xref: /freebsd/sys/netinet6/nd6.c (revision a3cf0ef5a295c885c895fabfd56470c0d1db322d)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sysctl.h>
54 
55 #include <net/if.h>
56 #include <net/if_arc.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/iso88025.h>
60 #include <net/fddi.h>
61 #include <net/route.h>
62 #include <net/vnet.h>
63 
64 #include <netinet/in.h>
65 #include <net/if_llatbl.h>
66 #define	L3_ADDR_SIN6(le)	((struct sockaddr_in6 *) L3_ADDR(le))
67 #include <netinet/if_ether.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet6/scope6_var.h>
72 #include <netinet6/nd6.h>
73 #include <netinet6/in6_ifattach.h>
74 #include <netinet/icmp6.h>
75 #include <netinet6/send.h>
76 
77 #include <sys/limits.h>
78 
79 #include <security/mac/mac_framework.h>
80 
81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
83 
84 #define SIN6(s) ((struct sockaddr_in6 *)s)
85 
86 /* timer values */
87 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
88 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
89 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
90 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
91 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
92 					 * local traffic */
93 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
94 					 * collection timer */
95 
96 /* preventing too many loops in ND option parsing */
97 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
98 
99 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
100 					 * layer hints */
101 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
102 					 * ND entries */
103 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
104 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
105 
106 #ifdef ND6_DEBUG
107 VNET_DEFINE(int, nd6_debug) = 1;
108 #else
109 VNET_DEFINE(int, nd6_debug) = 0;
110 #endif
111 
112 /* for debugging? */
113 #if 0
114 static int nd6_inuse, nd6_allocated;
115 #endif
116 
117 VNET_DEFINE(struct nd_drhead, nd_defrouter);
118 VNET_DEFINE(struct nd_prhead, nd_prefix);
119 
120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
121 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
122 
123 static struct sockaddr_in6 all1_sa;
124 
125 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
126 
127 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
128 	struct ifnet *));
129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
130 static void nd6_slowtimo(void *);
131 static int regen_tmpaddr(struct in6_ifaddr *);
132 static struct llentry *nd6_free(struct llentry *, int);
133 static void nd6_llinfo_timer(void *);
134 static void clear_llinfo_pqueue(struct llentry *);
135 
136 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
137 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
138 
139 VNET_DEFINE(struct callout, nd6_timer_ch);
140 
141 void
142 nd6_init(void)
143 {
144 	int i;
145 
146 	LIST_INIT(&V_nd_prefix);
147 
148 	all1_sa.sin6_family = AF_INET6;
149 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
150 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
151 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
152 
153 	/* initialization of the default router list */
154 	TAILQ_INIT(&V_nd_defrouter);
155 
156 	/* start timer */
157 	callout_init(&V_nd6_slowtimo_ch, 0);
158 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
159 	    nd6_slowtimo, curvnet);
160 }
161 
162 #ifdef VIMAGE
163 void
164 nd6_destroy()
165 {
166 
167 	callout_drain(&V_nd6_slowtimo_ch);
168 	callout_drain(&V_nd6_timer_ch);
169 }
170 #endif
171 
172 struct nd_ifinfo *
173 nd6_ifattach(struct ifnet *ifp)
174 {
175 	struct nd_ifinfo *nd;
176 
177 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
178 	bzero(nd, sizeof(*nd));
179 
180 	nd->initialized = 1;
181 
182 	nd->chlim = IPV6_DEFHLIM;
183 	nd->basereachable = REACHABLE_TIME;
184 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
185 	nd->retrans = RETRANS_TIMER;
186 
187 	nd->flags = ND6_IFF_PERFORMNUD;
188 
189 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */
190 	if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK))
191 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
192 
193 	/* A loopback interface does not need to accept RTADV. */
194 	if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK))
195 		nd->flags |= ND6_IFF_ACCEPT_RTADV;
196 
197 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
198 	nd6_setmtu0(ifp, nd);
199 
200 	return nd;
201 }
202 
203 void
204 nd6_ifdetach(struct nd_ifinfo *nd)
205 {
206 
207 	free(nd, M_IP6NDP);
208 }
209 
210 /*
211  * Reset ND level link MTU. This function is called when the physical MTU
212  * changes, which means we might have to adjust the ND level MTU.
213  */
214 void
215 nd6_setmtu(struct ifnet *ifp)
216 {
217 
218 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
219 }
220 
221 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
222 void
223 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
224 {
225 	u_int32_t omaxmtu;
226 
227 	omaxmtu = ndi->maxmtu;
228 
229 	switch (ifp->if_type) {
230 	case IFT_ARCNET:
231 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
232 		break;
233 	case IFT_FDDI:
234 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
235 		break;
236 	case IFT_ISO88025:
237 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
238 		 break;
239 	default:
240 		ndi->maxmtu = ifp->if_mtu;
241 		break;
242 	}
243 
244 	/*
245 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
246 	 * undesirable situation.  We thus notify the operator of the change
247 	 * explicitly.  The check for omaxmtu is necessary to restrict the
248 	 * log to the case of changing the MTU, not initializing it.
249 	 */
250 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
251 		log(LOG_NOTICE, "nd6_setmtu0: "
252 		    "new link MTU on %s (%lu) is too small for IPv6\n",
253 		    if_name(ifp), (unsigned long)ndi->maxmtu);
254 	}
255 
256 	if (ndi->maxmtu > V_in6_maxmtu)
257 		in6_setmaxmtu(); /* check all interfaces just in case */
258 
259 }
260 
261 void
262 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
263 {
264 
265 	bzero(ndopts, sizeof(*ndopts));
266 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
267 	ndopts->nd_opts_last
268 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
269 
270 	if (icmp6len == 0) {
271 		ndopts->nd_opts_done = 1;
272 		ndopts->nd_opts_search = NULL;
273 	}
274 }
275 
276 /*
277  * Take one ND option.
278  */
279 struct nd_opt_hdr *
280 nd6_option(union nd_opts *ndopts)
281 {
282 	struct nd_opt_hdr *nd_opt;
283 	int olen;
284 
285 	if (ndopts == NULL)
286 		panic("ndopts == NULL in nd6_option");
287 	if (ndopts->nd_opts_last == NULL)
288 		panic("uninitialized ndopts in nd6_option");
289 	if (ndopts->nd_opts_search == NULL)
290 		return NULL;
291 	if (ndopts->nd_opts_done)
292 		return NULL;
293 
294 	nd_opt = ndopts->nd_opts_search;
295 
296 	/* make sure nd_opt_len is inside the buffer */
297 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
298 		bzero(ndopts, sizeof(*ndopts));
299 		return NULL;
300 	}
301 
302 	olen = nd_opt->nd_opt_len << 3;
303 	if (olen == 0) {
304 		/*
305 		 * Message validation requires that all included
306 		 * options have a length that is greater than zero.
307 		 */
308 		bzero(ndopts, sizeof(*ndopts));
309 		return NULL;
310 	}
311 
312 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
313 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
314 		/* option overruns the end of buffer, invalid */
315 		bzero(ndopts, sizeof(*ndopts));
316 		return NULL;
317 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
318 		/* reached the end of options chain */
319 		ndopts->nd_opts_done = 1;
320 		ndopts->nd_opts_search = NULL;
321 	}
322 	return nd_opt;
323 }
324 
325 /*
326  * Parse multiple ND options.
327  * This function is much easier to use, for ND routines that do not need
328  * multiple options of the same type.
329  */
330 int
331 nd6_options(union nd_opts *ndopts)
332 {
333 	struct nd_opt_hdr *nd_opt;
334 	int i = 0;
335 
336 	if (ndopts == NULL)
337 		panic("ndopts == NULL in nd6_options");
338 	if (ndopts->nd_opts_last == NULL)
339 		panic("uninitialized ndopts in nd6_options");
340 	if (ndopts->nd_opts_search == NULL)
341 		return 0;
342 
343 	while (1) {
344 		nd_opt = nd6_option(ndopts);
345 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
346 			/*
347 			 * Message validation requires that all included
348 			 * options have a length that is greater than zero.
349 			 */
350 			ICMP6STAT_INC(icp6s_nd_badopt);
351 			bzero(ndopts, sizeof(*ndopts));
352 			return -1;
353 		}
354 
355 		if (nd_opt == NULL)
356 			goto skip1;
357 
358 		switch (nd_opt->nd_opt_type) {
359 		case ND_OPT_SOURCE_LINKADDR:
360 		case ND_OPT_TARGET_LINKADDR:
361 		case ND_OPT_MTU:
362 		case ND_OPT_REDIRECTED_HEADER:
363 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
364 				nd6log((LOG_INFO,
365 				    "duplicated ND6 option found (type=%d)\n",
366 				    nd_opt->nd_opt_type));
367 				/* XXX bark? */
368 			} else {
369 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
370 					= nd_opt;
371 			}
372 			break;
373 		case ND_OPT_PREFIX_INFORMATION:
374 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
375 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
376 					= nd_opt;
377 			}
378 			ndopts->nd_opts_pi_end =
379 				(struct nd_opt_prefix_info *)nd_opt;
380 			break;
381 		default:
382 			/*
383 			 * Unknown options must be silently ignored,
384 			 * to accomodate future extension to the protocol.
385 			 */
386 			nd6log((LOG_DEBUG,
387 			    "nd6_options: unsupported option %d - "
388 			    "option ignored\n", nd_opt->nd_opt_type));
389 		}
390 
391 skip1:
392 		i++;
393 		if (i > V_nd6_maxndopt) {
394 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
395 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
396 			break;
397 		}
398 
399 		if (ndopts->nd_opts_done)
400 			break;
401 	}
402 
403 	return 0;
404 }
405 
406 /*
407  * ND6 timer routine to handle ND6 entries
408  */
409 void
410 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
411 {
412 	int canceled;
413 
414 	if (tick < 0) {
415 		ln->la_expire = 0;
416 		ln->ln_ntick = 0;
417 		canceled = callout_stop(&ln->ln_timer_ch);
418 	} else {
419 		ln->la_expire = time_second + tick / hz;
420 		LLE_ADDREF(ln);
421 		if (tick > INT_MAX) {
422 			ln->ln_ntick = tick - INT_MAX;
423 			canceled = callout_reset(&ln->ln_timer_ch, INT_MAX,
424 			    nd6_llinfo_timer, ln);
425 		} else {
426 			ln->ln_ntick = 0;
427 			canceled = callout_reset(&ln->ln_timer_ch, tick,
428 			    nd6_llinfo_timer, ln);
429 		}
430 	}
431 	if (canceled)
432 		LLE_REMREF(ln);
433 }
434 
435 void
436 nd6_llinfo_settimer(struct llentry *ln, long tick)
437 {
438 
439 	LLE_WLOCK(ln);
440 	nd6_llinfo_settimer_locked(ln, tick);
441 	LLE_WUNLOCK(ln);
442 }
443 
444 static void
445 nd6_llinfo_timer(void *arg)
446 {
447 	struct llentry *ln;
448 	struct in6_addr *dst;
449 	struct ifnet *ifp;
450 	struct nd_ifinfo *ndi = NULL;
451 
452 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
453 	ln = (struct llentry *)arg;
454 	ifp = ln->lle_tbl->llt_ifp;
455 
456 	CURVNET_SET(ifp->if_vnet);
457 
458 	if (ln->ln_ntick > 0) {
459 		if (ln->ln_ntick > INT_MAX) {
460 			ln->ln_ntick -= INT_MAX;
461 			nd6_llinfo_settimer(ln, INT_MAX);
462 		} else {
463 			ln->ln_ntick = 0;
464 			nd6_llinfo_settimer(ln, ln->ln_ntick);
465 		}
466 		goto done;
467 	}
468 
469 	ndi = ND_IFINFO(ifp);
470 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
471 	if (ln->la_flags & LLE_STATIC) {
472 		goto done;
473 	}
474 
475 	if (ln->la_flags & LLE_DELETED) {
476 		(void)nd6_free(ln, 0);
477 		ln = NULL;
478 		goto done;
479 	}
480 
481 	switch (ln->ln_state) {
482 	case ND6_LLINFO_INCOMPLETE:
483 		if (ln->la_asked < V_nd6_mmaxtries) {
484 			ln->la_asked++;
485 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
486 			nd6_ns_output(ifp, NULL, dst, ln, 0);
487 		} else {
488 			struct mbuf *m = ln->la_hold;
489 			if (m) {
490 				struct mbuf *m0;
491 
492 				/*
493 				 * assuming every packet in la_hold has the
494 				 * same IP header
495 				 */
496 				m0 = m->m_nextpkt;
497 				m->m_nextpkt = NULL;
498 				icmp6_error2(m, ICMP6_DST_UNREACH,
499 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
500 
501 				ln->la_hold = m0;
502 				clear_llinfo_pqueue(ln);
503 			}
504 			(void)nd6_free(ln, 0);
505 			ln = NULL;
506 		}
507 		break;
508 	case ND6_LLINFO_REACHABLE:
509 		if (!ND6_LLINFO_PERMANENT(ln)) {
510 			ln->ln_state = ND6_LLINFO_STALE;
511 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
512 		}
513 		break;
514 
515 	case ND6_LLINFO_STALE:
516 		/* Garbage Collection(RFC 2461 5.3) */
517 		if (!ND6_LLINFO_PERMANENT(ln)) {
518 			(void)nd6_free(ln, 1);
519 			ln = NULL;
520 		}
521 		break;
522 
523 	case ND6_LLINFO_DELAY:
524 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
525 			/* We need NUD */
526 			ln->la_asked = 1;
527 			ln->ln_state = ND6_LLINFO_PROBE;
528 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
529 			nd6_ns_output(ifp, dst, dst, ln, 0);
530 		} else {
531 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
532 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
533 		}
534 		break;
535 	case ND6_LLINFO_PROBE:
536 		if (ln->la_asked < V_nd6_umaxtries) {
537 			ln->la_asked++;
538 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
539 			nd6_ns_output(ifp, dst, dst, ln, 0);
540 		} else {
541 			(void)nd6_free(ln, 0);
542 			ln = NULL;
543 		}
544 		break;
545 	}
546 done:
547 	if (ln != NULL)
548 		LLE_FREE(ln);
549 	CURVNET_RESTORE();
550 }
551 
552 
553 /*
554  * ND6 timer routine to expire default route list and prefix list
555  */
556 void
557 nd6_timer(void *arg)
558 {
559 	CURVNET_SET((struct vnet *) arg);
560 	int s;
561 	struct nd_defrouter *dr;
562 	struct nd_prefix *pr;
563 	struct in6_ifaddr *ia6, *nia6;
564 	struct in6_addrlifetime *lt6;
565 
566 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
567 	    nd6_timer, curvnet);
568 
569 	/* expire default router list */
570 	s = splnet();
571 	dr = TAILQ_FIRST(&V_nd_defrouter);
572 	while (dr) {
573 		if (dr->expire && dr->expire < time_second) {
574 			struct nd_defrouter *t;
575 			t = TAILQ_NEXT(dr, dr_entry);
576 			defrtrlist_del(dr);
577 			dr = t;
578 		} else {
579 			dr = TAILQ_NEXT(dr, dr_entry);
580 		}
581 	}
582 
583 	/*
584 	 * expire interface addresses.
585 	 * in the past the loop was inside prefix expiry processing.
586 	 * However, from a stricter speci-confrmance standpoint, we should
587 	 * rather separate address lifetimes and prefix lifetimes.
588 	 *
589 	 * XXXRW: in6_ifaddrhead locking.
590 	 */
591   addrloop:
592 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
593 		/* check address lifetime */
594 		lt6 = &ia6->ia6_lifetime;
595 		if (IFA6_IS_INVALID(ia6)) {
596 			int regen = 0;
597 
598 			/*
599 			 * If the expiring address is temporary, try
600 			 * regenerating a new one.  This would be useful when
601 			 * we suspended a laptop PC, then turned it on after a
602 			 * period that could invalidate all temporary
603 			 * addresses.  Although we may have to restart the
604 			 * loop (see below), it must be after purging the
605 			 * address.  Otherwise, we'd see an infinite loop of
606 			 * regeneration.
607 			 */
608 			if (V_ip6_use_tempaddr &&
609 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
610 				if (regen_tmpaddr(ia6) == 0)
611 					regen = 1;
612 			}
613 
614 			in6_purgeaddr(&ia6->ia_ifa);
615 
616 			if (regen)
617 				goto addrloop; /* XXX: see below */
618 		} else if (IFA6_IS_DEPRECATED(ia6)) {
619 			int oldflags = ia6->ia6_flags;
620 
621 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
622 
623 			/*
624 			 * If a temporary address has just become deprecated,
625 			 * regenerate a new one if possible.
626 			 */
627 			if (V_ip6_use_tempaddr &&
628 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
629 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
630 
631 				if (regen_tmpaddr(ia6) == 0) {
632 					/*
633 					 * A new temporary address is
634 					 * generated.
635 					 * XXX: this means the address chain
636 					 * has changed while we are still in
637 					 * the loop.  Although the change
638 					 * would not cause disaster (because
639 					 * it's not a deletion, but an
640 					 * addition,) we'd rather restart the
641 					 * loop just for safety.  Or does this
642 					 * significantly reduce performance??
643 					 */
644 					goto addrloop;
645 				}
646 			}
647 		} else {
648 			/*
649 			 * A new RA might have made a deprecated address
650 			 * preferred.
651 			 */
652 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
653 		}
654 	}
655 
656 	/* expire prefix list */
657 	pr = V_nd_prefix.lh_first;
658 	while (pr) {
659 		/*
660 		 * check prefix lifetime.
661 		 * since pltime is just for autoconf, pltime processing for
662 		 * prefix is not necessary.
663 		 */
664 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
665 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
666 			struct nd_prefix *t;
667 			t = pr->ndpr_next;
668 
669 			/*
670 			 * address expiration and prefix expiration are
671 			 * separate.  NEVER perform in6_purgeaddr here.
672 			 */
673 
674 			prelist_remove(pr);
675 			pr = t;
676 		} else
677 			pr = pr->ndpr_next;
678 	}
679 	splx(s);
680 	CURVNET_RESTORE();
681 }
682 
683 /*
684  * ia6 - deprecated/invalidated temporary address
685  */
686 static int
687 regen_tmpaddr(struct in6_ifaddr *ia6)
688 {
689 	struct ifaddr *ifa;
690 	struct ifnet *ifp;
691 	struct in6_ifaddr *public_ifa6 = NULL;
692 
693 	ifp = ia6->ia_ifa.ifa_ifp;
694 	IF_ADDR_LOCK(ifp);
695 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
696 		struct in6_ifaddr *it6;
697 
698 		if (ifa->ifa_addr->sa_family != AF_INET6)
699 			continue;
700 
701 		it6 = (struct in6_ifaddr *)ifa;
702 
703 		/* ignore no autoconf addresses. */
704 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
705 			continue;
706 
707 		/* ignore autoconf addresses with different prefixes. */
708 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
709 			continue;
710 
711 		/*
712 		 * Now we are looking at an autoconf address with the same
713 		 * prefix as ours.  If the address is temporary and is still
714 		 * preferred, do not create another one.  It would be rare, but
715 		 * could happen, for example, when we resume a laptop PC after
716 		 * a long period.
717 		 */
718 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
719 		    !IFA6_IS_DEPRECATED(it6)) {
720 			public_ifa6 = NULL;
721 			break;
722 		}
723 
724 		/*
725 		 * This is a public autoconf address that has the same prefix
726 		 * as ours.  If it is preferred, keep it.  We can't break the
727 		 * loop here, because there may be a still-preferred temporary
728 		 * address with the prefix.
729 		 */
730 		if (!IFA6_IS_DEPRECATED(it6))
731 		    public_ifa6 = it6;
732 
733 		if (public_ifa6 != NULL)
734 			ifa_ref(&public_ifa6->ia_ifa);
735 	}
736 	IF_ADDR_UNLOCK(ifp);
737 
738 	if (public_ifa6 != NULL) {
739 		int e;
740 
741 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
742 			ifa_free(&public_ifa6->ia_ifa);
743 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
744 			    " tmp addr,errno=%d\n", e);
745 			return (-1);
746 		}
747 		ifa_free(&public_ifa6->ia_ifa);
748 		return (0);
749 	}
750 
751 	return (-1);
752 }
753 
754 /*
755  * Nuke neighbor cache/prefix/default router management table, right before
756  * ifp goes away.
757  */
758 void
759 nd6_purge(struct ifnet *ifp)
760 {
761 	struct nd_defrouter *dr, *ndr;
762 	struct nd_prefix *pr, *npr;
763 
764 	/*
765 	 * Nuke default router list entries toward ifp.
766 	 * We defer removal of default router list entries that is installed
767 	 * in the routing table, in order to keep additional side effects as
768 	 * small as possible.
769 	 */
770 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
771 		ndr = TAILQ_NEXT(dr, dr_entry);
772 		if (dr->installed)
773 			continue;
774 
775 		if (dr->ifp == ifp)
776 			defrtrlist_del(dr);
777 	}
778 
779 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
780 		ndr = TAILQ_NEXT(dr, dr_entry);
781 		if (!dr->installed)
782 			continue;
783 
784 		if (dr->ifp == ifp)
785 			defrtrlist_del(dr);
786 	}
787 
788 	/* Nuke prefix list entries toward ifp */
789 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
790 		npr = pr->ndpr_next;
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 (!V_ip6_forwarding && 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 = 0;
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 	if (flags & ND6_CREATE)
849 	    llflags |= LLE_CREATE;
850 	if (flags & ND6_EXCLUSIVE)
851 	    llflags |= LLE_EXCLUSIVE;
852 
853 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
854 	if ((ln != NULL) && (flags & LLE_CREATE)) {
855 		ln->ln_state = ND6_LLINFO_NOSTATE;
856 		callout_init(&ln->ln_timer_ch, 0);
857 	}
858 
859 	return (ln);
860 }
861 
862 /*
863  * Test whether a given IPv6 address is a neighbor or not, ignoring
864  * the actual neighbor cache.  The neighbor cache is ignored in order
865  * to not reenter the routing code from within itself.
866  */
867 static int
868 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
869 {
870 	struct nd_prefix *pr;
871 	struct ifaddr *dstaddr;
872 
873 	/*
874 	 * A link-local address is always a neighbor.
875 	 * XXX: a link does not necessarily specify a single interface.
876 	 */
877 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
878 		struct sockaddr_in6 sin6_copy;
879 		u_int32_t zone;
880 
881 		/*
882 		 * We need sin6_copy since sa6_recoverscope() may modify the
883 		 * content (XXX).
884 		 */
885 		sin6_copy = *addr;
886 		if (sa6_recoverscope(&sin6_copy))
887 			return (0); /* XXX: should be impossible */
888 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
889 			return (0);
890 		if (sin6_copy.sin6_scope_id == zone)
891 			return (1);
892 		else
893 			return (0);
894 	}
895 
896 	/*
897 	 * If the address matches one of our addresses,
898 	 * it should be a neighbor.
899 	 * If the address matches one of our on-link prefixes, it should be a
900 	 * neighbor.
901 	 */
902 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
903 		if (pr->ndpr_ifp != ifp)
904 			continue;
905 
906 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
907 			struct rtentry *rt;
908 			rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0);
909 			if (rt == NULL)
910 				continue;
911 			/*
912 			 * This is the case where multiple interfaces
913 			 * have the same prefix, but only one is installed
914 			 * into the routing table and that prefix entry
915 			 * is not the one being examined here. In the case
916 			 * where RADIX_MPATH is enabled, multiple route
917 			 * entries (of the same rt_key value) will be
918 			 * installed because the interface addresses all
919 			 * differ.
920 			 */
921 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
922 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
923 				RTFREE_LOCKED(rt);
924 				continue;
925 			}
926 			RTFREE_LOCKED(rt);
927 		}
928 
929 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
930 		    &addr->sin6_addr, &pr->ndpr_mask))
931 			return (1);
932 	}
933 
934 	/*
935 	 * If the address is assigned on the node of the other side of
936 	 * a p2p interface, the address should be a neighbor.
937 	 */
938 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
939 	if (dstaddr != NULL) {
940 		if (dstaddr->ifa_ifp == ifp) {
941 			ifa_free(dstaddr);
942 			return (1);
943 		}
944 		ifa_free(dstaddr);
945 	}
946 
947 	/*
948 	 * If the default router list is empty, all addresses are regarded
949 	 * as on-link, and thus, as a neighbor.
950 	 * XXX: we restrict the condition to hosts, because routers usually do
951 	 * not have the "default router list".
952 	 */
953 	if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
954 	    V_nd6_defifindex == ifp->if_index) {
955 		return (1);
956 	}
957 
958 	return (0);
959 }
960 
961 
962 /*
963  * Detect if a given IPv6 address identifies a neighbor on a given link.
964  * XXX: should take care of the destination of a p2p link?
965  */
966 int
967 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
968 {
969 	struct llentry *lle;
970 	int rc = 0;
971 
972 	IF_AFDATA_UNLOCK_ASSERT(ifp);
973 	if (nd6_is_new_addr_neighbor(addr, ifp))
974 		return (1);
975 
976 	/*
977 	 * Even if the address matches none of our addresses, it might be
978 	 * in the neighbor cache.
979 	 */
980 	IF_AFDATA_LOCK(ifp);
981 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
982 		LLE_RUNLOCK(lle);
983 		rc = 1;
984 	}
985 	IF_AFDATA_UNLOCK(ifp);
986 	return (rc);
987 }
988 
989 /*
990  * Free an nd6 llinfo entry.
991  * Since the function would cause significant changes in the kernel, DO NOT
992  * make it global, unless you have a strong reason for the change, and are sure
993  * that the change is safe.
994  */
995 static struct llentry *
996 nd6_free(struct llentry *ln, int gc)
997 {
998         struct llentry *next;
999 	struct nd_defrouter *dr;
1000 	struct ifnet *ifp=NULL;
1001 
1002 	/*
1003 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1004 	 * even though it is not harmful, it was not really necessary.
1005 	 */
1006 
1007 	/* cancel timer */
1008 	nd6_llinfo_settimer(ln, -1);
1009 
1010 	if (!V_ip6_forwarding) {
1011 		int s;
1012 		s = splnet();
1013 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1014 
1015 		if (dr != NULL && dr->expire &&
1016 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1017 			/*
1018 			 * If the reason for the deletion is just garbage
1019 			 * collection, and the neighbor is an active default
1020 			 * router, do not delete it.  Instead, reset the GC
1021 			 * timer using the router's lifetime.
1022 			 * Simply deleting the entry would affect default
1023 			 * router selection, which is not necessarily a good
1024 			 * thing, especially when we're using router preference
1025 			 * values.
1026 			 * XXX: the check for ln_state would be redundant,
1027 			 *      but we intentionally keep it just in case.
1028 			 */
1029 			if (dr->expire > time_second)
1030 				nd6_llinfo_settimer(ln,
1031 				    (dr->expire - time_second) * hz);
1032 			else
1033 				nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
1034 			splx(s);
1035 			LLE_WLOCK(ln);
1036 			LLE_REMREF(ln);
1037 			LLE_WUNLOCK(ln);
1038 			return (LIST_NEXT(ln, lle_next));
1039 		}
1040 
1041 		if (ln->ln_router || dr) {
1042 			/*
1043 			 * rt6_flush must be called whether or not the neighbor
1044 			 * is in the Default Router List.
1045 			 * See a corresponding comment in nd6_na_input().
1046 			 */
1047 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1048 		}
1049 
1050 		if (dr) {
1051 			/*
1052 			 * Unreachablity of a router might affect the default
1053 			 * router selection and on-link detection of advertised
1054 			 * prefixes.
1055 			 */
1056 
1057 			/*
1058 			 * Temporarily fake the state to choose a new default
1059 			 * router and to perform on-link determination of
1060 			 * prefixes correctly.
1061 			 * Below the state will be set correctly,
1062 			 * or the entry itself will be deleted.
1063 			 */
1064 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1065 
1066 			/*
1067 			 * Since defrouter_select() does not affect the
1068 			 * on-link determination and MIP6 needs the check
1069 			 * before the default router selection, we perform
1070 			 * the check now.
1071 			 */
1072 			pfxlist_onlink_check();
1073 
1074 			/*
1075 			 * refresh default router list
1076 			 */
1077 			defrouter_select();
1078 		}
1079 		splx(s);
1080 	}
1081 
1082 	/*
1083 	 * Before deleting the entry, remember the next entry as the
1084 	 * return value.  We need this because pfxlist_onlink_check() above
1085 	 * might have freed other entries (particularly the old next entry) as
1086 	 * a side effect (XXX).
1087 	 */
1088 	next = LIST_NEXT(ln, lle_next);
1089 
1090 	ifp = ln->lle_tbl->llt_ifp;
1091 	IF_AFDATA_LOCK(ifp);
1092 	LLE_WLOCK(ln);
1093 	LLE_REMREF(ln);
1094 	llentry_free(ln);
1095 	IF_AFDATA_UNLOCK(ifp);
1096 
1097 	return (next);
1098 }
1099 
1100 /*
1101  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1102  *
1103  * XXX cost-effective methods?
1104  */
1105 void
1106 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1107 {
1108 	struct llentry *ln;
1109 	struct ifnet *ifp;
1110 
1111 	if ((dst6 == NULL) || (rt == NULL))
1112 		return;
1113 
1114 	ifp = rt->rt_ifp;
1115 	IF_AFDATA_LOCK(ifp);
1116 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1117 	IF_AFDATA_UNLOCK(ifp);
1118 	if (ln == NULL)
1119 		return;
1120 
1121 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1122 		goto done;
1123 
1124 	/*
1125 	 * if we get upper-layer reachability confirmation many times,
1126 	 * it is possible we have false information.
1127 	 */
1128 	if (!force) {
1129 		ln->ln_byhint++;
1130 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1131 			goto done;
1132 		}
1133 	}
1134 
1135  	ln->ln_state = ND6_LLINFO_REACHABLE;
1136 	if (!ND6_LLINFO_PERMANENT(ln)) {
1137 		nd6_llinfo_settimer_locked(ln,
1138 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1139 	}
1140 done:
1141 	LLE_WUNLOCK(ln);
1142 }
1143 
1144 
1145 int
1146 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1147 {
1148 	struct in6_drlist *drl = (struct in6_drlist *)data;
1149 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1150 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1151 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1152 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1153 	struct nd_defrouter *dr;
1154 	struct nd_prefix *pr;
1155 	int i = 0, error = 0;
1156 	int s;
1157 
1158 	switch (cmd) {
1159 	case SIOCGDRLST_IN6:
1160 		/*
1161 		 * obsolete API, use sysctl under net.inet6.icmp6
1162 		 */
1163 		bzero(drl, sizeof(*drl));
1164 		s = splnet();
1165 		dr = TAILQ_FIRST(&V_nd_defrouter);
1166 		while (dr && i < DRLSTSIZ) {
1167 			drl->defrouter[i].rtaddr = dr->rtaddr;
1168 			in6_clearscope(&drl->defrouter[i].rtaddr);
1169 
1170 			drl->defrouter[i].flags = dr->flags;
1171 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1172 			drl->defrouter[i].expire = dr->expire;
1173 			drl->defrouter[i].if_index = dr->ifp->if_index;
1174 			i++;
1175 			dr = TAILQ_NEXT(dr, dr_entry);
1176 		}
1177 		splx(s);
1178 		break;
1179 	case SIOCGPRLST_IN6:
1180 		/*
1181 		 * obsolete API, use sysctl under net.inet6.icmp6
1182 		 *
1183 		 * XXX the structure in6_prlist was changed in backward-
1184 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1185 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1186 		 */
1187 		/*
1188 		 * XXX meaning of fields, especialy "raflags", is very
1189 		 * differnet between RA prefix list and RR/static prefix list.
1190 		 * how about separating ioctls into two?
1191 		 */
1192 		bzero(oprl, sizeof(*oprl));
1193 		s = splnet();
1194 		pr = V_nd_prefix.lh_first;
1195 		while (pr && i < PRLSTSIZ) {
1196 			struct nd_pfxrouter *pfr;
1197 			int j;
1198 
1199 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1200 			oprl->prefix[i].raflags = pr->ndpr_raf;
1201 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1202 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1203 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1204 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1205 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1206 				oprl->prefix[i].expire = 0;
1207 			else {
1208 				time_t maxexpire;
1209 
1210 				/* XXX: we assume time_t is signed. */
1211 				maxexpire = (-1) &
1212 				    ~((time_t)1 <<
1213 				    ((sizeof(maxexpire) * 8) - 1));
1214 				if (pr->ndpr_vltime <
1215 				    maxexpire - pr->ndpr_lastupdate) {
1216 					oprl->prefix[i].expire =
1217 					    pr->ndpr_lastupdate +
1218 					    pr->ndpr_vltime;
1219 				} else
1220 					oprl->prefix[i].expire = maxexpire;
1221 			}
1222 
1223 			pfr = pr->ndpr_advrtrs.lh_first;
1224 			j = 0;
1225 			while (pfr) {
1226 				if (j < DRLSTSIZ) {
1227 #define RTRADDR oprl->prefix[i].advrtr[j]
1228 					RTRADDR = pfr->router->rtaddr;
1229 					in6_clearscope(&RTRADDR);
1230 #undef RTRADDR
1231 				}
1232 				j++;
1233 				pfr = pfr->pfr_next;
1234 			}
1235 			oprl->prefix[i].advrtrs = j;
1236 			oprl->prefix[i].origin = PR_ORIG_RA;
1237 
1238 			i++;
1239 			pr = pr->ndpr_next;
1240 		}
1241 		splx(s);
1242 
1243 		break;
1244 	case OSIOCGIFINFO_IN6:
1245 #define ND	ndi->ndi
1246 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1247 		bzero(&ND, sizeof(ND));
1248 		ND.linkmtu = IN6_LINKMTU(ifp);
1249 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1250 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1251 		ND.reachable = ND_IFINFO(ifp)->reachable;
1252 		ND.retrans = ND_IFINFO(ifp)->retrans;
1253 		ND.flags = ND_IFINFO(ifp)->flags;
1254 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1255 		ND.chlim = ND_IFINFO(ifp)->chlim;
1256 		break;
1257 	case SIOCGIFINFO_IN6:
1258 		ND = *ND_IFINFO(ifp);
1259 		break;
1260 	case SIOCSIFINFO_IN6:
1261 		/*
1262 		 * used to change host variables from userland.
1263 		 * intented for a use on router to reflect RA configurations.
1264 		 */
1265 		/* 0 means 'unspecified' */
1266 		if (ND.linkmtu != 0) {
1267 			if (ND.linkmtu < IPV6_MMTU ||
1268 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1269 				error = EINVAL;
1270 				break;
1271 			}
1272 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1273 		}
1274 
1275 		if (ND.basereachable != 0) {
1276 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1277 
1278 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1279 			if (ND.basereachable != obasereachable)
1280 				ND_IFINFO(ifp)->reachable =
1281 				    ND_COMPUTE_RTIME(ND.basereachable);
1282 		}
1283 		if (ND.retrans != 0)
1284 			ND_IFINFO(ifp)->retrans = ND.retrans;
1285 		if (ND.chlim != 0)
1286 			ND_IFINFO(ifp)->chlim = ND.chlim;
1287 		/* FALLTHROUGH */
1288 	case SIOCSIFINFO_FLAGS:
1289 	{
1290 		struct ifaddr *ifa;
1291 		struct in6_ifaddr *ia;
1292 
1293 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1294 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1295 			/* ifdisabled 1->0 transision */
1296 
1297 			/*
1298 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1299 			 * has an link-local address with IN6_IFF_DUPLICATED,
1300 			 * do not clear ND6_IFF_IFDISABLED.
1301 			 * See RFC 4862, Section 5.4.5.
1302 			 */
1303 			int duplicated_linklocal = 0;
1304 
1305 			IF_ADDR_LOCK(ifp);
1306 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1307 				if (ifa->ifa_addr->sa_family != AF_INET6)
1308 					continue;
1309 				ia = (struct in6_ifaddr *)ifa;
1310 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1311 				    IN6_IS_ADDR_LINKLOCAL(&ia->ia_addr.sin6_addr)) {
1312 					duplicated_linklocal = 1;
1313 					break;
1314 				}
1315 			}
1316 			IF_ADDR_UNLOCK(ifp);
1317 
1318 			if (duplicated_linklocal) {
1319 				ND.flags |= ND6_IFF_IFDISABLED;
1320 				log(LOG_ERR, "Cannot enable an interface"
1321 				    " with a link-local address marked"
1322 				    " duplicate.\n");
1323 			} else {
1324 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1325 				in6_if_up(ifp);
1326 			}
1327 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1328 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1329 			/* ifdisabled 0->1 transision */
1330 			/* Mark all IPv6 address as tentative. */
1331 
1332 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1333 			IF_ADDR_LOCK(ifp);
1334 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1335 				if (ifa->ifa_addr->sa_family != AF_INET6)
1336 					continue;
1337 				ia = (struct in6_ifaddr *)ifa;
1338 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1339 			}
1340 			IF_ADDR_UNLOCK(ifp);
1341 		}
1342 
1343 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) &&
1344 		    (ND.flags & ND6_IFF_AUTO_LINKLOCAL)) {
1345 			/* auto_linklocal 0->1 transision */
1346 
1347 			/* If no link-local address on ifp, configure */
1348 			ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1349 			in6_ifattach(ifp, NULL);
1350 		}
1351 	}
1352 		ND_IFINFO(ifp)->flags = ND.flags;
1353 		break;
1354 #undef ND
1355 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1356 		/* sync kernel routing table with the default router list */
1357 		defrouter_reset();
1358 		defrouter_select();
1359 		break;
1360 	case SIOCSPFXFLUSH_IN6:
1361 	{
1362 		/* flush all the prefix advertised by routers */
1363 		struct nd_prefix *pr, *next;
1364 
1365 		s = splnet();
1366 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1367 			struct in6_ifaddr *ia, *ia_next;
1368 
1369 			next = pr->ndpr_next;
1370 
1371 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1372 				continue; /* XXX */
1373 
1374 			/* do we really have to remove addresses as well? */
1375 			/* XXXRW: in6_ifaddrhead locking. */
1376 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1377 			    ia_next) {
1378 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1379 					continue;
1380 
1381 				if (ia->ia6_ndpr == pr)
1382 					in6_purgeaddr(&ia->ia_ifa);
1383 			}
1384 			prelist_remove(pr);
1385 		}
1386 		splx(s);
1387 		break;
1388 	}
1389 	case SIOCSRTRFLUSH_IN6:
1390 	{
1391 		/* flush all the default routers */
1392 		struct nd_defrouter *dr, *next;
1393 
1394 		s = splnet();
1395 		defrouter_reset();
1396 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1397 			next = TAILQ_NEXT(dr, dr_entry);
1398 			defrtrlist_del(dr);
1399 		}
1400 		defrouter_select();
1401 		splx(s);
1402 		break;
1403 	}
1404 	case SIOCGNBRINFO_IN6:
1405 	{
1406 		struct llentry *ln;
1407 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1408 
1409 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1410 			return (error);
1411 
1412 		IF_AFDATA_LOCK(ifp);
1413 		ln = nd6_lookup(&nb_addr, 0, ifp);
1414 		IF_AFDATA_UNLOCK(ifp);
1415 
1416 		if (ln == NULL) {
1417 			error = EINVAL;
1418 			break;
1419 		}
1420 		nbi->state = ln->ln_state;
1421 		nbi->asked = ln->la_asked;
1422 		nbi->isrouter = ln->ln_router;
1423 		nbi->expire = ln->la_expire;
1424 		LLE_RUNLOCK(ln);
1425 		break;
1426 	}
1427 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1428 		ndif->ifindex = V_nd6_defifindex;
1429 		break;
1430 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1431 		return (nd6_setdefaultiface(ndif->ifindex));
1432 	}
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Create neighbor cache entry and cache link-layer address,
1438  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1439  *
1440  * type - ICMP6 type
1441  * code - type dependent information
1442  *
1443  * XXXXX
1444  *  The caller of this function already acquired the ndp
1445  *  cache table lock because the cache entry is returned.
1446  */
1447 struct llentry *
1448 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1449     int lladdrlen, int type, int code)
1450 {
1451 	struct llentry *ln = NULL;
1452 	int is_newentry;
1453 	int do_update;
1454 	int olladdr;
1455 	int llchange;
1456 	int flags = 0;
1457 	int newstate = 0;
1458 	uint16_t router = 0;
1459 	struct sockaddr_in6 sin6;
1460 	struct mbuf *chain = NULL;
1461 	int static_route = 0;
1462 
1463 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1464 
1465 	if (ifp == NULL)
1466 		panic("ifp == NULL in nd6_cache_lladdr");
1467 	if (from == NULL)
1468 		panic("from == NULL in nd6_cache_lladdr");
1469 
1470 	/* nothing must be updated for unspecified address */
1471 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1472 		return NULL;
1473 
1474 	/*
1475 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1476 	 * the caller.
1477 	 *
1478 	 * XXX If the link does not have link-layer adderss, what should
1479 	 * we do? (ifp->if_addrlen == 0)
1480 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1481 	 * description on it in NS section (RFC 2461 7.2.3).
1482 	 */
1483 	flags |= lladdr ? ND6_EXCLUSIVE : 0;
1484 	IF_AFDATA_LOCK(ifp);
1485 	ln = nd6_lookup(from, flags, ifp);
1486 
1487 	if (ln == NULL) {
1488 		flags |= LLE_EXCLUSIVE;
1489 		ln = nd6_lookup(from, flags |ND6_CREATE, ifp);
1490 		IF_AFDATA_UNLOCK(ifp);
1491 		is_newentry = 1;
1492 	} else {
1493 		IF_AFDATA_UNLOCK(ifp);
1494 		/* do nothing if static ndp is set */
1495 		if (ln->la_flags & LLE_STATIC) {
1496 			static_route = 1;
1497 			goto done;
1498 		}
1499 		is_newentry = 0;
1500 	}
1501 	if (ln == NULL)
1502 		return (NULL);
1503 
1504 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1505 	if (olladdr && lladdr) {
1506 		llchange = bcmp(lladdr, &ln->ll_addr,
1507 		    ifp->if_addrlen);
1508 	} else
1509 		llchange = 0;
1510 
1511 	/*
1512 	 * newentry olladdr  lladdr  llchange	(*=record)
1513 	 *	0	n	n	--	(1)
1514 	 *	0	y	n	--	(2)
1515 	 *	0	n	y	--	(3) * STALE
1516 	 *	0	y	y	n	(4) *
1517 	 *	0	y	y	y	(5) * STALE
1518 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1519 	 *	1	--	y	--	(7) * STALE
1520 	 */
1521 
1522 	if (lladdr) {		/* (3-5) and (7) */
1523 		/*
1524 		 * Record source link-layer address
1525 		 * XXX is it dependent to ifp->if_type?
1526 		 */
1527 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1528 		ln->la_flags |= LLE_VALID;
1529 	}
1530 
1531 	if (!is_newentry) {
1532 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1533 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1534 			do_update = 1;
1535 			newstate = ND6_LLINFO_STALE;
1536 		} else					/* (1-2,4) */
1537 			do_update = 0;
1538 	} else {
1539 		do_update = 1;
1540 		if (lladdr == NULL)			/* (6) */
1541 			newstate = ND6_LLINFO_NOSTATE;
1542 		else					/* (7) */
1543 			newstate = ND6_LLINFO_STALE;
1544 	}
1545 
1546 	if (do_update) {
1547 		/*
1548 		 * Update the state of the neighbor cache.
1549 		 */
1550 		ln->ln_state = newstate;
1551 
1552 		if (ln->ln_state == ND6_LLINFO_STALE) {
1553 			/*
1554 			 * XXX: since nd6_output() below will cause
1555 			 * state tansition to DELAY and reset the timer,
1556 			 * we must set the timer now, although it is actually
1557 			 * meaningless.
1558 			 */
1559 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1560 
1561 			if (ln->la_hold) {
1562 				struct mbuf *m_hold, *m_hold_next;
1563 
1564 				/*
1565 				 * reset the la_hold in advance, to explicitly
1566 				 * prevent a la_hold lookup in nd6_output()
1567 				 * (wouldn't happen, though...)
1568 				 */
1569 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1570 				    m_hold; m_hold = m_hold_next) {
1571 					m_hold_next = m_hold->m_nextpkt;
1572 					m_hold->m_nextpkt = NULL;
1573 
1574 					/*
1575 					 * we assume ifp is not a p2p here, so
1576 					 * just set the 2nd argument as the
1577 					 * 1st one.
1578 					 */
1579 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1580 				}
1581 				/*
1582 				 * If we have mbufs in the chain we need to do
1583 				 * deferred transmit. Copy the address from the
1584 				 * llentry before dropping the lock down below.
1585 				 */
1586 				if (chain != NULL)
1587 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1588 			}
1589 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1590 			/* probe right away */
1591 			nd6_llinfo_settimer_locked((void *)ln, 0);
1592 		}
1593 	}
1594 
1595 	/*
1596 	 * ICMP6 type dependent behavior.
1597 	 *
1598 	 * NS: clear IsRouter if new entry
1599 	 * RS: clear IsRouter
1600 	 * RA: set IsRouter if there's lladdr
1601 	 * redir: clear IsRouter if new entry
1602 	 *
1603 	 * RA case, (1):
1604 	 * The spec says that we must set IsRouter in the following cases:
1605 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1606 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1607 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1608 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1609 	 * neighbor cache, this is similar to (6).
1610 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1611 	 *
1612 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1613 	 *							D R
1614 	 *	0	n	n	--	(1)	c   ?     s
1615 	 *	0	y	n	--	(2)	c   s     s
1616 	 *	0	n	y	--	(3)	c   s     s
1617 	 *	0	y	y	n	(4)	c   s     s
1618 	 *	0	y	y	y	(5)	c   s     s
1619 	 *	1	--	n	--	(6) c	c	c s
1620 	 *	1	--	y	--	(7) c	c   s	c s
1621 	 *
1622 	 *					(c=clear s=set)
1623 	 */
1624 	switch (type & 0xff) {
1625 	case ND_NEIGHBOR_SOLICIT:
1626 		/*
1627 		 * New entry must have is_router flag cleared.
1628 		 */
1629 		if (is_newentry)	/* (6-7) */
1630 			ln->ln_router = 0;
1631 		break;
1632 	case ND_REDIRECT:
1633 		/*
1634 		 * If the icmp is a redirect to a better router, always set the
1635 		 * is_router flag.  Otherwise, if the entry is newly created,
1636 		 * clear the flag.  [RFC 2461, sec 8.3]
1637 		 */
1638 		if (code == ND_REDIRECT_ROUTER)
1639 			ln->ln_router = 1;
1640 		else if (is_newentry) /* (6-7) */
1641 			ln->ln_router = 0;
1642 		break;
1643 	case ND_ROUTER_SOLICIT:
1644 		/*
1645 		 * is_router flag must always be cleared.
1646 		 */
1647 		ln->ln_router = 0;
1648 		break;
1649 	case ND_ROUTER_ADVERT:
1650 		/*
1651 		 * Mark an entry with lladdr as a router.
1652 		 */
1653 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1654 		    (is_newentry && lladdr)) {			/* (7) */
1655 			ln->ln_router = 1;
1656 		}
1657 		break;
1658 	}
1659 
1660 	if (ln != NULL) {
1661 		static_route = (ln->la_flags & LLE_STATIC);
1662 		router = ln->ln_router;
1663 
1664 		if (flags & ND6_EXCLUSIVE)
1665 			LLE_WUNLOCK(ln);
1666 		else
1667 			LLE_RUNLOCK(ln);
1668 		if (static_route)
1669 			ln = NULL;
1670 	}
1671 	if (chain)
1672 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1673 
1674 	/*
1675 	 * When the link-layer address of a router changes, select the
1676 	 * best router again.  In particular, when the neighbor entry is newly
1677 	 * created, it might affect the selection policy.
1678 	 * Question: can we restrict the first condition to the "is_newentry"
1679 	 * case?
1680 	 * XXX: when we hear an RA from a new router with the link-layer
1681 	 * address option, defrouter_select() is called twice, since
1682 	 * defrtrlist_update called the function as well.  However, I believe
1683 	 * we can compromise the overhead, since it only happens the first
1684 	 * time.
1685 	 * XXX: although defrouter_select() should not have a bad effect
1686 	 * for those are not autoconfigured hosts, we explicitly avoid such
1687 	 * cases for safety.
1688 	 */
1689 	if (do_update && router && !V_ip6_forwarding &&
1690 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1691 		/*
1692 		 * guaranteed recursion
1693 		 */
1694 		defrouter_select();
1695 	}
1696 
1697 	return (ln);
1698 done:
1699 	if (ln != NULL) {
1700 		if (flags & ND6_EXCLUSIVE)
1701 			LLE_WUNLOCK(ln);
1702 		else
1703 			LLE_RUNLOCK(ln);
1704 		if (static_route)
1705 			ln = NULL;
1706 	}
1707 	return (ln);
1708 }
1709 
1710 static void
1711 nd6_slowtimo(void *arg)
1712 {
1713 	CURVNET_SET((struct vnet *) arg);
1714 	struct nd_ifinfo *nd6if;
1715 	struct ifnet *ifp;
1716 
1717 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1718 	    nd6_slowtimo, curvnet);
1719 	IFNET_RLOCK_NOSLEEP();
1720 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1721 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1722 		nd6if = ND_IFINFO(ifp);
1723 		if (nd6if->basereachable && /* already initialized */
1724 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1725 			/*
1726 			 * Since reachable time rarely changes by router
1727 			 * advertisements, we SHOULD insure that a new random
1728 			 * value gets recomputed at least once every few hours.
1729 			 * (RFC 2461, 6.3.4)
1730 			 */
1731 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1732 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1733 		}
1734 	}
1735 	IFNET_RUNLOCK_NOSLEEP();
1736 	CURVNET_RESTORE();
1737 }
1738 
1739 int
1740 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1741     struct sockaddr_in6 *dst, struct rtentry *rt0)
1742 {
1743 
1744 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1745 }
1746 
1747 
1748 /*
1749  * Note that I'm not enforcing any global serialization
1750  * lle state or asked changes here as the logic is too
1751  * complicated to avoid having to always acquire an exclusive
1752  * lock
1753  * KMM
1754  *
1755  */
1756 #define senderr(e) { error = (e); goto bad;}
1757 
1758 int
1759 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1760     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1761 	struct mbuf **chain)
1762 {
1763 	struct mbuf *m = m0;
1764 	struct m_tag *mtag;
1765 	struct llentry *ln = lle;
1766 	struct ip6_hdr *ip6;
1767 	int error = 0;
1768 	int flags = 0;
1769 	int ip6len;
1770 
1771 #ifdef INVARIANTS
1772 	if (lle != NULL) {
1773 
1774 		LLE_WLOCK_ASSERT(lle);
1775 
1776 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1777 	}
1778 #endif
1779 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1780 		goto sendpkt;
1781 
1782 	if (nd6_need_cache(ifp) == 0)
1783 		goto sendpkt;
1784 
1785 	/*
1786 	 * next hop determination.  This routine is derived from ether_output.
1787 	 */
1788 
1789 	/*
1790 	 * Address resolution or Neighbor Unreachability Detection
1791 	 * for the next hop.
1792 	 * At this point, the destination of the packet must be a unicast
1793 	 * or an anycast address(i.e. not a multicast).
1794 	 */
1795 
1796 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1797 	if (ln == NULL) {
1798 	retry:
1799 		IF_AFDATA_LOCK(ifp);
1800 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1801 		IF_AFDATA_UNLOCK(ifp);
1802 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1803 			/*
1804 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1805 			 * the condition below is not very efficient.  But we believe
1806 			 * it is tolerable, because this should be a rare case.
1807 			 */
1808 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1809 			IF_AFDATA_LOCK(ifp);
1810 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1811 			IF_AFDATA_UNLOCK(ifp);
1812 		}
1813 	}
1814 	if (ln == NULL) {
1815 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1816 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1817 			char ip6buf[INET6_ADDRSTRLEN];
1818 			log(LOG_DEBUG,
1819 			    "nd6_output: can't allocate llinfo for %s "
1820 			    "(ln=%p)\n",
1821 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1822 			senderr(EIO);	/* XXX: good error? */
1823 		}
1824 		goto sendpkt;	/* send anyway */
1825 	}
1826 
1827 	/* We don't have to do link-layer address resolution on a p2p link. */
1828 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1829 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1830 		if ((flags & LLE_EXCLUSIVE) == 0) {
1831 			flags |= LLE_EXCLUSIVE;
1832 			goto retry;
1833 		}
1834 		ln->ln_state = ND6_LLINFO_STALE;
1835 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1836 	}
1837 
1838 	/*
1839 	 * The first time we send a packet to a neighbor whose entry is
1840 	 * STALE, we have to change the state to DELAY and a sets a timer to
1841 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1842 	 * neighbor unreachability detection on expiration.
1843 	 * (RFC 2461 7.3.3)
1844 	 */
1845 	if (ln->ln_state == ND6_LLINFO_STALE) {
1846 		if ((flags & LLE_EXCLUSIVE) == 0) {
1847 			flags |= LLE_EXCLUSIVE;
1848 			LLE_RUNLOCK(ln);
1849 			goto retry;
1850 		}
1851 		ln->la_asked = 0;
1852 		ln->ln_state = ND6_LLINFO_DELAY;
1853 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1854 	}
1855 
1856 	/*
1857 	 * If the neighbor cache entry has a state other than INCOMPLETE
1858 	 * (i.e. its link-layer address is already resolved), just
1859 	 * send the packet.
1860 	 */
1861 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1862 		goto sendpkt;
1863 
1864 	/*
1865 	 * There is a neighbor cache entry, but no ethernet address
1866 	 * response yet.  Append this latest packet to the end of the
1867 	 * packet queue in the mbuf, unless the number of the packet
1868 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1869 	 * the oldest packet in the queue will be removed.
1870 	 */
1871 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1872 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1873 
1874 	if ((flags & LLE_EXCLUSIVE) == 0) {
1875 		flags |= LLE_EXCLUSIVE;
1876 		LLE_RUNLOCK(ln);
1877 		goto retry;
1878 	}
1879 	if (ln->la_hold) {
1880 		struct mbuf *m_hold;
1881 		int i;
1882 
1883 		i = 0;
1884 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1885 			i++;
1886 			if (m_hold->m_nextpkt == NULL) {
1887 				m_hold->m_nextpkt = m;
1888 				break;
1889 			}
1890 		}
1891 		while (i >= V_nd6_maxqueuelen) {
1892 			m_hold = ln->la_hold;
1893 			ln->la_hold = ln->la_hold->m_nextpkt;
1894 			m_freem(m_hold);
1895 			i--;
1896 		}
1897 	} else {
1898 		ln->la_hold = m;
1899 	}
1900 	/*
1901 	 * We did the lookup (no lle arg) so we
1902 	 * need to do the unlock here
1903 	 */
1904 	if (lle == NULL) {
1905 		if (flags & LLE_EXCLUSIVE)
1906 			LLE_WUNLOCK(ln);
1907 		else
1908 			LLE_RUNLOCK(ln);
1909 	}
1910 
1911 	/*
1912 	 * If there has been no NS for the neighbor after entering the
1913 	 * INCOMPLETE state, send the first solicitation.
1914 	 */
1915 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1916 		ln->la_asked++;
1917 
1918 		nd6_llinfo_settimer(ln,
1919 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1920 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1921 	}
1922 	return (0);
1923 
1924   sendpkt:
1925 	/* discard the packet if IPv6 operation is disabled on the interface */
1926 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1927 		error = ENETDOWN; /* better error? */
1928 		goto bad;
1929 	}
1930 	/*
1931 	 * ln is valid and the caller did not pass in
1932 	 * an llentry
1933 	 */
1934 	if ((ln != NULL) && (lle == NULL)) {
1935 		if (flags & LLE_EXCLUSIVE)
1936 			LLE_WUNLOCK(ln);
1937 		else
1938 			LLE_RUNLOCK(ln);
1939 	}
1940 
1941 #ifdef MAC
1942 	mac_netinet6_nd6_send(ifp, m);
1943 #endif
1944 
1945 	/*
1946 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
1947 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
1948 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
1949 	 * to be diverted to user space.  When re-injected into the kernel,
1950 	 * send_output() will directly dispatch them to the outgoing interface.
1951 	 */
1952 	if (send_sendso_input_hook != NULL) {
1953 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
1954 		if (mtag != NULL) {
1955 			ip6 = mtod(m, struct ip6_hdr *);
1956 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
1957 			/* Use the SEND socket */
1958 			error = send_sendso_input_hook(m, ifp, SND_OUT,
1959 			    ip6len);
1960 			/* -1 == no app on SEND socket */
1961 			if (error == 0 || error != -1)
1962 			    return (error);
1963 		}
1964 	}
1965 
1966 	/*
1967 	 * We were passed in a pointer to an lle with the lock held
1968 	 * this means that we can't call if_output as we will
1969 	 * recurse on the lle lock - so what we do is we create
1970 	 * a list of mbufs to send and transmit them in the caller
1971 	 * after the lock is dropped
1972 	 */
1973 	if (lle != NULL) {
1974 		if (*chain == NULL)
1975 			*chain = m;
1976 		else {
1977 			struct mbuf *m = *chain;
1978 
1979 			/*
1980 			 * append mbuf to end of deferred chain
1981 			 */
1982 			while (m->m_nextpkt != NULL)
1983 				m = m->m_nextpkt;
1984 			m->m_nextpkt = m;
1985 		}
1986 		return (error);
1987 	}
1988 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1989 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1990 		    NULL));
1991 	}
1992 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1993 	return (error);
1994 
1995   bad:
1996 	/*
1997 	 * ln is valid and the caller did not pass in
1998 	 * an llentry
1999 	 */
2000 	if ((ln != NULL) && (lle == NULL)) {
2001 		if (flags & LLE_EXCLUSIVE)
2002 			LLE_WUNLOCK(ln);
2003 		else
2004 			LLE_RUNLOCK(ln);
2005 	}
2006 	if (m)
2007 		m_freem(m);
2008 	return (error);
2009 }
2010 #undef senderr
2011 
2012 
2013 int
2014 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2015     struct sockaddr_in6 *dst, struct route *ro)
2016 {
2017 	struct mbuf *m, *m_head;
2018 	struct ifnet *outifp;
2019 	int error = 0;
2020 
2021 	m_head = chain;
2022 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2023 		outifp = origifp;
2024 	else
2025 		outifp = ifp;
2026 
2027 	while (m_head) {
2028 		m = m_head;
2029 		m_head = m_head->m_nextpkt;
2030 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2031 	}
2032 
2033 	/*
2034 	 * XXX
2035 	 * note that intermediate errors are blindly ignored - but this is
2036 	 * the same convention as used with nd6_output when called by
2037 	 * nd6_cache_lladdr
2038 	 */
2039 	return (error);
2040 }
2041 
2042 
2043 int
2044 nd6_need_cache(struct ifnet *ifp)
2045 {
2046 	/*
2047 	 * XXX: we currently do not make neighbor cache on any interface
2048 	 * other than ARCnet, Ethernet, FDDI and GIF.
2049 	 *
2050 	 * RFC2893 says:
2051 	 * - unidirectional tunnels needs no ND
2052 	 */
2053 	switch (ifp->if_type) {
2054 	case IFT_ARCNET:
2055 	case IFT_ETHER:
2056 	case IFT_FDDI:
2057 	case IFT_IEEE1394:
2058 #ifdef IFT_L2VLAN
2059 	case IFT_L2VLAN:
2060 #endif
2061 #ifdef IFT_IEEE80211
2062 	case IFT_IEEE80211:
2063 #endif
2064 #ifdef IFT_CARP
2065 	case IFT_CARP:
2066 #endif
2067 	case IFT_GIF:		/* XXX need more cases? */
2068 	case IFT_PPP:
2069 	case IFT_TUNNEL:
2070 	case IFT_BRIDGE:
2071 	case IFT_PROPVIRTUAL:
2072 		return (1);
2073 	default:
2074 		return (0);
2075 	}
2076 }
2077 
2078 /*
2079  * the callers of this function need to be re-worked to drop
2080  * the lle lock, drop here for now
2081  */
2082 int
2083 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2084     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2085 {
2086 	struct llentry *ln;
2087 
2088 	*lle = NULL;
2089 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2090 	if (m->m_flags & M_MCAST) {
2091 		int i;
2092 
2093 		switch (ifp->if_type) {
2094 		case IFT_ETHER:
2095 		case IFT_FDDI:
2096 #ifdef IFT_L2VLAN
2097 		case IFT_L2VLAN:
2098 #endif
2099 #ifdef IFT_IEEE80211
2100 		case IFT_IEEE80211:
2101 #endif
2102 		case IFT_BRIDGE:
2103 		case IFT_ISO88025:
2104 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2105 						 desten);
2106 			return (0);
2107 		case IFT_IEEE1394:
2108 			/*
2109 			 * netbsd can use if_broadcastaddr, but we don't do so
2110 			 * to reduce # of ifdef.
2111 			 */
2112 			for (i = 0; i < ifp->if_addrlen; i++)
2113 				desten[i] = ~0;
2114 			return (0);
2115 		case IFT_ARCNET:
2116 			*desten = 0;
2117 			return (0);
2118 		default:
2119 			m_freem(m);
2120 			return (EAFNOSUPPORT);
2121 		}
2122 	}
2123 
2124 
2125 	/*
2126 	 * the entry should have been created in nd6_store_lladdr
2127 	 */
2128 	IF_AFDATA_LOCK(ifp);
2129 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2130 	IF_AFDATA_UNLOCK(ifp);
2131 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2132 		if (ln != NULL)
2133 			LLE_RUNLOCK(ln);
2134 		/* this could happen, if we could not allocate memory */
2135 		m_freem(m);
2136 		return (1);
2137 	}
2138 
2139 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2140 	*lle = ln;
2141 	LLE_RUNLOCK(ln);
2142 	/*
2143 	 * A *small* use after free race exists here
2144 	 */
2145 	return (0);
2146 }
2147 
2148 static void
2149 clear_llinfo_pqueue(struct llentry *ln)
2150 {
2151 	struct mbuf *m_hold, *m_hold_next;
2152 
2153 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2154 		m_hold_next = m_hold->m_nextpkt;
2155 		m_hold->m_nextpkt = NULL;
2156 		m_freem(m_hold);
2157 	}
2158 
2159 	ln->la_hold = NULL;
2160 	return;
2161 }
2162 
2163 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2164 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2165 #ifdef SYSCTL_DECL
2166 SYSCTL_DECL(_net_inet6_icmp6);
2167 #endif
2168 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2169 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2170 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2171 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2172 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2173 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2174 
2175 static int
2176 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2177 {
2178 	int error;
2179 	char buf[1024] __aligned(4);
2180 	struct in6_defrouter *d, *de;
2181 	struct nd_defrouter *dr;
2182 
2183 	if (req->newptr)
2184 		return EPERM;
2185 	error = 0;
2186 
2187 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2188 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2189 		d = (struct in6_defrouter *)buf;
2190 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2191 
2192 		if (d + 1 <= de) {
2193 			bzero(d, sizeof(*d));
2194 			d->rtaddr.sin6_family = AF_INET6;
2195 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2196 			d->rtaddr.sin6_addr = dr->rtaddr;
2197 			error = sa6_recoverscope(&d->rtaddr);
2198 			if (error != 0)
2199 				return (error);
2200 			d->flags = dr->flags;
2201 			d->rtlifetime = dr->rtlifetime;
2202 			d->expire = dr->expire;
2203 			d->if_index = dr->ifp->if_index;
2204 		} else
2205 			panic("buffer too short");
2206 
2207 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2208 		if (error)
2209 			break;
2210 	}
2211 
2212 	return (error);
2213 }
2214 
2215 static int
2216 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2217 {
2218 	int error;
2219 	char buf[1024] __aligned(4);
2220 	struct in6_prefix *p, *pe;
2221 	struct nd_prefix *pr;
2222 	char ip6buf[INET6_ADDRSTRLEN];
2223 
2224 	if (req->newptr)
2225 		return EPERM;
2226 	error = 0;
2227 
2228 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2229 		u_short advrtrs;
2230 		size_t advance;
2231 		struct sockaddr_in6 *sin6, *s6;
2232 		struct nd_pfxrouter *pfr;
2233 
2234 		p = (struct in6_prefix *)buf;
2235 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2236 
2237 		if (p + 1 <= pe) {
2238 			bzero(p, sizeof(*p));
2239 			sin6 = (struct sockaddr_in6 *)(p + 1);
2240 
2241 			p->prefix = pr->ndpr_prefix;
2242 			if (sa6_recoverscope(&p->prefix)) {
2243 				log(LOG_ERR,
2244 				    "scope error in prefix list (%s)\n",
2245 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2246 				/* XXX: press on... */
2247 			}
2248 			p->raflags = pr->ndpr_raf;
2249 			p->prefixlen = pr->ndpr_plen;
2250 			p->vltime = pr->ndpr_vltime;
2251 			p->pltime = pr->ndpr_pltime;
2252 			p->if_index = pr->ndpr_ifp->if_index;
2253 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2254 				p->expire = 0;
2255 			else {
2256 				time_t maxexpire;
2257 
2258 				/* XXX: we assume time_t is signed. */
2259 				maxexpire = (-1) &
2260 				    ~((time_t)1 <<
2261 				    ((sizeof(maxexpire) * 8) - 1));
2262 				if (pr->ndpr_vltime <
2263 				    maxexpire - pr->ndpr_lastupdate) {
2264 				    p->expire = pr->ndpr_lastupdate +
2265 				        pr->ndpr_vltime;
2266 				} else
2267 					p->expire = maxexpire;
2268 			}
2269 			p->refcnt = pr->ndpr_refcnt;
2270 			p->flags = pr->ndpr_stateflags;
2271 			p->origin = PR_ORIG_RA;
2272 			advrtrs = 0;
2273 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2274 			     pfr = pfr->pfr_next) {
2275 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2276 					advrtrs++;
2277 					continue;
2278 				}
2279 				s6 = &sin6[advrtrs];
2280 				bzero(s6, sizeof(*s6));
2281 				s6->sin6_family = AF_INET6;
2282 				s6->sin6_len = sizeof(*sin6);
2283 				s6->sin6_addr = pfr->router->rtaddr;
2284 				if (sa6_recoverscope(s6)) {
2285 					log(LOG_ERR,
2286 					    "scope error in "
2287 					    "prefix list (%s)\n",
2288 					    ip6_sprintf(ip6buf,
2289 						    &pfr->router->rtaddr));
2290 				}
2291 				advrtrs++;
2292 			}
2293 			p->advrtrs = advrtrs;
2294 		} else
2295 			panic("buffer too short");
2296 
2297 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2298 		error = SYSCTL_OUT(req, buf, advance);
2299 		if (error)
2300 			break;
2301 	}
2302 
2303 	return (error);
2304 }
2305