xref: /freebsd/sys/netinet6/nd6.c (revision edca4938f74db18d091868237592abbf7e718669)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/eventhandler.h>
43 #include <sys/callout.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/mutex.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/time.h>
51 #include <sys/kernel.h>
52 #include <sys/protosw.h>
53 #include <sys/errno.h>
54 #include <sys/syslog.h>
55 #include <sys/rwlock.h>
56 #include <sys/queue.h>
57 #include <sys/sdt.h>
58 #include <sys/sysctl.h>
59 
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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, ifnet_link_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 VNET_DEFINE_STATIC(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 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
238 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
239 	}
240 }
241 
242 #ifdef VIMAGE
243 void
244 nd6_destroy()
245 {
246 
247 	callout_drain(&V_nd6_slowtimo_ch);
248 	callout_drain(&V_nd6_timer_ch);
249 	if (IS_DEFAULT_VNET(curvnet)) {
250 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
251 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
252 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
253 	}
254 	rw_destroy(&V_nd6_lock);
255 	mtx_destroy(&V_nd6_onlink_mtx);
256 }
257 #endif
258 
259 struct nd_ifinfo *
260 nd6_ifattach(struct ifnet *ifp)
261 {
262 	struct nd_ifinfo *nd;
263 
264 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
265 	nd->initialized = 1;
266 
267 	nd->chlim = IPV6_DEFHLIM;
268 	nd->basereachable = REACHABLE_TIME;
269 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
270 	nd->retrans = RETRANS_TIMER;
271 
272 	nd->flags = ND6_IFF_PERFORMNUD;
273 
274 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
275 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
276 	 * default regardless of the V_ip6_auto_linklocal configuration to
277 	 * give a reasonable default behavior.
278 	 */
279 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
280 	    (ifp->if_flags & IFF_LOOPBACK))
281 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
282 	/*
283 	 * A loopback interface does not need to accept RTADV.
284 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
285 	 * default regardless of the V_ip6_accept_rtadv configuration to
286 	 * prevent the interface from accepting RA messages arrived
287 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
288 	 */
289 	if (V_ip6_accept_rtadv &&
290 	    !(ifp->if_flags & IFF_LOOPBACK) &&
291 	    (ifp->if_type != IFT_BRIDGE))
292 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
293 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
294 		nd->flags |= ND6_IFF_NO_RADR;
295 
296 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
297 	nd6_setmtu0(ifp, nd);
298 
299 	return nd;
300 }
301 
302 void
303 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
304 {
305 	struct epoch_tracker et;
306 	struct ifaddr *ifa, *next;
307 
308 	NET_EPOCH_ENTER(et);
309 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
310 		if (ifa->ifa_addr->sa_family != AF_INET6)
311 			continue;
312 
313 		/* stop DAD processing */
314 		nd6_dad_stop(ifa);
315 	}
316 	NET_EPOCH_EXIT(et);
317 
318 	free(nd, M_IP6NDP);
319 }
320 
321 /*
322  * Reset ND level link MTU. This function is called when the physical MTU
323  * changes, which means we might have to adjust the ND level MTU.
324  */
325 void
326 nd6_setmtu(struct ifnet *ifp)
327 {
328 	if (ifp->if_afdata[AF_INET6] == NULL)
329 		return;
330 
331 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
332 }
333 
334 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
335 void
336 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
337 {
338 	u_int32_t omaxmtu;
339 
340 	omaxmtu = ndi->maxmtu;
341 	ndi->maxmtu = ifp->if_mtu;
342 
343 	/*
344 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
345 	 * undesirable situation.  We thus notify the operator of the change
346 	 * explicitly.  The check for omaxmtu is necessary to restrict the
347 	 * log to the case of changing the MTU, not initializing it.
348 	 */
349 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
350 		log(LOG_NOTICE, "nd6_setmtu0: "
351 		    "new link MTU on %s (%lu) is too small for IPv6\n",
352 		    if_name(ifp), (unsigned long)ndi->maxmtu);
353 	}
354 
355 	if (ndi->maxmtu > V_in6_maxmtu)
356 		in6_setmaxmtu(); /* check all interfaces just in case */
357 
358 }
359 
360 void
361 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
362 {
363 
364 	bzero(ndopts, sizeof(*ndopts));
365 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
366 	ndopts->nd_opts_last
367 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
368 
369 	if (icmp6len == 0) {
370 		ndopts->nd_opts_done = 1;
371 		ndopts->nd_opts_search = NULL;
372 	}
373 }
374 
375 /*
376  * Take one ND option.
377  */
378 struct nd_opt_hdr *
379 nd6_option(union nd_opts *ndopts)
380 {
381 	struct nd_opt_hdr *nd_opt;
382 	int olen;
383 
384 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
385 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
386 	    __func__));
387 	if (ndopts->nd_opts_search == NULL)
388 		return NULL;
389 	if (ndopts->nd_opts_done)
390 		return NULL;
391 
392 	nd_opt = ndopts->nd_opts_search;
393 
394 	/* make sure nd_opt_len is inside the buffer */
395 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
396 		bzero(ndopts, sizeof(*ndopts));
397 		return NULL;
398 	}
399 
400 	olen = nd_opt->nd_opt_len << 3;
401 	if (olen == 0) {
402 		/*
403 		 * Message validation requires that all included
404 		 * options have a length that is greater than zero.
405 		 */
406 		bzero(ndopts, sizeof(*ndopts));
407 		return NULL;
408 	}
409 
410 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
411 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
412 		/* option overruns the end of buffer, invalid */
413 		bzero(ndopts, sizeof(*ndopts));
414 		return NULL;
415 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
416 		/* reached the end of options chain */
417 		ndopts->nd_opts_done = 1;
418 		ndopts->nd_opts_search = NULL;
419 	}
420 	return nd_opt;
421 }
422 
423 /*
424  * Parse multiple ND options.
425  * This function is much easier to use, for ND routines that do not need
426  * multiple options of the same type.
427  */
428 int
429 nd6_options(union nd_opts *ndopts)
430 {
431 	struct nd_opt_hdr *nd_opt;
432 	int i = 0;
433 
434 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
435 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
436 	    __func__));
437 	if (ndopts->nd_opts_search == NULL)
438 		return 0;
439 
440 	while (1) {
441 		nd_opt = nd6_option(ndopts);
442 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
443 			/*
444 			 * Message validation requires that all included
445 			 * options have a length that is greater than zero.
446 			 */
447 			ICMP6STAT_INC(icp6s_nd_badopt);
448 			bzero(ndopts, sizeof(*ndopts));
449 			return -1;
450 		}
451 
452 		if (nd_opt == NULL)
453 			goto skip1;
454 
455 		switch (nd_opt->nd_opt_type) {
456 		case ND_OPT_SOURCE_LINKADDR:
457 		case ND_OPT_TARGET_LINKADDR:
458 		case ND_OPT_MTU:
459 		case ND_OPT_REDIRECTED_HEADER:
460 		case ND_OPT_NONCE:
461 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
462 				nd6log((LOG_INFO,
463 				    "duplicated ND6 option found (type=%d)\n",
464 				    nd_opt->nd_opt_type));
465 				/* XXX bark? */
466 			} else {
467 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
468 					= nd_opt;
469 			}
470 			break;
471 		case ND_OPT_PREFIX_INFORMATION:
472 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
473 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
474 					= nd_opt;
475 			}
476 			ndopts->nd_opts_pi_end =
477 				(struct nd_opt_prefix_info *)nd_opt;
478 			break;
479 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
480 		case ND_OPT_RDNSS:	/* RFC 6106 */
481 		case ND_OPT_DNSSL:	/* RFC 6106 */
482 			/*
483 			 * Silently ignore options we know and do not care about
484 			 * in the kernel.
485 			 */
486 			break;
487 		default:
488 			/*
489 			 * Unknown options must be silently ignored,
490 			 * to accommodate future extension to the protocol.
491 			 */
492 			nd6log((LOG_DEBUG,
493 			    "nd6_options: unsupported option %d - "
494 			    "option ignored\n", nd_opt->nd_opt_type));
495 		}
496 
497 skip1:
498 		i++;
499 		if (i > V_nd6_maxndopt) {
500 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
501 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
502 			break;
503 		}
504 
505 		if (ndopts->nd_opts_done)
506 			break;
507 	}
508 
509 	return 0;
510 }
511 
512 /*
513  * ND6 timer routine to handle ND6 entries
514  */
515 static void
516 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
517 {
518 	int canceled;
519 
520 	LLE_WLOCK_ASSERT(ln);
521 
522 	if (tick < 0) {
523 		ln->la_expire = 0;
524 		ln->ln_ntick = 0;
525 		canceled = callout_stop(&ln->lle_timer);
526 	} else {
527 		ln->la_expire = time_uptime + tick / hz;
528 		LLE_ADDREF(ln);
529 		if (tick > INT_MAX) {
530 			ln->ln_ntick = tick - INT_MAX;
531 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
532 			    nd6_llinfo_timer, ln);
533 		} else {
534 			ln->ln_ntick = 0;
535 			canceled = callout_reset(&ln->lle_timer, tick,
536 			    nd6_llinfo_timer, ln);
537 		}
538 	}
539 	if (canceled > 0)
540 		LLE_REMREF(ln);
541 }
542 
543 /*
544  * Gets source address of the first packet in hold queue
545  * and stores it in @src.
546  * Returns pointer to @src (if hold queue is not empty) or NULL.
547  *
548  * Set noinline to be dtrace-friendly
549  */
550 static __noinline struct in6_addr *
551 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
552 {
553 	struct ip6_hdr hdr;
554 	struct mbuf *m;
555 
556 	if (ln->la_hold == NULL)
557 		return (NULL);
558 
559 	/*
560 	 * assume every packet in la_hold has the same IP header
561 	 */
562 	m = ln->la_hold;
563 	if (sizeof(hdr) > m->m_len)
564 		return (NULL);
565 
566 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
567 	*src = hdr.ip6_src;
568 
569 	return (src);
570 }
571 
572 /*
573  * Checks if we need to switch from STALE state.
574  *
575  * RFC 4861 requires switching from STALE to DELAY state
576  * on first packet matching entry, waiting V_nd6_delay and
577  * transition to PROBE state (if upper layer confirmation was
578  * not received).
579  *
580  * This code performs a bit differently:
581  * On packet hit we don't change state (but desired state
582  * can be guessed by control plane). However, after V_nd6_delay
583  * seconds code will transition to PROBE state (so DELAY state
584  * is kinda skipped in most situations).
585  *
586  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
587  * we perform the following upon entering STALE state:
588  *
589  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
590  * if packet was transmitted at the start of given interval, we
591  * would be able to switch to PROBE state in V_nd6_delay seconds
592  * as user expects.
593  *
594  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
595  * lle in STALE state (remaining timer value stored in lle_remtime).
596  *
597  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
598  * seconds ago.
599  *
600  * Returns non-zero value if the entry is still STALE (storing
601  * the next timer interval in @pdelay).
602  *
603  * Returns zero value if original timer expired or we need to switch to
604  * PROBE (store that in @do_switch variable).
605  */
606 static int
607 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
608 {
609 	int nd_delay, nd_gctimer, r_skip_req;
610 	time_t lle_hittime;
611 	long delay;
612 
613 	*do_switch = 0;
614 	nd_gctimer = V_nd6_gctimer;
615 	nd_delay = V_nd6_delay;
616 
617 	LLE_REQ_LOCK(lle);
618 	r_skip_req = lle->r_skip_req;
619 	lle_hittime = lle->lle_hittime;
620 	LLE_REQ_UNLOCK(lle);
621 
622 	if (r_skip_req > 0) {
623 
624 		/*
625 		 * Nonzero r_skip_req value was set upon entering
626 		 * STALE state. Since value was not changed, no
627 		 * packets were passed using this lle. Ask for
628 		 * timer reschedule and keep STALE state.
629 		 */
630 		delay = (long)(MIN(nd_gctimer, nd_delay));
631 		delay *= hz;
632 		if (lle->lle_remtime > delay)
633 			lle->lle_remtime -= delay;
634 		else {
635 			delay = lle->lle_remtime;
636 			lle->lle_remtime = 0;
637 		}
638 
639 		if (delay == 0) {
640 
641 			/*
642 			 * The original ng6_gctime timeout ended,
643 			 * no more rescheduling.
644 			 */
645 			return (0);
646 		}
647 
648 		*pdelay = delay;
649 		return (1);
650 	}
651 
652 	/*
653 	 * Packet received. Verify timestamp
654 	 */
655 	delay = (long)(time_uptime - lle_hittime);
656 	if (delay < nd_delay) {
657 
658 		/*
659 		 * V_nd6_delay still not passed since the first
660 		 * hit in STALE state.
661 		 * Reshedule timer and return.
662 		 */
663 		*pdelay = (long)(nd_delay - delay) * hz;
664 		return (1);
665 	}
666 
667 	/* Request switching to probe */
668 	*do_switch = 1;
669 	return (0);
670 }
671 
672 
673 /*
674  * Switch @lle state to new state optionally arming timers.
675  *
676  * Set noinline to be dtrace-friendly
677  */
678 __noinline void
679 nd6_llinfo_setstate(struct llentry *lle, int newstate)
680 {
681 	struct ifnet *ifp;
682 	int nd_gctimer, nd_delay;
683 	long delay, remtime;
684 
685 	delay = 0;
686 	remtime = 0;
687 
688 	switch (newstate) {
689 	case ND6_LLINFO_INCOMPLETE:
690 		ifp = lle->lle_tbl->llt_ifp;
691 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
692 		break;
693 	case ND6_LLINFO_REACHABLE:
694 		if (!ND6_LLINFO_PERMANENT(lle)) {
695 			ifp = lle->lle_tbl->llt_ifp;
696 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
697 		}
698 		break;
699 	case ND6_LLINFO_STALE:
700 
701 		/*
702 		 * Notify fast path that we want to know if any packet
703 		 * is transmitted by setting r_skip_req.
704 		 */
705 		LLE_REQ_LOCK(lle);
706 		lle->r_skip_req = 1;
707 		LLE_REQ_UNLOCK(lle);
708 		nd_delay = V_nd6_delay;
709 		nd_gctimer = V_nd6_gctimer;
710 
711 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
712 		remtime = (long)nd_gctimer * hz - delay;
713 		break;
714 	case ND6_LLINFO_DELAY:
715 		lle->la_asked = 0;
716 		delay = (long)V_nd6_delay * hz;
717 		break;
718 	}
719 
720 	if (delay > 0)
721 		nd6_llinfo_settimer_locked(lle, delay);
722 
723 	lle->lle_remtime = remtime;
724 	lle->ln_state = newstate;
725 }
726 
727 /*
728  * Timer-dependent part of nd state machine.
729  *
730  * Set noinline to be dtrace-friendly
731  */
732 static __noinline void
733 nd6_llinfo_timer(void *arg)
734 {
735 	struct epoch_tracker et;
736 	struct llentry *ln;
737 	struct in6_addr *dst, *pdst, *psrc, src;
738 	struct ifnet *ifp;
739 	struct nd_ifinfo *ndi;
740 	int do_switch, send_ns;
741 	long delay;
742 
743 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
744 	ln = (struct llentry *)arg;
745 	ifp = lltable_get_ifp(ln->lle_tbl);
746 	CURVNET_SET(ifp->if_vnet);
747 
748 	ND6_RLOCK();
749 	LLE_WLOCK(ln);
750 	if (callout_pending(&ln->lle_timer)) {
751 		/*
752 		 * Here we are a bit odd here in the treatment of
753 		 * active/pending. If the pending bit is set, it got
754 		 * rescheduled before I ran. The active
755 		 * bit we ignore, since if it was stopped
756 		 * in ll_tablefree() and was currently running
757 		 * it would have return 0 so the code would
758 		 * not have deleted it since the callout could
759 		 * not be stopped so we want to go through
760 		 * with the delete here now. If the callout
761 		 * was restarted, the pending bit will be back on and
762 		 * we just want to bail since the callout_reset would
763 		 * return 1 and our reference would have been removed
764 		 * by nd6_llinfo_settimer_locked above since canceled
765 		 * would have been 1.
766 		 */
767 		LLE_WUNLOCK(ln);
768 		ND6_RUNLOCK();
769 		CURVNET_RESTORE();
770 		return;
771 	}
772 	NET_EPOCH_ENTER(et);
773 	ndi = ND_IFINFO(ifp);
774 	send_ns = 0;
775 	dst = &ln->r_l3addr.addr6;
776 	pdst = dst;
777 
778 	if (ln->ln_ntick > 0) {
779 		if (ln->ln_ntick > INT_MAX) {
780 			ln->ln_ntick -= INT_MAX;
781 			nd6_llinfo_settimer_locked(ln, INT_MAX);
782 		} else {
783 			ln->ln_ntick = 0;
784 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
785 		}
786 		goto done;
787 	}
788 
789 	if (ln->la_flags & LLE_STATIC) {
790 		goto done;
791 	}
792 
793 	if (ln->la_flags & LLE_DELETED) {
794 		nd6_free(&ln, 0);
795 		goto done;
796 	}
797 
798 	switch (ln->ln_state) {
799 	case ND6_LLINFO_INCOMPLETE:
800 		if (ln->la_asked < V_nd6_mmaxtries) {
801 			ln->la_asked++;
802 			send_ns = 1;
803 			/* Send NS to multicast address */
804 			pdst = NULL;
805 		} else {
806 			struct mbuf *m = ln->la_hold;
807 			if (m) {
808 				struct mbuf *m0;
809 
810 				/*
811 				 * assuming every packet in la_hold has the
812 				 * same IP header.  Send error after unlock.
813 				 */
814 				m0 = m->m_nextpkt;
815 				m->m_nextpkt = NULL;
816 				ln->la_hold = m0;
817 				clear_llinfo_pqueue(ln);
818 			}
819 			nd6_free(&ln, 0);
820 			if (m != NULL)
821 				icmp6_error2(m, ICMP6_DST_UNREACH,
822 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
823 		}
824 		break;
825 	case ND6_LLINFO_REACHABLE:
826 		if (!ND6_LLINFO_PERMANENT(ln))
827 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
828 		break;
829 
830 	case ND6_LLINFO_STALE:
831 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
832 
833 			/*
834 			 * No packet has used this entry and GC timeout
835 			 * has not been passed. Reshedule timer and
836 			 * return.
837 			 */
838 			nd6_llinfo_settimer_locked(ln, delay);
839 			break;
840 		}
841 
842 		if (do_switch == 0) {
843 
844 			/*
845 			 * GC timer has ended and entry hasn't been used.
846 			 * Run Garbage collector (RFC 4861, 5.3)
847 			 */
848 			if (!ND6_LLINFO_PERMANENT(ln))
849 				nd6_free(&ln, 1);
850 			break;
851 		}
852 
853 		/* Entry has been used AND delay timer has ended. */
854 
855 		/* FALLTHROUGH */
856 
857 	case ND6_LLINFO_DELAY:
858 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
859 			/* We need NUD */
860 			ln->la_asked = 1;
861 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
862 			send_ns = 1;
863 		} else
864 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
865 		break;
866 	case ND6_LLINFO_PROBE:
867 		if (ln->la_asked < V_nd6_umaxtries) {
868 			ln->la_asked++;
869 			send_ns = 1;
870 		} else {
871 			nd6_free(&ln, 0);
872 		}
873 		break;
874 	default:
875 		panic("%s: paths in a dark night can be confusing: %d",
876 		    __func__, ln->ln_state);
877 	}
878 done:
879 	if (ln != NULL)
880 		ND6_RUNLOCK();
881 	if (send_ns != 0) {
882 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
883 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
884 		LLE_FREE_LOCKED(ln);
885 		ln = NULL;
886 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
887 	}
888 
889 	if (ln != NULL)
890 		LLE_FREE_LOCKED(ln);
891 	NET_EPOCH_EXIT(et);
892 	CURVNET_RESTORE();
893 }
894 
895 
896 /*
897  * ND6 timer routine to expire default route list and prefix list
898  */
899 void
900 nd6_timer(void *arg)
901 {
902 	CURVNET_SET((struct vnet *) arg);
903 	struct epoch_tracker et;
904 	struct nd_drhead drq;
905 	struct nd_prhead prl;
906 	struct nd_defrouter *dr, *ndr;
907 	struct nd_prefix *pr, *npr;
908 	struct ifnet *ifp;
909 	struct in6_ifaddr *ia6, *nia6;
910 	uint64_t genid;
911 
912 	TAILQ_INIT(&drq);
913 	LIST_INIT(&prl);
914 
915 	ND6_WLOCK();
916 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
917 		if (dr->expire && dr->expire < time_uptime)
918 			defrouter_unlink(dr, &drq);
919 	ND6_WUNLOCK();
920 
921 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
922 		TAILQ_REMOVE(&drq, dr, dr_entry);
923 		defrouter_del(dr);
924 	}
925 
926 	/*
927 	 * expire interface addresses.
928 	 * in the past the loop was inside prefix expiry processing.
929 	 * However, from a stricter speci-confrmance standpoint, we should
930 	 * rather separate address lifetimes and prefix lifetimes.
931 	 *
932 	 * XXXRW: in6_ifaddrhead locking.
933 	 */
934 	NET_EPOCH_ENTER(et);
935   addrloop:
936 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
937 		/* check address lifetime */
938 		if (IFA6_IS_INVALID(ia6)) {
939 			int regen = 0;
940 
941 			/*
942 			 * If the expiring address is temporary, try
943 			 * regenerating a new one.  This would be useful when
944 			 * we suspended a laptop PC, then turned it on after a
945 			 * period that could invalidate all temporary
946 			 * addresses.  Although we may have to restart the
947 			 * loop (see below), it must be after purging the
948 			 * address.  Otherwise, we'd see an infinite loop of
949 			 * regeneration.
950 			 */
951 			if (V_ip6_use_tempaddr &&
952 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
953 				if (regen_tmpaddr(ia6) == 0)
954 					regen = 1;
955 			}
956 
957 			in6_purgeaddr(&ia6->ia_ifa);
958 
959 			if (regen)
960 				goto addrloop; /* XXX: see below */
961 		} else if (IFA6_IS_DEPRECATED(ia6)) {
962 			int oldflags = ia6->ia6_flags;
963 
964 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
965 
966 			/*
967 			 * If a temporary address has just become deprecated,
968 			 * regenerate a new one if possible.
969 			 */
970 			if (V_ip6_use_tempaddr &&
971 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
972 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
973 
974 				if (regen_tmpaddr(ia6) == 0) {
975 					/*
976 					 * A new temporary address is
977 					 * generated.
978 					 * XXX: this means the address chain
979 					 * has changed while we are still in
980 					 * the loop.  Although the change
981 					 * would not cause disaster (because
982 					 * it's not a deletion, but an
983 					 * addition,) we'd rather restart the
984 					 * loop just for safety.  Or does this
985 					 * significantly reduce performance??
986 					 */
987 					goto addrloop;
988 				}
989 			}
990 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
991 			/*
992 			 * Schedule DAD for a tentative address.  This happens
993 			 * if the interface was down or not running
994 			 * when the address was configured.
995 			 */
996 			int delay;
997 
998 			delay = arc4random() %
999 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1000 			nd6_dad_start((struct ifaddr *)ia6, delay);
1001 		} else {
1002 			/*
1003 			 * Check status of the interface.  If it is down,
1004 			 * mark the address as tentative for future DAD.
1005 			 */
1006 			ifp = ia6->ia_ifp;
1007 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1008 			    ((ifp->if_flags & IFF_UP) == 0 ||
1009 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1010 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1011 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1012 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1013 			}
1014 
1015 			/*
1016 			 * A new RA might have made a deprecated address
1017 			 * preferred.
1018 			 */
1019 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1020 		}
1021 	}
1022 	NET_EPOCH_EXIT(et);
1023 
1024 	ND6_WLOCK();
1025 restart:
1026 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1027 		/*
1028 		 * Expire prefixes. Since the pltime is only used for
1029 		 * autoconfigured addresses, pltime processing for prefixes is
1030 		 * not necessary.
1031 		 *
1032 		 * Only unlink after all derived addresses have expired. This
1033 		 * may not occur until two hours after the prefix has expired
1034 		 * per RFC 4862. If the prefix expires before its derived
1035 		 * addresses, mark it off-link. This will be done automatically
1036 		 * after unlinking if no address references remain.
1037 		 */
1038 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1039 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1040 			continue;
1041 
1042 		if (pr->ndpr_addrcnt == 0) {
1043 			nd6_prefix_unlink(pr, &prl);
1044 			continue;
1045 		}
1046 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1047 			genid = V_nd6_list_genid;
1048 			nd6_prefix_ref(pr);
1049 			ND6_WUNLOCK();
1050 			ND6_ONLINK_LOCK();
1051 			(void)nd6_prefix_offlink(pr);
1052 			ND6_ONLINK_UNLOCK();
1053 			ND6_WLOCK();
1054 			nd6_prefix_rele(pr);
1055 			if (genid != V_nd6_list_genid)
1056 				goto restart;
1057 		}
1058 	}
1059 	ND6_WUNLOCK();
1060 
1061 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1062 		LIST_REMOVE(pr, ndpr_entry);
1063 		nd6_prefix_del(pr);
1064 	}
1065 
1066 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1067 	    nd6_timer, curvnet);
1068 
1069 	CURVNET_RESTORE();
1070 }
1071 
1072 /*
1073  * ia6 - deprecated/invalidated temporary address
1074  */
1075 static int
1076 regen_tmpaddr(struct in6_ifaddr *ia6)
1077 {
1078 	struct ifaddr *ifa;
1079 	struct ifnet *ifp;
1080 	struct in6_ifaddr *public_ifa6 = NULL;
1081 
1082 	NET_EPOCH_ASSERT();
1083 
1084 	ifp = ia6->ia_ifa.ifa_ifp;
1085 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1086 		struct in6_ifaddr *it6;
1087 
1088 		if (ifa->ifa_addr->sa_family != AF_INET6)
1089 			continue;
1090 
1091 		it6 = (struct in6_ifaddr *)ifa;
1092 
1093 		/* ignore no autoconf addresses. */
1094 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1095 			continue;
1096 
1097 		/* ignore autoconf addresses with different prefixes. */
1098 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1099 			continue;
1100 
1101 		/*
1102 		 * Now we are looking at an autoconf address with the same
1103 		 * prefix as ours.  If the address is temporary and is still
1104 		 * preferred, do not create another one.  It would be rare, but
1105 		 * could happen, for example, when we resume a laptop PC after
1106 		 * a long period.
1107 		 */
1108 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1109 		    !IFA6_IS_DEPRECATED(it6)) {
1110 			public_ifa6 = NULL;
1111 			break;
1112 		}
1113 
1114 		/*
1115 		 * This is a public autoconf address that has the same prefix
1116 		 * as ours.  If it is preferred, keep it.  We can't break the
1117 		 * loop here, because there may be a still-preferred temporary
1118 		 * address with the prefix.
1119 		 */
1120 		if (!IFA6_IS_DEPRECATED(it6))
1121 			public_ifa6 = it6;
1122 	}
1123 	if (public_ifa6 != NULL)
1124 		ifa_ref(&public_ifa6->ia_ifa);
1125 
1126 	if (public_ifa6 != NULL) {
1127 		int e;
1128 
1129 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1130 			ifa_free(&public_ifa6->ia_ifa);
1131 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1132 			    " tmp addr,errno=%d\n", e);
1133 			return (-1);
1134 		}
1135 		ifa_free(&public_ifa6->ia_ifa);
1136 		return (0);
1137 	}
1138 
1139 	return (-1);
1140 }
1141 
1142 /*
1143  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1144  * cache entries are freed in in6_domifdetach().
1145  */
1146 void
1147 nd6_purge(struct ifnet *ifp)
1148 {
1149 	struct nd_drhead drq;
1150 	struct nd_prhead prl;
1151 	struct nd_defrouter *dr, *ndr;
1152 	struct nd_prefix *pr, *npr;
1153 
1154 	TAILQ_INIT(&drq);
1155 	LIST_INIT(&prl);
1156 
1157 	/*
1158 	 * Nuke default router list entries toward ifp.
1159 	 * We defer removal of default router list entries that is installed
1160 	 * in the routing table, in order to keep additional side effects as
1161 	 * small as possible.
1162 	 */
1163 	ND6_WLOCK();
1164 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1165 		if (dr->installed)
1166 			continue;
1167 		if (dr->ifp == ifp)
1168 			defrouter_unlink(dr, &drq);
1169 	}
1170 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1171 		if (!dr->installed)
1172 			continue;
1173 		if (dr->ifp == ifp)
1174 			defrouter_unlink(dr, &drq);
1175 	}
1176 
1177 	/*
1178 	 * Remove prefixes on ifp. We should have already removed addresses on
1179 	 * this interface, so no addresses should be referencing these prefixes.
1180 	 */
1181 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1182 		if (pr->ndpr_ifp == ifp)
1183 			nd6_prefix_unlink(pr, &prl);
1184 	}
1185 	ND6_WUNLOCK();
1186 
1187 	/* Delete the unlinked router and prefix objects. */
1188 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1189 		TAILQ_REMOVE(&drq, dr, dr_entry);
1190 		defrouter_del(dr);
1191 	}
1192 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1193 		LIST_REMOVE(pr, ndpr_entry);
1194 		nd6_prefix_del(pr);
1195 	}
1196 
1197 	/* cancel default outgoing interface setting */
1198 	if (V_nd6_defifindex == ifp->if_index)
1199 		nd6_setdefaultiface(0);
1200 
1201 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1202 		/* Refresh default router list. */
1203 		defrouter_select_fib(ifp->if_fib);
1204 	}
1205 }
1206 
1207 /*
1208  * the caller acquires and releases the lock on the lltbls
1209  * Returns the llentry locked
1210  */
1211 struct llentry *
1212 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1213 {
1214 	struct sockaddr_in6 sin6;
1215 	struct llentry *ln;
1216 
1217 	bzero(&sin6, sizeof(sin6));
1218 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1219 	sin6.sin6_family = AF_INET6;
1220 	sin6.sin6_addr = *addr6;
1221 
1222 	IF_AFDATA_LOCK_ASSERT(ifp);
1223 
1224 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1225 
1226 	return (ln);
1227 }
1228 
1229 struct llentry *
1230 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1231 {
1232 	struct sockaddr_in6 sin6;
1233 	struct llentry *ln;
1234 
1235 	bzero(&sin6, sizeof(sin6));
1236 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1237 	sin6.sin6_family = AF_INET6;
1238 	sin6.sin6_addr = *addr6;
1239 
1240 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1241 	if (ln != NULL)
1242 		ln->ln_state = ND6_LLINFO_NOSTATE;
1243 
1244 	return (ln);
1245 }
1246 
1247 /*
1248  * Test whether a given IPv6 address is a neighbor or not, ignoring
1249  * the actual neighbor cache.  The neighbor cache is ignored in order
1250  * to not reenter the routing code from within itself.
1251  */
1252 static int
1253 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1254 {
1255 	struct nd_prefix *pr;
1256 	struct ifaddr *ifa;
1257 	struct rt_addrinfo info;
1258 	struct sockaddr_in6 rt_key;
1259 	const struct sockaddr *dst6;
1260 	uint64_t genid;
1261 	int error, fibnum;
1262 
1263 	/*
1264 	 * A link-local address is always a neighbor.
1265 	 * XXX: a link does not necessarily specify a single interface.
1266 	 */
1267 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1268 		struct sockaddr_in6 sin6_copy;
1269 		u_int32_t zone;
1270 
1271 		/*
1272 		 * We need sin6_copy since sa6_recoverscope() may modify the
1273 		 * content (XXX).
1274 		 */
1275 		sin6_copy = *addr;
1276 		if (sa6_recoverscope(&sin6_copy))
1277 			return (0); /* XXX: should be impossible */
1278 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1279 			return (0);
1280 		if (sin6_copy.sin6_scope_id == zone)
1281 			return (1);
1282 		else
1283 			return (0);
1284 	}
1285 
1286 	bzero(&rt_key, sizeof(rt_key));
1287 	bzero(&info, sizeof(info));
1288 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1289 
1290 	/*
1291 	 * If the address matches one of our addresses,
1292 	 * it should be a neighbor.
1293 	 * If the address matches one of our on-link prefixes, it should be a
1294 	 * neighbor.
1295 	 */
1296 	ND6_RLOCK();
1297 restart:
1298 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1299 		if (pr->ndpr_ifp != ifp)
1300 			continue;
1301 
1302 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1303 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1304 
1305 			/*
1306 			 * We only need to check all FIBs if add_addr_allfibs
1307 			 * is unset. If set, checking any FIB will suffice.
1308 			 */
1309 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1310 			for (; fibnum < rt_numfibs; fibnum++) {
1311 				genid = V_nd6_list_genid;
1312 				ND6_RUNLOCK();
1313 
1314 				/*
1315 				 * Restore length field before
1316 				 * retrying lookup
1317 				 */
1318 				rt_key.sin6_len = sizeof(rt_key);
1319 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1320 						        &info);
1321 
1322 				ND6_RLOCK();
1323 				if (genid != V_nd6_list_genid)
1324 					goto restart;
1325 				if (error == 0)
1326 					break;
1327 			}
1328 			if (error != 0)
1329 				continue;
1330 
1331 			/*
1332 			 * This is the case where multiple interfaces
1333 			 * have the same prefix, but only one is installed
1334 			 * into the routing table and that prefix entry
1335 			 * is not the one being examined here. In the case
1336 			 * where RADIX_MPATH is enabled, multiple route
1337 			 * entries (of the same rt_key value) will be
1338 			 * installed because the interface addresses all
1339 			 * differ.
1340 			 */
1341 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1342 			    &rt_key.sin6_addr))
1343 				continue;
1344 		}
1345 
1346 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1347 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1348 			ND6_RUNLOCK();
1349 			return (1);
1350 		}
1351 	}
1352 	ND6_RUNLOCK();
1353 
1354 	/*
1355 	 * If the address is assigned on the node of the other side of
1356 	 * a p2p interface, the address should be a neighbor.
1357 	 */
1358 	if (ifp->if_flags & IFF_POINTOPOINT) {
1359 		struct epoch_tracker et;
1360 
1361 		NET_EPOCH_ENTER(et);
1362 		CK_STAILQ_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 				NET_EPOCH_EXIT(et);
1368 				return 1;
1369 			}
1370 		}
1371 		NET_EPOCH_EXIT(et);
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 	NET_EPOCH_ASSERT();
1399 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1400 	if (nd6_is_new_addr_neighbor(addr, ifp))
1401 		return (1);
1402 
1403 	/*
1404 	 * Even if the address matches none of our addresses, it might be
1405 	 * in the neighbor cache.
1406 	 */
1407 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1408 		LLE_RUNLOCK(lle);
1409 		rc = 1;
1410 	}
1411 	return (rc);
1412 }
1413 
1414 /*
1415  * Free an nd6 llinfo entry.
1416  * Since the function would cause significant changes in the kernel, DO NOT
1417  * make it global, unless you have a strong reason for the change, and are sure
1418  * that the change is safe.
1419  *
1420  * Set noinline to be dtrace-friendly
1421  */
1422 static __noinline void
1423 nd6_free(struct llentry **lnp, int gc)
1424 {
1425 	struct ifnet *ifp;
1426 	struct llentry *ln;
1427 	struct nd_defrouter *dr;
1428 
1429 	ln = *lnp;
1430 	*lnp = NULL;
1431 
1432 	LLE_WLOCK_ASSERT(ln);
1433 	ND6_RLOCK_ASSERT();
1434 
1435 	ifp = lltable_get_ifp(ln->lle_tbl);
1436 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1437 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1438 	else
1439 		dr = NULL;
1440 	ND6_RUNLOCK();
1441 
1442 	if ((ln->la_flags & LLE_DELETED) == 0)
1443 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1444 
1445 	/*
1446 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1447 	 * even though it is not harmful, it was not really necessary.
1448 	 */
1449 
1450 	/* cancel timer */
1451 	nd6_llinfo_settimer_locked(ln, -1);
1452 
1453 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1454 		if (dr != NULL && dr->expire &&
1455 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1456 			/*
1457 			 * If the reason for the deletion is just garbage
1458 			 * collection, and the neighbor is an active default
1459 			 * router, do not delete it.  Instead, reset the GC
1460 			 * timer using the router's lifetime.
1461 			 * Simply deleting the entry would affect default
1462 			 * router selection, which is not necessarily a good
1463 			 * thing, especially when we're using router preference
1464 			 * values.
1465 			 * XXX: the check for ln_state would be redundant,
1466 			 *      but we intentionally keep it just in case.
1467 			 */
1468 			if (dr->expire > time_uptime)
1469 				nd6_llinfo_settimer_locked(ln,
1470 				    (dr->expire - time_uptime) * hz);
1471 			else
1472 				nd6_llinfo_settimer_locked(ln,
1473 				    (long)V_nd6_gctimer * hz);
1474 
1475 			LLE_REMREF(ln);
1476 			LLE_WUNLOCK(ln);
1477 			defrouter_rele(dr);
1478 			return;
1479 		}
1480 
1481 		if (dr) {
1482 			/*
1483 			 * Unreachablity of a router might affect the default
1484 			 * router selection and on-link detection of advertised
1485 			 * prefixes.
1486 			 */
1487 
1488 			/*
1489 			 * Temporarily fake the state to choose a new default
1490 			 * router and to perform on-link determination of
1491 			 * prefixes correctly.
1492 			 * Below the state will be set correctly,
1493 			 * or the entry itself will be deleted.
1494 			 */
1495 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1496 		}
1497 
1498 		if (ln->ln_router || dr) {
1499 
1500 			/*
1501 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1502 			 * rnh and for the calls to pfxlist_onlink_check() and
1503 			 * defrouter_select_fib() in the block further down for calls
1504 			 * into nd6_lookup().  We still hold a ref.
1505 			 */
1506 			LLE_WUNLOCK(ln);
1507 
1508 			/*
1509 			 * rt6_flush must be called whether or not the neighbor
1510 			 * is in the Default Router List.
1511 			 * See a corresponding comment in nd6_na_input().
1512 			 */
1513 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1514 		}
1515 
1516 		if (dr) {
1517 			/*
1518 			 * Since defrouter_select_fib() does not affect the
1519 			 * on-link determination and MIP6 needs the check
1520 			 * before the default router selection, we perform
1521 			 * the check now.
1522 			 */
1523 			pfxlist_onlink_check();
1524 
1525 			/*
1526 			 * Refresh default router list.
1527 			 */
1528 			defrouter_select_fib(dr->ifp->if_fib);
1529 		}
1530 
1531 		/*
1532 		 * If this entry was added by an on-link redirect, remove the
1533 		 * corresponding host route.
1534 		 */
1535 		if (ln->la_flags & LLE_REDIRECT)
1536 			nd6_free_redirect(ln);
1537 
1538 		if (ln->ln_router || dr)
1539 			LLE_WLOCK(ln);
1540 	}
1541 
1542 	/*
1543 	 * Save to unlock. We still hold an extra reference and will not
1544 	 * free(9) in llentry_free() if someone else holds one as well.
1545 	 */
1546 	LLE_WUNLOCK(ln);
1547 	IF_AFDATA_LOCK(ifp);
1548 	LLE_WLOCK(ln);
1549 	/* Guard against race with other llentry_free(). */
1550 	if (ln->la_flags & LLE_LINKED) {
1551 		/* Remove callout reference */
1552 		LLE_REMREF(ln);
1553 		lltable_unlink_entry(ln->lle_tbl, ln);
1554 	}
1555 	IF_AFDATA_UNLOCK(ifp);
1556 
1557 	llentry_free(ln);
1558 	if (dr != NULL)
1559 		defrouter_rele(dr);
1560 }
1561 
1562 static int
1563 nd6_isdynrte(const struct rtentry *rt, void *xap)
1564 {
1565 
1566 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1567 		return (1);
1568 
1569 	return (0);
1570 }
1571 /*
1572  * Remove the rtentry for the given llentry,
1573  * both of which were installed by a redirect.
1574  */
1575 static void
1576 nd6_free_redirect(const struct llentry *ln)
1577 {
1578 	int fibnum;
1579 	struct sockaddr_in6 sin6;
1580 	struct rt_addrinfo info;
1581 
1582 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1583 	memset(&info, 0, sizeof(info));
1584 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1585 	info.rti_filter = nd6_isdynrte;
1586 
1587 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1588 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1589 }
1590 
1591 /*
1592  * Rejuvenate this function for routing operations related
1593  * processing.
1594  */
1595 void
1596 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1597 {
1598 	struct sockaddr_in6 *gateway;
1599 	struct nd_defrouter *dr;
1600 	struct ifnet *ifp;
1601 
1602 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1603 	ifp = rt->rt_ifp;
1604 
1605 	switch (req) {
1606 	case RTM_ADD:
1607 		break;
1608 
1609 	case RTM_DELETE:
1610 		if (!ifp)
1611 			return;
1612 		/*
1613 		 * Only indirect routes are interesting.
1614 		 */
1615 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1616 			return;
1617 		/*
1618 		 * check for default route
1619 		 */
1620 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1621 		    &SIN6(rt_key(rt))->sin6_addr)) {
1622 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1623 			if (dr != NULL) {
1624 				dr->installed = 0;
1625 				defrouter_rele(dr);
1626 			}
1627 		}
1628 		break;
1629 	}
1630 }
1631 
1632 
1633 int
1634 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1635 {
1636 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1637 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1638 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1639 	struct epoch_tracker et;
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 			NET_EPOCH_ENTER(et);
1705 			CK_STAILQ_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 			NET_EPOCH_EXIT(et);
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 				NET_EPOCH_ENTER(et);
1735 				CK_STAILQ_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 				NET_EPOCH_EXIT(et);
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 				NET_EPOCH_ENTER(et);
1763 				CK_STAILQ_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 				NET_EPOCH_EXIT(et);
1773 				if (ifa != NULL)
1774 					/* No LLA is configured. */
1775 					in6_ifattach(ifp, NULL);
1776 			}
1777 		}
1778 		ND_IFINFO(ifp)->flags = ND.flags;
1779 		break;
1780 	}
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 			CK_STAILQ_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 		NET_EPOCH_ENTER(et);
1850 		ln = nd6_lookup(&nb_addr, 0, ifp);
1851 		NET_EPOCH_EXIT(et);
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 	NET_EPOCH_ASSERT();
1982 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1983 
1984 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1985 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1986 
1987 	/* nothing must be updated for unspecified address */
1988 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1989 		return;
1990 
1991 	/*
1992 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1993 	 * the caller.
1994 	 *
1995 	 * XXX If the link does not have link-layer adderss, what should
1996 	 * we do? (ifp->if_addrlen == 0)
1997 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1998 	 * description on it in NS section (RFC 2461 7.2.3).
1999 	 */
2000 	flags = lladdr ? LLE_EXCLUSIVE : 0;
2001 	ln = nd6_lookup(from, flags, ifp);
2002 	is_newentry = 0;
2003 	if (ln == NULL) {
2004 		flags |= LLE_EXCLUSIVE;
2005 		ln = nd6_alloc(from, 0, ifp);
2006 		if (ln == NULL)
2007 			return;
2008 
2009 		/*
2010 		 * Since we already know all the data for the new entry,
2011 		 * fill it before insertion.
2012 		 */
2013 		if (lladdr != NULL) {
2014 			linkhdrsize = sizeof(linkhdr);
2015 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2016 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2017 				return;
2018 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2019 			    lladdr_off);
2020 		}
2021 
2022 		IF_AFDATA_WLOCK(ifp);
2023 		LLE_WLOCK(ln);
2024 		/* Prefer any existing lle over newly-created one */
2025 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
2026 		if (ln_tmp == NULL)
2027 			lltable_link_entry(LLTABLE6(ifp), ln);
2028 		IF_AFDATA_WUNLOCK(ifp);
2029 		if (ln_tmp == NULL) {
2030 			/* No existing lle, mark as new entry (6,7) */
2031 			is_newentry = 1;
2032 			if (lladdr != NULL) {	/* (7) */
2033 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2034 				EVENTHANDLER_INVOKE(lle_event, ln,
2035 				    LLENTRY_RESOLVED);
2036 			}
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, 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 	struct epoch_tracker et;
2144 	CURVNET_SET((struct vnet *) arg);
2145 	struct nd_ifinfo *nd6if;
2146 	struct ifnet *ifp;
2147 
2148 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2149 	    nd6_slowtimo, curvnet);
2150 	NET_EPOCH_ENTER(et);
2151 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2152 		if (ifp->if_afdata[AF_INET6] == NULL)
2153 			continue;
2154 		nd6if = ND_IFINFO(ifp);
2155 		if (nd6if->basereachable && /* already initialized */
2156 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2157 			/*
2158 			 * Since reachable time rarely changes by router
2159 			 * advertisements, we SHOULD insure that a new random
2160 			 * value gets recomputed at least once every few hours.
2161 			 * (RFC 2461, 6.3.4)
2162 			 */
2163 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2164 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2165 		}
2166 	}
2167 	NET_EPOCH_EXIT(et);
2168 	CURVNET_RESTORE();
2169 }
2170 
2171 void
2172 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2173     struct sockaddr_in6 *sin6)
2174 {
2175 
2176 	LLE_WLOCK_ASSERT(ln);
2177 
2178 	*chain = ln->la_hold;
2179 	ln->la_hold = NULL;
2180 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2181 
2182 	if (ln->ln_state == ND6_LLINFO_STALE) {
2183 
2184 		/*
2185 		 * The first time we send a packet to a
2186 		 * neighbor whose entry is STALE, we have
2187 		 * to change the state to DELAY and a sets
2188 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2189 		 * seconds to ensure do neighbor unreachability
2190 		 * detection on expiration.
2191 		 * (RFC 2461 7.3.3)
2192 		 */
2193 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2194 	}
2195 }
2196 
2197 int
2198 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2199     struct sockaddr_in6 *dst, struct route *ro)
2200 {
2201 	int error;
2202 	int ip6len;
2203 	struct ip6_hdr *ip6;
2204 	struct m_tag *mtag;
2205 
2206 #ifdef MAC
2207 	mac_netinet6_nd6_send(ifp, m);
2208 #endif
2209 
2210 	/*
2211 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2212 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2213 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2214 	 * to be diverted to user space.  When re-injected into the kernel,
2215 	 * send_output() will directly dispatch them to the outgoing interface.
2216 	 */
2217 	if (send_sendso_input_hook != NULL) {
2218 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2219 		if (mtag != NULL) {
2220 			ip6 = mtod(m, struct ip6_hdr *);
2221 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2222 			/* Use the SEND socket */
2223 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2224 			    ip6len);
2225 			/* -1 == no app on SEND socket */
2226 			if (error == 0 || error != -1)
2227 			    return (error);
2228 		}
2229 	}
2230 
2231 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2232 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2233 	    mtod(m, struct ip6_hdr *));
2234 
2235 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2236 		origifp = ifp;
2237 
2238 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2239 	return (error);
2240 }
2241 
2242 /*
2243  * Lookup link headerfor @sa_dst address. Stores found
2244  * data in @desten buffer. Copy of lle ln_flags can be also
2245  * saved in @pflags if @pflags is non-NULL.
2246  *
2247  * If destination LLE does not exists or lle state modification
2248  * is required, call "slow" version.
2249  *
2250  * Return values:
2251  * - 0 on success (address copied to buffer).
2252  * - EWOULDBLOCK (no local error, but address is still unresolved)
2253  * - other errors (alloc failure, etc)
2254  */
2255 int
2256 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2257     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2258     struct llentry **plle)
2259 {
2260 	struct llentry *ln = NULL;
2261 	const struct sockaddr_in6 *dst6;
2262 
2263 	NET_EPOCH_ASSERT();
2264 
2265 	if (pflags != NULL)
2266 		*pflags = 0;
2267 
2268 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2269 
2270 	/* discard the packet if IPv6 operation is disabled on the interface */
2271 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2272 		m_freem(m);
2273 		return (ENETDOWN); /* better error? */
2274 	}
2275 
2276 	if (m != NULL && m->m_flags & M_MCAST) {
2277 		switch (ifp->if_type) {
2278 		case IFT_ETHER:
2279 		case IFT_L2VLAN:
2280 		case IFT_BRIDGE:
2281 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2282 						 desten);
2283 			return (0);
2284 		default:
2285 			m_freem(m);
2286 			return (EAFNOSUPPORT);
2287 		}
2288 	}
2289 
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 		return (0);
2310 	} else if (plle && ln)
2311 		LLE_WUNLOCK(ln);
2312 
2313 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2314 }
2315 
2316 
2317 /*
2318  * Do L2 address resolution for @sa_dst address. Stores found
2319  * address in @desten buffer. Copy of lle ln_flags can be also
2320  * saved in @pflags if @pflags is non-NULL.
2321  *
2322  * Heavy version.
2323  * Function assume that destination LLE does not exist,
2324  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2325  *
2326  * Set noinline to be dtrace-friendly
2327  */
2328 static __noinline int
2329 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2330     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2331     struct llentry **plle)
2332 {
2333 	struct llentry *lle = NULL, *lle_tmp;
2334 	struct in6_addr *psrc, src;
2335 	int send_ns, ll_len;
2336 	char *lladdr;
2337 
2338 	NET_EPOCH_ASSERT();
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 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2348 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2349 			/*
2350 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2351 			 * the condition below is not very efficient.  But we believe
2352 			 * it is tolerable, because this should be a rare case.
2353 			 */
2354 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2355 			if (lle == NULL) {
2356 				char ip6buf[INET6_ADDRSTRLEN];
2357 				log(LOG_DEBUG,
2358 				    "nd6_output: can't allocate llinfo for %s "
2359 				    "(ln=%p)\n",
2360 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2361 				m_freem(m);
2362 				return (ENOBUFS);
2363 			}
2364 
2365 			IF_AFDATA_WLOCK(ifp);
2366 			LLE_WLOCK(lle);
2367 			/* Prefer any existing entry over newly-created one */
2368 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2369 			if (lle_tmp == NULL)
2370 				lltable_link_entry(LLTABLE6(ifp), lle);
2371 			IF_AFDATA_WUNLOCK(ifp);
2372 			if (lle_tmp != NULL) {
2373 				lltable_free_entry(LLTABLE6(ifp), lle);
2374 				lle = lle_tmp;
2375 				lle_tmp = NULL;
2376 			}
2377 		}
2378 	}
2379 	if (lle == NULL) {
2380 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2381 			m_freem(m);
2382 			return (ENOBUFS);
2383 		}
2384 
2385 		if (m != NULL)
2386 			m_freem(m);
2387 		return (ENOBUFS);
2388 	}
2389 
2390 	LLE_WLOCK_ASSERT(lle);
2391 
2392 	/*
2393 	 * The first time we send a packet to a neighbor whose entry is
2394 	 * STALE, we have to change the state to DELAY and a sets a timer to
2395 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2396 	 * neighbor unreachability detection on expiration.
2397 	 * (RFC 2461 7.3.3)
2398 	 */
2399 	if (lle->ln_state == ND6_LLINFO_STALE)
2400 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2401 
2402 	/*
2403 	 * If the neighbor cache entry has a state other than INCOMPLETE
2404 	 * (i.e. its link-layer address is already resolved), just
2405 	 * send the packet.
2406 	 */
2407 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2408 		if (flags & LLE_ADDRONLY) {
2409 			lladdr = lle->ll_addr;
2410 			ll_len = ifp->if_addrlen;
2411 		} else {
2412 			lladdr = lle->r_linkdata;
2413 			ll_len = lle->r_hdrlen;
2414 		}
2415 		bcopy(lladdr, desten, ll_len);
2416 		if (pflags != NULL)
2417 			*pflags = lle->la_flags;
2418 		if (plle) {
2419 			LLE_ADDREF(lle);
2420 			*plle = lle;
2421 		}
2422 		LLE_WUNLOCK(lle);
2423 		return (0);
2424 	}
2425 
2426 	/*
2427 	 * There is a neighbor cache entry, but no ethernet address
2428 	 * response yet.  Append this latest packet to the end of the
2429 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2430 	 * the oldest packet in the queue will be removed.
2431 	 */
2432 
2433 	if (lle->la_hold != NULL) {
2434 		struct mbuf *m_hold;
2435 		int i;
2436 
2437 		i = 0;
2438 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2439 			i++;
2440 			if (m_hold->m_nextpkt == NULL) {
2441 				m_hold->m_nextpkt = m;
2442 				break;
2443 			}
2444 		}
2445 		while (i >= V_nd6_maxqueuelen) {
2446 			m_hold = lle->la_hold;
2447 			lle->la_hold = lle->la_hold->m_nextpkt;
2448 			m_freem(m_hold);
2449 			i--;
2450 		}
2451 	} else {
2452 		lle->la_hold = m;
2453 	}
2454 
2455 	/*
2456 	 * If there has been no NS for the neighbor after entering the
2457 	 * INCOMPLETE state, send the first solicitation.
2458 	 * Note that for newly-created lle la_asked will be 0,
2459 	 * so we will transition from ND6_LLINFO_NOSTATE to
2460 	 * ND6_LLINFO_INCOMPLETE state here.
2461 	 */
2462 	psrc = NULL;
2463 	send_ns = 0;
2464 	if (lle->la_asked == 0) {
2465 		lle->la_asked++;
2466 		send_ns = 1;
2467 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2468 
2469 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2470 	}
2471 	LLE_WUNLOCK(lle);
2472 	if (send_ns != 0)
2473 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2474 
2475 	return (EWOULDBLOCK);
2476 }
2477 
2478 /*
2479  * Do L2 address resolution for @sa_dst address. Stores found
2480  * address in @desten buffer. Copy of lle ln_flags can be also
2481  * saved in @pflags if @pflags is non-NULL.
2482  *
2483  * Return values:
2484  * - 0 on success (address copied to buffer).
2485  * - EWOULDBLOCK (no local error, but address is still unresolved)
2486  * - other errors (alloc failure, etc)
2487  */
2488 int
2489 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2490     char *desten, uint32_t *pflags)
2491 {
2492 	int error;
2493 
2494 	flags |= LLE_ADDRONLY;
2495 	error = nd6_resolve_slow(ifp, flags, NULL,
2496 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2497 	return (error);
2498 }
2499 
2500 int
2501 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2502     struct sockaddr_in6 *dst)
2503 {
2504 	struct mbuf *m, *m_head;
2505 	int error = 0;
2506 
2507 	m_head = chain;
2508 
2509 	while (m_head) {
2510 		m = m_head;
2511 		m_head = m_head->m_nextpkt;
2512 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2513 	}
2514 
2515 	/*
2516 	 * XXX
2517 	 * note that intermediate errors are blindly ignored
2518 	 */
2519 	return (error);
2520 }
2521 
2522 static int
2523 nd6_need_cache(struct ifnet *ifp)
2524 {
2525 	/*
2526 	 * XXX: we currently do not make neighbor cache on any interface
2527 	 * other than Ethernet and GIF.
2528 	 *
2529 	 * RFC2893 says:
2530 	 * - unidirectional tunnels needs no ND
2531 	 */
2532 	switch (ifp->if_type) {
2533 	case IFT_ETHER:
2534 	case IFT_IEEE1394:
2535 	case IFT_L2VLAN:
2536 	case IFT_INFINIBAND:
2537 	case IFT_BRIDGE:
2538 	case IFT_PROPVIRTUAL:
2539 		return (1);
2540 	default:
2541 		return (0);
2542 	}
2543 }
2544 
2545 /*
2546  * Add pernament ND6 link-layer record for given
2547  * interface address.
2548  *
2549  * Very similar to IPv4 arp_ifinit(), but:
2550  * 1) IPv6 DAD is performed in different place
2551  * 2) It is called by IPv6 protocol stack in contrast to
2552  * arp_ifinit() which is typically called in SIOCSIFADDR
2553  * driver ioctl handler.
2554  *
2555  */
2556 int
2557 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2558 {
2559 	struct ifnet *ifp;
2560 	struct llentry *ln, *ln_tmp;
2561 	struct sockaddr *dst;
2562 
2563 	ifp = ia->ia_ifa.ifa_ifp;
2564 	if (nd6_need_cache(ifp) == 0)
2565 		return (0);
2566 
2567 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2568 	dst = (struct sockaddr *)&ia->ia_addr;
2569 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2570 	if (ln == NULL)
2571 		return (ENOBUFS);
2572 
2573 	IF_AFDATA_WLOCK(ifp);
2574 	LLE_WLOCK(ln);
2575 	/* Unlink any entry if exists */
2576 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2577 	if (ln_tmp != NULL)
2578 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2579 	lltable_link_entry(LLTABLE6(ifp), ln);
2580 	IF_AFDATA_WUNLOCK(ifp);
2581 
2582 	if (ln_tmp != NULL)
2583 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2584 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2585 
2586 	LLE_WUNLOCK(ln);
2587 	if (ln_tmp != NULL)
2588 		llentry_free(ln_tmp);
2589 
2590 	return (0);
2591 }
2592 
2593 /*
2594  * Removes either all lle entries for given @ia, or lle
2595  * corresponding to @ia address.
2596  */
2597 void
2598 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2599 {
2600 	struct sockaddr_in6 mask, addr;
2601 	struct sockaddr *saddr, *smask;
2602 	struct ifnet *ifp;
2603 
2604 	ifp = ia->ia_ifa.ifa_ifp;
2605 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2606 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2607 	saddr = (struct sockaddr *)&addr;
2608 	smask = (struct sockaddr *)&mask;
2609 
2610 	if (all != 0)
2611 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2612 	else
2613 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2614 }
2615 
2616 static void
2617 clear_llinfo_pqueue(struct llentry *ln)
2618 {
2619 	struct mbuf *m_hold, *m_hold_next;
2620 
2621 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2622 		m_hold_next = m_hold->m_nextpkt;
2623 		m_freem(m_hold);
2624 	}
2625 
2626 	ln->la_hold = NULL;
2627 }
2628 
2629 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2630 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2631 
2632 SYSCTL_DECL(_net_inet6_icmp6);
2633 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2634 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2635 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2636 	"NDP default router list");
2637 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2638 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2639 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2640 	"NDP prefix list");
2641 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2642 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2643 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2644 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2645 
2646 static int
2647 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2648 {
2649 	struct in6_defrouter d;
2650 	struct nd_defrouter *dr;
2651 	int error;
2652 
2653 	if (req->newptr != NULL)
2654 		return (EPERM);
2655 
2656 	error = sysctl_wire_old_buffer(req, 0);
2657 	if (error != 0)
2658 		return (error);
2659 
2660 	bzero(&d, sizeof(d));
2661 	d.rtaddr.sin6_family = AF_INET6;
2662 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2663 
2664 	ND6_RLOCK();
2665 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2666 		d.rtaddr.sin6_addr = dr->rtaddr;
2667 		error = sa6_recoverscope(&d.rtaddr);
2668 		if (error != 0)
2669 			break;
2670 		d.flags = dr->raflags;
2671 		d.rtlifetime = dr->rtlifetime;
2672 		d.expire = dr->expire + (time_second - time_uptime);
2673 		d.if_index = dr->ifp->if_index;
2674 		error = SYSCTL_OUT(req, &d, sizeof(d));
2675 		if (error != 0)
2676 			break;
2677 	}
2678 	ND6_RUNLOCK();
2679 	return (error);
2680 }
2681 
2682 static int
2683 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2684 {
2685 	struct in6_prefix p;
2686 	struct sockaddr_in6 s6;
2687 	struct nd_prefix *pr;
2688 	struct nd_pfxrouter *pfr;
2689 	time_t maxexpire;
2690 	int error;
2691 	char ip6buf[INET6_ADDRSTRLEN];
2692 
2693 	if (req->newptr)
2694 		return (EPERM);
2695 
2696 	error = sysctl_wire_old_buffer(req, 0);
2697 	if (error != 0)
2698 		return (error);
2699 
2700 	bzero(&p, sizeof(p));
2701 	p.origin = PR_ORIG_RA;
2702 	bzero(&s6, sizeof(s6));
2703 	s6.sin6_family = AF_INET6;
2704 	s6.sin6_len = sizeof(s6);
2705 
2706 	ND6_RLOCK();
2707 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2708 		p.prefix = pr->ndpr_prefix;
2709 		if (sa6_recoverscope(&p.prefix)) {
2710 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2711 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2712 			/* XXX: press on... */
2713 		}
2714 		p.raflags = pr->ndpr_raf;
2715 		p.prefixlen = pr->ndpr_plen;
2716 		p.vltime = pr->ndpr_vltime;
2717 		p.pltime = pr->ndpr_pltime;
2718 		p.if_index = pr->ndpr_ifp->if_index;
2719 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2720 			p.expire = 0;
2721 		else {
2722 			/* XXX: we assume time_t is signed. */
2723 			maxexpire = (-1) &
2724 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2725 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2726 				p.expire = pr->ndpr_lastupdate +
2727 				    pr->ndpr_vltime +
2728 				    (time_second - time_uptime);
2729 			else
2730 				p.expire = maxexpire;
2731 		}
2732 		p.refcnt = pr->ndpr_addrcnt;
2733 		p.flags = pr->ndpr_stateflags;
2734 		p.advrtrs = 0;
2735 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2736 			p.advrtrs++;
2737 		error = SYSCTL_OUT(req, &p, sizeof(p));
2738 		if (error != 0)
2739 			break;
2740 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2741 			s6.sin6_addr = pfr->router->rtaddr;
2742 			if (sa6_recoverscope(&s6))
2743 				log(LOG_ERR,
2744 				    "scope error in prefix list (%s)\n",
2745 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2746 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2747 			if (error != 0)
2748 				goto out;
2749 		}
2750 	}
2751 out:
2752 	ND6_RUNLOCK();
2753 	return (error);
2754 }
2755