xref: /freebsd/sys/netinet6/nd6.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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 llentry *ln;
736 	struct in6_addr *dst, *pdst, *psrc, src;
737 	struct ifnet *ifp;
738 	struct nd_ifinfo *ndi;
739 	int do_switch, send_ns;
740 	long delay;
741 
742 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
743 	ln = (struct llentry *)arg;
744 	ifp = lltable_get_ifp(ln->lle_tbl);
745 	CURVNET_SET(ifp->if_vnet);
746 
747 	ND6_RLOCK();
748 	LLE_WLOCK(ln);
749 	if (callout_pending(&ln->lle_timer)) {
750 		/*
751 		 * Here we are a bit odd here in the treatment of
752 		 * active/pending. If the pending bit is set, it got
753 		 * rescheduled before I ran. The active
754 		 * bit we ignore, since if it was stopped
755 		 * in ll_tablefree() and was currently running
756 		 * it would have return 0 so the code would
757 		 * not have deleted it since the callout could
758 		 * not be stopped so we want to go through
759 		 * with the delete here now. If the callout
760 		 * was restarted, the pending bit will be back on and
761 		 * we just want to bail since the callout_reset would
762 		 * return 1 and our reference would have been removed
763 		 * by nd6_llinfo_settimer_locked above since canceled
764 		 * would have been 1.
765 		 */
766 		LLE_WUNLOCK(ln);
767 		ND6_RUNLOCK();
768 		CURVNET_RESTORE();
769 		return;
770 	}
771 	ndi = ND_IFINFO(ifp);
772 	send_ns = 0;
773 	dst = &ln->r_l3addr.addr6;
774 	pdst = dst;
775 
776 	if (ln->ln_ntick > 0) {
777 		if (ln->ln_ntick > INT_MAX) {
778 			ln->ln_ntick -= INT_MAX;
779 			nd6_llinfo_settimer_locked(ln, INT_MAX);
780 		} else {
781 			ln->ln_ntick = 0;
782 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
783 		}
784 		goto done;
785 	}
786 
787 	if (ln->la_flags & LLE_STATIC) {
788 		goto done;
789 	}
790 
791 	if (ln->la_flags & LLE_DELETED) {
792 		nd6_free(&ln, 0);
793 		goto done;
794 	}
795 
796 	switch (ln->ln_state) {
797 	case ND6_LLINFO_INCOMPLETE:
798 		if (ln->la_asked < V_nd6_mmaxtries) {
799 			ln->la_asked++;
800 			send_ns = 1;
801 			/* Send NS to multicast address */
802 			pdst = NULL;
803 		} else {
804 			struct mbuf *m = ln->la_hold;
805 			if (m) {
806 				struct mbuf *m0;
807 
808 				/*
809 				 * assuming every packet in la_hold has the
810 				 * same IP header.  Send error after unlock.
811 				 */
812 				m0 = m->m_nextpkt;
813 				m->m_nextpkt = NULL;
814 				ln->la_hold = m0;
815 				clear_llinfo_pqueue(ln);
816 			}
817 			nd6_free(&ln, 0);
818 			if (m != NULL)
819 				icmp6_error2(m, ICMP6_DST_UNREACH,
820 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
821 		}
822 		break;
823 	case ND6_LLINFO_REACHABLE:
824 		if (!ND6_LLINFO_PERMANENT(ln))
825 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
826 		break;
827 
828 	case ND6_LLINFO_STALE:
829 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
830 
831 			/*
832 			 * No packet has used this entry and GC timeout
833 			 * has not been passed. Reshedule timer and
834 			 * return.
835 			 */
836 			nd6_llinfo_settimer_locked(ln, delay);
837 			break;
838 		}
839 
840 		if (do_switch == 0) {
841 
842 			/*
843 			 * GC timer has ended and entry hasn't been used.
844 			 * Run Garbage collector (RFC 4861, 5.3)
845 			 */
846 			if (!ND6_LLINFO_PERMANENT(ln))
847 				nd6_free(&ln, 1);
848 			break;
849 		}
850 
851 		/* Entry has been used AND delay timer has ended. */
852 
853 		/* FALLTHROUGH */
854 
855 	case ND6_LLINFO_DELAY:
856 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
857 			/* We need NUD */
858 			ln->la_asked = 1;
859 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
860 			send_ns = 1;
861 		} else
862 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
863 		break;
864 	case ND6_LLINFO_PROBE:
865 		if (ln->la_asked < V_nd6_umaxtries) {
866 			ln->la_asked++;
867 			send_ns = 1;
868 		} else {
869 			nd6_free(&ln, 0);
870 		}
871 		break;
872 	default:
873 		panic("%s: paths in a dark night can be confusing: %d",
874 		    __func__, ln->ln_state);
875 	}
876 done:
877 	if (ln != NULL)
878 		ND6_RUNLOCK();
879 	if (send_ns != 0) {
880 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
881 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
882 		LLE_FREE_LOCKED(ln);
883 		ln = NULL;
884 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
885 	}
886 
887 	if (ln != NULL)
888 		LLE_FREE_LOCKED(ln);
889 	CURVNET_RESTORE();
890 }
891 
892 
893 /*
894  * ND6 timer routine to expire default route list and prefix list
895  */
896 void
897 nd6_timer(void *arg)
898 {
899 	CURVNET_SET((struct vnet *) arg);
900 	struct nd_drhead drq;
901 	struct nd_prhead prl;
902 	struct nd_defrouter *dr, *ndr;
903 	struct nd_prefix *pr, *npr;
904 	struct ifnet *ifp;
905 	struct in6_ifaddr *ia6, *nia6;
906 	uint64_t genid;
907 
908 	TAILQ_INIT(&drq);
909 	LIST_INIT(&prl);
910 
911 	ND6_WLOCK();
912 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
913 		if (dr->expire && dr->expire < time_uptime)
914 			defrouter_unlink(dr, &drq);
915 	ND6_WUNLOCK();
916 
917 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
918 		TAILQ_REMOVE(&drq, dr, dr_entry);
919 		defrouter_del(dr);
920 	}
921 
922 	/*
923 	 * expire interface addresses.
924 	 * in the past the loop was inside prefix expiry processing.
925 	 * However, from a stricter speci-confrmance standpoint, we should
926 	 * rather separate address lifetimes and prefix lifetimes.
927 	 *
928 	 * XXXRW: in6_ifaddrhead locking.
929 	 */
930   addrloop:
931 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
932 		/* check address lifetime */
933 		if (IFA6_IS_INVALID(ia6)) {
934 			int regen = 0;
935 
936 			/*
937 			 * If the expiring address is temporary, try
938 			 * regenerating a new one.  This would be useful when
939 			 * we suspended a laptop PC, then turned it on after a
940 			 * period that could invalidate all temporary
941 			 * addresses.  Although we may have to restart the
942 			 * loop (see below), it must be after purging the
943 			 * address.  Otherwise, we'd see an infinite loop of
944 			 * regeneration.
945 			 */
946 			if (V_ip6_use_tempaddr &&
947 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
948 				if (regen_tmpaddr(ia6) == 0)
949 					regen = 1;
950 			}
951 
952 			in6_purgeaddr(&ia6->ia_ifa);
953 
954 			if (regen)
955 				goto addrloop; /* XXX: see below */
956 		} else if (IFA6_IS_DEPRECATED(ia6)) {
957 			int oldflags = ia6->ia6_flags;
958 
959 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
960 
961 			/*
962 			 * If a temporary address has just become deprecated,
963 			 * regenerate a new one if possible.
964 			 */
965 			if (V_ip6_use_tempaddr &&
966 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
967 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
968 
969 				if (regen_tmpaddr(ia6) == 0) {
970 					/*
971 					 * A new temporary address is
972 					 * generated.
973 					 * XXX: this means the address chain
974 					 * has changed while we are still in
975 					 * the loop.  Although the change
976 					 * would not cause disaster (because
977 					 * it's not a deletion, but an
978 					 * addition,) we'd rather restart the
979 					 * loop just for safety.  Or does this
980 					 * significantly reduce performance??
981 					 */
982 					goto addrloop;
983 				}
984 			}
985 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
986 			/*
987 			 * Schedule DAD for a tentative address.  This happens
988 			 * if the interface was down or not running
989 			 * when the address was configured.
990 			 */
991 			int delay;
992 
993 			delay = arc4random() %
994 			    (MAX_RTR_SOLICITATION_DELAY * hz);
995 			nd6_dad_start((struct ifaddr *)ia6, delay);
996 		} else {
997 			/*
998 			 * Check status of the interface.  If it is down,
999 			 * mark the address as tentative for future DAD.
1000 			 */
1001 			ifp = ia6->ia_ifp;
1002 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1003 			    ((ifp->if_flags & IFF_UP) == 0 ||
1004 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1005 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1006 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1007 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1008 			}
1009 
1010 			/*
1011 			 * A new RA might have made a deprecated address
1012 			 * preferred.
1013 			 */
1014 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1015 		}
1016 	}
1017 
1018 	ND6_WLOCK();
1019 restart:
1020 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1021 		/*
1022 		 * Expire prefixes. Since the pltime is only used for
1023 		 * autoconfigured addresses, pltime processing for prefixes is
1024 		 * not necessary.
1025 		 *
1026 		 * Only unlink after all derived addresses have expired. This
1027 		 * may not occur until two hours after the prefix has expired
1028 		 * per RFC 4862. If the prefix expires before its derived
1029 		 * addresses, mark it off-link. This will be done automatically
1030 		 * after unlinking if no address references remain.
1031 		 */
1032 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1033 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1034 			continue;
1035 
1036 		if (pr->ndpr_addrcnt == 0) {
1037 			nd6_prefix_unlink(pr, &prl);
1038 			continue;
1039 		}
1040 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1041 			genid = V_nd6_list_genid;
1042 			nd6_prefix_ref(pr);
1043 			ND6_WUNLOCK();
1044 			ND6_ONLINK_LOCK();
1045 			(void)nd6_prefix_offlink(pr);
1046 			ND6_ONLINK_UNLOCK();
1047 			ND6_WLOCK();
1048 			nd6_prefix_rele(pr);
1049 			if (genid != V_nd6_list_genid)
1050 				goto restart;
1051 		}
1052 	}
1053 	ND6_WUNLOCK();
1054 
1055 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1056 		LIST_REMOVE(pr, ndpr_entry);
1057 		nd6_prefix_del(pr);
1058 	}
1059 
1060 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1061 	    nd6_timer, curvnet);
1062 
1063 	CURVNET_RESTORE();
1064 }
1065 
1066 /*
1067  * ia6 - deprecated/invalidated temporary address
1068  */
1069 static int
1070 regen_tmpaddr(struct in6_ifaddr *ia6)
1071 {
1072 	struct epoch_tracker et;
1073 	struct ifaddr *ifa;
1074 	struct ifnet *ifp;
1075 	struct in6_ifaddr *public_ifa6 = NULL;
1076 
1077 	ifp = ia6->ia_ifa.ifa_ifp;
1078 	NET_EPOCH_ENTER(et);
1079 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1080 		struct in6_ifaddr *it6;
1081 
1082 		if (ifa->ifa_addr->sa_family != AF_INET6)
1083 			continue;
1084 
1085 		it6 = (struct in6_ifaddr *)ifa;
1086 
1087 		/* ignore no autoconf addresses. */
1088 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1089 			continue;
1090 
1091 		/* ignore autoconf addresses with different prefixes. */
1092 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1093 			continue;
1094 
1095 		/*
1096 		 * Now we are looking at an autoconf address with the same
1097 		 * prefix as ours.  If the address is temporary and is still
1098 		 * preferred, do not create another one.  It would be rare, but
1099 		 * could happen, for example, when we resume a laptop PC after
1100 		 * a long period.
1101 		 */
1102 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1103 		    !IFA6_IS_DEPRECATED(it6)) {
1104 			public_ifa6 = NULL;
1105 			break;
1106 		}
1107 
1108 		/*
1109 		 * This is a public autoconf address that has the same prefix
1110 		 * as ours.  If it is preferred, keep it.  We can't break the
1111 		 * loop here, because there may be a still-preferred temporary
1112 		 * address with the prefix.
1113 		 */
1114 		if (!IFA6_IS_DEPRECATED(it6))
1115 			public_ifa6 = it6;
1116 	}
1117 	if (public_ifa6 != NULL)
1118 		ifa_ref(&public_ifa6->ia_ifa);
1119 	NET_EPOCH_EXIT(et);
1120 
1121 	if (public_ifa6 != NULL) {
1122 		int e;
1123 
1124 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1125 			ifa_free(&public_ifa6->ia_ifa);
1126 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1127 			    " tmp addr,errno=%d\n", e);
1128 			return (-1);
1129 		}
1130 		ifa_free(&public_ifa6->ia_ifa);
1131 		return (0);
1132 	}
1133 
1134 	return (-1);
1135 }
1136 
1137 /*
1138  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1139  * cache entries are freed in in6_domifdetach().
1140  */
1141 void
1142 nd6_purge(struct ifnet *ifp)
1143 {
1144 	struct nd_drhead drq;
1145 	struct nd_prhead prl;
1146 	struct nd_defrouter *dr, *ndr;
1147 	struct nd_prefix *pr, *npr;
1148 
1149 	TAILQ_INIT(&drq);
1150 	LIST_INIT(&prl);
1151 
1152 	/*
1153 	 * Nuke default router list entries toward ifp.
1154 	 * We defer removal of default router list entries that is installed
1155 	 * in the routing table, in order to keep additional side effects as
1156 	 * small as possible.
1157 	 */
1158 	ND6_WLOCK();
1159 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1160 		if (dr->installed)
1161 			continue;
1162 		if (dr->ifp == ifp)
1163 			defrouter_unlink(dr, &drq);
1164 	}
1165 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1166 		if (!dr->installed)
1167 			continue;
1168 		if (dr->ifp == ifp)
1169 			defrouter_unlink(dr, &drq);
1170 	}
1171 
1172 	/*
1173 	 * Remove prefixes on ifp. We should have already removed addresses on
1174 	 * this interface, so no addresses should be referencing these prefixes.
1175 	 */
1176 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1177 		if (pr->ndpr_ifp == ifp)
1178 			nd6_prefix_unlink(pr, &prl);
1179 	}
1180 	ND6_WUNLOCK();
1181 
1182 	/* Delete the unlinked router and prefix objects. */
1183 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1184 		TAILQ_REMOVE(&drq, dr, dr_entry);
1185 		defrouter_del(dr);
1186 	}
1187 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1188 		LIST_REMOVE(pr, ndpr_entry);
1189 		nd6_prefix_del(pr);
1190 	}
1191 
1192 	/* cancel default outgoing interface setting */
1193 	if (V_nd6_defifindex == ifp->if_index)
1194 		nd6_setdefaultiface(0);
1195 
1196 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1197 		/* Refresh default router list. */
1198 		defrouter_select_fib(ifp->if_fib);
1199 	}
1200 }
1201 
1202 /*
1203  * the caller acquires and releases the lock on the lltbls
1204  * Returns the llentry locked
1205  */
1206 struct llentry *
1207 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1208 {
1209 	struct sockaddr_in6 sin6;
1210 	struct llentry *ln;
1211 
1212 	bzero(&sin6, sizeof(sin6));
1213 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1214 	sin6.sin6_family = AF_INET6;
1215 	sin6.sin6_addr = *addr6;
1216 
1217 	IF_AFDATA_LOCK_ASSERT(ifp);
1218 
1219 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1220 
1221 	return (ln);
1222 }
1223 
1224 struct llentry *
1225 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1226 {
1227 	struct sockaddr_in6 sin6;
1228 	struct llentry *ln;
1229 
1230 	bzero(&sin6, sizeof(sin6));
1231 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1232 	sin6.sin6_family = AF_INET6;
1233 	sin6.sin6_addr = *addr6;
1234 
1235 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1236 	if (ln != NULL)
1237 		ln->ln_state = ND6_LLINFO_NOSTATE;
1238 
1239 	return (ln);
1240 }
1241 
1242 /*
1243  * Test whether a given IPv6 address is a neighbor or not, ignoring
1244  * the actual neighbor cache.  The neighbor cache is ignored in order
1245  * to not reenter the routing code from within itself.
1246  */
1247 static int
1248 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1249 {
1250 	struct nd_prefix *pr;
1251 	struct ifaddr *ifa;
1252 	struct rt_addrinfo info;
1253 	struct sockaddr_in6 rt_key;
1254 	const struct sockaddr *dst6;
1255 	uint64_t genid;
1256 	int error, fibnum;
1257 
1258 	/*
1259 	 * A link-local address is always a neighbor.
1260 	 * XXX: a link does not necessarily specify a single interface.
1261 	 */
1262 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1263 		struct sockaddr_in6 sin6_copy;
1264 		u_int32_t zone;
1265 
1266 		/*
1267 		 * We need sin6_copy since sa6_recoverscope() may modify the
1268 		 * content (XXX).
1269 		 */
1270 		sin6_copy = *addr;
1271 		if (sa6_recoverscope(&sin6_copy))
1272 			return (0); /* XXX: should be impossible */
1273 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1274 			return (0);
1275 		if (sin6_copy.sin6_scope_id == zone)
1276 			return (1);
1277 		else
1278 			return (0);
1279 	}
1280 
1281 	bzero(&rt_key, sizeof(rt_key));
1282 	bzero(&info, sizeof(info));
1283 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1284 
1285 	/*
1286 	 * If the address matches one of our addresses,
1287 	 * it should be a neighbor.
1288 	 * If the address matches one of our on-link prefixes, it should be a
1289 	 * neighbor.
1290 	 */
1291 	ND6_RLOCK();
1292 restart:
1293 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1294 		if (pr->ndpr_ifp != ifp)
1295 			continue;
1296 
1297 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1298 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1299 
1300 			/*
1301 			 * We only need to check all FIBs if add_addr_allfibs
1302 			 * is unset. If set, checking any FIB will suffice.
1303 			 */
1304 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1305 			for (; fibnum < rt_numfibs; fibnum++) {
1306 				genid = V_nd6_list_genid;
1307 				ND6_RUNLOCK();
1308 
1309 				/*
1310 				 * Restore length field before
1311 				 * retrying lookup
1312 				 */
1313 				rt_key.sin6_len = sizeof(rt_key);
1314 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1315 						        &info);
1316 
1317 				ND6_RLOCK();
1318 				if (genid != V_nd6_list_genid)
1319 					goto restart;
1320 				if (error == 0)
1321 					break;
1322 			}
1323 			if (error != 0)
1324 				continue;
1325 
1326 			/*
1327 			 * This is the case where multiple interfaces
1328 			 * have the same prefix, but only one is installed
1329 			 * into the routing table and that prefix entry
1330 			 * is not the one being examined here. In the case
1331 			 * where RADIX_MPATH is enabled, multiple route
1332 			 * entries (of the same rt_key value) will be
1333 			 * installed because the interface addresses all
1334 			 * differ.
1335 			 */
1336 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1337 			    &rt_key.sin6_addr))
1338 				continue;
1339 		}
1340 
1341 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1342 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1343 			ND6_RUNLOCK();
1344 			return (1);
1345 		}
1346 	}
1347 	ND6_RUNLOCK();
1348 
1349 	/*
1350 	 * If the address is assigned on the node of the other side of
1351 	 * a p2p interface, the address should be a neighbor.
1352 	 */
1353 	if (ifp->if_flags & IFF_POINTOPOINT) {
1354 		struct epoch_tracker et;
1355 
1356 		NET_EPOCH_ENTER(et);
1357 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1358 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1359 				continue;
1360 			if (ifa->ifa_dstaddr != NULL &&
1361 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1362 				NET_EPOCH_EXIT(et);
1363 				return 1;
1364 			}
1365 		}
1366 		NET_EPOCH_EXIT(et);
1367 	}
1368 
1369 	/*
1370 	 * If the default router list is empty, all addresses are regarded
1371 	 * as on-link, and thus, as a neighbor.
1372 	 */
1373 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1374 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1375 	    V_nd6_defifindex == ifp->if_index) {
1376 		return (1);
1377 	}
1378 
1379 	return (0);
1380 }
1381 
1382 
1383 /*
1384  * Detect if a given IPv6 address identifies a neighbor on a given link.
1385  * XXX: should take care of the destination of a p2p link?
1386  */
1387 int
1388 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1389 {
1390 	struct epoch_tracker et;
1391 	struct llentry *lle;
1392 	int rc = 0;
1393 
1394 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1395 	if (nd6_is_new_addr_neighbor(addr, ifp))
1396 		return (1);
1397 
1398 	/*
1399 	 * Even if the address matches none of our addresses, it might be
1400 	 * in the neighbor cache.
1401 	 */
1402 	NET_EPOCH_ENTER(et);
1403 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1404 		LLE_RUNLOCK(lle);
1405 		rc = 1;
1406 	}
1407 	NET_EPOCH_EXIT(et);
1408 	return (rc);
1409 }
1410 
1411 /*
1412  * Free an nd6 llinfo entry.
1413  * Since the function would cause significant changes in the kernel, DO NOT
1414  * make it global, unless you have a strong reason for the change, and are sure
1415  * that the change is safe.
1416  *
1417  * Set noinline to be dtrace-friendly
1418  */
1419 static __noinline void
1420 nd6_free(struct llentry **lnp, int gc)
1421 {
1422 	struct ifnet *ifp;
1423 	struct llentry *ln;
1424 	struct nd_defrouter *dr;
1425 
1426 	ln = *lnp;
1427 	*lnp = NULL;
1428 
1429 	LLE_WLOCK_ASSERT(ln);
1430 	ND6_RLOCK_ASSERT();
1431 
1432 	ifp = lltable_get_ifp(ln->lle_tbl);
1433 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1434 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1435 	else
1436 		dr = NULL;
1437 	ND6_RUNLOCK();
1438 
1439 	if ((ln->la_flags & LLE_DELETED) == 0)
1440 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1441 
1442 	/*
1443 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1444 	 * even though it is not harmful, it was not really necessary.
1445 	 */
1446 
1447 	/* cancel timer */
1448 	nd6_llinfo_settimer_locked(ln, -1);
1449 
1450 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1451 		if (dr != NULL && dr->expire &&
1452 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1453 			/*
1454 			 * If the reason for the deletion is just garbage
1455 			 * collection, and the neighbor is an active default
1456 			 * router, do not delete it.  Instead, reset the GC
1457 			 * timer using the router's lifetime.
1458 			 * Simply deleting the entry would affect default
1459 			 * router selection, which is not necessarily a good
1460 			 * thing, especially when we're using router preference
1461 			 * values.
1462 			 * XXX: the check for ln_state would be redundant,
1463 			 *      but we intentionally keep it just in case.
1464 			 */
1465 			if (dr->expire > time_uptime)
1466 				nd6_llinfo_settimer_locked(ln,
1467 				    (dr->expire - time_uptime) * hz);
1468 			else
1469 				nd6_llinfo_settimer_locked(ln,
1470 				    (long)V_nd6_gctimer * hz);
1471 
1472 			LLE_REMREF(ln);
1473 			LLE_WUNLOCK(ln);
1474 			defrouter_rele(dr);
1475 			return;
1476 		}
1477 
1478 		if (dr) {
1479 			/*
1480 			 * Unreachablity of a router might affect the default
1481 			 * router selection and on-link detection of advertised
1482 			 * prefixes.
1483 			 */
1484 
1485 			/*
1486 			 * Temporarily fake the state to choose a new default
1487 			 * router and to perform on-link determination of
1488 			 * prefixes correctly.
1489 			 * Below the state will be set correctly,
1490 			 * or the entry itself will be deleted.
1491 			 */
1492 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1493 		}
1494 
1495 		if (ln->ln_router || dr) {
1496 
1497 			/*
1498 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1499 			 * rnh and for the calls to pfxlist_onlink_check() and
1500 			 * defrouter_select_fib() in the block further down for calls
1501 			 * into nd6_lookup().  We still hold a ref.
1502 			 */
1503 			LLE_WUNLOCK(ln);
1504 
1505 			/*
1506 			 * rt6_flush must be called whether or not the neighbor
1507 			 * is in the Default Router List.
1508 			 * See a corresponding comment in nd6_na_input().
1509 			 */
1510 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1511 		}
1512 
1513 		if (dr) {
1514 			/*
1515 			 * Since defrouter_select_fib() does not affect the
1516 			 * on-link determination and MIP6 needs the check
1517 			 * before the default router selection, we perform
1518 			 * the check now.
1519 			 */
1520 			pfxlist_onlink_check();
1521 
1522 			/*
1523 			 * Refresh default router list.
1524 			 */
1525 			defrouter_select_fib(dr->ifp->if_fib);
1526 		}
1527 
1528 		/*
1529 		 * If this entry was added by an on-link redirect, remove the
1530 		 * corresponding host route.
1531 		 */
1532 		if (ln->la_flags & LLE_REDIRECT)
1533 			nd6_free_redirect(ln);
1534 
1535 		if (ln->ln_router || dr)
1536 			LLE_WLOCK(ln);
1537 	}
1538 
1539 	/*
1540 	 * Save to unlock. We still hold an extra reference and will not
1541 	 * free(9) in llentry_free() if someone else holds one as well.
1542 	 */
1543 	LLE_WUNLOCK(ln);
1544 	IF_AFDATA_LOCK(ifp);
1545 	LLE_WLOCK(ln);
1546 	/* Guard against race with other llentry_free(). */
1547 	if (ln->la_flags & LLE_LINKED) {
1548 		/* Remove callout reference */
1549 		LLE_REMREF(ln);
1550 		lltable_unlink_entry(ln->lle_tbl, ln);
1551 	}
1552 	IF_AFDATA_UNLOCK(ifp);
1553 
1554 	llentry_free(ln);
1555 	if (dr != NULL)
1556 		defrouter_rele(dr);
1557 }
1558 
1559 static int
1560 nd6_isdynrte(const struct rtentry *rt, void *xap)
1561 {
1562 
1563 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1564 		return (1);
1565 
1566 	return (0);
1567 }
1568 /*
1569  * Remove the rtentry for the given llentry,
1570  * both of which were installed by a redirect.
1571  */
1572 static void
1573 nd6_free_redirect(const struct llentry *ln)
1574 {
1575 	int fibnum;
1576 	struct sockaddr_in6 sin6;
1577 	struct rt_addrinfo info;
1578 
1579 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1580 	memset(&info, 0, sizeof(info));
1581 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1582 	info.rti_filter = nd6_isdynrte;
1583 
1584 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1585 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1586 }
1587 
1588 /*
1589  * Rejuvenate this function for routing operations related
1590  * processing.
1591  */
1592 void
1593 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1594 {
1595 	struct sockaddr_in6 *gateway;
1596 	struct nd_defrouter *dr;
1597 	struct ifnet *ifp;
1598 
1599 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1600 	ifp = rt->rt_ifp;
1601 
1602 	switch (req) {
1603 	case RTM_ADD:
1604 		break;
1605 
1606 	case RTM_DELETE:
1607 		if (!ifp)
1608 			return;
1609 		/*
1610 		 * Only indirect routes are interesting.
1611 		 */
1612 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1613 			return;
1614 		/*
1615 		 * check for default route
1616 		 */
1617 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1618 		    &SIN6(rt_key(rt))->sin6_addr)) {
1619 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1620 			if (dr != NULL) {
1621 				dr->installed = 0;
1622 				defrouter_rele(dr);
1623 			}
1624 		}
1625 		break;
1626 	}
1627 }
1628 
1629 
1630 int
1631 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1632 {
1633 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1634 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1635 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1636 	struct epoch_tracker et;
1637 	int error = 0;
1638 
1639 	if (ifp->if_afdata[AF_INET6] == NULL)
1640 		return (EPFNOSUPPORT);
1641 	switch (cmd) {
1642 	case OSIOCGIFINFO_IN6:
1643 #define ND	ndi->ndi
1644 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1645 		bzero(&ND, sizeof(ND));
1646 		ND.linkmtu = IN6_LINKMTU(ifp);
1647 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1648 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1649 		ND.reachable = ND_IFINFO(ifp)->reachable;
1650 		ND.retrans = ND_IFINFO(ifp)->retrans;
1651 		ND.flags = ND_IFINFO(ifp)->flags;
1652 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1653 		ND.chlim = ND_IFINFO(ifp)->chlim;
1654 		break;
1655 	case SIOCGIFINFO_IN6:
1656 		ND = *ND_IFINFO(ifp);
1657 		break;
1658 	case SIOCSIFINFO_IN6:
1659 		/*
1660 		 * used to change host variables from userland.
1661 		 * intended for a use on router to reflect RA configurations.
1662 		 */
1663 		/* 0 means 'unspecified' */
1664 		if (ND.linkmtu != 0) {
1665 			if (ND.linkmtu < IPV6_MMTU ||
1666 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1667 				error = EINVAL;
1668 				break;
1669 			}
1670 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1671 		}
1672 
1673 		if (ND.basereachable != 0) {
1674 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1675 
1676 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1677 			if (ND.basereachable != obasereachable)
1678 				ND_IFINFO(ifp)->reachable =
1679 				    ND_COMPUTE_RTIME(ND.basereachable);
1680 		}
1681 		if (ND.retrans != 0)
1682 			ND_IFINFO(ifp)->retrans = ND.retrans;
1683 		if (ND.chlim != 0)
1684 			ND_IFINFO(ifp)->chlim = ND.chlim;
1685 		/* FALLTHROUGH */
1686 	case SIOCSIFINFO_FLAGS:
1687 	{
1688 		struct ifaddr *ifa;
1689 		struct in6_ifaddr *ia;
1690 
1691 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1692 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1693 			/* ifdisabled 1->0 transision */
1694 
1695 			/*
1696 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1697 			 * has an link-local address with IN6_IFF_DUPLICATED,
1698 			 * do not clear ND6_IFF_IFDISABLED.
1699 			 * See RFC 4862, Section 5.4.5.
1700 			 */
1701 			NET_EPOCH_ENTER(et);
1702 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1703 				if (ifa->ifa_addr->sa_family != AF_INET6)
1704 					continue;
1705 				ia = (struct in6_ifaddr *)ifa;
1706 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1707 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1708 					break;
1709 			}
1710 			NET_EPOCH_EXIT(et);
1711 
1712 			if (ifa != NULL) {
1713 				/* LLA is duplicated. */
1714 				ND.flags |= ND6_IFF_IFDISABLED;
1715 				log(LOG_ERR, "Cannot enable an interface"
1716 				    " with a link-local address marked"
1717 				    " duplicate.\n");
1718 			} else {
1719 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1720 				if (ifp->if_flags & IFF_UP)
1721 					in6_if_up(ifp);
1722 			}
1723 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1724 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1725 			/* ifdisabled 0->1 transision */
1726 			/* Mark all IPv6 address as tentative. */
1727 
1728 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1729 			if (V_ip6_dad_count > 0 &&
1730 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1731 				NET_EPOCH_ENTER(et);
1732 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1733 				    ifa_link) {
1734 					if (ifa->ifa_addr->sa_family !=
1735 					    AF_INET6)
1736 						continue;
1737 					ia = (struct in6_ifaddr *)ifa;
1738 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1739 				}
1740 				NET_EPOCH_EXIT(et);
1741 			}
1742 		}
1743 
1744 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1745 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1746 				/* auto_linklocal 0->1 transision */
1747 
1748 				/* If no link-local address on ifp, configure */
1749 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1750 				in6_ifattach(ifp, NULL);
1751 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1752 			    ifp->if_flags & IFF_UP) {
1753 				/*
1754 				 * When the IF already has
1755 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1756 				 * address is assigned, and IFF_UP, try to
1757 				 * assign one.
1758 				 */
1759 				NET_EPOCH_ENTER(et);
1760 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1761 				    ifa_link) {
1762 					if (ifa->ifa_addr->sa_family !=
1763 					    AF_INET6)
1764 						continue;
1765 					ia = (struct in6_ifaddr *)ifa;
1766 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1767 						break;
1768 				}
1769 				NET_EPOCH_EXIT(et);
1770 				if (ifa != NULL)
1771 					/* No LLA is configured. */
1772 					in6_ifattach(ifp, NULL);
1773 			}
1774 		}
1775 	}
1776 		ND_IFINFO(ifp)->flags = ND.flags;
1777 		break;
1778 #undef ND
1779 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1780 		/* sync kernel routing table with the default router list */
1781 		defrouter_reset();
1782 		defrouter_select();
1783 		break;
1784 	case SIOCSPFXFLUSH_IN6:
1785 	{
1786 		/* flush all the prefix advertised by routers */
1787 		struct in6_ifaddr *ia, *ia_next;
1788 		struct nd_prefix *pr, *next;
1789 		struct nd_prhead prl;
1790 
1791 		LIST_INIT(&prl);
1792 
1793 		ND6_WLOCK();
1794 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1795 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1796 				continue; /* XXX */
1797 			nd6_prefix_unlink(pr, &prl);
1798 		}
1799 		ND6_WUNLOCK();
1800 
1801 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1802 			LIST_REMOVE(pr, ndpr_entry);
1803 			/* XXXRW: in6_ifaddrhead locking. */
1804 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1805 			    ia_next) {
1806 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1807 					continue;
1808 
1809 				if (ia->ia6_ndpr == pr)
1810 					in6_purgeaddr(&ia->ia_ifa);
1811 			}
1812 			nd6_prefix_del(pr);
1813 		}
1814 		break;
1815 	}
1816 	case SIOCSRTRFLUSH_IN6:
1817 	{
1818 		/* flush all the default routers */
1819 		struct nd_drhead drq;
1820 		struct nd_defrouter *dr;
1821 
1822 		TAILQ_INIT(&drq);
1823 
1824 		defrouter_reset();
1825 
1826 		ND6_WLOCK();
1827 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1828 			defrouter_unlink(dr, &drq);
1829 		ND6_WUNLOCK();
1830 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1831 			TAILQ_REMOVE(&drq, dr, dr_entry);
1832 			defrouter_del(dr);
1833 		}
1834 
1835 		defrouter_select();
1836 		break;
1837 	}
1838 	case SIOCGNBRINFO_IN6:
1839 	{
1840 		struct llentry *ln;
1841 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1842 
1843 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1844 			return (error);
1845 
1846 		NET_EPOCH_ENTER(et);
1847 		ln = nd6_lookup(&nb_addr, 0, ifp);
1848 		NET_EPOCH_EXIT(et);
1849 
1850 		if (ln == NULL) {
1851 			error = EINVAL;
1852 			break;
1853 		}
1854 		nbi->state = ln->ln_state;
1855 		nbi->asked = ln->la_asked;
1856 		nbi->isrouter = ln->ln_router;
1857 		if (ln->la_expire == 0)
1858 			nbi->expire = 0;
1859 		else
1860 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1861 			    (time_second - time_uptime);
1862 		LLE_RUNLOCK(ln);
1863 		break;
1864 	}
1865 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1866 		ndif->ifindex = V_nd6_defifindex;
1867 		break;
1868 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1869 		return (nd6_setdefaultiface(ndif->ifindex));
1870 	}
1871 	return (error);
1872 }
1873 
1874 /*
1875  * Calculates new isRouter value based on provided parameters and
1876  * returns it.
1877  */
1878 static int
1879 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1880     int ln_router)
1881 {
1882 
1883 	/*
1884 	 * ICMP6 type dependent behavior.
1885 	 *
1886 	 * NS: clear IsRouter if new entry
1887 	 * RS: clear IsRouter
1888 	 * RA: set IsRouter if there's lladdr
1889 	 * redir: clear IsRouter if new entry
1890 	 *
1891 	 * RA case, (1):
1892 	 * The spec says that we must set IsRouter in the following cases:
1893 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1894 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1895 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1896 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1897 	 * neighbor cache, this is similar to (6).
1898 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1899 	 *
1900 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1901 	 *							D R
1902 	 *	0	n	n	(1)	c   ?     s
1903 	 *	0	y	n	(2)	c   s     s
1904 	 *	0	n	y	(3)	c   s     s
1905 	 *	0	y	y	(4)	c   s     s
1906 	 *	0	y	y	(5)	c   s     s
1907 	 *	1	--	n	(6) c	c	c s
1908 	 *	1	--	y	(7) c	c   s	c s
1909 	 *
1910 	 *					(c=clear s=set)
1911 	 */
1912 	switch (type & 0xff) {
1913 	case ND_NEIGHBOR_SOLICIT:
1914 		/*
1915 		 * New entry must have is_router flag cleared.
1916 		 */
1917 		if (is_new)					/* (6-7) */
1918 			ln_router = 0;
1919 		break;
1920 	case ND_REDIRECT:
1921 		/*
1922 		 * If the icmp is a redirect to a better router, always set the
1923 		 * is_router flag.  Otherwise, if the entry is newly created,
1924 		 * clear the flag.  [RFC 2461, sec 8.3]
1925 		 */
1926 		if (code == ND_REDIRECT_ROUTER)
1927 			ln_router = 1;
1928 		else {
1929 			if (is_new)				/* (6-7) */
1930 				ln_router = 0;
1931 		}
1932 		break;
1933 	case ND_ROUTER_SOLICIT:
1934 		/*
1935 		 * is_router flag must always be cleared.
1936 		 */
1937 		ln_router = 0;
1938 		break;
1939 	case ND_ROUTER_ADVERT:
1940 		/*
1941 		 * Mark an entry with lladdr as a router.
1942 		 */
1943 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1944 		    (is_new && new_addr)) {			/* (7) */
1945 			ln_router = 1;
1946 		}
1947 		break;
1948 	}
1949 
1950 	return (ln_router);
1951 }
1952 
1953 /*
1954  * Create neighbor cache entry and cache link-layer address,
1955  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1956  *
1957  * type - ICMP6 type
1958  * code - type dependent information
1959  *
1960  */
1961 void
1962 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1963     int lladdrlen, int type, int code)
1964 {
1965 	struct llentry *ln = NULL, *ln_tmp;
1966 	int is_newentry;
1967 	int do_update;
1968 	int olladdr;
1969 	int llchange;
1970 	int flags;
1971 	uint16_t router = 0;
1972 	struct sockaddr_in6 sin6;
1973 	struct epoch_tracker et;
1974 	struct mbuf *chain = NULL;
1975 	u_char linkhdr[LLE_MAX_LINKHDR];
1976 	size_t linkhdrsize;
1977 	int lladdr_off;
1978 
1979 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1980 
1981 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1982 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1983 
1984 	/* nothing must be updated for unspecified address */
1985 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1986 		return;
1987 
1988 	/*
1989 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1990 	 * the caller.
1991 	 *
1992 	 * XXX If the link does not have link-layer adderss, what should
1993 	 * we do? (ifp->if_addrlen == 0)
1994 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1995 	 * description on it in NS section (RFC 2461 7.2.3).
1996 	 */
1997 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1998 	NET_EPOCH_ENTER(et);
1999 	ln = nd6_lookup(from, flags, ifp);
2000 	NET_EPOCH_EXIT(et);
2001 	is_newentry = 0;
2002 	if (ln == NULL) {
2003 		flags |= LLE_EXCLUSIVE;
2004 		ln = nd6_alloc(from, 0, ifp);
2005 		if (ln == NULL)
2006 			return;
2007 
2008 		/*
2009 		 * Since we already know all the data for the new entry,
2010 		 * fill it before insertion.
2011 		 */
2012 		if (lladdr != NULL) {
2013 			linkhdrsize = sizeof(linkhdr);
2014 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2015 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2016 				return;
2017 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2018 			    lladdr_off);
2019 		}
2020 
2021 		IF_AFDATA_WLOCK(ifp);
2022 		LLE_WLOCK(ln);
2023 		/* Prefer any existing lle over newly-created one */
2024 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
2025 		if (ln_tmp == NULL)
2026 			lltable_link_entry(LLTABLE6(ifp), ln);
2027 		IF_AFDATA_WUNLOCK(ifp);
2028 		if (ln_tmp == NULL) {
2029 			/* No existing lle, mark as new entry (6,7) */
2030 			is_newentry = 1;
2031 			if (lladdr != NULL) {	/* (7) */
2032 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2033 				EVENTHANDLER_INVOKE(lle_event, ln,
2034 				    LLENTRY_RESOLVED);
2035 			}
2036 		} else {
2037 			lltable_free_entry(LLTABLE6(ifp), ln);
2038 			ln = ln_tmp;
2039 			ln_tmp = NULL;
2040 		}
2041 	}
2042 	/* do nothing if static ndp is set */
2043 	if ((ln->la_flags & LLE_STATIC)) {
2044 		if (flags & LLE_EXCLUSIVE)
2045 			LLE_WUNLOCK(ln);
2046 		else
2047 			LLE_RUNLOCK(ln);
2048 		return;
2049 	}
2050 
2051 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2052 	if (olladdr && lladdr) {
2053 		llchange = bcmp(lladdr, ln->ll_addr,
2054 		    ifp->if_addrlen);
2055 	} else if (!olladdr && lladdr)
2056 		llchange = 1;
2057 	else
2058 		llchange = 0;
2059 
2060 	/*
2061 	 * newentry olladdr  lladdr  llchange	(*=record)
2062 	 *	0	n	n	--	(1)
2063 	 *	0	y	n	--	(2)
2064 	 *	0	n	y	y	(3) * STALE
2065 	 *	0	y	y	n	(4) *
2066 	 *	0	y	y	y	(5) * STALE
2067 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2068 	 *	1	--	y	--	(7) * STALE
2069 	 */
2070 
2071 	do_update = 0;
2072 	if (is_newentry == 0 && llchange != 0) {
2073 		do_update = 1;	/* (3,5) */
2074 
2075 		/*
2076 		 * Record source link-layer address
2077 		 * XXX is it dependent to ifp->if_type?
2078 		 */
2079 		linkhdrsize = sizeof(linkhdr);
2080 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2081 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2082 			return;
2083 
2084 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2085 		    lladdr_off) == 0) {
2086 			/* Entry was deleted */
2087 			return;
2088 		}
2089 
2090 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2091 
2092 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2093 
2094 		if (ln->la_hold != NULL)
2095 			nd6_grab_holdchain(ln, &chain, &sin6);
2096 	}
2097 
2098 	/* Calculates new router status */
2099 	router = nd6_is_router(type, code, is_newentry, olladdr,
2100 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2101 
2102 	ln->ln_router = router;
2103 	/* Mark non-router redirects with special flag */
2104 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2105 		ln->la_flags |= LLE_REDIRECT;
2106 
2107 	if (flags & LLE_EXCLUSIVE)
2108 		LLE_WUNLOCK(ln);
2109 	else
2110 		LLE_RUNLOCK(ln);
2111 
2112 	if (chain != NULL)
2113 		nd6_flush_holdchain(ifp, chain, &sin6);
2114 
2115 	/*
2116 	 * When the link-layer address of a router changes, select the
2117 	 * best router again.  In particular, when the neighbor entry is newly
2118 	 * created, it might affect the selection policy.
2119 	 * Question: can we restrict the first condition to the "is_newentry"
2120 	 * case?
2121 	 * XXX: when we hear an RA from a new router with the link-layer
2122 	 * address option, defrouter_select_fib() is called twice, since
2123 	 * defrtrlist_update called the function as well.  However, I believe
2124 	 * we can compromise the overhead, since it only happens the first
2125 	 * time.
2126 	 * XXX: although defrouter_select_fib() should not have a bad effect
2127 	 * for those are not autoconfigured hosts, we explicitly avoid such
2128 	 * cases for safety.
2129 	 */
2130 	if ((do_update || is_newentry) && router &&
2131 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2132 		/*
2133 		 * guaranteed recursion
2134 		 */
2135 		defrouter_select_fib(ifp->if_fib);
2136 	}
2137 }
2138 
2139 static void
2140 nd6_slowtimo(void *arg)
2141 {
2142 	struct epoch_tracker et;
2143 	CURVNET_SET((struct vnet *) arg);
2144 	struct nd_ifinfo *nd6if;
2145 	struct ifnet *ifp;
2146 
2147 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2148 	    nd6_slowtimo, curvnet);
2149 	NET_EPOCH_ENTER(et);
2150 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2151 		if (ifp->if_afdata[AF_INET6] == NULL)
2152 			continue;
2153 		nd6if = ND_IFINFO(ifp);
2154 		if (nd6if->basereachable && /* already initialized */
2155 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2156 			/*
2157 			 * Since reachable time rarely changes by router
2158 			 * advertisements, we SHOULD insure that a new random
2159 			 * value gets recomputed at least once every few hours.
2160 			 * (RFC 2461, 6.3.4)
2161 			 */
2162 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2163 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2164 		}
2165 	}
2166 	NET_EPOCH_EXIT(et);
2167 	CURVNET_RESTORE();
2168 }
2169 
2170 void
2171 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2172     struct sockaddr_in6 *sin6)
2173 {
2174 
2175 	LLE_WLOCK_ASSERT(ln);
2176 
2177 	*chain = ln->la_hold;
2178 	ln->la_hold = NULL;
2179 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2180 
2181 	if (ln->ln_state == ND6_LLINFO_STALE) {
2182 
2183 		/*
2184 		 * The first time we send a packet to a
2185 		 * neighbor whose entry is STALE, we have
2186 		 * to change the state to DELAY and a sets
2187 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2188 		 * seconds to ensure do neighbor unreachability
2189 		 * detection on expiration.
2190 		 * (RFC 2461 7.3.3)
2191 		 */
2192 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2193 	}
2194 }
2195 
2196 int
2197 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2198     struct sockaddr_in6 *dst, struct route *ro)
2199 {
2200 	int error;
2201 	int ip6len;
2202 	struct ip6_hdr *ip6;
2203 	struct m_tag *mtag;
2204 
2205 #ifdef MAC
2206 	mac_netinet6_nd6_send(ifp, m);
2207 #endif
2208 
2209 	/*
2210 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2211 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2212 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2213 	 * to be diverted to user space.  When re-injected into the kernel,
2214 	 * send_output() will directly dispatch them to the outgoing interface.
2215 	 */
2216 	if (send_sendso_input_hook != NULL) {
2217 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2218 		if (mtag != NULL) {
2219 			ip6 = mtod(m, struct ip6_hdr *);
2220 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2221 			/* Use the SEND socket */
2222 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2223 			    ip6len);
2224 			/* -1 == no app on SEND socket */
2225 			if (error == 0 || error != -1)
2226 			    return (error);
2227 		}
2228 	}
2229 
2230 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2231 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2232 	    mtod(m, struct ip6_hdr *));
2233 
2234 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2235 		origifp = ifp;
2236 
2237 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2238 	return (error);
2239 }
2240 
2241 /*
2242  * Lookup link headerfor @sa_dst address. Stores found
2243  * data in @desten buffer. Copy of lle ln_flags can be also
2244  * saved in @pflags if @pflags is non-NULL.
2245  *
2246  * If destination LLE does not exists or lle state modification
2247  * is required, call "slow" version.
2248  *
2249  * Return values:
2250  * - 0 on success (address copied to buffer).
2251  * - EWOULDBLOCK (no local error, but address is still unresolved)
2252  * - other errors (alloc failure, etc)
2253  */
2254 int
2255 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2256     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2257     struct llentry **plle)
2258 {
2259 	struct epoch_tracker et;
2260 	struct llentry *ln = NULL;
2261 	const struct sockaddr_in6 *dst6;
2262 
2263 	if (pflags != NULL)
2264 		*pflags = 0;
2265 
2266 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2267 
2268 	/* discard the packet if IPv6 operation is disabled on the interface */
2269 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2270 		m_freem(m);
2271 		return (ENETDOWN); /* better error? */
2272 	}
2273 
2274 	if (m != NULL && m->m_flags & M_MCAST) {
2275 		switch (ifp->if_type) {
2276 		case IFT_ETHER:
2277 		case IFT_L2VLAN:
2278 		case IFT_BRIDGE:
2279 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2280 						 desten);
2281 			return (0);
2282 		default:
2283 			m_freem(m);
2284 			return (EAFNOSUPPORT);
2285 		}
2286 	}
2287 
2288 	NET_EPOCH_ENTER(et);
2289 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2290 	    ifp);
2291 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2292 		/* Entry found, let's copy lle info */
2293 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2294 		if (pflags != NULL)
2295 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2296 		/* Check if we have feedback request from nd6 timer */
2297 		if (ln->r_skip_req != 0) {
2298 			LLE_REQ_LOCK(ln);
2299 			ln->r_skip_req = 0; /* Notify that entry was used */
2300 			ln->lle_hittime = time_uptime;
2301 			LLE_REQ_UNLOCK(ln);
2302 		}
2303 		if (plle) {
2304 			LLE_ADDREF(ln);
2305 			*plle = ln;
2306 			LLE_WUNLOCK(ln);
2307 		}
2308 		NET_EPOCH_EXIT(et);
2309 		return (0);
2310 	} else if (plle && ln)
2311 		LLE_WUNLOCK(ln);
2312 	NET_EPOCH_EXIT(et);
2313 
2314 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2315 }
2316 
2317 
2318 /*
2319  * Do L2 address resolution for @sa_dst address. Stores found
2320  * address in @desten buffer. Copy of lle ln_flags can be also
2321  * saved in @pflags if @pflags is non-NULL.
2322  *
2323  * Heavy version.
2324  * Function assume that destination LLE does not exist,
2325  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2326  *
2327  * Set noinline to be dtrace-friendly
2328  */
2329 static __noinline int
2330 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2331     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2332     struct llentry **plle)
2333 {
2334 	struct llentry *lle = NULL, *lle_tmp;
2335 	struct in6_addr *psrc, src;
2336 	int send_ns, ll_len;
2337 	char *lladdr;
2338 
2339 	/*
2340 	 * Address resolution or Neighbor Unreachability Detection
2341 	 * for the next hop.
2342 	 * At this point, the destination of the packet must be a unicast
2343 	 * or an anycast address(i.e. not a multicast).
2344 	 */
2345 	if (lle == NULL) {
2346 		struct epoch_tracker et;
2347 
2348 		NET_EPOCH_ENTER(et);
2349 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2350 		NET_EPOCH_EXIT(et);
2351 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2352 			/*
2353 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2354 			 * the condition below is not very efficient.  But we believe
2355 			 * it is tolerable, because this should be a rare case.
2356 			 */
2357 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2358 			if (lle == NULL) {
2359 				char ip6buf[INET6_ADDRSTRLEN];
2360 				log(LOG_DEBUG,
2361 				    "nd6_output: can't allocate llinfo for %s "
2362 				    "(ln=%p)\n",
2363 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2364 				m_freem(m);
2365 				return (ENOBUFS);
2366 			}
2367 
2368 			IF_AFDATA_WLOCK(ifp);
2369 			LLE_WLOCK(lle);
2370 			/* Prefer any existing entry over newly-created one */
2371 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2372 			if (lle_tmp == NULL)
2373 				lltable_link_entry(LLTABLE6(ifp), lle);
2374 			IF_AFDATA_WUNLOCK(ifp);
2375 			if (lle_tmp != NULL) {
2376 				lltable_free_entry(LLTABLE6(ifp), lle);
2377 				lle = lle_tmp;
2378 				lle_tmp = NULL;
2379 			}
2380 		}
2381 	}
2382 	if (lle == NULL) {
2383 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2384 			m_freem(m);
2385 			return (ENOBUFS);
2386 		}
2387 
2388 		if (m != NULL)
2389 			m_freem(m);
2390 		return (ENOBUFS);
2391 	}
2392 
2393 	LLE_WLOCK_ASSERT(lle);
2394 
2395 	/*
2396 	 * The first time we send a packet to a neighbor whose entry is
2397 	 * STALE, we have to change the state to DELAY and a sets a timer to
2398 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2399 	 * neighbor unreachability detection on expiration.
2400 	 * (RFC 2461 7.3.3)
2401 	 */
2402 	if (lle->ln_state == ND6_LLINFO_STALE)
2403 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2404 
2405 	/*
2406 	 * If the neighbor cache entry has a state other than INCOMPLETE
2407 	 * (i.e. its link-layer address is already resolved), just
2408 	 * send the packet.
2409 	 */
2410 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2411 		if (flags & LLE_ADDRONLY) {
2412 			lladdr = lle->ll_addr;
2413 			ll_len = ifp->if_addrlen;
2414 		} else {
2415 			lladdr = lle->r_linkdata;
2416 			ll_len = lle->r_hdrlen;
2417 		}
2418 		bcopy(lladdr, desten, ll_len);
2419 		if (pflags != NULL)
2420 			*pflags = lle->la_flags;
2421 		if (plle) {
2422 			LLE_ADDREF(lle);
2423 			*plle = lle;
2424 		}
2425 		LLE_WUNLOCK(lle);
2426 		return (0);
2427 	}
2428 
2429 	/*
2430 	 * There is a neighbor cache entry, but no ethernet address
2431 	 * response yet.  Append this latest packet to the end of the
2432 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2433 	 * the oldest packet in the queue will be removed.
2434 	 */
2435 
2436 	if (lle->la_hold != NULL) {
2437 		struct mbuf *m_hold;
2438 		int i;
2439 
2440 		i = 0;
2441 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2442 			i++;
2443 			if (m_hold->m_nextpkt == NULL) {
2444 				m_hold->m_nextpkt = m;
2445 				break;
2446 			}
2447 		}
2448 		while (i >= V_nd6_maxqueuelen) {
2449 			m_hold = lle->la_hold;
2450 			lle->la_hold = lle->la_hold->m_nextpkt;
2451 			m_freem(m_hold);
2452 			i--;
2453 		}
2454 	} else {
2455 		lle->la_hold = m;
2456 	}
2457 
2458 	/*
2459 	 * If there has been no NS for the neighbor after entering the
2460 	 * INCOMPLETE state, send the first solicitation.
2461 	 * Note that for newly-created lle la_asked will be 0,
2462 	 * so we will transition from ND6_LLINFO_NOSTATE to
2463 	 * ND6_LLINFO_INCOMPLETE state here.
2464 	 */
2465 	psrc = NULL;
2466 	send_ns = 0;
2467 	if (lle->la_asked == 0) {
2468 		lle->la_asked++;
2469 		send_ns = 1;
2470 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2471 
2472 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2473 	}
2474 	LLE_WUNLOCK(lle);
2475 	if (send_ns != 0)
2476 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2477 
2478 	return (EWOULDBLOCK);
2479 }
2480 
2481 /*
2482  * Do L2 address resolution for @sa_dst address. Stores found
2483  * address in @desten buffer. Copy of lle ln_flags can be also
2484  * saved in @pflags if @pflags is non-NULL.
2485  *
2486  * Return values:
2487  * - 0 on success (address copied to buffer).
2488  * - EWOULDBLOCK (no local error, but address is still unresolved)
2489  * - other errors (alloc failure, etc)
2490  */
2491 int
2492 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2493     char *desten, uint32_t *pflags)
2494 {
2495 	int error;
2496 
2497 	flags |= LLE_ADDRONLY;
2498 	error = nd6_resolve_slow(ifp, flags, NULL,
2499 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2500 	return (error);
2501 }
2502 
2503 int
2504 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2505     struct sockaddr_in6 *dst)
2506 {
2507 	struct mbuf *m, *m_head;
2508 	int error = 0;
2509 
2510 	m_head = chain;
2511 
2512 	while (m_head) {
2513 		m = m_head;
2514 		m_head = m_head->m_nextpkt;
2515 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2516 	}
2517 
2518 	/*
2519 	 * XXX
2520 	 * note that intermediate errors are blindly ignored
2521 	 */
2522 	return (error);
2523 }
2524 
2525 static int
2526 nd6_need_cache(struct ifnet *ifp)
2527 {
2528 	/*
2529 	 * XXX: we currently do not make neighbor cache on any interface
2530 	 * other than Ethernet and GIF.
2531 	 *
2532 	 * RFC2893 says:
2533 	 * - unidirectional tunnels needs no ND
2534 	 */
2535 	switch (ifp->if_type) {
2536 	case IFT_ETHER:
2537 	case IFT_IEEE1394:
2538 	case IFT_L2VLAN:
2539 	case IFT_INFINIBAND:
2540 	case IFT_BRIDGE:
2541 	case IFT_PROPVIRTUAL:
2542 		return (1);
2543 	default:
2544 		return (0);
2545 	}
2546 }
2547 
2548 /*
2549  * Add pernament ND6 link-layer record for given
2550  * interface address.
2551  *
2552  * Very similar to IPv4 arp_ifinit(), but:
2553  * 1) IPv6 DAD is performed in different place
2554  * 2) It is called by IPv6 protocol stack in contrast to
2555  * arp_ifinit() which is typically called in SIOCSIFADDR
2556  * driver ioctl handler.
2557  *
2558  */
2559 int
2560 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2561 {
2562 	struct ifnet *ifp;
2563 	struct llentry *ln, *ln_tmp;
2564 	struct sockaddr *dst;
2565 
2566 	ifp = ia->ia_ifa.ifa_ifp;
2567 	if (nd6_need_cache(ifp) == 0)
2568 		return (0);
2569 
2570 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2571 	dst = (struct sockaddr *)&ia->ia_addr;
2572 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2573 	if (ln == NULL)
2574 		return (ENOBUFS);
2575 
2576 	IF_AFDATA_WLOCK(ifp);
2577 	LLE_WLOCK(ln);
2578 	/* Unlink any entry if exists */
2579 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2580 	if (ln_tmp != NULL)
2581 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2582 	lltable_link_entry(LLTABLE6(ifp), ln);
2583 	IF_AFDATA_WUNLOCK(ifp);
2584 
2585 	if (ln_tmp != NULL)
2586 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2587 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2588 
2589 	LLE_WUNLOCK(ln);
2590 	if (ln_tmp != NULL)
2591 		llentry_free(ln_tmp);
2592 
2593 	return (0);
2594 }
2595 
2596 /*
2597  * Removes either all lle entries for given @ia, or lle
2598  * corresponding to @ia address.
2599  */
2600 void
2601 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2602 {
2603 	struct sockaddr_in6 mask, addr;
2604 	struct sockaddr *saddr, *smask;
2605 	struct ifnet *ifp;
2606 
2607 	ifp = ia->ia_ifa.ifa_ifp;
2608 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2609 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2610 	saddr = (struct sockaddr *)&addr;
2611 	smask = (struct sockaddr *)&mask;
2612 
2613 	if (all != 0)
2614 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2615 	else
2616 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2617 }
2618 
2619 static void
2620 clear_llinfo_pqueue(struct llentry *ln)
2621 {
2622 	struct mbuf *m_hold, *m_hold_next;
2623 
2624 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2625 		m_hold_next = m_hold->m_nextpkt;
2626 		m_freem(m_hold);
2627 	}
2628 
2629 	ln->la_hold = NULL;
2630 }
2631 
2632 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2633 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2634 
2635 SYSCTL_DECL(_net_inet6_icmp6);
2636 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2637 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2638 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2639 	"NDP default router list");
2640 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2641 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2642 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2643 	"NDP prefix list");
2644 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2645 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2646 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2647 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2648 
2649 static int
2650 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2651 {
2652 	struct in6_defrouter d;
2653 	struct nd_defrouter *dr;
2654 	int error;
2655 
2656 	if (req->newptr != NULL)
2657 		return (EPERM);
2658 
2659 	error = sysctl_wire_old_buffer(req, 0);
2660 	if (error != 0)
2661 		return (error);
2662 
2663 	bzero(&d, sizeof(d));
2664 	d.rtaddr.sin6_family = AF_INET6;
2665 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2666 
2667 	ND6_RLOCK();
2668 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2669 		d.rtaddr.sin6_addr = dr->rtaddr;
2670 		error = sa6_recoverscope(&d.rtaddr);
2671 		if (error != 0)
2672 			break;
2673 		d.flags = dr->raflags;
2674 		d.rtlifetime = dr->rtlifetime;
2675 		d.expire = dr->expire + (time_second - time_uptime);
2676 		d.if_index = dr->ifp->if_index;
2677 		error = SYSCTL_OUT(req, &d, sizeof(d));
2678 		if (error != 0)
2679 			break;
2680 	}
2681 	ND6_RUNLOCK();
2682 	return (error);
2683 }
2684 
2685 static int
2686 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2687 {
2688 	struct in6_prefix p;
2689 	struct sockaddr_in6 s6;
2690 	struct nd_prefix *pr;
2691 	struct nd_pfxrouter *pfr;
2692 	time_t maxexpire;
2693 	int error;
2694 	char ip6buf[INET6_ADDRSTRLEN];
2695 
2696 	if (req->newptr)
2697 		return (EPERM);
2698 
2699 	error = sysctl_wire_old_buffer(req, 0);
2700 	if (error != 0)
2701 		return (error);
2702 
2703 	bzero(&p, sizeof(p));
2704 	p.origin = PR_ORIG_RA;
2705 	bzero(&s6, sizeof(s6));
2706 	s6.sin6_family = AF_INET6;
2707 	s6.sin6_len = sizeof(s6);
2708 
2709 	ND6_RLOCK();
2710 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2711 		p.prefix = pr->ndpr_prefix;
2712 		if (sa6_recoverscope(&p.prefix)) {
2713 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2714 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2715 			/* XXX: press on... */
2716 		}
2717 		p.raflags = pr->ndpr_raf;
2718 		p.prefixlen = pr->ndpr_plen;
2719 		p.vltime = pr->ndpr_vltime;
2720 		p.pltime = pr->ndpr_pltime;
2721 		p.if_index = pr->ndpr_ifp->if_index;
2722 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2723 			p.expire = 0;
2724 		else {
2725 			/* XXX: we assume time_t is signed. */
2726 			maxexpire = (-1) &
2727 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2728 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2729 				p.expire = pr->ndpr_lastupdate +
2730 				    pr->ndpr_vltime +
2731 				    (time_second - time_uptime);
2732 			else
2733 				p.expire = maxexpire;
2734 		}
2735 		p.refcnt = pr->ndpr_addrcnt;
2736 		p.flags = pr->ndpr_stateflags;
2737 		p.advrtrs = 0;
2738 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2739 			p.advrtrs++;
2740 		error = SYSCTL_OUT(req, &p, sizeof(p));
2741 		if (error != 0)
2742 			break;
2743 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2744 			s6.sin6_addr = pfr->router->rtaddr;
2745 			if (sa6_recoverscope(&s6))
2746 				log(LOG_ERR,
2747 				    "scope error in prefix list (%s)\n",
2748 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2749 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2750 			if (error != 0)
2751 				goto out;
2752 		}
2753 	}
2754 out:
2755 	ND6_RUNLOCK();
2756 	return (error);
2757 }
2758