xref: /freebsd/sys/netinet6/nd6.c (revision 9a0c3479e22feda1bdb2db4b97f9deb1b5fa6269)
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) ((const 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 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
124 
125 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *,
126 	struct ifnet *);
127 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
128 static void nd6_slowtimo(void *);
129 static int regen_tmpaddr(struct in6_ifaddr *);
130 static struct llentry *nd6_free(struct llentry *, int);
131 static void nd6_llinfo_timer(void *);
132 static void clear_llinfo_pqueue(struct llentry *);
133 
134 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
135 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
136 
137 VNET_DEFINE(struct callout, nd6_timer_ch);
138 
139 void
140 nd6_init(void)
141 {
142 
143 	LIST_INIT(&V_nd_prefix);
144 
145 	/* initialization of the default router list */
146 	TAILQ_INIT(&V_nd_defrouter);
147 
148 	/* start timer */
149 	callout_init(&V_nd6_slowtimo_ch, 0);
150 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
151 	    nd6_slowtimo, curvnet);
152 }
153 
154 #ifdef VIMAGE
155 void
156 nd6_destroy()
157 {
158 
159 	callout_drain(&V_nd6_slowtimo_ch);
160 	callout_drain(&V_nd6_timer_ch);
161 }
162 #endif
163 
164 struct nd_ifinfo *
165 nd6_ifattach(struct ifnet *ifp)
166 {
167 	struct nd_ifinfo *nd;
168 
169 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO);
170 	nd->initialized = 1;
171 
172 	nd->chlim = IPV6_DEFHLIM;
173 	nd->basereachable = REACHABLE_TIME;
174 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
175 	nd->retrans = RETRANS_TIMER;
176 
177 	nd->flags = ND6_IFF_PERFORMNUD;
178 
179 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
180 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
181 	 * default regardless of the V_ip6_auto_linklocal configuration to
182 	 * give a reasonable default behavior.
183 	 */
184 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
185 	    (ifp->if_flags & IFF_LOOPBACK))
186 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
187 	/*
188 	 * A loopback interface does not need to accept RTADV.
189 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
190 	 * default regardless of the V_ip6_accept_rtadv configuration to
191 	 * prevent the interface from accepting RA messages arrived
192 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
193 	 */
194 	if (V_ip6_accept_rtadv &&
195 	    !(ifp->if_flags & IFF_LOOPBACK) &&
196 	    (ifp->if_type != IFT_BRIDGE))
197 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
198 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
199 		nd->flags |= ND6_IFF_NO_RADR;
200 
201 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
202 	nd6_setmtu0(ifp, nd);
203 
204 	return nd;
205 }
206 
207 void
208 nd6_ifdetach(struct nd_ifinfo *nd)
209 {
210 
211 	free(nd, M_IP6NDP);
212 }
213 
214 /*
215  * Reset ND level link MTU. This function is called when the physical MTU
216  * changes, which means we might have to adjust the ND level MTU.
217  */
218 void
219 nd6_setmtu(struct ifnet *ifp)
220 {
221 
222 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
223 }
224 
225 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
226 void
227 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
228 {
229 	u_int32_t omaxmtu;
230 
231 	omaxmtu = ndi->maxmtu;
232 
233 	switch (ifp->if_type) {
234 	case IFT_ARCNET:
235 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
236 		break;
237 	case IFT_FDDI:
238 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
239 		break;
240 	case IFT_ISO88025:
241 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
242 		 break;
243 	default:
244 		ndi->maxmtu = ifp->if_mtu;
245 		break;
246 	}
247 
248 	/*
249 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
250 	 * undesirable situation.  We thus notify the operator of the change
251 	 * explicitly.  The check for omaxmtu is necessary to restrict the
252 	 * log to the case of changing the MTU, not initializing it.
253 	 */
254 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
255 		log(LOG_NOTICE, "nd6_setmtu0: "
256 		    "new link MTU on %s (%lu) is too small for IPv6\n",
257 		    if_name(ifp), (unsigned long)ndi->maxmtu);
258 	}
259 
260 	if (ndi->maxmtu > V_in6_maxmtu)
261 		in6_setmaxmtu(); /* check all interfaces just in case */
262 
263 }
264 
265 void
266 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
267 {
268 
269 	bzero(ndopts, sizeof(*ndopts));
270 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
271 	ndopts->nd_opts_last
272 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
273 
274 	if (icmp6len == 0) {
275 		ndopts->nd_opts_done = 1;
276 		ndopts->nd_opts_search = NULL;
277 	}
278 }
279 
280 /*
281  * Take one ND option.
282  */
283 struct nd_opt_hdr *
284 nd6_option(union nd_opts *ndopts)
285 {
286 	struct nd_opt_hdr *nd_opt;
287 	int olen;
288 
289 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
290 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
291 	    __func__));
292 	if (ndopts->nd_opts_search == NULL)
293 		return NULL;
294 	if (ndopts->nd_opts_done)
295 		return NULL;
296 
297 	nd_opt = ndopts->nd_opts_search;
298 
299 	/* make sure nd_opt_len is inside the buffer */
300 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
301 		bzero(ndopts, sizeof(*ndopts));
302 		return NULL;
303 	}
304 
305 	olen = nd_opt->nd_opt_len << 3;
306 	if (olen == 0) {
307 		/*
308 		 * Message validation requires that all included
309 		 * options have a length that is greater than zero.
310 		 */
311 		bzero(ndopts, sizeof(*ndopts));
312 		return NULL;
313 	}
314 
315 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
316 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
317 		/* option overruns the end of buffer, invalid */
318 		bzero(ndopts, sizeof(*ndopts));
319 		return NULL;
320 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
321 		/* reached the end of options chain */
322 		ndopts->nd_opts_done = 1;
323 		ndopts->nd_opts_search = NULL;
324 	}
325 	return nd_opt;
326 }
327 
328 /*
329  * Parse multiple ND options.
330  * This function is much easier to use, for ND routines that do not need
331  * multiple options of the same type.
332  */
333 int
334 nd6_options(union nd_opts *ndopts)
335 {
336 	struct nd_opt_hdr *nd_opt;
337 	int i = 0;
338 
339 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
340 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
341 	    __func__));
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 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
384 		case ND_OPT_RDNSS:	/* RFC 6106 */
385 		case ND_OPT_DNSSL:	/* RFC 6106 */
386 			/*
387 			 * Silently ignore options we know and do not care about
388 			 * in the kernel.
389 			 */
390 			break;
391 		default:
392 			/*
393 			 * Unknown options must be silently ignored,
394 			 * to accomodate future extension to the protocol.
395 			 */
396 			nd6log((LOG_DEBUG,
397 			    "nd6_options: unsupported option %d - "
398 			    "option ignored\n", nd_opt->nd_opt_type));
399 		}
400 
401 skip1:
402 		i++;
403 		if (i > V_nd6_maxndopt) {
404 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
405 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
406 			break;
407 		}
408 
409 		if (ndopts->nd_opts_done)
410 			break;
411 	}
412 
413 	return 0;
414 }
415 
416 /*
417  * ND6 timer routine to handle ND6 entries
418  */
419 void
420 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
421 {
422 	int canceled;
423 
424 	LLE_WLOCK_ASSERT(ln);
425 
426 	if (tick < 0) {
427 		ln->la_expire = 0;
428 		ln->ln_ntick = 0;
429 		canceled = callout_stop(&ln->ln_timer_ch);
430 	} else {
431 		ln->la_expire = time_second + tick / hz;
432 		LLE_ADDREF(ln);
433 		if (tick > INT_MAX) {
434 			ln->ln_ntick = tick - INT_MAX;
435 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
436 			    nd6_llinfo_timer, ln);
437 		} else {
438 			ln->ln_ntick = 0;
439 			canceled = callout_reset(&ln->ln_timer_ch, tick,
440 			    nd6_llinfo_timer, ln);
441 		}
442 	}
443 	if (canceled)
444 		LLE_REMREF(ln);
445 }
446 
447 void
448 nd6_llinfo_settimer(struct llentry *ln, long tick)
449 {
450 
451 	LLE_WLOCK(ln);
452 	nd6_llinfo_settimer_locked(ln, tick);
453 	LLE_WUNLOCK(ln);
454 }
455 
456 static void
457 nd6_llinfo_timer(void *arg)
458 {
459 	struct llentry *ln;
460 	struct in6_addr *dst;
461 	struct ifnet *ifp;
462 	struct nd_ifinfo *ndi = NULL;
463 
464 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
465 	ln = (struct llentry *)arg;
466 	LLE_WLOCK_ASSERT(ln);
467 	ifp = ln->lle_tbl->llt_ifp;
468 
469 	CURVNET_SET(ifp->if_vnet);
470 
471 	if (ln->ln_ntick > 0) {
472 		if (ln->ln_ntick > INT_MAX) {
473 			ln->ln_ntick -= INT_MAX;
474 			nd6_llinfo_settimer_locked(ln, INT_MAX);
475 		} else {
476 			ln->ln_ntick = 0;
477 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
478 		}
479 		goto done;
480 	}
481 
482 	ndi = ND_IFINFO(ifp);
483 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
484 	if (ln->la_flags & LLE_STATIC) {
485 		goto done;
486 	}
487 
488 	if (ln->la_flags & LLE_DELETED) {
489 		(void)nd6_free(ln, 0);
490 		ln = NULL;
491 		goto done;
492 	}
493 
494 	switch (ln->ln_state) {
495 	case ND6_LLINFO_INCOMPLETE:
496 		if (ln->la_asked < V_nd6_mmaxtries) {
497 			ln->la_asked++;
498 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
499 			LLE_WUNLOCK(ln);
500 			nd6_ns_output(ifp, NULL, dst, ln, 0);
501 			LLE_WLOCK(ln);
502 		} else {
503 			struct mbuf *m = ln->la_hold;
504 			if (m) {
505 				struct mbuf *m0;
506 
507 				/*
508 				 * assuming every packet in la_hold has the
509 				 * same IP header.  Send error after unlock.
510 				 */
511 				m0 = m->m_nextpkt;
512 				m->m_nextpkt = NULL;
513 				ln->la_hold = m0;
514 				clear_llinfo_pqueue(ln);
515 			}
516 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
517 			(void)nd6_free(ln, 0);
518 			ln = NULL;
519 			if (m != NULL)
520 				icmp6_error2(m, ICMP6_DST_UNREACH,
521 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
522 		}
523 		break;
524 	case ND6_LLINFO_REACHABLE:
525 		if (!ND6_LLINFO_PERMANENT(ln)) {
526 			ln->ln_state = ND6_LLINFO_STALE;
527 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
528 		}
529 		break;
530 
531 	case ND6_LLINFO_STALE:
532 		/* Garbage Collection(RFC 2461 5.3) */
533 		if (!ND6_LLINFO_PERMANENT(ln)) {
534 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
535 			(void)nd6_free(ln, 1);
536 			ln = NULL;
537 		}
538 		break;
539 
540 	case ND6_LLINFO_DELAY:
541 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
542 			/* We need NUD */
543 			ln->la_asked = 1;
544 			ln->ln_state = ND6_LLINFO_PROBE;
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 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
551 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
552 		}
553 		break;
554 	case ND6_LLINFO_PROBE:
555 		if (ln->la_asked < V_nd6_umaxtries) {
556 			ln->la_asked++;
557 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
558 			LLE_WUNLOCK(ln);
559 			nd6_ns_output(ifp, dst, dst, ln, 0);
560 			LLE_WLOCK(ln);
561 		} else {
562 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
563 			(void)nd6_free(ln, 0);
564 			ln = NULL;
565 		}
566 		break;
567 	default:
568 		panic("%s: paths in a dark night can be confusing: %d",
569 		    __func__, ln->ln_state);
570 	}
571 done:
572 	if (ln != NULL)
573 		LLE_FREE_LOCKED(ln);
574 	CURVNET_RESTORE();
575 }
576 
577 
578 /*
579  * ND6 timer routine to expire default route list and prefix list
580  */
581 void
582 nd6_timer(void *arg)
583 {
584 	CURVNET_SET((struct vnet *) arg);
585 	struct nd_defrouter *dr, *ndr;
586 	struct nd_prefix *pr, *npr;
587 	struct in6_ifaddr *ia6, *nia6;
588 
589 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
590 	    nd6_timer, curvnet);
591 
592 	/* expire default router list */
593 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
594 		if (dr->expire && dr->expire < time_second)
595 			defrtrlist_del(dr);
596 	}
597 
598 	/*
599 	 * expire interface addresses.
600 	 * in the past the loop was inside prefix expiry processing.
601 	 * However, from a stricter speci-confrmance standpoint, we should
602 	 * rather separate address lifetimes and prefix lifetimes.
603 	 *
604 	 * XXXRW: in6_ifaddrhead locking.
605 	 */
606   addrloop:
607 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
608 		/* check address 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 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
672 		/*
673 		 * check prefix lifetime.
674 		 * since pltime is just for autoconf, pltime processing for
675 		 * prefix is not necessary.
676 		 */
677 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
678 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
679 
680 			/*
681 			 * address expiration and prefix expiration are
682 			 * separate.  NEVER perform in6_purgeaddr here.
683 			 */
684 			prelist_remove(pr);
685 		}
686 	}
687 	CURVNET_RESTORE();
688 }
689 
690 /*
691  * ia6 - deprecated/invalidated temporary address
692  */
693 static int
694 regen_tmpaddr(struct in6_ifaddr *ia6)
695 {
696 	struct ifaddr *ifa;
697 	struct ifnet *ifp;
698 	struct in6_ifaddr *public_ifa6 = NULL;
699 
700 	ifp = ia6->ia_ifa.ifa_ifp;
701 	IF_ADDR_RLOCK(ifp);
702 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
703 		struct in6_ifaddr *it6;
704 
705 		if (ifa->ifa_addr->sa_family != AF_INET6)
706 			continue;
707 
708 		it6 = (struct in6_ifaddr *)ifa;
709 
710 		/* ignore no autoconf addresses. */
711 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
712 			continue;
713 
714 		/* ignore autoconf addresses with different prefixes. */
715 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
716 			continue;
717 
718 		/*
719 		 * Now we are looking at an autoconf address with the same
720 		 * prefix as ours.  If the address is temporary and is still
721 		 * preferred, do not create another one.  It would be rare, but
722 		 * could happen, for example, when we resume a laptop PC after
723 		 * a long period.
724 		 */
725 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
726 		    !IFA6_IS_DEPRECATED(it6)) {
727 			public_ifa6 = NULL;
728 			break;
729 		}
730 
731 		/*
732 		 * This is a public autoconf address that has the same prefix
733 		 * as ours.  If it is preferred, keep it.  We can't break the
734 		 * loop here, because there may be a still-preferred temporary
735 		 * address with the prefix.
736 		 */
737 		if (!IFA6_IS_DEPRECATED(it6))
738 		    public_ifa6 = it6;
739 
740 		if (public_ifa6 != NULL)
741 			ifa_ref(&public_ifa6->ia_ifa);
742 	}
743 	IF_ADDR_RUNLOCK(ifp);
744 
745 	if (public_ifa6 != NULL) {
746 		int e;
747 
748 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
749 			ifa_free(&public_ifa6->ia_ifa);
750 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
751 			    " tmp addr,errno=%d\n", e);
752 			return (-1);
753 		}
754 		ifa_free(&public_ifa6->ia_ifa);
755 		return (0);
756 	}
757 
758 	return (-1);
759 }
760 
761 /*
762  * Nuke neighbor cache/prefix/default router management table, right before
763  * ifp goes away.
764  */
765 void
766 nd6_purge(struct ifnet *ifp)
767 {
768 	struct nd_defrouter *dr, *ndr;
769 	struct nd_prefix *pr, *npr;
770 
771 	/*
772 	 * Nuke default router list entries toward ifp.
773 	 * We defer removal of default router list entries that is installed
774 	 * in the routing table, in order to keep additional side effects as
775 	 * small as possible.
776 	 */
777 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
778 		if (dr->installed)
779 			continue;
780 
781 		if (dr->ifp == ifp)
782 			defrtrlist_del(dr);
783 	}
784 
785 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
786 		if (!dr->installed)
787 			continue;
788 
789 		if (dr->ifp == ifp)
790 			defrtrlist_del(dr);
791 	}
792 
793 	/* Nuke prefix list entries toward ifp */
794 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
795 		if (pr->ndpr_ifp == ifp) {
796 			/*
797 			 * Because if_detach() does *not* release prefixes
798 			 * while purging addresses the reference count will
799 			 * still be above zero. We therefore reset it to
800 			 * make sure that the prefix really gets purged.
801 			 */
802 			pr->ndpr_refcnt = 0;
803 
804 			/*
805 			 * Previously, pr->ndpr_addr is removed as well,
806 			 * but I strongly believe we don't have to do it.
807 			 * nd6_purge() is only called from in6_ifdetach(),
808 			 * which removes all the associated interface addresses
809 			 * by itself.
810 			 * (jinmei@kame.net 20010129)
811 			 */
812 			prelist_remove(pr);
813 		}
814 	}
815 
816 	/* cancel default outgoing interface setting */
817 	if (V_nd6_defifindex == ifp->if_index)
818 		nd6_setdefaultiface(0);
819 
820 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
821 		/* Refresh default router list. */
822 		defrouter_select();
823 	}
824 
825 	/* XXXXX
826 	 * We do not nuke the neighbor cache entries here any more
827 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
828 	 * nd6_purge() is invoked by in6_ifdetach() which is called
829 	 * from if_detach() where everything gets purged. So let
830 	 * in6_domifdetach() do the actual L2 table purging work.
831 	 */
832 }
833 
834 /*
835  * the caller acquires and releases the lock on the lltbls
836  * Returns the llentry locked
837  */
838 struct llentry *
839 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
840 {
841 	struct sockaddr_in6 sin6;
842 	struct llentry *ln;
843 	int llflags;
844 
845 	bzero(&sin6, sizeof(sin6));
846 	sin6.sin6_len = sizeof(struct sockaddr_in6);
847 	sin6.sin6_family = AF_INET6;
848 	sin6.sin6_addr = *addr6;
849 
850 	IF_AFDATA_LOCK_ASSERT(ifp);
851 
852 	llflags = 0;
853 	if (flags & ND6_CREATE)
854 	    llflags |= LLE_CREATE;
855 	if (flags & ND6_EXCLUSIVE)
856 	    llflags |= LLE_EXCLUSIVE;
857 
858 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
859 	if ((ln != NULL) && (llflags & LLE_CREATE))
860 		ln->ln_state = ND6_LLINFO_NOSTATE;
861 
862 	return (ln);
863 }
864 
865 /*
866  * Test whether a given IPv6 address is a neighbor or not, ignoring
867  * the actual neighbor cache.  The neighbor cache is ignored in order
868  * to not reenter the routing code from within itself.
869  */
870 static int
871 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
872 {
873 	struct nd_prefix *pr;
874 	struct ifaddr *dstaddr;
875 
876 	/*
877 	 * A link-local address is always a neighbor.
878 	 * XXX: a link does not necessarily specify a single interface.
879 	 */
880 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
881 		struct sockaddr_in6 sin6_copy;
882 		u_int32_t zone;
883 
884 		/*
885 		 * We need sin6_copy since sa6_recoverscope() may modify the
886 		 * content (XXX).
887 		 */
888 		sin6_copy = *addr;
889 		if (sa6_recoverscope(&sin6_copy))
890 			return (0); /* XXX: should be impossible */
891 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
892 			return (0);
893 		if (sin6_copy.sin6_scope_id == zone)
894 			return (1);
895 		else
896 			return (0);
897 	}
898 
899 	/*
900 	 * If the address matches one of our addresses,
901 	 * it should be a neighbor.
902 	 * If the address matches one of our on-link prefixes, it should be a
903 	 * neighbor.
904 	 */
905 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
906 		if (pr->ndpr_ifp != ifp)
907 			continue;
908 
909 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
910 			struct rtentry *rt;
911 
912 			/* Always use the default FIB here. */
913 			rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
914 			    0, 0, RT_DEFAULT_FIB);
915 			if (rt == NULL)
916 				continue;
917 			/*
918 			 * This is the case where multiple interfaces
919 			 * have the same prefix, but only one is installed
920 			 * into the routing table and that prefix entry
921 			 * is not the one being examined here. In the case
922 			 * where RADIX_MPATH is enabled, multiple route
923 			 * entries (of the same rt_key value) will be
924 			 * installed because the interface addresses all
925 			 * differ.
926 			 */
927 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
928 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
929 				RTFREE_LOCKED(rt);
930 				continue;
931 			}
932 			RTFREE_LOCKED(rt);
933 		}
934 
935 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
936 		    &addr->sin6_addr, &pr->ndpr_mask))
937 			return (1);
938 	}
939 
940 	/*
941 	 * If the address is assigned on the node of the other side of
942 	 * a p2p interface, the address should be a neighbor.
943 	 */
944 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
945 	if (dstaddr != NULL) {
946 		if (dstaddr->ifa_ifp == ifp) {
947 			ifa_free(dstaddr);
948 			return (1);
949 		}
950 		ifa_free(dstaddr);
951 	}
952 
953 	/*
954 	 * If the default router list is empty, all addresses are regarded
955 	 * as on-link, and thus, as a neighbor.
956 	 */
957 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
958 	    TAILQ_EMPTY(&V_nd_defrouter) &&
959 	    V_nd6_defifindex == ifp->if_index) {
960 		return (1);
961 	}
962 
963 	return (0);
964 }
965 
966 
967 /*
968  * Detect if a given IPv6 address identifies a neighbor on a given link.
969  * XXX: should take care of the destination of a p2p link?
970  */
971 int
972 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
973 {
974 	struct llentry *lle;
975 	int rc = 0;
976 
977 	IF_AFDATA_UNLOCK_ASSERT(ifp);
978 	if (nd6_is_new_addr_neighbor(addr, ifp))
979 		return (1);
980 
981 	/*
982 	 * Even if the address matches none of our addresses, it might be
983 	 * in the neighbor cache.
984 	 */
985 	IF_AFDATA_RLOCK(ifp);
986 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
987 		LLE_RUNLOCK(lle);
988 		rc = 1;
989 	}
990 	IF_AFDATA_RUNLOCK(ifp);
991 	return (rc);
992 }
993 
994 /*
995  * Free an nd6 llinfo entry.
996  * Since the function would cause significant changes in the kernel, DO NOT
997  * make it global, unless you have a strong reason for the change, and are sure
998  * that the change is safe.
999  */
1000 static struct llentry *
1001 nd6_free(struct llentry *ln, int gc)
1002 {
1003         struct llentry *next;
1004 	struct nd_defrouter *dr;
1005 	struct ifnet *ifp;
1006 
1007 	LLE_WLOCK_ASSERT(ln);
1008 
1009 	/*
1010 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1011 	 * even though it is not harmful, it was not really necessary.
1012 	 */
1013 
1014 	/* cancel timer */
1015 	nd6_llinfo_settimer_locked(ln, -1);
1016 
1017 	ifp = ln->lle_tbl->llt_ifp;
1018 
1019 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1020 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1021 
1022 		if (dr != NULL && dr->expire &&
1023 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1024 			/*
1025 			 * If the reason for the deletion is just garbage
1026 			 * collection, and the neighbor is an active default
1027 			 * router, do not delete it.  Instead, reset the GC
1028 			 * timer using the router's lifetime.
1029 			 * Simply deleting the entry would affect default
1030 			 * router selection, which is not necessarily a good
1031 			 * thing, especially when we're using router preference
1032 			 * values.
1033 			 * XXX: the check for ln_state would be redundant,
1034 			 *      but we intentionally keep it just in case.
1035 			 */
1036 			if (dr->expire > time_second)
1037 				nd6_llinfo_settimer_locked(ln,
1038 				    (dr->expire - time_second) * hz);
1039 			else
1040 				nd6_llinfo_settimer_locked(ln,
1041 				    (long)V_nd6_gctimer * hz);
1042 
1043 			next = LIST_NEXT(ln, lle_next);
1044 			LLE_REMREF(ln);
1045 			LLE_WUNLOCK(ln);
1046 			return (next);
1047 		}
1048 
1049 		if (dr) {
1050 			/*
1051 			 * Unreachablity of a router might affect the default
1052 			 * router selection and on-link detection of advertised
1053 			 * prefixes.
1054 			 */
1055 
1056 			/*
1057 			 * Temporarily fake the state to choose a new default
1058 			 * router and to perform on-link determination of
1059 			 * prefixes correctly.
1060 			 * Below the state will be set correctly,
1061 			 * or the entry itself will be deleted.
1062 			 */
1063 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1064 		}
1065 
1066 		if (ln->ln_router || dr) {
1067 
1068 			/*
1069 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1070 			 * rnh and for the calls to pfxlist_onlink_check() and
1071 			 * defrouter_select() in the block further down for calls
1072 			 * into nd6_lookup().  We still hold a ref.
1073 			 */
1074 			LLE_WUNLOCK(ln);
1075 
1076 			/*
1077 			 * rt6_flush must be called whether or not the neighbor
1078 			 * is in the Default Router List.
1079 			 * See a corresponding comment in nd6_na_input().
1080 			 */
1081 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1082 		}
1083 
1084 		if (dr) {
1085 			/*
1086 			 * Since defrouter_select() does not affect the
1087 			 * on-link determination and MIP6 needs the check
1088 			 * before the default router selection, we perform
1089 			 * the check now.
1090 			 */
1091 			pfxlist_onlink_check();
1092 
1093 			/*
1094 			 * Refresh default router list.
1095 			 */
1096 			defrouter_select();
1097 		}
1098 
1099 		if (ln->ln_router || dr)
1100 			LLE_WLOCK(ln);
1101 	}
1102 
1103 	/*
1104 	 * Before deleting the entry, remember the next entry as the
1105 	 * return value.  We need this because pfxlist_onlink_check() above
1106 	 * might have freed other entries (particularly the old next entry) as
1107 	 * a side effect (XXX).
1108 	 */
1109 	next = LIST_NEXT(ln, lle_next);
1110 
1111 	/*
1112 	 * Save to unlock. We still hold an extra reference and will not
1113 	 * free(9) in llentry_free() if someone else holds one as well.
1114 	 */
1115 	LLE_WUNLOCK(ln);
1116 	IF_AFDATA_LOCK(ifp);
1117 	LLE_WLOCK(ln);
1118 
1119 	/* Guard against race with other llentry_free(). */
1120 	if (ln->la_flags & LLE_LINKED) {
1121 		LLE_REMREF(ln);
1122 		llentry_free(ln);
1123 	} else
1124 		LLE_FREE_LOCKED(ln);
1125 
1126 	IF_AFDATA_UNLOCK(ifp);
1127 
1128 	return (next);
1129 }
1130 
1131 /*
1132  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1133  *
1134  * XXX cost-effective methods?
1135  */
1136 void
1137 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1138 {
1139 	struct llentry *ln;
1140 	struct ifnet *ifp;
1141 
1142 	if ((dst6 == NULL) || (rt == NULL))
1143 		return;
1144 
1145 	ifp = rt->rt_ifp;
1146 	IF_AFDATA_LOCK(ifp);
1147 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1148 	IF_AFDATA_UNLOCK(ifp);
1149 	if (ln == NULL)
1150 		return;
1151 
1152 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1153 		goto done;
1154 
1155 	/*
1156 	 * if we get upper-layer reachability confirmation many times,
1157 	 * it is possible we have false information.
1158 	 */
1159 	if (!force) {
1160 		ln->ln_byhint++;
1161 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1162 			goto done;
1163 		}
1164 	}
1165 
1166  	ln->ln_state = ND6_LLINFO_REACHABLE;
1167 	if (!ND6_LLINFO_PERMANENT(ln)) {
1168 		nd6_llinfo_settimer_locked(ln,
1169 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1170 	}
1171 done:
1172 	LLE_WUNLOCK(ln);
1173 }
1174 
1175 
1176 /*
1177  * Rejuvenate this function for routing operations related
1178  * processing.
1179  */
1180 void
1181 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1182 {
1183 	struct sockaddr_in6 *gateway;
1184 	struct nd_defrouter *dr;
1185 	struct ifnet *ifp;
1186 
1187 	RT_LOCK_ASSERT(rt);
1188 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1189 	ifp = rt->rt_ifp;
1190 
1191 	switch (req) {
1192 	case RTM_ADD:
1193 		break;
1194 
1195 	case RTM_DELETE:
1196 		if (!ifp)
1197 			return;
1198 		/*
1199 		 * Only indirect routes are interesting.
1200 		 */
1201 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1202 			return;
1203 		/*
1204 		 * check for default route
1205 		 */
1206 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1207 				       &SIN6(rt_key(rt))->sin6_addr)) {
1208 
1209 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1210 			if (dr != NULL)
1211 				dr->installed = 0;
1212 		}
1213 		break;
1214 	}
1215 }
1216 
1217 
1218 int
1219 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1220 {
1221 	struct in6_drlist *drl = (struct in6_drlist *)data;
1222 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1223 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1224 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1225 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1226 	struct nd_defrouter *dr;
1227 	struct nd_prefix *pr;
1228 	int i = 0, error = 0;
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 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1237 			if (i >= DRLSTSIZ)
1238 				break;
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 		}
1248 		break;
1249 	case SIOCGPRLST_IN6:
1250 		/*
1251 		 * obsolete API, use sysctl under net.inet6.icmp6
1252 		 *
1253 		 * XXX the structure in6_prlist was changed in backward-
1254 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1255 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1256 		 */
1257 		/*
1258 		 * XXX meaning of fields, especialy "raflags", is very
1259 		 * differnet between RA prefix list and RR/static prefix list.
1260 		 * how about separating ioctls into two?
1261 		 */
1262 		bzero(oprl, sizeof(*oprl));
1263 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1264 			struct nd_pfxrouter *pfr;
1265 			int j;
1266 
1267 			if (i >= PRLSTSIZ)
1268 				break;
1269 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1270 			oprl->prefix[i].raflags = pr->ndpr_raf;
1271 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1272 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1273 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1274 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1275 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1276 				oprl->prefix[i].expire = 0;
1277 			else {
1278 				time_t maxexpire;
1279 
1280 				/* XXX: we assume time_t is signed. */
1281 				maxexpire = (-1) &
1282 				    ~((time_t)1 <<
1283 				    ((sizeof(maxexpire) * 8) - 1));
1284 				if (pr->ndpr_vltime <
1285 				    maxexpire - pr->ndpr_lastupdate) {
1286 					oprl->prefix[i].expire =
1287 					    pr->ndpr_lastupdate +
1288 					    pr->ndpr_vltime;
1289 				} else
1290 					oprl->prefix[i].expire = maxexpire;
1291 			}
1292 
1293 			j = 0;
1294 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1295 				if (j < DRLSTSIZ) {
1296 #define RTRADDR oprl->prefix[i].advrtr[j]
1297 					RTRADDR = pfr->router->rtaddr;
1298 					in6_clearscope(&RTRADDR);
1299 #undef RTRADDR
1300 				}
1301 				j++;
1302 			}
1303 			oprl->prefix[i].advrtrs = j;
1304 			oprl->prefix[i].origin = PR_ORIG_RA;
1305 
1306 			i++;
1307 		}
1308 
1309 		break;
1310 	case OSIOCGIFINFO_IN6:
1311 #define ND	ndi->ndi
1312 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1313 		bzero(&ND, sizeof(ND));
1314 		ND.linkmtu = IN6_LINKMTU(ifp);
1315 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1316 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1317 		ND.reachable = ND_IFINFO(ifp)->reachable;
1318 		ND.retrans = ND_IFINFO(ifp)->retrans;
1319 		ND.flags = ND_IFINFO(ifp)->flags;
1320 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1321 		ND.chlim = ND_IFINFO(ifp)->chlim;
1322 		break;
1323 	case SIOCGIFINFO_IN6:
1324 		ND = *ND_IFINFO(ifp);
1325 		break;
1326 	case SIOCSIFINFO_IN6:
1327 		/*
1328 		 * used to change host variables from userland.
1329 		 * intented for a use on router to reflect RA configurations.
1330 		 */
1331 		/* 0 means 'unspecified' */
1332 		if (ND.linkmtu != 0) {
1333 			if (ND.linkmtu < IPV6_MMTU ||
1334 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1335 				error = EINVAL;
1336 				break;
1337 			}
1338 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1339 		}
1340 
1341 		if (ND.basereachable != 0) {
1342 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1343 
1344 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1345 			if (ND.basereachable != obasereachable)
1346 				ND_IFINFO(ifp)->reachable =
1347 				    ND_COMPUTE_RTIME(ND.basereachable);
1348 		}
1349 		if (ND.retrans != 0)
1350 			ND_IFINFO(ifp)->retrans = ND.retrans;
1351 		if (ND.chlim != 0)
1352 			ND_IFINFO(ifp)->chlim = ND.chlim;
1353 		/* FALLTHROUGH */
1354 	case SIOCSIFINFO_FLAGS:
1355 	{
1356 		struct ifaddr *ifa;
1357 		struct in6_ifaddr *ia;
1358 
1359 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1360 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1361 			/* ifdisabled 1->0 transision */
1362 
1363 			/*
1364 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1365 			 * has an link-local address with IN6_IFF_DUPLICATED,
1366 			 * do not clear ND6_IFF_IFDISABLED.
1367 			 * See RFC 4862, Section 5.4.5.
1368 			 */
1369 			int duplicated_linklocal = 0;
1370 
1371 			IF_ADDR_RLOCK(ifp);
1372 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1373 				if (ifa->ifa_addr->sa_family != AF_INET6)
1374 					continue;
1375 				ia = (struct in6_ifaddr *)ifa;
1376 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1377 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1378 					duplicated_linklocal = 1;
1379 					break;
1380 				}
1381 			}
1382 			IF_ADDR_RUNLOCK(ifp);
1383 
1384 			if (duplicated_linklocal) {
1385 				ND.flags |= ND6_IFF_IFDISABLED;
1386 				log(LOG_ERR, "Cannot enable an interface"
1387 				    " with a link-local address marked"
1388 				    " duplicate.\n");
1389 			} else {
1390 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1391 				if (ifp->if_flags & IFF_UP)
1392 					in6_if_up(ifp);
1393 			}
1394 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1395 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1396 			/* ifdisabled 0->1 transision */
1397 			/* Mark all IPv6 address as tentative. */
1398 
1399 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1400 			IF_ADDR_RLOCK(ifp);
1401 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1402 				if (ifa->ifa_addr->sa_family != AF_INET6)
1403 					continue;
1404 				ia = (struct in6_ifaddr *)ifa;
1405 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1406 			}
1407 			IF_ADDR_RUNLOCK(ifp);
1408 		}
1409 
1410 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1411 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1412 				/* auto_linklocal 0->1 transision */
1413 
1414 				/* If no link-local address on ifp, configure */
1415 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1416 				in6_ifattach(ifp, NULL);
1417 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1418 			    ifp->if_flags & IFF_UP) {
1419 				/*
1420 				 * When the IF already has
1421 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1422 				 * address is assigned, and IFF_UP, try to
1423 				 * assign one.
1424 				 */
1425 				int haslinklocal = 0;
1426 
1427 				IF_ADDR_RLOCK(ifp);
1428 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1429 					if (ifa->ifa_addr->sa_family != AF_INET6)
1430 						continue;
1431 					ia = (struct in6_ifaddr *)ifa;
1432 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1433 						haslinklocal = 1;
1434 						break;
1435 					}
1436 				}
1437 				IF_ADDR_RUNLOCK(ifp);
1438 				if (!haslinklocal)
1439 					in6_ifattach(ifp, NULL);
1440 			}
1441 		}
1442 	}
1443 		ND_IFINFO(ifp)->flags = ND.flags;
1444 		break;
1445 #undef ND
1446 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1447 		/* sync kernel routing table with the default router list */
1448 		defrouter_reset();
1449 		defrouter_select();
1450 		break;
1451 	case SIOCSPFXFLUSH_IN6:
1452 	{
1453 		/* flush all the prefix advertised by routers */
1454 		struct nd_prefix *pr, *next;
1455 
1456 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1457 			struct in6_ifaddr *ia, *ia_next;
1458 
1459 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1460 				continue; /* XXX */
1461 
1462 			/* do we really have to remove addresses as well? */
1463 			/* XXXRW: in6_ifaddrhead locking. */
1464 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1465 			    ia_next) {
1466 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1467 					continue;
1468 
1469 				if (ia->ia6_ndpr == pr)
1470 					in6_purgeaddr(&ia->ia_ifa);
1471 			}
1472 			prelist_remove(pr);
1473 		}
1474 		break;
1475 	}
1476 	case SIOCSRTRFLUSH_IN6:
1477 	{
1478 		/* flush all the default routers */
1479 		struct nd_defrouter *dr, *next;
1480 
1481 		defrouter_reset();
1482 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1483 			defrtrlist_del(dr);
1484 		}
1485 		defrouter_select();
1486 		break;
1487 	}
1488 	case SIOCGNBRINFO_IN6:
1489 	{
1490 		struct llentry *ln;
1491 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1492 
1493 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1494 			return (error);
1495 
1496 		IF_AFDATA_RLOCK(ifp);
1497 		ln = nd6_lookup(&nb_addr, 0, ifp);
1498 		IF_AFDATA_RUNLOCK(ifp);
1499 
1500 		if (ln == NULL) {
1501 			error = EINVAL;
1502 			break;
1503 		}
1504 		nbi->state = ln->ln_state;
1505 		nbi->asked = ln->la_asked;
1506 		nbi->isrouter = ln->ln_router;
1507 		nbi->expire = ln->la_expire;
1508 		LLE_RUNLOCK(ln);
1509 		break;
1510 	}
1511 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1512 		ndif->ifindex = V_nd6_defifindex;
1513 		break;
1514 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1515 		return (nd6_setdefaultiface(ndif->ifindex));
1516 	}
1517 	return (error);
1518 }
1519 
1520 /*
1521  * Create neighbor cache entry and cache link-layer address,
1522  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1523  *
1524  * type - ICMP6 type
1525  * code - type dependent information
1526  *
1527  * XXXXX
1528  *  The caller of this function already acquired the ndp
1529  *  cache table lock because the cache entry is returned.
1530  */
1531 struct llentry *
1532 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1533     int lladdrlen, int type, int code)
1534 {
1535 	struct llentry *ln = NULL;
1536 	int is_newentry;
1537 	int do_update;
1538 	int olladdr;
1539 	int llchange;
1540 	int flags;
1541 	int newstate = 0;
1542 	uint16_t router = 0;
1543 	struct sockaddr_in6 sin6;
1544 	struct mbuf *chain = NULL;
1545 	int static_route = 0;
1546 
1547 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1548 
1549 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1550 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1551 
1552 	/* nothing must be updated for unspecified address */
1553 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1554 		return NULL;
1555 
1556 	/*
1557 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1558 	 * the caller.
1559 	 *
1560 	 * XXX If the link does not have link-layer adderss, what should
1561 	 * we do? (ifp->if_addrlen == 0)
1562 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1563 	 * description on it in NS section (RFC 2461 7.2.3).
1564 	 */
1565 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1566 	IF_AFDATA_LOCK(ifp);
1567 	ln = nd6_lookup(from, flags, ifp);
1568 
1569 	if (ln == NULL) {
1570 		flags |= ND6_EXCLUSIVE;
1571 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1572 		IF_AFDATA_UNLOCK(ifp);
1573 		is_newentry = 1;
1574 	} else {
1575 		IF_AFDATA_UNLOCK(ifp);
1576 		/* do nothing if static ndp is set */
1577 		if (ln->la_flags & LLE_STATIC) {
1578 			static_route = 1;
1579 			goto done;
1580 		}
1581 		is_newentry = 0;
1582 	}
1583 	if (ln == NULL)
1584 		return (NULL);
1585 
1586 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1587 	if (olladdr && lladdr) {
1588 		llchange = bcmp(lladdr, &ln->ll_addr,
1589 		    ifp->if_addrlen);
1590 	} else
1591 		llchange = 0;
1592 
1593 	/*
1594 	 * newentry olladdr  lladdr  llchange	(*=record)
1595 	 *	0	n	n	--	(1)
1596 	 *	0	y	n	--	(2)
1597 	 *	0	n	y	--	(3) * STALE
1598 	 *	0	y	y	n	(4) *
1599 	 *	0	y	y	y	(5) * STALE
1600 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1601 	 *	1	--	y	--	(7) * STALE
1602 	 */
1603 
1604 	if (lladdr) {		/* (3-5) and (7) */
1605 		/*
1606 		 * Record source link-layer address
1607 		 * XXX is it dependent to ifp->if_type?
1608 		 */
1609 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1610 		ln->la_flags |= LLE_VALID;
1611 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
1612 	}
1613 
1614 	if (!is_newentry) {
1615 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1616 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1617 			do_update = 1;
1618 			newstate = ND6_LLINFO_STALE;
1619 		} else					/* (1-2,4) */
1620 			do_update = 0;
1621 	} else {
1622 		do_update = 1;
1623 		if (lladdr == NULL)			/* (6) */
1624 			newstate = ND6_LLINFO_NOSTATE;
1625 		else					/* (7) */
1626 			newstate = ND6_LLINFO_STALE;
1627 	}
1628 
1629 	if (do_update) {
1630 		/*
1631 		 * Update the state of the neighbor cache.
1632 		 */
1633 		ln->ln_state = newstate;
1634 
1635 		if (ln->ln_state == ND6_LLINFO_STALE) {
1636 			/*
1637 			 * XXX: since nd6_output() below will cause
1638 			 * state tansition to DELAY and reset the timer,
1639 			 * we must set the timer now, although it is actually
1640 			 * meaningless.
1641 			 */
1642 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1643 
1644 			if (ln->la_hold) {
1645 				struct mbuf *m_hold, *m_hold_next;
1646 
1647 				/*
1648 				 * reset the la_hold in advance, to explicitly
1649 				 * prevent a la_hold lookup in nd6_output()
1650 				 * (wouldn't happen, though...)
1651 				 */
1652 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1653 				    m_hold; m_hold = m_hold_next) {
1654 					m_hold_next = m_hold->m_nextpkt;
1655 					m_hold->m_nextpkt = NULL;
1656 
1657 					/*
1658 					 * we assume ifp is not a p2p here, so
1659 					 * just set the 2nd argument as the
1660 					 * 1st one.
1661 					 */
1662 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1663 				}
1664 				/*
1665 				 * If we have mbufs in the chain we need to do
1666 				 * deferred transmit. Copy the address from the
1667 				 * llentry before dropping the lock down below.
1668 				 */
1669 				if (chain != NULL)
1670 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1671 			}
1672 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1673 			/* probe right away */
1674 			nd6_llinfo_settimer_locked((void *)ln, 0);
1675 		}
1676 	}
1677 
1678 	/*
1679 	 * ICMP6 type dependent behavior.
1680 	 *
1681 	 * NS: clear IsRouter if new entry
1682 	 * RS: clear IsRouter
1683 	 * RA: set IsRouter if there's lladdr
1684 	 * redir: clear IsRouter if new entry
1685 	 *
1686 	 * RA case, (1):
1687 	 * The spec says that we must set IsRouter in the following cases:
1688 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1689 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1690 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1691 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1692 	 * neighbor cache, this is similar to (6).
1693 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1694 	 *
1695 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1696 	 *							D R
1697 	 *	0	n	n	--	(1)	c   ?     s
1698 	 *	0	y	n	--	(2)	c   s     s
1699 	 *	0	n	y	--	(3)	c   s     s
1700 	 *	0	y	y	n	(4)	c   s     s
1701 	 *	0	y	y	y	(5)	c   s     s
1702 	 *	1	--	n	--	(6) c	c	c s
1703 	 *	1	--	y	--	(7) c	c   s	c s
1704 	 *
1705 	 *					(c=clear s=set)
1706 	 */
1707 	switch (type & 0xff) {
1708 	case ND_NEIGHBOR_SOLICIT:
1709 		/*
1710 		 * New entry must have is_router flag cleared.
1711 		 */
1712 		if (is_newentry)	/* (6-7) */
1713 			ln->ln_router = 0;
1714 		break;
1715 	case ND_REDIRECT:
1716 		/*
1717 		 * If the icmp is a redirect to a better router, always set the
1718 		 * is_router flag.  Otherwise, if the entry is newly created,
1719 		 * clear the flag.  [RFC 2461, sec 8.3]
1720 		 */
1721 		if (code == ND_REDIRECT_ROUTER)
1722 			ln->ln_router = 1;
1723 		else if (is_newentry) /* (6-7) */
1724 			ln->ln_router = 0;
1725 		break;
1726 	case ND_ROUTER_SOLICIT:
1727 		/*
1728 		 * is_router flag must always be cleared.
1729 		 */
1730 		ln->ln_router = 0;
1731 		break;
1732 	case ND_ROUTER_ADVERT:
1733 		/*
1734 		 * Mark an entry with lladdr as a router.
1735 		 */
1736 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1737 		    (is_newentry && lladdr)) {			/* (7) */
1738 			ln->ln_router = 1;
1739 		}
1740 		break;
1741 	}
1742 
1743 	if (ln != NULL) {
1744 		static_route = (ln->la_flags & LLE_STATIC);
1745 		router = ln->ln_router;
1746 
1747 		if (flags & ND6_EXCLUSIVE)
1748 			LLE_WUNLOCK(ln);
1749 		else
1750 			LLE_RUNLOCK(ln);
1751 		if (static_route)
1752 			ln = NULL;
1753 	}
1754 	if (chain)
1755 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1756 
1757 	/*
1758 	 * When the link-layer address of a router changes, select the
1759 	 * best router again.  In particular, when the neighbor entry is newly
1760 	 * created, it might affect the selection policy.
1761 	 * Question: can we restrict the first condition to the "is_newentry"
1762 	 * case?
1763 	 * XXX: when we hear an RA from a new router with the link-layer
1764 	 * address option, defrouter_select() is called twice, since
1765 	 * defrtrlist_update called the function as well.  However, I believe
1766 	 * we can compromise the overhead, since it only happens the first
1767 	 * time.
1768 	 * XXX: although defrouter_select() should not have a bad effect
1769 	 * for those are not autoconfigured hosts, we explicitly avoid such
1770 	 * cases for safety.
1771 	 */
1772 	if (do_update && router &&
1773 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1774 		/*
1775 		 * guaranteed recursion
1776 		 */
1777 		defrouter_select();
1778 	}
1779 
1780 	return (ln);
1781 done:
1782 	if (ln != NULL) {
1783 		if (flags & ND6_EXCLUSIVE)
1784 			LLE_WUNLOCK(ln);
1785 		else
1786 			LLE_RUNLOCK(ln);
1787 		if (static_route)
1788 			ln = NULL;
1789 	}
1790 	return (ln);
1791 }
1792 
1793 static void
1794 nd6_slowtimo(void *arg)
1795 {
1796 	CURVNET_SET((struct vnet *) arg);
1797 	struct nd_ifinfo *nd6if;
1798 	struct ifnet *ifp;
1799 
1800 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1801 	    nd6_slowtimo, curvnet);
1802 	IFNET_RLOCK_NOSLEEP();
1803 	TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1804 		nd6if = ND_IFINFO(ifp);
1805 		if (nd6if->basereachable && /* already initialized */
1806 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1807 			/*
1808 			 * Since reachable time rarely changes by router
1809 			 * advertisements, we SHOULD insure that a new random
1810 			 * value gets recomputed at least once every few hours.
1811 			 * (RFC 2461, 6.3.4)
1812 			 */
1813 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1814 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1815 		}
1816 	}
1817 	IFNET_RUNLOCK_NOSLEEP();
1818 	CURVNET_RESTORE();
1819 }
1820 
1821 int
1822 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1823     struct sockaddr_in6 *dst, struct rtentry *rt0)
1824 {
1825 
1826 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1827 }
1828 
1829 
1830 /*
1831  * Note that I'm not enforcing any global serialization
1832  * lle state or asked changes here as the logic is too
1833  * complicated to avoid having to always acquire an exclusive
1834  * lock
1835  * KMM
1836  *
1837  */
1838 #define senderr(e) { error = (e); goto bad;}
1839 
1840 int
1841 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1842     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1843 	struct mbuf **chain)
1844 {
1845 	struct mbuf *m = m0;
1846 	struct m_tag *mtag;
1847 	struct llentry *ln = lle;
1848 	struct ip6_hdr *ip6;
1849 	int error = 0;
1850 	int flags = 0;
1851 	int ip6len;
1852 
1853 #ifdef INVARIANTS
1854 	if (lle != NULL) {
1855 
1856 		LLE_WLOCK_ASSERT(lle);
1857 
1858 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1859 	}
1860 #endif
1861 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1862 		goto sendpkt;
1863 
1864 	if (nd6_need_cache(ifp) == 0)
1865 		goto sendpkt;
1866 
1867 	/*
1868 	 * next hop determination.  This routine is derived from ether_output.
1869 	 */
1870 
1871 	/*
1872 	 * Address resolution or Neighbor Unreachability Detection
1873 	 * for the next hop.
1874 	 * At this point, the destination of the packet must be a unicast
1875 	 * or an anycast address(i.e. not a multicast).
1876 	 */
1877 
1878 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1879 	if (ln == NULL) {
1880 	retry:
1881 		IF_AFDATA_LOCK(ifp);
1882 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1883 		IF_AFDATA_UNLOCK(ifp);
1884 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1885 			/*
1886 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1887 			 * the condition below is not very efficient.  But we believe
1888 			 * it is tolerable, because this should be a rare case.
1889 			 */
1890 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1891 			IF_AFDATA_LOCK(ifp);
1892 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1893 			IF_AFDATA_UNLOCK(ifp);
1894 		}
1895 	}
1896 	if (ln == NULL) {
1897 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1898 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1899 			char ip6buf[INET6_ADDRSTRLEN];
1900 			log(LOG_DEBUG,
1901 			    "nd6_output: can't allocate llinfo for %s "
1902 			    "(ln=%p)\n",
1903 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1904 			senderr(EIO);	/* XXX: good error? */
1905 		}
1906 		goto sendpkt;	/* send anyway */
1907 	}
1908 
1909 	/* We don't have to do link-layer address resolution on a p2p link. */
1910 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1911 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1912 		if ((flags & LLE_EXCLUSIVE) == 0) {
1913 			flags |= LLE_EXCLUSIVE;
1914 			goto retry;
1915 		}
1916 		ln->ln_state = ND6_LLINFO_STALE;
1917 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1918 	}
1919 
1920 	/*
1921 	 * The first time we send a packet to a neighbor whose entry is
1922 	 * STALE, we have to change the state to DELAY and a sets a timer to
1923 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1924 	 * neighbor unreachability detection on expiration.
1925 	 * (RFC 2461 7.3.3)
1926 	 */
1927 	if (ln->ln_state == ND6_LLINFO_STALE) {
1928 		if ((flags & LLE_EXCLUSIVE) == 0) {
1929 			flags |= LLE_EXCLUSIVE;
1930 			LLE_RUNLOCK(ln);
1931 			goto retry;
1932 		}
1933 		ln->la_asked = 0;
1934 		ln->ln_state = ND6_LLINFO_DELAY;
1935 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1936 	}
1937 
1938 	/*
1939 	 * If the neighbor cache entry has a state other than INCOMPLETE
1940 	 * (i.e. its link-layer address is already resolved), just
1941 	 * send the packet.
1942 	 */
1943 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1944 		goto sendpkt;
1945 
1946 	/*
1947 	 * There is a neighbor cache entry, but no ethernet address
1948 	 * response yet.  Append this latest packet to the end of the
1949 	 * packet queue in the mbuf, unless the number of the packet
1950 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1951 	 * the oldest packet in the queue will be removed.
1952 	 */
1953 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1954 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1955 
1956 	if ((flags & LLE_EXCLUSIVE) == 0) {
1957 		flags |= LLE_EXCLUSIVE;
1958 		LLE_RUNLOCK(ln);
1959 		goto retry;
1960 	}
1961 
1962 	LLE_WLOCK_ASSERT(ln);
1963 
1964 	if (ln->la_hold) {
1965 		struct mbuf *m_hold;
1966 		int i;
1967 
1968 		i = 0;
1969 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1970 			i++;
1971 			if (m_hold->m_nextpkt == NULL) {
1972 				m_hold->m_nextpkt = m;
1973 				break;
1974 			}
1975 		}
1976 		while (i >= V_nd6_maxqueuelen) {
1977 			m_hold = ln->la_hold;
1978 			ln->la_hold = ln->la_hold->m_nextpkt;
1979 			m_freem(m_hold);
1980 			i--;
1981 		}
1982 	} else {
1983 		ln->la_hold = m;
1984 	}
1985 
1986 	/*
1987 	 * If there has been no NS for the neighbor after entering the
1988 	 * INCOMPLETE state, send the first solicitation.
1989 	 */
1990 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1991 		ln->la_asked++;
1992 
1993 		nd6_llinfo_settimer_locked(ln,
1994 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1995 		LLE_WUNLOCK(ln);
1996 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1997 		if (lle != NULL && ln == lle)
1998 			LLE_WLOCK(lle);
1999 
2000 	} else if (lle == NULL || ln != lle) {
2001 		/*
2002 		 * We did the lookup (no lle arg) so we
2003 		 * need to do the unlock here.
2004 		 */
2005 		LLE_WUNLOCK(ln);
2006 	}
2007 
2008 	return (0);
2009 
2010   sendpkt:
2011 	/* discard the packet if IPv6 operation is disabled on the interface */
2012 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2013 		error = ENETDOWN; /* better error? */
2014 		goto bad;
2015 	}
2016 	/*
2017 	 * ln is valid and the caller did not pass in
2018 	 * an llentry
2019 	 */
2020 	if ((ln != NULL) && (lle == NULL)) {
2021 		if (flags & LLE_EXCLUSIVE)
2022 			LLE_WUNLOCK(ln);
2023 		else
2024 			LLE_RUNLOCK(ln);
2025 	}
2026 
2027 #ifdef MAC
2028 	mac_netinet6_nd6_send(ifp, m);
2029 #endif
2030 
2031 	/*
2032 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2033 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2034 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2035 	 * to be diverted to user space.  When re-injected into the kernel,
2036 	 * send_output() will directly dispatch them to the outgoing interface.
2037 	 */
2038 	if (send_sendso_input_hook != NULL) {
2039 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2040 		if (mtag != NULL) {
2041 			ip6 = mtod(m, struct ip6_hdr *);
2042 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2043 			/* Use the SEND socket */
2044 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2045 			    ip6len);
2046 			/* -1 == no app on SEND socket */
2047 			if (error == 0 || error != -1)
2048 			    return (error);
2049 		}
2050 	}
2051 
2052 	/*
2053 	 * We were passed in a pointer to an lle with the lock held
2054 	 * this means that we can't call if_output as we will
2055 	 * recurse on the lle lock - so what we do is we create
2056 	 * a list of mbufs to send and transmit them in the caller
2057 	 * after the lock is dropped
2058 	 */
2059 	if (lle != NULL) {
2060 		if (*chain == NULL)
2061 			*chain = m;
2062 		else {
2063 			struct mbuf *mb;
2064 
2065 			/*
2066 			 * append mbuf to end of deferred chain
2067 			 */
2068 			mb = *chain;
2069 			while (mb->m_nextpkt != NULL)
2070 				mb = mb->m_nextpkt;
2071 			mb->m_nextpkt = m;
2072 		}
2073 		return (error);
2074 	}
2075 	/* Reset layer specific mbuf flags to avoid confusing lower layers. */
2076 	m->m_flags &= ~(M_PROTOFLAGS);
2077 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2078 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2079 		    NULL));
2080 	}
2081 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2082 	return (error);
2083 
2084   bad:
2085 	/*
2086 	 * ln is valid and the caller did not pass in
2087 	 * an llentry
2088 	 */
2089 	if ((ln != NULL) && (lle == NULL)) {
2090 		if (flags & LLE_EXCLUSIVE)
2091 			LLE_WUNLOCK(ln);
2092 		else
2093 			LLE_RUNLOCK(ln);
2094 	}
2095 	if (m)
2096 		m_freem(m);
2097 	return (error);
2098 }
2099 #undef senderr
2100 
2101 
2102 int
2103 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2104     struct sockaddr_in6 *dst, struct route *ro)
2105 {
2106 	struct mbuf *m, *m_head;
2107 	struct ifnet *outifp;
2108 	int error = 0;
2109 
2110 	m_head = chain;
2111 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2112 		outifp = origifp;
2113 	else
2114 		outifp = ifp;
2115 
2116 	while (m_head) {
2117 		m = m_head;
2118 		m_head = m_head->m_nextpkt;
2119 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2120 	}
2121 
2122 	/*
2123 	 * XXX
2124 	 * note that intermediate errors are blindly ignored - but this is
2125 	 * the same convention as used with nd6_output when called by
2126 	 * nd6_cache_lladdr
2127 	 */
2128 	return (error);
2129 }
2130 
2131 
2132 int
2133 nd6_need_cache(struct ifnet *ifp)
2134 {
2135 	/*
2136 	 * XXX: we currently do not make neighbor cache on any interface
2137 	 * other than ARCnet, Ethernet, FDDI and GIF.
2138 	 *
2139 	 * RFC2893 says:
2140 	 * - unidirectional tunnels needs no ND
2141 	 */
2142 	switch (ifp->if_type) {
2143 	case IFT_ARCNET:
2144 	case IFT_ETHER:
2145 	case IFT_FDDI:
2146 	case IFT_IEEE1394:
2147 #ifdef IFT_L2VLAN
2148 	case IFT_L2VLAN:
2149 #endif
2150 #ifdef IFT_IEEE80211
2151 	case IFT_IEEE80211:
2152 #endif
2153 	case IFT_INFINIBAND:
2154 	case IFT_GIF:		/* XXX need more cases? */
2155 	case IFT_PPP:
2156 	case IFT_TUNNEL:
2157 	case IFT_BRIDGE:
2158 	case IFT_PROPVIRTUAL:
2159 		return (1);
2160 	default:
2161 		return (0);
2162 	}
2163 }
2164 
2165 /*
2166  * the callers of this function need to be re-worked to drop
2167  * the lle lock, drop here for now
2168  */
2169 int
2170 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2171     const struct sockaddr *dst, u_char *desten, struct llentry **lle)
2172 {
2173 	struct llentry *ln;
2174 
2175 	*lle = NULL;
2176 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2177 	if (m != NULL && m->m_flags & M_MCAST) {
2178 		int i;
2179 
2180 		switch (ifp->if_type) {
2181 		case IFT_ETHER:
2182 		case IFT_FDDI:
2183 #ifdef IFT_L2VLAN
2184 		case IFT_L2VLAN:
2185 #endif
2186 #ifdef IFT_IEEE80211
2187 		case IFT_IEEE80211:
2188 #endif
2189 		case IFT_BRIDGE:
2190 		case IFT_ISO88025:
2191 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2192 						 desten);
2193 			return (0);
2194 		case IFT_IEEE1394:
2195 			/*
2196 			 * netbsd can use if_broadcastaddr, but we don't do so
2197 			 * to reduce # of ifdef.
2198 			 */
2199 			for (i = 0; i < ifp->if_addrlen; i++)
2200 				desten[i] = ~0;
2201 			return (0);
2202 		case IFT_ARCNET:
2203 			*desten = 0;
2204 			return (0);
2205 		default:
2206 			m_freem(m);
2207 			return (EAFNOSUPPORT);
2208 		}
2209 	}
2210 
2211 
2212 	/*
2213 	 * the entry should have been created in nd6_store_lladdr
2214 	 */
2215 	IF_AFDATA_RLOCK(ifp);
2216 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2217 	IF_AFDATA_RUNLOCK(ifp);
2218 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2219 		if (ln != NULL)
2220 			LLE_RUNLOCK(ln);
2221 		/* this could happen, if we could not allocate memory */
2222 		m_freem(m);
2223 		return (1);
2224 	}
2225 
2226 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2227 	*lle = ln;
2228 	LLE_RUNLOCK(ln);
2229 	/*
2230 	 * A *small* use after free race exists here
2231 	 */
2232 	return (0);
2233 }
2234 
2235 static void
2236 clear_llinfo_pqueue(struct llentry *ln)
2237 {
2238 	struct mbuf *m_hold, *m_hold_next;
2239 
2240 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2241 		m_hold_next = m_hold->m_nextpkt;
2242 		m_freem(m_hold);
2243 	}
2244 
2245 	ln->la_hold = NULL;
2246 	return;
2247 }
2248 
2249 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2250 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2251 #ifdef SYSCTL_DECL
2252 SYSCTL_DECL(_net_inet6_icmp6);
2253 #endif
2254 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2255 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2256 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2257 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2258 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2259 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2260 
2261 static int
2262 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2263 {
2264 	struct in6_defrouter d;
2265 	struct nd_defrouter *dr;
2266 	int error;
2267 
2268 	if (req->newptr)
2269 		return (EPERM);
2270 
2271 	bzero(&d, sizeof(d));
2272 	d.rtaddr.sin6_family = AF_INET6;
2273 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2274 
2275 	/*
2276 	 * XXX locking
2277 	 */
2278 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2279 		d.rtaddr.sin6_addr = dr->rtaddr;
2280 		error = sa6_recoverscope(&d.rtaddr);
2281 		if (error != 0)
2282 			return (error);
2283 		d.flags = dr->flags;
2284 		d.rtlifetime = dr->rtlifetime;
2285 		d.expire = dr->expire;
2286 		d.if_index = dr->ifp->if_index;
2287 		error = SYSCTL_OUT(req, &d, sizeof(d));
2288 		if (error != 0)
2289 			return (error);
2290 	}
2291 	return (0);
2292 }
2293 
2294 static int
2295 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2296 {
2297 	struct in6_prefix p;
2298 	struct sockaddr_in6 s6;
2299 	struct nd_prefix *pr;
2300 	struct nd_pfxrouter *pfr;
2301 	time_t maxexpire;
2302 	int error;
2303 	char ip6buf[INET6_ADDRSTRLEN];
2304 
2305 	if (req->newptr)
2306 		return (EPERM);
2307 
2308 	bzero(&p, sizeof(p));
2309 	p.origin = PR_ORIG_RA;
2310 	bzero(&s6, sizeof(s6));
2311 	s6.sin6_family = AF_INET6;
2312 	s6.sin6_len = sizeof(s6);
2313 
2314 	/*
2315 	 * XXX locking
2316 	 */
2317 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2318 		p.prefix = pr->ndpr_prefix;
2319 		if (sa6_recoverscope(&p.prefix)) {
2320 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2321 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2322 			/* XXX: press on... */
2323 		}
2324 		p.raflags = pr->ndpr_raf;
2325 		p.prefixlen = pr->ndpr_plen;
2326 		p.vltime = pr->ndpr_vltime;
2327 		p.pltime = pr->ndpr_pltime;
2328 		p.if_index = pr->ndpr_ifp->if_index;
2329 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2330 			p.expire = 0;
2331 		else {
2332 			/* XXX: we assume time_t is signed. */
2333 			maxexpire = (-1) &
2334 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2335 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2336 				p.expire = pr->ndpr_lastupdate +
2337 				    pr->ndpr_vltime;
2338 			else
2339 				p.expire = maxexpire;
2340 		}
2341 		p.refcnt = pr->ndpr_refcnt;
2342 		p.flags = pr->ndpr_stateflags;
2343 		p.advrtrs = 0;
2344 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2345 			p.advrtrs++;
2346 		error = SYSCTL_OUT(req, &p, sizeof(p));
2347 		if (error != 0)
2348 			return (error);
2349 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2350 			s6.sin6_addr = pfr->router->rtaddr;
2351 			if (sa6_recoverscope(&s6))
2352 				log(LOG_ERR,
2353 				    "scope error in prefix list (%s)\n",
2354 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2355 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2356 			if (error != 0)
2357 				return (error);
2358 		}
2359 	}
2360 	return (0);
2361 }
2362