xref: /freebsd/sys/netinet6/nd6.c (revision b24e6ac8b78ebf628872c821b7eb4f7ed008498c)
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/nhop.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_kdtrace.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #include <netinet6/in6_ifattach.h>
78 #include <netinet/icmp6.h>
79 #include <netinet6/send.h>
80 
81 #include <sys/limits.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
86 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
87 
88 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
89 
90 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
91 
92 /* timer values */
93 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
94 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
95 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
96 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
97 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
98 					 * local traffic */
99 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
100 					 * collection timer */
101 
102 /* preventing too many loops in ND option parsing */
103 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
104 
105 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
106 					 * layer hints */
107 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
108 					 * ND entries */
109 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
110 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
111 
112 #ifdef ND6_DEBUG
113 VNET_DEFINE(int, nd6_debug) = 1;
114 #else
115 VNET_DEFINE(int, nd6_debug) = 0;
116 #endif
117 
118 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
119 
120 VNET_DEFINE(struct nd_prhead, nd_prefix);
121 VNET_DEFINE(struct rwlock, nd6_lock);
122 VNET_DEFINE(uint64_t, nd6_list_genid);
123 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
124 
125 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
126 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
127 
128 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
129 
130 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
131 	struct ifnet *);
132 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
133 static void nd6_slowtimo(void *);
134 static int regen_tmpaddr(struct in6_ifaddr *);
135 static void nd6_free(struct llentry **, int);
136 static void nd6_free_redirect(const struct llentry *);
137 static void nd6_llinfo_timer(void *);
138 static void nd6_llinfo_settimer_locked(struct llentry *, long);
139 static void clear_llinfo_pqueue(struct llentry *);
140 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
141 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
142     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
143 static int nd6_need_cache(struct ifnet *);
144 
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 
212 	lltable_update_ifaddr(LLTABLE6(ifp));
213 }
214 
215 void
216 nd6_init(void)
217 {
218 
219 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
220 	rw_init(&V_nd6_lock, "nd6 list");
221 
222 	LIST_INIT(&V_nd_prefix);
223 	nd6_defrouter_init();
224 
225 	/* Start timers. */
226 	callout_init(&V_nd6_slowtimo_ch, 0);
227 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
228 	    nd6_slowtimo, curvnet);
229 
230 	callout_init(&V_nd6_timer_ch, 0);
231 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
232 
233 	nd6_dad_init();
234 	if (IS_DEFAULT_VNET(curvnet)) {
235 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
236 		    NULL, EVENTHANDLER_PRI_ANY);
237 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
238 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
239 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
240 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
241 	}
242 }
243 
244 #ifdef VIMAGE
245 void
246 nd6_destroy()
247 {
248 
249 	callout_drain(&V_nd6_slowtimo_ch);
250 	callout_drain(&V_nd6_timer_ch);
251 	if (IS_DEFAULT_VNET(curvnet)) {
252 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
253 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
254 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
255 	}
256 	rw_destroy(&V_nd6_lock);
257 	mtx_destroy(&V_nd6_onlink_mtx);
258 }
259 #endif
260 
261 struct nd_ifinfo *
262 nd6_ifattach(struct ifnet *ifp)
263 {
264 	struct nd_ifinfo *nd;
265 
266 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
267 	nd->initialized = 1;
268 
269 	nd->chlim = IPV6_DEFHLIM;
270 	nd->basereachable = REACHABLE_TIME;
271 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
272 	nd->retrans = RETRANS_TIMER;
273 
274 	nd->flags = ND6_IFF_PERFORMNUD;
275 
276 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
277 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
278 	 * default regardless of the V_ip6_auto_linklocal configuration to
279 	 * give a reasonable default behavior.
280 	 */
281 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
282 	    (ifp->if_flags & IFF_LOOPBACK))
283 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
284 	/*
285 	 * A loopback interface does not need to accept RTADV.
286 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
287 	 * default regardless of the V_ip6_accept_rtadv configuration to
288 	 * prevent the interface from accepting RA messages arrived
289 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
290 	 */
291 	if (V_ip6_accept_rtadv &&
292 	    !(ifp->if_flags & IFF_LOOPBACK) &&
293 	    (ifp->if_type != IFT_BRIDGE))
294 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
295 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
296 		nd->flags |= ND6_IFF_NO_RADR;
297 
298 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
299 	nd6_setmtu0(ifp, nd);
300 
301 	return nd;
302 }
303 
304 void
305 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
306 {
307 	struct epoch_tracker et;
308 	struct ifaddr *ifa, *next;
309 
310 	NET_EPOCH_ENTER(et);
311 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
312 		if (ifa->ifa_addr->sa_family != AF_INET6)
313 			continue;
314 
315 		/* stop DAD processing */
316 		nd6_dad_stop(ifa);
317 	}
318 	NET_EPOCH_EXIT(et);
319 
320 	free(nd, M_IP6NDP);
321 }
322 
323 /*
324  * Reset ND level link MTU. This function is called when the physical MTU
325  * changes, which means we might have to adjust the ND level MTU.
326  */
327 void
328 nd6_setmtu(struct ifnet *ifp)
329 {
330 	if (ifp->if_afdata[AF_INET6] == NULL)
331 		return;
332 
333 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
334 }
335 
336 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
337 void
338 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
339 {
340 	u_int32_t omaxmtu;
341 
342 	omaxmtu = ndi->maxmtu;
343 	ndi->maxmtu = ifp->if_mtu;
344 
345 	/*
346 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
347 	 * undesirable situation.  We thus notify the operator of the change
348 	 * explicitly.  The check for omaxmtu is necessary to restrict the
349 	 * log to the case of changing the MTU, not initializing it.
350 	 */
351 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
352 		log(LOG_NOTICE, "nd6_setmtu0: "
353 		    "new link MTU on %s (%lu) is too small for IPv6\n",
354 		    if_name(ifp), (unsigned long)ndi->maxmtu);
355 	}
356 
357 	if (ndi->maxmtu > V_in6_maxmtu)
358 		in6_setmaxmtu(); /* check all interfaces just in case */
359 
360 }
361 
362 void
363 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
364 {
365 
366 	bzero(ndopts, sizeof(*ndopts));
367 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
368 	ndopts->nd_opts_last
369 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
370 
371 	if (icmp6len == 0) {
372 		ndopts->nd_opts_done = 1;
373 		ndopts->nd_opts_search = NULL;
374 	}
375 }
376 
377 /*
378  * Take one ND option.
379  */
380 struct nd_opt_hdr *
381 nd6_option(union nd_opts *ndopts)
382 {
383 	struct nd_opt_hdr *nd_opt;
384 	int olen;
385 
386 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
387 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
388 	    __func__));
389 	if (ndopts->nd_opts_search == NULL)
390 		return NULL;
391 	if (ndopts->nd_opts_done)
392 		return NULL;
393 
394 	nd_opt = ndopts->nd_opts_search;
395 
396 	/* make sure nd_opt_len is inside the buffer */
397 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
398 		bzero(ndopts, sizeof(*ndopts));
399 		return NULL;
400 	}
401 
402 	olen = nd_opt->nd_opt_len << 3;
403 	if (olen == 0) {
404 		/*
405 		 * Message validation requires that all included
406 		 * options have a length that is greater than zero.
407 		 */
408 		bzero(ndopts, sizeof(*ndopts));
409 		return NULL;
410 	}
411 
412 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
413 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
414 		/* option overruns the end of buffer, invalid */
415 		bzero(ndopts, sizeof(*ndopts));
416 		return NULL;
417 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
418 		/* reached the end of options chain */
419 		ndopts->nd_opts_done = 1;
420 		ndopts->nd_opts_search = NULL;
421 	}
422 	return nd_opt;
423 }
424 
425 /*
426  * Parse multiple ND options.
427  * This function is much easier to use, for ND routines that do not need
428  * multiple options of the same type.
429  */
430 int
431 nd6_options(union nd_opts *ndopts)
432 {
433 	struct nd_opt_hdr *nd_opt;
434 	int i = 0;
435 
436 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
437 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
438 	    __func__));
439 	if (ndopts->nd_opts_search == NULL)
440 		return 0;
441 
442 	while (1) {
443 		nd_opt = nd6_option(ndopts);
444 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
445 			/*
446 			 * Message validation requires that all included
447 			 * options have a length that is greater than zero.
448 			 */
449 			ICMP6STAT_INC(icp6s_nd_badopt);
450 			bzero(ndopts, sizeof(*ndopts));
451 			return -1;
452 		}
453 
454 		if (nd_opt == NULL)
455 			goto skip1;
456 
457 		switch (nd_opt->nd_opt_type) {
458 		case ND_OPT_SOURCE_LINKADDR:
459 		case ND_OPT_TARGET_LINKADDR:
460 		case ND_OPT_MTU:
461 		case ND_OPT_REDIRECTED_HEADER:
462 		case ND_OPT_NONCE:
463 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
464 				nd6log((LOG_INFO,
465 				    "duplicated ND6 option found (type=%d)\n",
466 				    nd_opt->nd_opt_type));
467 				/* XXX bark? */
468 			} else {
469 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
470 					= nd_opt;
471 			}
472 			break;
473 		case ND_OPT_PREFIX_INFORMATION:
474 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
475 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
476 					= nd_opt;
477 			}
478 			ndopts->nd_opts_pi_end =
479 				(struct nd_opt_prefix_info *)nd_opt;
480 			break;
481 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
482 		case ND_OPT_RDNSS:	/* RFC 6106 */
483 		case ND_OPT_DNSSL:	/* RFC 6106 */
484 			/*
485 			 * Silently ignore options we know and do not care about
486 			 * in the kernel.
487 			 */
488 			break;
489 		default:
490 			/*
491 			 * Unknown options must be silently ignored,
492 			 * to accommodate future extension to the protocol.
493 			 */
494 			nd6log((LOG_DEBUG,
495 			    "nd6_options: unsupported option %d - "
496 			    "option ignored\n", nd_opt->nd_opt_type));
497 		}
498 
499 skip1:
500 		i++;
501 		if (i > V_nd6_maxndopt) {
502 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
503 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
504 			break;
505 		}
506 
507 		if (ndopts->nd_opts_done)
508 			break;
509 	}
510 
511 	return 0;
512 }
513 
514 /*
515  * ND6 timer routine to handle ND6 entries
516  */
517 static void
518 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
519 {
520 	int canceled;
521 
522 	LLE_WLOCK_ASSERT(ln);
523 
524 	if (tick < 0) {
525 		ln->la_expire = 0;
526 		ln->ln_ntick = 0;
527 		canceled = callout_stop(&ln->lle_timer);
528 	} else {
529 		ln->la_expire = time_uptime + tick / hz;
530 		LLE_ADDREF(ln);
531 		if (tick > INT_MAX) {
532 			ln->ln_ntick = tick - INT_MAX;
533 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
534 			    nd6_llinfo_timer, ln);
535 		} else {
536 			ln->ln_ntick = 0;
537 			canceled = callout_reset(&ln->lle_timer, tick,
538 			    nd6_llinfo_timer, ln);
539 		}
540 	}
541 	if (canceled > 0)
542 		LLE_REMREF(ln);
543 }
544 
545 /*
546  * Gets source address of the first packet in hold queue
547  * and stores it in @src.
548  * Returns pointer to @src (if hold queue is not empty) or NULL.
549  *
550  * Set noinline to be dtrace-friendly
551  */
552 static __noinline struct in6_addr *
553 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
554 {
555 	struct ip6_hdr hdr;
556 	struct mbuf *m;
557 
558 	if (ln->la_hold == NULL)
559 		return (NULL);
560 
561 	/*
562 	 * assume every packet in la_hold has the same IP header
563 	 */
564 	m = ln->la_hold;
565 	if (sizeof(hdr) > m->m_len)
566 		return (NULL);
567 
568 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
569 	*src = hdr.ip6_src;
570 
571 	return (src);
572 }
573 
574 /*
575  * Checks if we need to switch from STALE state.
576  *
577  * RFC 4861 requires switching from STALE to DELAY state
578  * on first packet matching entry, waiting V_nd6_delay and
579  * transition to PROBE state (if upper layer confirmation was
580  * not received).
581  *
582  * This code performs a bit differently:
583  * On packet hit we don't change state (but desired state
584  * can be guessed by control plane). However, after V_nd6_delay
585  * seconds code will transition to PROBE state (so DELAY state
586  * is kinda skipped in most situations).
587  *
588  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
589  * we perform the following upon entering STALE state:
590  *
591  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
592  * if packet was transmitted at the start of given interval, we
593  * would be able to switch to PROBE state in V_nd6_delay seconds
594  * as user expects.
595  *
596  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
597  * lle in STALE state (remaining timer value stored in lle_remtime).
598  *
599  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
600  * seconds ago.
601  *
602  * Returns non-zero value if the entry is still STALE (storing
603  * the next timer interval in @pdelay).
604  *
605  * Returns zero value if original timer expired or we need to switch to
606  * PROBE (store that in @do_switch variable).
607  */
608 static int
609 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
610 {
611 	int nd_delay, nd_gctimer, r_skip_req;
612 	time_t lle_hittime;
613 	long delay;
614 
615 	*do_switch = 0;
616 	nd_gctimer = V_nd6_gctimer;
617 	nd_delay = V_nd6_delay;
618 
619 	LLE_REQ_LOCK(lle);
620 	r_skip_req = lle->r_skip_req;
621 	lle_hittime = lle->lle_hittime;
622 	LLE_REQ_UNLOCK(lle);
623 
624 	if (r_skip_req > 0) {
625 
626 		/*
627 		 * Nonzero r_skip_req value was set upon entering
628 		 * STALE state. Since value was not changed, no
629 		 * packets were passed using this lle. Ask for
630 		 * timer reschedule and keep STALE state.
631 		 */
632 		delay = (long)(MIN(nd_gctimer, nd_delay));
633 		delay *= hz;
634 		if (lle->lle_remtime > delay)
635 			lle->lle_remtime -= delay;
636 		else {
637 			delay = lle->lle_remtime;
638 			lle->lle_remtime = 0;
639 		}
640 
641 		if (delay == 0) {
642 
643 			/*
644 			 * The original ng6_gctime timeout ended,
645 			 * no more rescheduling.
646 			 */
647 			return (0);
648 		}
649 
650 		*pdelay = delay;
651 		return (1);
652 	}
653 
654 	/*
655 	 * Packet received. Verify timestamp
656 	 */
657 	delay = (long)(time_uptime - lle_hittime);
658 	if (delay < nd_delay) {
659 
660 		/*
661 		 * V_nd6_delay still not passed since the first
662 		 * hit in STALE state.
663 		 * Reshedule timer and return.
664 		 */
665 		*pdelay = (long)(nd_delay - delay) * hz;
666 		return (1);
667 	}
668 
669 	/* Request switching to probe */
670 	*do_switch = 1;
671 	return (0);
672 }
673 
674 
675 /*
676  * Switch @lle state to new state optionally arming timers.
677  *
678  * Set noinline to be dtrace-friendly
679  */
680 __noinline void
681 nd6_llinfo_setstate(struct llentry *lle, int newstate)
682 {
683 	struct ifnet *ifp;
684 	int nd_gctimer, nd_delay;
685 	long delay, remtime;
686 
687 	delay = 0;
688 	remtime = 0;
689 
690 	switch (newstate) {
691 	case ND6_LLINFO_INCOMPLETE:
692 		ifp = lle->lle_tbl->llt_ifp;
693 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
694 		break;
695 	case ND6_LLINFO_REACHABLE:
696 		if (!ND6_LLINFO_PERMANENT(lle)) {
697 			ifp = lle->lle_tbl->llt_ifp;
698 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
699 		}
700 		break;
701 	case ND6_LLINFO_STALE:
702 
703 		/*
704 		 * Notify fast path that we want to know if any packet
705 		 * is transmitted by setting r_skip_req.
706 		 */
707 		LLE_REQ_LOCK(lle);
708 		lle->r_skip_req = 1;
709 		LLE_REQ_UNLOCK(lle);
710 		nd_delay = V_nd6_delay;
711 		nd_gctimer = V_nd6_gctimer;
712 
713 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
714 		remtime = (long)nd_gctimer * hz - delay;
715 		break;
716 	case ND6_LLINFO_DELAY:
717 		lle->la_asked = 0;
718 		delay = (long)V_nd6_delay * hz;
719 		break;
720 	}
721 
722 	if (delay > 0)
723 		nd6_llinfo_settimer_locked(lle, delay);
724 
725 	lle->lle_remtime = remtime;
726 	lle->ln_state = newstate;
727 }
728 
729 /*
730  * Timer-dependent part of nd state machine.
731  *
732  * Set noinline to be dtrace-friendly
733  */
734 static __noinline void
735 nd6_llinfo_timer(void *arg)
736 {
737 	struct epoch_tracker et;
738 	struct llentry *ln;
739 	struct in6_addr *dst, *pdst, *psrc, src;
740 	struct ifnet *ifp;
741 	struct nd_ifinfo *ndi;
742 	int do_switch, send_ns;
743 	long delay;
744 
745 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
746 	ln = (struct llentry *)arg;
747 	ifp = lltable_get_ifp(ln->lle_tbl);
748 	CURVNET_SET(ifp->if_vnet);
749 
750 	ND6_RLOCK();
751 	LLE_WLOCK(ln);
752 	if (callout_pending(&ln->lle_timer)) {
753 		/*
754 		 * Here we are a bit odd here in the treatment of
755 		 * active/pending. If the pending bit is set, it got
756 		 * rescheduled before I ran. The active
757 		 * bit we ignore, since if it was stopped
758 		 * in ll_tablefree() and was currently running
759 		 * it would have return 0 so the code would
760 		 * not have deleted it since the callout could
761 		 * not be stopped so we want to go through
762 		 * with the delete here now. If the callout
763 		 * was restarted, the pending bit will be back on and
764 		 * we just want to bail since the callout_reset would
765 		 * return 1 and our reference would have been removed
766 		 * by nd6_llinfo_settimer_locked above since canceled
767 		 * would have been 1.
768 		 */
769 		LLE_WUNLOCK(ln);
770 		ND6_RUNLOCK();
771 		CURVNET_RESTORE();
772 		return;
773 	}
774 	NET_EPOCH_ENTER(et);
775 	ndi = ND_IFINFO(ifp);
776 	send_ns = 0;
777 	dst = &ln->r_l3addr.addr6;
778 	pdst = dst;
779 
780 	if (ln->ln_ntick > 0) {
781 		if (ln->ln_ntick > INT_MAX) {
782 			ln->ln_ntick -= INT_MAX;
783 			nd6_llinfo_settimer_locked(ln, INT_MAX);
784 		} else {
785 			ln->ln_ntick = 0;
786 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
787 		}
788 		goto done;
789 	}
790 
791 	if (ln->la_flags & LLE_STATIC) {
792 		goto done;
793 	}
794 
795 	if (ln->la_flags & LLE_DELETED) {
796 		nd6_free(&ln, 0);
797 		goto done;
798 	}
799 
800 	switch (ln->ln_state) {
801 	case ND6_LLINFO_INCOMPLETE:
802 		if (ln->la_asked < V_nd6_mmaxtries) {
803 			ln->la_asked++;
804 			send_ns = 1;
805 			/* Send NS to multicast address */
806 			pdst = NULL;
807 		} else {
808 			struct mbuf *m = ln->la_hold;
809 			if (m) {
810 				struct mbuf *m0;
811 
812 				/*
813 				 * assuming every packet in la_hold has the
814 				 * same IP header.  Send error after unlock.
815 				 */
816 				m0 = m->m_nextpkt;
817 				m->m_nextpkt = NULL;
818 				ln->la_hold = m0;
819 				clear_llinfo_pqueue(ln);
820 			}
821 			nd6_free(&ln, 0);
822 			if (m != NULL)
823 				icmp6_error2(m, ICMP6_DST_UNREACH,
824 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
825 		}
826 		break;
827 	case ND6_LLINFO_REACHABLE:
828 		if (!ND6_LLINFO_PERMANENT(ln))
829 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
830 		break;
831 
832 	case ND6_LLINFO_STALE:
833 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
834 
835 			/*
836 			 * No packet has used this entry and GC timeout
837 			 * has not been passed. Reshedule timer and
838 			 * return.
839 			 */
840 			nd6_llinfo_settimer_locked(ln, delay);
841 			break;
842 		}
843 
844 		if (do_switch == 0) {
845 
846 			/*
847 			 * GC timer has ended and entry hasn't been used.
848 			 * Run Garbage collector (RFC 4861, 5.3)
849 			 */
850 			if (!ND6_LLINFO_PERMANENT(ln))
851 				nd6_free(&ln, 1);
852 			break;
853 		}
854 
855 		/* Entry has been used AND delay timer has ended. */
856 
857 		/* FALLTHROUGH */
858 
859 	case ND6_LLINFO_DELAY:
860 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
861 			/* We need NUD */
862 			ln->la_asked = 1;
863 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
864 			send_ns = 1;
865 		} else
866 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
867 		break;
868 	case ND6_LLINFO_PROBE:
869 		if (ln->la_asked < V_nd6_umaxtries) {
870 			ln->la_asked++;
871 			send_ns = 1;
872 		} else {
873 			nd6_free(&ln, 0);
874 		}
875 		break;
876 	default:
877 		panic("%s: paths in a dark night can be confusing: %d",
878 		    __func__, ln->ln_state);
879 	}
880 done:
881 	if (ln != NULL)
882 		ND6_RUNLOCK();
883 	if (send_ns != 0) {
884 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
885 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
886 		LLE_FREE_LOCKED(ln);
887 		ln = NULL;
888 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
889 	}
890 
891 	if (ln != NULL)
892 		LLE_FREE_LOCKED(ln);
893 	NET_EPOCH_EXIT(et);
894 	CURVNET_RESTORE();
895 }
896 
897 
898 /*
899  * ND6 timer routine to expire default route list and prefix list
900  */
901 void
902 nd6_timer(void *arg)
903 {
904 	CURVNET_SET((struct vnet *) arg);
905 	struct epoch_tracker et;
906 	struct nd_prhead prl;
907 	struct nd_prefix *pr, *npr;
908 	struct ifnet *ifp;
909 	struct in6_ifaddr *ia6, *nia6;
910 	uint64_t genid;
911 
912 	LIST_INIT(&prl);
913 
914 	NET_EPOCH_ENTER(et);
915 	nd6_defrouter_timer();
916 
917 	/*
918 	 * expire interface addresses.
919 	 * in the past the loop was inside prefix expiry processing.
920 	 * However, from a stricter speci-confrmance standpoint, we should
921 	 * rather separate address lifetimes and prefix lifetimes.
922 	 *
923 	 * XXXRW: in6_ifaddrhead locking.
924 	 */
925   addrloop:
926 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
927 		/* check address lifetime */
928 		if (IFA6_IS_INVALID(ia6)) {
929 			int regen = 0;
930 
931 			/*
932 			 * If the expiring address is temporary, try
933 			 * regenerating a new one.  This would be useful when
934 			 * we suspended a laptop PC, then turned it on after a
935 			 * period that could invalidate all temporary
936 			 * addresses.  Although we may have to restart the
937 			 * loop (see below), it must be after purging the
938 			 * address.  Otherwise, we'd see an infinite loop of
939 			 * regeneration.
940 			 */
941 			if (V_ip6_use_tempaddr &&
942 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
943 				if (regen_tmpaddr(ia6) == 0)
944 					regen = 1;
945 			}
946 
947 			in6_purgeaddr(&ia6->ia_ifa);
948 
949 			if (regen)
950 				goto addrloop; /* XXX: see below */
951 		} else if (IFA6_IS_DEPRECATED(ia6)) {
952 			int oldflags = ia6->ia6_flags;
953 
954 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
955 
956 			/*
957 			 * If a temporary address has just become deprecated,
958 			 * regenerate a new one if possible.
959 			 */
960 			if (V_ip6_use_tempaddr &&
961 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
962 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
963 
964 				if (regen_tmpaddr(ia6) == 0) {
965 					/*
966 					 * A new temporary address is
967 					 * generated.
968 					 * XXX: this means the address chain
969 					 * has changed while we are still in
970 					 * the loop.  Although the change
971 					 * would not cause disaster (because
972 					 * it's not a deletion, but an
973 					 * addition,) we'd rather restart the
974 					 * loop just for safety.  Or does this
975 					 * significantly reduce performance??
976 					 */
977 					goto addrloop;
978 				}
979 			}
980 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
981 			/*
982 			 * Schedule DAD for a tentative address.  This happens
983 			 * if the interface was down or not running
984 			 * when the address was configured.
985 			 */
986 			int delay;
987 
988 			delay = arc4random() %
989 			    (MAX_RTR_SOLICITATION_DELAY * hz);
990 			nd6_dad_start((struct ifaddr *)ia6, delay);
991 		} else {
992 			/*
993 			 * Check status of the interface.  If it is down,
994 			 * mark the address as tentative for future DAD.
995 			 */
996 			ifp = ia6->ia_ifp;
997 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
998 			    ((ifp->if_flags & IFF_UP) == 0 ||
999 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1000 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1001 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1002 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1003 			}
1004 
1005 			/*
1006 			 * A new RA might have made a deprecated address
1007 			 * preferred.
1008 			 */
1009 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1010 		}
1011 	}
1012 	NET_EPOCH_EXIT(et);
1013 
1014 	ND6_WLOCK();
1015 restart:
1016 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1017 		/*
1018 		 * Expire prefixes. Since the pltime is only used for
1019 		 * autoconfigured addresses, pltime processing for prefixes is
1020 		 * not necessary.
1021 		 *
1022 		 * Only unlink after all derived addresses have expired. This
1023 		 * may not occur until two hours after the prefix has expired
1024 		 * per RFC 4862. If the prefix expires before its derived
1025 		 * addresses, mark it off-link. This will be done automatically
1026 		 * after unlinking if no address references remain.
1027 		 */
1028 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1029 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1030 			continue;
1031 
1032 		if (pr->ndpr_addrcnt == 0) {
1033 			nd6_prefix_unlink(pr, &prl);
1034 			continue;
1035 		}
1036 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1037 			genid = V_nd6_list_genid;
1038 			nd6_prefix_ref(pr);
1039 			ND6_WUNLOCK();
1040 			ND6_ONLINK_LOCK();
1041 			(void)nd6_prefix_offlink(pr);
1042 			ND6_ONLINK_UNLOCK();
1043 			ND6_WLOCK();
1044 			nd6_prefix_rele(pr);
1045 			if (genid != V_nd6_list_genid)
1046 				goto restart;
1047 		}
1048 	}
1049 	ND6_WUNLOCK();
1050 
1051 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1052 		LIST_REMOVE(pr, ndpr_entry);
1053 		nd6_prefix_del(pr);
1054 	}
1055 
1056 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1057 	    nd6_timer, curvnet);
1058 
1059 	CURVNET_RESTORE();
1060 }
1061 
1062 /*
1063  * ia6 - deprecated/invalidated temporary address
1064  */
1065 static int
1066 regen_tmpaddr(struct in6_ifaddr *ia6)
1067 {
1068 	struct ifaddr *ifa;
1069 	struct ifnet *ifp;
1070 	struct in6_ifaddr *public_ifa6 = NULL;
1071 
1072 	NET_EPOCH_ASSERT();
1073 
1074 	ifp = ia6->ia_ifa.ifa_ifp;
1075 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1076 		struct in6_ifaddr *it6;
1077 
1078 		if (ifa->ifa_addr->sa_family != AF_INET6)
1079 			continue;
1080 
1081 		it6 = (struct in6_ifaddr *)ifa;
1082 
1083 		/* ignore no autoconf addresses. */
1084 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1085 			continue;
1086 
1087 		/* ignore autoconf addresses with different prefixes. */
1088 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1089 			continue;
1090 
1091 		/*
1092 		 * Now we are looking at an autoconf address with the same
1093 		 * prefix as ours.  If the address is temporary and is still
1094 		 * preferred, do not create another one.  It would be rare, but
1095 		 * could happen, for example, when we resume a laptop PC after
1096 		 * a long period.
1097 		 */
1098 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1099 		    !IFA6_IS_DEPRECATED(it6)) {
1100 			public_ifa6 = NULL;
1101 			break;
1102 		}
1103 
1104 		/*
1105 		 * This is a public autoconf address that has the same prefix
1106 		 * as ours.  If it is preferred, keep it.  We can't break the
1107 		 * loop here, because there may be a still-preferred temporary
1108 		 * address with the prefix.
1109 		 */
1110 		if (!IFA6_IS_DEPRECATED(it6))
1111 			public_ifa6 = it6;
1112 	}
1113 	if (public_ifa6 != NULL)
1114 		ifa_ref(&public_ifa6->ia_ifa);
1115 
1116 	if (public_ifa6 != NULL) {
1117 		int e;
1118 
1119 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1120 			ifa_free(&public_ifa6->ia_ifa);
1121 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1122 			    " tmp addr,errno=%d\n", e);
1123 			return (-1);
1124 		}
1125 		ifa_free(&public_ifa6->ia_ifa);
1126 		return (0);
1127 	}
1128 
1129 	return (-1);
1130 }
1131 
1132 /*
1133  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1134  * cache entries are freed in in6_domifdetach().
1135  */
1136 void
1137 nd6_purge(struct ifnet *ifp)
1138 {
1139 	struct nd_prhead prl;
1140 	struct nd_prefix *pr, *npr;
1141 
1142 	LIST_INIT(&prl);
1143 
1144 	/* Purge default router list entries toward ifp. */
1145 	nd6_defrouter_purge(ifp);
1146 
1147 	ND6_WLOCK();
1148 	/*
1149 	 * Remove prefixes on ifp. We should have already removed addresses on
1150 	 * this interface, so no addresses should be referencing these prefixes.
1151 	 */
1152 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1153 		if (pr->ndpr_ifp == ifp)
1154 			nd6_prefix_unlink(pr, &prl);
1155 	}
1156 	ND6_WUNLOCK();
1157 
1158 	/* Delete the unlinked prefix objects. */
1159 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1160 		LIST_REMOVE(pr, ndpr_entry);
1161 		nd6_prefix_del(pr);
1162 	}
1163 
1164 	/* cancel default outgoing interface setting */
1165 	if (V_nd6_defifindex == ifp->if_index)
1166 		nd6_setdefaultiface(0);
1167 
1168 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1169 		/* Refresh default router list. */
1170 		defrouter_select_fib(ifp->if_fib);
1171 	}
1172 }
1173 
1174 /*
1175  * the caller acquires and releases the lock on the lltbls
1176  * Returns the llentry locked
1177  */
1178 struct llentry *
1179 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1180 {
1181 	struct sockaddr_in6 sin6;
1182 	struct llentry *ln;
1183 
1184 	bzero(&sin6, sizeof(sin6));
1185 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1186 	sin6.sin6_family = AF_INET6;
1187 	sin6.sin6_addr = *addr6;
1188 
1189 	IF_AFDATA_LOCK_ASSERT(ifp);
1190 
1191 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1192 
1193 	return (ln);
1194 }
1195 
1196 static struct llentry *
1197 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1198 {
1199 	struct sockaddr_in6 sin6;
1200 	struct llentry *ln;
1201 
1202 	bzero(&sin6, sizeof(sin6));
1203 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1204 	sin6.sin6_family = AF_INET6;
1205 	sin6.sin6_addr = *addr6;
1206 
1207 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1208 	if (ln != NULL)
1209 		ln->ln_state = ND6_LLINFO_NOSTATE;
1210 
1211 	return (ln);
1212 }
1213 
1214 /*
1215  * Test whether a given IPv6 address is a neighbor or not, ignoring
1216  * the actual neighbor cache.  The neighbor cache is ignored in order
1217  * to not reenter the routing code from within itself.
1218  */
1219 static int
1220 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1221 {
1222 	struct nd_prefix *pr;
1223 	struct ifaddr *ifa;
1224 	struct rt_addrinfo info;
1225 	struct sockaddr_in6 rt_key;
1226 	const struct sockaddr *dst6;
1227 	uint64_t genid;
1228 	int error, fibnum;
1229 
1230 	/*
1231 	 * A link-local address is always a neighbor.
1232 	 * XXX: a link does not necessarily specify a single interface.
1233 	 */
1234 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1235 		struct sockaddr_in6 sin6_copy;
1236 		u_int32_t zone;
1237 
1238 		/*
1239 		 * We need sin6_copy since sa6_recoverscope() may modify the
1240 		 * content (XXX).
1241 		 */
1242 		sin6_copy = *addr;
1243 		if (sa6_recoverscope(&sin6_copy))
1244 			return (0); /* XXX: should be impossible */
1245 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1246 			return (0);
1247 		if (sin6_copy.sin6_scope_id == zone)
1248 			return (1);
1249 		else
1250 			return (0);
1251 	}
1252 
1253 	bzero(&rt_key, sizeof(rt_key));
1254 	bzero(&info, sizeof(info));
1255 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1256 
1257 	/*
1258 	 * If the address matches one of our addresses,
1259 	 * it should be a neighbor.
1260 	 * If the address matches one of our on-link prefixes, it should be a
1261 	 * neighbor.
1262 	 */
1263 	ND6_RLOCK();
1264 restart:
1265 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1266 		if (pr->ndpr_ifp != ifp)
1267 			continue;
1268 
1269 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1270 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1271 
1272 			/*
1273 			 * We only need to check all FIBs if add_addr_allfibs
1274 			 * is unset. If set, checking any FIB will suffice.
1275 			 */
1276 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1277 			for (; fibnum < rt_numfibs; fibnum++) {
1278 				genid = V_nd6_list_genid;
1279 				ND6_RUNLOCK();
1280 
1281 				/*
1282 				 * Restore length field before
1283 				 * retrying lookup
1284 				 */
1285 				rt_key.sin6_len = sizeof(rt_key);
1286 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1287 						        &info);
1288 
1289 				ND6_RLOCK();
1290 				if (genid != V_nd6_list_genid)
1291 					goto restart;
1292 				if (error == 0)
1293 					break;
1294 			}
1295 			if (error != 0)
1296 				continue;
1297 
1298 			/*
1299 			 * This is the case where multiple interfaces
1300 			 * have the same prefix, but only one is installed
1301 			 * into the routing table and that prefix entry
1302 			 * is not the one being examined here. In the case
1303 			 * where RADIX_MPATH is enabled, multiple route
1304 			 * entries (of the same rt_key value) will be
1305 			 * installed because the interface addresses all
1306 			 * differ.
1307 			 */
1308 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1309 			    &rt_key.sin6_addr))
1310 				continue;
1311 		}
1312 
1313 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1314 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1315 			ND6_RUNLOCK();
1316 			return (1);
1317 		}
1318 	}
1319 	ND6_RUNLOCK();
1320 
1321 	/*
1322 	 * If the address is assigned on the node of the other side of
1323 	 * a p2p interface, the address should be a neighbor.
1324 	 */
1325 	if (ifp->if_flags & IFF_POINTOPOINT) {
1326 		struct epoch_tracker et;
1327 
1328 		NET_EPOCH_ENTER(et);
1329 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1330 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1331 				continue;
1332 			if (ifa->ifa_dstaddr != NULL &&
1333 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1334 				NET_EPOCH_EXIT(et);
1335 				return 1;
1336 			}
1337 		}
1338 		NET_EPOCH_EXIT(et);
1339 	}
1340 
1341 	/*
1342 	 * If the default router list is empty, all addresses are regarded
1343 	 * as on-link, and thus, as a neighbor.
1344 	 */
1345 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1346 	    nd6_defrouter_list_empty() &&
1347 	    V_nd6_defifindex == ifp->if_index) {
1348 		return (1);
1349 	}
1350 
1351 	return (0);
1352 }
1353 
1354 
1355 /*
1356  * Detect if a given IPv6 address identifies a neighbor on a given link.
1357  * XXX: should take care of the destination of a p2p link?
1358  */
1359 int
1360 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1361 {
1362 	struct llentry *lle;
1363 	int rc = 0;
1364 
1365 	NET_EPOCH_ASSERT();
1366 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1367 	if (nd6_is_new_addr_neighbor(addr, ifp))
1368 		return (1);
1369 
1370 	/*
1371 	 * Even if the address matches none of our addresses, it might be
1372 	 * in the neighbor cache.
1373 	 */
1374 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1375 		LLE_RUNLOCK(lle);
1376 		rc = 1;
1377 	}
1378 	return (rc);
1379 }
1380 
1381 /*
1382  * Free an nd6 llinfo entry.
1383  * Since the function would cause significant changes in the kernel, DO NOT
1384  * make it global, unless you have a strong reason for the change, and are sure
1385  * that the change is safe.
1386  *
1387  * Set noinline to be dtrace-friendly
1388  */
1389 static __noinline void
1390 nd6_free(struct llentry **lnp, int gc)
1391 {
1392 	struct ifnet *ifp;
1393 	struct llentry *ln;
1394 	struct nd_defrouter *dr;
1395 
1396 	ln = *lnp;
1397 	*lnp = NULL;
1398 
1399 	LLE_WLOCK_ASSERT(ln);
1400 	ND6_RLOCK_ASSERT();
1401 
1402 	ifp = lltable_get_ifp(ln->lle_tbl);
1403 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1404 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1405 	else
1406 		dr = NULL;
1407 	ND6_RUNLOCK();
1408 
1409 	if ((ln->la_flags & LLE_DELETED) == 0)
1410 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1411 
1412 	/*
1413 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1414 	 * even though it is not harmful, it was not really necessary.
1415 	 */
1416 
1417 	/* cancel timer */
1418 	nd6_llinfo_settimer_locked(ln, -1);
1419 
1420 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1421 		if (dr != NULL && dr->expire &&
1422 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1423 			/*
1424 			 * If the reason for the deletion is just garbage
1425 			 * collection, and the neighbor is an active default
1426 			 * router, do not delete it.  Instead, reset the GC
1427 			 * timer using the router's lifetime.
1428 			 * Simply deleting the entry would affect default
1429 			 * router selection, which is not necessarily a good
1430 			 * thing, especially when we're using router preference
1431 			 * values.
1432 			 * XXX: the check for ln_state would be redundant,
1433 			 *      but we intentionally keep it just in case.
1434 			 */
1435 			if (dr->expire > time_uptime)
1436 				nd6_llinfo_settimer_locked(ln,
1437 				    (dr->expire - time_uptime) * hz);
1438 			else
1439 				nd6_llinfo_settimer_locked(ln,
1440 				    (long)V_nd6_gctimer * hz);
1441 
1442 			LLE_REMREF(ln);
1443 			LLE_WUNLOCK(ln);
1444 			defrouter_rele(dr);
1445 			return;
1446 		}
1447 
1448 		if (dr) {
1449 			/*
1450 			 * Unreachablity of a router might affect the default
1451 			 * router selection and on-link detection of advertised
1452 			 * prefixes.
1453 			 */
1454 
1455 			/*
1456 			 * Temporarily fake the state to choose a new default
1457 			 * router and to perform on-link determination of
1458 			 * prefixes correctly.
1459 			 * Below the state will be set correctly,
1460 			 * or the entry itself will be deleted.
1461 			 */
1462 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1463 		}
1464 
1465 		if (ln->ln_router || dr) {
1466 
1467 			/*
1468 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1469 			 * rnh and for the calls to pfxlist_onlink_check() and
1470 			 * defrouter_select_fib() in the block further down for calls
1471 			 * into nd6_lookup().  We still hold a ref.
1472 			 */
1473 			LLE_WUNLOCK(ln);
1474 
1475 			/*
1476 			 * rt6_flush must be called whether or not the neighbor
1477 			 * is in the Default Router List.
1478 			 * See a corresponding comment in nd6_na_input().
1479 			 */
1480 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1481 		}
1482 
1483 		if (dr) {
1484 			/*
1485 			 * Since defrouter_select_fib() does not affect the
1486 			 * on-link determination and MIP6 needs the check
1487 			 * before the default router selection, we perform
1488 			 * the check now.
1489 			 */
1490 			pfxlist_onlink_check();
1491 
1492 			/*
1493 			 * Refresh default router list.
1494 			 */
1495 			defrouter_select_fib(dr->ifp->if_fib);
1496 		}
1497 
1498 		/*
1499 		 * If this entry was added by an on-link redirect, remove the
1500 		 * corresponding host route.
1501 		 */
1502 		if (ln->la_flags & LLE_REDIRECT)
1503 			nd6_free_redirect(ln);
1504 
1505 		if (ln->ln_router || dr)
1506 			LLE_WLOCK(ln);
1507 	}
1508 
1509 	/*
1510 	 * Save to unlock. We still hold an extra reference and will not
1511 	 * free(9) in llentry_free() if someone else holds one as well.
1512 	 */
1513 	LLE_WUNLOCK(ln);
1514 	IF_AFDATA_LOCK(ifp);
1515 	LLE_WLOCK(ln);
1516 	/* Guard against race with other llentry_free(). */
1517 	if (ln->la_flags & LLE_LINKED) {
1518 		/* Remove callout reference */
1519 		LLE_REMREF(ln);
1520 		lltable_unlink_entry(ln->lle_tbl, ln);
1521 	}
1522 	IF_AFDATA_UNLOCK(ifp);
1523 
1524 	llentry_free(ln);
1525 	if (dr != NULL)
1526 		defrouter_rele(dr);
1527 }
1528 
1529 static int
1530 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1531 {
1532 
1533 	if (nh->nh_flags & NHF_REDIRECT)
1534 		return (1);
1535 
1536 	return (0);
1537 }
1538 
1539 /*
1540  * Remove the rtentry for the given llentry,
1541  * both of which were installed by a redirect.
1542  */
1543 static void
1544 nd6_free_redirect(const struct llentry *ln)
1545 {
1546 	int fibnum;
1547 	struct sockaddr_in6 sin6;
1548 	struct rt_addrinfo info;
1549 
1550 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1551 	memset(&info, 0, sizeof(info));
1552 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1553 	info.rti_filter = nd6_isdynrte;
1554 
1555 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1556 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1557 }
1558 
1559 /*
1560  * Rejuvenate this function for routing operations related
1561  * processing.
1562  */
1563 void
1564 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1565 {
1566 	struct sockaddr_in6 *gateway;
1567 	struct nd_defrouter *dr;
1568 	struct ifnet *ifp;
1569 
1570 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1571 	ifp = rt->rt_ifp;
1572 
1573 	switch (req) {
1574 	case RTM_ADD:
1575 		break;
1576 
1577 	case RTM_DELETE:
1578 		if (!ifp)
1579 			return;
1580 		/*
1581 		 * Only indirect routes are interesting.
1582 		 */
1583 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1584 			return;
1585 		/*
1586 		 * check for default route
1587 		 */
1588 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1589 		    &SIN6(rt_key(rt))->sin6_addr)) {
1590 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1591 			if (dr != NULL) {
1592 				dr->installed = 0;
1593 				defrouter_rele(dr);
1594 			}
1595 		}
1596 		break;
1597 	}
1598 }
1599 
1600 
1601 int
1602 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1603 {
1604 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1605 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1606 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1607 	struct epoch_tracker et;
1608 	int error = 0;
1609 
1610 	if (ifp->if_afdata[AF_INET6] == NULL)
1611 		return (EPFNOSUPPORT);
1612 	switch (cmd) {
1613 	case OSIOCGIFINFO_IN6:
1614 #define ND	ndi->ndi
1615 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1616 		bzero(&ND, sizeof(ND));
1617 		ND.linkmtu = IN6_LINKMTU(ifp);
1618 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1619 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1620 		ND.reachable = ND_IFINFO(ifp)->reachable;
1621 		ND.retrans = ND_IFINFO(ifp)->retrans;
1622 		ND.flags = ND_IFINFO(ifp)->flags;
1623 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1624 		ND.chlim = ND_IFINFO(ifp)->chlim;
1625 		break;
1626 	case SIOCGIFINFO_IN6:
1627 		ND = *ND_IFINFO(ifp);
1628 		break;
1629 	case SIOCSIFINFO_IN6:
1630 		/*
1631 		 * used to change host variables from userland.
1632 		 * intended for a use on router to reflect RA configurations.
1633 		 */
1634 		/* 0 means 'unspecified' */
1635 		if (ND.linkmtu != 0) {
1636 			if (ND.linkmtu < IPV6_MMTU ||
1637 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1638 				error = EINVAL;
1639 				break;
1640 			}
1641 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1642 		}
1643 
1644 		if (ND.basereachable != 0) {
1645 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1646 
1647 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1648 			if (ND.basereachable != obasereachable)
1649 				ND_IFINFO(ifp)->reachable =
1650 				    ND_COMPUTE_RTIME(ND.basereachable);
1651 		}
1652 		if (ND.retrans != 0)
1653 			ND_IFINFO(ifp)->retrans = ND.retrans;
1654 		if (ND.chlim != 0)
1655 			ND_IFINFO(ifp)->chlim = ND.chlim;
1656 		/* FALLTHROUGH */
1657 	case SIOCSIFINFO_FLAGS:
1658 	{
1659 		struct ifaddr *ifa;
1660 		struct in6_ifaddr *ia;
1661 
1662 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1663 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1664 			/* ifdisabled 1->0 transision */
1665 
1666 			/*
1667 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1668 			 * has an link-local address with IN6_IFF_DUPLICATED,
1669 			 * do not clear ND6_IFF_IFDISABLED.
1670 			 * See RFC 4862, Section 5.4.5.
1671 			 */
1672 			NET_EPOCH_ENTER(et);
1673 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1674 				if (ifa->ifa_addr->sa_family != AF_INET6)
1675 					continue;
1676 				ia = (struct in6_ifaddr *)ifa;
1677 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1678 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1679 					break;
1680 			}
1681 			NET_EPOCH_EXIT(et);
1682 
1683 			if (ifa != NULL) {
1684 				/* LLA is duplicated. */
1685 				ND.flags |= ND6_IFF_IFDISABLED;
1686 				log(LOG_ERR, "Cannot enable an interface"
1687 				    " with a link-local address marked"
1688 				    " duplicate.\n");
1689 			} else {
1690 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1691 				if (ifp->if_flags & IFF_UP)
1692 					in6_if_up(ifp);
1693 			}
1694 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1695 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1696 			/* ifdisabled 0->1 transision */
1697 			/* Mark all IPv6 address as tentative. */
1698 
1699 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1700 			if (V_ip6_dad_count > 0 &&
1701 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1702 				NET_EPOCH_ENTER(et);
1703 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1704 				    ifa_link) {
1705 					if (ifa->ifa_addr->sa_family !=
1706 					    AF_INET6)
1707 						continue;
1708 					ia = (struct in6_ifaddr *)ifa;
1709 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1710 				}
1711 				NET_EPOCH_EXIT(et);
1712 			}
1713 		}
1714 
1715 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1716 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1717 				/* auto_linklocal 0->1 transision */
1718 
1719 				/* If no link-local address on ifp, configure */
1720 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1721 				in6_ifattach(ifp, NULL);
1722 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1723 			    ifp->if_flags & IFF_UP) {
1724 				/*
1725 				 * When the IF already has
1726 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1727 				 * address is assigned, and IFF_UP, try to
1728 				 * assign one.
1729 				 */
1730 				NET_EPOCH_ENTER(et);
1731 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1732 				    ifa_link) {
1733 					if (ifa->ifa_addr->sa_family !=
1734 					    AF_INET6)
1735 						continue;
1736 					ia = (struct in6_ifaddr *)ifa;
1737 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1738 						break;
1739 				}
1740 				NET_EPOCH_EXIT(et);
1741 				if (ifa != NULL)
1742 					/* No LLA is configured. */
1743 					in6_ifattach(ifp, NULL);
1744 			}
1745 		}
1746 		ND_IFINFO(ifp)->flags = ND.flags;
1747 		break;
1748 	}
1749 #undef ND
1750 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1751 		/* sync kernel routing table with the default router list */
1752 		defrouter_reset();
1753 		defrouter_select_fib(RT_ALL_FIBS);
1754 		break;
1755 	case SIOCSPFXFLUSH_IN6:
1756 	{
1757 		/* flush all the prefix advertised by routers */
1758 		struct in6_ifaddr *ia, *ia_next;
1759 		struct nd_prefix *pr, *next;
1760 		struct nd_prhead prl;
1761 
1762 		LIST_INIT(&prl);
1763 
1764 		ND6_WLOCK();
1765 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1766 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1767 				continue; /* XXX */
1768 			nd6_prefix_unlink(pr, &prl);
1769 		}
1770 		ND6_WUNLOCK();
1771 
1772 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1773 			LIST_REMOVE(pr, ndpr_entry);
1774 			/* XXXRW: in6_ifaddrhead locking. */
1775 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1776 			    ia_next) {
1777 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1778 					continue;
1779 
1780 				if (ia->ia6_ndpr == pr)
1781 					in6_purgeaddr(&ia->ia_ifa);
1782 			}
1783 			nd6_prefix_del(pr);
1784 		}
1785 		break;
1786 	}
1787 	case SIOCSRTRFLUSH_IN6:
1788 	{
1789 		/* flush all the default routers */
1790 
1791 		defrouter_reset();
1792 		nd6_defrouter_flush_all();
1793 		defrouter_select_fib(RT_ALL_FIBS);
1794 		break;
1795 	}
1796 	case SIOCGNBRINFO_IN6:
1797 	{
1798 		struct llentry *ln;
1799 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1800 
1801 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1802 			return (error);
1803 
1804 		NET_EPOCH_ENTER(et);
1805 		ln = nd6_lookup(&nb_addr, 0, ifp);
1806 		NET_EPOCH_EXIT(et);
1807 
1808 		if (ln == NULL) {
1809 			error = EINVAL;
1810 			break;
1811 		}
1812 		nbi->state = ln->ln_state;
1813 		nbi->asked = ln->la_asked;
1814 		nbi->isrouter = ln->ln_router;
1815 		if (ln->la_expire == 0)
1816 			nbi->expire = 0;
1817 		else
1818 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1819 			    (time_second - time_uptime);
1820 		LLE_RUNLOCK(ln);
1821 		break;
1822 	}
1823 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1824 		ndif->ifindex = V_nd6_defifindex;
1825 		break;
1826 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1827 		return (nd6_setdefaultiface(ndif->ifindex));
1828 	}
1829 	return (error);
1830 }
1831 
1832 /*
1833  * Calculates new isRouter value based on provided parameters and
1834  * returns it.
1835  */
1836 static int
1837 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1838     int ln_router)
1839 {
1840 
1841 	/*
1842 	 * ICMP6 type dependent behavior.
1843 	 *
1844 	 * NS: clear IsRouter if new entry
1845 	 * RS: clear IsRouter
1846 	 * RA: set IsRouter if there's lladdr
1847 	 * redir: clear IsRouter if new entry
1848 	 *
1849 	 * RA case, (1):
1850 	 * The spec says that we must set IsRouter in the following cases:
1851 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1852 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1853 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1854 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1855 	 * neighbor cache, this is similar to (6).
1856 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1857 	 *
1858 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1859 	 *							D R
1860 	 *	0	n	n	(1)	c   ?     s
1861 	 *	0	y	n	(2)	c   s     s
1862 	 *	0	n	y	(3)	c   s     s
1863 	 *	0	y	y	(4)	c   s     s
1864 	 *	0	y	y	(5)	c   s     s
1865 	 *	1	--	n	(6) c	c	c s
1866 	 *	1	--	y	(7) c	c   s	c s
1867 	 *
1868 	 *					(c=clear s=set)
1869 	 */
1870 	switch (type & 0xff) {
1871 	case ND_NEIGHBOR_SOLICIT:
1872 		/*
1873 		 * New entry must have is_router flag cleared.
1874 		 */
1875 		if (is_new)					/* (6-7) */
1876 			ln_router = 0;
1877 		break;
1878 	case ND_REDIRECT:
1879 		/*
1880 		 * If the icmp is a redirect to a better router, always set the
1881 		 * is_router flag.  Otherwise, if the entry is newly created,
1882 		 * clear the flag.  [RFC 2461, sec 8.3]
1883 		 */
1884 		if (code == ND_REDIRECT_ROUTER)
1885 			ln_router = 1;
1886 		else {
1887 			if (is_new)				/* (6-7) */
1888 				ln_router = 0;
1889 		}
1890 		break;
1891 	case ND_ROUTER_SOLICIT:
1892 		/*
1893 		 * is_router flag must always be cleared.
1894 		 */
1895 		ln_router = 0;
1896 		break;
1897 	case ND_ROUTER_ADVERT:
1898 		/*
1899 		 * Mark an entry with lladdr as a router.
1900 		 */
1901 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1902 		    (is_new && new_addr)) {			/* (7) */
1903 			ln_router = 1;
1904 		}
1905 		break;
1906 	}
1907 
1908 	return (ln_router);
1909 }
1910 
1911 /*
1912  * Create neighbor cache entry and cache link-layer address,
1913  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1914  *
1915  * type - ICMP6 type
1916  * code - type dependent information
1917  *
1918  */
1919 void
1920 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1921     int lladdrlen, int type, int code)
1922 {
1923 	struct llentry *ln = NULL, *ln_tmp;
1924 	int is_newentry;
1925 	int do_update;
1926 	int olladdr;
1927 	int llchange;
1928 	int flags;
1929 	uint16_t router = 0;
1930 	struct sockaddr_in6 sin6;
1931 	struct mbuf *chain = NULL;
1932 	u_char linkhdr[LLE_MAX_LINKHDR];
1933 	size_t linkhdrsize;
1934 	int lladdr_off;
1935 
1936 	NET_EPOCH_ASSERT();
1937 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1938 
1939 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1940 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1941 
1942 	/* nothing must be updated for unspecified address */
1943 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1944 		return;
1945 
1946 	/*
1947 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1948 	 * the caller.
1949 	 *
1950 	 * XXX If the link does not have link-layer adderss, what should
1951 	 * we do? (ifp->if_addrlen == 0)
1952 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1953 	 * description on it in NS section (RFC 2461 7.2.3).
1954 	 */
1955 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1956 	ln = nd6_lookup(from, flags, ifp);
1957 	is_newentry = 0;
1958 	if (ln == NULL) {
1959 		flags |= LLE_EXCLUSIVE;
1960 		ln = nd6_alloc(from, 0, ifp);
1961 		if (ln == NULL)
1962 			return;
1963 
1964 		/*
1965 		 * Since we already know all the data for the new entry,
1966 		 * fill it before insertion.
1967 		 */
1968 		if (lladdr != NULL) {
1969 			linkhdrsize = sizeof(linkhdr);
1970 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1971 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1972 				return;
1973 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1974 			    lladdr_off);
1975 		}
1976 
1977 		IF_AFDATA_WLOCK(ifp);
1978 		LLE_WLOCK(ln);
1979 		/* Prefer any existing lle over newly-created one */
1980 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1981 		if (ln_tmp == NULL)
1982 			lltable_link_entry(LLTABLE6(ifp), ln);
1983 		IF_AFDATA_WUNLOCK(ifp);
1984 		if (ln_tmp == NULL) {
1985 			/* No existing lle, mark as new entry (6,7) */
1986 			is_newentry = 1;
1987 			if (lladdr != NULL) {	/* (7) */
1988 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1989 				EVENTHANDLER_INVOKE(lle_event, ln,
1990 				    LLENTRY_RESOLVED);
1991 			}
1992 		} else {
1993 			lltable_free_entry(LLTABLE6(ifp), ln);
1994 			ln = ln_tmp;
1995 			ln_tmp = NULL;
1996 		}
1997 	}
1998 	/* do nothing if static ndp is set */
1999 	if ((ln->la_flags & LLE_STATIC)) {
2000 		if (flags & LLE_EXCLUSIVE)
2001 			LLE_WUNLOCK(ln);
2002 		else
2003 			LLE_RUNLOCK(ln);
2004 		return;
2005 	}
2006 
2007 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2008 	if (olladdr && lladdr) {
2009 		llchange = bcmp(lladdr, ln->ll_addr,
2010 		    ifp->if_addrlen);
2011 	} else if (!olladdr && lladdr)
2012 		llchange = 1;
2013 	else
2014 		llchange = 0;
2015 
2016 	/*
2017 	 * newentry olladdr  lladdr  llchange	(*=record)
2018 	 *	0	n	n	--	(1)
2019 	 *	0	y	n	--	(2)
2020 	 *	0	n	y	y	(3) * STALE
2021 	 *	0	y	y	n	(4) *
2022 	 *	0	y	y	y	(5) * STALE
2023 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2024 	 *	1	--	y	--	(7) * STALE
2025 	 */
2026 
2027 	do_update = 0;
2028 	if (is_newentry == 0 && llchange != 0) {
2029 		do_update = 1;	/* (3,5) */
2030 
2031 		/*
2032 		 * Record source link-layer address
2033 		 * XXX is it dependent to ifp->if_type?
2034 		 */
2035 		linkhdrsize = sizeof(linkhdr);
2036 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2037 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2038 			return;
2039 
2040 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2041 		    lladdr_off) == 0) {
2042 			/* Entry was deleted */
2043 			return;
2044 		}
2045 
2046 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2047 
2048 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2049 
2050 		if (ln->la_hold != NULL)
2051 			nd6_grab_holdchain(ln, &chain, &sin6);
2052 	}
2053 
2054 	/* Calculates new router status */
2055 	router = nd6_is_router(type, code, is_newentry, olladdr,
2056 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2057 
2058 	ln->ln_router = router;
2059 	/* Mark non-router redirects with special flag */
2060 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2061 		ln->la_flags |= LLE_REDIRECT;
2062 
2063 	if (flags & LLE_EXCLUSIVE)
2064 		LLE_WUNLOCK(ln);
2065 	else
2066 		LLE_RUNLOCK(ln);
2067 
2068 	if (chain != NULL)
2069 		nd6_flush_holdchain(ifp, chain, &sin6);
2070 
2071 	/*
2072 	 * When the link-layer address of a router changes, select the
2073 	 * best router again.  In particular, when the neighbor entry is newly
2074 	 * created, it might affect the selection policy.
2075 	 * Question: can we restrict the first condition to the "is_newentry"
2076 	 * case?
2077 	 * XXX: when we hear an RA from a new router with the link-layer
2078 	 * address option, defrouter_select_fib() is called twice, since
2079 	 * defrtrlist_update called the function as well.  However, I believe
2080 	 * we can compromise the overhead, since it only happens the first
2081 	 * time.
2082 	 * XXX: although defrouter_select_fib() should not have a bad effect
2083 	 * for those are not autoconfigured hosts, we explicitly avoid such
2084 	 * cases for safety.
2085 	 */
2086 	if ((do_update || is_newentry) && router &&
2087 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2088 		/*
2089 		 * guaranteed recursion
2090 		 */
2091 		defrouter_select_fib(ifp->if_fib);
2092 	}
2093 }
2094 
2095 static void
2096 nd6_slowtimo(void *arg)
2097 {
2098 	struct epoch_tracker et;
2099 	CURVNET_SET((struct vnet *) arg);
2100 	struct nd_ifinfo *nd6if;
2101 	struct ifnet *ifp;
2102 
2103 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2104 	    nd6_slowtimo, curvnet);
2105 	NET_EPOCH_ENTER(et);
2106 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2107 		if (ifp->if_afdata[AF_INET6] == NULL)
2108 			continue;
2109 		nd6if = ND_IFINFO(ifp);
2110 		if (nd6if->basereachable && /* already initialized */
2111 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2112 			/*
2113 			 * Since reachable time rarely changes by router
2114 			 * advertisements, we SHOULD insure that a new random
2115 			 * value gets recomputed at least once every few hours.
2116 			 * (RFC 2461, 6.3.4)
2117 			 */
2118 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2119 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2120 		}
2121 	}
2122 	NET_EPOCH_EXIT(et);
2123 	CURVNET_RESTORE();
2124 }
2125 
2126 void
2127 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2128     struct sockaddr_in6 *sin6)
2129 {
2130 
2131 	LLE_WLOCK_ASSERT(ln);
2132 
2133 	*chain = ln->la_hold;
2134 	ln->la_hold = NULL;
2135 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2136 
2137 	if (ln->ln_state == ND6_LLINFO_STALE) {
2138 
2139 		/*
2140 		 * The first time we send a packet to a
2141 		 * neighbor whose entry is STALE, we have
2142 		 * to change the state to DELAY and a sets
2143 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2144 		 * seconds to ensure do neighbor unreachability
2145 		 * detection on expiration.
2146 		 * (RFC 2461 7.3.3)
2147 		 */
2148 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2149 	}
2150 }
2151 
2152 int
2153 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2154     struct sockaddr_in6 *dst, struct route *ro)
2155 {
2156 	int error;
2157 	int ip6len;
2158 	struct ip6_hdr *ip6;
2159 	struct m_tag *mtag;
2160 
2161 #ifdef MAC
2162 	mac_netinet6_nd6_send(ifp, m);
2163 #endif
2164 
2165 	/*
2166 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2167 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2168 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2169 	 * to be diverted to user space.  When re-injected into the kernel,
2170 	 * send_output() will directly dispatch them to the outgoing interface.
2171 	 */
2172 	if (send_sendso_input_hook != NULL) {
2173 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2174 		if (mtag != NULL) {
2175 			ip6 = mtod(m, struct ip6_hdr *);
2176 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2177 			/* Use the SEND socket */
2178 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2179 			    ip6len);
2180 			/* -1 == no app on SEND socket */
2181 			if (error == 0 || error != -1)
2182 			    return (error);
2183 		}
2184 	}
2185 
2186 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2187 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2188 	    mtod(m, struct ip6_hdr *));
2189 
2190 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2191 		origifp = ifp;
2192 
2193 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2194 	return (error);
2195 }
2196 
2197 /*
2198  * Lookup link headerfor @sa_dst address. Stores found
2199  * data in @desten buffer. Copy of lle ln_flags can be also
2200  * saved in @pflags if @pflags is non-NULL.
2201  *
2202  * If destination LLE does not exists or lle state modification
2203  * is required, call "slow" version.
2204  *
2205  * Return values:
2206  * - 0 on success (address copied to buffer).
2207  * - EWOULDBLOCK (no local error, but address is still unresolved)
2208  * - other errors (alloc failure, etc)
2209  */
2210 int
2211 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2212     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2213     struct llentry **plle)
2214 {
2215 	struct llentry *ln = NULL;
2216 	const struct sockaddr_in6 *dst6;
2217 
2218 	NET_EPOCH_ASSERT();
2219 
2220 	if (pflags != NULL)
2221 		*pflags = 0;
2222 
2223 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2224 
2225 	/* discard the packet if IPv6 operation is disabled on the interface */
2226 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2227 		m_freem(m);
2228 		return (ENETDOWN); /* better error? */
2229 	}
2230 
2231 	if (m != NULL && m->m_flags & M_MCAST) {
2232 		switch (ifp->if_type) {
2233 		case IFT_ETHER:
2234 		case IFT_L2VLAN:
2235 		case IFT_BRIDGE:
2236 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2237 						 desten);
2238 			return (0);
2239 		default:
2240 			m_freem(m);
2241 			return (EAFNOSUPPORT);
2242 		}
2243 	}
2244 
2245 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2246 	    ifp);
2247 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2248 		/* Entry found, let's copy lle info */
2249 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2250 		if (pflags != NULL)
2251 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2252 		/* Check if we have feedback request from nd6 timer */
2253 		if (ln->r_skip_req != 0) {
2254 			LLE_REQ_LOCK(ln);
2255 			ln->r_skip_req = 0; /* Notify that entry was used */
2256 			ln->lle_hittime = time_uptime;
2257 			LLE_REQ_UNLOCK(ln);
2258 		}
2259 		if (plle) {
2260 			LLE_ADDREF(ln);
2261 			*plle = ln;
2262 			LLE_WUNLOCK(ln);
2263 		}
2264 		return (0);
2265 	} else if (plle && ln)
2266 		LLE_WUNLOCK(ln);
2267 
2268 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2269 }
2270 
2271 
2272 /*
2273  * Do L2 address resolution for @sa_dst address. Stores found
2274  * address in @desten buffer. Copy of lle ln_flags can be also
2275  * saved in @pflags if @pflags is non-NULL.
2276  *
2277  * Heavy version.
2278  * Function assume that destination LLE does not exist,
2279  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2280  *
2281  * Set noinline to be dtrace-friendly
2282  */
2283 static __noinline int
2284 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2285     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2286     struct llentry **plle)
2287 {
2288 	struct llentry *lle = NULL, *lle_tmp;
2289 	struct in6_addr *psrc, src;
2290 	int send_ns, ll_len;
2291 	char *lladdr;
2292 
2293 	NET_EPOCH_ASSERT();
2294 
2295 	/*
2296 	 * Address resolution or Neighbor Unreachability Detection
2297 	 * for the next hop.
2298 	 * At this point, the destination of the packet must be a unicast
2299 	 * or an anycast address(i.e. not a multicast).
2300 	 */
2301 	if (lle == NULL) {
2302 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2303 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2304 			/*
2305 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2306 			 * the condition below is not very efficient.  But we believe
2307 			 * it is tolerable, because this should be a rare case.
2308 			 */
2309 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2310 			if (lle == NULL) {
2311 				char ip6buf[INET6_ADDRSTRLEN];
2312 				log(LOG_DEBUG,
2313 				    "nd6_output: can't allocate llinfo for %s "
2314 				    "(ln=%p)\n",
2315 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2316 				m_freem(m);
2317 				return (ENOBUFS);
2318 			}
2319 
2320 			IF_AFDATA_WLOCK(ifp);
2321 			LLE_WLOCK(lle);
2322 			/* Prefer any existing entry over newly-created one */
2323 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2324 			if (lle_tmp == NULL)
2325 				lltable_link_entry(LLTABLE6(ifp), lle);
2326 			IF_AFDATA_WUNLOCK(ifp);
2327 			if (lle_tmp != NULL) {
2328 				lltable_free_entry(LLTABLE6(ifp), lle);
2329 				lle = lle_tmp;
2330 				lle_tmp = NULL;
2331 			}
2332 		}
2333 	}
2334 	if (lle == NULL) {
2335 		m_freem(m);
2336 		return (ENOBUFS);
2337 	}
2338 
2339 	LLE_WLOCK_ASSERT(lle);
2340 
2341 	/*
2342 	 * The first time we send a packet to a neighbor whose entry is
2343 	 * STALE, we have to change the state to DELAY and a sets a timer to
2344 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2345 	 * neighbor unreachability detection on expiration.
2346 	 * (RFC 2461 7.3.3)
2347 	 */
2348 	if (lle->ln_state == ND6_LLINFO_STALE)
2349 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2350 
2351 	/*
2352 	 * If the neighbor cache entry has a state other than INCOMPLETE
2353 	 * (i.e. its link-layer address is already resolved), just
2354 	 * send the packet.
2355 	 */
2356 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2357 		if (flags & LLE_ADDRONLY) {
2358 			lladdr = lle->ll_addr;
2359 			ll_len = ifp->if_addrlen;
2360 		} else {
2361 			lladdr = lle->r_linkdata;
2362 			ll_len = lle->r_hdrlen;
2363 		}
2364 		bcopy(lladdr, desten, ll_len);
2365 		if (pflags != NULL)
2366 			*pflags = lle->la_flags;
2367 		if (plle) {
2368 			LLE_ADDREF(lle);
2369 			*plle = lle;
2370 		}
2371 		LLE_WUNLOCK(lle);
2372 		return (0);
2373 	}
2374 
2375 	/*
2376 	 * There is a neighbor cache entry, but no ethernet address
2377 	 * response yet.  Append this latest packet to the end of the
2378 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2379 	 * the oldest packet in the queue will be removed.
2380 	 */
2381 
2382 	if (lle->la_hold != NULL) {
2383 		struct mbuf *m_hold;
2384 		int i;
2385 
2386 		i = 0;
2387 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2388 			i++;
2389 			if (m_hold->m_nextpkt == NULL) {
2390 				m_hold->m_nextpkt = m;
2391 				break;
2392 			}
2393 		}
2394 		while (i >= V_nd6_maxqueuelen) {
2395 			m_hold = lle->la_hold;
2396 			lle->la_hold = lle->la_hold->m_nextpkt;
2397 			m_freem(m_hold);
2398 			i--;
2399 		}
2400 	} else {
2401 		lle->la_hold = m;
2402 	}
2403 
2404 	/*
2405 	 * If there has been no NS for the neighbor after entering the
2406 	 * INCOMPLETE state, send the first solicitation.
2407 	 * Note that for newly-created lle la_asked will be 0,
2408 	 * so we will transition from ND6_LLINFO_NOSTATE to
2409 	 * ND6_LLINFO_INCOMPLETE state here.
2410 	 */
2411 	psrc = NULL;
2412 	send_ns = 0;
2413 	if (lle->la_asked == 0) {
2414 		lle->la_asked++;
2415 		send_ns = 1;
2416 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2417 
2418 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2419 	}
2420 	LLE_WUNLOCK(lle);
2421 	if (send_ns != 0)
2422 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2423 
2424 	return (EWOULDBLOCK);
2425 }
2426 
2427 /*
2428  * Do L2 address resolution for @sa_dst address. Stores found
2429  * address in @desten buffer. Copy of lle ln_flags can be also
2430  * saved in @pflags if @pflags is non-NULL.
2431  *
2432  * Return values:
2433  * - 0 on success (address copied to buffer).
2434  * - EWOULDBLOCK (no local error, but address is still unresolved)
2435  * - other errors (alloc failure, etc)
2436  */
2437 int
2438 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2439     char *desten, uint32_t *pflags)
2440 {
2441 	int error;
2442 
2443 	flags |= LLE_ADDRONLY;
2444 	error = nd6_resolve_slow(ifp, flags, NULL,
2445 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2446 	return (error);
2447 }
2448 
2449 int
2450 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2451     struct sockaddr_in6 *dst)
2452 {
2453 	struct mbuf *m, *m_head;
2454 	int error = 0;
2455 
2456 	m_head = chain;
2457 
2458 	while (m_head) {
2459 		m = m_head;
2460 		m_head = m_head->m_nextpkt;
2461 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2462 	}
2463 
2464 	/*
2465 	 * XXX
2466 	 * note that intermediate errors are blindly ignored
2467 	 */
2468 	return (error);
2469 }
2470 
2471 static int
2472 nd6_need_cache(struct ifnet *ifp)
2473 {
2474 	/*
2475 	 * XXX: we currently do not make neighbor cache on any interface
2476 	 * other than Ethernet and GIF.
2477 	 *
2478 	 * RFC2893 says:
2479 	 * - unidirectional tunnels needs no ND
2480 	 */
2481 	switch (ifp->if_type) {
2482 	case IFT_ETHER:
2483 	case IFT_IEEE1394:
2484 	case IFT_L2VLAN:
2485 	case IFT_INFINIBAND:
2486 	case IFT_BRIDGE:
2487 	case IFT_PROPVIRTUAL:
2488 		return (1);
2489 	default:
2490 		return (0);
2491 	}
2492 }
2493 
2494 /*
2495  * Add pernament ND6 link-layer record for given
2496  * interface address.
2497  *
2498  * Very similar to IPv4 arp_ifinit(), but:
2499  * 1) IPv6 DAD is performed in different place
2500  * 2) It is called by IPv6 protocol stack in contrast to
2501  * arp_ifinit() which is typically called in SIOCSIFADDR
2502  * driver ioctl handler.
2503  *
2504  */
2505 int
2506 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2507 {
2508 	struct ifnet *ifp;
2509 	struct llentry *ln, *ln_tmp;
2510 	struct sockaddr *dst;
2511 
2512 	ifp = ia->ia_ifa.ifa_ifp;
2513 	if (nd6_need_cache(ifp) == 0)
2514 		return (0);
2515 
2516 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2517 	dst = (struct sockaddr *)&ia->ia_addr;
2518 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2519 	if (ln == NULL)
2520 		return (ENOBUFS);
2521 
2522 	IF_AFDATA_WLOCK(ifp);
2523 	LLE_WLOCK(ln);
2524 	/* Unlink any entry if exists */
2525 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2526 	if (ln_tmp != NULL)
2527 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2528 	lltable_link_entry(LLTABLE6(ifp), ln);
2529 	IF_AFDATA_WUNLOCK(ifp);
2530 
2531 	if (ln_tmp != NULL)
2532 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2533 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2534 
2535 	LLE_WUNLOCK(ln);
2536 	if (ln_tmp != NULL)
2537 		llentry_free(ln_tmp);
2538 
2539 	return (0);
2540 }
2541 
2542 /*
2543  * Removes either all lle entries for given @ia, or lle
2544  * corresponding to @ia address.
2545  */
2546 void
2547 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2548 {
2549 	struct sockaddr_in6 mask, addr;
2550 	struct sockaddr *saddr, *smask;
2551 	struct ifnet *ifp;
2552 
2553 	ifp = ia->ia_ifa.ifa_ifp;
2554 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2555 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2556 	saddr = (struct sockaddr *)&addr;
2557 	smask = (struct sockaddr *)&mask;
2558 
2559 	if (all != 0)
2560 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2561 	else
2562 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2563 }
2564 
2565 static void
2566 clear_llinfo_pqueue(struct llentry *ln)
2567 {
2568 	struct mbuf *m_hold, *m_hold_next;
2569 
2570 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2571 		m_hold_next = m_hold->m_nextpkt;
2572 		m_freem(m_hold);
2573 	}
2574 
2575 	ln->la_hold = NULL;
2576 }
2577 
2578 static int
2579 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2580 {
2581 	struct in6_prefix p;
2582 	struct sockaddr_in6 s6;
2583 	struct nd_prefix *pr;
2584 	struct nd_pfxrouter *pfr;
2585 	time_t maxexpire;
2586 	int error;
2587 	char ip6buf[INET6_ADDRSTRLEN];
2588 
2589 	if (req->newptr)
2590 		return (EPERM);
2591 
2592 	error = sysctl_wire_old_buffer(req, 0);
2593 	if (error != 0)
2594 		return (error);
2595 
2596 	bzero(&p, sizeof(p));
2597 	p.origin = PR_ORIG_RA;
2598 	bzero(&s6, sizeof(s6));
2599 	s6.sin6_family = AF_INET6;
2600 	s6.sin6_len = sizeof(s6);
2601 
2602 	ND6_RLOCK();
2603 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2604 		p.prefix = pr->ndpr_prefix;
2605 		if (sa6_recoverscope(&p.prefix)) {
2606 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2607 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2608 			/* XXX: press on... */
2609 		}
2610 		p.raflags = pr->ndpr_raf;
2611 		p.prefixlen = pr->ndpr_plen;
2612 		p.vltime = pr->ndpr_vltime;
2613 		p.pltime = pr->ndpr_pltime;
2614 		p.if_index = pr->ndpr_ifp->if_index;
2615 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2616 			p.expire = 0;
2617 		else {
2618 			/* XXX: we assume time_t is signed. */
2619 			maxexpire = (-1) &
2620 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2621 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2622 				p.expire = pr->ndpr_lastupdate +
2623 				    pr->ndpr_vltime +
2624 				    (time_second - time_uptime);
2625 			else
2626 				p.expire = maxexpire;
2627 		}
2628 		p.refcnt = pr->ndpr_addrcnt;
2629 		p.flags = pr->ndpr_stateflags;
2630 		p.advrtrs = 0;
2631 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2632 			p.advrtrs++;
2633 		error = SYSCTL_OUT(req, &p, sizeof(p));
2634 		if (error != 0)
2635 			break;
2636 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2637 			s6.sin6_addr = pfr->router->rtaddr;
2638 			if (sa6_recoverscope(&s6))
2639 				log(LOG_ERR,
2640 				    "scope error in prefix list (%s)\n",
2641 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2642 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2643 			if (error != 0)
2644 				goto out;
2645 		}
2646 	}
2647 out:
2648 	ND6_RUNLOCK();
2649 	return (error);
2650 }
2651 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2652 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2653 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2654 	"NDP prefix list");
2655 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2656 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2657 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2658 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2659