xref: /freebsd/sys/netinet6/nd6.c (revision 195ebc7e9e4b129de810833791a19dfb4349d6a9)
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 #include "opt_mac.h"
38 #include "opt_route.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/socket.h>
46 #include <sys/sockio.h>
47 #include <sys/time.h>
48 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/errno.h>
51 #include <sys/syslog.h>
52 #include <sys/lock.h>
53 #include <sys/rwlock.h>
54 #include <sys/queue.h>
55 #include <sys/sysctl.h>
56 #include <sys/vimage.h>
57 
58 #include <net/if.h>
59 #include <net/if_arc.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/iso88025.h>
63 #include <net/fddi.h>
64 #include <net/route.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <net/if_llatbl.h>
69 #define	L3_ADDR_SIN6(le)	((struct sockaddr_in6 *) L3_ADDR(le))
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/vinet6.h>
78 
79 #include <sys/limits.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
85 
86 #define SIN6(s) ((struct sockaddr_in6 *)s)
87 
88 #ifdef VIMAGE_GLOBALS
89 int nd6_prune;
90 int nd6_delay;
91 int nd6_umaxtries;
92 int nd6_mmaxtries;
93 int nd6_useloopback;
94 int nd6_gctimer;
95 
96 /* preventing too many loops in ND option parsing */
97 int nd6_maxndopt;
98 
99 int nd6_maxnudhint;
100 int nd6_maxqueuelen;
101 
102 int nd6_debug;
103 
104 /* for debugging? */
105 #if 0
106 static int nd6_inuse, nd6_allocated;
107 #endif
108 
109 struct nd_drhead nd_defrouter;
110 struct nd_prhead nd_prefix;
111 
112 int nd6_recalc_reachtm_interval;
113 #endif /* VIMAGE_GLOBALS */
114 
115 static struct sockaddr_in6 all1_sa;
116 
117 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
118 	struct ifnet *));
119 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
120 static void nd6_slowtimo(void *);
121 static int regen_tmpaddr(struct in6_ifaddr *);
122 static struct llentry *nd6_free(struct llentry *, int);
123 static void nd6_llinfo_timer(void *);
124 static void clear_llinfo_pqueue(struct llentry *);
125 
126 #ifdef VIMAGE_GLOBALS
127 struct callout nd6_slowtimo_ch;
128 struct callout nd6_timer_ch;
129 extern struct callout in6_tmpaddrtimer_ch;
130 extern int dad_ignore_ns;
131 extern int dad_maxtry;
132 #endif
133 
134 void
135 nd6_init(void)
136 {
137 	INIT_VNET_INET6(curvnet);
138 	static int nd6_init_done = 0;
139 	int i;
140 
141 	if (nd6_init_done) {
142 		log(LOG_NOTICE, "nd6_init called more than once(ignored)\n");
143 		return;
144 	}
145 
146 	V_nd6_prune	= 1;	/* walk list every 1 seconds */
147 	V_nd6_delay	= 5;	/* delay first probe time 5 second */
148 	V_nd6_umaxtries	= 3;	/* maximum unicast query */
149 	V_nd6_mmaxtries	= 3;	/* maximum multicast query */
150 	V_nd6_useloopback = 1;	/* use loopback interface for local traffic */
151 	V_nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
152 
153 	/* preventing too many loops in ND option parsing */
154 	V_nd6_maxndopt = 10;	/* max # of ND options allowed */
155 
156 	V_nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
157 	V_nd6_maxqueuelen = 1;	/* max pkts cached in unresolved ND entries */
158 
159 #ifdef ND6_DEBUG
160 	V_nd6_debug = 1;
161 #else
162 	V_nd6_debug = 0;
163 #endif
164 
165 	V_nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
166 
167 	V_dad_ignore_ns = 0;	/* ignore NS in DAD - specwise incorrect*/
168 	V_dad_maxtry = 15;	/* max # of *tries* to transmit DAD packet */
169 
170 	/*
171 	 * XXX just to get this to compile KMM
172 	 */
173 #ifdef notyet
174 	V_llinfo_nd6.ln_next = &V_llinfo_nd6;
175 	V_llinfo_nd6.ln_prev = &V_llinfo_nd6;
176 #endif
177 	LIST_INIT(&V_nd_prefix);
178 
179 	V_ip6_use_tempaddr = 0;
180 	V_ip6_temp_preferred_lifetime = DEF_TEMP_PREFERRED_LIFETIME;
181 	V_ip6_temp_valid_lifetime = DEF_TEMP_VALID_LIFETIME;
182 	V_ip6_temp_regen_advance = TEMPADDR_REGEN_ADVANCE;
183 
184 	all1_sa.sin6_family = AF_INET6;
185 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
186 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
187 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
188 
189 	/* initialization of the default router list */
190 	TAILQ_INIT(&V_nd_defrouter);
191 	/* start timer */
192 	callout_init(&V_nd6_slowtimo_ch, 0);
193 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
194 	    nd6_slowtimo, curvnet);
195 
196 	nd6_init_done = 1;
197 
198 }
199 
200 struct nd_ifinfo *
201 nd6_ifattach(struct ifnet *ifp)
202 {
203 	struct nd_ifinfo *nd;
204 
205 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
206 	bzero(nd, sizeof(*nd));
207 
208 	nd->initialized = 1;
209 
210 	nd->chlim = IPV6_DEFHLIM;
211 	nd->basereachable = REACHABLE_TIME;
212 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
213 	nd->retrans = RETRANS_TIMER;
214 	/*
215 	 * Note that the default value of ip6_accept_rtadv is 0, which means
216 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
217 	 * here.
218 	 */
219 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
220 
221 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
222 	nd6_setmtu0(ifp, nd);
223 
224 	return nd;
225 }
226 
227 void
228 nd6_ifdetach(struct nd_ifinfo *nd)
229 {
230 
231 	free(nd, M_IP6NDP);
232 }
233 
234 /*
235  * Reset ND level link MTU. This function is called when the physical MTU
236  * changes, which means we might have to adjust the ND level MTU.
237  */
238 void
239 nd6_setmtu(struct ifnet *ifp)
240 {
241 
242 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
243 }
244 
245 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
246 void
247 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
248 {
249 	INIT_VNET_INET6(ifp->if_vnet);
250 	u_int32_t omaxmtu;
251 
252 	omaxmtu = ndi->maxmtu;
253 
254 	switch (ifp->if_type) {
255 	case IFT_ARCNET:
256 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
257 		break;
258 	case IFT_FDDI:
259 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
260 		break;
261 	case IFT_ISO88025:
262 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
263 		 break;
264 	default:
265 		ndi->maxmtu = ifp->if_mtu;
266 		break;
267 	}
268 
269 	/*
270 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
271 	 * undesirable situation.  We thus notify the operator of the change
272 	 * explicitly.  The check for omaxmtu is necessary to restrict the
273 	 * log to the case of changing the MTU, not initializing it.
274 	 */
275 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
276 		log(LOG_NOTICE, "nd6_setmtu0: "
277 		    "new link MTU on %s (%lu) is too small for IPv6\n",
278 		    if_name(ifp), (unsigned long)ndi->maxmtu);
279 	}
280 
281 	if (ndi->maxmtu > V_in6_maxmtu)
282 		in6_setmaxmtu(); /* check all interfaces just in case */
283 
284 }
285 
286 void
287 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
288 {
289 
290 	bzero(ndopts, sizeof(*ndopts));
291 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
292 	ndopts->nd_opts_last
293 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
294 
295 	if (icmp6len == 0) {
296 		ndopts->nd_opts_done = 1;
297 		ndopts->nd_opts_search = NULL;
298 	}
299 }
300 
301 /*
302  * Take one ND option.
303  */
304 struct nd_opt_hdr *
305 nd6_option(union nd_opts *ndopts)
306 {
307 	struct nd_opt_hdr *nd_opt;
308 	int olen;
309 
310 	if (ndopts == NULL)
311 		panic("ndopts == NULL in nd6_option");
312 	if (ndopts->nd_opts_last == NULL)
313 		panic("uninitialized ndopts in nd6_option");
314 	if (ndopts->nd_opts_search == NULL)
315 		return NULL;
316 	if (ndopts->nd_opts_done)
317 		return NULL;
318 
319 	nd_opt = ndopts->nd_opts_search;
320 
321 	/* make sure nd_opt_len is inside the buffer */
322 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
323 		bzero(ndopts, sizeof(*ndopts));
324 		return NULL;
325 	}
326 
327 	olen = nd_opt->nd_opt_len << 3;
328 	if (olen == 0) {
329 		/*
330 		 * Message validation requires that all included
331 		 * options have a length that is greater than zero.
332 		 */
333 		bzero(ndopts, sizeof(*ndopts));
334 		return NULL;
335 	}
336 
337 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
338 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
339 		/* option overruns the end of buffer, invalid */
340 		bzero(ndopts, sizeof(*ndopts));
341 		return NULL;
342 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
343 		/* reached the end of options chain */
344 		ndopts->nd_opts_done = 1;
345 		ndopts->nd_opts_search = NULL;
346 	}
347 	return nd_opt;
348 }
349 
350 /*
351  * Parse multiple ND options.
352  * This function is much easier to use, for ND routines that do not need
353  * multiple options of the same type.
354  */
355 int
356 nd6_options(union nd_opts *ndopts)
357 {
358 	INIT_VNET_INET6(curvnet);
359 	struct nd_opt_hdr *nd_opt;
360 	int i = 0;
361 
362 	if (ndopts == NULL)
363 		panic("ndopts == NULL in nd6_options");
364 	if (ndopts->nd_opts_last == NULL)
365 		panic("uninitialized ndopts in nd6_options");
366 	if (ndopts->nd_opts_search == NULL)
367 		return 0;
368 
369 	while (1) {
370 		nd_opt = nd6_option(ndopts);
371 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
372 			/*
373 			 * Message validation requires that all included
374 			 * options have a length that is greater than zero.
375 			 */
376 			ICMP6STAT_INC(icp6s_nd_badopt);
377 			bzero(ndopts, sizeof(*ndopts));
378 			return -1;
379 		}
380 
381 		if (nd_opt == NULL)
382 			goto skip1;
383 
384 		switch (nd_opt->nd_opt_type) {
385 		case ND_OPT_SOURCE_LINKADDR:
386 		case ND_OPT_TARGET_LINKADDR:
387 		case ND_OPT_MTU:
388 		case ND_OPT_REDIRECTED_HEADER:
389 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
390 				nd6log((LOG_INFO,
391 				    "duplicated ND6 option found (type=%d)\n",
392 				    nd_opt->nd_opt_type));
393 				/* XXX bark? */
394 			} else {
395 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
396 					= nd_opt;
397 			}
398 			break;
399 		case ND_OPT_PREFIX_INFORMATION:
400 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
401 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
402 					= nd_opt;
403 			}
404 			ndopts->nd_opts_pi_end =
405 				(struct nd_opt_prefix_info *)nd_opt;
406 			break;
407 		default:
408 			/*
409 			 * Unknown options must be silently ignored,
410 			 * to accomodate future extension to the protocol.
411 			 */
412 			nd6log((LOG_DEBUG,
413 			    "nd6_options: unsupported option %d - "
414 			    "option ignored\n", nd_opt->nd_opt_type));
415 		}
416 
417 skip1:
418 		i++;
419 		if (i > V_nd6_maxndopt) {
420 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
421 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
422 			break;
423 		}
424 
425 		if (ndopts->nd_opts_done)
426 			break;
427 	}
428 
429 	return 0;
430 }
431 
432 /*
433  * ND6 timer routine to handle ND6 entries
434  */
435 void
436 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
437 {
438 	if (tick < 0) {
439 		ln->la_expire = 0;
440 		ln->ln_ntick = 0;
441 		callout_stop(&ln->ln_timer_ch);
442 		/*
443 		 * XXX - do we know that there is
444 		 * callout installed? i.e. are we
445 		 * guaranteed that we're not dropping
446 		 * a reference that we did not add?
447 		 * KMM
448 		 */
449 		LLE_REMREF(ln);
450 	} else {
451 		ln->la_expire = time_second + tick / hz;
452 		LLE_ADDREF(ln);
453 		if (tick > INT_MAX) {
454 			ln->ln_ntick = tick - INT_MAX;
455 			callout_reset(&ln->ln_timer_ch, INT_MAX,
456 			    nd6_llinfo_timer, ln);
457 		} else {
458 			ln->ln_ntick = 0;
459 			callout_reset(&ln->ln_timer_ch, tick,
460 			    nd6_llinfo_timer, ln);
461 		}
462 	}
463 }
464 
465 void
466 nd6_llinfo_settimer(struct llentry *ln, long tick)
467 {
468 
469 	LLE_WLOCK(ln);
470 	nd6_llinfo_settimer_locked(ln, tick);
471 	LLE_WUNLOCK(ln);
472 }
473 
474 static void
475 nd6_llinfo_timer(void *arg)
476 {
477 	struct llentry *ln;
478 	struct in6_addr *dst;
479 	struct ifnet *ifp;
480 	struct nd_ifinfo *ndi = NULL;
481 
482 	ln = (struct llentry *)arg;
483 	if (ln == NULL) {
484 		panic("%s: NULL entry!\n", __func__);
485 		return;
486 	}
487 
488 	if ((ifp = ((ln->lle_tbl != NULL) ? ln->lle_tbl->llt_ifp : NULL)) == NULL)
489 		panic("ln ifp == NULL");
490 
491 	CURVNET_SET(ifp->if_vnet);
492 	INIT_VNET_INET6(curvnet);
493 
494 	if (ln->ln_ntick > 0) {
495 		if (ln->ln_ntick > INT_MAX) {
496 			ln->ln_ntick -= INT_MAX;
497 			nd6_llinfo_settimer(ln, INT_MAX);
498 		} else {
499 			ln->ln_ntick = 0;
500 			nd6_llinfo_settimer(ln, ln->ln_ntick);
501 		}
502 		goto done;
503 	}
504 
505 	ndi = ND_IFINFO(ifp);
506 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
507 	if ((ln->la_flags & LLE_STATIC) || (ln->la_expire > time_second)) {
508 		goto done;
509 	}
510 
511 	if (ln->la_flags & LLE_DELETED) {
512 		(void)nd6_free(ln, 0);
513 		goto done;
514 	}
515 
516 	switch (ln->ln_state) {
517 	case ND6_LLINFO_INCOMPLETE:
518 		if (ln->la_asked < V_nd6_mmaxtries) {
519 			ln->la_asked++;
520 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
521 			nd6_ns_output(ifp, NULL, dst, ln, 0);
522 		} else {
523 			struct mbuf *m = ln->la_hold;
524 			if (m) {
525 				struct mbuf *m0;
526 
527 				/*
528 				 * assuming every packet in la_hold has the
529 				 * same IP header
530 				 */
531 				m0 = m->m_nextpkt;
532 				m->m_nextpkt = NULL;
533 				icmp6_error2(m, ICMP6_DST_UNREACH,
534 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
535 
536 				ln->la_hold = m0;
537 				clear_llinfo_pqueue(ln);
538 			}
539 			(void)nd6_free(ln, 0);
540 			ln = NULL;
541 		}
542 		break;
543 	case ND6_LLINFO_REACHABLE:
544 		if (!ND6_LLINFO_PERMANENT(ln)) {
545 			ln->ln_state = ND6_LLINFO_STALE;
546 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
547 		}
548 		break;
549 
550 	case ND6_LLINFO_STALE:
551 		/* Garbage Collection(RFC 2461 5.3) */
552 		if (!ND6_LLINFO_PERMANENT(ln)) {
553 			(void)nd6_free(ln, 1);
554 			ln = NULL;
555 		}
556 		break;
557 
558 	case ND6_LLINFO_DELAY:
559 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
560 			/* We need NUD */
561 			ln->la_asked = 1;
562 			ln->ln_state = ND6_LLINFO_PROBE;
563 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
564 			nd6_ns_output(ifp, dst, dst, ln, 0);
565 		} else {
566 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
567 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
568 		}
569 		break;
570 	case ND6_LLINFO_PROBE:
571 		if (ln->la_asked < V_nd6_umaxtries) {
572 			ln->la_asked++;
573 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
574 			nd6_ns_output(ifp, dst, dst, ln, 0);
575 		} else {
576 			(void)nd6_free(ln, 0);
577 			ln = NULL;
578 		}
579 		break;
580 	}
581 	CURVNET_RESTORE();
582 done:
583 	if (ln != NULL)
584 		LLE_FREE(ln);
585 }
586 
587 
588 /*
589  * ND6 timer routine to expire default route list and prefix list
590  */
591 void
592 nd6_timer(void *arg)
593 {
594 	CURVNET_SET((struct vnet *) arg);
595 	INIT_VNET_INET6(curvnet);
596 	int s;
597 	struct nd_defrouter *dr;
598 	struct nd_prefix *pr;
599 	struct in6_ifaddr *ia6, *nia6;
600 	struct in6_addrlifetime *lt6;
601 
602 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
603 	    nd6_timer, curvnet);
604 
605 	/* expire default router list */
606 	s = splnet();
607 	dr = TAILQ_FIRST(&V_nd_defrouter);
608 	while (dr) {
609 		if (dr->expire && dr->expire < time_second) {
610 			struct nd_defrouter *t;
611 			t = TAILQ_NEXT(dr, dr_entry);
612 			defrtrlist_del(dr);
613 			dr = t;
614 		} else {
615 			dr = TAILQ_NEXT(dr, dr_entry);
616 		}
617 	}
618 
619 	/*
620 	 * expire interface addresses.
621 	 * in the past the loop was inside prefix expiry processing.
622 	 * However, from a stricter speci-confrmance standpoint, we should
623 	 * rather separate address lifetimes and prefix lifetimes.
624 	 */
625   addrloop:
626 	for (ia6 = V_in6_ifaddr; ia6; ia6 = nia6) {
627 		nia6 = ia6->ia_next;
628 		/* check address lifetime */
629 		lt6 = &ia6->ia6_lifetime;
630 		if (IFA6_IS_INVALID(ia6)) {
631 			int regen = 0;
632 
633 			/*
634 			 * If the expiring address is temporary, try
635 			 * regenerating a new one.  This would be useful when
636 			 * we suspended a laptop PC, then turned it on after a
637 			 * period that could invalidate all temporary
638 			 * addresses.  Although we may have to restart the
639 			 * loop (see below), it must be after purging the
640 			 * address.  Otherwise, we'd see an infinite loop of
641 			 * regeneration.
642 			 */
643 			if (V_ip6_use_tempaddr &&
644 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
645 				if (regen_tmpaddr(ia6) == 0)
646 					regen = 1;
647 			}
648 
649 			in6_purgeaddr(&ia6->ia_ifa);
650 
651 			if (regen)
652 				goto addrloop; /* XXX: see below */
653 		} else if (IFA6_IS_DEPRECATED(ia6)) {
654 			int oldflags = ia6->ia6_flags;
655 
656 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
657 
658 			/*
659 			 * If a temporary address has just become deprecated,
660 			 * regenerate a new one if possible.
661 			 */
662 			if (V_ip6_use_tempaddr &&
663 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
664 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
665 
666 				if (regen_tmpaddr(ia6) == 0) {
667 					/*
668 					 * A new temporary address is
669 					 * generated.
670 					 * XXX: this means the address chain
671 					 * has changed while we are still in
672 					 * the loop.  Although the change
673 					 * would not cause disaster (because
674 					 * it's not a deletion, but an
675 					 * addition,) we'd rather restart the
676 					 * loop just for safety.  Or does this
677 					 * significantly reduce performance??
678 					 */
679 					goto addrloop;
680 				}
681 			}
682 		} else {
683 			/*
684 			 * A new RA might have made a deprecated address
685 			 * preferred.
686 			 */
687 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
688 		}
689 	}
690 
691 	/* expire prefix list */
692 	pr = V_nd_prefix.lh_first;
693 	while (pr) {
694 		/*
695 		 * check prefix lifetime.
696 		 * since pltime is just for autoconf, pltime processing for
697 		 * prefix is not necessary.
698 		 */
699 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
700 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
701 			struct nd_prefix *t;
702 			t = pr->ndpr_next;
703 
704 			/*
705 			 * address expiration and prefix expiration are
706 			 * separate.  NEVER perform in6_purgeaddr here.
707 			 */
708 
709 			prelist_remove(pr);
710 			pr = t;
711 		} else
712 			pr = pr->ndpr_next;
713 	}
714 	splx(s);
715 	CURVNET_RESTORE();
716 }
717 
718 /*
719  * ia6 - deprecated/invalidated temporary address
720  */
721 static int
722 regen_tmpaddr(struct in6_ifaddr *ia6)
723 {
724 	struct ifaddr *ifa;
725 	struct ifnet *ifp;
726 	struct in6_ifaddr *public_ifa6 = NULL;
727 
728 	ifp = ia6->ia_ifa.ifa_ifp;
729 	IF_ADDR_LOCK(ifp);
730 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
731 		struct in6_ifaddr *it6;
732 
733 		if (ifa->ifa_addr->sa_family != AF_INET6)
734 			continue;
735 
736 		it6 = (struct in6_ifaddr *)ifa;
737 
738 		/* ignore no autoconf addresses. */
739 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
740 			continue;
741 
742 		/* ignore autoconf addresses with different prefixes. */
743 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
744 			continue;
745 
746 		/*
747 		 * Now we are looking at an autoconf address with the same
748 		 * prefix as ours.  If the address is temporary and is still
749 		 * preferred, do not create another one.  It would be rare, but
750 		 * could happen, for example, when we resume a laptop PC after
751 		 * a long period.
752 		 */
753 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
754 		    !IFA6_IS_DEPRECATED(it6)) {
755 			public_ifa6 = NULL;
756 			break;
757 		}
758 
759 		/*
760 		 * This is a public autoconf address that has the same prefix
761 		 * as ours.  If it is preferred, keep it.  We can't break the
762 		 * loop here, because there may be a still-preferred temporary
763 		 * address with the prefix.
764 		 */
765 		if (!IFA6_IS_DEPRECATED(it6))
766 		    public_ifa6 = it6;
767 	}
768 
769 	if (public_ifa6 != NULL) {
770 		int e;
771 
772 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
773 			IF_ADDR_UNLOCK(ifp);
774 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
775 			    " tmp addr,errno=%d\n", e);
776 			return (-1);
777 		}
778 		IF_ADDR_UNLOCK(ifp);
779 		return (0);
780 	}
781 
782 	IF_ADDR_UNLOCK(ifp);
783 	return (-1);
784 }
785 
786 /*
787  * Nuke neighbor cache/prefix/default router management table, right before
788  * ifp goes away.
789  */
790 void
791 nd6_purge(struct ifnet *ifp)
792 {
793 	INIT_VNET_INET6(ifp->if_vnet);
794 	struct nd_defrouter *dr, *ndr;
795 	struct nd_prefix *pr, *npr;
796 
797 	/*
798 	 * Nuke default router list entries toward ifp.
799 	 * We defer removal of default router list entries that is installed
800 	 * in the routing table, in order to keep additional side effects as
801 	 * small as possible.
802 	 */
803 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
804 		ndr = TAILQ_NEXT(dr, dr_entry);
805 		if (dr->installed)
806 			continue;
807 
808 		if (dr->ifp == ifp)
809 			defrtrlist_del(dr);
810 	}
811 
812 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
813 		ndr = TAILQ_NEXT(dr, dr_entry);
814 		if (!dr->installed)
815 			continue;
816 
817 		if (dr->ifp == ifp)
818 			defrtrlist_del(dr);
819 	}
820 
821 	/* Nuke prefix list entries toward ifp */
822 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
823 		npr = pr->ndpr_next;
824 		if (pr->ndpr_ifp == ifp) {
825 			/*
826 			 * Because if_detach() does *not* release prefixes
827 			 * while purging addresses the reference count will
828 			 * still be above zero. We therefore reset it to
829 			 * make sure that the prefix really gets purged.
830 			 */
831 			pr->ndpr_refcnt = 0;
832 
833 			/*
834 			 * Previously, pr->ndpr_addr is removed as well,
835 			 * but I strongly believe we don't have to do it.
836 			 * nd6_purge() is only called from in6_ifdetach(),
837 			 * which removes all the associated interface addresses
838 			 * by itself.
839 			 * (jinmei@kame.net 20010129)
840 			 */
841 			prelist_remove(pr);
842 		}
843 	}
844 
845 	/* cancel default outgoing interface setting */
846 	if (V_nd6_defifindex == ifp->if_index)
847 		nd6_setdefaultiface(0);
848 
849 	if (!V_ip6_forwarding && V_ip6_accept_rtadv) { /* XXX: too restrictive? */
850 		/* refresh default router list
851 		 *
852 		 *
853 		 */
854 		defrouter_select();
855 
856 	}
857 
858 	/* XXXXX
859 	 * We do not nuke the neighbor cache entries here any more
860 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
861 	 * nd6_purge() is invoked by in6_ifdetach() which is called
862 	 * from if_detach() where everything gets purged. So let
863 	 * in6_domifdetach() do the actual L2 table purging work.
864 	 */
865 }
866 
867 /*
868  * the caller acquires and releases the lock on the lltbls
869  * Returns the llentry locked
870  */
871 struct llentry *
872 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
873 {
874 	struct sockaddr_in6 sin6;
875 	struct llentry *ln;
876 	int llflags = 0;
877 
878 	bzero(&sin6, sizeof(sin6));
879 	sin6.sin6_len = sizeof(struct sockaddr_in6);
880 	sin6.sin6_family = AF_INET6;
881 	sin6.sin6_addr = *addr6;
882 
883 	IF_AFDATA_LOCK_ASSERT(ifp);
884 
885 	if (flags & ND6_CREATE)
886 	    llflags |= LLE_CREATE;
887 	if (flags & ND6_EXCLUSIVE)
888 	    llflags |= LLE_EXCLUSIVE;
889 
890 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
891 	if ((ln != NULL) && (flags & LLE_CREATE)) {
892 		ln->ln_state = ND6_LLINFO_NOSTATE;
893 		callout_init(&ln->ln_timer_ch, 0);
894 	}
895 
896 	return (ln);
897 }
898 
899 /*
900  * Test whether a given IPv6 address is a neighbor or not, ignoring
901  * the actual neighbor cache.  The neighbor cache is ignored in order
902  * to not reenter the routing code from within itself.
903  */
904 static int
905 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
906 {
907 	INIT_VNET_INET6(ifp->if_vnet);
908 	struct nd_prefix *pr;
909 	struct ifaddr *dstaddr;
910 
911 	/*
912 	 * A link-local address is always a neighbor.
913 	 * XXX: a link does not necessarily specify a single interface.
914 	 */
915 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
916 		struct sockaddr_in6 sin6_copy;
917 		u_int32_t zone;
918 
919 		/*
920 		 * We need sin6_copy since sa6_recoverscope() may modify the
921 		 * content (XXX).
922 		 */
923 		sin6_copy = *addr;
924 		if (sa6_recoverscope(&sin6_copy))
925 			return (0); /* XXX: should be impossible */
926 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
927 			return (0);
928 		if (sin6_copy.sin6_scope_id == zone)
929 			return (1);
930 		else
931 			return (0);
932 	}
933 
934 	/*
935 	 * If the address matches one of our addresses,
936 	 * it should be a neighbor.
937 	 * If the address matches one of our on-link prefixes, it should be a
938 	 * neighbor.
939 	 */
940 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
941 		if (pr->ndpr_ifp != ifp)
942 			continue;
943 
944 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
945 			continue;
946 
947 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
948 		    &addr->sin6_addr, &pr->ndpr_mask))
949 			return (1);
950 	}
951 
952 	/*
953 	 * If the address is assigned on the node of the other side of
954 	 * a p2p interface, the address should be a neighbor.
955 	 */
956 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
957 	if ((dstaddr != NULL) && (dstaddr->ifa_ifp == ifp))
958 		return (1);
959 
960 	/*
961 	 * If the default router list is empty, all addresses are regarded
962 	 * as on-link, and thus, as a neighbor.
963 	 * XXX: we restrict the condition to hosts, because routers usually do
964 	 * not have the "default router list".
965 	 */
966 	if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
967 	    V_nd6_defifindex == ifp->if_index) {
968 		return (1);
969 	}
970 
971 	return (0);
972 }
973 
974 
975 /*
976  * Detect if a given IPv6 address identifies a neighbor on a given link.
977  * XXX: should take care of the destination of a p2p link?
978  */
979 int
980 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
981 {
982 	struct llentry *lle;
983 	int rc = 0;
984 
985 	IF_AFDATA_UNLOCK_ASSERT(ifp);
986 	if (nd6_is_new_addr_neighbor(addr, ifp))
987 		return (1);
988 
989 	/*
990 	 * Even if the address matches none of our addresses, it might be
991 	 * in the neighbor cache.
992 	 */
993 	IF_AFDATA_LOCK(ifp);
994 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
995 		LLE_RUNLOCK(lle);
996 		rc = 1;
997 	}
998 	IF_AFDATA_UNLOCK(ifp);
999 	return (rc);
1000 }
1001 
1002 /*
1003  * Free an nd6 llinfo entry.
1004  * Since the function would cause significant changes in the kernel, DO NOT
1005  * make it global, unless you have a strong reason for the change, and are sure
1006  * that the change is safe.
1007  */
1008 static struct llentry *
1009 nd6_free(struct llentry *ln, int gc)
1010 {
1011 	INIT_VNET_INET6(curvnet);
1012         struct llentry *next;
1013 	struct nd_defrouter *dr;
1014 	struct ifnet *ifp=NULL;
1015 
1016 	/*
1017 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1018 	 * even though it is not harmful, it was not really necessary.
1019 	 */
1020 
1021 	/* cancel timer */
1022 	nd6_llinfo_settimer(ln, -1);
1023 
1024 	if (!V_ip6_forwarding) {
1025 		int s;
1026 		s = splnet();
1027 		dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1028 
1029 		if (dr != NULL && dr->expire &&
1030 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1031 			/*
1032 			 * If the reason for the deletion is just garbage
1033 			 * collection, and the neighbor is an active default
1034 			 * router, do not delete it.  Instead, reset the GC
1035 			 * timer using the router's lifetime.
1036 			 * Simply deleting the entry would affect default
1037 			 * router selection, which is not necessarily a good
1038 			 * thing, especially when we're using router preference
1039 			 * values.
1040 			 * XXX: the check for ln_state would be redundant,
1041 			 *      but we intentionally keep it just in case.
1042 			 */
1043 			if (dr->expire > time_second)
1044 				nd6_llinfo_settimer(ln,
1045 				    (dr->expire - time_second) * hz);
1046 			else
1047 				nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
1048 			splx(s);
1049 			return (LIST_NEXT(ln, lle_next));
1050 		}
1051 
1052 		if (ln->ln_router || dr) {
1053 			/*
1054 			 * rt6_flush must be called whether or not the neighbor
1055 			 * is in the Default Router List.
1056 			 * See a corresponding comment in nd6_na_input().
1057 			 */
1058 			rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ln->lle_tbl->llt_ifp);
1059 		}
1060 
1061 		if (dr) {
1062 			/*
1063 			 * Unreachablity of a router might affect the default
1064 			 * router selection and on-link detection of advertised
1065 			 * prefixes.
1066 			 */
1067 
1068 			/*
1069 			 * Temporarily fake the state to choose a new default
1070 			 * router and to perform on-link determination of
1071 			 * prefixes correctly.
1072 			 * Below the state will be set correctly,
1073 			 * or the entry itself will be deleted.
1074 			 */
1075 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1076 
1077 			/*
1078 			 * Since defrouter_select() does not affect the
1079 			 * on-link determination and MIP6 needs the check
1080 			 * before the default router selection, we perform
1081 			 * the check now.
1082 			 */
1083 			pfxlist_onlink_check();
1084 
1085 			/*
1086 			 * refresh default router list
1087 			 */
1088 			defrouter_select();
1089 		}
1090 		splx(s);
1091 	}
1092 
1093 	/*
1094 	 * Before deleting the entry, remember the next entry as the
1095 	 * return value.  We need this because pfxlist_onlink_check() above
1096 	 * might have freed other entries (particularly the old next entry) as
1097 	 * a side effect (XXX).
1098 	 */
1099 	next = LIST_NEXT(ln, lle_next);
1100 
1101 	ifp = ln->lle_tbl->llt_ifp;
1102 	IF_AFDATA_LOCK(ifp);
1103 	LLE_WLOCK(ln);
1104 	llentry_free(ln);
1105 	IF_AFDATA_UNLOCK(ifp);
1106 
1107 	return (next);
1108 }
1109 
1110 /*
1111  * Upper-layer reachability hint for Neighbor Unreachability Detection.
1112  *
1113  * XXX cost-effective methods?
1114  */
1115 void
1116 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force)
1117 {
1118 	INIT_VNET_INET6(curvnet);
1119 	struct llentry *ln;
1120 	struct ifnet *ifp;
1121 
1122 	if ((dst6 == NULL) || (rt == NULL))
1123 		return;
1124 
1125 	ifp = rt->rt_ifp;
1126 	IF_AFDATA_LOCK(ifp);
1127 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1128 	IF_AFDATA_UNLOCK(ifp);
1129 	if (ln == NULL)
1130 		return;
1131 
1132 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1133 		goto done;
1134 
1135 	/*
1136 	 * if we get upper-layer reachability confirmation many times,
1137 	 * it is possible we have false information.
1138 	 */
1139 	if (!force) {
1140 		ln->ln_byhint++;
1141 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1142 			goto done;
1143 		}
1144 	}
1145 
1146  	ln->ln_state = ND6_LLINFO_REACHABLE;
1147 	if (!ND6_LLINFO_PERMANENT(ln)) {
1148 		nd6_llinfo_settimer(ln,
1149 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1150 	}
1151 done:
1152 	LLE_WUNLOCK(ln);
1153 }
1154 
1155 
1156 int
1157 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1158 {
1159 	INIT_VNET_INET6(ifp->if_vnet);
1160 	struct in6_drlist *drl = (struct in6_drlist *)data;
1161 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1162 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1163 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1164 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1165 	struct nd_defrouter *dr;
1166 	struct nd_prefix *pr;
1167 	int i = 0, error = 0;
1168 	int s;
1169 
1170 	switch (cmd) {
1171 	case SIOCGDRLST_IN6:
1172 		/*
1173 		 * obsolete API, use sysctl under net.inet6.icmp6
1174 		 */
1175 		bzero(drl, sizeof(*drl));
1176 		s = splnet();
1177 		dr = TAILQ_FIRST(&V_nd_defrouter);
1178 		while (dr && i < DRLSTSIZ) {
1179 			drl->defrouter[i].rtaddr = dr->rtaddr;
1180 			in6_clearscope(&drl->defrouter[i].rtaddr);
1181 
1182 			drl->defrouter[i].flags = dr->flags;
1183 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1184 			drl->defrouter[i].expire = dr->expire;
1185 			drl->defrouter[i].if_index = dr->ifp->if_index;
1186 			i++;
1187 			dr = TAILQ_NEXT(dr, dr_entry);
1188 		}
1189 		splx(s);
1190 		break;
1191 	case SIOCGPRLST_IN6:
1192 		/*
1193 		 * obsolete API, use sysctl under net.inet6.icmp6
1194 		 *
1195 		 * XXX the structure in6_prlist was changed in backward-
1196 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1197 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1198 		 */
1199 		/*
1200 		 * XXX meaning of fields, especialy "raflags", is very
1201 		 * differnet between RA prefix list and RR/static prefix list.
1202 		 * how about separating ioctls into two?
1203 		 */
1204 		bzero(oprl, sizeof(*oprl));
1205 		s = splnet();
1206 		pr = V_nd_prefix.lh_first;
1207 		while (pr && i < PRLSTSIZ) {
1208 			struct nd_pfxrouter *pfr;
1209 			int j;
1210 
1211 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1212 			oprl->prefix[i].raflags = pr->ndpr_raf;
1213 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1214 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1215 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1216 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1217 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1218 				oprl->prefix[i].expire = 0;
1219 			else {
1220 				time_t maxexpire;
1221 
1222 				/* XXX: we assume time_t is signed. */
1223 				maxexpire = (-1) &
1224 				    ~((time_t)1 <<
1225 				    ((sizeof(maxexpire) * 8) - 1));
1226 				if (pr->ndpr_vltime <
1227 				    maxexpire - pr->ndpr_lastupdate) {
1228 					oprl->prefix[i].expire =
1229 					    pr->ndpr_lastupdate +
1230 					    pr->ndpr_vltime;
1231 				} else
1232 					oprl->prefix[i].expire = maxexpire;
1233 			}
1234 
1235 			pfr = pr->ndpr_advrtrs.lh_first;
1236 			j = 0;
1237 			while (pfr) {
1238 				if (j < DRLSTSIZ) {
1239 #define RTRADDR oprl->prefix[i].advrtr[j]
1240 					RTRADDR = pfr->router->rtaddr;
1241 					in6_clearscope(&RTRADDR);
1242 #undef RTRADDR
1243 				}
1244 				j++;
1245 				pfr = pfr->pfr_next;
1246 			}
1247 			oprl->prefix[i].advrtrs = j;
1248 			oprl->prefix[i].origin = PR_ORIG_RA;
1249 
1250 			i++;
1251 			pr = pr->ndpr_next;
1252 		}
1253 		splx(s);
1254 
1255 		break;
1256 	case OSIOCGIFINFO_IN6:
1257 #define ND	ndi->ndi
1258 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1259 		bzero(&ND, sizeof(ND));
1260 		ND.linkmtu = IN6_LINKMTU(ifp);
1261 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1262 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1263 		ND.reachable = ND_IFINFO(ifp)->reachable;
1264 		ND.retrans = ND_IFINFO(ifp)->retrans;
1265 		ND.flags = ND_IFINFO(ifp)->flags;
1266 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1267 		ND.chlim = ND_IFINFO(ifp)->chlim;
1268 		break;
1269 	case SIOCGIFINFO_IN6:
1270 		ND = *ND_IFINFO(ifp);
1271 		break;
1272 	case SIOCSIFINFO_IN6:
1273 		/*
1274 		 * used to change host variables from userland.
1275 		 * intented for a use on router to reflect RA configurations.
1276 		 */
1277 		/* 0 means 'unspecified' */
1278 		if (ND.linkmtu != 0) {
1279 			if (ND.linkmtu < IPV6_MMTU ||
1280 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1281 				error = EINVAL;
1282 				break;
1283 			}
1284 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1285 		}
1286 
1287 		if (ND.basereachable != 0) {
1288 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1289 
1290 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1291 			if (ND.basereachable != obasereachable)
1292 				ND_IFINFO(ifp)->reachable =
1293 				    ND_COMPUTE_RTIME(ND.basereachable);
1294 		}
1295 		if (ND.retrans != 0)
1296 			ND_IFINFO(ifp)->retrans = ND.retrans;
1297 		if (ND.chlim != 0)
1298 			ND_IFINFO(ifp)->chlim = ND.chlim;
1299 		/* FALLTHROUGH */
1300 	case SIOCSIFINFO_FLAGS:
1301 		ND_IFINFO(ifp)->flags = ND.flags;
1302 		break;
1303 #undef ND
1304 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1305 		/* sync kernel routing table with the default router list */
1306 		defrouter_reset();
1307 		defrouter_select();
1308 		break;
1309 	case SIOCSPFXFLUSH_IN6:
1310 	{
1311 		/* flush all the prefix advertised by routers */
1312 		struct nd_prefix *pr, *next;
1313 
1314 		s = splnet();
1315 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1316 			struct in6_ifaddr *ia, *ia_next;
1317 
1318 			next = pr->ndpr_next;
1319 
1320 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1321 				continue; /* XXX */
1322 
1323 			/* do we really have to remove addresses as well? */
1324 			for (ia = V_in6_ifaddr; ia; ia = ia_next) {
1325 				/* ia might be removed.  keep the next ptr. */
1326 				ia_next = ia->ia_next;
1327 
1328 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1329 					continue;
1330 
1331 				if (ia->ia6_ndpr == pr)
1332 					in6_purgeaddr(&ia->ia_ifa);
1333 			}
1334 			prelist_remove(pr);
1335 		}
1336 		splx(s);
1337 		break;
1338 	}
1339 	case SIOCSRTRFLUSH_IN6:
1340 	{
1341 		/* flush all the default routers */
1342 		struct nd_defrouter *dr, *next;
1343 
1344 		s = splnet();
1345 		defrouter_reset();
1346 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1347 			next = TAILQ_NEXT(dr, dr_entry);
1348 			defrtrlist_del(dr);
1349 		}
1350 		defrouter_select();
1351 		splx(s);
1352 		break;
1353 	}
1354 	case SIOCGNBRINFO_IN6:
1355 	{
1356 		struct llentry *ln;
1357 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1358 
1359 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1360 			return (error);
1361 
1362 		IF_AFDATA_LOCK(ifp);
1363 		ln = nd6_lookup(&nb_addr, 0, ifp);
1364 		IF_AFDATA_UNLOCK(ifp);
1365 
1366 		if (ln == NULL) {
1367 			error = EINVAL;
1368 			break;
1369 		}
1370 		nbi->state = ln->ln_state;
1371 		nbi->asked = ln->la_asked;
1372 		nbi->isrouter = ln->ln_router;
1373 		nbi->expire = ln->la_expire;
1374 		LLE_RUNLOCK(ln);
1375 		break;
1376 	}
1377 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1378 		ndif->ifindex = V_nd6_defifindex;
1379 		break;
1380 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1381 		return (nd6_setdefaultiface(ndif->ifindex));
1382 	}
1383 	return (error);
1384 }
1385 
1386 /*
1387  * Create neighbor cache entry and cache link-layer address,
1388  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1389  *
1390  * type - ICMP6 type
1391  * code - type dependent information
1392  *
1393  * XXXXX
1394  *  The caller of this function already acquired the ndp
1395  *  cache table lock because the cache entry is returned.
1396  */
1397 struct llentry *
1398 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1399     int lladdrlen, int type, int code)
1400 {
1401 	INIT_VNET_INET6(curvnet);
1402 	struct llentry *ln = NULL;
1403 	int is_newentry;
1404 	int do_update;
1405 	int olladdr;
1406 	int llchange;
1407 	int flags = 0;
1408 	int newstate = 0;
1409 	uint16_t router = 0;
1410 	struct sockaddr_in6 sin6;
1411 	struct mbuf *chain = NULL;
1412 	int static_route = 0;
1413 
1414 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1415 
1416 	if (ifp == NULL)
1417 		panic("ifp == NULL in nd6_cache_lladdr");
1418 	if (from == NULL)
1419 		panic("from == NULL in nd6_cache_lladdr");
1420 
1421 	/* nothing must be updated for unspecified address */
1422 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1423 		return NULL;
1424 
1425 	/*
1426 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1427 	 * the caller.
1428 	 *
1429 	 * XXX If the link does not have link-layer adderss, what should
1430 	 * we do? (ifp->if_addrlen == 0)
1431 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1432 	 * description on it in NS section (RFC 2461 7.2.3).
1433 	 */
1434 	flags |= lladdr ? ND6_EXCLUSIVE : 0;
1435 	IF_AFDATA_LOCK(ifp);
1436 	ln = nd6_lookup(from, flags, ifp);
1437 
1438 	if (ln == NULL) {
1439 		flags |= LLE_EXCLUSIVE;
1440 		ln = nd6_lookup(from, flags |ND6_CREATE, ifp);
1441 		IF_AFDATA_UNLOCK(ifp);
1442 		is_newentry = 1;
1443 	} else {
1444 		IF_AFDATA_UNLOCK(ifp);
1445 		/* do nothing if static ndp is set */
1446 		if (ln->la_flags & LLE_STATIC) {
1447 			static_route = 1;
1448 			goto done;
1449 		}
1450 		is_newentry = 0;
1451 	}
1452 	if (ln == NULL)
1453 		return (NULL);
1454 
1455 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1456 	if (olladdr && lladdr) {
1457 		llchange = bcmp(lladdr, &ln->ll_addr,
1458 		    ifp->if_addrlen);
1459 	} else
1460 		llchange = 0;
1461 
1462 	/*
1463 	 * newentry olladdr  lladdr  llchange	(*=record)
1464 	 *	0	n	n	--	(1)
1465 	 *	0	y	n	--	(2)
1466 	 *	0	n	y	--	(3) * STALE
1467 	 *	0	y	y	n	(4) *
1468 	 *	0	y	y	y	(5) * STALE
1469 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1470 	 *	1	--	y	--	(7) * STALE
1471 	 */
1472 
1473 	if (lladdr) {		/* (3-5) and (7) */
1474 		/*
1475 		 * Record source link-layer address
1476 		 * XXX is it dependent to ifp->if_type?
1477 		 */
1478 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1479 		ln->la_flags |= LLE_VALID;
1480 	}
1481 
1482 	if (!is_newentry) {
1483 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1484 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1485 			do_update = 1;
1486 			newstate = ND6_LLINFO_STALE;
1487 		} else					/* (1-2,4) */
1488 			do_update = 0;
1489 	} else {
1490 		do_update = 1;
1491 		if (lladdr == NULL)			/* (6) */
1492 			newstate = ND6_LLINFO_NOSTATE;
1493 		else					/* (7) */
1494 			newstate = ND6_LLINFO_STALE;
1495 	}
1496 
1497 	if (do_update) {
1498 		/*
1499 		 * Update the state of the neighbor cache.
1500 		 */
1501 		ln->ln_state = newstate;
1502 
1503 		if (ln->ln_state == ND6_LLINFO_STALE) {
1504 			/*
1505 			 * XXX: since nd6_output() below will cause
1506 			 * state tansition to DELAY and reset the timer,
1507 			 * we must set the timer now, although it is actually
1508 			 * meaningless.
1509 			 */
1510 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1511 
1512 			if (ln->la_hold) {
1513 				struct mbuf *m_hold, *m_hold_next;
1514 
1515 				/*
1516 				 * reset the la_hold in advance, to explicitly
1517 				 * prevent a la_hold lookup in nd6_output()
1518 				 * (wouldn't happen, though...)
1519 				 */
1520 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1521 				    m_hold; m_hold = m_hold_next) {
1522 					m_hold_next = m_hold->m_nextpkt;
1523 					m_hold->m_nextpkt = NULL;
1524 
1525 					/*
1526 					 * we assume ifp is not a p2p here, so
1527 					 * just set the 2nd argument as the
1528 					 * 1st one.
1529 					 */
1530 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1531 				}
1532 				/*
1533 				 * If we have mbufs in the chain we need to do
1534 				 * deferred transmit. Copy the address from the
1535 				 * llentry before dropping the lock down below.
1536 				 */
1537 				if (chain != NULL)
1538 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1539 			}
1540 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1541 			/* probe right away */
1542 			nd6_llinfo_settimer_locked((void *)ln, 0);
1543 		}
1544 	}
1545 
1546 	/*
1547 	 * ICMP6 type dependent behavior.
1548 	 *
1549 	 * NS: clear IsRouter if new entry
1550 	 * RS: clear IsRouter
1551 	 * RA: set IsRouter if there's lladdr
1552 	 * redir: clear IsRouter if new entry
1553 	 *
1554 	 * RA case, (1):
1555 	 * The spec says that we must set IsRouter in the following cases:
1556 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1557 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1558 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1559 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1560 	 * neighbor cache, this is similar to (6).
1561 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1562 	 *
1563 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1564 	 *							D R
1565 	 *	0	n	n	--	(1)	c   ?     s
1566 	 *	0	y	n	--	(2)	c   s     s
1567 	 *	0	n	y	--	(3)	c   s     s
1568 	 *	0	y	y	n	(4)	c   s     s
1569 	 *	0	y	y	y	(5)	c   s     s
1570 	 *	1	--	n	--	(6) c	c	c s
1571 	 *	1	--	y	--	(7) c	c   s	c s
1572 	 *
1573 	 *					(c=clear s=set)
1574 	 */
1575 	switch (type & 0xff) {
1576 	case ND_NEIGHBOR_SOLICIT:
1577 		/*
1578 		 * New entry must have is_router flag cleared.
1579 		 */
1580 		if (is_newentry)	/* (6-7) */
1581 			ln->ln_router = 0;
1582 		break;
1583 	case ND_REDIRECT:
1584 		/*
1585 		 * If the icmp is a redirect to a better router, always set the
1586 		 * is_router flag.  Otherwise, if the entry is newly created,
1587 		 * clear the flag.  [RFC 2461, sec 8.3]
1588 		 */
1589 		if (code == ND_REDIRECT_ROUTER)
1590 			ln->ln_router = 1;
1591 		else if (is_newentry) /* (6-7) */
1592 			ln->ln_router = 0;
1593 		break;
1594 	case ND_ROUTER_SOLICIT:
1595 		/*
1596 		 * is_router flag must always be cleared.
1597 		 */
1598 		ln->ln_router = 0;
1599 		break;
1600 	case ND_ROUTER_ADVERT:
1601 		/*
1602 		 * Mark an entry with lladdr as a router.
1603 		 */
1604 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1605 		    (is_newentry && lladdr)) {			/* (7) */
1606 			ln->ln_router = 1;
1607 		}
1608 		break;
1609 	}
1610 
1611 	if (ln != NULL) {
1612 		static_route = (ln->la_flags & LLE_STATIC);
1613 		router = ln->ln_router;
1614 
1615 		if (flags & ND6_EXCLUSIVE)
1616 			LLE_WUNLOCK(ln);
1617 		else
1618 			LLE_RUNLOCK(ln);
1619 		if (static_route)
1620 			ln = NULL;
1621 	}
1622 	if (chain)
1623 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1624 
1625 	/*
1626 	 * When the link-layer address of a router changes, select the
1627 	 * best router again.  In particular, when the neighbor entry is newly
1628 	 * created, it might affect the selection policy.
1629 	 * Question: can we restrict the first condition to the "is_newentry"
1630 	 * case?
1631 	 * XXX: when we hear an RA from a new router with the link-layer
1632 	 * address option, defrouter_select() is called twice, since
1633 	 * defrtrlist_update called the function as well.  However, I believe
1634 	 * we can compromise the overhead, since it only happens the first
1635 	 * time.
1636 	 * XXX: although defrouter_select() should not have a bad effect
1637 	 * for those are not autoconfigured hosts, we explicitly avoid such
1638 	 * cases for safety.
1639 	 */
1640 	if (do_update && router && !V_ip6_forwarding && V_ip6_accept_rtadv) {
1641 		/*
1642 		 * guaranteed recursion
1643 		 */
1644 		defrouter_select();
1645 	}
1646 
1647 	return (ln);
1648 done:
1649 	if (ln != NULL) {
1650 		if (flags & ND6_EXCLUSIVE)
1651 			LLE_WUNLOCK(ln);
1652 		else
1653 			LLE_RUNLOCK(ln);
1654 		if (static_route)
1655 			ln = NULL;
1656 	}
1657 	return (ln);
1658 }
1659 
1660 static void
1661 nd6_slowtimo(void *arg)
1662 {
1663 	CURVNET_SET((struct vnet *) arg);
1664 	INIT_VNET_NET((struct vnet *) arg);
1665 	INIT_VNET_INET6((struct vnet *) arg);
1666 	struct nd_ifinfo *nd6if;
1667 	struct ifnet *ifp;
1668 
1669 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1670 	    nd6_slowtimo, curvnet);
1671 	IFNET_RLOCK();
1672 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1673 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1674 		nd6if = ND_IFINFO(ifp);
1675 		if (nd6if->basereachable && /* already initialized */
1676 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1677 			/*
1678 			 * Since reachable time rarely changes by router
1679 			 * advertisements, we SHOULD insure that a new random
1680 			 * value gets recomputed at least once every few hours.
1681 			 * (RFC 2461, 6.3.4)
1682 			 */
1683 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1684 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1685 		}
1686 	}
1687 	IFNET_RUNLOCK();
1688 	CURVNET_RESTORE();
1689 }
1690 
1691 int
1692 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1693     struct sockaddr_in6 *dst, struct rtentry *rt0)
1694 {
1695 
1696 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1697 }
1698 
1699 
1700 /*
1701  * Note that I'm not enforcing any global serialization
1702  * lle state or asked changes here as the logic is too
1703  * complicated to avoid having to always acquire an exclusive
1704  * lock
1705  * KMM
1706  *
1707  */
1708 #define senderr(e) { error = (e); goto bad;}
1709 
1710 int
1711 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1712     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1713 	struct mbuf **chain)
1714 {
1715 	INIT_VNET_INET6(curvnet);
1716 	struct mbuf *m = m0;
1717 	struct llentry *ln = lle;
1718 	int error = 0;
1719 	int flags = 0;
1720 
1721 #ifdef INVARIANTS
1722 	if (lle != NULL) {
1723 
1724 		LLE_WLOCK_ASSERT(lle);
1725 
1726 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1727 	}
1728 #endif
1729 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1730 		goto sendpkt;
1731 
1732 	if (nd6_need_cache(ifp) == 0)
1733 		goto sendpkt;
1734 
1735 	/*
1736 	 * next hop determination.  This routine is derived from ether_output.
1737 	 */
1738 
1739 	/*
1740 	 * Address resolution or Neighbor Unreachability Detection
1741 	 * for the next hop.
1742 	 * At this point, the destination of the packet must be a unicast
1743 	 * or an anycast address(i.e. not a multicast).
1744 	 */
1745 
1746 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1747 	if (ln == NULL) {
1748 	retry:
1749 		IF_AFDATA_LOCK(ifp);
1750 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1751 		IF_AFDATA_UNLOCK(ifp);
1752 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1753 			/*
1754 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1755 			 * the condition below is not very efficient.  But we believe
1756 			 * it is tolerable, because this should be a rare case.
1757 			 */
1758 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1759 			IF_AFDATA_LOCK(ifp);
1760 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1761 			IF_AFDATA_UNLOCK(ifp);
1762 		}
1763 	}
1764 	if (ln == NULL) {
1765 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1766 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1767 			char ip6buf[INET6_ADDRSTRLEN];
1768 			log(LOG_DEBUG,
1769 			    "nd6_output: can't allocate llinfo for %s "
1770 			    "(ln=%p)\n",
1771 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1772 			senderr(EIO);	/* XXX: good error? */
1773 		}
1774 		goto sendpkt;	/* send anyway */
1775 	}
1776 
1777 	/* We don't have to do link-layer address resolution on a p2p link. */
1778 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1779 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1780 		if ((flags & LLE_EXCLUSIVE) == 0) {
1781 			flags |= LLE_EXCLUSIVE;
1782 			goto retry;
1783 		}
1784 		ln->ln_state = ND6_LLINFO_STALE;
1785 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1786 	}
1787 
1788 	/*
1789 	 * The first time we send a packet to a neighbor whose entry is
1790 	 * STALE, we have to change the state to DELAY and a sets a timer to
1791 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1792 	 * neighbor unreachability detection on expiration.
1793 	 * (RFC 2461 7.3.3)
1794 	 */
1795 	if (ln->ln_state == ND6_LLINFO_STALE) {
1796 		if ((flags & LLE_EXCLUSIVE) == 0) {
1797 			flags |= LLE_EXCLUSIVE;
1798 			LLE_RUNLOCK(ln);
1799 			goto retry;
1800 		}
1801 		ln->la_asked = 0;
1802 		ln->ln_state = ND6_LLINFO_DELAY;
1803 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1804 	}
1805 
1806 	/*
1807 	 * If the neighbor cache entry has a state other than INCOMPLETE
1808 	 * (i.e. its link-layer address is already resolved), just
1809 	 * send the packet.
1810 	 */
1811 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1812 		goto sendpkt;
1813 
1814 	/*
1815 	 * There is a neighbor cache entry, but no ethernet address
1816 	 * response yet.  Append this latest packet to the end of the
1817 	 * packet queue in the mbuf, unless the number of the packet
1818 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1819 	 * the oldest packet in the queue will be removed.
1820 	 */
1821 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1822 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1823 
1824 	if ((flags & LLE_EXCLUSIVE) == 0) {
1825 		flags |= LLE_EXCLUSIVE;
1826 		LLE_RUNLOCK(ln);
1827 		goto retry;
1828 	}
1829 	if (ln->la_hold) {
1830 		struct mbuf *m_hold;
1831 		int i;
1832 
1833 		i = 0;
1834 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1835 			i++;
1836 			if (m_hold->m_nextpkt == NULL) {
1837 				m_hold->m_nextpkt = m;
1838 				break;
1839 			}
1840 		}
1841 		while (i >= V_nd6_maxqueuelen) {
1842 			m_hold = ln->la_hold;
1843 			ln->la_hold = ln->la_hold->m_nextpkt;
1844 			m_freem(m_hold);
1845 			i--;
1846 		}
1847 	} else {
1848 		ln->la_hold = m;
1849 	}
1850 	/*
1851 	 * We did the lookup (no lle arg) so we
1852 	 * need to do the unlock here
1853 	 */
1854 	if (lle == NULL) {
1855 		if (flags & LLE_EXCLUSIVE)
1856 			LLE_WUNLOCK(ln);
1857 		else
1858 			LLE_RUNLOCK(ln);
1859 	}
1860 
1861 	/*
1862 	 * If there has been no NS for the neighbor after entering the
1863 	 * INCOMPLETE state, send the first solicitation.
1864 	 */
1865 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1866 		ln->la_asked++;
1867 
1868 		nd6_llinfo_settimer(ln,
1869 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1870 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1871 	}
1872 	return (0);
1873 
1874   sendpkt:
1875 	/* discard the packet if IPv6 operation is disabled on the interface */
1876 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1877 		error = ENETDOWN; /* better error? */
1878 		goto bad;
1879 	}
1880 	/*
1881 	 * ln is valid and the caller did not pass in
1882 	 * an llentry
1883 	 */
1884 	if ((ln != NULL) && (lle == NULL)) {
1885 		if (flags & LLE_EXCLUSIVE)
1886 			LLE_WUNLOCK(ln);
1887 		else
1888 			LLE_RUNLOCK(ln);
1889 	}
1890 
1891 #ifdef MAC
1892 	mac_netinet6_nd6_send(ifp, m);
1893 #endif
1894 	/*
1895 	 * We were passed in a pointer to an lle with the lock held
1896 	 * this means that we can't call if_output as we will
1897 	 * recurse on the lle lock - so what we do is we create
1898 	 * a list of mbufs to send and transmit them in the caller
1899 	 * after the lock is dropped
1900 	 */
1901 	if (lle != NULL) {
1902 		if (*chain == NULL)
1903 			*chain = m;
1904 		else {
1905 			struct mbuf *m = *chain;
1906 
1907 			/*
1908 			 * append mbuf to end of deferred chain
1909 			 */
1910 			while (m->m_nextpkt != NULL)
1911 				m = m->m_nextpkt;
1912 			m->m_nextpkt = m;
1913 		}
1914 		return (error);
1915 	}
1916 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1917 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1918 		    NULL));
1919 	}
1920 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1921 	return (error);
1922 
1923   bad:
1924 	/*
1925 	 * ln is valid and the caller did not pass in
1926 	 * an llentry
1927 	 */
1928 	if ((ln != NULL) && (lle == NULL)) {
1929 		if (flags & LLE_EXCLUSIVE)
1930 			LLE_WUNLOCK(ln);
1931 		else
1932 			LLE_RUNLOCK(ln);
1933 	}
1934 	if (m)
1935 		m_freem(m);
1936 	return (error);
1937 }
1938 #undef senderr
1939 
1940 
1941 int
1942 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
1943     struct sockaddr_in6 *dst, struct route *ro)
1944 {
1945 	struct mbuf *m, *m_head;
1946 	struct ifnet *outifp;
1947 	int error = 0;
1948 
1949 	m_head = chain;
1950 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
1951 		outifp = origifp;
1952 	else
1953 		outifp = ifp;
1954 
1955 	while (m_head) {
1956 		m = m_head;
1957 		m_head = m_head->m_nextpkt;
1958 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
1959 	}
1960 
1961 	/*
1962 	 * XXX
1963 	 * note that intermediate errors are blindly ignored - but this is
1964 	 * the same convention as used with nd6_output when called by
1965 	 * nd6_cache_lladdr
1966 	 */
1967 	return (error);
1968 }
1969 
1970 
1971 int
1972 nd6_need_cache(struct ifnet *ifp)
1973 {
1974 	/*
1975 	 * XXX: we currently do not make neighbor cache on any interface
1976 	 * other than ARCnet, Ethernet, FDDI and GIF.
1977 	 *
1978 	 * RFC2893 says:
1979 	 * - unidirectional tunnels needs no ND
1980 	 */
1981 	switch (ifp->if_type) {
1982 	case IFT_ARCNET:
1983 	case IFT_ETHER:
1984 	case IFT_FDDI:
1985 	case IFT_IEEE1394:
1986 #ifdef IFT_L2VLAN
1987 	case IFT_L2VLAN:
1988 #endif
1989 #ifdef IFT_IEEE80211
1990 	case IFT_IEEE80211:
1991 #endif
1992 #ifdef IFT_CARP
1993 	case IFT_CARP:
1994 #endif
1995 	case IFT_GIF:		/* XXX need more cases? */
1996 	case IFT_PPP:
1997 	case IFT_TUNNEL:
1998 	case IFT_BRIDGE:
1999 	case IFT_PROPVIRTUAL:
2000 		return (1);
2001 	default:
2002 		return (0);
2003 	}
2004 }
2005 
2006 /*
2007  * the callers of this function need to be re-worked to drop
2008  * the lle lock, drop here for now
2009  */
2010 int
2011 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2012     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2013 {
2014 	struct llentry *ln;
2015 
2016 	*lle = NULL;
2017 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2018 	if (m->m_flags & M_MCAST) {
2019 		int i;
2020 
2021 		switch (ifp->if_type) {
2022 		case IFT_ETHER:
2023 		case IFT_FDDI:
2024 #ifdef IFT_L2VLAN
2025 		case IFT_L2VLAN:
2026 #endif
2027 #ifdef IFT_IEEE80211
2028 		case IFT_IEEE80211:
2029 #endif
2030 		case IFT_BRIDGE:
2031 		case IFT_ISO88025:
2032 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2033 						 desten);
2034 			return (0);
2035 		case IFT_IEEE1394:
2036 			/*
2037 			 * netbsd can use if_broadcastaddr, but we don't do so
2038 			 * to reduce # of ifdef.
2039 			 */
2040 			for (i = 0; i < ifp->if_addrlen; i++)
2041 				desten[i] = ~0;
2042 			return (0);
2043 		case IFT_ARCNET:
2044 			*desten = 0;
2045 			return (0);
2046 		default:
2047 			m_freem(m);
2048 			return (EAFNOSUPPORT);
2049 		}
2050 	}
2051 
2052 
2053 	/*
2054 	 * the entry should have been created in nd6_store_lladdr
2055 	 */
2056 	IF_AFDATA_LOCK(ifp);
2057 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2058 	IF_AFDATA_UNLOCK(ifp);
2059 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2060 		if (ln != NULL)
2061 			LLE_RUNLOCK(ln);
2062 		/* this could happen, if we could not allocate memory */
2063 		m_freem(m);
2064 		return (1);
2065 	}
2066 
2067 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2068 	*lle = ln;
2069 	LLE_RUNLOCK(ln);
2070 	/*
2071 	 * A *small* use after free race exists here
2072 	 */
2073 	return (0);
2074 }
2075 
2076 static void
2077 clear_llinfo_pqueue(struct llentry *ln)
2078 {
2079 	struct mbuf *m_hold, *m_hold_next;
2080 
2081 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2082 		m_hold_next = m_hold->m_nextpkt;
2083 		m_hold->m_nextpkt = NULL;
2084 		m_freem(m_hold);
2085 	}
2086 
2087 	ln->la_hold = NULL;
2088 	return;
2089 }
2090 
2091 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2092 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2093 #ifdef SYSCTL_DECL
2094 SYSCTL_DECL(_net_inet6_icmp6);
2095 #endif
2096 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2097 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2098 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2099 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2100 SYSCTL_V_INT(V_NET, vnet_inet6, _net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN,
2101 	nd6_maxqueuelen, CTLFLAG_RW, nd6_maxqueuelen, 1, "");
2102 
2103 static int
2104 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2105 {
2106 	INIT_VNET_INET6(curvnet);
2107 	int error;
2108 	char buf[1024] __aligned(4);
2109 	struct in6_defrouter *d, *de;
2110 	struct nd_defrouter *dr;
2111 
2112 	if (req->newptr)
2113 		return EPERM;
2114 	error = 0;
2115 
2116 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2117 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2118 		d = (struct in6_defrouter *)buf;
2119 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2120 
2121 		if (d + 1 <= de) {
2122 			bzero(d, sizeof(*d));
2123 			d->rtaddr.sin6_family = AF_INET6;
2124 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2125 			d->rtaddr.sin6_addr = dr->rtaddr;
2126 			error = sa6_recoverscope(&d->rtaddr);
2127 			if (error != 0)
2128 				return (error);
2129 			d->flags = dr->flags;
2130 			d->rtlifetime = dr->rtlifetime;
2131 			d->expire = dr->expire;
2132 			d->if_index = dr->ifp->if_index;
2133 		} else
2134 			panic("buffer too short");
2135 
2136 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2137 		if (error)
2138 			break;
2139 	}
2140 
2141 	return (error);
2142 }
2143 
2144 static int
2145 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2146 {
2147 	INIT_VNET_INET6(curvnet);
2148 	int error;
2149 	char buf[1024] __aligned(4);
2150 	struct in6_prefix *p, *pe;
2151 	struct nd_prefix *pr;
2152 	char ip6buf[INET6_ADDRSTRLEN];
2153 
2154 	if (req->newptr)
2155 		return EPERM;
2156 	error = 0;
2157 
2158 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2159 		u_short advrtrs;
2160 		size_t advance;
2161 		struct sockaddr_in6 *sin6, *s6;
2162 		struct nd_pfxrouter *pfr;
2163 
2164 		p = (struct in6_prefix *)buf;
2165 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2166 
2167 		if (p + 1 <= pe) {
2168 			bzero(p, sizeof(*p));
2169 			sin6 = (struct sockaddr_in6 *)(p + 1);
2170 
2171 			p->prefix = pr->ndpr_prefix;
2172 			if (sa6_recoverscope(&p->prefix)) {
2173 				log(LOG_ERR,
2174 				    "scope error in prefix list (%s)\n",
2175 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2176 				/* XXX: press on... */
2177 			}
2178 			p->raflags = pr->ndpr_raf;
2179 			p->prefixlen = pr->ndpr_plen;
2180 			p->vltime = pr->ndpr_vltime;
2181 			p->pltime = pr->ndpr_pltime;
2182 			p->if_index = pr->ndpr_ifp->if_index;
2183 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2184 				p->expire = 0;
2185 			else {
2186 				time_t maxexpire;
2187 
2188 				/* XXX: we assume time_t is signed. */
2189 				maxexpire = (-1) &
2190 				    ~((time_t)1 <<
2191 				    ((sizeof(maxexpire) * 8) - 1));
2192 				if (pr->ndpr_vltime <
2193 				    maxexpire - pr->ndpr_lastupdate) {
2194 				    p->expire = pr->ndpr_lastupdate +
2195 				        pr->ndpr_vltime;
2196 				} else
2197 					p->expire = maxexpire;
2198 			}
2199 			p->refcnt = pr->ndpr_refcnt;
2200 			p->flags = pr->ndpr_stateflags;
2201 			p->origin = PR_ORIG_RA;
2202 			advrtrs = 0;
2203 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2204 			     pfr = pfr->pfr_next) {
2205 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2206 					advrtrs++;
2207 					continue;
2208 				}
2209 				s6 = &sin6[advrtrs];
2210 				bzero(s6, sizeof(*s6));
2211 				s6->sin6_family = AF_INET6;
2212 				s6->sin6_len = sizeof(*sin6);
2213 				s6->sin6_addr = pfr->router->rtaddr;
2214 				if (sa6_recoverscope(s6)) {
2215 					log(LOG_ERR,
2216 					    "scope error in "
2217 					    "prefix list (%s)\n",
2218 					    ip6_sprintf(ip6buf,
2219 						    &pfr->router->rtaddr));
2220 				}
2221 				advrtrs++;
2222 			}
2223 			p->advrtrs = advrtrs;
2224 		} else
2225 			panic("buffer too short");
2226 
2227 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2228 		error = SYSCTL_OUT(req, buf, advance);
2229 		if (error)
2230 			break;
2231 	}
2232 
2233 	return (error);
2234 }
2235