xref: /freebsd/sys/netinet6/nd6.c (revision 2e1417489338b971e5fd599ff48b5f65df9e8d3b)
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;
576 	struct nd_prefix *pr;
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 	dr = TAILQ_FIRST(&V_nd_defrouter);
586 	while (dr) {
587 		if (dr->expire && dr->expire < time_second) {
588 			struct nd_defrouter *t;
589 			t = TAILQ_NEXT(dr, dr_entry);
590 			defrtrlist_del(dr);
591 			dr = t;
592 		} else {
593 			dr = TAILQ_NEXT(dr, dr_entry);
594 		}
595 	}
596 
597 	/*
598 	 * expire interface addresses.
599 	 * in the past the loop was inside prefix expiry processing.
600 	 * However, from a stricter speci-confrmance standpoint, we should
601 	 * rather separate address lifetimes and prefix lifetimes.
602 	 *
603 	 * XXXRW: in6_ifaddrhead locking.
604 	 */
605   addrloop:
606 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
607 		/* check address lifetime */
608 		lt6 = &ia6->ia6_lifetime;
609 		if (IFA6_IS_INVALID(ia6)) {
610 			int regen = 0;
611 
612 			/*
613 			 * If the expiring address is temporary, try
614 			 * regenerating a new one.  This would be useful when
615 			 * we suspended a laptop PC, then turned it on after a
616 			 * period that could invalidate all temporary
617 			 * addresses.  Although we may have to restart the
618 			 * loop (see below), it must be after purging the
619 			 * address.  Otherwise, we'd see an infinite loop of
620 			 * regeneration.
621 			 */
622 			if (V_ip6_use_tempaddr &&
623 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
624 				if (regen_tmpaddr(ia6) == 0)
625 					regen = 1;
626 			}
627 
628 			in6_purgeaddr(&ia6->ia_ifa);
629 
630 			if (regen)
631 				goto addrloop; /* XXX: see below */
632 		} else if (IFA6_IS_DEPRECATED(ia6)) {
633 			int oldflags = ia6->ia6_flags;
634 
635 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
636 
637 			/*
638 			 * If a temporary address has just become deprecated,
639 			 * regenerate a new one if possible.
640 			 */
641 			if (V_ip6_use_tempaddr &&
642 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
643 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
644 
645 				if (regen_tmpaddr(ia6) == 0) {
646 					/*
647 					 * A new temporary address is
648 					 * generated.
649 					 * XXX: this means the address chain
650 					 * has changed while we are still in
651 					 * the loop.  Although the change
652 					 * would not cause disaster (because
653 					 * it's not a deletion, but an
654 					 * addition,) we'd rather restart the
655 					 * loop just for safety.  Or does this
656 					 * significantly reduce performance??
657 					 */
658 					goto addrloop;
659 				}
660 			}
661 		} else {
662 			/*
663 			 * A new RA might have made a deprecated address
664 			 * preferred.
665 			 */
666 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
667 		}
668 	}
669 
670 	/* expire prefix list */
671 	pr = V_nd_prefix.lh_first;
672 	while (pr) {
673 		/*
674 		 * check prefix lifetime.
675 		 * since pltime is just for autoconf, pltime processing for
676 		 * prefix is not necessary.
677 		 */
678 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
679 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
680 			struct nd_prefix *t;
681 			t = pr->ndpr_next;
682 
683 			/*
684 			 * address expiration and prefix expiration are
685 			 * separate.  NEVER perform in6_purgeaddr here.
686 			 */
687 
688 			prelist_remove(pr);
689 			pr = t;
690 		} else
691 			pr = pr->ndpr_next;
692 	}
693 	splx(s);
694 	CURVNET_RESTORE();
695 }
696 
697 /*
698  * ia6 - deprecated/invalidated temporary address
699  */
700 static int
701 regen_tmpaddr(struct in6_ifaddr *ia6)
702 {
703 	struct ifaddr *ifa;
704 	struct ifnet *ifp;
705 	struct in6_ifaddr *public_ifa6 = NULL;
706 
707 	ifp = ia6->ia_ifa.ifa_ifp;
708 	IF_ADDR_LOCK(ifp);
709 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
710 		struct in6_ifaddr *it6;
711 
712 		if (ifa->ifa_addr->sa_family != AF_INET6)
713 			continue;
714 
715 		it6 = (struct in6_ifaddr *)ifa;
716 
717 		/* ignore no autoconf addresses. */
718 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
719 			continue;
720 
721 		/* ignore autoconf addresses with different prefixes. */
722 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
723 			continue;
724 
725 		/*
726 		 * Now we are looking at an autoconf address with the same
727 		 * prefix as ours.  If the address is temporary and is still
728 		 * preferred, do not create another one.  It would be rare, but
729 		 * could happen, for example, when we resume a laptop PC after
730 		 * a long period.
731 		 */
732 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
733 		    !IFA6_IS_DEPRECATED(it6)) {
734 			public_ifa6 = NULL;
735 			break;
736 		}
737 
738 		/*
739 		 * This is a public autoconf address that has the same prefix
740 		 * as ours.  If it is preferred, keep it.  We can't break the
741 		 * loop here, because there may be a still-preferred temporary
742 		 * address with the prefix.
743 		 */
744 		if (!IFA6_IS_DEPRECATED(it6))
745 		    public_ifa6 = it6;
746 
747 		if (public_ifa6 != NULL)
748 			ifa_ref(&public_ifa6->ia_ifa);
749 	}
750 	IF_ADDR_UNLOCK(ifp);
751 
752 	if (public_ifa6 != NULL) {
753 		int e;
754 
755 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
756 			ifa_free(&public_ifa6->ia_ifa);
757 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
758 			    " tmp addr,errno=%d\n", e);
759 			return (-1);
760 		}
761 		ifa_free(&public_ifa6->ia_ifa);
762 		return (0);
763 	}
764 
765 	return (-1);
766 }
767 
768 /*
769  * Nuke neighbor cache/prefix/default router management table, right before
770  * ifp goes away.
771  */
772 void
773 nd6_purge(struct ifnet *ifp)
774 {
775 	struct nd_defrouter *dr, *ndr;
776 	struct nd_prefix *pr, *npr;
777 
778 	/*
779 	 * Nuke default router list entries toward ifp.
780 	 * We defer removal of default router list entries that is installed
781 	 * in the routing table, in order to keep additional side effects as
782 	 * small as possible.
783 	 */
784 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
785 		ndr = TAILQ_NEXT(dr, dr_entry);
786 		if (dr->installed)
787 			continue;
788 
789 		if (dr->ifp == ifp)
790 			defrtrlist_del(dr);
791 	}
792 
793 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
794 		ndr = TAILQ_NEXT(dr, dr_entry);
795 		if (!dr->installed)
796 			continue;
797 
798 		if (dr->ifp == ifp)
799 			defrtrlist_del(dr);
800 	}
801 
802 	/* Nuke prefix list entries toward ifp */
803 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
804 		npr = pr->ndpr_next;
805 		if (pr->ndpr_ifp == ifp) {
806 			/*
807 			 * Because if_detach() does *not* release prefixes
808 			 * while purging addresses the reference count will
809 			 * still be above zero. We therefore reset it to
810 			 * make sure that the prefix really gets purged.
811 			 */
812 			pr->ndpr_refcnt = 0;
813 
814 			/*
815 			 * Previously, pr->ndpr_addr is removed as well,
816 			 * but I strongly believe we don't have to do it.
817 			 * nd6_purge() is only called from in6_ifdetach(),
818 			 * which removes all the associated interface addresses
819 			 * by itself.
820 			 * (jinmei@kame.net 20010129)
821 			 */
822 			prelist_remove(pr);
823 		}
824 	}
825 
826 	/* cancel default outgoing interface setting */
827 	if (V_nd6_defifindex == ifp->if_index)
828 		nd6_setdefaultiface(0);
829 
830 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
831 		/* Refresh default router list. */
832 		defrouter_select();
833 	}
834 
835 	/* XXXXX
836 	 * We do not nuke the neighbor cache entries here any more
837 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
838 	 * nd6_purge() is invoked by in6_ifdetach() which is called
839 	 * from if_detach() where everything gets purged. So let
840 	 * in6_domifdetach() do the actual L2 table purging work.
841 	 */
842 }
843 
844 /*
845  * the caller acquires and releases the lock on the lltbls
846  * Returns the llentry locked
847  */
848 struct llentry *
849 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
850 {
851 	struct sockaddr_in6 sin6;
852 	struct llentry *ln;
853 	int llflags;
854 
855 	bzero(&sin6, sizeof(sin6));
856 	sin6.sin6_len = sizeof(struct sockaddr_in6);
857 	sin6.sin6_family = AF_INET6;
858 	sin6.sin6_addr = *addr6;
859 
860 	IF_AFDATA_LOCK_ASSERT(ifp);
861 
862 	llflags = 0;
863 	if (flags & ND6_CREATE)
864 	    llflags |= LLE_CREATE;
865 	if (flags & ND6_EXCLUSIVE)
866 	    llflags |= LLE_EXCLUSIVE;
867 
868 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
869 	if ((ln != NULL) && (llflags & LLE_CREATE))
870 		ln->ln_state = ND6_LLINFO_NOSTATE;
871 
872 	return (ln);
873 }
874 
875 /*
876  * Test whether a given IPv6 address is a neighbor or not, ignoring
877  * the actual neighbor cache.  The neighbor cache is ignored in order
878  * to not reenter the routing code from within itself.
879  */
880 static int
881 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
882 {
883 	struct nd_prefix *pr;
884 	struct ifaddr *dstaddr;
885 
886 	/*
887 	 * A link-local address is always a neighbor.
888 	 * XXX: a link does not necessarily specify a single interface.
889 	 */
890 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
891 		struct sockaddr_in6 sin6_copy;
892 		u_int32_t zone;
893 
894 		/*
895 		 * We need sin6_copy since sa6_recoverscope() may modify the
896 		 * content (XXX).
897 		 */
898 		sin6_copy = *addr;
899 		if (sa6_recoverscope(&sin6_copy))
900 			return (0); /* XXX: should be impossible */
901 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
902 			return (0);
903 		if (sin6_copy.sin6_scope_id == zone)
904 			return (1);
905 		else
906 			return (0);
907 	}
908 
909 	/*
910 	 * If the address matches one of our addresses,
911 	 * it should be a neighbor.
912 	 * If the address matches one of our on-link prefixes, it should be a
913 	 * neighbor.
914 	 */
915 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
916 		if (pr->ndpr_ifp != ifp)
917 			continue;
918 
919 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
920 			struct rtentry *rt;
921 			rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
922 			if (rt == NULL)
923 				continue;
924 			/*
925 			 * This is the case where multiple interfaces
926 			 * have the same prefix, but only one is installed
927 			 * into the routing table and that prefix entry
928 			 * is not the one being examined here. In the case
929 			 * where RADIX_MPATH is enabled, multiple route
930 			 * entries (of the same rt_key value) will be
931 			 * installed because the interface addresses all
932 			 * differ.
933 			 */
934 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
935 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
936 				RTFREE_LOCKED(rt);
937 				continue;
938 			}
939 			RTFREE_LOCKED(rt);
940 		}
941 
942 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
943 		    &addr->sin6_addr, &pr->ndpr_mask))
944 			return (1);
945 	}
946 
947 	/*
948 	 * If the address is assigned on the node of the other side of
949 	 * a p2p interface, the address should be a neighbor.
950 	 */
951 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
952 	if (dstaddr != NULL) {
953 		if (dstaddr->ifa_ifp == ifp) {
954 			ifa_free(dstaddr);
955 			return (1);
956 		}
957 		ifa_free(dstaddr);
958 	}
959 
960 	/*
961 	 * If the default router list is empty, all addresses are regarded
962 	 * as on-link, and thus, as a neighbor.
963 	 */
964 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
965 	    TAILQ_FIRST(&V_nd_defrouter) == NULL &&
966 	    V_nd6_defifindex == ifp->if_index) {
967 		return (1);
968 	}
969 
970 	return (0);
971 }
972 
973 
974 /*
975  * Detect if a given IPv6 address identifies a neighbor on a given link.
976  * XXX: should take care of the destination of a p2p link?
977  */
978 int
979 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
980 {
981 	struct llentry *lle;
982 	int rc = 0;
983 
984 	IF_AFDATA_UNLOCK_ASSERT(ifp);
985 	if (nd6_is_new_addr_neighbor(addr, ifp))
986 		return (1);
987 
988 	/*
989 	 * Even if the address matches none of our addresses, it might be
990 	 * in the neighbor cache.
991 	 */
992 	IF_AFDATA_LOCK(ifp);
993 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
994 		LLE_RUNLOCK(lle);
995 		rc = 1;
996 	}
997 	IF_AFDATA_UNLOCK(ifp);
998 	return (rc);
999 }
1000 
1001 /*
1002  * Free an nd6 llinfo entry.
1003  * Since the function would cause significant changes in the kernel, DO NOT
1004  * make it global, unless you have a strong reason for the change, and are sure
1005  * that the change is safe.
1006  */
1007 static struct llentry *
1008 nd6_free(struct llentry *ln, int gc)
1009 {
1010         struct llentry *next;
1011 	struct nd_defrouter *dr;
1012 	struct ifnet *ifp;
1013 
1014 	LLE_WLOCK_ASSERT(ln);
1015 
1016 	/*
1017 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1018 	 * even though it is not harmful, it was not really necessary.
1019 	 */
1020 
1021 	/* cancel timer */
1022 	nd6_llinfo_settimer_locked(ln, -1);
1023 
1024 	ifp = ln->lle_tbl->llt_ifp;
1025 
1026 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
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 /*
1178  * Rejuvenate this function for routing operations related
1179  * processing.
1180  */
1181 void
1182 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1183 {
1184 	struct sockaddr_in6 *gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1185 	struct nd_defrouter *dr;
1186 	struct ifnet *ifp = rt->rt_ifp;
1187 
1188 	RT_LOCK_ASSERT(rt);
1189 
1190 	switch (req) {
1191 	case RTM_ADD:
1192 		break;
1193 
1194 	case RTM_DELETE:
1195 		if (!ifp)
1196 			return;
1197 		/*
1198 		 * Only indirect routes are interesting.
1199 		 */
1200 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1201 			return;
1202 		/*
1203 		 * check for default route
1204 		 */
1205 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1206 				       &SIN6(rt_key(rt))->sin6_addr)) {
1207 
1208 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1209 			if (dr != NULL)
1210 				dr->installed = 0;
1211 		}
1212 		break;
1213 	}
1214 }
1215 
1216 
1217 int
1218 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1219 {
1220 	struct in6_drlist *drl = (struct in6_drlist *)data;
1221 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1222 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1223 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1224 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1225 	struct nd_defrouter *dr;
1226 	struct nd_prefix *pr;
1227 	int i = 0, error = 0;
1228 	int s;
1229 
1230 	switch (cmd) {
1231 	case SIOCGDRLST_IN6:
1232 		/*
1233 		 * obsolete API, use sysctl under net.inet6.icmp6
1234 		 */
1235 		bzero(drl, sizeof(*drl));
1236 		s = splnet();
1237 		dr = TAILQ_FIRST(&V_nd_defrouter);
1238 		while (dr && i < DRLSTSIZ) {
1239 			drl->defrouter[i].rtaddr = dr->rtaddr;
1240 			in6_clearscope(&drl->defrouter[i].rtaddr);
1241 
1242 			drl->defrouter[i].flags = dr->flags;
1243 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1244 			drl->defrouter[i].expire = dr->expire;
1245 			drl->defrouter[i].if_index = dr->ifp->if_index;
1246 			i++;
1247 			dr = TAILQ_NEXT(dr, dr_entry);
1248 		}
1249 		splx(s);
1250 		break;
1251 	case SIOCGPRLST_IN6:
1252 		/*
1253 		 * obsolete API, use sysctl under net.inet6.icmp6
1254 		 *
1255 		 * XXX the structure in6_prlist was changed in backward-
1256 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1257 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1258 		 */
1259 		/*
1260 		 * XXX meaning of fields, especialy "raflags", is very
1261 		 * differnet between RA prefix list and RR/static prefix list.
1262 		 * how about separating ioctls into two?
1263 		 */
1264 		bzero(oprl, sizeof(*oprl));
1265 		s = splnet();
1266 		pr = V_nd_prefix.lh_first;
1267 		while (pr && i < PRLSTSIZ) {
1268 			struct nd_pfxrouter *pfr;
1269 			int j;
1270 
1271 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1272 			oprl->prefix[i].raflags = pr->ndpr_raf;
1273 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1274 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1275 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1276 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1277 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1278 				oprl->prefix[i].expire = 0;
1279 			else {
1280 				time_t maxexpire;
1281 
1282 				/* XXX: we assume time_t is signed. */
1283 				maxexpire = (-1) &
1284 				    ~((time_t)1 <<
1285 				    ((sizeof(maxexpire) * 8) - 1));
1286 				if (pr->ndpr_vltime <
1287 				    maxexpire - pr->ndpr_lastupdate) {
1288 					oprl->prefix[i].expire =
1289 					    pr->ndpr_lastupdate +
1290 					    pr->ndpr_vltime;
1291 				} else
1292 					oprl->prefix[i].expire = maxexpire;
1293 			}
1294 
1295 			pfr = pr->ndpr_advrtrs.lh_first;
1296 			j = 0;
1297 			while (pfr) {
1298 				if (j < DRLSTSIZ) {
1299 #define RTRADDR oprl->prefix[i].advrtr[j]
1300 					RTRADDR = pfr->router->rtaddr;
1301 					in6_clearscope(&RTRADDR);
1302 #undef RTRADDR
1303 				}
1304 				j++;
1305 				pfr = pfr->pfr_next;
1306 			}
1307 			oprl->prefix[i].advrtrs = j;
1308 			oprl->prefix[i].origin = PR_ORIG_RA;
1309 
1310 			i++;
1311 			pr = pr->ndpr_next;
1312 		}
1313 		splx(s);
1314 
1315 		break;
1316 	case OSIOCGIFINFO_IN6:
1317 #define ND	ndi->ndi
1318 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1319 		bzero(&ND, sizeof(ND));
1320 		ND.linkmtu = IN6_LINKMTU(ifp);
1321 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1322 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1323 		ND.reachable = ND_IFINFO(ifp)->reachable;
1324 		ND.retrans = ND_IFINFO(ifp)->retrans;
1325 		ND.flags = ND_IFINFO(ifp)->flags;
1326 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1327 		ND.chlim = ND_IFINFO(ifp)->chlim;
1328 		break;
1329 	case SIOCGIFINFO_IN6:
1330 		ND = *ND_IFINFO(ifp);
1331 		break;
1332 	case SIOCSIFINFO_IN6:
1333 		/*
1334 		 * used to change host variables from userland.
1335 		 * intented for a use on router to reflect RA configurations.
1336 		 */
1337 		/* 0 means 'unspecified' */
1338 		if (ND.linkmtu != 0) {
1339 			if (ND.linkmtu < IPV6_MMTU ||
1340 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1341 				error = EINVAL;
1342 				break;
1343 			}
1344 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1345 		}
1346 
1347 		if (ND.basereachable != 0) {
1348 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1349 
1350 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1351 			if (ND.basereachable != obasereachable)
1352 				ND_IFINFO(ifp)->reachable =
1353 				    ND_COMPUTE_RTIME(ND.basereachable);
1354 		}
1355 		if (ND.retrans != 0)
1356 			ND_IFINFO(ifp)->retrans = ND.retrans;
1357 		if (ND.chlim != 0)
1358 			ND_IFINFO(ifp)->chlim = ND.chlim;
1359 		/* FALLTHROUGH */
1360 	case SIOCSIFINFO_FLAGS:
1361 	{
1362 		struct ifaddr *ifa;
1363 		struct in6_ifaddr *ia;
1364 
1365 		/*
1366 		 * Try to clear ifdisabled flag when enabling
1367 		 * accept_rtadv or auto_linklocal.
1368 		 */
1369 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1370 		    !(ND.flags & ND6_IFF_IFDISABLED) &&
1371 		    (ND.flags & (ND6_IFF_ACCEPT_RTADV |
1372 		    ND6_IFF_AUTO_LINKLOCAL)))
1373 			ND.flags &= ~ND6_IFF_IFDISABLED;
1374 
1375 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1376 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1377 			/* ifdisabled 1->0 transision */
1378 
1379 			/*
1380 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1381 			 * has an link-local address with IN6_IFF_DUPLICATED,
1382 			 * do not clear ND6_IFF_IFDISABLED.
1383 			 * See RFC 4862, Section 5.4.5.
1384 			 */
1385 			int duplicated_linklocal = 0;
1386 
1387 			IF_ADDR_LOCK(ifp);
1388 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1389 				if (ifa->ifa_addr->sa_family != AF_INET6)
1390 					continue;
1391 				ia = (struct in6_ifaddr *)ifa;
1392 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1393 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1394 					duplicated_linklocal = 1;
1395 					break;
1396 				}
1397 			}
1398 			IF_ADDR_UNLOCK(ifp);
1399 
1400 			if (duplicated_linklocal) {
1401 				ND.flags |= ND6_IFF_IFDISABLED;
1402 				log(LOG_ERR, "Cannot enable an interface"
1403 				    " with a link-local address marked"
1404 				    " duplicate.\n");
1405 			} else {
1406 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1407 				if (ifp->if_flags & IFF_UP)
1408 					in6_if_up(ifp);
1409 			}
1410 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1411 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1412 			/* ifdisabled 0->1 transision */
1413 			/* Mark all IPv6 address as tentative. */
1414 
1415 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1416 			IF_ADDR_LOCK(ifp);
1417 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1418 				if (ifa->ifa_addr->sa_family != AF_INET6)
1419 					continue;
1420 				ia = (struct in6_ifaddr *)ifa;
1421 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1422 			}
1423 			IF_ADDR_UNLOCK(ifp);
1424 		}
1425 
1426 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1427 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1428 				/* auto_linklocal 0->1 transision */
1429 
1430 				/* If no link-local address on ifp, configure */
1431 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1432 				in6_ifattach(ifp, NULL);
1433 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1434 			    ifp->if_flags & IFF_UP) {
1435 				/*
1436 				 * When the IF already has
1437 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1438 				 * address is assigned, and IFF_UP, try to
1439 				 * assign one.
1440 				 */
1441 				int haslinklocal = 0;
1442 
1443 				IF_ADDR_LOCK(ifp);
1444 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1445 					if (ifa->ifa_addr->sa_family != AF_INET6)
1446 						continue;
1447 					ia = (struct in6_ifaddr *)ifa;
1448 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1449 						haslinklocal = 1;
1450 						break;
1451 					}
1452 				}
1453 				IF_ADDR_UNLOCK(ifp);
1454 				if (!haslinklocal)
1455 					in6_ifattach(ifp, NULL);
1456 			}
1457 		}
1458 	}
1459 		ND_IFINFO(ifp)->flags = ND.flags;
1460 		break;
1461 #undef ND
1462 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1463 		/* sync kernel routing table with the default router list */
1464 		defrouter_reset();
1465 		defrouter_select();
1466 		break;
1467 	case SIOCSPFXFLUSH_IN6:
1468 	{
1469 		/* flush all the prefix advertised by routers */
1470 		struct nd_prefix *pr, *next;
1471 
1472 		s = splnet();
1473 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1474 			struct in6_ifaddr *ia, *ia_next;
1475 
1476 			next = pr->ndpr_next;
1477 
1478 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1479 				continue; /* XXX */
1480 
1481 			/* do we really have to remove addresses as well? */
1482 			/* XXXRW: in6_ifaddrhead locking. */
1483 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1484 			    ia_next) {
1485 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1486 					continue;
1487 
1488 				if (ia->ia6_ndpr == pr)
1489 					in6_purgeaddr(&ia->ia_ifa);
1490 			}
1491 			prelist_remove(pr);
1492 		}
1493 		splx(s);
1494 		break;
1495 	}
1496 	case SIOCSRTRFLUSH_IN6:
1497 	{
1498 		/* flush all the default routers */
1499 		struct nd_defrouter *dr, *next;
1500 
1501 		s = splnet();
1502 		defrouter_reset();
1503 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1504 			next = TAILQ_NEXT(dr, dr_entry);
1505 			defrtrlist_del(dr);
1506 		}
1507 		defrouter_select();
1508 		splx(s);
1509 		break;
1510 	}
1511 	case SIOCGNBRINFO_IN6:
1512 	{
1513 		struct llentry *ln;
1514 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1515 
1516 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1517 			return (error);
1518 
1519 		IF_AFDATA_LOCK(ifp);
1520 		ln = nd6_lookup(&nb_addr, 0, ifp);
1521 		IF_AFDATA_UNLOCK(ifp);
1522 
1523 		if (ln == NULL) {
1524 			error = EINVAL;
1525 			break;
1526 		}
1527 		nbi->state = ln->ln_state;
1528 		nbi->asked = ln->la_asked;
1529 		nbi->isrouter = ln->ln_router;
1530 		nbi->expire = ln->la_expire;
1531 		LLE_RUNLOCK(ln);
1532 		break;
1533 	}
1534 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1535 		ndif->ifindex = V_nd6_defifindex;
1536 		break;
1537 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1538 		return (nd6_setdefaultiface(ndif->ifindex));
1539 	}
1540 	return (error);
1541 }
1542 
1543 /*
1544  * Create neighbor cache entry and cache link-layer address,
1545  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1546  *
1547  * type - ICMP6 type
1548  * code - type dependent information
1549  *
1550  * XXXXX
1551  *  The caller of this function already acquired the ndp
1552  *  cache table lock because the cache entry is returned.
1553  */
1554 struct llentry *
1555 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1556     int lladdrlen, int type, int code)
1557 {
1558 	struct llentry *ln = NULL;
1559 	int is_newentry;
1560 	int do_update;
1561 	int olladdr;
1562 	int llchange;
1563 	int flags;
1564 	int newstate = 0;
1565 	uint16_t router = 0;
1566 	struct sockaddr_in6 sin6;
1567 	struct mbuf *chain = NULL;
1568 	int static_route = 0;
1569 
1570 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1571 
1572 	if (ifp == NULL)
1573 		panic("ifp == NULL in nd6_cache_lladdr");
1574 	if (from == NULL)
1575 		panic("from == NULL in nd6_cache_lladdr");
1576 
1577 	/* nothing must be updated for unspecified address */
1578 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1579 		return NULL;
1580 
1581 	/*
1582 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1583 	 * the caller.
1584 	 *
1585 	 * XXX If the link does not have link-layer adderss, what should
1586 	 * we do? (ifp->if_addrlen == 0)
1587 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1588 	 * description on it in NS section (RFC 2461 7.2.3).
1589 	 */
1590 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1591 	IF_AFDATA_LOCK(ifp);
1592 	ln = nd6_lookup(from, flags, ifp);
1593 
1594 	if (ln == NULL) {
1595 		flags |= ND6_EXCLUSIVE;
1596 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1597 		IF_AFDATA_UNLOCK(ifp);
1598 		is_newentry = 1;
1599 	} else {
1600 		IF_AFDATA_UNLOCK(ifp);
1601 		/* do nothing if static ndp is set */
1602 		if (ln->la_flags & LLE_STATIC) {
1603 			static_route = 1;
1604 			goto done;
1605 		}
1606 		is_newentry = 0;
1607 	}
1608 	if (ln == NULL)
1609 		return (NULL);
1610 
1611 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1612 	if (olladdr && lladdr) {
1613 		llchange = bcmp(lladdr, &ln->ll_addr,
1614 		    ifp->if_addrlen);
1615 	} else
1616 		llchange = 0;
1617 
1618 	/*
1619 	 * newentry olladdr  lladdr  llchange	(*=record)
1620 	 *	0	n	n	--	(1)
1621 	 *	0	y	n	--	(2)
1622 	 *	0	n	y	--	(3) * STALE
1623 	 *	0	y	y	n	(4) *
1624 	 *	0	y	y	y	(5) * STALE
1625 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1626 	 *	1	--	y	--	(7) * STALE
1627 	 */
1628 
1629 	if (lladdr) {		/* (3-5) and (7) */
1630 		/*
1631 		 * Record source link-layer address
1632 		 * XXX is it dependent to ifp->if_type?
1633 		 */
1634 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1635 		ln->la_flags |= LLE_VALID;
1636 	}
1637 
1638 	if (!is_newentry) {
1639 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1640 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1641 			do_update = 1;
1642 			newstate = ND6_LLINFO_STALE;
1643 		} else					/* (1-2,4) */
1644 			do_update = 0;
1645 	} else {
1646 		do_update = 1;
1647 		if (lladdr == NULL)			/* (6) */
1648 			newstate = ND6_LLINFO_NOSTATE;
1649 		else					/* (7) */
1650 			newstate = ND6_LLINFO_STALE;
1651 	}
1652 
1653 	if (do_update) {
1654 		/*
1655 		 * Update the state of the neighbor cache.
1656 		 */
1657 		ln->ln_state = newstate;
1658 
1659 		if (ln->ln_state == ND6_LLINFO_STALE) {
1660 			/*
1661 			 * XXX: since nd6_output() below will cause
1662 			 * state tansition to DELAY and reset the timer,
1663 			 * we must set the timer now, although it is actually
1664 			 * meaningless.
1665 			 */
1666 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1667 
1668 			if (ln->la_hold) {
1669 				struct mbuf *m_hold, *m_hold_next;
1670 
1671 				/*
1672 				 * reset the la_hold in advance, to explicitly
1673 				 * prevent a la_hold lookup in nd6_output()
1674 				 * (wouldn't happen, though...)
1675 				 */
1676 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1677 				    m_hold; m_hold = m_hold_next) {
1678 					m_hold_next = m_hold->m_nextpkt;
1679 					m_hold->m_nextpkt = NULL;
1680 
1681 					/*
1682 					 * we assume ifp is not a p2p here, so
1683 					 * just set the 2nd argument as the
1684 					 * 1st one.
1685 					 */
1686 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1687 				}
1688 				/*
1689 				 * If we have mbufs in the chain we need to do
1690 				 * deferred transmit. Copy the address from the
1691 				 * llentry before dropping the lock down below.
1692 				 */
1693 				if (chain != NULL)
1694 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1695 			}
1696 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1697 			/* probe right away */
1698 			nd6_llinfo_settimer_locked((void *)ln, 0);
1699 		}
1700 	}
1701 
1702 	/*
1703 	 * ICMP6 type dependent behavior.
1704 	 *
1705 	 * NS: clear IsRouter if new entry
1706 	 * RS: clear IsRouter
1707 	 * RA: set IsRouter if there's lladdr
1708 	 * redir: clear IsRouter if new entry
1709 	 *
1710 	 * RA case, (1):
1711 	 * The spec says that we must set IsRouter in the following cases:
1712 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1713 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1714 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1715 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1716 	 * neighbor cache, this is similar to (6).
1717 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1718 	 *
1719 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1720 	 *							D R
1721 	 *	0	n	n	--	(1)	c   ?     s
1722 	 *	0	y	n	--	(2)	c   s     s
1723 	 *	0	n	y	--	(3)	c   s     s
1724 	 *	0	y	y	n	(4)	c   s     s
1725 	 *	0	y	y	y	(5)	c   s     s
1726 	 *	1	--	n	--	(6) c	c	c s
1727 	 *	1	--	y	--	(7) c	c   s	c s
1728 	 *
1729 	 *					(c=clear s=set)
1730 	 */
1731 	switch (type & 0xff) {
1732 	case ND_NEIGHBOR_SOLICIT:
1733 		/*
1734 		 * New entry must have is_router flag cleared.
1735 		 */
1736 		if (is_newentry)	/* (6-7) */
1737 			ln->ln_router = 0;
1738 		break;
1739 	case ND_REDIRECT:
1740 		/*
1741 		 * If the icmp is a redirect to a better router, always set the
1742 		 * is_router flag.  Otherwise, if the entry is newly created,
1743 		 * clear the flag.  [RFC 2461, sec 8.3]
1744 		 */
1745 		if (code == ND_REDIRECT_ROUTER)
1746 			ln->ln_router = 1;
1747 		else if (is_newentry) /* (6-7) */
1748 			ln->ln_router = 0;
1749 		break;
1750 	case ND_ROUTER_SOLICIT:
1751 		/*
1752 		 * is_router flag must always be cleared.
1753 		 */
1754 		ln->ln_router = 0;
1755 		break;
1756 	case ND_ROUTER_ADVERT:
1757 		/*
1758 		 * Mark an entry with lladdr as a router.
1759 		 */
1760 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1761 		    (is_newentry && lladdr)) {			/* (7) */
1762 			ln->ln_router = 1;
1763 		}
1764 		break;
1765 	}
1766 
1767 	if (ln != NULL) {
1768 		static_route = (ln->la_flags & LLE_STATIC);
1769 		router = ln->ln_router;
1770 
1771 		if (flags & ND6_EXCLUSIVE)
1772 			LLE_WUNLOCK(ln);
1773 		else
1774 			LLE_RUNLOCK(ln);
1775 		if (static_route)
1776 			ln = NULL;
1777 	}
1778 	if (chain)
1779 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1780 
1781 	/*
1782 	 * When the link-layer address of a router changes, select the
1783 	 * best router again.  In particular, when the neighbor entry is newly
1784 	 * created, it might affect the selection policy.
1785 	 * Question: can we restrict the first condition to the "is_newentry"
1786 	 * case?
1787 	 * XXX: when we hear an RA from a new router with the link-layer
1788 	 * address option, defrouter_select() is called twice, since
1789 	 * defrtrlist_update called the function as well.  However, I believe
1790 	 * we can compromise the overhead, since it only happens the first
1791 	 * time.
1792 	 * XXX: although defrouter_select() should not have a bad effect
1793 	 * for those are not autoconfigured hosts, we explicitly avoid such
1794 	 * cases for safety.
1795 	 */
1796 	if (do_update && router &&
1797 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1798 		/*
1799 		 * guaranteed recursion
1800 		 */
1801 		defrouter_select();
1802 	}
1803 
1804 	return (ln);
1805 done:
1806 	if (ln != NULL) {
1807 		if (flags & ND6_EXCLUSIVE)
1808 			LLE_WUNLOCK(ln);
1809 		else
1810 			LLE_RUNLOCK(ln);
1811 		if (static_route)
1812 			ln = NULL;
1813 	}
1814 	return (ln);
1815 }
1816 
1817 static void
1818 nd6_slowtimo(void *arg)
1819 {
1820 	CURVNET_SET((struct vnet *) arg);
1821 	struct nd_ifinfo *nd6if;
1822 	struct ifnet *ifp;
1823 
1824 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1825 	    nd6_slowtimo, curvnet);
1826 	IFNET_RLOCK_NOSLEEP();
1827 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1828 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1829 		nd6if = ND_IFINFO(ifp);
1830 		if (nd6if->basereachable && /* already initialized */
1831 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1832 			/*
1833 			 * Since reachable time rarely changes by router
1834 			 * advertisements, we SHOULD insure that a new random
1835 			 * value gets recomputed at least once every few hours.
1836 			 * (RFC 2461, 6.3.4)
1837 			 */
1838 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1839 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1840 		}
1841 	}
1842 	IFNET_RUNLOCK_NOSLEEP();
1843 	CURVNET_RESTORE();
1844 }
1845 
1846 int
1847 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1848     struct sockaddr_in6 *dst, struct rtentry *rt0)
1849 {
1850 
1851 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1852 }
1853 
1854 
1855 /*
1856  * Note that I'm not enforcing any global serialization
1857  * lle state or asked changes here as the logic is too
1858  * complicated to avoid having to always acquire an exclusive
1859  * lock
1860  * KMM
1861  *
1862  */
1863 #define senderr(e) { error = (e); goto bad;}
1864 
1865 int
1866 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1867     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1868 	struct mbuf **chain)
1869 {
1870 	struct mbuf *m = m0;
1871 	struct m_tag *mtag;
1872 	struct llentry *ln = lle;
1873 	struct ip6_hdr *ip6;
1874 	int error = 0;
1875 	int flags = 0;
1876 	int ip6len;
1877 
1878 #ifdef INVARIANTS
1879 	if (lle != NULL) {
1880 
1881 		LLE_WLOCK_ASSERT(lle);
1882 
1883 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1884 	}
1885 #endif
1886 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1887 		goto sendpkt;
1888 
1889 	if (nd6_need_cache(ifp) == 0)
1890 		goto sendpkt;
1891 
1892 	/*
1893 	 * next hop determination.  This routine is derived from ether_output.
1894 	 */
1895 
1896 	/*
1897 	 * Address resolution or Neighbor Unreachability Detection
1898 	 * for the next hop.
1899 	 * At this point, the destination of the packet must be a unicast
1900 	 * or an anycast address(i.e. not a multicast).
1901 	 */
1902 
1903 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1904 	if (ln == NULL) {
1905 	retry:
1906 		IF_AFDATA_LOCK(ifp);
1907 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1908 		IF_AFDATA_UNLOCK(ifp);
1909 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1910 			/*
1911 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1912 			 * the condition below is not very efficient.  But we believe
1913 			 * it is tolerable, because this should be a rare case.
1914 			 */
1915 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1916 			IF_AFDATA_LOCK(ifp);
1917 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1918 			IF_AFDATA_UNLOCK(ifp);
1919 		}
1920 	}
1921 	if (ln == NULL) {
1922 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1923 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1924 			char ip6buf[INET6_ADDRSTRLEN];
1925 			log(LOG_DEBUG,
1926 			    "nd6_output: can't allocate llinfo for %s "
1927 			    "(ln=%p)\n",
1928 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1929 			senderr(EIO);	/* XXX: good error? */
1930 		}
1931 		goto sendpkt;	/* send anyway */
1932 	}
1933 
1934 	/* We don't have to do link-layer address resolution on a p2p link. */
1935 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1936 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1937 		if ((flags & LLE_EXCLUSIVE) == 0) {
1938 			flags |= LLE_EXCLUSIVE;
1939 			goto retry;
1940 		}
1941 		ln->ln_state = ND6_LLINFO_STALE;
1942 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1943 	}
1944 
1945 	/*
1946 	 * The first time we send a packet to a neighbor whose entry is
1947 	 * STALE, we have to change the state to DELAY and a sets a timer to
1948 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1949 	 * neighbor unreachability detection on expiration.
1950 	 * (RFC 2461 7.3.3)
1951 	 */
1952 	if (ln->ln_state == ND6_LLINFO_STALE) {
1953 		if ((flags & LLE_EXCLUSIVE) == 0) {
1954 			flags |= LLE_EXCLUSIVE;
1955 			LLE_RUNLOCK(ln);
1956 			goto retry;
1957 		}
1958 		ln->la_asked = 0;
1959 		ln->ln_state = ND6_LLINFO_DELAY;
1960 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1961 	}
1962 
1963 	/*
1964 	 * If the neighbor cache entry has a state other than INCOMPLETE
1965 	 * (i.e. its link-layer address is already resolved), just
1966 	 * send the packet.
1967 	 */
1968 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1969 		goto sendpkt;
1970 
1971 	/*
1972 	 * There is a neighbor cache entry, but no ethernet address
1973 	 * response yet.  Append this latest packet to the end of the
1974 	 * packet queue in the mbuf, unless the number of the packet
1975 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1976 	 * the oldest packet in the queue will be removed.
1977 	 */
1978 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1979 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1980 
1981 	if ((flags & LLE_EXCLUSIVE) == 0) {
1982 		flags |= LLE_EXCLUSIVE;
1983 		LLE_RUNLOCK(ln);
1984 		goto retry;
1985 	}
1986 
1987 	LLE_WLOCK_ASSERT(ln);
1988 
1989 	if (ln->la_hold) {
1990 		struct mbuf *m_hold;
1991 		int i;
1992 
1993 		i = 0;
1994 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1995 			i++;
1996 			if (m_hold->m_nextpkt == NULL) {
1997 				m_hold->m_nextpkt = m;
1998 				break;
1999 			}
2000 		}
2001 		while (i >= V_nd6_maxqueuelen) {
2002 			m_hold = ln->la_hold;
2003 			ln->la_hold = ln->la_hold->m_nextpkt;
2004 			m_freem(m_hold);
2005 			i--;
2006 		}
2007 	} else {
2008 		ln->la_hold = m;
2009 	}
2010 
2011 	/*
2012 	 * If there has been no NS for the neighbor after entering the
2013 	 * INCOMPLETE state, send the first solicitation.
2014 	 */
2015 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
2016 		ln->la_asked++;
2017 
2018 		nd6_llinfo_settimer_locked(ln,
2019 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
2020 		LLE_WUNLOCK(ln);
2021 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
2022 		if (lle != NULL && ln == lle)
2023 			LLE_WLOCK(lle);
2024 
2025 	} else if (lle == NULL || ln != lle) {
2026 		/*
2027 		 * We did the lookup (no lle arg) so we
2028 		 * need to do the unlock here.
2029 		 */
2030 		LLE_WUNLOCK(ln);
2031 	}
2032 
2033 	return (0);
2034 
2035   sendpkt:
2036 	/* discard the packet if IPv6 operation is disabled on the interface */
2037 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2038 		error = ENETDOWN; /* better error? */
2039 		goto bad;
2040 	}
2041 	/*
2042 	 * ln is valid and the caller did not pass in
2043 	 * an llentry
2044 	 */
2045 	if ((ln != NULL) && (lle == NULL)) {
2046 		if (flags & LLE_EXCLUSIVE)
2047 			LLE_WUNLOCK(ln);
2048 		else
2049 			LLE_RUNLOCK(ln);
2050 	}
2051 
2052 #ifdef MAC
2053 	mac_netinet6_nd6_send(ifp, m);
2054 #endif
2055 
2056 	/*
2057 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2058 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2059 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2060 	 * to be diverted to user space.  When re-injected into the kernel,
2061 	 * send_output() will directly dispatch them to the outgoing interface.
2062 	 */
2063 	if (send_sendso_input_hook != NULL) {
2064 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2065 		if (mtag != NULL) {
2066 			ip6 = mtod(m, struct ip6_hdr *);
2067 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2068 			/* Use the SEND socket */
2069 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2070 			    ip6len);
2071 			/* -1 == no app on SEND socket */
2072 			if (error == 0 || error != -1)
2073 			    return (error);
2074 		}
2075 	}
2076 
2077 	/*
2078 	 * We were passed in a pointer to an lle with the lock held
2079 	 * this means that we can't call if_output as we will
2080 	 * recurse on the lle lock - so what we do is we create
2081 	 * a list of mbufs to send and transmit them in the caller
2082 	 * after the lock is dropped
2083 	 */
2084 	if (lle != NULL) {
2085 		if (*chain == NULL)
2086 			*chain = m;
2087 		else {
2088 			struct mbuf *mb;
2089 
2090 			/*
2091 			 * append mbuf to end of deferred chain
2092 			 */
2093 			mb = *chain;
2094 			while (mb->m_nextpkt != NULL)
2095 				mb = mb->m_nextpkt;
2096 			mb->m_nextpkt = m;
2097 		}
2098 		return (error);
2099 	}
2100 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2101 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2102 		    NULL));
2103 	}
2104 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2105 	return (error);
2106 
2107   bad:
2108 	/*
2109 	 * ln is valid and the caller did not pass in
2110 	 * an llentry
2111 	 */
2112 	if ((ln != NULL) && (lle == NULL)) {
2113 		if (flags & LLE_EXCLUSIVE)
2114 			LLE_WUNLOCK(ln);
2115 		else
2116 			LLE_RUNLOCK(ln);
2117 	}
2118 	if (m)
2119 		m_freem(m);
2120 	return (error);
2121 }
2122 #undef senderr
2123 
2124 
2125 int
2126 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2127     struct sockaddr_in6 *dst, struct route *ro)
2128 {
2129 	struct mbuf *m, *m_head;
2130 	struct ifnet *outifp;
2131 	int error = 0;
2132 
2133 	m_head = chain;
2134 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2135 		outifp = origifp;
2136 	else
2137 		outifp = ifp;
2138 
2139 	while (m_head) {
2140 		m = m_head;
2141 		m_head = m_head->m_nextpkt;
2142 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2143 	}
2144 
2145 	/*
2146 	 * XXX
2147 	 * note that intermediate errors are blindly ignored - but this is
2148 	 * the same convention as used with nd6_output when called by
2149 	 * nd6_cache_lladdr
2150 	 */
2151 	return (error);
2152 }
2153 
2154 
2155 int
2156 nd6_need_cache(struct ifnet *ifp)
2157 {
2158 	/*
2159 	 * XXX: we currently do not make neighbor cache on any interface
2160 	 * other than ARCnet, Ethernet, FDDI and GIF.
2161 	 *
2162 	 * RFC2893 says:
2163 	 * - unidirectional tunnels needs no ND
2164 	 */
2165 	switch (ifp->if_type) {
2166 	case IFT_ARCNET:
2167 	case IFT_ETHER:
2168 	case IFT_FDDI:
2169 	case IFT_IEEE1394:
2170 #ifdef IFT_L2VLAN
2171 	case IFT_L2VLAN:
2172 #endif
2173 #ifdef IFT_IEEE80211
2174 	case IFT_IEEE80211:
2175 #endif
2176 #ifdef IFT_CARP
2177 	case IFT_CARP:
2178 #endif
2179 	case IFT_INFINIBAND:
2180 	case IFT_GIF:		/* XXX need more cases? */
2181 	case IFT_PPP:
2182 	case IFT_TUNNEL:
2183 	case IFT_BRIDGE:
2184 	case IFT_PROPVIRTUAL:
2185 		return (1);
2186 	default:
2187 		return (0);
2188 	}
2189 }
2190 
2191 /*
2192  * the callers of this function need to be re-worked to drop
2193  * the lle lock, drop here for now
2194  */
2195 int
2196 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2197     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2198 {
2199 	struct llentry *ln;
2200 
2201 	*lle = NULL;
2202 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2203 	if (m->m_flags & M_MCAST) {
2204 		int i;
2205 
2206 		switch (ifp->if_type) {
2207 		case IFT_ETHER:
2208 		case IFT_FDDI:
2209 #ifdef IFT_L2VLAN
2210 		case IFT_L2VLAN:
2211 #endif
2212 #ifdef IFT_IEEE80211
2213 		case IFT_IEEE80211:
2214 #endif
2215 		case IFT_BRIDGE:
2216 		case IFT_ISO88025:
2217 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2218 						 desten);
2219 			return (0);
2220 		case IFT_IEEE1394:
2221 			/*
2222 			 * netbsd can use if_broadcastaddr, but we don't do so
2223 			 * to reduce # of ifdef.
2224 			 */
2225 			for (i = 0; i < ifp->if_addrlen; i++)
2226 				desten[i] = ~0;
2227 			return (0);
2228 		case IFT_ARCNET:
2229 			*desten = 0;
2230 			return (0);
2231 		default:
2232 			m_freem(m);
2233 			return (EAFNOSUPPORT);
2234 		}
2235 	}
2236 
2237 
2238 	/*
2239 	 * the entry should have been created in nd6_store_lladdr
2240 	 */
2241 	IF_AFDATA_LOCK(ifp);
2242 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2243 	IF_AFDATA_UNLOCK(ifp);
2244 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2245 		if (ln != NULL)
2246 			LLE_RUNLOCK(ln);
2247 		/* this could happen, if we could not allocate memory */
2248 		m_freem(m);
2249 		return (1);
2250 	}
2251 
2252 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2253 	*lle = ln;
2254 	LLE_RUNLOCK(ln);
2255 	/*
2256 	 * A *small* use after free race exists here
2257 	 */
2258 	return (0);
2259 }
2260 
2261 static void
2262 clear_llinfo_pqueue(struct llentry *ln)
2263 {
2264 	struct mbuf *m_hold, *m_hold_next;
2265 
2266 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2267 		m_hold_next = m_hold->m_nextpkt;
2268 		m_hold->m_nextpkt = NULL;
2269 		m_freem(m_hold);
2270 	}
2271 
2272 	ln->la_hold = NULL;
2273 	return;
2274 }
2275 
2276 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2277 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2278 #ifdef SYSCTL_DECL
2279 SYSCTL_DECL(_net_inet6_icmp6);
2280 #endif
2281 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2282 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2283 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2284 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2285 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2286 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2287 
2288 static int
2289 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2290 {
2291 	int error;
2292 	char buf[1024] __aligned(4);
2293 	struct in6_defrouter *d, *de;
2294 	struct nd_defrouter *dr;
2295 
2296 	if (req->newptr)
2297 		return EPERM;
2298 	error = 0;
2299 
2300 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2301 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2302 		d = (struct in6_defrouter *)buf;
2303 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2304 
2305 		if (d + 1 <= de) {
2306 			bzero(d, sizeof(*d));
2307 			d->rtaddr.sin6_family = AF_INET6;
2308 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2309 			d->rtaddr.sin6_addr = dr->rtaddr;
2310 			error = sa6_recoverscope(&d->rtaddr);
2311 			if (error != 0)
2312 				return (error);
2313 			d->flags = dr->flags;
2314 			d->rtlifetime = dr->rtlifetime;
2315 			d->expire = dr->expire;
2316 			d->if_index = dr->ifp->if_index;
2317 		} else
2318 			panic("buffer too short");
2319 
2320 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2321 		if (error)
2322 			break;
2323 	}
2324 
2325 	return (error);
2326 }
2327 
2328 static int
2329 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2330 {
2331 	int error;
2332 	char buf[1024] __aligned(4);
2333 	struct in6_prefix *p, *pe;
2334 	struct nd_prefix *pr;
2335 	char ip6buf[INET6_ADDRSTRLEN];
2336 
2337 	if (req->newptr)
2338 		return EPERM;
2339 	error = 0;
2340 
2341 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2342 		u_short advrtrs;
2343 		size_t advance;
2344 		struct sockaddr_in6 *sin6, *s6;
2345 		struct nd_pfxrouter *pfr;
2346 
2347 		p = (struct in6_prefix *)buf;
2348 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2349 
2350 		if (p + 1 <= pe) {
2351 			bzero(p, sizeof(*p));
2352 			sin6 = (struct sockaddr_in6 *)(p + 1);
2353 
2354 			p->prefix = pr->ndpr_prefix;
2355 			if (sa6_recoverscope(&p->prefix)) {
2356 				log(LOG_ERR,
2357 				    "scope error in prefix list (%s)\n",
2358 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2359 				/* XXX: press on... */
2360 			}
2361 			p->raflags = pr->ndpr_raf;
2362 			p->prefixlen = pr->ndpr_plen;
2363 			p->vltime = pr->ndpr_vltime;
2364 			p->pltime = pr->ndpr_pltime;
2365 			p->if_index = pr->ndpr_ifp->if_index;
2366 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2367 				p->expire = 0;
2368 			else {
2369 				time_t maxexpire;
2370 
2371 				/* XXX: we assume time_t is signed. */
2372 				maxexpire = (-1) &
2373 				    ~((time_t)1 <<
2374 				    ((sizeof(maxexpire) * 8) - 1));
2375 				if (pr->ndpr_vltime <
2376 				    maxexpire - pr->ndpr_lastupdate) {
2377 				    p->expire = pr->ndpr_lastupdate +
2378 				        pr->ndpr_vltime;
2379 				} else
2380 					p->expire = maxexpire;
2381 			}
2382 			p->refcnt = pr->ndpr_refcnt;
2383 			p->flags = pr->ndpr_stateflags;
2384 			p->origin = PR_ORIG_RA;
2385 			advrtrs = 0;
2386 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2387 			     pfr = pfr->pfr_next) {
2388 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2389 					advrtrs++;
2390 					continue;
2391 				}
2392 				s6 = &sin6[advrtrs];
2393 				bzero(s6, sizeof(*s6));
2394 				s6->sin6_family = AF_INET6;
2395 				s6->sin6_len = sizeof(*sin6);
2396 				s6->sin6_addr = pfr->router->rtaddr;
2397 				if (sa6_recoverscope(s6)) {
2398 					log(LOG_ERR,
2399 					    "scope error in "
2400 					    "prefix list (%s)\n",
2401 					    ip6_sprintf(ip6buf,
2402 						    &pfr->router->rtaddr));
2403 				}
2404 				advrtrs++;
2405 			}
2406 			p->advrtrs = advrtrs;
2407 		} else
2408 			panic("buffer too short");
2409 
2410 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2411 		error = SYSCTL_OUT(req, buf, advance);
2412 		if (error)
2413 			break;
2414 	}
2415 
2416 	return (error);
2417 }
2418