xref: /freebsd/sys/netinet6/nd6.c (revision e0c4386e7e71d93b0edc0c8fa156263fc4a8b0b6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_route.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/eventhandler.h>
42 #include <sys/callout.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/protosw.h>
52 #include <sys/errno.h>
53 #include <sys/syslog.h>
54 #include <sys/rwlock.h>
55 #include <sys/queue.h>
56 #include <sys/sdt.h>
57 #include <sys/sysctl.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_private.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/route/route_ctl.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_kdtrace.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #include <netinet6/in6_fib.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82 
83 #include <sys/limits.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
88 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
89 
90 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
91 
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93 
94 /* timer values */
95 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
96 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
97 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
98 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
99 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
100 					 * local traffic */
101 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
102 					 * collection timer */
103 
104 /* preventing too many loops in ND option parsing */
105 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
106 
107 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
108 					 * layer hints */
109 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
110 					 * ND entries */
111 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
112 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
113 
114 #ifdef ND6_DEBUG
115 VNET_DEFINE(int, nd6_debug) = 1;
116 #else
117 VNET_DEFINE(int, nd6_debug) = 0;
118 #endif
119 
120 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
121 
122 VNET_DEFINE(struct nd_prhead, nd_prefix);
123 VNET_DEFINE(struct rwlock, nd6_lock);
124 VNET_DEFINE(uint64_t, nd6_list_genid);
125 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
126 
127 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
128 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
129 
130 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
131 
132 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
133 	struct ifnet *);
134 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
135 static void nd6_slowtimo(void *);
136 static int regen_tmpaddr(struct in6_ifaddr *);
137 static void nd6_free(struct llentry **, int);
138 static void nd6_free_redirect(const struct llentry *);
139 static void nd6_llinfo_timer(void *);
140 static void nd6_llinfo_settimer_locked(struct llentry *, long);
141 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
142     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
143 static int nd6_need_cache(struct ifnet *);
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 	if (ifp->if_afdata[AF_INET6] == NULL)
211 		return;
212 
213 	lltable_update_ifaddr(LLTABLE6(ifp));
214 }
215 
216 void
217 nd6_init(void)
218 {
219 
220 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
221 	rw_init(&V_nd6_lock, "nd6 list");
222 
223 	LIST_INIT(&V_nd_prefix);
224 	nd6_defrouter_init();
225 
226 	/* Start timers. */
227 	callout_init(&V_nd6_slowtimo_ch, 1);
228 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
229 	    nd6_slowtimo, curvnet);
230 
231 	callout_init(&V_nd6_timer_ch, 1);
232 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
233 
234 	nd6_dad_init();
235 	if (IS_DEFAULT_VNET(curvnet)) {
236 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
237 		    NULL, EVENTHANDLER_PRI_ANY);
238 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
239 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
240 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
241 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
242 	}
243 }
244 
245 #ifdef VIMAGE
246 void
247 nd6_destroy(void)
248 {
249 
250 	callout_drain(&V_nd6_slowtimo_ch);
251 	callout_drain(&V_nd6_timer_ch);
252 	if (IS_DEFAULT_VNET(curvnet)) {
253 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
254 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
255 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
256 	}
257 	rw_destroy(&V_nd6_lock);
258 	mtx_destroy(&V_nd6_onlink_mtx);
259 }
260 #endif
261 
262 struct nd_ifinfo *
263 nd6_ifattach(struct ifnet *ifp)
264 {
265 	struct nd_ifinfo *nd;
266 
267 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
268 	nd->initialized = 1;
269 
270 	nd->chlim = IPV6_DEFHLIM;
271 	nd->basereachable = REACHABLE_TIME;
272 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
273 	nd->retrans = RETRANS_TIMER;
274 
275 	nd->flags = ND6_IFF_PERFORMNUD;
276 
277 	/* Set IPv6 disabled on all interfaces but loopback by default. */
278 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
279 		nd->flags |= ND6_IFF_IFDISABLED;
280 
281 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
282 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
283 	 * default regardless of the V_ip6_auto_linklocal configuration to
284 	 * give a reasonable default behavior.
285 	 */
286 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
287 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
288 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
289 	/*
290 	 * A loopback interface does not need to accept RTADV.
291 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
292 	 * default regardless of the V_ip6_accept_rtadv configuration to
293 	 * prevent the interface from accepting RA messages arrived
294 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
295 	 */
296 	if (V_ip6_accept_rtadv &&
297 	    !(ifp->if_flags & IFF_LOOPBACK) &&
298 	    (ifp->if_type != IFT_BRIDGE)) {
299 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
300 			/* If we globally accept rtadv, assume IPv6 on. */
301 			nd->flags &= ~ND6_IFF_IFDISABLED;
302 	}
303 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
304 		nd->flags |= ND6_IFF_NO_RADR;
305 
306 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
307 	nd6_setmtu0(ifp, nd);
308 
309 	return nd;
310 }
311 
312 void
313 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
314 {
315 	struct epoch_tracker et;
316 	struct ifaddr *ifa, *next;
317 
318 	NET_EPOCH_ENTER(et);
319 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
320 		if (ifa->ifa_addr->sa_family != AF_INET6)
321 			continue;
322 
323 		/* stop DAD processing */
324 		nd6_dad_stop(ifa);
325 	}
326 	NET_EPOCH_EXIT(et);
327 
328 	free(nd, M_IP6NDP);
329 }
330 
331 /*
332  * Reset ND level link MTU. This function is called when the physical MTU
333  * changes, which means we might have to adjust the ND level MTU.
334  */
335 void
336 nd6_setmtu(struct ifnet *ifp)
337 {
338 	if (ifp->if_afdata[AF_INET6] == NULL)
339 		return;
340 
341 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
342 }
343 
344 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
345 void
346 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
347 {
348 	u_int32_t omaxmtu;
349 
350 	omaxmtu = ndi->maxmtu;
351 	ndi->maxmtu = ifp->if_mtu;
352 
353 	/*
354 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
355 	 * undesirable situation.  We thus notify the operator of the change
356 	 * explicitly.  The check for omaxmtu is necessary to restrict the
357 	 * log to the case of changing the MTU, not initializing it.
358 	 */
359 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
360 		log(LOG_NOTICE, "nd6_setmtu0: "
361 		    "new link MTU on %s (%lu) is too small for IPv6\n",
362 		    if_name(ifp), (unsigned long)ndi->maxmtu);
363 	}
364 
365 	if (ndi->maxmtu > V_in6_maxmtu)
366 		in6_setmaxmtu(); /* check all interfaces just in case */
367 
368 }
369 
370 void
371 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
372 {
373 
374 	bzero(ndopts, sizeof(*ndopts));
375 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
376 	ndopts->nd_opts_last
377 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
378 
379 	if (icmp6len == 0) {
380 		ndopts->nd_opts_done = 1;
381 		ndopts->nd_opts_search = NULL;
382 	}
383 }
384 
385 /*
386  * Take one ND option.
387  */
388 struct nd_opt_hdr *
389 nd6_option(union nd_opts *ndopts)
390 {
391 	struct nd_opt_hdr *nd_opt;
392 	int olen;
393 
394 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
395 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
396 	    __func__));
397 	if (ndopts->nd_opts_search == NULL)
398 		return NULL;
399 	if (ndopts->nd_opts_done)
400 		return NULL;
401 
402 	nd_opt = ndopts->nd_opts_search;
403 
404 	/* make sure nd_opt_len is inside the buffer */
405 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
406 		bzero(ndopts, sizeof(*ndopts));
407 		return NULL;
408 	}
409 
410 	olen = nd_opt->nd_opt_len << 3;
411 	if (olen == 0) {
412 		/*
413 		 * Message validation requires that all included
414 		 * options have a length that is greater than zero.
415 		 */
416 		bzero(ndopts, sizeof(*ndopts));
417 		return NULL;
418 	}
419 
420 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
421 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
422 		/* option overruns the end of buffer, invalid */
423 		bzero(ndopts, sizeof(*ndopts));
424 		return NULL;
425 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
426 		/* reached the end of options chain */
427 		ndopts->nd_opts_done = 1;
428 		ndopts->nd_opts_search = NULL;
429 	}
430 	return nd_opt;
431 }
432 
433 /*
434  * Parse multiple ND options.
435  * This function is much easier to use, for ND routines that do not need
436  * multiple options of the same type.
437  */
438 int
439 nd6_options(union nd_opts *ndopts)
440 {
441 	struct nd_opt_hdr *nd_opt;
442 	int i = 0;
443 
444 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
445 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
446 	    __func__));
447 	if (ndopts->nd_opts_search == NULL)
448 		return 0;
449 
450 	while (1) {
451 		nd_opt = nd6_option(ndopts);
452 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
453 			/*
454 			 * Message validation requires that all included
455 			 * options have a length that is greater than zero.
456 			 */
457 			ICMP6STAT_INC(icp6s_nd_badopt);
458 			bzero(ndopts, sizeof(*ndopts));
459 			return -1;
460 		}
461 
462 		if (nd_opt == NULL)
463 			goto skip1;
464 
465 		switch (nd_opt->nd_opt_type) {
466 		case ND_OPT_SOURCE_LINKADDR:
467 		case ND_OPT_TARGET_LINKADDR:
468 		case ND_OPT_MTU:
469 		case ND_OPT_REDIRECTED_HEADER:
470 		case ND_OPT_NONCE:
471 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
472 				nd6log((LOG_INFO,
473 				    "duplicated ND6 option found (type=%d)\n",
474 				    nd_opt->nd_opt_type));
475 				/* XXX bark? */
476 			} else {
477 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
478 					= nd_opt;
479 			}
480 			break;
481 		case ND_OPT_PREFIX_INFORMATION:
482 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
483 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
484 					= nd_opt;
485 			}
486 			ndopts->nd_opts_pi_end =
487 				(struct nd_opt_prefix_info *)nd_opt;
488 			break;
489 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
490 		case ND_OPT_RDNSS:	/* RFC 6106 */
491 		case ND_OPT_DNSSL:	/* RFC 6106 */
492 			/*
493 			 * Silently ignore options we know and do not care about
494 			 * in the kernel.
495 			 */
496 			break;
497 		default:
498 			/*
499 			 * Unknown options must be silently ignored,
500 			 * to accommodate future extension to the protocol.
501 			 */
502 			nd6log((LOG_DEBUG,
503 			    "nd6_options: unsupported option %d - "
504 			    "option ignored\n", nd_opt->nd_opt_type));
505 		}
506 
507 skip1:
508 		i++;
509 		if (i > V_nd6_maxndopt) {
510 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
511 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
512 			break;
513 		}
514 
515 		if (ndopts->nd_opts_done)
516 			break;
517 	}
518 
519 	return 0;
520 }
521 
522 /*
523  * ND6 timer routine to handle ND6 entries
524  */
525 static void
526 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
527 {
528 	int canceled;
529 
530 	LLE_WLOCK_ASSERT(ln);
531 
532 	/* Do not schedule timers for child LLEs. */
533 	if (ln->la_flags & LLE_CHILD)
534 		return;
535 
536 	if (tick < 0) {
537 		ln->la_expire = 0;
538 		ln->ln_ntick = 0;
539 		canceled = callout_stop(&ln->lle_timer);
540 	} else {
541 		ln->la_expire = time_uptime + tick / hz;
542 		LLE_ADDREF(ln);
543 		if (tick > INT_MAX) {
544 			ln->ln_ntick = tick - INT_MAX;
545 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
546 			    nd6_llinfo_timer, ln);
547 		} else {
548 			ln->ln_ntick = 0;
549 			canceled = callout_reset(&ln->lle_timer, tick,
550 			    nd6_llinfo_timer, ln);
551 		}
552 	}
553 	if (canceled > 0)
554 		LLE_REMREF(ln);
555 }
556 
557 /*
558  * Gets source address of the first packet in hold queue
559  * and stores it in @src.
560  * Returns pointer to @src (if hold queue is not empty) or NULL.
561  *
562  * Set noinline to be dtrace-friendly
563  */
564 static __noinline struct in6_addr *
565 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
566 {
567 	struct ip6_hdr hdr;
568 	struct mbuf *m;
569 
570 	if (ln->la_hold == NULL)
571 		return (NULL);
572 
573 	/*
574 	 * assume every packet in la_hold has the same IP header
575 	 */
576 	m = ln->la_hold;
577 	if (sizeof(hdr) > m->m_len)
578 		return (NULL);
579 
580 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
581 	*src = hdr.ip6_src;
582 
583 	return (src);
584 }
585 
586 /*
587  * Checks if we need to switch from STALE state.
588  *
589  * RFC 4861 requires switching from STALE to DELAY state
590  * on first packet matching entry, waiting V_nd6_delay and
591  * transition to PROBE state (if upper layer confirmation was
592  * not received).
593  *
594  * This code performs a bit differently:
595  * On packet hit we don't change state (but desired state
596  * can be guessed by control plane). However, after V_nd6_delay
597  * seconds code will transition to PROBE state (so DELAY state
598  * is kinda skipped in most situations).
599  *
600  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
601  * we perform the following upon entering STALE state:
602  *
603  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
604  * if packet was transmitted at the start of given interval, we
605  * would be able to switch to PROBE state in V_nd6_delay seconds
606  * as user expects.
607  *
608  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
609  * lle in STALE state (remaining timer value stored in lle_remtime).
610  *
611  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
612  * seconds ago.
613  *
614  * Returns non-zero value if the entry is still STALE (storing
615  * the next timer interval in @pdelay).
616  *
617  * Returns zero value if original timer expired or we need to switch to
618  * PROBE (store that in @do_switch variable).
619  */
620 static int
621 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
622 {
623 	int nd_delay, nd_gctimer;
624 	time_t lle_hittime;
625 	long delay;
626 
627 	*do_switch = 0;
628 	nd_gctimer = V_nd6_gctimer;
629 	nd_delay = V_nd6_delay;
630 
631 	lle_hittime = llentry_get_hittime(lle);
632 
633 	if (lle_hittime == 0) {
634 		/*
635 		 * Datapath feedback has been requested upon entering
636 		 * STALE state. No packets has been passed using this lle.
637 		 * Ask for the timer reschedule and keep STALE state.
638 		 */
639 		delay = (long)(MIN(nd_gctimer, nd_delay));
640 		delay *= hz;
641 		if (lle->lle_remtime > delay)
642 			lle->lle_remtime -= delay;
643 		else {
644 			delay = lle->lle_remtime;
645 			lle->lle_remtime = 0;
646 		}
647 
648 		if (delay == 0) {
649 			/*
650 			 * The original ng6_gctime timeout ended,
651 			 * no more rescheduling.
652 			 */
653 			return (0);
654 		}
655 
656 		*pdelay = delay;
657 		return (1);
658 	}
659 
660 	/*
661 	 * Packet received. Verify timestamp
662 	 */
663 	delay = (long)(time_uptime - lle_hittime);
664 	if (delay < nd_delay) {
665 		/*
666 		 * V_nd6_delay still not passed since the first
667 		 * hit in STALE state.
668 		 * Reschedule timer and return.
669 		 */
670 		*pdelay = (long)(nd_delay - delay) * hz;
671 		return (1);
672 	}
673 
674 	/* Request switching to probe */
675 	*do_switch = 1;
676 	return (0);
677 }
678 
679 /*
680  * Switch @lle state to new state optionally arming timers.
681  *
682  * Set noinline to be dtrace-friendly
683  */
684 __noinline void
685 nd6_llinfo_setstate(struct llentry *lle, int newstate)
686 {
687 	struct ifnet *ifp;
688 	int nd_gctimer, nd_delay;
689 	long delay, remtime;
690 
691 	delay = 0;
692 	remtime = 0;
693 
694 	switch (newstate) {
695 	case ND6_LLINFO_INCOMPLETE:
696 		ifp = lle->lle_tbl->llt_ifp;
697 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
698 		break;
699 	case ND6_LLINFO_REACHABLE:
700 		if (!ND6_LLINFO_PERMANENT(lle)) {
701 			ifp = lle->lle_tbl->llt_ifp;
702 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
703 		}
704 		break;
705 	case ND6_LLINFO_STALE:
706 
707 		llentry_request_feedback(lle);
708 		nd_delay = V_nd6_delay;
709 		nd_gctimer = V_nd6_gctimer;
710 
711 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
712 		remtime = (long)nd_gctimer * hz - delay;
713 		break;
714 	case ND6_LLINFO_DELAY:
715 		lle->la_asked = 0;
716 		delay = (long)V_nd6_delay * hz;
717 		break;
718 	}
719 
720 	if (delay > 0)
721 		nd6_llinfo_settimer_locked(lle, delay);
722 
723 	lle->lle_remtime = remtime;
724 	lle->ln_state = newstate;
725 }
726 
727 /*
728  * Timer-dependent part of nd state machine.
729  *
730  * Set noinline to be dtrace-friendly
731  */
732 static __noinline void
733 nd6_llinfo_timer(void *arg)
734 {
735 	struct epoch_tracker et;
736 	struct llentry *ln;
737 	struct in6_addr *dst, *pdst, *psrc, src;
738 	struct ifnet *ifp;
739 	struct nd_ifinfo *ndi;
740 	int do_switch, send_ns;
741 	long delay;
742 
743 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
744 	ln = (struct llentry *)arg;
745 	ifp = lltable_get_ifp(ln->lle_tbl);
746 	CURVNET_SET(ifp->if_vnet);
747 
748 	ND6_RLOCK();
749 	LLE_WLOCK(ln);
750 	if (callout_pending(&ln->lle_timer)) {
751 		/*
752 		 * Here we are a bit odd here in the treatment of
753 		 * active/pending. If the pending bit is set, it got
754 		 * rescheduled before I ran. The active
755 		 * bit we ignore, since if it was stopped
756 		 * in ll_tablefree() and was currently running
757 		 * it would have return 0 so the code would
758 		 * not have deleted it since the callout could
759 		 * not be stopped so we want to go through
760 		 * with the delete here now. If the callout
761 		 * was restarted, the pending bit will be back on and
762 		 * we just want to bail since the callout_reset would
763 		 * return 1 and our reference would have been removed
764 		 * by nd6_llinfo_settimer_locked above since canceled
765 		 * would have been 1.
766 		 */
767 		LLE_WUNLOCK(ln);
768 		ND6_RUNLOCK();
769 		CURVNET_RESTORE();
770 		return;
771 	}
772 	NET_EPOCH_ENTER(et);
773 	ndi = ND_IFINFO(ifp);
774 	send_ns = 0;
775 	dst = &ln->r_l3addr.addr6;
776 	pdst = dst;
777 
778 	if (ln->ln_ntick > 0) {
779 		if (ln->ln_ntick > INT_MAX) {
780 			ln->ln_ntick -= INT_MAX;
781 			nd6_llinfo_settimer_locked(ln, INT_MAX);
782 		} else {
783 			ln->ln_ntick = 0;
784 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
785 		}
786 		goto done;
787 	}
788 
789 	if (ln->la_flags & LLE_STATIC) {
790 		goto done;
791 	}
792 
793 	if (ln->la_flags & LLE_DELETED) {
794 		nd6_free(&ln, 0);
795 		goto done;
796 	}
797 
798 	switch (ln->ln_state) {
799 	case ND6_LLINFO_INCOMPLETE:
800 		if (ln->la_asked < V_nd6_mmaxtries) {
801 			ln->la_asked++;
802 			send_ns = 1;
803 			/* Send NS to multicast address */
804 			pdst = NULL;
805 		} else {
806 			struct mbuf *m;
807 
808 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
809 
810 			m = ln->la_hold;
811 			if (m != NULL) {
812 				/*
813 				 * assuming every packet in la_hold has the
814 				 * same IP header.  Send error after unlock.
815 				 */
816 				ln->la_hold = m->m_nextpkt;
817 				m->m_nextpkt = NULL;
818 				ln->la_numheld--;
819 			}
820 			nd6_free(&ln, 0);
821 			if (m != NULL) {
822 				struct mbuf *n = m;
823 
824 				/*
825 				 * if there are any ummapped mbufs, we
826 				 * must free them, rather than using
827 				 * them for an ICMP, as they cannot be
828 				 * checksummed.
829 				 */
830 				while ((n = n->m_next) != NULL) {
831 					if (n->m_flags & M_EXTPG)
832 						break;
833 				}
834 				if (n != NULL) {
835 					m_freem(m);
836 					m = NULL;
837 				} else {
838 					icmp6_error2(m, ICMP6_DST_UNREACH,
839 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
840 				}
841 			}
842 		}
843 		break;
844 	case ND6_LLINFO_REACHABLE:
845 		if (!ND6_LLINFO_PERMANENT(ln))
846 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
847 		break;
848 
849 	case ND6_LLINFO_STALE:
850 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
851 			/*
852 			 * No packet has used this entry and GC timeout
853 			 * has not been passed. Reschedule timer and
854 			 * return.
855 			 */
856 			nd6_llinfo_settimer_locked(ln, delay);
857 			break;
858 		}
859 
860 		if (do_switch == 0) {
861 			/*
862 			 * GC timer has ended and entry hasn't been used.
863 			 * Run Garbage collector (RFC 4861, 5.3)
864 			 */
865 			if (!ND6_LLINFO_PERMANENT(ln))
866 				nd6_free(&ln, 1);
867 			break;
868 		}
869 
870 		/* Entry has been used AND delay timer has ended. */
871 
872 		/* FALLTHROUGH */
873 
874 	case ND6_LLINFO_DELAY:
875 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
876 			/* We need NUD */
877 			ln->la_asked = 1;
878 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
879 			send_ns = 1;
880 		} else
881 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
882 		break;
883 	case ND6_LLINFO_PROBE:
884 		if (ln->la_asked < V_nd6_umaxtries) {
885 			ln->la_asked++;
886 			send_ns = 1;
887 		} else {
888 			nd6_free(&ln, 0);
889 		}
890 		break;
891 	default:
892 		panic("%s: paths in a dark night can be confusing: %d",
893 		    __func__, ln->ln_state);
894 	}
895 done:
896 	if (ln != NULL)
897 		ND6_RUNLOCK();
898 	if (send_ns != 0) {
899 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
900 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
901 		LLE_FREE_LOCKED(ln);
902 		ln = NULL;
903 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
904 	}
905 
906 	if (ln != NULL)
907 		LLE_FREE_LOCKED(ln);
908 	NET_EPOCH_EXIT(et);
909 	CURVNET_RESTORE();
910 }
911 
912 /*
913  * ND6 timer routine to expire default route list and prefix list
914  */
915 void
916 nd6_timer(void *arg)
917 {
918 	CURVNET_SET((struct vnet *) arg);
919 	struct epoch_tracker et;
920 	struct nd_prhead prl;
921 	struct nd_prefix *pr, *npr;
922 	struct ifnet *ifp;
923 	struct in6_ifaddr *ia6, *nia6;
924 	uint64_t genid;
925 
926 	LIST_INIT(&prl);
927 
928 	NET_EPOCH_ENTER(et);
929 	nd6_defrouter_timer();
930 
931 	/*
932 	 * expire interface addresses.
933 	 * in the past the loop was inside prefix expiry processing.
934 	 * However, from a stricter speci-confrmance standpoint, we should
935 	 * rather separate address lifetimes and prefix lifetimes.
936 	 *
937 	 * XXXRW: in6_ifaddrhead locking.
938 	 */
939   addrloop:
940 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
941 		/* check address lifetime */
942 		if (IFA6_IS_INVALID(ia6)) {
943 			int regen = 0;
944 
945 			/*
946 			 * If the expiring address is temporary, try
947 			 * regenerating a new one.  This would be useful when
948 			 * we suspended a laptop PC, then turned it on after a
949 			 * period that could invalidate all temporary
950 			 * addresses.  Although we may have to restart the
951 			 * loop (see below), it must be after purging the
952 			 * address.  Otherwise, we'd see an infinite loop of
953 			 * regeneration.
954 			 */
955 			if (V_ip6_use_tempaddr &&
956 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
957 				if (regen_tmpaddr(ia6) == 0)
958 					regen = 1;
959 			}
960 
961 			in6_purgeaddr(&ia6->ia_ifa);
962 
963 			if (regen)
964 				goto addrloop; /* XXX: see below */
965 		} else if (IFA6_IS_DEPRECATED(ia6)) {
966 			int oldflags = ia6->ia6_flags;
967 
968 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
969 
970 			/*
971 			 * If a temporary address has just become deprecated,
972 			 * regenerate a new one if possible.
973 			 */
974 			if (V_ip6_use_tempaddr &&
975 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
976 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
977 				if (regen_tmpaddr(ia6) == 0) {
978 					/*
979 					 * A new temporary address is
980 					 * generated.
981 					 * XXX: this means the address chain
982 					 * has changed while we are still in
983 					 * the loop.  Although the change
984 					 * would not cause disaster (because
985 					 * it's not a deletion, but an
986 					 * addition,) we'd rather restart the
987 					 * loop just for safety.  Or does this
988 					 * significantly reduce performance??
989 					 */
990 					goto addrloop;
991 				}
992 			}
993 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
994 			/*
995 			 * Schedule DAD for a tentative address.  This happens
996 			 * if the interface was down or not running
997 			 * when the address was configured.
998 			 */
999 			int delay;
1000 
1001 			delay = arc4random() %
1002 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1003 			nd6_dad_start((struct ifaddr *)ia6, delay);
1004 		} else {
1005 			/*
1006 			 * Check status of the interface.  If it is down,
1007 			 * mark the address as tentative for future DAD.
1008 			 */
1009 			ifp = ia6->ia_ifp;
1010 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1011 			    ((ifp->if_flags & IFF_UP) == 0 ||
1012 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1013 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1014 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1015 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1016 			}
1017 
1018 			/*
1019 			 * A new RA might have made a deprecated address
1020 			 * preferred.
1021 			 */
1022 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1023 		}
1024 	}
1025 	NET_EPOCH_EXIT(et);
1026 
1027 	ND6_WLOCK();
1028 restart:
1029 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1030 		/*
1031 		 * Expire prefixes. Since the pltime is only used for
1032 		 * autoconfigured addresses, pltime processing for prefixes is
1033 		 * not necessary.
1034 		 *
1035 		 * Only unlink after all derived addresses have expired. This
1036 		 * may not occur until two hours after the prefix has expired
1037 		 * per RFC 4862. If the prefix expires before its derived
1038 		 * addresses, mark it off-link. This will be done automatically
1039 		 * after unlinking if no address references remain.
1040 		 */
1041 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1042 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1043 			continue;
1044 
1045 		if (pr->ndpr_addrcnt == 0) {
1046 			nd6_prefix_unlink(pr, &prl);
1047 			continue;
1048 		}
1049 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1050 			genid = V_nd6_list_genid;
1051 			nd6_prefix_ref(pr);
1052 			ND6_WUNLOCK();
1053 			ND6_ONLINK_LOCK();
1054 			(void)nd6_prefix_offlink(pr);
1055 			ND6_ONLINK_UNLOCK();
1056 			ND6_WLOCK();
1057 			nd6_prefix_rele(pr);
1058 			if (genid != V_nd6_list_genid)
1059 				goto restart;
1060 		}
1061 	}
1062 	ND6_WUNLOCK();
1063 
1064 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1065 		LIST_REMOVE(pr, ndpr_entry);
1066 		nd6_prefix_del(pr);
1067 	}
1068 
1069 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1070 	    nd6_timer, curvnet);
1071 
1072 	CURVNET_RESTORE();
1073 }
1074 
1075 /*
1076  * ia6 - deprecated/invalidated temporary address
1077  */
1078 static int
1079 regen_tmpaddr(struct in6_ifaddr *ia6)
1080 {
1081 	struct ifaddr *ifa;
1082 	struct ifnet *ifp;
1083 	struct in6_ifaddr *public_ifa6 = NULL;
1084 
1085 	NET_EPOCH_ASSERT();
1086 
1087 	ifp = ia6->ia_ifa.ifa_ifp;
1088 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1089 		struct in6_ifaddr *it6;
1090 
1091 		if (ifa->ifa_addr->sa_family != AF_INET6)
1092 			continue;
1093 
1094 		it6 = (struct in6_ifaddr *)ifa;
1095 
1096 		/* ignore no autoconf addresses. */
1097 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1098 			continue;
1099 
1100 		/* ignore autoconf addresses with different prefixes. */
1101 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1102 			continue;
1103 
1104 		/*
1105 		 * Now we are looking at an autoconf address with the same
1106 		 * prefix as ours.  If the address is temporary and is still
1107 		 * preferred, do not create another one.  It would be rare, but
1108 		 * could happen, for example, when we resume a laptop PC after
1109 		 * a long period.
1110 		 */
1111 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1112 		    !IFA6_IS_DEPRECATED(it6)) {
1113 			public_ifa6 = NULL;
1114 			break;
1115 		}
1116 
1117 		/*
1118 		 * This is a public autoconf address that has the same prefix
1119 		 * as ours.  If it is preferred, keep it.  We can't break the
1120 		 * loop here, because there may be a still-preferred temporary
1121 		 * address with the prefix.
1122 		 */
1123 		if (!IFA6_IS_DEPRECATED(it6))
1124 			public_ifa6 = it6;
1125 	}
1126 	if (public_ifa6 != NULL)
1127 		ifa_ref(&public_ifa6->ia_ifa);
1128 
1129 	if (public_ifa6 != NULL) {
1130 		int e;
1131 
1132 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1133 			ifa_free(&public_ifa6->ia_ifa);
1134 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1135 			    " tmp addr,errno=%d\n", e);
1136 			return (-1);
1137 		}
1138 		ifa_free(&public_ifa6->ia_ifa);
1139 		return (0);
1140 	}
1141 
1142 	return (-1);
1143 }
1144 
1145 /*
1146  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1147  * cache entries are freed in in6_domifdetach().
1148  */
1149 void
1150 nd6_purge(struct ifnet *ifp)
1151 {
1152 	struct nd_prhead prl;
1153 	struct nd_prefix *pr, *npr;
1154 
1155 	LIST_INIT(&prl);
1156 
1157 	/* Purge default router list entries toward ifp. */
1158 	nd6_defrouter_purge(ifp);
1159 
1160 	ND6_WLOCK();
1161 	/*
1162 	 * Remove prefixes on ifp. We should have already removed addresses on
1163 	 * this interface, so no addresses should be referencing these prefixes.
1164 	 */
1165 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1166 		if (pr->ndpr_ifp == ifp)
1167 			nd6_prefix_unlink(pr, &prl);
1168 	}
1169 	ND6_WUNLOCK();
1170 
1171 	/* Delete the unlinked prefix objects. */
1172 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1173 		LIST_REMOVE(pr, ndpr_entry);
1174 		nd6_prefix_del(pr);
1175 	}
1176 
1177 	/* cancel default outgoing interface setting */
1178 	if (V_nd6_defifindex == ifp->if_index)
1179 		nd6_setdefaultiface(0);
1180 
1181 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1182 		/* Refresh default router list. */
1183 		defrouter_select_fib(ifp->if_fib);
1184 	}
1185 }
1186 
1187 /*
1188  * the caller acquires and releases the lock on the lltbls
1189  * Returns the llentry locked
1190  */
1191 struct llentry *
1192 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1193 {
1194 	struct sockaddr_in6 sin6;
1195 	struct llentry *ln;
1196 
1197 	bzero(&sin6, sizeof(sin6));
1198 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1199 	sin6.sin6_family = AF_INET6;
1200 	sin6.sin6_addr = *addr6;
1201 
1202 	IF_AFDATA_LOCK_ASSERT(ifp);
1203 
1204 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1205 
1206 	return (ln);
1207 }
1208 
1209 static struct llentry *
1210 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1211 {
1212 	struct sockaddr_in6 sin6;
1213 	struct llentry *ln;
1214 
1215 	bzero(&sin6, sizeof(sin6));
1216 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1217 	sin6.sin6_family = AF_INET6;
1218 	sin6.sin6_addr = *addr6;
1219 
1220 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1221 	if (ln != NULL)
1222 		ln->ln_state = ND6_LLINFO_NOSTATE;
1223 
1224 	return (ln);
1225 }
1226 
1227 /*
1228  * Test whether a given IPv6 address can be a neighbor.
1229  */
1230 static bool
1231 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1232 {
1233 
1234 	/*
1235 	 * A link-local address is always a neighbor.
1236 	 * XXX: a link does not necessarily specify a single interface.
1237 	 */
1238 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1239 		struct sockaddr_in6 sin6_copy;
1240 		u_int32_t zone;
1241 
1242 		/*
1243 		 * We need sin6_copy since sa6_recoverscope() may modify the
1244 		 * content (XXX).
1245 		 */
1246 		sin6_copy = *addr;
1247 		if (sa6_recoverscope(&sin6_copy))
1248 			return (0); /* XXX: should be impossible */
1249 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1250 			return (0);
1251 		if (sin6_copy.sin6_scope_id == zone)
1252 			return (1);
1253 		else
1254 			return (0);
1255 	}
1256 	/* Checking global unicast */
1257 
1258 	/* If an address is directly reachable, it is a neigbor */
1259 	struct nhop_object *nh;
1260 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1261 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1262 		return (true);
1263 
1264 	/*
1265 	 * Check prefixes with desired on-link state, as some may be not
1266 	 * installed in the routing table.
1267 	 */
1268 	bool matched = false;
1269 	struct nd_prefix *pr;
1270 	ND6_RLOCK();
1271 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1272 		if (pr->ndpr_ifp != ifp)
1273 			continue;
1274 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1275 			continue;
1276 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1277 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1278 			matched = true;
1279 			break;
1280 		}
1281 	}
1282 	ND6_RUNLOCK();
1283 	if (matched)
1284 		return (true);
1285 
1286 	/*
1287 	 * If the address is assigned on the node of the other side of
1288 	 * a p2p interface, the address should be a neighbor.
1289 	 */
1290 	if (ifp->if_flags & IFF_POINTOPOINT) {
1291 		struct ifaddr *ifa;
1292 
1293 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1294 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1295 				continue;
1296 			if (ifa->ifa_dstaddr != NULL &&
1297 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1298 				return (true);
1299 			}
1300 		}
1301 	}
1302 
1303 	/*
1304 	 * If the default router list is empty, all addresses are regarded
1305 	 * as on-link, and thus, as a neighbor.
1306 	 */
1307 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1308 	    nd6_defrouter_list_empty() &&
1309 	    V_nd6_defifindex == ifp->if_index) {
1310 		return (1);
1311 	}
1312 
1313 	return (0);
1314 }
1315 
1316 /*
1317  * Detect if a given IPv6 address identifies a neighbor on a given link.
1318  * XXX: should take care of the destination of a p2p link?
1319  */
1320 int
1321 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1322 {
1323 	struct llentry *lle;
1324 	int rc = 0;
1325 
1326 	NET_EPOCH_ASSERT();
1327 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1328 	if (nd6_is_new_addr_neighbor(addr, ifp))
1329 		return (1);
1330 
1331 	/*
1332 	 * Even if the address matches none of our addresses, it might be
1333 	 * in the neighbor cache.
1334 	 */
1335 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1336 		LLE_RUNLOCK(lle);
1337 		rc = 1;
1338 	}
1339 	return (rc);
1340 }
1341 
1342 static __noinline void
1343 nd6_free_children(struct llentry *lle)
1344 {
1345 	struct llentry *child_lle;
1346 
1347 	NET_EPOCH_ASSERT();
1348 	LLE_WLOCK_ASSERT(lle);
1349 
1350 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1351 		LLE_WLOCK(child_lle);
1352 		lltable_unlink_child_entry(child_lle);
1353 		llentry_free(child_lle);
1354 	}
1355 }
1356 
1357 /*
1358  * Tries to update @lle address/prepend data with new @lladdr.
1359  *
1360  * Returns true on success.
1361  * In any case, @lle is returned wlocked.
1362  */
1363 static __noinline bool
1364 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1365 {
1366 	u_char buf[LLE_MAX_LINKHDR];
1367 	int fam, off;
1368 	size_t sz;
1369 
1370 	sz = sizeof(buf);
1371 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1372 		return (false);
1373 
1374 	/* Update data */
1375 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1376 
1377 	struct llentry *child_lle;
1378 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1379 		LLE_WLOCK(child_lle);
1380 		fam = child_lle->r_family;
1381 		sz = sizeof(buf);
1382 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1383 			/* success */
1384 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1385 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1386 		}
1387 		LLE_WUNLOCK(child_lle);
1388 	}
1389 
1390 	return (true);
1391 }
1392 
1393 bool
1394 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1395 {
1396 	NET_EPOCH_ASSERT();
1397 	LLE_WLOCK_ASSERT(lle);
1398 
1399 	if (!lltable_acquire_wlock(ifp, lle))
1400 		return (false);
1401 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1402 	IF_AFDATA_WUNLOCK(ifp);
1403 
1404 	return (ret);
1405 }
1406 
1407 /*
1408  * Free an nd6 llinfo entry.
1409  * Since the function would cause significant changes in the kernel, DO NOT
1410  * make it global, unless you have a strong reason for the change, and are sure
1411  * that the change is safe.
1412  *
1413  * Set noinline to be dtrace-friendly
1414  */
1415 static __noinline void
1416 nd6_free(struct llentry **lnp, int gc)
1417 {
1418 	struct ifnet *ifp;
1419 	struct llentry *ln;
1420 	struct nd_defrouter *dr;
1421 
1422 	ln = *lnp;
1423 	*lnp = NULL;
1424 
1425 	LLE_WLOCK_ASSERT(ln);
1426 	ND6_RLOCK_ASSERT();
1427 
1428 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1429 
1430 	ifp = lltable_get_ifp(ln->lle_tbl);
1431 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1432 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1433 	else
1434 		dr = NULL;
1435 	ND6_RUNLOCK();
1436 
1437 	if ((ln->la_flags & LLE_DELETED) == 0)
1438 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1439 
1440 	/*
1441 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1442 	 * even though it is not harmful, it was not really necessary.
1443 	 */
1444 
1445 	/* cancel timer */
1446 	nd6_llinfo_settimer_locked(ln, -1);
1447 
1448 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1449 		if (dr != NULL && dr->expire &&
1450 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1451 			/*
1452 			 * If the reason for the deletion is just garbage
1453 			 * collection, and the neighbor is an active default
1454 			 * router, do not delete it.  Instead, reset the GC
1455 			 * timer using the router's lifetime.
1456 			 * Simply deleting the entry would affect default
1457 			 * router selection, which is not necessarily a good
1458 			 * thing, especially when we're using router preference
1459 			 * values.
1460 			 * XXX: the check for ln_state would be redundant,
1461 			 *      but we intentionally keep it just in case.
1462 			 */
1463 			if (dr->expire > time_uptime)
1464 				nd6_llinfo_settimer_locked(ln,
1465 				    (dr->expire - time_uptime) * hz);
1466 			else
1467 				nd6_llinfo_settimer_locked(ln,
1468 				    (long)V_nd6_gctimer * hz);
1469 
1470 			LLE_REMREF(ln);
1471 			LLE_WUNLOCK(ln);
1472 			defrouter_rele(dr);
1473 			return;
1474 		}
1475 
1476 		if (dr) {
1477 			/*
1478 			 * Unreachability of a router might affect the default
1479 			 * router selection and on-link detection of advertised
1480 			 * prefixes.
1481 			 */
1482 
1483 			/*
1484 			 * Temporarily fake the state to choose a new default
1485 			 * router and to perform on-link determination of
1486 			 * prefixes correctly.
1487 			 * Below the state will be set correctly,
1488 			 * or the entry itself will be deleted.
1489 			 */
1490 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1491 		}
1492 
1493 		if (ln->ln_router || dr) {
1494 			/*
1495 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1496 			 * rnh and for the calls to pfxlist_onlink_check() and
1497 			 * defrouter_select_fib() in the block further down for calls
1498 			 * into nd6_lookup().  We still hold a ref.
1499 			 */
1500 			LLE_WUNLOCK(ln);
1501 
1502 			/*
1503 			 * rt6_flush must be called whether or not the neighbor
1504 			 * is in the Default Router List.
1505 			 * See a corresponding comment in nd6_na_input().
1506 			 */
1507 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1508 		}
1509 
1510 		if (dr) {
1511 			/*
1512 			 * Since defrouter_select_fib() does not affect the
1513 			 * on-link determination and MIP6 needs the check
1514 			 * before the default router selection, we perform
1515 			 * the check now.
1516 			 */
1517 			pfxlist_onlink_check();
1518 
1519 			/*
1520 			 * Refresh default router list.
1521 			 */
1522 			defrouter_select_fib(dr->ifp->if_fib);
1523 		}
1524 
1525 		/*
1526 		 * If this entry was added by an on-link redirect, remove the
1527 		 * corresponding host route.
1528 		 */
1529 		if (ln->la_flags & LLE_REDIRECT)
1530 			nd6_free_redirect(ln);
1531 
1532 		if (ln->ln_router || dr)
1533 			LLE_WLOCK(ln);
1534 	}
1535 
1536 	/*
1537 	 * Save to unlock. We still hold an extra reference and will not
1538 	 * free(9) in llentry_free() if someone else holds one as well.
1539 	 */
1540 	LLE_WUNLOCK(ln);
1541 	IF_AFDATA_LOCK(ifp);
1542 	LLE_WLOCK(ln);
1543 	/* Guard against race with other llentry_free(). */
1544 	if (ln->la_flags & LLE_LINKED) {
1545 		/* Remove callout reference */
1546 		LLE_REMREF(ln);
1547 		lltable_unlink_entry(ln->lle_tbl, ln);
1548 	}
1549 	IF_AFDATA_UNLOCK(ifp);
1550 
1551 	nd6_free_children(ln);
1552 
1553 	llentry_free(ln);
1554 	if (dr != NULL)
1555 		defrouter_rele(dr);
1556 }
1557 
1558 static int
1559 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1560 {
1561 
1562 	if (nh->nh_flags & NHF_REDIRECT)
1563 		return (1);
1564 
1565 	return (0);
1566 }
1567 
1568 /*
1569  * Remove the rtentry for the given llentry,
1570  * both of which were installed by a redirect.
1571  */
1572 static void
1573 nd6_free_redirect(const struct llentry *ln)
1574 {
1575 	int fibnum;
1576 	struct sockaddr_in6 sin6;
1577 	struct rib_cmd_info rc;
1578 	struct epoch_tracker et;
1579 
1580 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1581 
1582 	NET_EPOCH_ENTER(et);
1583 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1584 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1585 		    nd6_isdynrte, NULL, 0, &rc);
1586 	NET_EPOCH_EXIT(et);
1587 }
1588 
1589 /*
1590  * Updates status of the default router route.
1591  */
1592 static void
1593 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1594 {
1595 	struct nd_defrouter *dr;
1596 	struct nhop_object *nh;
1597 
1598 	nh = rc->rc_nh_old;
1599 
1600 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1601 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1602 		if (dr != NULL) {
1603 			dr->installed = 0;
1604 			defrouter_rele(dr);
1605 		}
1606 	}
1607 }
1608 
1609 void
1610 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1611 {
1612 
1613 #ifdef ROUTE_MPATH
1614 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1615 #else
1616 	check_release_defrouter(rc, NULL);
1617 #endif
1618 }
1619 
1620 int
1621 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1622 {
1623 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1624 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1625 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1626 	struct epoch_tracker et;
1627 	int error = 0;
1628 
1629 	if (ifp->if_afdata[AF_INET6] == NULL)
1630 		return (EPFNOSUPPORT);
1631 	switch (cmd) {
1632 	case OSIOCGIFINFO_IN6:
1633 #define ND	ndi->ndi
1634 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1635 		bzero(&ND, sizeof(ND));
1636 		ND.linkmtu = IN6_LINKMTU(ifp);
1637 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1638 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1639 		ND.reachable = ND_IFINFO(ifp)->reachable;
1640 		ND.retrans = ND_IFINFO(ifp)->retrans;
1641 		ND.flags = ND_IFINFO(ifp)->flags;
1642 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1643 		ND.chlim = ND_IFINFO(ifp)->chlim;
1644 		break;
1645 	case SIOCGIFINFO_IN6:
1646 		ND = *ND_IFINFO(ifp);
1647 		break;
1648 	case SIOCSIFINFO_IN6:
1649 		/*
1650 		 * used to change host variables from userland.
1651 		 * intended for a use on router to reflect RA configurations.
1652 		 */
1653 		/* 0 means 'unspecified' */
1654 		if (ND.linkmtu != 0) {
1655 			if (ND.linkmtu < IPV6_MMTU ||
1656 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1657 				error = EINVAL;
1658 				break;
1659 			}
1660 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1661 		}
1662 
1663 		if (ND.basereachable != 0) {
1664 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1665 
1666 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1667 			if (ND.basereachable != obasereachable)
1668 				ND_IFINFO(ifp)->reachable =
1669 				    ND_COMPUTE_RTIME(ND.basereachable);
1670 		}
1671 		if (ND.retrans != 0)
1672 			ND_IFINFO(ifp)->retrans = ND.retrans;
1673 		if (ND.chlim != 0)
1674 			ND_IFINFO(ifp)->chlim = ND.chlim;
1675 		/* FALLTHROUGH */
1676 	case SIOCSIFINFO_FLAGS:
1677 	{
1678 		struct ifaddr *ifa;
1679 		struct in6_ifaddr *ia;
1680 
1681 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1682 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1683 			/* ifdisabled 1->0 transision */
1684 
1685 			/*
1686 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1687 			 * has an link-local address with IN6_IFF_DUPLICATED,
1688 			 * do not clear ND6_IFF_IFDISABLED.
1689 			 * See RFC 4862, Section 5.4.5.
1690 			 */
1691 			NET_EPOCH_ENTER(et);
1692 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1693 				if (ifa->ifa_addr->sa_family != AF_INET6)
1694 					continue;
1695 				ia = (struct in6_ifaddr *)ifa;
1696 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1697 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1698 					break;
1699 			}
1700 			NET_EPOCH_EXIT(et);
1701 
1702 			if (ifa != NULL) {
1703 				/* LLA is duplicated. */
1704 				ND.flags |= ND6_IFF_IFDISABLED;
1705 				log(LOG_ERR, "Cannot enable an interface"
1706 				    " with a link-local address marked"
1707 				    " duplicate.\n");
1708 			} else {
1709 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1710 				if (ifp->if_flags & IFF_UP)
1711 					in6_if_up(ifp);
1712 			}
1713 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1714 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1715 			/* ifdisabled 0->1 transision */
1716 			/* Mark all IPv6 address as tentative. */
1717 
1718 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1719 			if (V_ip6_dad_count > 0 &&
1720 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1721 				NET_EPOCH_ENTER(et);
1722 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1723 				    ifa_link) {
1724 					if (ifa->ifa_addr->sa_family !=
1725 					    AF_INET6)
1726 						continue;
1727 					ia = (struct in6_ifaddr *)ifa;
1728 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1729 				}
1730 				NET_EPOCH_EXIT(et);
1731 			}
1732 		}
1733 
1734 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1735 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1736 				/* auto_linklocal 0->1 transision */
1737 
1738 				/* If no link-local address on ifp, configure */
1739 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1740 				in6_ifattach(ifp, NULL);
1741 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1742 			    ifp->if_flags & IFF_UP) {
1743 				/*
1744 				 * When the IF already has
1745 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1746 				 * address is assigned, and IFF_UP, try to
1747 				 * assign one.
1748 				 */
1749 				NET_EPOCH_ENTER(et);
1750 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1751 				    ifa_link) {
1752 					if (ifa->ifa_addr->sa_family !=
1753 					    AF_INET6)
1754 						continue;
1755 					ia = (struct in6_ifaddr *)ifa;
1756 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1757 						break;
1758 				}
1759 				NET_EPOCH_EXIT(et);
1760 				if (ifa != NULL)
1761 					/* No LLA is configured. */
1762 					in6_ifattach(ifp, NULL);
1763 			}
1764 		}
1765 		ND_IFINFO(ifp)->flags = ND.flags;
1766 		break;
1767 	}
1768 #undef ND
1769 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1770 		/* sync kernel routing table with the default router list */
1771 		defrouter_reset();
1772 		defrouter_select_fib(RT_ALL_FIBS);
1773 		break;
1774 	case SIOCSPFXFLUSH_IN6:
1775 	{
1776 		/* flush all the prefix advertised by routers */
1777 		struct in6_ifaddr *ia, *ia_next;
1778 		struct nd_prefix *pr, *next;
1779 		struct nd_prhead prl;
1780 
1781 		LIST_INIT(&prl);
1782 
1783 		ND6_WLOCK();
1784 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1785 			if (pr->ndpr_raf_ra_derived)
1786 				nd6_prefix_unlink(pr, &prl);
1787 		}
1788 		ND6_WUNLOCK();
1789 
1790 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1791 			LIST_REMOVE(pr, ndpr_entry);
1792 			/* XXXRW: in6_ifaddrhead locking. */
1793 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1794 			    ia_next) {
1795 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1796 					continue;
1797 
1798 				if (ia->ia6_ndpr == pr)
1799 					in6_purgeaddr(&ia->ia_ifa);
1800 			}
1801 			nd6_prefix_del(pr);
1802 		}
1803 		break;
1804 	}
1805 	case SIOCSRTRFLUSH_IN6:
1806 	{
1807 		/* flush all the default routers */
1808 
1809 		defrouter_reset();
1810 		nd6_defrouter_flush_all();
1811 		defrouter_select_fib(RT_ALL_FIBS);
1812 		break;
1813 	}
1814 	case SIOCGNBRINFO_IN6:
1815 	{
1816 		struct llentry *ln;
1817 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1818 
1819 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1820 			return (error);
1821 
1822 		NET_EPOCH_ENTER(et);
1823 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1824 		NET_EPOCH_EXIT(et);
1825 
1826 		if (ln == NULL) {
1827 			error = EINVAL;
1828 			break;
1829 		}
1830 		nbi->state = ln->ln_state;
1831 		nbi->asked = ln->la_asked;
1832 		nbi->isrouter = ln->ln_router;
1833 		if (ln->la_expire == 0)
1834 			nbi->expire = 0;
1835 		else
1836 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1837 			    (time_second - time_uptime);
1838 		LLE_RUNLOCK(ln);
1839 		break;
1840 	}
1841 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1842 		ndif->ifindex = V_nd6_defifindex;
1843 		break;
1844 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1845 		return (nd6_setdefaultiface(ndif->ifindex));
1846 	}
1847 	return (error);
1848 }
1849 
1850 /*
1851  * Calculates new isRouter value based on provided parameters and
1852  * returns it.
1853  */
1854 static int
1855 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1856     int ln_router)
1857 {
1858 
1859 	/*
1860 	 * ICMP6 type dependent behavior.
1861 	 *
1862 	 * NS: clear IsRouter if new entry
1863 	 * RS: clear IsRouter
1864 	 * RA: set IsRouter if there's lladdr
1865 	 * redir: clear IsRouter if new entry
1866 	 *
1867 	 * RA case, (1):
1868 	 * The spec says that we must set IsRouter in the following cases:
1869 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1870 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1871 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1872 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1873 	 * neighbor cache, this is similar to (6).
1874 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1875 	 *
1876 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1877 	 *							D R
1878 	 *	0	n	n	(1)	c   ?     s
1879 	 *	0	y	n	(2)	c   s     s
1880 	 *	0	n	y	(3)	c   s     s
1881 	 *	0	y	y	(4)	c   s     s
1882 	 *	0	y	y	(5)	c   s     s
1883 	 *	1	--	n	(6) c	c	c s
1884 	 *	1	--	y	(7) c	c   s	c s
1885 	 *
1886 	 *					(c=clear s=set)
1887 	 */
1888 	switch (type & 0xff) {
1889 	case ND_NEIGHBOR_SOLICIT:
1890 		/*
1891 		 * New entry must have is_router flag cleared.
1892 		 */
1893 		if (is_new)					/* (6-7) */
1894 			ln_router = 0;
1895 		break;
1896 	case ND_REDIRECT:
1897 		/*
1898 		 * If the icmp is a redirect to a better router, always set the
1899 		 * is_router flag.  Otherwise, if the entry is newly created,
1900 		 * clear the flag.  [RFC 2461, sec 8.3]
1901 		 */
1902 		if (code == ND_REDIRECT_ROUTER)
1903 			ln_router = 1;
1904 		else {
1905 			if (is_new)				/* (6-7) */
1906 				ln_router = 0;
1907 		}
1908 		break;
1909 	case ND_ROUTER_SOLICIT:
1910 		/*
1911 		 * is_router flag must always be cleared.
1912 		 */
1913 		ln_router = 0;
1914 		break;
1915 	case ND_ROUTER_ADVERT:
1916 		/*
1917 		 * Mark an entry with lladdr as a router.
1918 		 */
1919 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1920 		    (is_new && new_addr)) {			/* (7) */
1921 			ln_router = 1;
1922 		}
1923 		break;
1924 	}
1925 
1926 	return (ln_router);
1927 }
1928 
1929 /*
1930  * Create neighbor cache entry and cache link-layer address,
1931  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1932  *
1933  * type - ICMP6 type
1934  * code - type dependent information
1935  *
1936  */
1937 void
1938 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1939     int lladdrlen, int type, int code)
1940 {
1941 	struct llentry *ln = NULL, *ln_tmp;
1942 	int is_newentry;
1943 	int do_update;
1944 	int olladdr;
1945 	int llchange;
1946 	int flags;
1947 	uint16_t router = 0;
1948 	struct mbuf *chain = NULL;
1949 	u_char linkhdr[LLE_MAX_LINKHDR];
1950 	size_t linkhdrsize;
1951 	int lladdr_off;
1952 
1953 	NET_EPOCH_ASSERT();
1954 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1955 
1956 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1957 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1958 
1959 	/* nothing must be updated for unspecified address */
1960 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1961 		return;
1962 
1963 	/*
1964 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1965 	 * the caller.
1966 	 *
1967 	 * XXX If the link does not have link-layer adderss, what should
1968 	 * we do? (ifp->if_addrlen == 0)
1969 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1970 	 * description on it in NS section (RFC 2461 7.2.3).
1971 	 */
1972 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1973 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1974 	is_newentry = 0;
1975 	if (ln == NULL) {
1976 		flags |= LLE_EXCLUSIVE;
1977 		ln = nd6_alloc(from, 0, ifp);
1978 		if (ln == NULL)
1979 			return;
1980 
1981 		/*
1982 		 * Since we already know all the data for the new entry,
1983 		 * fill it before insertion.
1984 		 */
1985 		if (lladdr != NULL) {
1986 			linkhdrsize = sizeof(linkhdr);
1987 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1988 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
1989 				lltable_free_entry(LLTABLE6(ifp), ln);
1990 				return;
1991 			}
1992 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1993 			    lladdr_off);
1994 		}
1995 
1996 		IF_AFDATA_WLOCK(ifp);
1997 		LLE_WLOCK(ln);
1998 		/* Prefer any existing lle over newly-created one */
1999 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2000 		if (ln_tmp == NULL)
2001 			lltable_link_entry(LLTABLE6(ifp), ln);
2002 		IF_AFDATA_WUNLOCK(ifp);
2003 		if (ln_tmp == NULL) {
2004 			/* No existing lle, mark as new entry (6,7) */
2005 			is_newentry = 1;
2006 			if (lladdr != NULL) {	/* (7) */
2007 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2008 				EVENTHANDLER_INVOKE(lle_event, ln,
2009 				    LLENTRY_RESOLVED);
2010 			}
2011 		} else {
2012 			lltable_free_entry(LLTABLE6(ifp), ln);
2013 			ln = ln_tmp;
2014 			ln_tmp = NULL;
2015 		}
2016 	}
2017 	/* do nothing if static ndp is set */
2018 	if ((ln->la_flags & LLE_STATIC)) {
2019 		if (flags & LLE_EXCLUSIVE)
2020 			LLE_WUNLOCK(ln);
2021 		else
2022 			LLE_RUNLOCK(ln);
2023 		return;
2024 	}
2025 
2026 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2027 	if (olladdr && lladdr) {
2028 		llchange = bcmp(lladdr, ln->ll_addr,
2029 		    ifp->if_addrlen);
2030 	} else if (!olladdr && lladdr)
2031 		llchange = 1;
2032 	else
2033 		llchange = 0;
2034 
2035 	/*
2036 	 * newentry olladdr  lladdr  llchange	(*=record)
2037 	 *	0	n	n	--	(1)
2038 	 *	0	y	n	--	(2)
2039 	 *	0	n	y	y	(3) * STALE
2040 	 *	0	y	y	n	(4) *
2041 	 *	0	y	y	y	(5) * STALE
2042 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2043 	 *	1	--	y	--	(7) * STALE
2044 	 */
2045 
2046 	do_update = 0;
2047 	if (is_newentry == 0 && llchange != 0) {
2048 		do_update = 1;	/* (3,5) */
2049 
2050 		/*
2051 		 * Record source link-layer address
2052 		 * XXX is it dependent to ifp->if_type?
2053 		 */
2054 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2055 			/* Entry was deleted */
2056 			LLE_WUNLOCK(ln);
2057 			return;
2058 		}
2059 
2060 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2061 
2062 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2063 
2064 		if (ln->la_hold != NULL)
2065 			chain = nd6_grab_holdchain(ln);
2066 	}
2067 
2068 	/* Calculates new router status */
2069 	router = nd6_is_router(type, code, is_newentry, olladdr,
2070 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2071 
2072 	ln->ln_router = router;
2073 	/* Mark non-router redirects with special flag */
2074 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2075 		ln->la_flags |= LLE_REDIRECT;
2076 
2077 	if (flags & LLE_EXCLUSIVE)
2078 		LLE_WUNLOCK(ln);
2079 	else
2080 		LLE_RUNLOCK(ln);
2081 
2082 	if (chain != NULL)
2083 		nd6_flush_holdchain(ifp, ln, chain);
2084 	if (do_update)
2085 		nd6_flush_children_holdchain(ifp, ln);
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 struct mbuf *
2143 nd6_grab_holdchain(struct llentry *ln)
2144 {
2145 	struct mbuf *chain;
2146 
2147 	LLE_WLOCK_ASSERT(ln);
2148 
2149 	chain = ln->la_hold;
2150 	ln->la_hold = NULL;
2151 	ln->la_numheld = 0;
2152 
2153 	if (ln->ln_state == ND6_LLINFO_STALE) {
2154 		/*
2155 		 * The first time we send a packet to a
2156 		 * neighbor whose entry is STALE, we have
2157 		 * to change the state to DELAY and a sets
2158 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2159 		 * seconds to ensure do neighbor unreachability
2160 		 * detection on expiration.
2161 		 * (RFC 2461 7.3.3)
2162 		 */
2163 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2164 	}
2165 
2166 	return (chain);
2167 }
2168 
2169 int
2170 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2171     struct sockaddr_in6 *dst, struct route *ro)
2172 {
2173 	int error;
2174 	int ip6len;
2175 	struct ip6_hdr *ip6;
2176 	struct m_tag *mtag;
2177 
2178 #ifdef MAC
2179 	mac_netinet6_nd6_send(ifp, m);
2180 #endif
2181 
2182 	/*
2183 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2184 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2185 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2186 	 * to be diverted to user space.  When re-injected into the kernel,
2187 	 * send_output() will directly dispatch them to the outgoing interface.
2188 	 */
2189 	if (send_sendso_input_hook != NULL) {
2190 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2191 		if (mtag != NULL) {
2192 			ip6 = mtod(m, struct ip6_hdr *);
2193 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2194 			/* Use the SEND socket */
2195 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2196 			    ip6len);
2197 			/* -1 == no app on SEND socket */
2198 			if (error == 0 || error != -1)
2199 			    return (error);
2200 		}
2201 	}
2202 
2203 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2204 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2205 	    mtod(m, struct ip6_hdr *));
2206 
2207 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2208 		origifp = ifp;
2209 
2210 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2211 	return (error);
2212 }
2213 
2214 /*
2215  * Lookup link headerfor @sa_dst address. Stores found
2216  * data in @desten buffer. Copy of lle ln_flags can be also
2217  * saved in @pflags if @pflags is non-NULL.
2218  *
2219  * If destination LLE does not exists or lle state modification
2220  * is required, call "slow" version.
2221  *
2222  * Return values:
2223  * - 0 on success (address copied to buffer).
2224  * - EWOULDBLOCK (no local error, but address is still unresolved)
2225  * - other errors (alloc failure, etc)
2226  */
2227 int
2228 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2229     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2230     struct llentry **plle)
2231 {
2232 	struct llentry *ln = NULL;
2233 	const struct sockaddr_in6 *dst6;
2234 
2235 	NET_EPOCH_ASSERT();
2236 
2237 	if (pflags != NULL)
2238 		*pflags = 0;
2239 
2240 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2241 
2242 	/* discard the packet if IPv6 operation is disabled on the interface */
2243 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2244 		m_freem(m);
2245 		return (ENETDOWN); /* better error? */
2246 	}
2247 
2248 	if (m != NULL && m->m_flags & M_MCAST) {
2249 		switch (ifp->if_type) {
2250 		case IFT_ETHER:
2251 		case IFT_L2VLAN:
2252 		case IFT_BRIDGE:
2253 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2254 						 desten);
2255 			return (0);
2256 		default:
2257 			m_freem(m);
2258 			return (EAFNOSUPPORT);
2259 		}
2260 	}
2261 
2262 	int family = gw_flags >> 16;
2263 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2264 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2265 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2266 		/* Entry found, let's copy lle info */
2267 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2268 		if (pflags != NULL)
2269 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2270 		llentry_provide_feedback(ln);
2271 		if (plle) {
2272 			LLE_ADDREF(ln);
2273 			*plle = ln;
2274 			LLE_WUNLOCK(ln);
2275 		}
2276 		return (0);
2277 	} else if (plle && ln)
2278 		LLE_WUNLOCK(ln);
2279 
2280 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2281 }
2282 
2283 /*
2284  * Finds or creates a new llentry for @addr and @family.
2285  * Returns wlocked llentry or NULL.
2286  *
2287  *
2288  * Child LLEs.
2289  *
2290  * Do not have their own state machine (gets marked as static)
2291  *  settimer bails out for child LLEs just in case.
2292  *
2293  * Locking order: parent lle gets locked first, chen goes the child.
2294  */
2295 static __noinline struct llentry *
2296 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2297 {
2298 	struct llentry *child_lle = NULL;
2299 	struct llentry *lle, *lle_tmp;
2300 
2301 	lle = nd6_alloc(addr, 0, ifp);
2302 	if (lle != NULL && family != AF_INET6) {
2303 		child_lle = nd6_alloc(addr, 0, ifp);
2304 		if (child_lle == NULL) {
2305 			lltable_free_entry(LLTABLE6(ifp), lle);
2306 			return (NULL);
2307 		}
2308 		child_lle->r_family = family;
2309 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2310 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2311 	}
2312 
2313 	if (lle == NULL) {
2314 		char ip6buf[INET6_ADDRSTRLEN];
2315 		log(LOG_DEBUG,
2316 		    "nd6_get_llentry: can't allocate llinfo for %s "
2317 		    "(ln=%p)\n",
2318 		    ip6_sprintf(ip6buf, addr), lle);
2319 		return (NULL);
2320 	}
2321 
2322 	IF_AFDATA_WLOCK(ifp);
2323 	LLE_WLOCK(lle);
2324 	/* Prefer any existing entry over newly-created one */
2325 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2326 	if (lle_tmp == NULL)
2327 		lltable_link_entry(LLTABLE6(ifp), lle);
2328 	else {
2329 		lltable_free_entry(LLTABLE6(ifp), lle);
2330 		lle = lle_tmp;
2331 	}
2332 	if (child_lle != NULL) {
2333 		/* Check if child lle for the same family exists */
2334 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2335 		LLE_WLOCK(child_lle);
2336 		if (lle_tmp == NULL) {
2337 			/* Attach */
2338 			lltable_link_child_entry(lle, child_lle);
2339 		} else {
2340 			/* child lle already exists, free newly-created one */
2341 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2342 			child_lle = lle_tmp;
2343 		}
2344 		LLE_WUNLOCK(lle);
2345 		lle = child_lle;
2346 	}
2347 	IF_AFDATA_WUNLOCK(ifp);
2348 	return (lle);
2349 }
2350 
2351 /*
2352  * Do L2 address resolution for @sa_dst address. Stores found
2353  * address in @desten buffer. Copy of lle ln_flags can be also
2354  * saved in @pflags if @pflags is non-NULL.
2355  *
2356  * Heavy version.
2357  * Function assume that destination LLE does not exist,
2358  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2359  *
2360  * Set noinline to be dtrace-friendly
2361  */
2362 static __noinline int
2363 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2364     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2365     struct llentry **plle)
2366 {
2367 	struct llentry *lle = NULL;
2368 	struct in6_addr *psrc, src;
2369 	int send_ns, ll_len;
2370 	char *lladdr;
2371 
2372 	NET_EPOCH_ASSERT();
2373 
2374 	/*
2375 	 * Address resolution or Neighbor Unreachability Detection
2376 	 * for the next hop.
2377 	 * At this point, the destination of the packet must be a unicast
2378 	 * or an anycast address(i.e. not a multicast).
2379 	 */
2380 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2381 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2382 		/*
2383 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2384 		 * the condition below is not very efficient.  But we believe
2385 		 * it is tolerable, because this should be a rare case.
2386 		 */
2387 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2388 	}
2389 
2390 	if (lle == NULL) {
2391 		m_freem(m);
2392 		return (ENOBUFS);
2393 	}
2394 
2395 	LLE_WLOCK_ASSERT(lle);
2396 
2397 	/*
2398 	 * The first time we send a packet to a neighbor whose entry is
2399 	 * STALE, we have to change the state to DELAY and a sets a timer to
2400 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2401 	 * neighbor unreachability detection on expiration.
2402 	 * (RFC 2461 7.3.3)
2403 	 */
2404 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2405 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2406 
2407 	/*
2408 	 * If the neighbor cache entry has a state other than INCOMPLETE
2409 	 * (i.e. its link-layer address is already resolved), just
2410 	 * send the packet.
2411 	 */
2412 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2413 		if (flags & LLE_ADDRONLY) {
2414 			lladdr = lle->ll_addr;
2415 			ll_len = ifp->if_addrlen;
2416 		} else {
2417 			lladdr = lle->r_linkdata;
2418 			ll_len = lle->r_hdrlen;
2419 		}
2420 		bcopy(lladdr, desten, ll_len);
2421 		if (pflags != NULL)
2422 			*pflags = lle->la_flags;
2423 		if (plle) {
2424 			LLE_ADDREF(lle);
2425 			*plle = lle;
2426 		}
2427 		LLE_WUNLOCK(lle);
2428 		return (0);
2429 	}
2430 
2431 	/*
2432 	 * There is a neighbor cache entry, but no ethernet address
2433 	 * response yet.  Append this latest packet to the end of the
2434 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2435 	 * the oldest packet in the queue will be removed.
2436 	 */
2437 	if (m != NULL) {
2438 		size_t dropped;
2439 
2440 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2441 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2442 	}
2443 
2444 	/*
2445 	 * If there has been no NS for the neighbor after entering the
2446 	 * INCOMPLETE state, send the first solicitation.
2447 	 * Note that for newly-created lle la_asked will be 0,
2448 	 * so we will transition from ND6_LLINFO_NOSTATE to
2449 	 * ND6_LLINFO_INCOMPLETE state here.
2450 	 */
2451 	psrc = NULL;
2452 	send_ns = 0;
2453 
2454 	/* If we have child lle, switch to the parent to send NS */
2455 	if (lle->la_flags & LLE_CHILD) {
2456 		struct llentry *lle_parent = lle->lle_parent;
2457 		LLE_WUNLOCK(lle);
2458 		lle = lle_parent;
2459 		LLE_WLOCK(lle);
2460 	}
2461 	if (lle->la_asked == 0) {
2462 		lle->la_asked++;
2463 		send_ns = 1;
2464 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2465 
2466 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2467 	}
2468 	LLE_WUNLOCK(lle);
2469 	if (send_ns != 0)
2470 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2471 
2472 	return (EWOULDBLOCK);
2473 }
2474 
2475 /*
2476  * Do L2 address resolution for @sa_dst address. Stores found
2477  * address in @desten buffer. Copy of lle ln_flags can be also
2478  * saved in @pflags if @pflags is non-NULL.
2479  *
2480  * Return values:
2481  * - 0 on success (address copied to buffer).
2482  * - EWOULDBLOCK (no local error, but address is still unresolved)
2483  * - other errors (alloc failure, etc)
2484  */
2485 int
2486 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2487     char *desten, uint32_t *pflags)
2488 {
2489 	int error;
2490 
2491 	flags |= LLE_ADDRONLY;
2492 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2493 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2494 	return (error);
2495 }
2496 
2497 int
2498 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2499 {
2500 	struct mbuf *m, *m_head;
2501 	struct sockaddr_in6 dst6;
2502 	int error = 0;
2503 
2504 	NET_EPOCH_ASSERT();
2505 
2506 	struct route_in6 ro = {
2507 		.ro_prepend = lle->r_linkdata,
2508 		.ro_plen = lle->r_hdrlen,
2509 	};
2510 
2511 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2512 	m_head = chain;
2513 
2514 	while (m_head) {
2515 		m = m_head;
2516 		m_head = m_head->m_nextpkt;
2517 		m->m_nextpkt = NULL;
2518 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2519 	}
2520 
2521 	/*
2522 	 * XXX
2523 	 * note that intermediate errors are blindly ignored
2524 	 */
2525 	return (error);
2526 }
2527 
2528 __noinline void
2529 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2530 {
2531 	struct llentry *child_lle;
2532 	struct mbuf *chain;
2533 
2534 	NET_EPOCH_ASSERT();
2535 
2536 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2537 		LLE_WLOCK(child_lle);
2538 		chain = nd6_grab_holdchain(child_lle);
2539 		LLE_WUNLOCK(child_lle);
2540 		nd6_flush_holdchain(ifp, child_lle, chain);
2541 	}
2542 }
2543 
2544 static int
2545 nd6_need_cache(struct ifnet *ifp)
2546 {
2547 	/*
2548 	 * XXX: we currently do not make neighbor cache on any interface
2549 	 * other than Ethernet and GIF.
2550 	 *
2551 	 * RFC2893 says:
2552 	 * - unidirectional tunnels needs no ND
2553 	 */
2554 	switch (ifp->if_type) {
2555 	case IFT_ETHER:
2556 	case IFT_IEEE1394:
2557 	case IFT_L2VLAN:
2558 	case IFT_INFINIBAND:
2559 	case IFT_BRIDGE:
2560 	case IFT_PROPVIRTUAL:
2561 		return (1);
2562 	default:
2563 		return (0);
2564 	}
2565 }
2566 
2567 /*
2568  * Add pernament ND6 link-layer record for given
2569  * interface address.
2570  *
2571  * Very similar to IPv4 arp_ifinit(), but:
2572  * 1) IPv6 DAD is performed in different place
2573  * 2) It is called by IPv6 protocol stack in contrast to
2574  * arp_ifinit() which is typically called in SIOCSIFADDR
2575  * driver ioctl handler.
2576  *
2577  */
2578 int
2579 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2580 {
2581 	struct ifnet *ifp;
2582 	struct llentry *ln, *ln_tmp;
2583 	struct sockaddr *dst;
2584 
2585 	ifp = ia->ia_ifa.ifa_ifp;
2586 	if (nd6_need_cache(ifp) == 0)
2587 		return (0);
2588 
2589 	dst = (struct sockaddr *)&ia->ia_addr;
2590 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2591 	if (ln == NULL)
2592 		return (ENOBUFS);
2593 
2594 	IF_AFDATA_WLOCK(ifp);
2595 	LLE_WLOCK(ln);
2596 	/* Unlink any entry if exists */
2597 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2598 	if (ln_tmp != NULL)
2599 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2600 	lltable_link_entry(LLTABLE6(ifp), ln);
2601 	IF_AFDATA_WUNLOCK(ifp);
2602 
2603 	if (ln_tmp != NULL)
2604 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2605 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2606 
2607 	LLE_WUNLOCK(ln);
2608 	if (ln_tmp != NULL)
2609 		llentry_free(ln_tmp);
2610 
2611 	return (0);
2612 }
2613 
2614 /*
2615  * Removes either all lle entries for given @ia, or lle
2616  * corresponding to @ia address.
2617  */
2618 void
2619 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2620 {
2621 	struct sockaddr_in6 mask, addr;
2622 	struct sockaddr *saddr, *smask;
2623 	struct ifnet *ifp;
2624 
2625 	ifp = ia->ia_ifa.ifa_ifp;
2626 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2627 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2628 	saddr = (struct sockaddr *)&addr;
2629 	smask = (struct sockaddr *)&mask;
2630 
2631 	if (all != 0)
2632 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2633 	else
2634 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2635 }
2636 
2637 static int
2638 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2639 {
2640 	struct in6_prefix p;
2641 	struct sockaddr_in6 s6;
2642 	struct nd_prefix *pr;
2643 	struct nd_pfxrouter *pfr;
2644 	time_t maxexpire;
2645 	int error;
2646 	char ip6buf[INET6_ADDRSTRLEN];
2647 
2648 	if (req->newptr)
2649 		return (EPERM);
2650 
2651 	error = sysctl_wire_old_buffer(req, 0);
2652 	if (error != 0)
2653 		return (error);
2654 
2655 	bzero(&p, sizeof(p));
2656 	p.origin = PR_ORIG_RA;
2657 	bzero(&s6, sizeof(s6));
2658 	s6.sin6_family = AF_INET6;
2659 	s6.sin6_len = sizeof(s6);
2660 
2661 	ND6_RLOCK();
2662 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2663 		if (!pr->ndpr_raf_ra_derived)
2664 			continue;
2665 		p.prefix = pr->ndpr_prefix;
2666 		if (sa6_recoverscope(&p.prefix)) {
2667 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2668 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2669 			/* XXX: press on... */
2670 		}
2671 		p.raflags = pr->ndpr_raf;
2672 		p.prefixlen = pr->ndpr_plen;
2673 		p.vltime = pr->ndpr_vltime;
2674 		p.pltime = pr->ndpr_pltime;
2675 		p.if_index = pr->ndpr_ifp->if_index;
2676 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2677 			p.expire = 0;
2678 		else {
2679 			/* XXX: we assume time_t is signed. */
2680 			maxexpire = (-1) &
2681 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2682 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2683 				p.expire = pr->ndpr_lastupdate +
2684 				    pr->ndpr_vltime +
2685 				    (time_second - time_uptime);
2686 			else
2687 				p.expire = maxexpire;
2688 		}
2689 		p.refcnt = pr->ndpr_addrcnt;
2690 		p.flags = pr->ndpr_stateflags;
2691 		p.advrtrs = 0;
2692 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2693 			p.advrtrs++;
2694 		error = SYSCTL_OUT(req, &p, sizeof(p));
2695 		if (error != 0)
2696 			break;
2697 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2698 			s6.sin6_addr = pfr->router->rtaddr;
2699 			if (sa6_recoverscope(&s6))
2700 				log(LOG_ERR,
2701 				    "scope error in prefix list (%s)\n",
2702 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2703 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2704 			if (error != 0)
2705 				goto out;
2706 		}
2707 	}
2708 out:
2709 	ND6_RUNLOCK();
2710 	return (error);
2711 }
2712 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2713 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2714 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2715 	"NDP prefix list");
2716 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2717 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2718 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2719 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2720