xref: /freebsd/sys/netinet6/nd6.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 
197 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
198 	nd6_setmtu0(ifp, nd);
199 
200 	return nd;
201 }
202 
203 void
204 nd6_ifdetach(struct nd_ifinfo *nd)
205 {
206 
207 	free(nd, M_IP6NDP);
208 }
209 
210 /*
211  * Reset ND level link MTU. This function is called when the physical MTU
212  * changes, which means we might have to adjust the ND level MTU.
213  */
214 void
215 nd6_setmtu(struct ifnet *ifp)
216 {
217 
218 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
219 }
220 
221 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
222 void
223 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
224 {
225 	u_int32_t omaxmtu;
226 
227 	omaxmtu = ndi->maxmtu;
228 
229 	switch (ifp->if_type) {
230 	case IFT_ARCNET:
231 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
232 		break;
233 	case IFT_FDDI:
234 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
235 		break;
236 	case IFT_ISO88025:
237 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
238 		 break;
239 	default:
240 		ndi->maxmtu = ifp->if_mtu;
241 		break;
242 	}
243 
244 	/*
245 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
246 	 * undesirable situation.  We thus notify the operator of the change
247 	 * explicitly.  The check for omaxmtu is necessary to restrict the
248 	 * log to the case of changing the MTU, not initializing it.
249 	 */
250 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
251 		log(LOG_NOTICE, "nd6_setmtu0: "
252 		    "new link MTU on %s (%lu) is too small for IPv6\n",
253 		    if_name(ifp), (unsigned long)ndi->maxmtu);
254 	}
255 
256 	if (ndi->maxmtu > V_in6_maxmtu)
257 		in6_setmaxmtu(); /* check all interfaces just in case */
258 
259 }
260 
261 void
262 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
263 {
264 
265 	bzero(ndopts, sizeof(*ndopts));
266 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
267 	ndopts->nd_opts_last
268 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
269 
270 	if (icmp6len == 0) {
271 		ndopts->nd_opts_done = 1;
272 		ndopts->nd_opts_search = NULL;
273 	}
274 }
275 
276 /*
277  * Take one ND option.
278  */
279 struct nd_opt_hdr *
280 nd6_option(union nd_opts *ndopts)
281 {
282 	struct nd_opt_hdr *nd_opt;
283 	int olen;
284 
285 	if (ndopts == NULL)
286 		panic("ndopts == NULL in nd6_option");
287 	if (ndopts->nd_opts_last == NULL)
288 		panic("uninitialized ndopts in nd6_option");
289 	if (ndopts->nd_opts_search == NULL)
290 		return NULL;
291 	if (ndopts->nd_opts_done)
292 		return NULL;
293 
294 	nd_opt = ndopts->nd_opts_search;
295 
296 	/* make sure nd_opt_len is inside the buffer */
297 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
298 		bzero(ndopts, sizeof(*ndopts));
299 		return NULL;
300 	}
301 
302 	olen = nd_opt->nd_opt_len << 3;
303 	if (olen == 0) {
304 		/*
305 		 * Message validation requires that all included
306 		 * options have a length that is greater than zero.
307 		 */
308 		bzero(ndopts, sizeof(*ndopts));
309 		return NULL;
310 	}
311 
312 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
313 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
314 		/* option overruns the end of buffer, invalid */
315 		bzero(ndopts, sizeof(*ndopts));
316 		return NULL;
317 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
318 		/* reached the end of options chain */
319 		ndopts->nd_opts_done = 1;
320 		ndopts->nd_opts_search = NULL;
321 	}
322 	return nd_opt;
323 }
324 
325 /*
326  * Parse multiple ND options.
327  * This function is much easier to use, for ND routines that do not need
328  * multiple options of the same type.
329  */
330 int
331 nd6_options(union nd_opts *ndopts)
332 {
333 	struct nd_opt_hdr *nd_opt;
334 	int i = 0;
335 
336 	if (ndopts == NULL)
337 		panic("ndopts == NULL in nd6_options");
338 	if (ndopts->nd_opts_last == NULL)
339 		panic("uninitialized ndopts in nd6_options");
340 	if (ndopts->nd_opts_search == NULL)
341 		return 0;
342 
343 	while (1) {
344 		nd_opt = nd6_option(ndopts);
345 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
346 			/*
347 			 * Message validation requires that all included
348 			 * options have a length that is greater than zero.
349 			 */
350 			ICMP6STAT_INC(icp6s_nd_badopt);
351 			bzero(ndopts, sizeof(*ndopts));
352 			return -1;
353 		}
354 
355 		if (nd_opt == NULL)
356 			goto skip1;
357 
358 		switch (nd_opt->nd_opt_type) {
359 		case ND_OPT_SOURCE_LINKADDR:
360 		case ND_OPT_TARGET_LINKADDR:
361 		case ND_OPT_MTU:
362 		case ND_OPT_REDIRECTED_HEADER:
363 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
364 				nd6log((LOG_INFO,
365 				    "duplicated ND6 option found (type=%d)\n",
366 				    nd_opt->nd_opt_type));
367 				/* XXX bark? */
368 			} else {
369 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
370 					= nd_opt;
371 			}
372 			break;
373 		case ND_OPT_PREFIX_INFORMATION:
374 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
375 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
376 					= nd_opt;
377 			}
378 			ndopts->nd_opts_pi_end =
379 				(struct nd_opt_prefix_info *)nd_opt;
380 			break;
381 		default:
382 			/*
383 			 * Unknown options must be silently ignored,
384 			 * to accomodate future extension to the protocol.
385 			 */
386 			nd6log((LOG_DEBUG,
387 			    "nd6_options: unsupported option %d - "
388 			    "option ignored\n", nd_opt->nd_opt_type));
389 		}
390 
391 skip1:
392 		i++;
393 		if (i > V_nd6_maxndopt) {
394 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
395 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
396 			break;
397 		}
398 
399 		if (ndopts->nd_opts_done)
400 			break;
401 	}
402 
403 	return 0;
404 }
405 
406 /*
407  * ND6 timer routine to handle ND6 entries
408  */
409 void
410 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
411 {
412 	int canceled;
413 
414 	LLE_WLOCK_ASSERT(ln);
415 
416 	if (tick < 0) {
417 		ln->la_expire = 0;
418 		ln->ln_ntick = 0;
419 		canceled = callout_stop(&ln->ln_timer_ch);
420 	} else {
421 		ln->la_expire = time_second + tick / hz;
422 		LLE_ADDREF(ln);
423 		if (tick > INT_MAX) {
424 			ln->ln_ntick = tick - INT_MAX;
425 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
426 			    nd6_llinfo_timer, ln);
427 		} else {
428 			ln->ln_ntick = 0;
429 			canceled = callout_reset(&ln->ln_timer_ch, tick,
430 			    nd6_llinfo_timer, ln);
431 		}
432 	}
433 	if (canceled)
434 		LLE_REMREF(ln);
435 }
436 
437 void
438 nd6_llinfo_settimer(struct llentry *ln, long tick)
439 {
440 
441 	LLE_WLOCK(ln);
442 	nd6_llinfo_settimer_locked(ln, tick);
443 	LLE_WUNLOCK(ln);
444 }
445 
446 static void
447 nd6_llinfo_timer(void *arg)
448 {
449 	struct llentry *ln;
450 	struct in6_addr *dst;
451 	struct ifnet *ifp;
452 	struct nd_ifinfo *ndi = NULL;
453 
454 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
455 	ln = (struct llentry *)arg;
456 	LLE_WLOCK_ASSERT(ln);
457 	ifp = ln->lle_tbl->llt_ifp;
458 
459 	CURVNET_SET(ifp->if_vnet);
460 
461 	if (ln->ln_ntick > 0) {
462 		if (ln->ln_ntick > INT_MAX) {
463 			ln->ln_ntick -= INT_MAX;
464 			nd6_llinfo_settimer_locked(ln, INT_MAX);
465 		} else {
466 			ln->ln_ntick = 0;
467 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
468 		}
469 		goto done;
470 	}
471 
472 	ndi = ND_IFINFO(ifp);
473 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
474 	if (ln->la_flags & LLE_STATIC) {
475 		goto done;
476 	}
477 
478 	if (ln->la_flags & LLE_DELETED) {
479 		(void)nd6_free(ln, 0);
480 		ln = NULL;
481 		goto done;
482 	}
483 
484 	switch (ln->ln_state) {
485 	case ND6_LLINFO_INCOMPLETE:
486 		if (ln->la_asked < V_nd6_mmaxtries) {
487 			ln->la_asked++;
488 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
489 			LLE_WUNLOCK(ln);
490 			nd6_ns_output(ifp, NULL, dst, ln, 0);
491 			LLE_WLOCK(ln);
492 		} else {
493 			struct mbuf *m = ln->la_hold;
494 			if (m) {
495 				struct mbuf *m0;
496 
497 				/*
498 				 * assuming every packet in la_hold has the
499 				 * same IP header.  Send error after unlock.
500 				 */
501 				m0 = m->m_nextpkt;
502 				m->m_nextpkt = NULL;
503 				ln->la_hold = m0;
504 				clear_llinfo_pqueue(ln);
505 			}
506 			(void)nd6_free(ln, 0);
507 			ln = NULL;
508 			if (m != NULL)
509 				icmp6_error2(m, ICMP6_DST_UNREACH,
510 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
511 		}
512 		break;
513 	case ND6_LLINFO_REACHABLE:
514 		if (!ND6_LLINFO_PERMANENT(ln)) {
515 			ln->ln_state = ND6_LLINFO_STALE;
516 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
517 		}
518 		break;
519 
520 	case ND6_LLINFO_STALE:
521 		/* Garbage Collection(RFC 2461 5.3) */
522 		if (!ND6_LLINFO_PERMANENT(ln)) {
523 			(void)nd6_free(ln, 1);
524 			ln = NULL;
525 		}
526 		break;
527 
528 	case ND6_LLINFO_DELAY:
529 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
530 			/* We need NUD */
531 			ln->la_asked = 1;
532 			ln->ln_state = ND6_LLINFO_PROBE;
533 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
534 			LLE_WUNLOCK(ln);
535 			nd6_ns_output(ifp, dst, dst, ln, 0);
536 			LLE_WLOCK(ln);
537 		} else {
538 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
539 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
540 		}
541 		break;
542 	case ND6_LLINFO_PROBE:
543 		if (ln->la_asked < V_nd6_umaxtries) {
544 			ln->la_asked++;
545 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
546 			LLE_WUNLOCK(ln);
547 			nd6_ns_output(ifp, dst, dst, ln, 0);
548 			LLE_WLOCK(ln);
549 		} else {
550 			(void)nd6_free(ln, 0);
551 			ln = NULL;
552 		}
553 		break;
554 	default:
555 		panic("%s: paths in a dark night can be confusing: %d",
556 		    __func__, ln->ln_state);
557 	}
558 done:
559 	if (ln != NULL)
560 		LLE_FREE_LOCKED(ln);
561 	CURVNET_RESTORE();
562 }
563 
564 
565 /*
566  * ND6 timer routine to expire default route list and prefix list
567  */
568 void
569 nd6_timer(void *arg)
570 {
571 	CURVNET_SET((struct vnet *) arg);
572 	int s;
573 	struct nd_defrouter *dr;
574 	struct nd_prefix *pr;
575 	struct in6_ifaddr *ia6, *nia6;
576 	struct in6_addrlifetime *lt6;
577 
578 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
579 	    nd6_timer, curvnet);
580 
581 	/* expire default router list */
582 	s = splnet();
583 	dr = TAILQ_FIRST(&V_nd_defrouter);
584 	while (dr) {
585 		if (dr->expire && dr->expire < time_second) {
586 			struct nd_defrouter *t;
587 			t = TAILQ_NEXT(dr, dr_entry);
588 			defrtrlist_del(dr);
589 			dr = t;
590 		} else {
591 			dr = TAILQ_NEXT(dr, dr_entry);
592 		}
593 	}
594 
595 	/*
596 	 * expire interface addresses.
597 	 * in the past the loop was inside prefix expiry processing.
598 	 * However, from a stricter speci-confrmance standpoint, we should
599 	 * rather separate address lifetimes and prefix lifetimes.
600 	 *
601 	 * XXXRW: in6_ifaddrhead locking.
602 	 */
603   addrloop:
604 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
605 		/* check address lifetime */
606 		lt6 = &ia6->ia6_lifetime;
607 		if (IFA6_IS_INVALID(ia6)) {
608 			int regen = 0;
609 
610 			/*
611 			 * If the expiring address is temporary, try
612 			 * regenerating a new one.  This would be useful when
613 			 * we suspended a laptop PC, then turned it on after a
614 			 * period that could invalidate all temporary
615 			 * addresses.  Although we may have to restart the
616 			 * loop (see below), it must be after purging the
617 			 * address.  Otherwise, we'd see an infinite loop of
618 			 * regeneration.
619 			 */
620 			if (V_ip6_use_tempaddr &&
621 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
622 				if (regen_tmpaddr(ia6) == 0)
623 					regen = 1;
624 			}
625 
626 			in6_purgeaddr(&ia6->ia_ifa);
627 
628 			if (regen)
629 				goto addrloop; /* XXX: see below */
630 		} else if (IFA6_IS_DEPRECATED(ia6)) {
631 			int oldflags = ia6->ia6_flags;
632 
633 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
634 
635 			/*
636 			 * If a temporary address has just become deprecated,
637 			 * regenerate a new one if possible.
638 			 */
639 			if (V_ip6_use_tempaddr &&
640 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
641 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
642 
643 				if (regen_tmpaddr(ia6) == 0) {
644 					/*
645 					 * A new temporary address is
646 					 * generated.
647 					 * XXX: this means the address chain
648 					 * has changed while we are still in
649 					 * the loop.  Although the change
650 					 * would not cause disaster (because
651 					 * it's not a deletion, but an
652 					 * addition,) we'd rather restart the
653 					 * loop just for safety.  Or does this
654 					 * significantly reduce performance??
655 					 */
656 					goto addrloop;
657 				}
658 			}
659 		} else {
660 			/*
661 			 * A new RA might have made a deprecated address
662 			 * preferred.
663 			 */
664 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
665 		}
666 	}
667 
668 	/* expire prefix list */
669 	pr = V_nd_prefix.lh_first;
670 	while (pr) {
671 		/*
672 		 * check prefix lifetime.
673 		 * since pltime is just for autoconf, pltime processing for
674 		 * prefix is not necessary.
675 		 */
676 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
677 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
678 			struct nd_prefix *t;
679 			t = pr->ndpr_next;
680 
681 			/*
682 			 * address expiration and prefix expiration are
683 			 * separate.  NEVER perform in6_purgeaddr here.
684 			 */
685 
686 			prelist_remove(pr);
687 			pr = t;
688 		} else
689 			pr = pr->ndpr_next;
690 	}
691 	splx(s);
692 	CURVNET_RESTORE();
693 }
694 
695 /*
696  * ia6 - deprecated/invalidated temporary address
697  */
698 static int
699 regen_tmpaddr(struct in6_ifaddr *ia6)
700 {
701 	struct ifaddr *ifa;
702 	struct ifnet *ifp;
703 	struct in6_ifaddr *public_ifa6 = NULL;
704 
705 	ifp = ia6->ia_ifa.ifa_ifp;
706 	IF_ADDR_LOCK(ifp);
707 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
708 		struct in6_ifaddr *it6;
709 
710 		if (ifa->ifa_addr->sa_family != AF_INET6)
711 			continue;
712 
713 		it6 = (struct in6_ifaddr *)ifa;
714 
715 		/* ignore no autoconf addresses. */
716 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
717 			continue;
718 
719 		/* ignore autoconf addresses with different prefixes. */
720 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
721 			continue;
722 
723 		/*
724 		 * Now we are looking at an autoconf address with the same
725 		 * prefix as ours.  If the address is temporary and is still
726 		 * preferred, do not create another one.  It would be rare, but
727 		 * could happen, for example, when we resume a laptop PC after
728 		 * a long period.
729 		 */
730 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
731 		    !IFA6_IS_DEPRECATED(it6)) {
732 			public_ifa6 = NULL;
733 			break;
734 		}
735 
736 		/*
737 		 * This is a public autoconf address that has the same prefix
738 		 * as ours.  If it is preferred, keep it.  We can't break the
739 		 * loop here, because there may be a still-preferred temporary
740 		 * address with the prefix.
741 		 */
742 		if (!IFA6_IS_DEPRECATED(it6))
743 		    public_ifa6 = it6;
744 
745 		if (public_ifa6 != NULL)
746 			ifa_ref(&public_ifa6->ia_ifa);
747 	}
748 	IF_ADDR_UNLOCK(ifp);
749 
750 	if (public_ifa6 != NULL) {
751 		int e;
752 
753 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
754 			ifa_free(&public_ifa6->ia_ifa);
755 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
756 			    " tmp addr,errno=%d\n", e);
757 			return (-1);
758 		}
759 		ifa_free(&public_ifa6->ia_ifa);
760 		return (0);
761 	}
762 
763 	return (-1);
764 }
765 
766 /*
767  * Nuke neighbor cache/prefix/default router management table, right before
768  * ifp goes away.
769  */
770 void
771 nd6_purge(struct ifnet *ifp)
772 {
773 	struct nd_defrouter *dr, *ndr;
774 	struct nd_prefix *pr, *npr;
775 
776 	/*
777 	 * Nuke default router list entries toward ifp.
778 	 * We defer removal of default router list entries that is installed
779 	 * in the routing table, in order to keep additional side effects as
780 	 * small as possible.
781 	 */
782 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
783 		ndr = TAILQ_NEXT(dr, dr_entry);
784 		if (dr->installed)
785 			continue;
786 
787 		if (dr->ifp == ifp)
788 			defrtrlist_del(dr);
789 	}
790 
791 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
792 		ndr = TAILQ_NEXT(dr, dr_entry);
793 		if (!dr->installed)
794 			continue;
795 
796 		if (dr->ifp == ifp)
797 			defrtrlist_del(dr);
798 	}
799 
800 	/* Nuke prefix list entries toward ifp */
801 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
802 		npr = pr->ndpr_next;
803 		if (pr->ndpr_ifp == ifp) {
804 			/*
805 			 * Because if_detach() does *not* release prefixes
806 			 * while purging addresses the reference count will
807 			 * still be above zero. We therefore reset it to
808 			 * make sure that the prefix really gets purged.
809 			 */
810 			pr->ndpr_refcnt = 0;
811 
812 			/*
813 			 * Previously, pr->ndpr_addr is removed as well,
814 			 * but I strongly believe we don't have to do it.
815 			 * nd6_purge() is only called from in6_ifdetach(),
816 			 * which removes all the associated interface addresses
817 			 * by itself.
818 			 * (jinmei@kame.net 20010129)
819 			 */
820 			prelist_remove(pr);
821 		}
822 	}
823 
824 	/* cancel default outgoing interface setting */
825 	if (V_nd6_defifindex == ifp->if_index)
826 		nd6_setdefaultiface(0);
827 
828 	if (!V_ip6_forwarding && ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
829 		/* Refresh default router list. */
830 		defrouter_select();
831 	}
832 
833 	/* XXXXX
834 	 * We do not nuke the neighbor cache entries here any more
835 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
836 	 * nd6_purge() is invoked by in6_ifdetach() which is called
837 	 * from if_detach() where everything gets purged. So let
838 	 * in6_domifdetach() do the actual L2 table purging work.
839 	 */
840 }
841 
842 /*
843  * the caller acquires and releases the lock on the lltbls
844  * Returns the llentry locked
845  */
846 struct llentry *
847 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
848 {
849 	struct sockaddr_in6 sin6;
850 	struct llentry *ln;
851 	int llflags;
852 
853 	bzero(&sin6, sizeof(sin6));
854 	sin6.sin6_len = sizeof(struct sockaddr_in6);
855 	sin6.sin6_family = AF_INET6;
856 	sin6.sin6_addr = *addr6;
857 
858 	IF_AFDATA_LOCK_ASSERT(ifp);
859 
860 	llflags = 0;
861 	if (flags & ND6_CREATE)
862 	    llflags |= LLE_CREATE;
863 	if (flags & ND6_EXCLUSIVE)
864 	    llflags |= LLE_EXCLUSIVE;
865 
866 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
867 	if ((ln != NULL) && (llflags & LLE_CREATE))
868 		ln->ln_state = ND6_LLINFO_NOSTATE;
869 
870 	return (ln);
871 }
872 
873 /*
874  * Test whether a given IPv6 address is a neighbor or not, ignoring
875  * the actual neighbor cache.  The neighbor cache is ignored in order
876  * to not reenter the routing code from within itself.
877  */
878 static int
879 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
880 {
881 	struct nd_prefix *pr;
882 	struct ifaddr *dstaddr;
883 
884 	/*
885 	 * A link-local address is always a neighbor.
886 	 * XXX: a link does not necessarily specify a single interface.
887 	 */
888 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
889 		struct sockaddr_in6 sin6_copy;
890 		u_int32_t zone;
891 
892 		/*
893 		 * We need sin6_copy since sa6_recoverscope() may modify the
894 		 * content (XXX).
895 		 */
896 		sin6_copy = *addr;
897 		if (sa6_recoverscope(&sin6_copy))
898 			return (0); /* XXX: should be impossible */
899 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
900 			return (0);
901 		if (sin6_copy.sin6_scope_id == zone)
902 			return (1);
903 		else
904 			return (0);
905 	}
906 
907 	/*
908 	 * If the address matches one of our addresses,
909 	 * it should be a neighbor.
910 	 * If the address matches one of our on-link prefixes, it should be a
911 	 * neighbor.
912 	 */
913 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
914 		if (pr->ndpr_ifp != ifp)
915 			continue;
916 
917 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
918 			struct rtentry *rt;
919 			rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
920 			if (rt == NULL)
921 				continue;
922 			/*
923 			 * This is the case where multiple interfaces
924 			 * have the same prefix, but only one is installed
925 			 * into the routing table and that prefix entry
926 			 * is not the one being examined here. In the case
927 			 * where RADIX_MPATH is enabled, multiple route
928 			 * entries (of the same rt_key value) will be
929 			 * installed because the interface addresses all
930 			 * differ.
931 			 */
932 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
933 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
934 				RTFREE_LOCKED(rt);
935 				continue;
936 			}
937 			RTFREE_LOCKED(rt);
938 		}
939 
940 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
941 		    &addr->sin6_addr, &pr->ndpr_mask))
942 			return (1);
943 	}
944 
945 	/*
946 	 * If the address is assigned on the node of the other side of
947 	 * a p2p interface, the address should be a neighbor.
948 	 */
949 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
950 	if (dstaddr != NULL) {
951 		if (dstaddr->ifa_ifp == ifp) {
952 			ifa_free(dstaddr);
953 			return (1);
954 		}
955 		ifa_free(dstaddr);
956 	}
957 
958 	/*
959 	 * If the default router list is empty, all addresses are regarded
960 	 * as on-link, and thus, as a neighbor.
961 	 * XXX: we restrict the condition to hosts, because routers usually do
962 	 * not have the "default router list".
963 	 */
964 	if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
965 	    V_nd6_defifindex == ifp->if_index) {
966 		return (1);
967 	}
968 
969 	return (0);
970 }
971 
972 
973 /*
974  * Detect if a given IPv6 address identifies a neighbor on a given link.
975  * XXX: should take care of the destination of a p2p link?
976  */
977 int
978 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
979 {
980 	struct llentry *lle;
981 	int rc = 0;
982 
983 	IF_AFDATA_UNLOCK_ASSERT(ifp);
984 	if (nd6_is_new_addr_neighbor(addr, ifp))
985 		return (1);
986 
987 	/*
988 	 * Even if the address matches none of our addresses, it might be
989 	 * in the neighbor cache.
990 	 */
991 	IF_AFDATA_LOCK(ifp);
992 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
993 		LLE_RUNLOCK(lle);
994 		rc = 1;
995 	}
996 	IF_AFDATA_UNLOCK(ifp);
997 	return (rc);
998 }
999 
1000 /*
1001  * Free an nd6 llinfo entry.
1002  * Since the function would cause significant changes in the kernel, DO NOT
1003  * make it global, unless you have a strong reason for the change, and are sure
1004  * that the change is safe.
1005  */
1006 static struct llentry *
1007 nd6_free(struct llentry *ln, int gc)
1008 {
1009         struct llentry *next;
1010 	struct nd_defrouter *dr;
1011 	struct ifnet *ifp;
1012 
1013 	LLE_WLOCK_ASSERT(ln);
1014 
1015 	/*
1016 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1017 	 * even though it is not harmful, it was not really necessary.
1018 	 */
1019 
1020 	/* cancel timer */
1021 	nd6_llinfo_settimer_locked(ln, -1);
1022 
1023 	ifp = ln->lle_tbl->llt_ifp;
1024 
1025 	if (!V_ip6_forwarding) {
1026 
1027 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1028 
1029 		if (dr != NULL && dr->expire &&
1030 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1031 			/*
1032 			 * If the reason for the deletion is just garbage
1033 			 * collection, and the neighbor is an active default
1034 			 * router, do not delete it.  Instead, reset the GC
1035 			 * timer using the router's lifetime.
1036 			 * Simply deleting the entry would affect default
1037 			 * router selection, which is not necessarily a good
1038 			 * thing, especially when we're using router preference
1039 			 * values.
1040 			 * XXX: the check for ln_state would be redundant,
1041 			 *      but we intentionally keep it just in case.
1042 			 */
1043 			if (dr->expire > time_second)
1044 				nd6_llinfo_settimer_locked(ln,
1045 				    (dr->expire - time_second) * hz);
1046 			else
1047 				nd6_llinfo_settimer_locked(ln,
1048 				    (long)V_nd6_gctimer * hz);
1049 
1050 			next = LIST_NEXT(ln, lle_next);
1051 			LLE_REMREF(ln);
1052 			LLE_WUNLOCK(ln);
1053 			return (next);
1054 		}
1055 
1056 		if (dr) {
1057 			/*
1058 			 * Unreachablity of a router might affect the default
1059 			 * router selection and on-link detection of advertised
1060 			 * prefixes.
1061 			 */
1062 
1063 			/*
1064 			 * Temporarily fake the state to choose a new default
1065 			 * router and to perform on-link determination of
1066 			 * prefixes correctly.
1067 			 * Below the state will be set correctly,
1068 			 * or the entry itself will be deleted.
1069 			 */
1070 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1071 		}
1072 
1073 		if (ln->ln_router || dr) {
1074 
1075 			/*
1076 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1077 			 * rnh and for the calls to pfxlist_onlink_check() and
1078 			 * defrouter_select() in the block further down for calls
1079 			 * into nd6_lookup().  We still hold a ref.
1080 			 */
1081 			LLE_WUNLOCK(ln);
1082 
1083 			/*
1084 			 * rt6_flush must be called whether or not the neighbor
1085 			 * is in the Default Router List.
1086 			 * See a corresponding comment in nd6_na_input().
1087 			 */
1088 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1089 		}
1090 
1091 		if (dr) {
1092 			/*
1093 			 * Since defrouter_select() does not affect the
1094 			 * on-link determination and MIP6 needs the check
1095 			 * before the default router selection, we perform
1096 			 * the check now.
1097 			 */
1098 			pfxlist_onlink_check();
1099 
1100 			/*
1101 			 * Refresh default router list.
1102 			 */
1103 			defrouter_select();
1104 		}
1105 
1106 		if (ln->ln_router || dr)
1107 			LLE_WLOCK(ln);
1108 	}
1109 
1110 	/*
1111 	 * Before deleting the entry, remember the next entry as the
1112 	 * return value.  We need this because pfxlist_onlink_check() above
1113 	 * might have freed other entries (particularly the old next entry) as
1114 	 * a side effect (XXX).
1115 	 */
1116 	next = LIST_NEXT(ln, lle_next);
1117 
1118 	/*
1119 	 * Save to unlock. We still hold an extra reference and will not
1120 	 * free(9) in llentry_free() if someone else holds one as well.
1121 	 */
1122 	LLE_WUNLOCK(ln);
1123 	IF_AFDATA_LOCK(ifp);
1124 	LLE_WLOCK(ln);
1125 	LLE_REMREF(ln);
1126 	llentry_free(ln);
1127 	IF_AFDATA_UNLOCK(ifp);
1128 
1129 	return (next);
1130 }
1131 
1132 /*
1133  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1134  *
1135  * XXX cost-effective methods?
1136  */
1137 void
1138 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1139 {
1140 	struct llentry *ln;
1141 	struct ifnet *ifp;
1142 
1143 	if ((dst6 == NULL) || (rt == NULL))
1144 		return;
1145 
1146 	ifp = rt->rt_ifp;
1147 	IF_AFDATA_LOCK(ifp);
1148 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1149 	IF_AFDATA_UNLOCK(ifp);
1150 	if (ln == NULL)
1151 		return;
1152 
1153 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1154 		goto done;
1155 
1156 	/*
1157 	 * if we get upper-layer reachability confirmation many times,
1158 	 * it is possible we have false information.
1159 	 */
1160 	if (!force) {
1161 		ln->ln_byhint++;
1162 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1163 			goto done;
1164 		}
1165 	}
1166 
1167  	ln->ln_state = ND6_LLINFO_REACHABLE;
1168 	if (!ND6_LLINFO_PERMANENT(ln)) {
1169 		nd6_llinfo_settimer_locked(ln,
1170 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1171 	}
1172 done:
1173 	LLE_WUNLOCK(ln);
1174 }
1175 
1176 
1177 int
1178 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1179 {
1180 	struct in6_drlist *drl = (struct in6_drlist *)data;
1181 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1182 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1183 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1184 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1185 	struct nd_defrouter *dr;
1186 	struct nd_prefix *pr;
1187 	int i = 0, error = 0;
1188 	int s;
1189 
1190 	switch (cmd) {
1191 	case SIOCGDRLST_IN6:
1192 		/*
1193 		 * obsolete API, use sysctl under net.inet6.icmp6
1194 		 */
1195 		bzero(drl, sizeof(*drl));
1196 		s = splnet();
1197 		dr = TAILQ_FIRST(&V_nd_defrouter);
1198 		while (dr && i < DRLSTSIZ) {
1199 			drl->defrouter[i].rtaddr = dr->rtaddr;
1200 			in6_clearscope(&drl->defrouter[i].rtaddr);
1201 
1202 			drl->defrouter[i].flags = dr->flags;
1203 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1204 			drl->defrouter[i].expire = dr->expire;
1205 			drl->defrouter[i].if_index = dr->ifp->if_index;
1206 			i++;
1207 			dr = TAILQ_NEXT(dr, dr_entry);
1208 		}
1209 		splx(s);
1210 		break;
1211 	case SIOCGPRLST_IN6:
1212 		/*
1213 		 * obsolete API, use sysctl under net.inet6.icmp6
1214 		 *
1215 		 * XXX the structure in6_prlist was changed in backward-
1216 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1217 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1218 		 */
1219 		/*
1220 		 * XXX meaning of fields, especialy "raflags", is very
1221 		 * differnet between RA prefix list and RR/static prefix list.
1222 		 * how about separating ioctls into two?
1223 		 */
1224 		bzero(oprl, sizeof(*oprl));
1225 		s = splnet();
1226 		pr = V_nd_prefix.lh_first;
1227 		while (pr && i < PRLSTSIZ) {
1228 			struct nd_pfxrouter *pfr;
1229 			int j;
1230 
1231 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1232 			oprl->prefix[i].raflags = pr->ndpr_raf;
1233 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1234 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1235 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1236 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1237 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1238 				oprl->prefix[i].expire = 0;
1239 			else {
1240 				time_t maxexpire;
1241 
1242 				/* XXX: we assume time_t is signed. */
1243 				maxexpire = (-1) &
1244 				    ~((time_t)1 <<
1245 				    ((sizeof(maxexpire) * 8) - 1));
1246 				if (pr->ndpr_vltime <
1247 				    maxexpire - pr->ndpr_lastupdate) {
1248 					oprl->prefix[i].expire =
1249 					    pr->ndpr_lastupdate +
1250 					    pr->ndpr_vltime;
1251 				} else
1252 					oprl->prefix[i].expire = maxexpire;
1253 			}
1254 
1255 			pfr = pr->ndpr_advrtrs.lh_first;
1256 			j = 0;
1257 			while (pfr) {
1258 				if (j < DRLSTSIZ) {
1259 #define RTRADDR oprl->prefix[i].advrtr[j]
1260 					RTRADDR = pfr->router->rtaddr;
1261 					in6_clearscope(&RTRADDR);
1262 #undef RTRADDR
1263 				}
1264 				j++;
1265 				pfr = pfr->pfr_next;
1266 			}
1267 			oprl->prefix[i].advrtrs = j;
1268 			oprl->prefix[i].origin = PR_ORIG_RA;
1269 
1270 			i++;
1271 			pr = pr->ndpr_next;
1272 		}
1273 		splx(s);
1274 
1275 		break;
1276 	case OSIOCGIFINFO_IN6:
1277 #define ND	ndi->ndi
1278 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1279 		bzero(&ND, sizeof(ND));
1280 		ND.linkmtu = IN6_LINKMTU(ifp);
1281 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1282 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1283 		ND.reachable = ND_IFINFO(ifp)->reachable;
1284 		ND.retrans = ND_IFINFO(ifp)->retrans;
1285 		ND.flags = ND_IFINFO(ifp)->flags;
1286 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1287 		ND.chlim = ND_IFINFO(ifp)->chlim;
1288 		break;
1289 	case SIOCGIFINFO_IN6:
1290 		ND = *ND_IFINFO(ifp);
1291 		break;
1292 	case SIOCSIFINFO_IN6:
1293 		/*
1294 		 * used to change host variables from userland.
1295 		 * intented for a use on router to reflect RA configurations.
1296 		 */
1297 		/* 0 means 'unspecified' */
1298 		if (ND.linkmtu != 0) {
1299 			if (ND.linkmtu < IPV6_MMTU ||
1300 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1301 				error = EINVAL;
1302 				break;
1303 			}
1304 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1305 		}
1306 
1307 		if (ND.basereachable != 0) {
1308 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1309 
1310 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1311 			if (ND.basereachable != obasereachable)
1312 				ND_IFINFO(ifp)->reachable =
1313 				    ND_COMPUTE_RTIME(ND.basereachable);
1314 		}
1315 		if (ND.retrans != 0)
1316 			ND_IFINFO(ifp)->retrans = ND.retrans;
1317 		if (ND.chlim != 0)
1318 			ND_IFINFO(ifp)->chlim = ND.chlim;
1319 		/* FALLTHROUGH */
1320 	case SIOCSIFINFO_FLAGS:
1321 	{
1322 		struct ifaddr *ifa;
1323 		struct in6_ifaddr *ia;
1324 
1325 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1326 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1327 			/* ifdisabled 1->0 transision */
1328 
1329 			/*
1330 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1331 			 * has an link-local address with IN6_IFF_DUPLICATED,
1332 			 * do not clear ND6_IFF_IFDISABLED.
1333 			 * See RFC 4862, Section 5.4.5.
1334 			 */
1335 			int duplicated_linklocal = 0;
1336 
1337 			IF_ADDR_LOCK(ifp);
1338 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1339 				if (ifa->ifa_addr->sa_family != AF_INET6)
1340 					continue;
1341 				ia = (struct in6_ifaddr *)ifa;
1342 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1343 				    IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1344 					duplicated_linklocal = 1;
1345 					break;
1346 				}
1347 			}
1348 			IF_ADDR_UNLOCK(ifp);
1349 
1350 			if (duplicated_linklocal) {
1351 				ND.flags |= ND6_IFF_IFDISABLED;
1352 				log(LOG_ERR, "Cannot enable an interface"
1353 				    " with a link-local address marked"
1354 				    " duplicate.\n");
1355 			} else {
1356 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1357 				in6_if_up(ifp);
1358 			}
1359 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1360 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1361 			/* ifdisabled 0->1 transision */
1362 			/* Mark all IPv6 address as tentative. */
1363 
1364 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1365 			IF_ADDR_LOCK(ifp);
1366 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1367 				if (ifa->ifa_addr->sa_family != AF_INET6)
1368 					continue;
1369 				ia = (struct in6_ifaddr *)ifa;
1370 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1371 			}
1372 			IF_ADDR_UNLOCK(ifp);
1373 		}
1374 
1375 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) &&
1376 		    (ND.flags & ND6_IFF_AUTO_LINKLOCAL)) {
1377 			/* auto_linklocal 0->1 transision */
1378 
1379 			/* If no link-local address on ifp, configure */
1380 			ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1381 			in6_ifattach(ifp, NULL);
1382 		}
1383 	}
1384 		ND_IFINFO(ifp)->flags = ND.flags;
1385 		break;
1386 #undef ND
1387 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1388 		/* sync kernel routing table with the default router list */
1389 		defrouter_reset();
1390 		defrouter_select();
1391 		break;
1392 	case SIOCSPFXFLUSH_IN6:
1393 	{
1394 		/* flush all the prefix advertised by routers */
1395 		struct nd_prefix *pr, *next;
1396 
1397 		s = splnet();
1398 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1399 			struct in6_ifaddr *ia, *ia_next;
1400 
1401 			next = pr->ndpr_next;
1402 
1403 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1404 				continue; /* XXX */
1405 
1406 			/* do we really have to remove addresses as well? */
1407 			/* XXXRW: in6_ifaddrhead locking. */
1408 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1409 			    ia_next) {
1410 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1411 					continue;
1412 
1413 				if (ia->ia6_ndpr == pr)
1414 					in6_purgeaddr(&ia->ia_ifa);
1415 			}
1416 			prelist_remove(pr);
1417 		}
1418 		splx(s);
1419 		break;
1420 	}
1421 	case SIOCSRTRFLUSH_IN6:
1422 	{
1423 		/* flush all the default routers */
1424 		struct nd_defrouter *dr, *next;
1425 
1426 		s = splnet();
1427 		defrouter_reset();
1428 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1429 			next = TAILQ_NEXT(dr, dr_entry);
1430 			defrtrlist_del(dr);
1431 		}
1432 		defrouter_select();
1433 		splx(s);
1434 		break;
1435 	}
1436 	case SIOCGNBRINFO_IN6:
1437 	{
1438 		struct llentry *ln;
1439 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1440 
1441 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1442 			return (error);
1443 
1444 		IF_AFDATA_LOCK(ifp);
1445 		ln = nd6_lookup(&nb_addr, 0, ifp);
1446 		IF_AFDATA_UNLOCK(ifp);
1447 
1448 		if (ln == NULL) {
1449 			error = EINVAL;
1450 			break;
1451 		}
1452 		nbi->state = ln->ln_state;
1453 		nbi->asked = ln->la_asked;
1454 		nbi->isrouter = ln->ln_router;
1455 		nbi->expire = ln->la_expire;
1456 		LLE_RUNLOCK(ln);
1457 		break;
1458 	}
1459 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1460 		ndif->ifindex = V_nd6_defifindex;
1461 		break;
1462 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1463 		return (nd6_setdefaultiface(ndif->ifindex));
1464 	}
1465 	return (error);
1466 }
1467 
1468 /*
1469  * Create neighbor cache entry and cache link-layer address,
1470  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1471  *
1472  * type - ICMP6 type
1473  * code - type dependent information
1474  *
1475  * XXXXX
1476  *  The caller of this function already acquired the ndp
1477  *  cache table lock because the cache entry is returned.
1478  */
1479 struct llentry *
1480 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1481     int lladdrlen, int type, int code)
1482 {
1483 	struct llentry *ln = NULL;
1484 	int is_newentry;
1485 	int do_update;
1486 	int olladdr;
1487 	int llchange;
1488 	int flags;
1489 	int newstate = 0;
1490 	uint16_t router = 0;
1491 	struct sockaddr_in6 sin6;
1492 	struct mbuf *chain = NULL;
1493 	int static_route = 0;
1494 
1495 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1496 
1497 	if (ifp == NULL)
1498 		panic("ifp == NULL in nd6_cache_lladdr");
1499 	if (from == NULL)
1500 		panic("from == NULL in nd6_cache_lladdr");
1501 
1502 	/* nothing must be updated for unspecified address */
1503 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1504 		return NULL;
1505 
1506 	/*
1507 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1508 	 * the caller.
1509 	 *
1510 	 * XXX If the link does not have link-layer adderss, what should
1511 	 * we do? (ifp->if_addrlen == 0)
1512 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1513 	 * description on it in NS section (RFC 2461 7.2.3).
1514 	 */
1515 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1516 	IF_AFDATA_LOCK(ifp);
1517 	ln = nd6_lookup(from, flags, ifp);
1518 
1519 	if (ln == NULL) {
1520 		flags |= ND6_EXCLUSIVE;
1521 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1522 		IF_AFDATA_UNLOCK(ifp);
1523 		is_newentry = 1;
1524 	} else {
1525 		IF_AFDATA_UNLOCK(ifp);
1526 		/* do nothing if static ndp is set */
1527 		if (ln->la_flags & LLE_STATIC) {
1528 			static_route = 1;
1529 			goto done;
1530 		}
1531 		is_newentry = 0;
1532 	}
1533 	if (ln == NULL)
1534 		return (NULL);
1535 
1536 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1537 	if (olladdr && lladdr) {
1538 		llchange = bcmp(lladdr, &ln->ll_addr,
1539 		    ifp->if_addrlen);
1540 	} else
1541 		llchange = 0;
1542 
1543 	/*
1544 	 * newentry olladdr  lladdr  llchange	(*=record)
1545 	 *	0	n	n	--	(1)
1546 	 *	0	y	n	--	(2)
1547 	 *	0	n	y	--	(3) * STALE
1548 	 *	0	y	y	n	(4) *
1549 	 *	0	y	y	y	(5) * STALE
1550 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1551 	 *	1	--	y	--	(7) * STALE
1552 	 */
1553 
1554 	if (lladdr) {		/* (3-5) and (7) */
1555 		/*
1556 		 * Record source link-layer address
1557 		 * XXX is it dependent to ifp->if_type?
1558 		 */
1559 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1560 		ln->la_flags |= LLE_VALID;
1561 	}
1562 
1563 	if (!is_newentry) {
1564 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1565 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1566 			do_update = 1;
1567 			newstate = ND6_LLINFO_STALE;
1568 		} else					/* (1-2,4) */
1569 			do_update = 0;
1570 	} else {
1571 		do_update = 1;
1572 		if (lladdr == NULL)			/* (6) */
1573 			newstate = ND6_LLINFO_NOSTATE;
1574 		else					/* (7) */
1575 			newstate = ND6_LLINFO_STALE;
1576 	}
1577 
1578 	if (do_update) {
1579 		/*
1580 		 * Update the state of the neighbor cache.
1581 		 */
1582 		ln->ln_state = newstate;
1583 
1584 		if (ln->ln_state == ND6_LLINFO_STALE) {
1585 			/*
1586 			 * XXX: since nd6_output() below will cause
1587 			 * state tansition to DELAY and reset the timer,
1588 			 * we must set the timer now, although it is actually
1589 			 * meaningless.
1590 			 */
1591 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1592 
1593 			if (ln->la_hold) {
1594 				struct mbuf *m_hold, *m_hold_next;
1595 
1596 				/*
1597 				 * reset the la_hold in advance, to explicitly
1598 				 * prevent a la_hold lookup in nd6_output()
1599 				 * (wouldn't happen, though...)
1600 				 */
1601 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1602 				    m_hold; m_hold = m_hold_next) {
1603 					m_hold_next = m_hold->m_nextpkt;
1604 					m_hold->m_nextpkt = NULL;
1605 
1606 					/*
1607 					 * we assume ifp is not a p2p here, so
1608 					 * just set the 2nd argument as the
1609 					 * 1st one.
1610 					 */
1611 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1612 				}
1613 				/*
1614 				 * If we have mbufs in the chain we need to do
1615 				 * deferred transmit. Copy the address from the
1616 				 * llentry before dropping the lock down below.
1617 				 */
1618 				if (chain != NULL)
1619 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1620 			}
1621 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1622 			/* probe right away */
1623 			nd6_llinfo_settimer_locked((void *)ln, 0);
1624 		}
1625 	}
1626 
1627 	/*
1628 	 * ICMP6 type dependent behavior.
1629 	 *
1630 	 * NS: clear IsRouter if new entry
1631 	 * RS: clear IsRouter
1632 	 * RA: set IsRouter if there's lladdr
1633 	 * redir: clear IsRouter if new entry
1634 	 *
1635 	 * RA case, (1):
1636 	 * The spec says that we must set IsRouter in the following cases:
1637 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1638 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1639 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1640 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1641 	 * neighbor cache, this is similar to (6).
1642 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1643 	 *
1644 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1645 	 *							D R
1646 	 *	0	n	n	--	(1)	c   ?     s
1647 	 *	0	y	n	--	(2)	c   s     s
1648 	 *	0	n	y	--	(3)	c   s     s
1649 	 *	0	y	y	n	(4)	c   s     s
1650 	 *	0	y	y	y	(5)	c   s     s
1651 	 *	1	--	n	--	(6) c	c	c s
1652 	 *	1	--	y	--	(7) c	c   s	c s
1653 	 *
1654 	 *					(c=clear s=set)
1655 	 */
1656 	switch (type & 0xff) {
1657 	case ND_NEIGHBOR_SOLICIT:
1658 		/*
1659 		 * New entry must have is_router flag cleared.
1660 		 */
1661 		if (is_newentry)	/* (6-7) */
1662 			ln->ln_router = 0;
1663 		break;
1664 	case ND_REDIRECT:
1665 		/*
1666 		 * If the icmp is a redirect to a better router, always set the
1667 		 * is_router flag.  Otherwise, if the entry is newly created,
1668 		 * clear the flag.  [RFC 2461, sec 8.3]
1669 		 */
1670 		if (code == ND_REDIRECT_ROUTER)
1671 			ln->ln_router = 1;
1672 		else if (is_newentry) /* (6-7) */
1673 			ln->ln_router = 0;
1674 		break;
1675 	case ND_ROUTER_SOLICIT:
1676 		/*
1677 		 * is_router flag must always be cleared.
1678 		 */
1679 		ln->ln_router = 0;
1680 		break;
1681 	case ND_ROUTER_ADVERT:
1682 		/*
1683 		 * Mark an entry with lladdr as a router.
1684 		 */
1685 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1686 		    (is_newentry && lladdr)) {			/* (7) */
1687 			ln->ln_router = 1;
1688 		}
1689 		break;
1690 	}
1691 
1692 	if (ln != NULL) {
1693 		static_route = (ln->la_flags & LLE_STATIC);
1694 		router = ln->ln_router;
1695 
1696 		if (flags & ND6_EXCLUSIVE)
1697 			LLE_WUNLOCK(ln);
1698 		else
1699 			LLE_RUNLOCK(ln);
1700 		if (static_route)
1701 			ln = NULL;
1702 	}
1703 	if (chain)
1704 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1705 
1706 	/*
1707 	 * When the link-layer address of a router changes, select the
1708 	 * best router again.  In particular, when the neighbor entry is newly
1709 	 * created, it might affect the selection policy.
1710 	 * Question: can we restrict the first condition to the "is_newentry"
1711 	 * case?
1712 	 * XXX: when we hear an RA from a new router with the link-layer
1713 	 * address option, defrouter_select() is called twice, since
1714 	 * defrtrlist_update called the function as well.  However, I believe
1715 	 * we can compromise the overhead, since it only happens the first
1716 	 * time.
1717 	 * XXX: although defrouter_select() should not have a bad effect
1718 	 * for those are not autoconfigured hosts, we explicitly avoid such
1719 	 * cases for safety.
1720 	 */
1721 	if (do_update && router && !V_ip6_forwarding &&
1722 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1723 		/*
1724 		 * guaranteed recursion
1725 		 */
1726 		defrouter_select();
1727 	}
1728 
1729 	return (ln);
1730 done:
1731 	if (ln != NULL) {
1732 		if (flags & ND6_EXCLUSIVE)
1733 			LLE_WUNLOCK(ln);
1734 		else
1735 			LLE_RUNLOCK(ln);
1736 		if (static_route)
1737 			ln = NULL;
1738 	}
1739 	return (ln);
1740 }
1741 
1742 static void
1743 nd6_slowtimo(void *arg)
1744 {
1745 	CURVNET_SET((struct vnet *) arg);
1746 	struct nd_ifinfo *nd6if;
1747 	struct ifnet *ifp;
1748 
1749 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1750 	    nd6_slowtimo, curvnet);
1751 	IFNET_RLOCK_NOSLEEP();
1752 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1753 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1754 		nd6if = ND_IFINFO(ifp);
1755 		if (nd6if->basereachable && /* already initialized */
1756 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1757 			/*
1758 			 * Since reachable time rarely changes by router
1759 			 * advertisements, we SHOULD insure that a new random
1760 			 * value gets recomputed at least once every few hours.
1761 			 * (RFC 2461, 6.3.4)
1762 			 */
1763 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1764 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1765 		}
1766 	}
1767 	IFNET_RUNLOCK_NOSLEEP();
1768 	CURVNET_RESTORE();
1769 }
1770 
1771 int
1772 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1773     struct sockaddr_in6 *dst, struct rtentry *rt0)
1774 {
1775 
1776 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1777 }
1778 
1779 
1780 /*
1781  * Note that I'm not enforcing any global serialization
1782  * lle state or asked changes here as the logic is too
1783  * complicated to avoid having to always acquire an exclusive
1784  * lock
1785  * KMM
1786  *
1787  */
1788 #define senderr(e) { error = (e); goto bad;}
1789 
1790 int
1791 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1792     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1793 	struct mbuf **chain)
1794 {
1795 	struct mbuf *m = m0;
1796 	struct m_tag *mtag;
1797 	struct llentry *ln = lle;
1798 	struct ip6_hdr *ip6;
1799 	int error = 0;
1800 	int flags = 0;
1801 	int ip6len;
1802 
1803 #ifdef INVARIANTS
1804 	if (lle != NULL) {
1805 
1806 		LLE_WLOCK_ASSERT(lle);
1807 
1808 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1809 	}
1810 #endif
1811 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1812 		goto sendpkt;
1813 
1814 	if (nd6_need_cache(ifp) == 0)
1815 		goto sendpkt;
1816 
1817 	/*
1818 	 * next hop determination.  This routine is derived from ether_output.
1819 	 */
1820 
1821 	/*
1822 	 * Address resolution or Neighbor Unreachability Detection
1823 	 * for the next hop.
1824 	 * At this point, the destination of the packet must be a unicast
1825 	 * or an anycast address(i.e. not a multicast).
1826 	 */
1827 
1828 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1829 	if (ln == NULL) {
1830 	retry:
1831 		IF_AFDATA_LOCK(ifp);
1832 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1833 		IF_AFDATA_UNLOCK(ifp);
1834 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1835 			/*
1836 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1837 			 * the condition below is not very efficient.  But we believe
1838 			 * it is tolerable, because this should be a rare case.
1839 			 */
1840 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1841 			IF_AFDATA_LOCK(ifp);
1842 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1843 			IF_AFDATA_UNLOCK(ifp);
1844 		}
1845 	}
1846 	if (ln == NULL) {
1847 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1848 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1849 			char ip6buf[INET6_ADDRSTRLEN];
1850 			log(LOG_DEBUG,
1851 			    "nd6_output: can't allocate llinfo for %s "
1852 			    "(ln=%p)\n",
1853 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1854 			senderr(EIO);	/* XXX: good error? */
1855 		}
1856 		goto sendpkt;	/* send anyway */
1857 	}
1858 
1859 	/* We don't have to do link-layer address resolution on a p2p link. */
1860 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1861 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1862 		if ((flags & LLE_EXCLUSIVE) == 0) {
1863 			flags |= LLE_EXCLUSIVE;
1864 			goto retry;
1865 		}
1866 		ln->ln_state = ND6_LLINFO_STALE;
1867 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1868 	}
1869 
1870 	/*
1871 	 * The first time we send a packet to a neighbor whose entry is
1872 	 * STALE, we have to change the state to DELAY and a sets a timer to
1873 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1874 	 * neighbor unreachability detection on expiration.
1875 	 * (RFC 2461 7.3.3)
1876 	 */
1877 	if (ln->ln_state == ND6_LLINFO_STALE) {
1878 		if ((flags & LLE_EXCLUSIVE) == 0) {
1879 			flags |= LLE_EXCLUSIVE;
1880 			LLE_RUNLOCK(ln);
1881 			goto retry;
1882 		}
1883 		ln->la_asked = 0;
1884 		ln->ln_state = ND6_LLINFO_DELAY;
1885 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1886 	}
1887 
1888 	/*
1889 	 * If the neighbor cache entry has a state other than INCOMPLETE
1890 	 * (i.e. its link-layer address is already resolved), just
1891 	 * send the packet.
1892 	 */
1893 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1894 		goto sendpkt;
1895 
1896 	/*
1897 	 * There is a neighbor cache entry, but no ethernet address
1898 	 * response yet.  Append this latest packet to the end of the
1899 	 * packet queue in the mbuf, unless the number of the packet
1900 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1901 	 * the oldest packet in the queue will be removed.
1902 	 */
1903 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1904 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1905 
1906 	if ((flags & LLE_EXCLUSIVE) == 0) {
1907 		flags |= LLE_EXCLUSIVE;
1908 		LLE_RUNLOCK(ln);
1909 		goto retry;
1910 	}
1911 
1912 	LLE_WLOCK_ASSERT(ln);
1913 
1914 	if (ln->la_hold) {
1915 		struct mbuf *m_hold;
1916 		int i;
1917 
1918 		i = 0;
1919 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1920 			i++;
1921 			if (m_hold->m_nextpkt == NULL) {
1922 				m_hold->m_nextpkt = m;
1923 				break;
1924 			}
1925 		}
1926 		while (i >= V_nd6_maxqueuelen) {
1927 			m_hold = ln->la_hold;
1928 			ln->la_hold = ln->la_hold->m_nextpkt;
1929 			m_freem(m_hold);
1930 			i--;
1931 		}
1932 	} else {
1933 		ln->la_hold = m;
1934 	}
1935 
1936 	/*
1937 	 * If there has been no NS for the neighbor after entering the
1938 	 * INCOMPLETE state, send the first solicitation.
1939 	 */
1940 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1941 		ln->la_asked++;
1942 
1943 		nd6_llinfo_settimer_locked(ln,
1944 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1945 		LLE_WUNLOCK(ln);
1946 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1947 		if (lle != NULL && ln == lle)
1948 			LLE_WLOCK(lle);
1949 
1950 	} else if (lle == NULL || ln != lle) {
1951 		/*
1952 		 * We did the lookup (no lle arg) so we
1953 		 * need to do the unlock here.
1954 		 */
1955 		LLE_WUNLOCK(ln);
1956 	}
1957 
1958 	return (0);
1959 
1960   sendpkt:
1961 	/* discard the packet if IPv6 operation is disabled on the interface */
1962 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1963 		error = ENETDOWN; /* better error? */
1964 		goto bad;
1965 	}
1966 	/*
1967 	 * ln is valid and the caller did not pass in
1968 	 * an llentry
1969 	 */
1970 	if ((ln != NULL) && (lle == NULL)) {
1971 		if (flags & LLE_EXCLUSIVE)
1972 			LLE_WUNLOCK(ln);
1973 		else
1974 			LLE_RUNLOCK(ln);
1975 	}
1976 
1977 #ifdef MAC
1978 	mac_netinet6_nd6_send(ifp, m);
1979 #endif
1980 
1981 	/*
1982 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1983 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1984 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1985 	 * to be diverted to user space.  When re-injected into the kernel,
1986 	 * send_output() will directly dispatch them to the outgoing interface.
1987 	 */
1988 	if (send_sendso_input_hook != NULL) {
1989 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1990 		if (mtag != NULL) {
1991 			ip6 = mtod(m, struct ip6_hdr *);
1992 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1993 			/* Use the SEND socket */
1994 			error = send_sendso_input_hook(m, ifp, SND_OUT,
1995 			    ip6len);
1996 			/* -1 == no app on SEND socket */
1997 			if (error == 0 || error != -1)
1998 			    return (error);
1999 		}
2000 	}
2001 
2002 	/*
2003 	 * We were passed in a pointer to an lle with the lock held
2004 	 * this means that we can't call if_output as we will
2005 	 * recurse on the lle lock - so what we do is we create
2006 	 * a list of mbufs to send and transmit them in the caller
2007 	 * after the lock is dropped
2008 	 */
2009 	if (lle != NULL) {
2010 		if (*chain == NULL)
2011 			*chain = m;
2012 		else {
2013 			struct mbuf *m = *chain;
2014 
2015 			/*
2016 			 * append mbuf to end of deferred chain
2017 			 */
2018 			while (m->m_nextpkt != NULL)
2019 				m = m->m_nextpkt;
2020 			m->m_nextpkt = m;
2021 		}
2022 		return (error);
2023 	}
2024 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2025 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2026 		    NULL));
2027 	}
2028 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2029 	return (error);
2030 
2031   bad:
2032 	/*
2033 	 * ln is valid and the caller did not pass in
2034 	 * an llentry
2035 	 */
2036 	if ((ln != NULL) && (lle == NULL)) {
2037 		if (flags & LLE_EXCLUSIVE)
2038 			LLE_WUNLOCK(ln);
2039 		else
2040 			LLE_RUNLOCK(ln);
2041 	}
2042 	if (m)
2043 		m_freem(m);
2044 	return (error);
2045 }
2046 #undef senderr
2047 
2048 
2049 int
2050 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2051     struct sockaddr_in6 *dst, struct route *ro)
2052 {
2053 	struct mbuf *m, *m_head;
2054 	struct ifnet *outifp;
2055 	int error = 0;
2056 
2057 	m_head = chain;
2058 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2059 		outifp = origifp;
2060 	else
2061 		outifp = ifp;
2062 
2063 	while (m_head) {
2064 		m = m_head;
2065 		m_head = m_head->m_nextpkt;
2066 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2067 	}
2068 
2069 	/*
2070 	 * XXX
2071 	 * note that intermediate errors are blindly ignored - but this is
2072 	 * the same convention as used with nd6_output when called by
2073 	 * nd6_cache_lladdr
2074 	 */
2075 	return (error);
2076 }
2077 
2078 
2079 int
2080 nd6_need_cache(struct ifnet *ifp)
2081 {
2082 	/*
2083 	 * XXX: we currently do not make neighbor cache on any interface
2084 	 * other than ARCnet, Ethernet, FDDI and GIF.
2085 	 *
2086 	 * RFC2893 says:
2087 	 * - unidirectional tunnels needs no ND
2088 	 */
2089 	switch (ifp->if_type) {
2090 	case IFT_ARCNET:
2091 	case IFT_ETHER:
2092 	case IFT_FDDI:
2093 	case IFT_IEEE1394:
2094 #ifdef IFT_L2VLAN
2095 	case IFT_L2VLAN:
2096 #endif
2097 #ifdef IFT_IEEE80211
2098 	case IFT_IEEE80211:
2099 #endif
2100 #ifdef IFT_CARP
2101 	case IFT_CARP:
2102 #endif
2103 	case IFT_INFINIBAND:
2104 	case IFT_GIF:		/* XXX need more cases? */
2105 	case IFT_PPP:
2106 	case IFT_TUNNEL:
2107 	case IFT_BRIDGE:
2108 	case IFT_PROPVIRTUAL:
2109 		return (1);
2110 	default:
2111 		return (0);
2112 	}
2113 }
2114 
2115 /*
2116  * the callers of this function need to be re-worked to drop
2117  * the lle lock, drop here for now
2118  */
2119 int
2120 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2121     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2122 {
2123 	struct llentry *ln;
2124 
2125 	*lle = NULL;
2126 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2127 	if (m->m_flags & M_MCAST) {
2128 		int i;
2129 
2130 		switch (ifp->if_type) {
2131 		case IFT_ETHER:
2132 		case IFT_FDDI:
2133 #ifdef IFT_L2VLAN
2134 		case IFT_L2VLAN:
2135 #endif
2136 #ifdef IFT_IEEE80211
2137 		case IFT_IEEE80211:
2138 #endif
2139 		case IFT_BRIDGE:
2140 		case IFT_ISO88025:
2141 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2142 						 desten);
2143 			return (0);
2144 		case IFT_IEEE1394:
2145 			/*
2146 			 * netbsd can use if_broadcastaddr, but we don't do so
2147 			 * to reduce # of ifdef.
2148 			 */
2149 			for (i = 0; i < ifp->if_addrlen; i++)
2150 				desten[i] = ~0;
2151 			return (0);
2152 		case IFT_ARCNET:
2153 			*desten = 0;
2154 			return (0);
2155 		default:
2156 			m_freem(m);
2157 			return (EAFNOSUPPORT);
2158 		}
2159 	}
2160 
2161 
2162 	/*
2163 	 * the entry should have been created in nd6_store_lladdr
2164 	 */
2165 	IF_AFDATA_LOCK(ifp);
2166 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2167 	IF_AFDATA_UNLOCK(ifp);
2168 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2169 		if (ln != NULL)
2170 			LLE_RUNLOCK(ln);
2171 		/* this could happen, if we could not allocate memory */
2172 		m_freem(m);
2173 		return (1);
2174 	}
2175 
2176 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2177 	*lle = ln;
2178 	LLE_RUNLOCK(ln);
2179 	/*
2180 	 * A *small* use after free race exists here
2181 	 */
2182 	return (0);
2183 }
2184 
2185 static void
2186 clear_llinfo_pqueue(struct llentry *ln)
2187 {
2188 	struct mbuf *m_hold, *m_hold_next;
2189 
2190 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2191 		m_hold_next = m_hold->m_nextpkt;
2192 		m_hold->m_nextpkt = NULL;
2193 		m_freem(m_hold);
2194 	}
2195 
2196 	ln->la_hold = NULL;
2197 	return;
2198 }
2199 
2200 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2201 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2202 #ifdef SYSCTL_DECL
2203 SYSCTL_DECL(_net_inet6_icmp6);
2204 #endif
2205 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2206 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2207 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2208 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2209 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2210 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2211 
2212 static int
2213 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2214 {
2215 	int error;
2216 	char buf[1024] __aligned(4);
2217 	struct in6_defrouter *d, *de;
2218 	struct nd_defrouter *dr;
2219 
2220 	if (req->newptr)
2221 		return EPERM;
2222 	error = 0;
2223 
2224 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2225 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2226 		d = (struct in6_defrouter *)buf;
2227 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2228 
2229 		if (d + 1 <= de) {
2230 			bzero(d, sizeof(*d));
2231 			d->rtaddr.sin6_family = AF_INET6;
2232 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2233 			d->rtaddr.sin6_addr = dr->rtaddr;
2234 			error = sa6_recoverscope(&d->rtaddr);
2235 			if (error != 0)
2236 				return (error);
2237 			d->flags = dr->flags;
2238 			d->rtlifetime = dr->rtlifetime;
2239 			d->expire = dr->expire;
2240 			d->if_index = dr->ifp->if_index;
2241 		} else
2242 			panic("buffer too short");
2243 
2244 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2245 		if (error)
2246 			break;
2247 	}
2248 
2249 	return (error);
2250 }
2251 
2252 static int
2253 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2254 {
2255 	int error;
2256 	char buf[1024] __aligned(4);
2257 	struct in6_prefix *p, *pe;
2258 	struct nd_prefix *pr;
2259 	char ip6buf[INET6_ADDRSTRLEN];
2260 
2261 	if (req->newptr)
2262 		return EPERM;
2263 	error = 0;
2264 
2265 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2266 		u_short advrtrs;
2267 		size_t advance;
2268 		struct sockaddr_in6 *sin6, *s6;
2269 		struct nd_pfxrouter *pfr;
2270 
2271 		p = (struct in6_prefix *)buf;
2272 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2273 
2274 		if (p + 1 <= pe) {
2275 			bzero(p, sizeof(*p));
2276 			sin6 = (struct sockaddr_in6 *)(p + 1);
2277 
2278 			p->prefix = pr->ndpr_prefix;
2279 			if (sa6_recoverscope(&p->prefix)) {
2280 				log(LOG_ERR,
2281 				    "scope error in prefix list (%s)\n",
2282 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2283 				/* XXX: press on... */
2284 			}
2285 			p->raflags = pr->ndpr_raf;
2286 			p->prefixlen = pr->ndpr_plen;
2287 			p->vltime = pr->ndpr_vltime;
2288 			p->pltime = pr->ndpr_pltime;
2289 			p->if_index = pr->ndpr_ifp->if_index;
2290 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2291 				p->expire = 0;
2292 			else {
2293 				time_t maxexpire;
2294 
2295 				/* XXX: we assume time_t is signed. */
2296 				maxexpire = (-1) &
2297 				    ~((time_t)1 <<
2298 				    ((sizeof(maxexpire) * 8) - 1));
2299 				if (pr->ndpr_vltime <
2300 				    maxexpire - pr->ndpr_lastupdate) {
2301 				    p->expire = pr->ndpr_lastupdate +
2302 				        pr->ndpr_vltime;
2303 				} else
2304 					p->expire = maxexpire;
2305 			}
2306 			p->refcnt = pr->ndpr_refcnt;
2307 			p->flags = pr->ndpr_stateflags;
2308 			p->origin = PR_ORIG_RA;
2309 			advrtrs = 0;
2310 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2311 			     pfr = pfr->pfr_next) {
2312 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2313 					advrtrs++;
2314 					continue;
2315 				}
2316 				s6 = &sin6[advrtrs];
2317 				bzero(s6, sizeof(*s6));
2318 				s6->sin6_family = AF_INET6;
2319 				s6->sin6_len = sizeof(*sin6);
2320 				s6->sin6_addr = pfr->router->rtaddr;
2321 				if (sa6_recoverscope(s6)) {
2322 					log(LOG_ERR,
2323 					    "scope error in "
2324 					    "prefix list (%s)\n",
2325 					    ip6_sprintf(ip6buf,
2326 						    &pfr->router->rtaddr));
2327 				}
2328 				advrtrs++;
2329 			}
2330 			p->advrtrs = advrtrs;
2331 		} else
2332 			panic("buffer too short");
2333 
2334 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2335 		error = SYSCTL_OUT(req, buf, advance);
2336 		if (error)
2337 			break;
2338 	}
2339 
2340 	return (error);
2341 }
2342