xref: /freebsd/sys/netinet6/nd6.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 <netinet/icmp6.h>
74 
75 #include <sys/limits.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
80 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
81 
82 #define SIN6(s) ((struct sockaddr_in6 *)s)
83 
84 VNET_DEFINE(int, nd6_prune);
85 VNET_DEFINE(int, nd6_delay);
86 VNET_DEFINE(int, nd6_umaxtries);
87 VNET_DEFINE(int, nd6_mmaxtries);
88 VNET_DEFINE(int, nd6_useloopback);
89 VNET_DEFINE(int, nd6_gctimer);
90 
91 /* preventing too many loops in ND option parsing */
92 static VNET_DEFINE(int, nd6_maxndopt);
93 VNET_DEFINE(int, nd6_maxnudhint);
94 static VNET_DEFINE(int, nd6_maxqueuelen);
95 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
96 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
97 
98 VNET_DEFINE(int, nd6_debug);
99 
100 /* for debugging? */
101 #if 0
102 static int nd6_inuse, nd6_allocated;
103 #endif
104 
105 VNET_DEFINE(struct nd_drhead, nd_defrouter);
106 VNET_DEFINE(struct nd_prhead, nd_prefix);
107 
108 VNET_DEFINE(int, nd6_recalc_reachtm_interval);
109 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
110 
111 static struct sockaddr_in6 all1_sa;
112 
113 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *,
114 	struct ifnet *));
115 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
116 static void nd6_slowtimo(void *);
117 static int regen_tmpaddr(struct in6_ifaddr *);
118 static struct llentry *nd6_free(struct llentry *, int);
119 static void nd6_llinfo_timer(void *);
120 static void clear_llinfo_pqueue(struct llentry *);
121 
122 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
123 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
124 
125 VNET_DEFINE(struct callout, nd6_timer_ch);
126 
127 VNET_DECLARE(int, dad_ignore_ns);
128 VNET_DECLARE(int, dad_maxtry);
129 #define	V_dad_ignore_ns			VNET(dad_ignore_ns)
130 #define	V_dad_maxtry			VNET(dad_maxtry)
131 
132 void
133 nd6_init(void)
134 {
135 	int i;
136 
137 	V_nd6_prune	= 1;	/* walk list every 1 seconds */
138 	V_nd6_delay	= 5;	/* delay first probe time 5 second */
139 	V_nd6_umaxtries	= 3;	/* maximum unicast query */
140 	V_nd6_mmaxtries	= 3;	/* maximum multicast query */
141 	V_nd6_useloopback = 1;	/* use loopback interface for local traffic */
142 	V_nd6_gctimer	= (60 * 60 * 24); /* 1 day: garbage collection timer */
143 
144 	/* preventing too many loops in ND option parsing */
145 	V_nd6_maxndopt = 10;	/* max # of ND options allowed */
146 
147 	V_nd6_maxnudhint = 0;	/* max # of subsequent upper layer hints */
148 	V_nd6_maxqueuelen = 1;	/* max pkts cached in unresolved ND entries */
149 
150 #ifdef ND6_DEBUG
151 	V_nd6_debug = 1;
152 #else
153 	V_nd6_debug = 0;
154 #endif
155 
156 	V_nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
157 
158 	V_dad_ignore_ns = 0;	/* ignore NS in DAD - specwise incorrect*/
159 	V_dad_maxtry = 15;	/* max # of *tries* to transmit DAD packet */
160 
161 	/*
162 	 * XXX just to get this to compile KMM
163 	 */
164 #ifdef notyet
165 	V_llinfo_nd6.ln_next = &V_llinfo_nd6;
166 	V_llinfo_nd6.ln_prev = &V_llinfo_nd6;
167 #endif
168 	LIST_INIT(&V_nd_prefix);
169 
170 	V_ip6_use_tempaddr = 0;
171 	V_ip6_temp_preferred_lifetime = DEF_TEMP_PREFERRED_LIFETIME;
172 	V_ip6_temp_valid_lifetime = DEF_TEMP_VALID_LIFETIME;
173 	V_ip6_temp_regen_advance = TEMPADDR_REGEN_ADVANCE;
174 
175 	V_ip6_desync_factor = 0;
176 
177 	all1_sa.sin6_family = AF_INET6;
178 	all1_sa.sin6_len = sizeof(struct sockaddr_in6);
179 	for (i = 0; i < sizeof(all1_sa.sin6_addr); i++)
180 		all1_sa.sin6_addr.s6_addr[i] = 0xff;
181 
182 	/* initialization of the default router list */
183 	TAILQ_INIT(&V_nd_defrouter);
184 	/* start timer */
185 	callout_init(&V_nd6_slowtimo_ch, 0);
186 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
187 	    nd6_slowtimo, curvnet);
188 }
189 
190 
191 #ifdef VIMAGE
192 void
193 nd6_destroy()
194 {
195 
196 	callout_drain(&V_nd6_slowtimo_ch);
197 	callout_drain(&V_nd6_timer_ch);
198 }
199 #endif
200 
201 struct nd_ifinfo *
202 nd6_ifattach(struct ifnet *ifp)
203 {
204 	struct nd_ifinfo *nd;
205 
206 	nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK);
207 	bzero(nd, sizeof(*nd));
208 
209 	nd->initialized = 1;
210 
211 	nd->chlim = IPV6_DEFHLIM;
212 	nd->basereachable = REACHABLE_TIME;
213 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
214 	nd->retrans = RETRANS_TIMER;
215 	/*
216 	 * Note that the default value of ip6_accept_rtadv is 0, which means
217 	 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV
218 	 * here.
219 	 */
220 	nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV);
221 
222 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
223 	nd6_setmtu0(ifp, nd);
224 
225 	return nd;
226 }
227 
228 void
229 nd6_ifdetach(struct nd_ifinfo *nd)
230 {
231 
232 	free(nd, M_IP6NDP);
233 }
234 
235 /*
236  * Reset ND level link MTU. This function is called when the physical MTU
237  * changes, which means we might have to adjust the ND level MTU.
238  */
239 void
240 nd6_setmtu(struct ifnet *ifp)
241 {
242 
243 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
244 }
245 
246 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
247 void
248 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
249 {
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 	struct nd_opt_hdr *nd_opt;
359 	int i = 0;
360 
361 	if (ndopts == NULL)
362 		panic("ndopts == NULL in nd6_options");
363 	if (ndopts->nd_opts_last == NULL)
364 		panic("uninitialized ndopts in nd6_options");
365 	if (ndopts->nd_opts_search == NULL)
366 		return 0;
367 
368 	while (1) {
369 		nd_opt = nd6_option(ndopts);
370 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
371 			/*
372 			 * Message validation requires that all included
373 			 * options have a length that is greater than zero.
374 			 */
375 			ICMP6STAT_INC(icp6s_nd_badopt);
376 			bzero(ndopts, sizeof(*ndopts));
377 			return -1;
378 		}
379 
380 		if (nd_opt == NULL)
381 			goto skip1;
382 
383 		switch (nd_opt->nd_opt_type) {
384 		case ND_OPT_SOURCE_LINKADDR:
385 		case ND_OPT_TARGET_LINKADDR:
386 		case ND_OPT_MTU:
387 		case ND_OPT_REDIRECTED_HEADER:
388 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
389 				nd6log((LOG_INFO,
390 				    "duplicated ND6 option found (type=%d)\n",
391 				    nd_opt->nd_opt_type));
392 				/* XXX bark? */
393 			} else {
394 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
395 					= nd_opt;
396 			}
397 			break;
398 		case ND_OPT_PREFIX_INFORMATION:
399 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
400 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
401 					= nd_opt;
402 			}
403 			ndopts->nd_opts_pi_end =
404 				(struct nd_opt_prefix_info *)nd_opt;
405 			break;
406 		default:
407 			/*
408 			 * Unknown options must be silently ignored,
409 			 * to accomodate future extension to the protocol.
410 			 */
411 			nd6log((LOG_DEBUG,
412 			    "nd6_options: unsupported option %d - "
413 			    "option ignored\n", nd_opt->nd_opt_type));
414 		}
415 
416 skip1:
417 		i++;
418 		if (i > V_nd6_maxndopt) {
419 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
420 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
421 			break;
422 		}
423 
424 		if (ndopts->nd_opts_done)
425 			break;
426 	}
427 
428 	return 0;
429 }
430 
431 /*
432  * ND6 timer routine to handle ND6 entries
433  */
434 void
435 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
436 {
437 	if (tick < 0) {
438 		ln->la_expire = 0;
439 		ln->ln_ntick = 0;
440 		callout_stop(&ln->ln_timer_ch);
441 		/*
442 		 * XXX - do we know that there is
443 		 * callout installed? i.e. are we
444 		 * guaranteed that we're not dropping
445 		 * a reference that we did not add?
446 		 * KMM
447 		 */
448 		LLE_REMREF(ln);
449 	} else {
450 		ln->la_expire = time_second + tick / hz;
451 		LLE_ADDREF(ln);
452 		if (tick > INT_MAX) {
453 			ln->ln_ntick = tick - INT_MAX;
454 			callout_reset(&ln->ln_timer_ch, INT_MAX,
455 			    nd6_llinfo_timer, ln);
456 		} else {
457 			ln->ln_ntick = 0;
458 			callout_reset(&ln->ln_timer_ch, tick,
459 			    nd6_llinfo_timer, ln);
460 		}
461 	}
462 }
463 
464 void
465 nd6_llinfo_settimer(struct llentry *ln, long tick)
466 {
467 
468 	LLE_WLOCK(ln);
469 	nd6_llinfo_settimer_locked(ln, tick);
470 	LLE_WUNLOCK(ln);
471 }
472 
473 static void
474 nd6_llinfo_timer(void *arg)
475 {
476 	struct llentry *ln;
477 	struct in6_addr *dst;
478 	struct ifnet *ifp;
479 	struct nd_ifinfo *ndi = NULL;
480 
481 	ln = (struct llentry *)arg;
482 	if (ln == NULL) {
483 		panic("%s: NULL entry!\n", __func__);
484 		return;
485 	}
486 
487 	if ((ifp = ((ln->lle_tbl != NULL) ? ln->lle_tbl->llt_ifp : NULL)) == NULL)
488 		panic("ln ifp == NULL");
489 
490 	CURVNET_SET(ifp->if_vnet);
491 
492 	if (ln->ln_ntick > 0) {
493 		if (ln->ln_ntick > INT_MAX) {
494 			ln->ln_ntick -= INT_MAX;
495 			nd6_llinfo_settimer(ln, INT_MAX);
496 		} else {
497 			ln->ln_ntick = 0;
498 			nd6_llinfo_settimer(ln, ln->ln_ntick);
499 		}
500 		goto done;
501 	}
502 
503 	ndi = ND_IFINFO(ifp);
504 	dst = &L3_ADDR_SIN6(ln)->sin6_addr;
505 	if ((ln->la_flags & LLE_STATIC) || (ln->la_expire > time_second)) {
506 		goto done;
507 	}
508 
509 	if (ln->la_flags & LLE_DELETED) {
510 		(void)nd6_free(ln, 0);
511 		goto done;
512 	}
513 
514 	switch (ln->ln_state) {
515 	case ND6_LLINFO_INCOMPLETE:
516 		if (ln->la_asked < V_nd6_mmaxtries) {
517 			ln->la_asked++;
518 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
519 			nd6_ns_output(ifp, NULL, dst, ln, 0);
520 		} else {
521 			struct mbuf *m = ln->la_hold;
522 			if (m) {
523 				struct mbuf *m0;
524 
525 				/*
526 				 * assuming every packet in la_hold has the
527 				 * same IP header
528 				 */
529 				m0 = m->m_nextpkt;
530 				m->m_nextpkt = NULL;
531 				icmp6_error2(m, ICMP6_DST_UNREACH,
532 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
533 
534 				ln->la_hold = m0;
535 				clear_llinfo_pqueue(ln);
536 			}
537 			(void)nd6_free(ln, 0);
538 			ln = NULL;
539 		}
540 		break;
541 	case ND6_LLINFO_REACHABLE:
542 		if (!ND6_LLINFO_PERMANENT(ln)) {
543 			ln->ln_state = ND6_LLINFO_STALE;
544 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
545 		}
546 		break;
547 
548 	case ND6_LLINFO_STALE:
549 		/* Garbage Collection(RFC 2461 5.3) */
550 		if (!ND6_LLINFO_PERMANENT(ln)) {
551 			(void)nd6_free(ln, 1);
552 			ln = NULL;
553 		}
554 		break;
555 
556 	case ND6_LLINFO_DELAY:
557 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
558 			/* We need NUD */
559 			ln->la_asked = 1;
560 			ln->ln_state = ND6_LLINFO_PROBE;
561 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
562 			nd6_ns_output(ifp, dst, dst, ln, 0);
563 		} else {
564 			ln->ln_state = ND6_LLINFO_STALE; /* XXX */
565 			nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz);
566 		}
567 		break;
568 	case ND6_LLINFO_PROBE:
569 		if (ln->la_asked < V_nd6_umaxtries) {
570 			ln->la_asked++;
571 			nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000);
572 			nd6_ns_output(ifp, dst, dst, ln, 0);
573 		} else {
574 			(void)nd6_free(ln, 0);
575 			ln = NULL;
576 		}
577 		break;
578 	}
579 	CURVNET_RESTORE();
580 done:
581 	if (ln != NULL)
582 		LLE_FREE(ln);
583 }
584 
585 
586 /*
587  * ND6 timer routine to expire default route list and prefix list
588  */
589 void
590 nd6_timer(void *arg)
591 {
592 	CURVNET_SET((struct vnet *) arg);
593 	int s;
594 	struct nd_defrouter *dr;
595 	struct nd_prefix *pr;
596 	struct in6_ifaddr *ia6, *nia6;
597 	struct in6_addrlifetime *lt6;
598 
599 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
600 	    nd6_timer, curvnet);
601 
602 	/* expire default router list */
603 	s = splnet();
604 	dr = TAILQ_FIRST(&V_nd_defrouter);
605 	while (dr) {
606 		if (dr->expire && dr->expire < time_second) {
607 			struct nd_defrouter *t;
608 			t = TAILQ_NEXT(dr, dr_entry);
609 			defrtrlist_del(dr);
610 			dr = t;
611 		} else {
612 			dr = TAILQ_NEXT(dr, dr_entry);
613 		}
614 	}
615 
616 	/*
617 	 * expire interface addresses.
618 	 * in the past the loop was inside prefix expiry processing.
619 	 * However, from a stricter speci-confrmance standpoint, we should
620 	 * rather separate address lifetimes and prefix lifetimes.
621 	 *
622 	 * XXXRW: in6_ifaddrhead locking.
623 	 */
624   addrloop:
625 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
626 		/* check address lifetime */
627 		lt6 = &ia6->ia6_lifetime;
628 		if (IFA6_IS_INVALID(ia6)) {
629 			int regen = 0;
630 
631 			/*
632 			 * If the expiring address is temporary, try
633 			 * regenerating a new one.  This would be useful when
634 			 * we suspended a laptop PC, then turned it on after a
635 			 * period that could invalidate all temporary
636 			 * addresses.  Although we may have to restart the
637 			 * loop (see below), it must be after purging the
638 			 * address.  Otherwise, we'd see an infinite loop of
639 			 * regeneration.
640 			 */
641 			if (V_ip6_use_tempaddr &&
642 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
643 				if (regen_tmpaddr(ia6) == 0)
644 					regen = 1;
645 			}
646 
647 			in6_purgeaddr(&ia6->ia_ifa);
648 
649 			if (regen)
650 				goto addrloop; /* XXX: see below */
651 		} else if (IFA6_IS_DEPRECATED(ia6)) {
652 			int oldflags = ia6->ia6_flags;
653 
654 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
655 
656 			/*
657 			 * If a temporary address has just become deprecated,
658 			 * regenerate a new one if possible.
659 			 */
660 			if (V_ip6_use_tempaddr &&
661 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
662 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
663 
664 				if (regen_tmpaddr(ia6) == 0) {
665 					/*
666 					 * A new temporary address is
667 					 * generated.
668 					 * XXX: this means the address chain
669 					 * has changed while we are still in
670 					 * the loop.  Although the change
671 					 * would not cause disaster (because
672 					 * it's not a deletion, but an
673 					 * addition,) we'd rather restart the
674 					 * loop just for safety.  Or does this
675 					 * significantly reduce performance??
676 					 */
677 					goto addrloop;
678 				}
679 			}
680 		} else {
681 			/*
682 			 * A new RA might have made a deprecated address
683 			 * preferred.
684 			 */
685 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
686 		}
687 	}
688 
689 	/* expire prefix list */
690 	pr = V_nd_prefix.lh_first;
691 	while (pr) {
692 		/*
693 		 * check prefix lifetime.
694 		 * since pltime is just for autoconf, pltime processing for
695 		 * prefix is not necessary.
696 		 */
697 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
698 		    time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) {
699 			struct nd_prefix *t;
700 			t = pr->ndpr_next;
701 
702 			/*
703 			 * address expiration and prefix expiration are
704 			 * separate.  NEVER perform in6_purgeaddr here.
705 			 */
706 
707 			prelist_remove(pr);
708 			pr = t;
709 		} else
710 			pr = pr->ndpr_next;
711 	}
712 	splx(s);
713 	CURVNET_RESTORE();
714 }
715 
716 /*
717  * ia6 - deprecated/invalidated temporary address
718  */
719 static int
720 regen_tmpaddr(struct in6_ifaddr *ia6)
721 {
722 	struct ifaddr *ifa;
723 	struct ifnet *ifp;
724 	struct in6_ifaddr *public_ifa6 = NULL;
725 
726 	ifp = ia6->ia_ifa.ifa_ifp;
727 	IF_ADDR_LOCK(ifp);
728 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
729 		struct in6_ifaddr *it6;
730 
731 		if (ifa->ifa_addr->sa_family != AF_INET6)
732 			continue;
733 
734 		it6 = (struct in6_ifaddr *)ifa;
735 
736 		/* ignore no autoconf addresses. */
737 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
738 			continue;
739 
740 		/* ignore autoconf addresses with different prefixes. */
741 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
742 			continue;
743 
744 		/*
745 		 * Now we are looking at an autoconf address with the same
746 		 * prefix as ours.  If the address is temporary and is still
747 		 * preferred, do not create another one.  It would be rare, but
748 		 * could happen, for example, when we resume a laptop PC after
749 		 * a long period.
750 		 */
751 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
752 		    !IFA6_IS_DEPRECATED(it6)) {
753 			public_ifa6 = NULL;
754 			break;
755 		}
756 
757 		/*
758 		 * This is a public autoconf address that has the same prefix
759 		 * as ours.  If it is preferred, keep it.  We can't break the
760 		 * loop here, because there may be a still-preferred temporary
761 		 * address with the prefix.
762 		 */
763 		if (!IFA6_IS_DEPRECATED(it6))
764 		    public_ifa6 = it6;
765 	}
766 
767 	if (public_ifa6 != NULL) {
768 		int e;
769 
770 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
771 			IF_ADDR_UNLOCK(ifp);
772 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
773 			    " tmp addr,errno=%d\n", e);
774 			return (-1);
775 		}
776 		IF_ADDR_UNLOCK(ifp);
777 		return (0);
778 	}
779 
780 	IF_ADDR_UNLOCK(ifp);
781 	return (-1);
782 }
783 
784 /*
785  * Nuke neighbor cache/prefix/default router management table, right before
786  * ifp goes away.
787  */
788 void
789 nd6_purge(struct ifnet *ifp)
790 {
791 	struct nd_defrouter *dr, *ndr;
792 	struct nd_prefix *pr, *npr;
793 
794 	/*
795 	 * Nuke default router list entries toward ifp.
796 	 * We defer removal of default router list entries that is installed
797 	 * in the routing table, in order to keep additional side effects as
798 	 * small as possible.
799 	 */
800 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
801 		ndr = TAILQ_NEXT(dr, dr_entry);
802 		if (dr->installed)
803 			continue;
804 
805 		if (dr->ifp == ifp)
806 			defrtrlist_del(dr);
807 	}
808 
809 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) {
810 		ndr = TAILQ_NEXT(dr, dr_entry);
811 		if (!dr->installed)
812 			continue;
813 
814 		if (dr->ifp == ifp)
815 			defrtrlist_del(dr);
816 	}
817 
818 	/* Nuke prefix list entries toward ifp */
819 	for (pr = V_nd_prefix.lh_first; pr; pr = npr) {
820 		npr = pr->ndpr_next;
821 		if (pr->ndpr_ifp == ifp) {
822 			/*
823 			 * Because if_detach() does *not* release prefixes
824 			 * while purging addresses the reference count will
825 			 * still be above zero. We therefore reset it to
826 			 * make sure that the prefix really gets purged.
827 			 */
828 			pr->ndpr_refcnt = 0;
829 
830 			/*
831 			 * Previously, pr->ndpr_addr is removed as well,
832 			 * but I strongly believe we don't have to do it.
833 			 * nd6_purge() is only called from in6_ifdetach(),
834 			 * which removes all the associated interface addresses
835 			 * by itself.
836 			 * (jinmei@kame.net 20010129)
837 			 */
838 			prelist_remove(pr);
839 		}
840 	}
841 
842 	/* cancel default outgoing interface setting */
843 	if (V_nd6_defifindex == ifp->if_index)
844 		nd6_setdefaultiface(0);
845 
846 	if (!V_ip6_forwarding && V_ip6_accept_rtadv) { /* XXX: too restrictive? */
847 		/* refresh default router list
848 		 *
849 		 *
850 		 */
851 		defrouter_select();
852 
853 	}
854 
855 	/* XXXXX
856 	 * We do not nuke the neighbor cache entries here any more
857 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
858 	 * nd6_purge() is invoked by in6_ifdetach() which is called
859 	 * from if_detach() where everything gets purged. So let
860 	 * in6_domifdetach() do the actual L2 table purging work.
861 	 */
862 }
863 
864 /*
865  * the caller acquires and releases the lock on the lltbls
866  * Returns the llentry locked
867  */
868 struct llentry *
869 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp)
870 {
871 	struct sockaddr_in6 sin6;
872 	struct llentry *ln;
873 	int llflags = 0;
874 
875 	bzero(&sin6, sizeof(sin6));
876 	sin6.sin6_len = sizeof(struct sockaddr_in6);
877 	sin6.sin6_family = AF_INET6;
878 	sin6.sin6_addr = *addr6;
879 
880 	IF_AFDATA_LOCK_ASSERT(ifp);
881 
882 	if (flags & ND6_CREATE)
883 	    llflags |= LLE_CREATE;
884 	if (flags & ND6_EXCLUSIVE)
885 	    llflags |= LLE_EXCLUSIVE;
886 
887 	ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6);
888 	if ((ln != NULL) && (flags & LLE_CREATE)) {
889 		ln->ln_state = ND6_LLINFO_NOSTATE;
890 		callout_init(&ln->ln_timer_ch, 0);
891 	}
892 
893 	return (ln);
894 }
895 
896 /*
897  * Test whether a given IPv6 address is a neighbor or not, ignoring
898  * the actual neighbor cache.  The neighbor cache is ignored in order
899  * to not reenter the routing code from within itself.
900  */
901 static int
902 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
903 {
904 	struct nd_prefix *pr;
905 	struct ifaddr *dstaddr;
906 
907 	/*
908 	 * A link-local address is always a neighbor.
909 	 * XXX: a link does not necessarily specify a single interface.
910 	 */
911 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
912 		struct sockaddr_in6 sin6_copy;
913 		u_int32_t zone;
914 
915 		/*
916 		 * We need sin6_copy since sa6_recoverscope() may modify the
917 		 * content (XXX).
918 		 */
919 		sin6_copy = *addr;
920 		if (sa6_recoverscope(&sin6_copy))
921 			return (0); /* XXX: should be impossible */
922 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
923 			return (0);
924 		if (sin6_copy.sin6_scope_id == zone)
925 			return (1);
926 		else
927 			return (0);
928 	}
929 
930 	/*
931 	 * If the address matches one of our addresses,
932 	 * it should be a neighbor.
933 	 * If the address matches one of our on-link prefixes, it should be a
934 	 * neighbor.
935 	 */
936 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
937 		if (pr->ndpr_ifp != ifp)
938 			continue;
939 
940 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK))
941 			continue;
942 
943 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
944 		    &addr->sin6_addr, &pr->ndpr_mask))
945 			return (1);
946 	}
947 
948 	/*
949 	 * If the address is assigned on the node of the other side of
950 	 * a p2p interface, the address should be a neighbor.
951 	 */
952 	dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr);
953 	if (dstaddr != NULL) {
954 		if (dstaddr->ifa_ifp == ifp) {
955 			ifa_free(dstaddr);
956 			return (1);
957 		}
958 		ifa_free(dstaddr);
959 	}
960 
961 	/*
962 	 * If the default router list is empty, all addresses are regarded
963 	 * as on-link, and thus, as a neighbor.
964 	 * XXX: we restrict the condition to hosts, because routers usually do
965 	 * not have the "default router list".
966 	 */
967 	if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL &&
968 	    V_nd6_defifindex == ifp->if_index) {
969 		return (1);
970 	}
971 
972 	return (0);
973 }
974 
975 
976 /*
977  * Detect if a given IPv6 address identifies a neighbor on a given link.
978  * XXX: should take care of the destination of a p2p link?
979  */
980 int
981 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp)
982 {
983 	struct llentry *lle;
984 	int rc = 0;
985 
986 	IF_AFDATA_UNLOCK_ASSERT(ifp);
987 	if (nd6_is_new_addr_neighbor(addr, ifp))
988 		return (1);
989 
990 	/*
991 	 * Even if the address matches none of our addresses, it might be
992 	 * in the neighbor cache.
993 	 */
994 	IF_AFDATA_LOCK(ifp);
995 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
996 		LLE_RUNLOCK(lle);
997 		rc = 1;
998 	}
999 	IF_AFDATA_UNLOCK(ifp);
1000 	return (rc);
1001 }
1002 
1003 /*
1004  * Free an nd6 llinfo entry.
1005  * Since the function would cause significant changes in the kernel, DO NOT
1006  * make it global, unless you have a strong reason for the change, and are sure
1007  * that the change is safe.
1008  */
1009 static struct llentry *
1010 nd6_free(struct llentry *ln, int gc)
1011 {
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 	struct llentry *ln;
1119 	struct ifnet *ifp;
1120 
1121 	if ((dst6 == NULL) || (rt == NULL))
1122 		return;
1123 
1124 	ifp = rt->rt_ifp;
1125 	IF_AFDATA_LOCK(ifp);
1126 	ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL);
1127 	IF_AFDATA_UNLOCK(ifp);
1128 	if (ln == NULL)
1129 		return;
1130 
1131 	if (ln->ln_state < ND6_LLINFO_REACHABLE)
1132 		goto done;
1133 
1134 	/*
1135 	 * if we get upper-layer reachability confirmation many times,
1136 	 * it is possible we have false information.
1137 	 */
1138 	if (!force) {
1139 		ln->ln_byhint++;
1140 		if (ln->ln_byhint > V_nd6_maxnudhint) {
1141 			goto done;
1142 		}
1143 	}
1144 
1145  	ln->ln_state = ND6_LLINFO_REACHABLE;
1146 	if (!ND6_LLINFO_PERMANENT(ln)) {
1147 		nd6_llinfo_settimer(ln,
1148 		    (long)ND_IFINFO(rt->rt_ifp)->reachable * hz);
1149 	}
1150 done:
1151 	LLE_WUNLOCK(ln);
1152 }
1153 
1154 
1155 int
1156 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1157 {
1158 	struct in6_drlist *drl = (struct in6_drlist *)data;
1159 	struct in6_oprlist *oprl = (struct in6_oprlist *)data;
1160 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1161 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1162 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1163 	struct nd_defrouter *dr;
1164 	struct nd_prefix *pr;
1165 	int i = 0, error = 0;
1166 	int s;
1167 
1168 	switch (cmd) {
1169 	case SIOCGDRLST_IN6:
1170 		/*
1171 		 * obsolete API, use sysctl under net.inet6.icmp6
1172 		 */
1173 		bzero(drl, sizeof(*drl));
1174 		s = splnet();
1175 		dr = TAILQ_FIRST(&V_nd_defrouter);
1176 		while (dr && i < DRLSTSIZ) {
1177 			drl->defrouter[i].rtaddr = dr->rtaddr;
1178 			in6_clearscope(&drl->defrouter[i].rtaddr);
1179 
1180 			drl->defrouter[i].flags = dr->flags;
1181 			drl->defrouter[i].rtlifetime = dr->rtlifetime;
1182 			drl->defrouter[i].expire = dr->expire;
1183 			drl->defrouter[i].if_index = dr->ifp->if_index;
1184 			i++;
1185 			dr = TAILQ_NEXT(dr, dr_entry);
1186 		}
1187 		splx(s);
1188 		break;
1189 	case SIOCGPRLST_IN6:
1190 		/*
1191 		 * obsolete API, use sysctl under net.inet6.icmp6
1192 		 *
1193 		 * XXX the structure in6_prlist was changed in backward-
1194 		 * incompatible manner.  in6_oprlist is used for SIOCGPRLST_IN6,
1195 		 * in6_prlist is used for nd6_sysctl() - fill_prlist().
1196 		 */
1197 		/*
1198 		 * XXX meaning of fields, especialy "raflags", is very
1199 		 * differnet between RA prefix list and RR/static prefix list.
1200 		 * how about separating ioctls into two?
1201 		 */
1202 		bzero(oprl, sizeof(*oprl));
1203 		s = splnet();
1204 		pr = V_nd_prefix.lh_first;
1205 		while (pr && i < PRLSTSIZ) {
1206 			struct nd_pfxrouter *pfr;
1207 			int j;
1208 
1209 			oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr;
1210 			oprl->prefix[i].raflags = pr->ndpr_raf;
1211 			oprl->prefix[i].prefixlen = pr->ndpr_plen;
1212 			oprl->prefix[i].vltime = pr->ndpr_vltime;
1213 			oprl->prefix[i].pltime = pr->ndpr_pltime;
1214 			oprl->prefix[i].if_index = pr->ndpr_ifp->if_index;
1215 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
1216 				oprl->prefix[i].expire = 0;
1217 			else {
1218 				time_t maxexpire;
1219 
1220 				/* XXX: we assume time_t is signed. */
1221 				maxexpire = (-1) &
1222 				    ~((time_t)1 <<
1223 				    ((sizeof(maxexpire) * 8) - 1));
1224 				if (pr->ndpr_vltime <
1225 				    maxexpire - pr->ndpr_lastupdate) {
1226 					oprl->prefix[i].expire =
1227 					    pr->ndpr_lastupdate +
1228 					    pr->ndpr_vltime;
1229 				} else
1230 					oprl->prefix[i].expire = maxexpire;
1231 			}
1232 
1233 			pfr = pr->ndpr_advrtrs.lh_first;
1234 			j = 0;
1235 			while (pfr) {
1236 				if (j < DRLSTSIZ) {
1237 #define RTRADDR oprl->prefix[i].advrtr[j]
1238 					RTRADDR = pfr->router->rtaddr;
1239 					in6_clearscope(&RTRADDR);
1240 #undef RTRADDR
1241 				}
1242 				j++;
1243 				pfr = pfr->pfr_next;
1244 			}
1245 			oprl->prefix[i].advrtrs = j;
1246 			oprl->prefix[i].origin = PR_ORIG_RA;
1247 
1248 			i++;
1249 			pr = pr->ndpr_next;
1250 		}
1251 		splx(s);
1252 
1253 		break;
1254 	case OSIOCGIFINFO_IN6:
1255 #define ND	ndi->ndi
1256 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1257 		bzero(&ND, sizeof(ND));
1258 		ND.linkmtu = IN6_LINKMTU(ifp);
1259 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1260 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1261 		ND.reachable = ND_IFINFO(ifp)->reachable;
1262 		ND.retrans = ND_IFINFO(ifp)->retrans;
1263 		ND.flags = ND_IFINFO(ifp)->flags;
1264 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1265 		ND.chlim = ND_IFINFO(ifp)->chlim;
1266 		break;
1267 	case SIOCGIFINFO_IN6:
1268 		ND = *ND_IFINFO(ifp);
1269 		break;
1270 	case SIOCSIFINFO_IN6:
1271 		/*
1272 		 * used to change host variables from userland.
1273 		 * intented for a use on router to reflect RA configurations.
1274 		 */
1275 		/* 0 means 'unspecified' */
1276 		if (ND.linkmtu != 0) {
1277 			if (ND.linkmtu < IPV6_MMTU ||
1278 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1279 				error = EINVAL;
1280 				break;
1281 			}
1282 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1283 		}
1284 
1285 		if (ND.basereachable != 0) {
1286 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1287 
1288 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1289 			if (ND.basereachable != obasereachable)
1290 				ND_IFINFO(ifp)->reachable =
1291 				    ND_COMPUTE_RTIME(ND.basereachable);
1292 		}
1293 		if (ND.retrans != 0)
1294 			ND_IFINFO(ifp)->retrans = ND.retrans;
1295 		if (ND.chlim != 0)
1296 			ND_IFINFO(ifp)->chlim = ND.chlim;
1297 		/* FALLTHROUGH */
1298 	case SIOCSIFINFO_FLAGS:
1299 		ND_IFINFO(ifp)->flags = ND.flags;
1300 		break;
1301 #undef ND
1302 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1303 		/* sync kernel routing table with the default router list */
1304 		defrouter_reset();
1305 		defrouter_select();
1306 		break;
1307 	case SIOCSPFXFLUSH_IN6:
1308 	{
1309 		/* flush all the prefix advertised by routers */
1310 		struct nd_prefix *pr, *next;
1311 
1312 		s = splnet();
1313 		for (pr = V_nd_prefix.lh_first; pr; pr = next) {
1314 			struct in6_ifaddr *ia, *ia_next;
1315 
1316 			next = pr->ndpr_next;
1317 
1318 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1319 				continue; /* XXX */
1320 
1321 			/* do we really have to remove addresses as well? */
1322 			/* XXXRW: in6_ifaddrhead locking. */
1323 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1324 			    ia_next) {
1325 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1326 					continue;
1327 
1328 				if (ia->ia6_ndpr == pr)
1329 					in6_purgeaddr(&ia->ia_ifa);
1330 			}
1331 			prelist_remove(pr);
1332 		}
1333 		splx(s);
1334 		break;
1335 	}
1336 	case SIOCSRTRFLUSH_IN6:
1337 	{
1338 		/* flush all the default routers */
1339 		struct nd_defrouter *dr, *next;
1340 
1341 		s = splnet();
1342 		defrouter_reset();
1343 		for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) {
1344 			next = TAILQ_NEXT(dr, dr_entry);
1345 			defrtrlist_del(dr);
1346 		}
1347 		defrouter_select();
1348 		splx(s);
1349 		break;
1350 	}
1351 	case SIOCGNBRINFO_IN6:
1352 	{
1353 		struct llentry *ln;
1354 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1355 
1356 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1357 			return (error);
1358 
1359 		IF_AFDATA_LOCK(ifp);
1360 		ln = nd6_lookup(&nb_addr, 0, ifp);
1361 		IF_AFDATA_UNLOCK(ifp);
1362 
1363 		if (ln == NULL) {
1364 			error = EINVAL;
1365 			break;
1366 		}
1367 		nbi->state = ln->ln_state;
1368 		nbi->asked = ln->la_asked;
1369 		nbi->isrouter = ln->ln_router;
1370 		nbi->expire = ln->la_expire;
1371 		LLE_RUNLOCK(ln);
1372 		break;
1373 	}
1374 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1375 		ndif->ifindex = V_nd6_defifindex;
1376 		break;
1377 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1378 		return (nd6_setdefaultiface(ndif->ifindex));
1379 	}
1380 	return (error);
1381 }
1382 
1383 /*
1384  * Create neighbor cache entry and cache link-layer address,
1385  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1386  *
1387  * type - ICMP6 type
1388  * code - type dependent information
1389  *
1390  * XXXXX
1391  *  The caller of this function already acquired the ndp
1392  *  cache table lock because the cache entry is returned.
1393  */
1394 struct llentry *
1395 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1396     int lladdrlen, int type, int code)
1397 {
1398 	struct llentry *ln = NULL;
1399 	int is_newentry;
1400 	int do_update;
1401 	int olladdr;
1402 	int llchange;
1403 	int flags = 0;
1404 	int newstate = 0;
1405 	uint16_t router = 0;
1406 	struct sockaddr_in6 sin6;
1407 	struct mbuf *chain = NULL;
1408 	int static_route = 0;
1409 
1410 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1411 
1412 	if (ifp == NULL)
1413 		panic("ifp == NULL in nd6_cache_lladdr");
1414 	if (from == NULL)
1415 		panic("from == NULL in nd6_cache_lladdr");
1416 
1417 	/* nothing must be updated for unspecified address */
1418 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1419 		return NULL;
1420 
1421 	/*
1422 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1423 	 * the caller.
1424 	 *
1425 	 * XXX If the link does not have link-layer adderss, what should
1426 	 * we do? (ifp->if_addrlen == 0)
1427 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1428 	 * description on it in NS section (RFC 2461 7.2.3).
1429 	 */
1430 	flags |= lladdr ? ND6_EXCLUSIVE : 0;
1431 	IF_AFDATA_LOCK(ifp);
1432 	ln = nd6_lookup(from, flags, ifp);
1433 
1434 	if (ln == NULL) {
1435 		flags |= LLE_EXCLUSIVE;
1436 		ln = nd6_lookup(from, flags |ND6_CREATE, ifp);
1437 		IF_AFDATA_UNLOCK(ifp);
1438 		is_newentry = 1;
1439 	} else {
1440 		IF_AFDATA_UNLOCK(ifp);
1441 		/* do nothing if static ndp is set */
1442 		if (ln->la_flags & LLE_STATIC) {
1443 			static_route = 1;
1444 			goto done;
1445 		}
1446 		is_newentry = 0;
1447 	}
1448 	if (ln == NULL)
1449 		return (NULL);
1450 
1451 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1452 	if (olladdr && lladdr) {
1453 		llchange = bcmp(lladdr, &ln->ll_addr,
1454 		    ifp->if_addrlen);
1455 	} else
1456 		llchange = 0;
1457 
1458 	/*
1459 	 * newentry olladdr  lladdr  llchange	(*=record)
1460 	 *	0	n	n	--	(1)
1461 	 *	0	y	n	--	(2)
1462 	 *	0	n	y	--	(3) * STALE
1463 	 *	0	y	y	n	(4) *
1464 	 *	0	y	y	y	(5) * STALE
1465 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1466 	 *	1	--	y	--	(7) * STALE
1467 	 */
1468 
1469 	if (lladdr) {		/* (3-5) and (7) */
1470 		/*
1471 		 * Record source link-layer address
1472 		 * XXX is it dependent to ifp->if_type?
1473 		 */
1474 		bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen);
1475 		ln->la_flags |= LLE_VALID;
1476 	}
1477 
1478 	if (!is_newentry) {
1479 		if ((!olladdr && lladdr != NULL) ||	/* (3) */
1480 		    (olladdr && lladdr != NULL && llchange)) {	/* (5) */
1481 			do_update = 1;
1482 			newstate = ND6_LLINFO_STALE;
1483 		} else					/* (1-2,4) */
1484 			do_update = 0;
1485 	} else {
1486 		do_update = 1;
1487 		if (lladdr == NULL)			/* (6) */
1488 			newstate = ND6_LLINFO_NOSTATE;
1489 		else					/* (7) */
1490 			newstate = ND6_LLINFO_STALE;
1491 	}
1492 
1493 	if (do_update) {
1494 		/*
1495 		 * Update the state of the neighbor cache.
1496 		 */
1497 		ln->ln_state = newstate;
1498 
1499 		if (ln->ln_state == ND6_LLINFO_STALE) {
1500 			/*
1501 			 * XXX: since nd6_output() below will cause
1502 			 * state tansition to DELAY and reset the timer,
1503 			 * we must set the timer now, although it is actually
1504 			 * meaningless.
1505 			 */
1506 			nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1507 
1508 			if (ln->la_hold) {
1509 				struct mbuf *m_hold, *m_hold_next;
1510 
1511 				/*
1512 				 * reset the la_hold in advance, to explicitly
1513 				 * prevent a la_hold lookup in nd6_output()
1514 				 * (wouldn't happen, though...)
1515 				 */
1516 				for (m_hold = ln->la_hold, ln->la_hold = NULL;
1517 				    m_hold; m_hold = m_hold_next) {
1518 					m_hold_next = m_hold->m_nextpkt;
1519 					m_hold->m_nextpkt = NULL;
1520 
1521 					/*
1522 					 * we assume ifp is not a p2p here, so
1523 					 * just set the 2nd argument as the
1524 					 * 1st one.
1525 					 */
1526 					nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain);
1527 				}
1528 				/*
1529 				 * If we have mbufs in the chain we need to do
1530 				 * deferred transmit. Copy the address from the
1531 				 * llentry before dropping the lock down below.
1532 				 */
1533 				if (chain != NULL)
1534 					memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6));
1535 			}
1536 		} else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) {
1537 			/* probe right away */
1538 			nd6_llinfo_settimer_locked((void *)ln, 0);
1539 		}
1540 	}
1541 
1542 	/*
1543 	 * ICMP6 type dependent behavior.
1544 	 *
1545 	 * NS: clear IsRouter if new entry
1546 	 * RS: clear IsRouter
1547 	 * RA: set IsRouter if there's lladdr
1548 	 * redir: clear IsRouter if new entry
1549 	 *
1550 	 * RA case, (1):
1551 	 * The spec says that we must set IsRouter in the following cases:
1552 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1553 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1554 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1555 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1556 	 * neighbor cache, this is similar to (6).
1557 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1558 	 *
1559 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1560 	 *							D R
1561 	 *	0	n	n	--	(1)	c   ?     s
1562 	 *	0	y	n	--	(2)	c   s     s
1563 	 *	0	n	y	--	(3)	c   s     s
1564 	 *	0	y	y	n	(4)	c   s     s
1565 	 *	0	y	y	y	(5)	c   s     s
1566 	 *	1	--	n	--	(6) c	c	c s
1567 	 *	1	--	y	--	(7) c	c   s	c s
1568 	 *
1569 	 *					(c=clear s=set)
1570 	 */
1571 	switch (type & 0xff) {
1572 	case ND_NEIGHBOR_SOLICIT:
1573 		/*
1574 		 * New entry must have is_router flag cleared.
1575 		 */
1576 		if (is_newentry)	/* (6-7) */
1577 			ln->ln_router = 0;
1578 		break;
1579 	case ND_REDIRECT:
1580 		/*
1581 		 * If the icmp is a redirect to a better router, always set the
1582 		 * is_router flag.  Otherwise, if the entry is newly created,
1583 		 * clear the flag.  [RFC 2461, sec 8.3]
1584 		 */
1585 		if (code == ND_REDIRECT_ROUTER)
1586 			ln->ln_router = 1;
1587 		else if (is_newentry) /* (6-7) */
1588 			ln->ln_router = 0;
1589 		break;
1590 	case ND_ROUTER_SOLICIT:
1591 		/*
1592 		 * is_router flag must always be cleared.
1593 		 */
1594 		ln->ln_router = 0;
1595 		break;
1596 	case ND_ROUTER_ADVERT:
1597 		/*
1598 		 * Mark an entry with lladdr as a router.
1599 		 */
1600 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1601 		    (is_newentry && lladdr)) {			/* (7) */
1602 			ln->ln_router = 1;
1603 		}
1604 		break;
1605 	}
1606 
1607 	if (ln != NULL) {
1608 		static_route = (ln->la_flags & LLE_STATIC);
1609 		router = ln->ln_router;
1610 
1611 		if (flags & ND6_EXCLUSIVE)
1612 			LLE_WUNLOCK(ln);
1613 		else
1614 			LLE_RUNLOCK(ln);
1615 		if (static_route)
1616 			ln = NULL;
1617 	}
1618 	if (chain)
1619 		nd6_output_flush(ifp, ifp, chain, &sin6, NULL);
1620 
1621 	/*
1622 	 * When the link-layer address of a router changes, select the
1623 	 * best router again.  In particular, when the neighbor entry is newly
1624 	 * created, it might affect the selection policy.
1625 	 * Question: can we restrict the first condition to the "is_newentry"
1626 	 * case?
1627 	 * XXX: when we hear an RA from a new router with the link-layer
1628 	 * address option, defrouter_select() is called twice, since
1629 	 * defrtrlist_update called the function as well.  However, I believe
1630 	 * we can compromise the overhead, since it only happens the first
1631 	 * time.
1632 	 * XXX: although defrouter_select() should not have a bad effect
1633 	 * for those are not autoconfigured hosts, we explicitly avoid such
1634 	 * cases for safety.
1635 	 */
1636 	if (do_update && router && !V_ip6_forwarding && V_ip6_accept_rtadv) {
1637 		/*
1638 		 * guaranteed recursion
1639 		 */
1640 		defrouter_select();
1641 	}
1642 
1643 	return (ln);
1644 done:
1645 	if (ln != NULL) {
1646 		if (flags & ND6_EXCLUSIVE)
1647 			LLE_WUNLOCK(ln);
1648 		else
1649 			LLE_RUNLOCK(ln);
1650 		if (static_route)
1651 			ln = NULL;
1652 	}
1653 	return (ln);
1654 }
1655 
1656 static void
1657 nd6_slowtimo(void *arg)
1658 {
1659 	CURVNET_SET((struct vnet *) arg);
1660 	struct nd_ifinfo *nd6if;
1661 	struct ifnet *ifp;
1662 
1663 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1664 	    nd6_slowtimo, curvnet);
1665 	IFNET_RLOCK();
1666 	for (ifp = TAILQ_FIRST(&V_ifnet); ifp;
1667 	    ifp = TAILQ_NEXT(ifp, if_list)) {
1668 		nd6if = ND_IFINFO(ifp);
1669 		if (nd6if->basereachable && /* already initialized */
1670 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1671 			/*
1672 			 * Since reachable time rarely changes by router
1673 			 * advertisements, we SHOULD insure that a new random
1674 			 * value gets recomputed at least once every few hours.
1675 			 * (RFC 2461, 6.3.4)
1676 			 */
1677 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
1678 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
1679 		}
1680 	}
1681 	IFNET_RUNLOCK();
1682 	CURVNET_RESTORE();
1683 }
1684 
1685 int
1686 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1687     struct sockaddr_in6 *dst, struct rtentry *rt0)
1688 {
1689 
1690 	return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL));
1691 }
1692 
1693 
1694 /*
1695  * Note that I'm not enforcing any global serialization
1696  * lle state or asked changes here as the logic is too
1697  * complicated to avoid having to always acquire an exclusive
1698  * lock
1699  * KMM
1700  *
1701  */
1702 #define senderr(e) { error = (e); goto bad;}
1703 
1704 int
1705 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0,
1706     struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle,
1707 	struct mbuf **chain)
1708 {
1709 	struct mbuf *m = m0;
1710 	struct llentry *ln = lle;
1711 	int error = 0;
1712 	int flags = 0;
1713 
1714 #ifdef INVARIANTS
1715 	if (lle != NULL) {
1716 
1717 		LLE_WLOCK_ASSERT(lle);
1718 
1719 		KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed"));
1720 	}
1721 #endif
1722 	if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr))
1723 		goto sendpkt;
1724 
1725 	if (nd6_need_cache(ifp) == 0)
1726 		goto sendpkt;
1727 
1728 	/*
1729 	 * next hop determination.  This routine is derived from ether_output.
1730 	 */
1731 
1732 	/*
1733 	 * Address resolution or Neighbor Unreachability Detection
1734 	 * for the next hop.
1735 	 * At this point, the destination of the packet must be a unicast
1736 	 * or an anycast address(i.e. not a multicast).
1737 	 */
1738 
1739 	flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0;
1740 	if (ln == NULL) {
1741 	retry:
1742 		IF_AFDATA_LOCK(ifp);
1743 		ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst);
1744 		IF_AFDATA_UNLOCK(ifp);
1745 		if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
1746 			/*
1747 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1748 			 * the condition below is not very efficient.  But we believe
1749 			 * it is tolerable, because this should be a rare case.
1750 			 */
1751 			flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0);
1752 			IF_AFDATA_LOCK(ifp);
1753 			ln = nd6_lookup(&dst->sin6_addr, flags, ifp);
1754 			IF_AFDATA_UNLOCK(ifp);
1755 		}
1756 	}
1757 	if (ln == NULL) {
1758 		if ((ifp->if_flags & IFF_POINTOPOINT) == 0 &&
1759 		    !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
1760 			char ip6buf[INET6_ADDRSTRLEN];
1761 			log(LOG_DEBUG,
1762 			    "nd6_output: can't allocate llinfo for %s "
1763 			    "(ln=%p)\n",
1764 			    ip6_sprintf(ip6buf, &dst->sin6_addr), ln);
1765 			senderr(EIO);	/* XXX: good error? */
1766 		}
1767 		goto sendpkt;	/* send anyway */
1768 	}
1769 
1770 	/* We don't have to do link-layer address resolution on a p2p link. */
1771 	if ((ifp->if_flags & IFF_POINTOPOINT) != 0 &&
1772 	    ln->ln_state < ND6_LLINFO_REACHABLE) {
1773 		if ((flags & LLE_EXCLUSIVE) == 0) {
1774 			flags |= LLE_EXCLUSIVE;
1775 			goto retry;
1776 		}
1777 		ln->ln_state = ND6_LLINFO_STALE;
1778 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz);
1779 	}
1780 
1781 	/*
1782 	 * The first time we send a packet to a neighbor whose entry is
1783 	 * STALE, we have to change the state to DELAY and a sets a timer to
1784 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
1785 	 * neighbor unreachability detection on expiration.
1786 	 * (RFC 2461 7.3.3)
1787 	 */
1788 	if (ln->ln_state == ND6_LLINFO_STALE) {
1789 		if ((flags & LLE_EXCLUSIVE) == 0) {
1790 			flags |= LLE_EXCLUSIVE;
1791 			LLE_RUNLOCK(ln);
1792 			goto retry;
1793 		}
1794 		ln->la_asked = 0;
1795 		ln->ln_state = ND6_LLINFO_DELAY;
1796 		nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz);
1797 	}
1798 
1799 	/*
1800 	 * If the neighbor cache entry has a state other than INCOMPLETE
1801 	 * (i.e. its link-layer address is already resolved), just
1802 	 * send the packet.
1803 	 */
1804 	if (ln->ln_state > ND6_LLINFO_INCOMPLETE)
1805 		goto sendpkt;
1806 
1807 	/*
1808 	 * There is a neighbor cache entry, but no ethernet address
1809 	 * response yet.  Append this latest packet to the end of the
1810 	 * packet queue in the mbuf, unless the number of the packet
1811 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
1812 	 * the oldest packet in the queue will be removed.
1813 	 */
1814 	if (ln->ln_state == ND6_LLINFO_NOSTATE)
1815 		ln->ln_state = ND6_LLINFO_INCOMPLETE;
1816 
1817 	if ((flags & LLE_EXCLUSIVE) == 0) {
1818 		flags |= LLE_EXCLUSIVE;
1819 		LLE_RUNLOCK(ln);
1820 		goto retry;
1821 	}
1822 	if (ln->la_hold) {
1823 		struct mbuf *m_hold;
1824 		int i;
1825 
1826 		i = 0;
1827 		for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) {
1828 			i++;
1829 			if (m_hold->m_nextpkt == NULL) {
1830 				m_hold->m_nextpkt = m;
1831 				break;
1832 			}
1833 		}
1834 		while (i >= V_nd6_maxqueuelen) {
1835 			m_hold = ln->la_hold;
1836 			ln->la_hold = ln->la_hold->m_nextpkt;
1837 			m_freem(m_hold);
1838 			i--;
1839 		}
1840 	} else {
1841 		ln->la_hold = m;
1842 	}
1843 	/*
1844 	 * We did the lookup (no lle arg) so we
1845 	 * need to do the unlock here
1846 	 */
1847 	if (lle == NULL) {
1848 		if (flags & LLE_EXCLUSIVE)
1849 			LLE_WUNLOCK(ln);
1850 		else
1851 			LLE_RUNLOCK(ln);
1852 	}
1853 
1854 	/*
1855 	 * If there has been no NS for the neighbor after entering the
1856 	 * INCOMPLETE state, send the first solicitation.
1857 	 */
1858 	if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) {
1859 		ln->la_asked++;
1860 
1861 		nd6_llinfo_settimer(ln,
1862 		    (long)ND_IFINFO(ifp)->retrans * hz / 1000);
1863 		nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0);
1864 	}
1865 	return (0);
1866 
1867   sendpkt:
1868 	/* discard the packet if IPv6 operation is disabled on the interface */
1869 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
1870 		error = ENETDOWN; /* better error? */
1871 		goto bad;
1872 	}
1873 	/*
1874 	 * ln is valid and the caller did not pass in
1875 	 * an llentry
1876 	 */
1877 	if ((ln != NULL) && (lle == NULL)) {
1878 		if (flags & LLE_EXCLUSIVE)
1879 			LLE_WUNLOCK(ln);
1880 		else
1881 			LLE_RUNLOCK(ln);
1882 	}
1883 
1884 #ifdef MAC
1885 	mac_netinet6_nd6_send(ifp, m);
1886 #endif
1887 	/*
1888 	 * We were passed in a pointer to an lle with the lock held
1889 	 * this means that we can't call if_output as we will
1890 	 * recurse on the lle lock - so what we do is we create
1891 	 * a list of mbufs to send and transmit them in the caller
1892 	 * after the lock is dropped
1893 	 */
1894 	if (lle != NULL) {
1895 		if (*chain == NULL)
1896 			*chain = m;
1897 		else {
1898 			struct mbuf *m = *chain;
1899 
1900 			/*
1901 			 * append mbuf to end of deferred chain
1902 			 */
1903 			while (m->m_nextpkt != NULL)
1904 				m = m->m_nextpkt;
1905 			m->m_nextpkt = m;
1906 		}
1907 		return (error);
1908 	}
1909 	if ((ifp->if_flags & IFF_LOOPBACK) != 0) {
1910 		return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst,
1911 		    NULL));
1912 	}
1913 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL);
1914 	return (error);
1915 
1916   bad:
1917 	/*
1918 	 * ln is valid and the caller did not pass in
1919 	 * an llentry
1920 	 */
1921 	if ((ln != NULL) && (lle == NULL)) {
1922 		if (flags & LLE_EXCLUSIVE)
1923 			LLE_WUNLOCK(ln);
1924 		else
1925 			LLE_RUNLOCK(ln);
1926 	}
1927 	if (m)
1928 		m_freem(m);
1929 	return (error);
1930 }
1931 #undef senderr
1932 
1933 
1934 int
1935 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
1936     struct sockaddr_in6 *dst, struct route *ro)
1937 {
1938 	struct mbuf *m, *m_head;
1939 	struct ifnet *outifp;
1940 	int error = 0;
1941 
1942 	m_head = chain;
1943 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
1944 		outifp = origifp;
1945 	else
1946 		outifp = ifp;
1947 
1948 	while (m_head) {
1949 		m = m_head;
1950 		m_head = m_head->m_nextpkt;
1951 		error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro);
1952 	}
1953 
1954 	/*
1955 	 * XXX
1956 	 * note that intermediate errors are blindly ignored - but this is
1957 	 * the same convention as used with nd6_output when called by
1958 	 * nd6_cache_lladdr
1959 	 */
1960 	return (error);
1961 }
1962 
1963 
1964 int
1965 nd6_need_cache(struct ifnet *ifp)
1966 {
1967 	/*
1968 	 * XXX: we currently do not make neighbor cache on any interface
1969 	 * other than ARCnet, Ethernet, FDDI and GIF.
1970 	 *
1971 	 * RFC2893 says:
1972 	 * - unidirectional tunnels needs no ND
1973 	 */
1974 	switch (ifp->if_type) {
1975 	case IFT_ARCNET:
1976 	case IFT_ETHER:
1977 	case IFT_FDDI:
1978 	case IFT_IEEE1394:
1979 #ifdef IFT_L2VLAN
1980 	case IFT_L2VLAN:
1981 #endif
1982 #ifdef IFT_IEEE80211
1983 	case IFT_IEEE80211:
1984 #endif
1985 #ifdef IFT_CARP
1986 	case IFT_CARP:
1987 #endif
1988 	case IFT_GIF:		/* XXX need more cases? */
1989 	case IFT_PPP:
1990 	case IFT_TUNNEL:
1991 	case IFT_BRIDGE:
1992 	case IFT_PROPVIRTUAL:
1993 		return (1);
1994 	default:
1995 		return (0);
1996 	}
1997 }
1998 
1999 /*
2000  * the callers of this function need to be re-worked to drop
2001  * the lle lock, drop here for now
2002  */
2003 int
2004 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m,
2005     struct sockaddr *dst, u_char *desten, struct llentry **lle)
2006 {
2007 	struct llentry *ln;
2008 
2009 	*lle = NULL;
2010 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2011 	if (m->m_flags & M_MCAST) {
2012 		int i;
2013 
2014 		switch (ifp->if_type) {
2015 		case IFT_ETHER:
2016 		case IFT_FDDI:
2017 #ifdef IFT_L2VLAN
2018 		case IFT_L2VLAN:
2019 #endif
2020 #ifdef IFT_IEEE80211
2021 		case IFT_IEEE80211:
2022 #endif
2023 		case IFT_BRIDGE:
2024 		case IFT_ISO88025:
2025 			ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr,
2026 						 desten);
2027 			return (0);
2028 		case IFT_IEEE1394:
2029 			/*
2030 			 * netbsd can use if_broadcastaddr, but we don't do so
2031 			 * to reduce # of ifdef.
2032 			 */
2033 			for (i = 0; i < ifp->if_addrlen; i++)
2034 				desten[i] = ~0;
2035 			return (0);
2036 		case IFT_ARCNET:
2037 			*desten = 0;
2038 			return (0);
2039 		default:
2040 			m_freem(m);
2041 			return (EAFNOSUPPORT);
2042 		}
2043 	}
2044 
2045 
2046 	/*
2047 	 * the entry should have been created in nd6_store_lladdr
2048 	 */
2049 	IF_AFDATA_LOCK(ifp);
2050 	ln = lla_lookup(LLTABLE6(ifp), 0, dst);
2051 	IF_AFDATA_UNLOCK(ifp);
2052 	if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) {
2053 		if (ln != NULL)
2054 			LLE_RUNLOCK(ln);
2055 		/* this could happen, if we could not allocate memory */
2056 		m_freem(m);
2057 		return (1);
2058 	}
2059 
2060 	bcopy(&ln->ll_addr, desten, ifp->if_addrlen);
2061 	*lle = ln;
2062 	LLE_RUNLOCK(ln);
2063 	/*
2064 	 * A *small* use after free race exists here
2065 	 */
2066 	return (0);
2067 }
2068 
2069 static void
2070 clear_llinfo_pqueue(struct llentry *ln)
2071 {
2072 	struct mbuf *m_hold, *m_hold_next;
2073 
2074 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2075 		m_hold_next = m_hold->m_nextpkt;
2076 		m_hold->m_nextpkt = NULL;
2077 		m_freem(m_hold);
2078 	}
2079 
2080 	ln->la_hold = NULL;
2081 	return;
2082 }
2083 
2084 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2085 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2086 #ifdef SYSCTL_DECL
2087 SYSCTL_DECL(_net_inet6_icmp6);
2088 #endif
2089 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2090 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2091 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2092 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2093 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2094 	CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2095 
2096 static int
2097 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2098 {
2099 	int error;
2100 	char buf[1024] __aligned(4);
2101 	struct in6_defrouter *d, *de;
2102 	struct nd_defrouter *dr;
2103 
2104 	if (req->newptr)
2105 		return EPERM;
2106 	error = 0;
2107 
2108 	for (dr = TAILQ_FIRST(&V_nd_defrouter); dr;
2109 	     dr = TAILQ_NEXT(dr, dr_entry)) {
2110 		d = (struct in6_defrouter *)buf;
2111 		de = (struct in6_defrouter *)(buf + sizeof(buf));
2112 
2113 		if (d + 1 <= de) {
2114 			bzero(d, sizeof(*d));
2115 			d->rtaddr.sin6_family = AF_INET6;
2116 			d->rtaddr.sin6_len = sizeof(d->rtaddr);
2117 			d->rtaddr.sin6_addr = dr->rtaddr;
2118 			error = sa6_recoverscope(&d->rtaddr);
2119 			if (error != 0)
2120 				return (error);
2121 			d->flags = dr->flags;
2122 			d->rtlifetime = dr->rtlifetime;
2123 			d->expire = dr->expire;
2124 			d->if_index = dr->ifp->if_index;
2125 		} else
2126 			panic("buffer too short");
2127 
2128 		error = SYSCTL_OUT(req, buf, sizeof(*d));
2129 		if (error)
2130 			break;
2131 	}
2132 
2133 	return (error);
2134 }
2135 
2136 static int
2137 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2138 {
2139 	int error;
2140 	char buf[1024] __aligned(4);
2141 	struct in6_prefix *p, *pe;
2142 	struct nd_prefix *pr;
2143 	char ip6buf[INET6_ADDRSTRLEN];
2144 
2145 	if (req->newptr)
2146 		return EPERM;
2147 	error = 0;
2148 
2149 	for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) {
2150 		u_short advrtrs;
2151 		size_t advance;
2152 		struct sockaddr_in6 *sin6, *s6;
2153 		struct nd_pfxrouter *pfr;
2154 
2155 		p = (struct in6_prefix *)buf;
2156 		pe = (struct in6_prefix *)(buf + sizeof(buf));
2157 
2158 		if (p + 1 <= pe) {
2159 			bzero(p, sizeof(*p));
2160 			sin6 = (struct sockaddr_in6 *)(p + 1);
2161 
2162 			p->prefix = pr->ndpr_prefix;
2163 			if (sa6_recoverscope(&p->prefix)) {
2164 				log(LOG_ERR,
2165 				    "scope error in prefix list (%s)\n",
2166 				    ip6_sprintf(ip6buf, &p->prefix.sin6_addr));
2167 				/* XXX: press on... */
2168 			}
2169 			p->raflags = pr->ndpr_raf;
2170 			p->prefixlen = pr->ndpr_plen;
2171 			p->vltime = pr->ndpr_vltime;
2172 			p->pltime = pr->ndpr_pltime;
2173 			p->if_index = pr->ndpr_ifp->if_index;
2174 			if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2175 				p->expire = 0;
2176 			else {
2177 				time_t maxexpire;
2178 
2179 				/* XXX: we assume time_t is signed. */
2180 				maxexpire = (-1) &
2181 				    ~((time_t)1 <<
2182 				    ((sizeof(maxexpire) * 8) - 1));
2183 				if (pr->ndpr_vltime <
2184 				    maxexpire - pr->ndpr_lastupdate) {
2185 				    p->expire = pr->ndpr_lastupdate +
2186 				        pr->ndpr_vltime;
2187 				} else
2188 					p->expire = maxexpire;
2189 			}
2190 			p->refcnt = pr->ndpr_refcnt;
2191 			p->flags = pr->ndpr_stateflags;
2192 			p->origin = PR_ORIG_RA;
2193 			advrtrs = 0;
2194 			for (pfr = pr->ndpr_advrtrs.lh_first; pfr;
2195 			     pfr = pfr->pfr_next) {
2196 				if ((void *)&sin6[advrtrs + 1] > (void *)pe) {
2197 					advrtrs++;
2198 					continue;
2199 				}
2200 				s6 = &sin6[advrtrs];
2201 				bzero(s6, sizeof(*s6));
2202 				s6->sin6_family = AF_INET6;
2203 				s6->sin6_len = sizeof(*sin6);
2204 				s6->sin6_addr = pfr->router->rtaddr;
2205 				if (sa6_recoverscope(s6)) {
2206 					log(LOG_ERR,
2207 					    "scope error in "
2208 					    "prefix list (%s)\n",
2209 					    ip6_sprintf(ip6buf,
2210 						    &pfr->router->rtaddr));
2211 				}
2212 				advrtrs++;
2213 			}
2214 			p->advrtrs = advrtrs;
2215 		} else
2216 			panic("buffer too short");
2217 
2218 		advance = sizeof(*p) + sizeof(*sin6) * advrtrs;
2219 		error = SYSCTL_OUT(req, buf, advance);
2220 		if (error)
2221 			break;
2222 	}
2223 
2224 	return (error);
2225 }
2226