xref: /freebsd/sys/netinet6/nd6.c (revision 65a1d63665bf8b9cd2596d4319b31efc2232b722)
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/route/route_var.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_var.h>
74 #include <netinet/ip6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/scope6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet6/in6_ifattach.h>
79 #include <netinet/icmp6.h>
80 #include <netinet6/send.h>
81 
82 #include <sys/limits.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
87 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
88 
89 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
90 
91 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
92 
93 /* timer values */
94 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
95 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
96 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
97 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
98 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
99 					 * local traffic */
100 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
101 					 * collection timer */
102 
103 /* preventing too many loops in ND option parsing */
104 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
105 
106 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
107 					 * layer hints */
108 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
109 					 * ND entries */
110 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
111 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
112 
113 #ifdef ND6_DEBUG
114 VNET_DEFINE(int, nd6_debug) = 1;
115 #else
116 VNET_DEFINE(int, nd6_debug) = 0;
117 #endif
118 
119 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
120 
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
122 VNET_DEFINE(struct rwlock, nd6_lock);
123 VNET_DEFINE(uint64_t, nd6_list_genid);
124 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
125 
126 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
127 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
128 
129 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
130 
131 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
132 	struct ifnet *);
133 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
134 static void nd6_slowtimo(void *);
135 static int regen_tmpaddr(struct in6_ifaddr *);
136 static void nd6_free(struct llentry **, int);
137 static void nd6_free_redirect(const struct llentry *);
138 static void nd6_llinfo_timer(void *);
139 static void nd6_llinfo_settimer_locked(struct llentry *, long);
140 static void clear_llinfo_pqueue(struct llentry *);
141 static void nd6_rtrequest(int, struct rtentry *, struct nhop_object *,
142     struct rt_addrinfo *);
143 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
144     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
145 static int nd6_need_cache(struct ifnet *);
146 
147 
148 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
149 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
150 
151 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
152 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
153 
154 SYSCTL_DECL(_net_inet6_icmp6);
155 
156 static void
157 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
158 {
159 	struct rt_addrinfo rtinfo;
160 	struct sockaddr_in6 dst;
161 	struct sockaddr_dl gw;
162 	struct ifnet *ifp;
163 	int type;
164 	int fibnum;
165 
166 	LLE_WLOCK_ASSERT(lle);
167 
168 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
169 		return;
170 
171 	switch (evt) {
172 	case LLENTRY_RESOLVED:
173 		type = RTM_ADD;
174 		KASSERT(lle->la_flags & LLE_VALID,
175 		    ("%s: %p resolved but not valid?", __func__, lle));
176 		break;
177 	case LLENTRY_EXPIRED:
178 		type = RTM_DELETE;
179 		break;
180 	default:
181 		return;
182 	}
183 
184 	ifp = lltable_get_ifp(lle->lle_tbl);
185 
186 	bzero(&dst, sizeof(dst));
187 	bzero(&gw, sizeof(gw));
188 	bzero(&rtinfo, sizeof(rtinfo));
189 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
190 	dst.sin6_scope_id = in6_getscopezone(ifp,
191 	    in6_addrscope(&dst.sin6_addr));
192 	gw.sdl_len = sizeof(struct sockaddr_dl);
193 	gw.sdl_family = AF_LINK;
194 	gw.sdl_alen = ifp->if_addrlen;
195 	gw.sdl_index = ifp->if_index;
196 	gw.sdl_type = ifp->if_type;
197 	if (evt == LLENTRY_RESOLVED)
198 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
199 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
200 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
201 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
202 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
203 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
204 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
205 }
206 
207 /*
208  * A handler for interface link layer address change event.
209  */
210 static void
211 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
212 {
213 
214 	lltable_update_ifaddr(LLTABLE6(ifp));
215 }
216 
217 void
218 nd6_init(void)
219 {
220 
221 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
222 	rw_init(&V_nd6_lock, "nd6 list");
223 
224 	LIST_INIT(&V_nd_prefix);
225 	nd6_defrouter_init();
226 
227 	/* Start timers. */
228 	callout_init(&V_nd6_slowtimo_ch, 0);
229 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
230 	    nd6_slowtimo, curvnet);
231 
232 	callout_init(&V_nd6_timer_ch, 0);
233 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
234 
235 	nd6_dad_init();
236 	if (IS_DEFAULT_VNET(curvnet)) {
237 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
238 		    NULL, EVENTHANDLER_PRI_ANY);
239 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
240 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
241 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
242 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
243 	}
244 }
245 
246 #ifdef VIMAGE
247 void
248 nd6_destroy()
249 {
250 
251 	callout_drain(&V_nd6_slowtimo_ch);
252 	callout_drain(&V_nd6_timer_ch);
253 	if (IS_DEFAULT_VNET(curvnet)) {
254 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
255 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
256 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
257 	}
258 	rw_destroy(&V_nd6_lock);
259 	mtx_destroy(&V_nd6_onlink_mtx);
260 }
261 #endif
262 
263 struct nd_ifinfo *
264 nd6_ifattach(struct ifnet *ifp)
265 {
266 	struct nd_ifinfo *nd;
267 
268 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
269 	nd->initialized = 1;
270 
271 	nd->chlim = IPV6_DEFHLIM;
272 	nd->basereachable = REACHABLE_TIME;
273 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
274 	nd->retrans = RETRANS_TIMER;
275 
276 	nd->flags = ND6_IFF_PERFORMNUD;
277 
278 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
279 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
280 	 * default regardless of the V_ip6_auto_linklocal configuration to
281 	 * give a reasonable default behavior.
282 	 */
283 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
284 	    (ifp->if_flags & IFF_LOOPBACK))
285 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
286 	/*
287 	 * A loopback interface does not need to accept RTADV.
288 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
289 	 * default regardless of the V_ip6_accept_rtadv configuration to
290 	 * prevent the interface from accepting RA messages arrived
291 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
292 	 */
293 	if (V_ip6_accept_rtadv &&
294 	    !(ifp->if_flags & IFF_LOOPBACK) &&
295 	    (ifp->if_type != IFT_BRIDGE))
296 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
297 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
298 		nd->flags |= ND6_IFF_NO_RADR;
299 
300 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
301 	nd6_setmtu0(ifp, nd);
302 
303 	return nd;
304 }
305 
306 void
307 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
308 {
309 	struct epoch_tracker et;
310 	struct ifaddr *ifa, *next;
311 
312 	NET_EPOCH_ENTER(et);
313 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
314 		if (ifa->ifa_addr->sa_family != AF_INET6)
315 			continue;
316 
317 		/* stop DAD processing */
318 		nd6_dad_stop(ifa);
319 	}
320 	NET_EPOCH_EXIT(et);
321 
322 	free(nd, M_IP6NDP);
323 }
324 
325 /*
326  * Reset ND level link MTU. This function is called when the physical MTU
327  * changes, which means we might have to adjust the ND level MTU.
328  */
329 void
330 nd6_setmtu(struct ifnet *ifp)
331 {
332 	if (ifp->if_afdata[AF_INET6] == NULL)
333 		return;
334 
335 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
336 }
337 
338 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
339 void
340 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
341 {
342 	u_int32_t omaxmtu;
343 
344 	omaxmtu = ndi->maxmtu;
345 	ndi->maxmtu = ifp->if_mtu;
346 
347 	/*
348 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
349 	 * undesirable situation.  We thus notify the operator of the change
350 	 * explicitly.  The check for omaxmtu is necessary to restrict the
351 	 * log to the case of changing the MTU, not initializing it.
352 	 */
353 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
354 		log(LOG_NOTICE, "nd6_setmtu0: "
355 		    "new link MTU on %s (%lu) is too small for IPv6\n",
356 		    if_name(ifp), (unsigned long)ndi->maxmtu);
357 	}
358 
359 	if (ndi->maxmtu > V_in6_maxmtu)
360 		in6_setmaxmtu(); /* check all interfaces just in case */
361 
362 }
363 
364 void
365 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
366 {
367 
368 	bzero(ndopts, sizeof(*ndopts));
369 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
370 	ndopts->nd_opts_last
371 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
372 
373 	if (icmp6len == 0) {
374 		ndopts->nd_opts_done = 1;
375 		ndopts->nd_opts_search = NULL;
376 	}
377 }
378 
379 /*
380  * Take one ND option.
381  */
382 struct nd_opt_hdr *
383 nd6_option(union nd_opts *ndopts)
384 {
385 	struct nd_opt_hdr *nd_opt;
386 	int olen;
387 
388 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
389 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
390 	    __func__));
391 	if (ndopts->nd_opts_search == NULL)
392 		return NULL;
393 	if (ndopts->nd_opts_done)
394 		return NULL;
395 
396 	nd_opt = ndopts->nd_opts_search;
397 
398 	/* make sure nd_opt_len is inside the buffer */
399 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
400 		bzero(ndopts, sizeof(*ndopts));
401 		return NULL;
402 	}
403 
404 	olen = nd_opt->nd_opt_len << 3;
405 	if (olen == 0) {
406 		/*
407 		 * Message validation requires that all included
408 		 * options have a length that is greater than zero.
409 		 */
410 		bzero(ndopts, sizeof(*ndopts));
411 		return NULL;
412 	}
413 
414 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
415 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
416 		/* option overruns the end of buffer, invalid */
417 		bzero(ndopts, sizeof(*ndopts));
418 		return NULL;
419 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
420 		/* reached the end of options chain */
421 		ndopts->nd_opts_done = 1;
422 		ndopts->nd_opts_search = NULL;
423 	}
424 	return nd_opt;
425 }
426 
427 /*
428  * Parse multiple ND options.
429  * This function is much easier to use, for ND routines that do not need
430  * multiple options of the same type.
431  */
432 int
433 nd6_options(union nd_opts *ndopts)
434 {
435 	struct nd_opt_hdr *nd_opt;
436 	int i = 0;
437 
438 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
439 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
440 	    __func__));
441 	if (ndopts->nd_opts_search == NULL)
442 		return 0;
443 
444 	while (1) {
445 		nd_opt = nd6_option(ndopts);
446 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
447 			/*
448 			 * Message validation requires that all included
449 			 * options have a length that is greater than zero.
450 			 */
451 			ICMP6STAT_INC(icp6s_nd_badopt);
452 			bzero(ndopts, sizeof(*ndopts));
453 			return -1;
454 		}
455 
456 		if (nd_opt == NULL)
457 			goto skip1;
458 
459 		switch (nd_opt->nd_opt_type) {
460 		case ND_OPT_SOURCE_LINKADDR:
461 		case ND_OPT_TARGET_LINKADDR:
462 		case ND_OPT_MTU:
463 		case ND_OPT_REDIRECTED_HEADER:
464 		case ND_OPT_NONCE:
465 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
466 				nd6log((LOG_INFO,
467 				    "duplicated ND6 option found (type=%d)\n",
468 				    nd_opt->nd_opt_type));
469 				/* XXX bark? */
470 			} else {
471 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
472 					= nd_opt;
473 			}
474 			break;
475 		case ND_OPT_PREFIX_INFORMATION:
476 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
477 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
478 					= nd_opt;
479 			}
480 			ndopts->nd_opts_pi_end =
481 				(struct nd_opt_prefix_info *)nd_opt;
482 			break;
483 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
484 		case ND_OPT_RDNSS:	/* RFC 6106 */
485 		case ND_OPT_DNSSL:	/* RFC 6106 */
486 			/*
487 			 * Silently ignore options we know and do not care about
488 			 * in the kernel.
489 			 */
490 			break;
491 		default:
492 			/*
493 			 * Unknown options must be silently ignored,
494 			 * to accommodate future extension to the protocol.
495 			 */
496 			nd6log((LOG_DEBUG,
497 			    "nd6_options: unsupported option %d - "
498 			    "option ignored\n", nd_opt->nd_opt_type));
499 		}
500 
501 skip1:
502 		i++;
503 		if (i > V_nd6_maxndopt) {
504 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
505 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
506 			break;
507 		}
508 
509 		if (ndopts->nd_opts_done)
510 			break;
511 	}
512 
513 	return 0;
514 }
515 
516 /*
517  * ND6 timer routine to handle ND6 entries
518  */
519 static void
520 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
521 {
522 	int canceled;
523 
524 	LLE_WLOCK_ASSERT(ln);
525 
526 	if (tick < 0) {
527 		ln->la_expire = 0;
528 		ln->ln_ntick = 0;
529 		canceled = callout_stop(&ln->lle_timer);
530 	} else {
531 		ln->la_expire = time_uptime + tick / hz;
532 		LLE_ADDREF(ln);
533 		if (tick > INT_MAX) {
534 			ln->ln_ntick = tick - INT_MAX;
535 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
536 			    nd6_llinfo_timer, ln);
537 		} else {
538 			ln->ln_ntick = 0;
539 			canceled = callout_reset(&ln->lle_timer, tick,
540 			    nd6_llinfo_timer, ln);
541 		}
542 	}
543 	if (canceled > 0)
544 		LLE_REMREF(ln);
545 }
546 
547 /*
548  * Gets source address of the first packet in hold queue
549  * and stores it in @src.
550  * Returns pointer to @src (if hold queue is not empty) or NULL.
551  *
552  * Set noinline to be dtrace-friendly
553  */
554 static __noinline struct in6_addr *
555 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
556 {
557 	struct ip6_hdr hdr;
558 	struct mbuf *m;
559 
560 	if (ln->la_hold == NULL)
561 		return (NULL);
562 
563 	/*
564 	 * assume every packet in la_hold has the same IP header
565 	 */
566 	m = ln->la_hold;
567 	if (sizeof(hdr) > m->m_len)
568 		return (NULL);
569 
570 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
571 	*src = hdr.ip6_src;
572 
573 	return (src);
574 }
575 
576 /*
577  * Checks if we need to switch from STALE state.
578  *
579  * RFC 4861 requires switching from STALE to DELAY state
580  * on first packet matching entry, waiting V_nd6_delay and
581  * transition to PROBE state (if upper layer confirmation was
582  * not received).
583  *
584  * This code performs a bit differently:
585  * On packet hit we don't change state (but desired state
586  * can be guessed by control plane). However, after V_nd6_delay
587  * seconds code will transition to PROBE state (so DELAY state
588  * is kinda skipped in most situations).
589  *
590  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
591  * we perform the following upon entering STALE state:
592  *
593  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
594  * if packet was transmitted at the start of given interval, we
595  * would be able to switch to PROBE state in V_nd6_delay seconds
596  * as user expects.
597  *
598  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
599  * lle in STALE state (remaining timer value stored in lle_remtime).
600  *
601  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
602  * seconds ago.
603  *
604  * Returns non-zero value if the entry is still STALE (storing
605  * the next timer interval in @pdelay).
606  *
607  * Returns zero value if original timer expired or we need to switch to
608  * PROBE (store that in @do_switch variable).
609  */
610 static int
611 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
612 {
613 	int nd_delay, nd_gctimer, r_skip_req;
614 	time_t lle_hittime;
615 	long delay;
616 
617 	*do_switch = 0;
618 	nd_gctimer = V_nd6_gctimer;
619 	nd_delay = V_nd6_delay;
620 
621 	LLE_REQ_LOCK(lle);
622 	r_skip_req = lle->r_skip_req;
623 	lle_hittime = lle->lle_hittime;
624 	LLE_REQ_UNLOCK(lle);
625 
626 	if (r_skip_req > 0) {
627 
628 		/*
629 		 * Nonzero r_skip_req value was set upon entering
630 		 * STALE state. Since value was not changed, no
631 		 * packets were passed using this lle. Ask for
632 		 * timer reschedule and keep STALE state.
633 		 */
634 		delay = (long)(MIN(nd_gctimer, nd_delay));
635 		delay *= hz;
636 		if (lle->lle_remtime > delay)
637 			lle->lle_remtime -= delay;
638 		else {
639 			delay = lle->lle_remtime;
640 			lle->lle_remtime = 0;
641 		}
642 
643 		if (delay == 0) {
644 
645 			/*
646 			 * The original ng6_gctime timeout ended,
647 			 * no more rescheduling.
648 			 */
649 			return (0);
650 		}
651 
652 		*pdelay = delay;
653 		return (1);
654 	}
655 
656 	/*
657 	 * Packet received. Verify timestamp
658 	 */
659 	delay = (long)(time_uptime - lle_hittime);
660 	if (delay < nd_delay) {
661 
662 		/*
663 		 * V_nd6_delay still not passed since the first
664 		 * hit in STALE state.
665 		 * Reshedule timer and return.
666 		 */
667 		*pdelay = (long)(nd_delay - delay) * hz;
668 		return (1);
669 	}
670 
671 	/* Request switching to probe */
672 	*do_switch = 1;
673 	return (0);
674 }
675 
676 
677 /*
678  * Switch @lle state to new state optionally arming timers.
679  *
680  * Set noinline to be dtrace-friendly
681  */
682 __noinline void
683 nd6_llinfo_setstate(struct llentry *lle, int newstate)
684 {
685 	struct ifnet *ifp;
686 	int nd_gctimer, nd_delay;
687 	long delay, remtime;
688 
689 	delay = 0;
690 	remtime = 0;
691 
692 	switch (newstate) {
693 	case ND6_LLINFO_INCOMPLETE:
694 		ifp = lle->lle_tbl->llt_ifp;
695 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
696 		break;
697 	case ND6_LLINFO_REACHABLE:
698 		if (!ND6_LLINFO_PERMANENT(lle)) {
699 			ifp = lle->lle_tbl->llt_ifp;
700 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
701 		}
702 		break;
703 	case ND6_LLINFO_STALE:
704 
705 		/*
706 		 * Notify fast path that we want to know if any packet
707 		 * is transmitted by setting r_skip_req.
708 		 */
709 		LLE_REQ_LOCK(lle);
710 		lle->r_skip_req = 1;
711 		LLE_REQ_UNLOCK(lle);
712 		nd_delay = V_nd6_delay;
713 		nd_gctimer = V_nd6_gctimer;
714 
715 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
716 		remtime = (long)nd_gctimer * hz - delay;
717 		break;
718 	case ND6_LLINFO_DELAY:
719 		lle->la_asked = 0;
720 		delay = (long)V_nd6_delay * hz;
721 		break;
722 	}
723 
724 	if (delay > 0)
725 		nd6_llinfo_settimer_locked(lle, delay);
726 
727 	lle->lle_remtime = remtime;
728 	lle->ln_state = newstate;
729 }
730 
731 /*
732  * Timer-dependent part of nd state machine.
733  *
734  * Set noinline to be dtrace-friendly
735  */
736 static __noinline void
737 nd6_llinfo_timer(void *arg)
738 {
739 	struct epoch_tracker et;
740 	struct llentry *ln;
741 	struct in6_addr *dst, *pdst, *psrc, src;
742 	struct ifnet *ifp;
743 	struct nd_ifinfo *ndi;
744 	int do_switch, send_ns;
745 	long delay;
746 
747 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
748 	ln = (struct llentry *)arg;
749 	ifp = lltable_get_ifp(ln->lle_tbl);
750 	CURVNET_SET(ifp->if_vnet);
751 
752 	ND6_RLOCK();
753 	LLE_WLOCK(ln);
754 	if (callout_pending(&ln->lle_timer)) {
755 		/*
756 		 * Here we are a bit odd here in the treatment of
757 		 * active/pending. If the pending bit is set, it got
758 		 * rescheduled before I ran. The active
759 		 * bit we ignore, since if it was stopped
760 		 * in ll_tablefree() and was currently running
761 		 * it would have return 0 so the code would
762 		 * not have deleted it since the callout could
763 		 * not be stopped so we want to go through
764 		 * with the delete here now. If the callout
765 		 * was restarted, the pending bit will be back on and
766 		 * we just want to bail since the callout_reset would
767 		 * return 1 and our reference would have been removed
768 		 * by nd6_llinfo_settimer_locked above since canceled
769 		 * would have been 1.
770 		 */
771 		LLE_WUNLOCK(ln);
772 		ND6_RUNLOCK();
773 		CURVNET_RESTORE();
774 		return;
775 	}
776 	NET_EPOCH_ENTER(et);
777 	ndi = ND_IFINFO(ifp);
778 	send_ns = 0;
779 	dst = &ln->r_l3addr.addr6;
780 	pdst = dst;
781 
782 	if (ln->ln_ntick > 0) {
783 		if (ln->ln_ntick > INT_MAX) {
784 			ln->ln_ntick -= INT_MAX;
785 			nd6_llinfo_settimer_locked(ln, INT_MAX);
786 		} else {
787 			ln->ln_ntick = 0;
788 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
789 		}
790 		goto done;
791 	}
792 
793 	if (ln->la_flags & LLE_STATIC) {
794 		goto done;
795 	}
796 
797 	if (ln->la_flags & LLE_DELETED) {
798 		nd6_free(&ln, 0);
799 		goto done;
800 	}
801 
802 	switch (ln->ln_state) {
803 	case ND6_LLINFO_INCOMPLETE:
804 		if (ln->la_asked < V_nd6_mmaxtries) {
805 			ln->la_asked++;
806 			send_ns = 1;
807 			/* Send NS to multicast address */
808 			pdst = NULL;
809 		} else {
810 			struct mbuf *m = ln->la_hold;
811 			if (m) {
812 				struct mbuf *m0;
813 
814 				/*
815 				 * assuming every packet in la_hold has the
816 				 * same IP header.  Send error after unlock.
817 				 */
818 				m0 = m->m_nextpkt;
819 				m->m_nextpkt = NULL;
820 				ln->la_hold = m0;
821 				clear_llinfo_pqueue(ln);
822 			}
823 			nd6_free(&ln, 0);
824 			if (m != NULL) {
825 				struct mbuf *n = m;
826 
827 				/*
828 				 * if there are any ummapped mbufs, we
829 				 * must free them, rather than using
830 				 * them for an ICMP, as they cannot be
831 				 * checksummed.
832 				 */
833 				while ((n = n->m_next) != NULL) {
834 					if (n->m_flags & M_EXTPG)
835 						break;
836 				}
837 				if (n != NULL) {
838 					m_freem(m);
839 					m = NULL;
840 				} else {
841 					icmp6_error2(m, ICMP6_DST_UNREACH,
842 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
843 				}
844 			}
845 		}
846 		break;
847 	case ND6_LLINFO_REACHABLE:
848 		if (!ND6_LLINFO_PERMANENT(ln))
849 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
850 		break;
851 
852 	case ND6_LLINFO_STALE:
853 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
854 
855 			/*
856 			 * No packet has used this entry and GC timeout
857 			 * has not been passed. Reshedule timer and
858 			 * return.
859 			 */
860 			nd6_llinfo_settimer_locked(ln, delay);
861 			break;
862 		}
863 
864 		if (do_switch == 0) {
865 
866 			/*
867 			 * GC timer has ended and entry hasn't been used.
868 			 * Run Garbage collector (RFC 4861, 5.3)
869 			 */
870 			if (!ND6_LLINFO_PERMANENT(ln))
871 				nd6_free(&ln, 1);
872 			break;
873 		}
874 
875 		/* Entry has been used AND delay timer has ended. */
876 
877 		/* FALLTHROUGH */
878 
879 	case ND6_LLINFO_DELAY:
880 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
881 			/* We need NUD */
882 			ln->la_asked = 1;
883 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
884 			send_ns = 1;
885 		} else
886 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
887 		break;
888 	case ND6_LLINFO_PROBE:
889 		if (ln->la_asked < V_nd6_umaxtries) {
890 			ln->la_asked++;
891 			send_ns = 1;
892 		} else {
893 			nd6_free(&ln, 0);
894 		}
895 		break;
896 	default:
897 		panic("%s: paths in a dark night can be confusing: %d",
898 		    __func__, ln->ln_state);
899 	}
900 done:
901 	if (ln != NULL)
902 		ND6_RUNLOCK();
903 	if (send_ns != 0) {
904 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
905 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
906 		LLE_FREE_LOCKED(ln);
907 		ln = NULL;
908 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
909 	}
910 
911 	if (ln != NULL)
912 		LLE_FREE_LOCKED(ln);
913 	NET_EPOCH_EXIT(et);
914 	CURVNET_RESTORE();
915 }
916 
917 
918 /*
919  * ND6 timer routine to expire default route list and prefix list
920  */
921 void
922 nd6_timer(void *arg)
923 {
924 	CURVNET_SET((struct vnet *) arg);
925 	struct epoch_tracker et;
926 	struct nd_prhead prl;
927 	struct nd_prefix *pr, *npr;
928 	struct ifnet *ifp;
929 	struct in6_ifaddr *ia6, *nia6;
930 	uint64_t genid;
931 
932 	LIST_INIT(&prl);
933 
934 	NET_EPOCH_ENTER(et);
935 	nd6_defrouter_timer();
936 
937 	/*
938 	 * expire interface addresses.
939 	 * in the past the loop was inside prefix expiry processing.
940 	 * However, from a stricter speci-confrmance standpoint, we should
941 	 * rather separate address lifetimes and prefix lifetimes.
942 	 *
943 	 * XXXRW: in6_ifaddrhead locking.
944 	 */
945   addrloop:
946 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
947 		/* check address lifetime */
948 		if (IFA6_IS_INVALID(ia6)) {
949 			int regen = 0;
950 
951 			/*
952 			 * If the expiring address is temporary, try
953 			 * regenerating a new one.  This would be useful when
954 			 * we suspended a laptop PC, then turned it on after a
955 			 * period that could invalidate all temporary
956 			 * addresses.  Although we may have to restart the
957 			 * loop (see below), it must be after purging the
958 			 * address.  Otherwise, we'd see an infinite loop of
959 			 * regeneration.
960 			 */
961 			if (V_ip6_use_tempaddr &&
962 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
963 				if (regen_tmpaddr(ia6) == 0)
964 					regen = 1;
965 			}
966 
967 			in6_purgeaddr(&ia6->ia_ifa);
968 
969 			if (regen)
970 				goto addrloop; /* XXX: see below */
971 		} else if (IFA6_IS_DEPRECATED(ia6)) {
972 			int oldflags = ia6->ia6_flags;
973 
974 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
975 
976 			/*
977 			 * If a temporary address has just become deprecated,
978 			 * regenerate a new one if possible.
979 			 */
980 			if (V_ip6_use_tempaddr &&
981 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
982 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
983 
984 				if (regen_tmpaddr(ia6) == 0) {
985 					/*
986 					 * A new temporary address is
987 					 * generated.
988 					 * XXX: this means the address chain
989 					 * has changed while we are still in
990 					 * the loop.  Although the change
991 					 * would not cause disaster (because
992 					 * it's not a deletion, but an
993 					 * addition,) we'd rather restart the
994 					 * loop just for safety.  Or does this
995 					 * significantly reduce performance??
996 					 */
997 					goto addrloop;
998 				}
999 			}
1000 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1001 			/*
1002 			 * Schedule DAD for a tentative address.  This happens
1003 			 * if the interface was down or not running
1004 			 * when the address was configured.
1005 			 */
1006 			int delay;
1007 
1008 			delay = arc4random() %
1009 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1010 			nd6_dad_start((struct ifaddr *)ia6, delay);
1011 		} else {
1012 			/*
1013 			 * Check status of the interface.  If it is down,
1014 			 * mark the address as tentative for future DAD.
1015 			 */
1016 			ifp = ia6->ia_ifp;
1017 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1018 			    ((ifp->if_flags & IFF_UP) == 0 ||
1019 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1020 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1021 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1022 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1023 			}
1024 
1025 			/*
1026 			 * A new RA might have made a deprecated address
1027 			 * preferred.
1028 			 */
1029 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1030 		}
1031 	}
1032 	NET_EPOCH_EXIT(et);
1033 
1034 	ND6_WLOCK();
1035 restart:
1036 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1037 		/*
1038 		 * Expire prefixes. Since the pltime is only used for
1039 		 * autoconfigured addresses, pltime processing for prefixes is
1040 		 * not necessary.
1041 		 *
1042 		 * Only unlink after all derived addresses have expired. This
1043 		 * may not occur until two hours after the prefix has expired
1044 		 * per RFC 4862. If the prefix expires before its derived
1045 		 * addresses, mark it off-link. This will be done automatically
1046 		 * after unlinking if no address references remain.
1047 		 */
1048 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1049 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1050 			continue;
1051 
1052 		if (pr->ndpr_addrcnt == 0) {
1053 			nd6_prefix_unlink(pr, &prl);
1054 			continue;
1055 		}
1056 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1057 			genid = V_nd6_list_genid;
1058 			nd6_prefix_ref(pr);
1059 			ND6_WUNLOCK();
1060 			ND6_ONLINK_LOCK();
1061 			(void)nd6_prefix_offlink(pr);
1062 			ND6_ONLINK_UNLOCK();
1063 			ND6_WLOCK();
1064 			nd6_prefix_rele(pr);
1065 			if (genid != V_nd6_list_genid)
1066 				goto restart;
1067 		}
1068 	}
1069 	ND6_WUNLOCK();
1070 
1071 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1072 		LIST_REMOVE(pr, ndpr_entry);
1073 		nd6_prefix_del(pr);
1074 	}
1075 
1076 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1077 	    nd6_timer, curvnet);
1078 
1079 	CURVNET_RESTORE();
1080 }
1081 
1082 /*
1083  * ia6 - deprecated/invalidated temporary address
1084  */
1085 static int
1086 regen_tmpaddr(struct in6_ifaddr *ia6)
1087 {
1088 	struct ifaddr *ifa;
1089 	struct ifnet *ifp;
1090 	struct in6_ifaddr *public_ifa6 = NULL;
1091 
1092 	NET_EPOCH_ASSERT();
1093 
1094 	ifp = ia6->ia_ifa.ifa_ifp;
1095 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1096 		struct in6_ifaddr *it6;
1097 
1098 		if (ifa->ifa_addr->sa_family != AF_INET6)
1099 			continue;
1100 
1101 		it6 = (struct in6_ifaddr *)ifa;
1102 
1103 		/* ignore no autoconf addresses. */
1104 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1105 			continue;
1106 
1107 		/* ignore autoconf addresses with different prefixes. */
1108 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1109 			continue;
1110 
1111 		/*
1112 		 * Now we are looking at an autoconf address with the same
1113 		 * prefix as ours.  If the address is temporary and is still
1114 		 * preferred, do not create another one.  It would be rare, but
1115 		 * could happen, for example, when we resume a laptop PC after
1116 		 * a long period.
1117 		 */
1118 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1119 		    !IFA6_IS_DEPRECATED(it6)) {
1120 			public_ifa6 = NULL;
1121 			break;
1122 		}
1123 
1124 		/*
1125 		 * This is a public autoconf address that has the same prefix
1126 		 * as ours.  If it is preferred, keep it.  We can't break the
1127 		 * loop here, because there may be a still-preferred temporary
1128 		 * address with the prefix.
1129 		 */
1130 		if (!IFA6_IS_DEPRECATED(it6))
1131 			public_ifa6 = it6;
1132 	}
1133 	if (public_ifa6 != NULL)
1134 		ifa_ref(&public_ifa6->ia_ifa);
1135 
1136 	if (public_ifa6 != NULL) {
1137 		int e;
1138 
1139 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1140 			ifa_free(&public_ifa6->ia_ifa);
1141 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1142 			    " tmp addr,errno=%d\n", e);
1143 			return (-1);
1144 		}
1145 		ifa_free(&public_ifa6->ia_ifa);
1146 		return (0);
1147 	}
1148 
1149 	return (-1);
1150 }
1151 
1152 /*
1153  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1154  * cache entries are freed in in6_domifdetach().
1155  */
1156 void
1157 nd6_purge(struct ifnet *ifp)
1158 {
1159 	struct nd_prhead prl;
1160 	struct nd_prefix *pr, *npr;
1161 
1162 	LIST_INIT(&prl);
1163 
1164 	/* Purge default router list entries toward ifp. */
1165 	nd6_defrouter_purge(ifp);
1166 
1167 	ND6_WLOCK();
1168 	/*
1169 	 * Remove prefixes on ifp. We should have already removed addresses on
1170 	 * this interface, so no addresses should be referencing these prefixes.
1171 	 */
1172 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1173 		if (pr->ndpr_ifp == ifp)
1174 			nd6_prefix_unlink(pr, &prl);
1175 	}
1176 	ND6_WUNLOCK();
1177 
1178 	/* Delete the unlinked prefix objects. */
1179 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1180 		LIST_REMOVE(pr, ndpr_entry);
1181 		nd6_prefix_del(pr);
1182 	}
1183 
1184 	/* cancel default outgoing interface setting */
1185 	if (V_nd6_defifindex == ifp->if_index)
1186 		nd6_setdefaultiface(0);
1187 
1188 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1189 		/* Refresh default router list. */
1190 		defrouter_select_fib(ifp->if_fib);
1191 	}
1192 }
1193 
1194 /*
1195  * the caller acquires and releases the lock on the lltbls
1196  * Returns the llentry locked
1197  */
1198 struct llentry *
1199 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1200 {
1201 	struct sockaddr_in6 sin6;
1202 	struct llentry *ln;
1203 
1204 	bzero(&sin6, sizeof(sin6));
1205 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1206 	sin6.sin6_family = AF_INET6;
1207 	sin6.sin6_addr = *addr6;
1208 
1209 	IF_AFDATA_LOCK_ASSERT(ifp);
1210 
1211 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1212 
1213 	return (ln);
1214 }
1215 
1216 static struct llentry *
1217 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1218 {
1219 	struct sockaddr_in6 sin6;
1220 	struct llentry *ln;
1221 
1222 	bzero(&sin6, sizeof(sin6));
1223 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1224 	sin6.sin6_family = AF_INET6;
1225 	sin6.sin6_addr = *addr6;
1226 
1227 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1228 	if (ln != NULL)
1229 		ln->ln_state = ND6_LLINFO_NOSTATE;
1230 
1231 	return (ln);
1232 }
1233 
1234 /*
1235  * Test whether a given IPv6 address is a neighbor or not, ignoring
1236  * the actual neighbor cache.  The neighbor cache is ignored in order
1237  * to not reenter the routing code from within itself.
1238  */
1239 static int
1240 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1241 {
1242 	struct nd_prefix *pr;
1243 	struct ifaddr *ifa;
1244 	struct rt_addrinfo info;
1245 	struct sockaddr_in6 rt_key;
1246 	const struct sockaddr *dst6;
1247 	uint64_t genid;
1248 	int error, fibnum;
1249 
1250 	/*
1251 	 * A link-local address is always a neighbor.
1252 	 * XXX: a link does not necessarily specify a single interface.
1253 	 */
1254 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1255 		struct sockaddr_in6 sin6_copy;
1256 		u_int32_t zone;
1257 
1258 		/*
1259 		 * We need sin6_copy since sa6_recoverscope() may modify the
1260 		 * content (XXX).
1261 		 */
1262 		sin6_copy = *addr;
1263 		if (sa6_recoverscope(&sin6_copy))
1264 			return (0); /* XXX: should be impossible */
1265 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1266 			return (0);
1267 		if (sin6_copy.sin6_scope_id == zone)
1268 			return (1);
1269 		else
1270 			return (0);
1271 	}
1272 
1273 	bzero(&rt_key, sizeof(rt_key));
1274 	bzero(&info, sizeof(info));
1275 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1276 
1277 	/*
1278 	 * If the address matches one of our addresses,
1279 	 * it should be a neighbor.
1280 	 * If the address matches one of our on-link prefixes, it should be a
1281 	 * neighbor.
1282 	 */
1283 	ND6_RLOCK();
1284 restart:
1285 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1286 		if (pr->ndpr_ifp != ifp)
1287 			continue;
1288 
1289 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1290 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1291 
1292 			/*
1293 			 * We only need to check all FIBs if add_addr_allfibs
1294 			 * is unset. If set, checking any FIB will suffice.
1295 			 */
1296 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1297 			for (; fibnum < rt_numfibs; fibnum++) {
1298 				genid = V_nd6_list_genid;
1299 				ND6_RUNLOCK();
1300 
1301 				/*
1302 				 * Restore length field before
1303 				 * retrying lookup
1304 				 */
1305 				rt_key.sin6_len = sizeof(rt_key);
1306 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1307 						        &info);
1308 
1309 				ND6_RLOCK();
1310 				if (genid != V_nd6_list_genid)
1311 					goto restart;
1312 				if (error == 0)
1313 					break;
1314 			}
1315 			if (error != 0)
1316 				continue;
1317 
1318 			/*
1319 			 * This is the case where multiple interfaces
1320 			 * have the same prefix, but only one is installed
1321 			 * into the routing table and that prefix entry
1322 			 * is not the one being examined here. In the case
1323 			 * where RADIX_MPATH is enabled, multiple route
1324 			 * entries (of the same rt_key value) will be
1325 			 * installed because the interface addresses all
1326 			 * differ.
1327 			 */
1328 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1329 			    &rt_key.sin6_addr))
1330 				continue;
1331 		}
1332 
1333 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1334 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1335 			ND6_RUNLOCK();
1336 			return (1);
1337 		}
1338 	}
1339 	ND6_RUNLOCK();
1340 
1341 	/*
1342 	 * If the address is assigned on the node of the other side of
1343 	 * a p2p interface, the address should be a neighbor.
1344 	 */
1345 	if (ifp->if_flags & IFF_POINTOPOINT) {
1346 		struct epoch_tracker et;
1347 
1348 		NET_EPOCH_ENTER(et);
1349 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1350 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1351 				continue;
1352 			if (ifa->ifa_dstaddr != NULL &&
1353 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1354 				NET_EPOCH_EXIT(et);
1355 				return 1;
1356 			}
1357 		}
1358 		NET_EPOCH_EXIT(et);
1359 	}
1360 
1361 	/*
1362 	 * If the default router list is empty, all addresses are regarded
1363 	 * as on-link, and thus, as a neighbor.
1364 	 */
1365 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1366 	    nd6_defrouter_list_empty() &&
1367 	    V_nd6_defifindex == ifp->if_index) {
1368 		return (1);
1369 	}
1370 
1371 	return (0);
1372 }
1373 
1374 
1375 /*
1376  * Detect if a given IPv6 address identifies a neighbor on a given link.
1377  * XXX: should take care of the destination of a p2p link?
1378  */
1379 int
1380 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1381 {
1382 	struct llentry *lle;
1383 	int rc = 0;
1384 
1385 	NET_EPOCH_ASSERT();
1386 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1387 	if (nd6_is_new_addr_neighbor(addr, ifp))
1388 		return (1);
1389 
1390 	/*
1391 	 * Even if the address matches none of our addresses, it might be
1392 	 * in the neighbor cache.
1393 	 */
1394 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1395 		LLE_RUNLOCK(lle);
1396 		rc = 1;
1397 	}
1398 	return (rc);
1399 }
1400 
1401 /*
1402  * Free an nd6 llinfo entry.
1403  * Since the function would cause significant changes in the kernel, DO NOT
1404  * make it global, unless you have a strong reason for the change, and are sure
1405  * that the change is safe.
1406  *
1407  * Set noinline to be dtrace-friendly
1408  */
1409 static __noinline void
1410 nd6_free(struct llentry **lnp, int gc)
1411 {
1412 	struct ifnet *ifp;
1413 	struct llentry *ln;
1414 	struct nd_defrouter *dr;
1415 
1416 	ln = *lnp;
1417 	*lnp = NULL;
1418 
1419 	LLE_WLOCK_ASSERT(ln);
1420 	ND6_RLOCK_ASSERT();
1421 
1422 	ifp = lltable_get_ifp(ln->lle_tbl);
1423 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1424 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1425 	else
1426 		dr = NULL;
1427 	ND6_RUNLOCK();
1428 
1429 	if ((ln->la_flags & LLE_DELETED) == 0)
1430 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1431 
1432 	/*
1433 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1434 	 * even though it is not harmful, it was not really necessary.
1435 	 */
1436 
1437 	/* cancel timer */
1438 	nd6_llinfo_settimer_locked(ln, -1);
1439 
1440 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1441 		if (dr != NULL && dr->expire &&
1442 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1443 			/*
1444 			 * If the reason for the deletion is just garbage
1445 			 * collection, and the neighbor is an active default
1446 			 * router, do not delete it.  Instead, reset the GC
1447 			 * timer using the router's lifetime.
1448 			 * Simply deleting the entry would affect default
1449 			 * router selection, which is not necessarily a good
1450 			 * thing, especially when we're using router preference
1451 			 * values.
1452 			 * XXX: the check for ln_state would be redundant,
1453 			 *      but we intentionally keep it just in case.
1454 			 */
1455 			if (dr->expire > time_uptime)
1456 				nd6_llinfo_settimer_locked(ln,
1457 				    (dr->expire - time_uptime) * hz);
1458 			else
1459 				nd6_llinfo_settimer_locked(ln,
1460 				    (long)V_nd6_gctimer * hz);
1461 
1462 			LLE_REMREF(ln);
1463 			LLE_WUNLOCK(ln);
1464 			defrouter_rele(dr);
1465 			return;
1466 		}
1467 
1468 		if (dr) {
1469 			/*
1470 			 * Unreachablity of a router might affect the default
1471 			 * router selection and on-link detection of advertised
1472 			 * prefixes.
1473 			 */
1474 
1475 			/*
1476 			 * Temporarily fake the state to choose a new default
1477 			 * router and to perform on-link determination of
1478 			 * prefixes correctly.
1479 			 * Below the state will be set correctly,
1480 			 * or the entry itself will be deleted.
1481 			 */
1482 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1483 		}
1484 
1485 		if (ln->ln_router || dr) {
1486 
1487 			/*
1488 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1489 			 * rnh and for the calls to pfxlist_onlink_check() and
1490 			 * defrouter_select_fib() in the block further down for calls
1491 			 * into nd6_lookup().  We still hold a ref.
1492 			 */
1493 			LLE_WUNLOCK(ln);
1494 
1495 			/*
1496 			 * rt6_flush must be called whether or not the neighbor
1497 			 * is in the Default Router List.
1498 			 * See a corresponding comment in nd6_na_input().
1499 			 */
1500 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1501 		}
1502 
1503 		if (dr) {
1504 			/*
1505 			 * Since defrouter_select_fib() does not affect the
1506 			 * on-link determination and MIP6 needs the check
1507 			 * before the default router selection, we perform
1508 			 * the check now.
1509 			 */
1510 			pfxlist_onlink_check();
1511 
1512 			/*
1513 			 * Refresh default router list.
1514 			 */
1515 			defrouter_select_fib(dr->ifp->if_fib);
1516 		}
1517 
1518 		/*
1519 		 * If this entry was added by an on-link redirect, remove the
1520 		 * corresponding host route.
1521 		 */
1522 		if (ln->la_flags & LLE_REDIRECT)
1523 			nd6_free_redirect(ln);
1524 
1525 		if (ln->ln_router || dr)
1526 			LLE_WLOCK(ln);
1527 	}
1528 
1529 	/*
1530 	 * Save to unlock. We still hold an extra reference and will not
1531 	 * free(9) in llentry_free() if someone else holds one as well.
1532 	 */
1533 	LLE_WUNLOCK(ln);
1534 	IF_AFDATA_LOCK(ifp);
1535 	LLE_WLOCK(ln);
1536 	/* Guard against race with other llentry_free(). */
1537 	if (ln->la_flags & LLE_LINKED) {
1538 		/* Remove callout reference */
1539 		LLE_REMREF(ln);
1540 		lltable_unlink_entry(ln->lle_tbl, ln);
1541 	}
1542 	IF_AFDATA_UNLOCK(ifp);
1543 
1544 	llentry_free(ln);
1545 	if (dr != NULL)
1546 		defrouter_rele(dr);
1547 }
1548 
1549 static int
1550 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1551 {
1552 
1553 	if (nh->nh_flags & NHF_REDIRECT)
1554 		return (1);
1555 
1556 	return (0);
1557 }
1558 
1559 /*
1560  * Remove the rtentry for the given llentry,
1561  * both of which were installed by a redirect.
1562  */
1563 static void
1564 nd6_free_redirect(const struct llentry *ln)
1565 {
1566 	int fibnum;
1567 	struct sockaddr_in6 sin6;
1568 	struct rt_addrinfo info;
1569 
1570 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1571 	memset(&info, 0, sizeof(info));
1572 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1573 	info.rti_filter = nd6_isdynrte;
1574 
1575 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1576 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1577 }
1578 
1579 /*
1580  * Rejuvenate this function for routing operations related
1581  * processing.
1582  */
1583 void
1584 nd6_rtrequest(int req, struct rtentry *rt, struct nhop_object *nh,
1585     struct rt_addrinfo *info)
1586 {
1587 	struct sockaddr_in6 *gateway;
1588 	struct nd_defrouter *dr;
1589 
1590 	gateway = &nh->gw6_sa;
1591 
1592 	switch (req) {
1593 	case RTM_ADD:
1594 		break;
1595 
1596 	case RTM_DELETE:
1597 		/*
1598 		 * Only indirect routes are interesting.
1599 		 */
1600 		if ((nh->nh_flags & NHF_GATEWAY) == 0)
1601 			return;
1602 		/*
1603 		 * check for default route
1604 		 */
1605 		if (nh->nh_flags & NHF_DEFAULT) {
1606 			dr = defrouter_lookup(&gateway->sin6_addr, nh->nh_ifp);
1607 			if (dr != NULL) {
1608 				dr->installed = 0;
1609 				defrouter_rele(dr);
1610 			}
1611 		}
1612 		break;
1613 	}
1614 }
1615 
1616 
1617 int
1618 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1619 {
1620 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1621 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1622 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1623 	struct epoch_tracker et;
1624 	int error = 0;
1625 
1626 	if (ifp->if_afdata[AF_INET6] == NULL)
1627 		return (EPFNOSUPPORT);
1628 	switch (cmd) {
1629 	case OSIOCGIFINFO_IN6:
1630 #define ND	ndi->ndi
1631 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1632 		bzero(&ND, sizeof(ND));
1633 		ND.linkmtu = IN6_LINKMTU(ifp);
1634 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1635 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1636 		ND.reachable = ND_IFINFO(ifp)->reachable;
1637 		ND.retrans = ND_IFINFO(ifp)->retrans;
1638 		ND.flags = ND_IFINFO(ifp)->flags;
1639 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1640 		ND.chlim = ND_IFINFO(ifp)->chlim;
1641 		break;
1642 	case SIOCGIFINFO_IN6:
1643 		ND = *ND_IFINFO(ifp);
1644 		break;
1645 	case SIOCSIFINFO_IN6:
1646 		/*
1647 		 * used to change host variables from userland.
1648 		 * intended for a use on router to reflect RA configurations.
1649 		 */
1650 		/* 0 means 'unspecified' */
1651 		if (ND.linkmtu != 0) {
1652 			if (ND.linkmtu < IPV6_MMTU ||
1653 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1654 				error = EINVAL;
1655 				break;
1656 			}
1657 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1658 		}
1659 
1660 		if (ND.basereachable != 0) {
1661 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1662 
1663 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1664 			if (ND.basereachable != obasereachable)
1665 				ND_IFINFO(ifp)->reachable =
1666 				    ND_COMPUTE_RTIME(ND.basereachable);
1667 		}
1668 		if (ND.retrans != 0)
1669 			ND_IFINFO(ifp)->retrans = ND.retrans;
1670 		if (ND.chlim != 0)
1671 			ND_IFINFO(ifp)->chlim = ND.chlim;
1672 		/* FALLTHROUGH */
1673 	case SIOCSIFINFO_FLAGS:
1674 	{
1675 		struct ifaddr *ifa;
1676 		struct in6_ifaddr *ia;
1677 
1678 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1679 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1680 			/* ifdisabled 1->0 transision */
1681 
1682 			/*
1683 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1684 			 * has an link-local address with IN6_IFF_DUPLICATED,
1685 			 * do not clear ND6_IFF_IFDISABLED.
1686 			 * See RFC 4862, Section 5.4.5.
1687 			 */
1688 			NET_EPOCH_ENTER(et);
1689 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1690 				if (ifa->ifa_addr->sa_family != AF_INET6)
1691 					continue;
1692 				ia = (struct in6_ifaddr *)ifa;
1693 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1694 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1695 					break;
1696 			}
1697 			NET_EPOCH_EXIT(et);
1698 
1699 			if (ifa != NULL) {
1700 				/* LLA is duplicated. */
1701 				ND.flags |= ND6_IFF_IFDISABLED;
1702 				log(LOG_ERR, "Cannot enable an interface"
1703 				    " with a link-local address marked"
1704 				    " duplicate.\n");
1705 			} else {
1706 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1707 				if (ifp->if_flags & IFF_UP)
1708 					in6_if_up(ifp);
1709 			}
1710 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1711 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1712 			/* ifdisabled 0->1 transision */
1713 			/* Mark all IPv6 address as tentative. */
1714 
1715 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1716 			if (V_ip6_dad_count > 0 &&
1717 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1718 				NET_EPOCH_ENTER(et);
1719 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1720 				    ifa_link) {
1721 					if (ifa->ifa_addr->sa_family !=
1722 					    AF_INET6)
1723 						continue;
1724 					ia = (struct in6_ifaddr *)ifa;
1725 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1726 				}
1727 				NET_EPOCH_EXIT(et);
1728 			}
1729 		}
1730 
1731 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1732 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1733 				/* auto_linklocal 0->1 transision */
1734 
1735 				/* If no link-local address on ifp, configure */
1736 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1737 				in6_ifattach(ifp, NULL);
1738 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1739 			    ifp->if_flags & IFF_UP) {
1740 				/*
1741 				 * When the IF already has
1742 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1743 				 * address is assigned, and IFF_UP, try to
1744 				 * assign one.
1745 				 */
1746 				NET_EPOCH_ENTER(et);
1747 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1748 				    ifa_link) {
1749 					if (ifa->ifa_addr->sa_family !=
1750 					    AF_INET6)
1751 						continue;
1752 					ia = (struct in6_ifaddr *)ifa;
1753 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1754 						break;
1755 				}
1756 				NET_EPOCH_EXIT(et);
1757 				if (ifa != NULL)
1758 					/* No LLA is configured. */
1759 					in6_ifattach(ifp, NULL);
1760 			}
1761 		}
1762 		ND_IFINFO(ifp)->flags = ND.flags;
1763 		break;
1764 	}
1765 #undef ND
1766 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1767 		/* sync kernel routing table with the default router list */
1768 		defrouter_reset();
1769 		defrouter_select_fib(RT_ALL_FIBS);
1770 		break;
1771 	case SIOCSPFXFLUSH_IN6:
1772 	{
1773 		/* flush all the prefix advertised by routers */
1774 		struct in6_ifaddr *ia, *ia_next;
1775 		struct nd_prefix *pr, *next;
1776 		struct nd_prhead prl;
1777 
1778 		LIST_INIT(&prl);
1779 
1780 		ND6_WLOCK();
1781 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1782 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1783 				continue; /* XXX */
1784 			nd6_prefix_unlink(pr, &prl);
1785 		}
1786 		ND6_WUNLOCK();
1787 
1788 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1789 			LIST_REMOVE(pr, ndpr_entry);
1790 			/* XXXRW: in6_ifaddrhead locking. */
1791 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1792 			    ia_next) {
1793 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1794 					continue;
1795 
1796 				if (ia->ia6_ndpr == pr)
1797 					in6_purgeaddr(&ia->ia_ifa);
1798 			}
1799 			nd6_prefix_del(pr);
1800 		}
1801 		break;
1802 	}
1803 	case SIOCSRTRFLUSH_IN6:
1804 	{
1805 		/* flush all the default routers */
1806 
1807 		defrouter_reset();
1808 		nd6_defrouter_flush_all();
1809 		defrouter_select_fib(RT_ALL_FIBS);
1810 		break;
1811 	}
1812 	case SIOCGNBRINFO_IN6:
1813 	{
1814 		struct llentry *ln;
1815 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1816 
1817 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1818 			return (error);
1819 
1820 		NET_EPOCH_ENTER(et);
1821 		ln = nd6_lookup(&nb_addr, 0, ifp);
1822 		NET_EPOCH_EXIT(et);
1823 
1824 		if (ln == NULL) {
1825 			error = EINVAL;
1826 			break;
1827 		}
1828 		nbi->state = ln->ln_state;
1829 		nbi->asked = ln->la_asked;
1830 		nbi->isrouter = ln->ln_router;
1831 		if (ln->la_expire == 0)
1832 			nbi->expire = 0;
1833 		else
1834 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1835 			    (time_second - time_uptime);
1836 		LLE_RUNLOCK(ln);
1837 		break;
1838 	}
1839 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1840 		ndif->ifindex = V_nd6_defifindex;
1841 		break;
1842 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1843 		return (nd6_setdefaultiface(ndif->ifindex));
1844 	}
1845 	return (error);
1846 }
1847 
1848 /*
1849  * Calculates new isRouter value based on provided parameters and
1850  * returns it.
1851  */
1852 static int
1853 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1854     int ln_router)
1855 {
1856 
1857 	/*
1858 	 * ICMP6 type dependent behavior.
1859 	 *
1860 	 * NS: clear IsRouter if new entry
1861 	 * RS: clear IsRouter
1862 	 * RA: set IsRouter if there's lladdr
1863 	 * redir: clear IsRouter if new entry
1864 	 *
1865 	 * RA case, (1):
1866 	 * The spec says that we must set IsRouter in the following cases:
1867 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1868 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1869 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1870 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1871 	 * neighbor cache, this is similar to (6).
1872 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1873 	 *
1874 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1875 	 *							D R
1876 	 *	0	n	n	(1)	c   ?     s
1877 	 *	0	y	n	(2)	c   s     s
1878 	 *	0	n	y	(3)	c   s     s
1879 	 *	0	y	y	(4)	c   s     s
1880 	 *	0	y	y	(5)	c   s     s
1881 	 *	1	--	n	(6) c	c	c s
1882 	 *	1	--	y	(7) c	c   s	c s
1883 	 *
1884 	 *					(c=clear s=set)
1885 	 */
1886 	switch (type & 0xff) {
1887 	case ND_NEIGHBOR_SOLICIT:
1888 		/*
1889 		 * New entry must have is_router flag cleared.
1890 		 */
1891 		if (is_new)					/* (6-7) */
1892 			ln_router = 0;
1893 		break;
1894 	case ND_REDIRECT:
1895 		/*
1896 		 * If the icmp is a redirect to a better router, always set the
1897 		 * is_router flag.  Otherwise, if the entry is newly created,
1898 		 * clear the flag.  [RFC 2461, sec 8.3]
1899 		 */
1900 		if (code == ND_REDIRECT_ROUTER)
1901 			ln_router = 1;
1902 		else {
1903 			if (is_new)				/* (6-7) */
1904 				ln_router = 0;
1905 		}
1906 		break;
1907 	case ND_ROUTER_SOLICIT:
1908 		/*
1909 		 * is_router flag must always be cleared.
1910 		 */
1911 		ln_router = 0;
1912 		break;
1913 	case ND_ROUTER_ADVERT:
1914 		/*
1915 		 * Mark an entry with lladdr as a router.
1916 		 */
1917 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1918 		    (is_new && new_addr)) {			/* (7) */
1919 			ln_router = 1;
1920 		}
1921 		break;
1922 	}
1923 
1924 	return (ln_router);
1925 }
1926 
1927 /*
1928  * Create neighbor cache entry and cache link-layer address,
1929  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1930  *
1931  * type - ICMP6 type
1932  * code - type dependent information
1933  *
1934  */
1935 void
1936 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1937     int lladdrlen, int type, int code)
1938 {
1939 	struct llentry *ln = NULL, *ln_tmp;
1940 	int is_newentry;
1941 	int do_update;
1942 	int olladdr;
1943 	int llchange;
1944 	int flags;
1945 	uint16_t router = 0;
1946 	struct sockaddr_in6 sin6;
1947 	struct mbuf *chain = NULL;
1948 	u_char linkhdr[LLE_MAX_LINKHDR];
1949 	size_t linkhdrsize;
1950 	int lladdr_off;
1951 
1952 	NET_EPOCH_ASSERT();
1953 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1954 
1955 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1956 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1957 
1958 	/* nothing must be updated for unspecified address */
1959 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1960 		return;
1961 
1962 	/*
1963 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1964 	 * the caller.
1965 	 *
1966 	 * XXX If the link does not have link-layer adderss, what should
1967 	 * we do? (ifp->if_addrlen == 0)
1968 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1969 	 * description on it in NS section (RFC 2461 7.2.3).
1970 	 */
1971 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1972 	ln = nd6_lookup(from, flags, ifp);
1973 	is_newentry = 0;
1974 	if (ln == NULL) {
1975 		flags |= LLE_EXCLUSIVE;
1976 		ln = nd6_alloc(from, 0, ifp);
1977 		if (ln == NULL)
1978 			return;
1979 
1980 		/*
1981 		 * Since we already know all the data for the new entry,
1982 		 * fill it before insertion.
1983 		 */
1984 		if (lladdr != NULL) {
1985 			linkhdrsize = sizeof(linkhdr);
1986 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1987 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1988 				return;
1989 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1990 			    lladdr_off);
1991 		}
1992 
1993 		IF_AFDATA_WLOCK(ifp);
1994 		LLE_WLOCK(ln);
1995 		/* Prefer any existing lle over newly-created one */
1996 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1997 		if (ln_tmp == NULL)
1998 			lltable_link_entry(LLTABLE6(ifp), ln);
1999 		IF_AFDATA_WUNLOCK(ifp);
2000 		if (ln_tmp == NULL) {
2001 			/* No existing lle, mark as new entry (6,7) */
2002 			is_newentry = 1;
2003 			if (lladdr != NULL) {	/* (7) */
2004 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2005 				EVENTHANDLER_INVOKE(lle_event, ln,
2006 				    LLENTRY_RESOLVED);
2007 			}
2008 		} else {
2009 			lltable_free_entry(LLTABLE6(ifp), ln);
2010 			ln = ln_tmp;
2011 			ln_tmp = NULL;
2012 		}
2013 	}
2014 	/* do nothing if static ndp is set */
2015 	if ((ln->la_flags & LLE_STATIC)) {
2016 		if (flags & LLE_EXCLUSIVE)
2017 			LLE_WUNLOCK(ln);
2018 		else
2019 			LLE_RUNLOCK(ln);
2020 		return;
2021 	}
2022 
2023 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2024 	if (olladdr && lladdr) {
2025 		llchange = bcmp(lladdr, ln->ll_addr,
2026 		    ifp->if_addrlen);
2027 	} else if (!olladdr && lladdr)
2028 		llchange = 1;
2029 	else
2030 		llchange = 0;
2031 
2032 	/*
2033 	 * newentry olladdr  lladdr  llchange	(*=record)
2034 	 *	0	n	n	--	(1)
2035 	 *	0	y	n	--	(2)
2036 	 *	0	n	y	y	(3) * STALE
2037 	 *	0	y	y	n	(4) *
2038 	 *	0	y	y	y	(5) * STALE
2039 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2040 	 *	1	--	y	--	(7) * STALE
2041 	 */
2042 
2043 	do_update = 0;
2044 	if (is_newentry == 0 && llchange != 0) {
2045 		do_update = 1;	/* (3,5) */
2046 
2047 		/*
2048 		 * Record source link-layer address
2049 		 * XXX is it dependent to ifp->if_type?
2050 		 */
2051 		linkhdrsize = sizeof(linkhdr);
2052 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2053 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2054 			return;
2055 
2056 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2057 		    lladdr_off) == 0) {
2058 			/* Entry was deleted */
2059 			return;
2060 		}
2061 
2062 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2063 
2064 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2065 
2066 		if (ln->la_hold != NULL)
2067 			nd6_grab_holdchain(ln, &chain, &sin6);
2068 	}
2069 
2070 	/* Calculates new router status */
2071 	router = nd6_is_router(type, code, is_newentry, olladdr,
2072 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2073 
2074 	ln->ln_router = router;
2075 	/* Mark non-router redirects with special flag */
2076 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2077 		ln->la_flags |= LLE_REDIRECT;
2078 
2079 	if (flags & LLE_EXCLUSIVE)
2080 		LLE_WUNLOCK(ln);
2081 	else
2082 		LLE_RUNLOCK(ln);
2083 
2084 	if (chain != NULL)
2085 		nd6_flush_holdchain(ifp, chain, &sin6);
2086 
2087 	/*
2088 	 * When the link-layer address of a router changes, select the
2089 	 * best router again.  In particular, when the neighbor entry is newly
2090 	 * created, it might affect the selection policy.
2091 	 * Question: can we restrict the first condition to the "is_newentry"
2092 	 * case?
2093 	 * XXX: when we hear an RA from a new router with the link-layer
2094 	 * address option, defrouter_select_fib() is called twice, since
2095 	 * defrtrlist_update called the function as well.  However, I believe
2096 	 * we can compromise the overhead, since it only happens the first
2097 	 * time.
2098 	 * XXX: although defrouter_select_fib() should not have a bad effect
2099 	 * for those are not autoconfigured hosts, we explicitly avoid such
2100 	 * cases for safety.
2101 	 */
2102 	if ((do_update || is_newentry) && router &&
2103 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2104 		/*
2105 		 * guaranteed recursion
2106 		 */
2107 		defrouter_select_fib(ifp->if_fib);
2108 	}
2109 }
2110 
2111 static void
2112 nd6_slowtimo(void *arg)
2113 {
2114 	struct epoch_tracker et;
2115 	CURVNET_SET((struct vnet *) arg);
2116 	struct nd_ifinfo *nd6if;
2117 	struct ifnet *ifp;
2118 
2119 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2120 	    nd6_slowtimo, curvnet);
2121 	NET_EPOCH_ENTER(et);
2122 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2123 		if (ifp->if_afdata[AF_INET6] == NULL)
2124 			continue;
2125 		nd6if = ND_IFINFO(ifp);
2126 		if (nd6if->basereachable && /* already initialized */
2127 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2128 			/*
2129 			 * Since reachable time rarely changes by router
2130 			 * advertisements, we SHOULD insure that a new random
2131 			 * value gets recomputed at least once every few hours.
2132 			 * (RFC 2461, 6.3.4)
2133 			 */
2134 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2135 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2136 		}
2137 	}
2138 	NET_EPOCH_EXIT(et);
2139 	CURVNET_RESTORE();
2140 }
2141 
2142 void
2143 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2144     struct sockaddr_in6 *sin6)
2145 {
2146 
2147 	LLE_WLOCK_ASSERT(ln);
2148 
2149 	*chain = ln->la_hold;
2150 	ln->la_hold = NULL;
2151 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2152 
2153 	if (ln->ln_state == ND6_LLINFO_STALE) {
2154 
2155 		/*
2156 		 * The first time we send a packet to a
2157 		 * neighbor whose entry is STALE, we have
2158 		 * to change the state to DELAY and a sets
2159 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2160 		 * seconds to ensure do neighbor unreachability
2161 		 * detection on expiration.
2162 		 * (RFC 2461 7.3.3)
2163 		 */
2164 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2165 	}
2166 }
2167 
2168 int
2169 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2170     struct sockaddr_in6 *dst, struct route *ro)
2171 {
2172 	int error;
2173 	int ip6len;
2174 	struct ip6_hdr *ip6;
2175 	struct m_tag *mtag;
2176 
2177 #ifdef MAC
2178 	mac_netinet6_nd6_send(ifp, m);
2179 #endif
2180 
2181 	/*
2182 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2183 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2184 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2185 	 * to be diverted to user space.  When re-injected into the kernel,
2186 	 * send_output() will directly dispatch them to the outgoing interface.
2187 	 */
2188 	if (send_sendso_input_hook != NULL) {
2189 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2190 		if (mtag != NULL) {
2191 			ip6 = mtod(m, struct ip6_hdr *);
2192 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2193 			/* Use the SEND socket */
2194 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2195 			    ip6len);
2196 			/* -1 == no app on SEND socket */
2197 			if (error == 0 || error != -1)
2198 			    return (error);
2199 		}
2200 	}
2201 
2202 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2203 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2204 	    mtod(m, struct ip6_hdr *));
2205 
2206 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2207 		origifp = ifp;
2208 
2209 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2210 	return (error);
2211 }
2212 
2213 /*
2214  * Lookup link headerfor @sa_dst address. Stores found
2215  * data in @desten buffer. Copy of lle ln_flags can be also
2216  * saved in @pflags if @pflags is non-NULL.
2217  *
2218  * If destination LLE does not exists or lle state modification
2219  * is required, call "slow" version.
2220  *
2221  * Return values:
2222  * - 0 on success (address copied to buffer).
2223  * - EWOULDBLOCK (no local error, but address is still unresolved)
2224  * - other errors (alloc failure, etc)
2225  */
2226 int
2227 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2228     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2229     struct llentry **plle)
2230 {
2231 	struct llentry *ln = NULL;
2232 	const struct sockaddr_in6 *dst6;
2233 
2234 	NET_EPOCH_ASSERT();
2235 
2236 	if (pflags != NULL)
2237 		*pflags = 0;
2238 
2239 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2240 
2241 	/* discard the packet if IPv6 operation is disabled on the interface */
2242 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2243 		m_freem(m);
2244 		return (ENETDOWN); /* better error? */
2245 	}
2246 
2247 	if (m != NULL && m->m_flags & M_MCAST) {
2248 		switch (ifp->if_type) {
2249 		case IFT_ETHER:
2250 		case IFT_L2VLAN:
2251 		case IFT_BRIDGE:
2252 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2253 						 desten);
2254 			return (0);
2255 		default:
2256 			m_freem(m);
2257 			return (EAFNOSUPPORT);
2258 		}
2259 	}
2260 
2261 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2262 	    ifp);
2263 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2264 		/* Entry found, let's copy lle info */
2265 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2266 		if (pflags != NULL)
2267 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2268 		/* Check if we have feedback request from nd6 timer */
2269 		if (ln->r_skip_req != 0) {
2270 			LLE_REQ_LOCK(ln);
2271 			ln->r_skip_req = 0; /* Notify that entry was used */
2272 			ln->lle_hittime = time_uptime;
2273 			LLE_REQ_UNLOCK(ln);
2274 		}
2275 		if (plle) {
2276 			LLE_ADDREF(ln);
2277 			*plle = ln;
2278 			LLE_WUNLOCK(ln);
2279 		}
2280 		return (0);
2281 	} else if (plle && ln)
2282 		LLE_WUNLOCK(ln);
2283 
2284 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2285 }
2286 
2287 
2288 /*
2289  * Do L2 address resolution for @sa_dst address. Stores found
2290  * address in @desten buffer. Copy of lle ln_flags can be also
2291  * saved in @pflags if @pflags is non-NULL.
2292  *
2293  * Heavy version.
2294  * Function assume that destination LLE does not exist,
2295  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2296  *
2297  * Set noinline to be dtrace-friendly
2298  */
2299 static __noinline int
2300 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2301     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2302     struct llentry **plle)
2303 {
2304 	struct llentry *lle = NULL, *lle_tmp;
2305 	struct in6_addr *psrc, src;
2306 	int send_ns, ll_len;
2307 	char *lladdr;
2308 
2309 	NET_EPOCH_ASSERT();
2310 
2311 	/*
2312 	 * Address resolution or Neighbor Unreachability Detection
2313 	 * for the next hop.
2314 	 * At this point, the destination of the packet must be a unicast
2315 	 * or an anycast address(i.e. not a multicast).
2316 	 */
2317 	if (lle == NULL) {
2318 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2319 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2320 			/*
2321 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2322 			 * the condition below is not very efficient.  But we believe
2323 			 * it is tolerable, because this should be a rare case.
2324 			 */
2325 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2326 			if (lle == NULL) {
2327 				char ip6buf[INET6_ADDRSTRLEN];
2328 				log(LOG_DEBUG,
2329 				    "nd6_output: can't allocate llinfo for %s "
2330 				    "(ln=%p)\n",
2331 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2332 				m_freem(m);
2333 				return (ENOBUFS);
2334 			}
2335 
2336 			IF_AFDATA_WLOCK(ifp);
2337 			LLE_WLOCK(lle);
2338 			/* Prefer any existing entry over newly-created one */
2339 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2340 			if (lle_tmp == NULL)
2341 				lltable_link_entry(LLTABLE6(ifp), lle);
2342 			IF_AFDATA_WUNLOCK(ifp);
2343 			if (lle_tmp != NULL) {
2344 				lltable_free_entry(LLTABLE6(ifp), lle);
2345 				lle = lle_tmp;
2346 				lle_tmp = NULL;
2347 			}
2348 		}
2349 	}
2350 	if (lle == NULL) {
2351 		m_freem(m);
2352 		return (ENOBUFS);
2353 	}
2354 
2355 	LLE_WLOCK_ASSERT(lle);
2356 
2357 	/*
2358 	 * The first time we send a packet to a neighbor whose entry is
2359 	 * STALE, we have to change the state to DELAY and a sets a timer to
2360 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2361 	 * neighbor unreachability detection on expiration.
2362 	 * (RFC 2461 7.3.3)
2363 	 */
2364 	if (lle->ln_state == ND6_LLINFO_STALE)
2365 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2366 
2367 	/*
2368 	 * If the neighbor cache entry has a state other than INCOMPLETE
2369 	 * (i.e. its link-layer address is already resolved), just
2370 	 * send the packet.
2371 	 */
2372 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2373 		if (flags & LLE_ADDRONLY) {
2374 			lladdr = lle->ll_addr;
2375 			ll_len = ifp->if_addrlen;
2376 		} else {
2377 			lladdr = lle->r_linkdata;
2378 			ll_len = lle->r_hdrlen;
2379 		}
2380 		bcopy(lladdr, desten, ll_len);
2381 		if (pflags != NULL)
2382 			*pflags = lle->la_flags;
2383 		if (plle) {
2384 			LLE_ADDREF(lle);
2385 			*plle = lle;
2386 		}
2387 		LLE_WUNLOCK(lle);
2388 		return (0);
2389 	}
2390 
2391 	/*
2392 	 * There is a neighbor cache entry, but no ethernet address
2393 	 * response yet.  Append this latest packet to the end of the
2394 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2395 	 * the oldest packet in the queue will be removed.
2396 	 */
2397 
2398 	if (lle->la_hold != NULL) {
2399 		struct mbuf *m_hold;
2400 		int i;
2401 
2402 		i = 0;
2403 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2404 			i++;
2405 			if (m_hold->m_nextpkt == NULL) {
2406 				m_hold->m_nextpkt = m;
2407 				break;
2408 			}
2409 		}
2410 		while (i >= V_nd6_maxqueuelen) {
2411 			m_hold = lle->la_hold;
2412 			lle->la_hold = lle->la_hold->m_nextpkt;
2413 			m_freem(m_hold);
2414 			i--;
2415 		}
2416 	} else {
2417 		lle->la_hold = m;
2418 	}
2419 
2420 	/*
2421 	 * If there has been no NS for the neighbor after entering the
2422 	 * INCOMPLETE state, send the first solicitation.
2423 	 * Note that for newly-created lle la_asked will be 0,
2424 	 * so we will transition from ND6_LLINFO_NOSTATE to
2425 	 * ND6_LLINFO_INCOMPLETE state here.
2426 	 */
2427 	psrc = NULL;
2428 	send_ns = 0;
2429 	if (lle->la_asked == 0) {
2430 		lle->la_asked++;
2431 		send_ns = 1;
2432 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2433 
2434 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2435 	}
2436 	LLE_WUNLOCK(lle);
2437 	if (send_ns != 0)
2438 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2439 
2440 	return (EWOULDBLOCK);
2441 }
2442 
2443 /*
2444  * Do L2 address resolution for @sa_dst address. Stores found
2445  * address in @desten buffer. Copy of lle ln_flags can be also
2446  * saved in @pflags if @pflags is non-NULL.
2447  *
2448  * Return values:
2449  * - 0 on success (address copied to buffer).
2450  * - EWOULDBLOCK (no local error, but address is still unresolved)
2451  * - other errors (alloc failure, etc)
2452  */
2453 int
2454 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2455     char *desten, uint32_t *pflags)
2456 {
2457 	int error;
2458 
2459 	flags |= LLE_ADDRONLY;
2460 	error = nd6_resolve_slow(ifp, flags, NULL,
2461 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2462 	return (error);
2463 }
2464 
2465 int
2466 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2467     struct sockaddr_in6 *dst)
2468 {
2469 	struct mbuf *m, *m_head;
2470 	int error = 0;
2471 
2472 	m_head = chain;
2473 
2474 	while (m_head) {
2475 		m = m_head;
2476 		m_head = m_head->m_nextpkt;
2477 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2478 	}
2479 
2480 	/*
2481 	 * XXX
2482 	 * note that intermediate errors are blindly ignored
2483 	 */
2484 	return (error);
2485 }
2486 
2487 static int
2488 nd6_need_cache(struct ifnet *ifp)
2489 {
2490 	/*
2491 	 * XXX: we currently do not make neighbor cache on any interface
2492 	 * other than Ethernet and GIF.
2493 	 *
2494 	 * RFC2893 says:
2495 	 * - unidirectional tunnels needs no ND
2496 	 */
2497 	switch (ifp->if_type) {
2498 	case IFT_ETHER:
2499 	case IFT_IEEE1394:
2500 	case IFT_L2VLAN:
2501 	case IFT_INFINIBAND:
2502 	case IFT_BRIDGE:
2503 	case IFT_PROPVIRTUAL:
2504 		return (1);
2505 	default:
2506 		return (0);
2507 	}
2508 }
2509 
2510 /*
2511  * Add pernament ND6 link-layer record for given
2512  * interface address.
2513  *
2514  * Very similar to IPv4 arp_ifinit(), but:
2515  * 1) IPv6 DAD is performed in different place
2516  * 2) It is called by IPv6 protocol stack in contrast to
2517  * arp_ifinit() which is typically called in SIOCSIFADDR
2518  * driver ioctl handler.
2519  *
2520  */
2521 int
2522 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2523 {
2524 	struct ifnet *ifp;
2525 	struct llentry *ln, *ln_tmp;
2526 	struct sockaddr *dst;
2527 
2528 	ifp = ia->ia_ifa.ifa_ifp;
2529 	if (nd6_need_cache(ifp) == 0)
2530 		return (0);
2531 
2532 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2533 	dst = (struct sockaddr *)&ia->ia_addr;
2534 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2535 	if (ln == NULL)
2536 		return (ENOBUFS);
2537 
2538 	IF_AFDATA_WLOCK(ifp);
2539 	LLE_WLOCK(ln);
2540 	/* Unlink any entry if exists */
2541 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2542 	if (ln_tmp != NULL)
2543 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2544 	lltable_link_entry(LLTABLE6(ifp), ln);
2545 	IF_AFDATA_WUNLOCK(ifp);
2546 
2547 	if (ln_tmp != NULL)
2548 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2549 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2550 
2551 	LLE_WUNLOCK(ln);
2552 	if (ln_tmp != NULL)
2553 		llentry_free(ln_tmp);
2554 
2555 	return (0);
2556 }
2557 
2558 /*
2559  * Removes either all lle entries for given @ia, or lle
2560  * corresponding to @ia address.
2561  */
2562 void
2563 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2564 {
2565 	struct sockaddr_in6 mask, addr;
2566 	struct sockaddr *saddr, *smask;
2567 	struct ifnet *ifp;
2568 
2569 	ifp = ia->ia_ifa.ifa_ifp;
2570 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2571 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2572 	saddr = (struct sockaddr *)&addr;
2573 	smask = (struct sockaddr *)&mask;
2574 
2575 	if (all != 0)
2576 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2577 	else
2578 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2579 }
2580 
2581 static void
2582 clear_llinfo_pqueue(struct llentry *ln)
2583 {
2584 	struct mbuf *m_hold, *m_hold_next;
2585 
2586 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2587 		m_hold_next = m_hold->m_nextpkt;
2588 		m_freem(m_hold);
2589 	}
2590 
2591 	ln->la_hold = NULL;
2592 }
2593 
2594 static int
2595 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2596 {
2597 	struct in6_prefix p;
2598 	struct sockaddr_in6 s6;
2599 	struct nd_prefix *pr;
2600 	struct nd_pfxrouter *pfr;
2601 	time_t maxexpire;
2602 	int error;
2603 	char ip6buf[INET6_ADDRSTRLEN];
2604 
2605 	if (req->newptr)
2606 		return (EPERM);
2607 
2608 	error = sysctl_wire_old_buffer(req, 0);
2609 	if (error != 0)
2610 		return (error);
2611 
2612 	bzero(&p, sizeof(p));
2613 	p.origin = PR_ORIG_RA;
2614 	bzero(&s6, sizeof(s6));
2615 	s6.sin6_family = AF_INET6;
2616 	s6.sin6_len = sizeof(s6);
2617 
2618 	ND6_RLOCK();
2619 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2620 		p.prefix = pr->ndpr_prefix;
2621 		if (sa6_recoverscope(&p.prefix)) {
2622 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2623 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2624 			/* XXX: press on... */
2625 		}
2626 		p.raflags = pr->ndpr_raf;
2627 		p.prefixlen = pr->ndpr_plen;
2628 		p.vltime = pr->ndpr_vltime;
2629 		p.pltime = pr->ndpr_pltime;
2630 		p.if_index = pr->ndpr_ifp->if_index;
2631 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2632 			p.expire = 0;
2633 		else {
2634 			/* XXX: we assume time_t is signed. */
2635 			maxexpire = (-1) &
2636 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2637 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2638 				p.expire = pr->ndpr_lastupdate +
2639 				    pr->ndpr_vltime +
2640 				    (time_second - time_uptime);
2641 			else
2642 				p.expire = maxexpire;
2643 		}
2644 		p.refcnt = pr->ndpr_addrcnt;
2645 		p.flags = pr->ndpr_stateflags;
2646 		p.advrtrs = 0;
2647 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2648 			p.advrtrs++;
2649 		error = SYSCTL_OUT(req, &p, sizeof(p));
2650 		if (error != 0)
2651 			break;
2652 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2653 			s6.sin6_addr = pfr->router->rtaddr;
2654 			if (sa6_recoverscope(&s6))
2655 				log(LOG_ERR,
2656 				    "scope error in prefix list (%s)\n",
2657 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2658 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2659 			if (error != 0)
2660 				goto out;
2661 		}
2662 	}
2663 out:
2664 	ND6_RUNLOCK();
2665 	return (error);
2666 }
2667 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2668 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2669 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2670 	"NDP prefix list");
2671 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2672 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2673 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2674 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2675