xref: /freebsd/sys/netinet6/nd6.c (revision 724b4bfdf1306e4f2c451b6d146fe0fe0353b2c8)
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(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 	struct nd_defrouter *dr, *ndr;
579 	struct nd_prefix *pr, *npr;
580 	struct in6_ifaddr *ia6, *nia6;
581 
582 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
583 	    nd6_timer, curvnet);
584 
585 	/* expire default router list */
586 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
587 		if (dr->expire && dr->expire < time_second)
588 			defrtrlist_del(dr);
589 	}
590 
591 	/*
592 	 * expire interface addresses.
593 	 * in the past the loop was inside prefix expiry processing.
594 	 * However, from a stricter speci-confrmance standpoint, we should
595 	 * rather separate address lifetimes and prefix lifetimes.
596 	 *
597 	 * XXXRW: in6_ifaddrhead locking.
598 	 */
599   addrloop:
600 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
601 		/* check address lifetime */
602 		if (IFA6_IS_INVALID(ia6)) {
603 			int regen = 0;
604 
605 			/*
606 			 * If the expiring address is temporary, try
607 			 * regenerating a new one.  This would be useful when
608 			 * we suspended a laptop PC, then turned it on after a
609 			 * period that could invalidate all temporary
610 			 * addresses.  Although we may have to restart the
611 			 * loop (see below), it must be after purging the
612 			 * address.  Otherwise, we'd see an infinite loop of
613 			 * regeneration.
614 			 */
615 			if (V_ip6_use_tempaddr &&
616 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
617 				if (regen_tmpaddr(ia6) == 0)
618 					regen = 1;
619 			}
620 
621 			in6_purgeaddr(&ia6->ia_ifa);
622 
623 			if (regen)
624 				goto addrloop; /* XXX: see below */
625 		} else if (IFA6_IS_DEPRECATED(ia6)) {
626 			int oldflags = ia6->ia6_flags;
627 
628 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
629 
630 			/*
631 			 * If a temporary address has just become deprecated,
632 			 * regenerate a new one if possible.
633 			 */
634 			if (V_ip6_use_tempaddr &&
635 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
636 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
637 
638 				if (regen_tmpaddr(ia6) == 0) {
639 					/*
640 					 * A new temporary address is
641 					 * generated.
642 					 * XXX: this means the address chain
643 					 * has changed while we are still in
644 					 * the loop.  Although the change
645 					 * would not cause disaster (because
646 					 * it's not a deletion, but an
647 					 * addition,) we'd rather restart the
648 					 * loop just for safety.  Or does this
649 					 * significantly reduce performance??
650 					 */
651 					goto addrloop;
652 				}
653 			}
654 		} else {
655 			/*
656 			 * A new RA might have made a deprecated address
657 			 * preferred.
658 			 */
659 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
660 		}
661 	}
662 
663 	/* expire prefix list */
664 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
665 		/*
666 		 * check prefix lifetime.
667 		 * since pltime is just for autoconf, pltime processing for
668 		 * prefix is not necessary.
669 		 */
670 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
671 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
672 
673 			/*
674 			 * address expiration and prefix expiration are
675 			 * separate.  NEVER perform in6_purgeaddr here.
676 			 */
677 			prelist_remove(pr);
678 		}
679 	}
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_RLOCK(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_RUNLOCK(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 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
771 		if (dr->installed)
772 			continue;
773 
774 		if (dr->ifp == ifp)
775 			defrtrlist_del(dr);
776 	}
777 
778 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
779 		if (!dr->installed)
780 			continue;
781 
782 		if (dr->ifp == ifp)
783 			defrtrlist_del(dr);
784 	}
785 
786 	/* Nuke prefix list entries toward ifp */
787 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
788 		if (pr->ndpr_ifp == ifp) {
789 			/*
790 			 * Because if_detach() does *not* release prefixes
791 			 * while purging addresses the reference count will
792 			 * still be above zero. We therefore reset it to
793 			 * make sure that the prefix really gets purged.
794 			 */
795 			pr->ndpr_refcnt = 0;
796 
797 			/*
798 			 * Previously, pr->ndpr_addr is removed as well,
799 			 * but I strongly believe we don't have to do it.
800 			 * nd6_purge() is only called from in6_ifdetach(),
801 			 * which removes all the associated interface addresses
802 			 * by itself.
803 			 * (jinmei@kame.net 20010129)
804 			 */
805 			prelist_remove(pr);
806 		}
807 	}
808 
809 	/* cancel default outgoing interface setting */
810 	if (V_nd6_defifindex == ifp->if_index)
811 		nd6_setdefaultiface(0);
812 
813 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
814 		/* Refresh default router list. */
815 		defrouter_select();
816 	}
817 
818 	/* XXXXX
819 	 * We do not nuke the neighbor cache entries here any more
820 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
821 	 * nd6_purge() is invoked by in6_ifdetach() which is called
822 	 * from if_detach() where everything gets purged. So let
823 	 * in6_domifdetach() do the actual L2 table purging work.
824 	 */
825 }
826 
827 /*
828  * the caller acquires and releases the lock on the lltbls
829  * Returns the llentry locked
830  */
831 struct llentry *
832 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
833 {
834 	struct sockaddr_in6 sin6;
835 	struct llentry *ln;
836 	int llflags;
837 
838 	bzero(&sin6, sizeof(sin6));
839 	sin6.sin6_len = sizeof(struct sockaddr_in6);
840 	sin6.sin6_family = AF_INET6;
841 	sin6.sin6_addr = *addr6;
842 
843 	IF_AFDATA_LOCK_ASSERT(ifp);
844 
845 	llflags = 0;
846 	if (flags & ND6_CREATE)
847 	    llflags |= LLE_CREATE;
848 	if (flags & ND6_EXCLUSIVE)
849 	    llflags |= LLE_EXCLUSIVE;
850 
851 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
852 	if ((ln != NULL) && (llflags & LLE_CREATE))
853 		ln->ln_state = ND6_LLINFO_NOSTATE;
854 
855 	return (ln);
856 }
857 
858 /*
859  * Test whether a given IPv6 address is a neighbor or not, ignoring
860  * the actual neighbor cache.  The neighbor cache is ignored in order
861  * to not reenter the routing code from within itself.
862  */
863 static int
864 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
865 {
866 	struct nd_prefix *pr;
867 	struct ifaddr *dstaddr;
868 
869 	/*
870 	 * A link-local address is always a neighbor.
871 	 * XXX: a link does not necessarily specify a single interface.
872 	 */
873 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
874 		struct sockaddr_in6 sin6_copy;
875 		u_int32_t zone;
876 
877 		/*
878 		 * We need sin6_copy since sa6_recoverscope() may modify the
879 		 * content (XXX).
880 		 */
881 		sin6_copy = *addr;
882 		if (sa6_recoverscope(&sin6_copy))
883 			return (0); /* XXX: should be impossible */
884 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
885 			return (0);
886 		if (sin6_copy.sin6_scope_id == zone)
887 			return (1);
888 		else
889 			return (0);
890 	}
891 
892 	/*
893 	 * If the address matches one of our addresses,
894 	 * it should be a neighbor.
895 	 * If the address matches one of our on-link prefixes, it should be a
896 	 * neighbor.
897 	 */
898 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
899 		if (pr->ndpr_ifp != ifp)
900 			continue;
901 
902 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
903 			struct rtentry *rt;
904 
905 			/* Always use the default FIB here. */
906 			rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix,
907 			    0, 0, RT_DEFAULT_FIB);
908 			if (rt == NULL)
909 				continue;
910 			/*
911 			 * This is the case where multiple interfaces
912 			 * have the same prefix, but only one is installed
913 			 * into the routing table and that prefix entry
914 			 * is not the one being examined here. In the case
915 			 * where RADIX_MPATH is enabled, multiple route
916 			 * entries (of the same rt_key value) will be
917 			 * installed because the interface addresses all
918 			 * differ.
919 			 */
920 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
921 			       &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) {
922 				RTFREE_LOCKED(rt);
923 				continue;
924 			}
925 			RTFREE_LOCKED(rt);
926 		}
927 
928 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
929 		    &addr->sin6_addr, &pr->ndpr_mask))
930 			return (1);
931 	}
932 
933 	/*
934 	 * If the address is assigned on the node of the other side of
935 	 * a p2p interface, the address should be a neighbor.
936 	 */
937 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
938 	if (dstaddr != NULL) {
939 		if (dstaddr->ifa_ifp == ifp) {
940 			ifa_free(dstaddr);
941 			return (1);
942 		}
943 		ifa_free(dstaddr);
944 	}
945 
946 	/*
947 	 * If the default router list is empty, all addresses are regarded
948 	 * as on-link, and thus, as a neighbor.
949 	 */
950 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
951 	    TAILQ_EMPTY(&V_nd_defrouter) &&
952 	    V_nd6_defifindex == ifp->if_index) {
953 		return (1);
954 	}
955 
956 	return (0);
957 }
958 
959 
960 /*
961  * Detect if a given IPv6 address identifies a neighbor on a given link.
962  * XXX: should take care of the destination of a p2p link?
963  */
964 int
965 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
966 {
967 	struct llentry *lle;
968 	int rc = 0;
969 
970 	IF_AFDATA_UNLOCK_ASSERT(ifp);
971 	if (nd6_is_new_addr_neighbor(addr, ifp))
972 		return (1);
973 
974 	/*
975 	 * Even if the address matches none of our addresses, it might be
976 	 * in the neighbor cache.
977 	 */
978 	IF_AFDATA_LOCK(ifp);
979 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
980 		LLE_RUNLOCK(lle);
981 		rc = 1;
982 	}
983 	IF_AFDATA_UNLOCK(ifp);
984 	return (rc);
985 }
986 
987 /*
988  * Free an nd6 llinfo entry.
989  * Since the function would cause significant changes in the kernel, DO NOT
990  * make it global, unless you have a strong reason for the change, and are sure
991  * that the change is safe.
992  */
993 static struct llentry *
994 nd6_free(struct llentry *ln, int gc)
995 {
996         struct llentry *next;
997 	struct nd_defrouter *dr;
998 	struct ifnet *ifp;
999 
1000 	LLE_WLOCK_ASSERT(ln);
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_locked(ln, -1);
1009 
1010 	ifp = ln->lle_tbl->llt_ifp;
1011 
1012 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1013 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, 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_locked(ln,
1031 				    (dr->expire - time_second) * hz);
1032 			else
1033 				nd6_llinfo_settimer_locked(ln,
1034 				    (long)V_nd6_gctimer * hz);
1035 
1036 			next = LIST_NEXT(ln, lle_next);
1037 			LLE_REMREF(ln);
1038 			LLE_WUNLOCK(ln);
1039 			return (next);
1040 		}
1041 
1042 		if (dr) {
1043 			/*
1044 			 * Unreachablity of a router might affect the default
1045 			 * router selection and on-link detection of advertised
1046 			 * prefixes.
1047 			 */
1048 
1049 			/*
1050 			 * Temporarily fake the state to choose a new default
1051 			 * router and to perform on-link determination of
1052 			 * prefixes correctly.
1053 			 * Below the state will be set correctly,
1054 			 * or the entry itself will be deleted.
1055 			 */
1056 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1057 		}
1058 
1059 		if (ln->ln_router || dr) {
1060 
1061 			/*
1062 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1063 			 * rnh and for the calls to pfxlist_onlink_check() and
1064 			 * defrouter_select() in the block further down for calls
1065 			 * into nd6_lookup().  We still hold a ref.
1066 			 */
1067 			LLE_WUNLOCK(ln);
1068 
1069 			/*
1070 			 * rt6_flush must be called whether or not the neighbor
1071 			 * is in the Default Router List.
1072 			 * See a corresponding comment in nd6_na_input().
1073 			 */
1074 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp);
1075 		}
1076 
1077 		if (dr) {
1078 			/*
1079 			 * Since defrouter_select() does not affect the
1080 			 * on-link determination and MIP6 needs the check
1081 			 * before the default router selection, we perform
1082 			 * the check now.
1083 			 */
1084 			pfxlist_onlink_check();
1085 
1086 			/*
1087 			 * Refresh default router list.
1088 			 */
1089 			defrouter_select();
1090 		}
1091 
1092 		if (ln->ln_router || dr)
1093 			LLE_WLOCK(ln);
1094 	}
1095 
1096 	/*
1097 	 * Before deleting the entry, remember the next entry as the
1098 	 * return value.  We need this because pfxlist_onlink_check() above
1099 	 * might have freed other entries (particularly the old next entry) as
1100 	 * a side effect (XXX).
1101 	 */
1102 	next = LIST_NEXT(ln, lle_next);
1103 
1104 	/*
1105 	 * Save to unlock. We still hold an extra reference and will not
1106 	 * free(9) in llentry_free() if someone else holds one as well.
1107 	 */
1108 	LLE_WUNLOCK(ln);
1109 	IF_AFDATA_LOCK(ifp);
1110 	LLE_WLOCK(ln);
1111 	LLE_REMREF(ln);
1112 	llentry_free(ln);
1113 	IF_AFDATA_UNLOCK(ifp);
1114 
1115 	return (next);
1116 }
1117 
1118 /*
1119  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1120  *
1121  * XXX cost-effective methods?
1122  */
1123 void
1124 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1125 {
1126 	struct llentry *ln;
1127 	struct ifnet *ifp;
1128 
1129 	if ((dst6 == NULL) || (rt == NULL))
1130 		return;
1131 
1132 	ifp = rt->rt_ifp;
1133 	IF_AFDATA_LOCK(ifp);
1134 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1135 	IF_AFDATA_UNLOCK(ifp);
1136 	if (ln == NULL)
1137 		return;
1138 
1139 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1140 		goto done;
1141 
1142 	/*
1143 	 * if we get upper-layer reachability confirmation many times,
1144 	 * it is possible we have false information.
1145 	 */
1146 	if (!force) {
1147 		ln->ln_byhint++;
1148 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1149 			goto done;
1150 		}
1151 	}
1152 
1153  	ln->ln_state = ND6_LLINFO_REACHABLE;
1154 	if (!ND6_LLINFO_PERMANENT(ln)) {
1155 		nd6_llinfo_settimer_locked(ln,
1156 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1157 	}
1158 done:
1159 	LLE_WUNLOCK(ln);
1160 }
1161 
1162 
1163 /*
1164  * Rejuvenate this function for routing operations related
1165  * processing.
1166  */
1167 void
1168 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1169 {
1170 	struct sockaddr_in6 *gateway;
1171 	struct nd_defrouter *dr;
1172 	struct ifnet *ifp;
1173 
1174 	RT_LOCK_ASSERT(rt);
1175 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1176 	ifp = rt->rt_ifp;
1177 
1178 	switch (req) {
1179 	case RTM_ADD:
1180 		break;
1181 
1182 	case RTM_DELETE:
1183 		if (!ifp)
1184 			return;
1185 		/*
1186 		 * Only indirect routes are interesting.
1187 		 */
1188 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1189 			return;
1190 		/*
1191 		 * check for default route
1192 		 */
1193 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1194 				       &SIN6(rt_key(rt))->sin6_addr)) {
1195 
1196 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1197 			if (dr != NULL)
1198 				dr->installed = 0;
1199 		}
1200 		break;
1201 	}
1202 }
1203 
1204 
1205 int
1206 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1207 {
1208 	struct in6_drlist *drl = (struct in6_drlist *)data;
1209 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1210 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1211 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1212 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1213 	struct nd_defrouter *dr;
1214 	struct nd_prefix *pr;
1215 	int i = 0, error = 0;
1216 
1217 	switch (cmd) {
1218 	case SIOCGDRLST_IN6:
1219 		/*
1220 		 * obsolete API, use sysctl under net.inet6.icmp6
1221 		 */
1222 		bzero(drl, sizeof(*drl));
1223 		TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
1224 			if (i >= DRLSTSIZ)
1225 				break;
1226 			drl->defrouter[i].rtaddr = dr->rtaddr;
1227 			in6_clearscope(&drl->defrouter[i].rtaddr);
1228 
1229 			drl->defrouter[i].flags = dr->flags;
1230 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1231 			drl->defrouter[i].expire = dr->expire;
1232 			drl->defrouter[i].if_index = dr->ifp->if_index;
1233 			i++;
1234 		}
1235 		break;
1236 	case SIOCGPRLST_IN6:
1237 		/*
1238 		 * obsolete API, use sysctl under net.inet6.icmp6
1239 		 *
1240 		 * XXX the structure in6_prlist was changed in backward-
1241 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1242 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1243 		 */
1244 		/*
1245 		 * XXX meaning of fields, especialy "raflags", is very
1246 		 * differnet between RA prefix list and RR/static prefix list.
1247 		 * how about separating ioctls into two?
1248 		 */
1249 		bzero(oprl, sizeof(*oprl));
1250 		LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1251 			struct nd_pfxrouter *pfr;
1252 			int j;
1253 
1254 			if (i >= PRLSTSIZ)
1255 				break;
1256 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1257 			oprl->prefix[i].raflags = pr->ndpr_raf;
1258 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1259 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1260 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1261 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1262 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1263 				oprl->prefix[i].expire = 0;
1264 			else {
1265 				time_t maxexpire;
1266 
1267 				/* XXX: we assume time_t is signed. */
1268 				maxexpire = (-1) &
1269 				    ~((time_t)1 <<
1270 				    ((sizeof(maxexpire) * 8) - 1));
1271 				if (pr->ndpr_vltime <
1272 				    maxexpire - pr->ndpr_lastupdate) {
1273 					oprl->prefix[i].expire =
1274 					    pr->ndpr_lastupdate +
1275 					    pr->ndpr_vltime;
1276 				} else
1277 					oprl->prefix[i].expire = maxexpire;
1278 			}
1279 
1280 			j = 0;
1281 			LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
1282 				if (j < DRLSTSIZ) {
1283 #define RTRADDR oprl->prefix[i].advrtr[j]
1284 					RTRADDR = pfr->router->rtaddr;
1285 					in6_clearscope(&RTRADDR);
1286 #undef RTRADDR
1287 				}
1288 				j++;
1289 			}
1290 			oprl->prefix[i].advrtrs = j;
1291 			oprl->prefix[i].origin = PR_ORIG_RA;
1292 
1293 			i++;
1294 		}
1295 
1296 		break;
1297 	case OSIOCGIFINFO_IN6:
1298 #define ND	ndi->ndi
1299 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1300 		bzero(&ND, sizeof(ND));
1301 		ND.linkmtu = IN6_LINKMTU(ifp);
1302 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1303 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1304 		ND.reachable = ND_IFINFO(ifp)->reachable;
1305 		ND.retrans = ND_IFINFO(ifp)->retrans;
1306 		ND.flags = ND_IFINFO(ifp)->flags;
1307 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1308 		ND.chlim = ND_IFINFO(ifp)->chlim;
1309 		break;
1310 	case SIOCGIFINFO_IN6:
1311 		ND = *ND_IFINFO(ifp);
1312 		break;
1313 	case SIOCSIFINFO_IN6:
1314 		/*
1315 		 * used to change host variables from userland.
1316 		 * intented for a use on router to reflect RA configurations.
1317 		 */
1318 		/* 0 means 'unspecified' */
1319 		if (ND.linkmtu != 0) {
1320 			if (ND.linkmtu < IPV6_MMTU ||
1321 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1322 				error = EINVAL;
1323 				break;
1324 			}
1325 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1326 		}
1327 
1328 		if (ND.basereachable != 0) {
1329 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1330 
1331 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1332 			if (ND.basereachable != obasereachable)
1333 				ND_IFINFO(ifp)->reachable =
1334 				    ND_COMPUTE_RTIME(ND.basereachable);
1335 		}
1336 		if (ND.retrans != 0)
1337 			ND_IFINFO(ifp)->retrans = ND.retrans;
1338 		if (ND.chlim != 0)
1339 			ND_IFINFO(ifp)->chlim = ND.chlim;
1340 		/* FALLTHROUGH */
1341 	case SIOCSIFINFO_FLAGS:
1342 	{
1343 		struct ifaddr *ifa;
1344 		struct in6_ifaddr *ia;
1345 
1346 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1347 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1348 			/* ifdisabled 1->0 transision */
1349 
1350 			/*
1351 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1352 			 * has an link-local address with IN6_IFF_DUPLICATED,
1353 			 * do not clear ND6_IFF_IFDISABLED.
1354 			 * See RFC 4862, Section 5.4.5.
1355 			 */
1356 			int duplicated_linklocal = 0;
1357 
1358 			IF_ADDR_RLOCK(ifp);
1359 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1360 				if (ifa->ifa_addr->sa_family != AF_INET6)
1361 					continue;
1362 				ia = (struct in6_ifaddr *)ifa;
1363 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1364 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1365 					duplicated_linklocal = 1;
1366 					break;
1367 				}
1368 			}
1369 			IF_ADDR_RUNLOCK(ifp);
1370 
1371 			if (duplicated_linklocal) {
1372 				ND.flags |= ND6_IFF_IFDISABLED;
1373 				log(LOG_ERR, "Cannot enable an interface"
1374 				    " with a link-local address marked"
1375 				    " duplicate.\n");
1376 			} else {
1377 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1378 				if (ifp->if_flags & IFF_UP)
1379 					in6_if_up(ifp);
1380 			}
1381 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1382 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1383 			/* ifdisabled 0->1 transision */
1384 			/* Mark all IPv6 address as tentative. */
1385 
1386 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1387 			IF_ADDR_RLOCK(ifp);
1388 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1389 				if (ifa->ifa_addr->sa_family != AF_INET6)
1390 					continue;
1391 				ia = (struct in6_ifaddr *)ifa;
1392 				ia->ia6_flags |= IN6_IFF_TENTATIVE;
1393 			}
1394 			IF_ADDR_RUNLOCK(ifp);
1395 		}
1396 
1397 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1398 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1399 				/* auto_linklocal 0->1 transision */
1400 
1401 				/* If no link-local address on ifp, configure */
1402 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1403 				in6_ifattach(ifp, NULL);
1404 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1405 			    ifp->if_flags & IFF_UP) {
1406 				/*
1407 				 * When the IF already has
1408 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1409 				 * address is assigned, and IFF_UP, try to
1410 				 * assign one.
1411 				 */
1412 				int haslinklocal = 0;
1413 
1414 				IF_ADDR_RLOCK(ifp);
1415 				TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1416 					if (ifa->ifa_addr->sa_family != AF_INET6)
1417 						continue;
1418 					ia = (struct in6_ifaddr *)ifa;
1419 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) {
1420 						haslinklocal = 1;
1421 						break;
1422 					}
1423 				}
1424 				IF_ADDR_RUNLOCK(ifp);
1425 				if (!haslinklocal)
1426 					in6_ifattach(ifp, NULL);
1427 			}
1428 		}
1429 	}
1430 		ND_IFINFO(ifp)->flags = ND.flags;
1431 		break;
1432 #undef ND
1433 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1434 		/* sync kernel routing table with the default router list */
1435 		defrouter_reset();
1436 		defrouter_select();
1437 		break;
1438 	case SIOCSPFXFLUSH_IN6:
1439 	{
1440 		/* flush all the prefix advertised by routers */
1441 		struct nd_prefix *pr, *next;
1442 
1443 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1444 			struct in6_ifaddr *ia, *ia_next;
1445 
1446 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1447 				continue; /* XXX */
1448 
1449 			/* do we really have to remove addresses as well? */
1450 			/* XXXRW: in6_ifaddrhead locking. */
1451 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1452 			    ia_next) {
1453 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1454 					continue;
1455 
1456 				if (ia->ia6_ndpr == pr)
1457 					in6_purgeaddr(&ia->ia_ifa);
1458 			}
1459 			prelist_remove(pr);
1460 		}
1461 		break;
1462 	}
1463 	case SIOCSRTRFLUSH_IN6:
1464 	{
1465 		/* flush all the default routers */
1466 		struct nd_defrouter *dr, *next;
1467 
1468 		defrouter_reset();
1469 		TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) {
1470 			defrtrlist_del(dr);
1471 		}
1472 		defrouter_select();
1473 		break;
1474 	}
1475 	case SIOCGNBRINFO_IN6:
1476 	{
1477 		struct llentry *ln;
1478 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1479 
1480 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1481 			return (error);
1482 
1483 		IF_AFDATA_LOCK(ifp);
1484 		ln = nd6_lookup(&nb_addr, 0, ifp);
1485 		IF_AFDATA_UNLOCK(ifp);
1486 
1487 		if (ln == NULL) {
1488 			error = EINVAL;
1489 			break;
1490 		}
1491 		nbi->state = ln->ln_state;
1492 		nbi->asked = ln->la_asked;
1493 		nbi->isrouter = ln->ln_router;
1494 		nbi->expire = ln->la_expire;
1495 		LLE_RUNLOCK(ln);
1496 		break;
1497 	}
1498 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1499 		ndif->ifindex = V_nd6_defifindex;
1500 		break;
1501 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1502 		return (nd6_setdefaultiface(ndif->ifindex));
1503 	}
1504 	return (error);
1505 }
1506 
1507 /*
1508  * Create neighbor cache entry and cache link-layer address,
1509  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1510  *
1511  * type - ICMP6 type
1512  * code - type dependent information
1513  *
1514  * XXXXX
1515  *  The caller of this function already acquired the ndp
1516  *  cache table lock because the cache entry is returned.
1517  */
1518 struct llentry *
1519 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1520     int lladdrlen, int type, int code)
1521 {
1522 	struct llentry *ln = NULL;
1523 	int is_newentry;
1524 	int do_update;
1525 	int olladdr;
1526 	int llchange;
1527 	int flags;
1528 	int newstate = 0;
1529 	uint16_t router = 0;
1530 	struct sockaddr_in6 sin6;
1531 	struct mbuf *chain = NULL;
1532 	int static_route = 0;
1533 
1534 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1535 
1536 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1537 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1538 
1539 	/* nothing must be updated for unspecified address */
1540 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1541 		return NULL;
1542 
1543 	/*
1544 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1545 	 * the caller.
1546 	 *
1547 	 * XXX If the link does not have link-layer adderss, what should
1548 	 * we do? (ifp->if_addrlen == 0)
1549 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1550 	 * description on it in NS section (RFC 2461 7.2.3).
1551 	 */
1552 	flags = lladdr ? ND6_EXCLUSIVE : 0;
1553 	IF_AFDATA_LOCK(ifp);
1554 	ln = nd6_lookup(from, flags, ifp);
1555 
1556 	if (ln == NULL) {
1557 		flags |= ND6_EXCLUSIVE;
1558 		ln = nd6_lookup(from, flags | ND6_CREATE, ifp);
1559 		IF_AFDATA_UNLOCK(ifp);
1560 		is_newentry = 1;
1561 	} else {
1562 		IF_AFDATA_UNLOCK(ifp);
1563 		/* do nothing if static ndp is set */
1564 		if (ln->la_flags & LLE_STATIC) {
1565 			static_route = 1;
1566 			goto done;
1567 		}
1568 		is_newentry = 0;
1569 	}
1570 	if (ln == NULL)
1571 		return (NULL);
1572 
1573 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1574 	if (olladdr && lladdr) {
1575 		llchange = bcmp(lladdr, &ln->ll_addr,
1576 		    ifp->if_addrlen);
1577 	} else
1578 		llchange = 0;
1579 
1580 	/*
1581 	 * newentry olladdr  lladdr  llchange	(*=record)
1582 	 *	0	n	n	--	(1)
1583 	 *	0	y	n	--	(2)
1584 	 *	0	n	y	--	(3) * STALE
1585 	 *	0	y	y	n	(4) *
1586 	 *	0	y	y	y	(5) * STALE
1587 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1588 	 *	1	--	y	--	(7) * STALE
1589 	 */
1590 
1591 	if (lladdr) {		/* (3-5) and (7) */
1592 		/*
1593 		 * Record source link-layer address
1594 		 * XXX is it dependent to ifp->if_type?
1595 		 */
1596 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1597 		ln->la_flags |= LLE_VALID;
1598 	}
1599 
1600 	if (!is_newentry) {
1601 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1602 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1603 			do_update = 1;
1604 			newstate = ND6_LLINFO_STALE;
1605 		} else					/* (1-2,4) */
1606 			do_update = 0;
1607 	} else {
1608 		do_update = 1;
1609 		if (lladdr == NULL)			/* (6) */
1610 			newstate = ND6_LLINFO_NOSTATE;
1611 		else					/* (7) */
1612 			newstate = ND6_LLINFO_STALE;
1613 	}
1614 
1615 	if (do_update) {
1616 		/*
1617 		 * Update the state of the neighbor cache.
1618 		 */
1619 		ln->ln_state = newstate;
1620 
1621 		if (ln->ln_state == ND6_LLINFO_STALE) {
1622 			/*
1623 			 * XXX: since nd6_output() below will cause
1624 			 * state tansition to DELAY and reset the timer,
1625 			 * we must set the timer now, although it is actually
1626 			 * meaningless.
1627 			 */
1628 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1629 
1630 			if (ln->la_hold) {
1631 				struct mbuf *m_hold, *m_hold_next;
1632 
1633 				/*
1634 				 * reset the la_hold in advance, to explicitly
1635 				 * prevent a la_hold lookup in nd6_output()
1636 				 * (wouldn't happen, though...)
1637 				 */
1638 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1639 				    m_hold; m_hold = m_hold_next) {
1640 					m_hold_next = m_hold->m_nextpkt;
1641 					m_hold->m_nextpkt = NULL;
1642 
1643 					/*
1644 					 * we assume ifp is not a p2p here, so
1645 					 * just set the 2nd argument as the
1646 					 * 1st one.
1647 					 */
1648 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1649 				}
1650 				/*
1651 				 * If we have mbufs in the chain we need to do
1652 				 * deferred transmit. Copy the address from the
1653 				 * llentry before dropping the lock down below.
1654 				 */
1655 				if (chain != NULL)
1656 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1657 			}
1658 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1659 			/* probe right away */
1660 			nd6_llinfo_settimer_locked((void *)ln, 0);
1661 		}
1662 	}
1663 
1664 	/*
1665 	 * ICMP6 type dependent behavior.
1666 	 *
1667 	 * NS: clear IsRouter if new entry
1668 	 * RS: clear IsRouter
1669 	 * RA: set IsRouter if there's lladdr
1670 	 * redir: clear IsRouter if new entry
1671 	 *
1672 	 * RA case, (1):
1673 	 * The spec says that we must set IsRouter in the following cases:
1674 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1675 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1676 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1677 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1678 	 * neighbor cache, this is similar to (6).
1679 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1680 	 *
1681 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1682 	 *							D R
1683 	 *	0	n	n	--	(1)	c   ?     s
1684 	 *	0	y	n	--	(2)	c   s     s
1685 	 *	0	n	y	--	(3)	c   s     s
1686 	 *	0	y	y	n	(4)	c   s     s
1687 	 *	0	y	y	y	(5)	c   s     s
1688 	 *	1	--	n	--	(6) c	c	c s
1689 	 *	1	--	y	--	(7) c	c   s	c s
1690 	 *
1691 	 *					(c=clear s=set)
1692 	 */
1693 	switch (type & 0xff) {
1694 	case ND_NEIGHBOR_SOLICIT:
1695 		/*
1696 		 * New entry must have is_router flag cleared.
1697 		 */
1698 		if (is_newentry)	/* (6-7) */
1699 			ln->ln_router = 0;
1700 		break;
1701 	case ND_REDIRECT:
1702 		/*
1703 		 * If the icmp is a redirect to a better router, always set the
1704 		 * is_router flag.  Otherwise, if the entry is newly created,
1705 		 * clear the flag.  [RFC 2461, sec 8.3]
1706 		 */
1707 		if (code == ND_REDIRECT_ROUTER)
1708 			ln->ln_router = 1;
1709 		else if (is_newentry) /* (6-7) */
1710 			ln->ln_router = 0;
1711 		break;
1712 	case ND_ROUTER_SOLICIT:
1713 		/*
1714 		 * is_router flag must always be cleared.
1715 		 */
1716 		ln->ln_router = 0;
1717 		break;
1718 	case ND_ROUTER_ADVERT:
1719 		/*
1720 		 * Mark an entry with lladdr as a router.
1721 		 */
1722 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1723 		    (is_newentry && lladdr)) {			/* (7) */
1724 			ln->ln_router = 1;
1725 		}
1726 		break;
1727 	}
1728 
1729 	if (ln != NULL) {
1730 		static_route = (ln->la_flags & LLE_STATIC);
1731 		router = ln->ln_router;
1732 
1733 		if (flags & ND6_EXCLUSIVE)
1734 			LLE_WUNLOCK(ln);
1735 		else
1736 			LLE_RUNLOCK(ln);
1737 		if (static_route)
1738 			ln = NULL;
1739 	}
1740 	if (chain)
1741 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1742 
1743 	/*
1744 	 * When the link-layer address of a router changes, select the
1745 	 * best router again.  In particular, when the neighbor entry is newly
1746 	 * created, it might affect the selection policy.
1747 	 * Question: can we restrict the first condition to the "is_newentry"
1748 	 * case?
1749 	 * XXX: when we hear an RA from a new router with the link-layer
1750 	 * address option, defrouter_select() is called twice, since
1751 	 * defrtrlist_update called the function as well.  However, I believe
1752 	 * we can compromise the overhead, since it only happens the first
1753 	 * time.
1754 	 * XXX: although defrouter_select() should not have a bad effect
1755 	 * for those are not autoconfigured hosts, we explicitly avoid such
1756 	 * cases for safety.
1757 	 */
1758 	if (do_update && router &&
1759 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1760 		/*
1761 		 * guaranteed recursion
1762 		 */
1763 		defrouter_select();
1764 	}
1765 
1766 	return (ln);
1767 done:
1768 	if (ln != NULL) {
1769 		if (flags & ND6_EXCLUSIVE)
1770 			LLE_WUNLOCK(ln);
1771 		else
1772 			LLE_RUNLOCK(ln);
1773 		if (static_route)
1774 			ln = NULL;
1775 	}
1776 	return (ln);
1777 }
1778 
1779 static void
1780 nd6_slowtimo(void *arg)
1781 {
1782 	CURVNET_SET((struct vnet *) arg);
1783 	struct nd_ifinfo *nd6if;
1784 	struct ifnet *ifp;
1785 
1786 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1787 	    nd6_slowtimo, curvnet);
1788 	IFNET_RLOCK_NOSLEEP();
1789 	TAILQ_FOREACH(ifp, &V_ifnet, if_list) {
1790 		nd6if = ND_IFINFO(ifp);
1791 		if (nd6if->basereachable && /* already initialized */
1792 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1793 			/*
1794 			 * Since reachable time rarely changes by router
1795 			 * advertisements, we SHOULD insure that a new random
1796 			 * value gets recomputed at least once every few hours.
1797 			 * (RFC 2461, 6.3.4)
1798 			 */
1799 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1800 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1801 		}
1802 	}
1803 	IFNET_RUNLOCK_NOSLEEP();
1804 	CURVNET_RESTORE();
1805 }
1806 
1807 int
1808 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1809     struct sockaddr_in6 *dst, struct rtentry *rt0)
1810 {
1811 
1812 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1813 }
1814 
1815 
1816 /*
1817  * Note that I'm not enforcing any global serialization
1818  * lle state or asked changes here as the logic is too
1819  * complicated to avoid having to always acquire an exclusive
1820  * lock
1821  * KMM
1822  *
1823  */
1824 #define senderr(e) { error = (e); goto bad;}
1825 
1826 int
1827 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1828     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1829 	struct mbuf **chain)
1830 {
1831 	struct mbuf *m = m0;
1832 	struct m_tag *mtag;
1833 	struct llentry *ln = lle;
1834 	struct ip6_hdr *ip6;
1835 	int error = 0;
1836 	int flags = 0;
1837 	int ip6len;
1838 
1839 #ifdef INVARIANTS
1840 	if (lle != NULL) {
1841 
1842 		LLE_WLOCK_ASSERT(lle);
1843 
1844 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1845 	}
1846 #endif
1847 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1848 		goto sendpkt;
1849 
1850 	if (nd6_need_cache(ifp) == 0)
1851 		goto sendpkt;
1852 
1853 	/*
1854 	 * next hop determination.  This routine is derived from ether_output.
1855 	 */
1856 
1857 	/*
1858 	 * Address resolution or Neighbor Unreachability Detection
1859 	 * for the next hop.
1860 	 * At this point, the destination of the packet must be a unicast
1861 	 * or an anycast address(i.e. not a multicast).
1862 	 */
1863 
1864 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1865 	if (ln == NULL) {
1866 	retry:
1867 		IF_AFDATA_LOCK(ifp);
1868 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1869 		IF_AFDATA_UNLOCK(ifp);
1870 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1871 			/*
1872 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1873 			 * the condition below is not very efficient.  But we believe
1874 			 * it is tolerable, because this should be a rare case.
1875 			 */
1876 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1877 			IF_AFDATA_LOCK(ifp);
1878 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1879 			IF_AFDATA_UNLOCK(ifp);
1880 		}
1881 	}
1882 	if (ln == NULL) {
1883 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1884 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1885 			char ip6buf[INET6_ADDRSTRLEN];
1886 			log(LOG_DEBUG,
1887 			    "nd6_output: can't allocate llinfo for %s "
1888 			    "(ln=%p)\n",
1889 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1890 			senderr(EIO);	/* XXX: good error? */
1891 		}
1892 		goto sendpkt;	/* send anyway */
1893 	}
1894 
1895 	/* We don't have to do link-layer address resolution on a p2p link. */
1896 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1897 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1898 		if ((flags & LLE_EXCLUSIVE) == 0) {
1899 			flags |= LLE_EXCLUSIVE;
1900 			goto retry;
1901 		}
1902 		ln->ln_state = ND6_LLINFO_STALE;
1903 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1904 	}
1905 
1906 	/*
1907 	 * The first time we send a packet to a neighbor whose entry is
1908 	 * STALE, we have to change the state to DELAY and a sets a timer to
1909 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1910 	 * neighbor unreachability detection on expiration.
1911 	 * (RFC 2461 7.3.3)
1912 	 */
1913 	if (ln->ln_state == ND6_LLINFO_STALE) {
1914 		if ((flags & LLE_EXCLUSIVE) == 0) {
1915 			flags |= LLE_EXCLUSIVE;
1916 			LLE_RUNLOCK(ln);
1917 			goto retry;
1918 		}
1919 		ln->la_asked = 0;
1920 		ln->ln_state = ND6_LLINFO_DELAY;
1921 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1922 	}
1923 
1924 	/*
1925 	 * If the neighbor cache entry has a state other than INCOMPLETE
1926 	 * (i.e. its link-layer address is already resolved), just
1927 	 * send the packet.
1928 	 */
1929 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1930 		goto sendpkt;
1931 
1932 	/*
1933 	 * There is a neighbor cache entry, but no ethernet address
1934 	 * response yet.  Append this latest packet to the end of the
1935 	 * packet queue in the mbuf, unless the number of the packet
1936 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1937 	 * the oldest packet in the queue will be removed.
1938 	 */
1939 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1940 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1941 
1942 	if ((flags & LLE_EXCLUSIVE) == 0) {
1943 		flags |= LLE_EXCLUSIVE;
1944 		LLE_RUNLOCK(ln);
1945 		goto retry;
1946 	}
1947 
1948 	LLE_WLOCK_ASSERT(ln);
1949 
1950 	if (ln->la_hold) {
1951 		struct mbuf *m_hold;
1952 		int i;
1953 
1954 		i = 0;
1955 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1956 			i++;
1957 			if (m_hold->m_nextpkt == NULL) {
1958 				m_hold->m_nextpkt = m;
1959 				break;
1960 			}
1961 		}
1962 		while (i >= V_nd6_maxqueuelen) {
1963 			m_hold = ln->la_hold;
1964 			ln->la_hold = ln->la_hold->m_nextpkt;
1965 			m_freem(m_hold);
1966 			i--;
1967 		}
1968 	} else {
1969 		ln->la_hold = m;
1970 	}
1971 
1972 	/*
1973 	 * If there has been no NS for the neighbor after entering the
1974 	 * INCOMPLETE state, send the first solicitation.
1975 	 */
1976 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1977 		ln->la_asked++;
1978 
1979 		nd6_llinfo_settimer_locked(ln,
1980 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1981 		LLE_WUNLOCK(ln);
1982 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1983 		if (lle != NULL && ln == lle)
1984 			LLE_WLOCK(lle);
1985 
1986 	} else if (lle == NULL || ln != lle) {
1987 		/*
1988 		 * We did the lookup (no lle arg) so we
1989 		 * need to do the unlock here.
1990 		 */
1991 		LLE_WUNLOCK(ln);
1992 	}
1993 
1994 	return (0);
1995 
1996   sendpkt:
1997 	/* discard the packet if IPv6 operation is disabled on the interface */
1998 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1999 		error = ENETDOWN; /* better error? */
2000 		goto bad;
2001 	}
2002 	/*
2003 	 * ln is valid and the caller did not pass in
2004 	 * an llentry
2005 	 */
2006 	if ((ln != NULL) && (lle == NULL)) {
2007 		if (flags & LLE_EXCLUSIVE)
2008 			LLE_WUNLOCK(ln);
2009 		else
2010 			LLE_RUNLOCK(ln);
2011 	}
2012 
2013 #ifdef MAC
2014 	mac_netinet6_nd6_send(ifp, m);
2015 #endif
2016 
2017 	/*
2018 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2019 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2020 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2021 	 * to be diverted to user space.  When re-injected into the kernel,
2022 	 * send_output() will directly dispatch them to the outgoing interface.
2023 	 */
2024 	if (send_sendso_input_hook != NULL) {
2025 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2026 		if (mtag != NULL) {
2027 			ip6 = mtod(m, struct ip6_hdr *);
2028 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2029 			/* Use the SEND socket */
2030 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2031 			    ip6len);
2032 			/* -1 == no app on SEND socket */
2033 			if (error == 0 || error != -1)
2034 			    return (error);
2035 		}
2036 	}
2037 
2038 	/*
2039 	 * We were passed in a pointer to an lle with the lock held
2040 	 * this means that we can't call if_output as we will
2041 	 * recurse on the lle lock - so what we do is we create
2042 	 * a list of mbufs to send and transmit them in the caller
2043 	 * after the lock is dropped
2044 	 */
2045 	if (lle != NULL) {
2046 		if (*chain == NULL)
2047 			*chain = m;
2048 		else {
2049 			struct mbuf *mb;
2050 
2051 			/*
2052 			 * append mbuf to end of deferred chain
2053 			 */
2054 			mb = *chain;
2055 			while (mb->m_nextpkt != NULL)
2056 				mb = mb->m_nextpkt;
2057 			mb->m_nextpkt = m;
2058 		}
2059 		return (error);
2060 	}
2061 	/* Reset layer specific mbuf flags to avoid confusing lower layers. */
2062 	m->m_flags &= ~(M_PROTOFLAGS);
2063 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
2064 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
2065 		    NULL));
2066 	}
2067 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
2068 	return (error);
2069 
2070   bad:
2071 	/*
2072 	 * ln is valid and the caller did not pass in
2073 	 * an llentry
2074 	 */
2075 	if ((ln != NULL) && (lle == NULL)) {
2076 		if (flags & LLE_EXCLUSIVE)
2077 			LLE_WUNLOCK(ln);
2078 		else
2079 			LLE_RUNLOCK(ln);
2080 	}
2081 	if (m)
2082 		m_freem(m);
2083 	return (error);
2084 }
2085 #undef senderr
2086 
2087 
2088 int
2089 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2090     struct sockaddr_in6 *dst, struct route *ro)
2091 {
2092 	struct mbuf *m, *m_head;
2093 	struct ifnet *outifp;
2094 	int error = 0;
2095 
2096 	m_head = chain;
2097 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2098 		outifp = origifp;
2099 	else
2100 		outifp = ifp;
2101 
2102 	while (m_head) {
2103 		m = m_head;
2104 		m_head = m_head->m_nextpkt;
2105 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
2106 	}
2107 
2108 	/*
2109 	 * XXX
2110 	 * note that intermediate errors are blindly ignored - but this is
2111 	 * the same convention as used with nd6_output when called by
2112 	 * nd6_cache_lladdr
2113 	 */
2114 	return (error);
2115 }
2116 
2117 
2118 int
2119 nd6_need_cache(struct ifnet *ifp)
2120 {
2121 	/*
2122 	 * XXX: we currently do not make neighbor cache on any interface
2123 	 * other than ARCnet, Ethernet, FDDI and GIF.
2124 	 *
2125 	 * RFC2893 says:
2126 	 * - unidirectional tunnels needs no ND
2127 	 */
2128 	switch (ifp->if_type) {
2129 	case IFT_ARCNET:
2130 	case IFT_ETHER:
2131 	case IFT_FDDI:
2132 	case IFT_IEEE1394:
2133 #ifdef IFT_L2VLAN
2134 	case IFT_L2VLAN:
2135 #endif
2136 #ifdef IFT_IEEE80211
2137 	case IFT_IEEE80211:
2138 #endif
2139 	case IFT_INFINIBAND:
2140 	case IFT_GIF:		/* XXX need more cases? */
2141 	case IFT_PPP:
2142 	case IFT_TUNNEL:
2143 	case IFT_BRIDGE:
2144 	case IFT_PROPVIRTUAL:
2145 		return (1);
2146 	default:
2147 		return (0);
2148 	}
2149 }
2150 
2151 /*
2152  * the callers of this function need to be re-worked to drop
2153  * the lle lock, drop here for now
2154  */
2155 int
2156 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2157     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2158 {
2159 	struct llentry *ln;
2160 
2161 	*lle = NULL;
2162 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2163 	if (m->m_flags & M_MCAST) {
2164 		int i;
2165 
2166 		switch (ifp->if_type) {
2167 		case IFT_ETHER:
2168 		case IFT_FDDI:
2169 #ifdef IFT_L2VLAN
2170 		case IFT_L2VLAN:
2171 #endif
2172 #ifdef IFT_IEEE80211
2173 		case IFT_IEEE80211:
2174 #endif
2175 		case IFT_BRIDGE:
2176 		case IFT_ISO88025:
2177 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2178 						 desten);
2179 			return (0);
2180 		case IFT_IEEE1394:
2181 			/*
2182 			 * netbsd can use if_broadcastaddr, but we don't do so
2183 			 * to reduce # of ifdef.
2184 			 */
2185 			for (i = 0; i < ifp->if_addrlen; i++)
2186 				desten[i] = ~0;
2187 			return (0);
2188 		case IFT_ARCNET:
2189 			*desten = 0;
2190 			return (0);
2191 		default:
2192 			m_freem(m);
2193 			return (EAFNOSUPPORT);
2194 		}
2195 	}
2196 
2197 
2198 	/*
2199 	 * the entry should have been created in nd6_store_lladdr
2200 	 */
2201 	IF_AFDATA_LOCK(ifp);
2202 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2203 	IF_AFDATA_UNLOCK(ifp);
2204 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2205 		if (ln != NULL)
2206 			LLE_RUNLOCK(ln);
2207 		/* this could happen, if we could not allocate memory */
2208 		m_freem(m);
2209 		return (1);
2210 	}
2211 
2212 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2213 	*lle = ln;
2214 	LLE_RUNLOCK(ln);
2215 	/*
2216 	 * A *small* use after free race exists here
2217 	 */
2218 	return (0);
2219 }
2220 
2221 static void
2222 clear_llinfo_pqueue(struct llentry *ln)
2223 {
2224 	struct mbuf *m_hold, *m_hold_next;
2225 
2226 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2227 		m_hold_next = m_hold->m_nextpkt;
2228 		m_freem(m_hold);
2229 	}
2230 
2231 	ln->la_hold = NULL;
2232 	return;
2233 }
2234 
2235 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2236 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2237 #ifdef SYSCTL_DECL
2238 SYSCTL_DECL(_net_inet6_icmp6);
2239 #endif
2240 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2241 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2242 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2243 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2244 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2245 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2246 
2247 static int
2248 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2249 {
2250 	struct in6_defrouter d;
2251 	struct nd_defrouter *dr;
2252 	int error;
2253 
2254 	if (req->newptr)
2255 		return (EPERM);
2256 
2257 	bzero(&d, sizeof(d));
2258 	d.rtaddr.sin6_family = AF_INET6;
2259 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2260 
2261 	/*
2262 	 * XXX locking
2263 	 */
2264 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2265 		d.rtaddr.sin6_addr = dr->rtaddr;
2266 		error = sa6_recoverscope(&d.rtaddr);
2267 		if (error != 0)
2268 			return (error);
2269 		d.flags = dr->flags;
2270 		d.rtlifetime = dr->rtlifetime;
2271 		d.expire = dr->expire;
2272 		d.if_index = dr->ifp->if_index;
2273 		error = SYSCTL_OUT(req, &d, sizeof(d));
2274 		if (error != 0)
2275 			return (error);
2276 	}
2277 	return (0);
2278 }
2279 
2280 static int
2281 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2282 {
2283 	struct in6_prefix p;
2284 	struct sockaddr_in6 s6;
2285 	struct nd_prefix *pr;
2286 	struct nd_pfxrouter *pfr;
2287 	time_t maxexpire;
2288 	int error;
2289 	char ip6buf[INET6_ADDRSTRLEN];
2290 
2291 	if (req->newptr)
2292 		return (EPERM);
2293 
2294 	bzero(&p, sizeof(p));
2295 	p.origin = PR_ORIG_RA;
2296 	bzero(&s6, sizeof(s6));
2297 	s6.sin6_family = AF_INET6;
2298 	s6.sin6_len = sizeof(s6);
2299 
2300 	/*
2301 	 * XXX locking
2302 	 */
2303 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2304 		p.prefix = pr->ndpr_prefix;
2305 		if (sa6_recoverscope(&p.prefix)) {
2306 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2307 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2308 			/* XXX: press on... */
2309 		}
2310 		p.raflags = pr->ndpr_raf;
2311 		p.prefixlen = pr->ndpr_plen;
2312 		p.vltime = pr->ndpr_vltime;
2313 		p.pltime = pr->ndpr_pltime;
2314 		p.if_index = pr->ndpr_ifp->if_index;
2315 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2316 			p.expire = 0;
2317 		else {
2318 			/* XXX: we assume time_t is signed. */
2319 			maxexpire = (-1) &
2320 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2321 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2322 				p.expire = pr->ndpr_lastupdate +
2323 				    pr->ndpr_vltime;
2324 			else
2325 				p.expire = maxexpire;
2326 		}
2327 		p.refcnt = pr->ndpr_refcnt;
2328 		p.flags = pr->ndpr_stateflags;
2329 		p.advrtrs = 0;
2330 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2331 			p.advrtrs++;
2332 		error = SYSCTL_OUT(req, &p, sizeof(p));
2333 		if (error != 0)
2334 			return (error);
2335 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2336 			s6.sin6_addr = pfr->router->rtaddr;
2337 			if (sa6_recoverscope(&s6))
2338 				log(LOG_ERR,
2339 				    "scope error in prefix list (%s)\n",
2340 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2341 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2342 			if (error != 0)
2343 				return (error);
2344 		}
2345 	}
2346 	return (0);
2347 }
2348