xref: /freebsd/sys/netinet6/nd6.c (revision 75a813087a5bba4504efe6695411715f6352a8a8)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/eventhandler.h>
43 #include <sys/callout.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/mutex.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/time.h>
51 #include <sys/kernel.h>
52 #include <sys/protosw.h>
53 #include <sys/errno.h>
54 #include <sys/syslog.h>
55 #include <sys/rwlock.h>
56 #include <sys/queue.h>
57 #include <sys/sdt.h>
58 #include <sys/sysctl.h>
59 
60 #include <net/if.h>
61 #include <net/if_var.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/route_var.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_kdtrace.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #include <netinet6/in6_var.h>
74 #include <netinet/ip6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/scope6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet6/in6_ifattach.h>
79 #include <netinet/icmp6.h>
80 #include <netinet6/send.h>
81 
82 #include <sys/limits.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
87 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
88 
89 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
90 
91 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
92 
93 /* timer values */
94 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
95 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
96 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
97 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
98 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
99 					 * local traffic */
100 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
101 					 * collection timer */
102 
103 /* preventing too many loops in ND option parsing */
104 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
105 
106 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
107 					 * layer hints */
108 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
109 					 * ND entries */
110 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
111 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
112 
113 #ifdef ND6_DEBUG
114 VNET_DEFINE(int, nd6_debug) = 1;
115 #else
116 VNET_DEFINE(int, nd6_debug) = 0;
117 #endif
118 
119 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
120 
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
122 VNET_DEFINE(struct rwlock, nd6_lock);
123 VNET_DEFINE(uint64_t, nd6_list_genid);
124 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
125 
126 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
127 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
128 
129 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
130 
131 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
132 	struct ifnet *);
133 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
134 static void nd6_slowtimo(void *);
135 static int regen_tmpaddr(struct in6_ifaddr *);
136 static void nd6_free(struct llentry **, int);
137 static void nd6_free_redirect(const struct llentry *);
138 static void nd6_llinfo_timer(void *);
139 static void nd6_llinfo_settimer_locked(struct llentry *, long);
140 static void clear_llinfo_pqueue(struct llentry *);
141 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
142 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
143     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
144 static int nd6_need_cache(struct ifnet *);
145 
146 
147 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
148 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
149 
150 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
151 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
152 
153 SYSCTL_DECL(_net_inet6_icmp6);
154 
155 static void
156 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
157 {
158 	struct rt_addrinfo rtinfo;
159 	struct sockaddr_in6 dst;
160 	struct sockaddr_dl gw;
161 	struct ifnet *ifp;
162 	int type;
163 	int fibnum;
164 
165 	LLE_WLOCK_ASSERT(lle);
166 
167 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
168 		return;
169 
170 	switch (evt) {
171 	case LLENTRY_RESOLVED:
172 		type = RTM_ADD;
173 		KASSERT(lle->la_flags & LLE_VALID,
174 		    ("%s: %p resolved but not valid?", __func__, lle));
175 		break;
176 	case LLENTRY_EXPIRED:
177 		type = RTM_DELETE;
178 		break;
179 	default:
180 		return;
181 	}
182 
183 	ifp = lltable_get_ifp(lle->lle_tbl);
184 
185 	bzero(&dst, sizeof(dst));
186 	bzero(&gw, sizeof(gw));
187 	bzero(&rtinfo, sizeof(rtinfo));
188 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
189 	dst.sin6_scope_id = in6_getscopezone(ifp,
190 	    in6_addrscope(&dst.sin6_addr));
191 	gw.sdl_len = sizeof(struct sockaddr_dl);
192 	gw.sdl_family = AF_LINK;
193 	gw.sdl_alen = ifp->if_addrlen;
194 	gw.sdl_index = ifp->if_index;
195 	gw.sdl_type = ifp->if_type;
196 	if (evt == LLENTRY_RESOLVED)
197 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
198 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
199 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
200 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
201 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
202 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
203 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
204 }
205 
206 /*
207  * A handler for interface link layer address change event.
208  */
209 static void
210 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
211 {
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, 0);
228 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
229 	    nd6_slowtimo, curvnet);
230 
231 	callout_init(&V_nd6_timer_ch, 0);
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()
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 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
278 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
279 	 * default regardless of the V_ip6_auto_linklocal configuration to
280 	 * give a reasonable default behavior.
281 	 */
282 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
283 	    (ifp->if_flags & IFF_LOOPBACK))
284 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
285 	/*
286 	 * A loopback interface does not need to accept RTADV.
287 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
288 	 * default regardless of the V_ip6_accept_rtadv configuration to
289 	 * prevent the interface from accepting RA messages arrived
290 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
291 	 */
292 	if (V_ip6_accept_rtadv &&
293 	    !(ifp->if_flags & IFF_LOOPBACK) &&
294 	    (ifp->if_type != IFT_BRIDGE))
295 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
296 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
297 		nd->flags |= ND6_IFF_NO_RADR;
298 
299 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
300 	nd6_setmtu0(ifp, nd);
301 
302 	return nd;
303 }
304 
305 void
306 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
307 {
308 	struct epoch_tracker et;
309 	struct ifaddr *ifa, *next;
310 
311 	NET_EPOCH_ENTER(et);
312 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
313 		if (ifa->ifa_addr->sa_family != AF_INET6)
314 			continue;
315 
316 		/* stop DAD processing */
317 		nd6_dad_stop(ifa);
318 	}
319 	NET_EPOCH_EXIT(et);
320 
321 	free(nd, M_IP6NDP);
322 }
323 
324 /*
325  * Reset ND level link MTU. This function is called when the physical MTU
326  * changes, which means we might have to adjust the ND level MTU.
327  */
328 void
329 nd6_setmtu(struct ifnet *ifp)
330 {
331 	if (ifp->if_afdata[AF_INET6] == NULL)
332 		return;
333 
334 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
335 }
336 
337 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
338 void
339 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
340 {
341 	u_int32_t omaxmtu;
342 
343 	omaxmtu = ndi->maxmtu;
344 	ndi->maxmtu = ifp->if_mtu;
345 
346 	/*
347 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
348 	 * undesirable situation.  We thus notify the operator of the change
349 	 * explicitly.  The check for omaxmtu is necessary to restrict the
350 	 * log to the case of changing the MTU, not initializing it.
351 	 */
352 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
353 		log(LOG_NOTICE, "nd6_setmtu0: "
354 		    "new link MTU on %s (%lu) is too small for IPv6\n",
355 		    if_name(ifp), (unsigned long)ndi->maxmtu);
356 	}
357 
358 	if (ndi->maxmtu > V_in6_maxmtu)
359 		in6_setmaxmtu(); /* check all interfaces just in case */
360 
361 }
362 
363 void
364 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
365 {
366 
367 	bzero(ndopts, sizeof(*ndopts));
368 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
369 	ndopts->nd_opts_last
370 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
371 
372 	if (icmp6len == 0) {
373 		ndopts->nd_opts_done = 1;
374 		ndopts->nd_opts_search = NULL;
375 	}
376 }
377 
378 /*
379  * Take one ND option.
380  */
381 struct nd_opt_hdr *
382 nd6_option(union nd_opts *ndopts)
383 {
384 	struct nd_opt_hdr *nd_opt;
385 	int olen;
386 
387 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
388 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
389 	    __func__));
390 	if (ndopts->nd_opts_search == NULL)
391 		return NULL;
392 	if (ndopts->nd_opts_done)
393 		return NULL;
394 
395 	nd_opt = ndopts->nd_opts_search;
396 
397 	/* make sure nd_opt_len is inside the buffer */
398 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
399 		bzero(ndopts, sizeof(*ndopts));
400 		return NULL;
401 	}
402 
403 	olen = nd_opt->nd_opt_len << 3;
404 	if (olen == 0) {
405 		/*
406 		 * Message validation requires that all included
407 		 * options have a length that is greater than zero.
408 		 */
409 		bzero(ndopts, sizeof(*ndopts));
410 		return NULL;
411 	}
412 
413 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
414 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
415 		/* option overruns the end of buffer, invalid */
416 		bzero(ndopts, sizeof(*ndopts));
417 		return NULL;
418 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
419 		/* reached the end of options chain */
420 		ndopts->nd_opts_done = 1;
421 		ndopts->nd_opts_search = NULL;
422 	}
423 	return nd_opt;
424 }
425 
426 /*
427  * Parse multiple ND options.
428  * This function is much easier to use, for ND routines that do not need
429  * multiple options of the same type.
430  */
431 int
432 nd6_options(union nd_opts *ndopts)
433 {
434 	struct nd_opt_hdr *nd_opt;
435 	int i = 0;
436 
437 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
438 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
439 	    __func__));
440 	if (ndopts->nd_opts_search == NULL)
441 		return 0;
442 
443 	while (1) {
444 		nd_opt = nd6_option(ndopts);
445 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
446 			/*
447 			 * Message validation requires that all included
448 			 * options have a length that is greater than zero.
449 			 */
450 			ICMP6STAT_INC(icp6s_nd_badopt);
451 			bzero(ndopts, sizeof(*ndopts));
452 			return -1;
453 		}
454 
455 		if (nd_opt == NULL)
456 			goto skip1;
457 
458 		switch (nd_opt->nd_opt_type) {
459 		case ND_OPT_SOURCE_LINKADDR:
460 		case ND_OPT_TARGET_LINKADDR:
461 		case ND_OPT_MTU:
462 		case ND_OPT_REDIRECTED_HEADER:
463 		case ND_OPT_NONCE:
464 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
465 				nd6log((LOG_INFO,
466 				    "duplicated ND6 option found (type=%d)\n",
467 				    nd_opt->nd_opt_type));
468 				/* XXX bark? */
469 			} else {
470 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
471 					= nd_opt;
472 			}
473 			break;
474 		case ND_OPT_PREFIX_INFORMATION:
475 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
476 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
477 					= nd_opt;
478 			}
479 			ndopts->nd_opts_pi_end =
480 				(struct nd_opt_prefix_info *)nd_opt;
481 			break;
482 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
483 		case ND_OPT_RDNSS:	/* RFC 6106 */
484 		case ND_OPT_DNSSL:	/* RFC 6106 */
485 			/*
486 			 * Silently ignore options we know and do not care about
487 			 * in the kernel.
488 			 */
489 			break;
490 		default:
491 			/*
492 			 * Unknown options must be silently ignored,
493 			 * to accommodate future extension to the protocol.
494 			 */
495 			nd6log((LOG_DEBUG,
496 			    "nd6_options: unsupported option %d - "
497 			    "option ignored\n", nd_opt->nd_opt_type));
498 		}
499 
500 skip1:
501 		i++;
502 		if (i > V_nd6_maxndopt) {
503 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
504 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
505 			break;
506 		}
507 
508 		if (ndopts->nd_opts_done)
509 			break;
510 	}
511 
512 	return 0;
513 }
514 
515 /*
516  * ND6 timer routine to handle ND6 entries
517  */
518 static void
519 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
520 {
521 	int canceled;
522 
523 	LLE_WLOCK_ASSERT(ln);
524 
525 	if (tick < 0) {
526 		ln->la_expire = 0;
527 		ln->ln_ntick = 0;
528 		canceled = callout_stop(&ln->lle_timer);
529 	} else {
530 		ln->la_expire = time_uptime + tick / hz;
531 		LLE_ADDREF(ln);
532 		if (tick > INT_MAX) {
533 			ln->ln_ntick = tick - INT_MAX;
534 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
535 			    nd6_llinfo_timer, ln);
536 		} else {
537 			ln->ln_ntick = 0;
538 			canceled = callout_reset(&ln->lle_timer, tick,
539 			    nd6_llinfo_timer, ln);
540 		}
541 	}
542 	if (canceled > 0)
543 		LLE_REMREF(ln);
544 }
545 
546 /*
547  * Gets source address of the first packet in hold queue
548  * and stores it in @src.
549  * Returns pointer to @src (if hold queue is not empty) or NULL.
550  *
551  * Set noinline to be dtrace-friendly
552  */
553 static __noinline struct in6_addr *
554 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
555 {
556 	struct ip6_hdr hdr;
557 	struct mbuf *m;
558 
559 	if (ln->la_hold == NULL)
560 		return (NULL);
561 
562 	/*
563 	 * assume every packet in la_hold has the same IP header
564 	 */
565 	m = ln->la_hold;
566 	if (sizeof(hdr) > m->m_len)
567 		return (NULL);
568 
569 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
570 	*src = hdr.ip6_src;
571 
572 	return (src);
573 }
574 
575 /*
576  * Checks if we need to switch from STALE state.
577  *
578  * RFC 4861 requires switching from STALE to DELAY state
579  * on first packet matching entry, waiting V_nd6_delay and
580  * transition to PROBE state (if upper layer confirmation was
581  * not received).
582  *
583  * This code performs a bit differently:
584  * On packet hit we don't change state (but desired state
585  * can be guessed by control plane). However, after V_nd6_delay
586  * seconds code will transition to PROBE state (so DELAY state
587  * is kinda skipped in most situations).
588  *
589  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
590  * we perform the following upon entering STALE state:
591  *
592  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
593  * if packet was transmitted at the start of given interval, we
594  * would be able to switch to PROBE state in V_nd6_delay seconds
595  * as user expects.
596  *
597  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
598  * lle in STALE state (remaining timer value stored in lle_remtime).
599  *
600  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
601  * seconds ago.
602  *
603  * Returns non-zero value if the entry is still STALE (storing
604  * the next timer interval in @pdelay).
605  *
606  * Returns zero value if original timer expired or we need to switch to
607  * PROBE (store that in @do_switch variable).
608  */
609 static int
610 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
611 {
612 	int nd_delay, nd_gctimer, r_skip_req;
613 	time_t lle_hittime;
614 	long delay;
615 
616 	*do_switch = 0;
617 	nd_gctimer = V_nd6_gctimer;
618 	nd_delay = V_nd6_delay;
619 
620 	LLE_REQ_LOCK(lle);
621 	r_skip_req = lle->r_skip_req;
622 	lle_hittime = lle->lle_hittime;
623 	LLE_REQ_UNLOCK(lle);
624 
625 	if (r_skip_req > 0) {
626 
627 		/*
628 		 * Nonzero r_skip_req value was set upon entering
629 		 * STALE state. Since value was not changed, no
630 		 * packets were passed using this lle. Ask for
631 		 * timer reschedule and keep STALE state.
632 		 */
633 		delay = (long)(MIN(nd_gctimer, nd_delay));
634 		delay *= hz;
635 		if (lle->lle_remtime > delay)
636 			lle->lle_remtime -= delay;
637 		else {
638 			delay = lle->lle_remtime;
639 			lle->lle_remtime = 0;
640 		}
641 
642 		if (delay == 0) {
643 
644 			/*
645 			 * The original ng6_gctime timeout ended,
646 			 * no more rescheduling.
647 			 */
648 			return (0);
649 		}
650 
651 		*pdelay = delay;
652 		return (1);
653 	}
654 
655 	/*
656 	 * Packet received. Verify timestamp
657 	 */
658 	delay = (long)(time_uptime - lle_hittime);
659 	if (delay < nd_delay) {
660 
661 		/*
662 		 * V_nd6_delay still not passed since the first
663 		 * hit in STALE state.
664 		 * Reshedule timer and return.
665 		 */
666 		*pdelay = (long)(nd_delay - delay) * hz;
667 		return (1);
668 	}
669 
670 	/* Request switching to probe */
671 	*do_switch = 1;
672 	return (0);
673 }
674 
675 
676 /*
677  * Switch @lle state to new state optionally arming timers.
678  *
679  * Set noinline to be dtrace-friendly
680  */
681 __noinline void
682 nd6_llinfo_setstate(struct llentry *lle, int newstate)
683 {
684 	struct ifnet *ifp;
685 	int nd_gctimer, nd_delay;
686 	long delay, remtime;
687 
688 	delay = 0;
689 	remtime = 0;
690 
691 	switch (newstate) {
692 	case ND6_LLINFO_INCOMPLETE:
693 		ifp = lle->lle_tbl->llt_ifp;
694 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
695 		break;
696 	case ND6_LLINFO_REACHABLE:
697 		if (!ND6_LLINFO_PERMANENT(lle)) {
698 			ifp = lle->lle_tbl->llt_ifp;
699 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
700 		}
701 		break;
702 	case ND6_LLINFO_STALE:
703 
704 		/*
705 		 * Notify fast path that we want to know if any packet
706 		 * is transmitted by setting r_skip_req.
707 		 */
708 		LLE_REQ_LOCK(lle);
709 		lle->r_skip_req = 1;
710 		LLE_REQ_UNLOCK(lle);
711 		nd_delay = V_nd6_delay;
712 		nd_gctimer = V_nd6_gctimer;
713 
714 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
715 		remtime = (long)nd_gctimer * hz - delay;
716 		break;
717 	case ND6_LLINFO_DELAY:
718 		lle->la_asked = 0;
719 		delay = (long)V_nd6_delay * hz;
720 		break;
721 	}
722 
723 	if (delay > 0)
724 		nd6_llinfo_settimer_locked(lle, delay);
725 
726 	lle->lle_remtime = remtime;
727 	lle->ln_state = newstate;
728 }
729 
730 /*
731  * Timer-dependent part of nd state machine.
732  *
733  * Set noinline to be dtrace-friendly
734  */
735 static __noinline void
736 nd6_llinfo_timer(void *arg)
737 {
738 	struct epoch_tracker et;
739 	struct llentry *ln;
740 	struct in6_addr *dst, *pdst, *psrc, src;
741 	struct ifnet *ifp;
742 	struct nd_ifinfo *ndi;
743 	int do_switch, send_ns;
744 	long delay;
745 
746 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
747 	ln = (struct llentry *)arg;
748 	ifp = lltable_get_ifp(ln->lle_tbl);
749 	CURVNET_SET(ifp->if_vnet);
750 
751 	ND6_RLOCK();
752 	LLE_WLOCK(ln);
753 	if (callout_pending(&ln->lle_timer)) {
754 		/*
755 		 * Here we are a bit odd here in the treatment of
756 		 * active/pending. If the pending bit is set, it got
757 		 * rescheduled before I ran. The active
758 		 * bit we ignore, since if it was stopped
759 		 * in ll_tablefree() and was currently running
760 		 * it would have return 0 so the code would
761 		 * not have deleted it since the callout could
762 		 * not be stopped so we want to go through
763 		 * with the delete here now. If the callout
764 		 * was restarted, the pending bit will be back on and
765 		 * we just want to bail since the callout_reset would
766 		 * return 1 and our reference would have been removed
767 		 * by nd6_llinfo_settimer_locked above since canceled
768 		 * would have been 1.
769 		 */
770 		LLE_WUNLOCK(ln);
771 		ND6_RUNLOCK();
772 		CURVNET_RESTORE();
773 		return;
774 	}
775 	NET_EPOCH_ENTER(et);
776 	ndi = ND_IFINFO(ifp);
777 	send_ns = 0;
778 	dst = &ln->r_l3addr.addr6;
779 	pdst = dst;
780 
781 	if (ln->ln_ntick > 0) {
782 		if (ln->ln_ntick > INT_MAX) {
783 			ln->ln_ntick -= INT_MAX;
784 			nd6_llinfo_settimer_locked(ln, INT_MAX);
785 		} else {
786 			ln->ln_ntick = 0;
787 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
788 		}
789 		goto done;
790 	}
791 
792 	if (ln->la_flags & LLE_STATIC) {
793 		goto done;
794 	}
795 
796 	if (ln->la_flags & LLE_DELETED) {
797 		nd6_free(&ln, 0);
798 		goto done;
799 	}
800 
801 	switch (ln->ln_state) {
802 	case ND6_LLINFO_INCOMPLETE:
803 		if (ln->la_asked < V_nd6_mmaxtries) {
804 			ln->la_asked++;
805 			send_ns = 1;
806 			/* Send NS to multicast address */
807 			pdst = NULL;
808 		} else {
809 			struct mbuf *m = ln->la_hold;
810 			if (m) {
811 				struct mbuf *m0;
812 
813 				/*
814 				 * assuming every packet in la_hold has the
815 				 * same IP header.  Send error after unlock.
816 				 */
817 				m0 = m->m_nextpkt;
818 				m->m_nextpkt = NULL;
819 				ln->la_hold = m0;
820 				clear_llinfo_pqueue(ln);
821 			}
822 			nd6_free(&ln, 0);
823 			if (m != NULL)
824 				icmp6_error2(m, ICMP6_DST_UNREACH,
825 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
826 		}
827 		break;
828 	case ND6_LLINFO_REACHABLE:
829 		if (!ND6_LLINFO_PERMANENT(ln))
830 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
831 		break;
832 
833 	case ND6_LLINFO_STALE:
834 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
835 
836 			/*
837 			 * No packet has used this entry and GC timeout
838 			 * has not been passed. Reshedule timer and
839 			 * return.
840 			 */
841 			nd6_llinfo_settimer_locked(ln, delay);
842 			break;
843 		}
844 
845 		if (do_switch == 0) {
846 
847 			/*
848 			 * GC timer has ended and entry hasn't been used.
849 			 * Run Garbage collector (RFC 4861, 5.3)
850 			 */
851 			if (!ND6_LLINFO_PERMANENT(ln))
852 				nd6_free(&ln, 1);
853 			break;
854 		}
855 
856 		/* Entry has been used AND delay timer has ended. */
857 
858 		/* FALLTHROUGH */
859 
860 	case ND6_LLINFO_DELAY:
861 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
862 			/* We need NUD */
863 			ln->la_asked = 1;
864 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
865 			send_ns = 1;
866 		} else
867 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
868 		break;
869 	case ND6_LLINFO_PROBE:
870 		if (ln->la_asked < V_nd6_umaxtries) {
871 			ln->la_asked++;
872 			send_ns = 1;
873 		} else {
874 			nd6_free(&ln, 0);
875 		}
876 		break;
877 	default:
878 		panic("%s: paths in a dark night can be confusing: %d",
879 		    __func__, ln->ln_state);
880 	}
881 done:
882 	if (ln != NULL)
883 		ND6_RUNLOCK();
884 	if (send_ns != 0) {
885 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
886 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
887 		LLE_FREE_LOCKED(ln);
888 		ln = NULL;
889 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
890 	}
891 
892 	if (ln != NULL)
893 		LLE_FREE_LOCKED(ln);
894 	NET_EPOCH_EXIT(et);
895 	CURVNET_RESTORE();
896 }
897 
898 
899 /*
900  * ND6 timer routine to expire default route list and prefix list
901  */
902 void
903 nd6_timer(void *arg)
904 {
905 	CURVNET_SET((struct vnet *) arg);
906 	struct epoch_tracker et;
907 	struct nd_prhead prl;
908 	struct nd_prefix *pr, *npr;
909 	struct ifnet *ifp;
910 	struct in6_ifaddr *ia6, *nia6;
911 	uint64_t genid;
912 
913 	LIST_INIT(&prl);
914 
915 	NET_EPOCH_ENTER(et);
916 	nd6_defrouter_timer();
917 
918 	/*
919 	 * expire interface addresses.
920 	 * in the past the loop was inside prefix expiry processing.
921 	 * However, from a stricter speci-confrmance standpoint, we should
922 	 * rather separate address lifetimes and prefix lifetimes.
923 	 *
924 	 * XXXRW: in6_ifaddrhead locking.
925 	 */
926   addrloop:
927 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
928 		/* check address lifetime */
929 		if (IFA6_IS_INVALID(ia6)) {
930 			int regen = 0;
931 
932 			/*
933 			 * If the expiring address is temporary, try
934 			 * regenerating a new one.  This would be useful when
935 			 * we suspended a laptop PC, then turned it on after a
936 			 * period that could invalidate all temporary
937 			 * addresses.  Although we may have to restart the
938 			 * loop (see below), it must be after purging the
939 			 * address.  Otherwise, we'd see an infinite loop of
940 			 * regeneration.
941 			 */
942 			if (V_ip6_use_tempaddr &&
943 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
944 				if (regen_tmpaddr(ia6) == 0)
945 					regen = 1;
946 			}
947 
948 			in6_purgeaddr(&ia6->ia_ifa);
949 
950 			if (regen)
951 				goto addrloop; /* XXX: see below */
952 		} else if (IFA6_IS_DEPRECATED(ia6)) {
953 			int oldflags = ia6->ia6_flags;
954 
955 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
956 
957 			/*
958 			 * If a temporary address has just become deprecated,
959 			 * regenerate a new one if possible.
960 			 */
961 			if (V_ip6_use_tempaddr &&
962 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
963 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
964 
965 				if (regen_tmpaddr(ia6) == 0) {
966 					/*
967 					 * A new temporary address is
968 					 * generated.
969 					 * XXX: this means the address chain
970 					 * has changed while we are still in
971 					 * the loop.  Although the change
972 					 * would not cause disaster (because
973 					 * it's not a deletion, but an
974 					 * addition,) we'd rather restart the
975 					 * loop just for safety.  Or does this
976 					 * significantly reduce performance??
977 					 */
978 					goto addrloop;
979 				}
980 			}
981 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
982 			/*
983 			 * Schedule DAD for a tentative address.  This happens
984 			 * if the interface was down or not running
985 			 * when the address was configured.
986 			 */
987 			int delay;
988 
989 			delay = arc4random() %
990 			    (MAX_RTR_SOLICITATION_DELAY * hz);
991 			nd6_dad_start((struct ifaddr *)ia6, delay);
992 		} else {
993 			/*
994 			 * Check status of the interface.  If it is down,
995 			 * mark the address as tentative for future DAD.
996 			 */
997 			ifp = ia6->ia_ifp;
998 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
999 			    ((ifp->if_flags & IFF_UP) == 0 ||
1000 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1001 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1002 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1003 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1004 			}
1005 
1006 			/*
1007 			 * A new RA might have made a deprecated address
1008 			 * preferred.
1009 			 */
1010 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1011 		}
1012 	}
1013 	NET_EPOCH_EXIT(et);
1014 
1015 	ND6_WLOCK();
1016 restart:
1017 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1018 		/*
1019 		 * Expire prefixes. Since the pltime is only used for
1020 		 * autoconfigured addresses, pltime processing for prefixes is
1021 		 * not necessary.
1022 		 *
1023 		 * Only unlink after all derived addresses have expired. This
1024 		 * may not occur until two hours after the prefix has expired
1025 		 * per RFC 4862. If the prefix expires before its derived
1026 		 * addresses, mark it off-link. This will be done automatically
1027 		 * after unlinking if no address references remain.
1028 		 */
1029 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1030 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1031 			continue;
1032 
1033 		if (pr->ndpr_addrcnt == 0) {
1034 			nd6_prefix_unlink(pr, &prl);
1035 			continue;
1036 		}
1037 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1038 			genid = V_nd6_list_genid;
1039 			nd6_prefix_ref(pr);
1040 			ND6_WUNLOCK();
1041 			ND6_ONLINK_LOCK();
1042 			(void)nd6_prefix_offlink(pr);
1043 			ND6_ONLINK_UNLOCK();
1044 			ND6_WLOCK();
1045 			nd6_prefix_rele(pr);
1046 			if (genid != V_nd6_list_genid)
1047 				goto restart;
1048 		}
1049 	}
1050 	ND6_WUNLOCK();
1051 
1052 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1053 		LIST_REMOVE(pr, ndpr_entry);
1054 		nd6_prefix_del(pr);
1055 	}
1056 
1057 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1058 	    nd6_timer, curvnet);
1059 
1060 	CURVNET_RESTORE();
1061 }
1062 
1063 /*
1064  * ia6 - deprecated/invalidated temporary address
1065  */
1066 static int
1067 regen_tmpaddr(struct in6_ifaddr *ia6)
1068 {
1069 	struct ifaddr *ifa;
1070 	struct ifnet *ifp;
1071 	struct in6_ifaddr *public_ifa6 = NULL;
1072 
1073 	NET_EPOCH_ASSERT();
1074 
1075 	ifp = ia6->ia_ifa.ifa_ifp;
1076 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1077 		struct in6_ifaddr *it6;
1078 
1079 		if (ifa->ifa_addr->sa_family != AF_INET6)
1080 			continue;
1081 
1082 		it6 = (struct in6_ifaddr *)ifa;
1083 
1084 		/* ignore no autoconf addresses. */
1085 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1086 			continue;
1087 
1088 		/* ignore autoconf addresses with different prefixes. */
1089 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1090 			continue;
1091 
1092 		/*
1093 		 * Now we are looking at an autoconf address with the same
1094 		 * prefix as ours.  If the address is temporary and is still
1095 		 * preferred, do not create another one.  It would be rare, but
1096 		 * could happen, for example, when we resume a laptop PC after
1097 		 * a long period.
1098 		 */
1099 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1100 		    !IFA6_IS_DEPRECATED(it6)) {
1101 			public_ifa6 = NULL;
1102 			break;
1103 		}
1104 
1105 		/*
1106 		 * This is a public autoconf address that has the same prefix
1107 		 * as ours.  If it is preferred, keep it.  We can't break the
1108 		 * loop here, because there may be a still-preferred temporary
1109 		 * address with the prefix.
1110 		 */
1111 		if (!IFA6_IS_DEPRECATED(it6))
1112 			public_ifa6 = it6;
1113 	}
1114 	if (public_ifa6 != NULL)
1115 		ifa_ref(&public_ifa6->ia_ifa);
1116 
1117 	if (public_ifa6 != NULL) {
1118 		int e;
1119 
1120 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1121 			ifa_free(&public_ifa6->ia_ifa);
1122 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1123 			    " tmp addr,errno=%d\n", e);
1124 			return (-1);
1125 		}
1126 		ifa_free(&public_ifa6->ia_ifa);
1127 		return (0);
1128 	}
1129 
1130 	return (-1);
1131 }
1132 
1133 /*
1134  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1135  * cache entries are freed in in6_domifdetach().
1136  */
1137 void
1138 nd6_purge(struct ifnet *ifp)
1139 {
1140 	struct nd_prhead prl;
1141 	struct nd_prefix *pr, *npr;
1142 
1143 	LIST_INIT(&prl);
1144 
1145 	/* Purge default router list entries toward ifp. */
1146 	nd6_defrouter_purge(ifp);
1147 
1148 	ND6_WLOCK();
1149 	/*
1150 	 * Remove prefixes on ifp. We should have already removed addresses on
1151 	 * this interface, so no addresses should be referencing these prefixes.
1152 	 */
1153 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1154 		if (pr->ndpr_ifp == ifp)
1155 			nd6_prefix_unlink(pr, &prl);
1156 	}
1157 	ND6_WUNLOCK();
1158 
1159 	/* Delete the unlinked prefix objects. */
1160 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1161 		LIST_REMOVE(pr, ndpr_entry);
1162 		nd6_prefix_del(pr);
1163 	}
1164 
1165 	/* cancel default outgoing interface setting */
1166 	if (V_nd6_defifindex == ifp->if_index)
1167 		nd6_setdefaultiface(0);
1168 
1169 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1170 		/* Refresh default router list. */
1171 		defrouter_select_fib(ifp->if_fib);
1172 	}
1173 }
1174 
1175 /*
1176  * the caller acquires and releases the lock on the lltbls
1177  * Returns the llentry locked
1178  */
1179 struct llentry *
1180 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1181 {
1182 	struct sockaddr_in6 sin6;
1183 	struct llentry *ln;
1184 
1185 	bzero(&sin6, sizeof(sin6));
1186 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1187 	sin6.sin6_family = AF_INET6;
1188 	sin6.sin6_addr = *addr6;
1189 
1190 	IF_AFDATA_LOCK_ASSERT(ifp);
1191 
1192 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1193 
1194 	return (ln);
1195 }
1196 
1197 static struct llentry *
1198 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1199 {
1200 	struct sockaddr_in6 sin6;
1201 	struct llentry *ln;
1202 
1203 	bzero(&sin6, sizeof(sin6));
1204 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1205 	sin6.sin6_family = AF_INET6;
1206 	sin6.sin6_addr = *addr6;
1207 
1208 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1209 	if (ln != NULL)
1210 		ln->ln_state = ND6_LLINFO_NOSTATE;
1211 
1212 	return (ln);
1213 }
1214 
1215 /*
1216  * Test whether a given IPv6 address is a neighbor or not, ignoring
1217  * the actual neighbor cache.  The neighbor cache is ignored in order
1218  * to not reenter the routing code from within itself.
1219  */
1220 static int
1221 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1222 {
1223 	struct nd_prefix *pr;
1224 	struct ifaddr *ifa;
1225 	struct rt_addrinfo info;
1226 	struct sockaddr_in6 rt_key;
1227 	const struct sockaddr *dst6;
1228 	uint64_t genid;
1229 	int error, fibnum;
1230 
1231 	/*
1232 	 * A link-local address is always a neighbor.
1233 	 * XXX: a link does not necessarily specify a single interface.
1234 	 */
1235 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1236 		struct sockaddr_in6 sin6_copy;
1237 		u_int32_t zone;
1238 
1239 		/*
1240 		 * We need sin6_copy since sa6_recoverscope() may modify the
1241 		 * content (XXX).
1242 		 */
1243 		sin6_copy = *addr;
1244 		if (sa6_recoverscope(&sin6_copy))
1245 			return (0); /* XXX: should be impossible */
1246 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1247 			return (0);
1248 		if (sin6_copy.sin6_scope_id == zone)
1249 			return (1);
1250 		else
1251 			return (0);
1252 	}
1253 
1254 	bzero(&rt_key, sizeof(rt_key));
1255 	bzero(&info, sizeof(info));
1256 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1257 
1258 	/*
1259 	 * If the address matches one of our addresses,
1260 	 * it should be a neighbor.
1261 	 * If the address matches one of our on-link prefixes, it should be a
1262 	 * neighbor.
1263 	 */
1264 	ND6_RLOCK();
1265 restart:
1266 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1267 		if (pr->ndpr_ifp != ifp)
1268 			continue;
1269 
1270 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1271 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1272 
1273 			/*
1274 			 * We only need to check all FIBs if add_addr_allfibs
1275 			 * is unset. If set, checking any FIB will suffice.
1276 			 */
1277 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1278 			for (; fibnum < rt_numfibs; fibnum++) {
1279 				genid = V_nd6_list_genid;
1280 				ND6_RUNLOCK();
1281 
1282 				/*
1283 				 * Restore length field before
1284 				 * retrying lookup
1285 				 */
1286 				rt_key.sin6_len = sizeof(rt_key);
1287 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1288 						        &info);
1289 
1290 				ND6_RLOCK();
1291 				if (genid != V_nd6_list_genid)
1292 					goto restart;
1293 				if (error == 0)
1294 					break;
1295 			}
1296 			if (error != 0)
1297 				continue;
1298 
1299 			/*
1300 			 * This is the case where multiple interfaces
1301 			 * have the same prefix, but only one is installed
1302 			 * into the routing table and that prefix entry
1303 			 * is not the one being examined here. In the case
1304 			 * where RADIX_MPATH is enabled, multiple route
1305 			 * entries (of the same rt_key value) will be
1306 			 * installed because the interface addresses all
1307 			 * differ.
1308 			 */
1309 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1310 			    &rt_key.sin6_addr))
1311 				continue;
1312 		}
1313 
1314 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1315 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1316 			ND6_RUNLOCK();
1317 			return (1);
1318 		}
1319 	}
1320 	ND6_RUNLOCK();
1321 
1322 	/*
1323 	 * If the address is assigned on the node of the other side of
1324 	 * a p2p interface, the address should be a neighbor.
1325 	 */
1326 	if (ifp->if_flags & IFF_POINTOPOINT) {
1327 		struct epoch_tracker et;
1328 
1329 		NET_EPOCH_ENTER(et);
1330 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1331 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1332 				continue;
1333 			if (ifa->ifa_dstaddr != NULL &&
1334 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1335 				NET_EPOCH_EXIT(et);
1336 				return 1;
1337 			}
1338 		}
1339 		NET_EPOCH_EXIT(et);
1340 	}
1341 
1342 	/*
1343 	 * If the default router list is empty, all addresses are regarded
1344 	 * as on-link, and thus, as a neighbor.
1345 	 */
1346 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1347 	    nd6_defrouter_list_empty() &&
1348 	    V_nd6_defifindex == ifp->if_index) {
1349 		return (1);
1350 	}
1351 
1352 	return (0);
1353 }
1354 
1355 
1356 /*
1357  * Detect if a given IPv6 address identifies a neighbor on a given link.
1358  * XXX: should take care of the destination of a p2p link?
1359  */
1360 int
1361 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1362 {
1363 	struct llentry *lle;
1364 	int rc = 0;
1365 
1366 	NET_EPOCH_ASSERT();
1367 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1368 	if (nd6_is_new_addr_neighbor(addr, ifp))
1369 		return (1);
1370 
1371 	/*
1372 	 * Even if the address matches none of our addresses, it might be
1373 	 * in the neighbor cache.
1374 	 */
1375 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1376 		LLE_RUNLOCK(lle);
1377 		rc = 1;
1378 	}
1379 	return (rc);
1380 }
1381 
1382 /*
1383  * Free an nd6 llinfo entry.
1384  * Since the function would cause significant changes in the kernel, DO NOT
1385  * make it global, unless you have a strong reason for the change, and are sure
1386  * that the change is safe.
1387  *
1388  * Set noinline to be dtrace-friendly
1389  */
1390 static __noinline void
1391 nd6_free(struct llentry **lnp, int gc)
1392 {
1393 	struct ifnet *ifp;
1394 	struct llentry *ln;
1395 	struct nd_defrouter *dr;
1396 
1397 	ln = *lnp;
1398 	*lnp = NULL;
1399 
1400 	LLE_WLOCK_ASSERT(ln);
1401 	ND6_RLOCK_ASSERT();
1402 
1403 	ifp = lltable_get_ifp(ln->lle_tbl);
1404 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1405 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1406 	else
1407 		dr = NULL;
1408 	ND6_RUNLOCK();
1409 
1410 	if ((ln->la_flags & LLE_DELETED) == 0)
1411 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1412 
1413 	/*
1414 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1415 	 * even though it is not harmful, it was not really necessary.
1416 	 */
1417 
1418 	/* cancel timer */
1419 	nd6_llinfo_settimer_locked(ln, -1);
1420 
1421 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1422 		if (dr != NULL && dr->expire &&
1423 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1424 			/*
1425 			 * If the reason for the deletion is just garbage
1426 			 * collection, and the neighbor is an active default
1427 			 * router, do not delete it.  Instead, reset the GC
1428 			 * timer using the router's lifetime.
1429 			 * Simply deleting the entry would affect default
1430 			 * router selection, which is not necessarily a good
1431 			 * thing, especially when we're using router preference
1432 			 * values.
1433 			 * XXX: the check for ln_state would be redundant,
1434 			 *      but we intentionally keep it just in case.
1435 			 */
1436 			if (dr->expire > time_uptime)
1437 				nd6_llinfo_settimer_locked(ln,
1438 				    (dr->expire - time_uptime) * hz);
1439 			else
1440 				nd6_llinfo_settimer_locked(ln,
1441 				    (long)V_nd6_gctimer * hz);
1442 
1443 			LLE_REMREF(ln);
1444 			LLE_WUNLOCK(ln);
1445 			defrouter_rele(dr);
1446 			return;
1447 		}
1448 
1449 		if (dr) {
1450 			/*
1451 			 * Unreachablity of a router might affect the default
1452 			 * router selection and on-link detection of advertised
1453 			 * prefixes.
1454 			 */
1455 
1456 			/*
1457 			 * Temporarily fake the state to choose a new default
1458 			 * router and to perform on-link determination of
1459 			 * prefixes correctly.
1460 			 * Below the state will be set correctly,
1461 			 * or the entry itself will be deleted.
1462 			 */
1463 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1464 		}
1465 
1466 		if (ln->ln_router || dr) {
1467 
1468 			/*
1469 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1470 			 * rnh and for the calls to pfxlist_onlink_check() and
1471 			 * defrouter_select_fib() in the block further down for calls
1472 			 * into nd6_lookup().  We still hold a ref.
1473 			 */
1474 			LLE_WUNLOCK(ln);
1475 
1476 			/*
1477 			 * rt6_flush must be called whether or not the neighbor
1478 			 * is in the Default Router List.
1479 			 * See a corresponding comment in nd6_na_input().
1480 			 */
1481 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1482 		}
1483 
1484 		if (dr) {
1485 			/*
1486 			 * Since defrouter_select_fib() does not affect the
1487 			 * on-link determination and MIP6 needs the check
1488 			 * before the default router selection, we perform
1489 			 * the check now.
1490 			 */
1491 			pfxlist_onlink_check();
1492 
1493 			/*
1494 			 * Refresh default router list.
1495 			 */
1496 			defrouter_select_fib(dr->ifp->if_fib);
1497 		}
1498 
1499 		/*
1500 		 * If this entry was added by an on-link redirect, remove the
1501 		 * corresponding host route.
1502 		 */
1503 		if (ln->la_flags & LLE_REDIRECT)
1504 			nd6_free_redirect(ln);
1505 
1506 		if (ln->ln_router || dr)
1507 			LLE_WLOCK(ln);
1508 	}
1509 
1510 	/*
1511 	 * Save to unlock. We still hold an extra reference and will not
1512 	 * free(9) in llentry_free() if someone else holds one as well.
1513 	 */
1514 	LLE_WUNLOCK(ln);
1515 	IF_AFDATA_LOCK(ifp);
1516 	LLE_WLOCK(ln);
1517 	/* Guard against race with other llentry_free(). */
1518 	if (ln->la_flags & LLE_LINKED) {
1519 		/* Remove callout reference */
1520 		LLE_REMREF(ln);
1521 		lltable_unlink_entry(ln->lle_tbl, ln);
1522 	}
1523 	IF_AFDATA_UNLOCK(ifp);
1524 
1525 	llentry_free(ln);
1526 	if (dr != NULL)
1527 		defrouter_rele(dr);
1528 }
1529 
1530 static int
1531 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1532 {
1533 
1534 	if (nh->nh_flags & NHF_REDIRECT)
1535 		return (1);
1536 
1537 	return (0);
1538 }
1539 
1540 /*
1541  * Remove the rtentry for the given llentry,
1542  * both of which were installed by a redirect.
1543  */
1544 static void
1545 nd6_free_redirect(const struct llentry *ln)
1546 {
1547 	int fibnum;
1548 	struct sockaddr_in6 sin6;
1549 	struct rt_addrinfo info;
1550 
1551 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1552 	memset(&info, 0, sizeof(info));
1553 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1554 	info.rti_filter = nd6_isdynrte;
1555 
1556 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1557 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1558 }
1559 
1560 /*
1561  * Rejuvenate this function for routing operations related
1562  * processing.
1563  */
1564 void
1565 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1566 {
1567 	struct sockaddr_in6 *gateway;
1568 	struct nd_defrouter *dr;
1569 	struct nhop_object *nh;
1570 
1571 	nh = rt->rt_nhop;
1572 	gateway = &nh->gw6_sa;
1573 
1574 	switch (req) {
1575 	case RTM_ADD:
1576 		break;
1577 
1578 	case RTM_DELETE:
1579 		/*
1580 		 * Only indirect routes are interesting.
1581 		 */
1582 		if ((nh->nh_flags & NHF_GATEWAY) == 0)
1583 			return;
1584 		/*
1585 		 * check for default route
1586 		 */
1587 		if (nh->nh_flags & NHF_DEFAULT) {
1588 			dr = defrouter_lookup(&gateway->sin6_addr, nh->nh_ifp);
1589 			if (dr != NULL) {
1590 				dr->installed = 0;
1591 				defrouter_rele(dr);
1592 			}
1593 		}
1594 		break;
1595 	}
1596 }
1597 
1598 
1599 int
1600 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1601 {
1602 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1603 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1604 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1605 	struct epoch_tracker et;
1606 	int error = 0;
1607 
1608 	if (ifp->if_afdata[AF_INET6] == NULL)
1609 		return (EPFNOSUPPORT);
1610 	switch (cmd) {
1611 	case OSIOCGIFINFO_IN6:
1612 #define ND	ndi->ndi
1613 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1614 		bzero(&ND, sizeof(ND));
1615 		ND.linkmtu = IN6_LINKMTU(ifp);
1616 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1617 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1618 		ND.reachable = ND_IFINFO(ifp)->reachable;
1619 		ND.retrans = ND_IFINFO(ifp)->retrans;
1620 		ND.flags = ND_IFINFO(ifp)->flags;
1621 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1622 		ND.chlim = ND_IFINFO(ifp)->chlim;
1623 		break;
1624 	case SIOCGIFINFO_IN6:
1625 		ND = *ND_IFINFO(ifp);
1626 		break;
1627 	case SIOCSIFINFO_IN6:
1628 		/*
1629 		 * used to change host variables from userland.
1630 		 * intended for a use on router to reflect RA configurations.
1631 		 */
1632 		/* 0 means 'unspecified' */
1633 		if (ND.linkmtu != 0) {
1634 			if (ND.linkmtu < IPV6_MMTU ||
1635 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1636 				error = EINVAL;
1637 				break;
1638 			}
1639 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1640 		}
1641 
1642 		if (ND.basereachable != 0) {
1643 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1644 
1645 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1646 			if (ND.basereachable != obasereachable)
1647 				ND_IFINFO(ifp)->reachable =
1648 				    ND_COMPUTE_RTIME(ND.basereachable);
1649 		}
1650 		if (ND.retrans != 0)
1651 			ND_IFINFO(ifp)->retrans = ND.retrans;
1652 		if (ND.chlim != 0)
1653 			ND_IFINFO(ifp)->chlim = ND.chlim;
1654 		/* FALLTHROUGH */
1655 	case SIOCSIFINFO_FLAGS:
1656 	{
1657 		struct ifaddr *ifa;
1658 		struct in6_ifaddr *ia;
1659 
1660 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1661 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1662 			/* ifdisabled 1->0 transision */
1663 
1664 			/*
1665 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1666 			 * has an link-local address with IN6_IFF_DUPLICATED,
1667 			 * do not clear ND6_IFF_IFDISABLED.
1668 			 * See RFC 4862, Section 5.4.5.
1669 			 */
1670 			NET_EPOCH_ENTER(et);
1671 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1672 				if (ifa->ifa_addr->sa_family != AF_INET6)
1673 					continue;
1674 				ia = (struct in6_ifaddr *)ifa;
1675 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1676 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1677 					break;
1678 			}
1679 			NET_EPOCH_EXIT(et);
1680 
1681 			if (ifa != NULL) {
1682 				/* LLA is duplicated. */
1683 				ND.flags |= ND6_IFF_IFDISABLED;
1684 				log(LOG_ERR, "Cannot enable an interface"
1685 				    " with a link-local address marked"
1686 				    " duplicate.\n");
1687 			} else {
1688 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1689 				if (ifp->if_flags & IFF_UP)
1690 					in6_if_up(ifp);
1691 			}
1692 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1693 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1694 			/* ifdisabled 0->1 transision */
1695 			/* Mark all IPv6 address as tentative. */
1696 
1697 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1698 			if (V_ip6_dad_count > 0 &&
1699 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1700 				NET_EPOCH_ENTER(et);
1701 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1702 				    ifa_link) {
1703 					if (ifa->ifa_addr->sa_family !=
1704 					    AF_INET6)
1705 						continue;
1706 					ia = (struct in6_ifaddr *)ifa;
1707 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1708 				}
1709 				NET_EPOCH_EXIT(et);
1710 			}
1711 		}
1712 
1713 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1714 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1715 				/* auto_linklocal 0->1 transision */
1716 
1717 				/* If no link-local address on ifp, configure */
1718 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1719 				in6_ifattach(ifp, NULL);
1720 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1721 			    ifp->if_flags & IFF_UP) {
1722 				/*
1723 				 * When the IF already has
1724 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1725 				 * address is assigned, and IFF_UP, try to
1726 				 * assign one.
1727 				 */
1728 				NET_EPOCH_ENTER(et);
1729 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1730 				    ifa_link) {
1731 					if (ifa->ifa_addr->sa_family !=
1732 					    AF_INET6)
1733 						continue;
1734 					ia = (struct in6_ifaddr *)ifa;
1735 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1736 						break;
1737 				}
1738 				NET_EPOCH_EXIT(et);
1739 				if (ifa != NULL)
1740 					/* No LLA is configured. */
1741 					in6_ifattach(ifp, NULL);
1742 			}
1743 		}
1744 		ND_IFINFO(ifp)->flags = ND.flags;
1745 		break;
1746 	}
1747 #undef ND
1748 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1749 		/* sync kernel routing table with the default router list */
1750 		defrouter_reset();
1751 		defrouter_select_fib(RT_ALL_FIBS);
1752 		break;
1753 	case SIOCSPFXFLUSH_IN6:
1754 	{
1755 		/* flush all the prefix advertised by routers */
1756 		struct in6_ifaddr *ia, *ia_next;
1757 		struct nd_prefix *pr, *next;
1758 		struct nd_prhead prl;
1759 
1760 		LIST_INIT(&prl);
1761 
1762 		ND6_WLOCK();
1763 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1764 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1765 				continue; /* XXX */
1766 			nd6_prefix_unlink(pr, &prl);
1767 		}
1768 		ND6_WUNLOCK();
1769 
1770 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1771 			LIST_REMOVE(pr, ndpr_entry);
1772 			/* XXXRW: in6_ifaddrhead locking. */
1773 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1774 			    ia_next) {
1775 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1776 					continue;
1777 
1778 				if (ia->ia6_ndpr == pr)
1779 					in6_purgeaddr(&ia->ia_ifa);
1780 			}
1781 			nd6_prefix_del(pr);
1782 		}
1783 		break;
1784 	}
1785 	case SIOCSRTRFLUSH_IN6:
1786 	{
1787 		/* flush all the default routers */
1788 
1789 		defrouter_reset();
1790 		nd6_defrouter_flush_all();
1791 		defrouter_select_fib(RT_ALL_FIBS);
1792 		break;
1793 	}
1794 	case SIOCGNBRINFO_IN6:
1795 	{
1796 		struct llentry *ln;
1797 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1798 
1799 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1800 			return (error);
1801 
1802 		NET_EPOCH_ENTER(et);
1803 		ln = nd6_lookup(&nb_addr, 0, ifp);
1804 		NET_EPOCH_EXIT(et);
1805 
1806 		if (ln == NULL) {
1807 			error = EINVAL;
1808 			break;
1809 		}
1810 		nbi->state = ln->ln_state;
1811 		nbi->asked = ln->la_asked;
1812 		nbi->isrouter = ln->ln_router;
1813 		if (ln->la_expire == 0)
1814 			nbi->expire = 0;
1815 		else
1816 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1817 			    (time_second - time_uptime);
1818 		LLE_RUNLOCK(ln);
1819 		break;
1820 	}
1821 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1822 		ndif->ifindex = V_nd6_defifindex;
1823 		break;
1824 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1825 		return (nd6_setdefaultiface(ndif->ifindex));
1826 	}
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Calculates new isRouter value based on provided parameters and
1832  * returns it.
1833  */
1834 static int
1835 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1836     int ln_router)
1837 {
1838 
1839 	/*
1840 	 * ICMP6 type dependent behavior.
1841 	 *
1842 	 * NS: clear IsRouter if new entry
1843 	 * RS: clear IsRouter
1844 	 * RA: set IsRouter if there's lladdr
1845 	 * redir: clear IsRouter if new entry
1846 	 *
1847 	 * RA case, (1):
1848 	 * The spec says that we must set IsRouter in the following cases:
1849 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1850 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1851 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1852 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1853 	 * neighbor cache, this is similar to (6).
1854 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1855 	 *
1856 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1857 	 *							D R
1858 	 *	0	n	n	(1)	c   ?     s
1859 	 *	0	y	n	(2)	c   s     s
1860 	 *	0	n	y	(3)	c   s     s
1861 	 *	0	y	y	(4)	c   s     s
1862 	 *	0	y	y	(5)	c   s     s
1863 	 *	1	--	n	(6) c	c	c s
1864 	 *	1	--	y	(7) c	c   s	c s
1865 	 *
1866 	 *					(c=clear s=set)
1867 	 */
1868 	switch (type & 0xff) {
1869 	case ND_NEIGHBOR_SOLICIT:
1870 		/*
1871 		 * New entry must have is_router flag cleared.
1872 		 */
1873 		if (is_new)					/* (6-7) */
1874 			ln_router = 0;
1875 		break;
1876 	case ND_REDIRECT:
1877 		/*
1878 		 * If the icmp is a redirect to a better router, always set the
1879 		 * is_router flag.  Otherwise, if the entry is newly created,
1880 		 * clear the flag.  [RFC 2461, sec 8.3]
1881 		 */
1882 		if (code == ND_REDIRECT_ROUTER)
1883 			ln_router = 1;
1884 		else {
1885 			if (is_new)				/* (6-7) */
1886 				ln_router = 0;
1887 		}
1888 		break;
1889 	case ND_ROUTER_SOLICIT:
1890 		/*
1891 		 * is_router flag must always be cleared.
1892 		 */
1893 		ln_router = 0;
1894 		break;
1895 	case ND_ROUTER_ADVERT:
1896 		/*
1897 		 * Mark an entry with lladdr as a router.
1898 		 */
1899 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1900 		    (is_new && new_addr)) {			/* (7) */
1901 			ln_router = 1;
1902 		}
1903 		break;
1904 	}
1905 
1906 	return (ln_router);
1907 }
1908 
1909 /*
1910  * Create neighbor cache entry and cache link-layer address,
1911  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1912  *
1913  * type - ICMP6 type
1914  * code - type dependent information
1915  *
1916  */
1917 void
1918 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1919     int lladdrlen, int type, int code)
1920 {
1921 	struct llentry *ln = NULL, *ln_tmp;
1922 	int is_newentry;
1923 	int do_update;
1924 	int olladdr;
1925 	int llchange;
1926 	int flags;
1927 	uint16_t router = 0;
1928 	struct sockaddr_in6 sin6;
1929 	struct mbuf *chain = NULL;
1930 	u_char linkhdr[LLE_MAX_LINKHDR];
1931 	size_t linkhdrsize;
1932 	int lladdr_off;
1933 
1934 	NET_EPOCH_ASSERT();
1935 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1936 
1937 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1938 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1939 
1940 	/* nothing must be updated for unspecified address */
1941 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1942 		return;
1943 
1944 	/*
1945 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1946 	 * the caller.
1947 	 *
1948 	 * XXX If the link does not have link-layer adderss, what should
1949 	 * we do? (ifp->if_addrlen == 0)
1950 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1951 	 * description on it in NS section (RFC 2461 7.2.3).
1952 	 */
1953 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1954 	ln = nd6_lookup(from, flags, ifp);
1955 	is_newentry = 0;
1956 	if (ln == NULL) {
1957 		flags |= LLE_EXCLUSIVE;
1958 		ln = nd6_alloc(from, 0, ifp);
1959 		if (ln == NULL)
1960 			return;
1961 
1962 		/*
1963 		 * Since we already know all the data for the new entry,
1964 		 * fill it before insertion.
1965 		 */
1966 		if (lladdr != NULL) {
1967 			linkhdrsize = sizeof(linkhdr);
1968 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1969 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1970 				return;
1971 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1972 			    lladdr_off);
1973 		}
1974 
1975 		IF_AFDATA_WLOCK(ifp);
1976 		LLE_WLOCK(ln);
1977 		/* Prefer any existing lle over newly-created one */
1978 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1979 		if (ln_tmp == NULL)
1980 			lltable_link_entry(LLTABLE6(ifp), ln);
1981 		IF_AFDATA_WUNLOCK(ifp);
1982 		if (ln_tmp == NULL) {
1983 			/* No existing lle, mark as new entry (6,7) */
1984 			is_newentry = 1;
1985 			if (lladdr != NULL) {	/* (7) */
1986 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1987 				EVENTHANDLER_INVOKE(lle_event, ln,
1988 				    LLENTRY_RESOLVED);
1989 			}
1990 		} else {
1991 			lltable_free_entry(LLTABLE6(ifp), ln);
1992 			ln = ln_tmp;
1993 			ln_tmp = NULL;
1994 		}
1995 	}
1996 	/* do nothing if static ndp is set */
1997 	if ((ln->la_flags & LLE_STATIC)) {
1998 		if (flags & LLE_EXCLUSIVE)
1999 			LLE_WUNLOCK(ln);
2000 		else
2001 			LLE_RUNLOCK(ln);
2002 		return;
2003 	}
2004 
2005 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2006 	if (olladdr && lladdr) {
2007 		llchange = bcmp(lladdr, ln->ll_addr,
2008 		    ifp->if_addrlen);
2009 	} else if (!olladdr && lladdr)
2010 		llchange = 1;
2011 	else
2012 		llchange = 0;
2013 
2014 	/*
2015 	 * newentry olladdr  lladdr  llchange	(*=record)
2016 	 *	0	n	n	--	(1)
2017 	 *	0	y	n	--	(2)
2018 	 *	0	n	y	y	(3) * STALE
2019 	 *	0	y	y	n	(4) *
2020 	 *	0	y	y	y	(5) * STALE
2021 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2022 	 *	1	--	y	--	(7) * STALE
2023 	 */
2024 
2025 	do_update = 0;
2026 	if (is_newentry == 0 && llchange != 0) {
2027 		do_update = 1;	/* (3,5) */
2028 
2029 		/*
2030 		 * Record source link-layer address
2031 		 * XXX is it dependent to ifp->if_type?
2032 		 */
2033 		linkhdrsize = sizeof(linkhdr);
2034 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2035 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2036 			return;
2037 
2038 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2039 		    lladdr_off) == 0) {
2040 			/* Entry was deleted */
2041 			return;
2042 		}
2043 
2044 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2045 
2046 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2047 
2048 		if (ln->la_hold != NULL)
2049 			nd6_grab_holdchain(ln, &chain, &sin6);
2050 	}
2051 
2052 	/* Calculates new router status */
2053 	router = nd6_is_router(type, code, is_newentry, olladdr,
2054 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2055 
2056 	ln->ln_router = router;
2057 	/* Mark non-router redirects with special flag */
2058 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2059 		ln->la_flags |= LLE_REDIRECT;
2060 
2061 	if (flags & LLE_EXCLUSIVE)
2062 		LLE_WUNLOCK(ln);
2063 	else
2064 		LLE_RUNLOCK(ln);
2065 
2066 	if (chain != NULL)
2067 		nd6_flush_holdchain(ifp, chain, &sin6);
2068 
2069 	/*
2070 	 * When the link-layer address of a router changes, select the
2071 	 * best router again.  In particular, when the neighbor entry is newly
2072 	 * created, it might affect the selection policy.
2073 	 * Question: can we restrict the first condition to the "is_newentry"
2074 	 * case?
2075 	 * XXX: when we hear an RA from a new router with the link-layer
2076 	 * address option, defrouter_select_fib() is called twice, since
2077 	 * defrtrlist_update called the function as well.  However, I believe
2078 	 * we can compromise the overhead, since it only happens the first
2079 	 * time.
2080 	 * XXX: although defrouter_select_fib() should not have a bad effect
2081 	 * for those are not autoconfigured hosts, we explicitly avoid such
2082 	 * cases for safety.
2083 	 */
2084 	if ((do_update || is_newentry) && router &&
2085 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2086 		/*
2087 		 * guaranteed recursion
2088 		 */
2089 		defrouter_select_fib(ifp->if_fib);
2090 	}
2091 }
2092 
2093 static void
2094 nd6_slowtimo(void *arg)
2095 {
2096 	struct epoch_tracker et;
2097 	CURVNET_SET((struct vnet *) arg);
2098 	struct nd_ifinfo *nd6if;
2099 	struct ifnet *ifp;
2100 
2101 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2102 	    nd6_slowtimo, curvnet);
2103 	NET_EPOCH_ENTER(et);
2104 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2105 		if (ifp->if_afdata[AF_INET6] == NULL)
2106 			continue;
2107 		nd6if = ND_IFINFO(ifp);
2108 		if (nd6if->basereachable && /* already initialized */
2109 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2110 			/*
2111 			 * Since reachable time rarely changes by router
2112 			 * advertisements, we SHOULD insure that a new random
2113 			 * value gets recomputed at least once every few hours.
2114 			 * (RFC 2461, 6.3.4)
2115 			 */
2116 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2117 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2118 		}
2119 	}
2120 	NET_EPOCH_EXIT(et);
2121 	CURVNET_RESTORE();
2122 }
2123 
2124 void
2125 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2126     struct sockaddr_in6 *sin6)
2127 {
2128 
2129 	LLE_WLOCK_ASSERT(ln);
2130 
2131 	*chain = ln->la_hold;
2132 	ln->la_hold = NULL;
2133 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2134 
2135 	if (ln->ln_state == ND6_LLINFO_STALE) {
2136 
2137 		/*
2138 		 * The first time we send a packet to a
2139 		 * neighbor whose entry is STALE, we have
2140 		 * to change the state to DELAY and a sets
2141 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2142 		 * seconds to ensure do neighbor unreachability
2143 		 * detection on expiration.
2144 		 * (RFC 2461 7.3.3)
2145 		 */
2146 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2147 	}
2148 }
2149 
2150 int
2151 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2152     struct sockaddr_in6 *dst, struct route *ro)
2153 {
2154 	int error;
2155 	int ip6len;
2156 	struct ip6_hdr *ip6;
2157 	struct m_tag *mtag;
2158 
2159 #ifdef MAC
2160 	mac_netinet6_nd6_send(ifp, m);
2161 #endif
2162 
2163 	/*
2164 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2165 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2166 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2167 	 * to be diverted to user space.  When re-injected into the kernel,
2168 	 * send_output() will directly dispatch them to the outgoing interface.
2169 	 */
2170 	if (send_sendso_input_hook != NULL) {
2171 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2172 		if (mtag != NULL) {
2173 			ip6 = mtod(m, struct ip6_hdr *);
2174 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2175 			/* Use the SEND socket */
2176 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2177 			    ip6len);
2178 			/* -1 == no app on SEND socket */
2179 			if (error == 0 || error != -1)
2180 			    return (error);
2181 		}
2182 	}
2183 
2184 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2185 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2186 	    mtod(m, struct ip6_hdr *));
2187 
2188 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2189 		origifp = ifp;
2190 
2191 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2192 	return (error);
2193 }
2194 
2195 /*
2196  * Lookup link headerfor @sa_dst address. Stores found
2197  * data in @desten buffer. Copy of lle ln_flags can be also
2198  * saved in @pflags if @pflags is non-NULL.
2199  *
2200  * If destination LLE does not exists or lle state modification
2201  * is required, call "slow" version.
2202  *
2203  * Return values:
2204  * - 0 on success (address copied to buffer).
2205  * - EWOULDBLOCK (no local error, but address is still unresolved)
2206  * - other errors (alloc failure, etc)
2207  */
2208 int
2209 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2210     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2211     struct llentry **plle)
2212 {
2213 	struct llentry *ln = NULL;
2214 	const struct sockaddr_in6 *dst6;
2215 
2216 	NET_EPOCH_ASSERT();
2217 
2218 	if (pflags != NULL)
2219 		*pflags = 0;
2220 
2221 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2222 
2223 	/* discard the packet if IPv6 operation is disabled on the interface */
2224 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2225 		m_freem(m);
2226 		return (ENETDOWN); /* better error? */
2227 	}
2228 
2229 	if (m != NULL && m->m_flags & M_MCAST) {
2230 		switch (ifp->if_type) {
2231 		case IFT_ETHER:
2232 		case IFT_L2VLAN:
2233 		case IFT_BRIDGE:
2234 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2235 						 desten);
2236 			return (0);
2237 		default:
2238 			m_freem(m);
2239 			return (EAFNOSUPPORT);
2240 		}
2241 	}
2242 
2243 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2244 	    ifp);
2245 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2246 		/* Entry found, let's copy lle info */
2247 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2248 		if (pflags != NULL)
2249 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2250 		/* Check if we have feedback request from nd6 timer */
2251 		if (ln->r_skip_req != 0) {
2252 			LLE_REQ_LOCK(ln);
2253 			ln->r_skip_req = 0; /* Notify that entry was used */
2254 			ln->lle_hittime = time_uptime;
2255 			LLE_REQ_UNLOCK(ln);
2256 		}
2257 		if (plle) {
2258 			LLE_ADDREF(ln);
2259 			*plle = ln;
2260 			LLE_WUNLOCK(ln);
2261 		}
2262 		return (0);
2263 	} else if (plle && ln)
2264 		LLE_WUNLOCK(ln);
2265 
2266 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2267 }
2268 
2269 
2270 /*
2271  * Do L2 address resolution for @sa_dst address. Stores found
2272  * address in @desten buffer. Copy of lle ln_flags can be also
2273  * saved in @pflags if @pflags is non-NULL.
2274  *
2275  * Heavy version.
2276  * Function assume that destination LLE does not exist,
2277  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2278  *
2279  * Set noinline to be dtrace-friendly
2280  */
2281 static __noinline int
2282 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2283     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2284     struct llentry **plle)
2285 {
2286 	struct llentry *lle = NULL, *lle_tmp;
2287 	struct in6_addr *psrc, src;
2288 	int send_ns, ll_len;
2289 	char *lladdr;
2290 
2291 	NET_EPOCH_ASSERT();
2292 
2293 	/*
2294 	 * Address resolution or Neighbor Unreachability Detection
2295 	 * for the next hop.
2296 	 * At this point, the destination of the packet must be a unicast
2297 	 * or an anycast address(i.e. not a multicast).
2298 	 */
2299 	if (lle == NULL) {
2300 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2301 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2302 			/*
2303 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2304 			 * the condition below is not very efficient.  But we believe
2305 			 * it is tolerable, because this should be a rare case.
2306 			 */
2307 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2308 			if (lle == NULL) {
2309 				char ip6buf[INET6_ADDRSTRLEN];
2310 				log(LOG_DEBUG,
2311 				    "nd6_output: can't allocate llinfo for %s "
2312 				    "(ln=%p)\n",
2313 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2314 				m_freem(m);
2315 				return (ENOBUFS);
2316 			}
2317 
2318 			IF_AFDATA_WLOCK(ifp);
2319 			LLE_WLOCK(lle);
2320 			/* Prefer any existing entry over newly-created one */
2321 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2322 			if (lle_tmp == NULL)
2323 				lltable_link_entry(LLTABLE6(ifp), lle);
2324 			IF_AFDATA_WUNLOCK(ifp);
2325 			if (lle_tmp != NULL) {
2326 				lltable_free_entry(LLTABLE6(ifp), lle);
2327 				lle = lle_tmp;
2328 				lle_tmp = NULL;
2329 			}
2330 		}
2331 	}
2332 	if (lle == NULL) {
2333 		m_freem(m);
2334 		return (ENOBUFS);
2335 	}
2336 
2337 	LLE_WLOCK_ASSERT(lle);
2338 
2339 	/*
2340 	 * The first time we send a packet to a neighbor whose entry is
2341 	 * STALE, we have to change the state to DELAY and a sets a timer to
2342 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2343 	 * neighbor unreachability detection on expiration.
2344 	 * (RFC 2461 7.3.3)
2345 	 */
2346 	if (lle->ln_state == ND6_LLINFO_STALE)
2347 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2348 
2349 	/*
2350 	 * If the neighbor cache entry has a state other than INCOMPLETE
2351 	 * (i.e. its link-layer address is already resolved), just
2352 	 * send the packet.
2353 	 */
2354 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2355 		if (flags & LLE_ADDRONLY) {
2356 			lladdr = lle->ll_addr;
2357 			ll_len = ifp->if_addrlen;
2358 		} else {
2359 			lladdr = lle->r_linkdata;
2360 			ll_len = lle->r_hdrlen;
2361 		}
2362 		bcopy(lladdr, desten, ll_len);
2363 		if (pflags != NULL)
2364 			*pflags = lle->la_flags;
2365 		if (plle) {
2366 			LLE_ADDREF(lle);
2367 			*plle = lle;
2368 		}
2369 		LLE_WUNLOCK(lle);
2370 		return (0);
2371 	}
2372 
2373 	/*
2374 	 * There is a neighbor cache entry, but no ethernet address
2375 	 * response yet.  Append this latest packet to the end of the
2376 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2377 	 * the oldest packet in the queue will be removed.
2378 	 */
2379 
2380 	if (lle->la_hold != NULL) {
2381 		struct mbuf *m_hold;
2382 		int i;
2383 
2384 		i = 0;
2385 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2386 			i++;
2387 			if (m_hold->m_nextpkt == NULL) {
2388 				m_hold->m_nextpkt = m;
2389 				break;
2390 			}
2391 		}
2392 		while (i >= V_nd6_maxqueuelen) {
2393 			m_hold = lle->la_hold;
2394 			lle->la_hold = lle->la_hold->m_nextpkt;
2395 			m_freem(m_hold);
2396 			i--;
2397 		}
2398 	} else {
2399 		lle->la_hold = m;
2400 	}
2401 
2402 	/*
2403 	 * If there has been no NS for the neighbor after entering the
2404 	 * INCOMPLETE state, send the first solicitation.
2405 	 * Note that for newly-created lle la_asked will be 0,
2406 	 * so we will transition from ND6_LLINFO_NOSTATE to
2407 	 * ND6_LLINFO_INCOMPLETE state here.
2408 	 */
2409 	psrc = NULL;
2410 	send_ns = 0;
2411 	if (lle->la_asked == 0) {
2412 		lle->la_asked++;
2413 		send_ns = 1;
2414 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2415 
2416 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2417 	}
2418 	LLE_WUNLOCK(lle);
2419 	if (send_ns != 0)
2420 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2421 
2422 	return (EWOULDBLOCK);
2423 }
2424 
2425 /*
2426  * Do L2 address resolution for @sa_dst address. Stores found
2427  * address in @desten buffer. Copy of lle ln_flags can be also
2428  * saved in @pflags if @pflags is non-NULL.
2429  *
2430  * Return values:
2431  * - 0 on success (address copied to buffer).
2432  * - EWOULDBLOCK (no local error, but address is still unresolved)
2433  * - other errors (alloc failure, etc)
2434  */
2435 int
2436 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2437     char *desten, uint32_t *pflags)
2438 {
2439 	int error;
2440 
2441 	flags |= LLE_ADDRONLY;
2442 	error = nd6_resolve_slow(ifp, flags, NULL,
2443 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2444 	return (error);
2445 }
2446 
2447 int
2448 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2449     struct sockaddr_in6 *dst)
2450 {
2451 	struct mbuf *m, *m_head;
2452 	int error = 0;
2453 
2454 	m_head = chain;
2455 
2456 	while (m_head) {
2457 		m = m_head;
2458 		m_head = m_head->m_nextpkt;
2459 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2460 	}
2461 
2462 	/*
2463 	 * XXX
2464 	 * note that intermediate errors are blindly ignored
2465 	 */
2466 	return (error);
2467 }
2468 
2469 static int
2470 nd6_need_cache(struct ifnet *ifp)
2471 {
2472 	/*
2473 	 * XXX: we currently do not make neighbor cache on any interface
2474 	 * other than Ethernet and GIF.
2475 	 *
2476 	 * RFC2893 says:
2477 	 * - unidirectional tunnels needs no ND
2478 	 */
2479 	switch (ifp->if_type) {
2480 	case IFT_ETHER:
2481 	case IFT_IEEE1394:
2482 	case IFT_L2VLAN:
2483 	case IFT_INFINIBAND:
2484 	case IFT_BRIDGE:
2485 	case IFT_PROPVIRTUAL:
2486 		return (1);
2487 	default:
2488 		return (0);
2489 	}
2490 }
2491 
2492 /*
2493  * Add pernament ND6 link-layer record for given
2494  * interface address.
2495  *
2496  * Very similar to IPv4 arp_ifinit(), but:
2497  * 1) IPv6 DAD is performed in different place
2498  * 2) It is called by IPv6 protocol stack in contrast to
2499  * arp_ifinit() which is typically called in SIOCSIFADDR
2500  * driver ioctl handler.
2501  *
2502  */
2503 int
2504 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2505 {
2506 	struct ifnet *ifp;
2507 	struct llentry *ln, *ln_tmp;
2508 	struct sockaddr *dst;
2509 
2510 	ifp = ia->ia_ifa.ifa_ifp;
2511 	if (nd6_need_cache(ifp) == 0)
2512 		return (0);
2513 
2514 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2515 	dst = (struct sockaddr *)&ia->ia_addr;
2516 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2517 	if (ln == NULL)
2518 		return (ENOBUFS);
2519 
2520 	IF_AFDATA_WLOCK(ifp);
2521 	LLE_WLOCK(ln);
2522 	/* Unlink any entry if exists */
2523 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2524 	if (ln_tmp != NULL)
2525 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2526 	lltable_link_entry(LLTABLE6(ifp), ln);
2527 	IF_AFDATA_WUNLOCK(ifp);
2528 
2529 	if (ln_tmp != NULL)
2530 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2531 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2532 
2533 	LLE_WUNLOCK(ln);
2534 	if (ln_tmp != NULL)
2535 		llentry_free(ln_tmp);
2536 
2537 	return (0);
2538 }
2539 
2540 /*
2541  * Removes either all lle entries for given @ia, or lle
2542  * corresponding to @ia address.
2543  */
2544 void
2545 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2546 {
2547 	struct sockaddr_in6 mask, addr;
2548 	struct sockaddr *saddr, *smask;
2549 	struct ifnet *ifp;
2550 
2551 	ifp = ia->ia_ifa.ifa_ifp;
2552 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2553 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2554 	saddr = (struct sockaddr *)&addr;
2555 	smask = (struct sockaddr *)&mask;
2556 
2557 	if (all != 0)
2558 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2559 	else
2560 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2561 }
2562 
2563 static void
2564 clear_llinfo_pqueue(struct llentry *ln)
2565 {
2566 	struct mbuf *m_hold, *m_hold_next;
2567 
2568 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2569 		m_hold_next = m_hold->m_nextpkt;
2570 		m_freem(m_hold);
2571 	}
2572 
2573 	ln->la_hold = NULL;
2574 }
2575 
2576 static int
2577 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2578 {
2579 	struct in6_prefix p;
2580 	struct sockaddr_in6 s6;
2581 	struct nd_prefix *pr;
2582 	struct nd_pfxrouter *pfr;
2583 	time_t maxexpire;
2584 	int error;
2585 	char ip6buf[INET6_ADDRSTRLEN];
2586 
2587 	if (req->newptr)
2588 		return (EPERM);
2589 
2590 	error = sysctl_wire_old_buffer(req, 0);
2591 	if (error != 0)
2592 		return (error);
2593 
2594 	bzero(&p, sizeof(p));
2595 	p.origin = PR_ORIG_RA;
2596 	bzero(&s6, sizeof(s6));
2597 	s6.sin6_family = AF_INET6;
2598 	s6.sin6_len = sizeof(s6);
2599 
2600 	ND6_RLOCK();
2601 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2602 		p.prefix = pr->ndpr_prefix;
2603 		if (sa6_recoverscope(&p.prefix)) {
2604 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2605 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2606 			/* XXX: press on... */
2607 		}
2608 		p.raflags = pr->ndpr_raf;
2609 		p.prefixlen = pr->ndpr_plen;
2610 		p.vltime = pr->ndpr_vltime;
2611 		p.pltime = pr->ndpr_pltime;
2612 		p.if_index = pr->ndpr_ifp->if_index;
2613 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2614 			p.expire = 0;
2615 		else {
2616 			/* XXX: we assume time_t is signed. */
2617 			maxexpire = (-1) &
2618 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2619 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2620 				p.expire = pr->ndpr_lastupdate +
2621 				    pr->ndpr_vltime +
2622 				    (time_second - time_uptime);
2623 			else
2624 				p.expire = maxexpire;
2625 		}
2626 		p.refcnt = pr->ndpr_addrcnt;
2627 		p.flags = pr->ndpr_stateflags;
2628 		p.advrtrs = 0;
2629 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2630 			p.advrtrs++;
2631 		error = SYSCTL_OUT(req, &p, sizeof(p));
2632 		if (error != 0)
2633 			break;
2634 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2635 			s6.sin6_addr = pfr->router->rtaddr;
2636 			if (sa6_recoverscope(&s6))
2637 				log(LOG_ERR,
2638 				    "scope error in prefix list (%s)\n",
2639 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2640 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2641 			if (error != 0)
2642 				goto out;
2643 		}
2644 	}
2645 out:
2646 	ND6_RUNLOCK();
2647 	return (error);
2648 }
2649 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2650 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2651 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2652 	"NDP prefix list");
2653 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2654 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2655 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2656 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2657