xref: /freebsd/sys/netinet6/nd6.c (revision a48df53e4249499be3e8779dd30888a405aa81ae)
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 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_route.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/eventhandler.h>
42 #include <sys/callout.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/protosw.h>
52 #include <sys/errno.h>
53 #include <sys/syslog.h>
54 #include <sys/rwlock.h>
55 #include <sys/queue.h>
56 #include <sys/sdt.h>
57 #include <sys/sysctl.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_private.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/route/route_ctl.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_kdtrace.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #include <netinet6/in6_fib.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82 
83 #include <sys/limits.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
88 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
89 
90 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
91 
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93 
94 VNET_DEFINE_STATIC(int, nd6_prune) = 1;
95 #define	V_nd6_prune	VNET(nd6_prune)
96 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_PRUNE, nd6_prune,
97     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_prune), 0,
98     "Frequency in seconds of checks for expired prefixes and routers");
99 
100 VNET_DEFINE_STATIC(int, nd6_delay) = 5;
101 #define	V_nd6_delay	VNET(nd6_delay)
102 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DELAY, nd6_delay,
103     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_delay), 0,
104     "Delay in seconds before probing for reachability");
105 
106 VNET_DEFINE_STATIC(int, nd6_umaxtries) = 3;
107 #define	V_nd6_umaxtries	VNET(nd6_umaxtries)
108 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_UMAXTRIES, nd6_umaxtries,
109     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_umaxtries), 0,
110     "Number of ICMPv6 NS messages sent during reachability detection");
111 
112 VNET_DEFINE(int, nd6_mmaxtries) = 3;
113 #define	V_nd6_mmaxtries	VNET(nd6_mmaxtries)
114 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MMAXTRIES, nd6_mmaxtries,
115     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_mmaxtries), 0,
116     "Number of ICMPv6 NS messages sent during address resolution");
117 
118 VNET_DEFINE_STATIC(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
119 							* collection timer */
120 #define	V_nd6_gctimer	VNET(nd6_gctimer)
121 
122 /* preventing too many loops in ND option parsing */
123 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
124 
125 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
126 					 * ND entries */
127 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
128 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
129 
130 #ifdef ND6_DEBUG
131 VNET_DEFINE(int, nd6_debug) = 1;
132 #else
133 VNET_DEFINE(int, nd6_debug) = 0;
134 #endif
135 #define	V_nd6_debug	VNET(nd6_debug)
136 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DEBUG, nd6_debug,
137     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_debug), 0,
138     "Log NDP debug messages");
139 
140 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
141 
142 VNET_DEFINE(struct nd_prhead, nd_prefix);
143 VNET_DEFINE(struct rwlock, nd6_lock);
144 VNET_DEFINE(uint64_t, nd6_list_genid);
145 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
146 
147 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
148 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
149 
150 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
151 
152 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
153 	struct ifnet *);
154 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
155 static void nd6_slowtimo(void *);
156 static int regen_tmpaddr(struct in6_ifaddr *);
157 static void nd6_free(struct llentry **, int);
158 static void nd6_free_redirect(const struct llentry *);
159 static void nd6_llinfo_timer(void *);
160 static void nd6_llinfo_settimer_locked(struct llentry *, long);
161 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
162     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
163 static int nd6_need_cache(struct ifnet *);
164 
165 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
166 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
167 
168 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
169 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
170 
171 static void
172 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
173 {
174 	struct rt_addrinfo rtinfo;
175 	struct sockaddr_in6 dst;
176 	struct sockaddr_dl gw;
177 	struct ifnet *ifp;
178 	int type;
179 	int fibnum;
180 
181 	LLE_WLOCK_ASSERT(lle);
182 
183 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
184 		return;
185 
186 	switch (evt) {
187 	case LLENTRY_RESOLVED:
188 		type = RTM_ADD;
189 		KASSERT(lle->la_flags & LLE_VALID,
190 		    ("%s: %p resolved but not valid?", __func__, lle));
191 		break;
192 	case LLENTRY_EXPIRED:
193 		type = RTM_DELETE;
194 		break;
195 	default:
196 		return;
197 	}
198 
199 	ifp = lltable_get_ifp(lle->lle_tbl);
200 
201 	bzero(&dst, sizeof(dst));
202 	bzero(&gw, sizeof(gw));
203 	bzero(&rtinfo, sizeof(rtinfo));
204 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
205 	dst.sin6_scope_id = in6_getscopezone(ifp,
206 	    in6_addrscope(&dst.sin6_addr));
207 	gw.sdl_len = sizeof(struct sockaddr_dl);
208 	gw.sdl_family = AF_LINK;
209 	gw.sdl_alen = ifp->if_addrlen;
210 	gw.sdl_index = ifp->if_index;
211 	gw.sdl_type = ifp->if_type;
212 	if (evt == LLENTRY_RESOLVED)
213 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
214 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
215 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
216 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
217 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
218 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
219 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
220 }
221 
222 /*
223  * A handler for interface link layer address change event.
224  */
225 static void
226 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
227 {
228 	if (ifp->if_afdata[AF_INET6] == NULL)
229 		return;
230 
231 	lltable_update_ifaddr(LLTABLE6(ifp));
232 }
233 
234 void
235 nd6_init(void)
236 {
237 
238 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
239 	rw_init(&V_nd6_lock, "nd6 list");
240 
241 	LIST_INIT(&V_nd_prefix);
242 	nd6_defrouter_init();
243 
244 	/* Start timers. */
245 	callout_init(&V_nd6_slowtimo_ch, 1);
246 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
247 	    nd6_slowtimo, curvnet);
248 
249 	callout_init(&V_nd6_timer_ch, 1);
250 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
251 
252 	nd6_dad_init();
253 	if (IS_DEFAULT_VNET(curvnet)) {
254 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
255 		    NULL, EVENTHANDLER_PRI_ANY);
256 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
257 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
258 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
259 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
260 	}
261 }
262 
263 #ifdef VIMAGE
264 void
265 nd6_destroy(void)
266 {
267 
268 	callout_drain(&V_nd6_slowtimo_ch);
269 	callout_drain(&V_nd6_timer_ch);
270 	if (IS_DEFAULT_VNET(curvnet)) {
271 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
272 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
273 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
274 	}
275 	rw_destroy(&V_nd6_lock);
276 	mtx_destroy(&V_nd6_onlink_mtx);
277 }
278 #endif
279 
280 struct nd_ifinfo *
281 nd6_ifattach(struct ifnet *ifp)
282 {
283 	struct nd_ifinfo *nd;
284 
285 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
286 	nd->initialized = 1;
287 
288 	nd->chlim = IPV6_DEFHLIM;
289 	nd->basereachable = REACHABLE_TIME;
290 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
291 	nd->retrans = RETRANS_TIMER;
292 
293 	nd->flags = ND6_IFF_PERFORMNUD;
294 
295 	/* Set IPv6 disabled on all interfaces but loopback by default. */
296 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
297 		nd->flags |= ND6_IFF_IFDISABLED;
298 
299 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
300 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
301 	 * default regardless of the V_ip6_auto_linklocal configuration to
302 	 * give a reasonable default behavior.
303 	 */
304 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
305 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
306 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
307 	/*
308 	 * A loopback interface does not need to accept RTADV.
309 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
310 	 * default regardless of the V_ip6_accept_rtadv configuration to
311 	 * prevent the interface from accepting RA messages arrived
312 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
313 	 */
314 	if (V_ip6_accept_rtadv &&
315 	    !(ifp->if_flags & IFF_LOOPBACK) &&
316 	    (ifp->if_type != IFT_BRIDGE)) {
317 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
318 			/* If we globally accept rtadv, assume IPv6 on. */
319 			nd->flags &= ~ND6_IFF_IFDISABLED;
320 	}
321 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
322 		nd->flags |= ND6_IFF_NO_RADR;
323 
324 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
325 	nd6_setmtu0(ifp, nd);
326 
327 	return nd;
328 }
329 
330 void
331 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
332 {
333 	struct epoch_tracker et;
334 	struct ifaddr *ifa, *next;
335 
336 	NET_EPOCH_ENTER(et);
337 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
338 		if (ifa->ifa_addr->sa_family != AF_INET6)
339 			continue;
340 
341 		/* stop DAD processing */
342 		nd6_dad_stop(ifa);
343 	}
344 	NET_EPOCH_EXIT(et);
345 
346 	free(nd, M_IP6NDP);
347 }
348 
349 /*
350  * Reset ND level link MTU. This function is called when the physical MTU
351  * changes, which means we might have to adjust the ND level MTU.
352  */
353 void
354 nd6_setmtu(struct ifnet *ifp)
355 {
356 	if (ifp->if_afdata[AF_INET6] == NULL)
357 		return;
358 
359 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
360 }
361 
362 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
363 void
364 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
365 {
366 	u_int32_t omaxmtu;
367 
368 	omaxmtu = ndi->maxmtu;
369 	ndi->maxmtu = ifp->if_mtu;
370 
371 	/*
372 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
373 	 * undesirable situation.  We thus notify the operator of the change
374 	 * explicitly.  The check for omaxmtu is necessary to restrict the
375 	 * log to the case of changing the MTU, not initializing it.
376 	 */
377 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
378 		log(LOG_NOTICE, "nd6_setmtu0: "
379 		    "new link MTU on %s (%lu) is too small for IPv6\n",
380 		    if_name(ifp), (unsigned long)ndi->maxmtu);
381 	}
382 
383 	if (ndi->maxmtu > V_in6_maxmtu)
384 		in6_setmaxmtu(); /* check all interfaces just in case */
385 
386 }
387 
388 void
389 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
390 {
391 
392 	bzero(ndopts, sizeof(*ndopts));
393 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
394 	ndopts->nd_opts_last
395 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
396 
397 	if (icmp6len == 0) {
398 		ndopts->nd_opts_done = 1;
399 		ndopts->nd_opts_search = NULL;
400 	}
401 }
402 
403 /*
404  * Take one ND option.
405  */
406 struct nd_opt_hdr *
407 nd6_option(union nd_opts *ndopts)
408 {
409 	struct nd_opt_hdr *nd_opt;
410 	int olen;
411 
412 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
413 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
414 	    __func__));
415 	if (ndopts->nd_opts_search == NULL)
416 		return NULL;
417 	if (ndopts->nd_opts_done)
418 		return NULL;
419 
420 	nd_opt = ndopts->nd_opts_search;
421 
422 	/* make sure nd_opt_len is inside the buffer */
423 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
424 		bzero(ndopts, sizeof(*ndopts));
425 		return NULL;
426 	}
427 
428 	olen = nd_opt->nd_opt_len << 3;
429 	if (olen == 0) {
430 		/*
431 		 * Message validation requires that all included
432 		 * options have a length that is greater than zero.
433 		 */
434 		bzero(ndopts, sizeof(*ndopts));
435 		return NULL;
436 	}
437 
438 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
439 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
440 		/* option overruns the end of buffer, invalid */
441 		bzero(ndopts, sizeof(*ndopts));
442 		return NULL;
443 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
444 		/* reached the end of options chain */
445 		ndopts->nd_opts_done = 1;
446 		ndopts->nd_opts_search = NULL;
447 	}
448 	return nd_opt;
449 }
450 
451 /*
452  * Parse multiple ND options.
453  * This function is much easier to use, for ND routines that do not need
454  * multiple options of the same type.
455  */
456 int
457 nd6_options(union nd_opts *ndopts)
458 {
459 	struct nd_opt_hdr *nd_opt;
460 	int i = 0;
461 
462 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
463 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
464 	    __func__));
465 	if (ndopts->nd_opts_search == NULL)
466 		return 0;
467 
468 	while (1) {
469 		nd_opt = nd6_option(ndopts);
470 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
471 			/*
472 			 * Message validation requires that all included
473 			 * options have a length that is greater than zero.
474 			 */
475 			ICMP6STAT_INC(icp6s_nd_badopt);
476 			bzero(ndopts, sizeof(*ndopts));
477 			return -1;
478 		}
479 
480 		if (nd_opt == NULL)
481 			goto skip1;
482 
483 		switch (nd_opt->nd_opt_type) {
484 		case ND_OPT_SOURCE_LINKADDR:
485 		case ND_OPT_TARGET_LINKADDR:
486 		case ND_OPT_MTU:
487 		case ND_OPT_REDIRECTED_HEADER:
488 		case ND_OPT_NONCE:
489 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
490 				nd6log((LOG_INFO,
491 				    "duplicated ND6 option found (type=%d)\n",
492 				    nd_opt->nd_opt_type));
493 				/* XXX bark? */
494 			} else {
495 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
496 					= nd_opt;
497 			}
498 			break;
499 		case ND_OPT_PREFIX_INFORMATION:
500 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
501 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
502 					= nd_opt;
503 			}
504 			ndopts->nd_opts_pi_end =
505 				(struct nd_opt_prefix_info *)nd_opt;
506 			break;
507 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
508 		case ND_OPT_RDNSS:	/* RFC 6106 */
509 		case ND_OPT_DNSSL:	/* RFC 6106 */
510 			/*
511 			 * Silently ignore options we know and do not care about
512 			 * in the kernel.
513 			 */
514 			break;
515 		default:
516 			/*
517 			 * Unknown options must be silently ignored,
518 			 * to accommodate future extension to the protocol.
519 			 */
520 			nd6log((LOG_DEBUG,
521 			    "nd6_options: unsupported option %d - "
522 			    "option ignored\n", nd_opt->nd_opt_type));
523 		}
524 
525 skip1:
526 		i++;
527 		if (i > V_nd6_maxndopt) {
528 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
529 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
530 			break;
531 		}
532 
533 		if (ndopts->nd_opts_done)
534 			break;
535 	}
536 
537 	return 0;
538 }
539 
540 /*
541  * ND6 timer routine to handle ND6 entries
542  */
543 static void
544 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
545 {
546 	int canceled;
547 
548 	LLE_WLOCK_ASSERT(ln);
549 
550 	/* Do not schedule timers for child LLEs. */
551 	if (ln->la_flags & LLE_CHILD)
552 		return;
553 
554 	if (tick < 0) {
555 		ln->la_expire = 0;
556 		ln->ln_ntick = 0;
557 		canceled = callout_stop(&ln->lle_timer);
558 	} else {
559 		ln->la_expire = time_uptime + tick / hz;
560 		LLE_ADDREF(ln);
561 		if (tick > INT_MAX) {
562 			ln->ln_ntick = tick - INT_MAX;
563 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
564 			    nd6_llinfo_timer, ln);
565 		} else {
566 			ln->ln_ntick = 0;
567 			canceled = callout_reset(&ln->lle_timer, tick,
568 			    nd6_llinfo_timer, ln);
569 		}
570 	}
571 	if (canceled > 0)
572 		LLE_REMREF(ln);
573 }
574 
575 /*
576  * Gets source address of the first packet in hold queue
577  * and stores it in @src.
578  * Returns pointer to @src (if hold queue is not empty) or NULL.
579  *
580  * Set noinline to be dtrace-friendly
581  */
582 static __noinline struct in6_addr *
583 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
584 {
585 	struct ip6_hdr hdr;
586 	struct mbuf *m;
587 
588 	if (ln->la_hold == NULL)
589 		return (NULL);
590 
591 	/*
592 	 * assume every packet in la_hold has the same IP header
593 	 */
594 	m = ln->la_hold;
595 	if (sizeof(hdr) > m->m_len)
596 		return (NULL);
597 
598 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
599 	*src = hdr.ip6_src;
600 
601 	return (src);
602 }
603 
604 /*
605  * Checks if we need to switch from STALE state.
606  *
607  * RFC 4861 requires switching from STALE to DELAY state
608  * on first packet matching entry, waiting V_nd6_delay and
609  * transition to PROBE state (if upper layer confirmation was
610  * not received).
611  *
612  * This code performs a bit differently:
613  * On packet hit we don't change state (but desired state
614  * can be guessed by control plane). However, after V_nd6_delay
615  * seconds code will transition to PROBE state (so DELAY state
616  * is kinda skipped in most situations).
617  *
618  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
619  * we perform the following upon entering STALE state:
620  *
621  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
622  * if packet was transmitted at the start of given interval, we
623  * would be able to switch to PROBE state in V_nd6_delay seconds
624  * as user expects.
625  *
626  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
627  * lle in STALE state (remaining timer value stored in lle_remtime).
628  *
629  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
630  * seconds ago.
631  *
632  * Returns non-zero value if the entry is still STALE (storing
633  * the next timer interval in @pdelay).
634  *
635  * Returns zero value if original timer expired or we need to switch to
636  * PROBE (store that in @do_switch variable).
637  */
638 static int
639 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
640 {
641 	int nd_delay, nd_gctimer;
642 	time_t lle_hittime;
643 	long delay;
644 
645 	*do_switch = 0;
646 	nd_gctimer = V_nd6_gctimer;
647 	nd_delay = V_nd6_delay;
648 
649 	lle_hittime = llentry_get_hittime(lle);
650 
651 	if (lle_hittime == 0) {
652 		/*
653 		 * Datapath feedback has been requested upon entering
654 		 * STALE state. No packets has been passed using this lle.
655 		 * Ask for the timer reschedule and keep STALE state.
656 		 */
657 		delay = (long)(MIN(nd_gctimer, nd_delay));
658 		delay *= hz;
659 		if (lle->lle_remtime > delay)
660 			lle->lle_remtime -= delay;
661 		else {
662 			delay = lle->lle_remtime;
663 			lle->lle_remtime = 0;
664 		}
665 
666 		if (delay == 0) {
667 			/*
668 			 * The original ng6_gctime timeout ended,
669 			 * no more rescheduling.
670 			 */
671 			return (0);
672 		}
673 
674 		*pdelay = delay;
675 		return (1);
676 	}
677 
678 	/*
679 	 * Packet received. Verify timestamp
680 	 */
681 	delay = (long)(time_uptime - lle_hittime);
682 	if (delay < nd_delay) {
683 		/*
684 		 * V_nd6_delay still not passed since the first
685 		 * hit in STALE state.
686 		 * Reschedule timer and return.
687 		 */
688 		*pdelay = (long)(nd_delay - delay) * hz;
689 		return (1);
690 	}
691 
692 	/* Request switching to probe */
693 	*do_switch = 1;
694 	return (0);
695 }
696 
697 /*
698  * Switch @lle state to new state optionally arming timers.
699  *
700  * Set noinline to be dtrace-friendly
701  */
702 __noinline void
703 nd6_llinfo_setstate(struct llentry *lle, int newstate)
704 {
705 	struct ifnet *ifp;
706 	int nd_gctimer, nd_delay;
707 	long delay, remtime;
708 
709 	delay = 0;
710 	remtime = 0;
711 
712 	switch (newstate) {
713 	case ND6_LLINFO_INCOMPLETE:
714 		ifp = lle->lle_tbl->llt_ifp;
715 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
716 		break;
717 	case ND6_LLINFO_REACHABLE:
718 		if (!ND6_LLINFO_PERMANENT(lle)) {
719 			ifp = lle->lle_tbl->llt_ifp;
720 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
721 		}
722 		break;
723 	case ND6_LLINFO_STALE:
724 
725 		llentry_request_feedback(lle);
726 		nd_delay = V_nd6_delay;
727 		nd_gctimer = V_nd6_gctimer;
728 
729 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
730 		remtime = (long)nd_gctimer * hz - delay;
731 		break;
732 	case ND6_LLINFO_DELAY:
733 		lle->la_asked = 0;
734 		delay = (long)V_nd6_delay * hz;
735 		break;
736 	}
737 
738 	if (delay > 0)
739 		nd6_llinfo_settimer_locked(lle, delay);
740 
741 	lle->lle_remtime = remtime;
742 	lle->ln_state = newstate;
743 }
744 
745 /*
746  * Timer-dependent part of nd state machine.
747  *
748  * Set noinline to be dtrace-friendly
749  */
750 static __noinline void
751 nd6_llinfo_timer(void *arg)
752 {
753 	struct epoch_tracker et;
754 	struct llentry *ln;
755 	struct in6_addr *dst, *pdst, *psrc, src;
756 	struct ifnet *ifp;
757 	struct nd_ifinfo *ndi;
758 	int do_switch, send_ns;
759 	long delay;
760 
761 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
762 	ln = (struct llentry *)arg;
763 	ifp = lltable_get_ifp(ln->lle_tbl);
764 	CURVNET_SET(ifp->if_vnet);
765 
766 	ND6_RLOCK();
767 	LLE_WLOCK(ln);
768 	if (callout_pending(&ln->lle_timer)) {
769 		/*
770 		 * Here we are a bit odd here in the treatment of
771 		 * active/pending. If the pending bit is set, it got
772 		 * rescheduled before I ran. The active
773 		 * bit we ignore, since if it was stopped
774 		 * in ll_tablefree() and was currently running
775 		 * it would have return 0 so the code would
776 		 * not have deleted it since the callout could
777 		 * not be stopped so we want to go through
778 		 * with the delete here now. If the callout
779 		 * was restarted, the pending bit will be back on and
780 		 * we just want to bail since the callout_reset would
781 		 * return 1 and our reference would have been removed
782 		 * by nd6_llinfo_settimer_locked above since canceled
783 		 * would have been 1.
784 		 */
785 		LLE_WUNLOCK(ln);
786 		ND6_RUNLOCK();
787 		CURVNET_RESTORE();
788 		return;
789 	}
790 	NET_EPOCH_ENTER(et);
791 	ndi = ND_IFINFO(ifp);
792 	send_ns = 0;
793 	dst = &ln->r_l3addr.addr6;
794 	pdst = dst;
795 
796 	if (ln->ln_ntick > 0) {
797 		if (ln->ln_ntick > INT_MAX) {
798 			ln->ln_ntick -= INT_MAX;
799 			nd6_llinfo_settimer_locked(ln, INT_MAX);
800 		} else {
801 			ln->ln_ntick = 0;
802 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
803 		}
804 		goto done;
805 	}
806 
807 	if (ln->la_flags & LLE_STATIC) {
808 		goto done;
809 	}
810 
811 	if (ln->la_flags & LLE_DELETED) {
812 		nd6_free(&ln, 0);
813 		goto done;
814 	}
815 
816 	switch (ln->ln_state) {
817 	case ND6_LLINFO_INCOMPLETE:
818 		if (ln->la_asked < V_nd6_mmaxtries) {
819 			ln->la_asked++;
820 			send_ns = 1;
821 			/* Send NS to multicast address */
822 			pdst = NULL;
823 		} else {
824 			struct mbuf *m;
825 
826 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
827 
828 			m = ln->la_hold;
829 			if (m != NULL) {
830 				/*
831 				 * assuming every packet in la_hold has the
832 				 * same IP header.  Send error after unlock.
833 				 */
834 				ln->la_hold = m->m_nextpkt;
835 				m->m_nextpkt = NULL;
836 				ln->la_numheld--;
837 			}
838 			nd6_free(&ln, 0);
839 			if (m != NULL) {
840 				struct mbuf *n = m;
841 
842 				/*
843 				 * if there are any ummapped mbufs, we
844 				 * must free them, rather than using
845 				 * them for an ICMP, as they cannot be
846 				 * checksummed.
847 				 */
848 				while ((n = n->m_next) != NULL) {
849 					if (n->m_flags & M_EXTPG)
850 						break;
851 				}
852 				if (n != NULL) {
853 					m_freem(m);
854 					m = NULL;
855 				} else {
856 					icmp6_error2(m, ICMP6_DST_UNREACH,
857 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
858 				}
859 			}
860 		}
861 		break;
862 	case ND6_LLINFO_REACHABLE:
863 		if (!ND6_LLINFO_PERMANENT(ln))
864 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
865 		break;
866 
867 	case ND6_LLINFO_STALE:
868 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
869 			/*
870 			 * No packet has used this entry and GC timeout
871 			 * has not been passed. Reschedule timer and
872 			 * return.
873 			 */
874 			nd6_llinfo_settimer_locked(ln, delay);
875 			break;
876 		}
877 
878 		if (do_switch == 0) {
879 			/*
880 			 * GC timer has ended and entry hasn't been used.
881 			 * Run Garbage collector (RFC 4861, 5.3)
882 			 */
883 			if (!ND6_LLINFO_PERMANENT(ln))
884 				nd6_free(&ln, 1);
885 			break;
886 		}
887 
888 		/* Entry has been used AND delay timer has ended. */
889 
890 		/* FALLTHROUGH */
891 
892 	case ND6_LLINFO_DELAY:
893 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
894 			/* We need NUD */
895 			ln->la_asked = 1;
896 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
897 			send_ns = 1;
898 		} else
899 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
900 		break;
901 	case ND6_LLINFO_PROBE:
902 		if (ln->la_asked < V_nd6_umaxtries) {
903 			ln->la_asked++;
904 			send_ns = 1;
905 		} else {
906 			nd6_free(&ln, 0);
907 		}
908 		break;
909 	default:
910 		panic("%s: paths in a dark night can be confusing: %d",
911 		    __func__, ln->ln_state);
912 	}
913 done:
914 	if (ln != NULL)
915 		ND6_RUNLOCK();
916 	if (send_ns != 0) {
917 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
918 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
919 		LLE_FREE_LOCKED(ln);
920 		ln = NULL;
921 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
922 	}
923 
924 	if (ln != NULL)
925 		LLE_FREE_LOCKED(ln);
926 	NET_EPOCH_EXIT(et);
927 	CURVNET_RESTORE();
928 }
929 
930 /*
931  * ND6 timer routine to expire default route list and prefix list
932  */
933 void
934 nd6_timer(void *arg)
935 {
936 	CURVNET_SET((struct vnet *) arg);
937 	struct epoch_tracker et;
938 	struct nd_prhead prl;
939 	struct nd_prefix *pr, *npr;
940 	struct ifnet *ifp;
941 	struct in6_ifaddr *ia6, *nia6;
942 	uint64_t genid;
943 
944 	LIST_INIT(&prl);
945 
946 	NET_EPOCH_ENTER(et);
947 	nd6_defrouter_timer();
948 
949 	/*
950 	 * expire interface addresses.
951 	 * in the past the loop was inside prefix expiry processing.
952 	 * However, from a stricter speci-confrmance standpoint, we should
953 	 * rather separate address lifetimes and prefix lifetimes.
954 	 *
955 	 * XXXRW: in6_ifaddrhead locking.
956 	 */
957   addrloop:
958 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
959 		/* check address lifetime */
960 		if (IFA6_IS_INVALID(ia6)) {
961 			int regen = 0;
962 
963 			/*
964 			 * If the expiring address is temporary, try
965 			 * regenerating a new one.  This would be useful when
966 			 * we suspended a laptop PC, then turned it on after a
967 			 * period that could invalidate all temporary
968 			 * addresses.  Although we may have to restart the
969 			 * loop (see below), it must be after purging the
970 			 * address.  Otherwise, we'd see an infinite loop of
971 			 * regeneration.
972 			 */
973 			if (V_ip6_use_tempaddr &&
974 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
975 				if (regen_tmpaddr(ia6) == 0)
976 					regen = 1;
977 			}
978 
979 			in6_purgeaddr(&ia6->ia_ifa);
980 
981 			if (regen)
982 				goto addrloop; /* XXX: see below */
983 		} else if (IFA6_IS_DEPRECATED(ia6)) {
984 			int oldflags = ia6->ia6_flags;
985 
986 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
987 
988 			/*
989 			 * If a temporary address has just become deprecated,
990 			 * regenerate a new one if possible.
991 			 */
992 			if (V_ip6_use_tempaddr &&
993 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
994 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
995 				if (regen_tmpaddr(ia6) == 0) {
996 					/*
997 					 * A new temporary address is
998 					 * generated.
999 					 * XXX: this means the address chain
1000 					 * has changed while we are still in
1001 					 * the loop.  Although the change
1002 					 * would not cause disaster (because
1003 					 * it's not a deletion, but an
1004 					 * addition,) we'd rather restart the
1005 					 * loop just for safety.  Or does this
1006 					 * significantly reduce performance??
1007 					 */
1008 					goto addrloop;
1009 				}
1010 			}
1011 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1012 			/*
1013 			 * Schedule DAD for a tentative address.  This happens
1014 			 * if the interface was down or not running
1015 			 * when the address was configured.
1016 			 */
1017 			int delay;
1018 
1019 			delay = arc4random() %
1020 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1021 			nd6_dad_start((struct ifaddr *)ia6, delay);
1022 		} else {
1023 			/*
1024 			 * Check status of the interface.  If it is down,
1025 			 * mark the address as tentative for future DAD.
1026 			 */
1027 			ifp = ia6->ia_ifp;
1028 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1029 			    ((ifp->if_flags & IFF_UP) == 0 ||
1030 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1031 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1032 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1033 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1034 			}
1035 
1036 			/*
1037 			 * A new RA might have made a deprecated address
1038 			 * preferred.
1039 			 */
1040 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1041 		}
1042 	}
1043 	NET_EPOCH_EXIT(et);
1044 
1045 	ND6_WLOCK();
1046 restart:
1047 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1048 		/*
1049 		 * Expire prefixes. Since the pltime is only used for
1050 		 * autoconfigured addresses, pltime processing for prefixes is
1051 		 * not necessary.
1052 		 *
1053 		 * Only unlink after all derived addresses have expired. This
1054 		 * may not occur until two hours after the prefix has expired
1055 		 * per RFC 4862. If the prefix expires before its derived
1056 		 * addresses, mark it off-link. This will be done automatically
1057 		 * after unlinking if no address references remain.
1058 		 */
1059 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1060 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1061 			continue;
1062 
1063 		if (pr->ndpr_addrcnt == 0) {
1064 			nd6_prefix_unlink(pr, &prl);
1065 			continue;
1066 		}
1067 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1068 			genid = V_nd6_list_genid;
1069 			nd6_prefix_ref(pr);
1070 			ND6_WUNLOCK();
1071 			ND6_ONLINK_LOCK();
1072 			(void)nd6_prefix_offlink(pr);
1073 			ND6_ONLINK_UNLOCK();
1074 			ND6_WLOCK();
1075 			nd6_prefix_rele(pr);
1076 			if (genid != V_nd6_list_genid)
1077 				goto restart;
1078 		}
1079 	}
1080 	ND6_WUNLOCK();
1081 
1082 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1083 		LIST_REMOVE(pr, ndpr_entry);
1084 		nd6_prefix_del(pr);
1085 	}
1086 
1087 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1088 	    nd6_timer, curvnet);
1089 
1090 	CURVNET_RESTORE();
1091 }
1092 
1093 /*
1094  * ia6 - deprecated/invalidated temporary address
1095  */
1096 static int
1097 regen_tmpaddr(struct in6_ifaddr *ia6)
1098 {
1099 	struct ifaddr *ifa;
1100 	struct ifnet *ifp;
1101 	struct in6_ifaddr *public_ifa6 = NULL;
1102 
1103 	NET_EPOCH_ASSERT();
1104 
1105 	ifp = ia6->ia_ifa.ifa_ifp;
1106 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1107 		struct in6_ifaddr *it6;
1108 
1109 		if (ifa->ifa_addr->sa_family != AF_INET6)
1110 			continue;
1111 
1112 		it6 = (struct in6_ifaddr *)ifa;
1113 
1114 		/* ignore no autoconf addresses. */
1115 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1116 			continue;
1117 
1118 		/* ignore autoconf addresses with different prefixes. */
1119 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1120 			continue;
1121 
1122 		/*
1123 		 * Now we are looking at an autoconf address with the same
1124 		 * prefix as ours.  If the address is temporary and is still
1125 		 * preferred, do not create another one.  It would be rare, but
1126 		 * could happen, for example, when we resume a laptop PC after
1127 		 * a long period.
1128 		 */
1129 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1130 		    !IFA6_IS_DEPRECATED(it6)) {
1131 			public_ifa6 = NULL;
1132 			break;
1133 		}
1134 
1135 		/*
1136 		 * This is a public autoconf address that has the same prefix
1137 		 * as ours.  If it is preferred, keep it.  We can't break the
1138 		 * loop here, because there may be a still-preferred temporary
1139 		 * address with the prefix.
1140 		 */
1141 		if (!IFA6_IS_DEPRECATED(it6))
1142 			public_ifa6 = it6;
1143 	}
1144 	if (public_ifa6 != NULL)
1145 		ifa_ref(&public_ifa6->ia_ifa);
1146 
1147 	if (public_ifa6 != NULL) {
1148 		int e;
1149 
1150 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1151 			ifa_free(&public_ifa6->ia_ifa);
1152 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1153 			    " tmp addr,errno=%d\n", e);
1154 			return (-1);
1155 		}
1156 		ifa_free(&public_ifa6->ia_ifa);
1157 		return (0);
1158 	}
1159 
1160 	return (-1);
1161 }
1162 
1163 /*
1164  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1165  * cache entries are freed in in6_domifdetach().
1166  */
1167 void
1168 nd6_purge(struct ifnet *ifp)
1169 {
1170 	struct nd_prhead prl;
1171 	struct nd_prefix *pr, *npr;
1172 
1173 	LIST_INIT(&prl);
1174 
1175 	/* Purge default router list entries toward ifp. */
1176 	nd6_defrouter_purge(ifp);
1177 
1178 	ND6_WLOCK();
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 prefix objects. */
1190 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1191 		LIST_REMOVE(pr, ndpr_entry);
1192 		nd6_prefix_del(pr);
1193 	}
1194 
1195 	/* cancel default outgoing interface setting */
1196 	if (V_nd6_defifindex == ifp->if_index)
1197 		nd6_setdefaultiface(0);
1198 
1199 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1200 		/* Refresh default router list. */
1201 		defrouter_select_fib(ifp->if_fib);
1202 	}
1203 }
1204 
1205 /*
1206  * the caller acquires and releases the lock on the lltbls
1207  * Returns the llentry locked
1208  */
1209 struct llentry *
1210 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1211 {
1212 	struct sockaddr_in6 sin6;
1213 	struct llentry *ln;
1214 
1215 	bzero(&sin6, sizeof(sin6));
1216 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1217 	sin6.sin6_family = AF_INET6;
1218 	sin6.sin6_addr = *addr6;
1219 
1220 	IF_AFDATA_LOCK_ASSERT(ifp);
1221 
1222 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1223 
1224 	return (ln);
1225 }
1226 
1227 static struct llentry *
1228 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1229 {
1230 	struct sockaddr_in6 sin6;
1231 	struct llentry *ln;
1232 
1233 	bzero(&sin6, sizeof(sin6));
1234 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1235 	sin6.sin6_family = AF_INET6;
1236 	sin6.sin6_addr = *addr6;
1237 
1238 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1239 	if (ln != NULL)
1240 		ln->ln_state = ND6_LLINFO_NOSTATE;
1241 
1242 	return (ln);
1243 }
1244 
1245 /*
1246  * Test whether a given IPv6 address can be a neighbor.
1247  */
1248 static bool
1249 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1250 {
1251 
1252 	/*
1253 	 * A link-local address is always a neighbor.
1254 	 * XXX: a link does not necessarily specify a single interface.
1255 	 */
1256 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1257 		struct sockaddr_in6 sin6_copy;
1258 		u_int32_t zone;
1259 
1260 		/*
1261 		 * We need sin6_copy since sa6_recoverscope() may modify the
1262 		 * content (XXX).
1263 		 */
1264 		sin6_copy = *addr;
1265 		if (sa6_recoverscope(&sin6_copy))
1266 			return (0); /* XXX: should be impossible */
1267 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1268 			return (0);
1269 		if (sin6_copy.sin6_scope_id == zone)
1270 			return (1);
1271 		else
1272 			return (0);
1273 	}
1274 	/* Checking global unicast */
1275 
1276 	/* If an address is directly reachable, it is a neigbor */
1277 	struct nhop_object *nh;
1278 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1279 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1280 		return (true);
1281 
1282 	/*
1283 	 * Check prefixes with desired on-link state, as some may be not
1284 	 * installed in the routing table.
1285 	 */
1286 	bool matched = false;
1287 	struct nd_prefix *pr;
1288 	ND6_RLOCK();
1289 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1290 		if (pr->ndpr_ifp != ifp)
1291 			continue;
1292 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1293 			continue;
1294 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1295 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1296 			matched = true;
1297 			break;
1298 		}
1299 	}
1300 	ND6_RUNLOCK();
1301 	if (matched)
1302 		return (true);
1303 
1304 	/*
1305 	 * If the address is assigned on the node of the other side of
1306 	 * a p2p interface, the address should be a neighbor.
1307 	 */
1308 	if (ifp->if_flags & IFF_POINTOPOINT) {
1309 		struct ifaddr *ifa;
1310 
1311 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1312 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1313 				continue;
1314 			if (ifa->ifa_dstaddr != NULL &&
1315 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1316 				return (true);
1317 			}
1318 		}
1319 	}
1320 
1321 	/*
1322 	 * If the default router list is empty, all addresses are regarded
1323 	 * as on-link, and thus, as a neighbor.
1324 	 */
1325 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1326 	    nd6_defrouter_list_empty() &&
1327 	    V_nd6_defifindex == ifp->if_index) {
1328 		return (1);
1329 	}
1330 
1331 	return (0);
1332 }
1333 
1334 /*
1335  * Detect if a given IPv6 address identifies a neighbor on a given link.
1336  * XXX: should take care of the destination of a p2p link?
1337  */
1338 int
1339 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1340 {
1341 	struct llentry *lle;
1342 	int rc = 0;
1343 
1344 	NET_EPOCH_ASSERT();
1345 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1346 	if (nd6_is_new_addr_neighbor(addr, ifp))
1347 		return (1);
1348 
1349 	/*
1350 	 * Even if the address matches none of our addresses, it might be
1351 	 * in the neighbor cache.
1352 	 */
1353 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1354 		LLE_RUNLOCK(lle);
1355 		rc = 1;
1356 	}
1357 	return (rc);
1358 }
1359 
1360 static __noinline void
1361 nd6_free_children(struct llentry *lle)
1362 {
1363 	struct llentry *child_lle;
1364 
1365 	NET_EPOCH_ASSERT();
1366 	LLE_WLOCK_ASSERT(lle);
1367 
1368 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1369 		LLE_WLOCK(child_lle);
1370 		lltable_unlink_child_entry(child_lle);
1371 		llentry_free(child_lle);
1372 	}
1373 }
1374 
1375 /*
1376  * Tries to update @lle address/prepend data with new @lladdr.
1377  *
1378  * Returns true on success.
1379  * In any case, @lle is returned wlocked.
1380  */
1381 static __noinline bool
1382 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1383 {
1384 	u_char buf[LLE_MAX_LINKHDR];
1385 	int fam, off;
1386 	size_t sz;
1387 
1388 	sz = sizeof(buf);
1389 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1390 		return (false);
1391 
1392 	/* Update data */
1393 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1394 
1395 	struct llentry *child_lle;
1396 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1397 		LLE_WLOCK(child_lle);
1398 		fam = child_lle->r_family;
1399 		sz = sizeof(buf);
1400 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1401 			/* success */
1402 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1403 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1404 		}
1405 		LLE_WUNLOCK(child_lle);
1406 	}
1407 
1408 	return (true);
1409 }
1410 
1411 bool
1412 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1413 {
1414 	NET_EPOCH_ASSERT();
1415 	LLE_WLOCK_ASSERT(lle);
1416 
1417 	if (!lltable_acquire_wlock(ifp, lle))
1418 		return (false);
1419 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1420 	IF_AFDATA_WUNLOCK(ifp);
1421 
1422 	return (ret);
1423 }
1424 
1425 /*
1426  * Free an nd6 llinfo entry.
1427  * Since the function would cause significant changes in the kernel, DO NOT
1428  * make it global, unless you have a strong reason for the change, and are sure
1429  * that the change is safe.
1430  *
1431  * Set noinline to be dtrace-friendly
1432  */
1433 static __noinline void
1434 nd6_free(struct llentry **lnp, int gc)
1435 {
1436 	struct ifnet *ifp;
1437 	struct llentry *ln;
1438 	struct nd_defrouter *dr;
1439 
1440 	ln = *lnp;
1441 	*lnp = NULL;
1442 
1443 	LLE_WLOCK_ASSERT(ln);
1444 	ND6_RLOCK_ASSERT();
1445 
1446 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1447 
1448 	ifp = lltable_get_ifp(ln->lle_tbl);
1449 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1450 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1451 	else
1452 		dr = NULL;
1453 	ND6_RUNLOCK();
1454 
1455 	if ((ln->la_flags & LLE_DELETED) == 0)
1456 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1457 
1458 	/*
1459 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1460 	 * even though it is not harmful, it was not really necessary.
1461 	 */
1462 
1463 	/* cancel timer */
1464 	nd6_llinfo_settimer_locked(ln, -1);
1465 
1466 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1467 		if (dr != NULL && dr->expire &&
1468 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1469 			/*
1470 			 * If the reason for the deletion is just garbage
1471 			 * collection, and the neighbor is an active default
1472 			 * router, do not delete it.  Instead, reset the GC
1473 			 * timer using the router's lifetime.
1474 			 * Simply deleting the entry would affect default
1475 			 * router selection, which is not necessarily a good
1476 			 * thing, especially when we're using router preference
1477 			 * values.
1478 			 * XXX: the check for ln_state would be redundant,
1479 			 *      but we intentionally keep it just in case.
1480 			 */
1481 			if (dr->expire > time_uptime)
1482 				nd6_llinfo_settimer_locked(ln,
1483 				    (dr->expire - time_uptime) * hz);
1484 			else
1485 				nd6_llinfo_settimer_locked(ln,
1486 				    (long)V_nd6_gctimer * hz);
1487 
1488 			LLE_REMREF(ln);
1489 			LLE_WUNLOCK(ln);
1490 			defrouter_rele(dr);
1491 			return;
1492 		}
1493 
1494 		if (dr) {
1495 			/*
1496 			 * Unreachability of a router might affect the default
1497 			 * router selection and on-link detection of advertised
1498 			 * prefixes.
1499 			 */
1500 
1501 			/*
1502 			 * Temporarily fake the state to choose a new default
1503 			 * router and to perform on-link determination of
1504 			 * prefixes correctly.
1505 			 * Below the state will be set correctly,
1506 			 * or the entry itself will be deleted.
1507 			 */
1508 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1509 		}
1510 
1511 		if (ln->ln_router || dr) {
1512 			/*
1513 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1514 			 * rnh and for the calls to pfxlist_onlink_check() and
1515 			 * defrouter_select_fib() in the block further down for calls
1516 			 * into nd6_lookup().  We still hold a ref.
1517 			 */
1518 			LLE_WUNLOCK(ln);
1519 
1520 			/*
1521 			 * rt6_flush must be called whether or not the neighbor
1522 			 * is in the Default Router List.
1523 			 * See a corresponding comment in nd6_na_input().
1524 			 */
1525 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1526 		}
1527 
1528 		if (dr) {
1529 			/*
1530 			 * Since defrouter_select_fib() does not affect the
1531 			 * on-link determination and MIP6 needs the check
1532 			 * before the default router selection, we perform
1533 			 * the check now.
1534 			 */
1535 			pfxlist_onlink_check();
1536 
1537 			/*
1538 			 * Refresh default router list.
1539 			 */
1540 			defrouter_select_fib(dr->ifp->if_fib);
1541 		}
1542 
1543 		/*
1544 		 * If this entry was added by an on-link redirect, remove the
1545 		 * corresponding host route.
1546 		 */
1547 		if (ln->la_flags & LLE_REDIRECT)
1548 			nd6_free_redirect(ln);
1549 
1550 		if (ln->ln_router || dr)
1551 			LLE_WLOCK(ln);
1552 	}
1553 
1554 	/*
1555 	 * Save to unlock. We still hold an extra reference and will not
1556 	 * free(9) in llentry_free() if someone else holds one as well.
1557 	 */
1558 	LLE_WUNLOCK(ln);
1559 	IF_AFDATA_LOCK(ifp);
1560 	LLE_WLOCK(ln);
1561 	/* Guard against race with other llentry_free(). */
1562 	if (ln->la_flags & LLE_LINKED) {
1563 		/* Remove callout reference */
1564 		LLE_REMREF(ln);
1565 		lltable_unlink_entry(ln->lle_tbl, ln);
1566 	}
1567 	IF_AFDATA_UNLOCK(ifp);
1568 
1569 	nd6_free_children(ln);
1570 
1571 	llentry_free(ln);
1572 	if (dr != NULL)
1573 		defrouter_rele(dr);
1574 }
1575 
1576 static int
1577 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1578 {
1579 
1580 	if (nh->nh_flags & NHF_REDIRECT)
1581 		return (1);
1582 
1583 	return (0);
1584 }
1585 
1586 /*
1587  * Remove the rtentry for the given llentry,
1588  * both of which were installed by a redirect.
1589  */
1590 static void
1591 nd6_free_redirect(const struct llentry *ln)
1592 {
1593 	int fibnum;
1594 	struct sockaddr_in6 sin6;
1595 	struct rib_cmd_info rc;
1596 	struct epoch_tracker et;
1597 
1598 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1599 
1600 	NET_EPOCH_ENTER(et);
1601 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1602 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1603 		    nd6_isdynrte, NULL, 0, &rc);
1604 	NET_EPOCH_EXIT(et);
1605 }
1606 
1607 /*
1608  * Updates status of the default router route.
1609  */
1610 static void
1611 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1612 {
1613 	struct nd_defrouter *dr;
1614 	struct nhop_object *nh;
1615 
1616 	nh = rc->rc_nh_old;
1617 	if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1618 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1619 		if (dr != NULL) {
1620 			dr->installed = 0;
1621 			defrouter_rele(dr);
1622 		}
1623 	}
1624 }
1625 
1626 void
1627 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1628 {
1629 #ifdef ROUTE_MPATH
1630 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1631 	if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1632 		check_release_defrouter(rc, NULL);
1633 #else
1634 	check_release_defrouter(rc, NULL);
1635 #endif
1636 }
1637 
1638 int
1639 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1640 {
1641 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1642 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1643 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1644 	struct epoch_tracker et;
1645 	int error = 0;
1646 
1647 	if (ifp->if_afdata[AF_INET6] == NULL)
1648 		return (EPFNOSUPPORT);
1649 	switch (cmd) {
1650 	case OSIOCGIFINFO_IN6:
1651 #define ND	ndi->ndi
1652 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1653 		bzero(&ND, sizeof(ND));
1654 		ND.linkmtu = IN6_LINKMTU(ifp);
1655 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1656 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1657 		ND.reachable = ND_IFINFO(ifp)->reachable;
1658 		ND.retrans = ND_IFINFO(ifp)->retrans;
1659 		ND.flags = ND_IFINFO(ifp)->flags;
1660 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1661 		ND.chlim = ND_IFINFO(ifp)->chlim;
1662 		break;
1663 	case SIOCGIFINFO_IN6:
1664 		ND = *ND_IFINFO(ifp);
1665 		break;
1666 	case SIOCSIFINFO_IN6:
1667 		/*
1668 		 * used to change host variables from userland.
1669 		 * intended for a use on router to reflect RA configurations.
1670 		 */
1671 		/* 0 means 'unspecified' */
1672 		if (ND.linkmtu != 0) {
1673 			if (ND.linkmtu < IPV6_MMTU ||
1674 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1675 				error = EINVAL;
1676 				break;
1677 			}
1678 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1679 		}
1680 
1681 		if (ND.basereachable != 0) {
1682 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1683 
1684 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1685 			if (ND.basereachable != obasereachable)
1686 				ND_IFINFO(ifp)->reachable =
1687 				    ND_COMPUTE_RTIME(ND.basereachable);
1688 		}
1689 		if (ND.retrans != 0)
1690 			ND_IFINFO(ifp)->retrans = ND.retrans;
1691 		if (ND.chlim != 0)
1692 			ND_IFINFO(ifp)->chlim = ND.chlim;
1693 		/* FALLTHROUGH */
1694 	case SIOCSIFINFO_FLAGS:
1695 	{
1696 		struct ifaddr *ifa;
1697 		struct in6_ifaddr *ia;
1698 
1699 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1700 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1701 			/* ifdisabled 1->0 transision */
1702 
1703 			/*
1704 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1705 			 * has an link-local address with IN6_IFF_DUPLICATED,
1706 			 * do not clear ND6_IFF_IFDISABLED.
1707 			 * See RFC 4862, Section 5.4.5.
1708 			 */
1709 			NET_EPOCH_ENTER(et);
1710 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1711 				if (ifa->ifa_addr->sa_family != AF_INET6)
1712 					continue;
1713 				ia = (struct in6_ifaddr *)ifa;
1714 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1715 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1716 					break;
1717 			}
1718 			NET_EPOCH_EXIT(et);
1719 
1720 			if (ifa != NULL) {
1721 				/* LLA is duplicated. */
1722 				ND.flags |= ND6_IFF_IFDISABLED;
1723 				log(LOG_ERR, "Cannot enable an interface"
1724 				    " with a link-local address marked"
1725 				    " duplicate.\n");
1726 			} else {
1727 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1728 				if (ifp->if_flags & IFF_UP)
1729 					in6_if_up(ifp);
1730 			}
1731 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1732 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1733 			/* ifdisabled 0->1 transision */
1734 			/* Mark all IPv6 address as tentative. */
1735 
1736 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1737 			if (V_ip6_dad_count > 0 &&
1738 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1739 				NET_EPOCH_ENTER(et);
1740 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1741 				    ifa_link) {
1742 					if (ifa->ifa_addr->sa_family !=
1743 					    AF_INET6)
1744 						continue;
1745 					ia = (struct in6_ifaddr *)ifa;
1746 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1747 				}
1748 				NET_EPOCH_EXIT(et);
1749 			}
1750 		}
1751 
1752 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1753 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1754 				/* auto_linklocal 0->1 transision */
1755 
1756 				/* If no link-local address on ifp, configure */
1757 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1758 				in6_ifattach(ifp, NULL);
1759 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1760 			    ifp->if_flags & IFF_UP) {
1761 				/*
1762 				 * When the IF already has
1763 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1764 				 * address is assigned, and IFF_UP, try to
1765 				 * assign one.
1766 				 */
1767 				NET_EPOCH_ENTER(et);
1768 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1769 				    ifa_link) {
1770 					if (ifa->ifa_addr->sa_family !=
1771 					    AF_INET6)
1772 						continue;
1773 					ia = (struct in6_ifaddr *)ifa;
1774 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1775 						break;
1776 				}
1777 				NET_EPOCH_EXIT(et);
1778 				if (ifa != NULL)
1779 					/* No LLA is configured. */
1780 					in6_ifattach(ifp, NULL);
1781 			}
1782 		}
1783 		ND_IFINFO(ifp)->flags = ND.flags;
1784 		break;
1785 	}
1786 #undef ND
1787 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1788 		/* sync kernel routing table with the default router list */
1789 		defrouter_reset();
1790 		defrouter_select_fib(RT_ALL_FIBS);
1791 		break;
1792 	case SIOCSPFXFLUSH_IN6:
1793 	{
1794 		/* flush all the prefix advertised by routers */
1795 		struct in6_ifaddr *ia, *ia_next;
1796 		struct nd_prefix *pr, *next;
1797 		struct nd_prhead prl;
1798 
1799 		LIST_INIT(&prl);
1800 
1801 		ND6_WLOCK();
1802 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1803 			if (pr->ndpr_raf_ra_derived)
1804 				nd6_prefix_unlink(pr, &prl);
1805 		}
1806 		ND6_WUNLOCK();
1807 
1808 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1809 			LIST_REMOVE(pr, ndpr_entry);
1810 			/* XXXRW: in6_ifaddrhead locking. */
1811 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1812 			    ia_next) {
1813 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1814 					continue;
1815 
1816 				if (ia->ia6_ndpr == pr)
1817 					in6_purgeaddr(&ia->ia_ifa);
1818 			}
1819 			nd6_prefix_del(pr);
1820 		}
1821 		break;
1822 	}
1823 	case SIOCSRTRFLUSH_IN6:
1824 	{
1825 		/* flush all the default routers */
1826 
1827 		defrouter_reset();
1828 		nd6_defrouter_flush_all();
1829 		defrouter_select_fib(RT_ALL_FIBS);
1830 		break;
1831 	}
1832 	case SIOCGNBRINFO_IN6:
1833 	{
1834 		struct llentry *ln;
1835 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1836 
1837 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1838 			return (error);
1839 
1840 		NET_EPOCH_ENTER(et);
1841 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1842 		NET_EPOCH_EXIT(et);
1843 
1844 		if (ln == NULL) {
1845 			error = EINVAL;
1846 			break;
1847 		}
1848 		nbi->state = ln->ln_state;
1849 		nbi->asked = ln->la_asked;
1850 		nbi->isrouter = ln->ln_router;
1851 		if (ln->la_expire == 0)
1852 			nbi->expire = 0;
1853 		else
1854 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1855 			    (time_second - time_uptime);
1856 		LLE_RUNLOCK(ln);
1857 		break;
1858 	}
1859 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1860 		ndif->ifindex = V_nd6_defifindex;
1861 		break;
1862 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1863 		return (nd6_setdefaultiface(ndif->ifindex));
1864 	}
1865 	return (error);
1866 }
1867 
1868 /*
1869  * Calculates new isRouter value based on provided parameters and
1870  * returns it.
1871  */
1872 static int
1873 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1874     int ln_router)
1875 {
1876 
1877 	/*
1878 	 * ICMP6 type dependent behavior.
1879 	 *
1880 	 * NS: clear IsRouter if new entry
1881 	 * RS: clear IsRouter
1882 	 * RA: set IsRouter if there's lladdr
1883 	 * redir: clear IsRouter if new entry
1884 	 *
1885 	 * RA case, (1):
1886 	 * The spec says that we must set IsRouter in the following cases:
1887 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1888 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1889 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1890 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1891 	 * neighbor cache, this is similar to (6).
1892 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1893 	 *
1894 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1895 	 *							D R
1896 	 *	0	n	n	(1)	c   ?     s
1897 	 *	0	y	n	(2)	c   s     s
1898 	 *	0	n	y	(3)	c   s     s
1899 	 *	0	y	y	(4)	c   s     s
1900 	 *	0	y	y	(5)	c   s     s
1901 	 *	1	--	n	(6) c	c	c s
1902 	 *	1	--	y	(7) c	c   s	c s
1903 	 *
1904 	 *					(c=clear s=set)
1905 	 */
1906 	switch (type & 0xff) {
1907 	case ND_NEIGHBOR_SOLICIT:
1908 		/*
1909 		 * New entry must have is_router flag cleared.
1910 		 */
1911 		if (is_new)					/* (6-7) */
1912 			ln_router = 0;
1913 		break;
1914 	case ND_REDIRECT:
1915 		/*
1916 		 * If the icmp is a redirect to a better router, always set the
1917 		 * is_router flag.  Otherwise, if the entry is newly created,
1918 		 * clear the flag.  [RFC 2461, sec 8.3]
1919 		 */
1920 		if (code == ND_REDIRECT_ROUTER)
1921 			ln_router = 1;
1922 		else {
1923 			if (is_new)				/* (6-7) */
1924 				ln_router = 0;
1925 		}
1926 		break;
1927 	case ND_ROUTER_SOLICIT:
1928 		/*
1929 		 * is_router flag must always be cleared.
1930 		 */
1931 		ln_router = 0;
1932 		break;
1933 	case ND_ROUTER_ADVERT:
1934 		/*
1935 		 * Mark an entry with lladdr as a router.
1936 		 */
1937 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1938 		    (is_new && new_addr)) {			/* (7) */
1939 			ln_router = 1;
1940 		}
1941 		break;
1942 	}
1943 
1944 	return (ln_router);
1945 }
1946 
1947 /*
1948  * Create neighbor cache entry and cache link-layer address,
1949  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1950  *
1951  * type - ICMP6 type
1952  * code - type dependent information
1953  *
1954  */
1955 void
1956 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1957     int lladdrlen, int type, int code)
1958 {
1959 	struct llentry *ln = NULL, *ln_tmp;
1960 	int is_newentry;
1961 	int do_update;
1962 	int olladdr;
1963 	int llchange;
1964 	int flags;
1965 	uint16_t router = 0;
1966 	struct mbuf *chain = NULL;
1967 	u_char linkhdr[LLE_MAX_LINKHDR];
1968 	size_t linkhdrsize;
1969 	int lladdr_off;
1970 
1971 	NET_EPOCH_ASSERT();
1972 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1973 
1974 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1975 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1976 
1977 	/* nothing must be updated for unspecified address */
1978 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1979 		return;
1980 
1981 	/*
1982 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1983 	 * the caller.
1984 	 *
1985 	 * XXX If the link does not have link-layer adderss, what should
1986 	 * we do? (ifp->if_addrlen == 0)
1987 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1988 	 * description on it in NS section (RFC 2461 7.2.3).
1989 	 */
1990 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1991 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1992 	is_newentry = 0;
1993 	if (ln == NULL) {
1994 		flags |= LLE_EXCLUSIVE;
1995 		ln = nd6_alloc(from, 0, ifp);
1996 		if (ln == NULL)
1997 			return;
1998 
1999 		/*
2000 		 * Since we already know all the data for the new entry,
2001 		 * fill it before insertion.
2002 		 */
2003 		if (lladdr != NULL) {
2004 			linkhdrsize = sizeof(linkhdr);
2005 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2006 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2007 				lltable_free_entry(LLTABLE6(ifp), ln);
2008 				return;
2009 			}
2010 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2011 			    lladdr_off);
2012 		}
2013 
2014 		IF_AFDATA_WLOCK(ifp);
2015 		LLE_WLOCK(ln);
2016 		/* Prefer any existing lle over newly-created one */
2017 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2018 		if (ln_tmp == NULL)
2019 			lltable_link_entry(LLTABLE6(ifp), ln);
2020 		IF_AFDATA_WUNLOCK(ifp);
2021 		if (ln_tmp == NULL) {
2022 			/* No existing lle, mark as new entry (6,7) */
2023 			is_newentry = 1;
2024 			if (lladdr != NULL) {	/* (7) */
2025 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2026 				EVENTHANDLER_INVOKE(lle_event, ln,
2027 				    LLENTRY_RESOLVED);
2028 			}
2029 		} else {
2030 			lltable_free_entry(LLTABLE6(ifp), ln);
2031 			ln = ln_tmp;
2032 			ln_tmp = NULL;
2033 		}
2034 	}
2035 	/* do nothing if static ndp is set */
2036 	if ((ln->la_flags & LLE_STATIC)) {
2037 		if (flags & LLE_EXCLUSIVE)
2038 			LLE_WUNLOCK(ln);
2039 		else
2040 			LLE_RUNLOCK(ln);
2041 		return;
2042 	}
2043 
2044 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2045 	if (olladdr && lladdr) {
2046 		llchange = bcmp(lladdr, ln->ll_addr,
2047 		    ifp->if_addrlen);
2048 	} else if (!olladdr && lladdr)
2049 		llchange = 1;
2050 	else
2051 		llchange = 0;
2052 
2053 	/*
2054 	 * newentry olladdr  lladdr  llchange	(*=record)
2055 	 *	0	n	n	--	(1)
2056 	 *	0	y	n	--	(2)
2057 	 *	0	n	y	y	(3) * STALE
2058 	 *	0	y	y	n	(4) *
2059 	 *	0	y	y	y	(5) * STALE
2060 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2061 	 *	1	--	y	--	(7) * STALE
2062 	 */
2063 
2064 	do_update = 0;
2065 	if (is_newentry == 0 && llchange != 0) {
2066 		do_update = 1;	/* (3,5) */
2067 
2068 		/*
2069 		 * Record source link-layer address
2070 		 * XXX is it dependent to ifp->if_type?
2071 		 */
2072 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2073 			/* Entry was deleted */
2074 			LLE_WUNLOCK(ln);
2075 			return;
2076 		}
2077 
2078 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2079 
2080 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2081 
2082 		if (ln->la_hold != NULL)
2083 			chain = nd6_grab_holdchain(ln);
2084 	}
2085 
2086 	/* Calculates new router status */
2087 	router = nd6_is_router(type, code, is_newentry, olladdr,
2088 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2089 
2090 	ln->ln_router = router;
2091 	/* Mark non-router redirects with special flag */
2092 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2093 		ln->la_flags |= LLE_REDIRECT;
2094 
2095 	if (flags & LLE_EXCLUSIVE)
2096 		LLE_WUNLOCK(ln);
2097 	else
2098 		LLE_RUNLOCK(ln);
2099 
2100 	if (chain != NULL)
2101 		nd6_flush_holdchain(ifp, ln, chain);
2102 	if (do_update)
2103 		nd6_flush_children_holdchain(ifp, ln);
2104 
2105 	/*
2106 	 * When the link-layer address of a router changes, select the
2107 	 * best router again.  In particular, when the neighbor entry is newly
2108 	 * created, it might affect the selection policy.
2109 	 * Question: can we restrict the first condition to the "is_newentry"
2110 	 * case?
2111 	 * XXX: when we hear an RA from a new router with the link-layer
2112 	 * address option, defrouter_select_fib() is called twice, since
2113 	 * defrtrlist_update called the function as well.  However, I believe
2114 	 * we can compromise the overhead, since it only happens the first
2115 	 * time.
2116 	 * XXX: although defrouter_select_fib() should not have a bad effect
2117 	 * for those are not autoconfigured hosts, we explicitly avoid such
2118 	 * cases for safety.
2119 	 */
2120 	if ((do_update || is_newentry) && router &&
2121 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2122 		/*
2123 		 * guaranteed recursion
2124 		 */
2125 		defrouter_select_fib(ifp->if_fib);
2126 	}
2127 }
2128 
2129 static void
2130 nd6_slowtimo(void *arg)
2131 {
2132 	struct epoch_tracker et;
2133 	CURVNET_SET((struct vnet *) arg);
2134 	struct nd_ifinfo *nd6if;
2135 	struct ifnet *ifp;
2136 
2137 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2138 	    nd6_slowtimo, curvnet);
2139 	NET_EPOCH_ENTER(et);
2140 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2141 		if (ifp->if_afdata[AF_INET6] == NULL)
2142 			continue;
2143 		nd6if = ND_IFINFO(ifp);
2144 		if (nd6if->basereachable && /* already initialized */
2145 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2146 			/*
2147 			 * Since reachable time rarely changes by router
2148 			 * advertisements, we SHOULD insure that a new random
2149 			 * value gets recomputed at least once every few hours.
2150 			 * (RFC 2461, 6.3.4)
2151 			 */
2152 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2153 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2154 		}
2155 	}
2156 	NET_EPOCH_EXIT(et);
2157 	CURVNET_RESTORE();
2158 }
2159 
2160 struct mbuf *
2161 nd6_grab_holdchain(struct llentry *ln)
2162 {
2163 	struct mbuf *chain;
2164 
2165 	LLE_WLOCK_ASSERT(ln);
2166 
2167 	chain = ln->la_hold;
2168 	ln->la_hold = NULL;
2169 	ln->la_numheld = 0;
2170 
2171 	if (ln->ln_state == ND6_LLINFO_STALE) {
2172 		/*
2173 		 * The first time we send a packet to a
2174 		 * neighbor whose entry is STALE, we have
2175 		 * to change the state to DELAY and a sets
2176 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2177 		 * seconds to ensure do neighbor unreachability
2178 		 * detection on expiration.
2179 		 * (RFC 2461 7.3.3)
2180 		 */
2181 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2182 	}
2183 
2184 	return (chain);
2185 }
2186 
2187 int
2188 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2189     struct sockaddr_in6 *dst, struct route *ro)
2190 {
2191 	int error;
2192 	int ip6len;
2193 	struct ip6_hdr *ip6;
2194 	struct m_tag *mtag;
2195 
2196 #ifdef MAC
2197 	mac_netinet6_nd6_send(ifp, m);
2198 #endif
2199 
2200 	/*
2201 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2202 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2203 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2204 	 * to be diverted to user space.  When re-injected into the kernel,
2205 	 * send_output() will directly dispatch them to the outgoing interface.
2206 	 */
2207 	if (send_sendso_input_hook != NULL) {
2208 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2209 		if (mtag != NULL) {
2210 			ip6 = mtod(m, struct ip6_hdr *);
2211 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2212 			/* Use the SEND socket */
2213 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2214 			    ip6len);
2215 			/* -1 == no app on SEND socket */
2216 			if (error == 0 || error != -1)
2217 			    return (error);
2218 		}
2219 	}
2220 
2221 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2222 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2223 	    mtod(m, struct ip6_hdr *));
2224 
2225 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2226 		origifp = ifp;
2227 
2228 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2229 	return (error);
2230 }
2231 
2232 /*
2233  * Lookup link headerfor @sa_dst address. Stores found
2234  * data in @desten buffer. Copy of lle ln_flags can be also
2235  * saved in @pflags if @pflags is non-NULL.
2236  *
2237  * If destination LLE does not exists or lle state modification
2238  * is required, call "slow" version.
2239  *
2240  * Return values:
2241  * - 0 on success (address copied to buffer).
2242  * - EWOULDBLOCK (no local error, but address is still unresolved)
2243  * - other errors (alloc failure, etc)
2244  */
2245 int
2246 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2247     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2248     struct llentry **plle)
2249 {
2250 	struct llentry *ln = NULL;
2251 	const struct sockaddr_in6 *dst6;
2252 
2253 	NET_EPOCH_ASSERT();
2254 
2255 	if (pflags != NULL)
2256 		*pflags = 0;
2257 
2258 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2259 
2260 	/* discard the packet if IPv6 operation is disabled on the interface */
2261 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2262 		m_freem(m);
2263 		return (ENETDOWN); /* better error? */
2264 	}
2265 
2266 	if (m != NULL && m->m_flags & M_MCAST) {
2267 		switch (ifp->if_type) {
2268 		case IFT_ETHER:
2269 		case IFT_L2VLAN:
2270 		case IFT_BRIDGE:
2271 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2272 						 desten);
2273 			return (0);
2274 		default:
2275 			m_freem(m);
2276 			return (EAFNOSUPPORT);
2277 		}
2278 	}
2279 
2280 	int family = gw_flags >> 16;
2281 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2282 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2283 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2284 		/* Entry found, let's copy lle info */
2285 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2286 		if (pflags != NULL)
2287 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2288 		llentry_provide_feedback(ln);
2289 		if (plle) {
2290 			LLE_ADDREF(ln);
2291 			*plle = ln;
2292 			LLE_WUNLOCK(ln);
2293 		}
2294 		return (0);
2295 	} else if (plle && ln)
2296 		LLE_WUNLOCK(ln);
2297 
2298 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2299 }
2300 
2301 /*
2302  * Finds or creates a new llentry for @addr and @family.
2303  * Returns wlocked llentry or NULL.
2304  *
2305  *
2306  * Child LLEs.
2307  *
2308  * Do not have their own state machine (gets marked as static)
2309  *  settimer bails out for child LLEs just in case.
2310  *
2311  * Locking order: parent lle gets locked first, chen goes the child.
2312  */
2313 static __noinline struct llentry *
2314 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2315 {
2316 	struct llentry *child_lle = NULL;
2317 	struct llentry *lle, *lle_tmp;
2318 
2319 	lle = nd6_alloc(addr, 0, ifp);
2320 	if (lle != NULL && family != AF_INET6) {
2321 		child_lle = nd6_alloc(addr, 0, ifp);
2322 		if (child_lle == NULL) {
2323 			lltable_free_entry(LLTABLE6(ifp), lle);
2324 			return (NULL);
2325 		}
2326 		child_lle->r_family = family;
2327 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2328 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2329 	}
2330 
2331 	if (lle == NULL) {
2332 		char ip6buf[INET6_ADDRSTRLEN];
2333 		log(LOG_DEBUG,
2334 		    "nd6_get_llentry: can't allocate llinfo for %s "
2335 		    "(ln=%p)\n",
2336 		    ip6_sprintf(ip6buf, addr), lle);
2337 		return (NULL);
2338 	}
2339 
2340 	IF_AFDATA_WLOCK(ifp);
2341 	LLE_WLOCK(lle);
2342 	/* Prefer any existing entry over newly-created one */
2343 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2344 	if (lle_tmp == NULL)
2345 		lltable_link_entry(LLTABLE6(ifp), lle);
2346 	else {
2347 		lltable_free_entry(LLTABLE6(ifp), lle);
2348 		lle = lle_tmp;
2349 	}
2350 	if (child_lle != NULL) {
2351 		/* Check if child lle for the same family exists */
2352 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2353 		LLE_WLOCK(child_lle);
2354 		if (lle_tmp == NULL) {
2355 			/* Attach */
2356 			lltable_link_child_entry(lle, child_lle);
2357 		} else {
2358 			/* child lle already exists, free newly-created one */
2359 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2360 			child_lle = lle_tmp;
2361 		}
2362 		LLE_WUNLOCK(lle);
2363 		lle = child_lle;
2364 	}
2365 	IF_AFDATA_WUNLOCK(ifp);
2366 	return (lle);
2367 }
2368 
2369 /*
2370  * Do L2 address resolution for @sa_dst address. Stores found
2371  * address in @desten buffer. Copy of lle ln_flags can be also
2372  * saved in @pflags if @pflags is non-NULL.
2373  *
2374  * Heavy version.
2375  * Function assume that destination LLE does not exist,
2376  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2377  *
2378  * Set noinline to be dtrace-friendly
2379  */
2380 static __noinline int
2381 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2382     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2383     struct llentry **plle)
2384 {
2385 	struct llentry *lle = NULL;
2386 	struct in6_addr *psrc, src;
2387 	int send_ns, ll_len;
2388 	char *lladdr;
2389 
2390 	NET_EPOCH_ASSERT();
2391 
2392 	/*
2393 	 * Address resolution or Neighbor Unreachability Detection
2394 	 * for the next hop.
2395 	 * At this point, the destination of the packet must be a unicast
2396 	 * or an anycast address(i.e. not a multicast).
2397 	 */
2398 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2399 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2400 		/*
2401 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2402 		 * the condition below is not very efficient.  But we believe
2403 		 * it is tolerable, because this should be a rare case.
2404 		 */
2405 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2406 	}
2407 
2408 	if (lle == NULL) {
2409 		m_freem(m);
2410 		return (ENOBUFS);
2411 	}
2412 
2413 	LLE_WLOCK_ASSERT(lle);
2414 
2415 	/*
2416 	 * The first time we send a packet to a neighbor whose entry is
2417 	 * STALE, we have to change the state to DELAY and a sets a timer to
2418 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2419 	 * neighbor unreachability detection on expiration.
2420 	 * (RFC 2461 7.3.3)
2421 	 */
2422 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2423 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2424 
2425 	/*
2426 	 * If the neighbor cache entry has a state other than INCOMPLETE
2427 	 * (i.e. its link-layer address is already resolved), just
2428 	 * send the packet.
2429 	 */
2430 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2431 		if (flags & LLE_ADDRONLY) {
2432 			lladdr = lle->ll_addr;
2433 			ll_len = ifp->if_addrlen;
2434 		} else {
2435 			lladdr = lle->r_linkdata;
2436 			ll_len = lle->r_hdrlen;
2437 		}
2438 		bcopy(lladdr, desten, ll_len);
2439 		if (pflags != NULL)
2440 			*pflags = lle->la_flags;
2441 		if (plle) {
2442 			LLE_ADDREF(lle);
2443 			*plle = lle;
2444 		}
2445 		LLE_WUNLOCK(lle);
2446 		return (0);
2447 	}
2448 
2449 	/*
2450 	 * There is a neighbor cache entry, but no ethernet address
2451 	 * response yet.  Append this latest packet to the end of the
2452 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2453 	 * the oldest packet in the queue will be removed.
2454 	 */
2455 	if (m != NULL) {
2456 		size_t dropped;
2457 
2458 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2459 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2460 	}
2461 
2462 	/*
2463 	 * If there has been no NS for the neighbor after entering the
2464 	 * INCOMPLETE state, send the first solicitation.
2465 	 * Note that for newly-created lle la_asked will be 0,
2466 	 * so we will transition from ND6_LLINFO_NOSTATE to
2467 	 * ND6_LLINFO_INCOMPLETE state here.
2468 	 */
2469 	psrc = NULL;
2470 	send_ns = 0;
2471 
2472 	/* If we have child lle, switch to the parent to send NS */
2473 	if (lle->la_flags & LLE_CHILD) {
2474 		struct llentry *lle_parent = lle->lle_parent;
2475 		LLE_WUNLOCK(lle);
2476 		lle = lle_parent;
2477 		LLE_WLOCK(lle);
2478 	}
2479 	if (lle->la_asked == 0) {
2480 		lle->la_asked++;
2481 		send_ns = 1;
2482 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2483 
2484 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2485 	}
2486 	LLE_WUNLOCK(lle);
2487 	if (send_ns != 0)
2488 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2489 
2490 	return (EWOULDBLOCK);
2491 }
2492 
2493 /*
2494  * Do L2 address resolution for @sa_dst address. Stores found
2495  * address in @desten buffer. Copy of lle ln_flags can be also
2496  * saved in @pflags if @pflags is non-NULL.
2497  *
2498  * Return values:
2499  * - 0 on success (address copied to buffer).
2500  * - EWOULDBLOCK (no local error, but address is still unresolved)
2501  * - other errors (alloc failure, etc)
2502  */
2503 int
2504 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2505     char *desten, uint32_t *pflags)
2506 {
2507 	int error;
2508 
2509 	flags |= LLE_ADDRONLY;
2510 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2511 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2512 	return (error);
2513 }
2514 
2515 int
2516 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2517 {
2518 	struct mbuf *m, *m_head;
2519 	struct sockaddr_in6 dst6;
2520 	int error = 0;
2521 
2522 	NET_EPOCH_ASSERT();
2523 
2524 	struct route_in6 ro = {
2525 		.ro_prepend = lle->r_linkdata,
2526 		.ro_plen = lle->r_hdrlen,
2527 	};
2528 
2529 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2530 	m_head = chain;
2531 
2532 	while (m_head) {
2533 		m = m_head;
2534 		m_head = m_head->m_nextpkt;
2535 		m->m_nextpkt = NULL;
2536 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2537 	}
2538 
2539 	/*
2540 	 * XXX
2541 	 * note that intermediate errors are blindly ignored
2542 	 */
2543 	return (error);
2544 }
2545 
2546 __noinline void
2547 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2548 {
2549 	struct llentry *child_lle;
2550 	struct mbuf *chain;
2551 
2552 	NET_EPOCH_ASSERT();
2553 
2554 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2555 		LLE_WLOCK(child_lle);
2556 		chain = nd6_grab_holdchain(child_lle);
2557 		LLE_WUNLOCK(child_lle);
2558 		nd6_flush_holdchain(ifp, child_lle, chain);
2559 	}
2560 }
2561 
2562 static int
2563 nd6_need_cache(struct ifnet *ifp)
2564 {
2565 	/*
2566 	 * XXX: we currently do not make neighbor cache on any interface
2567 	 * other than Ethernet and GIF.
2568 	 *
2569 	 * RFC2893 says:
2570 	 * - unidirectional tunnels needs no ND
2571 	 */
2572 	switch (ifp->if_type) {
2573 	case IFT_ETHER:
2574 	case IFT_IEEE1394:
2575 	case IFT_L2VLAN:
2576 	case IFT_INFINIBAND:
2577 	case IFT_BRIDGE:
2578 	case IFT_PROPVIRTUAL:
2579 		return (1);
2580 	default:
2581 		return (0);
2582 	}
2583 }
2584 
2585 /*
2586  * Add pernament ND6 link-layer record for given
2587  * interface address.
2588  *
2589  * Very similar to IPv4 arp_ifinit(), but:
2590  * 1) IPv6 DAD is performed in different place
2591  * 2) It is called by IPv6 protocol stack in contrast to
2592  * arp_ifinit() which is typically called in SIOCSIFADDR
2593  * driver ioctl handler.
2594  *
2595  */
2596 int
2597 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2598 {
2599 	struct ifnet *ifp;
2600 	struct llentry *ln, *ln_tmp;
2601 	struct sockaddr *dst;
2602 
2603 	ifp = ia->ia_ifa.ifa_ifp;
2604 	if (nd6_need_cache(ifp) == 0)
2605 		return (0);
2606 
2607 	dst = (struct sockaddr *)&ia->ia_addr;
2608 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2609 	if (ln == NULL)
2610 		return (ENOBUFS);
2611 
2612 	IF_AFDATA_WLOCK(ifp);
2613 	LLE_WLOCK(ln);
2614 	/* Unlink any entry if exists */
2615 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2616 	if (ln_tmp != NULL)
2617 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2618 	lltable_link_entry(LLTABLE6(ifp), ln);
2619 	IF_AFDATA_WUNLOCK(ifp);
2620 
2621 	if (ln_tmp != NULL)
2622 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2623 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2624 
2625 	LLE_WUNLOCK(ln);
2626 	if (ln_tmp != NULL)
2627 		llentry_free(ln_tmp);
2628 
2629 	return (0);
2630 }
2631 
2632 /*
2633  * Removes either all lle entries for given @ia, or lle
2634  * corresponding to @ia address.
2635  */
2636 void
2637 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2638 {
2639 	struct sockaddr_in6 mask, addr;
2640 	struct sockaddr *saddr, *smask;
2641 	struct ifnet *ifp;
2642 
2643 	ifp = ia->ia_ifa.ifa_ifp;
2644 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2645 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2646 	saddr = (struct sockaddr *)&addr;
2647 	smask = (struct sockaddr *)&mask;
2648 
2649 	if (all != 0)
2650 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2651 	else
2652 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2653 }
2654 
2655 static int
2656 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2657 {
2658 	struct in6_prefix p;
2659 	struct sockaddr_in6 s6;
2660 	struct nd_prefix *pr;
2661 	struct nd_pfxrouter *pfr;
2662 	time_t maxexpire;
2663 	int error;
2664 	char ip6buf[INET6_ADDRSTRLEN];
2665 
2666 	if (req->newptr)
2667 		return (EPERM);
2668 
2669 	error = sysctl_wire_old_buffer(req, 0);
2670 	if (error != 0)
2671 		return (error);
2672 
2673 	bzero(&p, sizeof(p));
2674 	p.origin = PR_ORIG_RA;
2675 	bzero(&s6, sizeof(s6));
2676 	s6.sin6_family = AF_INET6;
2677 	s6.sin6_len = sizeof(s6);
2678 
2679 	ND6_RLOCK();
2680 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2681 		if (!pr->ndpr_raf_ra_derived)
2682 			continue;
2683 		p.prefix = pr->ndpr_prefix;
2684 		if (sa6_recoverscope(&p.prefix)) {
2685 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2686 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2687 			/* XXX: press on... */
2688 		}
2689 		p.raflags = pr->ndpr_raf;
2690 		p.prefixlen = pr->ndpr_plen;
2691 		p.vltime = pr->ndpr_vltime;
2692 		p.pltime = pr->ndpr_pltime;
2693 		p.if_index = pr->ndpr_ifp->if_index;
2694 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2695 			p.expire = 0;
2696 		else {
2697 			/* XXX: we assume time_t is signed. */
2698 			maxexpire = (-1) &
2699 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2700 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2701 				p.expire = pr->ndpr_lastupdate +
2702 				    pr->ndpr_vltime +
2703 				    (time_second - time_uptime);
2704 			else
2705 				p.expire = maxexpire;
2706 		}
2707 		p.refcnt = pr->ndpr_addrcnt;
2708 		p.flags = pr->ndpr_stateflags;
2709 		p.advrtrs = 0;
2710 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2711 			p.advrtrs++;
2712 		error = SYSCTL_OUT(req, &p, sizeof(p));
2713 		if (error != 0)
2714 			break;
2715 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2716 			s6.sin6_addr = pfr->router->rtaddr;
2717 			if (sa6_recoverscope(&s6))
2718 				log(LOG_ERR,
2719 				    "scope error in prefix list (%s)\n",
2720 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2721 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2722 			if (error != 0)
2723 				goto out;
2724 		}
2725 	}
2726 out:
2727 	ND6_RUNLOCK();
2728 	return (error);
2729 }
2730 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2731 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2732 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2733 	"NDP prefix list");
2734 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2735 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2736 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2737 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2738