xref: /freebsd/sys/netinet6/nd6.c (revision 8881d206f4e68b564c2c5f50fc717086fc3e827a)
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 #include "opt_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/eventhandler.h>
44 #include <sys/callout.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/mutex.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/rwlock.h>
57 #include <sys/queue.h>
58 #include <sys/sdt.h>
59 #include <sys/sysctl.h>
60 
61 #include <net/if.h>
62 #include <net/if_var.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/route/route_ctl.h>
67 #include <net/route/nhop.h>
68 #include <net/vnet.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.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 int 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 void clear_llinfo_pqueue(struct llentry *);
142 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
143     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
144 static int nd6_need_cache(struct ifnet *);
145 
146 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
147 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
148 
149 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
150 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
151 
152 SYSCTL_DECL(_net_inet6_icmp6);
153 
154 static void
155 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
156 {
157 	struct rt_addrinfo rtinfo;
158 	struct sockaddr_in6 dst;
159 	struct sockaddr_dl gw;
160 	struct ifnet *ifp;
161 	int type;
162 	int fibnum;
163 
164 	LLE_WLOCK_ASSERT(lle);
165 
166 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
167 		return;
168 
169 	switch (evt) {
170 	case LLENTRY_RESOLVED:
171 		type = RTM_ADD;
172 		KASSERT(lle->la_flags & LLE_VALID,
173 		    ("%s: %p resolved but not valid?", __func__, lle));
174 		break;
175 	case LLENTRY_EXPIRED:
176 		type = RTM_DELETE;
177 		break;
178 	default:
179 		return;
180 	}
181 
182 	ifp = lltable_get_ifp(lle->lle_tbl);
183 
184 	bzero(&dst, sizeof(dst));
185 	bzero(&gw, sizeof(gw));
186 	bzero(&rtinfo, sizeof(rtinfo));
187 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
188 	dst.sin6_scope_id = in6_getscopezone(ifp,
189 	    in6_addrscope(&dst.sin6_addr));
190 	gw.sdl_len = sizeof(struct sockaddr_dl);
191 	gw.sdl_family = AF_LINK;
192 	gw.sdl_alen = ifp->if_addrlen;
193 	gw.sdl_index = ifp->if_index;
194 	gw.sdl_type = ifp->if_type;
195 	if (evt == LLENTRY_RESOLVED)
196 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
197 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
198 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
199 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
200 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
201 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
202 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
203 }
204 
205 /*
206  * A handler for interface link layer address change event.
207  */
208 static void
209 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
210 {
211 	if (ifp->if_afdata[AF_INET6] == NULL)
212 		return;
213 
214 	lltable_update_ifaddr(LLTABLE6(ifp));
215 }
216 
217 void
218 nd6_init(void)
219 {
220 
221 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
222 	rw_init(&V_nd6_lock, "nd6 list");
223 
224 	LIST_INIT(&V_nd_prefix);
225 	nd6_defrouter_init();
226 
227 	/* Start timers. */
228 	callout_init(&V_nd6_slowtimo_ch, 1);
229 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
230 	    nd6_slowtimo, curvnet);
231 
232 	callout_init(&V_nd6_timer_ch, 1);
233 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
234 
235 	nd6_dad_init();
236 	if (IS_DEFAULT_VNET(curvnet)) {
237 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
238 		    NULL, EVENTHANDLER_PRI_ANY);
239 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
240 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
241 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
242 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
243 	}
244 }
245 
246 #ifdef VIMAGE
247 void
248 nd6_destroy()
249 {
250 
251 	callout_drain(&V_nd6_slowtimo_ch);
252 	callout_drain(&V_nd6_timer_ch);
253 	if (IS_DEFAULT_VNET(curvnet)) {
254 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
255 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
256 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
257 	}
258 	rw_destroy(&V_nd6_lock);
259 	mtx_destroy(&V_nd6_onlink_mtx);
260 }
261 #endif
262 
263 struct nd_ifinfo *
264 nd6_ifattach(struct ifnet *ifp)
265 {
266 	struct nd_ifinfo *nd;
267 
268 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
269 	nd->initialized = 1;
270 
271 	nd->chlim = IPV6_DEFHLIM;
272 	nd->basereachable = REACHABLE_TIME;
273 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
274 	nd->retrans = RETRANS_TIMER;
275 
276 	nd->flags = ND6_IFF_PERFORMNUD;
277 
278 	/* Set IPv6 disabled on all interfaces but loopback by default. */
279 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
280 		nd->flags |= ND6_IFF_IFDISABLED;
281 
282 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
283 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
284 	 * default regardless of the V_ip6_auto_linklocal configuration to
285 	 * give a reasonable default behavior.
286 	 */
287 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
288 	    (ifp->if_flags & IFF_LOOPBACK))
289 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
290 	/*
291 	 * A loopback interface does not need to accept RTADV.
292 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
293 	 * default regardless of the V_ip6_accept_rtadv configuration to
294 	 * prevent the interface from accepting RA messages arrived
295 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
296 	 */
297 	if (V_ip6_accept_rtadv &&
298 	    !(ifp->if_flags & IFF_LOOPBACK) &&
299 	    (ifp->if_type != IFT_BRIDGE)) {
300 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
301 			/* If we globally accept rtadv, assume IPv6 on. */
302 			nd->flags &= ~ND6_IFF_IFDISABLED;
303 	}
304 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
305 		nd->flags |= ND6_IFF_NO_RADR;
306 
307 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
308 	nd6_setmtu0(ifp, nd);
309 
310 	return nd;
311 }
312 
313 void
314 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
315 {
316 	struct epoch_tracker et;
317 	struct ifaddr *ifa, *next;
318 
319 	NET_EPOCH_ENTER(et);
320 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
321 		if (ifa->ifa_addr->sa_family != AF_INET6)
322 			continue;
323 
324 		/* stop DAD processing */
325 		nd6_dad_stop(ifa);
326 	}
327 	NET_EPOCH_EXIT(et);
328 
329 	free(nd, M_IP6NDP);
330 }
331 
332 /*
333  * Reset ND level link MTU. This function is called when the physical MTU
334  * changes, which means we might have to adjust the ND level MTU.
335  */
336 void
337 nd6_setmtu(struct ifnet *ifp)
338 {
339 	if (ifp->if_afdata[AF_INET6] == NULL)
340 		return;
341 
342 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
343 }
344 
345 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
346 void
347 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
348 {
349 	u_int32_t omaxmtu;
350 
351 	omaxmtu = ndi->maxmtu;
352 	ndi->maxmtu = ifp->if_mtu;
353 
354 	/*
355 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
356 	 * undesirable situation.  We thus notify the operator of the change
357 	 * explicitly.  The check for omaxmtu is necessary to restrict the
358 	 * log to the case of changing the MTU, not initializing it.
359 	 */
360 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
361 		log(LOG_NOTICE, "nd6_setmtu0: "
362 		    "new link MTU on %s (%lu) is too small for IPv6\n",
363 		    if_name(ifp), (unsigned long)ndi->maxmtu);
364 	}
365 
366 	if (ndi->maxmtu > V_in6_maxmtu)
367 		in6_setmaxmtu(); /* check all interfaces just in case */
368 
369 }
370 
371 void
372 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
373 {
374 
375 	bzero(ndopts, sizeof(*ndopts));
376 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
377 	ndopts->nd_opts_last
378 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
379 
380 	if (icmp6len == 0) {
381 		ndopts->nd_opts_done = 1;
382 		ndopts->nd_opts_search = NULL;
383 	}
384 }
385 
386 /*
387  * Take one ND option.
388  */
389 struct nd_opt_hdr *
390 nd6_option(union nd_opts *ndopts)
391 {
392 	struct nd_opt_hdr *nd_opt;
393 	int olen;
394 
395 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
396 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
397 	    __func__));
398 	if (ndopts->nd_opts_search == NULL)
399 		return NULL;
400 	if (ndopts->nd_opts_done)
401 		return NULL;
402 
403 	nd_opt = ndopts->nd_opts_search;
404 
405 	/* make sure nd_opt_len is inside the buffer */
406 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
407 		bzero(ndopts, sizeof(*ndopts));
408 		return NULL;
409 	}
410 
411 	olen = nd_opt->nd_opt_len << 3;
412 	if (olen == 0) {
413 		/*
414 		 * Message validation requires that all included
415 		 * options have a length that is greater than zero.
416 		 */
417 		bzero(ndopts, sizeof(*ndopts));
418 		return NULL;
419 	}
420 
421 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
422 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
423 		/* option overruns the end of buffer, invalid */
424 		bzero(ndopts, sizeof(*ndopts));
425 		return NULL;
426 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
427 		/* reached the end of options chain */
428 		ndopts->nd_opts_done = 1;
429 		ndopts->nd_opts_search = NULL;
430 	}
431 	return nd_opt;
432 }
433 
434 /*
435  * Parse multiple ND options.
436  * This function is much easier to use, for ND routines that do not need
437  * multiple options of the same type.
438  */
439 int
440 nd6_options(union nd_opts *ndopts)
441 {
442 	struct nd_opt_hdr *nd_opt;
443 	int i = 0;
444 
445 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
446 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
447 	    __func__));
448 	if (ndopts->nd_opts_search == NULL)
449 		return 0;
450 
451 	while (1) {
452 		nd_opt = nd6_option(ndopts);
453 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
454 			/*
455 			 * Message validation requires that all included
456 			 * options have a length that is greater than zero.
457 			 */
458 			ICMP6STAT_INC(icp6s_nd_badopt);
459 			bzero(ndopts, sizeof(*ndopts));
460 			return -1;
461 		}
462 
463 		if (nd_opt == NULL)
464 			goto skip1;
465 
466 		switch (nd_opt->nd_opt_type) {
467 		case ND_OPT_SOURCE_LINKADDR:
468 		case ND_OPT_TARGET_LINKADDR:
469 		case ND_OPT_MTU:
470 		case ND_OPT_REDIRECTED_HEADER:
471 		case ND_OPT_NONCE:
472 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
473 				nd6log((LOG_INFO,
474 				    "duplicated ND6 option found (type=%d)\n",
475 				    nd_opt->nd_opt_type));
476 				/* XXX bark? */
477 			} else {
478 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
479 					= nd_opt;
480 			}
481 			break;
482 		case ND_OPT_PREFIX_INFORMATION:
483 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
484 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
485 					= nd_opt;
486 			}
487 			ndopts->nd_opts_pi_end =
488 				(struct nd_opt_prefix_info *)nd_opt;
489 			break;
490 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
491 		case ND_OPT_RDNSS:	/* RFC 6106 */
492 		case ND_OPT_DNSSL:	/* RFC 6106 */
493 			/*
494 			 * Silently ignore options we know and do not care about
495 			 * in the kernel.
496 			 */
497 			break;
498 		default:
499 			/*
500 			 * Unknown options must be silently ignored,
501 			 * to accommodate future extension to the protocol.
502 			 */
503 			nd6log((LOG_DEBUG,
504 			    "nd6_options: unsupported option %d - "
505 			    "option ignored\n", nd_opt->nd_opt_type));
506 		}
507 
508 skip1:
509 		i++;
510 		if (i > V_nd6_maxndopt) {
511 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
512 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
513 			break;
514 		}
515 
516 		if (ndopts->nd_opts_done)
517 			break;
518 	}
519 
520 	return 0;
521 }
522 
523 /*
524  * ND6 timer routine to handle ND6 entries
525  */
526 static void
527 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
528 {
529 	int canceled;
530 
531 	LLE_WLOCK_ASSERT(ln);
532 
533 	/* Do not schedule timers for child LLEs. */
534 	if (ln->la_flags & LLE_CHILD)
535 		return;
536 
537 	if (tick < 0) {
538 		ln->la_expire = 0;
539 		ln->ln_ntick = 0;
540 		canceled = callout_stop(&ln->lle_timer);
541 	} else {
542 		ln->la_expire = time_uptime + tick / hz;
543 		LLE_ADDREF(ln);
544 		if (tick > INT_MAX) {
545 			ln->ln_ntick = tick - INT_MAX;
546 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
547 			    nd6_llinfo_timer, ln);
548 		} else {
549 			ln->ln_ntick = 0;
550 			canceled = callout_reset(&ln->lle_timer, tick,
551 			    nd6_llinfo_timer, ln);
552 		}
553 	}
554 	if (canceled > 0)
555 		LLE_REMREF(ln);
556 }
557 
558 /*
559  * Gets source address of the first packet in hold queue
560  * and stores it in @src.
561  * Returns pointer to @src (if hold queue is not empty) or NULL.
562  *
563  * Set noinline to be dtrace-friendly
564  */
565 static __noinline struct in6_addr *
566 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
567 {
568 	struct ip6_hdr hdr;
569 	struct mbuf *m;
570 
571 	if (ln->la_hold == NULL)
572 		return (NULL);
573 
574 	/*
575 	 * assume every packet in la_hold has the same IP header
576 	 */
577 	m = ln->la_hold;
578 	if (sizeof(hdr) > m->m_len)
579 		return (NULL);
580 
581 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
582 	*src = hdr.ip6_src;
583 
584 	return (src);
585 }
586 
587 /*
588  * Checks if we need to switch from STALE state.
589  *
590  * RFC 4861 requires switching from STALE to DELAY state
591  * on first packet matching entry, waiting V_nd6_delay and
592  * transition to PROBE state (if upper layer confirmation was
593  * not received).
594  *
595  * This code performs a bit differently:
596  * On packet hit we don't change state (but desired state
597  * can be guessed by control plane). However, after V_nd6_delay
598  * seconds code will transition to PROBE state (so DELAY state
599  * is kinda skipped in most situations).
600  *
601  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
602  * we perform the following upon entering STALE state:
603  *
604  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
605  * if packet was transmitted at the start of given interval, we
606  * would be able to switch to PROBE state in V_nd6_delay seconds
607  * as user expects.
608  *
609  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
610  * lle in STALE state (remaining timer value stored in lle_remtime).
611  *
612  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
613  * seconds ago.
614  *
615  * Returns non-zero value if the entry is still STALE (storing
616  * the next timer interval in @pdelay).
617  *
618  * Returns zero value if original timer expired or we need to switch to
619  * PROBE (store that in @do_switch variable).
620  */
621 static int
622 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
623 {
624 	int nd_delay, nd_gctimer;
625 	time_t lle_hittime;
626 	long delay;
627 
628 	*do_switch = 0;
629 	nd_gctimer = V_nd6_gctimer;
630 	nd_delay = V_nd6_delay;
631 
632 	lle_hittime = llentry_get_hittime(lle);
633 
634 	if (lle_hittime == 0) {
635 		/*
636 		 * Datapath feedback has been requested upon entering
637 		 * STALE state. No packets has been passed using this lle.
638 		 * Ask for the timer reschedule and keep STALE state.
639 		 */
640 		delay = (long)(MIN(nd_gctimer, nd_delay));
641 		delay *= hz;
642 		if (lle->lle_remtime > delay)
643 			lle->lle_remtime -= delay;
644 		else {
645 			delay = lle->lle_remtime;
646 			lle->lle_remtime = 0;
647 		}
648 
649 		if (delay == 0) {
650 			/*
651 			 * The original ng6_gctime timeout ended,
652 			 * no more rescheduling.
653 			 */
654 			return (0);
655 		}
656 
657 		*pdelay = delay;
658 		return (1);
659 	}
660 
661 	/*
662 	 * Packet received. Verify timestamp
663 	 */
664 	delay = (long)(time_uptime - lle_hittime);
665 	if (delay < nd_delay) {
666 		/*
667 		 * V_nd6_delay still not passed since the first
668 		 * hit in STALE state.
669 		 * Reschedule timer and return.
670 		 */
671 		*pdelay = (long)(nd_delay - delay) * hz;
672 		return (1);
673 	}
674 
675 	/* Request switching to probe */
676 	*do_switch = 1;
677 	return (0);
678 }
679 
680 /*
681  * Switch @lle state to new state optionally arming timers.
682  *
683  * Set noinline to be dtrace-friendly
684  */
685 __noinline void
686 nd6_llinfo_setstate(struct llentry *lle, int newstate)
687 {
688 	struct ifnet *ifp;
689 	int nd_gctimer, nd_delay;
690 	long delay, remtime;
691 
692 	delay = 0;
693 	remtime = 0;
694 
695 	switch (newstate) {
696 	case ND6_LLINFO_INCOMPLETE:
697 		ifp = lle->lle_tbl->llt_ifp;
698 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
699 		break;
700 	case ND6_LLINFO_REACHABLE:
701 		if (!ND6_LLINFO_PERMANENT(lle)) {
702 			ifp = lle->lle_tbl->llt_ifp;
703 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
704 		}
705 		break;
706 	case ND6_LLINFO_STALE:
707 
708 		llentry_request_feedback(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 				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 is a neighbor or not, ignoring
1229  * the actual neighbor cache.  The neighbor cache is ignored in order
1230  * to not reenter the routing code from within itself.
1231  */
1232 static int
1233 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1234 {
1235 	struct nd_prefix *pr;
1236 	struct ifaddr *ifa;
1237 	struct rt_addrinfo info;
1238 	struct sockaddr_in6 rt_key;
1239 	const struct sockaddr *dst6;
1240 	uint64_t genid;
1241 	int error, fibnum;
1242 
1243 	/*
1244 	 * A link-local address is always a neighbor.
1245 	 * XXX: a link does not necessarily specify a single interface.
1246 	 */
1247 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1248 		struct sockaddr_in6 sin6_copy;
1249 		u_int32_t zone;
1250 
1251 		/*
1252 		 * We need sin6_copy since sa6_recoverscope() may modify the
1253 		 * content (XXX).
1254 		 */
1255 		sin6_copy = *addr;
1256 		if (sa6_recoverscope(&sin6_copy))
1257 			return (0); /* XXX: should be impossible */
1258 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1259 			return (0);
1260 		if (sin6_copy.sin6_scope_id == zone)
1261 			return (1);
1262 		else
1263 			return (0);
1264 	}
1265 
1266 	bzero(&rt_key, sizeof(rt_key));
1267 	bzero(&info, sizeof(info));
1268 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1269 
1270 	/*
1271 	 * If the address matches one of our addresses,
1272 	 * it should be a neighbor.
1273 	 * If the address matches one of our on-link prefixes, it should be a
1274 	 * neighbor.
1275 	 */
1276 	ND6_RLOCK();
1277 restart:
1278 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1279 		if (pr->ndpr_ifp != ifp)
1280 			continue;
1281 
1282 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1283 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1284 
1285 			/*
1286 			 * We only need to check all FIBs if add_addr_allfibs
1287 			 * is unset. If set, checking any FIB will suffice.
1288 			 */
1289 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1290 			for (; fibnum < rt_numfibs; fibnum++) {
1291 				genid = V_nd6_list_genid;
1292 				ND6_RUNLOCK();
1293 
1294 				/*
1295 				 * Restore length field before
1296 				 * retrying lookup
1297 				 */
1298 				rt_key.sin6_len = sizeof(rt_key);
1299 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1300 						        &info);
1301 
1302 				ND6_RLOCK();
1303 				if (genid != V_nd6_list_genid)
1304 					goto restart;
1305 				if (error == 0)
1306 					break;
1307 			}
1308 			if (error != 0)
1309 				continue;
1310 
1311 			/*
1312 			 * This is the case where multiple interfaces
1313 			 * have the same prefix, but only one is installed
1314 			 * into the routing table and that prefix entry
1315 			 * is not the one being examined here.
1316 			 */
1317 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1318 			    &rt_key.sin6_addr))
1319 				continue;
1320 		}
1321 
1322 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1323 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1324 			ND6_RUNLOCK();
1325 			return (1);
1326 		}
1327 	}
1328 	ND6_RUNLOCK();
1329 
1330 	/*
1331 	 * If the address is assigned on the node of the other side of
1332 	 * a p2p interface, the address should be a neighbor.
1333 	 */
1334 	if (ifp->if_flags & IFF_POINTOPOINT) {
1335 		struct epoch_tracker et;
1336 
1337 		NET_EPOCH_ENTER(et);
1338 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1339 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1340 				continue;
1341 			if (ifa->ifa_dstaddr != NULL &&
1342 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1343 				NET_EPOCH_EXIT(et);
1344 				return 1;
1345 			}
1346 		}
1347 		NET_EPOCH_EXIT(et);
1348 	}
1349 
1350 	/*
1351 	 * If the default router list is empty, all addresses are regarded
1352 	 * as on-link, and thus, as a neighbor.
1353 	 */
1354 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1355 	    nd6_defrouter_list_empty() &&
1356 	    V_nd6_defifindex == ifp->if_index) {
1357 		return (1);
1358 	}
1359 
1360 	return (0);
1361 }
1362 
1363 /*
1364  * Detect if a given IPv6 address identifies a neighbor on a given link.
1365  * XXX: should take care of the destination of a p2p link?
1366  */
1367 int
1368 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1369 {
1370 	struct llentry *lle;
1371 	int rc = 0;
1372 
1373 	NET_EPOCH_ASSERT();
1374 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1375 	if (nd6_is_new_addr_neighbor(addr, ifp))
1376 		return (1);
1377 
1378 	/*
1379 	 * Even if the address matches none of our addresses, it might be
1380 	 * in the neighbor cache.
1381 	 */
1382 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1383 		LLE_RUNLOCK(lle);
1384 		rc = 1;
1385 	}
1386 	return (rc);
1387 }
1388 
1389 static __noinline void
1390 nd6_free_children(struct llentry *lle)
1391 {
1392 	struct llentry *child_lle;
1393 
1394 	NET_EPOCH_ASSERT();
1395 	LLE_WLOCK_ASSERT(lle);
1396 
1397 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1398 		LLE_WLOCK(child_lle);
1399 		lltable_unlink_child_entry(child_lle);
1400 		llentry_free(child_lle);
1401 	}
1402 }
1403 
1404 /*
1405  * Tries to update @lle address/prepend data with new @lladdr.
1406  *
1407  * Returns true on success.
1408  * In any case, @lle is returned wlocked.
1409  */
1410 static __noinline bool
1411 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1412 {
1413 	u_char buf[LLE_MAX_LINKHDR];
1414 	int fam, off;
1415 	size_t sz;
1416 
1417 	sz = sizeof(buf);
1418 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1419 		return (false);
1420 
1421 	/* Update data */
1422 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1423 
1424 	struct llentry *child_lle;
1425 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1426 		LLE_WLOCK(child_lle);
1427 		fam = child_lle->r_family;
1428 		sz = sizeof(buf);
1429 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1430 			/* success */
1431 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1432 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1433 		}
1434 		LLE_WUNLOCK(child_lle);
1435 	}
1436 
1437 	return (true);
1438 }
1439 
1440 bool
1441 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1442 {
1443 	NET_EPOCH_ASSERT();
1444 	LLE_WLOCK_ASSERT(lle);
1445 
1446 	if (!lltable_acquire_wlock(ifp, lle))
1447 		return (false);
1448 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1449 	IF_AFDATA_WUNLOCK(ifp);
1450 
1451 	return (ret);
1452 }
1453 
1454 /*
1455  * Free an nd6 llinfo entry.
1456  * Since the function would cause significant changes in the kernel, DO NOT
1457  * make it global, unless you have a strong reason for the change, and are sure
1458  * that the change is safe.
1459  *
1460  * Set noinline to be dtrace-friendly
1461  */
1462 static __noinline void
1463 nd6_free(struct llentry **lnp, int gc)
1464 {
1465 	struct ifnet *ifp;
1466 	struct llentry *ln;
1467 	struct nd_defrouter *dr;
1468 
1469 	ln = *lnp;
1470 	*lnp = NULL;
1471 
1472 	LLE_WLOCK_ASSERT(ln);
1473 	ND6_RLOCK_ASSERT();
1474 
1475 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1476 
1477 	ifp = lltable_get_ifp(ln->lle_tbl);
1478 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1479 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1480 	else
1481 		dr = NULL;
1482 	ND6_RUNLOCK();
1483 
1484 	if ((ln->la_flags & LLE_DELETED) == 0)
1485 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1486 
1487 	/*
1488 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1489 	 * even though it is not harmful, it was not really necessary.
1490 	 */
1491 
1492 	/* cancel timer */
1493 	nd6_llinfo_settimer_locked(ln, -1);
1494 
1495 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1496 		if (dr != NULL && dr->expire &&
1497 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1498 			/*
1499 			 * If the reason for the deletion is just garbage
1500 			 * collection, and the neighbor is an active default
1501 			 * router, do not delete it.  Instead, reset the GC
1502 			 * timer using the router's lifetime.
1503 			 * Simply deleting the entry would affect default
1504 			 * router selection, which is not necessarily a good
1505 			 * thing, especially when we're using router preference
1506 			 * values.
1507 			 * XXX: the check for ln_state would be redundant,
1508 			 *      but we intentionally keep it just in case.
1509 			 */
1510 			if (dr->expire > time_uptime)
1511 				nd6_llinfo_settimer_locked(ln,
1512 				    (dr->expire - time_uptime) * hz);
1513 			else
1514 				nd6_llinfo_settimer_locked(ln,
1515 				    (long)V_nd6_gctimer * hz);
1516 
1517 			LLE_REMREF(ln);
1518 			LLE_WUNLOCK(ln);
1519 			defrouter_rele(dr);
1520 			return;
1521 		}
1522 
1523 		if (dr) {
1524 			/*
1525 			 * Unreachablity of a router might affect the default
1526 			 * router selection and on-link detection of advertised
1527 			 * prefixes.
1528 			 */
1529 
1530 			/*
1531 			 * Temporarily fake the state to choose a new default
1532 			 * router and to perform on-link determination of
1533 			 * prefixes correctly.
1534 			 * Below the state will be set correctly,
1535 			 * or the entry itself will be deleted.
1536 			 */
1537 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1538 		}
1539 
1540 		if (ln->ln_router || dr) {
1541 			/*
1542 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1543 			 * rnh and for the calls to pfxlist_onlink_check() and
1544 			 * defrouter_select_fib() in the block further down for calls
1545 			 * into nd6_lookup().  We still hold a ref.
1546 			 */
1547 			LLE_WUNLOCK(ln);
1548 
1549 			/*
1550 			 * rt6_flush must be called whether or not the neighbor
1551 			 * is in the Default Router List.
1552 			 * See a corresponding comment in nd6_na_input().
1553 			 */
1554 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1555 		}
1556 
1557 		if (dr) {
1558 			/*
1559 			 * Since defrouter_select_fib() does not affect the
1560 			 * on-link determination and MIP6 needs the check
1561 			 * before the default router selection, we perform
1562 			 * the check now.
1563 			 */
1564 			pfxlist_onlink_check();
1565 
1566 			/*
1567 			 * Refresh default router list.
1568 			 */
1569 			defrouter_select_fib(dr->ifp->if_fib);
1570 		}
1571 
1572 		/*
1573 		 * If this entry was added by an on-link redirect, remove the
1574 		 * corresponding host route.
1575 		 */
1576 		if (ln->la_flags & LLE_REDIRECT)
1577 			nd6_free_redirect(ln);
1578 
1579 		if (ln->ln_router || dr)
1580 			LLE_WLOCK(ln);
1581 	}
1582 
1583 	/*
1584 	 * Save to unlock. We still hold an extra reference and will not
1585 	 * free(9) in llentry_free() if someone else holds one as well.
1586 	 */
1587 	LLE_WUNLOCK(ln);
1588 	IF_AFDATA_LOCK(ifp);
1589 	LLE_WLOCK(ln);
1590 	/* Guard against race with other llentry_free(). */
1591 	if (ln->la_flags & LLE_LINKED) {
1592 		/* Remove callout reference */
1593 		LLE_REMREF(ln);
1594 		lltable_unlink_entry(ln->lle_tbl, ln);
1595 	}
1596 	IF_AFDATA_UNLOCK(ifp);
1597 
1598 	nd6_free_children(ln);
1599 
1600 	llentry_free(ln);
1601 	if (dr != NULL)
1602 		defrouter_rele(dr);
1603 }
1604 
1605 static int
1606 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1607 {
1608 
1609 	if (nh->nh_flags & NHF_REDIRECT)
1610 		return (1);
1611 
1612 	return (0);
1613 }
1614 
1615 /*
1616  * Remove the rtentry for the given llentry,
1617  * both of which were installed by a redirect.
1618  */
1619 static void
1620 nd6_free_redirect(const struct llentry *ln)
1621 {
1622 	int fibnum;
1623 	struct sockaddr_in6 sin6;
1624 	struct rt_addrinfo info;
1625 	struct rib_cmd_info rc;
1626 	struct epoch_tracker et;
1627 
1628 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1629 	memset(&info, 0, sizeof(info));
1630 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1631 	info.rti_filter = nd6_isdynrte;
1632 
1633 	NET_EPOCH_ENTER(et);
1634 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1635 		rib_action(fibnum, RTM_DELETE, &info, &rc);
1636 	NET_EPOCH_EXIT(et);
1637 }
1638 
1639 /*
1640  * Updates status of the default router route.
1641  */
1642 static void
1643 check_release_defrouter(struct rib_cmd_info *rc, void *_cbdata)
1644 {
1645 	struct nd_defrouter *dr;
1646 	struct nhop_object *nh;
1647 
1648 	nh = rc->rc_nh_old;
1649 
1650 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1651 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1652 		if (dr != NULL) {
1653 			dr->installed = 0;
1654 			defrouter_rele(dr);
1655 		}
1656 	}
1657 }
1658 
1659 void
1660 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1661 {
1662 
1663 #ifdef ROUTE_MPATH
1664 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1665 #else
1666 	check_release_defrouter(rc, NULL);
1667 #endif
1668 }
1669 
1670 int
1671 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1672 {
1673 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1674 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1675 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1676 	struct epoch_tracker et;
1677 	int error = 0;
1678 
1679 	if (ifp->if_afdata[AF_INET6] == NULL)
1680 		return (EPFNOSUPPORT);
1681 	switch (cmd) {
1682 	case OSIOCGIFINFO_IN6:
1683 #define ND	ndi->ndi
1684 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1685 		bzero(&ND, sizeof(ND));
1686 		ND.linkmtu = IN6_LINKMTU(ifp);
1687 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1688 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1689 		ND.reachable = ND_IFINFO(ifp)->reachable;
1690 		ND.retrans = ND_IFINFO(ifp)->retrans;
1691 		ND.flags = ND_IFINFO(ifp)->flags;
1692 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1693 		ND.chlim = ND_IFINFO(ifp)->chlim;
1694 		break;
1695 	case SIOCGIFINFO_IN6:
1696 		ND = *ND_IFINFO(ifp);
1697 		break;
1698 	case SIOCSIFINFO_IN6:
1699 		/*
1700 		 * used to change host variables from userland.
1701 		 * intended for a use on router to reflect RA configurations.
1702 		 */
1703 		/* 0 means 'unspecified' */
1704 		if (ND.linkmtu != 0) {
1705 			if (ND.linkmtu < IPV6_MMTU ||
1706 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1707 				error = EINVAL;
1708 				break;
1709 			}
1710 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1711 		}
1712 
1713 		if (ND.basereachable != 0) {
1714 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1715 
1716 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1717 			if (ND.basereachable != obasereachable)
1718 				ND_IFINFO(ifp)->reachable =
1719 				    ND_COMPUTE_RTIME(ND.basereachable);
1720 		}
1721 		if (ND.retrans != 0)
1722 			ND_IFINFO(ifp)->retrans = ND.retrans;
1723 		if (ND.chlim != 0)
1724 			ND_IFINFO(ifp)->chlim = ND.chlim;
1725 		/* FALLTHROUGH */
1726 	case SIOCSIFINFO_FLAGS:
1727 	{
1728 		struct ifaddr *ifa;
1729 		struct in6_ifaddr *ia;
1730 
1731 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1732 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1733 			/* ifdisabled 1->0 transision */
1734 
1735 			/*
1736 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1737 			 * has an link-local address with IN6_IFF_DUPLICATED,
1738 			 * do not clear ND6_IFF_IFDISABLED.
1739 			 * See RFC 4862, Section 5.4.5.
1740 			 */
1741 			NET_EPOCH_ENTER(et);
1742 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1743 				if (ifa->ifa_addr->sa_family != AF_INET6)
1744 					continue;
1745 				ia = (struct in6_ifaddr *)ifa;
1746 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1747 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1748 					break;
1749 			}
1750 			NET_EPOCH_EXIT(et);
1751 
1752 			if (ifa != NULL) {
1753 				/* LLA is duplicated. */
1754 				ND.flags |= ND6_IFF_IFDISABLED;
1755 				log(LOG_ERR, "Cannot enable an interface"
1756 				    " with a link-local address marked"
1757 				    " duplicate.\n");
1758 			} else {
1759 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1760 				if (ifp->if_flags & IFF_UP)
1761 					in6_if_up(ifp);
1762 			}
1763 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1764 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1765 			/* ifdisabled 0->1 transision */
1766 			/* Mark all IPv6 address as tentative. */
1767 
1768 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1769 			if (V_ip6_dad_count > 0 &&
1770 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1771 				NET_EPOCH_ENTER(et);
1772 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1773 				    ifa_link) {
1774 					if (ifa->ifa_addr->sa_family !=
1775 					    AF_INET6)
1776 						continue;
1777 					ia = (struct in6_ifaddr *)ifa;
1778 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1779 				}
1780 				NET_EPOCH_EXIT(et);
1781 			}
1782 		}
1783 
1784 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1785 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1786 				/* auto_linklocal 0->1 transision */
1787 
1788 				/* If no link-local address on ifp, configure */
1789 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1790 				in6_ifattach(ifp, NULL);
1791 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1792 			    ifp->if_flags & IFF_UP) {
1793 				/*
1794 				 * When the IF already has
1795 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1796 				 * address is assigned, and IFF_UP, try to
1797 				 * assign one.
1798 				 */
1799 				NET_EPOCH_ENTER(et);
1800 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1801 				    ifa_link) {
1802 					if (ifa->ifa_addr->sa_family !=
1803 					    AF_INET6)
1804 						continue;
1805 					ia = (struct in6_ifaddr *)ifa;
1806 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1807 						break;
1808 				}
1809 				NET_EPOCH_EXIT(et);
1810 				if (ifa != NULL)
1811 					/* No LLA is configured. */
1812 					in6_ifattach(ifp, NULL);
1813 			}
1814 		}
1815 		ND_IFINFO(ifp)->flags = ND.flags;
1816 		break;
1817 	}
1818 #undef ND
1819 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1820 		/* sync kernel routing table with the default router list */
1821 		defrouter_reset();
1822 		defrouter_select_fib(RT_ALL_FIBS);
1823 		break;
1824 	case SIOCSPFXFLUSH_IN6:
1825 	{
1826 		/* flush all the prefix advertised by routers */
1827 		struct in6_ifaddr *ia, *ia_next;
1828 		struct nd_prefix *pr, *next;
1829 		struct nd_prhead prl;
1830 
1831 		LIST_INIT(&prl);
1832 
1833 		ND6_WLOCK();
1834 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1835 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1836 				continue; /* XXX */
1837 			nd6_prefix_unlink(pr, &prl);
1838 		}
1839 		ND6_WUNLOCK();
1840 
1841 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1842 			LIST_REMOVE(pr, ndpr_entry);
1843 			/* XXXRW: in6_ifaddrhead locking. */
1844 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1845 			    ia_next) {
1846 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1847 					continue;
1848 
1849 				if (ia->ia6_ndpr == pr)
1850 					in6_purgeaddr(&ia->ia_ifa);
1851 			}
1852 			nd6_prefix_del(pr);
1853 		}
1854 		break;
1855 	}
1856 	case SIOCSRTRFLUSH_IN6:
1857 	{
1858 		/* flush all the default routers */
1859 
1860 		defrouter_reset();
1861 		nd6_defrouter_flush_all();
1862 		defrouter_select_fib(RT_ALL_FIBS);
1863 		break;
1864 	}
1865 	case SIOCGNBRINFO_IN6:
1866 	{
1867 		struct llentry *ln;
1868 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1869 
1870 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1871 			return (error);
1872 
1873 		NET_EPOCH_ENTER(et);
1874 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1875 		NET_EPOCH_EXIT(et);
1876 
1877 		if (ln == NULL) {
1878 			error = EINVAL;
1879 			break;
1880 		}
1881 		nbi->state = ln->ln_state;
1882 		nbi->asked = ln->la_asked;
1883 		nbi->isrouter = ln->ln_router;
1884 		if (ln->la_expire == 0)
1885 			nbi->expire = 0;
1886 		else
1887 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1888 			    (time_second - time_uptime);
1889 		LLE_RUNLOCK(ln);
1890 		break;
1891 	}
1892 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1893 		ndif->ifindex = V_nd6_defifindex;
1894 		break;
1895 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1896 		return (nd6_setdefaultiface(ndif->ifindex));
1897 	}
1898 	return (error);
1899 }
1900 
1901 /*
1902  * Calculates new isRouter value based on provided parameters and
1903  * returns it.
1904  */
1905 static int
1906 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1907     int ln_router)
1908 {
1909 
1910 	/*
1911 	 * ICMP6 type dependent behavior.
1912 	 *
1913 	 * NS: clear IsRouter if new entry
1914 	 * RS: clear IsRouter
1915 	 * RA: set IsRouter if there's lladdr
1916 	 * redir: clear IsRouter if new entry
1917 	 *
1918 	 * RA case, (1):
1919 	 * The spec says that we must set IsRouter in the following cases:
1920 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1921 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1922 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1923 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1924 	 * neighbor cache, this is similar to (6).
1925 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1926 	 *
1927 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1928 	 *							D R
1929 	 *	0	n	n	(1)	c   ?     s
1930 	 *	0	y	n	(2)	c   s     s
1931 	 *	0	n	y	(3)	c   s     s
1932 	 *	0	y	y	(4)	c   s     s
1933 	 *	0	y	y	(5)	c   s     s
1934 	 *	1	--	n	(6) c	c	c s
1935 	 *	1	--	y	(7) c	c   s	c s
1936 	 *
1937 	 *					(c=clear s=set)
1938 	 */
1939 	switch (type & 0xff) {
1940 	case ND_NEIGHBOR_SOLICIT:
1941 		/*
1942 		 * New entry must have is_router flag cleared.
1943 		 */
1944 		if (is_new)					/* (6-7) */
1945 			ln_router = 0;
1946 		break;
1947 	case ND_REDIRECT:
1948 		/*
1949 		 * If the icmp is a redirect to a better router, always set the
1950 		 * is_router flag.  Otherwise, if the entry is newly created,
1951 		 * clear the flag.  [RFC 2461, sec 8.3]
1952 		 */
1953 		if (code == ND_REDIRECT_ROUTER)
1954 			ln_router = 1;
1955 		else {
1956 			if (is_new)				/* (6-7) */
1957 				ln_router = 0;
1958 		}
1959 		break;
1960 	case ND_ROUTER_SOLICIT:
1961 		/*
1962 		 * is_router flag must always be cleared.
1963 		 */
1964 		ln_router = 0;
1965 		break;
1966 	case ND_ROUTER_ADVERT:
1967 		/*
1968 		 * Mark an entry with lladdr as a router.
1969 		 */
1970 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1971 		    (is_new && new_addr)) {			/* (7) */
1972 			ln_router = 1;
1973 		}
1974 		break;
1975 	}
1976 
1977 	return (ln_router);
1978 }
1979 
1980 /*
1981  * Create neighbor cache entry and cache link-layer address,
1982  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1983  *
1984  * type - ICMP6 type
1985  * code - type dependent information
1986  *
1987  */
1988 void
1989 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1990     int lladdrlen, int type, int code)
1991 {
1992 	struct llentry *ln = NULL, *ln_tmp;
1993 	int is_newentry;
1994 	int do_update;
1995 	int olladdr;
1996 	int llchange;
1997 	int flags;
1998 	uint16_t router = 0;
1999 	struct mbuf *chain = NULL;
2000 	u_char linkhdr[LLE_MAX_LINKHDR];
2001 	size_t linkhdrsize;
2002 	int lladdr_off;
2003 
2004 	NET_EPOCH_ASSERT();
2005 	IF_AFDATA_UNLOCK_ASSERT(ifp);
2006 
2007 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
2008 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
2009 
2010 	/* nothing must be updated for unspecified address */
2011 	if (IN6_IS_ADDR_UNSPECIFIED(from))
2012 		return;
2013 
2014 	/*
2015 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
2016 	 * the caller.
2017 	 *
2018 	 * XXX If the link does not have link-layer adderss, what should
2019 	 * we do? (ifp->if_addrlen == 0)
2020 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2021 	 * description on it in NS section (RFC 2461 7.2.3).
2022 	 */
2023 	flags = lladdr ? LLE_EXCLUSIVE : 0;
2024 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
2025 	is_newentry = 0;
2026 	if (ln == NULL) {
2027 		flags |= LLE_EXCLUSIVE;
2028 		ln = nd6_alloc(from, 0, ifp);
2029 		if (ln == NULL)
2030 			return;
2031 
2032 		/*
2033 		 * Since we already know all the data for the new entry,
2034 		 * fill it before insertion.
2035 		 */
2036 		if (lladdr != NULL) {
2037 			linkhdrsize = sizeof(linkhdr);
2038 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2039 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2040 				return;
2041 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2042 			    lladdr_off);
2043 		}
2044 
2045 		IF_AFDATA_WLOCK(ifp);
2046 		LLE_WLOCK(ln);
2047 		/* Prefer any existing lle over newly-created one */
2048 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2049 		if (ln_tmp == NULL)
2050 			lltable_link_entry(LLTABLE6(ifp), ln);
2051 		IF_AFDATA_WUNLOCK(ifp);
2052 		if (ln_tmp == NULL) {
2053 			/* No existing lle, mark as new entry (6,7) */
2054 			is_newentry = 1;
2055 			if (lladdr != NULL) {	/* (7) */
2056 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2057 				EVENTHANDLER_INVOKE(lle_event, ln,
2058 				    LLENTRY_RESOLVED);
2059 			}
2060 		} else {
2061 			lltable_free_entry(LLTABLE6(ifp), ln);
2062 			ln = ln_tmp;
2063 			ln_tmp = NULL;
2064 		}
2065 	}
2066 	/* do nothing if static ndp is set */
2067 	if ((ln->la_flags & LLE_STATIC)) {
2068 		if (flags & LLE_EXCLUSIVE)
2069 			LLE_WUNLOCK(ln);
2070 		else
2071 			LLE_RUNLOCK(ln);
2072 		return;
2073 	}
2074 
2075 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2076 	if (olladdr && lladdr) {
2077 		llchange = bcmp(lladdr, ln->ll_addr,
2078 		    ifp->if_addrlen);
2079 	} else if (!olladdr && lladdr)
2080 		llchange = 1;
2081 	else
2082 		llchange = 0;
2083 
2084 	/*
2085 	 * newentry olladdr  lladdr  llchange	(*=record)
2086 	 *	0	n	n	--	(1)
2087 	 *	0	y	n	--	(2)
2088 	 *	0	n	y	y	(3) * STALE
2089 	 *	0	y	y	n	(4) *
2090 	 *	0	y	y	y	(5) * STALE
2091 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2092 	 *	1	--	y	--	(7) * STALE
2093 	 */
2094 
2095 	do_update = 0;
2096 	if (is_newentry == 0 && llchange != 0) {
2097 		do_update = 1;	/* (3,5) */
2098 
2099 		/*
2100 		 * Record source link-layer address
2101 		 * XXX is it dependent to ifp->if_type?
2102 		 */
2103 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2104 			/* Entry was deleted */
2105 			LLE_WUNLOCK(ln);
2106 			return;
2107 		}
2108 
2109 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2110 
2111 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2112 
2113 		if (ln->la_hold != NULL)
2114 			chain = nd6_grab_holdchain(ln);
2115 	}
2116 
2117 	/* Calculates new router status */
2118 	router = nd6_is_router(type, code, is_newentry, olladdr,
2119 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2120 
2121 	ln->ln_router = router;
2122 	/* Mark non-router redirects with special flag */
2123 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2124 		ln->la_flags |= LLE_REDIRECT;
2125 
2126 	if (flags & LLE_EXCLUSIVE)
2127 		LLE_WUNLOCK(ln);
2128 	else
2129 		LLE_RUNLOCK(ln);
2130 
2131 	if (chain != NULL)
2132 		nd6_flush_holdchain(ifp, ln, chain);
2133 	if (do_update)
2134 		nd6_flush_children_holdchain(ifp, ln);
2135 
2136 	/*
2137 	 * When the link-layer address of a router changes, select the
2138 	 * best router again.  In particular, when the neighbor entry is newly
2139 	 * created, it might affect the selection policy.
2140 	 * Question: can we restrict the first condition to the "is_newentry"
2141 	 * case?
2142 	 * XXX: when we hear an RA from a new router with the link-layer
2143 	 * address option, defrouter_select_fib() is called twice, since
2144 	 * defrtrlist_update called the function as well.  However, I believe
2145 	 * we can compromise the overhead, since it only happens the first
2146 	 * time.
2147 	 * XXX: although defrouter_select_fib() should not have a bad effect
2148 	 * for those are not autoconfigured hosts, we explicitly avoid such
2149 	 * cases for safety.
2150 	 */
2151 	if ((do_update || is_newentry) && router &&
2152 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2153 		/*
2154 		 * guaranteed recursion
2155 		 */
2156 		defrouter_select_fib(ifp->if_fib);
2157 	}
2158 }
2159 
2160 static void
2161 nd6_slowtimo(void *arg)
2162 {
2163 	struct epoch_tracker et;
2164 	CURVNET_SET((struct vnet *) arg);
2165 	struct nd_ifinfo *nd6if;
2166 	struct ifnet *ifp;
2167 
2168 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2169 	    nd6_slowtimo, curvnet);
2170 	NET_EPOCH_ENTER(et);
2171 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2172 		if (ifp->if_afdata[AF_INET6] == NULL)
2173 			continue;
2174 		nd6if = ND_IFINFO(ifp);
2175 		if (nd6if->basereachable && /* already initialized */
2176 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2177 			/*
2178 			 * Since reachable time rarely changes by router
2179 			 * advertisements, we SHOULD insure that a new random
2180 			 * value gets recomputed at least once every few hours.
2181 			 * (RFC 2461, 6.3.4)
2182 			 */
2183 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2184 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2185 		}
2186 	}
2187 	NET_EPOCH_EXIT(et);
2188 	CURVNET_RESTORE();
2189 }
2190 
2191 struct mbuf *
2192 nd6_grab_holdchain(struct llentry *ln)
2193 {
2194 	struct mbuf *chain;
2195 
2196 	LLE_WLOCK_ASSERT(ln);
2197 
2198 	chain = ln->la_hold;
2199 	ln->la_hold = NULL;
2200 
2201 	if (ln->ln_state == ND6_LLINFO_STALE) {
2202 		/*
2203 		 * The first time we send a packet to a
2204 		 * neighbor whose entry is STALE, we have
2205 		 * to change the state to DELAY and a sets
2206 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2207 		 * seconds to ensure do neighbor unreachability
2208 		 * detection on expiration.
2209 		 * (RFC 2461 7.3.3)
2210 		 */
2211 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2212 	}
2213 
2214 	return (chain);
2215 }
2216 
2217 int
2218 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2219     struct sockaddr_in6 *dst, struct route *ro)
2220 {
2221 	int error;
2222 	int ip6len;
2223 	struct ip6_hdr *ip6;
2224 	struct m_tag *mtag;
2225 
2226 #ifdef MAC
2227 	mac_netinet6_nd6_send(ifp, m);
2228 #endif
2229 
2230 	/*
2231 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2232 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2233 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2234 	 * to be diverted to user space.  When re-injected into the kernel,
2235 	 * send_output() will directly dispatch them to the outgoing interface.
2236 	 */
2237 	if (send_sendso_input_hook != NULL) {
2238 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2239 		if (mtag != NULL) {
2240 			ip6 = mtod(m, struct ip6_hdr *);
2241 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2242 			/* Use the SEND socket */
2243 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2244 			    ip6len);
2245 			/* -1 == no app on SEND socket */
2246 			if (error == 0 || error != -1)
2247 			    return (error);
2248 		}
2249 	}
2250 
2251 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2252 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2253 	    mtod(m, struct ip6_hdr *));
2254 
2255 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2256 		origifp = ifp;
2257 
2258 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2259 	return (error);
2260 }
2261 
2262 /*
2263  * Lookup link headerfor @sa_dst address. Stores found
2264  * data in @desten buffer. Copy of lle ln_flags can be also
2265  * saved in @pflags if @pflags is non-NULL.
2266  *
2267  * If destination LLE does not exists or lle state modification
2268  * is required, call "slow" version.
2269  *
2270  * Return values:
2271  * - 0 on success (address copied to buffer).
2272  * - EWOULDBLOCK (no local error, but address is still unresolved)
2273  * - other errors (alloc failure, etc)
2274  */
2275 int
2276 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2277     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2278     struct llentry **plle)
2279 {
2280 	struct llentry *ln = NULL;
2281 	const struct sockaddr_in6 *dst6;
2282 
2283 	NET_EPOCH_ASSERT();
2284 
2285 	if (pflags != NULL)
2286 		*pflags = 0;
2287 
2288 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2289 
2290 	/* discard the packet if IPv6 operation is disabled on the interface */
2291 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2292 		m_freem(m);
2293 		return (ENETDOWN); /* better error? */
2294 	}
2295 
2296 	if (m != NULL && m->m_flags & M_MCAST) {
2297 		switch (ifp->if_type) {
2298 		case IFT_ETHER:
2299 		case IFT_L2VLAN:
2300 		case IFT_BRIDGE:
2301 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2302 						 desten);
2303 			return (0);
2304 		default:
2305 			m_freem(m);
2306 			return (EAFNOSUPPORT);
2307 		}
2308 	}
2309 
2310 	int family = gw_flags >> 16;
2311 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2312 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2313 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2314 		/* Entry found, let's copy lle info */
2315 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2316 		if (pflags != NULL)
2317 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2318 		llentry_provide_feedback(ln);
2319 		if (plle) {
2320 			LLE_ADDREF(ln);
2321 			*plle = ln;
2322 			LLE_WUNLOCK(ln);
2323 		}
2324 		return (0);
2325 	} else if (plle && ln)
2326 		LLE_WUNLOCK(ln);
2327 
2328 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2329 }
2330 
2331 /*
2332  * Finds or creates a new llentry for @addr and @family.
2333  * Returns wlocked llentry or NULL.
2334  *
2335  *
2336  * Child LLEs.
2337  *
2338  * Do not have their own state machine (gets marked as static)
2339  *  settimer bails out for child LLEs just in case.
2340  *
2341  * Locking order: parent lle gets locked first, chen goes the child.
2342  */
2343 static __noinline struct llentry *
2344 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2345 {
2346 	struct llentry *child_lle = NULL;
2347 	struct llentry *lle, *lle_tmp;
2348 
2349 	lle = nd6_alloc(addr, 0, ifp);
2350 	if (lle != NULL && family != AF_INET6) {
2351 		child_lle = nd6_alloc(addr, 0, ifp);
2352 		if (child_lle == NULL) {
2353 			lltable_free_entry(LLTABLE6(ifp), lle);
2354 			return (NULL);
2355 		}
2356 		child_lle->r_family = family;
2357 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2358 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2359 	}
2360 
2361 	if (lle == NULL) {
2362 		char ip6buf[INET6_ADDRSTRLEN];
2363 		log(LOG_DEBUG,
2364 		    "nd6_get_llentry: can't allocate llinfo for %s "
2365 		    "(ln=%p)\n",
2366 		    ip6_sprintf(ip6buf, addr), lle);
2367 		return (NULL);
2368 	}
2369 
2370 	IF_AFDATA_WLOCK(ifp);
2371 	LLE_WLOCK(lle);
2372 	/* Prefer any existing entry over newly-created one */
2373 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2374 	if (lle_tmp == NULL)
2375 		lltable_link_entry(LLTABLE6(ifp), lle);
2376 	else {
2377 		lltable_free_entry(LLTABLE6(ifp), lle);
2378 		lle = lle_tmp;
2379 	}
2380 	if (child_lle != NULL) {
2381 		/* Check if child lle for the same family exists */
2382 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2383 		LLE_WLOCK(child_lle);
2384 		if (lle_tmp == NULL) {
2385 			/* Attach */
2386 			lltable_link_child_entry(lle, child_lle);
2387 		} else {
2388 			/* child lle already exists, free newly-created one */
2389 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2390 			child_lle = lle_tmp;
2391 		}
2392 		LLE_WUNLOCK(lle);
2393 		lle = child_lle;
2394 	}
2395 	IF_AFDATA_WUNLOCK(ifp);
2396 	return (lle);
2397 }
2398 
2399 /*
2400  * Do L2 address resolution for @sa_dst address. Stores found
2401  * address in @desten buffer. Copy of lle ln_flags can be also
2402  * saved in @pflags if @pflags is non-NULL.
2403  *
2404  * Heavy version.
2405  * Function assume that destination LLE does not exist,
2406  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2407  *
2408  * Set noinline to be dtrace-friendly
2409  */
2410 static __noinline int
2411 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2412     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2413     struct llentry **plle)
2414 {
2415 	struct llentry *lle = NULL;
2416 	struct in6_addr *psrc, src;
2417 	int send_ns, ll_len;
2418 	char *lladdr;
2419 
2420 	NET_EPOCH_ASSERT();
2421 
2422 	/*
2423 	 * Address resolution or Neighbor Unreachability Detection
2424 	 * for the next hop.
2425 	 * At this point, the destination of the packet must be a unicast
2426 	 * or an anycast address(i.e. not a multicast).
2427 	 */
2428 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2429 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2430 		/*
2431 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2432 		 * the condition below is not very efficient.  But we believe
2433 		 * it is tolerable, because this should be a rare case.
2434 		 */
2435 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2436 	}
2437 
2438 	if (lle == NULL) {
2439 		m_freem(m);
2440 		return (ENOBUFS);
2441 	}
2442 
2443 	LLE_WLOCK_ASSERT(lle);
2444 
2445 	/*
2446 	 * The first time we send a packet to a neighbor whose entry is
2447 	 * STALE, we have to change the state to DELAY and a sets a timer to
2448 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2449 	 * neighbor unreachability detection on expiration.
2450 	 * (RFC 2461 7.3.3)
2451 	 */
2452 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2453 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2454 
2455 	/*
2456 	 * If the neighbor cache entry has a state other than INCOMPLETE
2457 	 * (i.e. its link-layer address is already resolved), just
2458 	 * send the packet.
2459 	 */
2460 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2461 		if (flags & LLE_ADDRONLY) {
2462 			lladdr = lle->ll_addr;
2463 			ll_len = ifp->if_addrlen;
2464 		} else {
2465 			lladdr = lle->r_linkdata;
2466 			ll_len = lle->r_hdrlen;
2467 		}
2468 		bcopy(lladdr, desten, ll_len);
2469 		if (pflags != NULL)
2470 			*pflags = lle->la_flags;
2471 		if (plle) {
2472 			LLE_ADDREF(lle);
2473 			*plle = lle;
2474 		}
2475 		LLE_WUNLOCK(lle);
2476 		return (0);
2477 	}
2478 
2479 	/*
2480 	 * There is a neighbor cache entry, but no ethernet address
2481 	 * response yet.  Append this latest packet to the end of the
2482 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2483 	 * the oldest packet in the queue will be removed.
2484 	 */
2485 
2486 	if (lle->la_hold != NULL) {
2487 		struct mbuf *m_hold;
2488 		int i;
2489 
2490 		i = 0;
2491 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2492 			i++;
2493 			if (m_hold->m_nextpkt == NULL) {
2494 				m_hold->m_nextpkt = m;
2495 				break;
2496 			}
2497 		}
2498 		while (i >= V_nd6_maxqueuelen) {
2499 			m_hold = lle->la_hold;
2500 			lle->la_hold = lle->la_hold->m_nextpkt;
2501 			m_freem(m_hold);
2502 			i--;
2503 		}
2504 	} else {
2505 		lle->la_hold = m;
2506 	}
2507 
2508 	/*
2509 	 * If there has been no NS for the neighbor after entering the
2510 	 * INCOMPLETE state, send the first solicitation.
2511 	 * Note that for newly-created lle la_asked will be 0,
2512 	 * so we will transition from ND6_LLINFO_NOSTATE to
2513 	 * ND6_LLINFO_INCOMPLETE state here.
2514 	 */
2515 	psrc = NULL;
2516 	send_ns = 0;
2517 
2518 	/* If we have child lle, switch to the parent to send NS */
2519 	if (lle->la_flags & LLE_CHILD) {
2520 		struct llentry *lle_parent = lle->lle_parent;
2521 		LLE_WUNLOCK(lle);
2522 		lle = lle_parent;
2523 		LLE_WLOCK(lle);
2524 	}
2525 	if (lle->la_asked == 0) {
2526 		lle->la_asked++;
2527 		send_ns = 1;
2528 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2529 
2530 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2531 	}
2532 	LLE_WUNLOCK(lle);
2533 	if (send_ns != 0)
2534 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2535 
2536 	return (EWOULDBLOCK);
2537 }
2538 
2539 /*
2540  * Do L2 address resolution for @sa_dst address. Stores found
2541  * address in @desten buffer. Copy of lle ln_flags can be also
2542  * saved in @pflags if @pflags is non-NULL.
2543  *
2544  * Return values:
2545  * - 0 on success (address copied to buffer).
2546  * - EWOULDBLOCK (no local error, but address is still unresolved)
2547  * - other errors (alloc failure, etc)
2548  */
2549 int
2550 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2551     char *desten, uint32_t *pflags)
2552 {
2553 	int error;
2554 
2555 	flags |= LLE_ADDRONLY;
2556 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2557 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2558 	return (error);
2559 }
2560 
2561 int
2562 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2563 {
2564 	struct mbuf *m, *m_head;
2565 	struct sockaddr_in6 dst6;
2566 	int error = 0;
2567 
2568 	NET_EPOCH_ASSERT();
2569 
2570 	struct route_in6 ro = {
2571 		.ro_prepend = lle->r_linkdata,
2572 		.ro_plen = lle->r_hdrlen,
2573 	};
2574 
2575 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2576 	m_head = chain;
2577 
2578 	while (m_head) {
2579 		m = m_head;
2580 		m_head = m_head->m_nextpkt;
2581 		m->m_nextpkt = NULL;
2582 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2583 	}
2584 
2585 	/*
2586 	 * XXX
2587 	 * note that intermediate errors are blindly ignored
2588 	 */
2589 	return (error);
2590 }
2591 
2592 __noinline void
2593 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2594 {
2595 	struct llentry *child_lle;
2596 	struct mbuf *chain;
2597 
2598 	NET_EPOCH_ASSERT();
2599 
2600 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2601 		LLE_WLOCK(child_lle);
2602 		chain = nd6_grab_holdchain(child_lle);
2603 		LLE_WUNLOCK(child_lle);
2604 		nd6_flush_holdchain(ifp, child_lle, chain);
2605 	}
2606 }
2607 
2608 static int
2609 nd6_need_cache(struct ifnet *ifp)
2610 {
2611 	/*
2612 	 * XXX: we currently do not make neighbor cache on any interface
2613 	 * other than Ethernet and GIF.
2614 	 *
2615 	 * RFC2893 says:
2616 	 * - unidirectional tunnels needs no ND
2617 	 */
2618 	switch (ifp->if_type) {
2619 	case IFT_ETHER:
2620 	case IFT_IEEE1394:
2621 	case IFT_L2VLAN:
2622 	case IFT_INFINIBAND:
2623 	case IFT_BRIDGE:
2624 	case IFT_PROPVIRTUAL:
2625 		return (1);
2626 	default:
2627 		return (0);
2628 	}
2629 }
2630 
2631 /*
2632  * Add pernament ND6 link-layer record for given
2633  * interface address.
2634  *
2635  * Very similar to IPv4 arp_ifinit(), but:
2636  * 1) IPv6 DAD is performed in different place
2637  * 2) It is called by IPv6 protocol stack in contrast to
2638  * arp_ifinit() which is typically called in SIOCSIFADDR
2639  * driver ioctl handler.
2640  *
2641  */
2642 int
2643 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2644 {
2645 	struct ifnet *ifp;
2646 	struct llentry *ln, *ln_tmp;
2647 	struct sockaddr *dst;
2648 
2649 	ifp = ia->ia_ifa.ifa_ifp;
2650 	if (nd6_need_cache(ifp) == 0)
2651 		return (0);
2652 
2653 	dst = (struct sockaddr *)&ia->ia_addr;
2654 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2655 	if (ln == NULL)
2656 		return (ENOBUFS);
2657 
2658 	IF_AFDATA_WLOCK(ifp);
2659 	LLE_WLOCK(ln);
2660 	/* Unlink any entry if exists */
2661 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2662 	if (ln_tmp != NULL)
2663 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2664 	lltable_link_entry(LLTABLE6(ifp), ln);
2665 	IF_AFDATA_WUNLOCK(ifp);
2666 
2667 	if (ln_tmp != NULL)
2668 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2669 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2670 
2671 	LLE_WUNLOCK(ln);
2672 	if (ln_tmp != NULL)
2673 		llentry_free(ln_tmp);
2674 
2675 	return (0);
2676 }
2677 
2678 /*
2679  * Removes either all lle entries for given @ia, or lle
2680  * corresponding to @ia address.
2681  */
2682 void
2683 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2684 {
2685 	struct sockaddr_in6 mask, addr;
2686 	struct sockaddr *saddr, *smask;
2687 	struct ifnet *ifp;
2688 
2689 	ifp = ia->ia_ifa.ifa_ifp;
2690 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2691 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2692 	saddr = (struct sockaddr *)&addr;
2693 	smask = (struct sockaddr *)&mask;
2694 
2695 	if (all != 0)
2696 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2697 	else
2698 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2699 }
2700 
2701 static void
2702 clear_llinfo_pqueue(struct llentry *ln)
2703 {
2704 	struct mbuf *m_hold, *m_hold_next;
2705 
2706 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2707 		m_hold_next = m_hold->m_nextpkt;
2708 		m_freem(m_hold);
2709 	}
2710 
2711 	ln->la_hold = NULL;
2712 }
2713 
2714 static int
2715 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2716 {
2717 	struct in6_prefix p;
2718 	struct sockaddr_in6 s6;
2719 	struct nd_prefix *pr;
2720 	struct nd_pfxrouter *pfr;
2721 	time_t maxexpire;
2722 	int error;
2723 	char ip6buf[INET6_ADDRSTRLEN];
2724 
2725 	if (req->newptr)
2726 		return (EPERM);
2727 
2728 	error = sysctl_wire_old_buffer(req, 0);
2729 	if (error != 0)
2730 		return (error);
2731 
2732 	bzero(&p, sizeof(p));
2733 	p.origin = PR_ORIG_RA;
2734 	bzero(&s6, sizeof(s6));
2735 	s6.sin6_family = AF_INET6;
2736 	s6.sin6_len = sizeof(s6);
2737 
2738 	ND6_RLOCK();
2739 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2740 		p.prefix = pr->ndpr_prefix;
2741 		if (sa6_recoverscope(&p.prefix)) {
2742 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2743 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2744 			/* XXX: press on... */
2745 		}
2746 		p.raflags = pr->ndpr_raf;
2747 		p.prefixlen = pr->ndpr_plen;
2748 		p.vltime = pr->ndpr_vltime;
2749 		p.pltime = pr->ndpr_pltime;
2750 		p.if_index = pr->ndpr_ifp->if_index;
2751 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2752 			p.expire = 0;
2753 		else {
2754 			/* XXX: we assume time_t is signed. */
2755 			maxexpire = (-1) &
2756 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2757 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2758 				p.expire = pr->ndpr_lastupdate +
2759 				    pr->ndpr_vltime +
2760 				    (time_second - time_uptime);
2761 			else
2762 				p.expire = maxexpire;
2763 		}
2764 		p.refcnt = pr->ndpr_addrcnt;
2765 		p.flags = pr->ndpr_stateflags;
2766 		p.advrtrs = 0;
2767 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2768 			p.advrtrs++;
2769 		error = SYSCTL_OUT(req, &p, sizeof(p));
2770 		if (error != 0)
2771 			break;
2772 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2773 			s6.sin6_addr = pfr->router->rtaddr;
2774 			if (sa6_recoverscope(&s6))
2775 				log(LOG_ERR,
2776 				    "scope error in prefix list (%s)\n",
2777 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2778 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2779 			if (error != 0)
2780 				goto out;
2781 		}
2782 	}
2783 out:
2784 	ND6_RUNLOCK();
2785 	return (error);
2786 }
2787 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2788 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2789 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2790 	"NDP prefix list");
2791 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2792 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2793 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2794 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2795