xref: /freebsd/sys/netinet6/nd6.c (revision ef36b3f75658d201edb495068db5e1be49593de5)
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/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.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/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/sdt.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/if.h>
58 #include <net/if_var.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 <netinet/in_kdtrace.h>
69 #include <net/if_llatbl.h>
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 <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/send.h>
79 
80 #include <sys/limits.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
86 
87 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
88 
89 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
90 
91 /* timer values */
92 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
93 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
94 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
95 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
96 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
97 					 * local traffic */
98 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
99 					 * collection timer */
100 
101 /* preventing too many loops in ND option parsing */
102 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
103 
104 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
105 					 * layer hints */
106 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
107 					 * ND entries */
108 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
109 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
110 
111 #ifdef ND6_DEBUG
112 VNET_DEFINE(int, nd6_debug) = 1;
113 #else
114 VNET_DEFINE(int, nd6_debug) = 0;
115 #endif
116 
117 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
118 
119 VNET_DEFINE(struct nd_drhead, nd_defrouter);
120 VNET_DEFINE(struct nd_prhead, nd_prefix);
121 VNET_DEFINE(struct rwlock, nd6_lock);
122 VNET_DEFINE(uint64_t, nd6_list_genid);
123 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
124 
125 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
126 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
127 
128 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
129 
130 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
131 	struct ifnet *);
132 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
133 static void nd6_slowtimo(void *);
134 static int regen_tmpaddr(struct in6_ifaddr *);
135 static void nd6_free(struct llentry **, int);
136 static void nd6_free_redirect(const struct llentry *);
137 static void nd6_llinfo_timer(void *);
138 static void nd6_llinfo_settimer_locked(struct llentry *, long);
139 static void clear_llinfo_pqueue(struct llentry *);
140 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
141 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
142     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
143 static int nd6_need_cache(struct ifnet *);
144 
145 
146 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
147 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
148 
149 VNET_DEFINE(struct callout, nd6_timer_ch);
150 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
151 
152 static void
153 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
154 {
155 	struct rt_addrinfo rtinfo;
156 	struct sockaddr_in6 dst;
157 	struct sockaddr_dl gw;
158 	struct ifnet *ifp;
159 	int type;
160 	int fibnum;
161 
162 	LLE_WLOCK_ASSERT(lle);
163 
164 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
165 		return;
166 
167 	switch (evt) {
168 	case LLENTRY_RESOLVED:
169 		type = RTM_ADD;
170 		KASSERT(lle->la_flags & LLE_VALID,
171 		    ("%s: %p resolved but not valid?", __func__, lle));
172 		break;
173 	case LLENTRY_EXPIRED:
174 		type = RTM_DELETE;
175 		break;
176 	default:
177 		return;
178 	}
179 
180 	ifp = lltable_get_ifp(lle->lle_tbl);
181 
182 	bzero(&dst, sizeof(dst));
183 	bzero(&gw, sizeof(gw));
184 	bzero(&rtinfo, sizeof(rtinfo));
185 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
186 	dst.sin6_scope_id = in6_getscopezone(ifp,
187 	    in6_addrscope(&dst.sin6_addr));
188 	gw.sdl_len = sizeof(struct sockaddr_dl);
189 	gw.sdl_family = AF_LINK;
190 	gw.sdl_alen = ifp->if_addrlen;
191 	gw.sdl_index = ifp->if_index;
192 	gw.sdl_type = ifp->if_type;
193 	if (evt == LLENTRY_RESOLVED)
194 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
195 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
196 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
197 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
198 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
199 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
200 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
201 }
202 
203 /*
204  * A handler for interface link layer address change event.
205  */
206 static void
207 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
208 {
209 
210 	lltable_update_ifaddr(LLTABLE6(ifp));
211 }
212 
213 void
214 nd6_init(void)
215 {
216 
217 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
218 	rw_init(&V_nd6_lock, "nd6 list");
219 
220 	LIST_INIT(&V_nd_prefix);
221 	TAILQ_INIT(&V_nd_defrouter);
222 
223 	/* Start timers. */
224 	callout_init(&V_nd6_slowtimo_ch, 0);
225 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
226 	    nd6_slowtimo, curvnet);
227 
228 	callout_init(&V_nd6_timer_ch, 0);
229 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
230 
231 	nd6_dad_init();
232 	if (IS_DEFAULT_VNET(curvnet)) {
233 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
234 		    NULL, EVENTHANDLER_PRI_ANY);
235 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
236 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
237 	}
238 }
239 
240 #ifdef VIMAGE
241 void
242 nd6_destroy()
243 {
244 
245 	callout_drain(&V_nd6_slowtimo_ch);
246 	callout_drain(&V_nd6_timer_ch);
247 	if (IS_DEFAULT_VNET(curvnet)) {
248 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
249 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
250 	}
251 	rw_destroy(&V_nd6_lock);
252 	mtx_destroy(&V_nd6_onlink_mtx);
253 }
254 #endif
255 
256 struct nd_ifinfo *
257 nd6_ifattach(struct ifnet *ifp)
258 {
259 	struct nd_ifinfo *nd;
260 
261 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
262 	nd->initialized = 1;
263 
264 	nd->chlim = IPV6_DEFHLIM;
265 	nd->basereachable = REACHABLE_TIME;
266 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
267 	nd->retrans = RETRANS_TIMER;
268 
269 	nd->flags = ND6_IFF_PERFORMNUD;
270 
271 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
272 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
273 	 * default regardless of the V_ip6_auto_linklocal configuration to
274 	 * give a reasonable default behavior.
275 	 */
276 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
277 	    (ifp->if_flags & IFF_LOOPBACK))
278 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
279 	/*
280 	 * A loopback interface does not need to accept RTADV.
281 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
282 	 * default regardless of the V_ip6_accept_rtadv configuration to
283 	 * prevent the interface from accepting RA messages arrived
284 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
285 	 */
286 	if (V_ip6_accept_rtadv &&
287 	    !(ifp->if_flags & IFF_LOOPBACK) &&
288 	    (ifp->if_type != IFT_BRIDGE))
289 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
290 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
291 		nd->flags |= ND6_IFF_NO_RADR;
292 
293 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
294 	nd6_setmtu0(ifp, nd);
295 
296 	return nd;
297 }
298 
299 void
300 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
301 {
302 	struct ifaddr *ifa, *next;
303 
304 	IF_ADDR_RLOCK(ifp);
305 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
306 		if (ifa->ifa_addr->sa_family != AF_INET6)
307 			continue;
308 
309 		/* stop DAD processing */
310 		nd6_dad_stop(ifa);
311 	}
312 	IF_ADDR_RUNLOCK(ifp);
313 
314 	free(nd, M_IP6NDP);
315 }
316 
317 /*
318  * Reset ND level link MTU. This function is called when the physical MTU
319  * changes, which means we might have to adjust the ND level MTU.
320  */
321 void
322 nd6_setmtu(struct ifnet *ifp)
323 {
324 	if (ifp->if_afdata[AF_INET6] == NULL)
325 		return;
326 
327 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
328 }
329 
330 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
331 void
332 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
333 {
334 	u_int32_t omaxmtu;
335 
336 	omaxmtu = ndi->maxmtu;
337 
338 	switch (ifp->if_type) {
339 	case IFT_ARCNET:
340 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
341 		break;
342 	case IFT_FDDI:
343 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
344 		break;
345 	case IFT_ISO88025:
346 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
347 		 break;
348 	default:
349 		ndi->maxmtu = ifp->if_mtu;
350 		break;
351 	}
352 
353 	/*
354 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
355 	 * undesirable situation.  We thus notify the operator of the change
356 	 * explicitly.  The check for omaxmtu is necessary to restrict the
357 	 * log to the case of changing the MTU, not initializing it.
358 	 */
359 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
360 		log(LOG_NOTICE, "nd6_setmtu0: "
361 		    "new link MTU on %s (%lu) is too small for IPv6\n",
362 		    if_name(ifp), (unsigned long)ndi->maxmtu);
363 	}
364 
365 	if (ndi->maxmtu > V_in6_maxmtu)
366 		in6_setmaxmtu(); /* check all interfaces just in case */
367 
368 }
369 
370 void
371 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
372 {
373 
374 	bzero(ndopts, sizeof(*ndopts));
375 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
376 	ndopts->nd_opts_last
377 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
378 
379 	if (icmp6len == 0) {
380 		ndopts->nd_opts_done = 1;
381 		ndopts->nd_opts_search = NULL;
382 	}
383 }
384 
385 /*
386  * Take one ND option.
387  */
388 struct nd_opt_hdr *
389 nd6_option(union nd_opts *ndopts)
390 {
391 	struct nd_opt_hdr *nd_opt;
392 	int olen;
393 
394 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
395 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
396 	    __func__));
397 	if (ndopts->nd_opts_search == NULL)
398 		return NULL;
399 	if (ndopts->nd_opts_done)
400 		return NULL;
401 
402 	nd_opt = ndopts->nd_opts_search;
403 
404 	/* make sure nd_opt_len is inside the buffer */
405 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
406 		bzero(ndopts, sizeof(*ndopts));
407 		return NULL;
408 	}
409 
410 	olen = nd_opt->nd_opt_len << 3;
411 	if (olen == 0) {
412 		/*
413 		 * Message validation requires that all included
414 		 * options have a length that is greater than zero.
415 		 */
416 		bzero(ndopts, sizeof(*ndopts));
417 		return NULL;
418 	}
419 
420 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
421 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
422 		/* option overruns the end of buffer, invalid */
423 		bzero(ndopts, sizeof(*ndopts));
424 		return NULL;
425 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
426 		/* reached the end of options chain */
427 		ndopts->nd_opts_done = 1;
428 		ndopts->nd_opts_search = NULL;
429 	}
430 	return nd_opt;
431 }
432 
433 /*
434  * Parse multiple ND options.
435  * This function is much easier to use, for ND routines that do not need
436  * multiple options of the same type.
437  */
438 int
439 nd6_options(union nd_opts *ndopts)
440 {
441 	struct nd_opt_hdr *nd_opt;
442 	int i = 0;
443 
444 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
445 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
446 	    __func__));
447 	if (ndopts->nd_opts_search == NULL)
448 		return 0;
449 
450 	while (1) {
451 		nd_opt = nd6_option(ndopts);
452 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
453 			/*
454 			 * Message validation requires that all included
455 			 * options have a length that is greater than zero.
456 			 */
457 			ICMP6STAT_INC(icp6s_nd_badopt);
458 			bzero(ndopts, sizeof(*ndopts));
459 			return -1;
460 		}
461 
462 		if (nd_opt == NULL)
463 			goto skip1;
464 
465 		switch (nd_opt->nd_opt_type) {
466 		case ND_OPT_SOURCE_LINKADDR:
467 		case ND_OPT_TARGET_LINKADDR:
468 		case ND_OPT_MTU:
469 		case ND_OPT_REDIRECTED_HEADER:
470 		case ND_OPT_NONCE:
471 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
472 				nd6log((LOG_INFO,
473 				    "duplicated ND6 option found (type=%d)\n",
474 				    nd_opt->nd_opt_type));
475 				/* XXX bark? */
476 			} else {
477 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
478 					= nd_opt;
479 			}
480 			break;
481 		case ND_OPT_PREFIX_INFORMATION:
482 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
483 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
484 					= nd_opt;
485 			}
486 			ndopts->nd_opts_pi_end =
487 				(struct nd_opt_prefix_info *)nd_opt;
488 			break;
489 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
490 		case ND_OPT_RDNSS:	/* RFC 6106 */
491 		case ND_OPT_DNSSL:	/* RFC 6106 */
492 			/*
493 			 * Silently ignore options we know and do not care about
494 			 * in the kernel.
495 			 */
496 			break;
497 		default:
498 			/*
499 			 * Unknown options must be silently ignored,
500 			 * to accommodate future extension to the protocol.
501 			 */
502 			nd6log((LOG_DEBUG,
503 			    "nd6_options: unsupported option %d - "
504 			    "option ignored\n", nd_opt->nd_opt_type));
505 		}
506 
507 skip1:
508 		i++;
509 		if (i > V_nd6_maxndopt) {
510 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
511 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
512 			break;
513 		}
514 
515 		if (ndopts->nd_opts_done)
516 			break;
517 	}
518 
519 	return 0;
520 }
521 
522 /*
523  * ND6 timer routine to handle ND6 entries
524  */
525 static void
526 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
527 {
528 	int canceled;
529 
530 	LLE_WLOCK_ASSERT(ln);
531 
532 	if (tick < 0) {
533 		ln->la_expire = 0;
534 		ln->ln_ntick = 0;
535 		canceled = callout_stop(&ln->lle_timer);
536 	} else {
537 		ln->la_expire = time_uptime + tick / hz;
538 		LLE_ADDREF(ln);
539 		if (tick > INT_MAX) {
540 			ln->ln_ntick = tick - INT_MAX;
541 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
542 			    nd6_llinfo_timer, ln);
543 		} else {
544 			ln->ln_ntick = 0;
545 			canceled = callout_reset(&ln->lle_timer, tick,
546 			    nd6_llinfo_timer, ln);
547 		}
548 	}
549 	if (canceled > 0)
550 		LLE_REMREF(ln);
551 }
552 
553 /*
554  * Gets source address of the first packet in hold queue
555  * and stores it in @src.
556  * Returns pointer to @src (if hold queue is not empty) or NULL.
557  *
558  * Set noinline to be dtrace-friendly
559  */
560 static __noinline struct in6_addr *
561 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
562 {
563 	struct ip6_hdr hdr;
564 	struct mbuf *m;
565 
566 	if (ln->la_hold == NULL)
567 		return (NULL);
568 
569 	/*
570 	 * assume every packet in la_hold has the same IP header
571 	 */
572 	m = ln->la_hold;
573 	if (sizeof(hdr) > m->m_len)
574 		return (NULL);
575 
576 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
577 	*src = hdr.ip6_src;
578 
579 	return (src);
580 }
581 
582 /*
583  * Checks if we need to switch from STALE state.
584  *
585  * RFC 4861 requires switching from STALE to DELAY state
586  * on first packet matching entry, waiting V_nd6_delay and
587  * transition to PROBE state (if upper layer confirmation was
588  * not received).
589  *
590  * This code performs a bit differently:
591  * On packet hit we don't change state (but desired state
592  * can be guessed by control plane). However, after V_nd6_delay
593  * seconds code will transition to PROBE state (so DELAY state
594  * is kinda skipped in most situations).
595  *
596  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
597  * we perform the following upon entering STALE state:
598  *
599  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
600  * if packet was transmitted at the start of given interval, we
601  * would be able to switch to PROBE state in V_nd6_delay seconds
602  * as user expects.
603  *
604  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
605  * lle in STALE state (remaining timer value stored in lle_remtime).
606  *
607  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
608  * seconds ago.
609  *
610  * Returns non-zero value if the entry is still STALE (storing
611  * the next timer interval in @pdelay).
612  *
613  * Returns zero value if original timer expired or we need to switch to
614  * PROBE (store that in @do_switch variable).
615  */
616 static int
617 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
618 {
619 	int nd_delay, nd_gctimer, r_skip_req;
620 	time_t lle_hittime;
621 	long delay;
622 
623 	*do_switch = 0;
624 	nd_gctimer = V_nd6_gctimer;
625 	nd_delay = V_nd6_delay;
626 
627 	LLE_REQ_LOCK(lle);
628 	r_skip_req = lle->r_skip_req;
629 	lle_hittime = lle->lle_hittime;
630 	LLE_REQ_UNLOCK(lle);
631 
632 	if (r_skip_req > 0) {
633 
634 		/*
635 		 * Nonzero r_skip_req value was set upon entering
636 		 * STALE state. Since value was not changed, no
637 		 * packets were passed using this lle. Ask for
638 		 * timer reschedule and keep STALE state.
639 		 */
640 		delay = (long)(MIN(nd_gctimer, nd_delay));
641 		delay *= hz;
642 		if (lle->lle_remtime > delay)
643 			lle->lle_remtime -= delay;
644 		else {
645 			delay = lle->lle_remtime;
646 			lle->lle_remtime = 0;
647 		}
648 
649 		if (delay == 0) {
650 
651 			/*
652 			 * The original ng6_gctime timeout ended,
653 			 * no more rescheduling.
654 			 */
655 			return (0);
656 		}
657 
658 		*pdelay = delay;
659 		return (1);
660 	}
661 
662 	/*
663 	 * Packet received. Verify timestamp
664 	 */
665 	delay = (long)(time_uptime - lle_hittime);
666 	if (delay < nd_delay) {
667 
668 		/*
669 		 * V_nd6_delay still not passed since the first
670 		 * hit in STALE state.
671 		 * Reshedule timer and return.
672 		 */
673 		*pdelay = (long)(nd_delay - delay) * hz;
674 		return (1);
675 	}
676 
677 	/* Request switching to probe */
678 	*do_switch = 1;
679 	return (0);
680 }
681 
682 
683 /*
684  * Switch @lle state to new state optionally arming timers.
685  *
686  * Set noinline to be dtrace-friendly
687  */
688 __noinline void
689 nd6_llinfo_setstate(struct llentry *lle, int newstate)
690 {
691 	struct ifnet *ifp;
692 	int nd_gctimer, nd_delay;
693 	long delay, remtime;
694 
695 	delay = 0;
696 	remtime = 0;
697 
698 	switch (newstate) {
699 	case ND6_LLINFO_INCOMPLETE:
700 		ifp = lle->lle_tbl->llt_ifp;
701 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
702 		break;
703 	case ND6_LLINFO_REACHABLE:
704 		if (!ND6_LLINFO_PERMANENT(lle)) {
705 			ifp = lle->lle_tbl->llt_ifp;
706 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
707 		}
708 		break;
709 	case ND6_LLINFO_STALE:
710 
711 		/*
712 		 * Notify fast path that we want to know if any packet
713 		 * is transmitted by setting r_skip_req.
714 		 */
715 		LLE_REQ_LOCK(lle);
716 		lle->r_skip_req = 1;
717 		LLE_REQ_UNLOCK(lle);
718 		nd_delay = V_nd6_delay;
719 		nd_gctimer = V_nd6_gctimer;
720 
721 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
722 		remtime = (long)nd_gctimer * hz - delay;
723 		break;
724 	case ND6_LLINFO_DELAY:
725 		lle->la_asked = 0;
726 		delay = (long)V_nd6_delay * hz;
727 		break;
728 	}
729 
730 	if (delay > 0)
731 		nd6_llinfo_settimer_locked(lle, delay);
732 
733 	lle->lle_remtime = remtime;
734 	lle->ln_state = newstate;
735 }
736 
737 /*
738  * Timer-dependent part of nd state machine.
739  *
740  * Set noinline to be dtrace-friendly
741  */
742 static __noinline void
743 nd6_llinfo_timer(void *arg)
744 {
745 	struct llentry *ln;
746 	struct in6_addr *dst, *pdst, *psrc, src;
747 	struct ifnet *ifp;
748 	struct nd_ifinfo *ndi;
749 	int do_switch, send_ns;
750 	long delay;
751 
752 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
753 	ln = (struct llentry *)arg;
754 	ifp = lltable_get_ifp(ln->lle_tbl);
755 	CURVNET_SET(ifp->if_vnet);
756 
757 	ND6_RLOCK();
758 	LLE_WLOCK(ln);
759 	if (callout_pending(&ln->lle_timer)) {
760 		/*
761 		 * Here we are a bit odd here in the treatment of
762 		 * active/pending. If the pending bit is set, it got
763 		 * rescheduled before I ran. The active
764 		 * bit we ignore, since if it was stopped
765 		 * in ll_tablefree() and was currently running
766 		 * it would have return 0 so the code would
767 		 * not have deleted it since the callout could
768 		 * not be stopped so we want to go through
769 		 * with the delete here now. If the callout
770 		 * was restarted, the pending bit will be back on and
771 		 * we just want to bail since the callout_reset would
772 		 * return 1 and our reference would have been removed
773 		 * by nd6_llinfo_settimer_locked above since canceled
774 		 * would have been 1.
775 		 */
776 		LLE_WUNLOCK(ln);
777 		ND6_RUNLOCK();
778 		CURVNET_RESTORE();
779 		return;
780 	}
781 	ndi = ND_IFINFO(ifp);
782 	send_ns = 0;
783 	dst = &ln->r_l3addr.addr6;
784 	pdst = dst;
785 
786 	if (ln->ln_ntick > 0) {
787 		if (ln->ln_ntick > INT_MAX) {
788 			ln->ln_ntick -= INT_MAX;
789 			nd6_llinfo_settimer_locked(ln, INT_MAX);
790 		} else {
791 			ln->ln_ntick = 0;
792 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
793 		}
794 		goto done;
795 	}
796 
797 	if (ln->la_flags & LLE_STATIC) {
798 		goto done;
799 	}
800 
801 	if (ln->la_flags & LLE_DELETED) {
802 		nd6_free(&ln, 0);
803 		goto done;
804 	}
805 
806 	switch (ln->ln_state) {
807 	case ND6_LLINFO_INCOMPLETE:
808 		if (ln->la_asked < V_nd6_mmaxtries) {
809 			ln->la_asked++;
810 			send_ns = 1;
811 			/* Send NS to multicast address */
812 			pdst = NULL;
813 		} else {
814 			struct mbuf *m = ln->la_hold;
815 			if (m) {
816 				struct mbuf *m0;
817 
818 				/*
819 				 * assuming every packet in la_hold has the
820 				 * same IP header.  Send error after unlock.
821 				 */
822 				m0 = m->m_nextpkt;
823 				m->m_nextpkt = NULL;
824 				ln->la_hold = m0;
825 				clear_llinfo_pqueue(ln);
826 			}
827 			nd6_free(&ln, 0);
828 			if (m != NULL)
829 				icmp6_error2(m, ICMP6_DST_UNREACH,
830 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
831 		}
832 		break;
833 	case ND6_LLINFO_REACHABLE:
834 		if (!ND6_LLINFO_PERMANENT(ln))
835 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
836 		break;
837 
838 	case ND6_LLINFO_STALE:
839 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
840 
841 			/*
842 			 * No packet has used this entry and GC timeout
843 			 * has not been passed. Reshedule timer and
844 			 * return.
845 			 */
846 			nd6_llinfo_settimer_locked(ln, delay);
847 			break;
848 		}
849 
850 		if (do_switch == 0) {
851 
852 			/*
853 			 * GC timer has ended and entry hasn't been used.
854 			 * Run Garbage collector (RFC 4861, 5.3)
855 			 */
856 			if (!ND6_LLINFO_PERMANENT(ln))
857 				nd6_free(&ln, 1);
858 			break;
859 		}
860 
861 		/* Entry has been used AND delay timer has ended. */
862 
863 		/* FALLTHROUGH */
864 
865 	case ND6_LLINFO_DELAY:
866 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
867 			/* We need NUD */
868 			ln->la_asked = 1;
869 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
870 			send_ns = 1;
871 		} else
872 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
873 		break;
874 	case ND6_LLINFO_PROBE:
875 		if (ln->la_asked < V_nd6_umaxtries) {
876 			ln->la_asked++;
877 			send_ns = 1;
878 		} else {
879 			nd6_free(&ln, 0);
880 		}
881 		break;
882 	default:
883 		panic("%s: paths in a dark night can be confusing: %d",
884 		    __func__, ln->ln_state);
885 	}
886 done:
887 	if (ln != NULL)
888 		ND6_RUNLOCK();
889 	if (send_ns != 0) {
890 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
891 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
892 		LLE_FREE_LOCKED(ln);
893 		ln = NULL;
894 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
895 	}
896 
897 	if (ln != NULL)
898 		LLE_FREE_LOCKED(ln);
899 	CURVNET_RESTORE();
900 }
901 
902 
903 /*
904  * ND6 timer routine to expire default route list and prefix list
905  */
906 void
907 nd6_timer(void *arg)
908 {
909 	CURVNET_SET((struct vnet *) arg);
910 	struct nd_drhead drq;
911 	struct nd_prhead prl;
912 	struct nd_defrouter *dr, *ndr;
913 	struct nd_prefix *pr, *npr;
914 	struct in6_ifaddr *ia6, *nia6;
915 	uint64_t genid;
916 
917 	TAILQ_INIT(&drq);
918 	LIST_INIT(&prl);
919 
920 	ND6_WLOCK();
921 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
922 		if (dr->expire && dr->expire < time_uptime)
923 			defrouter_unlink(dr, &drq);
924 	ND6_WUNLOCK();
925 
926 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
927 		TAILQ_REMOVE(&drq, dr, dr_entry);
928 		defrouter_del(dr);
929 	}
930 
931 	/*
932 	 * expire interface addresses.
933 	 * in the past the loop was inside prefix expiry processing.
934 	 * However, from a stricter speci-confrmance standpoint, we should
935 	 * rather separate address lifetimes and prefix lifetimes.
936 	 *
937 	 * XXXRW: in6_ifaddrhead locking.
938 	 */
939   addrloop:
940 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
941 		/* check address lifetime */
942 		if (IFA6_IS_INVALID(ia6)) {
943 			int regen = 0;
944 
945 			/*
946 			 * If the expiring address is temporary, try
947 			 * regenerating a new one.  This would be useful when
948 			 * we suspended a laptop PC, then turned it on after a
949 			 * period that could invalidate all temporary
950 			 * addresses.  Although we may have to restart the
951 			 * loop (see below), it must be after purging the
952 			 * address.  Otherwise, we'd see an infinite loop of
953 			 * regeneration.
954 			 */
955 			if (V_ip6_use_tempaddr &&
956 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
957 				if (regen_tmpaddr(ia6) == 0)
958 					regen = 1;
959 			}
960 
961 			in6_purgeaddr(&ia6->ia_ifa);
962 
963 			if (regen)
964 				goto addrloop; /* XXX: see below */
965 		} else if (IFA6_IS_DEPRECATED(ia6)) {
966 			int oldflags = ia6->ia6_flags;
967 
968 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
969 
970 			/*
971 			 * If a temporary address has just become deprecated,
972 			 * regenerate a new one if possible.
973 			 */
974 			if (V_ip6_use_tempaddr &&
975 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
976 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
977 
978 				if (regen_tmpaddr(ia6) == 0) {
979 					/*
980 					 * A new temporary address is
981 					 * generated.
982 					 * XXX: this means the address chain
983 					 * has changed while we are still in
984 					 * the loop.  Although the change
985 					 * would not cause disaster (because
986 					 * it's not a deletion, but an
987 					 * addition,) we'd rather restart the
988 					 * loop just for safety.  Or does this
989 					 * significantly reduce performance??
990 					 */
991 					goto addrloop;
992 				}
993 			}
994 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
995 			/*
996 			 * Schedule DAD for a tentative address.  This happens
997 			 * if the interface was down or not running
998 			 * when the address was configured.
999 			 */
1000 			int delay;
1001 
1002 			delay = arc4random() %
1003 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1004 			nd6_dad_start((struct ifaddr *)ia6, delay);
1005 		} else {
1006 			/*
1007 			 * Check status of the interface.  If it is down,
1008 			 * mark the address as tentative for future DAD.
1009 			 */
1010 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
1011 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
1012 				== 0 ||
1013 			    (ND_IFINFO(ia6->ia_ifp)->flags &
1014 				ND6_IFF_IFDISABLED) != 0) {
1015 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1016 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1017 			}
1018 			/*
1019 			 * A new RA might have made a deprecated address
1020 			 * preferred.
1021 			 */
1022 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1023 		}
1024 	}
1025 
1026 	ND6_WLOCK();
1027 restart:
1028 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1029 		/*
1030 		 * Expire prefixes. Since the pltime is only used for
1031 		 * autoconfigured addresses, pltime processing for prefixes is
1032 		 * not necessary.
1033 		 *
1034 		 * Only unlink after all derived addresses have expired. This
1035 		 * may not occur until two hours after the prefix has expired
1036 		 * per RFC 4862. If the prefix expires before its derived
1037 		 * addresses, mark it off-link. This will be done automatically
1038 		 * after unlinking if no address references remain.
1039 		 */
1040 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1041 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1042 			continue;
1043 
1044 		if (pr->ndpr_addrcnt == 0) {
1045 			nd6_prefix_unlink(pr, &prl);
1046 			continue;
1047 		}
1048 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1049 			genid = V_nd6_list_genid;
1050 			nd6_prefix_ref(pr);
1051 			ND6_WUNLOCK();
1052 			ND6_ONLINK_LOCK();
1053 			(void)nd6_prefix_offlink(pr);
1054 			ND6_ONLINK_UNLOCK();
1055 			ND6_WLOCK();
1056 			nd6_prefix_rele(pr);
1057 			if (genid != V_nd6_list_genid)
1058 				goto restart;
1059 		}
1060 	}
1061 	ND6_WUNLOCK();
1062 
1063 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1064 		LIST_REMOVE(pr, ndpr_entry);
1065 		nd6_prefix_del(pr);
1066 	}
1067 
1068 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1069 	    nd6_timer, curvnet);
1070 
1071 	CURVNET_RESTORE();
1072 }
1073 
1074 /*
1075  * ia6 - deprecated/invalidated temporary address
1076  */
1077 static int
1078 regen_tmpaddr(struct in6_ifaddr *ia6)
1079 {
1080 	struct ifaddr *ifa;
1081 	struct ifnet *ifp;
1082 	struct in6_ifaddr *public_ifa6 = NULL;
1083 
1084 	ifp = ia6->ia_ifa.ifa_ifp;
1085 	IF_ADDR_RLOCK(ifp);
1086 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1087 		struct in6_ifaddr *it6;
1088 
1089 		if (ifa->ifa_addr->sa_family != AF_INET6)
1090 			continue;
1091 
1092 		it6 = (struct in6_ifaddr *)ifa;
1093 
1094 		/* ignore no autoconf addresses. */
1095 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1096 			continue;
1097 
1098 		/* ignore autoconf addresses with different prefixes. */
1099 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1100 			continue;
1101 
1102 		/*
1103 		 * Now we are looking at an autoconf address with the same
1104 		 * prefix as ours.  If the address is temporary and is still
1105 		 * preferred, do not create another one.  It would be rare, but
1106 		 * could happen, for example, when we resume a laptop PC after
1107 		 * a long period.
1108 		 */
1109 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1110 		    !IFA6_IS_DEPRECATED(it6)) {
1111 			public_ifa6 = NULL;
1112 			break;
1113 		}
1114 
1115 		/*
1116 		 * This is a public autoconf address that has the same prefix
1117 		 * as ours.  If it is preferred, keep it.  We can't break the
1118 		 * loop here, because there may be a still-preferred temporary
1119 		 * address with the prefix.
1120 		 */
1121 		if (!IFA6_IS_DEPRECATED(it6))
1122 			public_ifa6 = it6;
1123 	}
1124 	if (public_ifa6 != NULL)
1125 		ifa_ref(&public_ifa6->ia_ifa);
1126 	IF_ADDR_RUNLOCK(ifp);
1127 
1128 	if (public_ifa6 != NULL) {
1129 		int e;
1130 
1131 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1132 			ifa_free(&public_ifa6->ia_ifa);
1133 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1134 			    " tmp addr,errno=%d\n", e);
1135 			return (-1);
1136 		}
1137 		ifa_free(&public_ifa6->ia_ifa);
1138 		return (0);
1139 	}
1140 
1141 	return (-1);
1142 }
1143 
1144 /*
1145  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1146  * cache entries are freed in in6_domifdetach().
1147  */
1148 void
1149 nd6_purge(struct ifnet *ifp)
1150 {
1151 	struct nd_drhead drq;
1152 	struct nd_prhead prl;
1153 	struct nd_defrouter *dr, *ndr;
1154 	struct nd_prefix *pr, *npr;
1155 
1156 	TAILQ_INIT(&drq);
1157 	LIST_INIT(&prl);
1158 
1159 	/*
1160 	 * Nuke default router list entries toward ifp.
1161 	 * We defer removal of default router list entries that is installed
1162 	 * in the routing table, in order to keep additional side effects as
1163 	 * small as possible.
1164 	 */
1165 	ND6_WLOCK();
1166 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1167 		if (dr->installed)
1168 			continue;
1169 		if (dr->ifp == ifp)
1170 			defrouter_unlink(dr, &drq);
1171 	}
1172 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1173 		if (!dr->installed)
1174 			continue;
1175 		if (dr->ifp == ifp)
1176 			defrouter_unlink(dr, &drq);
1177 	}
1178 
1179 	/*
1180 	 * Remove prefixes on ifp. We should have already removed addresses on
1181 	 * this interface, so no addresses should be referencing these prefixes.
1182 	 */
1183 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1184 		if (pr->ndpr_ifp == ifp)
1185 			nd6_prefix_unlink(pr, &prl);
1186 	}
1187 	ND6_WUNLOCK();
1188 
1189 	/* Delete the unlinked router and prefix objects. */
1190 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1191 		TAILQ_REMOVE(&drq, dr, dr_entry);
1192 		defrouter_del(dr);
1193 	}
1194 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1195 		LIST_REMOVE(pr, ndpr_entry);
1196 		nd6_prefix_del(pr);
1197 	}
1198 
1199 	/* cancel default outgoing interface setting */
1200 	if (V_nd6_defifindex == ifp->if_index)
1201 		nd6_setdefaultiface(0);
1202 
1203 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1204 		/* Refresh default router list. */
1205 		defrouter_select_fib(ifp->if_fib);
1206 	}
1207 }
1208 
1209 /*
1210  * the caller acquires and releases the lock on the lltbls
1211  * Returns the llentry locked
1212  */
1213 struct llentry *
1214 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1215 {
1216 	struct sockaddr_in6 sin6;
1217 	struct llentry *ln;
1218 
1219 	bzero(&sin6, sizeof(sin6));
1220 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1221 	sin6.sin6_family = AF_INET6;
1222 	sin6.sin6_addr = *addr6;
1223 
1224 	IF_AFDATA_LOCK_ASSERT(ifp);
1225 
1226 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1227 
1228 	return (ln);
1229 }
1230 
1231 struct llentry *
1232 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1233 {
1234 	struct sockaddr_in6 sin6;
1235 	struct llentry *ln;
1236 
1237 	bzero(&sin6, sizeof(sin6));
1238 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1239 	sin6.sin6_family = AF_INET6;
1240 	sin6.sin6_addr = *addr6;
1241 
1242 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1243 	if (ln != NULL)
1244 		ln->ln_state = ND6_LLINFO_NOSTATE;
1245 
1246 	return (ln);
1247 }
1248 
1249 /*
1250  * Test whether a given IPv6 address is a neighbor or not, ignoring
1251  * the actual neighbor cache.  The neighbor cache is ignored in order
1252  * to not reenter the routing code from within itself.
1253  */
1254 static int
1255 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1256 {
1257 	struct nd_prefix *pr;
1258 	struct ifaddr *ifa;
1259 	struct rt_addrinfo info;
1260 	struct sockaddr_in6 rt_key;
1261 	const struct sockaddr *dst6;
1262 	uint64_t genid;
1263 	int error, fibnum;
1264 
1265 	/*
1266 	 * A link-local address is always a neighbor.
1267 	 * XXX: a link does not necessarily specify a single interface.
1268 	 */
1269 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1270 		struct sockaddr_in6 sin6_copy;
1271 		u_int32_t zone;
1272 
1273 		/*
1274 		 * We need sin6_copy since sa6_recoverscope() may modify the
1275 		 * content (XXX).
1276 		 */
1277 		sin6_copy = *addr;
1278 		if (sa6_recoverscope(&sin6_copy))
1279 			return (0); /* XXX: should be impossible */
1280 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1281 			return (0);
1282 		if (sin6_copy.sin6_scope_id == zone)
1283 			return (1);
1284 		else
1285 			return (0);
1286 	}
1287 
1288 	bzero(&rt_key, sizeof(rt_key));
1289 	bzero(&info, sizeof(info));
1290 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1291 
1292 	/*
1293 	 * If the address matches one of our addresses,
1294 	 * it should be a neighbor.
1295 	 * If the address matches one of our on-link prefixes, it should be a
1296 	 * neighbor.
1297 	 */
1298 	ND6_RLOCK();
1299 restart:
1300 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1301 		if (pr->ndpr_ifp != ifp)
1302 			continue;
1303 
1304 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1305 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1306 
1307 			/*
1308 			 * We only need to check all FIBs if add_addr_allfibs
1309 			 * is unset. If set, checking any FIB will suffice.
1310 			 */
1311 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1312 			for (; fibnum < rt_numfibs; fibnum++) {
1313 				genid = V_nd6_list_genid;
1314 				ND6_RUNLOCK();
1315 
1316 				/*
1317 				 * Restore length field before
1318 				 * retrying lookup
1319 				 */
1320 				rt_key.sin6_len = sizeof(rt_key);
1321 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1322 						        &info);
1323 
1324 				ND6_RLOCK();
1325 				if (genid != V_nd6_list_genid)
1326 					goto restart;
1327 				if (error == 0)
1328 					break;
1329 			}
1330 			if (error != 0)
1331 				continue;
1332 
1333 			/*
1334 			 * This is the case where multiple interfaces
1335 			 * have the same prefix, but only one is installed
1336 			 * into the routing table and that prefix entry
1337 			 * is not the one being examined here. In the case
1338 			 * where RADIX_MPATH is enabled, multiple route
1339 			 * entries (of the same rt_key value) will be
1340 			 * installed because the interface addresses all
1341 			 * differ.
1342 			 */
1343 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1344 			    &rt_key.sin6_addr))
1345 				continue;
1346 		}
1347 
1348 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1349 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1350 			ND6_RUNLOCK();
1351 			return (1);
1352 		}
1353 	}
1354 	ND6_RUNLOCK();
1355 
1356 	/*
1357 	 * If the address is assigned on the node of the other side of
1358 	 * a p2p interface, the address should be a neighbor.
1359 	 */
1360 	if (ifp->if_flags & IFF_POINTOPOINT) {
1361 		IF_ADDR_RLOCK(ifp);
1362 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1363 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1364 				continue;
1365 			if (ifa->ifa_dstaddr != NULL &&
1366 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1367 				IF_ADDR_RUNLOCK(ifp);
1368 				return 1;
1369 			}
1370 		}
1371 		IF_ADDR_RUNLOCK(ifp);
1372 	}
1373 
1374 	/*
1375 	 * If the default router list is empty, all addresses are regarded
1376 	 * as on-link, and thus, as a neighbor.
1377 	 */
1378 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1379 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1380 	    V_nd6_defifindex == ifp->if_index) {
1381 		return (1);
1382 	}
1383 
1384 	return (0);
1385 }
1386 
1387 
1388 /*
1389  * Detect if a given IPv6 address identifies a neighbor on a given link.
1390  * XXX: should take care of the destination of a p2p link?
1391  */
1392 int
1393 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1394 {
1395 	struct llentry *lle;
1396 	int rc = 0;
1397 
1398 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1399 	if (nd6_is_new_addr_neighbor(addr, ifp))
1400 		return (1);
1401 
1402 	/*
1403 	 * Even if the address matches none of our addresses, it might be
1404 	 * in the neighbor cache.
1405 	 */
1406 	IF_AFDATA_RLOCK(ifp);
1407 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1408 		LLE_RUNLOCK(lle);
1409 		rc = 1;
1410 	}
1411 	IF_AFDATA_RUNLOCK(ifp);
1412 	return (rc);
1413 }
1414 
1415 /*
1416  * Free an nd6 llinfo entry.
1417  * Since the function would cause significant changes in the kernel, DO NOT
1418  * make it global, unless you have a strong reason for the change, and are sure
1419  * that the change is safe.
1420  *
1421  * Set noinline to be dtrace-friendly
1422  */
1423 static __noinline void
1424 nd6_free(struct llentry **lnp, int gc)
1425 {
1426 	struct ifnet *ifp;
1427 	struct llentry *ln;
1428 	struct nd_defrouter *dr;
1429 
1430 	ln = *lnp;
1431 	*lnp = NULL;
1432 
1433 	LLE_WLOCK_ASSERT(ln);
1434 	ND6_RLOCK_ASSERT();
1435 
1436 	ifp = lltable_get_ifp(ln->lle_tbl);
1437 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1438 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1439 	else
1440 		dr = NULL;
1441 	ND6_RUNLOCK();
1442 
1443 	if ((ln->la_flags & LLE_DELETED) == 0)
1444 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1445 
1446 	/*
1447 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1448 	 * even though it is not harmful, it was not really necessary.
1449 	 */
1450 
1451 	/* cancel timer */
1452 	nd6_llinfo_settimer_locked(ln, -1);
1453 
1454 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1455 		if (dr != NULL && dr->expire &&
1456 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1457 			/*
1458 			 * If the reason for the deletion is just garbage
1459 			 * collection, and the neighbor is an active default
1460 			 * router, do not delete it.  Instead, reset the GC
1461 			 * timer using the router's lifetime.
1462 			 * Simply deleting the entry would affect default
1463 			 * router selection, which is not necessarily a good
1464 			 * thing, especially when we're using router preference
1465 			 * values.
1466 			 * XXX: the check for ln_state would be redundant,
1467 			 *      but we intentionally keep it just in case.
1468 			 */
1469 			if (dr->expire > time_uptime)
1470 				nd6_llinfo_settimer_locked(ln,
1471 				    (dr->expire - time_uptime) * hz);
1472 			else
1473 				nd6_llinfo_settimer_locked(ln,
1474 				    (long)V_nd6_gctimer * hz);
1475 
1476 			LLE_REMREF(ln);
1477 			LLE_WUNLOCK(ln);
1478 			defrouter_rele(dr);
1479 			return;
1480 		}
1481 
1482 		if (dr) {
1483 			/*
1484 			 * Unreachablity of a router might affect the default
1485 			 * router selection and on-link detection of advertised
1486 			 * prefixes.
1487 			 */
1488 
1489 			/*
1490 			 * Temporarily fake the state to choose a new default
1491 			 * router and to perform on-link determination of
1492 			 * prefixes correctly.
1493 			 * Below the state will be set correctly,
1494 			 * or the entry itself will be deleted.
1495 			 */
1496 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1497 		}
1498 
1499 		if (ln->ln_router || dr) {
1500 
1501 			/*
1502 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1503 			 * rnh and for the calls to pfxlist_onlink_check() and
1504 			 * defrouter_select_fib() in the block further down for calls
1505 			 * into nd6_lookup().  We still hold a ref.
1506 			 */
1507 			LLE_WUNLOCK(ln);
1508 
1509 			/*
1510 			 * rt6_flush must be called whether or not the neighbor
1511 			 * is in the Default Router List.
1512 			 * See a corresponding comment in nd6_na_input().
1513 			 */
1514 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1515 		}
1516 
1517 		if (dr) {
1518 			/*
1519 			 * Since defrouter_select_fib() does not affect the
1520 			 * on-link determination and MIP6 needs the check
1521 			 * before the default router selection, we perform
1522 			 * the check now.
1523 			 */
1524 			pfxlist_onlink_check();
1525 
1526 			/*
1527 			 * Refresh default router list.
1528 			 */
1529 			defrouter_select_fib(dr->ifp->if_fib);
1530 		}
1531 
1532 		/*
1533 		 * If this entry was added by an on-link redirect, remove the
1534 		 * corresponding host route.
1535 		 */
1536 		if (ln->la_flags & LLE_REDIRECT)
1537 			nd6_free_redirect(ln);
1538 
1539 		if (ln->ln_router || dr)
1540 			LLE_WLOCK(ln);
1541 	}
1542 
1543 	/*
1544 	 * Save to unlock. We still hold an extra reference and will not
1545 	 * free(9) in llentry_free() if someone else holds one as well.
1546 	 */
1547 	LLE_WUNLOCK(ln);
1548 	IF_AFDATA_LOCK(ifp);
1549 	LLE_WLOCK(ln);
1550 	/* Guard against race with other llentry_free(). */
1551 	if (ln->la_flags & LLE_LINKED) {
1552 		/* Remove callout reference */
1553 		LLE_REMREF(ln);
1554 		lltable_unlink_entry(ln->lle_tbl, ln);
1555 	}
1556 	IF_AFDATA_UNLOCK(ifp);
1557 
1558 	llentry_free(ln);
1559 	if (dr != NULL)
1560 		defrouter_rele(dr);
1561 }
1562 
1563 static int
1564 nd6_isdynrte(const struct rtentry *rt, void *xap)
1565 {
1566 
1567 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1568 		return (1);
1569 
1570 	return (0);
1571 }
1572 /*
1573  * Remove the rtentry for the given llentry,
1574  * both of which were installed by a redirect.
1575  */
1576 static void
1577 nd6_free_redirect(const struct llentry *ln)
1578 {
1579 	int fibnum;
1580 	struct sockaddr_in6 sin6;
1581 	struct rt_addrinfo info;
1582 
1583 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1584 	memset(&info, 0, sizeof(info));
1585 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1586 	info.rti_filter = nd6_isdynrte;
1587 
1588 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1589 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1590 }
1591 
1592 /*
1593  * Rejuvenate this function for routing operations related
1594  * processing.
1595  */
1596 void
1597 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1598 {
1599 	struct sockaddr_in6 *gateway;
1600 	struct nd_defrouter *dr;
1601 	struct ifnet *ifp;
1602 
1603 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1604 	ifp = rt->rt_ifp;
1605 
1606 	switch (req) {
1607 	case RTM_ADD:
1608 		break;
1609 
1610 	case RTM_DELETE:
1611 		if (!ifp)
1612 			return;
1613 		/*
1614 		 * Only indirect routes are interesting.
1615 		 */
1616 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1617 			return;
1618 		/*
1619 		 * check for default route
1620 		 */
1621 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1622 		    &SIN6(rt_key(rt))->sin6_addr)) {
1623 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1624 			if (dr != NULL) {
1625 				dr->installed = 0;
1626 				defrouter_rele(dr);
1627 			}
1628 		}
1629 		break;
1630 	}
1631 }
1632 
1633 
1634 int
1635 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1636 {
1637 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1638 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1639 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1640 	int error = 0;
1641 
1642 	if (ifp->if_afdata[AF_INET6] == NULL)
1643 		return (EPFNOSUPPORT);
1644 	switch (cmd) {
1645 	case OSIOCGIFINFO_IN6:
1646 #define ND	ndi->ndi
1647 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1648 		bzero(&ND, sizeof(ND));
1649 		ND.linkmtu = IN6_LINKMTU(ifp);
1650 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1651 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1652 		ND.reachable = ND_IFINFO(ifp)->reachable;
1653 		ND.retrans = ND_IFINFO(ifp)->retrans;
1654 		ND.flags = ND_IFINFO(ifp)->flags;
1655 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1656 		ND.chlim = ND_IFINFO(ifp)->chlim;
1657 		break;
1658 	case SIOCGIFINFO_IN6:
1659 		ND = *ND_IFINFO(ifp);
1660 		break;
1661 	case SIOCSIFINFO_IN6:
1662 		/*
1663 		 * used to change host variables from userland.
1664 		 * intended for a use on router to reflect RA configurations.
1665 		 */
1666 		/* 0 means 'unspecified' */
1667 		if (ND.linkmtu != 0) {
1668 			if (ND.linkmtu < IPV6_MMTU ||
1669 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1670 				error = EINVAL;
1671 				break;
1672 			}
1673 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1674 		}
1675 
1676 		if (ND.basereachable != 0) {
1677 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1678 
1679 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1680 			if (ND.basereachable != obasereachable)
1681 				ND_IFINFO(ifp)->reachable =
1682 				    ND_COMPUTE_RTIME(ND.basereachable);
1683 		}
1684 		if (ND.retrans != 0)
1685 			ND_IFINFO(ifp)->retrans = ND.retrans;
1686 		if (ND.chlim != 0)
1687 			ND_IFINFO(ifp)->chlim = ND.chlim;
1688 		/* FALLTHROUGH */
1689 	case SIOCSIFINFO_FLAGS:
1690 	{
1691 		struct ifaddr *ifa;
1692 		struct in6_ifaddr *ia;
1693 
1694 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1695 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1696 			/* ifdisabled 1->0 transision */
1697 
1698 			/*
1699 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1700 			 * has an link-local address with IN6_IFF_DUPLICATED,
1701 			 * do not clear ND6_IFF_IFDISABLED.
1702 			 * See RFC 4862, Section 5.4.5.
1703 			 */
1704 			IF_ADDR_RLOCK(ifp);
1705 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1706 				if (ifa->ifa_addr->sa_family != AF_INET6)
1707 					continue;
1708 				ia = (struct in6_ifaddr *)ifa;
1709 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1710 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1711 					break;
1712 			}
1713 			IF_ADDR_RUNLOCK(ifp);
1714 
1715 			if (ifa != NULL) {
1716 				/* LLA is duplicated. */
1717 				ND.flags |= ND6_IFF_IFDISABLED;
1718 				log(LOG_ERR, "Cannot enable an interface"
1719 				    " with a link-local address marked"
1720 				    " duplicate.\n");
1721 			} else {
1722 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1723 				if (ifp->if_flags & IFF_UP)
1724 					in6_if_up(ifp);
1725 			}
1726 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1727 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1728 			/* ifdisabled 0->1 transision */
1729 			/* Mark all IPv6 address as tentative. */
1730 
1731 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1732 			if (V_ip6_dad_count > 0 &&
1733 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1734 				IF_ADDR_RLOCK(ifp);
1735 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1736 				    ifa_link) {
1737 					if (ifa->ifa_addr->sa_family !=
1738 					    AF_INET6)
1739 						continue;
1740 					ia = (struct in6_ifaddr *)ifa;
1741 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1742 				}
1743 				IF_ADDR_RUNLOCK(ifp);
1744 			}
1745 		}
1746 
1747 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1748 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1749 				/* auto_linklocal 0->1 transision */
1750 
1751 				/* If no link-local address on ifp, configure */
1752 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1753 				in6_ifattach(ifp, NULL);
1754 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1755 			    ifp->if_flags & IFF_UP) {
1756 				/*
1757 				 * When the IF already has
1758 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1759 				 * address is assigned, and IFF_UP, try to
1760 				 * assign one.
1761 				 */
1762 				IF_ADDR_RLOCK(ifp);
1763 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1764 				    ifa_link) {
1765 					if (ifa->ifa_addr->sa_family !=
1766 					    AF_INET6)
1767 						continue;
1768 					ia = (struct in6_ifaddr *)ifa;
1769 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1770 						break;
1771 				}
1772 				IF_ADDR_RUNLOCK(ifp);
1773 				if (ifa != NULL)
1774 					/* No LLA is configured. */
1775 					in6_ifattach(ifp, NULL);
1776 			}
1777 		}
1778 	}
1779 		ND_IFINFO(ifp)->flags = ND.flags;
1780 		break;
1781 #undef ND
1782 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1783 		/* sync kernel routing table with the default router list */
1784 		defrouter_reset();
1785 		defrouter_select();
1786 		break;
1787 	case SIOCSPFXFLUSH_IN6:
1788 	{
1789 		/* flush all the prefix advertised by routers */
1790 		struct in6_ifaddr *ia, *ia_next;
1791 		struct nd_prefix *pr, *next;
1792 		struct nd_prhead prl;
1793 
1794 		LIST_INIT(&prl);
1795 
1796 		ND6_WLOCK();
1797 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1798 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1799 				continue; /* XXX */
1800 			nd6_prefix_unlink(pr, &prl);
1801 		}
1802 		ND6_WUNLOCK();
1803 
1804 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1805 			LIST_REMOVE(pr, ndpr_entry);
1806 			/* XXXRW: in6_ifaddrhead locking. */
1807 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1808 			    ia_next) {
1809 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1810 					continue;
1811 
1812 				if (ia->ia6_ndpr == pr)
1813 					in6_purgeaddr(&ia->ia_ifa);
1814 			}
1815 			nd6_prefix_del(pr);
1816 		}
1817 		break;
1818 	}
1819 	case SIOCSRTRFLUSH_IN6:
1820 	{
1821 		/* flush all the default routers */
1822 		struct nd_drhead drq;
1823 		struct nd_defrouter *dr;
1824 
1825 		TAILQ_INIT(&drq);
1826 
1827 		defrouter_reset();
1828 
1829 		ND6_WLOCK();
1830 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1831 			defrouter_unlink(dr, &drq);
1832 		ND6_WUNLOCK();
1833 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1834 			TAILQ_REMOVE(&drq, dr, dr_entry);
1835 			defrouter_del(dr);
1836 		}
1837 
1838 		defrouter_select();
1839 		break;
1840 	}
1841 	case SIOCGNBRINFO_IN6:
1842 	{
1843 		struct llentry *ln;
1844 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1845 
1846 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1847 			return (error);
1848 
1849 		IF_AFDATA_RLOCK(ifp);
1850 		ln = nd6_lookup(&nb_addr, 0, ifp);
1851 		IF_AFDATA_RUNLOCK(ifp);
1852 
1853 		if (ln == NULL) {
1854 			error = EINVAL;
1855 			break;
1856 		}
1857 		nbi->state = ln->ln_state;
1858 		nbi->asked = ln->la_asked;
1859 		nbi->isrouter = ln->ln_router;
1860 		if (ln->la_expire == 0)
1861 			nbi->expire = 0;
1862 		else
1863 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1864 			    (time_second - time_uptime);
1865 		LLE_RUNLOCK(ln);
1866 		break;
1867 	}
1868 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1869 		ndif->ifindex = V_nd6_defifindex;
1870 		break;
1871 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1872 		return (nd6_setdefaultiface(ndif->ifindex));
1873 	}
1874 	return (error);
1875 }
1876 
1877 /*
1878  * Calculates new isRouter value based on provided parameters and
1879  * returns it.
1880  */
1881 static int
1882 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1883     int ln_router)
1884 {
1885 
1886 	/*
1887 	 * ICMP6 type dependent behavior.
1888 	 *
1889 	 * NS: clear IsRouter if new entry
1890 	 * RS: clear IsRouter
1891 	 * RA: set IsRouter if there's lladdr
1892 	 * redir: clear IsRouter if new entry
1893 	 *
1894 	 * RA case, (1):
1895 	 * The spec says that we must set IsRouter in the following cases:
1896 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1897 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1898 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1899 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1900 	 * neighbor cache, this is similar to (6).
1901 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1902 	 *
1903 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1904 	 *							D R
1905 	 *	0	n	n	(1)	c   ?     s
1906 	 *	0	y	n	(2)	c   s     s
1907 	 *	0	n	y	(3)	c   s     s
1908 	 *	0	y	y	(4)	c   s     s
1909 	 *	0	y	y	(5)	c   s     s
1910 	 *	1	--	n	(6) c	c	c s
1911 	 *	1	--	y	(7) c	c   s	c s
1912 	 *
1913 	 *					(c=clear s=set)
1914 	 */
1915 	switch (type & 0xff) {
1916 	case ND_NEIGHBOR_SOLICIT:
1917 		/*
1918 		 * New entry must have is_router flag cleared.
1919 		 */
1920 		if (is_new)					/* (6-7) */
1921 			ln_router = 0;
1922 		break;
1923 	case ND_REDIRECT:
1924 		/*
1925 		 * If the icmp is a redirect to a better router, always set the
1926 		 * is_router flag.  Otherwise, if the entry is newly created,
1927 		 * clear the flag.  [RFC 2461, sec 8.3]
1928 		 */
1929 		if (code == ND_REDIRECT_ROUTER)
1930 			ln_router = 1;
1931 		else {
1932 			if (is_new)				/* (6-7) */
1933 				ln_router = 0;
1934 		}
1935 		break;
1936 	case ND_ROUTER_SOLICIT:
1937 		/*
1938 		 * is_router flag must always be cleared.
1939 		 */
1940 		ln_router = 0;
1941 		break;
1942 	case ND_ROUTER_ADVERT:
1943 		/*
1944 		 * Mark an entry with lladdr as a router.
1945 		 */
1946 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1947 		    (is_new && new_addr)) {			/* (7) */
1948 			ln_router = 1;
1949 		}
1950 		break;
1951 	}
1952 
1953 	return (ln_router);
1954 }
1955 
1956 /*
1957  * Create neighbor cache entry and cache link-layer address,
1958  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1959  *
1960  * type - ICMP6 type
1961  * code - type dependent information
1962  *
1963  */
1964 void
1965 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1966     int lladdrlen, int type, int code)
1967 {
1968 	struct llentry *ln = NULL, *ln_tmp;
1969 	int is_newentry;
1970 	int do_update;
1971 	int olladdr;
1972 	int llchange;
1973 	int flags;
1974 	uint16_t router = 0;
1975 	struct sockaddr_in6 sin6;
1976 	struct mbuf *chain = NULL;
1977 	u_char linkhdr[LLE_MAX_LINKHDR];
1978 	size_t linkhdrsize;
1979 	int lladdr_off;
1980 
1981 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1982 
1983 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1984 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1985 
1986 	/* nothing must be updated for unspecified address */
1987 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1988 		return;
1989 
1990 	/*
1991 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1992 	 * the caller.
1993 	 *
1994 	 * XXX If the link does not have link-layer adderss, what should
1995 	 * we do? (ifp->if_addrlen == 0)
1996 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1997 	 * description on it in NS section (RFC 2461 7.2.3).
1998 	 */
1999 	flags = lladdr ? LLE_EXCLUSIVE : 0;
2000 	IF_AFDATA_RLOCK(ifp);
2001 	ln = nd6_lookup(from, flags, ifp);
2002 	IF_AFDATA_RUNLOCK(ifp);
2003 	is_newentry = 0;
2004 	if (ln == NULL) {
2005 		flags |= LLE_EXCLUSIVE;
2006 		ln = nd6_alloc(from, 0, ifp);
2007 		if (ln == NULL)
2008 			return;
2009 
2010 		/*
2011 		 * Since we already know all the data for the new entry,
2012 		 * fill it before insertion.
2013 		 */
2014 		if (lladdr != NULL) {
2015 			linkhdrsize = sizeof(linkhdr);
2016 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2017 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2018 				return;
2019 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2020 			    lladdr_off);
2021 		}
2022 
2023 		IF_AFDATA_WLOCK(ifp);
2024 		LLE_WLOCK(ln);
2025 		/* Prefer any existing lle over newly-created one */
2026 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
2027 		if (ln_tmp == NULL)
2028 			lltable_link_entry(LLTABLE6(ifp), ln);
2029 		IF_AFDATA_WUNLOCK(ifp);
2030 		if (ln_tmp == NULL) {
2031 			/* No existing lle, mark as new entry (6,7) */
2032 			is_newentry = 1;
2033 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2034 			if (lladdr != NULL)	/* (7) */
2035 				EVENTHANDLER_INVOKE(lle_event, ln,
2036 				    LLENTRY_RESOLVED);
2037 		} else {
2038 			lltable_free_entry(LLTABLE6(ifp), ln);
2039 			ln = ln_tmp;
2040 			ln_tmp = NULL;
2041 		}
2042 	}
2043 	/* do nothing if static ndp is set */
2044 	if ((ln->la_flags & LLE_STATIC)) {
2045 		if (flags & LLE_EXCLUSIVE)
2046 			LLE_WUNLOCK(ln);
2047 		else
2048 			LLE_RUNLOCK(ln);
2049 		return;
2050 	}
2051 
2052 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2053 	if (olladdr && lladdr) {
2054 		llchange = bcmp(lladdr, ln->ll_addr,
2055 		    ifp->if_addrlen);
2056 	} else if (!olladdr && lladdr)
2057 		llchange = 1;
2058 	else
2059 		llchange = 0;
2060 
2061 	/*
2062 	 * newentry olladdr  lladdr  llchange	(*=record)
2063 	 *	0	n	n	--	(1)
2064 	 *	0	y	n	--	(2)
2065 	 *	0	n	y	y	(3) * STALE
2066 	 *	0	y	y	n	(4) *
2067 	 *	0	y	y	y	(5) * STALE
2068 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2069 	 *	1	--	y	--	(7) * STALE
2070 	 */
2071 
2072 	do_update = 0;
2073 	if (is_newentry == 0 && llchange != 0) {
2074 		do_update = 1;	/* (3,5) */
2075 
2076 		/*
2077 		 * Record source link-layer address
2078 		 * XXX is it dependent to ifp->if_type?
2079 		 */
2080 		linkhdrsize = sizeof(linkhdr);
2081 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2082 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2083 			return;
2084 
2085 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2086 		    lladdr_off) == 0) {
2087 			/* Entry was deleted */
2088 			return;
2089 		}
2090 
2091 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2092 
2093 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2094 
2095 		if (ln->la_hold != NULL)
2096 			nd6_grab_holdchain(ln, &chain, &sin6);
2097 	}
2098 
2099 	/* Calculates new router status */
2100 	router = nd6_is_router(type, code, is_newentry, olladdr,
2101 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2102 
2103 	ln->ln_router = router;
2104 	/* Mark non-router redirects with special flag */
2105 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2106 		ln->la_flags |= LLE_REDIRECT;
2107 
2108 	if (flags & LLE_EXCLUSIVE)
2109 		LLE_WUNLOCK(ln);
2110 	else
2111 		LLE_RUNLOCK(ln);
2112 
2113 	if (chain != NULL)
2114 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2115 
2116 	/*
2117 	 * When the link-layer address of a router changes, select the
2118 	 * best router again.  In particular, when the neighbor entry is newly
2119 	 * created, it might affect the selection policy.
2120 	 * Question: can we restrict the first condition to the "is_newentry"
2121 	 * case?
2122 	 * XXX: when we hear an RA from a new router with the link-layer
2123 	 * address option, defrouter_select_fib() is called twice, since
2124 	 * defrtrlist_update called the function as well.  However, I believe
2125 	 * we can compromise the overhead, since it only happens the first
2126 	 * time.
2127 	 * XXX: although defrouter_select_fib() should not have a bad effect
2128 	 * for those are not autoconfigured hosts, we explicitly avoid such
2129 	 * cases for safety.
2130 	 */
2131 	if ((do_update || is_newentry) && router &&
2132 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2133 		/*
2134 		 * guaranteed recursion
2135 		 */
2136 		defrouter_select_fib(ifp->if_fib);
2137 	}
2138 }
2139 
2140 static void
2141 nd6_slowtimo(void *arg)
2142 {
2143 	CURVNET_SET((struct vnet *) arg);
2144 	struct nd_ifinfo *nd6if;
2145 	struct ifnet *ifp;
2146 
2147 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2148 	    nd6_slowtimo, curvnet);
2149 	IFNET_RLOCK_NOSLEEP();
2150 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2151 		if (ifp->if_afdata[AF_INET6] == NULL)
2152 			continue;
2153 		nd6if = ND_IFINFO(ifp);
2154 		if (nd6if->basereachable && /* already initialized */
2155 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2156 			/*
2157 			 * Since reachable time rarely changes by router
2158 			 * advertisements, we SHOULD insure that a new random
2159 			 * value gets recomputed at least once every few hours.
2160 			 * (RFC 2461, 6.3.4)
2161 			 */
2162 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2163 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2164 		}
2165 	}
2166 	IFNET_RUNLOCK_NOSLEEP();
2167 	CURVNET_RESTORE();
2168 }
2169 
2170 void
2171 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2172     struct sockaddr_in6 *sin6)
2173 {
2174 
2175 	LLE_WLOCK_ASSERT(ln);
2176 
2177 	*chain = ln->la_hold;
2178 	ln->la_hold = NULL;
2179 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2180 
2181 	if (ln->ln_state == ND6_LLINFO_STALE) {
2182 
2183 		/*
2184 		 * The first time we send a packet to a
2185 		 * neighbor whose entry is STALE, we have
2186 		 * to change the state to DELAY and a sets
2187 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2188 		 * seconds to ensure do neighbor unreachability
2189 		 * detection on expiration.
2190 		 * (RFC 2461 7.3.3)
2191 		 */
2192 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2193 	}
2194 }
2195 
2196 int
2197 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2198     struct sockaddr_in6 *dst, struct route *ro)
2199 {
2200 	int error;
2201 	int ip6len;
2202 	struct ip6_hdr *ip6;
2203 	struct m_tag *mtag;
2204 
2205 #ifdef MAC
2206 	mac_netinet6_nd6_send(ifp, m);
2207 #endif
2208 
2209 	/*
2210 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2211 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2212 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2213 	 * to be diverted to user space.  When re-injected into the kernel,
2214 	 * send_output() will directly dispatch them to the outgoing interface.
2215 	 */
2216 	if (send_sendso_input_hook != NULL) {
2217 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2218 		if (mtag != NULL) {
2219 			ip6 = mtod(m, struct ip6_hdr *);
2220 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2221 			/* Use the SEND socket */
2222 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2223 			    ip6len);
2224 			/* -1 == no app on SEND socket */
2225 			if (error == 0 || error != -1)
2226 			    return (error);
2227 		}
2228 	}
2229 
2230 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2231 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2232 	    mtod(m, struct ip6_hdr *));
2233 
2234 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2235 		origifp = ifp;
2236 
2237 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2238 	return (error);
2239 }
2240 
2241 /*
2242  * Lookup link headerfor @sa_dst address. Stores found
2243  * data in @desten buffer. Copy of lle ln_flags can be also
2244  * saved in @pflags if @pflags is non-NULL.
2245  *
2246  * If destination LLE does not exists or lle state modification
2247  * is required, call "slow" version.
2248  *
2249  * Return values:
2250  * - 0 on success (address copied to buffer).
2251  * - EWOULDBLOCK (no local error, but address is still unresolved)
2252  * - other errors (alloc failure, etc)
2253  */
2254 int
2255 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2256     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2257     struct llentry **plle)
2258 {
2259 	struct llentry *ln = NULL;
2260 	const struct sockaddr_in6 *dst6;
2261 
2262 	if (pflags != NULL)
2263 		*pflags = 0;
2264 
2265 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2266 
2267 	/* discard the packet if IPv6 operation is disabled on the interface */
2268 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2269 		m_freem(m);
2270 		return (ENETDOWN); /* better error? */
2271 	}
2272 
2273 	if (m != NULL && m->m_flags & M_MCAST) {
2274 		switch (ifp->if_type) {
2275 		case IFT_ETHER:
2276 		case IFT_FDDI:
2277 		case IFT_L2VLAN:
2278 		case IFT_BRIDGE:
2279 		case IFT_ISO88025:
2280 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2281 						 desten);
2282 			return (0);
2283 		default:
2284 			m_freem(m);
2285 			return (EAFNOSUPPORT);
2286 		}
2287 	}
2288 
2289 	IF_AFDATA_RLOCK(ifp);
2290 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2291 	    ifp);
2292 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2293 		/* Entry found, let's copy lle info */
2294 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2295 		if (pflags != NULL)
2296 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2297 		/* Check if we have feedback request from nd6 timer */
2298 		if (ln->r_skip_req != 0) {
2299 			LLE_REQ_LOCK(ln);
2300 			ln->r_skip_req = 0; /* Notify that entry was used */
2301 			ln->lle_hittime = time_uptime;
2302 			LLE_REQ_UNLOCK(ln);
2303 		}
2304 		if (plle) {
2305 			LLE_ADDREF(ln);
2306 			*plle = ln;
2307 			LLE_WUNLOCK(ln);
2308 		}
2309 		IF_AFDATA_RUNLOCK(ifp);
2310 		return (0);
2311 	} else if (plle && ln)
2312 		LLE_WUNLOCK(ln);
2313 	IF_AFDATA_RUNLOCK(ifp);
2314 
2315 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2316 }
2317 
2318 
2319 /*
2320  * Do L2 address resolution for @sa_dst address. Stores found
2321  * address in @desten buffer. Copy of lle ln_flags can be also
2322  * saved in @pflags if @pflags is non-NULL.
2323  *
2324  * Heavy version.
2325  * Function assume that destination LLE does not exist,
2326  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2327  *
2328  * Set noinline to be dtrace-friendly
2329  */
2330 static __noinline int
2331 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2332     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2333     struct llentry **plle)
2334 {
2335 	struct llentry *lle = NULL, *lle_tmp;
2336 	struct in6_addr *psrc, src;
2337 	int send_ns, ll_len;
2338 	char *lladdr;
2339 
2340 	/*
2341 	 * Address resolution or Neighbor Unreachability Detection
2342 	 * for the next hop.
2343 	 * At this point, the destination of the packet must be a unicast
2344 	 * or an anycast address(i.e. not a multicast).
2345 	 */
2346 	if (lle == NULL) {
2347 		IF_AFDATA_RLOCK(ifp);
2348 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2349 		IF_AFDATA_RUNLOCK(ifp);
2350 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2351 			/*
2352 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2353 			 * the condition below is not very efficient.  But we believe
2354 			 * it is tolerable, because this should be a rare case.
2355 			 */
2356 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2357 			if (lle == NULL) {
2358 				char ip6buf[INET6_ADDRSTRLEN];
2359 				log(LOG_DEBUG,
2360 				    "nd6_output: can't allocate llinfo for %s "
2361 				    "(ln=%p)\n",
2362 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2363 				m_freem(m);
2364 				return (ENOBUFS);
2365 			}
2366 
2367 			IF_AFDATA_WLOCK(ifp);
2368 			LLE_WLOCK(lle);
2369 			/* Prefer any existing entry over newly-created one */
2370 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2371 			if (lle_tmp == NULL)
2372 				lltable_link_entry(LLTABLE6(ifp), lle);
2373 			IF_AFDATA_WUNLOCK(ifp);
2374 			if (lle_tmp != NULL) {
2375 				lltable_free_entry(LLTABLE6(ifp), lle);
2376 				lle = lle_tmp;
2377 				lle_tmp = NULL;
2378 			}
2379 		}
2380 	}
2381 	if (lle == NULL) {
2382 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2383 			m_freem(m);
2384 			return (ENOBUFS);
2385 		}
2386 
2387 		if (m != NULL)
2388 			m_freem(m);
2389 		return (ENOBUFS);
2390 	}
2391 
2392 	LLE_WLOCK_ASSERT(lle);
2393 
2394 	/*
2395 	 * The first time we send a packet to a neighbor whose entry is
2396 	 * STALE, we have to change the state to DELAY and a sets a timer to
2397 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2398 	 * neighbor unreachability detection on expiration.
2399 	 * (RFC 2461 7.3.3)
2400 	 */
2401 	if (lle->ln_state == ND6_LLINFO_STALE)
2402 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2403 
2404 	/*
2405 	 * If the neighbor cache entry has a state other than INCOMPLETE
2406 	 * (i.e. its link-layer address is already resolved), just
2407 	 * send the packet.
2408 	 */
2409 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2410 		if (flags & LLE_ADDRONLY) {
2411 			lladdr = lle->ll_addr;
2412 			ll_len = ifp->if_addrlen;
2413 		} else {
2414 			lladdr = lle->r_linkdata;
2415 			ll_len = lle->r_hdrlen;
2416 		}
2417 		bcopy(lladdr, desten, ll_len);
2418 		if (pflags != NULL)
2419 			*pflags = lle->la_flags;
2420 		if (plle) {
2421 			LLE_ADDREF(lle);
2422 			*plle = lle;
2423 		}
2424 		LLE_WUNLOCK(lle);
2425 		return (0);
2426 	}
2427 
2428 	/*
2429 	 * There is a neighbor cache entry, but no ethernet address
2430 	 * response yet.  Append this latest packet to the end of the
2431 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2432 	 * the oldest packet in the queue will be removed.
2433 	 */
2434 
2435 	if (lle->la_hold != NULL) {
2436 		struct mbuf *m_hold;
2437 		int i;
2438 
2439 		i = 0;
2440 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2441 			i++;
2442 			if (m_hold->m_nextpkt == NULL) {
2443 				m_hold->m_nextpkt = m;
2444 				break;
2445 			}
2446 		}
2447 		while (i >= V_nd6_maxqueuelen) {
2448 			m_hold = lle->la_hold;
2449 			lle->la_hold = lle->la_hold->m_nextpkt;
2450 			m_freem(m_hold);
2451 			i--;
2452 		}
2453 	} else {
2454 		lle->la_hold = m;
2455 	}
2456 
2457 	/*
2458 	 * If there has been no NS for the neighbor after entering the
2459 	 * INCOMPLETE state, send the first solicitation.
2460 	 * Note that for newly-created lle la_asked will be 0,
2461 	 * so we will transition from ND6_LLINFO_NOSTATE to
2462 	 * ND6_LLINFO_INCOMPLETE state here.
2463 	 */
2464 	psrc = NULL;
2465 	send_ns = 0;
2466 	if (lle->la_asked == 0) {
2467 		lle->la_asked++;
2468 		send_ns = 1;
2469 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2470 
2471 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2472 	}
2473 	LLE_WUNLOCK(lle);
2474 	if (send_ns != 0)
2475 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2476 
2477 	return (EWOULDBLOCK);
2478 }
2479 
2480 /*
2481  * Do L2 address resolution for @sa_dst address. Stores found
2482  * address in @desten buffer. Copy of lle ln_flags can be also
2483  * saved in @pflags if @pflags is non-NULL.
2484  *
2485  * Return values:
2486  * - 0 on success (address copied to buffer).
2487  * - EWOULDBLOCK (no local error, but address is still unresolved)
2488  * - other errors (alloc failure, etc)
2489  */
2490 int
2491 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2492     char *desten, uint32_t *pflags)
2493 {
2494 	int error;
2495 
2496 	flags |= LLE_ADDRONLY;
2497 	error = nd6_resolve_slow(ifp, flags, NULL,
2498 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2499 	return (error);
2500 }
2501 
2502 int
2503 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2504     struct sockaddr_in6 *dst)
2505 {
2506 	struct mbuf *m, *m_head;
2507 	struct ifnet *outifp;
2508 	int error = 0;
2509 
2510 	m_head = chain;
2511 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2512 		outifp = origifp;
2513 	else
2514 		outifp = ifp;
2515 
2516 	while (m_head) {
2517 		m = m_head;
2518 		m_head = m_head->m_nextpkt;
2519 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2520 	}
2521 
2522 	/*
2523 	 * XXX
2524 	 * note that intermediate errors are blindly ignored
2525 	 */
2526 	return (error);
2527 }
2528 
2529 static int
2530 nd6_need_cache(struct ifnet *ifp)
2531 {
2532 	/*
2533 	 * XXX: we currently do not make neighbor cache on any interface
2534 	 * other than ARCnet, Ethernet, FDDI and GIF.
2535 	 *
2536 	 * RFC2893 says:
2537 	 * - unidirectional tunnels needs no ND
2538 	 */
2539 	switch (ifp->if_type) {
2540 	case IFT_ARCNET:
2541 	case IFT_ETHER:
2542 	case IFT_FDDI:
2543 	case IFT_IEEE1394:
2544 	case IFT_L2VLAN:
2545 	case IFT_INFINIBAND:
2546 	case IFT_BRIDGE:
2547 	case IFT_PROPVIRTUAL:
2548 		return (1);
2549 	default:
2550 		return (0);
2551 	}
2552 }
2553 
2554 /*
2555  * Add pernament ND6 link-layer record for given
2556  * interface address.
2557  *
2558  * Very similar to IPv4 arp_ifinit(), but:
2559  * 1) IPv6 DAD is performed in different place
2560  * 2) It is called by IPv6 protocol stack in contrast to
2561  * arp_ifinit() which is typically called in SIOCSIFADDR
2562  * driver ioctl handler.
2563  *
2564  */
2565 int
2566 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2567 {
2568 	struct ifnet *ifp;
2569 	struct llentry *ln, *ln_tmp;
2570 	struct sockaddr *dst;
2571 
2572 	ifp = ia->ia_ifa.ifa_ifp;
2573 	if (nd6_need_cache(ifp) == 0)
2574 		return (0);
2575 
2576 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2577 	dst = (struct sockaddr *)&ia->ia_addr;
2578 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2579 	if (ln == NULL)
2580 		return (ENOBUFS);
2581 
2582 	IF_AFDATA_WLOCK(ifp);
2583 	LLE_WLOCK(ln);
2584 	/* Unlink any entry if exists */
2585 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2586 	if (ln_tmp != NULL)
2587 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2588 	lltable_link_entry(LLTABLE6(ifp), ln);
2589 	IF_AFDATA_WUNLOCK(ifp);
2590 
2591 	if (ln_tmp != NULL)
2592 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2593 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2594 
2595 	LLE_WUNLOCK(ln);
2596 	if (ln_tmp != NULL)
2597 		llentry_free(ln_tmp);
2598 
2599 	return (0);
2600 }
2601 
2602 /*
2603  * Removes either all lle entries for given @ia, or lle
2604  * corresponding to @ia address.
2605  */
2606 void
2607 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2608 {
2609 	struct sockaddr_in6 mask, addr;
2610 	struct sockaddr *saddr, *smask;
2611 	struct ifnet *ifp;
2612 
2613 	ifp = ia->ia_ifa.ifa_ifp;
2614 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2615 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2616 	saddr = (struct sockaddr *)&addr;
2617 	smask = (struct sockaddr *)&mask;
2618 
2619 	if (all != 0)
2620 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2621 	else
2622 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2623 }
2624 
2625 static void
2626 clear_llinfo_pqueue(struct llentry *ln)
2627 {
2628 	struct mbuf *m_hold, *m_hold_next;
2629 
2630 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2631 		m_hold_next = m_hold->m_nextpkt;
2632 		m_freem(m_hold);
2633 	}
2634 
2635 	ln->la_hold = NULL;
2636 }
2637 
2638 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2639 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2640 
2641 SYSCTL_DECL(_net_inet6_icmp6);
2642 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2643 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2644 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2645 	"NDP default router list");
2646 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2647 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2648 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2649 	"NDP prefix list");
2650 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2651 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2652 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2653 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2654 
2655 static int
2656 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2657 {
2658 	struct in6_defrouter d;
2659 	struct nd_defrouter *dr;
2660 	int error;
2661 
2662 	if (req->newptr != NULL)
2663 		return (EPERM);
2664 
2665 	error = sysctl_wire_old_buffer(req, 0);
2666 	if (error != 0)
2667 		return (error);
2668 
2669 	bzero(&d, sizeof(d));
2670 	d.rtaddr.sin6_family = AF_INET6;
2671 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2672 
2673 	ND6_RLOCK();
2674 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2675 		d.rtaddr.sin6_addr = dr->rtaddr;
2676 		error = sa6_recoverscope(&d.rtaddr);
2677 		if (error != 0)
2678 			break;
2679 		d.flags = dr->raflags;
2680 		d.rtlifetime = dr->rtlifetime;
2681 		d.expire = dr->expire + (time_second - time_uptime);
2682 		d.if_index = dr->ifp->if_index;
2683 		error = SYSCTL_OUT(req, &d, sizeof(d));
2684 		if (error != 0)
2685 			break;
2686 	}
2687 	ND6_RUNLOCK();
2688 	return (error);
2689 }
2690 
2691 static int
2692 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2693 {
2694 	struct in6_prefix p;
2695 	struct sockaddr_in6 s6;
2696 	struct nd_prefix *pr;
2697 	struct nd_pfxrouter *pfr;
2698 	time_t maxexpire;
2699 	int error;
2700 	char ip6buf[INET6_ADDRSTRLEN];
2701 
2702 	if (req->newptr)
2703 		return (EPERM);
2704 
2705 	error = sysctl_wire_old_buffer(req, 0);
2706 	if (error != 0)
2707 		return (error);
2708 
2709 	bzero(&p, sizeof(p));
2710 	p.origin = PR_ORIG_RA;
2711 	bzero(&s6, sizeof(s6));
2712 	s6.sin6_family = AF_INET6;
2713 	s6.sin6_len = sizeof(s6);
2714 
2715 	ND6_RLOCK();
2716 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2717 		p.prefix = pr->ndpr_prefix;
2718 		if (sa6_recoverscope(&p.prefix)) {
2719 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2720 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2721 			/* XXX: press on... */
2722 		}
2723 		p.raflags = pr->ndpr_raf;
2724 		p.prefixlen = pr->ndpr_plen;
2725 		p.vltime = pr->ndpr_vltime;
2726 		p.pltime = pr->ndpr_pltime;
2727 		p.if_index = pr->ndpr_ifp->if_index;
2728 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2729 			p.expire = 0;
2730 		else {
2731 			/* XXX: we assume time_t is signed. */
2732 			maxexpire = (-1) &
2733 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2734 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2735 				p.expire = pr->ndpr_lastupdate +
2736 				    pr->ndpr_vltime +
2737 				    (time_second - time_uptime);
2738 			else
2739 				p.expire = maxexpire;
2740 		}
2741 		p.refcnt = pr->ndpr_addrcnt;
2742 		p.flags = pr->ndpr_stateflags;
2743 		p.advrtrs = 0;
2744 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2745 			p.advrtrs++;
2746 		error = SYSCTL_OUT(req, &p, sizeof(p));
2747 		if (error != 0)
2748 			break;
2749 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2750 			s6.sin6_addr = pfr->router->rtaddr;
2751 			if (sa6_recoverscope(&s6))
2752 				log(LOG_ERR,
2753 				    "scope error in prefix list (%s)\n",
2754 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2755 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2756 			if (error != 0)
2757 				goto out;
2758 		}
2759 	}
2760 out:
2761 	ND6_RUNLOCK();
2762 	return (error);
2763 }
2764