xref: /freebsd/sys/netinet6/nd6.c (revision a10cee30c94cf5944826d2a495e9cdf339dfbcc8)
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|M_ZERO);
178 	nd->initialized = 1;
179 
180 	nd->chlim = IPV6_DEFHLIM;
181 	nd->basereachable = REACHABLE_TIME;
182 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
183 	nd->retrans = RETRANS_TIMER;
184 
185 	nd->flags = ND6_IFF_PERFORMNUD;
186 
187 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
188 	if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
189 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
190 
191 	/* A loopback interface does not need to accept RTADV. */
192 	if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
193 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
194 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
195 		nd->flags |= ND6_IFF_NO_RADR;
196 
197 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
198 	nd6_setmtu0(ifp, nd);
199 
200 	return nd;
201 }
202 
203 void
204 nd6_ifdetach(struct nd_ifinfo *nd)
205 {
206 
207 	free(nd, M_IP6NDP);
208 }
209 
210 /*
211  * Reset ND level link MTU. This function is called when the physical MTU
212  * changes, which means we might have to adjust the ND level MTU.
213  */
214 void
215 nd6_setmtu(struct ifnet *ifp)
216 {
217 
218 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
219 }
220 
221 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
222 void
223 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
224 {
225 	u_int32_t omaxmtu;
226 
227 	omaxmtu = ndi->maxmtu;
228 
229 	switch (ifp->if_type) {
230 	case IFT_ARCNET:
231 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
232 		break;
233 	case IFT_FDDI:
234 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
235 		break;
236 	case IFT_ISO88025:
237 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
238 		 break;
239 	default:
240 		ndi->maxmtu = ifp->if_mtu;
241 		break;
242 	}
243 
244 	/*
245 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
246 	 * undesirable situation.  We thus notify the operator of the change
247 	 * explicitly.  The check for omaxmtu is necessary to restrict the
248 	 * log to the case of changing the MTU, not initializing it.
249 	 */
250 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
251 		log(LOG_NOTICE, "nd6_setmtu0: "
252 		    "new link MTU on %s (%lu) is too small for IPv6\n",
253 		    if_name(ifp), (unsigned long)ndi->maxmtu);
254 	}
255 
256 	if (ndi->maxmtu > V_in6_maxmtu)
257 		in6_setmaxmtu(); /* check all interfaces just in case */
258 
259 }
260 
261 void
262 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
263 {
264 
265 	bzero(ndopts, sizeof(*ndopts));
266 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
267 	ndopts->nd_opts_last
268 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
269 
270 	if (icmp6len == 0) {
271 		ndopts->nd_opts_done = 1;
272 		ndopts->nd_opts_search = NULL;
273 	}
274 }
275 
276 /*
277  * Take one ND option.
278  */
279 struct nd_opt_hdr *
280 nd6_option(union nd_opts *ndopts)
281 {
282 	struct nd_opt_hdr *nd_opt;
283 	int olen;
284 
285 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
286 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
287 	    __func__));
288 	if (ndopts->nd_opts_search == NULL)
289 		return NULL;
290 	if (ndopts->nd_opts_done)
291 		return NULL;
292 
293 	nd_opt = ndopts->nd_opts_search;
294 
295 	/* make sure nd_opt_len is inside the buffer */
296 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
297 		bzero(ndopts, sizeof(*ndopts));
298 		return NULL;
299 	}
300 
301 	olen = nd_opt->nd_opt_len << 3;
302 	if (olen == 0) {
303 		/*
304 		 * Message validation requires that all included
305 		 * options have a length that is greater than zero.
306 		 */
307 		bzero(ndopts, sizeof(*ndopts));
308 		return NULL;
309 	}
310 
311 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
312 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
313 		/* option overruns the end of buffer, invalid */
314 		bzero(ndopts, sizeof(*ndopts));
315 		return NULL;
316 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
317 		/* reached the end of options chain */
318 		ndopts->nd_opts_done = 1;
319 		ndopts->nd_opts_search = NULL;
320 	}
321 	return nd_opt;
322 }
323 
324 /*
325  * Parse multiple ND options.
326  * This function is much easier to use, for ND routines that do not need
327  * multiple options of the same type.
328  */
329 int
330 nd6_options(union nd_opts *ndopts)
331 {
332 	struct nd_opt_hdr *nd_opt;
333 	int i = 0;
334 
335 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
336 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
337 	    __func__));
338 	if (ndopts->nd_opts_search == NULL)
339 		return 0;
340 
341 	while (1) {
342 		nd_opt = nd6_option(ndopts);
343 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
344 			/*
345 			 * Message validation requires that all included
346 			 * options have a length that is greater than zero.
347 			 */
348 			ICMP6STAT_INC(icp6s_nd_badopt);
349 			bzero(ndopts, sizeof(*ndopts));
350 			return -1;
351 		}
352 
353 		if (nd_opt == NULL)
354 			goto skip1;
355 
356 		switch (nd_opt->nd_opt_type) {
357 		case ND_OPT_SOURCE_LINKADDR:
358 		case ND_OPT_TARGET_LINKADDR:
359 		case ND_OPT_MTU:
360 		case ND_OPT_REDIRECTED_HEADER:
361 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
362 				nd6log((LOG_INFO,
363 				    "duplicated ND6 option found (type=%d)\n",
364 				    nd_opt->nd_opt_type));
365 				/* XXX bark? */
366 			} else {
367 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
368 					= nd_opt;
369 			}
370 			break;
371 		case ND_OPT_PREFIX_INFORMATION:
372 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
373 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
374 					= nd_opt;
375 			}
376 			ndopts->nd_opts_pi_end =
377 				(struct nd_opt_prefix_info *)nd_opt;
378 			break;
379 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
380 		case ND_OPT_RDNSS:	/* RFC 6106 */
381 		case ND_OPT_DNSSL:	/* RFC 6106 */
382 			/*
383 			 * Silently ignore options we know and do not care about
384 			 * in the kernel.
385 			 */
386 			break;
387 		default:
388 			/*
389 			 * Unknown options must be silently ignored,
390 			 * to accomodate future extension to the protocol.
391 			 */
392 			nd6log((LOG_DEBUG,
393 			    "nd6_options: unsupported option %d - "
394 			    "option ignored\n", nd_opt->nd_opt_type));
395 		}
396 
397 skip1:
398 		i++;
399 		if (i > V_nd6_maxndopt) {
400 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
401 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
402 			break;
403 		}
404 
405 		if (ndopts->nd_opts_done)
406 			break;
407 	}
408 
409 	return 0;
410 }
411 
412 /*
413  * ND6 timer routine to handle ND6 entries
414  */
415 void
416 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
417 {
418 	int canceled;
419 
420 	LLE_WLOCK_ASSERT(ln);
421 
422 	if (tick < 0) {
423 		ln->la_expire = 0;
424 		ln->ln_ntick = 0;
425 		canceled = callout_stop(&ln->ln_timer_ch);
426 	} else {
427 		ln->la_expire = time_second + tick / hz;
428 		LLE_ADDREF(ln);
429 		if (tick > INT_MAX) {
430 			ln->ln_ntick = tick - INT_MAX;
431 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
432 			    nd6_llinfo_timer, ln);
433 		} else {
434 			ln->ln_ntick = 0;
435 			canceled = callout_reset(&ln->ln_timer_ch, tick,
436 			    nd6_llinfo_timer, ln);
437 		}
438 	}
439 	if (canceled)
440 		LLE_REMREF(ln);
441 }
442 
443 void
444 nd6_llinfo_settimer(struct llentry *ln, long tick)
445 {
446 
447 	LLE_WLOCK(ln);
448 	nd6_llinfo_settimer_locked(ln, tick);
449 	LLE_WUNLOCK(ln);
450 }
451 
452 static void
453 nd6_llinfo_timer(void *arg)
454 {
455 	struct llentry *ln;
456 	struct in6_addr *dst;
457 	struct ifnet *ifp;
458 	struct nd_ifinfo *ndi = NULL;
459 
460 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
461 	ln = (struct llentry *)arg;
462 	LLE_WLOCK_ASSERT(ln);
463 	ifp = ln->lle_tbl->llt_ifp;
464 
465 	CURVNET_SET(ifp->if_vnet);
466 
467 	if (ln->ln_ntick > 0) {
468 		if (ln->ln_ntick > INT_MAX) {
469 			ln->ln_ntick -= INT_MAX;
470 			nd6_llinfo_settimer_locked(ln, INT_MAX);
471 		} else {
472 			ln->ln_ntick = 0;
473 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
474 		}
475 		goto done;
476 	}
477 
478 	ndi = ND_IFINFO(ifp);
479 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
480 	if (ln->la_flags & LLE_STATIC) {
481 		goto done;
482 	}
483 
484 	if (ln->la_flags & LLE_DELETED) {
485 		(void)nd6_free(ln, 0);
486 		ln = NULL;
487 		goto done;
488 	}
489 
490 	switch (ln->ln_state) {
491 	case ND6_LLINFO_INCOMPLETE:
492 		if (ln->la_asked < V_nd6_mmaxtries) {
493 			ln->la_asked++;
494 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
495 			LLE_WUNLOCK(ln);
496 			nd6_ns_output(ifp, NULL, dst, ln, 0);
497 			LLE_WLOCK(ln);
498 		} else {
499 			struct mbuf *m = ln->la_hold;
500 			if (m) {
501 				struct mbuf *m0;
502 
503 				/*
504 				 * assuming every packet in la_hold has the
505 				 * same IP header.  Send error after unlock.
506 				 */
507 				m0 = m->m_nextpkt;
508 				m->m_nextpkt = NULL;
509 				ln->la_hold = m0;
510 				clear_llinfo_pqueue(ln);
511 			}
512 			(void)nd6_free(ln, 0);
513 			ln = NULL;
514 			if (m != NULL)
515 				icmp6_error2(m, ICMP6_DST_UNREACH,
516 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
517 		}
518 		break;
519 	case ND6_LLINFO_REACHABLE:
520 		if (!ND6_LLINFO_PERMANENT(ln)) {
521 			ln->ln_state = ND6_LLINFO_STALE;
522 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
523 		}
524 		break;
525 
526 	case ND6_LLINFO_STALE:
527 		/* Garbage Collection(RFC 2461 5.3) */
528 		if (!ND6_LLINFO_PERMANENT(ln)) {
529 			(void)nd6_free(ln, 1);
530 			ln = NULL;
531 		}
532 		break;
533 
534 	case ND6_LLINFO_DELAY:
535 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
536 			/* We need NUD */
537 			ln->la_asked = 1;
538 			ln->ln_state = ND6_LLINFO_PROBE;
539 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
540 			LLE_WUNLOCK(ln);
541 			nd6_ns_output(ifp, dst, dst, ln, 0);
542 			LLE_WLOCK(ln);
543 		} else {
544 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
545 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
546 		}
547 		break;
548 	case ND6_LLINFO_PROBE:
549 		if (ln->la_asked < V_nd6_umaxtries) {
550 			ln->la_asked++;
551 			nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
552 			LLE_WUNLOCK(ln);
553 			nd6_ns_output(ifp, dst, dst, ln, 0);
554 			LLE_WLOCK(ln);
555 		} else {
556 			(void)nd6_free(ln, 0);
557 			ln = NULL;
558 		}
559 		break;
560 	default:
561 		panic("%s: paths in a dark night can be confusing: %d",
562 		    __func__, ln->ln_state);
563 	}
564 done:
565 	if (ln != NULL)
566 		LLE_FREE_LOCKED(ln);
567 	CURVNET_RESTORE();
568 }
569 
570 
571 /*
572  * ND6 timer routine to expire default route list and prefix list
573  */
574 void
575 nd6_timer(void *arg)
576 {
577 	CURVNET_SET((struct vnet *) arg);
578 	int s;
579 	struct nd_defrouter *dr, *ndr;
580 	struct nd_prefix *pr, *npr;
581 	struct in6_ifaddr *ia6, *nia6;
582 
583 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
584 	    nd6_timer, curvnet);
585 
586 	/* expire default router list */
587 	s = splnet();
588 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
589 		if (dr->expire && dr->expire < time_second)
590 			defrtrlist_del(dr);
591 	}
592 
593 	/*
594 	 * expire interface addresses.
595 	 * in the past the loop was inside prefix expiry processing.
596 	 * However, from a stricter speci-confrmance standpoint, we should
597 	 * rather separate address lifetimes and prefix lifetimes.
598 	 *
599 	 * XXXRW: in6_ifaddrhead locking.
600 	 */
601   addrloop:
602 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
603 		/* check address lifetime */
604 		if (IFA6_IS_INVALID(ia6)) {
605 			int regen = 0;
606 
607 			/*
608 			 * If the expiring address is temporary, try
609 			 * regenerating a new one.  This would be useful when
610 			 * we suspended a laptop PC, then turned it on after a
611 			 * period that could invalidate all temporary
612 			 * addresses.  Although we may have to restart the
613 			 * loop (see below), it must be after purging the
614 			 * address.  Otherwise, we'd see an infinite loop of
615 			 * regeneration.
616 			 */
617 			if (V_ip6_use_tempaddr &&
618 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
619 				if (regen_tmpaddr(ia6) == 0)
620 					regen = 1;
621 			}
622 
623 			in6_purgeaddr(&ia6->ia_ifa);
624 
625 			if (regen)
626 				goto addrloop; /* XXX: see below */
627 		} else if (IFA6_IS_DEPRECATED(ia6)) {
628 			int oldflags = ia6->ia6_flags;
629 
630 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
631 
632 			/*
633 			 * If a temporary address has just become deprecated,
634 			 * regenerate a new one if possible.
635 			 */
636 			if (V_ip6_use_tempaddr &&
637 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
638 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
639 
640 				if (regen_tmpaddr(ia6) == 0) {
641 					/*
642 					 * A new temporary address is
643 					 * generated.
644 					 * XXX: this means the address chain
645 					 * has changed while we are still in
646 					 * the loop.  Although the change
647 					 * would not cause disaster (because
648 					 * it's not a deletion, but an
649 					 * addition,) we'd rather restart the
650 					 * loop just for safety.  Or does this
651 					 * significantly reduce performance??
652 					 */
653 					goto addrloop;
654 				}
655 			}
656 		} else {
657 			/*
658 			 * A new RA might have made a deprecated address
659 			 * preferred.
660 			 */
661 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
662 		}
663 	}
664 
665 	/* expire prefix list */
666 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
667 		/*
668 		 * check prefix lifetime.
669 		 * since pltime is just for autoconf, pltime processing for
670 		 * prefix is not necessary.
671 		 */
672 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
673 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
674 
675 			/*
676 			 * address expiration and prefix expiration are
677 			 * separate.  NEVER perform in6_purgeaddr here.
678 			 */
679 			prelist_remove(pr);
680 		}
681 	}
682 	splx(s);
683 	CURVNET_RESTORE();
684 }
685 
686 /*
687  * ia6 - deprecated/invalidated temporary address
688  */
689 static int
690 regen_tmpaddr(struct in6_ifaddr *ia6)
691 {
692 	struct ifaddr *ifa;
693 	struct ifnet *ifp;
694 	struct in6_ifaddr *public_ifa6 = NULL;
695 
696 	ifp = ia6->ia_ifa.ifa_ifp;
697 	IF_ADDR_RLOCK(ifp);
698 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
699 		struct in6_ifaddr *it6;
700 
701 		if (ifa->ifa_addr->sa_family != AF_INET6)
702 			continue;
703 
704 		it6 = (struct in6_ifaddr *)ifa;
705 
706 		/* ignore no autoconf addresses. */
707 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
708 			continue;
709 
710 		/* ignore autoconf addresses with different prefixes. */
711 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
712 			continue;
713 
714 		/*
715 		 * Now we are looking at an autoconf address with the same
716 		 * prefix as ours.  If the address is temporary and is still
717 		 * preferred, do not create another one.  It would be rare, but
718 		 * could happen, for example, when we resume a laptop PC after
719 		 * a long period.
720 		 */
721 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
722 		    !IFA6_IS_DEPRECATED(it6)) {
723 			public_ifa6 = NULL;
724 			break;
725 		}
726 
727 		/*
728 		 * This is a public autoconf address that has the same prefix
729 		 * as ours.  If it is preferred, keep it.  We can't break the
730 		 * loop here, because there may be a still-preferred temporary
731 		 * address with the prefix.
732 		 */
733 		if (!IFA6_IS_DEPRECATED(it6))
734 		    public_ifa6 = it6;
735 
736 		if (public_ifa6 != NULL)
737 			ifa_ref(&public_ifa6->ia_ifa);
738 	}
739 	IF_ADDR_RUNLOCK(ifp);
740 
741 	if (public_ifa6 != NULL) {
742 		int e;
743 
744 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
745 			ifa_free(&public_ifa6->ia_ifa);
746 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
747 			    " tmp addr,errno=%d\n", e);
748 			return (-1);
749 		}
750 		ifa_free(&public_ifa6->ia_ifa);
751 		return (0);
752 	}
753 
754 	return (-1);
755 }
756 
757 /*
758  * Nuke neighbor cache/prefix/default router management table, right before
759  * ifp goes away.
760  */
761 void
762 nd6_purge(struct ifnet *ifp)
763 {
764 	struct nd_defrouter *dr, *ndr;
765 	struct nd_prefix *pr, *npr;
766 
767 	/*
768 	 * Nuke default router list entries toward ifp.
769 	 * We defer removal of default router list entries that is installed
770 	 * in the routing table, in order to keep additional side effects as
771 	 * small as possible.
772 	 */
773 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
774 		if (dr->installed)
775 			continue;
776 
777 		if (dr->ifp == ifp)
778 			defrtrlist_del(dr);
779 	}
780 
781 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
782 		if (!dr->installed)
783 			continue;
784 
785 		if (dr->ifp == ifp)
786 			defrtrlist_del(dr);
787 	}
788 
789 	/* Nuke prefix list entries toward ifp */
790 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
791 		if (pr->ndpr_ifp == ifp) {
792 			/*
793 			 * Because if_detach() does *not* release prefixes
794 			 * while purging addresses the reference count will
795 			 * still be above zero. We therefore reset it to
796 			 * make sure that the prefix really gets purged.
797 			 */
798 			pr->ndpr_refcnt = 0;
799 
800 			/*
801 			 * Previously, pr->ndpr_addr is removed as well,
802 			 * but I strongly believe we don't have to do it.
803 			 * nd6_purge() is only called from in6_ifdetach(),
804 			 * which removes all the associated interface addresses
805 			 * by itself.
806 			 * (jinmei@kame.net 20010129)
807 			 */
808 			prelist_remove(pr);
809 		}
810 	}
811 
812 	/* cancel default outgoing interface setting */
813 	if (V_nd6_defifindex == ifp->if_index)
814 		nd6_setdefaultiface(0);
815 
816 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
817 		/* Refresh default router list. */
818 		defrouter_select();
819 	}
820 
821 	/* XXXXX
822 	 * We do not nuke the neighbor cache entries here any more
823 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
824 	 * nd6_purge() is invoked by in6_ifdetach() which is called
825 	 * from if_detach() where everything gets purged. So let
826 	 * in6_domifdetach() do the actual L2 table purging work.
827 	 */
828 }
829 
830 /*
831  * the caller acquires and releases the lock on the lltbls
832  * Returns the llentry locked
833  */
834 struct llentry *
835 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
836 {
837 	struct sockaddr_in6 sin6;
838 	struct llentry *ln;
839 	int llflags;
840 
841 	bzero(&sin6, sizeof(sin6));
842 	sin6.sin6_len = sizeof(struct sockaddr_in6);
843 	sin6.sin6_family = AF_INET6;
844 	sin6.sin6_addr = *addr6;
845 
846 	IF_AFDATA_LOCK_ASSERT(ifp);
847 
848 	llflags = 0;
849 	if (flags & ND6_CREATE)
850 	    llflags |= LLE_CREATE;
851 	if (flags & ND6_EXCLUSIVE)
852 	    llflags |= LLE_EXCLUSIVE;
853 
854 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
855 	if ((ln != NULL) && (llflags & LLE_CREATE))
856 		ln->ln_state = ND6_LLINFO_NOSTATE;
857 
858 	return (ln);
859 }
860 
861 /*
862  * Test whether a given IPv6 address is a neighbor or not, ignoring
863  * the actual neighbor cache.  The neighbor cache is ignored in order
864  * to not reenter the routing code from within itself.
865  */
866 static int
867 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
868 {
869 	struct nd_prefix *pr;
870 	struct ifaddr *dstaddr;
871 
872 	/*
873 	 * A link-local address is always a neighbor.
874 	 * XXX: a link does not necessarily specify a single interface.
875 	 */
876 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
877 		struct sockaddr_in6 sin6_copy;
878 		u_int32_t zone;
879 
880 		/*
881 		 * We need sin6_copy since sa6_recoverscope() may modify the
882 		 * content (XXX).
883 		 */
884 		sin6_copy = *addr;
885 		if (sa6_recoverscope(&sin6_copy))
886 			return (0); /* XXX: should be impossible */
887 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
888 			return (0);
889 		if (sin6_copy.sin6_scope_id == zone)
890 			return (1);
891 		else
892 			return (0);
893 	}
894 
895 	/*
896 	 * If the address matches one of our addresses,
897 	 * it should be a neighbor.
898 	 * If the address matches one of our on-link prefixes, it should be a
899 	 * neighbor.
900 	 */
901 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
902 		if (pr->ndpr_ifp != ifp)
903 			continue;
904 
905 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
906 			struct rtentry *rt;
907 
908 			/* Always use the default FIB here. */
909 			rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
910 			    0, 0, RT_DEFAULT_FIB);
911 			if (rt == NULL)
912 				continue;
913 			/*
914 			 * This is the case where multiple interfaces
915 			 * have the same prefix, but only one is installed
916 			 * into the routing table and that prefix entry
917 			 * is not the one being examined here. In the case
918 			 * where RADIX_MPATH is enabled, multiple route
919 			 * entries (of the same rt_key value) will be
920 			 * installed because the interface addresses all
921 			 * differ.
922 			 */
923 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
924 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
925 				RTFREE_LOCKED(rt);
926 				continue;
927 			}
928 			RTFREE_LOCKED(rt);
929 		}
930 
931 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
932 		    &addr->sin6_addr, &pr->ndpr_mask))
933 			return (1);
934 	}
935 
936 	/*
937 	 * If the address is assigned on the node of the other side of
938 	 * a p2p interface, the address should be a neighbor.
939 	 */
940 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
941 	if (dstaddr != NULL) {
942 		if (dstaddr->ifa_ifp == ifp) {
943 			ifa_free(dstaddr);
944 			return (1);
945 		}
946 		ifa_free(dstaddr);
947 	}
948 
949 	/*
950 	 * If the default router list is empty, all addresses are regarded
951 	 * as on-link, and thus, as a neighbor.
952 	 */
953 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
954 	    TAILQ_EMPTY(&V_nd_defrouter) &&
955 	    V_nd6_defifindex == ifp->if_index) {
956 		return (1);
957 	}
958 
959 	return (0);
960 }
961 
962 
963 /*
964  * Detect if a given IPv6 address identifies a neighbor on a given link.
965  * XXX: should take care of the destination of a p2p link?
966  */
967 int
968 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
969 {
970 	struct llentry *lle;
971 	int rc = 0;
972 
973 	IF_AFDATA_UNLOCK_ASSERT(ifp);
974 	if (nd6_is_new_addr_neighbor(addr, ifp))
975 		return (1);
976 
977 	/*
978 	 * Even if the address matches none of our addresses, it might be
979 	 * in the neighbor cache.
980 	 */
981 	IF_AFDATA_LOCK(ifp);
982 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
983 		LLE_RUNLOCK(lle);
984 		rc = 1;
985 	}
986 	IF_AFDATA_UNLOCK(ifp);
987 	return (rc);
988 }
989 
990 /*
991  * Free an nd6 llinfo entry.
992  * Since the function would cause significant changes in the kernel, DO NOT
993  * make it global, unless you have a strong reason for the change, and are sure
994  * that the change is safe.
995  */
996 static struct llentry *
997 nd6_free(struct llentry *ln, int gc)
998 {
999         struct llentry *next;
1000 	struct nd_defrouter *dr;
1001 	struct ifnet *ifp;
1002 
1003 	LLE_WLOCK_ASSERT(ln);
1004 
1005 	/*
1006 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1007 	 * even though it is not harmful, it was not really necessary.
1008 	 */
1009 
1010 	/* cancel timer */
1011 	nd6_llinfo_settimer_locked(ln, -1);
1012 
1013 	ifp = ln->lle_tbl->llt_ifp;
1014 
1015 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1016 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1017 
1018 		if (dr != NULL && dr->expire &&
1019 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1020 			/*
1021 			 * If the reason for the deletion is just garbage
1022 			 * collection, and the neighbor is an active default
1023 			 * router, do not delete it.  Instead, reset the GC
1024 			 * timer using the router's lifetime.
1025 			 * Simply deleting the entry would affect default
1026 			 * router selection, which is not necessarily a good
1027 			 * thing, especially when we're using router preference
1028 			 * values.
1029 			 * XXX: the check for ln_state would be redundant,
1030 			 *      but we intentionally keep it just in case.
1031 			 */
1032 			if (dr->expire > time_second)
1033 				nd6_llinfo_settimer_locked(ln,
1034 				    (dr->expire - time_second) * hz);
1035 			else
1036 				nd6_llinfo_settimer_locked(ln,
1037 				    (long)V_nd6_gctimer * hz);
1038 
1039 			next = LIST_NEXT(ln, lle_next);
1040 			LLE_REMREF(ln);
1041 			LLE_WUNLOCK(ln);
1042 			return (next);
1043 		}
1044 
1045 		if (dr) {
1046 			/*
1047 			 * Unreachablity of a router might affect the default
1048 			 * router selection and on-link detection of advertised
1049 			 * prefixes.
1050 			 */
1051 
1052 			/*
1053 			 * Temporarily fake the state to choose a new default
1054 			 * router and to perform on-link determination of
1055 			 * prefixes correctly.
1056 			 * Below the state will be set correctly,
1057 			 * or the entry itself will be deleted.
1058 			 */
1059 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1060 		}
1061 
1062 		if (ln->ln_router || dr) {
1063 
1064 			/*
1065 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1066 			 * rnh and for the calls to pfxlist_onlink_check() and
1067 			 * defrouter_select() in the block further down for calls
1068 			 * into nd6_lookup().  We still hold a ref.
1069 			 */
1070 			LLE_WUNLOCK(ln);
1071 
1072 			/*
1073 			 * rt6_flush must be called whether or not the neighbor
1074 			 * is in the Default Router List.
1075 			 * See a corresponding comment in nd6_na_input().
1076 			 */
1077 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1078 		}
1079 
1080 		if (dr) {
1081 			/*
1082 			 * Since defrouter_select() does not affect the
1083 			 * on-link determination and MIP6 needs the check
1084 			 * before the default router selection, we perform
1085 			 * the check now.
1086 			 */
1087 			pfxlist_onlink_check();
1088 
1089 			/*
1090 			 * Refresh default router list.
1091 			 */
1092 			defrouter_select();
1093 		}
1094 
1095 		if (ln->ln_router || dr)
1096 			LLE_WLOCK(ln);
1097 	}
1098 
1099 	/*
1100 	 * Before deleting the entry, remember the next entry as the
1101 	 * return value.  We need this because pfxlist_onlink_check() above
1102 	 * might have freed other entries (particularly the old next entry) as
1103 	 * a side effect (XXX).
1104 	 */
1105 	next = LIST_NEXT(ln, lle_next);
1106 
1107 	/*
1108 	 * Save to unlock. We still hold an extra reference and will not
1109 	 * free(9) in llentry_free() if someone else holds one as well.
1110 	 */
1111 	LLE_WUNLOCK(ln);
1112 	IF_AFDATA_LOCK(ifp);
1113 	LLE_WLOCK(ln);
1114 	LLE_REMREF(ln);
1115 	llentry_free(ln);
1116 	IF_AFDATA_UNLOCK(ifp);
1117 
1118 	return (next);
1119 }
1120 
1121 /*
1122  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1123  *
1124  * XXX cost-effective methods?
1125  */
1126 void
1127 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1128 {
1129 	struct llentry *ln;
1130 	struct ifnet *ifp;
1131 
1132 	if ((dst6 == NULL) || (rt == NULL))
1133 		return;
1134 
1135 	ifp = rt->rt_ifp;
1136 	IF_AFDATA_LOCK(ifp);
1137 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1138 	IF_AFDATA_UNLOCK(ifp);
1139 	if (ln == NULL)
1140 		return;
1141 
1142 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1143 		goto done;
1144 
1145 	/*
1146 	 * if we get upper-layer reachability confirmation many times,
1147 	 * it is possible we have false information.
1148 	 */
1149 	if (!force) {
1150 		ln->ln_byhint++;
1151 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1152 			goto done;
1153 		}
1154 	}
1155 
1156  	ln->ln_state = ND6_LLINFO_REACHABLE;
1157 	if (!ND6_LLINFO_PERMANENT(ln)) {
1158 		nd6_llinfo_settimer_locked(ln,
1159 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1160 	}
1161 done:
1162 	LLE_WUNLOCK(ln);
1163 }
1164 
1165 
1166 /*
1167  * Rejuvenate this function for routing operations related
1168  * processing.
1169  */
1170 void
1171 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1172 {
1173 	struct sockaddr_in6 *gateway;
1174 	struct nd_defrouter *dr;
1175 	struct ifnet *ifp;
1176 
1177 	RT_LOCK_ASSERT(rt);
1178 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1179 	ifp = rt->rt_ifp;
1180 
1181 	switch (req) {
1182 	case RTM_ADD:
1183 		break;
1184 
1185 	case RTM_DELETE:
1186 		if (!ifp)
1187 			return;
1188 		/*
1189 		 * Only indirect routes are interesting.
1190 		 */
1191 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1192 			return;
1193 		/*
1194 		 * check for default route
1195 		 */
1196 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1197 				       &SIN6(rt_key(rt))->sin6_addr)) {
1198 
1199 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1200 			if (dr != NULL)
1201 				dr->installed = 0;
1202 		}
1203 		break;
1204 	}
1205 }
1206 
1207 
1208 int
1209 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1210 {
1211 	struct in6_drlist *drl = (struct in6_drlist *)data;
1212 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1213 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1214 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1215 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1216 	struct nd_defrouter *dr;
1217 	struct nd_prefix *pr;
1218 	int i = 0, error = 0;
1219 	int s;
1220 
1221 	switch (cmd) {
1222 	case SIOCGDRLST_IN6:
1223 		/*
1224 		 * obsolete API, use sysctl under net.inet6.icmp6
1225 		 */
1226 		bzero(drl, sizeof(*drl));
1227 		s = splnet();
1228 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1229 			if (i >= DRLSTSIZ)
1230 				break;
1231 			drl->defrouter[i].rtaddr = dr->rtaddr;
1232 			in6_clearscope(&drl->defrouter[i].rtaddr);
1233 
1234 			drl->defrouter[i].flags = dr->flags;
1235 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1236 			drl->defrouter[i].expire = dr->expire;
1237 			drl->defrouter[i].if_index = dr->ifp->if_index;
1238 			i++;
1239 		}
1240 		splx(s);
1241 		break;
1242 	case SIOCGPRLST_IN6:
1243 		/*
1244 		 * obsolete API, use sysctl under net.inet6.icmp6
1245 		 *
1246 		 * XXX the structure in6_prlist was changed in backward-
1247 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1248 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1249 		 */
1250 		/*
1251 		 * XXX meaning of fields, especialy "raflags", is very
1252 		 * differnet between RA prefix list and RR/static prefix list.
1253 		 * how about separating ioctls into two?
1254 		 */
1255 		bzero(oprl, sizeof(*oprl));
1256 		s = splnet();
1257 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1258 			struct nd_pfxrouter *pfr;
1259 			int j;
1260 
1261 			if (i >= PRLSTSIZ)
1262 				break;
1263 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1264 			oprl->prefix[i].raflags = pr->ndpr_raf;
1265 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1266 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1267 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1268 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1269 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1270 				oprl->prefix[i].expire = 0;
1271 			else {
1272 				time_t maxexpire;
1273 
1274 				/* XXX: we assume time_t is signed. */
1275 				maxexpire = (-1) &
1276 				    ~((time_t)1 <<
1277 				    ((sizeof(maxexpire) * 8) - 1));
1278 				if (pr->ndpr_vltime <
1279 				    maxexpire - pr->ndpr_lastupdate) {
1280 					oprl->prefix[i].expire =
1281 					    pr->ndpr_lastupdate +
1282 					    pr->ndpr_vltime;
1283 				} else
1284 					oprl->prefix[i].expire = maxexpire;
1285 			}
1286 
1287 			j = 0;
1288 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1289 				if (j < DRLSTSIZ) {
1290 #define RTRADDR oprl->prefix[i].advrtr[j]
1291 					RTRADDR = pfr->router->rtaddr;
1292 					in6_clearscope(&RTRADDR);
1293 #undef RTRADDR
1294 				}
1295 				j++;
1296 			}
1297 			oprl->prefix[i].advrtrs = j;
1298 			oprl->prefix[i].origin = PR_ORIG_RA;
1299 
1300 			i++;
1301 		}
1302 		splx(s);
1303 
1304 		break;
1305 	case OSIOCGIFINFO_IN6:
1306 #define ND	ndi->ndi
1307 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1308 		bzero(&ND, sizeof(ND));
1309 		ND.linkmtu = IN6_LINKMTU(ifp);
1310 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1311 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1312 		ND.reachable = ND_IFINFO(ifp)->reachable;
1313 		ND.retrans = ND_IFINFO(ifp)->retrans;
1314 		ND.flags = ND_IFINFO(ifp)->flags;
1315 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1316 		ND.chlim = ND_IFINFO(ifp)->chlim;
1317 		break;
1318 	case SIOCGIFINFO_IN6:
1319 		ND = *ND_IFINFO(ifp);
1320 		break;
1321 	case SIOCSIFINFO_IN6:
1322 		/*
1323 		 * used to change host variables from userland.
1324 		 * intented for a use on router to reflect RA configurations.
1325 		 */
1326 		/* 0 means 'unspecified' */
1327 		if (ND.linkmtu != 0) {
1328 			if (ND.linkmtu < IPV6_MMTU ||
1329 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1330 				error = EINVAL;
1331 				break;
1332 			}
1333 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1334 		}
1335 
1336 		if (ND.basereachable != 0) {
1337 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1338 
1339 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1340 			if (ND.basereachable != obasereachable)
1341 				ND_IFINFO(ifp)->reachable =
1342 				    ND_COMPUTE_RTIME(ND.basereachable);
1343 		}
1344 		if (ND.retrans != 0)
1345 			ND_IFINFO(ifp)->retrans = ND.retrans;
1346 		if (ND.chlim != 0)
1347 			ND_IFINFO(ifp)->chlim = ND.chlim;
1348 		/* FALLTHROUGH */
1349 	case SIOCSIFINFO_FLAGS:
1350 	{
1351 		struct ifaddr *ifa;
1352 		struct in6_ifaddr *ia;
1353 
1354 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1355 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1356 			/* ifdisabled 1->0 transision */
1357 
1358 			/*
1359 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1360 			 * has an link-local address with IN6_IFF_DUPLICATED,
1361 			 * do not clear ND6_IFF_IFDISABLED.
1362 			 * See RFC 4862, Section 5.4.5.
1363 			 */
1364 			int duplicated_linklocal = 0;
1365 
1366 			IF_ADDR_RLOCK(ifp);
1367 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1368 				if (ifa->ifa_addr->sa_family != AF_INET6)
1369 					continue;
1370 				ia = (struct in6_ifaddr *)ifa;
1371 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1372 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1373 					duplicated_linklocal = 1;
1374 					break;
1375 				}
1376 			}
1377 			IF_ADDR_RUNLOCK(ifp);
1378 
1379 			if (duplicated_linklocal) {
1380 				ND.flags |= ND6_IFF_IFDISABLED;
1381 				log(LOG_ERR, "Cannot enable an interface"
1382 				    " with a link-local address marked"
1383 				    " duplicate.\n");
1384 			} else {
1385 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1386 				if (ifp->if_flags & IFF_UP)
1387 					in6_if_up(ifp);
1388 			}
1389 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1390 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1391 			/* ifdisabled 0->1 transision */
1392 			/* Mark all IPv6 address as tentative. */
1393 
1394 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1395 			IF_ADDR_RLOCK(ifp);
1396 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1397 				if (ifa->ifa_addr->sa_family != AF_INET6)
1398 					continue;
1399 				ia = (struct in6_ifaddr *)ifa;
1400 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1401 			}
1402 			IF_ADDR_RUNLOCK(ifp);
1403 		}
1404 
1405 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1406 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1407 				/* auto_linklocal 0->1 transision */
1408 
1409 				/* If no link-local address on ifp, configure */
1410 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1411 				in6_ifattach(ifp, NULL);
1412 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1413 			    ifp->if_flags & IFF_UP) {
1414 				/*
1415 				 * When the IF already has
1416 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1417 				 * address is assigned, and IFF_UP, try to
1418 				 * assign one.
1419 				 */
1420 				int haslinklocal = 0;
1421 
1422 				IF_ADDR_RLOCK(ifp);
1423 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1424 					if (ifa->ifa_addr->sa_family != AF_INET6)
1425 						continue;
1426 					ia = (struct in6_ifaddr *)ifa;
1427 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1428 						haslinklocal = 1;
1429 						break;
1430 					}
1431 				}
1432 				IF_ADDR_RUNLOCK(ifp);
1433 				if (!haslinklocal)
1434 					in6_ifattach(ifp, NULL);
1435 			}
1436 		}
1437 	}
1438 		ND_IFINFO(ifp)->flags = ND.flags;
1439 		break;
1440 #undef ND
1441 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1442 		/* sync kernel routing table with the default router list */
1443 		defrouter_reset();
1444 		defrouter_select();
1445 		break;
1446 	case SIOCSPFXFLUSH_IN6:
1447 	{
1448 		/* flush all the prefix advertised by routers */
1449 		struct nd_prefix *pr, *next;
1450 
1451 		s = splnet();
1452 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1453 			struct in6_ifaddr *ia, *ia_next;
1454 
1455 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1456 				continue; /* XXX */
1457 
1458 			/* do we really have to remove addresses as well? */
1459 			/* XXXRW: in6_ifaddrhead locking. */
1460 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1461 			    ia_next) {
1462 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1463 					continue;
1464 
1465 				if (ia->ia6_ndpr == pr)
1466 					in6_purgeaddr(&ia->ia_ifa);
1467 			}
1468 			prelist_remove(pr);
1469 		}
1470 		splx(s);
1471 		break;
1472 	}
1473 	case SIOCSRTRFLUSH_IN6:
1474 	{
1475 		/* flush all the default routers */
1476 		struct nd_defrouter *dr, *next;
1477 
1478 		s = splnet();
1479 		defrouter_reset();
1480 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1481 			defrtrlist_del(dr);
1482 		}
1483 		defrouter_select();
1484 		splx(s);
1485 		break;
1486 	}
1487 	case SIOCGNBRINFO_IN6:
1488 	{
1489 		struct llentry *ln;
1490 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1491 
1492 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1493 			return (error);
1494 
1495 		IF_AFDATA_LOCK(ifp);
1496 		ln = nd6_lookup(&nb_addr, 0, ifp);
1497 		IF_AFDATA_UNLOCK(ifp);
1498 
1499 		if (ln == NULL) {
1500 			error = EINVAL;
1501 			break;
1502 		}
1503 		nbi->state = ln->ln_state;
1504 		nbi->asked = ln->la_asked;
1505 		nbi->isrouter = ln->ln_router;
1506 		nbi->expire = ln->la_expire;
1507 		LLE_RUNLOCK(ln);
1508 		break;
1509 	}
1510 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1511 		ndif->ifindex = V_nd6_defifindex;
1512 		break;
1513 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1514 		return (nd6_setdefaultiface(ndif->ifindex));
1515 	}
1516 	return (error);
1517 }
1518 
1519 /*
1520  * Create neighbor cache entry and cache link-layer address,
1521  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1522  *
1523  * type - ICMP6 type
1524  * code - type dependent information
1525  *
1526  * XXXXX
1527  *  The caller of this function already acquired the ndp
1528  *  cache table lock because the cache entry is returned.
1529  */
1530 struct llentry *
1531 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1532     int lladdrlen, int type, int code)
1533 {
1534 	struct llentry *ln = NULL;
1535 	int is_newentry;
1536 	int do_update;
1537 	int olladdr;
1538 	int llchange;
1539 	int flags;
1540 	int newstate = 0;
1541 	uint16_t router = 0;
1542 	struct sockaddr_in6 sin6;
1543 	struct mbuf *chain = NULL;
1544 	int static_route = 0;
1545 
1546 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1547 
1548 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1549 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1550 
1551 	/* nothing must be updated for unspecified address */
1552 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1553 		return NULL;
1554 
1555 	/*
1556 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1557 	 * the caller.
1558 	 *
1559 	 * XXX If the link does not have link-layer adderss, what should
1560 	 * we do? (ifp->if_addrlen == 0)
1561 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1562 	 * description on it in NS section (RFC 2461 7.2.3).
1563 	 */
1564 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1565 	IF_AFDATA_LOCK(ifp);
1566 	ln = nd6_lookup(from, flags, ifp);
1567 
1568 	if (ln == NULL) {
1569 		flags |= ND6_EXCLUSIVE;
1570 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1571 		IF_AFDATA_UNLOCK(ifp);
1572 		is_newentry = 1;
1573 	} else {
1574 		IF_AFDATA_UNLOCK(ifp);
1575 		/* do nothing if static ndp is set */
1576 		if (ln->la_flags & LLE_STATIC) {
1577 			static_route = 1;
1578 			goto done;
1579 		}
1580 		is_newentry = 0;
1581 	}
1582 	if (ln == NULL)
1583 		return (NULL);
1584 
1585 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1586 	if (olladdr && lladdr) {
1587 		llchange = bcmp(lladdr, &ln->ll_addr,
1588 		    ifp->if_addrlen);
1589 	} else
1590 		llchange = 0;
1591 
1592 	/*
1593 	 * newentry olladdr  lladdr  llchange	(*=record)
1594 	 *	0	n	n	--	(1)
1595 	 *	0	y	n	--	(2)
1596 	 *	0	n	y	--	(3) * STALE
1597 	 *	0	y	y	n	(4) *
1598 	 *	0	y	y	y	(5) * STALE
1599 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1600 	 *	1	--	y	--	(7) * STALE
1601 	 */
1602 
1603 	if (lladdr) {		/* (3-5) and (7) */
1604 		/*
1605 		 * Record source link-layer address
1606 		 * XXX is it dependent to ifp->if_type?
1607 		 */
1608 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1609 		ln->la_flags |= LLE_VALID;
1610 	}
1611 
1612 	if (!is_newentry) {
1613 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1614 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1615 			do_update = 1;
1616 			newstate = ND6_LLINFO_STALE;
1617 		} else					/* (1-2,4) */
1618 			do_update = 0;
1619 	} else {
1620 		do_update = 1;
1621 		if (lladdr == NULL)			/* (6) */
1622 			newstate = ND6_LLINFO_NOSTATE;
1623 		else					/* (7) */
1624 			newstate = ND6_LLINFO_STALE;
1625 	}
1626 
1627 	if (do_update) {
1628 		/*
1629 		 * Update the state of the neighbor cache.
1630 		 */
1631 		ln->ln_state = newstate;
1632 
1633 		if (ln->ln_state == ND6_LLINFO_STALE) {
1634 			/*
1635 			 * XXX: since nd6_output() below will cause
1636 			 * state tansition to DELAY and reset the timer,
1637 			 * we must set the timer now, although it is actually
1638 			 * meaningless.
1639 			 */
1640 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1641 
1642 			if (ln->la_hold) {
1643 				struct mbuf *m_hold, *m_hold_next;
1644 
1645 				/*
1646 				 * reset the la_hold in advance, to explicitly
1647 				 * prevent a la_hold lookup in nd6_output()
1648 				 * (wouldn't happen, though...)
1649 				 */
1650 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1651 				    m_hold; m_hold = m_hold_next) {
1652 					m_hold_next = m_hold->m_nextpkt;
1653 					m_hold->m_nextpkt = NULL;
1654 
1655 					/*
1656 					 * we assume ifp is not a p2p here, so
1657 					 * just set the 2nd argument as the
1658 					 * 1st one.
1659 					 */
1660 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1661 				}
1662 				/*
1663 				 * If we have mbufs in the chain we need to do
1664 				 * deferred transmit. Copy the address from the
1665 				 * llentry before dropping the lock down below.
1666 				 */
1667 				if (chain != NULL)
1668 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1669 			}
1670 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1671 			/* probe right away */
1672 			nd6_llinfo_settimer_locked((void *)ln, 0);
1673 		}
1674 	}
1675 
1676 	/*
1677 	 * ICMP6 type dependent behavior.
1678 	 *
1679 	 * NS: clear IsRouter if new entry
1680 	 * RS: clear IsRouter
1681 	 * RA: set IsRouter if there's lladdr
1682 	 * redir: clear IsRouter if new entry
1683 	 *
1684 	 * RA case, (1):
1685 	 * The spec says that we must set IsRouter in the following cases:
1686 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1687 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1688 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1689 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1690 	 * neighbor cache, this is similar to (6).
1691 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1692 	 *
1693 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1694 	 *							D R
1695 	 *	0	n	n	--	(1)	c   ?     s
1696 	 *	0	y	n	--	(2)	c   s     s
1697 	 *	0	n	y	--	(3)	c   s     s
1698 	 *	0	y	y	n	(4)	c   s     s
1699 	 *	0	y	y	y	(5)	c   s     s
1700 	 *	1	--	n	--	(6) c	c	c s
1701 	 *	1	--	y	--	(7) c	c   s	c s
1702 	 *
1703 	 *					(c=clear s=set)
1704 	 */
1705 	switch (type & 0xff) {
1706 	case ND_NEIGHBOR_SOLICIT:
1707 		/*
1708 		 * New entry must have is_router flag cleared.
1709 		 */
1710 		if (is_newentry)	/* (6-7) */
1711 			ln->ln_router = 0;
1712 		break;
1713 	case ND_REDIRECT:
1714 		/*
1715 		 * If the icmp is a redirect to a better router, always set the
1716 		 * is_router flag.  Otherwise, if the entry is newly created,
1717 		 * clear the flag.  [RFC 2461, sec 8.3]
1718 		 */
1719 		if (code == ND_REDIRECT_ROUTER)
1720 			ln->ln_router = 1;
1721 		else if (is_newentry) /* (6-7) */
1722 			ln->ln_router = 0;
1723 		break;
1724 	case ND_ROUTER_SOLICIT:
1725 		/*
1726 		 * is_router flag must always be cleared.
1727 		 */
1728 		ln->ln_router = 0;
1729 		break;
1730 	case ND_ROUTER_ADVERT:
1731 		/*
1732 		 * Mark an entry with lladdr as a router.
1733 		 */
1734 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1735 		    (is_newentry && lladdr)) {			/* (7) */
1736 			ln->ln_router = 1;
1737 		}
1738 		break;
1739 	}
1740 
1741 	if (ln != NULL) {
1742 		static_route = (ln->la_flags & LLE_STATIC);
1743 		router = ln->ln_router;
1744 
1745 		if (flags & ND6_EXCLUSIVE)
1746 			LLE_WUNLOCK(ln);
1747 		else
1748 			LLE_RUNLOCK(ln);
1749 		if (static_route)
1750 			ln = NULL;
1751 	}
1752 	if (chain)
1753 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1754 
1755 	/*
1756 	 * When the link-layer address of a router changes, select the
1757 	 * best router again.  In particular, when the neighbor entry is newly
1758 	 * created, it might affect the selection policy.
1759 	 * Question: can we restrict the first condition to the "is_newentry"
1760 	 * case?
1761 	 * XXX: when we hear an RA from a new router with the link-layer
1762 	 * address option, defrouter_select() is called twice, since
1763 	 * defrtrlist_update called the function as well.  However, I believe
1764 	 * we can compromise the overhead, since it only happens the first
1765 	 * time.
1766 	 * XXX: although defrouter_select() should not have a bad effect
1767 	 * for those are not autoconfigured hosts, we explicitly avoid such
1768 	 * cases for safety.
1769 	 */
1770 	if (do_update && router &&
1771 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1772 		/*
1773 		 * guaranteed recursion
1774 		 */
1775 		defrouter_select();
1776 	}
1777 
1778 	return (ln);
1779 done:
1780 	if (ln != NULL) {
1781 		if (flags & ND6_EXCLUSIVE)
1782 			LLE_WUNLOCK(ln);
1783 		else
1784 			LLE_RUNLOCK(ln);
1785 		if (static_route)
1786 			ln = NULL;
1787 	}
1788 	return (ln);
1789 }
1790 
1791 static void
1792 nd6_slowtimo(void *arg)
1793 {
1794 	CURVNET_SET((struct vnet *) arg);
1795 	struct nd_ifinfo *nd6if;
1796 	struct ifnet *ifp;
1797 
1798 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1799 	    nd6_slowtimo, curvnet);
1800 	IFNET_RLOCK_NOSLEEP();
1801 	TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1802 		nd6if = ND_IFINFO(ifp);
1803 		if (nd6if->basereachable && /* already initialized */
1804 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1805 			/*
1806 			 * Since reachable time rarely changes by router
1807 			 * advertisements, we SHOULD insure that a new random
1808 			 * value gets recomputed at least once every few hours.
1809 			 * (RFC 2461, 6.3.4)
1810 			 */
1811 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1812 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1813 		}
1814 	}
1815 	IFNET_RUNLOCK_NOSLEEP();
1816 	CURVNET_RESTORE();
1817 }
1818 
1819 int
1820 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1821     struct sockaddr_in6 *dst, struct rtentry *rt0)
1822 {
1823 
1824 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1825 }
1826 
1827 
1828 /*
1829  * Note that I'm not enforcing any global serialization
1830  * lle state or asked changes here as the logic is too
1831  * complicated to avoid having to always acquire an exclusive
1832  * lock
1833  * KMM
1834  *
1835  */
1836 #define senderr(e) { error = (e); goto bad;}
1837 
1838 int
1839 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1840     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1841 	struct mbuf **chain)
1842 {
1843 	struct mbuf *m = m0;
1844 	struct m_tag *mtag;
1845 	struct llentry *ln = lle;
1846 	struct ip6_hdr *ip6;
1847 	int error = 0;
1848 	int flags = 0;
1849 	int ip6len;
1850 
1851 #ifdef INVARIANTS
1852 	if (lle != NULL) {
1853 
1854 		LLE_WLOCK_ASSERT(lle);
1855 
1856 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1857 	}
1858 #endif
1859 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1860 		goto sendpkt;
1861 
1862 	if (nd6_need_cache(ifp) == 0)
1863 		goto sendpkt;
1864 
1865 	/*
1866 	 * next hop determination.  This routine is derived from ether_output.
1867 	 */
1868 
1869 	/*
1870 	 * Address resolution or Neighbor Unreachability Detection
1871 	 * for the next hop.
1872 	 * At this point, the destination of the packet must be a unicast
1873 	 * or an anycast address(i.e. not a multicast).
1874 	 */
1875 
1876 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1877 	if (ln == NULL) {
1878 	retry:
1879 		IF_AFDATA_LOCK(ifp);
1880 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1881 		IF_AFDATA_UNLOCK(ifp);
1882 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1883 			/*
1884 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1885 			 * the condition below is not very efficient.  But we believe
1886 			 * it is tolerable, because this should be a rare case.
1887 			 */
1888 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1889 			IF_AFDATA_LOCK(ifp);
1890 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1891 			IF_AFDATA_UNLOCK(ifp);
1892 		}
1893 	}
1894 	if (ln == NULL) {
1895 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1896 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1897 			char ip6buf[INET6_ADDRSTRLEN];
1898 			log(LOG_DEBUG,
1899 			    "nd6_output: can't allocate llinfo for %s "
1900 			    "(ln=%p)\n",
1901 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1902 			senderr(EIO);	/* XXX: good error? */
1903 		}
1904 		goto sendpkt;	/* send anyway */
1905 	}
1906 
1907 	/* We don't have to do link-layer address resolution on a p2p link. */
1908 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1909 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1910 		if ((flags & LLE_EXCLUSIVE) == 0) {
1911 			flags |= LLE_EXCLUSIVE;
1912 			goto retry;
1913 		}
1914 		ln->ln_state = ND6_LLINFO_STALE;
1915 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1916 	}
1917 
1918 	/*
1919 	 * The first time we send a packet to a neighbor whose entry is
1920 	 * STALE, we have to change the state to DELAY and a sets a timer to
1921 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1922 	 * neighbor unreachability detection on expiration.
1923 	 * (RFC 2461 7.3.3)
1924 	 */
1925 	if (ln->ln_state == ND6_LLINFO_STALE) {
1926 		if ((flags & LLE_EXCLUSIVE) == 0) {
1927 			flags |= LLE_EXCLUSIVE;
1928 			LLE_RUNLOCK(ln);
1929 			goto retry;
1930 		}
1931 		ln->la_asked = 0;
1932 		ln->ln_state = ND6_LLINFO_DELAY;
1933 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1934 	}
1935 
1936 	/*
1937 	 * If the neighbor cache entry has a state other than INCOMPLETE
1938 	 * (i.e. its link-layer address is already resolved), just
1939 	 * send the packet.
1940 	 */
1941 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1942 		goto sendpkt;
1943 
1944 	/*
1945 	 * There is a neighbor cache entry, but no ethernet address
1946 	 * response yet.  Append this latest packet to the end of the
1947 	 * packet queue in the mbuf, unless the number of the packet
1948 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1949 	 * the oldest packet in the queue will be removed.
1950 	 */
1951 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1952 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1953 
1954 	if ((flags & LLE_EXCLUSIVE) == 0) {
1955 		flags |= LLE_EXCLUSIVE;
1956 		LLE_RUNLOCK(ln);
1957 		goto retry;
1958 	}
1959 
1960 	LLE_WLOCK_ASSERT(ln);
1961 
1962 	if (ln->la_hold) {
1963 		struct mbuf *m_hold;
1964 		int i;
1965 
1966 		i = 0;
1967 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1968 			i++;
1969 			if (m_hold->m_nextpkt == NULL) {
1970 				m_hold->m_nextpkt = m;
1971 				break;
1972 			}
1973 		}
1974 		while (i >= V_nd6_maxqueuelen) {
1975 			m_hold = ln->la_hold;
1976 			ln->la_hold = ln->la_hold->m_nextpkt;
1977 			m_freem(m_hold);
1978 			i--;
1979 		}
1980 	} else {
1981 		ln->la_hold = m;
1982 	}
1983 
1984 	/*
1985 	 * If there has been no NS for the neighbor after entering the
1986 	 * INCOMPLETE state, send the first solicitation.
1987 	 */
1988 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1989 		ln->la_asked++;
1990 
1991 		nd6_llinfo_settimer_locked(ln,
1992 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1993 		LLE_WUNLOCK(ln);
1994 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1995 		if (lle != NULL && ln == lle)
1996 			LLE_WLOCK(lle);
1997 
1998 	} else if (lle == NULL || ln != lle) {
1999 		/*
2000 		 * We did the lookup (no lle arg) so we
2001 		 * need to do the unlock here.
2002 		 */
2003 		LLE_WUNLOCK(ln);
2004 	}
2005 
2006 	return (0);
2007 
2008   sendpkt:
2009 	/* discard the packet if IPv6 operation is disabled on the interface */
2010 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2011 		error = ENETDOWN; /* better error? */
2012 		goto bad;
2013 	}
2014 	/*
2015 	 * ln is valid and the caller did not pass in
2016 	 * an llentry
2017 	 */
2018 	if ((ln != NULL) && (lle == NULL)) {
2019 		if (flags & LLE_EXCLUSIVE)
2020 			LLE_WUNLOCK(ln);
2021 		else
2022 			LLE_RUNLOCK(ln);
2023 	}
2024 
2025 #ifdef MAC
2026 	mac_netinet6_nd6_send(ifp, m);
2027 #endif
2028 
2029 	/*
2030 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2031 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2032 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2033 	 * to be diverted to user space.  When re-injected into the kernel,
2034 	 * send_output() will directly dispatch them to the outgoing interface.
2035 	 */
2036 	if (send_sendso_input_hook != NULL) {
2037 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2038 		if (mtag != NULL) {
2039 			ip6 = mtod(m, struct ip6_hdr *);
2040 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2041 			/* Use the SEND socket */
2042 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2043 			    ip6len);
2044 			/* -1 == no app on SEND socket */
2045 			if (error == 0 || error != -1)
2046 			    return (error);
2047 		}
2048 	}
2049 
2050 	/*
2051 	 * We were passed in a pointer to an lle with the lock held
2052 	 * this means that we can't call if_output as we will
2053 	 * recurse on the lle lock - so what we do is we create
2054 	 * a list of mbufs to send and transmit them in the caller
2055 	 * after the lock is dropped
2056 	 */
2057 	if (lle != NULL) {
2058 		if (*chain == NULL)
2059 			*chain = m;
2060 		else {
2061 			struct mbuf *mb;
2062 
2063 			/*
2064 			 * append mbuf to end of deferred chain
2065 			 */
2066 			mb = *chain;
2067 			while (mb->m_nextpkt != NULL)
2068 				mb = mb->m_nextpkt;
2069 			mb->m_nextpkt = m;
2070 		}
2071 		return (error);
2072 	}
2073 	/* Reset layer specific mbuf flags to avoid confusing lower layers. */
2074 	m->m_flags &= ~(M_PROTOFLAGS);
2075 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2076 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2077 		    NULL));
2078 	}
2079 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2080 	return (error);
2081 
2082   bad:
2083 	/*
2084 	 * ln is valid and the caller did not pass in
2085 	 * an llentry
2086 	 */
2087 	if ((ln != NULL) && (lle == NULL)) {
2088 		if (flags & LLE_EXCLUSIVE)
2089 			LLE_WUNLOCK(ln);
2090 		else
2091 			LLE_RUNLOCK(ln);
2092 	}
2093 	if (m)
2094 		m_freem(m);
2095 	return (error);
2096 }
2097 #undef senderr
2098 
2099 
2100 int
2101 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2102     struct sockaddr_in6 *dst, struct route *ro)
2103 {
2104 	struct mbuf *m, *m_head;
2105 	struct ifnet *outifp;
2106 	int error = 0;
2107 
2108 	m_head = chain;
2109 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2110 		outifp = origifp;
2111 	else
2112 		outifp = ifp;
2113 
2114 	while (m_head) {
2115 		m = m_head;
2116 		m_head = m_head->m_nextpkt;
2117 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2118 	}
2119 
2120 	/*
2121 	 * XXX
2122 	 * note that intermediate errors are blindly ignored - but this is
2123 	 * the same convention as used with nd6_output when called by
2124 	 * nd6_cache_lladdr
2125 	 */
2126 	return (error);
2127 }
2128 
2129 
2130 int
2131 nd6_need_cache(struct ifnet *ifp)
2132 {
2133 	/*
2134 	 * XXX: we currently do not make neighbor cache on any interface
2135 	 * other than ARCnet, Ethernet, FDDI and GIF.
2136 	 *
2137 	 * RFC2893 says:
2138 	 * - unidirectional tunnels needs no ND
2139 	 */
2140 	switch (ifp->if_type) {
2141 	case IFT_ARCNET:
2142 	case IFT_ETHER:
2143 	case IFT_FDDI:
2144 	case IFT_IEEE1394:
2145 #ifdef IFT_L2VLAN
2146 	case IFT_L2VLAN:
2147 #endif
2148 #ifdef IFT_IEEE80211
2149 	case IFT_IEEE80211:
2150 #endif
2151 	case IFT_INFINIBAND:
2152 	case IFT_GIF:		/* XXX need more cases? */
2153 	case IFT_PPP:
2154 	case IFT_TUNNEL:
2155 	case IFT_BRIDGE:
2156 	case IFT_PROPVIRTUAL:
2157 		return (1);
2158 	default:
2159 		return (0);
2160 	}
2161 }
2162 
2163 /*
2164  * the callers of this function need to be re-worked to drop
2165  * the lle lock, drop here for now
2166  */
2167 int
2168 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2169     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2170 {
2171 	struct llentry *ln;
2172 
2173 	*lle = NULL;
2174 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2175 	if (m->m_flags & M_MCAST) {
2176 		int i;
2177 
2178 		switch (ifp->if_type) {
2179 		case IFT_ETHER:
2180 		case IFT_FDDI:
2181 #ifdef IFT_L2VLAN
2182 		case IFT_L2VLAN:
2183 #endif
2184 #ifdef IFT_IEEE80211
2185 		case IFT_IEEE80211:
2186 #endif
2187 		case IFT_BRIDGE:
2188 		case IFT_ISO88025:
2189 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2190 						 desten);
2191 			return (0);
2192 		case IFT_IEEE1394:
2193 			/*
2194 			 * netbsd can use if_broadcastaddr, but we don't do so
2195 			 * to reduce # of ifdef.
2196 			 */
2197 			for (i = 0; i < ifp->if_addrlen; i++)
2198 				desten[i] = ~0;
2199 			return (0);
2200 		case IFT_ARCNET:
2201 			*desten = 0;
2202 			return (0);
2203 		default:
2204 			m_freem(m);
2205 			return (EAFNOSUPPORT);
2206 		}
2207 	}
2208 
2209 
2210 	/*
2211 	 * the entry should have been created in nd6_store_lladdr
2212 	 */
2213 	IF_AFDATA_LOCK(ifp);
2214 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2215 	IF_AFDATA_UNLOCK(ifp);
2216 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2217 		if (ln != NULL)
2218 			LLE_RUNLOCK(ln);
2219 		/* this could happen, if we could not allocate memory */
2220 		m_freem(m);
2221 		return (1);
2222 	}
2223 
2224 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2225 	*lle = ln;
2226 	LLE_RUNLOCK(ln);
2227 	/*
2228 	 * A *small* use after free race exists here
2229 	 */
2230 	return (0);
2231 }
2232 
2233 static void
2234 clear_llinfo_pqueue(struct llentry *ln)
2235 {
2236 	struct mbuf *m_hold, *m_hold_next;
2237 
2238 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2239 		m_hold_next = m_hold->m_nextpkt;
2240 		m_freem(m_hold);
2241 	}
2242 
2243 	ln->la_hold = NULL;
2244 	return;
2245 }
2246 
2247 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2248 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2249 #ifdef SYSCTL_DECL
2250 SYSCTL_DECL(_net_inet6_icmp6);
2251 #endif
2252 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2253 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2254 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2255 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2256 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2257 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2258 
2259 static int
2260 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2261 {
2262 	struct in6_defrouter d;
2263 	struct nd_defrouter *dr;
2264 	int error;
2265 
2266 	if (req->newptr)
2267 		return (EPERM);
2268 
2269 	bzero(&d, sizeof(d));
2270 	d.rtaddr.sin6_family = AF_INET6;
2271 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2272 
2273 	/*
2274 	 * XXX locking
2275 	 */
2276 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2277 		d.rtaddr.sin6_addr = dr->rtaddr;
2278 		error = sa6_recoverscope(&d.rtaddr);
2279 		if (error != 0)
2280 			return (error);
2281 		d.flags = dr->flags;
2282 		d.rtlifetime = dr->rtlifetime;
2283 		d.expire = dr->expire;
2284 		d.if_index = dr->ifp->if_index;
2285 		error = SYSCTL_OUT(req, &d, sizeof(d));
2286 		if (error != 0)
2287 			return (error);
2288 	}
2289 	return (0);
2290 }
2291 
2292 static int
2293 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2294 {
2295 	struct in6_prefix p;
2296 	struct sockaddr_in6 s6;
2297 	struct nd_prefix *pr;
2298 	struct nd_pfxrouter *pfr;
2299 	time_t maxexpire;
2300 	int error;
2301 	char ip6buf[INET6_ADDRSTRLEN];
2302 
2303 	if (req->newptr)
2304 		return (EPERM);
2305 
2306 	bzero(&p, sizeof(p));
2307 	p.origin = PR_ORIG_RA;
2308 	bzero(&s6, sizeof(s6));
2309 	s6.sin6_family = AF_INET6;
2310 	s6.sin6_len = sizeof(s6);
2311 
2312 	/*
2313 	 * XXX locking
2314 	 */
2315 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2316 		p.prefix = pr->ndpr_prefix;
2317 		if (sa6_recoverscope(&p.prefix)) {
2318 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2319 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2320 			/* XXX: press on... */
2321 		}
2322 		p.raflags = pr->ndpr_raf;
2323 		p.prefixlen = pr->ndpr_plen;
2324 		p.vltime = pr->ndpr_vltime;
2325 		p.pltime = pr->ndpr_pltime;
2326 		p.if_index = pr->ndpr_ifp->if_index;
2327 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2328 			p.expire = 0;
2329 		else {
2330 			/* XXX: we assume time_t is signed. */
2331 			maxexpire = (-1) &
2332 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2333 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2334 				p.expire = pr->ndpr_lastupdate +
2335 				    pr->ndpr_vltime;
2336 			else
2337 				p.expire = maxexpire;
2338 		}
2339 		p.refcnt = pr->ndpr_refcnt;
2340 		p.flags = pr->ndpr_stateflags;
2341 		p.advrtrs = 0;
2342 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2343 			p.advrtrs++;
2344 		error = SYSCTL_OUT(req, &p, sizeof(p));
2345 		if (error != 0)
2346 			return (error);
2347 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2348 			s6.sin6_addr = pfr->router->rtaddr;
2349 			if (sa6_recoverscope(&s6))
2350 				log(LOG_ERR,
2351 				    "scope error in prefix list (%s)\n",
2352 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2353 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2354 			if (error != 0)
2355 				return (error);
2356 		}
2357 	}
2358 	return (0);
2359 }
2360