229 	LLE_RUNLOCK(ln);
2230 	/*
2231 	 * A *small* use after free race exists here
2232 	 */
2233 	return (0);
2234 }
2235 
2236 static void
2237 clear_llinfo_pqueue(struct llentry *ln)
2238 {
2239 	struct mbuf *m_hold, *m_hold_next;
2240 
2241 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2242 		m_hold_next = m_hold->m_nextpkt;
2243 		m_hold->m_nextpkt = NULL;
2244 		m_freem(m_hold);
2245 	}
2246 
2247 	ln->la_hold = NULL;
2248 	return;
2249 }
2250 
2251 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2252 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2253 #ifdef SYSCTL_DECL
2254 SYSCTL_DECL(_net_inet6_icmp6);
2255 #endif
2256 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2257 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2258 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2259 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2260 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2261 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2262 
2263 static int
2264 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2265 {
2266 	int error;
2267 	char buf[1024] __aligned(4);
2268 	struct in6_defrouter *d, *de;
2269 	struct nd_defrouter *dr;
2270 
2271 	if (req->newptr)
2272 		return EPERM;
2273 	error = 0;
2274 
2275 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2276 		d = (struct in6_defrouter *)buf;
2277 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2278 
2279 		if (d + 1 <= de) {
2280 			bzero(d, sizeof(*d));
2281 			d->rtaddr.sin6_family = AF_INET6;
2282 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2283 			d->rtaddr.sin6_addr = dr->rtaddr;
2284 			error = sa6_recoverscope(&d->rtaddr);
2285 			if (error != 0)
2286 				return (error);
2287 			d->flags = dr->flags;
2288 			d->rtlifetime = dr->rtlifetime;
2289 			d->expire = dr->expire;
2290 			d->if_index = dr->ifp->if_index;
2291 		} else
2292 			panic("buffer too short");
2293 
2294 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2295 		if (error)
2296 			break;
2297 	}
2298 
2299 	return (error);
2300 }
2301 
2302 static int
2303 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2304 {
2305 	int error;
2306 	char buf[1024] __aligned(4);
2307 	struct in6_prefix *p, *pe;
2308 	struct nd_prefix *pr;
2309 	char ip6buf[INET6_ADDRSTRLEN];
2310 
2311 	if (req->newptr)
2312 		return EPERM;
2313 	error = 0;
2314 
2315 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2316 		u_short advrtrs;
2317 		size_t advance;
2318 		struct sockaddr_in6 *sin6, *s6;
2319 		struct nd_pfxrouter *pfr;
2320 
2321 		p = (struct in6_prefix *)buf;
2322 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2323 
2324 		if (p + 1 <= pe) {
2325 			bzero(p, sizeof(*p));
2326 			sin6 = (struct sockaddr_in6 *)(p + 1);
2327 
2328 			p->prefix = pr->ndpr_prefix;
2329 			if (sa6_recoverscope(&p->prefix)) {
2330 				log(LOG_ERR,
2331 				    "scope error in prefix list (%s)\n",
2332 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2333 				/* XXX: press on... */
2334 			}
2335 			p->raflags = pr->ndpr_raf;
2336 			p->prefixlen = pr->ndpr_plen;
2337 			p->vltime = pr->ndpr_vltime;
2338 			p->pltime = pr->ndpr_pltime;
2339 			p->if_index = pr->ndpr_ifp->if_index;
2340 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2341 				p->expire = 0;
2342 			else {
2343 				time_t maxexpire;
2344 
2345 				/* XXX: we assume time_t is signed. */
2346 				maxexpire = (-1) &
2347 				    ~((time_t)1 <<
2348 				    ((sizeof(maxexpire) * 8) - 1));
2349 				if (pr->ndpr_vltime <
2350 				    maxexpire - pr->ndpr_lastupdate) {
2351 				    p->expire = pr->ndpr_lastupdate +
2352 				        pr->ndpr_vltime;
2353 				} else
2354 					p->expire = maxexpire;
2355 			}
2356 			p->refcnt = pr->ndpr_refcnt;
2357 			p->flags = pr->ndpr_stateflags;
2358 			p->origin = PR_ORIG_RA;
2359 			advrtrs = 0;
2360 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2361 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2362 					advrtrs++;
2363 					continue;
2364 				}
2365 				s6 = &sin6[advrtrs];
2366 				bzero(s6, sizeof(*s6));
2367 				s6->sin6_family = AF_INET6;
2368 				s6->sin6_len = sizeof(*sin6);
2369 				s6->sin6_addr = pfr->router->rtaddr;
2370 				if (sa6_recoverscope(s6)) {
2371 					log(LOG_ERR,
2372 					    "scope error in "
2373 					    "prefix list (%s)\n",
2374 					    ip6_sprintf(ip6buf,
2375 						    &pfr->router->rtaddr));
2376 				}
2377 				advrtrs++;
2378 			}
2379 			p->advrtrs = advrtrs;
2380 		} else
2381 			panic("buffer too short");
2382 
2383 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2384 		error = SYSCTL_OUT(req, buf, advance);
2385 		if (error)
2386 			break;
2387 	}
2388 
2389 	return (error);
2390 }
2391