xref: /freebsd/sys/netinet6/nd6.c (revision b0d29bc47dba79f6f38e67eabadfb4b32ffd9390)
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_prhead, nd_prefix);
120 VNET_DEFINE(struct rwlock, nd6_lock);
121 VNET_DEFINE(uint64_t, nd6_list_genid);
122 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
123 
124 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
125 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
126 
127 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
128 
129 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
130 	struct ifnet *);
131 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
132 static void nd6_slowtimo(void *);
133 static int regen_tmpaddr(struct in6_ifaddr *);
134 static void nd6_free(struct llentry **, int);
135 static void nd6_free_redirect(const struct llentry *);
136 static void nd6_llinfo_timer(void *);
137 static void nd6_llinfo_settimer_locked(struct llentry *, long);
138 static void clear_llinfo_pqueue(struct llentry *);
139 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
140 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
141     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
142 static int nd6_need_cache(struct ifnet *);
143 
144 
145 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
146 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
147 
148 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
149 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
150 
151 SYSCTL_DECL(_net_inet6_icmp6);
152 
153 static void
154 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
155 {
156 	struct rt_addrinfo rtinfo;
157 	struct sockaddr_in6 dst;
158 	struct sockaddr_dl gw;
159 	struct ifnet *ifp;
160 	int type;
161 	int fibnum;
162 
163 	LLE_WLOCK_ASSERT(lle);
164 
165 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
166 		return;
167 
168 	switch (evt) {
169 	case LLENTRY_RESOLVED:
170 		type = RTM_ADD;
171 		KASSERT(lle->la_flags & LLE_VALID,
172 		    ("%s: %p resolved but not valid?", __func__, lle));
173 		break;
174 	case LLENTRY_EXPIRED:
175 		type = RTM_DELETE;
176 		break;
177 	default:
178 		return;
179 	}
180 
181 	ifp = lltable_get_ifp(lle->lle_tbl);
182 
183 	bzero(&dst, sizeof(dst));
184 	bzero(&gw, sizeof(gw));
185 	bzero(&rtinfo, sizeof(rtinfo));
186 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
187 	dst.sin6_scope_id = in6_getscopezone(ifp,
188 	    in6_addrscope(&dst.sin6_addr));
189 	gw.sdl_len = sizeof(struct sockaddr_dl);
190 	gw.sdl_family = AF_LINK;
191 	gw.sdl_alen = ifp->if_addrlen;
192 	gw.sdl_index = ifp->if_index;
193 	gw.sdl_type = ifp->if_type;
194 	if (evt == LLENTRY_RESOLVED)
195 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
196 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
197 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
198 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
199 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
200 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
201 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
202 }
203 
204 /*
205  * A handler for interface link layer address change event.
206  */
207 static void
208 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
209 {
210 
211 	lltable_update_ifaddr(LLTABLE6(ifp));
212 }
213 
214 void
215 nd6_init(void)
216 {
217 
218 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
219 	rw_init(&V_nd6_lock, "nd6 list");
220 
221 	LIST_INIT(&V_nd_prefix);
222 	nd6_defrouter_init();
223 
224 	/* Start timers. */
225 	callout_init(&V_nd6_slowtimo_ch, 0);
226 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
227 	    nd6_slowtimo, curvnet);
228 
229 	callout_init(&V_nd6_timer_ch, 0);
230 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
231 
232 	nd6_dad_init();
233 	if (IS_DEFAULT_VNET(curvnet)) {
234 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
235 		    NULL, EVENTHANDLER_PRI_ANY);
236 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
237 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
238 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
239 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
240 	}
241 }
242 
243 #ifdef VIMAGE
244 void
245 nd6_destroy()
246 {
247 
248 	callout_drain(&V_nd6_slowtimo_ch);
249 	callout_drain(&V_nd6_timer_ch);
250 	if (IS_DEFAULT_VNET(curvnet)) {
251 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
252 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
253 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
254 	}
255 	rw_destroy(&V_nd6_lock);
256 	mtx_destroy(&V_nd6_onlink_mtx);
257 }
258 #endif
259 
260 struct nd_ifinfo *
261 nd6_ifattach(struct ifnet *ifp)
262 {
263 	struct nd_ifinfo *nd;
264 
265 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
266 	nd->initialized = 1;
267 
268 	nd->chlim = IPV6_DEFHLIM;
269 	nd->basereachable = REACHABLE_TIME;
270 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
271 	nd->retrans = RETRANS_TIMER;
272 
273 	nd->flags = ND6_IFF_PERFORMNUD;
274 
275 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
276 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
277 	 * default regardless of the V_ip6_auto_linklocal configuration to
278 	 * give a reasonable default behavior.
279 	 */
280 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
281 	    (ifp->if_flags & IFF_LOOPBACK))
282 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
283 	/*
284 	 * A loopback interface does not need to accept RTADV.
285 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
286 	 * default regardless of the V_ip6_accept_rtadv configuration to
287 	 * prevent the interface from accepting RA messages arrived
288 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
289 	 */
290 	if (V_ip6_accept_rtadv &&
291 	    !(ifp->if_flags & IFF_LOOPBACK) &&
292 	    (ifp->if_type != IFT_BRIDGE))
293 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
294 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
295 		nd->flags |= ND6_IFF_NO_RADR;
296 
297 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
298 	nd6_setmtu0(ifp, nd);
299 
300 	return nd;
301 }
302 
303 void
304 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
305 {
306 	struct epoch_tracker et;
307 	struct ifaddr *ifa, *next;
308 
309 	NET_EPOCH_ENTER(et);
310 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
311 		if (ifa->ifa_addr->sa_family != AF_INET6)
312 			continue;
313 
314 		/* stop DAD processing */
315 		nd6_dad_stop(ifa);
316 	}
317 	NET_EPOCH_EXIT(et);
318 
319 	free(nd, M_IP6NDP);
320 }
321 
322 /*
323  * Reset ND level link MTU. This function is called when the physical MTU
324  * changes, which means we might have to adjust the ND level MTU.
325  */
326 void
327 nd6_setmtu(struct ifnet *ifp)
328 {
329 	if (ifp->if_afdata[AF_INET6] == NULL)
330 		return;
331 
332 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
333 }
334 
335 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
336 void
337 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
338 {
339 	u_int32_t omaxmtu;
340 
341 	omaxmtu = ndi->maxmtu;
342 	ndi->maxmtu = ifp->if_mtu;
343 
344 	/*
345 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
346 	 * undesirable situation.  We thus notify the operator of the change
347 	 * explicitly.  The check for omaxmtu is necessary to restrict the
348 	 * log to the case of changing the MTU, not initializing it.
349 	 */
350 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
351 		log(LOG_NOTICE, "nd6_setmtu0: "
352 		    "new link MTU on %s (%lu) is too small for IPv6\n",
353 		    if_name(ifp), (unsigned long)ndi->maxmtu);
354 	}
355 
356 	if (ndi->maxmtu > V_in6_maxmtu)
357 		in6_setmaxmtu(); /* check all interfaces just in case */
358 
359 }
360 
361 void
362 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
363 {
364 
365 	bzero(ndopts, sizeof(*ndopts));
366 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
367 	ndopts->nd_opts_last
368 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
369 
370 	if (icmp6len == 0) {
371 		ndopts->nd_opts_done = 1;
372 		ndopts->nd_opts_search = NULL;
373 	}
374 }
375 
376 /*
377  * Take one ND option.
378  */
379 struct nd_opt_hdr *
380 nd6_option(union nd_opts *ndopts)
381 {
382 	struct nd_opt_hdr *nd_opt;
383 	int olen;
384 
385 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
386 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
387 	    __func__));
388 	if (ndopts->nd_opts_search == NULL)
389 		return NULL;
390 	if (ndopts->nd_opts_done)
391 		return NULL;
392 
393 	nd_opt = ndopts->nd_opts_search;
394 
395 	/* make sure nd_opt_len is inside the buffer */
396 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
397 		bzero(ndopts, sizeof(*ndopts));
398 		return NULL;
399 	}
400 
401 	olen = nd_opt->nd_opt_len << 3;
402 	if (olen == 0) {
403 		/*
404 		 * Message validation requires that all included
405 		 * options have a length that is greater than zero.
406 		 */
407 		bzero(ndopts, sizeof(*ndopts));
408 		return NULL;
409 	}
410 
411 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
412 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
413 		/* option overruns the end of buffer, invalid */
414 		bzero(ndopts, sizeof(*ndopts));
415 		return NULL;
416 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
417 		/* reached the end of options chain */
418 		ndopts->nd_opts_done = 1;
419 		ndopts->nd_opts_search = NULL;
420 	}
421 	return nd_opt;
422 }
423 
424 /*
425  * Parse multiple ND options.
426  * This function is much easier to use, for ND routines that do not need
427  * multiple options of the same type.
428  */
429 int
430 nd6_options(union nd_opts *ndopts)
431 {
432 	struct nd_opt_hdr *nd_opt;
433 	int i = 0;
434 
435 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
436 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
437 	    __func__));
438 	if (ndopts->nd_opts_search == NULL)
439 		return 0;
440 
441 	while (1) {
442 		nd_opt = nd6_option(ndopts);
443 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
444 			/*
445 			 * Message validation requires that all included
446 			 * options have a length that is greater than zero.
447 			 */
448 			ICMP6STAT_INC(icp6s_nd_badopt);
449 			bzero(ndopts, sizeof(*ndopts));
450 			return -1;
451 		}
452 
453 		if (nd_opt == NULL)
454 			goto skip1;
455 
456 		switch (nd_opt->nd_opt_type) {
457 		case ND_OPT_SOURCE_LINKADDR:
458 		case ND_OPT_TARGET_LINKADDR:
459 		case ND_OPT_MTU:
460 		case ND_OPT_REDIRECTED_HEADER:
461 		case ND_OPT_NONCE:
462 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
463 				nd6log((LOG_INFO,
464 				    "duplicated ND6 option found (type=%d)\n",
465 				    nd_opt->nd_opt_type));
466 				/* XXX bark? */
467 			} else {
468 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
469 					= nd_opt;
470 			}
471 			break;
472 		case ND_OPT_PREFIX_INFORMATION:
473 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
474 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
475 					= nd_opt;
476 			}
477 			ndopts->nd_opts_pi_end =
478 				(struct nd_opt_prefix_info *)nd_opt;
479 			break;
480 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
481 		case ND_OPT_RDNSS:	/* RFC 6106 */
482 		case ND_OPT_DNSSL:	/* RFC 6106 */
483 			/*
484 			 * Silently ignore options we know and do not care about
485 			 * in the kernel.
486 			 */
487 			break;
488 		default:
489 			/*
490 			 * Unknown options must be silently ignored,
491 			 * to accommodate future extension to the protocol.
492 			 */
493 			nd6log((LOG_DEBUG,
494 			    "nd6_options: unsupported option %d - "
495 			    "option ignored\n", nd_opt->nd_opt_type));
496 		}
497 
498 skip1:
499 		i++;
500 		if (i > V_nd6_maxndopt) {
501 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
502 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
503 			break;
504 		}
505 
506 		if (ndopts->nd_opts_done)
507 			break;
508 	}
509 
510 	return 0;
511 }
512 
513 /*
514  * ND6 timer routine to handle ND6 entries
515  */
516 static void
517 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
518 {
519 	int canceled;
520 
521 	LLE_WLOCK_ASSERT(ln);
522 
523 	if (tick < 0) {
524 		ln->la_expire = 0;
525 		ln->ln_ntick = 0;
526 		canceled = callout_stop(&ln->lle_timer);
527 	} else {
528 		ln->la_expire = time_uptime + tick / hz;
529 		LLE_ADDREF(ln);
530 		if (tick > INT_MAX) {
531 			ln->ln_ntick = tick - INT_MAX;
532 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
533 			    nd6_llinfo_timer, ln);
534 		} else {
535 			ln->ln_ntick = 0;
536 			canceled = callout_reset(&ln->lle_timer, tick,
537 			    nd6_llinfo_timer, ln);
538 		}
539 	}
540 	if (canceled > 0)
541 		LLE_REMREF(ln);
542 }
543 
544 /*
545  * Gets source address of the first packet in hold queue
546  * and stores it in @src.
547  * Returns pointer to @src (if hold queue is not empty) or NULL.
548  *
549  * Set noinline to be dtrace-friendly
550  */
551 static __noinline struct in6_addr *
552 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
553 {
554 	struct ip6_hdr hdr;
555 	struct mbuf *m;
556 
557 	if (ln->la_hold == NULL)
558 		return (NULL);
559 
560 	/*
561 	 * assume every packet in la_hold has the same IP header
562 	 */
563 	m = ln->la_hold;
564 	if (sizeof(hdr) > m->m_len)
565 		return (NULL);
566 
567 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
568 	*src = hdr.ip6_src;
569 
570 	return (src);
571 }
572 
573 /*
574  * Checks if we need to switch from STALE state.
575  *
576  * RFC 4861 requires switching from STALE to DELAY state
577  * on first packet matching entry, waiting V_nd6_delay and
578  * transition to PROBE state (if upper layer confirmation was
579  * not received).
580  *
581  * This code performs a bit differently:
582  * On packet hit we don't change state (but desired state
583  * can be guessed by control plane). However, after V_nd6_delay
584  * seconds code will transition to PROBE state (so DELAY state
585  * is kinda skipped in most situations).
586  *
587  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
588  * we perform the following upon entering STALE state:
589  *
590  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
591  * if packet was transmitted at the start of given interval, we
592  * would be able to switch to PROBE state in V_nd6_delay seconds
593  * as user expects.
594  *
595  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
596  * lle in STALE state (remaining timer value stored in lle_remtime).
597  *
598  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
599  * seconds ago.
600  *
601  * Returns non-zero value if the entry is still STALE (storing
602  * the next timer interval in @pdelay).
603  *
604  * Returns zero value if original timer expired or we need to switch to
605  * PROBE (store that in @do_switch variable).
606  */
607 static int
608 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
609 {
610 	int nd_delay, nd_gctimer, r_skip_req;
611 	time_t lle_hittime;
612 	long delay;
613 
614 	*do_switch = 0;
615 	nd_gctimer = V_nd6_gctimer;
616 	nd_delay = V_nd6_delay;
617 
618 	LLE_REQ_LOCK(lle);
619 	r_skip_req = lle->r_skip_req;
620 	lle_hittime = lle->lle_hittime;
621 	LLE_REQ_UNLOCK(lle);
622 
623 	if (r_skip_req > 0) {
624 
625 		/*
626 		 * Nonzero r_skip_req value was set upon entering
627 		 * STALE state. Since value was not changed, no
628 		 * packets were passed using this lle. Ask for
629 		 * timer reschedule and keep STALE state.
630 		 */
631 		delay = (long)(MIN(nd_gctimer, nd_delay));
632 		delay *= hz;
633 		if (lle->lle_remtime > delay)
634 			lle->lle_remtime -= delay;
635 		else {
636 			delay = lle->lle_remtime;
637 			lle->lle_remtime = 0;
638 		}
639 
640 		if (delay == 0) {
641 
642 			/*
643 			 * The original ng6_gctime timeout ended,
644 			 * no more rescheduling.
645 			 */
646 			return (0);
647 		}
648 
649 		*pdelay = delay;
650 		return (1);
651 	}
652 
653 	/*
654 	 * Packet received. Verify timestamp
655 	 */
656 	delay = (long)(time_uptime - lle_hittime);
657 	if (delay < nd_delay) {
658 
659 		/*
660 		 * V_nd6_delay still not passed since the first
661 		 * hit in STALE state.
662 		 * Reshedule timer and return.
663 		 */
664 		*pdelay = (long)(nd_delay - delay) * hz;
665 		return (1);
666 	}
667 
668 	/* Request switching to probe */
669 	*do_switch = 1;
670 	return (0);
671 }
672 
673 
674 /*
675  * Switch @lle state to new state optionally arming timers.
676  *
677  * Set noinline to be dtrace-friendly
678  */
679 __noinline void
680 nd6_llinfo_setstate(struct llentry *lle, int newstate)
681 {
682 	struct ifnet *ifp;
683 	int nd_gctimer, nd_delay;
684 	long delay, remtime;
685 
686 	delay = 0;
687 	remtime = 0;
688 
689 	switch (newstate) {
690 	case ND6_LLINFO_INCOMPLETE:
691 		ifp = lle->lle_tbl->llt_ifp;
692 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
693 		break;
694 	case ND6_LLINFO_REACHABLE:
695 		if (!ND6_LLINFO_PERMANENT(lle)) {
696 			ifp = lle->lle_tbl->llt_ifp;
697 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
698 		}
699 		break;
700 	case ND6_LLINFO_STALE:
701 
702 		/*
703 		 * Notify fast path that we want to know if any packet
704 		 * is transmitted by setting r_skip_req.
705 		 */
706 		LLE_REQ_LOCK(lle);
707 		lle->r_skip_req = 1;
708 		LLE_REQ_UNLOCK(lle);
709 		nd_delay = V_nd6_delay;
710 		nd_gctimer = V_nd6_gctimer;
711 
712 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
713 		remtime = (long)nd_gctimer * hz - delay;
714 		break;
715 	case ND6_LLINFO_DELAY:
716 		lle->la_asked = 0;
717 		delay = (long)V_nd6_delay * hz;
718 		break;
719 	}
720 
721 	if (delay > 0)
722 		nd6_llinfo_settimer_locked(lle, delay);
723 
724 	lle->lle_remtime = remtime;
725 	lle->ln_state = newstate;
726 }
727 
728 /*
729  * Timer-dependent part of nd state machine.
730  *
731  * Set noinline to be dtrace-friendly
732  */
733 static __noinline void
734 nd6_llinfo_timer(void *arg)
735 {
736 	struct epoch_tracker et;
737 	struct llentry *ln;
738 	struct in6_addr *dst, *pdst, *psrc, src;
739 	struct ifnet *ifp;
740 	struct nd_ifinfo *ndi;
741 	int do_switch, send_ns;
742 	long delay;
743 
744 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
745 	ln = (struct llentry *)arg;
746 	ifp = lltable_get_ifp(ln->lle_tbl);
747 	CURVNET_SET(ifp->if_vnet);
748 
749 	ND6_RLOCK();
750 	LLE_WLOCK(ln);
751 	if (callout_pending(&ln->lle_timer)) {
752 		/*
753 		 * Here we are a bit odd here in the treatment of
754 		 * active/pending. If the pending bit is set, it got
755 		 * rescheduled before I ran. The active
756 		 * bit we ignore, since if it was stopped
757 		 * in ll_tablefree() and was currently running
758 		 * it would have return 0 so the code would
759 		 * not have deleted it since the callout could
760 		 * not be stopped so we want to go through
761 		 * with the delete here now. If the callout
762 		 * was restarted, the pending bit will be back on and
763 		 * we just want to bail since the callout_reset would
764 		 * return 1 and our reference would have been removed
765 		 * by nd6_llinfo_settimer_locked above since canceled
766 		 * would have been 1.
767 		 */
768 		LLE_WUNLOCK(ln);
769 		ND6_RUNLOCK();
770 		CURVNET_RESTORE();
771 		return;
772 	}
773 	NET_EPOCH_ENTER(et);
774 	ndi = ND_IFINFO(ifp);
775 	send_ns = 0;
776 	dst = &ln->r_l3addr.addr6;
777 	pdst = dst;
778 
779 	if (ln->ln_ntick > 0) {
780 		if (ln->ln_ntick > INT_MAX) {
781 			ln->ln_ntick -= INT_MAX;
782 			nd6_llinfo_settimer_locked(ln, INT_MAX);
783 		} else {
784 			ln->ln_ntick = 0;
785 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
786 		}
787 		goto done;
788 	}
789 
790 	if (ln->la_flags & LLE_STATIC) {
791 		goto done;
792 	}
793 
794 	if (ln->la_flags & LLE_DELETED) {
795 		nd6_free(&ln, 0);
796 		goto done;
797 	}
798 
799 	switch (ln->ln_state) {
800 	case ND6_LLINFO_INCOMPLETE:
801 		if (ln->la_asked < V_nd6_mmaxtries) {
802 			ln->la_asked++;
803 			send_ns = 1;
804 			/* Send NS to multicast address */
805 			pdst = NULL;
806 		} else {
807 			struct mbuf *m = ln->la_hold;
808 			if (m) {
809 				struct mbuf *m0;
810 
811 				/*
812 				 * assuming every packet in la_hold has the
813 				 * same IP header.  Send error after unlock.
814 				 */
815 				m0 = m->m_nextpkt;
816 				m->m_nextpkt = NULL;
817 				ln->la_hold = m0;
818 				clear_llinfo_pqueue(ln);
819 			}
820 			nd6_free(&ln, 0);
821 			if (m != NULL)
822 				icmp6_error2(m, ICMP6_DST_UNREACH,
823 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
824 		}
825 		break;
826 	case ND6_LLINFO_REACHABLE:
827 		if (!ND6_LLINFO_PERMANENT(ln))
828 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
829 		break;
830 
831 	case ND6_LLINFO_STALE:
832 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
833 
834 			/*
835 			 * No packet has used this entry and GC timeout
836 			 * has not been passed. Reshedule timer and
837 			 * return.
838 			 */
839 			nd6_llinfo_settimer_locked(ln, delay);
840 			break;
841 		}
842 
843 		if (do_switch == 0) {
844 
845 			/*
846 			 * GC timer has ended and entry hasn't been used.
847 			 * Run Garbage collector (RFC 4861, 5.3)
848 			 */
849 			if (!ND6_LLINFO_PERMANENT(ln))
850 				nd6_free(&ln, 1);
851 			break;
852 		}
853 
854 		/* Entry has been used AND delay timer has ended. */
855 
856 		/* FALLTHROUGH */
857 
858 	case ND6_LLINFO_DELAY:
859 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
860 			/* We need NUD */
861 			ln->la_asked = 1;
862 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
863 			send_ns = 1;
864 		} else
865 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
866 		break;
867 	case ND6_LLINFO_PROBE:
868 		if (ln->la_asked < V_nd6_umaxtries) {
869 			ln->la_asked++;
870 			send_ns = 1;
871 		} else {
872 			nd6_free(&ln, 0);
873 		}
874 		break;
875 	default:
876 		panic("%s: paths in a dark night can be confusing: %d",
877 		    __func__, ln->ln_state);
878 	}
879 done:
880 	if (ln != NULL)
881 		ND6_RUNLOCK();
882 	if (send_ns != 0) {
883 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
884 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
885 		LLE_FREE_LOCKED(ln);
886 		ln = NULL;
887 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
888 	}
889 
890 	if (ln != NULL)
891 		LLE_FREE_LOCKED(ln);
892 	NET_EPOCH_EXIT(et);
893 	CURVNET_RESTORE();
894 }
895 
896 
897 /*
898  * ND6 timer routine to expire default route list and prefix list
899  */
900 void
901 nd6_timer(void *arg)
902 {
903 	CURVNET_SET((struct vnet *) arg);
904 	struct epoch_tracker et;
905 	struct nd_prhead prl;
906 	struct nd_prefix *pr, *npr;
907 	struct ifnet *ifp;
908 	struct in6_ifaddr *ia6, *nia6;
909 	uint64_t genid;
910 
911 	LIST_INIT(&prl);
912 
913 	NET_EPOCH_ENTER(et);
914 	nd6_defrouter_timer();
915 
916 	/*
917 	 * expire interface addresses.
918 	 * in the past the loop was inside prefix expiry processing.
919 	 * However, from a stricter speci-confrmance standpoint, we should
920 	 * rather separate address lifetimes and prefix lifetimes.
921 	 *
922 	 * XXXRW: in6_ifaddrhead locking.
923 	 */
924   addrloop:
925 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
926 		/* check address lifetime */
927 		if (IFA6_IS_INVALID(ia6)) {
928 			int regen = 0;
929 
930 			/*
931 			 * If the expiring address is temporary, try
932 			 * regenerating a new one.  This would be useful when
933 			 * we suspended a laptop PC, then turned it on after a
934 			 * period that could invalidate all temporary
935 			 * addresses.  Although we may have to restart the
936 			 * loop (see below), it must be after purging the
937 			 * address.  Otherwise, we'd see an infinite loop of
938 			 * regeneration.
939 			 */
940 			if (V_ip6_use_tempaddr &&
941 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
942 				if (regen_tmpaddr(ia6) == 0)
943 					regen = 1;
944 			}
945 
946 			in6_purgeaddr(&ia6->ia_ifa);
947 
948 			if (regen)
949 				goto addrloop; /* XXX: see below */
950 		} else if (IFA6_IS_DEPRECATED(ia6)) {
951 			int oldflags = ia6->ia6_flags;
952 
953 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
954 
955 			/*
956 			 * If a temporary address has just become deprecated,
957 			 * regenerate a new one if possible.
958 			 */
959 			if (V_ip6_use_tempaddr &&
960 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
961 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
962 
963 				if (regen_tmpaddr(ia6) == 0) {
964 					/*
965 					 * A new temporary address is
966 					 * generated.
967 					 * XXX: this means the address chain
968 					 * has changed while we are still in
969 					 * the loop.  Although the change
970 					 * would not cause disaster (because
971 					 * it's not a deletion, but an
972 					 * addition,) we'd rather restart the
973 					 * loop just for safety.  Or does this
974 					 * significantly reduce performance??
975 					 */
976 					goto addrloop;
977 				}
978 			}
979 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
980 			/*
981 			 * Schedule DAD for a tentative address.  This happens
982 			 * if the interface was down or not running
983 			 * when the address was configured.
984 			 */
985 			int delay;
986 
987 			delay = arc4random() %
988 			    (MAX_RTR_SOLICITATION_DELAY * hz);
989 			nd6_dad_start((struct ifaddr *)ia6, delay);
990 		} else {
991 			/*
992 			 * Check status of the interface.  If it is down,
993 			 * mark the address as tentative for future DAD.
994 			 */
995 			ifp = ia6->ia_ifp;
996 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
997 			    ((ifp->if_flags & IFF_UP) == 0 ||
998 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
999 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1000 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1001 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1002 			}
1003 
1004 			/*
1005 			 * A new RA might have made a deprecated address
1006 			 * preferred.
1007 			 */
1008 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1009 		}
1010 	}
1011 	NET_EPOCH_EXIT(et);
1012 
1013 	ND6_WLOCK();
1014 restart:
1015 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1016 		/*
1017 		 * Expire prefixes. Since the pltime is only used for
1018 		 * autoconfigured addresses, pltime processing for prefixes is
1019 		 * not necessary.
1020 		 *
1021 		 * Only unlink after all derived addresses have expired. This
1022 		 * may not occur until two hours after the prefix has expired
1023 		 * per RFC 4862. If the prefix expires before its derived
1024 		 * addresses, mark it off-link. This will be done automatically
1025 		 * after unlinking if no address references remain.
1026 		 */
1027 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1028 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1029 			continue;
1030 
1031 		if (pr->ndpr_addrcnt == 0) {
1032 			nd6_prefix_unlink(pr, &prl);
1033 			continue;
1034 		}
1035 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1036 			genid = V_nd6_list_genid;
1037 			nd6_prefix_ref(pr);
1038 			ND6_WUNLOCK();
1039 			ND6_ONLINK_LOCK();
1040 			(void)nd6_prefix_offlink(pr);
1041 			ND6_ONLINK_UNLOCK();
1042 			ND6_WLOCK();
1043 			nd6_prefix_rele(pr);
1044 			if (genid != V_nd6_list_genid)
1045 				goto restart;
1046 		}
1047 	}
1048 	ND6_WUNLOCK();
1049 
1050 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1051 		LIST_REMOVE(pr, ndpr_entry);
1052 		nd6_prefix_del(pr);
1053 	}
1054 
1055 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1056 	    nd6_timer, curvnet);
1057 
1058 	CURVNET_RESTORE();
1059 }
1060 
1061 /*
1062  * ia6 - deprecated/invalidated temporary address
1063  */
1064 static int
1065 regen_tmpaddr(struct in6_ifaddr *ia6)
1066 {
1067 	struct ifaddr *ifa;
1068 	struct ifnet *ifp;
1069 	struct in6_ifaddr *public_ifa6 = NULL;
1070 
1071 	NET_EPOCH_ASSERT();
1072 
1073 	ifp = ia6->ia_ifa.ifa_ifp;
1074 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1075 		struct in6_ifaddr *it6;
1076 
1077 		if (ifa->ifa_addr->sa_family != AF_INET6)
1078 			continue;
1079 
1080 		it6 = (struct in6_ifaddr *)ifa;
1081 
1082 		/* ignore no autoconf addresses. */
1083 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1084 			continue;
1085 
1086 		/* ignore autoconf addresses with different prefixes. */
1087 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1088 			continue;
1089 
1090 		/*
1091 		 * Now we are looking at an autoconf address with the same
1092 		 * prefix as ours.  If the address is temporary and is still
1093 		 * preferred, do not create another one.  It would be rare, but
1094 		 * could happen, for example, when we resume a laptop PC after
1095 		 * a long period.
1096 		 */
1097 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1098 		    !IFA6_IS_DEPRECATED(it6)) {
1099 			public_ifa6 = NULL;
1100 			break;
1101 		}
1102 
1103 		/*
1104 		 * This is a public autoconf address that has the same prefix
1105 		 * as ours.  If it is preferred, keep it.  We can't break the
1106 		 * loop here, because there may be a still-preferred temporary
1107 		 * address with the prefix.
1108 		 */
1109 		if (!IFA6_IS_DEPRECATED(it6))
1110 			public_ifa6 = it6;
1111 	}
1112 	if (public_ifa6 != NULL)
1113 		ifa_ref(&public_ifa6->ia_ifa);
1114 
1115 	if (public_ifa6 != NULL) {
1116 		int e;
1117 
1118 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1119 			ifa_free(&public_ifa6->ia_ifa);
1120 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1121 			    " tmp addr,errno=%d\n", e);
1122 			return (-1);
1123 		}
1124 		ifa_free(&public_ifa6->ia_ifa);
1125 		return (0);
1126 	}
1127 
1128 	return (-1);
1129 }
1130 
1131 /*
1132  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1133  * cache entries are freed in in6_domifdetach().
1134  */
1135 void
1136 nd6_purge(struct ifnet *ifp)
1137 {
1138 	struct nd_prhead prl;
1139 	struct nd_prefix *pr, *npr;
1140 
1141 	LIST_INIT(&prl);
1142 
1143 	/* Purge default router list entries toward ifp. */
1144 	nd6_defrouter_purge(ifp);
1145 
1146 	ND6_WLOCK();
1147 	/*
1148 	 * Remove prefixes on ifp. We should have already removed addresses on
1149 	 * this interface, so no addresses should be referencing these prefixes.
1150 	 */
1151 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1152 		if (pr->ndpr_ifp == ifp)
1153 			nd6_prefix_unlink(pr, &prl);
1154 	}
1155 	ND6_WUNLOCK();
1156 
1157 	/* Delete the unlinked prefix objects. */
1158 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1159 		LIST_REMOVE(pr, ndpr_entry);
1160 		nd6_prefix_del(pr);
1161 	}
1162 
1163 	/* cancel default outgoing interface setting */
1164 	if (V_nd6_defifindex == ifp->if_index)
1165 		nd6_setdefaultiface(0);
1166 
1167 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1168 		/* Refresh default router list. */
1169 		defrouter_select_fib(ifp->if_fib);
1170 	}
1171 }
1172 
1173 /*
1174  * the caller acquires and releases the lock on the lltbls
1175  * Returns the llentry locked
1176  */
1177 struct llentry *
1178 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1179 {
1180 	struct sockaddr_in6 sin6;
1181 	struct llentry *ln;
1182 
1183 	bzero(&sin6, sizeof(sin6));
1184 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1185 	sin6.sin6_family = AF_INET6;
1186 	sin6.sin6_addr = *addr6;
1187 
1188 	IF_AFDATA_LOCK_ASSERT(ifp);
1189 
1190 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1191 
1192 	return (ln);
1193 }
1194 
1195 static struct llentry *
1196 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1197 {
1198 	struct sockaddr_in6 sin6;
1199 	struct llentry *ln;
1200 
1201 	bzero(&sin6, sizeof(sin6));
1202 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1203 	sin6.sin6_family = AF_INET6;
1204 	sin6.sin6_addr = *addr6;
1205 
1206 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1207 	if (ln != NULL)
1208 		ln->ln_state = ND6_LLINFO_NOSTATE;
1209 
1210 	return (ln);
1211 }
1212 
1213 /*
1214  * Test whether a given IPv6 address is a neighbor or not, ignoring
1215  * the actual neighbor cache.  The neighbor cache is ignored in order
1216  * to not reenter the routing code from within itself.
1217  */
1218 static int
1219 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1220 {
1221 	struct nd_prefix *pr;
1222 	struct ifaddr *ifa;
1223 	struct rt_addrinfo info;
1224 	struct sockaddr_in6 rt_key;
1225 	const struct sockaddr *dst6;
1226 	uint64_t genid;
1227 	int error, fibnum;
1228 
1229 	/*
1230 	 * A link-local address is always a neighbor.
1231 	 * XXX: a link does not necessarily specify a single interface.
1232 	 */
1233 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1234 		struct sockaddr_in6 sin6_copy;
1235 		u_int32_t zone;
1236 
1237 		/*
1238 		 * We need sin6_copy since sa6_recoverscope() may modify the
1239 		 * content (XXX).
1240 		 */
1241 		sin6_copy = *addr;
1242 		if (sa6_recoverscope(&sin6_copy))
1243 			return (0); /* XXX: should be impossible */
1244 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1245 			return (0);
1246 		if (sin6_copy.sin6_scope_id == zone)
1247 			return (1);
1248 		else
1249 			return (0);
1250 	}
1251 
1252 	bzero(&rt_key, sizeof(rt_key));
1253 	bzero(&info, sizeof(info));
1254 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1255 
1256 	/*
1257 	 * If the address matches one of our addresses,
1258 	 * it should be a neighbor.
1259 	 * If the address matches one of our on-link prefixes, it should be a
1260 	 * neighbor.
1261 	 */
1262 	ND6_RLOCK();
1263 restart:
1264 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1265 		if (pr->ndpr_ifp != ifp)
1266 			continue;
1267 
1268 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1269 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1270 
1271 			/*
1272 			 * We only need to check all FIBs if add_addr_allfibs
1273 			 * is unset. If set, checking any FIB will suffice.
1274 			 */
1275 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1276 			for (; fibnum < rt_numfibs; fibnum++) {
1277 				genid = V_nd6_list_genid;
1278 				ND6_RUNLOCK();
1279 
1280 				/*
1281 				 * Restore length field before
1282 				 * retrying lookup
1283 				 */
1284 				rt_key.sin6_len = sizeof(rt_key);
1285 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1286 						        &info);
1287 
1288 				ND6_RLOCK();
1289 				if (genid != V_nd6_list_genid)
1290 					goto restart;
1291 				if (error == 0)
1292 					break;
1293 			}
1294 			if (error != 0)
1295 				continue;
1296 
1297 			/*
1298 			 * This is the case where multiple interfaces
1299 			 * have the same prefix, but only one is installed
1300 			 * into the routing table and that prefix entry
1301 			 * is not the one being examined here. In the case
1302 			 * where RADIX_MPATH is enabled, multiple route
1303 			 * entries (of the same rt_key value) will be
1304 			 * installed because the interface addresses all
1305 			 * differ.
1306 			 */
1307 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1308 			    &rt_key.sin6_addr))
1309 				continue;
1310 		}
1311 
1312 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1313 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1314 			ND6_RUNLOCK();
1315 			return (1);
1316 		}
1317 	}
1318 	ND6_RUNLOCK();
1319 
1320 	/*
1321 	 * If the address is assigned on the node of the other side of
1322 	 * a p2p interface, the address should be a neighbor.
1323 	 */
1324 	if (ifp->if_flags & IFF_POINTOPOINT) {
1325 		struct epoch_tracker et;
1326 
1327 		NET_EPOCH_ENTER(et);
1328 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1329 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1330 				continue;
1331 			if (ifa->ifa_dstaddr != NULL &&
1332 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1333 				NET_EPOCH_EXIT(et);
1334 				return 1;
1335 			}
1336 		}
1337 		NET_EPOCH_EXIT(et);
1338 	}
1339 
1340 	/*
1341 	 * If the default router list is empty, all addresses are regarded
1342 	 * as on-link, and thus, as a neighbor.
1343 	 */
1344 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1345 	    nd6_defrouter_list_empty() &&
1346 	    V_nd6_defifindex == ifp->if_index) {
1347 		return (1);
1348 	}
1349 
1350 	return (0);
1351 }
1352 
1353 
1354 /*
1355  * Detect if a given IPv6 address identifies a neighbor on a given link.
1356  * XXX: should take care of the destination of a p2p link?
1357  */
1358 int
1359 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1360 {
1361 	struct llentry *lle;
1362 	int rc = 0;
1363 
1364 	NET_EPOCH_ASSERT();
1365 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1366 	if (nd6_is_new_addr_neighbor(addr, ifp))
1367 		return (1);
1368 
1369 	/*
1370 	 * Even if the address matches none of our addresses, it might be
1371 	 * in the neighbor cache.
1372 	 */
1373 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1374 		LLE_RUNLOCK(lle);
1375 		rc = 1;
1376 	}
1377 	return (rc);
1378 }
1379 
1380 /*
1381  * Free an nd6 llinfo entry.
1382  * Since the function would cause significant changes in the kernel, DO NOT
1383  * make it global, unless you have a strong reason for the change, and are sure
1384  * that the change is safe.
1385  *
1386  * Set noinline to be dtrace-friendly
1387  */
1388 static __noinline void
1389 nd6_free(struct llentry **lnp, int gc)
1390 {
1391 	struct ifnet *ifp;
1392 	struct llentry *ln;
1393 	struct nd_defrouter *dr;
1394 
1395 	ln = *lnp;
1396 	*lnp = NULL;
1397 
1398 	LLE_WLOCK_ASSERT(ln);
1399 	ND6_RLOCK_ASSERT();
1400 
1401 	ifp = lltable_get_ifp(ln->lle_tbl);
1402 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1403 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1404 	else
1405 		dr = NULL;
1406 	ND6_RUNLOCK();
1407 
1408 	if ((ln->la_flags & LLE_DELETED) == 0)
1409 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1410 
1411 	/*
1412 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1413 	 * even though it is not harmful, it was not really necessary.
1414 	 */
1415 
1416 	/* cancel timer */
1417 	nd6_llinfo_settimer_locked(ln, -1);
1418 
1419 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1420 		if (dr != NULL && dr->expire &&
1421 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1422 			/*
1423 			 * If the reason for the deletion is just garbage
1424 			 * collection, and the neighbor is an active default
1425 			 * router, do not delete it.  Instead, reset the GC
1426 			 * timer using the router's lifetime.
1427 			 * Simply deleting the entry would affect default
1428 			 * router selection, which is not necessarily a good
1429 			 * thing, especially when we're using router preference
1430 			 * values.
1431 			 * XXX: the check for ln_state would be redundant,
1432 			 *      but we intentionally keep it just in case.
1433 			 */
1434 			if (dr->expire > time_uptime)
1435 				nd6_llinfo_settimer_locked(ln,
1436 				    (dr->expire - time_uptime) * hz);
1437 			else
1438 				nd6_llinfo_settimer_locked(ln,
1439 				    (long)V_nd6_gctimer * hz);
1440 
1441 			LLE_REMREF(ln);
1442 			LLE_WUNLOCK(ln);
1443 			defrouter_rele(dr);
1444 			return;
1445 		}
1446 
1447 		if (dr) {
1448 			/*
1449 			 * Unreachablity of a router might affect the default
1450 			 * router selection and on-link detection of advertised
1451 			 * prefixes.
1452 			 */
1453 
1454 			/*
1455 			 * Temporarily fake the state to choose a new default
1456 			 * router and to perform on-link determination of
1457 			 * prefixes correctly.
1458 			 * Below the state will be set correctly,
1459 			 * or the entry itself will be deleted.
1460 			 */
1461 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1462 		}
1463 
1464 		if (ln->ln_router || dr) {
1465 
1466 			/*
1467 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1468 			 * rnh and for the calls to pfxlist_onlink_check() and
1469 			 * defrouter_select_fib() in the block further down for calls
1470 			 * into nd6_lookup().  We still hold a ref.
1471 			 */
1472 			LLE_WUNLOCK(ln);
1473 
1474 			/*
1475 			 * rt6_flush must be called whether or not the neighbor
1476 			 * is in the Default Router List.
1477 			 * See a corresponding comment in nd6_na_input().
1478 			 */
1479 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1480 		}
1481 
1482 		if (dr) {
1483 			/*
1484 			 * Since defrouter_select_fib() does not affect the
1485 			 * on-link determination and MIP6 needs the check
1486 			 * before the default router selection, we perform
1487 			 * the check now.
1488 			 */
1489 			pfxlist_onlink_check();
1490 
1491 			/*
1492 			 * Refresh default router list.
1493 			 */
1494 			defrouter_select_fib(dr->ifp->if_fib);
1495 		}
1496 
1497 		/*
1498 		 * If this entry was added by an on-link redirect, remove the
1499 		 * corresponding host route.
1500 		 */
1501 		if (ln->la_flags & LLE_REDIRECT)
1502 			nd6_free_redirect(ln);
1503 
1504 		if (ln->ln_router || dr)
1505 			LLE_WLOCK(ln);
1506 	}
1507 
1508 	/*
1509 	 * Save to unlock. We still hold an extra reference and will not
1510 	 * free(9) in llentry_free() if someone else holds one as well.
1511 	 */
1512 	LLE_WUNLOCK(ln);
1513 	IF_AFDATA_LOCK(ifp);
1514 	LLE_WLOCK(ln);
1515 	/* Guard against race with other llentry_free(). */
1516 	if (ln->la_flags & LLE_LINKED) {
1517 		/* Remove callout reference */
1518 		LLE_REMREF(ln);
1519 		lltable_unlink_entry(ln->lle_tbl, ln);
1520 	}
1521 	IF_AFDATA_UNLOCK(ifp);
1522 
1523 	llentry_free(ln);
1524 	if (dr != NULL)
1525 		defrouter_rele(dr);
1526 }
1527 
1528 static int
1529 nd6_isdynrte(const struct rtentry *rt, void *xap)
1530 {
1531 
1532 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1533 		return (1);
1534 
1535 	return (0);
1536 }
1537 /*
1538  * Remove the rtentry for the given llentry,
1539  * both of which were installed by a redirect.
1540  */
1541 static void
1542 nd6_free_redirect(const struct llentry *ln)
1543 {
1544 	int fibnum;
1545 	struct sockaddr_in6 sin6;
1546 	struct rt_addrinfo info;
1547 
1548 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1549 	memset(&info, 0, sizeof(info));
1550 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1551 	info.rti_filter = nd6_isdynrte;
1552 
1553 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1554 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1555 }
1556 
1557 /*
1558  * Rejuvenate this function for routing operations related
1559  * processing.
1560  */
1561 void
1562 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1563 {
1564 	struct sockaddr_in6 *gateway;
1565 	struct nd_defrouter *dr;
1566 	struct ifnet *ifp;
1567 
1568 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1569 	ifp = rt->rt_ifp;
1570 
1571 	switch (req) {
1572 	case RTM_ADD:
1573 		break;
1574 
1575 	case RTM_DELETE:
1576 		if (!ifp)
1577 			return;
1578 		/*
1579 		 * Only indirect routes are interesting.
1580 		 */
1581 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1582 			return;
1583 		/*
1584 		 * check for default route
1585 		 */
1586 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1587 		    &SIN6(rt_key(rt))->sin6_addr)) {
1588 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1589 			if (dr != NULL) {
1590 				dr->installed = 0;
1591 				defrouter_rele(dr);
1592 			}
1593 		}
1594 		break;
1595 	}
1596 }
1597 
1598 
1599 int
1600 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1601 {
1602 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1603 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1604 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1605 	struct epoch_tracker et;
1606 	int error = 0;
1607 
1608 	if (ifp->if_afdata[AF_INET6] == NULL)
1609 		return (EPFNOSUPPORT);
1610 	switch (cmd) {
1611 	case OSIOCGIFINFO_IN6:
1612 #define ND	ndi->ndi
1613 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1614 		bzero(&ND, sizeof(ND));
1615 		ND.linkmtu = IN6_LINKMTU(ifp);
1616 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1617 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1618 		ND.reachable = ND_IFINFO(ifp)->reachable;
1619 		ND.retrans = ND_IFINFO(ifp)->retrans;
1620 		ND.flags = ND_IFINFO(ifp)->flags;
1621 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1622 		ND.chlim = ND_IFINFO(ifp)->chlim;
1623 		break;
1624 	case SIOCGIFINFO_IN6:
1625 		ND = *ND_IFINFO(ifp);
1626 		break;
1627 	case SIOCSIFINFO_IN6:
1628 		/*
1629 		 * used to change host variables from userland.
1630 		 * intended for a use on router to reflect RA configurations.
1631 		 */
1632 		/* 0 means 'unspecified' */
1633 		if (ND.linkmtu != 0) {
1634 			if (ND.linkmtu < IPV6_MMTU ||
1635 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1636 				error = EINVAL;
1637 				break;
1638 			}
1639 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1640 		}
1641 
1642 		if (ND.basereachable != 0) {
1643 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1644 
1645 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1646 			if (ND.basereachable != obasereachable)
1647 				ND_IFINFO(ifp)->reachable =
1648 				    ND_COMPUTE_RTIME(ND.basereachable);
1649 		}
1650 		if (ND.retrans != 0)
1651 			ND_IFINFO(ifp)->retrans = ND.retrans;
1652 		if (ND.chlim != 0)
1653 			ND_IFINFO(ifp)->chlim = ND.chlim;
1654 		/* FALLTHROUGH */
1655 	case SIOCSIFINFO_FLAGS:
1656 	{
1657 		struct ifaddr *ifa;
1658 		struct in6_ifaddr *ia;
1659 
1660 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1661 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1662 			/* ifdisabled 1->0 transision */
1663 
1664 			/*
1665 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1666 			 * has an link-local address with IN6_IFF_DUPLICATED,
1667 			 * do not clear ND6_IFF_IFDISABLED.
1668 			 * See RFC 4862, Section 5.4.5.
1669 			 */
1670 			NET_EPOCH_ENTER(et);
1671 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1672 				if (ifa->ifa_addr->sa_family != AF_INET6)
1673 					continue;
1674 				ia = (struct in6_ifaddr *)ifa;
1675 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1676 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1677 					break;
1678 			}
1679 			NET_EPOCH_EXIT(et);
1680 
1681 			if (ifa != NULL) {
1682 				/* LLA is duplicated. */
1683 				ND.flags |= ND6_IFF_IFDISABLED;
1684 				log(LOG_ERR, "Cannot enable an interface"
1685 				    " with a link-local address marked"
1686 				    " duplicate.\n");
1687 			} else {
1688 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1689 				if (ifp->if_flags & IFF_UP)
1690 					in6_if_up(ifp);
1691 			}
1692 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1693 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1694 			/* ifdisabled 0->1 transision */
1695 			/* Mark all IPv6 address as tentative. */
1696 
1697 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1698 			if (V_ip6_dad_count > 0 &&
1699 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1700 				NET_EPOCH_ENTER(et);
1701 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1702 				    ifa_link) {
1703 					if (ifa->ifa_addr->sa_family !=
1704 					    AF_INET6)
1705 						continue;
1706 					ia = (struct in6_ifaddr *)ifa;
1707 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1708 				}
1709 				NET_EPOCH_EXIT(et);
1710 			}
1711 		}
1712 
1713 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1714 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1715 				/* auto_linklocal 0->1 transision */
1716 
1717 				/* If no link-local address on ifp, configure */
1718 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1719 				in6_ifattach(ifp, NULL);
1720 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1721 			    ifp->if_flags & IFF_UP) {
1722 				/*
1723 				 * When the IF already has
1724 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1725 				 * address is assigned, and IFF_UP, try to
1726 				 * assign one.
1727 				 */
1728 				NET_EPOCH_ENTER(et);
1729 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1730 				    ifa_link) {
1731 					if (ifa->ifa_addr->sa_family !=
1732 					    AF_INET6)
1733 						continue;
1734 					ia = (struct in6_ifaddr *)ifa;
1735 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1736 						break;
1737 				}
1738 				NET_EPOCH_EXIT(et);
1739 				if (ifa != NULL)
1740 					/* No LLA is configured. */
1741 					in6_ifattach(ifp, NULL);
1742 			}
1743 		}
1744 		ND_IFINFO(ifp)->flags = ND.flags;
1745 		break;
1746 	}
1747 #undef ND
1748 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1749 		/* sync kernel routing table with the default router list */
1750 		defrouter_reset();
1751 		defrouter_select_fib(RT_ALL_FIBS);
1752 		break;
1753 	case SIOCSPFXFLUSH_IN6:
1754 	{
1755 		/* flush all the prefix advertised by routers */
1756 		struct in6_ifaddr *ia, *ia_next;
1757 		struct nd_prefix *pr, *next;
1758 		struct nd_prhead prl;
1759 
1760 		LIST_INIT(&prl);
1761 
1762 		ND6_WLOCK();
1763 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1764 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1765 				continue; /* XXX */
1766 			nd6_prefix_unlink(pr, &prl);
1767 		}
1768 		ND6_WUNLOCK();
1769 
1770 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1771 			LIST_REMOVE(pr, ndpr_entry);
1772 			/* XXXRW: in6_ifaddrhead locking. */
1773 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1774 			    ia_next) {
1775 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1776 					continue;
1777 
1778 				if (ia->ia6_ndpr == pr)
1779 					in6_purgeaddr(&ia->ia_ifa);
1780 			}
1781 			nd6_prefix_del(pr);
1782 		}
1783 		break;
1784 	}
1785 	case SIOCSRTRFLUSH_IN6:
1786 	{
1787 		/* flush all the default routers */
1788 
1789 		defrouter_reset();
1790 		nd6_defrouter_flush_all();
1791 		defrouter_select_fib(RT_ALL_FIBS);
1792 		break;
1793 	}
1794 	case SIOCGNBRINFO_IN6:
1795 	{
1796 		struct llentry *ln;
1797 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1798 
1799 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1800 			return (error);
1801 
1802 		NET_EPOCH_ENTER(et);
1803 		ln = nd6_lookup(&nb_addr, 0, ifp);
1804 		NET_EPOCH_EXIT(et);
1805 
1806 		if (ln == NULL) {
1807 			error = EINVAL;
1808 			break;
1809 		}
1810 		nbi->state = ln->ln_state;
1811 		nbi->asked = ln->la_asked;
1812 		nbi->isrouter = ln->ln_router;
1813 		if (ln->la_expire == 0)
1814 			nbi->expire = 0;
1815 		else
1816 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1817 			    (time_second - time_uptime);
1818 		LLE_RUNLOCK(ln);
1819 		break;
1820 	}
1821 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1822 		ndif->ifindex = V_nd6_defifindex;
1823 		break;
1824 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1825 		return (nd6_setdefaultiface(ndif->ifindex));
1826 	}
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Calculates new isRouter value based on provided parameters and
1832  * returns it.
1833  */
1834 static int
1835 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1836     int ln_router)
1837 {
1838 
1839 	/*
1840 	 * ICMP6 type dependent behavior.
1841 	 *
1842 	 * NS: clear IsRouter if new entry
1843 	 * RS: clear IsRouter
1844 	 * RA: set IsRouter if there's lladdr
1845 	 * redir: clear IsRouter if new entry
1846 	 *
1847 	 * RA case, (1):
1848 	 * The spec says that we must set IsRouter in the following cases:
1849 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1850 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1851 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1852 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1853 	 * neighbor cache, this is similar to (6).
1854 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1855 	 *
1856 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1857 	 *							D R
1858 	 *	0	n	n	(1)	c   ?     s
1859 	 *	0	y	n	(2)	c   s     s
1860 	 *	0	n	y	(3)	c   s     s
1861 	 *	0	y	y	(4)	c   s     s
1862 	 *	0	y	y	(5)	c   s     s
1863 	 *	1	--	n	(6) c	c	c s
1864 	 *	1	--	y	(7) c	c   s	c s
1865 	 *
1866 	 *					(c=clear s=set)
1867 	 */
1868 	switch (type & 0xff) {
1869 	case ND_NEIGHBOR_SOLICIT:
1870 		/*
1871 		 * New entry must have is_router flag cleared.
1872 		 */
1873 		if (is_new)					/* (6-7) */
1874 			ln_router = 0;
1875 		break;
1876 	case ND_REDIRECT:
1877 		/*
1878 		 * If the icmp is a redirect to a better router, always set the
1879 		 * is_router flag.  Otherwise, if the entry is newly created,
1880 		 * clear the flag.  [RFC 2461, sec 8.3]
1881 		 */
1882 		if (code == ND_REDIRECT_ROUTER)
1883 			ln_router = 1;
1884 		else {
1885 			if (is_new)				/* (6-7) */
1886 				ln_router = 0;
1887 		}
1888 		break;
1889 	case ND_ROUTER_SOLICIT:
1890 		/*
1891 		 * is_router flag must always be cleared.
1892 		 */
1893 		ln_router = 0;
1894 		break;
1895 	case ND_ROUTER_ADVERT:
1896 		/*
1897 		 * Mark an entry with lladdr as a router.
1898 		 */
1899 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1900 		    (is_new && new_addr)) {			/* (7) */
1901 			ln_router = 1;
1902 		}
1903 		break;
1904 	}
1905 
1906 	return (ln_router);
1907 }
1908 
1909 /*
1910  * Create neighbor cache entry and cache link-layer address,
1911  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1912  *
1913  * type - ICMP6 type
1914  * code - type dependent information
1915  *
1916  */
1917 void
1918 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1919     int lladdrlen, int type, int code)
1920 {
1921 	struct llentry *ln = NULL, *ln_tmp;
1922 	int is_newentry;
1923 	int do_update;
1924 	int olladdr;
1925 	int llchange;
1926 	int flags;
1927 	uint16_t router = 0;
1928 	struct sockaddr_in6 sin6;
1929 	struct mbuf *chain = NULL;
1930 	u_char linkhdr[LLE_MAX_LINKHDR];
1931 	size_t linkhdrsize;
1932 	int lladdr_off;
1933 
1934 	NET_EPOCH_ASSERT();
1935 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1936 
1937 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1938 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1939 
1940 	/* nothing must be updated for unspecified address */
1941 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1942 		return;
1943 
1944 	/*
1945 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1946 	 * the caller.
1947 	 *
1948 	 * XXX If the link does not have link-layer adderss, what should
1949 	 * we do? (ifp->if_addrlen == 0)
1950 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1951 	 * description on it in NS section (RFC 2461 7.2.3).
1952 	 */
1953 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1954 	ln = nd6_lookup(from, flags, ifp);
1955 	is_newentry = 0;
1956 	if (ln == NULL) {
1957 		flags |= LLE_EXCLUSIVE;
1958 		ln = nd6_alloc(from, 0, ifp);
1959 		if (ln == NULL)
1960 			return;
1961 
1962 		/*
1963 		 * Since we already know all the data for the new entry,
1964 		 * fill it before insertion.
1965 		 */
1966 		if (lladdr != NULL) {
1967 			linkhdrsize = sizeof(linkhdr);
1968 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1969 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1970 				return;
1971 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1972 			    lladdr_off);
1973 		}
1974 
1975 		IF_AFDATA_WLOCK(ifp);
1976 		LLE_WLOCK(ln);
1977 		/* Prefer any existing lle over newly-created one */
1978 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1979 		if (ln_tmp == NULL)
1980 			lltable_link_entry(LLTABLE6(ifp), ln);
1981 		IF_AFDATA_WUNLOCK(ifp);
1982 		if (ln_tmp == NULL) {
1983 			/* No existing lle, mark as new entry (6,7) */
1984 			is_newentry = 1;
1985 			if (lladdr != NULL) {	/* (7) */
1986 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1987 				EVENTHANDLER_INVOKE(lle_event, ln,
1988 				    LLENTRY_RESOLVED);
1989 			}
1990 		} else {
1991 			lltable_free_entry(LLTABLE6(ifp), ln);
1992 			ln = ln_tmp;
1993 			ln_tmp = NULL;
1994 		}
1995 	}
1996 	/* do nothing if static ndp is set */
1997 	if ((ln->la_flags & LLE_STATIC)) {
1998 		if (flags & LLE_EXCLUSIVE)
1999 			LLE_WUNLOCK(ln);
2000 		else
2001 			LLE_RUNLOCK(ln);
2002 		return;
2003 	}
2004 
2005 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2006 	if (olladdr && lladdr) {
2007 		llchange = bcmp(lladdr, ln->ll_addr,
2008 		    ifp->if_addrlen);
2009 	} else if (!olladdr && lladdr)
2010 		llchange = 1;
2011 	else
2012 		llchange = 0;
2013 
2014 	/*
2015 	 * newentry olladdr  lladdr  llchange	(*=record)
2016 	 *	0	n	n	--	(1)
2017 	 *	0	y	n	--	(2)
2018 	 *	0	n	y	y	(3) * STALE
2019 	 *	0	y	y	n	(4) *
2020 	 *	0	y	y	y	(5) * STALE
2021 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2022 	 *	1	--	y	--	(7) * STALE
2023 	 */
2024 
2025 	do_update = 0;
2026 	if (is_newentry == 0 && llchange != 0) {
2027 		do_update = 1;	/* (3,5) */
2028 
2029 		/*
2030 		 * Record source link-layer address
2031 		 * XXX is it dependent to ifp->if_type?
2032 		 */
2033 		linkhdrsize = sizeof(linkhdr);
2034 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2035 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2036 			return;
2037 
2038 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2039 		    lladdr_off) == 0) {
2040 			/* Entry was deleted */
2041 			return;
2042 		}
2043 
2044 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2045 
2046 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2047 
2048 		if (ln->la_hold != NULL)
2049 			nd6_grab_holdchain(ln, &chain, &sin6);
2050 	}
2051 
2052 	/* Calculates new router status */
2053 	router = nd6_is_router(type, code, is_newentry, olladdr,
2054 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2055 
2056 	ln->ln_router = router;
2057 	/* Mark non-router redirects with special flag */
2058 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2059 		ln->la_flags |= LLE_REDIRECT;
2060 
2061 	if (flags & LLE_EXCLUSIVE)
2062 		LLE_WUNLOCK(ln);
2063 	else
2064 		LLE_RUNLOCK(ln);
2065 
2066 	if (chain != NULL)
2067 		nd6_flush_holdchain(ifp, chain, &sin6);
2068 
2069 	/*
2070 	 * When the link-layer address of a router changes, select the
2071 	 * best router again.  In particular, when the neighbor entry is newly
2072 	 * created, it might affect the selection policy.
2073 	 * Question: can we restrict the first condition to the "is_newentry"
2074 	 * case?
2075 	 * XXX: when we hear an RA from a new router with the link-layer
2076 	 * address option, defrouter_select_fib() is called twice, since
2077 	 * defrtrlist_update called the function as well.  However, I believe
2078 	 * we can compromise the overhead, since it only happens the first
2079 	 * time.
2080 	 * XXX: although defrouter_select_fib() should not have a bad effect
2081 	 * for those are not autoconfigured hosts, we explicitly avoid such
2082 	 * cases for safety.
2083 	 */
2084 	if ((do_update || is_newentry) && router &&
2085 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2086 		/*
2087 		 * guaranteed recursion
2088 		 */
2089 		defrouter_select_fib(ifp->if_fib);
2090 	}
2091 }
2092 
2093 static void
2094 nd6_slowtimo(void *arg)
2095 {
2096 	struct epoch_tracker et;
2097 	CURVNET_SET((struct vnet *) arg);
2098 	struct nd_ifinfo *nd6if;
2099 	struct ifnet *ifp;
2100 
2101 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2102 	    nd6_slowtimo, curvnet);
2103 	NET_EPOCH_ENTER(et);
2104 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2105 		if (ifp->if_afdata[AF_INET6] == NULL)
2106 			continue;
2107 		nd6if = ND_IFINFO(ifp);
2108 		if (nd6if->basereachable && /* already initialized */
2109 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2110 			/*
2111 			 * Since reachable time rarely changes by router
2112 			 * advertisements, we SHOULD insure that a new random
2113 			 * value gets recomputed at least once every few hours.
2114 			 * (RFC 2461, 6.3.4)
2115 			 */
2116 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2117 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2118 		}
2119 	}
2120 	NET_EPOCH_EXIT(et);
2121 	CURVNET_RESTORE();
2122 }
2123 
2124 void
2125 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2126     struct sockaddr_in6 *sin6)
2127 {
2128 
2129 	LLE_WLOCK_ASSERT(ln);
2130 
2131 	*chain = ln->la_hold;
2132 	ln->la_hold = NULL;
2133 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2134 
2135 	if (ln->ln_state == ND6_LLINFO_STALE) {
2136 
2137 		/*
2138 		 * The first time we send a packet to a
2139 		 * neighbor whose entry is STALE, we have
2140 		 * to change the state to DELAY and a sets
2141 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2142 		 * seconds to ensure do neighbor unreachability
2143 		 * detection on expiration.
2144 		 * (RFC 2461 7.3.3)
2145 		 */
2146 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2147 	}
2148 }
2149 
2150 int
2151 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2152     struct sockaddr_in6 *dst, struct route *ro)
2153 {
2154 	int error;
2155 	int ip6len;
2156 	struct ip6_hdr *ip6;
2157 	struct m_tag *mtag;
2158 
2159 #ifdef MAC
2160 	mac_netinet6_nd6_send(ifp, m);
2161 #endif
2162 
2163 	/*
2164 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2165 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2166 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2167 	 * to be diverted to user space.  When re-injected into the kernel,
2168 	 * send_output() will directly dispatch them to the outgoing interface.
2169 	 */
2170 	if (send_sendso_input_hook != NULL) {
2171 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2172 		if (mtag != NULL) {
2173 			ip6 = mtod(m, struct ip6_hdr *);
2174 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2175 			/* Use the SEND socket */
2176 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2177 			    ip6len);
2178 			/* -1 == no app on SEND socket */
2179 			if (error == 0 || error != -1)
2180 			    return (error);
2181 		}
2182 	}
2183 
2184 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2185 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2186 	    mtod(m, struct ip6_hdr *));
2187 
2188 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2189 		origifp = ifp;
2190 
2191 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2192 	return (error);
2193 }
2194 
2195 /*
2196  * Lookup link headerfor @sa_dst address. Stores found
2197  * data in @desten buffer. Copy of lle ln_flags can be also
2198  * saved in @pflags if @pflags is non-NULL.
2199  *
2200  * If destination LLE does not exists or lle state modification
2201  * is required, call "slow" version.
2202  *
2203  * Return values:
2204  * - 0 on success (address copied to buffer).
2205  * - EWOULDBLOCK (no local error, but address is still unresolved)
2206  * - other errors (alloc failure, etc)
2207  */
2208 int
2209 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2210     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2211     struct llentry **plle)
2212 {
2213 	struct llentry *ln = NULL;
2214 	const struct sockaddr_in6 *dst6;
2215 
2216 	NET_EPOCH_ASSERT();
2217 
2218 	if (pflags != NULL)
2219 		*pflags = 0;
2220 
2221 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2222 
2223 	/* discard the packet if IPv6 operation is disabled on the interface */
2224 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2225 		m_freem(m);
2226 		return (ENETDOWN); /* better error? */
2227 	}
2228 
2229 	if (m != NULL && m->m_flags & M_MCAST) {
2230 		switch (ifp->if_type) {
2231 		case IFT_ETHER:
2232 		case IFT_L2VLAN:
2233 		case IFT_BRIDGE:
2234 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2235 						 desten);
2236 			return (0);
2237 		default:
2238 			m_freem(m);
2239 			return (EAFNOSUPPORT);
2240 		}
2241 	}
2242 
2243 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2244 	    ifp);
2245 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2246 		/* Entry found, let's copy lle info */
2247 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2248 		if (pflags != NULL)
2249 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2250 		/* Check if we have feedback request from nd6 timer */
2251 		if (ln->r_skip_req != 0) {
2252 			LLE_REQ_LOCK(ln);
2253 			ln->r_skip_req = 0; /* Notify that entry was used */
2254 			ln->lle_hittime = time_uptime;
2255 			LLE_REQ_UNLOCK(ln);
2256 		}
2257 		if (plle) {
2258 			LLE_ADDREF(ln);
2259 			*plle = ln;
2260 			LLE_WUNLOCK(ln);
2261 		}
2262 		return (0);
2263 	} else if (plle && ln)
2264 		LLE_WUNLOCK(ln);
2265 
2266 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2267 }
2268 
2269 
2270 /*
2271  * Do L2 address resolution for @sa_dst address. Stores found
2272  * address in @desten buffer. Copy of lle ln_flags can be also
2273  * saved in @pflags if @pflags is non-NULL.
2274  *
2275  * Heavy version.
2276  * Function assume that destination LLE does not exist,
2277  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2278  *
2279  * Set noinline to be dtrace-friendly
2280  */
2281 static __noinline int
2282 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2283     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2284     struct llentry **plle)
2285 {
2286 	struct llentry *lle = NULL, *lle_tmp;
2287 	struct in6_addr *psrc, src;
2288 	int send_ns, ll_len;
2289 	char *lladdr;
2290 
2291 	NET_EPOCH_ASSERT();
2292 
2293 	/*
2294 	 * Address resolution or Neighbor Unreachability Detection
2295 	 * for the next hop.
2296 	 * At this point, the destination of the packet must be a unicast
2297 	 * or an anycast address(i.e. not a multicast).
2298 	 */
2299 	if (lle == NULL) {
2300 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2301 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2302 			/*
2303 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2304 			 * the condition below is not very efficient.  But we believe
2305 			 * it is tolerable, because this should be a rare case.
2306 			 */
2307 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2308 			if (lle == NULL) {
2309 				char ip6buf[INET6_ADDRSTRLEN];
2310 				log(LOG_DEBUG,
2311 				    "nd6_output: can't allocate llinfo for %s "
2312 				    "(ln=%p)\n",
2313 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2314 				m_freem(m);
2315 				return (ENOBUFS);
2316 			}
2317 
2318 			IF_AFDATA_WLOCK(ifp);
2319 			LLE_WLOCK(lle);
2320 			/* Prefer any existing entry over newly-created one */
2321 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2322 			if (lle_tmp == NULL)
2323 				lltable_link_entry(LLTABLE6(ifp), lle);
2324 			IF_AFDATA_WUNLOCK(ifp);
2325 			if (lle_tmp != NULL) {
2326 				lltable_free_entry(LLTABLE6(ifp), lle);
2327 				lle = lle_tmp;
2328 				lle_tmp = NULL;
2329 			}
2330 		}
2331 	}
2332 	if (lle == NULL) {
2333 		m_freem(m);
2334 		return (ENOBUFS);
2335 	}
2336 
2337 	LLE_WLOCK_ASSERT(lle);
2338 
2339 	/*
2340 	 * The first time we send a packet to a neighbor whose entry is
2341 	 * STALE, we have to change the state to DELAY and a sets a timer to
2342 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2343 	 * neighbor unreachability detection on expiration.
2344 	 * (RFC 2461 7.3.3)
2345 	 */
2346 	if (lle->ln_state == ND6_LLINFO_STALE)
2347 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2348 
2349 	/*
2350 	 * If the neighbor cache entry has a state other than INCOMPLETE
2351 	 * (i.e. its link-layer address is already resolved), just
2352 	 * send the packet.
2353 	 */
2354 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2355 		if (flags & LLE_ADDRONLY) {
2356 			lladdr = lle->ll_addr;
2357 			ll_len = ifp->if_addrlen;
2358 		} else {
2359 			lladdr = lle->r_linkdata;
2360 			ll_len = lle->r_hdrlen;
2361 		}
2362 		bcopy(lladdr, desten, ll_len);
2363 		if (pflags != NULL)
2364 			*pflags = lle->la_flags;
2365 		if (plle) {
2366 			LLE_ADDREF(lle);
2367 			*plle = lle;
2368 		}
2369 		LLE_WUNLOCK(lle);
2370 		return (0);
2371 	}
2372 
2373 	/*
2374 	 * There is a neighbor cache entry, but no ethernet address
2375 	 * response yet.  Append this latest packet to the end of the
2376 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2377 	 * the oldest packet in the queue will be removed.
2378 	 */
2379 
2380 	if (lle->la_hold != NULL) {
2381 		struct mbuf *m_hold;
2382 		int i;
2383 
2384 		i = 0;
2385 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2386 			i++;
2387 			if (m_hold->m_nextpkt == NULL) {
2388 				m_hold->m_nextpkt = m;
2389 				break;
2390 			}
2391 		}
2392 		while (i >= V_nd6_maxqueuelen) {
2393 			m_hold = lle->la_hold;
2394 			lle->la_hold = lle->la_hold->m_nextpkt;
2395 			m_freem(m_hold);
2396 			i--;
2397 		}
2398 	} else {
2399 		lle->la_hold = m;
2400 	}
2401 
2402 	/*
2403 	 * If there has been no NS for the neighbor after entering the
2404 	 * INCOMPLETE state, send the first solicitation.
2405 	 * Note that for newly-created lle la_asked will be 0,
2406 	 * so we will transition from ND6_LLINFO_NOSTATE to
2407 	 * ND6_LLINFO_INCOMPLETE state here.
2408 	 */
2409 	psrc = NULL;
2410 	send_ns = 0;
2411 	if (lle->la_asked == 0) {
2412 		lle->la_asked++;
2413 		send_ns = 1;
2414 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2415 
2416 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2417 	}
2418 	LLE_WUNLOCK(lle);
2419 	if (send_ns != 0)
2420 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2421 
2422 	return (EWOULDBLOCK);
2423 }
2424 
2425 /*
2426  * Do L2 address resolution for @sa_dst address. Stores found
2427  * address in @desten buffer. Copy of lle ln_flags can be also
2428  * saved in @pflags if @pflags is non-NULL.
2429  *
2430  * Return values:
2431  * - 0 on success (address copied to buffer).
2432  * - EWOULDBLOCK (no local error, but address is still unresolved)
2433  * - other errors (alloc failure, etc)
2434  */
2435 int
2436 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2437     char *desten, uint32_t *pflags)
2438 {
2439 	int error;
2440 
2441 	flags |= LLE_ADDRONLY;
2442 	error = nd6_resolve_slow(ifp, flags, NULL,
2443 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2444 	return (error);
2445 }
2446 
2447 int
2448 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2449     struct sockaddr_in6 *dst)
2450 {
2451 	struct mbuf *m, *m_head;
2452 	int error = 0;
2453 
2454 	m_head = chain;
2455 
2456 	while (m_head) {
2457 		m = m_head;
2458 		m_head = m_head->m_nextpkt;
2459 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2460 	}
2461 
2462 	/*
2463 	 * XXX
2464 	 * note that intermediate errors are blindly ignored
2465 	 */
2466 	return (error);
2467 }
2468 
2469 static int
2470 nd6_need_cache(struct ifnet *ifp)
2471 {
2472 	/*
2473 	 * XXX: we currently do not make neighbor cache on any interface
2474 	 * other than Ethernet and GIF.
2475 	 *
2476 	 * RFC2893 says:
2477 	 * - unidirectional tunnels needs no ND
2478 	 */
2479 	switch (ifp->if_type) {
2480 	case IFT_ETHER:
2481 	case IFT_IEEE1394:
2482 	case IFT_L2VLAN:
2483 	case IFT_INFINIBAND:
2484 	case IFT_BRIDGE:
2485 	case IFT_PROPVIRTUAL:
2486 		return (1);
2487 	default:
2488 		return (0);
2489 	}
2490 }
2491 
2492 /*
2493  * Add pernament ND6 link-layer record for given
2494  * interface address.
2495  *
2496  * Very similar to IPv4 arp_ifinit(), but:
2497  * 1) IPv6 DAD is performed in different place
2498  * 2) It is called by IPv6 protocol stack in contrast to
2499  * arp_ifinit() which is typically called in SIOCSIFADDR
2500  * driver ioctl handler.
2501  *
2502  */
2503 int
2504 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2505 {
2506 	struct ifnet *ifp;
2507 	struct llentry *ln, *ln_tmp;
2508 	struct sockaddr *dst;
2509 
2510 	ifp = ia->ia_ifa.ifa_ifp;
2511 	if (nd6_need_cache(ifp) == 0)
2512 		return (0);
2513 
2514 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2515 	dst = (struct sockaddr *)&ia->ia_addr;
2516 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2517 	if (ln == NULL)
2518 		return (ENOBUFS);
2519 
2520 	IF_AFDATA_WLOCK(ifp);
2521 	LLE_WLOCK(ln);
2522 	/* Unlink any entry if exists */
2523 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2524 	if (ln_tmp != NULL)
2525 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2526 	lltable_link_entry(LLTABLE6(ifp), ln);
2527 	IF_AFDATA_WUNLOCK(ifp);
2528 
2529 	if (ln_tmp != NULL)
2530 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2531 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2532 
2533 	LLE_WUNLOCK(ln);
2534 	if (ln_tmp != NULL)
2535 		llentry_free(ln_tmp);
2536 
2537 	return (0);
2538 }
2539 
2540 /*
2541  * Removes either all lle entries for given @ia, or lle
2542  * corresponding to @ia address.
2543  */
2544 void
2545 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2546 {
2547 	struct sockaddr_in6 mask, addr;
2548 	struct sockaddr *saddr, *smask;
2549 	struct ifnet *ifp;
2550 
2551 	ifp = ia->ia_ifa.ifa_ifp;
2552 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2553 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2554 	saddr = (struct sockaddr *)&addr;
2555 	smask = (struct sockaddr *)&mask;
2556 
2557 	if (all != 0)
2558 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2559 	else
2560 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2561 }
2562 
2563 static void
2564 clear_llinfo_pqueue(struct llentry *ln)
2565 {
2566 	struct mbuf *m_hold, *m_hold_next;
2567 
2568 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2569 		m_hold_next = m_hold->m_nextpkt;
2570 		m_freem(m_hold);
2571 	}
2572 
2573 	ln->la_hold = NULL;
2574 }
2575 
2576 static int
2577 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2578 {
2579 	struct in6_prefix p;
2580 	struct sockaddr_in6 s6;
2581 	struct nd_prefix *pr;
2582 	struct nd_pfxrouter *pfr;
2583 	time_t maxexpire;
2584 	int error;
2585 	char ip6buf[INET6_ADDRSTRLEN];
2586 
2587 	if (req->newptr)
2588 		return (EPERM);
2589 
2590 	error = sysctl_wire_old_buffer(req, 0);
2591 	if (error != 0)
2592 		return (error);
2593 
2594 	bzero(&p, sizeof(p));
2595 	p.origin = PR_ORIG_RA;
2596 	bzero(&s6, sizeof(s6));
2597 	s6.sin6_family = AF_INET6;
2598 	s6.sin6_len = sizeof(s6);
2599 
2600 	ND6_RLOCK();
2601 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2602 		p.prefix = pr->ndpr_prefix;
2603 		if (sa6_recoverscope(&p.prefix)) {
2604 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2605 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2606 			/* XXX: press on... */
2607 		}
2608 		p.raflags = pr->ndpr_raf;
2609 		p.prefixlen = pr->ndpr_plen;
2610 		p.vltime = pr->ndpr_vltime;
2611 		p.pltime = pr->ndpr_pltime;
2612 		p.if_index = pr->ndpr_ifp->if_index;
2613 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2614 			p.expire = 0;
2615 		else {
2616 			/* XXX: we assume time_t is signed. */
2617 			maxexpire = (-1) &
2618 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2619 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2620 				p.expire = pr->ndpr_lastupdate +
2621 				    pr->ndpr_vltime +
2622 				    (time_second - time_uptime);
2623 			else
2624 				p.expire = maxexpire;
2625 		}
2626 		p.refcnt = pr->ndpr_addrcnt;
2627 		p.flags = pr->ndpr_stateflags;
2628 		p.advrtrs = 0;
2629 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2630 			p.advrtrs++;
2631 		error = SYSCTL_OUT(req, &p, sizeof(p));
2632 		if (error != 0)
2633 			break;
2634 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2635 			s6.sin6_addr = pfr->router->rtaddr;
2636 			if (sa6_recoverscope(&s6))
2637 				log(LOG_ERR,
2638 				    "scope error in prefix list (%s)\n",
2639 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2640 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2641 			if (error != 0)
2642 				goto out;
2643 		}
2644 	}
2645 out:
2646 	ND6_RUNLOCK();
2647 	return (error);
2648 }
2649 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2650 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2651 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2652 	"NDP prefix list");
2653 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2654 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2655 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2656 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2657